Ma

Saw:

1-! U A...
“a: P. . .ﬁeir.
an!!! .
2| . ~ n. ¢v
a. 31...; He
.1- 31. L I

. a.
1.. r
5:3»...

.5...
I...

{grail-1....
25‘ no...“

thy-.59.? .
9...: {WEE-1.9
.0

n“ .i

 

2&3.

2
5:.

t.
nib..."
x. 2 vs

.

s. .
«we: 54...: . »

«H‘h‘it. 1.1.2.3

a3. 15.3. hi...”

 

 

 

T .: ...'. An.nnpa£§!:i...~ql
12.83..

;.. h... a. “ﬁngﬁﬁmﬁmﬁ

a?

‘31
.. :4
(JD

; LIBRARY
-- o c» ”-i Michigan §tate
Univemty

This is to certify that the
dissertation entitled

VIRTUAL ORGANIZATION BASED DISTRIBUTED
ENVIRONMENTAL SPATIAL DECISION SUPPORT
SYSTEMS: APPLICATIONS IN WATERSHED
MANAGEMENT

presented by

Yl SHI

has been accepted towards fulﬁllment
of the requirements for the

 

 

    

 

Ph.D. degree in Geography
/ / ’/ if 4/.- /"J
/ -. {‘1 71"}; \4.’ A
m. «A ~ 4M- are

‘b ' Major Professor’s yﬁnature

/.Z V / / ’ if) 33

 

Date

MSU is an Afﬁnnative Action/Equal Opportunity Employer

._ -.-a------------.-v--- --

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 Kthrolecc8-Pres/CIRC/DaIeDue.indd

 

VIRTUAL ORGANIZATION BASED DISTRIBUTED ENVIRONMENTAL SPATIAL
DECISION SUPPORT SYSTEMS: APPLICATIONS IN WATERSHED
MANAGEMENT

By

Yi Shi

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Geography

2008

ABSTRACT

VIRTUAL ORGANIZATION BASED DISTRIBUTED ENVIRONMENTAL SPATIAL

DECISION SUPPORT SYSTEMS: APPLICATIONS IN WATERSHED

MANAGEMENT
By
Yi Shi

Watershed management as a consensus-building process demands a holistic
integrative approach to consider the impacts of various human activities in a watershed
on water resources and other related natural resources. The challenge of effectively
integrating scientiﬁc information from different disciplines into watershed environmental
decision making process remains an outstanding challenge for environmental
practitioners. In this research, the author argues that a collaborative online Virtual
Organization (V 0) based Watershed Management Spatial Decision Support System
(SDSS) capable of integrating distributed databases and computational models offers
superior opportunities for holistic watershed management. This system can utilize the
latest Grid computing technology and provide an effective way to embed science in the
decision making process by improving the communication between the scientiﬁc and the
policy sector and between decision-makers and involved stakeholders and translating
oﬁen complicated modeling results into an easy to understand format for environmental
decision makers. The requirement analysis for a VO-based watershed management SDSS
is conducted ﬁrst. A general Intemet-based integration framework for the VO-based
watershed management SDSS is then conceptualized based on the requirements speciﬁed.
The implementation of an example VO-based watershed management SDSS within this

framework for sediment runoff reduction in two watersheds in northern Indiana is

described in detail to illustrate the advantages of VO-based watershed management SDSS.
The parallelization of a selected water quality model and subsequent performance
analysis demonstrates how Grid computing technology can be utilized as underlying
technical infrastructure for a VO-based watershed management SDSS. Finally, design,
implementation and application of a new fundamental DEM-based ﬂow analysis
algorithm are reported and statistical analysis of application results are done to show its
improvement over existing methods. The integration of this new algorithm into VO-based

watershed management SDSS can beneﬁt decision makers and other users immediately.

ACKNOWLEDGEMENTS

This dissertation could not have come to fruition without helps from many people
around me. I would like to thank Dr. Ashton Shortridge for serving as my supervisor and
guiding me throughout my academic program. He and the other faculty members, Dr. Jon
Bartholic, Dr. Jiaguo Qi and Dr. David Lusch, patiently directed me through the
dissertation process. I thank them all. My special thanks go to Dr. Jon Bartholic for
providing me with an excellent working environment at the Institute of Water Research
and kindly allowing me to combine work and dissertation research. I am also very
grateful to all my colleagues at the Institute of Water Research for all the assistance and
friendship they gave me through the years. Finally, I want to thank my parents in China

and my family here in United States for all they have done for me in my life.

iv

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................ vii
LIST OF FIGURES ......................................................................................................... viii
Chapter 1. Challenges in Distributed Environmental Decision Support Systems .............. 1
1 .1 . Introduction .............................................................................................................. l
1.2. Distributed Computing, Geoprocessing and Spatial Decision Support ................... 8
1.2.1. Distributed Computing ..................................................................................... 8
1.2.2. Geoprocessing and Spatial Decision Support ................................................. 10
1.2.3. GRID-based Geospatial Applications ............................................................. 13
1.3. Watershed Management ......................................................................................... 19
1.3.1. Watersheds ...................................................................................................... 19
1.3.2. Watershed Management Process .................................................................... 20
1.3.3. A F our-phase Watershed Management Process and its Beneﬁts .................... 22
1.3.4. Watershed Partnership .................................................................................... 26
1.4. Objectives and Organization .................................................................................. 30
Chapter 2. Requirement Analysis for a VO-based Watershed Management SDSS ........ 32
2.1 . Introduction ............................................................................................................ 32
2.2. Actor Network Theory (ANT) ............................................................................... 36
2.3. Virtual Organization (V O) ..................................................................................... 39
2.4. Spatial Decision Support System for Watershed Management ............................. 43
2.4.1. Overview ......................................................................................................... 43
2.4.2. Components of SDSS ..................................................................................... 47
2.5. Requirement Analysis for a VO-based Watershed Management SDSS ................ 53
2.5.1. Organizational Requirement Analysis ............................................................ 53
2.5.2. Technical Requirement Analysis .................................................................... 60
2.6. Conclusions ............................................................................................................ 69
Chapter 3. Grid Computing and VO-based Watershed Management SDSS .................... 71
3.1 . Introduction ............................................................................................................ 71
3.2. A Grid based Integration Framework for the VO-based Watershed Management
SDSS ............................................................................................................................. 72
3.3. An Example VO-based Watershed Management SDSS ........................................ 75
3.3.1. Erosion modeling ............................................................................................ 82
3.3.2. Web-GIS based SDSS for erosion and water quality management ................ 84
3.3.3. Interoperability ................................................................................................ 86
3.3.4. Burns Ditch Trail Creek Watershed Management VO ................................... 88
3.3.5. Use Case .......................................................................................................... 90
3.3.6. Training Workshop ......................................................................................... 94
3.3.7. Validation of VO-based SDSS Approach for Watershed Management ......... 99
3.4. Parallelization of a Water Quality Model ............................................................ 102

3.5. Conclusions .......................................................................................................... 108

Chapter 4. A New DEM-based Flow Analysis Algorithm ............................................. 110
4.1. Introduction .......................................................................................................... 110
4.2. Hydrologic Modeling Using Digital Elevation Data ........................................... 111
4.3. New Flow Routing Algorithm ............................................................................. 119
4.4. Application to Different Physiographic Regions ................................................. 123
4.5. Spatial Variation in Flow Accumulation ............................................................. 129
4.6. Contrasting Algorithm Performance on Hydrologic Applications ...................... 152
4.7. Validation of FlowNet Algorithm ........................................................................ 162
4.8. Conclusions .......................................................................................................... 17 3

Chapter 5. Summary and Conclusions ............................................................................ 176
5.1 . Introduction .......................................................................................................... 176
5.2. Major Contributions ............................................................................................. 177
5.3. Major Conclusions ............................................................................................... 180
5.4. Recommendations for Future Research Directions ............................................. 181

Appendix A. Statistical Test Results from R: ................................................................ 184

Bibliography ................................................................................................................... 193

vi

 

LIST OF TABLES

Table 1.1 Agencies involved in Great Lakes Environmental Management ....................... 4
Table 1.2 Major GRID enabled Geospatial Applications ................................................. 14
Table 3.1 User Feedback Summary .................................................................................. 96

Table 3.2 Afﬁliation of “Changing Landscapes” Workshop Participants (Percentages are
rounded to the nearest whole number and may not sum to 100) (Lucero et al., 2004). . 100

Table 3.3 What makes a tool useful? Characteristics derived from Wisconsin DNR’s

evaluative workshops (Watermolen, 2008). ................................................................... 100
Table 4.1 Ten-step procedure for a GIS hydrology study (David Maidment 1996) ....... 112
Table 4.2 Summary Table for 21 Selected Watersheds .................................................. 128
Table 4.3 Mean Values of Flow Accumulation Grids .................................................... 153
Table 4.4 Mean Values of Erosion Potential Grids from RUSLE .................................. 155

Table 4.5 Linear Model Residuals for Watersheds in Different Physiographic Regions 158

Table 4.6 Paired T-Test Results between FlowNet and D8 Flow Accumulation Samples
for both Wetland and Non-Wetland Areas ..................................................................... 172

vii

LIST OF FIGURES

Figure 1.1 Watershed Management Process (Davenport, 2002) ...................................... 23
Figure 2.1 Chapter 2 Organization .................................................................................... 35
Figure 2.2 Watershed Management Virtual Organization Diagram ................................. 56
Figure 2.3 Actor Network for the Midwest Partnership for a Watershed Management
Spatial Decision Support System ...................................................................................... 58
Figure 3.1 Grid based Integration Framework for Watershed Management V0 (Derived
from Figure 2.1) ................................................................................................................ 73
Figure 3.2 Digital Watershed System Architecture .......................................................... 76
Figure 3.3 Purdue SDSS System Architecture ................................................................. 80
Figure 3.4 Burns Ditch Trail Creek Web-based SDSS System Architecture ................... 85
Figure 3.5 Interoperability ................................................................................................ 87
Figure 3.6 Stakeholders Involved in Burns Ditch and Trail Creek Watershed Management
SDSS ................................................................................................................................. 89
Figure 3.7a Use Case Part A ............................................................................................. 91
Figure 3.7b Use Case Part B ............................................................................................. 92
Figure 3.7c Use Case Part C ............................................................................................. 93
Figure 3.7d Use Case Part D ............................................................................................. 94
Figure 3.8 Prototype system architecture for parallel processing application. ............... 105
Figure 4.1 Flow direction on planar triangular facets in a block-centered grid (from
Tarboton 1997) ................................................................................................................ l 18
Figure 4.2 Illustration of the Sweeping Plane Concept .................................................. 122
Figure 4.3 Physiographic Provinces ............................................................................... 124
Figure 4.4 Watershed Boundary Dataset Status (N RCS) ............................................... 125

viii

 

Figure 4.5 Locations of Selected Watersheds in Michigan ............................................ 126

Figure 4.6 Locations of Selected Watersheds in New Mexico ....................................... 126
Figure 4.7 Locations of Selected Watersheds in Virginia .............................................. 127
Figure 4.8 South Fork Hayworth Creek Watershed in Clinton, MI ................................ 130
Figure 4.9 North Fork South Branch Potomac Watershed in Highland, VA ................. 131
Figure 4.10 Upper Bear Creek Watershed in Grant, NM ............................................... 132
Figure 4.11 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

FlowNet Algorithm ......................................................................................................... 133
Figure 4.12 Local Blowup of Figure 4.11 ....................................................................... 134
Figure 4.13 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

Doc Algorithm ................................................................................................................. 135
Figure 4.14 Local Blowup of Figure 4.13 ....................................................................... 136
Figure 4.15 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

D8 Algorithm .................................................................................................................. 137
Figure 4.16 Local Blowup of Figure 4.15 ....................................................................... 138
Figure 4.17 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

F lowNet Algorithm ......................................................................................................... 140
Figure 4.18 Local Blowup of Figure 4.17 ....................................................................... 141
Figure 4.19 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

Duo Algorithm ................................................................................................................. 142
Figure 4.20 Local Blowup of Figure 4.19 ....................................................................... 143
Figure 4.21 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

D8 Algorithm .................................................................................................................. 144
Figure 4.22 Local Blowup of Figure 4.21 ....................................................................... 145
Figure 4.23 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

FlowNet Algorithm ......................................................................................................... 146
Figure 4.24 Local Blowup of Figure 4.23 ....................................................................... 147

ix

Figure 4.25 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

Doc Algorithm ................................................................................................................. 148
Figure 4.26 Local Blowup of Figure 4.25 ....................................................................... 149
Figure 4.27 Stream Cells Derived from Flow Accumulation Grid Calculated Using the

D8 Algorithm .................................................................................................................. 150
Figure 4.28 Local Blowup of Figure 4.27 ....................................................................... 151

Figure 4.29 Relationship between FlowNet algorithm and D8 erosion potential means 157
Figure 4.30 Difference Grid for South Fork Hayworth Creek Watershed in MI ........... 160
Figure 4.31 Difference Grid for North Fork South Branch Potomac Watershed in VA 160
Figure 4.32 Difference Grid for Upper Bear Creek Watershed in NM .......................... 161
Figure 4.33 Wetlands in South Fork Hayworth Creek Watershed, MI .......................... 163

Figure 4.34 FlowNet Wetland Flow Accumulation Grid for South Fork Hayworth Creek
Watershed, MI ................................................................................................................. 166

Figure 4.35 D8 Wetland Flow Accumulation Grid for South F ork Hayworth Creek
Watershed, MI ................................................................................................................. 167

Figure 4.36 FlowNet Non-Wetland Flow Accumulation Grid for South Fork Hayworth
Creek Watershed, MI ...................................................................................................... 168

Figure 4.37 D8 Non-Wetland Flow Accumulation Grid for South Fork Hayworth Creek
Watershed, MI ................................................................................................................. 169

Figure 4.38 Wetland Grid Sampling Points for South Fork Hayworth Creek Watershed,
Ml .................................................................................................................................... 170

Figure 4.39 Non-Wetland Grid Sampling Points for South Fork Hayworth Creek
Watershed, MI ................................................................................................................. 171

 

Chapter 1. Challenges in Distributed Environmental Decision Support Systems

1.1. Introduction

This research addresses the long standing issue of effective integration of science
and decision making in environmental management using the latest Virtual Organization
(V0) based Grid Computing technology coupled with advanced geospatial technology
within watershed management context. The related technical and organizational
interoperability problems in distributed geoprocessing and modeling are also discussed in
the watershed management context. Environmental managers have been increasingly
calling on science to provide relevant information for environmental decision making
(e.g., Browning-Aiken et al., 2004, 2006; NRC, 1999; Matthieset al., 2007). However,
despite the recent remarkable growth in availability of relevant knowledge, data, and
information in environmental science, how science can effectively support environmental
decision making remains an unresolved question (Cash et al., 2003; Lee, 1993; Reichert
et al., 2007; van der Sluijs, 2007).

The objective of this research is to extend the virtual organization paradigm to the
context of watershed management, and to evaluate its effectiveness for that context. My
hypothesis is that watershed management can beneﬁt from V0 if particular developments
are realized:

0 Integration of decision-making and communication into a web-based
distributed environmental decision support system. For example,
policymakers at multiple organizations should be able to work with one

another remotely, sharing not only data and models, but also decisions.

0 Distribution of environmental models. For example, actors in one location
and agency should be able to execute watershed models with elements on
remote servers in different organizations.

0 Distribution of computing resources. Complex models require much
processing time, and distributing those in a Grid-based manner may
integrate model runs into real time or near-real time decision support.

0 Modularity of environmental model components. For example, analysts
should be able to plug different functions into a model and have it run
seamlessly. A watershed model might have one ﬂow accumulation
algorithm replaced by another.

This dissertation presents my progress in each of these areas, with original
contributions illustrating how V0 can be utilized by organizations involved in watershed
management. Furthermore, it evaluates the performance and efﬁcacy of each
contribution, to help identify whether the challenges of this model of decision-making are
worth the effort.

This research is clearly within the realm of geography because it studies
geographic phenomena and their relations within watersheds and human interactions with
watershed environments. As Turner (2002) pointed out, geography has two basic
identities: geography as a spatial-chorological approach and geography as the human-
environrnent subject. The spatial-chorological branch of geography investigates the
geographic phenomena and their relations within a particular region while the human
environment branch deals with human environment interactions (Turner, 2002). By

furthering our understanding of flow phenomena within watersheds and developing a

coherent organizational and technical framework for integrating scientiﬁc knowledge into
decision making process, this research conforms to both traditions of geography
discipline.

The rest of this section describes the background of this research ﬁrst and then
deﬁnes the overall research approach. The second section of this chapter reviews
distributed computing, geoprocessing and spatial decision support. The third section
examines the various aspects of watershed management. The ﬁnal section lists speciﬁc
research objectives and lays out the organization of this manuscript.

Environmental systems are driven by various natural and human processes at
multiple spatial and temporal scales (Steyaert, 1993). Even though science has been
increasingly brought in to provide critical information for complex environmental
decision making (Browning-Aiken et al., 2004, 2006; NRC, 1999; Matthies et al., 2007),
our knowledge of complex environmental systems relative to the holistic nature of these
systems, however, tends to be fragmented frequently resulting in narrowly-focused, or
myopic, fragmented management approaches and strategies (Savory et al., 1999). Federal
environmental statutes and permitting programs, many of which are implemented by state
agencies, focus on one environmental medium at a time—the Clean Water Act, Clean Air
Act, Resource Conservation and Recovery Act (RCRA), Superfund (CERCLA), and the
Endangered Species Act to name but a few. Not surprisingly, then, environmental data
are collected and stored in different formats by different agencies pursuing different
missions. Different agencies also produce different computational environmental models
that are often not interoperable. In addition to different agencies being responsible for

different environmental media, there is limited vertical integration between federal,

regional, state, and local government agencies. The vertical as well as horizontal range
of agencies can be broad and extremely diverse.

In the case of the Great Lakes Basin, for example, many government agencies in
the US. with environmental responsibilities are involved in data collection and other

aspects of environmental management (See table 1.1).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Agency Name Agency Level i
Environmental Protection Agency Federal E
US Geological Survey Federal I
USDA Natural Resources Conservation Service Federal ,1
US Army Corps of Engineers Federal
US Coast Guard Federal
US Fish and Wildlife Service Federal
National Park Service Federal
US Forest Service Federal
National Oceanic and Atmospheric Administration Federal
Regional Planning Agencies Regional
State Departments of Natural Resources State
State Departments of Environmental Quality State
State Departments of Agriculture State
Soil Conservation Districts State
Regional Great Lakes Commission Regional
Great Lakes Fisheries Commission Regional
International Joint Commission International
Local (City, County, and Township) Governments Local

 

 

Table 1.1 Agencies involved in Great Lakes Environmental Management

All of these agencies collect a myriad of environmental data in a variety of
formats for purposes associated with their respective missions. Some of these agencies
also sponsor or produce different computational environmental models or decision
support systems for various environmental management purposes. But unfortunately,

these models and systems are often difficult to use and don’t work with each other so

they are rarely used in the real world decision making process (Acreman, 2005; Cash et
al., 2003; NRC, 1999).

There have been a number of recent discussions on how to best connect
environmental science and decision making, so that science can generate useful
information in the context of decision making (Cash, 2002;Cash et al., 2003; Dilling,
2007; Logar and Conant, 2007; McNie, 2007; Sarevvitz and Pielke, 2007). While
scientists often say their opinions have been ignored by policy makers, the latter have
also indicated that critical information required for decision making is often not readily
available or not presented in a usable form (Jacobs, 2002). So clearly a better interface
and communication mechanism between science and policy is needed for improved
environmental decision making (Parker et al., 2002). To deal with the management, data
and modeling fragmentation issues mentioned above and improve environmental decision
making with sound science, a holistic systematic approach is needed based on: 1) a
coherent set of multi-institutional relationships and 2) technical consistency and quality
(Chi & Holsapple, 2005; Castelletti and Soncini-Sessa, 2006,2007; Gaddis et al., 2007).

Multi-institutional coherence can be provided through strategic coordination and
timely communication among those collecting and storing data, those doing modeling
and decision support system work such as researchers, scientists and IT professionals,
environmental managers and policy makers, and others (Castelletti and Soncini-Sessa,
2006,2007; Gaddis et al., 2007). Developing and understanding these multi-institutional
and multi-disciplinary relationships is critical to building a common base of knowledge
that supports well-informed, science-based environmental and natural resource

management decisions (Castelletti and Soncini-Sessa, 2006,2007; Gaddis et al., 2007).

Technical consistency and quality and can be achieved through effective use of
technology and modeling—particularly information technology, geospatial technology,
and environmental modeling (Parson, 1995; Dilling, 2007; Xu et al., 2007). Recent
advances in organizational theories, geospatial technology, and information management
technology, grid computing in particular, have contributed to the development of new
types of comprehensive decision support systems (DSS) (Birkin et al., 2005). These
systems have the capacity to integrate multi-disciplinary, multi-institutional sets of
diverse environmental data and computational models, and provide rapid translation of
research results into practical methods to understand and manage complex environmental
systems in a virtual organization context. A virtual organization is “an identiﬁable group
of people or organizations that makes substantially more use of Information and
Communication Technologies than physical presence to interact, conduct business and
operate together, in order to achieve their objectives” (Sieber et al., 1998).

By combining Grid computing capacity with a virtual organization, a new
paradigm is being created. This powerful paradigm takes advantage of multiple
networked computers to form virtual computer architecture with the capacity to distribute
process execution across a parallel infrastructure and produce greater computation
throughput (Foster, 2003; Foster & Kesselman, 2003; Foster, 2005; Foster & Tuecke,
2005; Stockinger, 2007). Grid computing employs the resources of many computers
located at different sites but connected through a digital network (the Internet, usually) to
solve large-scale computation problems (Foster & Tuecke, 2005 ; Stockinger, 2007). The
Grid computing approach reﬂects a conceptual framework rather than an actual physical

structure. Using multiple physically and administratively-separated resources, Grid

 

computing enables a single virtual organization to perform complex sets of computational
tasks (Jacob et al., 2005). The one important aspect of Grid Computing is to emphasize
and enable organizational interoperability ﬁrst before engaging in detailed technical
interoperability (Foster & Tuecke, 2005). This is exactly what the multi-organizational
and multi-disciplinary environmental DSS needs. Grid computing can not only integrate
system components across organizational boundaries, but it can also enable high-
performance computing within the DSS.

A case study approach is used in this research to test the hypothesis stated above.
This case study investigates the functionality requirements of a virtual framework based
on an integrated grid computing network coupled with advanced geospatial technology
for a comprehensive watershed management decision support system (DS S) at multiple
watershed scales. Then an example VO-based distributed watershed management
decision support system within this framework is built to demonstrate the beneﬁts of V0
for watershed management. Distribution of environmental models and computing
resources including the technical and organizational interoperability issues in distributed
environments are also discussed in this context. An innovative DEM based ﬂow
modeling algorithm for such a watershed management V0 is also developed to show how
modularity of environmental model components can help improve watershed
management V0.

In the next two sub-sections distributed computing, spatial decision support and

watershed management are reviewed in the context of this research.

 

1.2. Distributed Computing, Geoprocessing and Spatial Decision Support

1.2.1. Distributed Computing

The term distributed computing was used to describe a type of computing where
different processing tasks and data are distributed among separate computers connected
to a network (Attiya & Welch, 2004). Distributed systems today generally have following
characteristics: (1) greater access to data and computing resources stored and managed on
distributed servers on the network; (2) faster response time due to local computing; (3)
tighter system security; (4) less complexity; and (5) in many cases, lower-cost computing
solutions than more traditional mainframe or minicomputer solutions (Coleman, 1999;
Attiya & Welch, 2004). Distributed systems are pervasive today throughout business,
academia, government, and the home. They provide ways to share resources such as
printers and scanners, and to share data and computational services. The concept of
distributed computing often provides a better ﬁt to the complex structures and often
multidisciplinary nature of modern organizations and offers greater user involvement in
information management activities (Coleman, 1999). The rapid advances of distributed
computing technologies and the grth of high—speed Internet connections has resulted in
an Internet-driven, online computing era (often referred to as Web 2.0). The software
industry is moving towards the Internet platform while the new distributed technologies
like XML and Web services have emerged to provide a solid foundation for this seismic
shift (Spector, 2003). Internet-based software and Service Oriented Architecture (SOA)
has globalized information and content while catalyzing formation of various online
communities dedicated to a bewildering array of different topics and subjects (Dempsey,

2000; Purcell, 2006; Lawler & Howell-barber, 2007).

The emerging Grid computing technology is a result of natural evolution of
distributed computing and the Internet. Grid computing is based on the notion of
computational Grids (Foster & Kesselman, 2003) where computing resources are
available as universally and as easily as electric power. Grid computing focuses on large-
scale resource sharing, innovative applications, and, in some cases, high-perfonnance
orientation within a Virtual Organization (V 0) (Foster & Tuecke, 2005; Stockinger,
2007). Grid computing is capable of linking multiple organizations through a network of
shared data, software programs, and powerful processors. This capability is a function of
integrated organizational and technical components. The degree of integration that a V0
employing Grid computing technologies can achieve is unprecedented (Shi et al., 2002;
Foster & Kesselman, 2003; Foster & Tuecke, 2005; Stockinger, 2007). Grid computing
also has two major integrated technical components: distributed computing and
parallelization of intensive computation. These two components are also very important
for Geographic Information Technology (Coleman, 1999; Healy et al., 1997; Armstrong
et al., 2005).

The Geographic Information Science (GIScience) community has long recognized
the value of distributed computing technologies for societal use and diffusion of
geospatial technologies (Coleman, 1999). The problems and applications addressed by
GISs are particularly well suited to distributed computing (UCGIS, 1996, 1998, 2002;
Coleman, 1999), and as a result, distributed computing is a research priority in the ﬁeld
of GIScience. Geographic information technologies allow demonstration of vital linkages
between what otherwise appear to be unrelated activities. The identiﬁcation of

characteristics and attributes with a common geographic location has frequently led to

more extensive sharing and integration between what were previously rigidly-separate

organizations or parts of the same organization (U CGIS, 1996).

1.2.2. Geoprocessing and Spatial Decision Support

Geographic decisions supported by 0185 often involve multiple stakeholder
groups widely distributed across geographic locations and social strata. Stakeholders are
also frequently located in different administrative tiers in an organization hierarchy. In the
past, this distribution has precluded any integration or coordination (Shi et al., 2002).

Data sources and geoprocessing power residing in sophisticated software and
hardware are often widely distributed. Geospatial operations and related decisions are
ﬁ’equently made in the ﬁeld and thus require ﬂexible access to remote databases and
models. Computationally-intensive spatial applications with ever-ﬁner data resolution
often tax the capacity of single machines (UCGIS 2002, Shi et al., 2002, Xue et al., 2002;
Armstrong et al., 2005; Wang et al., 2005).

Spatial Decision Support Systems (SDSS) are a type of computer systems that
utilize both geospatial technologies and decision support system to assist decision-makers
with problems that have spatial dimensions (Walsh 1993, Prato and Hajkowicz, 1999).
SDSS have been used in a variety of applications such as ﬂood prediction (Al-Sabhan et
al., 2003), river water quality evaluation (Wang et al., 2005), conservation program
management and best management practices assessment (Rao et al., 2007). But most of
these applications are stand-alone ones that don’t use distributed computing technologies.
An early representative GIS application of distributed computing and spatial decision

support system was a project called Sequoia 2000 that linked researchers throughout the

10

University of California system who were investigating global climate change via an

integrated computer environment (Stonebreaker, 1995). Scores of earth scientists,

atmospheric scientists, and computer scientists as well as industry representatives worked

together and created a system to manage vast environmental databases that rapidly

provide data and visualizations across a dedicated wide-area network.

The experiences of Sequoia 2000 and similar projects indicated the presence of

both opportunities for and impediments to the distributed computing applications.

Impedirnents may be either technological or organizational. Technological challenges

include:

Integrating data from different sources, stored in different databases
Transmitting very large quantities of information over the network

Developing ﬂexible interfaces for different users

Building high performance and, if necessary, parallel geoprocessing functions

for real time applications and very large datasets with ﬁner resolution.

Organizational challenges include:

Combining members of highly disparate research communities like computer

science, hydrology, agricultural and environmental engineering, earth science, etc.

Communicating clearly research objectives and requirements

Orchestrating linked but independent environmental models developed in

different laboratories (Shi et al., 2002)

Forging geographic information partnerships for local, regional, national,

international, and global spatial data infrastructures.

ll

 

The academic GIScience community, government agencies, and commercial GIS
software vendors have worked diligently to address these challenges since the Sequoia
project. A variety of GIScience initiatives (see e.g., OGIS, 2002; Hecht, 2001; UCGIS,
2002) were intended to improve the utility of distributed computing for geospatial
applications. These initiatives focused on technical aspects although organizational
challenges associated with large projects have long been acknowledged. Several U.S.
initiatives have focused on building nationwide partnerships and data infrastructures
(e.g., e-Govermnent (eG), National Spatial Data Infrastructure (N SDI), The National
Map (TNM), and GeoSpatial One-Stop (GOS)). These initiatives call for development of
an "information highway" to connect diverse spatial data producers and users in the
private sector, academic institutions, and all levels of government (UCGIS 2002; NSDI
future directions, 2004; Crompvoets et al., 2004).

A critical issue related to applications of distributed computing in geospatial
applications is how to achieve interoperability in a distributed environment.
Interoperability refers to the ability to move easily from one system to another.
Interoperability has been a serious concern for sharing of geographic information in
GIScience since the late 19705 (Sondheim et al., 1999; Harvey et al., 1999, Riedemann
and Kuhn, 1999). The explosive growth of the Internet and Web has facilitated the rapid
advances in the interoperability research. As a result, ontology and semantics have
become important topic in both general IT and GIScience community. Ontology has been
viewed in GIScience as a standardization process through which easier semantic
interoperability can be achieved between different information systems (Gruber, 1991;

Chandrasekaran et al. 1999, Smith 1999, Fonseca et al. 2002, 2003). Semantics focuses

l2

on the study of meanings. Semantic interoperability is the capability of two or more
computer information systems to exchange information and have the meaning of that
information automatically interpreted by each other (Selvage et al., 2006). Because of the
inherent indeterminacy and vagueness associated with geographic concepts and
categories, no consensual shareable ontology for the geospatial domain has been
produced even though some speciﬁc ontologies exist for certain applications (Agarwal,
2005; Fonseca et al. 2002, 2003).

GIS communities’ interest in distributed computing and interoperability has
historically focused almost entirely on sharing data. Effective sharing of geoprocessing
methods and tools has a potentially higher return than sharing data because these methods
and tools are generic with a broader range of users than a given set of data (Peng & Tsou,
2003). Over the past few years, major GIS software vendors have recognized the
importance of the Internet as a computing platform, and they re-engineered their products
for an Internet platform. This new generation of Intemet-based GIS software products not
only facilitates sharing of spatial datasets but also facilitates advanced geoprocessing
functions across the Internet based on service oriented architecture (Smith & Sheahen,

2008)

1.2.3. GRID-based Geospatial Applications

The new generation of Internet GIS, however, has not met all the technical and
organizational challenges cited in the previous section. New innovations in distributed
computing are needed to tackle those challenges. The emerging Grid computing

technology has great potential to solve these problems and challenges. In fact, the

13

GIScience and environmental sciences community in general have already recognized the
importance of Grid computing technology and started to apply it in various geospatial
applications (e.g., Shi et al., 2002; Wang et al., 2002, 2005; Armstrong et al., 2005).
Beven (2003) realized the potential of GRID to change the way that environmental
models are constructed and used. Hluchy et al. (2004) introduced a ﬂood forecasting
system in a Grid computing environment. Arat'rjo et al. (2005) described a GRID-enabled
Hydro-Meteorological Scientiﬁc Network. Zhang and Tsou (2005) presented a new
framework for Grid-enabled GIService web portals to facilitate the building of high-level
intelligent Internet GIServices. Shu et al. (2006) explored integrating OGC (Open
Geospatial Consortium) web services and Grid technologies for geospatial data sharing.
Several major research efforts that involve developing GRID enabled geospatial

applications are listed in table 1.2.

 

 

 

 

 

 

 

Application Title Application Type State of
Development
GEON Online Portal w/ Access to Data, Tools & Applications Ongoing
GISolve Online Portal w/ Access to Data, Tools & Applications Ongoing
WATERS Online Portal w/ Access to Data, Tools 8 Applications Ongoing

 

 

Table 1.2 Major GRID enabled Geospatial Applications

As a multi-disciplinary research initiative, GEON (GEOscience Network) was
formed by earth and computer scientists in 2003 to build a cyberinfrastructure for the
geosciences. In order to better understand the dynamics of complex earth systems,
geoscientists need an integrative computational infrastructure to link, share and utilize
massive multi-disciplinary data sets and tools they developed over the years. The main

goal of the GEON is to provide the critical infrastructure necessary to facilitate

14

collaborative interdisciplinary earth science research (Seber et al., 2003; Keller et al.,
2003, 2005). By integrating datasets at semantic levels, GEON can provide users with
information on datasets, tools and resources at the conceptual level and thus enable them
to access and utilize data and tools from allied disciplines with minimal effort. However,
construction of this cyberinfrastructure capable of integrating, analyzing, and modeling
dynamic heterogeneous spatial information is not an easy task. Major obstacles include
different characteristics of geoscience data formats, storage protocols, and computing
systems and, most importantly, the differing conventions, terminologies, and ontological
frameworks across disciplines (Seber et al., 2003). Because of these obstacles, many
researches within the GEON project have been focused on ontology and semantic
interoperability (e. g., Bowers & Ludaescher 2003, 2004; Lin & Ludaescher 2004; Sinha
& Lin et al., 2005; Sinha & Zendel et al., 2006; Sinha & Lin et al., 2006; Brodaric &.
Gahegan, 2006).

Grid technology is used in GEON to develop its cyberinfrastructure. Grids
provide sustainable and distributed architecture to facilitate seamless data and
computational resources access over the Internet using secure links and connections. The
GEON grid and portal were built as open source platforms so the user community can
develop their own tools and access mechanisms (Seber et al., 2003; Keller et al., 2003,
2005). A lot of development efforts within the GEON project are geared towards
middleware development because it is the critical component for connecting different
parts of the infrastructure (e.g., Bhatia & Memon et al., 2003; Memon et al., 2004; Bhatia
& Chandra et al., 2005; Mueller, 2005; Youn et al., 2006). According to Carnegie Mellon

Software Engineering Institute (1997), middleware is “connectivity software that consists

15

of a set of enabling services that allow multiple processes running on one or more
machines to interact across a network”.

GEON’s earth science research has been tested in two regions: the Rocky
Mountains and the Mid-Atlantic region. Intracontinental deformation and terrane
recognition and analysis were studied in each region, respectively (Seber et al., 2003).
Most of applications developed within the GEON are centered on geological areas (e. g.,
Sinha et al., 2003; Salayandia et al., 2006; Rees et al., 2007) and LiDAR (Light Detection
and Ranging) data distribution and processing (e. g., Crosby & Arrowsmith, 2004; Crosby
et al., 2006; Jaeger-Frank et al., 2006). The future goals of GEON network include
improving existing GEON systems and middleware infrastructure; building new useful
tools on top of existing framework; encouraging software development and resource
integration with partner sites; and integrating more data, more applications (Chandra,
2007)

As a problem solving environment, GISolve was created based on Grid
architecture to provide grid services for computationally intensive geospatial information
analysis. Many of GISolve projects are focused on the so-called geo-middleware
development. The geo-middleware is a piece of software that sits between existing grid
middleware and geographic information analysis applications to manage heterogeneous
and dynamic resources on behalf of analysis applications. GISolve can also decompose
certain parallel geographic information analysis problems into several sub-tasks that can
be scheduled for processing using multiple Grid resources. Data transfer management and
distributed task scheduling are all supported by GISolve. All functions in GISolve are

exposed as grid services that are compliant with the open grid service architecture

16

(OGSA) through grid portal. Grid portal technologies provide access to Grid services for
security management, remote data and job management, resource information access,
application interface speciﬁcations, and access to collaboration services. The examples
within GISolve include a computationally intensive spatial statistics application and
spatial temporal data explorer etc (Wang et al., 2005).

WATERS (WATer and Environmental Research Systems) network was planned by
CUAHSI (Consortium of Universities for the Advancement of Hydrologic Sciences) and
CLEANER (Collaborative Large-scale Engineering Analysis Network for Environmental
Research) project ofﬁce in 2005. It is funded by National Science Foundation (NSF)
Engineering and Geosciences Directorates. The main goal of the WATERS Network is to
build a national capacity to better understand and predict processes at multiple spatial and
temporal scales coupling water with natural and human systems. It’s a bold
environmental observatory initiative to transform water research through new
investments in infrastructure. It recognizes that the advances in water-related sciences
and engineering can only be based on new measurements (WATERS Network, 2008).
According to its website, the WATERS Network will include the following elements:

1. “A national Observational Network of highly instrumented experimental facilities

for testing hypotheses, with:

0 sensor networks within the natural, built, and managed environments for
acquisition and analysis of high-frequency and spatially dense
environmental data in real time

o representative coverage of the diverse conditions of the United States

0 high-resolution characterization data”

17

2. “A Sensor and Measurement Facility to provide specialized instrumentation,
support and training in instrumentation use to address WATERS Network
questions. The Sensor and Measurement Facility will include:

o specialized support personnel and facilities for sensor and measurement
support

0 expertise on recent advances in environmental sensors, wireless
communications and remotely sensed data”

3. “A Cyberinfrastructure, Modeling, and Synthesis Facility to provide a shared-use
network as the framework for collaborative analysis, including:

0 an integrated model of the hydrologic system to support adaptive sampling
0 a digital watershed to access all available data (including those collected
by other organizations and mission agencies)”

WATERS network cyberinfrastructure demo site include some preliminary
applications such as hydrological data access system and streamﬂow analyst. A review of
the project timeline at WATERS network website indicates that even though the
WATERS network project ofﬁce was established in 2005, the project is still at planning
and strategy design stage as of 2008. Various planning documents regarding different
components of the network such as sensors, cyberinfrastructure and modeling etc. can be
found at WATERS network website.

All three major research efforts mentioned above attempt to develop
cyberinfrastructure based on Grid technology in support of sciences. GEON is focused on
facilitating collaborative and interdisciplinary research in geosciences. GISolve bridges

the gap between computationally intensive spatial information analysis applications and

18

the Grid environment. WATERS network strives to build a national capacity for
advancing water-related sciences. But one thing worth notice is that all these research
efforts have largely been conducted in the academia with very few linkages to
environmental agencies and user communities. These linkages are considered critical in
the research presented in this dissertation even though it also aims at applying Grid

technology to build a better technical infrastructure for environmental decision making.

1.3. Watershed Management

1.3.1. Watersheds

Water is essential to life. The quantity and quality of water is critical to all human
systems and those of other natural organisms. Every living organism on the Earth
depends on water. Water, therefore, regulates population growth, affects world health and
living conditions, and determines biodiversity (Newson 1992). Yet until the 19803, most
water management practices focused on solving singular, localized problems without
considering the impacts of changes in human behaviors and the landscape on the
biophysical, economic, and social dimensions of a watershed system (Schramm 1980).
Over the past thirty years, a global consensus has emerged to treat watersheds as the
appropriate units for the management of water and related aquatic resources (McDonald
and Kay 1988; Koudstaal et al. 1992; Lee 1992; Newson 1992; Heathcote 1998;
Gregersen et al., 2008).

A watershed is the area of land which drains runoff (from rain, snow, and ﬂowing
springs) to a body of water (Gregersen et al., 2008). In each watershed, water ﬂows

downward to a common receiving body such as a river, lake, wetland, or bay. Water

19

travels across land surfaces used for farming, forests, suburban lawns and golf courses,
city streets and other impervious surfaces before reaching a surface water body. Some

water inﬁltrates the soil and becomes groundwater (Ward et al., 2003).

1.3.2. Watershed Management Process

Watershed management is the process of organizing and guiding the uses of land
and other resources at a watershed scale to provide desired goods and services without
adversely affecting water and related natural resources (Brooks et al., 1997). Watershed
management recognizes the interrelationships among land use, soil, and water, and the
linkages between uplands and downstream areas. A watershed perspective drives the
consideration of the multiple and cumulative effects of changes and actions in a
biophysical area and the need to identify the nature, attributes, and characteristics of key
linkages between terrestrial and aquatic ecosystems (Gregersen et al., 2008). Watershed
management entails an integrated approach in which social, economic, and environmental
factors are considered in determining appropriate actions to manage water resources
(Heathcote 1998; Gregersen et al., 2008). This holistic approach requires consideration of
the interaction of ground and surface water. Effective integrated watershed management
is an ongoing, iterative process, and ﬂexibility is needed to adapt management practices
to the unique characteristics of each watershed as well as those changes in human
populations and behaviors and changes on landscapes in a watershed over time (Brooks
etal., 1997; Heathcote 1998; Gregersen et al., 2008).

Watershed management relies in large part on the active involvement of local

agencies, economic interests, NGOs, and citizens. Bishop (1970) observes that water

20

management planning is a process to achieve social change. As a consensus-building
process, then, it is categorically not a one-dimensional pure scientiﬁc exercise.
Davenport, an EPA watershed management expert with two decades of ﬁeld experience,
indicates that watershed management is mainly a process of working with people to solve
identiﬁed problems (Davenport, 2002). An effective watershed management plan must
reﬂect societal consensus regarding the value of water as a resource, social attitudes, and
stewardship responsibilities, and the community’s vision of an ideal watershed and water
quality (Nickum and Easter, 1990; Davenport, 2002; Gregersen et al., 2008).

Practices associated with watershed management are directed at guiding changes
in land use and vegetative covers as well as other structural and nonstructural actions to
achieve water quality goals and objectives (Reimold, 1998). Management tools include
nonstructural (vegetation management) practices as well as an array of structural
(engineering) practices depending on speciﬁc conditions. Soil conservation practices and
land use planning activities are tools that can be employed in watershed management.
Land use practices include establishment of protected sensitive areas; regulations on road
building, timber harvesting, and agro-forestry practices; storm water runoff management;
and other activities (Reimold, 1998). The focus of these practices is their impacts on
water resources and related aquatic resources in a watershed.

The institutional and policy framework of watershed management provides an
integrative methodology to consider the impacts of various human activities in a given
unit of land (the watershed) on water resources and related aquatic resources (Reimold,
1998). Management practices include the use of physical, regulatory, or economic tools

and techniques to address identiﬁed and potential problems involving relationships

21

between water resources and land uses. These tools and techniques are employed by
farmers, foresters, soil conservation staff, engineers, citizens, and others. Watershed
management requires active collaboration among specialists in a broad range of
disciplines — engineering, biological sciences, hydrology, economics, sociology, law,
institutional policy, and ethics — as well as government agencies, private industries,
nongovernmental organizations, and the public (Brooks et al., 1997; Heathcote, 1998;
Reirnold, 1998; Gregersen et al., 2008 ).

The process of watershed assessment and management decision making requires
biophysical, social, and economic information that ideally is augmented by local
knowledge and insights. This process is built on stakeholder involvement, social capacity
building, and adequate monitoring (Heathcote, 1998; Davenport, 2002; Gregersen et al.,

2008)

1.3.3. A Four-phase Watershed Management Process and its Beneﬁts

Davenport (2002) describes a four-phase watershed management process based
on collaboration that responds to common, stated needs and goals. The four phases are
assessment; planning; implementation; and evaluation. This watershed management

process is multi-phase and iterative (Figure 1.1).

22

Watershed Management Process

 

 

Phase 1 Phase 2
Setting Goals
Defining The And Identifying

Problem KN Solutions

'\ /
3 Building a 9%
PrOJect Team Q

Phase 4 A/ \A Phase 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Measuring Success v
And Making Implementing
Adjustments Controls
m

Figure 1.1 Watershed Management Process (Davenport, 2002)

In phase 1 of watershed assessment/problem identiﬁcation, knowledge of the
biophysical, economic, and social processes and interactions in and around the watershed
is acquired, including stakeholder concerns and interests, institutional concerns and
constraints, and socio-economic characteristics. This phase is subdivided into four
components that include: (1) inventory/mapping; (2) obtain information and data
analysis; (3) identify problems (stressors and their sources) based on the analysis; and (4)
determine the overall goal (Davenport, 2002).

In phase 2, planning refers to development of an implementation strategy based
on a diagnostic analysis that uses information from inventory assessments and

stakeholder input. Selection of strategic options from among identiﬁed alternatives and

23

documenting the decision making process is based on this analysis. To achieve a valid
planning outcome, a mere document is not sufficient. An effective implementation
strategy solves identiﬁed problems and resolves outstanding data needs. An effective plan
is “bottom-up” rather than “top-down” (Davenport & Wilson, 2001). Prescribed actions
are based on well-defmed, water-quality standards, voluntary incentives, education and
information, and regulation/enforcement (Davenport, 2002; Gregersen et al., 2008). It
should be noted that a watershed management plan need not be completed before
implementation begins. Problems identiﬁed with known solutions should be addressed
immediately. With an early start in building a successful track record, success of the
overall planning and management process is more likely (Davenport, 2002).

In phase 3, implementation represents the culmination of the initial assessment
and planning efforts, organizing, and information/education activities — bringing together
previous efforts and actually doing the work. Implementation actions include steps to
address pollution prevention, pollution controls and reduction, and restoration of
ecological quality and functions (Davenport, 2002). Implementation builds on the historic
point source controls in which compliance with efﬂuent discharge standards (maximum
acceptable levels of pollutant loadings) achieved signiﬁcant and even dramatic
improvements in water quality with the reduction of pollutant loadings (Gregersen et al.,
2008). Reducing nonpoint source pollution (as distinguished from “poin ” source
discharges from industrial and municipal treatment facilities regulated by the National
Pollution Discharge Elimination System [NPDES] under the 1972 Clean Water Act) has
been a steep challenge given the diffuse nature of nonpoint pollution sources—farm

ﬁelds, septic tanks, residential lawn fertilizers and pesticides, water runoff from streets

24

and parking lots, and many other multiple small sources (Ritter and Shirmohammadi,
2000). Although the aggregate impact of nonpoint pollution is a highly signiﬁcant cause
of degradation of water quality, precise identiﬁcation of the location of these highly
diffuse sources has been extremely difﬁcult (Nikolaidis et al., 1998; Ritter and
Shirmohammadi, 2000; Zhang and Wang, 2002; Zhang and Jargensen, 2005). Such
identiﬁcation is critical to targeting appropriate best management practices (BMPs) at
those sources with the highest risks of generating pollutant loadings.

In phase 4, evaluation builds on assessment data and will continue after
completion of the implementation phase. Evaluation provides indications of whether
implementation has achieved stated goals and objectives. In addition, evaluations
conducted during implementation can provide feedback to guide midcourse corrections
(Davenport, 2002). Evaluations can also signal opportunities to celebrate
accomplishments and this kind of positive feedback contributes to a sense of community
ownership in the process of enhancing responsible environmental stewardship.

There are three categories of beneﬁts for this watershed management process.
First, it provides a context for multi-institutional integration by using a practical, tangible
management unit. The size of the watershed addressed is tied to identiﬁed problems and
local capabilities and interests. A process that includes a high level of stakeholder
engagement facilitates focusing and coordinating federal and state efforts to address local
problems (Davenport, 2002). Second, it promotes learning, understanding, and an
appreciation of natural ecological functions and processes. Appreciating how natural
processes and ecological functions beneﬁt humans is critical to the identiﬁcation and

implementation of actions to protect and preserve valuable ecological functions

25

(Gregersen et al., 2008). Stakeholders learn to link the impacts human activities on
ecological changes in water quality and quantity. Third, this process promotes better

management (Davenport, 2002).

1.3.4. Watershed Partnership

The watershed planning process requires the creation of a partnership that
manages this process to ensure stakeholder involvement, adequate monitoring, and an
outreach program (Walters, 2000; Davenport, 2002). Stakeholder participation early in
the process is critical. Extensive interaction with stakeholders is imperative to develop
the most appropriate plan to manage the watershed and fulﬁll stakeholder goals (CTIC,
1995; Born and Genskow, 1999). Since stakeholders’ opinions on watershed issues
frequently diverge or clash, negotiation and compromise may be necessary to agree on
appropriate actions. Effective compromises depend on the active involvement of multiple
stakeholders with easy access to accurate data and information (CTIC, 1995). Active
stakeholder involvement creates a context in which consensus and support needed to
implement watershed management plan can be obtained (CTIC, 1995; Born and
Genskow, 1999). It is obviously true that information is power. And distributed
information can provide a mechanism to disseminate information across conventional
barriers and enable effective democratic decision making (Voinov and Costanza 1999).

A partnership is critical to the process of developing an effective watershed
management plan (Walters, 2000; Davenport, 2002). A partnership is deﬁned here as a
collection of agencies, organizations, and individuals associated in a geographically-

based undertaking as shareholders and/or partakers to address speciﬁc problems or issues.

26

Watershed partnerships embrace a place-based focus and rely in large part on science to
support decision making (Born and Genskow, 1999; Walters, 2000; Davenport, 2002). A
partnership bridges the capabilities, assets, and resources of multiple agencies,
organizations, and individuals. The development and implementation of a watershed plan
by a broadly-based partnership creates local ownership in the watershed and provides an
effective mechanism to reach consensus (Davenport, 2002). A partnership may also
provide a means for more efﬁcient use of ﬁnancial resources; more creative and
acceptable ways to manage and protect natural resources; and a broad commitment to the
protection and preservation of natural resources (Walters, 2000; Davenport, 2002). A
partnership-building process includes the identiﬁcation and engagement of interests with
a stake in the management of the watershed, establishment of an organization,
determination of a goal and plan for the watershed, implementation of the plan, and
evaluation of its impact (Davenport, 2002). Davenport (2002) describes four
characteristics for successful watershed management partnerships: one, a clearly-stated
goal; two, a strong management plan with identiﬁed outcomes; three, leadership
representing multiple interests, and four, local support. Holdren et al. (2001) identiﬁes
key institutional characteristics of a successful watershed organization that include full-
time staff; ofﬁce space and equipment; a water-quality monitoring program; public
outreach program; access to water quality information; concern for water quality; and a
commitment to citizen participation.

There is no single organizational model for watershed partnerships. The form and
structure of a watershed partnership can range from an informal organization to a

complex, formally-organized structure. Davenport (2002) recommends a minimum

27

organizational structure that includes steering, planning, technical advisory, citizen
advisory, and operations committees. A steering committee provides leadership for the
overall watershed management effort and identiﬁes program resources available to
address watershed management needs. A planning committee develops the watershed
management plan working closely with a technical advisory committee (TAC). A TAC
conducts the watershed assessment to provide a technical basis for the watershed
management plan. A planning committee determines watershed plan objectives after
identifying the resource issues and concerns in the watershed. A TAC uses these
objectives to identify possible solutions to the identiﬁed problems and meet the
objectives. A TAC is typically comprised of 20-25 professionals and interested
stakeholders from nonproﬁt, nongovememental organizations (N GOs), local
governments, state and federal agencies, and public and private universities. A TAC is
interdisciplinary with expertise in biology, hydrology, etc. to perform the
assessment/problem identiﬁcation and formulate alternative strategies (Davenport, 2002).
Some watershed partnerships have a citizen advisory committee (CAC) to provide
advice. A CAC provides input, inﬂuence, and active involvement of watershed residents.
An operations committee is responsible for implementation, evaluation, outreach, and
monitoring. Depending on the scope and complexity of implementation activities, various
operations teams or workgroups may be formed to focus on speciﬁc activities. A
planning committee determines the primary outreach message, and an outreach team
identiﬁes target groups and appropriate communications strategies (Davenport, 2002). An
important part of the outreach component is convincing stakeholders that they are

stakeholders (Walters, 2000; Davenport, 2002). Outreach must also be coordinated with

28

TAC and planning committee activities (Davenport, 2002). A docrunent/information
repository should be created to house completed studies, reports, and other information
and be made available to the public. An operations committee should be responsible for
the repository (Davenport, 2002).

Watershed management partnerships have organizational and technical needs that
correspond to the capacity of VO-based watershed management DSS. An Internet-based
infrastructure can provide critical support in forming partnerships and subsequent
collaboration and ongoing networking. A watershed management VO extends the utility
of irrtemet partnerships in two general areas: ﬁrst, it offers a much more powerful
knowledge-sharing mechanism, and second, it provides a natural extension to the
decision-making functionality of traditional watershed partnerships. Even though many
researchers have utilized the web and Internet for environmental DSS development (e.g.,
Watson et al., 2002; Cheng et al., 2004; Dymond et al., 2004; Hluchy et al., 2004; Xie
and Yapa, 2006; Wang and Cheng, 2007), only a few realized the value of V0 and
associated Grid computing technology (e. g., Hluchy et al., 2006). A VO-based watershed
management DSS can effectively function to integrate information from multiple
organizations and disciplines. It can also assist in providing watershed inventories,
assessments, problem identiﬁcation, modeling, and analyses of alternative scenarios for
stakeholders so they can understand issues at hand. This system can be used as a
communication tool and assist dissemination of project results to maintain active public
engagement and support. By providing access to anyone at any time, in, any location with

a computer, a powerful tool is created. Development of VO-based watershed DSS may

29

result in a distributed system that participants can continue to use across multiple
locations over time, as demonstrated later in this dissertation.
The next section describes the speciﬁc objectives of this research and the outlines

of the dissertation organization.

1.4. Objectives and Organization

A case study approach is used here to test the hypothesis that V0 model can
beneﬁt watershed management by providing effective ways to integrate science and
decision making and to operationalize a multi-organizational and multi-disciplinary,
comprehensive, watershed management SDSS. Given its multi-organizational structure
and multi-functional management nature, the integrated watershed management SDSS
has multiple information and analysis requirements. It also requires intensive computing
capabilities that both Grid computing and geospatial computing provide. The multi-
organizational nature of the system also demands organizational theories like Virtual
Organization and Actor Network Theory (ANT) for requirement analysis.

The main goal of this research is both to conceptualize a watershed management
V0 and to discuss certain implementation aspects of this concept, including exploring
technical issues related to hydrologic modeling within the context of such a V0.

The speciﬁc objectives include:

0 Conduct a requirements analysis for a VO-based watershed management DSS;

o Conceptualize, partially implement and evaluate an Intemet-based integration

framework for VO-based watershed management SDSS;

o Parallelize a water-quality model to utilize distributed processors; and

30

0 Develop and evaluate a new DEM-based ﬂow analysis algorithm.

The following chapters focus on different issues and objectives outlined above.
Chapter 2 discusses V0 and ANT organizational theories and applies them to a
requirements analysis for a VO-based watershed management SDSS. Chapter 3 discusses
Grid computing technology and the conceptualization, partial implementation and
evaluation of an Intemet-based integration framework for the VO-based watershed
management SDSS. The parallelization of a water-quality model is also developed to
demonstrate how to utilize distributed computing resources. Successful watershed
management depends not only on organizational aspects described in the previous
section, but also technical aspects such as data sampling, modeling, monitoring etc. In
this research, a detailed study on fundamental DEM (Digital Elevation Model) based
ﬂow modeling algorithms is also conducted. Reviews of various ﬂow modeling
algorithms and details of this study can be found in Chapter 4. As a result of this study, a
new ﬂow routing algorithm based on principles in ﬂuid mechanics called FlowNet is
created and evaluated. This new F lowNet algorithm is applied to 21 watersheds in three
different physiographic regions along with two other algorithms for comparison
purposes. It is also validated against wetland data available for Michigan watersheds. The
comparison and validation results in Chapter 4 show advantages of the FlowNet
algorithm for watershed modeling. This algorithm can clearly be integrated into the
modular VO-based watershed management SDSS to provide better information for
decision making. Chapter 5 presents conclusions on the research and future research

recommendations.

31

Chapter 2. Requirement Analysis for a VO-based Watershed Management SDSS

2.1. Introduction

Water is a renewable but ﬁnite resource. It’s an important resource necessary for
sustainability of all socio-economic development activities including industry and
agriculture. When water and land become limited resources, competition among various
stakeholders for use of fresh water is bound to increase (CTIC, 1995). To ensure long-
term sustainability, watershed managers recognize the need for a holistic watershed
management approach that involves coordination among different organizations and
collective decision-making based upon information from different sources (Gregersen et
al., 2008). As indicated in Chapter One, the ultimate goal of watershed management is to
improve watershed ecosystem health and support economic developments such as
agricultural production on a sustained basis (Brooks et al., 1997; Heathcote 1998;
Gregersen et al., 2008). Therefore, the needs and impacts of every stakeholder on the
natural functions of a watershed need to be understood for making decisions regarding
resource management and sustainable development (Davenport, 2002). There is also a
need for a clear identiﬁcation of the steps involved and the detailing of the requirements
that adhere to speciﬁc policies all of which are a requirement for an effective
management plan. So a key to effective water management is to have in place an enabling
framework at a watershed/river basin scale and multi-stakeholder partnerships at all
levels (Walters, 2000; Davenport, 2002). This framework should reinforce the fact that
nothing happens in isolation and that everything is connected by the land and water
within the watershed. Given the relationships and interdependencies that exist between

land, water, and various stakeholders, a comprehensive, all-inclusive approach to

32

considering the factors affecting water and other natural resources within a watershed
needs to be clearly understood (Gregersen et al., 2008). Any decisions regarding water
resource management must be done in a socially, environmentally, and economically
sustainable manner (Chander 2005). Davenport’s partnership driven four-phase
watershed management approach described in Chapter One clearly meets all these
requirements.

A critical aspect of implementing the above framework is an application of a
dynamic integrated assessment system. The system should contribute to provide
methodological support to deal with the general problem of integrated watershed
management implementation. In particular, the system should support the management of
large amount of multi-sectoral and multidisciplinary information, and the communication
between the scientiﬁc and the policy sector and between decision-makers and involved
stakeholders (Chen et al., 2004). The system should also support public participation and
social learning. Social learning here refers to the growing capacity of a multiple actors
network (those concerned by the watershed) to develop and perform collective actions
related to integrated watershed management (Pahl-Wostl et al., 2007).

Integrated Watershed Management is a social-relational activity (interests, water
practices, information, knowledge, funds are spread over many actors) and a complex
technical task; both cannot be separated (Gregersen et al., 2008). Social learning
corresponds both to this participatory social/technical process as well as to the outcomes
of this process. This collective problem solving approach requires that the actors meet
each other, develop relational practices. The quality of these relational practices is

fundamental from a social learning perspective: it is based on reﬂectivity, reciprocity and

33

respect for diversity (Bandura, 1977; Pahl-Wostl et al., 2007). The quality of these social
interactions is supposed to determine the awareness for interdependencies between the
participants and the acceptance of a diversity of interests, of mental frames, of
knowledge. The different stakeholder groups in a basin have to understand that a complex
issue such as integrated watershed management can be better resolved in a collective
way, relying on disseminated information and knowledge (Pahl-Wostl et al., 2007).

Considering the huge number of people concerned by integrated watershed
management, traditional interactions between experts and decision-makers are not
sufﬁcient any more and other additional relational mechanisms have to be considered to
go across geographical and organizational scales, including the public at large (Maurel et
al., 2007).

In this context the author argues that a Virtual Organization (V 0) based
watershed management Spatial Decision Support System (SDSS) capable of linking
multiple organizations, databases and modeling at the watershed or sub-watershed level
can play a crucial role to support the implementation of the enabling framework and the
social learning dimension of public participation through two-ways communication
processes for the integrated watershed management. A sub-watershed is a smaller basin
of a larger drainage area that all drains to a central point of the larger watershed
(Gregersen et al., 2008). But as indicated by Zimmerman and Nardi (2006), requirement
analysis for a multi-sited, multi-user cyberinfrastructure can be a difﬁcult task. Therefore,
both organizational and technical elements are used in this chapter to tackle the

requirement analysis for a VO-based watershed management SDSS.

34

In the following sections, Actor Network Theory (ANT), Virtual Organization
(V O) concept and Spatial Decision Support System (SDSS) are introduced ﬁrst. Essential
elements of ANT, V0 and SDSS are described and then applied to the requirement
analysis of a VO-based watershed management SDSS. Both organizational and technical
requirements are analyzed. An example watershed management V0 is employed to
facilitate the organizational requirement analysis based on ANT and V0 concept. The
integration of Grid computing technology with various components of SDSS is discussed
for technical requirement analysis. Other technical requirements are also identiﬁed.

Please refer to Figure 2.1 for the organization of chapter two.

. ,

 

 

 

 

 

 

Requirement Analysis Organizational & Technical
For VO-based Requirements
0 Watershed Management For VO-based
‘ 5033 Watershed Management
SDSS /

 

® % \
Figure 2.1 Chapter 2 Organization

The goal of this chapter is to collect requirements for a VO-based watershed management

SDSS to guide the conceptualization and partial implementation of an Intemet-based

integration framework for the VO-based watershed management SDSS in the next

chapter.

35

2.2. Actor Network Theory (ANT)

ANT is a social theory and a framework for investigating society-technology
interactions that evolved from the work of Michel Callon and Bruno Latour. ANT is very
useful for analyzing and understanding the social and technical nature of the watershed
management process. The ANT framework was developed by tracing heterogeneous
networks of actors and their interactions involved in the production of science and
technology (Latour, 1987, Callon et al., 1996). ANT offers explanations of how
technology becomes acceptable and is taken up by groups in society. It suggests how
technology is socially constructed. ANT provides a ﬁne-grained approach to analyze the
mechanism by which social action shapes technology and technology shapes social
action. The primary focus is on actors and how they are involved in the shaping of
technology. ANT considers the whole world as patterned networks of heterogeneous
entities containing both human and non-human elements (Latour, 1999). An actor
network is a network where elements of any kind may be included: humans,
technological artifacts, organizations, institutions, etc. ANT denies purely technical or
purely social relations. It describes the society as heterogeneous. ANT considers both
social and technical determinism to be ﬂawed and proposes instead a socio-technical
account in which neither social nor technical positions are privileged. Since all
interactions between humans are mediated through objects of one type or another (Law,
1992), ANT accepts humans, non-humans and their intermediaries as actors and
considers all actors equal (Latour, 2005). An actor is any entity that interacts with other
actors or serves as an intermediary between actors. By removing the analytical divide

between humans and objects, the ANT framework enhances researchers’ ability to

36

examine the nature of interactions that are the building blocks of networks both within
and beyond organizations (Latour, 1991, 1992; Law, 1992; Martin, 2000). The ANT
framework is very useful for analyzing the partnership-driven watershed management
process and the requirements for a V0- based Watershed Management Spatial Decision
Support System.

Interactions between actors are the primary building blocks of actor-networks and
their many forms are called ‘translations’ (Callon, 1985; Latour, 1987; Latour, 1997;
Latour, 2005). Between human actors, the translation of interests is roughly similar to
persuasion and the negotiation of common interests. Between humans and objects,
translation occurs during design stage when the object is given with its purpose, program
or script in how it interacts or affects other actors (Akrich 1992). Further translation takes
place between the object and the actors it encounters as the initial program or script is
changed through interaction.

When actors and their interactions are taken together, they form a network. Latour
(1997) cautions about the differences between an ANT network and technical or social
networks. Social networks exclude non-human actors just as technical networks exclude
the human. ANT networks incorporate both with the linkages consisting of stabilized
translations and interactions between actors. Actor-network theory assumes the
heterogeneous nature of these linkages and seeks to understand the reasons for this
heterogeneity (Law, 1992). This uneven distribution in networks requires two types of
consideration: temporal and spatial. The temporal considers the durability of the network
and its effects and the spatial considers the nature of circulation within a network. Both of

these characteristics become central focuses in the tracing or delineating networks.

37

Interaction and circulation by four action intermediaries: control, money, people’s
abilities and information, can be utilized to delineate relationships among actors (Martin,
2000)

Convergence is the degree to which the processes of translation and circulation
lead to agreement (Callon, 1991). Convergence is a special case of translation that aligns
the elements in a network. Alignment and durability can lead to punctualization, a point
where the network supporting an actor disappears from view. This takes place when the
network components that are responsible for the production of objects or performance of
functions are summed up in symbols or artifacts that encapsulate the network (Callon,
1991). In the case of technology, punctualization takes the form of a ‘black box’ (Latour,
1987).

Actor Network Theory has been widely used in information system research to
study the implementation of information systems and in other situations involving
technological innovation. Adams and Berg (2004) used ANT to show how reliability of
health information sites on the Internet is currently being negotiated. They described
some of the central actors and initiatives, including certiﬁcation schemes, portals, and
rating systems, and attempts to analyze which issues are at stake, and what reliability in
this novel setting ultimately comes to mean. Faraj et al. (2004) described the evolution of
the Web browser technology in the context of commercial actor-networks of innovation.
They attempted to take the actor network as a unit of analysis and thus reject the subject-
object distinction. In the GIS area, Martin (2000) applied ANT to study conservation GIS

implementation in Ecuador. Later in this chapter, ANT will be employed to analyze the

38

social technical watershed management process in assistance of requirements analysis for

a VO-based watershed management SDSS.

2.3. Virtual Organization (V 0)

As deﬁned in Chapter One, a Virtual Organization is “an identiﬁable group of
people or organizations that makes substantially more use of Information and
Communication Technologies than physical presence to interact, conduct business and
operate together, in order to achieve their objectives” (Sieber et al., 1998). The V0 is a
powerful model for accomplishing group objectives in a dynamic manner independent of
the geographic location of group members (Palmer & Speier, 1997). A typology proposed
by Shao et al. (1998) identiﬁes four characteristic features of a virtual organization:

0 Connectivity, which establishes linkages where none previously existed
0 Purpose, which creates a common incentive for agents in the virtual organization
0 Technology, which enables connectivity

0 Boundary, which separates those who are part of the V0 from those who aren't

The role of trust is critical, since decisions must be made without the same degree
of face-to-face interaction in more traditional settings, and because data, strategic
decisions, and objectives must be shared with semi-independent participants in the V0.
Because of this greater autonomy, V0 participants tend to be more equal, with greater
independence than is typical in a hierarchical organization.

The V0 paradigm has been highly useful for certain business activities,
particularly those that are technologically oriented and require ﬂexibility and rapid

adaptation (Jagers et al., 1998). It has also been adapted in academic settings at large and

39

small scales. At large scales, it is common for multidisciplinary research teams with
members in several academic institutions to collaborate on projects. At small scales,
multiple-authored papers are typically developed in concert, with portions of work done
by different authors in physically different places. In either case, electronic transmission
of documents and data, as well as ideas, keeps the work coordinated between multiple
parties. The network enables the rapid ﬂow of communication, which is critical for the
virtual organization. Strader et al. (1998) indicated that an information infrastructure
based on the Internet and Intranet technology can support the communication required for
effective virtual organization management.

In the business world, because of the growing demands of the global market,
small and medium sized business enterprises see virtual organization type of
collaboration between companies as a powerful option to give companies new, successful
opportunities and ways to stand the pressures of global competition.

Harmus et al. (2005) listed a number of beneﬁts of V0 for business enterprises:

0 Companies can easily access and use competencies not available internally.

0 As consequence, a virtual organization can go for larger, more complex and

thus higher value opportunities, than individual companies.

0 The V0 can more ﬂexibly balance bottlenecks.

o The risks for the projects are effectively shared among the participating

companies.

0 In trustful relationships there is a strong incentive for joint problem solving

and customer satisfaction.

40

0 VO members can have access to very different (international) markets or
customers bringing in new opportunities.

0 The combined mass of companies allows stronger market presence, e.g., on
trade fairs or through advertisement than an individual company.

0 V0 serves as a very good training ground for management skills. People
become more aware of different working practices across the companies,
develop an outlook for new opportunities, and sharpen their skills in
collaboration and project management.

0 V0 also enables companies to align investment decisions, strengthening their
core competencies or production facilities, while relying for other facilities on
other partners. This can signiﬁcantly increase efﬁciency of investments and
utilization of resources.

They also mentioned signiﬁcant risks and hurdles for the build-up of a V0:

0 Customers often like to have a single company being liable for an order — and
also many banks to do not yet have experiences in ﬁnancing such projects.

0 Companies need to build the trust to rely on cooperation.

o The competence mix and market orientation of a V0 needs to be carefully
evaluated and jointly deﬁned.

0 Setting-up and running VOs and projects for business opportunities requires
managerial skills — up to the point that the level of managerial competence can
be assessed by the level of activity in such networked organizations.

In order to use the beneﬁts and avoid the hurdles, the V0 needs to be well

planned and implemented. Required competencies and skills as well as methods,

41

processes and conditions need to be understood. Lawson et a1. (2005) investigated a
virtual organization for Australian tool making industry and indicated the trust is a crucial
underlying aspect of successful collaboration. Joia and Neto (2004) studied the
implementation of virtual enterprises for government agencies in Brazil and identiﬁed
security, organizational culture and training as critical success factors for such endeavor.
Dunne and Browne (2005) examined four virtual organization cases and found the main
reasons for their failures are: insufﬁcient emphasis on initial planning; informal and
unstructured sourcing and selection methods for potential partner identiﬁcation;
communication barriers and inadequate information technology support.

The V0 theory proves beneﬁcial not only for the business community, but also
for the environmental management community. Arar’rjo et al. (2005) presented a virtual
Hydro-Meteorological Scientiﬁc Network that connects data, computing power and
human expertise together in a productive way. Hluchy et al. (2006) described a virtual
organization for ﬂood forecasting. Their V0 includes computation cycle providers,
storage providers, data providers, experts, developers and end users. It utilizes the
computational grid to run a large cascade of meteorologic, hydrologic and hydraulic
models for ﬂood forecasting purpose. Visualization and workﬂow management functions
are also available in the grid environment to facilitate modeling process. The user
friendly graphical interface allows users to interact with the system without knowing
anything about grid. One important fact about this system is that it is designed to be used
solely by meteorologic and hydrologic experts within the V0. But in a watershed
management V0, a variety of users such as decision makers, experts and members of the

public etc. have to be considered. The V0 theory can certainly help to further our

42

understanding of different needs of these users. For this reason, it will be applied later in
this chapter for the requirement analysis of VO-based watershed management SDSS. The
above mentioned beneﬁts and hurdles of V0 clearly shed some lights on how to build
successful Watershed Management VOs.

As indicated in Chapter One, an important technical aspect associated with V0 is
the Grid Computing technology. It will be explored in more details in the requirement
analysis section.

The ANT and V0 theories described above form a solid organizational
foundation for the requirement analysis of a VO-based Watershed Management SDSS. In
the next section, Spatial Decision Support System, a major technical component for a

VO-based Watershed Management SDSS, will be discussed.

2.4. Spatial Decision Support System for Watershed Management

2.4.]. Overview

Walsh (1993) deﬁned Decision Support System (DSS) as a computer system,
hardware and software, designed to support decision makers interactively in thinking
about and making decision about relatively unstructured problems. A DSS provides a
framework for integrating modeling capabilities with database resources to improve
decision-making processes. Decision makers can interact with the system using
intuitively designed easy-to-use graphical user interfaces. Historically, development of
the D885 can be attributed to the ineffectiveness of the old-fashioned Management
Information Systems (MIS) available in the 1960s and 1970s. They were data oriented,

most of which simply retrieved data from large databases on selected queries. The

43

demand for better modeling facilities and a greater degree of interaction with solution
processes spurred development of the new tool known as the DSS (Armstrong et al.
1990). Decision support systems have progressed from tools that simply provide users
with the resources to formulate, assess, and compare alternative solutions to applications
that educate users about the problem context and how the problem has come into
existence (Bellamy et al., 1996). With education as the primary focus, managers and
planners can more easily adapt to changing situations through understanding of the causal
relationships of the situation at hand (Climaco et al., 1995; Bellamy et al., 1996).

Spatial Decision Support Systems (SDSS) are a type of computer system that
combine the technologies of GIS and DSS to assist decision-makers with problems that
have spatial dimensions (Walsh 1993). Spatial Decision Support Systems (SDSS) are
developed to integrate data, knowledge, and modeling results to identify, evaluate, and
recommend alternative solutions to spatially distributed problems (Djokic, 1996; Bellamy
et al., 1996; Prato and Hajkowicz, 1999). A SDSS focuses on a limited problem domain,
utilizes a variety of data, and brings analytical and statistical modeling capabilities to
solve the problems. It further depends on graphical displays to convey information to the
users. It can be adapted to decision-maker’s style of problem solving, and can easily be
extended to include new capabilities as needed (Densham et al. 1989, Armstrong et al.
1990). Leipnik et al. (1993) deﬁned SDSSs as integrated environments, which make use
of the databases that are both spatial and non-spatial, models, decision support tools like
eXpert systems, statistical packages, optimization software, and enhanced graphics to

Offer the decision makers a new paradigm for analysis and problem solving.

44

Densham et al. (1989) suggested the following three key characteristics of

effective decision processes, which can be used to guide the identiﬁcation of the goals of

a SDSS.

“Iterative: Decision problem is iterative because decision makers generate and
evaluate a set of alternative solutions, gaining insights which are input to, and
used to deﬁne, further analyses.”

“Integrative: Integration occurs because decision-makers, who hold the expert
knowledge that must be incorporated into the analysis with the quantitative
data in the models, evaluate alternatives across a broad range of pertinent
criteria, making value judgments that materially affect the ﬁnal outcome.”
“Participative: The participation by decision-makers returns control over the

decision-making process to them, enhancing the quality of that process.”

Based on Geoffrion’s (1983) suggested ﬁve distinguished styles in a DSS design,

Densham et al. (1989) further simpliﬁed them as the following six characteristics of

SDSS.

SDSSs are designed to tackle semi or ill-structured problems where either the
problems or the objectives or both are not fully and coherently speciﬁed.

They often adopt interactive and recursive ways of system development
known as multipass approach, which contrasts the more traditional serial
approach involving clearly deﬁned phases like requirements speciﬁcations,
detailed design, programming, testing and implementation.

The designs place high value on the ﬂexibility of system use and ease of

adaptation to the evolving needs of the users.

45

0 They strive for a genuine integration of data sources and models, including
appropriate interfaces to transaction processing and database management
systems.

0 Users are of prime importance during DSS design. The underlying technology
comes second. That is why much emphasis is given on the interface to be
user-friendly.

o The users should be able to generate a series of possible solutions by running
different ‘what-if scenarios in the models.

In natural resource management, SDSS have proven to be effective in a variety of
applications such as ﬂood prediction (Al-Sabhan et al., 2003), river water quality
evaluation (Wang et al., 2005), conservation program management and best management
practices assessment (Rao et al., 2007). Al-Sabhan et al. (2003) listed issues associated
with stand—alone or GIS based hydrologic modeling practices: difficult interface; GIS
knowledge requirements; platform dependency; programming knowledge needs;
customization difficulty; limited accessibility by non-experts and the public; lack of
collaboration support; and ﬁnally, costly data acquisition and communications. They
further argued that a web-based hydrologic modeling SDSS can help solve most of these
problems. They indicated such system can offer openness, user friendly interface,
transparency, interactivity, ﬂexibility, and fast communication and be directly accessible
to a broad audience including decision makers, stakeholders and the general public.
Using a regional river water quality model from EPA and web-based GIS software, Wang

et al. (2005) created a web-based river water quality SDSS with similar advantages.

46

Even though a lot of effort, time, and money were spent on their development, the
adoption and success rates of decision support systems and spatial decision support
systems have been relatively low (Uran and Janssen, 2003). Newman et al. (2000)
attributed this lack of adoption to the complexity and quantity of data inputs, limited
computer ownership, and a lack of understanding by potential users of the underlying
modeling theory (Newman et al., 2000). Uran and Janssen (2003) examined ﬁve spatial
decision support systems for coastal zone and water management and found that
difﬁculties in specifying alternatives, complexity in navigation resulting from a large
number of options, and lack of adequate support to the decision process are major reasons
for the low adoption rates. Furthermore, they indicated that a closer involvement of users
during development process would potentially lead to higher adoption rates in the future.
An important observation made by the National Research Council’s Committee on
Watershed Management (1999) is that the difficulty in developing a DSS is not a lack of
available simulation models, but rather making these models available and
understandable to decision makers. Over the past few decades, federal government
agencies have spent millions of dollars on model development. These simulation models
are used extensively in research settings, but they are rarely incorporated into the

decision-making process in an easy to understand and easy to use fashion.

2.4.2. Components of SDSS

Spatial Decision Support Systems are created either by writing the entire program
from scratch, or by linking existing applications that provide the necessary tools (Djokic,

1996). Since writing the application from scratch is complex and often reinvents

47

procedures to provide required functionality, SDSS shells are usually developed by
integrating existing applications (Djokic, 1996). In a watershed management context,
Spatial Decision Support Systems typically include a user interface, geographic
information systems, hydrologic simulation models, and database management systems
(Sugumaran, 2002; Densham, 1991; Fedra, 1991). In an actor network, all these
components can be considered as actors.

In a watershed management SDSS, linkages are often made between hydrologic/water
quality simulation models, a GIS, and a relational database management system which
provides an efﬁcient means to store, analyze, and visualize results from the model (Yoon,

1996).

2.4.2.1. User Interface

Historically decision support system research has focused on data, procedures,
rule sets, text, forms, and spreadsheets associated with the problem decision domain
(Sankar et al., 1995). However, the user interface has been considered to be the most
important aspect (Sprague and Carlson, 1982) because acceptance of decision support
systems is largely dependent on their ease of use (Uran and Jassen, 2003), which is often
determined by the user interface. The user interface controls the communication between
the user and the application. A good dialog should be error tolerant and provide user help
as carefully phrased informative messages (Molich and Nielsen, 1990). The user interface
should be easy to learn, allow graceful shifting from one task to the next, and provide a
high level of guidance and feedback based on a user’s interactions while giving the user

the sense of being in control (Holsapple and Whinston, 2001).

48

2.4.2.2. Geographic Information Systems

Geographic Information Systems are used as integration platform for Spatial
Decision Support Systems. GIS can incorporate data from a variety of sources addressing
multi-ownership, infrastructure, and economic concerns. GIS offers the functionality to
explicitly model the spatial heterogeneity in landscapes, pasture utilization, and
distribution of management units (Bellamy et al., 1996; Bellamy and Lowes, 1999) and
effectively analyze non-point source pollution problems (Fraser et al., 1996; Yoon, 1996;
Fraser et al., 1998; Guertin et al., 1998; Basnyat et al., 2000).

As indicated in Chapter One, the GIScience community has long been interested
in integrating distributed computing capability into GIS. In the past few years,
commercial GIS vendors have developed server based GIS software products capable of
utilizing Internet technology. This new breed of server based GIS has the potential for
rapid increases in the efﬁciency and effectiveness of the ways in which we share
geographic data and geoprocessing capabilities (Peng & Tsou, 2003).

GIS are widely used in watershed management for reporting and technical
analysis. It can also be used to support public participation. Speciﬁcally in a watershed
management context, GIS can be used to identify legitimate stakeholders; manage shared
geo-information; communicate geo-information; support GIS-based watershed modeling;
collect and communicate public knowledge, perceptions and comments; and bring people

together.

49

2.4.2.3. Hydrologic Simulation Models

Hydrologic models have long been used in watershed management. They provide
an important tool for evaluating and assessing hydrologic systems, and environmental
managers are increasingly reliant on this technology to support decision making. The
classiﬁcation, application, and development of available models have been reviewed in
great detail by others (see Singh, 1995; Maidment, 1993). The majority of models applied
today perform simulations using methods derived in the early 20th century.

Even though the ﬁrst watershed-scale, computer based simulation model was
developed almost forty years ago in the Stanford Watershed Model (Crawford and
Linsley, 1966), simulating watershed scale processes continues to be an extremely
challenging activity, in spite of recent advances in data quantity and quality, and
technologies to manage the spatial attributes of watersheds. The National Research
Council (1999) recommends that tools be developed to facilitate the transfer of
simulation modeling technology, which will provide modeling results to managers for
decision making in an easy to understand fashion even when they are based on imperfect
information.

Recently GIS has been utilized widely in hydrologic modeling to develop
parameter sets and visualize simulation results. Sui and Maggio (1999) identiﬁed four
approaches for integrating models and GIS: embedding GIS in the hydrologic model and
vice versa, and loose and tight coupling between the components, with each approach
having advantages and disadvantages. Embedding the GIS ﬁrnctionality in the hydrologic
model provides the most ﬂexibility for application design, eliminating dependencies on

previous GIS data structures, but most hydrologic modeling packages do not have the

50

visualization capabilities of commercial geographic information systems (Sui and
Maggio, 1999). Embedding hydrologic modeling functionality in GIS has recently been
conducted by vendors such as ESRI and Intergraph, but the modeling capabilities are
usually simplistic and calibration and veriﬁcation must be conducted outside the GIS (Sui
and Maggio, 1999). Loose coupling is completed using “stand alone” GISs and
hydrologic models that exchange data using an ASCII or binary data format. Loose
coupling relies on existing components, therefore reducing the programming required to
develop these technologies, but data conversion between the components can be tedious
sometimes (Sui and Maggio, 1999). The ﬁnal approach, tight coupling embeds 3
hydrologic model within a commercial GIS utilizing its custorrrization capability with
scripting languages such as VB Script, ESRI Avenue or AML etc. (Sui and Maggio,
1999)

These approaches have produced numerous applications that utilize a combination
of loose or tight coupling methods (Sui and Maggio, 1999). The Automated Geospatial
Watershed Assessment Tool (AGWA; Miller et al., 2002; Miller et al., 2006) uses a
hybrid between the loose and tight coupling where specialized routines are created using
ESRI’s Avenue programming language to prepare input ﬁles, but the communication
between the GIS and hydrologic models are performed using an ASCII text format. A
similar approach was used in developing a generic object-oriented modeling framework
(McKinney and Cai, 2002) and integrating GIS into Agricultural Nonpoint Source
Pollution Model (AGNPS: He, 2003; He et al., 2001), Better Assessment Science
Integrating Point and Nonpoint Sources (BASIN S: Lahlou et al., 1998), Areal Nonpoint

Source Watershed Environmental Response Simulation (ANSWERS: De Roo et al.,

51

1989), and the Soil and Water Assessment Tool (SWAT: Srinivasan and Arnold, 1994).
The US Army Corps of Engineers also used a loosely coupled approach integrating GIS
and their Hydrologic Engineering Center River Analysis System (HEC-GeoRAS:
Ackerman, 2002). AGNPS and ANSWERS were integrated into open source GRASS
(Geographic Resources Analysis Support System) GIS system as extra modules
(Srinivasan & Engel, 1994; Rewerts & Engel, 1991). A SWAT/GRASS interface was
built into the earlier version of SWAT package (Srinivasan and Arnold, 1994). Special
routines for soil erosion/deposition modeling were developed in GRASS by Mitasova and
others (Mitasova et al., 1996; Mitas & Mitasova, 1998). Other hydrologic models such as
KINEROS, TOPMODEL, FEA and CASCZD were also incorporated into GRASS as
add-on modules (see Woollriser et al., 1990; Chariat & Delleur, 1993; Vieux et al., 1990;
Saghaﬁan, 1993).

Intemet-based hydrologic modeling applications such as Michigan State
University’s Online RUSLE and Purdue University’s L-THIA have also been realized
(Ouyang and Bartholic, 2001; Choi et al., 2005). Cate et al. (2007) also implemented
DOTAGWA, the web-based AGWA tool for a selected pilot watershed. These
applications offer several advantages over traditional stand-alone computer applications.
A typical Internet application offers a centralized simulation model that does not require
installation on local computers and provides access to the latest version of the data and
software at all times. Internet applications do not require advanced software or hardware
for the end user, since these applications operate through a web browser, with most of the
processing conducted on the server. However, deploying applications over the Internet

may alienate user groups since access to the Internet is not ubiquitous.

52

2.4.2.4. Relational Database Management Systems

A relational database management system is a database management system
based on the relational model in which data is stored in the form of tables and the
relationships among the data is also stored in the form of tables (Rob and Coronel, 2006).
The relational database management system for a SDSS must support integrated spatial
data storage, spatial query, spatial analysis and cartographic display. Because the data
requirement for SDSS can be very large, relational database management systems are
considered better than the traditional ﬂat-ﬁle database systems in a SDSS. Most GIS and
database vendors today provide strong spatial data support in mainstream relational

database management systems (Zlatanova and Stoter, 2006).

2.5. Requirement Analysis for a VO-based Watershed Management SDSS

In the previous sections, Actor Network Theory (ANT), the Virtual Organization
(V 0) concept and Spatial Decision Support System (SDSS) technology have been
introduced separately. In this section, they are applied together to the requirements
analysis for a VO-based Watershed Management SDSS. Both organizational and

technical requirements are investigated.

2.5.1. Organizational Requirement Analysis

2.5.1.1. Overview
The complexity and uncertainty of current water resource issues require new

forms of governance that replace the traditional control-oriented hierarchical systems by

53

participatory and ﬂexible systems, based on experimenting and social leaming between
multiple actors (Doppelt, 2000; Gregory, 2000; Tabara, 2003; Woodhill, 2003). This
social learning approach supposedly can contribute to a more integrated and sustainable
way of managing water resources on a watershed basis. Integrated Watershed
Management is the process of formulating and implementing a course of action involving
natural and human resources in a watershed, taking into account social, economic,
political and institutional factors operating within the watershed and the surrounding river
basins to achieve speciﬁc social objectives (UN ESCO, 1993; F alkenmark, 2002).
Integrated Watershed Management involves multiple stakeholders and the public
at large. Stakeholders include all individuals, groups or organizations that are directly
concerned by actions that others take to solve the problem/deal with the issue (Gray,
1989). Green (2003) distinguished the following criteria for identifying stakeholders in
watershed management:
0 Those whose actions can signiﬁcantly promote or inhibit the achievement of
watershed management in the case of a particular watershed.
0 Those who have knowledge or experience that can contribute to that achievement.
0 Those who will be affected in one way or another by the outcome of the particular
choice (including those who will bear the costs).
0 Those who have an interest in the watershed in question.
Typical stakeholders involved in watershed management include: professionals,
authorities and elected ofﬁcials, local groups and non-professional organized entities
(broken down into groups focusing on a place such as a group resident association, and

those focusing on an interest, such as ﬁshermen), individual citizens, farmers and

54

companies representing themselves and ﬁnally, the experts (government and water
authority staff, academics, private consultants, etc.). The main challenge for watershed
management can be formulated as follows: how can different stakeholders learn to take
joint decisions related to water and other resources in a watershed, in which each has a
speciﬁc stake and interest, in order to arrive at collective sustainable solutions? So the
key is about people learning how to deal with each other and their interdependence, while
they are learning together to deal with the interconnected issues of their watersheds
(Pahl-Wostl etal., 2007).
To help people understand the complex dynanrics of the environmental system in

a watershed, efforts need to be devoted to developing appropriate spatial decision support
systems for decision makers, citizens and lay people. Information contained in the system
should be timely, comprehensive, meaningful, and uncertainties must be clearly
expressed. The system should have the capability to integrate knowledge from different
disciplines about an environmental problem along the whole chain of causes and effects
to provide useful information for decision makers. Integrated Watershed Management
often involves the following disciplines:

o biophysical sciences: hydrologists, soil scientists, geologists, agronomists, etc.

0 engineering sciences: civil, environmental, mechanical, etc.

o computational sciences: GIS specialists, numerical modelers, etc.

0 social, political and economical sciences

o managerial sciences

Such a system should support and facilitate the critical communications among

various stakeholders with different backgrounds, so they can better understand each

55

other’s view on the watershed. Actually in all areas of environmental policy making,
including land use and water resources management, the call for improved
communication between scientists and decision makers is ubiquitous (Callahan et al.,
1999; Schiller et al., 2001; Siepen and Westrup, 2002).

Therefore, the ﬁrst essential requirement for a watershed management SDSS is to
support multi-organizational and multi-disciplinary integration and communication.

According to Actor Network Theory, all stakeholders and technical components
of a SDSS involved in a watershed management context can be equally treated as actors
in a watershed management actor network. Based on their relationships and the
connectivity under the common watershed management goals, they also form a watershed
management V0. Figure 2.2 describes the general structure of a watershed management

virtual organization that is applicable to any watershed management context.

 

 

 

Watershed Groups Government Agencies/Ofﬁces
Data/GIS/Model Server Data/GIS/Model Server

 

 

 

 

 

 

    
 
  

 

 

 
  
 

 

 

Visualization Server Wireless Digital

for Virtual Reality Agreements Sensors and Devices

Teleirnmersion INetwork/DSS for Ground Truth
Data Collection

 

 

 

 

 

 

 

 

 

 

 

 

 

General . . Other Stakeholders
. Unrversrty Researchers 1
PM” Data/GIS/Model Server Data/GIS/Mc’d" sew"
Knowledge Base in
Virtual University

 

 

 

 

 

 

Figure 2.2 Watershed Management Virtual Organization Diagram

56

This diagram includes both organizational and technical elements of a watershed
management V0. The central circle deﬁnes the different connectivity between any
elements and it does not imply a central power that controls the relationships. A
combination of organizational agreement, technical network such as the Internet and
distributed decision support system can be used to deﬁne the relationship between any
two members. In the meantime, any member is an independent entity and can provide its
own services to watershed management. But as we know, understanding the inter-
relationships among different stakeholders is the key to successful watershed
management. So such VO focusing on connecting different organizational and technical
elements can be very beneﬁcial to watershed management. The purpose of this general
VO structure is to provide comprehensive support for watershed management practices.
What ﬂows across the lines in the diagram could be any combination of action
intermediaries such as control, skills/abilities, money or information. An example
watershed management V0 is described in the next section with Actor Network Theory

applied to it to identify actors and their relationships.

2.5.1.2. Organizational Requirement Analysis for an Example Watershed
Management V0

The Midwest Partnership for a Watershed Management Spatial Decision Support
System was created in April 2002. The goal of the partnership is to develop, promote, and
disseminate web-based spatial decision support systems to help manage watersheds in the
Midwest region. In particular, the main objective is to make these systems freely

available via the Internet to local officials, natural resource managers, and the general

57

public. The major organizations involved in the partnership include EPA Region 5, EPA
Ofﬁce of Research and Development, Michigan State University, Purdue University and
the Wisconsin Department of Natural Resources. Because of the nature of this
partnership, it can be considered as a typical example of watershed management VO.
Figure 2.3 describes the actor network for this example partnership watershed

management V0.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Michigan State University
Web-based DSS Watershed Groups

7 I

/
/ |
l
l
\ I
\ I
\ l
X Y

Purdue University General Public
Web-based Modeling
System
Action Intermediaries:
Control Skills/Abilities Money Information
‘——’ ‘ ................ > ‘_ . . — . ’ ‘— — _ ,

 

 

 

Figure 2.3 Actor Network for the Midwest Partnership for a Watershed Management
Spatial Decision Support System
As illustrated in the diagram, all partnership members can be treated as actors and
they all have different roles. Public Universities - Michigan State University and Purdue
University - lead the main effort in tool development. EPA Region 5 and the EPA Ofﬁce
of Research and Development are responsible for planning the efforts and the general

search for funds. The main task of the Wisconsin DNR is outreach and capacity building.

58

The relationships among partnership actors are described by four action intermediaries in
the diagram. EPA as a regulatory and funding agency provides grant money and
associated controls to the two universities and the Wisconsin DNR. It also provides
environmental information and modeling tools as an information action intermediary.
The two universities can share information and people’s skills/abilities. The Wisconsin
DNR as a state governmental organization may also provide grant money and associated
controls to the two universities. Information also ﬂows both ways among the EPA, the
universities and the Wisconsin DNR. As a major outreach/education partner, the
Wisconsin DNR provides information to watershed groups and the general public for
capacity building and outreach. In the meantime, it also collects user evaluation and
feedback information that tool developers can use to improve their products. Of course,
EPA and the two universities also have outreach/education capabilities and they too can
provide information to watershed groups and the general public, even though these are
not explicitly described in the diagram. Watershed groups may also provide important
local information to state and federal government agencies and the universities. The
following example helps to explain the relationships among partners in this V0. The
partnership was recently funded by EPA to enable watershed quality information to ﬂow
between state and federal government agencies, based on web services technology. In this
grant, the Wisconsin DNR is the main funding receiver from EPA. The two universities
are subcontractors of the Wisconsin DNR to carry out major developmental work. The
end results, such as a water-quality information query and reporting tool and integrated

modeling functions, will be made available to watershed groups and the general public.

59

Strong alignment between actors is the key for success and sustainability of any
actor network. In the case of the Midwest Partnership for a Watershed Management
Spatial Decision Support System, strong ﬁnancial and technical alignments and sharable
credits among actors helped its steady growth over the past few years.

In terms of a general Watershed Management VO, sharable common goals and
credits are the most important factors that determine the sustainability of the V0. An
expandable computational infrastructure that can meet watershed management computing
needs is also critical. The relationship between VO members can be deﬁned by physical
network protocols and organizational agreements. Grid computing technology is an
excellent ﬁt to achieve this goal. Of course, steady, diverse funding sources also lead to
longevity of watershed management VO.

For a VO-based Watershed Management SDSS, the most important requirement
is the capability to support the partnership-driven watershed management process. It’s
obvious that considerable information and communication technological infrastructure is
critical for the Watershed Management VO, especially as more advanced communication
forms are enabled. Grid computing, as an emerging information technology, can provide

such infrastructure.

2.5.2. Technical Requirement Analysis

2.5.2.1. Grid Computing
The Grid Computing concept is intended to enable coordinated resource sharing
and problem solving in dynamic, multi-organizational virtual organizations. The basic

idea is to provide computation power to everyone who can access it, like the power grid

60

providing electricity. In this view of grid computing, computing becomes ubiquitous and
individual users (or client applications) gain access to computing resources (CPUs,
storage, data, applications and so on) as needed with little or no knowledge of where
those resources are located or what the underlying technologies, hardware, operating
system, and so on are (Foster & Kesselman, 2003). The key values of Grid computing are
in the underlying distributed computing infrastructure technologies that are evolving in
support of cross-organizational applications and resource sharing - in a word,
virtualization - including virtualization across technologies, platforms, and organizations
(Joseph et al., 2004; Jacob et al., 2005). Use of open standards is critical in this kind of
virtualization, because it enables interoperability among heterogeneous resources and
platforms. Grid Computing involves an evolving set of open standards for Web services
and interfaces that make services, or computing resources, available over the Internet
(Foster and Kishimoto et al., 2005; Baker et al., 2005).

The sharing in the computational grid refers to not only ﬁle exchanges, but also to
direct access to a spectrum of computers, software, data, and other resources, as is
required by a range of collaborative, problem-solving and resource-brokering strategies
emerging in industry, science, and engineering. This sharing is necessarily highly
controlled, with resource providers and consumers clearly and carefully deﬁning just
what is shared, who is allowed to share, and the conditions under which sharing occurs
(Foster & Kesselman, 2003; Stockinger, 2007).

Grid Computing enables organizations (real and virtual) to take advantage of
various computing resources in ways not previously possible. They can take advantage of

under-utilized resources to meet user requirements while minimizing additional costs

61

(Foster, 2003). The nature of a computing grid allows organizations to take advantage of
parallel processing, making many applications ﬁnancially feasible, as well as allowing
them to complete sooner (Foster & Kesselman, 2003).

Grid Computing makes more resources available to more people and
organizations while allowing those responsible for the IT infrastructure to enhance
resource balancing, reliability, and manageability (Jacob et al., 2005).

Since its emergence in 1997, Grid Computing technology has grown rapidly.
Scientiﬁc communities around the world have built numerous Grid systems for scientiﬁc
applications. Arar'rjo et al. (2005) built a virtual Hydro-Meteorological Scientiﬁc Network
based on Grid technology. Hluchy et al. (2006) created a ﬂood forecasting application
using a computational grid. Major commercial software vendors like IBM and Oracle
offer Grid solutions for various industries. Organizations such as The Global Grid Forum
have been formed to deﬁne speciﬁcations for Grid Computing. Grid technology is also
being converged with Service Oriented Architecture (Foster and Tuecke, 2005). Foster
(2005) also described the notion of service oriented science based on Grid technology.

To support VO-based watershed management SDSS, Grid technology has to
integrate with different components like GIS, Hydrologic Models, RDBMS and User

Interfaces.

2.5.2.2. Grid Computing & GIS
The notion of Grid Computing matches very well with the argument that GIS will
ﬁnally become part of the urban infrastructure and real-time delivery of data and GIS

functions through the network will be passed from professional use to lower level routine

62

use (Batty, 1999). How to utilize Grid Computing technologies for geospatial
applications is a new and challenging topic in Geographic Information Sciences. The
UCGIS has identiﬁed pervasive computing as a major research priority in the Geographic
Information Sciences (UCGIS 2002). Since Grid Computing deals with the integration
problem from both organizational and technical aspects, research on its utility for
geospatial applications is also beneﬁcial for two other research priorities in Geographic
Information Sciences identiﬁed by UCGIS: GI Partnering and Institutional GIS (UCGIS
2002). Shen et al. (2004) analyzed the weaknesses and problems of traditional GIS and
proposed methods to solve these problems with the technology provided by Grid
Computing and web services. Wang and Armstrong (2005) demonstrated a Grid-based,
problem-solving environment for computationally intensive geographic information
analysis based on geo-middleware. Wang and Zhu (2008) indicated that coupling Grid-
based cyberinfrastructure and GI S can facilitate computational thinking to analyze
massive quantities of spatiotemporal data rapidly and economically. As described in
Chapter One, researchers involved the GEON network also published a series of papers
on the Grid-based cyber infrastructure for the earth sciences. Data interoperability issues
within a Grid environment and integration problems with ESRI software components
such as ArcIMS and ArcWEB services have also been studied (Arrowsmith et al., 2004;
Memon et al., 2004). The GIS industry is well poised to utilize the potential of Grid
Computing because of its mature, web-service culture, interoperability data standards, a
process-oriented environment, high performance computing needs and the characteristics
of spatial data. But Grid Computing technology has not been widely adopted by

commercial GIS software vendors yet.

63

2.5.2.3. Grid Computing and Hydrologic Models

Hydrologic modeling is a powerful technique of hydrologic system investigation
for both the research hydrologists and the practicing water resources engineers involved
in the planning and development of an integrated approach for management of water
resources (Gregersen et al., 2008). The availability of remotely sensed and other
geospatial data provide very useful input data required for physically based hydrological
models (Bhaskar et al., 1992; Browne, 1995; Fortin et al., 2001). The use of remote
sensing and GIS facilitates the analysis of large scale, complex and spatially distributed
hydrological processes (Maidment, 1996; Sui and Maggio, 1999).

The complex, distributed watershed models often require intensive computational
resources, particularly when the resolution of the underlying data keeps increasing. This
is where Grid technology can provide viable solutions. The high performance computing
capability of Grid can help speed up the modeling process. But the parallelization of
hydrological models has to be done ﬁrst. Shi et al. (2002) experimented with the
parallelization of a soil erosion model in a Grid environment. Hluchy et al. (2006) used
Grid-based parallelization techniques to solve problems of precise ﬂood prediction and
potential damage assessment.

Another area of hydrological modeling that can beneﬁt from Grid Computing is
uncertainty analysis. For example, a Monte Carlo simulation for a hydrological model
can be run on hundreds of computers within a Grid and, hence, increase the speed of the
whole process. As an example, Lei et al. (2006) utilized Grid technology for reservoir

uncertainty analysis.

64

2.5.2.4. Grid Computing and RDBMS

RDBMS (relational database management system) is a type of database
management system (DBMS) that stores data in the form of related tables. An important
feature of relational systems is that a single database can be spread across several tables.
Relational databases are powerful because they require few assumptions about how data
is related or how it will be extracted from the database. As a result, the same database can
be viewed in many different ways (Allen and Creary, 2003). Today, almost all full-scale
database systems are relational.

The performance and scalability requirements of relational databases have always
been critical for end users. Grid technology can improve both the performance and
scalability of a database (Segal, 2003). In fact, the commercial relational database
providers like Oracle and IBM have realized the beneﬁts of Grid Computing and have
integrated Grid technology into their products. Sterck et al. (2007) integrated database

and Grid technology for a bioinformatics application.

2.5.2.5. Grid Computing and User Interfaces

The user interface is the aggregate of means by which users interact with a
particular machine, device, computer program or other complex system. The user
interface provides means of input, allowing the users to manipulate a system, and output,
allowing the system to produce the effects of the users’ manipulation (Galitz, 2007).
Many technological innovations rely upon User Interface Design to elevate their
technical complexity to a usable product. Grid technology is no exception. The idea

behind the GRID IT environment is to make the complicated inner workings of the

65

system transparent to the end users. The user interface is obviously the key here. Wood et

al. (2004) have presented a Graphical User Interface for Grid Services.

2.5.2.6. Other Technical Requirements

Besides the V0 support requirement, other user requirements were collected

based on reviewing the literature, the watershed management decision-making process

and discussions with watershed groups and government officials. Based on these inputs,

the following requirements were identiﬁed.

The system must support multiple organizations and public participation in a
watershed management V0.

The social and technical nature of watershed management demands
technology that can support communication and collaboration among multiple
organizations. Public participation implies public communication. That means
that government agencies, at a minimum, have to inform the public at large
about issues concerning water and watershed management. Every higher level
of participation, such as consultation, etc. implies ever higher demands on the
quantity and quality of the information to be given. It implies also that this
communication becomes more bi-directional and interactive. Nevertheless, it
is all but evident to decide in which way and to what extent the public has to
be informed (Webler and Tuler, 2001). A VO-based watershed management
SDSS has to support all these communication and collaboration needs. A
Grid-based IT inﬁastructure can provide such support.

The system must provide interactive spatial data display services.

66

Users of the system require extensive interaction with spatial data through an
appropriate user interface. The required operations should include zoorrring in
and out, panning, and toggling spatial layers on and off (Densham, 1991).
This provides users with basic data-viewing functionality found in typical GIS
applications.

The system must provide interactive non-spatial data display services.

Users of the system require extensive interaction with non-spatial data through
a proper user interface. The required operations should include table data and
text displays (Djokic, D. 1996).

The system must provide distributed data access and geoprocessing
capabilities.

Users of the system require access to distributed data and geoprocessing
methods made available by different organizations as web services and the
capability to display data and use methods locally (Dangennond 2007). This is
very important in an increasingly serviced-oriented IT environment.

The system must provide access to simulation models.

Users of the system require access to different simulation models and the
capability to run them for scenario analysis. These models could be
watershed-based hydrological models such as SWAT, AGNPS, etc. or
economic analysis models. These models should be integrated with the system
in a proper way so users can use them through a uniform interface (Sui and

Maggio, 1999).

67

The system must provide alternative scenario generation and analysis
ﬁmctions.

Users of the system require the capability to generate alternative scenarios and
analyze the pros and cons of each output (Choi et al., 2005).

The system must be extensible, so new data and models can be easily added
incrementally.

The system should be extensible and scalable. As new data and models
become available, it should be possible to integrate them into the system
without causing major changes to the existing architecture (Walsh, 1993).

The heterogeneous components within the system must be interoperable with
each other and with external systems.

The components within the system, such as GIS, models and databases,
should be interoperable. The system should also be interoperable with external
systems based on existing standards (Densham, 1991).

The system must provide a user-ﬁ'iendly interface.

Users of the system require an intuitive interface that can help them learn to
use the system easily and heuristically (Densham, 1991; Walsh, 1993).

The system must be web-enabled and easily accessible.

The system must be web accessible to users. The Internet provides an

excellent platform for technology transfer (Lane et al., 1999).

68

2.6. Conclusions

As illustrated in this chapter, the social/technical nature of watershed management
demands a multi-organizational and multi-disciplinary VO-based Watershed
Management SDSS. Such a system should support the management of large volumes of
multi-sectoral and multidisciplinary information in a distributed environment, and the
communication between the scientiﬁc and policy sectors and between decision-makers
and involved stakeholders. The system should translate often complicated modeling
results into an easy-to-understand format for environmental decision makers.

This chapter suggests that Grid Computing, combined with advanced geospatial
technology, can provide an effective organizational and computational architecture for
such a VO-based Watershed Management SDSS. After a conceptual watershed
management VO was described, an existing watershed management V0 was used to
facilitate the requirements analysis for a medium-sized, VO-based Watershed
Management SDSS incorporating actor from federal and state government, academia and
environmental groups. Concepts of Actor Network Theory were applied to the
requirements analysis process and they proved to be very useful for such an endeavor.
Detailed requirements were speciﬁed in the previous section. The integration of decision
making and communication into a web-based environmental decision support system
within a watershed management V0 presents the best opportunity for effectively
integrating sciences into real-world watershed management decision-making processes
because of the social/technical nature of watershed management. In the next chapter, the

requirements deﬁned in this chapter will be used to guide the design and partial

69

implementation of a Grid-based integration framework for the VO-based Watershed

Management SDSS.

70

Chapter 3. Grid Computing and VO-based Watershed Management SDSS

3.1. Introduction

As indicated in Chapter Two, the multi-organizational and multi-disciplinary
nature of watershed management demands an enabling decision support framework that
can assist such cross-organizational and cross disciplinary integration and collaboration.
The Grid Computing technology based on the Virtual Organization model clearly meets
this requirement. In fact, Grid technology and the V0 concept have been used in a variety
of environmental applications. For example, Araujo et al. (2005) described a successful
Grid Empowered Hydro-Meteorological Scientiﬁc Network to support a cascade of
models that simulates the behavior of the atmosphere, hydrographic basins, aquifers and
reservoirs. Hluchy et al. (2006) presented 3 Grid based ﬂood prediction V0. The Earth
System Grid (ESG) project targeted their efforts toward the problems faced by the
climate modeling community (Bemholdt et al, 2005; Williams et al., 2007). Kendall et al.
(2008) built a web-enabled collaborative climate visualization application within the ESG.
But one important limitation of all these research efforts is that they have largely been
conducted in the academia with very few linkages to environmental agencies and user
communities and the applications built within these endeavors are mostly used by experts
and researchers only.

In this chapter, the author addresses this limitation by conceptualizing a grid
based integration framework with connections to government agencies and broad user
communities for the VO-based Watershed Management SDSS. The implementation of
selected components for this framework is also presented in a project context for the

Army Corp. of Engineers. The main goal of this project is to create a sediment runoff

71

analysis SDSS for two local watersheds in northern Indiana by combining and
customizing existing online tools from Michigan State University and Purdue University.
The author is responsible for overall system architecture design and development of the
modeling system on MSU’s end.

Section 3.2 describes the details of the integration framework. The design and
implementation of an example VO-based watershed management SDSS within the
framework is reported in section 3.3. Section 3.4 explains the parallelization of a water

quality model in a grid environment. Section 3.5 concludes the chapter.

3.2. A Grid based Integration Framework for the VO-based Watershed
Management SDSS

The social and technical characteristics of watershed management demand a
SDSS that can support Virtual Organization environment and provide high performance
computing capabilities. Based on the requirement analysis for VO-based watershed
management SDSS done in Chapter Two, the author designs the following conceptual
Grid based integration framework for the VO-based Watershed Management SDSS

(Figure 3.1) as a general architecture for such a system.

72

 

 

 

 

Data/GlS/Model Server

Watershed Groups Government

 

 

Data/GIS/Model Server

Agencies/Offices

 

 

 

 

Visualization Server
for Virtual Reality
Teleirnmersion

 

 

 
     

VO-based Watershed
Management SDSS

 

  

 

 
 

\ /

        
 

 

Wireless Digital
‘ Sensors and Devices for
Ground Truth Data
Collection

 

 

 

 

 

General
Public

 

 

 

University Researchers
Data/GIS/Model Server

 

 

 

 

Other Stakeholders
Data/GIS/Model Server

 

 

Knowledge Base in
Virtual University

 

 

 

 

Figure 3.1 Grid based Integration Framework for Watershed Management VO (Derived

In this conceptual framework, all components except for the box labeled ‘General
Public’ are connected as a loosely coupled extensible, interoperable and scalable
hierarchical VO-based spatial decision support system for watershed management. Every
square box except for the General Public one can be considered as a V0 member with its
own spatial decision support system. The General Public is also a V0 member but
without its own spatial decision support system. So this framework is actually a system of
systems in a distributed network environment. Technically all these systems are made
interoperable with one another through web services or Grid protocols. The speciﬁc
interoperability capabilities are developed incrementally for different user needs and
requirements within the V0. This framework connects different watershed management

related organizations in a watershed management V0. These organizations participate in

from Figure 2.1)

the V0 by different means.

73

 

University researchers usually have strong technical capabilities so they may
contribute both database and modeling services. They may also provide training services
because of the educational nature of the university. The university researchers are often
domain experts too so they can also provide applied consulting services. The government
agencies are typically responsible for public environmental information dissemination so
they often provide database services. Some government agencies such as USGS or EPA
may also make various environmental models available to the public, but in general these
models are not provided as online web services for various reasons (EPA, 2008; USGS,
2008). Of course the government agencies are also charged with other non-technical
responsibilities such as policy making and enforcement etc. Watershed groups are grass-
roots organizations founded and maintained by citizens or nonproﬁt groups for the
purpose of protecting particular watersheds (Davenport, 2002). Some watershed groups
may have their own spatial decision support system but most of them don’t have the
technical resources to maintain an online system that can provide services to other
entities in the watershed management V0 (Davenport, 2002). Therefore most watershed
groups are pure end users of the system within the V0. Even though watershed groups
may not contribute technical resources, they often play a very important role in building
essential organizational connections among actors within the V0. Other stakeholders
with their own interests in watershed management may also have their own database or
modeling services for others to use. And at the meantime, they are end users of the
system as well. The wireless and digital sensors within the system are used for ﬁeld data
collection purposes. These data may be used to calibrate model or as model input. The

visualization server is used for data and model result visualization to help people

74

understand issues at hand better. The General Public within the V0 is often viewed as
people who just want to access public information. Anyone can connect to the system
through a user-friendly interface that provides a single entry point to the whole system.
And the system architecture is transparent to the end users so they may not realize that
they are utilizing different computing resources within the system from different

organizations in the V0.

3.3. An Example VO-based Watershed Management SDSS

To further illustrate this conceptual ﬁamework, let’s take a look at an example
watershed management SDSS designed and developed by the author at Michigan State
University’s Institute of Water Research. This system has the potential to be Grid-
enabled. It’s also part of the Midwest Spatial Decision Support System described in
Chapter Two. The system is called Digital Watershed (DW). The DW system connects
federal agencies; universities, state government and other organizations to form a very
effective watershed management V0. The goal of this V0 is to develop, promote, and
disseminate web-based spatial decision support systems to help manage watersheds in the
Midwest region. In particular, the main objective is to make these systems freely
available via the Internet to local officials, natural resource managers, and the general
public.

Digital Watershed provides a user-friendly spatial decision support system that
utilizes the national 8-digit watershed database within the EPA BASINS program.
Watersheds are delineated by USGS using a Hydrologic Unit Code (HUC) system based

on surface hydrologic features. This system divides the country into 21 regions (2-digit),

75

222 subregions (4-digit), 352 accounting units (6-digit), and 2,262 cataloging units (8-
digit) (Seaber et al., 1987; USGS, 2007). An 8-digit watershed is one of these 2,262
cataloging units. This comprehensive 8-digit watershed database contains all regulated
facilities, river networks, DEM, state soils, and other relevant data layers. The overall

system architecture of DW is illustrated in Figure 3.2.

 

 

 

 

 

 

 

 

 

 

 

 

 

Server i
r
I
I
~ I
Web GIS '
3 Server Spatial :
E A Database .
Web Client «0 , '
.9 (Soil, '
cur g Modeling Landuse, :
Server DEM .
E.
fé’fée) etc.) :
r
l
I

 

 

 

Figure 3.2 Digital Watershed System Architecture

The Digital Watershed system provides end users with three forms of access to an
interactive GIS for speciﬁc 8-digit watersheds. The address-based entry allows users to
locate their watersheds with a street address or by clicking on a location on a regional
map. The search entry allows users to search for their watershed by name or Hydrologic
Unit Code (HUC) code. The interactive GIS provides basic spatial functions such as

zoom and pan. A visualization function allows end users to generate a 3-D view of any 8-

76

digit watershed. Using the USPED modeling function, one can run an erosion and
deposition model to calculate erosion and deposition pattern in a watershed. USPED
stands for Unit Stream Power - based Erosion Deposition, a simple model predicting the
spatial distribution of erosion and deposition rates for a steady state overland ﬂow under
uniform rainfall excess conditions (Mitasova et al. 1996, Mitas and Mitasova 1998). The
system also includes a precalculated 10-meter resolution sediment data layer based on the
SEDMOD model for the selected watersheds in the Midwest region to help end users
identify areas at high risk of erosion. SEDMOD, an acronym for Spatially Explicit
Delivery MODel, calculates a Sediment Delivery Ratio (SDR) which can be useful to
calculate the amount of eroded material that would be available for transport and that is
deposited along hillslopes and streams (Fraser 1999). Total amount deposited along
hillslopes and streams is equal to the amount transported. The interoperability features of
Digital Watershed include seamless links (a single click) to the TerraServer (Barclay et
al., 2000) for aerial imagery and Google Earth and Maps (Google, 2008). The
TerraServer Web site developed by Microsoft and USGS is one of the world's largest
online databases, providing free public access to a vast data store of maps and aerial
photographs of the United States (Barclay et al., 2000). The link to TerraServer is done
via OGC WMS (Open Geospatial Consortium Web Map Service) protocol. The link to
Google Maps and Google Earth is implemented by passing watershed boundary data to
the Google Mapping interface and generating a KML ﬁle from the same watershed
boundary data. KML is the data format Google Earth uses for spatial data. The system is
also connected with Purdue University’s online L-THIA modeling system through web

services. L-THIA, Long-Term Hydrologic Impact Assessment, is designed to help these

77

people to quantify the impact of land use change on the quantity and quality of their
water. This tool employs land use and soil characteristics from the user along with thirty
years of precipitation data to determine the average impact that a particular land use
change or set of changes will have on both the annual runoff and the average amount of
several non-point source pollutants (Engel et al., 2003).

The DW system is extensible because new data layers and new models can be
added over time. It’s also designed to be interoperable with other online SDSS systems
within a V0 as mentioned above. The system provides scalability in that it contains a
hierarchical database for watersheds at different spatial scales. A recently completed
project by Michigan State University and Purdue University involving the author for the
US Army Corps of Engineers can help illustrate more details about how a loosely
coupled distributed Spatial Decision Support System of multiple online systems might
actually work.

The major goal of the project is to develop a fully Web-based watershed
management SDSS for two critical watersheds in Indiana, the Burns Ditch and Trail
Creek watersheds, on the southern end of Lake Michigan for erosion and sediment
control purposes. Erosion is a major environmental concern in the areas surrounding
Great Lakes waters in the United States (and Canada). Great Lakes Tributary Modeling
Program was established by the US. Army Corps of Engineers (USACE) to develop
computer modeling tools for Great Lakes watersheds to facilitate management planning
by various stakeholders to reduce erosion and associated water quality problems. This
web-based SDSS was developed as part of the program. The management system is ﬁrlly

Web-based and was built by linking two existing Web-GIS applications hosted separately

78

at Purdue University (the Long-Tenn Hydrologic Impact Assessment, L-THIA, Web
modeling tool) and Michigan State University (Digital Watershed Web mapping tool).
Interoperability between the two systems was implemented by extending both systems
and passing dynamically re-projected vector GIS data and modeling results between the
two sites. The integrated system takes advantage of complementary data and modeling
capabilities from both applications to construct a complete SDSS to facilitate
management decision making for erosion and nonpoint source pollution reduction. It
allows end users to browse GIS data, dynamically delineate watershed boundaries, make
changes in land use and/or apply Best Management Practices (BMPs) within a watershed,
and run hydrologic and erosion models to assess different management scenario impacts
on hydrology, water quality, and sediment yield. It also allows preliminary sizing and
cost estimation for building a number of erosion and sediment control structures. The
Web-GIS was advocated as a spatial decision support system (SDSS) to support state and
local measures that are designed to reduce tributary loadings of sediments and pollutants.
The purpose of this work is to help USACE reduce the need for, and costs of, navigation
dredging, while promoting actions to delist Great Lakes’ area of concerns (AOCs).
According to the user survey results (see Table 3.1 in section 3.3.5), the system was well
received during a stakeholder workshop for its user-ﬁiendly graphical user interface
(GUI) design and usefulness in potential local level management efforts to combat
erosion and nonpoint source (N PS) pollution in Great Lakes region.

Speciﬁc objectives of the project include: 1) extend two existing systems to allow
BMP application and erosion modeling; 2) establish interoperability between MSU’s

Digital Watershed system and Purdue’s watershed modeling system to allow seamless

79

integration of the various mapping and modeling components for the resulting watershed
management system; and 3) disseminate the integrated modeling system through
workshops and evaluate the system based on user feedback.

Since MSU’s DW system was already described above, here only a general
description of Purdue’s SDSS is given. The overall system architecture of Purdue’s SDSS

is illustrated in Figure 3.3.

 

 

 

 
   
 

 

 

 

 

 

 

 

 

...., V

I

I

I

I

I

I

Web GIS I

._ Spatial .

g Server Database I

3 ll, I

Web Client (3 it La(n:use, I
a - ~ DEM I

GUI g Modeling etc-l :
3 ism? Weather '

SEDSPEC Database :

I

I

I

 

 

 

Figure 3.3 Purdue SDSS System Architecture

Purdue University’s Web-GIS based spatial decision support system (SDSS)
encompass the following modeling tools: 1) Long-Term Hydrologic Impact Assessment
(L-THIA) model for small watersheds (Engel et al., 2003); 2) Sediment and Erosion
Control Planning, Design and SPECiﬁcation Information and Guidance Tool (SEDSPEC,

Tang et al., 2004); and 3) the real-time watershed delineation and land use change impact

80

assessment system (Engel et. al., 2003, Choi et al., 2005) that includes two additional
tools for estimating imperviousness and peak runoff. The Purdue SDSS’s gee-database
covers the ﬁve states in US EPA region 5 (Indiana, Illinois, Michigan, Ohio, and
Wisconsin). Data include aerial photos, digital elevation model (DEM), land use,
topographic maps, soils, streams, and other civil features. These modeling tools are
introduced below.

The L-THIA model uses long term (30+ years) daily weather, soil type, and land
use data to simulate runoff volumes and loadings in the runoff for 13 NPS pollutants
including nitrates, phosphorus, heavy metals, and fecal coliform. Long-term daily outputs
are presented in terms of loadings for each land use type and the probability of
exceedance curves. Details of the model were described previously (Engel et al., 2003).

The SEDSPEC model estimates small watershed peak runoff that can be used for
the preliminary design of hydraulic and erosion control structures. For peak runoff
estimation, two standard hydrologic models (the Rational Method and TR-55) were
implemented to simulate short-term peak runoff based on site-speciﬁc hydrologic soil
groups and land uses. The hydrologic models estimate peak runoff using storm data
stored in a database. SEDSPEC can be used to design or provide recommendations for
the structural dimensions of channels, culverts, grass-lined channels, level terraces, low
water crossings, runoff diversions, sediment basins, and storm water detention basins.
Estimates of construction costs for these structures are also provided. Details of the
model were described previously (Tang et al., 2004).

The above-mentioned modeling tools and a tool that estimates imperviousness

using the mean imperviousness of each land use type in percentages and the overall

81

imperviousness in a watershed or area of concern were integrated into a Web-GIS based
SDSS (Choi et al., 2005). The online SDSS allows users to delineate a watershed, extract
land use, soil, and weather data from backend databases to form model inputs, and run
hydrologic models and analyses results based on land use changes. Delineation of
watershed boundaries can be done with a user-speciﬁed drainage point on a stream
segment, ﬂow accumulation, and ﬂow path derived from a digital elevation model
(DEM).

It is noteworthy that both Purdue’s and MSU’s Web-GIS systems have been
widely used by local, state government agencies, as well as environmental consulting
professionals (Watermolen, 2008).

The following three sections provide detail on particular system components: how
they functioned to model important hydrologic elements, and how these components

were integrated for the project.

3.3.1. Erosion modeling

To estimate soil erosion, sediment yield, and the impact of implementing BMPs,
the Revised Universal Soil Loss Equation model (RUSLE) was chosen. RUSLE is an
erosion prediction model that estimates long-term average annual soil loss resulting from
the detachment of soil due to raindrop splash and overland runoff from ﬁeld slopes in
speciﬁc cropping and management systems and from rangeland (Renard et al., 1997;
Renard and Ferreira, 1993). RUSLE is a replacement for the Universal Soil Loss
Equation (U SLE) and retains its six factors in that equation, as shown below.

A=R°K'LS°C°P

82

where A is the long-term average annual soil loss (ton acre'l yr '1), R is rainfall erosivity
in [(hundreds of ft-ton) inch acre'lhr "'yr ’1], K is the soil credibility in [ton acre'l
(hundreds of ft—ton)"inch'l acre hr], LS is the dimensionless slope length and steepness
factor, and C and P represent the dimensionless impacts of cropping and management
systems and of erosion control practices, respectively (Renard et al., 1997). The RUSLE
model was ﬁrst developed by the USDA-Agricultural Research Service and was ﬁrst
released in 1993 (Renard et al., 1997). It has been widely used by USDA-Natural
Resources Conservation Service (NRCS) nationally, and it has been adopted
internationally as well. There is a wealth of information and data available for its
application for many locations (see e.g., Ouyang and Bartholic, 1997, 2001; Nyakatawa
et al., 2001; Angirna et al., 2003; Fu et al., 2006; Nyakatawa et al., 2007; Schiettecatte et
al., 2008).

The RUSLE model predicts long-term average annual erosion (Renard et al.,
1997). In this project, however, the desired erosion related estimate is the sediment yield.
Soil erosion refers to the soil dislodged from its original location due to rainfall and/or
overland runoff. Not all of the dislodged soil, however, is transported in runoff water to a
nearby stream or lake. A portion of the eroded soil is deposited at lower points in the
watershed whenever runoff slows down. The amount of eroded soil that actually reaches
a stream or other water body is called sediment (Fraser, 1999). Hence, for a given
watershed, the long-term average annual sediment yield can be estimated by multiplying
the long-term average annual soil erosion potential by a sediment delivery ratio. The

sediment delivery ratio is the ratio between the actual lost sediment to the total erosion

(detached soil) potential from a watershed (Novoty, 1980). The sediment delivery ratio

83

varies between 0 and 1. There are different ways to determine sediment delivery ratio for
a watershed. In this project, a relationship between watershed size and sediment delivery
ratio is used. In other words, the sediment delivery ratio for a watershed is determined as
a function of the watershed size (Roehl, 1962; Renfro, 1975).

In this project, the RUSLE equation is applied to a watershed by way of
multiplying the raster (or grid) data layers (10 meter resolution) for the factors in the
RUSLE equation in a watershed. Then, total watershed soil loss is calculated by summing
up soil loss from all cells in the watershed. Finally, the sum is multiplied by the sediment
delivery ratio for the watershed to arrive at the sediment yield value in tons yr '1.

The erosion BMPs considered for this project include both structural and non-
structural BMPs. Non-structural BMPs include no till, reduced till, and conservation
tillage on agricultural ﬁeld and riparian buffer strip. No till refers to the total cover (100
percent) of soil surface with crop residue. Conservation tillage leaves at least 30 percent
of the soil covered by crop residues. Reduced tillage is an in-between tillage type.
Structural BMPs include sediment basins and grassed waterways. To represent the
different types of BMPs in the RUSLE equation, the C and P factors are adjusted for

each of the BMPs accordingly.

3.3.2. Web-GIS based SDSS for erosion and water quality management

The overall architecture of the Web-GIS SDSS is shown in Figure 3.4. There are
three common components in any Web-based modeling system: the user interface,

backend server databases and modeling programs, and the Web server situated in

84

between handling Hypertext Transfer Protocol (HTTP) connection and Web Service

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

calls.
________________________ .1
Purdue SDSS :
Web-GIS '
User Web 5 DB I
Interface Server M"”9"“9 I
to Server I
8 I
"E A I
Web Portal 3 Web Services :
'8 Y I
3 MSU DW :
Web-GIS '
User Web 3' DB I
Interface Server "M91109 .
Server '
I

 

 

 

 

 

 

Figure 3.4 Burns Ditch Trail Creek Web-based SDSS System Architecture

The development team includes the author and Tong Zhai of Purdue University.
The author is the overall designer of this architecture and developer of the modeling
system on MSU’s end. Tong Zhai is responsible for interface and BMP parameterization
implementation on Purdue’s end.

The Purdue Web-GIS interface is built using the open source MapServer
(http://mapserver.gis.umn.edu/) software with a Java applet front end. It handles
watershed delineation based on a user-speciﬁed outlet point and user digitization of areas
within a delineated watershed for land use change or erosion BMP implementation. Its
hydrologic models, introduced earlier, can provide before and after land use change

hydrologic impact assessment for the delineated watershed.

85

The MSU Digital Watershed Web-GIS system is built using Internet Mapping
software from ESRI. It stores the raster data layers for the K, LS, and C factors for
RUSLE simulation for the project area. The default P factor is assumed to be 1.

Web portal page provides all different entry points to the system. Through the
interoperable approach, described in the next section, watershed and BMP area
boundaries are ﬁrst delineated by the Purdue Web-GIS system and sent to the MSU
Digital Watershed system, which are used to clip raster layers of the erosion factors.
BMP type speciﬁc C or P factors are then incorporated into the corresponding raster data
layers for the user-deﬁned areas. Then, the RUSLE model is run for the watershed to
calculate total erosion, which is then modiﬁed by a sediment delivery ratio to arrive at
long-term average annual sediment yield for the watershed. The results are then displayed

back in the user’s Web browser.

3.3.3. Interoperability

The interoperability operations of data passing and other related operations
between the Purdue Web-GIS and MSU’s Digital Watershed are carried out behind the
scenes without the need of explicit intervention by the user. This ensures seamless
integration of the two Web-GIS systems. The integrated SDSS links the two physically
separate Web-GIS systems by passing dynamically re-projected vector GIS data and

modeling results between them, as shown in Figure 3.5.

86

 

 

_ -_P.Urd.UeWeb-GIS _ MSU Digltelwatershed

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

Watershed 0 Identify watershed outlet
delineation rte-projection [Li i} (Lat- Long point)
/ \
Boundaries t' Web . .
v

re-projection

 

 

 

 

 

 

 

Figure 3.5 Interoperability

As shown in Figure 3.5, users ﬁrst identify a drainage outlet point on a stream line
within the MSU Digital Watershed Web-GIS environment. The outlet point’s latitude and
longitude coordinates are sent to the Purdue Web-GIS system, where they are re-
projected to the Universal Transverse Mercator (UTM) Coordinate System, Zone 16
coordinates for the project area. Then, the Purdue Web-GIS uses the point to delineate a
watershed based on DEM derived ﬂow accumulation and ﬂow path data. Users can also
initiate watershed delineation starting from the Purdue Web-GIS system. Either way, a
watershed will be delineated, upon which a summary of land use and soil group of the
delineated watershed is displayed within the Purdue Web-GIS system. After reviewing
the watershed land use and soil summary, users can proceed to digitize areas within the
watershed to assign land use changes and/or apply erosion BMPs. The boundaries of the

watershed and the digitized areas are saved on the Purdue Web-GIS as ESRI shapeﬁles

87

ﬁrst. Then, they are re—projected from UTM zone 16 coordinates to Latitude-Longitude
coordinates. The locations of the shapeﬁles in the Purdue Web-GIS system’s ﬁle
structure are then sent to the MSU Digital Watershed, which in turn retrieves the
shapeﬁles and uses them as masks for clipping grid data layers for erosion calculations
based on GIS-based RUSLE model. A sediment delivery ratio (SDR) value generated by
Purdue Web-GIS system based on the area of the delineated watershed also gets passed
on to MSU’s Digital Watershed system. Map algebra function is used to multiply
multiple RUSLE factor grid layers together to get the ﬁnal erosion amount. This erosion
value is then multiplied by SDR value to get the actual sediment number. The
information (latitude-longitude coordinates or Web address of the boundary shapeﬁles) is
passed through web service calls on programs that reside on destination Web Servers.

This description of interoperability also illustrates the distributed modeling process.

3.3.4. Burns Ditch Trail Creek Watershed Management V0

To further explain this example VO-based Watershed Management SDSS, all

stakeholders involved are put into different boxes in Figure 3.6 for illustration purposes.

88

 

 

 

Local Planning and US Army Corp. of Engineers
Watershed Groups Chicago District

 

 

 

 

 

 

  

 

 

 
 
 

  

 

 

 

 

 

  

 

 

 

 

General Burns Ditch Trail
Publ' Creek Watershed Wisconsin DNR
1c Management SDSS
Michigan State
University Purdue University

 

 

 

 

 

 

 

 

 

Figure 3.6 Stakeholders Involved in Burns Ditch and Trail Creek Watershed Management
SDSS

As shown in Figure 3.6, multiple institutions are participating in this Virtual
Organization for the common objective of reducing sediment runoff in the two
watersheds. US Army Corp. of Engineers (U SACE) provides funding sources and control
for the project. The two universities are responsible for the development of the watershed
management SDSS. Speciﬁcally, Michigan State University brings resource development
expertise and perspective while Purdue University introduces engineering expertise and
perspective into the system. Previous user evaluation and feedback information on Digital
Watershed and L-THIA collected by Wisconsin DNR provides valuable input and
validation for system development. Local planning and watershed groups are primary

users of the system and they can also help with outreach and education.

89

3.3.5. Use Case

The implemented watershed management system for the Burns Ditch and Trail
Creek watersheds is available online at http://danpatch.ecn.purdue.edu/~ecLip/erosion/.
This system fulﬁlls most of the functional requirements speciﬁed in Chapter Two. Users
can access the SDSS from either the Purdue Web-GIS interface or MSU Digital
Watershed interface based on their preferences. Both ways will lead to the same system
capability.

The general procedure for watershed management using the SDSS is shown in
Figure 3.7 (a-d). This ﬁgure contains screen shots from the actual system. From the initial
entry page, either from Purdue Web-GIS or MSU Digital Watershed, the user would
zoom in to identify the area of interest and nearby stream, and then initiate watershed
delineation by a single click on the stream (Figure 3.7a). A watershed is delineated in
approximately ten seconds based on the user speciﬁed outlet point and underlying DEM
data. The user can then activate the online digitizing interface to either manually digitize
areas for BMPs or allow the system to determine the contributing areas in the case of
grassed waterways or sediment basin structural BMPs. Then, the user speciﬁes the type
of BMP for the digitized area using the land use/BMP dialog box (Figure 3.7b). For
grassed waterways, the user also needs to digitize a line inside the contributing area to
deﬁne the location of the waterway. Tillage BMPs can only be applied to agricultural
land uses. Once the changes are made and saved by the online digitizing tool, a before
and after land use and BMP summary is given, along with a modeling toolbox for
hydrologic and erosion modeling (Figure 3.7c). The available models can then be used to

obtain a quantitative estimate of the impact from the land use changes made or BMPs

90

applied (Figure 3.7d). The whole process can be repeated for the same delineated

watershed as many times as the user would like. This allows multiple management

scenarios to be evaluated and compared.

 

 

   

: ammonium .

 

 

 

 

11' ..—
’43.
l

3

 

 

 

 

 

 

 

 

 

 

 

Figure 3.7a Use Case Part A

91

Density Residential
Density Residential

Wndow

 

Figure 3.7b Use Case Part B

92

 

 

 

Modeling Toolbox

 

 

 

Land use summary table View watershed / Apply BMPS

 

 

Contains land use information . .
Estrrnate Sediment w / BMP

 

 

before and after modiﬁcation
Estimate Imperviousness

 

 

Estimate Peak Runoff

 

 

Run L-THIA Model

 

 

 

 

 

Run SEDSPEC Model

 

BMP summary table

 

 

 

 

 

Figure 3.70 Use Case Part C

93

 

 

 

 

 

Figure 3.7d Use Case Part D

3.3.6. 'Ihrining Workshop

The Web-GIS based watershed hydrologic and sediment modeling SDSS was
delivered at a stakeholder workshop on December 19, 2006 at the Northwestern Indiana
Regional Planning Commission (NIRPC) building located in Portage, Indiana. The
workshop was organized and conducted by Army Corp. of Engineers and Great Lakes
Commission (Bucaro, 2006). The Great Lakes Commission was directly contracted by
Army Corp. of Engineers for workshop arrangement and coordination (Bucaro, 2006).
Participants of the workshop represented a diverse set of stakeholder groups, including
local water resource managers (Save the Dunes Council, Great Lakes Commission, Little

Calumet River Basin Comnrission, Michigan City Port Authority), staff from several

94

state agencies (Indiana Department of Environmental Management, Indiana Department
of Natural Resources), county environmental planners, consultants, and US EPA Region
5 personnel.

The workshop started with an overview of the Web-GIS SDSS, including its work
ﬂow, modeling capability, and underlying theories for quantifying impacts of land
management decisions and BMPS on sedimentation within the study area of the Burns
Ditch and Trail Creek watersheds. Then, a hands-on training session was conducted for
three hours. Case exercises were provided to the workshop participants who used their
own laptop computers to access the Web-GIS SDSS via a wireless Internet connection.
At the conclusion of the workshop, various aspects of the system were discussed and
feedback forms were distributed among the participants.

User feedback is summarized in Table 3.1. Overall, feedback from participants
was very positive. The workshop was rated excellent or good for its content and
presentation by six out of seven and ﬁve out of seven completed user surveys,

respectively.

95

 

 

Evaluation of the ACOE Sediment Runoff Predictive Tool and Training Session

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TrainlILg Material Total number of response received: 7
Excellent Good Fair
- content 4 out of 7 2 out of 7 1 out of 7
- presentation 4 out of 7 1 out of 7 2 out of 7
Web Site Deslgp and Content Scale: 1: poor, 5= exceptional
Navigability is good. Links are clearly labeled. Can move from page to
4.0 page easily.
4.0 This site offers interactivity. The visitor is engaged usinLthe site.
4.0 This site uses appropriate page format. Pages are not inordinately logg._
4.0 Can easily ﬁnd information
4.1 This site is aesthetically appealing. Good use of graphics and color.
4.0 Additional resource links are included.
4.0 lnforrnation is useful
4.2 Rich content and will likely be revisited.
4.1 How this website compares in content to similar websites
4.1 Please indicate the usefulness of the Watershed Tools and Summary?
Modeling tools in the SDSS Scale: 1: poor, 5: exceptional
Score User-envisioned usage
View watershed/Apply 1 Target speciﬁc problems
BMPs 4.3 1 This can provide a “ﬁrst peek” into land use changes
1 Look at and compare drainage calculations for new projects
Estimate sediment 4.0 1 to protect coldwater ﬁshery
Estimate 1 Look at and compare drainage calculations for new projects
lmgrviousness 4.2 1 to protect coldwater ﬁshery
1 Look at and compare drainage calculations for new projects
1 calculate rough number for peak discharge
Estimate Peak Runoff 4.0 1 Impacts on ﬂooding and water quality impacts
Run L-THIA Model 4.0 1 Changi_ng land use—for watershed planning
Run SEDSPEC Model 3.8 1 To identify areas of concern

 

 

 

Table 3.1 User Feedback Summary

Participants also viewed design of the website and content displayed throughout

the site favorably. Design elements such as layout, navigability, interactivity, formatting,

aesthetics, and similarity to other websites covering similar topics received consistent

scores between 4.0 and 4.1 on a scale of 1 to 5 (1 being poor and 5 being excellent).

Combined with participants’ responses to usefulness of the information (4.0) and

likelihood of revisiting the site (4.2), this part of the survey results suggest that the site’s

96

design and content is user-friendly, meaningful and allows users to focus on making
decisions, which is the purpose of an SDSS.

The last section of the survey queried participants about speciﬁc models and tools
that constitute the SDSS. The ﬁrst three tools — view watershed/apply BMPS, estimate
sediment, and estimate imperviousness — received very favorable scores. Participants
rated the “view watershed/apply BMPS” tool the highest (4.3). The importance of this
score cannot be understated since this speciﬁc tool assumes users understand their water
quality problem, use the GIS map service to learn about existing land cover and other
resource conditions, know about or can quickly learn about agricultural BMPS from the
website, construct scenarios for improving water quality and input required data. A score
of 4.3 out of a possible 5 suggests we are meeting our goals of developing a user-ﬁiendly,
meaningful decision support system.

Direct feedback ﬁ'om participants during and after the workshop also indicate
acceptance of the SDSS. Participants stated that they would continue to use the Web-GIS
SDSS for watershed management planning and implementation in target areas and for
quantifying changes in water quality and evaluating nutrient reductions from various
BMPs (using the NPS results from L-THIA) and consider the integration of this
information into future enhancements. One participant, for example, stated the following
regarding the expansion of the L-THIA model: “Would like the ability to change
assumption to consider agricultural land with land application of manure”. Another
participant encouraged expansion of the SDSS to include more agricultural BMPS as well
as urban BMPS by stating the following: “Conversion of agricultural land to urban

development is one of our biggest threats. For ﬁtture updates, it would be nice to

97

incorporate various development types (traditional, low impact development (LID),
conservation design, etc) to provide planners with additional discussion. Also expand
BMP options if possible”. Finally, tool developers have continued to answer questions in
the months following the workshop about the SDSS from the participants as they apply
the tools.

In summary, the survey results and user feedback echo ﬁndings from other DSS
researchers such as Jarupatlrirun and Zahedi (2007). In their effort to study the inﬂuence
of users’ perceptual factors in the success of web-based SDSS, they found that well-
designed, user friendly, and focused SDSS can help improve its perceived “task-
technology ﬁt” (TTF) and “perceived goal commitment”, which in turn improves users’
“self-efficacy” and their satisfaction in decision quality from using the SDSS. This would
lead to greater adoption of the SDSS in real life decision making processes. Other
researchers also attributed the greater community acceptance and participation of a Web-
based SDSS to developing customized, focused, and user friendly tools for community
users who do not have extensive technical skills (Rattray, 2006). From the outset, Web-
based GIS improves the availability of geospatial data and the adoption of spatially
explicit analysis, and, at the same time, incurs no cost to the end-user through the use of
web clients (Peng and Tsou, 2003). Indeed, after the conclusion of the modeling system
development and the workshop, users continued communicating with the development
team to provide feedback on their experiences using the tool and provide suggestions for
tool enhancements.

There are some weaknesses in this watershed management SDSS. First, as

pointed out by a number of users after the workshop, the system only considers a limited

98

list of erosion BMPS. Second, the land use classiﬁcation considered by the current SDSS
is overly broad, lacking the representation of variations within some of the land use types,
e.g., residential area of different densities. These deﬁciencies are largely due to the lack
of site speciﬁc information and data for models to develop such capabilities.

This system proves that integrating multiple SDSS within a V0 according to
speciﬁc user needs and requirements is an excellent way to facilitate the watershed
management tasks. To assess whether the VO-based SDSS approach provides better tools
for watershed management, user evaluations collected by Wisconsin DNR during two
workshops in 2003 and 2004 and two papers (Lucero et al., 2004; Watermolen, 2008)
published by its staff members are used in the following sub-section to illustrate the

point.

3.3.7. Validation of VO-based SDSS Approach for Watershed Management

Wisconsin DNR, as a member of Midwest Spatial Decision Support System
Partnership, assembled over two hundred representatives from diverse agencies and
organizations that make or inﬂuence land use decisions for two “Changing Landscapes”
workshops in 2003 and 2004. Table 3.2 shows a breakdown of workshop participants by
affiliation. Workshop attendees examined and evaluated several tools including Digital
Watershed and L-THIA. They were asked to provide feedback regarding the tools’ utility
and accessibility and evaluate them against a number of measures to help Wisconsin
DNR identify the factors that make tools particularly useful. Table 3.3 contains a list of

criteria derived from the workshops that characterize tools regarded as being most useful

99

to local decision makers. The original raw user survey results can be found in Lucero

 

 

 

 

 

 

 

 

 

 

2004.

Affiliation Percent
Federal Government 4%
State Agency 25%
Tribal Government 1%
Local Government 21%
Regional Planning

Commission 6%
University/Extension 1 8%
Private Firm 18%
Nonproﬁt 7%
Other 1%

 

 

 

 

Table 3.2 Afﬁliation of “Changing Landscapes” Workshop Participants (Percentages are
rounded to the nearest whole number and may not sum to 100) (Lucero et al., 2004).

 

Characteristic Comments

Accessible via the Internet; only required software or
Web-based hardware is an Internet browser.

Housed within the public domain; no purchase cost. Our
research indicates that tools that perform basic ﬁmctions like
data access, interactive mapping, and routine modeling
Cost—free increasingly will be made available in the public domain.
Data required for the tool to function is implicit to the tool.
For example, all mapping tools contain spatial data sets that
can be customized and displayed to illustrate local conditions.
For modeling tools, only the most basic inputs are required.
Thus, there is no cost to create unique scenarios when using
Data included the tools.

Data are accessible at various spatial scales. Tool allows user
Scalable to assess local conditions within a regional context.

Users can address speciﬁc needs through features inherent in
Customizable the tool or through “plug-in” components.

With a user ﬁiendly interface. As users and tool developers
increasingly rely on Intemet-based services for their daily
activities (e. g., travel arrangements, news sources, search
Relatively engines, etc.), consistent, intuitive navigation features are
intuitive becoming increasingly common.

 

 

 

 

 

 

 

 

 

 

Table 3.3 What makes a tool useful? Characteristics derived from Wisconsin DNR’s
evaluative workshops (Waterrnolen, 2008).

100

Watermolen (2008) also noted that participants in Wisconsin DNR’s technical
assistance program often indicate that the “tools do not ‘talk’ to each other.” These users
also repeatedly ask for the capability of two or more tools to interoperate with each other
so they don’t have to worry about the compatibility issues among the tools. Watermolen
(2008) further stated that such interoperability can help narrow the choice of tools and
create integrated decision support systems allowing users to answer questions in a more
holistic manner.

Based on these criteria identiﬁed by Wisconsin DNR, it is very clear that VO-
based SDSS approach such as the Burns Ditch Trail Creek example system described
above offers better tools for watershed management for four general reasons. First,
technical components within a VO-based SDSS are already made interoperable so users
can utilize them in a more integrated manner. Second, such a VO—based SDSS is also
web-based, cost free and doesn’t require users to provide data to use them. Third, as the
Burns Ditch Trail Creek example system demonstrated, a VO-based SDSS can be made
relatively intuitive, scalable and customizable to meet watershed management needs.
Finally, a VO-based SDSS can support capacity building on a broader scale because of
the involvement of multiple organizations in a VO-based SDSS for watershed
management.

In the Burns Ditch Trail Creek example system described above, the author and
his collaborators were able to implement and support highly interactive modeling because
of relatively low resolution of data and simplicity of the model. But modern data
collection technologies now enable researchers to collect large quantities of

geographically referenced data at low cost. Increased levels of investment in the public

101

and private sectors also have led to expansions in the areal coverage of spatial databases
as well as to increases in their levels of details (spatial resolution) (Dowman, 2005).
Spatial models linked to these databases, such as those used for environmental simulation
modeling, spatial statistics, interpolation, and network optimization also continue to
increase in computational intensity (Armstrong, 1994; Wang et al., 2005). As a
consequence of these increases in database size and model complexity, interactive
problem-solving and modeling can be difﬁcult to support. Grid-based high performance
computing can be used to overcome the computational intractability of large, complex
spatial analysis problems. The following section demonstrates the parallelization of a
water quality model in a Grid environment. The parallelized model can be integrated into
the VO-based watershed management SDSS to utilize distributed computing resources

for real time modeling and scenario analysis purposes.

3.4. Parallelization of a Water Quality Model

To support complex real time modeling and scenario analysis in a VO-based
SDSS, Grid-based parallel computing technology can be used to improve the
computational efﬁciency of environmental models in general. But such parallelization
work for different models is more likely to be done individually because a single
generalizable parallelization solution doesn’t exist for different computing algorithms
used in these models (Magillo and Puppo, 1998; Shi et al., 2002). This is also why a V0-
based SDSS using Grid technology can play a very important role in socially oriented
watershed management. Once a complex model is parallelized, it can be integrated into

the system and become a part of the watershed computing infrastructure. As more models

102

are being parallelized and integrated over time, the VO-based watershed computing
infrastructure also grows. This collective effort will beneﬁt generations to come in the
watershed or more broadly environmental computing area.

Many spatial modeling applications utilizing geoprocessing are extremely
resource-intensive due to large volumes of data and intensive algorithms (Armstrong,
1994). For such applications, parallel processing enables the division of the work onto
multiple machines, with potentially large reductions in time to complete the task. This
goal has attracted the attention of researchers involved in diverse, high-load
geoprocessing activities. Hunsaker et al. (1996) developed a parallel implementation for
spatial data error simulation. The task required the production of many equiprobable
realizations of a land cover data set using a complex spatial statistical model. Terrain
modeling involves operations to generate terrain meshes from scattered points, detecting
local surface properties, and calculating viewsheds. Executing these operations on
multiple processors is especially useful when grids are large or dense. Magillo and Puppo
(1998) summarized progress towards implementing parallel terrain modeling. Spatial
environmental modeling of snow conditions across the United States involves intensive
use of resources. Research by El Haddi et al. (1996) sought to improve model
performance by parallelizing the application. Spatial data mining that involves
knowledge extraction from vast stores of spatial data could also beneﬁt from
parallelization. Sorokine et al. (2005) presented parallel visualization techniques for large
spatial dataset. In all the cases described here, parallelization involves some combination

of algorithm development and data model architecture design.

103

An interesting and relevant technical issue for intensive scientiﬁc computation
concerns the different rates of technological change for computing power, storage, and
network speed, typically measured in doubling time. These rates are roughly 18 months,
12 months, and 9 months, respectively (Foster, 2002). Foster notes that petabyte-
magnitude data archives are in the planning stages based on the assumption that this trend
will continue. However, processing capacity is not keeping pace. The implication is that
local computer power will not be up to the task of processing this volume of data.
Network speed, on the other hand, is increasing even more rapidly than storage, as is
exempliﬁed by the 40 Gbs/second rate for the Terragrid network (Benner, 2001). As a
result, communication-intensive tasks will become relatively more efficient than
processor-intensive tasks, encouraging network-distributed solutions for major,
processor-intensive tasks.

In general, two separate problems exist for distributing spatial modeling. The ﬁrst
is the problem of parallelizing the relevant algorithms so that they may be employed on
multiple machines. The second is concerned with identifying an optimal spatial partition
of the database so that the load for each processor is roughly equivalent. Researchers
developed parallel algorithms and spatial declustering techniques (Hoel & Sarnet, 1994;
Shekhar et al., 1996). Spatial operations that may be amenable to parallelization possess
several qualities. The algorithms may have regularities for parallel processing so they can
efficiently use idle cycles on different computers within a V0. Second, the problem may
be one that can be spatially partitioned into smaller geographic units and distributed to
different processors. Third, although spatial analysis may be intensive due to large data

volume rather than complex processing, the results of the analysis may be much smaller

104

than the data volume used for it. For intensive spatial analysis with parallel regularities in
the algorithms, the data domain can be divided spatially. Then subsets of data are
transmitted along with the (sub) algorithm to different machines. Once these smaller,
more tractable subtasks are done, the outcomes are collected, reassembled spatially, and
presented or stored.

A substantial challenge for the parallelization of distributed spatial modeling
applications has been implementing the low-level processing necessary for enabling
multiple processors to work in concert. The Grid facilitates parallel processing by
handling much of this. Toolkits have been developed that supply protocols for data
handling and communications, fault detection, and cross-platform portability (Foster,
2005). This frees the programmer to concentrate on higher-level issues, such as speciﬁc
parallelization implementations, improved algorithm design, or speedier heuristics.

To illustrate how Grid-based parallel computing techniques can be used to meet
the real time modeling needs of watershed modeling, a parallelized spatially explicit soil
loss model was developed using the Message Passing Interface (MP1) standard library
that underlies the Grid. The model was run on both a CrayT3E supercomputer located in
University of Texas and local PC clusters built by the author at Michigan State University

using Grid packages. The architecture of the system is illustrated in Figure 3.8.

 

 

 

 

Internet “9“ Cb'ﬁ’mm
End Users llntemet I “Server 0,
FCClusters

 

 

 

 

 

 

 

 

 

 

Figure 3.8 Prototype system architecture for parallel processing application.

105

The soil loss model is the RUSLE (Revised Universal Soil Loss Equation)
introduced earlier in this Chapter. The RUSLE model was typically applied in a
watershed context using an integrated GIS approach in a raster environment so as to
obtain maps for each RUSLE factor. In other words, every factor was obtained or derived
from source data as a raster layer. The ﬁnal result of the model is also a raster layer from
map algebra calculation (Fernandez et al., 2003).

In this experiment, RUSLE factors for a 1000x1000 cell raster grid were
developed for the calculation of erosion result. This model was run on a single processor
on the Cray T3E; processing required 6.5 seconds. The parallel model was then executed
on 10 processors, which required 0.83 seconds. Performance gain is signiﬁcant, although
this improvement must be weighed against network transfer time.

This spatial application employs purely local map algebra operations upon one or
more input raster layers; that is, the resulting value in cell [i,j] in the output raster is
solely a function of values at [i,j] in the input raster(s). In general, for applications

involving local operations, processing time may be calculated for a single processor:

TS = nRows x nCols x T c (1)

where T S is the time to complete the task on a single processor, nRows and nCols are the
dimensions of the raster, and T c is the time to perform the local function on a single cell.
If the operation is parallelized, processing speed is:

R C l b t
(n owsx n o s)x y esc + nRows xnColsx Tc (2)
NetworkSpeed (bytes / s) nProcs

 

Tp =2anayersx

where nLayers is the number of input raster layers, bytes, is the number of bytes per cell
value that need to be transmitted, and nProcs is the number of distributed processors

handling spatial subsets of the operation. Processor speed is held constant for both single

106

and parallel implementations, which is highly conservative (typically one would take
advantage of very powerful distributed processors). Setting Eqs. (1) and (2) equal to one
another and simplifying results in the following expression:

_ 2 x nLayersx bytesc + TC
0 NetworkSpeed(bytes/ s) nProcs '

 

(3)

The right-hand side of Eq. (3) consists of two elements: the added cost of parallelization,
which is the time required to transfer the data, and the beneﬁt, which is the reduction in
processing time. As nProcs increases, the processing time drops. For a given application
then, one can identify the number of processors required to offset the network speed
bottleneck.

This example shows the great potential of Grid based high performance
computing for environmental applications. The complexity and urgency of the world's
environmental issues demands us to improve our collective problem solving capabilities.
The author believes that building a distributed environmental software system capable of
utilizing distributed computing resources is the most effective way to achieve this goal.
Such a system can record patterns at different scales and understand different processes
that shape these patterns. This system should also be able to grow over time with the
ideal of once developed; new knowledge of our planet in the form of databases and
models can be integrated into the system. Based on state of the art Grid computing and
advanced GIS technology, this system can become part of a national enviromnental
computing infrastructure and will also be able to answer questions about the environment

based on information entered.

107

3.5. Conclusions

Modern watershed management practices require participatory decision making
processes involving multiple shareholders due to increasing complexity, interdependency
and fragmentation of interests and identities within a watershed. The research in this
chapter demonstrates that a multi-organizational VO-based watershed management SDSS
based on advanced information and communication technology and geospatial
technology offers an effective way to facilitate this process.

Based on the requirement analysis conducted in chapter two, a general Grid based
integration framework for the VO-based watershed management SDSS was
conceptualized to provide guidance on how to set up the institutional and technical
framework for such a system. Then a speciﬁc VO-based watershed management SDSS
application was described in detail. This application engages government agencies,
universities and local planning and watershed groups for the common purpose of
reducing sediment runoff into the streams in two relatively small watersheds in northern
Indiana. It differs from most stand-alone or single organizational watershed management
SDSS seen today because it integrates two separate web-based watershed management
spatial decision support systems from two universities to serve the needs of local
stakeholders and managers. It also demonstrates how distributed modeling and
modularity of environmental model components can be implemented within a V0
environment for watershed management. This has rarely been done in the watershed
management community. Most of the requirements speciﬁed in chapter two were fulﬁlled
in this system of systems. The positive feedbacks from end users of the system

demonstrated the effectiveness of VO-based watershed management SDSS approach.

108

User evaluation results collected by Wisconsin DNR are used to validate that a VO-based
SDSS approach offers better tools for watershed management.

As geospatial data resolution continues to increase and spatial modeling becomes
more complex and computationally intensive, a more efﬁcient computing infrastructure is
needed to improve the performance of complex environmental models so end users can
utilize distributed computing resources to conduct what-if scenario analysis in a more
timely fashion. The author argues that VO-based Grid Computing technology is an
excellent ﬁt for this requirement. A water quality model was parallelized using basic Grid
techniques in this chapter to demonstrate this point. The parallelized model can take
advantage of distributed computing resources in 3 Grid environment. A VO-based
watershed management SDSS built on top of the Grid technOlogy capable of growing
over time offers the best opportunity for participatory decision making process in the
watershed management context.

In the next chapter, a new fundamental DEM-based ﬂow analysis algorithm will
be designed, implemented and applied to different physiographic regions in the US. This
statistically shown better algorithm could be integrated into an evolving VO-based

watershed management SDSS to provide better information for decision makers.

109

Chapter 4. A New DEM-based Flow Analysis Algorithm

4.1. Introduction

It is well known that a number of the algorithms associated with DEM-based ﬂow
analysis in most commercially available software are not ideal and they may contain
inherent systematic errors. A speciﬁc example is the so-called D8 ﬂow routing algorithm.
Zhou and Liu (2002) evaluated the errors for various ﬂow routing algorithms based on
synthetic, mathematically generated surfaces and found that the D8 algorithm is the worst
performer of all. Turcotte et al. (2001) identiﬁed three major problem areas for the D8
approach: (1) representation of ﬂow by only eight possible directions with no ﬂow
divergence consideration; (2) presence of ﬂat areas and pits, and (3) the absence of
information about lakes. But despite all these issues with the D8 method, it is still the
most widely used ﬂow routing algorithm in most GIS-based distributed hydrological
models today because of its easy availability in commercial software packages. Even
though researchers around the world have developed better algorithms for ﬂow routing,
they are not easily accessible to most end users. This is where a VO-based Watershed
Management SDSS can come into play. In a VO-based Watershed Management SDSS,
better models incorporating better algorithms can be integrated into the system and
applied to real-world applications in a more timely fashion because of the ubiquitous
accessibility of the system. For example, if a new erosion model using a better ﬂow-
routing algorithm is developed and integrated into the online system, end users of the
VO-based Watershed Management SDSS will have instant access to the new model. This
chapter presents a new DEM-based ﬂow routing analysis algorithm based on principles in

ﬂuid mechanics. Various existing DEM-based ﬂow modeling methods and their

110

limitations are reviewed in section 4.2. Section 4.3 introduces the new ﬂow routing
algorithm called FlowNet. The applications of FlowNet, along with two other popular
algorithms, in different physiographic regions are described in section 4.4. Section 4.5
examines the spatial variation in ﬂow accumulation derived from different algorithms.
Section 4.6 contrasts algorithm performance on hydrologic applications. Section 4.7

concludes the chapter.

4.2. Hydrologic Modeling Using Digital Elevation Data

GIS-based hydrological modeling has come a long way since the early 1990s.
Even though the ten-step modeling procedure presented by David Maidment (1996) still
holds true (see Table 4.1), much progress has been made towards the tighter and
smoother integration of hydrological models and core GIS functionality and more
accurate and realistic representation of physical and man-made features within a

watershed.

111

 

1. Study design: Objectives and scope of study; spatial and time domain; process
models needed, variables to be computed.

2. Terrain analysis: Deriving a watershed and stream network layout from digital
elevation data and mapped streams.

3. Land surface: Describing soils, land cover, land use, cities, and roads.
4. Subsurface: Hydrogeologic description of aquifers

5. Hydrologic data: Locating point gages, attaching time series and their average
values, interpolating point climatic data onto grids.

 

 

 

 

 

6. Soil water balance: Partitioning precipitation into evaporation, groundwater
recharge and surface runoff; partitioning of chemicals applied to the land surface.

7. Water ﬂow: Movement of water through the landscape in streams and aquifers.
Computing streamﬂow and groundwater ﬂow rates.

8. Constituent transport: Transport of sediment and contaminants in water as it
ﬂows. Computing concentrations and loadings.

9. Impact of water utilization: Locating reservoirs, water withdrawals and
discharges in rivers, and aquifer pumping. Their effects on water ﬂow and
constituent transport.

10. Presentation of results: Developing visual and tabular presentation of the study
results. Use of Internet and CD-ROM to transmit results.

 

 

 

 

 

 

 

Table 4.1 Ten-step procedure for a GIS hydrology study (David Maidment 1996).

Terrain analysis based on digital elevation data is a fundamental step in GIS-
based hydrological modeling. A digital elevation model (DEM), also widely known as a
digital terrain model (DTM), is a digital representation of ground surface topography or
terrain. A DEM can be represented as a raster (a grid of squares) or as a triangular
irregular network (TIN). DEMs are commonly built using remote sensing techniques;
however, they may also be built from land surveying and secondary sources such as maps
and contours etc. (Wilson and Gallant 2000). DEMs are being increasingly used in GIS
for hydrologic analysis. For example, in the popular watershed modeling package SWAT,
DEMs are used as a fundamental data layer for watershed modeling (Diluzio et al. 2001,

Arnold et al. 1998). This is due to the general availability of digital elevation data,

112

nationally from the USGS (2003) and worldwide including the space-based data available
from the NASA Shuttle Radar Topography Mission (SRTM). The increasing computing
power in personal computers has also made DEM processing functions more readily
accessible and easy to use. DEM-based ﬂow routing analysis has been a traditional
research topic for decades in the Geographic Information Sciences (Wilson et al. 2000).

The function of a ﬂow algorithm in a GIS is to transfer ﬂow (water, sediment,
nutrients) to lower, adjacent points or areas in a landscape (Desmet and Govers 1996a).
Hence, a raster-based, ﬂow-routing algorithm determines the way in which the outﬂow
from a given cell will be distributed to one or more downslope cells. The choice of ﬂow
routing algorithm is important and affects the calculation of the upslope contributing
area, speciﬁc catchment area, stream power index, and several other topographic
attributes. Zhou and Liu (2002) used mathematical surfaces to calculate true values of
speciﬁc catchment area and compared the results with ones generated using different
grid-based ﬂow routing algorithms. They found all these algorithms have some level of
error associated with them. They indicated that the major cause of these processing errors
is the nature of the grid data structure and over-simpliﬁed assumptions about the behavior
of water ﬂow controlled by the surface morphology. Clarke and Lee (2007) investigated
the impact of spatial resolution and algorithm choice on the computation of downslope
ﬂow from DEMs. Their analysis revealed only minor differences among the algorithms
used. They also found that these algorithms are fraught with critical assumptions and
scale effects.

Many different raster-based ﬂow routing algorithms have been implemented to

simulate water ﬂow in speciﬁc environments. The guiding principles and decision rules

113

employed in ﬁve of the most commonly adopted algorithms —— the D8 (O’Callaghan and
Mark 1984), Rho8 (Fairﬁeld and Leymarie 1991), FD8 (Quinn et al. 1991), DEMON
(Lea 1992; Costa-Cabral and Burges 1994), and D00 (Tarboton 1997) algorithms are
reviewed in this section, along with several studies that have compared their performance
in different environments.

The deterministic eight-node (D8) single-ﬂow-direction algorithm, the earliest
and simplest method for specifying ﬂow directions, assigns ﬂow from each pixel to one
of its eight neighbors, either adjacent or diagonally, in the direction with steepest
downward slope. This method was introduced by O'Callaghan and Mark (1984). The D8
algorithm works well to mimic the ﬂow of rivers and streams, and ﬂow convergence in
valleys (Chorowicz et al. 1992; Quinn et al. 1991; Fairﬁeld and Leymarie 1991; Costa-
Cabral and Burges 1994; Tarboton 1997). However, the D8 approach oversirnpliﬁes the
possible ﬂow direction from a grid center by limiting ﬂow to one grid cell, and as a result,
is unable to simulate divergent ﬂows (Holrngren 1994). These limitations are most often
expressed by the presence of multiple parallel ﬂow paths in either the cardinal or
diagonal directions (i.e., multiples of 45°) that are produced with this algorithm (Fairﬁeld
and Leymarie, 1991; Quinn et al., 1991; Costa-Cabral and Borges, 1994). Other sources
of error for D8 include the presence of ﬂat areas and pits, and the lack of information on
the locations of lakes (Turcotte et al. 2001). An advantage of the method is that it is
relatively easy to calculate upslope contributing area and speciﬁc catchment area since all
ﬂow in one pixel drains into the steepest downslope pixel. Upslope contributing area,
therefore, is the number of pixels whose ﬂow reaches the pixel of interest multiplied by

the pixel area (Costa-Cabral and Burges 1994).

114

F airﬁeld and Leymarie (1991) developed the random eight-node (Rho8)
algorithm to break up the parallel ﬂow paths that D8 produces by randomly assigning a
ﬂow direction to one of the downslope neighbors, with the probability proportional to
slope. This algorithm starts by identifying all the downslope neighboring cells, then
calculating the slope gradients in each of these directions, and ﬁnally choosing a number
from a table of random numbers to direct the ﬂow to one of these candidate cells. The
numbers are allocated on a slope-weighted basis such that the potential ﬂow paths with
the steepest gradients have the greatest probability of being selected and the overall
pattern more or less matches that which would have been produced with the D8 algorithm.
However, this algorithm may still calculate unrealistic ﬂow directions in upslope areas
and there is now the added problem that a different ﬂow network will be produced each
time the algorithm is used because of the reliance on a table of random numbers to
allocate ﬂow among multiple downslope cells (Wilson and Gallant 2000). The upslope
contributing areas for each cell are calculated using the ﬂow accumulation approaches
adopted for D8; however, the degree of randomness implemented in Rho8 means that this
attribute will be overestimated or underestimated for numerous cells (Costa-Cabral and
Burges 1994).

FD8 multiple ﬂow direction algorithms (Quinn et al., 1991; Freeman, 1991) have
also been developed to address the limitations of D8. These methods direct water to every
adjacent downslope cell on a slope-weighted basis. Each cell will receive only a fraction
of the discharge from each upslope cell, and therefore, the upslope contributing area of
the receiving cell is composed of fractional contributions from different cells (Costa-

Cabral and Burges 1994).

115

Lea (1992) developed an algorithm that uses the aspect associated with each pixel
to specify ﬂow directions. Flow is routed as though it were a ball rolling on a plane
released from the center of each grid cell. A plane is ﬁt to the elevations of pixel comers,
these comer elevations being estimated by averaging the elevations of adjoining pixel
center elevations. This procedure has the advantage of specifying ﬂow direction
continuously (as an angle between 0 and 2a). Costa-Cabral and Burges (1994) presented
an elaborate set of procedures named DEMON (Digital Elevation Model Network) that
extend the ideas of Lea (1992). Grid elevation values are used as pixel comers, rather
than block centered, and a plane surface is ﬁtted for each pixel. The upslope contributing
area for each cell in DEMON is computed by successive addition of the inﬂuence matrix
of every pixel in the DEM (Costa-Cabral and Burges 1994).

Tarboton (1997) proposed the D00 method that incorporates several ideas from
DEMON to assign multiple ﬂow directions to selected cells. This single ﬂow direction
(represented as a continuous quantity between 0 and 2 it radians) is determined in the
direction of the steepest downward slope on the eight triangular facets formed in a 3 x 3
pixel window centered on the pixel of interest (Figure 4.1). Each downslope vector is
drawn outward from the center and may be at an angle that lies within or outside the 45°
vector (or a multiple of li/4 radians). If the slope vector angle falls within the facet, it
represents the steepest ﬂow direction of that facet. If the slope vector angle lies outside
the facet, the steepest ﬂow occurs along the steepest edge. The slope for each vector is
calculated and used to determine the ordering of the facet numbers from 1 to 8 (Figure
4.1). If vectors do not ﬂow downslope, a ﬂow direction angle of -1 is assigned to signify

ﬂat areas or pits. The ﬂow direction is forced to ﬂow toward a neighbor of equal

116

elevation that has a ﬂow direction resolved in these instances. This ensures that ﬂat pixels
drain to a neighbor that ultimately drains to a lower elevation, and eliminates loops in the
ﬂow direction angles, so that the user does not need to worry about spurious pits in the
DEM. The upslope area of each pixel is taken as its own area plus the fractional areas of
upslope neighbors that drain into the pixel of interest, similar to F D8 and DEMON
(Tarboton 1997). If the angle falls on a cardinal or diagonal direction, then the ﬂow from
each cell drains to one neighbor. If the ﬂow direction falls between the direct angles to
two adjacent neighbors, the ﬂow is proportioned between the two neighbor pixels
according to how close the ﬂow direction angle is to the direct angle for those pixels

(Tarboton 1997).

117

 

 

Row indices

1+1

Proportion of flow to pixel

Column indices

j j+l

(i-l, j) is A2 / (Al+A2)

 

-----------------

 

 

I
I
I
I
I
I
I
I
I;
I
I
I
I
I
I
I
I

 

 

 

 

 

.—

k

/

Proportion of
flow to pixel
(i-l, j+l) is
Al / (Al-I-A2)

Flow direction
measured as
counter-clockwise
angle from east

Steepest downslope
direction

Facet numbering

 

Figure 4.1 Flow direction on planar triangular facets in a block-centered grid (from

ﬂow across the land surface represented as a square-grid DEM. These rules vary in two

key ways: (1) the method or granularity used for aspect calculations, and (2) the number

The ﬁve algorithms described in this section provide a series of rules for directing

Despite all the advances in the DEM-based ﬂow routing analysis, all the above—

Tarboton 1997).

of downslope cells that can receive ﬂow from each upslope cell.

118

mentioned algorithms do not consider or utilize the physical laws that govern water

movement as a ﬂuid through the landscape. These algorithms oversimplify water

 

movement and often treat water as a solid instead of a ﬂuid moving through the

landscape.

4.3. New Flow Routing Algorithm

Here the author proposes a new method for the representation and calculation of
ﬂow direction and accumulation based on mathematical principles and physical laws that
govern water movement through the landscape. In the study of ﬂuid dynamics, we often
consider that ﬂuids form a continuum wherein the motion of individual particles is not
traced. The focus is on a control volume, a ﬁxed frame in space through which the ﬂuid
passes. This is also called the Eulerian view of motion. In the proposed method, this view
is adopted and every DEM cell is treated as a control volume. The assumption is that
water ﬂows through the saturated DEM surface, so Darcy’s law can be applied for the
water ﬂow. One implication of this new method is that only four direct neighboring cells
(directly above, below and directly left and right) instead of eight for every DEM cell are
considered when the water ﬂow is modeled. The four diagonal cells are not considered
because mathematically the common area between every DEM cell and its diagonal
neighboring cell is 0, so that there is no possible ﬂow between them. This forms the
foundation of the new algorithm. The actual ﬂow volume to one or more of the remaining
direct neighboring cells is determined by Darcy’s law. Darcy's Law is a generalized
relationship for ﬂow in a porous medium (Chow et al. 1988). It states that volumetric
ﬂow rate is a function of the ﬂow area, elevation, ﬂuid pressure and a proportionality

constant. Darcy's Law can be summarized as:

_ Ah
9'“? (1)

119

Where,

Q = voltunetric ﬂow rate (m3/s),

A = ﬂow area perpendicular to L (m2),
K = hydraulic conductivity (m/s),

L = ﬂow path length (m),

A h = change in hydraulic head (m)

Let’s assume that the total incoming ﬂow equals Qt for any center cell that has one or

more direct lower neighboring cells. The ﬂow proportion for cell m that is lower than the

center cell can be calculated as:

Slpm (2)

m= t
Q QSlp1+...+Slpm+...+Slpn

Where,

Qm = ﬂow proportion for the cell m that is lower than the center cell,
Qt = total incoming ﬂow for the center cell,

Slp, = slope between the center cell and the ﬁrst lower direct neighboring cell,
Slpm = slope between the center cell and the mth lower direct neighboring cell,
Slpn = slope between the center cell and the nth lower direct neighboring cell,

m = any integer number between 1 and n,

n = total number of cells that are lower than the center cell, note n <= 4.

The proposed algorithm routes the actual ﬂow according to formula (2) from the

highest cell to the lowest cell. Its output is a ﬂow accumulation grid A, in which each cell

120

value Aij is the sum of all upstream ﬂows that pass through cell (i, j). The algorithm works
as follows: First a list L is produced by sorting the elevation values in the DEM in
decreasing order, and every cell in A is initialized to one unit of ﬂow. Then the points are
visited in decreasing order of elevation by scanning through L, while ﬂow is distributed
to lower neighboring cells by updating entries in A. When point (i,]) is processed, Aij
contains the correct ﬁnal ﬂow accumulation value for (i, j), since all higher points (if any)
pushing ﬂow into (i, j) have already been processed. Conceptually the algorithm
corresponds to sweeping the terrain top-down with a horizontal plane, “pushing” ﬂow
down the terrain in front of the plane (See ﬁgure 4.2 for an illustration of this concept).
The actual ﬂow routing calculation of formula (2) is implemented in the following C++

function:

void FlowCalculator::moveFlow(int col, int row)

{

int i, cOffs, rOffs;
int offsresult; /*, IowCount;*/

stSortGC *tmp;
vector <stSortGC *>lowList; // we only need to know the contents

double ptVal = source->get(col, row);
double slopeTotal;

I/ determine which values are lower
for (i = 0; i < 4; i++){

cOffs = col;
rOffs = row;
offsresult = source->calc0ffs(i, cOffs, rOffs);
if (offsresult) {
if (source->get(cOffs, rOffs) < ptVal){
llIowCount++;
llisLow[i] = TRUE;
tmp = new stSortGC;
tmp->dZ = source->get(cOffs, rOffs);
tmp->iCol = cOffs;
tmp—>iRow = rOffs;
lowList.push_back(tmp);
}

121

}
”calculate slopeTotal
slopeTotal = 0;

for (i = O; i < IowList.size(); i++){
slopeTotal += (ptVal - lowList[i]->dZ) / source->cellSize();
}

double propoltion, curSlope;
for (i = O; i < IowList.size(); i++){
curSlope = (ptVaI - lowList[i]->dZ) / source->cellSize();
proportion = curSlope/slopeTotal;
result->get(lowList[i]->iCol, lowList[i]->iRow) += result->get(col, row) * proportion;

ll Done, cleanup
for (i = 0; i < IowList.size(); i++){
if (lowList[i] != NULL) {
delete lowList[i];
lowList[i] = NULL;
}

lowList.clear();

 

—-— Sweeping Plane
— DEM Proﬁle

 

 

 

 

Figure 4.2 Illustration of the Sweeping Plane Concept

This algorithm requires that the DEM data set be preprocessed to make sure every
cell except for boundary cells has at least one lower direct neighboring cell. A method
based on mathematical morphology called lower complete transformation (Soille 1994) is

used to achieve this goal. The lower complete transformation removes all pits and ﬁat

122

areas in a DEM and mathematically guarantees every cell except for those on the
boundary has at least one lower direct neighboring cell. This new algorithm shall be

referred to as the “FlowNet” method in this chapter and beyond.

4.4. Application to Different Physiographic Regions

To evaluate the performance and effectiveness of the F lowNet algorithm on a
variety of real-world topographic surfaces, a sampling process based on physiographic
provinces, state boundaries and available 12-digit watershed HUC (Hydrologic Unit
Code) boundaries is used to select sample watersheds for testing purposes. The purpose is
to test the algorithm on a number of watersheds in different types of landscapes. The core
sampling unit is the elevation of the 12-digit watershed. All DEM data were downloaded
from USGS National Map Seamless Server (USGS 2003). The horizontal resolution of
these DEMs is 1/3 are second, which were then projected into UTM projections with a
horizontal spacing of 10 meters. The physiographic provinces were used to ensure that
different types of landscapes are sampled. Figure 4.3 is a map of United States
physiographic provinces based on data from the USGS (U SGS 2002). Different provinces
are expected to have terrain with varying surface properties; therefore they were used to
stratify the study sites. Since the 12-digit watershed boundary data are made available by
state for a limited number of states, both 12-digit watershed boundary and state boundary
data are used to select the sample dataset. The United States is divided and sub-divided
into successively smaller watersheds which are classiﬁed into different levels. Each
watershed is identiﬁed by a unique hydrologic unit code (HUC) consisting of numbers

based on the different levels of classiﬁcation in the hierarchical coding system. A typical

123

12 digit HUC level watershed is a relatively small (10,000 — 40,000 acres) local
watershed. As of November 2008, the 12 digit HUC level watershed boundary data are
not available across the US. Figure 4.4 is a map showing availability of 12-digit
watershed boundary data at the time when data for this research were collected. The gray
shaded states have certiﬁed data available for download.

This experimental application started with sample data selection in three different
physiographic provinces (Central Lowland of Interior Plains, Valley and Ridge of
Appalachian Highlands, and Basin and Range of Intermontane Plateaus), and then three
ﬂow routing algorithms (FlowNet, Duo and D8) were applied to all data in the sample.

Finally the results were collected and analyzed. The following is the detailed process.

 

Physiographic Provinces

- ADIRONDACK

APPALACHIAN PLATEAUS
BASIN AND RANGE

BLUE RIDGE

CASCADE-SIERRA MOUNTAINS

   
 
  
    
  

COLORADO PLATEAUS
COLUMBIA PLATEAU
GREAT PLAINS
m INTERIOR Low PLATEAUS
LOWER CAUFORNIAN
MIDDLE ROCKY MOUNTAINS
m NEW ENGLAND
NORTHERN ROCKY MOUNTAINS
[:3 OUACHITA

OZARK PLATEAUS
- PACIFIC BORDER
IIII] PIEDMONT
SOUTHERN ROCKY MOUNTAINS
ST. LAWRENCE VALLEY
- SUPERIOR UPLAND
VALLEY AND RIDGE
WYOMING BASIN

/(;€Z? ¢¢ﬂy

  

\\
\\\\\\\
s\\\

 

 

 

Figure 4.3 Physiographic Provinces

124

 

WATERSHED BOUNDARY DATASET (WBD) STATUS

     

         
       

 
 
   

G A” ’
.,..g.:5?;.;.;::j,~. ‘~ .
, In. 1 e1 e .
9.93.5. 5' .-.* ,. 2.1.;

.4-

 
  

  

  
    
    
   
 

r I ._ .
«:3? pr: w”;
as . 2., .
I‘- fd. 5", ,. #5?
5 . a . "‘?§’
‘24?" .‘auan-at- Te

- . . 31"".
I ' 7 '19. ' .'
‘.

 
  

   

 

 

 

Figure 4.4 Watershed Boundary Dataset Status (NRCS)

According to the map overlay of available 12-digit watershed boundaries, state
boundaries and physiographic provinces, a total of 21 digit watersheds in three
physiographic provinces (Central Lowland of Interior Plains, Valley and Ridge of
Appalachian Highlands, and Basin and Range of Intermontane Plateaus) located in three
states (Michigan, New Mexico and Virginia) were selected. Each state or physiographic
province has exactly seven representative 12-digit watersheds selected. The selection
process was not random. Visual judgment was used to select representative watersheds in
the three physiographic provinces. Figure 4.5 — 4.7 show locations of selected

watersheds.

125

 

 

Selected Watersheds in Michigan

/ - Selected Watersheds
" [:3 State Boundaries

 

3

 

Figure 4.5 Locations of Selected Watersheds in Michigan

 

 

Selected Watersheds in New Mexico

 

  
 

- Selected Watersheds
E State Boundaries

 

Figure 4.6 Locations of Selected Watersheds in New Mexico

126

 

 

 

Selected Watersheds in Virginia

 

- Selected Watersheds
[:3 State Boundaries

 

 

 

Figure 4.7 Locations of Selected Watersheds in Virginia

Table 4.2 lists common descriptive metrics for all 21 selected watersheds. These
metrics include area, average slope, minimum elevation, maximum elevation, mean
elevation and standard deviation of elevation. A comparison of these metrics indicates
that all watersheds have relatively similar sizes. Selected watersheds in Michigan are
very ﬂat with small elevation ranges, while the New Mexico and Virginia watersheds

have rugged terrain with relatively large elevation ranges.

127

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Standard
Watershed Area Average Minimum Maximum Mean Deviation
(Acres) Slope Elevation Elevation Elevation Of

(dggree) Elevation

Ml WS 1 23,843 1.0 215 256 236.2 6.5
MI WS 2 17,509 0.8 210 245 225.0 5.1
MI WS 3 18,958 1.1 222 263 244.9 6.9
MI WS 4 20,735 0.9 260 301 271.7 5.3
MI WS 5 10,424 1.2 274 316 285.1 5.4
MI WS 6 13,201 1.2 274 305 285.8 4.5
MI WS 7 14,597 0.8 208 247 223.9 4.9
NM WS 1 23,171 18.9 2,083 3,210 2,566.2 189.2
NM WS 2 32,955 28.3 1,455 2,671 2,110.6 220.8
NM WS 3 20,211 21.2 1,783 2,821 2,147.1 192.5
NM WS 4 11,472 26.2 1,863 2,969 2,323.1 227.2
NM WS 5 34,729 21.3 1,965 3,100 2,448.4 248.3
NM WS 6 16,512 6.7 1,653 2,319 1,868.7 122.4
NM WS 7 38,336 17.5 L651 2,751 2,115.7 205.5
VA WS 1 9,495 18.0 574 1,012 745.4 89.5
VA WS 2 16,235 21.1 474 1,337 824.5 172.3
VA WS 3 22,785 14.2 416 1,148 616.4 149.4
VA WS 4 14,685 6.8 330 523 405.9 33.3
VA WS 5 29,115 7.5 395 678 514.2 63.4
VA WS 6 22,797 15.4 278 1,111 517.8 146.1
VA WS 7 24,021 17.0 775 1,362 1,084.1 101.1

 

 

 

 

 

 

 

 

Table 4.2 Summary Table for 21 Selected Watersheds

Three ﬂow routing algorithms, FlowNet, D8 and D00, were applied to all 21
watersheds, and ﬂow accumulations for each cell were calculated. The FlowNet
algorithm was implemented in C++ as a command line tool for this experiment. The Doc
algorithm was implemented in C++ as a command line tool by David Tarboton. It was
downloaded from Tarboton’s website (Tarboton 2008). Both FlowNet and Doc command
line tools were used to batch process the elevation data and generate results. The D8
algorithm is available in ArcGIS, so ArcGIS was used to calculate D8 results. Results
were compared using qualitative and quantitative techniques. Qualitative approaches

included looking at output maps and contrasting method performance. The following

128

section presents detailed results on a subset of three watersheds, one from each
physiographic region. Quantitative approaches included:
e conducting difference of means testing on average ﬂow accumulation values and
erosion potential erosion values for each method
e evaluating the linear relationship between D8 and F lowNet methods via linear
regression
e assessing whether signiﬁcant differences in model performance occurred between
physiographic provinces

Quantitative results are reported and discussed in section 4.5.

4.5. Spatial Variation in Flow Accumulation

The ﬁrst watershed, called South Fork Hayworth Creek, is located in Clinton
County, Michigan, in the “Central Lowland of Interior Plains” province. The area of this
watershed is 14,597 acres. it’s very ﬂat with an average slope of 0.8 degrees, and drains
to the north, with relatively higher elevations in the southeast. The minimum and
maximum elevation values are 208 and 247 meters respectively. The following is a map

of the watershed.

129

 

South Fork Hayworth Creek Watershed

Rivers
Watershed Boundary
DEM

- 208-212
- 212-215

   

 

 

 

Figure 4.8 South Fork Hayworth Creek Watershed in Clinton, MI

The second watershed, called North Fork South Branch Potomac, is located in

Highland County, Virginia. This watershed falls in the physiographic province of “Valley

and Ridge of Appalachian Highlands.” The area of this watershed is 24,021 acres. The

average slope is 17.0 degrees. The minimum and maximum elevation values are 775 and

1362 meters respectively. It drains to the north with higher elevations in the southwest.

The following is a map of the watershed. Note that this 12-digit watershed actually

contains two subwatersheds. This situation may occur for other 12—digit watersheds as

well.

130

 

 

 

  

1012 Miles
E

North Fork South Branch Potomac Watershed

Rivers
Watershed Boundary

-922-971
[:1971 -1019
E] 1019- 1068
1:] 1068- 1117
[:3 1117- 1166
1166-1215
— 1215- 1264
- 1264-1313
S 1313-1362

E: No Data

 

 

Figure 4.9 North Fork South Branch Potomac Watershed in Highland, VA

The third watershed, called Upper Bear Creek, is located in Grant County, New
Mexico. This watershed falls in the physiographic province of “Basin and Range of
Intermontane Plateaus.” The area of this watershed is 38,336 acres. The average slope for
this watershed is 17.5 degrees. The minimum and maximum elevation values are 1651

and 2751 meters respectively. It drains to the west with higher elevations in the northeast.

The following is a map of the watershed.

131

 

 

Upper Bear Creek Watershed

Rivers
Watershed Boundary
E

1651 - 1742

1742- 1834
- 1834-1926
1926- 2017
[:1 2017- 2109
[:| 2109-2201
[:1 2201- 2293
2293-2334
2384- 2476

2559 - 2751
No Data

 

1 0 1 2Miles
E

 

 

 

Figure 4.10 Upper Bear Creek Watershed in Grant, NM

Now let’s take a look at the detailed results for three ﬂow routing algorithms. The
ﬁrst set of six ﬁgures (4.11 — 4.16) is for South Fork Hayworth Creek watershed in
Michigan. All stream cells are derived from the ﬂow accumulation grid with a threshold
value of 100 cells for comparison purposes; they do not necessarily reﬂect permanent

stream locations.

132

 

Stream Cells Derived from FlowNet Algorithm
for South Fork Hayworth Creek Watershed

  
   
 

[:j Blowup Area
- Dfn Stream Cells
[:j Watershed Boundary

 

 

   

0.5 0 0.5 1 Miles
E

 

 

 

Figure 4.11 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
FlowNet Algorithm

133

 

 

Stream Cells Derived from FlowNet Algorithm (Blowup)
for South Fork Hayworth Creek Watershed

 

 

 

 

 

 

 

 

Figure 4.12 Local Blowup of Figure 4.11

134

 

 

 

Stream Cells Derived from Dinf Algorithm
for South Fork Hayworth Creek Watershed

 
   
   
 

|:| Blowup Area
- Dinf Stream Cells

[:3 Watershed Boundary

0.5 O 0.5 1 Miles

 

Figure 4.13 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
Doc Algorithm

135

 

 

 

Stream Cells Derived from Dinf Algorithm (Blowup)
for South Fork Hayworth Creek Watershed

‘91 VII—'— " .
e <

' WL

 

LXI
I
.-

 

 

Figure 4.14 Local Blowup of Figure 4.13

136

 

 

Stream Cells Derived from D8 Algorithm
for South Fork Hayworth Creek Watershed

\

   
  
 

 
 
  

[:1 Blowup Area
- D8 Stream Cells
|:] Watershed Boundary

0.5 0 0.5 1 Miles

 

 

 

Figure 4.15 Stream Cells Derived from F low Accumulation Grid Calculated Using the
D8 Algorithm

137

 

Stream Cells Derived from DB Algorithm (Blowup)
for South Fork Hayworth Creek Watershed

W "——'

”7";

 

,\

x!
Y
A‘-

.L
‘\

 

is L

 

 

 

Figure 4.16 Local Blowup of Figure 4.15

The ﬂow accumulation ﬁgures for the Michigan watershed illustrate that the
stream cell patterns derived from D8 and D00 are very similar. Two factors may explain
this. The ﬁrst is that Doo also uses single ﬂow direction, even though the ﬂow direction is
not limited to 8 possible directions like in D8. The second is that Doo uses the same
method as D8 when dealing with relatively ﬂat areas like the majority of this watershed.
Flow directions in ﬂat areas are determined iteratively by making them ﬂow towards a
neighbor of equal elevation that has a ﬂow direction resolved. A very obvious difference
is that the FlowNet algorithm produces ‘thick’ streams. Doo also produces ‘thick’ streams,

but not as many as the F lowNet algorithm does. D8 does not produce ‘thick’ streams at

138

all. This is due to how ﬂow is routed in different algorithms. In the FlowNet algorithm.
which is based on physical laws that govern water movement through the landscape, ﬂow
can be routed to up to four neighboring cells. In Doo, which is based on the geometric
conﬁguration of neighboring grid cells, ﬂow can be routed to a maximum of two
neighboring cells. In D8, ﬂow can only be routed to one of eight neighboring cells based

on slopes.

Similar observations can be made for the other 6 watersheds in Michigan after
examining their respective results derived from using the three algorithms.

Now results for the other two watersheds are considered.

139

 

Stream Cells Derived from FlowNet Algorithm
for North Fork South Branch Potomac Watershed

  

[:1 Blowup Area
- Dfn Stream Cells

[:l Watershed Boundary

 
  

1 0 1 2Miles
SE

 

 

 

Figure 4.17 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
FlowNet Algorithm

140

 

 

Stream Cells Derived from FlowNet Algorithm (Blowup)
for North Fork South Branch Potomac Watershed

I
“ A - n.—

 

\ » - ,
is.

 

I
I

\

\\\‘~\ /

 

 

Figure 4.18 Local Blowup of Figure 4.17

141

 

 

 

Stream Cells Derived from Dinf Algorithm
for North Fork South Branch Potomac Watershed

    

 

Blowup Area
Dinf Stream Cells
[:] Watershed Boundary

1 0 1 2 Miles
E

 

Figure 4.19 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
Doc Algorithm

142

 

 

 

Stream Cells Derived from Dinf Algorithm (Blowup)
for North Fork South Branch Potomac Watershed

,

 

.1;

 

 

Figure 4.20 Local Blowup of Figure 4.19

143

 

 

 

Stream Cells Derived from D8 Algorithm
for North Fork South Branch Potomac Watershed

  

 

[:3 Blowup Area
- D8 Stream Cells

[: Watershed Boundary

u—l

0 1 2 Miles
55

 

Figure 4.21 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
D8 Algorithm

144

 

 

 

Stream Cells Derived from DB Algorithm (Blowup)
for North Fork South Branch Potomac Watershed

.-

v.“"‘e I h j

 

 

Y
I.ib.

 

 

Figure 4.22 Local Blowup of Figure 4.21

145

 

 

 

Stream Cells Derived from FlowNet Algorithm
for Upper Bear Creek Watershed

 

    
  

[:1 Blowup Area
- Dfn Stream Cells

1:] Watershed Boundary

 

Figure 4.23 Stream Cells Derived from F low Accumulation Grid Calculated Using the
F lowNet Algorithm

146

 

 

 

Stream Cells Derived from FlowNet Algorithm (Blowup)
for Upper Bear Creek Watershed

q x. _ ‘N a
T e w‘

 

 

\‘-

 

C
I
I. L V F I L

J

 

Figure 4.24 Local Blowup of Figure 4.23

147

 

 

 

Stream Cells Derived from Dinf Algorithm
for Upper Bear Creek Watershed

 

    
  

Blowup Area
Dinf Stream Cells
[:] Watershed Boundary

 

 

Figure 4.25 Stream Cells Derived from Flow Accumulation Grid Calculated Using the
Doc Algorithm

148

 

 

 

Stream Cells Derived from Dinf Algorithm (Blowup)
for Upper Bear Creek Watershed

f- l A I ‘P‘ j;
1
\

" 9 1 s

 

J
II

 

 

C '. .
° I.
/ I '
A ,
r r e r I 1‘ 1 I I ‘

 

 

 

Figure 4.26 Local Blowup of Figure 4.25

149

 

 

 

Stream Cells Derived from D8 Algorithm
for Upper Bear Creek Watershed

 
 
 

 
   

[:j Blowup Area
- D8 Stream Cells

[:j Watershed Boundary

 

Figure 4.27 Stream Cells Derived from F low Accumulation Grid Calculated Using the
D8 Algorithm

150

 

 

Stream Cells Derived from D8 Algorithm (Blowup)
for Upper Bear Creek Watershed

 

 

 

J

47;“ _ J 1 l
V
ut' — 4r
‘\
‘
i ‘ r
:1
- A” ,-

 

 

 

 

 

J L 4'

 

Figure 4.28 Local Blowup of Figure 4.27

Similar observations can be made for these two watersheds in Virginia and New
Mexico as ones for the previous watershed in Michigan from this set of ﬁgures. But the
streams derived from the FlowNet algorithm in these mountainous areas are not as ‘thick’
as those derived from the FlowNet algorithm in ﬂat areas. The reason for this difference
is that in mountainous areas water ﬂow tends to disperse less because of steep slopes. In
other words, ﬂows tend to be more concentrated because of the rugged terrain. The better

channelization effects in mountainous areas also contribute to this fact.

151

Similar observations can be made for the other 12 watersheds in Virginia and

New Mexico after examining their respective results derived from using the three

algorithms.

4.6. Contrasting Algorithm Performance on Hydrologic Applications

The previous section provided a visual comparison of a few representative
watersheds. To get more information on how different algorithms perform, we also need
to ﬁnd a way to directly compare ﬂow accumulation grids generated from different
algorithms. Mean values of all 21 ﬂow accumulation grids were collected for this purpose

(table 4.3).

152

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

D8 Flow Doo Flow FlowNet Flow
Watershed Accumulation Mean Accumulation Mean Accumulation Mean
MI WS 1 1,023.9 1,024.7 1,378.9
MI WS 2 794.7 793.1 1,028.2
Ml WS 3 1,072.0 1,071.2 1,372.1
Ml WS 4 945.0 945.4 1,190.5
Ml WS 5 605.0 602.7 778.1
Ml WS 6 585.6 585.3 753.2
Ml WS 7 886.2 885.5 1,203.6
NM WS 1 945.8 949.3 1,378.6
NM WS 2 1,298.9 1,298.3 1,752.4
NM WS 3 1,050.8 786.0 1,485.4
NM WS 4 906.8 907.8 1,280.2
NM WS 5 1,315.6 1,316.4 1,866.5
NM WS 6 1,262.5 1,262.8 1,755.3
NM WS 7 1,231.8 1,236.4 1,769.8
VAWS1 761.4 761.1 1,111.4
VA WS 2 799.7 793.1 1,109.2
VA WS 3 1,048.5 1,044.7 1,462.7
VA WS 4 740.4 740.9 1,073.4
VA WS 5 1,054.2 1,050.8 1,498.1
VA WS 6 1,110.8 1,103.9 1,606.0
VA WS 7 963.7 962.0 1,352.9

 

From Table 4.3 we can see that ﬂow accumulation grid mean values for D8 and
Doc appear to be very similar, but the FlowNet algorithm seems to produce systematically
larger mean values. To test whether these differences are signiﬁcant, paired T—Tests were

conducted using the open-source statistical package R (R Development Core Team

Table 4.3 Mean Values of Flow Accumulation Grids

2005); output is provided in Appendix A. The results are listed as follows:

1. Paired T-Test between Doe and D8 ﬂow accumulation mean values:

H0: Actual difference in means between Doe and D8 ﬂow accumulation grids is equal to

O.

153

 

HA: Actual difference in means between Doe and D8 ﬂow accumulation grids is not
equal to O.

t = -1.0665, df = 20, p-value = 0.2989, mean of the differences = -13.42090
The calculated p-value exceeds 0.05, so we fail to reject the null hypothesis and conclude
that the D00 ﬂow accumulation grid mean value is the same as the D8 ﬂow accumulation
grid mean value.

2. Paired T-Test between F lowNet algorithm and D8 ﬂow accumulation mean values:
H0: Actual difference in means between FlowNet algorithm and D8 ﬂow accumulation
grids is equal to 0.

HA: Actual difference in means between F lowNet algorithm and D8 ﬂow accumulation
grids is greater than 0.

t = 15.4888, df = 20, p-value = 6.652e-13, mean of the differences = 371.5763
In this case, the calculated p-value is much smaller than 0.05, so we reject the null
hypothesis and conclude that the FlowNet algorithm ﬂow accumulation grid mean value
is greater than the D8 ﬂow accumulation grid mean value.

Flow accumulation has been widely used in a variety of GIS-based environmental
models. In order to ﬁnd out how the FlowNet algorithm will affect environmental models
that use ﬂow accumulation, the RUSLE model described in Chapter 3 is chosen and run
for all 21 watersheds using the three algorithms. The RUSLE is a set of mathematical
equations that estimate average annual soil loss and sediment yield resulting from interrill
and rill erosion by multiplying ﬁve factors such as slope length (LS) and soil erodibility

(K) (Renard et al. 1994). The erosion potential mean results are listed in the table 4.4.

154

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

D8 Erosion Doo Erosion FlowNet Erosion

Watershed Potential Mean Potential Mean Potential Mean

Ml WS 1 0.96 1.01 1.98
Ml WS 2 0.65 0.69 1.25
Ml WS 3 0.87 0.94 1.58
MI WS 4 0.75 0.80 1.58
MI WS 5 0.75 0.82 1.17
MI WS 6 0.80 0.85 1.36
MI WS 7 0.74 0.78 1.51
NM WS 1 33.73 36.01 49.80
NM WS 2 46.10 48.99 68.52
NM WS 3 29.30 30.77 42.55
NM WS 4 40.92 43.23 60.16
NM WS 5 32.32 34.38 48.04
NM WS 6 8.08 8.74 11.70
NM WS 7 28.61 30.72 42.01
VA WS 1 23.59 24.96 34.20
VA WS 2 34.26 36.26 52.82
VA WS 3 19.04 20.03 27.35
VA WS 4 6.81 7.22 10.68
VA WS 5 9.26 9.78 16.47
VA WS 6 20.81 22.15 30.70
VA WS 7 26.53 28.40 38.74

 

 

 

 

Table 4.4 Mean Values of Erosion Potential Grids ﬁ'om RUSLE

 

By examining the results in the table it seems that similar observations can be
made about erosion potential grid mean as for ﬂow accumulation grid mean previously.
Therefore two paired T-Tests are run for the erosion potential means.

1. Paired T-Test between D00 and D8 erosion potential mean values:
H0: Actual difference in means between D00 and D8 erosion potential grids is equal to 0.
HA: Actual difference in means between Doe and D8 erosion potential grids is not equal

to O.

t = 5.1517, df = 20, p-value = 4.854e-05, mean of the differences = 1.079048

155

The calculated p-value is much smaller than 0.05, so we reject the null hypothesis and
conclude that the D00 erosion potential grid mean value is not the same as D8 erosion
potential grid mean value. This is not what we expected even though the difference does
not appear to be large.
2. Paired T-Test between FlowNet algorithm and D8 erosion potential mean values:
H0: Actual difference in means between FlowNet algorithm and D8 erosion potential
grids is equal to 0.
HA: Actual difference in means between FlowNet algorithm and D8 erosion potential
grids is greater than 0.
t = 5.3756, df = 20, p-value = 1.458e-05, mean of the differences = 8.537714

The calculated p-value is much smaller than 0.05, so we reject the null hypothesis and
conclude that that FlowNet erosion potential grid mean value is larger than the D8
erosion potential grid mean value. This difference favors the proposed algorithm because
the current RUSLE method based on D8 tends to underestimate the overall erosion
potential of a watershed (USEPA Region 7 and IOWA Dept. of Natural Resources 2006).
The preceding results demonstrate signiﬁcant differences in erosion potential for the
commonly employed D8 algorithm and FlowNet algorithm. Is this difference
predictable? To evaluate this, a linear regression model was developed in R to predict
F lowNet erosion potential grid mean value from the D8 erosion potential grid mean
value. The model is listed as follows and detailed R output can be found in appendix A:

Epna = A + B * Epd8

Where A = 0.38890, t = 1.33, Pr(>|t|) = 0.199

B = 1.46902, t = 115.45, Pr(>|t|) <2e-16

156

Adjusted R2 = 0.9985
From the adjusted R-squared value, as well as a scatterplot of the two variables (ﬁgure
4.28), it is clear that a strong linear relationship exists between the two variables. The
intercept coefﬁcient is 0.38890; meaning even if there is no erosion predicted by D8
algorithm, the F lowNet algorithm will have an estimated erosion value of 0.38890. The
slope coefﬁcient is 1.46902, indicating a strongly positive relationship between the
variables. An increase of 1 in Epd8 corresponds to a rise of 1.46902 in Epna. This
coefﬁcient differs signiﬁcantly from zero (t=115.45, P << 0.0001). Standard regression

diagnostics revealed no problems with this model.

Erosion Potential Means

 

FlowNet
10 20 30 40 50 60 70

 

 

0
l

 

0 10 20 30 40
D8

Figure 4.29 Relationship between FlowNet algorithm and D8 erosion potential means

157

To evaluate the variability of this relationship between different physiographic
regions, residuals of the linear model for each watershed were collected (table 4.5) and T-
Tests were conducted to determine whether model differences varied signiﬁcantly among

regions. More detailed results are listed in Appendix A.

 

 

 

 

 

 

 

 

 

 

 

 

No. MI Residual NM Residual VA Residual
1 0.173957 -0. 12971 -0.84904
2 -0.10218 0.412 2.105704
3 -0.08442 -0.87894 -1 .0027
4 0.089859 —0.34046 0.285516
5 -0.31 18 0.175202 2.47694
6 -0.19425 -0.561 14 -0.26569
7 0.030549 -0.40919 -0.62022

 

 

Table 4.5 Linear Model Residuals for Watersheds in Different Physiographic Regions

1. Residual T-Test between MI and NM

t = 1.0715, df = 7.714, p-value = 0.3163, mean of x = -0.0569, mean of y = -0.2475

2. Residual T-Test between MI and VA

t = -0.6661, df = 6.167, p-value = 0.5294, mean of x = -0.0569, mean of y = 0.3044

3. Residual T-Test between NM and VA

t = -0.979, df = 7.132, p-value = 0.3596, mean of x = -O.2475, mean of y = 0.3044

These T-Test results tell us that the linear relationship between the FlowNet
algorithm erosion potential mean values and the D8 erosion potential mean values does
not vary signiﬁcantly among different physiographic regions. But Table 4.5 also shows

us that the absolute differences for the VA residuals are much bigger than the MI

158

residuals. This indicates that high terrain variability leads to high variability in residuals.
What this means for the linear model derived above is that it performs well in the ﬂat
areas such as Michigan with less variability in the results.

Even though we may be able to derive F lowNet algorithm erosion potential mean
values from D8 erosion potential mean values using this linear relationship, how do these
differences appear spatially within the watersheds? Are differences essentially constant
across each watershed, or are there important patterns of differences between them? To
answer these questions, map algebra based calculations were done for all watersheds to
derive mapped differences. First a predicted FlowNet algorithm erosion potential grid
from D8 erosion potential grid was calculated using the above linear model. Then a
difference grid was calculated using the actual FlowNet algorithm erosion potential grid
and the predicted FlowNet algorithm erosion potential grid. The detailed results for three

selected watersheds are shown below.

159

 

Difference Grid for South Fork Hayworth Creek Watershed

 
  
 
 
 
 
 
 
  
 
 

Difference Grid for MW 7
< -3 Std. Dev.
-3 - -2 Std. Dev.
-2 - -1 Std. Dev.
-1 - 0 Std. Dev.
[:1 Mean
0- 1 Std. Dev.
a"; 1 - 2 Std. Dev.
- 2 - 3 Std. Dev.
- > 3 Std. Dev.

[:1 No Data

0.5 0 0.5 1 Miles
E

 

 

 

Figure 4.30 Difference Grid for South Fork Hayworth Creek Watershed in MI

 

Difference Grid for North Fork South Branch Potomac Watershed

  
 

Difference Grid for VW 7
< -3 Std. Dev.
= -3 - -2 Std. Dev.
- -2 - -1 Std. Dev.
“ -1 - 0 Std. Dev.

  

0 1 2 Miles

 

 

 

Figure 4.31 Difference Grid for North Fork South Branch Potomac Watershed in VA

160

 

Difference Grid for Upper Bear Creek Watershed

Difference Grid for NW 7
- < -3 Std. Dev.
-3 - -2 Std. Dev.
-2 --1 Std. Dev.
1:] -1 - 0 Std. Dev.
[:1 Mean
0- 1 Std. Dev.
1- 2 Std. Dev.
- 2 - 3 Std. Dev.
- > 3 Std. Dev.
:1 No Data

 

 

 

 

Figure 4.32 Difference Grid for Upper Bear Creek Watershed in NM

Similar difference grid results were derived from all other watersheds as well.
There is tremendous spatial variability in the differences between the accumulation
algorithms. Both high and low difference values occur close to streams. The reason for
this is the FlowNet algorithm tends to generate ‘thick’ streams with more cells with
higher ﬂow accumulation values in the vicinity of the real stream than the D8 algorithm.
Another interesting observation is that D8 generates stream cells with higher ﬂow
accumulation value right on the stream than the FlowNet algorithm. This is due to the
fact that ﬂow is distributed more to cells near the stream in the F lowNet algorithm. From
these results we can conclude that even though the linear relationship can be used to

predict FlowNet algorithm erosion potential grid mean values from D8 erosion potential

161

grid mean values, it cannot be used to predict F lowNet algorithm erosion potential spatial
distributions from D8 erosion potential spatial distributions. This is particularly important
because in real world applications involving erosion potential, spatial distributions give
people speciﬁc information about where those might occur, and shape management
responses to the threat of erosion.

It could be argued that the ﬂow pattern generated by the F lowNet algorithm
should reﬂect the actual ﬂow pattern in the real world better than the other ﬂow-routing
algorithms because it is based on a physical law that governs the movement of water
through the landscape. Because of this attribute, it should be integrated into various
hydrological models in a more systematic way. It may also provide a better foundation
for combined surface and groundwater modeling as well. This FlowNet algorithm may
also form the basis of a more accurate wetness index. To assess whether FlowNet
produces more realistic ﬂow accumulation results than D8 method, wetland data
collected by Michigan state government were used to validate results generated by both
FlowNet and D8 algorithms. The details of this validation work are described in the next

section.

4.7. Validation of FlowNet Algorithm

To test whether the FlowNet algorithm matches reality better than the traditional
D8 approach, the independent wetland location data (e.g. Figure 4.3 3) available from
Michigan Center for Geographic Information (MI CGI Website, 2008) are used to collect
sampling data from ﬂow accumulation grids generated by both FlowNet and D8 methods

for seven Michigan watersheds for comparison purposes. The rationale is that if FlowNet

162

is a better algorithm, then wetland locations not in the stream channel should have higher
values than D8 for those same cells. At the same time, non-wetland locations should have
similar values between algorithms. The following sampling procedure and statistical tests

are designed based on this principle.

 

Wetlands in South Fork Hayworth Creek Watershed

/\/ Rivers

Wetlands
Watershed Boundary

 

  
  
 

 

 

 

 

 

0.5 0 0.5 1 Miles

 
 

 

 

 

Figure 4.33 Wetlands in South Fork Hayworth Creek Watershed, Ml

Speciﬁc steps for data sampling and statistical tests are described as follows:
1. Flow accumulation grids derived from both FlowNet and D8 for all seven
watersheds are pre-processed in ArcGIS Grid environment to exclude
stream cells derived from D8 algorithm using a threshold value of 500

cells. The threshold value is chosen by visually comparing the resulting

163

stream cells with national hydrography vector data. This exclusion is
necessary because these stream cells are not considered as part of the land
mass.

. The wetland polygons are overlaid on top of the ﬂow accrunulation grids
and then used to clip these grids to produce both wetland ﬂow
accumulation grids and non-wetland ﬂow accumulation grids respectively
using ArcGIS Grid functions. This process is repeated for each of the
seven watersheds in Michigan. In the end, for every watershed two
wetland ﬂow accumulation grids and two non-wetland ﬂow accumulation
grids are generated for FlowNet and D8 respectively (e.g. Figure 4.34 -
4.3 7).

. Hawth Analysis Tools for ArcGIS (Hawth Analysis Tools website, 2008)
is used to generate two random sampling point themes for each watershed,
one for wetland ﬂow accumulation grids and one for non-wetland ﬂow
accumulation grids (e.g. Figure 4.38 — 4.39).

. One thousand random sampling point locations are used to sample
accumulation values from both FlowNet and D8 grids in wetland areas for
each watershed. A total of 7 tables are generated and ready for use in
statistical tests.

. One thousand random sampling point locations are used to sample
accumulation values from both FlowNet and D8 grids in non-wetland
areas for each watershed. A total of 7 tables are generated and ready for

use in statistical tests.

164

6. Paired differences of means t-tests are conducted between FlowNet and
D8 wetland ﬂow accumulation samples for every watershed. The null
hypothesis (H0) for this test is: actual difference in means between
wetland ﬂow accumulation samples from FlowNet method and wetland
ﬂow accumulation samples from D8 method is equal to 0. The alternative
hypothesis (HA) is: actual difference in means between wetland ﬂow
accumulation samples from FlowNet method and wetland accumulation
samples from D8 method is greater than 0. Here one-sided tests are used
because our purpose is to show ﬂow accumulation values generated by
FlowNet are bigger than those generated by D8 in wetland areas. The
results of these tests are included in table 4.6. The raw outputs of these
tests from R are included in appendix A.

7. Statistical paired t-test is conducted between F lowNet and D8 non-wetland
ﬂow accumulation samples for every watershed. The null hypothesis (H0)
for this test is: actual difference in means between non-wetland ﬂow
accumulation samples from FlowNet method and non-wetland ﬂow
accumulation samples from D8 method is equal to 0. The alternative
hypothesis (HA) is: actual difference in means between non-wetland ﬂow
accumulation samples from F lowNet method and non-wetland
accumulation samples from D8 method is not equal to 0. Here two-sided
tests are used because our goal is to demonstrate ﬂow accumulation values
generated by FlowNet are similar than those generated by D8 in non-

wetland areas. And if they are not similar, we expect FlowNet samples

165

 

could either be bigger or smaller than D8 samples. The results of these
tests are also included in table 4.6. The raw outputs of these tests from R

are included in appendix A.

 

FlowNet Wetland Flow Accumulation Grid for
South Fork Hayworth Creek Watershed

   
  
 
  
  
 

[:1 Watershed Boundary
FlowNet Wetland Flow Accumulation Grid
-- --1 0 Std. Dev.

 

-2 38td. Dev
->38td. Dev.

|:] No Data

0.5 0 0.5 1 Miles
E

 

 

Figure 4.34 FlowNet Wetland F low Accumulation Grid for South Fork Hayworth Creek
Watershed, MI

166

 

 

D8 Wetland Flow Accumulation Grid for
South Fork Hayworth Creek Watershed

 
  
 
 
 
 
 
 
  
 
 

:1 Watershed Boundary

D8 Wetland Flow Accumulation Grid
-1 - 0 Std. Dev.
Mean

0 - 1 Std. Dev.
- 1 - 2 Std. Dev.
- 2 - 3 Std. Dev.
- > 3 Std. Dev.

1:] No Data

0.5 0 0.5 1 Miles
E

 

Figure 4.35 D8 Wetland Flow Accumulation Grid for South Fork Hayworth Creek
Watershed, MI

167

 

 

 

FlowNet Non-Wetland Flow Accumulation Grid for
South Fork Hayworth Creek Watershed

 
 
 
 
 
   
   

:1 Watershed Boundary

Mean

0 - 1 Std. Dev.
n 1 - 2 Std. Dev.
- 2- 3 Std. Dev.
- > 3 Std. Dev.

E) No Data

 

Figure 4.36 FlowNet Non-Wetland Flow Accumulation Grid for South Fork Hayworth
Creek Watershed, MI

168

 

 

 

D8 Non-Wetland Flow Accumulation Grid for
South Fork Hayworth Creek Watershed

 
   
 
  

l::| Watershed Boundary
DB Non-Wetland Flow Accumulation Grid
- -1 - 0 Std. Dev.

2 - 3 Std. Dev.
- > 3 Std. Dev.

[:1 No Data

 

 

Figure 4.37 D8 Non-Wetland Flow Accumulation Grid for South Fork Hayworth Creek
Watershed, MI

169

 

 

Wetland Grid Sampling Points for
South Fork Hayworth Creek Watershed

  
   
 
  
 
 
  
 
 

Wetland Grid Sampling Points
Watershed Boundary
D8 Wetland Flow Accumulation Grid
1.2-""22: -1 - 0 Std. Dev.
*2: Mean
0- 1 Std. Dev.
:- 1- 2 Std. Dev.
- 2 - 3 Std. Dev.
- > 3 Std. Dev.
[:I No Data

 

 

0.5 O 0.5 1 Miles
E

 

Figure 4.38 Wetland Grid Sampling Points for South Fork Hayworth Creek Watershed,
MI

170

 

 

 

Non-Wetland Grid Sampling Points for
South Fork Hayworth Creek Watershed

 
 
  
 
 
 
 
 
 
  

Non-Wetland Grid Sampling Points

[:1 Watershed Boundary
D8 Non-Wetland Flow Accumulation Grid
-1 - 0 Std. Dev.
Mean
0 - 1 Std. Dev.

1 - 2 Std. Dev.
- 2 - 3 Std. Dev.
- > 3 Std. Dev.
E] No Data

 

 

Figure 4.39 Non-Wetland Grid Sampling Points for South Fork Hayworth Creek
Watershed, MI

171

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Mean of the

Watershed Test T Df P-value differences

MI WS 1 Wetland 2.59 999 0.005 1820
MI WS 1 Non-Wetland 1.47 999 0.141 299
M1 WS 2 Wetland 3.89 999 <0.001 2174
M1 WS 2 Non-Wetland 1.18 999 0.240 169
M1 WS 3 Wetland 4.02 999 <0.001 3462
M1 WS 3 Non-Wetland 2.21 999 0.027 276
M1 WS 4 Wetland 2.73 999 0.003 3273
M1 WS 4 Non-Wetland 1.45 999 0.148 261
Ml WS 5 Wetland 3.73 999 <0.001 1805
M1 WS 5 Non-Wetland 1.82 999 0.068 71
MI WS 6 Wetland 4.07 999 <0.001 1828
M1 WS 6 Non-Wetland 2.22 999 0.027 88
MI WS 7 Wetland 2.93 999 0.002 776
Ml WS 7 Non-Wetland 1.80 999 0.072 178

 

 

Table 4.6 Paired T-Test Results between F lowNet and D8 F low Accumulation Samples
for both Wetland and Non-Wetland Areas

Considering the t-test results between F lowNet and D8 wetland ﬂow
accumulation samples listed in table 4.6 for all seven watersheds, it is clear that all
calculated p-values are much smaller than 0.05. Therefore, we can reject the null
hypothesis and conclude that the mean of FlowNet wetland ﬂow accumulation samples is
larger than the mean of D8 wetland ﬂow accumulation samples. While seeing t-test
results between F lowNet and D8 non-wetland ﬂow accumulation samples, ﬁve out of '
seven watersheds have calculated p-values bigger than 0.05. In those cases we fail to
reject the null hypothesis and conclude that mean of FlowNet non-wetland ﬂow
accumulations samples is the same as the mean of D8 non-wetland ﬂow accumulation
samples. For the remaining two watersheds 3 and 6, even though judging from the

calculated p-value we cannot accept the null hypothesis for non-wetland ﬂow

172

accumulation samples, there are big differences between p-values for wetland and non-
wetland tests. These results show that F lowNet ﬂow accumulation totals are higher in
wetland, non-channel locations than D8 ﬂow accumulation totals, and they are generally
not signiﬁcantly different in non-wetland locations. This means FlowNet ﬂow
accumulation results capture the wetland locations better than D8 ones because FlowNet
algorithm produces ‘wetter’ cells in wetland areas while still generates ‘dry’ cells in non-
wetland areas in comparison with D8 method. Therefore, it can still be concluded that

FlowNet algorithm is better than D8 method in terms of matching wetland locations.

4.8. Conclusions

Various ﬂow routing algorithms and their advantage and shortcomings were
reviewed in this chapter. The review found that the existing ﬂow routing algorithms tend
to oversimplify the water movement across the landscape without consideration of
physical laws that govern the ﬂuid movement. Based on this fact, a new ﬂow routing
algorithm called FlowNet was developed and applied to the total of 21 watersheds in
three different physiographic regions in the US. Two other algorithms, the popular D8
and Doc, were also applied to these watersheds for comparison purposes. RUSLE soil
erosion estimates based on the three ﬂow routing algorithms were also done for all these
watersheds. The ﬁnal results were collected and compared to show the advantages of the
FlowNet algorithm. Statistical analyses were conducted to ﬁnd out whether certain
relationships exist between the results using three algorithms. Wetland data collected
from Michigan Center for Geographic Information were also used to validate that
FlowNet algorithm performs better than D8 method in terms of matching wetland

locations.

173

Compared with D8 and Doc, the F lowNet algorithm tends to generate ‘thicker’
stream cells because of the ﬂow patterns governed by physical laws. FlowNet shares
important similarities with the FD8 algorithm because they both distribute ﬂow to lower
neighboring cells on a slope-weighted basis, but the key difference is that FlowNet only
distributes ﬂow to four possible orthogonal neighbors according to the physical law that
governs the ﬂow. Of course, if the FD8 algorithm routing scheme is limited to four
orthogonal neighbors as well, it should work the same way as the FlowNet algorithm.
However, FlowNet has the capability to incorporate other parameters for ﬂuid movement
besides topography and it can be further developed into a complete surface water ﬂow
model.

The ﬂow pattern generated by the FlowNet algorithm should reﬂect the actual
ﬂow pattern in the real world situations and hence make it a better basic ﬂow routing
algorithm for various hydrological models. The RUSLE soil erosion modeling results
seem to conﬁrm this because the RUSLE model based on D8 often underestimates
erosion while the results in this study show that RUSLE model based on FlowNet
systematically generate signiﬁcantly larger results than the model based on D8. The
statistical linear relationship found between RUSLE erosion based on FlowNet and
RUSLE erosion based on D8 is Eﬂownet = 0.38890 + 1.46902 * Ed8. The residuals
results from different physiographic regions for this linear model indicate it performs
well in the ﬂat areas such as Michigan with less variability in the results.

If the F lowNet algorithm is integrated into our VO-based Watershed Management
SDSS discussed in the previous chapters, it can be hosted on universities’ server nodes

along with all DEM datasets. Other server nodes in the system or end users could directly

174

access this new function over the network immediately for their own use and beneﬁt from
the new algorithm. Also in a modular environmental modeling environment within a V0-
based Watershed Management SDSS, this new algorithm can replace the problematic
method currently in place and improve the modeling capability of the overall system. But
to make this happen, FlowNet algorithm would have to be improved by parallelization
using Grid technology. This also illustrates how a living VO-based Watershed
Management SDSS capable of integrating new algorithms or models over time can help
hydrologists, information scientists, and policymakers extend their collective abilities for
watershed management.

Another recent advance in DEM-based ﬂow routing techniques that could
complement the FlowNet algorithm is the realistic representation of physical and man-
made features such as roads and bridges in a watershed and use this secondary
information to generate even more realistic ﬂow patterns. Turcotte et al. (2001) described
a new approach to couple the DRLN (digital river and lake network) to the modeled
drainage structure. Vogt et al. (2003) proposed a methodology to combine digital
elevation data and environmental characteristics to derive drainage networks and
catchment boundaries. Duke et al. (2003, 2006) reported a new approach to improve
overland ﬂow routing by incorporating ancillary road data into DEMs. Further research is

needed to ﬁnd out how to integrate this new advance with the FlowNet algorithm.

175

Chapter 5. Summary and Conclusions

5.1. Introduction

Watershed management is essentially a consensus-building process relying on the
active involvement of multiple stakeholders. A good watershed management plan must
reﬂect social consensus regarding water as a resource and the society’s vision of an ideal
watershed and water quality and quantity (Nickum and Easter, 1990; Davenport, 2002;
Gregersen et al., 2008). The nature of watershed management demands a holistic
integrative approach to consider the impacts of various human activities in a watershed
on water resources and other related natural resources (Reimold, 1998). Watershed
management oﬁen entails close collaboration among specialists from different disciplines
such as hydrology, ecology, biology, engineering, economics, management science, law
and policy, as well as government agencies, private industries, non proﬁt organizations,
and the public (Heathcote, 1998; Reimold, 1998; Gregersen et al., 2008). An effective
way to integrate scientiﬁc information from multiple disciplines into the real world
decision making process is mandatory for successful watershed management practices.

Even though the advances in information technology, geospatial technology and
environmental modeling techniques have dramatically changed watershed management
practices in the past few decades, the reality of current watershed management is still
highly fragmented in terms of the data collection, modeling activities and management
practices (Savory et al., 1999; Acreman, 2005; Cash et al., 2003). Due to the spatial
nature of watershed environmental data, information is collected and analyzed by
organizations and people who typically care about a speciﬁc location such as a state, a

county or a watershed. Various computer models that use these data to describe

176

watershed systems have been created by research organizations or private environmental
consulting companies. All these data, information, analytical methods and simulation
models are often scattered across different organizations with limited or no
communication between organizations because of fragmented management practices
(Acreman, 2005; Cash et al., 2003; NRC, 1999). Since many watershed environmental
problems require ongoing investigation, the next team or individual investigators who
want to study the same place usually spend signiﬁcant time and effort collecting existing
information from multiple sources. This is not an efficient way to build up our collective
watershed management capabilities.

In this research, the author argues that a collaborative online Virtual Organization
based Watershed Management Spatial Decision Support System capable of integrating
distributed databases and computational models offers superior opportunities for holistic
watershed management. This system can provide an effective way to embed science in
the decision making process by improving the communication between the scientiﬁc and
the policy sector and between decision-makers and involved stakeholders and translating
often complicated modeling results into an easy to understand format for environmental
decision makers. Based on the user evaluation results collected by Wisconsin DNR, it can
be concluded that this VO-based approach for spatial decision support offers better tools

for watershed management.

5.2. Major Contributions

The research conducted led to the conceptualization of an Internet based

integration framework for the VO-based watershed management SDSS based on

177

requirements analysis for the VO-based watershed management SDSS and
implementation of an example VO-based watershed management SDSS. This example
system engages government agencies, universities and local planning and watershed
groups for the common purpose of reducing sediment runoff into the streams of two
relatively small watersheds in northern Indiana. It integrates two separate web-based
watershed management spatial decision support systems ﬁom two universities to serve
the needs of local communities. The system allows end users to simulate different best
management practices (BMP) scenarios online in real time with an easy to use graphical
user interface. This system also demonstrates how distributed modeling and modularity
of environmental model components can be implemented in a V0 environment.

With continuing growth of availability of higher resolution geospatial data and
computational intensity of spatial and watershed models, more efﬁcient computing
infrastructure is needed for the improvement of complex modeling performance so end
users can carry out management scenario analyses in a real time fashion. The author
believes that VO-based Grid Computing technology can play a signiﬁcant role in
fulﬁlling this requirement. To demonstrate this point, a water quality model was
parallelized using basic Grid techniques to show how performance can be improved by
utilizing distributed computing resources. A VO-based watershed management SDSS
built on top of the Grid technology with the ability of evolving over time is a strong
architecture for participatory decision making process in the watershed management
context.

To illustrate the importance of technical aspects of watershed management

process, a new basic DEM-based ﬂow analysis algorithm called F lowNet based on the

178

physical law that governs water movement was designed, implemented and applied to
different physiographic regions in the US. The statistical results show this is a better
algorithm in terms of overcoming the erosion potential underestimation problem
associated with traditional D8 method (U SEPA Region 7 and IOWA Dept. of Natural
Resources 2006). This algorithm was also validated using independent wetland data
available in Michigan to illustrate that it performs better than D8 method in terms of
matching wetland locations. If F lowNet is integrated into an evolving modular VO-based

watershed management SDSS, it can help decision makers get better ﬂow information.

The major contributions of this research to the literature on spatial decision
support system, watershed management and geographic information science are:

0 Analysis and documentation of requirements for VO-based Watershed
Management Spatial Decision Support Systems. This is reported in Chapter two.

0 Conceptualization and docmnentation of an Internet—based integration framework
for the VO-based watershed management SDSS. This is reported in Chapter three.

0 Implementation of the example VO-based Watershed Management Spatial
Decision Support System for sediment runoff reduction in two watersheds in
northern Indiana and validation of VO-based SDSS Approach for Watershed
Management. These are documented in Chapter three.

- Demonstration of the utility of Grid Computing technology in VO-based
Watershed Management Spatial Decision Support System via parallelization of a
selected water quality model and subsequent performance analysis. This is

documented in Chapter three.

179

Design, implementation, application and validation of a fundamentally new
DEM-based ﬂow analysis algorithm and statistical analysis of application results.

This is described in Chapter four.

5.3. Major Conclusions

Results from this study produced several conclusions related to the ﬁelds of

spatial decision support systems, watershed management and geographic information

science:

The social technical nature of watershed management process requires a multi-
organizational and multi-disciplinary integration framework for the participatory
consensus building decision making process.

A VO-based Watershed Management Spatial Decision Support System utilizing
Grid Computing and advanced geospatial technology offers an effective way to
involve multiple shareholders and integrate scientiﬁc information from multiple
disciplines into participatory decision making process in watershed management.
Parallelization based on Grid technology provides an efﬁcient way to improve the
performance of watershed models so they can be integrated into the VO-based
Watershed Management Spatial Decision Support system in real time fashion.

A better basic DEM-based ﬂow analysis algorithm can be designed and
implemented by considering the physical laws that govern water ﬂow in the real

world landscape.

180

5.4. Recommendations for Future Research Directions

The research in this study suggests the following future research directions:

1). How to deﬁne the institutional agreement ﬁ'amework so organizations can be
encouraged to participate in the watershed management V0 and be willing to
integrate and share their resources. More organizational research needs to be done to
identify policies and incentives that will bring more related institutions into the
watershed management VO. Policies that encourage and reward VO participation and
incentives such as extra ﬁmding mechanism will likely help in this aspect.

2). How to integrate Grid Computing technology into the technical inﬁastructure
of major players such as government agencies and universities in a V0. Even though
Grid technology is increasingly gaining popularity, it still largely remains in the
academic and research communities. Further research needs to be done to develop
easy-to-implement methods for transforming existing facilities into or constructing
new Grid-enabled technical infrastructure in non-academic institutions such as
government agencies.

3). How to integrate data intensive and complex watershed simulation models into
Grid-based computing infrastructure so end users can conduct real or near real time
scenario analysis online. Data intensive and complex watershed simulation models
are often too time consuming for real or near real time scenario analysis. New
speciﬁc parallelization techniques need to be developed for various simulation
models by modeling communities.

4). How to calibrate and validate complex watershed models in a V0 and carry

out uncertainty analysis in a relatively short time period utilizing Grid infrastructure.

181

w-

3“,.

 

The calibration of complex watershed models is still a major challenge in using
watershed models for environmental decision making. Even though some auto
calibration modules have been developed to calibrate watershed models, most of
these tools currently run on a single computer requiring signiﬁcant CPU hours to
complete the model calibration. More research is needed to develop the model
calibration process utilizing Grid Computing so that the model calibration can be
performed in a relatively shorter time frame.

5). How to present and visualize environmental information in a more easy to
understand way so any stakeholder can quickly understand the complexities and make
an informed decision. Even though some animations on rainfall, temperature, and
graphs/maps on streamﬂow conditions exist, there are no other easy means of
creating visualizations for use as educational tools to both decision makers and
stakeholders. More research needs to be done to develop various educational tools
that will use public domain and data as well as user provided data and 2D/3D
visualization techniques to explain hydrologic/water quality processes to various
stakeholders in the Grid environment.

6). How to better organize watershed environmental information, engage more
people in the watershed decision making process and help them build social
consensus regarding solutions to environmental problems in a V0 environment.
Watershed environmental information within a V0 often comes in different format
from different sources. Organizing and presenting this information in a consistent,
transparent and holistic way is a major challenge that will demand more research

efforts in ontologies and semantics for environmental information. Getting more

182

people involved in the watershed decision making process is a very import goal of
any watershed management VO. New technology such as social networking is likely
to help with this objective. Further research is needed to investigate new ways for
integrating social networking technology into watershed management V0 to bring
together a diverse group of people (scientists, policymakers, NGOs, stakeholders,
students, educators, and the public) who are concerned with watershed environmental
issues and provides a platform that encourages them to voice their opinions, share

information, and make connections with other people.

183

Appendix A. Statistical Test Results from R:

1. Paired T-Test between Dinf and D8 ﬂow accumulation mean values:
H0: Actual difference in means between Dinf and D8 ﬂow accumulation grids is equal to
0.

HA: Actual difference in means between Dinf and D8 ﬂow accumulation grids is not
equal to 0.

> t.test(mydata$fadi,mydatanadB,paired=T)
Paired t-test

data: mydata$fadi and mydata$fad8
t = -1.0665, df = 20, p-value = 0.2989
alternative hypothesis: true difference in means is not equal to
O
95 percent confidence interval:
-39.67098 12.82917
sample estimates:
mean of the differences
—l3.42090

2. Paired T-Test between FlowNet algorithm and D8 ﬂow accumulation mean values:
H0: Actual difference in means between FlowNet algorithm and D8 ﬂow accumulation
grids is equal to 0.

HA: Actual difference in means between FlowNet algorithm and D8 ﬂow accumulation

grids is greater than 0.

>
t.test(mydatanana,mydata$fad8,alternative=c("greater"),paired=T)

Paired t-test

data: mydatanana and mydata$fad8
t = 15.4888, df = 20, p—value = 6.652e-13
alternative hypothesis: true difference in means is greater than
0
95 percent confidence interval:
330.2002 Inf
sample estimates:

184

mean of the differences
371.5763

3. Paired T-Test between Dinf and D8 erosion potential mean values:
H0: Actual difference in means between Dinf and D8 erosion potential grids is equal to 0.
HA: Actual difference in means between Dinf and D8 erosion potential grids is not equal

toO.
> t.test(mydata$epdi,mydataSede,paired=T)
Paired t-test

data: mydata$epdi and mydata$epd8
t = 5.1517, df = 20, p-value = 4.854e-05
alternative hypothesis: true difference in means is not equal to
0
95 percent confidence interval:
0.6421343 1.5159609
sample estimates:
mean of the differences
1.079048

4. Paired T-Test between F lowNet algorithm and D8 erosion potential mean values:

H0: Actual difference in means between FlowNet algorithm and D8 erosion potential
grids is equal to 0.

HA: Actual difference in means between FlowNet algorithm and D8 erosion potential

grids is greater than 0.

>
t.test(mydata$epna,mydata$epd8,alternative=c("greater"),paired=T)

Paired t-test

data: mydata$epna and mydata$epd8
t = 5.3756, df = 20, p-Value = 1.458e-O5
alternative hypothesis: true difference in means is greater than
0
95 percent confidence interval:
5.798465 Inf
sample estimates:

185

mean of the differences
8.537714

5. Linear regression model between FlowNet erosion potential grid mean value and D8

erosion potential grid mean value:

> regnad8<-1m(mydata$epna~mydata$epd8)
> summary(regnad8)

Call:
lm(formula = mydata$epna ~ mydata$epd8)

Residuals:
Min 1Q Median 3Q Max
—1.0027 -0.4092 —0.1297 0.1740 2.4769

Coefficients:

Estimate Std. Error t value Pr(>|tl)
(Intercept) 0.38890 0.29240 1.33 0.199
mydata$epd8 1.46902 0.01272 115.45 <2e-16 ***

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*' 0.05 ‘.’ 0.1 ‘ ' 1

Residual standard error: 0.877 on 19 degrees of freedom
Multiple R-Squared: 0.9986, Adjusted R-squared: 0.9985
F—statistic: 1.333e+04 on 1 and 19 DF, p-value: < 2.2e-16

6. Linear model residuals T-Test for different physiographic regions:
> t.test(epresidatast$residmi,epresidatast$residnm)
Welch Two Sample t—test

data: epresidatast$residmi and epresidatast$residnm
t = 1.0715, df = 7.714, p-value = 0.3163
alternative hypothesis: true difference in means is not equal to
0
95 percent confidence interval:
—0.2222181 0.6033515
sample estimates:
mean of x mean of y
—0.05689679 -0.24746349

> t.test(epresidatast$residmi,epresidatast$residva)

Welch Two Sample t-test

186

data: epresidatast$residmi and epresidatast$residva
t = -0.6661, df = 6.167, p-value = 0.5294
alternative hypothesis: true difference in means is not equal to
0
95 percent confidence interval:
-1.6796426 0.9571285
sample estimates:
mean of x mean of y
-0.05689679 0.30436027

> t.test(epresidatast$residnm,epresidatast$residva)
Welch Two Sample t—test
data: epresidatast$residnm and epresidatast$residva
t = -O.979, df = 7.132, p—value = 0.3596
alternative hypothesis: true difference in means is not equal to
0
95 percent confidence interval:
-1.879661 0.776013
sample estimates:

mean of x mean of y
—0.2474635 0.3043603

7. Paired T-Test between FlowNet and D8 wetland ﬂow accumulation samples for seven
watersheds in Michigan:
H0: actual difference in means between wetland ﬂow accumulation samples from
FlowNet method and wetland ﬂow accumulation samples from D8 method is equal to 0.
HA: actual difference in means between wetland ﬂow accumulation samples from
F lowNet method and wetland accumulation samples from D8 method is greater than 0.

>
t.test(mw1wgst$MW1FNWGNS,mw1 wgst$MW1D8WGNS,altemative=c("greater"),pair

ed=T)
Paired t-test

data: mwlwgst$MW1FNWGNS and mwlwgst$MW1D8WGNS
t = 2.5866, df = 999, p-value = 0.004917

alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

187

 

661.6799 Inf
sample estimates:

mean of the differences
1820.378

>
t.test(mw2wgst$MW2FNWGNS,mw2wgst$MW2D8WGNS,alternative=c("greater"),pair
ed=T)

Paired t-test

data: mw2wgst$MW2FNWGNS and mw2wgst$MW2D8WGNS
t = 3.8868, df = 999, p-value = 5.413e-05
alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

1253.360 Inf
sample estimates:

mean of the differences
2174.3 81

>
t.test(mw3wgst$MW3FNWGNS,mw3wgst$MW3D8WGNS,alternative=c("greater"),pair
ed=T)

Paired t-test

data: mw3Wgst$MW3FNWGNS and mw3wgst$MW3D8WGNS
t = 4.0211, df= 999, p-value = 3.114e-05

alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

2044.606 Inf

sample estimates:

mean of the differences
3462.1 11

>
t.test(mw4wgst$MW4FNWGNS,mw4wgst$MW4D8WGNS,altemative=c("greater"),pair
ed=T)

Paired t-test

data: mw4wgst$MW4FNWGNS and mw4wgst$MW4D8WGNS
t = 2.7332, df = 999, p-value = 0.003192
alternative hypothesis: true difference in means is greater than 0

95 percent conﬁdence interval:
1301.61 Inf

188

sample estimates:
mean of the differences
3273.375

>
t.test(mw5wgst$MW5FNWGNS,mw5wgst$MW5D8WGNS,altemative=c("greater"),pair
ed=T)

Paired t-test

data: mw5wgst$MW5FNWGNS and mw5wgst$MW5D8WGNS
t = 3.7341, df = 999, p-value = 9.951e-05
alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

1009.417 Inf
sample estimates:

mean of the differences
1805.439

>
t.test(mw6wgst$MW6FNWGNS,mw6wgst$MW6D8WGNS,altemative=c("greater"),pair
ed=T)

Paired t-test

data: mw6wgst$MW6FNWGNS and mw6wgst$MW6D8WGNS
t = 4.0705, df = 999, p-value = 2.531e-05
alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

1088.810 Inf
sample estimates:

mean of the differences
1 828.284

>
t.test(mw7wgst$MW7FNWGNS,mw7wgst$MW7D8WGNS,alternative=c("greater"),pair
ed=T)

Paired t-test

data: mw7wgst$MW7FNWGNS and mw7wgst$MW7D8WGNS
t = 2.9332, df = 999, p-value = 0.001716

alternative hypothesis: true difference in means is greater than 0
95 percent conﬁdence interval:

340.5154 Inf

sample estimates:

189

mean of the differences
776.181

8. Paired T-Test between FlowNet and D8 non-wetland ﬂow accumulation samples for
seven watersheds in Michigan:
H0: actual difference in means between non-wetland ﬂow accumulation samples from
F lowNet method and non-wetland ﬂow accumulation samples from D8 method is equal
to 0.
HA: actual difference in means between non-wetland ﬂow accumulation samples from
F lowNet method and non-wetland accumulation samples from D8 method is not equal to
0.
> t.test(mw1nwgst$MW1 FNNWGNS,mw1nwgst$MWl D8NWGNS,paired=T)

Paired t-test

data: mwlnwgst$MW1FNNWGNS and mwlnwgst$MW1D8NWGNS
t=1.4731, df= 999, p-value = 0.1411

alternative hypothesis: true difference in means is not equal to 0

95 percent conﬁdence interval:

-99.18173 696.37664

sample estimates:

mean of the differences
298.5975

> t.test(mw2nwgst$MW2FNNWGNS,mw2nwgst$MW2D8NWGNS,paired=T)
Paired t-test

data: mw2nwgst$MW2FNNWGNS and mw2nwgst$MW2D8NWGNS
t = 1.1765, df = 999, p-value = 0.2397

alternative hypothesis: true difference in means is not equal to 0

95 percent conﬁdence interval:

-112.9532 451.1824

sample estimates:

mean of the differences

190

169. 1 146
> t.test(mw3nwgst$MW3FNNWGNS,mw3nwgst$MW3D8NWGNS,paired=T)
Paired t-test

data: mw3nwgst$MW3FNNWGNS and mw3nwgst$MW3D8NWGNS
t = 2.2146, df = 999, p-value = 0.02701
alternative hypothesis: true difference in means is not equal to 0
95 percent conﬁdence interval:
31.40946 520.03519
sample estimates:

mean of the differences
275.7223

> t.test(mw4nwgst$MW4FNNWGNS,mw4nwgst$MW4D8NWGNS,paired=T)
Paired t-test

data: mw4nwgst$MW4FNNWGNS and mw4nwgst$MW4D8NWGNS
t = 1.4478, df = 999, p-value = 0.148

alternative hypothesis: true difference in means is not equal to 0

95 percent conﬁdence interval:

-92.74754 614.68799

sample estimates:

mean of the differences
260.9702

> t.test(mw5nwgst$MW5FNNWGNS,mw5nwgst$MW5D8NWGNS,paired=T)
Paired t-test

data: mw5nwgst$MW5FNNWGNS and mw5nwgst$MW5D8NWGNS
t = 1.8247, df = 999, p-value = 0.06834
alternative hypothesis: true difference in means is not equal to 0
95 percent conﬁdence interval:
-5.386305 148.204066
sample estimates:

mean of the differences
71.40888

> t.test(mw6nwgst$MW6FNNWGNS,mw6nwgst$MW6D8NWGNS,paired=T)
Paired t-test

data: mw6nwgst$MW6FNNWGNS and mw6nwgst$MW6D8NWGNS

191

t = 2.2213, df = 999, p-value = 0.02656
alternative hypothesis: true difference in means is not equal to 0
95 percent conﬁdence interval:
10.28564 166.17910
sample estimates:

mean of the differences
88.23237

> t.test(mw7nwgst$MW7FNNWGNS,mw7nwgst$MW7D8NWGNS,paired=T)
Paired t-test

data: mw7nwgst$MW7FNNWGNS and mw7nwgst$MW7D8NWGNS
t = 1.8013, df= 999, p-value = 0.07196

alternative hypothesis: true difference in means is not equal to 0

95 percent conﬁdence interval:

-15.94429 372.58035

sample estimates:

mean of the differences
178.3180

192

Bibliography

Ackerman, CT. 2002. An extension for support of HEC-RAS using ArcView GIS. US.
Army Corps of Engineers, Hydrologic Engineering Center. HEC. 609 Second St. Davis
CA 95616-4687.

Acreman, M., 2005. Linking science and decision-making: Features and experience from
environmental river ﬂow setting. Environmental Modelling & Software 20 (2), 99-109.

Adams, 8., Marc Berg, 2004. The nature of the Net: constructing reliability of health
information on the Web. Journal: Information Technology & People. 17 (2), 150-170.

Agarwal, P. 2005. Ontological considerations in GIScience, International Journal of
Geographical Information Science, 19(5):501-536.

Akrich, M., 1992, The description of technical objects. Shaping Technology/Building
Society: Studies in Sociotechnological Change, edited by W. E. Bijker and J. Law
(Cambridge: The MIT Press), pp. 205—224.

Allen, C., Catherine Creary, 2003. Introduction to Relational Database and SQL
Programming. McGRAW-HILL, 2003

Al-Sabhan W., M. Mulligan and G. A. Blackburn, 2003. A real-time hydrological model
for ﬂood prediction using GIS and the WWW. Computers, Environment and Urban
Systems, 27(1):9-32.

Angima, S.D., Stott, D.E., O'Neill, M.K., Ong, C.K., Weesies, G.A., 2003. Soil erosion
prediction using RUSLE for central Kenyan highland conditions. Agriculture,
Ecosystems and Environment 97, 295-308.

Araujo, E., Walfredo Cirne, Gustavo Wagner, Nigini Oliveira, Enio P. Souza, Carlos 0.
Galvo, Eduardo Sévio Martins, 2005. The SegHidro Experience: Using the Grid to
Empower a Hydro-Meteorological Scientiﬁc Network, e-science, pp. 64-71, First
International Conference on e-Science and Grid Computing (e-Science'05), 2005

Armstrong, MP. 1994. High performance computing and GIS. Proceedings of GIS/LIS’
94, Bethesda, MD: American Congress on Surveying and Mapping, pp. 4-13.

Armstrong, M. P., Densham, P. J. 1990. Database Organization Strategies for Spatial
Decision Support Systems. International Journal of Geographical Information Systems,
Vol. 4(1), pp. 3-20.

Armstrong, M.P., Wang, S., and Cowles, M. 2005. Using a computational grid for
geographic information analysis. The Professional Geographer 57(3): 365-375.

193

Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams. (1998). Large Area
Hydrologic Modeling and Assessment Part 1: Model Development. Journal of the
American Water Resources Association 34: 73-89.

Arrowsmith, J .R., G. R. Keller, J. Conner, J. Yoburn, L. Prashad, C. Eisinger, W.L.
Stefanov, M.J. Fouch, 2004. GEON: ArcIMS Online Mapping to Facilitate Integration of
Geoscience Data. Abstract submitted to Twenty-fourth Annual ESRI International User
Conference, August 9-13, 2004, San Diego, CA

Arunatileka, Shiromi; Ginige, Athula; Hol, Ana; and Lawson, Robyn, 2005. A Pilot
Project on eCollaboration in the Australian Toolmaking Industry 2005. BLED 2005
Proceedings. Paper 4. http://ais.bepress.com/bled2005/4

Attiya, H., Welch, J. 2004. Distributed Computing: Fundamentals, Simulations and
Advanced Topics. Published by: Wiley-Interscience. 2004

Baker, M., A. Apon, C. Femer, and J. Brown, Emerging Grid Standards, page.43-50,
IEEE Computer, April 2005.

Bandura, A. 1977. Social learning theory. Prentice-Hall, Englewood Cliffs, New Jersey,
USA.

Barclay, T., Gray, J. and Slutz, D., Microsoﬁ TerraServer: a spatial data warehouse. in
Proceedings of the 2000 ACM SIGMOD on Management of data, (Dallas, TX, 2000),
ACM, 307-318. http://citeseer.ist.psu.edu/barclay99microsoft.html [Last checked 8/5/08]

Basnyat, B., L.D. Teeter, B.G. Lockaby, and K.M. Flynn. 2000. The use of remote
sensing and GIS in watershed level analysis of non-point source pollution problems.
Forest Ecology and Management, Vol. 128, pp 65-73.

Batty, M. 1999. New technology and GIS, in Longley, P.A., Goodchild, M.F., Maguire,
D.J., and Rhind, D.W. (eds.), Geographic Information Systems, New York: Wiley, 309-
316.

Bellamy J .A., D. Lowes, A.J. Ash, J.G. Mclvor, and ND. MacLeod. 1996. A decision
support approach to sustainable grazing management for spatially heterogeneous
rangeland paddocks. Rangeland Journal, Vol. 18(2), pp. 370-391.

Bellamy J .A. and D. Lowes. 1999. Modeling change in state of complex ecological
systems in space and time: An application to sustainable grazing management.

Environmental International, Vol. 25(6/7), pp. 701-712.

Benner J, 2001. A grid of supercomputers, Wired News, August 9, 2001.
http://www.wired.com/science/discoveries/news/2001/08/45977 [Last checked 8/5/08]

194

Bemholdt, D., Shishir Bharathi, David Brown, Kasidit Chanchio, Meili Chen, Ann
Chervenak, Luca Cinquini, Bob Drach, Ian Foster, Peter Fox, Jose Garcia, Carl
Kesselman, Rob Markel, Don Middleton, Veronika Nefedova, Line Pouchard, Arie
Shoshani, Alex Sim, Gary Strand, and Dean Williams, 2005. The Earth System Grid:
Supporting the Next Generation of Climate Modeling Research, Proc. IEEE 93, 485
(2005), invited article.

Beven, K., 2003. On environmental models of everywhere on the GRID, Hydrological
Processes, V 17, N 1 (p 171-174)

Bhaskar, N.R., Wesley, P.J. and Devulapalli, RS, 1992. Hydrologic parameter
estimation using geographic information systems. Journal of Water Resources Planning
and Management 118, pp. 492—5 12

Bhatia, K., A. Memon, I. Zaslavsky, D. Seber, C. Baru, 2003. Creating Grid Services to
Enable Data Interaoperability: An Example from the GEON Project. Abstract submitted
to GSA Annual Meeting in Seattle, WA November 2-5, 2003

Bhatia, K., S. Chandra, K. Mueller, 2005. GAMA: Grid Account Management
Architecture. IEEE e-Science 2005

Birkin, M., McFarland, O., Dew, P., Hodrien, J. 2005. HYDRA: a prototype Grid-
enabled spatial decision support system. Published in the Proceedings of the 1st
International Conference on e-Social Science, Manchester, 22 - 24 June 2005.

Bishop, Bruce. 1970. Public Participation in Planning: A Multi-media Course. IWR
Report 70-7. Fort Belvoir, Va: US. Army Engineers Institute for Water Resources.

Born, S., and Genskow, K. D, 1999. Exploering the watershed approach: Critical
dimensions of state-local partnerships, the Four Corners Watershed Innovators Initiative,
ﬁnal report, River Network, Washington, d.C., 1999

Bowers, S., B. Ludaescher, 2003. Towards a Generic Framework for Semantic
Registration of Scientiﬁc Data. Semantic Web Technologies for Searching and
Retrieving Scientiﬁc Data (SCISW), Sanibel Island, Florida, 2003

Bowers, S., K. Lin, B. Ludaescher, 2004. On Integrating Scientiﬁc Resources through
Semantic Registration. 16th International Conference on Scientiﬁc and Statistical
Database Management (SSDBM 04), 21-23 June 2004, Santorini Island, Greece

Brodaric, B., M. Gahegan, 2006. Representing Geoscientiﬁc Knowledge in
Cyberinfrastructure: challenges, approaches and implementations. In: Sinha, A.K. (Ed),
Geolnforrnatics, Data to Knowledge, Geological Society of America Special Paper 397,
pp. 1-20. (2006)

195

Brooks K. N., P F Ffolliott, H M Gregerson, 1997. Hydrology and the management of
watersheds. Iowa State University Press, 1997

Browne, T. J. 1995. The role of geographical information systems in hydrology. In 1.
Foster, A. M. Gurnell, & G. E. Petts (Eds), Sediment and water quality in river
catchments (pp. 33—48). Chichester, UK: John Wiley & Sons.

Browning-Aiken, A., Richter, H., Goodrich, D., Strain, B., Varady, R., 2004. Upper San
Pedro Basin: Collaborative binational watershed managementlntemational Journal of
Water Resources and Development 20 (3), 353-367.

Browning-Aiken, A., Varady, R.G., Goodrich, D., Richter, H., Sprouse, T., Shuttleworth,
W.J., 2006. Integrating science and policy for water management: A case study of the
Upper San Pedro River Basin. In: Wallace, J .S., Wouters, P. (Ed8.), Hydrology and Water
Laderidging the Gap: A Case Study of HELP Basins. National Research Council
(NRC), 1999. New Strategies for America’s Watersheds. National Academy Press,
Washington, DC.

Bucaro, D, 2006. Personal Communication regarding Burns Ditch & Trail Creek Web-
based Sediment Modeling SDSS Workshop (David Bucaro is a program manager at
Army Corps of Engineers Chicago District).

Callahan, 3., Miles, E., Fluharty, D., 1999. Policy implications of climate forecasts for
water resources management in the Paciﬁc Northwest. Policy Science 32 (3), 269-293.

Callon, M., 1985, Some elements of a sociology of translation: Domestication of the
scallops and the Sherman of St. Brieuc Bay. In Power, Action and Belief, edited by J.
Law (London: Routledge & Kegan Paul).

Callon, M., 1991, Techno-economic networks and irreversibility. In A Sociology of
Monsters: Essays on Power, Technology and Domination edited by J. Law, Sociological
Review Monograph (London: Routledge), 38, 132—161.

Callon, M., Law, J ., and Rip, A., editors, 1996, Mapping the Dynamics of Science and
Technology: Sociology of Science in the Real World (London: Macmillan).

Carnegie Mellon Software Engineering Institute, 1997. Deﬁnition of middleware.
http://www.sei.cmu.edu/str/descriptions/middleware.html. last checked 09/ 14/08.

Cash, D., 2002. Salience, Credibility, Legitimacy and Boundaries Linking Research,
Assessment and Decision Making. Harvard University, John F. Kennedy School of
Government, Cambridge, MA.

Cash, D.W., Clark, W.C., Alcock, F ., Dickson, N.M., Eckley, N., Guston, D.H., Jager, J .,

Mitchell, RB, 2003. Knowledge systems for sustainable development. Proceedings of
the National Academy of Sciences USA 100 (14), 8086-8091.

196

Castelletti, A., Soncini-Sessa, R., 2006. A procedural approach to strengthening
integration and participation in water resource planning. Environmental Modelling &
Software 12 (10), 1455-1470.

Castelletti, A., Soncini-Sessa, R., 2007. Coupling real time control and socioeconomic
issues in participatory river basin planning. Environmental Modelling & Software 22 (8),
1114-1128.

Cate Jr, A, Semmens, D. J ., Guertin, D. P., Goodrich, D. C. 2007. DOTAGWA: a case
study in web-based architectures for connecting surface water models to spatially enabled
web applications. Proceedings of the 2007 summer computer simulation conference. San
Diego, California.

Chander S. 2005. Sustainable fresh water resources systems. In Proceedings of the
International Conference on hydrological perspectives for sustainable development, pp.
549—555 (vol. II), edited by M Perumal, D C Singhal, D S Arya, D K Srivastava, N K
Goel, B S Mathur, H J oshi, R Singh, M D Nautiyal. New Delhi: Allied Publishers Pvt.
Ltd. pp. 1099. Roorkee, 23-25 February 2005, organized by Department of Hydrology,
IIT

Chandra S., 2007. GEON Architecture II: Systems Components Overview. iGEON 2007
Workshop at the University of Hyderabad, India, August 3-8, 2007

Chandrasekaran, B., John R. Josephson, V. Richard Benjamins, 1999. What Are
Ontologies, and Why Do We Need Them?, IEEE Intelligent Systems, vol. 14, no. 1, pp.
20-26, Jan/Feb, 1999

Chariat, S., & Delleur, J. W. (1993). Integrated a physically based hydrological model
with GRASS. HydroGIS 93 (IAHS Publication, No. 211, pp. 143—150). Wallingford, CT:
International Association of Hydrological Sciences.

Chen, CW, Joel Herr; and Laura Weintraub, 2004. Decision Support System for
Stakeholder Involvement. Journal of Environmental Engineering. ASCE / JUNE 2004

Cheng, Chun-Tian, K. W. Chau, Xiang-Yang Li, Gang Li, 2004. Developing a Web-
based ﬂood forecasting system for reservoirs with J2EE, Hydrological Sciences Journal
49(6) 973-986

Chi, L., Holsapple, C. W., 2005. Understanding computer-mediated interorganizational
collaboration: a model and framework. JOURNAL OF KNOWLEDGE
MANAGEMENT VOL. 9 NO. 1 2005, pp. 53-7.

Choi J .Y., B.A. Engel, L. Theller, and J. Harbor, 2005. Utilizing Web-based GIS and

SDSS for hydrological land use change impact assessment. Transactions of the ASAE,
48(2):815-822

197

Chorowicz J ., Ichoku C., Riazanoff S., Kim YJ., and Cervelle B. (1992). A Combined
Algorithm for Automated Drainage Network Extraction. Water Resources Research
28:1293—1302.

Chow, V.T., Maidment, DR, and Mays, L.W. (1988) Applied Hydrology, McGraw-Hill,
New York.

Clarke, Keith C., and Su J in Lee.(2007) Spatial Resolution and Algorithm Choice as
Modiﬁers of Down-Slope Flow Computed from Digital Elevation Models, Cartography
and Geographic Information Science,34(3): 215-230.

Climaco, J. , C.H. Antunes, and J .P. Costa. 1995. Decision aiding in an organizational
context. In: C. Zucchermaglio, S. Bagnara, and S.U. Stucky (eds) Organizational
Learning and Technological Change, Springer, Germany pp. 125-137.

Coleman, DJ. 1999. Geographical information systems in networked environments, in
Longley, P.A., Goodchild, M.F., Maguire, D.J., and Rhind, D.W. (eds.), Geographic
Information Systems, New York: Wiley, 317-329.

Costa-Cabral, M. and S. J. Burges, (1994), Digital Elevation Model Networks
(DEMON): A Model of Flow Over Hillslopes for Computation of Contributing and
Dispersal Areas, Water Resources Research, 30(6): 1681-1692.

Crawford and Linsley, Digital simulation in hydrology: Stanford watershed model IV,
Stanford Univ., Dept. Civ. Eng. Tech. Rep. 39, 1966.

Crompvoets, J ., Bregt, A., Rajabifard, A., and Williamson, 1. 2004. Assessing the
worldwide developments of national spatial data clearinghouses, International Journal of
Geographical Information Science, 18(7): 665-689.

Crosby, C.J., Arrowsmith, J .R., 2004. LiDAR Data Distribution, Interpolation and
Analysis on the GEON Grid - A Conceptual Framework. Geological Society of America
Annual meeting, Paper No. 60-6, 2004

Crosby, C.J., J. Conner, E. Frank, J R. Arrowsmith, A. Memon, V. Nandigam, G.
Wurman, C. Baru, 2006. The GEON LiDAR Workﬂow: An Intemet-Based Tool for the
Distribution and Processing of LiDAR Point Cloud Data. Geoinforrnatics 2006, Reston,
VA May 11, 2006.

CTIC 1995, Managing conﬂict, a guide for watershed partnerships, CTIC, West
Lafayette, IN, 1995.

CTIC 1995. Putting together a watershed plan, a guide for watershed partnerships, ctic,
west lafayette, in, 1995.

Dangermond, Jack 2007 GeoWeb 2007 Keynote Speech

198

Davenport, T. and Wilson, D., 2001. Local solutions to regional problems: the watershed
approach. Presented at Earth Day Symp., St. Louis, April 26, 2001

Davenport, T., 2002. The Watershed Project Management Guide. Lewis Publishers, 2002

Dempsey, Bert J. 2000. Prospects for the global Internet: new techniques for delivering
rich digital collections to users world-wide, The Electronic Library Vol 18:4

Densham, P. J., Goodchild, M. F. 1989. Spatial Decision Support System: A Research
Agenda. Proceedings of GIS/LIS ’89, Florida, pp. 707-716.

Densham, P.J. 1991. Spatial decision support system, geographical information system.
In: Maguire, D.J., Goodchild, M.S., Rhind, D.W. (Eds), Principles and Applications.
Longrnan, London. pp. 403-412.

De Roo, A.P. Hazelhoff, L. and Burrough, PA. 1989. Soil erosion modelling using
ANSWERS and geographical information systems. Earth Surface Processes and
Landforms Vol. 14, 517-532.

Desmet, P.J. and G. Govers, (1996a). A GIS-procedure for the Automated Calculation of
the USLE LS-factor on Topographically Complex Landscape Units. Journal of Soil and
Water Conservation 51 (1996), pp. 427—433.

Dilling, L., 2007. Towards science in support of decision making: Characterizing the
supply of carbon cycle science. Environmental Science and Policy 10, 48-61.

DiLuzio, M., R. Srinivasan, and J. Arnold. 2001. Arcview Interface for SWAT2000.
User’s Guide. Temple, Texas: United States Department of Agriculture — Agriculture
Research Service. Available at: www.brc.tamus.edu/swat/soﬁ_model.html [Last Checked
08/05/08].

Djokic, D. 1996. Towards general purpose spatial decision support system using existing
technologies. In: GIS and Environmental Modeling: Progress and Research Issues. GIS
World Books, Ft. Collins, CO.

Doppelt, B. 2000, Emerging approaches to watershed governance. New approaches to
guidesustainable ecological, economic, and organizational performance within the
watershed context. Working draft, 4/1/2000.

Dowman, I., 2005. High resolution remotely sensed data and spatial data infrastructure

development. Eighth United Nations Regional Cartographic Conference for the Americas
New York, 27 June-1 July 2005.

199

Duke G. D., Kienzle S. W., Johnson D. L., Byrne J. M. (2003): Improving Overland Flow
Routing by Incorporating Ancillary Road Data into Digital Elevation Models. Journal of
Spatial Hydrology 3 (2003)

Duke G. D., Kienzle S. W., Johnson D. L., Byrne J. M. (2006): Incorporating Ancillary
Data to Reﬁne Anthropogenically Modiﬁed Overland Flow Paths. Hydrological
Processes 20 (2006) 1827-1843

Dunne, K., Jim Browne, 2005. Collaboration Initiation Practices — Experiences from
Case Studies. 24 November 2005. ICE Conference Proceedings. http://www.ve-
forum.org/proj ects/408/ice2004/ce%20methods%20and%20tools/5 1_collaboration%20in
itiation%20practices%20—%20experiences%20from%20case%20studies.pdf

Dymond,R. L., B. Regmi, V. K. Lohani, and R. Dietz, 2004. Interdisciplinary Web-
Enabled Spatial Decision Support System for Watershed Management Water Resour.
Plng. and Mgmt., Volume 130, Issue 4, pp. 290-300.

eG: e-Govemment, http://www.whitehouse.gov/omb/egov/ last checked 09/14/2008.

El Haddi AA, Shekhar S, Hills R, and McManamon A, 1996 Mirror-image round robin
spatial data partitioning: a case study with parallel SEUS, Third International
Conference/Workshop on Integrating GIS and Environmental Modeling, January 21-25,
1996, Santa Fe, New Mexico, USA.

Engel, B.A., J in-Yong Choi, J. Harbor and S. Pandey, 2003. Web-based DSS for
hydrologic impact evaluation of small watershed land use changes. Computers and
Electronics in Agriculture, 39(3):241-249

EPA, 2008 Office of Environmental Information
http://www.epa.gov/oei/basicinformation.htm

Fairﬁeld, J. and P. Leymarie, (1991), Drainage Networks from Grid Digital Elevation
Models, Water Resources Research, 27(5): 709-717.

Falkenmark, M., Gottschalk, L., Lundqvist, J. & P. Wouters 2002. Towards integrated
catchment management - Increasing the dialogue between scientists, policy makers and
stakeholders. Background document of HELP Symposium, Kalmar, Sweden, August 18-
22.

Faraj, S., Dowan Kwon, Stephanie Watts, 2004. Contested artifact: technology
sensemaking, actor networks, and the shaping of the Web browser. Journal: Information
Technology & People. 17 (2), 186 - 209.

F edra, K. 1991. A computer-based approach to environmental impact assessment. In:
Proceedings of the Indicators and Indices for Environmental Impact Assessment and Risk
Analysis. Joint Research Centre, Ispra, Italy. pp. 11-40.

200

Fernandez, C., Wu, J .Q., McCool, D.K., Stckle, CO, 2003. Estimating water erosion and
sediment yield with GIS, RUSLE, and SEDD. Journal of Soil and Water Conservation
58, 128-136.

Fonseca F.T., Egenhofer M.J., Agouris P., and Camara G., 2002. Using Ontologies for
Integrated Geographic Information Systems, Transactions in GIS, 6(3):231-257.

F onseca, F .T., Davis, C. and Carnara, G. 2003. Bridging Ontologies and Conceptual
Schemas in Geographic Information Integration, Geolnformatica, 7(4): 355-378.

Fortin, Jean-Pierre, Richard Turcotte, Serge Massicotte, Roger Moussa, Josée F itzback,
and Jean-Pierre Villeneuve, 2001. A Distributed Watershed Model Compatible with
Remote Sensing and GIS Data. 1: Description of Model. Journal of Hydrologic
Engineering, Vol. 6, No. 2, March/Apri12001, pp. 91-99.

Foster 1, 2002. The Grid: a new infrastructure for let Century Science. Physics Today
February 2002, 42-47

Foster, 1. 2003. The Grid: Computing Without Bounds, Scientiﬁc American, 288, 4, 78-
85.

Foster, 1. 2005. Service-Oriented Science, Science 308, 814 (2005)

Foster 1, 2005. Globus Toolkit Version 4: Software for Service-Oriented Systems. IFIP
International Conference on Network and Parallel Computing, Springer-Verlag LNCS
3779, pp 2-13, 2005.

Foster, 1. and Kesselman, C. eds. 2003. The Grid 2: Blueprint for a New Computing
Inﬁ'astructure, Published by Morgan Kaufrnann, 2003.

Foster, 1., H. Kishimoto, A. Savva, D. Berry, A. Djaoui, A. Grirnshaw, B. Horn, F.
Maciel, F. Siebenlist, R. Subrarnaniam, J. Treadwell and J. Von Reich, 2005. The Open
Grid Services Architecture, Version 1.0, http://forge.gridforum.org/projects/ogsa-wg,
January 2005.

Foster 1, Tuecke S (2005) Describing the elephant: the different faces of IT as services.
ACM Queue. 3(6):26—29.

Foster 1, 2005. Service-Oriented Science Science 6 May 2005: Vol. 308. no. 5723, pp.
814 — 817.

Fraser, RH. 1999. SEDMOD: a GIS-based delivery model for diffuse source pollutants.
New Haven, Conn., Yale University, Ph.D. dissertation, p. 7 6-99.

201

Fraser, R.H., P.K. Barten, and P.K. Tomlin. 1996. SEDMOD: A GIS-based method for
estimating distributed sediment delivery ratios. In: Hallam, C.A., J .M. Salisbury, K.J.
Lanfear, and BA. Battaglin (Eds), Proceedings of the AWRA Annual Symposium, GIS
and Water Resources. American Water Resources Association, Hemdon, Virginia, TPS-
96-3, 482 pp.

Fraser, R.H., P.K. Barten, and D.A.K. Pinney. 1998. Predicting stream pathogen loading
from livestock using geographical information system-based delivery model. Journal of
Environmental Quality, Vol. 27, pp 935-945.

Freeman, T. G., (1991), Calculating Catchment Area with Divergent Flow Based on a
Regular Grid, Computers & Geosciences, 17(3): 413-422.

Fu, G., Chen, S., McCool, D.K., 2006. Modeling the impacts of no-till practice on soil
erosion and sediment yield with RUSLE, SEDD, and ArcView GIS. Soil and Tillage
Research 85, 38-49.

Gaddis, E.J.B., Vladich, H., Voinov, A., 2007. Participatory modeling and the dilemma
of diffuse nitrogen management in a residential watershed. Environmental Modelling &
Software 22 (5), 619-629.

Galitz,Wilbert O. 2007. The Essential Guide to User Interface Design: An Introduction to
GUI Design Principles and Techniques, Wiley, John & Sons, Incorporated. 2007.

Google, 2008. Google Earth. http://earth.google.com/
Google, 2008. Google Maps. http://maps.google.com/maps

GOS: GeoSpatial One-Stop: http://gos2.geodata.gov/wps/portal/gos last checked
09/ 14/2008.

Gray, B., 1989. Collaborating: Finding Common Ground for Multiparty Problems,
Jossey-Bass Publishers, San Francisco.

Green, C. 2003, The Water Framework: a poisoned chalice? Paper given at the Annual
Conference of the Royal Geographical Society/Institute of British Geographers, London,
september 2003.

Gregersen, H., Ffolliott, P., Brookes, K., 2008. Integrated Watershed Management:
Connecting People to their Land and Water. Cabi Publishing. 2008.

Gregory, R. 2000, Using stakeholder values to make smarter environmental decisions.
Environment. (June 2000, ﬁndarticles.com).

202

Gruber, T. 1991. The Role of Common Ontology in Achieving Sharable, Reusable
Knowledge Bases, in Principles of Knowledge Representation and Reasoning,
Cambridge, MA, pp. 601-602.

Guertin, D.P., J .D. Womack, R. MacArthur, G.B. Ruyle. 1998. Geographic information
system based tool for integrated allotment and watershed management. In: D.F. Potts
(ed.) Proceedings of AWRA Specialty Conference, Rangeland Management and Water
Resources, American Water Resources Association, Hemdon, Virginia, TPS-98-1, pp.35-
44.

Harvey, F., W. Kuhn, Pundt, H., Bishr, Y. and Riedemann, C. 1999. Semantic
Interoperability: A Central Issue for Sharing Geographic Information. Annals of Regional
Science 33(2): 213—232.

Hawth Analysis Tools website, 2008.
http://www.spatialecology.com/htools/downloadphp last checked 12/08/2008.

Healey, R., Dowers, S., Gittings, B., and Mineter, M. (eds.) 1998 Parallel processing
algorithms for GIS, London: Taylor & Francis.

Heathcote I. 1998. Integrated watershed management: principles and practice. Wiley.
1998.

Hecht L (2001) OGC aims to enable information communities, GEO World, August 2001.
http://www.geoplace.com/gw/2001/0801/0801opng.asp. Last checked 3/10/02.

He, C. 2003. Integration of geographic information systems and simulation model for
watershed management. Environmental Modeling & Software. Volume 18, Issues 8-9,
October-November 2003, Pages 809-813

He, C., C. Shi, C. Yang, and HP Agosti. 2001. A Windows based GIS-AGNPS Interface.
Journal of the American Water Resources Association. Vol. 37(2), 395-406.

Holdren, C., Jones, W., and Taggart, J ., 2001. Managing Lakes and Reservoirs, N. Am.
Lake Manage. Soc. and Terrene Inst, Madison, WI, 2001.

Holmgren, P., (1994), Multiple Flow Direction Algorithms for Runoff Modelling in Grid
based Elevation Models: An Empirical Evaluation, Hydrological Processes, 8: 327—334.

Hluchy, L., Jan Astalos, Miroslav Dobrucky, Ondrej Habala, Branislav Simo and Viet D.
Tran, 2004. Flood Forecasting in a Grid Computing Environment, In Parallel Processing
and Applied Mathematics.

Hluchy, L., Ondrej Habala, Martin Maliska, Branislav Simo, Viet Tran, Jan Astalos,
Marian Babik, 2006. Grid Based Flood Prediction Virtual Organization, e-science, p. 4,

203

Second IEEE International Conference on e-Science and Grid Computing (e-Science'06),
2006.

Hoel EG, and Samet H, 1994. Performance of data-parallel spatial operations,
Proceedings 20th Intl. Conf. On Very Large Data Bases, September 1994, Santiago,
Chile, 156-167.

Holsapple, CW. and AB. Whinston. 2001. Decision support systems, A knowledge-
based approach. Thomson Learning Custom Publishing, Cincinnati, OH. 2001.

Hunsaker C, Ehlschlaeger CR, Davis F W, and Goodchild MF 1996 Estimating spatial
uncertainty as a function of scale: implications for landscape ecology, Third International
Conference/Workshop on Integrating GIS and Environmental Modeling, January 21-25,
1996, Santa Fe, New Mexico, USA.

Jacob, B., Brown, M., Fukui, K., Trivedi, N., 2005. Introduction to Grid Computing.
IBM. 2005

Jacobs, K., 2002. Connecting Science, Policy, and Decisionmaking: A Handbook for
Researchers and Science Agencies. National Oceanic and Atmospheric Administration.
Ofﬁce of Global Programs.

Jaeger-Frank, E., C.J. Crosby, A. Memon, V. Nandigam, J .R. Arrowsmith, J. Conner, I.
Altintas, C. Baru, 2006. A Three Tier Architecture for LiDAR Interpolation and
Analysis. Lecture Notes in Computer Science, 3993, Pages 920-927, 2006.

J agers H, Jansen W, Steenbakkers W., 1998. Characteristics of Virtual Organizations.
Organizational Virtualness: Proceedings of the VoNet Workshop, April 27-28, 1998,
Bern, Switzerland 65-76. http://www.virtual-organization.net/ﬁles/articles/an-l.pdf, last
checked 3/10/02.

Jarupathirun S. and FM. Zahedi, 2007. Exploring the inﬂuence of perceptual factors in
the success of web-based spatial DSS. Decision Support Systems, 43(3):933-951

J oia, Luiz and Neto, Aristides, 2004. Govemment-To-Govemment Enterprises In Brazil:
Key Success Factors Drawn From Two Case Studies 2004. BLED 2004 Proceedings.
Paper 20. http:l/ais.bepress.com/bled2004/20

Joseph, J. , M. Ernest, and C. F ellenstein, Evolution of grid computing architecture and
grid adoption models, IBM Systems Journal Vol 43, No 4, 2004.

Kebede, E., J ianbang Gan, and Zewdu Gebeyehu, 2005. Non-point source pollution and

land use pattern linkage: a watershed approach: An article from: Journal of the Alabama
Academy of Science (Digital - Aug 1, 2005)

204

Keller, R., D. Seber, A.K. Sinha, C. Baru. 2005. The Geosciences Network (GEON): one
step towards building cyberinfrastructure for the geosciences. European Geophysical
Union, Vienna

Keller, R., T.J. Owens, 2003. GEON (GEOscience Network) - A First Step In Creating
Cyberinfrastructure For The Geosceinces. Electronic Seismologist, July/August 2003,
Vol. 74, No. 4

Kendall, W., Markus Glatter, Jian Huang, Forrest Hoffman, and David E. Bemholdt,
2008, Web Enabled Collaborative Climate Visualization in the Earth System Grid, in The
2008 International Symposimn on Collaborative Technologies and Systems (CTS 2008),
2008, invited paper.

Koudstaal, Rob, Frank R. Rijsberman, and hubert Savenije. 1992. Water and sustainable
development. Natural Resources Forum 16(4): 277-290.

Lahlou, N., Shoemaker, L., Choudhury, 8., Elmer, R., Hu, A., Manguerra, H., Parker, A.
1998. BASINS V.2.0 User’s Manual. US Environmental Protection Agency Office of
Water, Washington DC (EPA-823-B-98-006).

Lane, L.J., Nichols, M.H., Levick, and LR. 1999. Information systems technology for
assessment of rangeland ecosystems. Proc. Intemat'l. Conf. on Water, Environment,
Ecology, Socio-economics and Health Engineering, Environmental Modeling, V.P. Singh,
Seo, W., and Sonu, J .H. (eds.), October, 1999, Seoul, Korea, pp. 329-340.

Latour, B., 1987, Science In Action (Cambridge: Harvard University Press).

Latour, B., 1991 , Technology is society made durable. A Sociology of Monsters: Essays
on Power, Technology and Domination, edited by J. Law (London: Routledge), pp. 103—
131.

Latour, B., 1992, Where are the missing masses? The sociology of a few mundane
artefacts. In Shaping Technology/Building Society: Studies in Sociotechnological
Change, edited by W. E. Bijker and J. Law (Cambridge: The MIT Press), pp. 205-224.

Latour, B., 1997, On actor-network theory: A few clariﬁcations. URL http:
//www.keele.ac.uk/ depts/stt/stt/ant/latour.htrn

Latour, B., 1999, On recalling ANT. Actor Network and After, edited by J. Law and J.
Hassard (Oxford: Blackwell and the Sociological Review), pp. 15—25.

Latour, B., 2005. Reassembling the Social: An Introduction to Actor-Network-Theory.
Oxford University Press. 2005.

Law, J ., 1992, Notes on the theory of the actor-network: ordering, strategy and
heterogeneity. Systems Practice, 5, 379—393.

205

Lawler, P., & Howell-barber, H., 2007. Service-Oriented Architecture: SOA Strategy,
Methodology, and Technology. Published by: CRC Press, 2007.

Lea, N. L., (1992), An Aspect Driven Kinematic Routing Algorithm, in Overland Flow:
Hydraulics and Erosion Mechanics, Edited by A. J. Parsons and A. D. Abrahams,
Chapman & Hall, New York.

Lee, Terence. 1992. Water management since the adoption of the Mar del Plata Action
Plan: Lessons for the 19903. Natural Resources Forum 16(3): 202-211.

Lee, K.N., 1993. Compass and Gyroscope: Integrating Science and Politics for the
Environment. Island Press, Washington, DC.

Lei, 2., D. Huang, et al. 2006. Leveraging Grid Technologies For Reservoir Uncertainty
Analysis. Proceedings of High Performance Computing Symposium (HPC06),
Huntsville, Alabama, April 2--6, 2006.

Leipnik, M. R., Kemp, K. K., Loaiciga, H. A., 1993. Implementation of GIS for Water
Resources Planning and Management. Journal of Water Resource Planning and
Management, Vol.119(2), pp. 184-205.

Lin, K., B. Ludaescher, 2004. GEON: Ontology-Enabled Map Integration. Abstract
submitted to Twenty-fourth Annual ESRI International User Conference, August 9-13,
2004, San Diego, CA

Liu, M., H. Zhang, Y. Yang, 2004. Integrating a community modeling environment in the
Geosciences cyberinfrastructure. American Geophysical Union annual meeting, San
Francisco, 12-10-2004

Logar, N.J., Conant, RT, 2007. Reconciling the supply of and demand for carbon cycle
science in the US. agriculture sector. Environmental Science & Policy 10, 75-84.

Lucero, D., D.J. Waterrnolen, and MD. Murrell. 2004. A State-Federal Partnership to
Understand the Usefulness of Decision Support Tools for Natural Resources
Management. "Watershed 2004," Water Environment Federation Annual Meeting. Hyatt
Regency, Dearbom, MI. (July 11-14, 2004).

Magillo P, and Puppo P 1998 Algorithms for parallel terrain modeling and visualization,
In: Healey R, Dowers S, Gittings B, and Mineter M (eds.) Parallel processing algorithms
for GIS, London: Taylor & Francis, 351-3 86.

Maidment, DR. 1993. Handbook of Hydrology. McGraw-Hill, Inc. New York, NY.

Maidment, D. R. 1996. GIS and hydrological modeling—An assessment of progress.
Paper presented at The Third International Conference on GIS and Environmental

206

Modeling, Santa Fe, NM. Available: maidment/GISHydro/meetings/santafe/santafe.htm.
[Last Checked 08/05/08].

Martin, Eugene W., 2000. Actor-networks and implementation: example from
conservation GIS in Ecuador, Int. J. GISci, 2000, Vol. 14, No.8, 715-738

Matti Hannus et al., 2005. Guidelines for Virtual Organisations. 04 November 2005
VOIG. http://www.ve-
forum.org/projects/280/industry%20guidelines/virtual%200rganisation%20practical%20g
uidelines.pdf

Matthies, M., Giupponi, C., Ostendorf, B., 2007. Environmental decision support
systems: Current issues, methods and tools. Environmental Modelling & Software 22 (2),
123-127.

Maurel, P., M. Craps, F. Cernesson, R. Raymond, P. Valkering, and N. Ferrand. 2007.
Concepts and methods for analysing the role of information and communication tools
(IC-tools) in social learning processes for river basin management. Environmental
Modelling and Software 22:630-639.

McKinney, DC. and X. Cai. 2002. Linking GIS and water resources management
models: an object-oriented method. Environmental Modeling & Software. Vol. 17, 413-
425.

McNie, EC, 2007. Reconciling the supply of scientiﬁc information with use demands:
an analysis of the problem and review of the literature. Environmental Science & Policy
10, 17-38.

Memon, A., C. Baru, I. Zaslavsky, S. Mock, A. Behere, 2004. GEON: Standards-based
Secure Invocation of ArcWeb Services. Abstract submitted to Twenty-fourth Annual
ESRI International User Conference, August 9-13, 2004, San Diego, CA.

MI CGI Website, 2008. http://www.michigan.gov/cgi last checked 12/08/2008.

Miller, S.N., R.C. Miller, D.J. Semmens, M. Hernandez, W.G Kepner, D. Ebert, W.P.
Miller, and DC. Goodrich. 2002. AGWA — Automated Geospatial Watershed Assessment:
A GIS-based hydrologic modeling tool. Application and User Manual. Available at
http:/lwww.tucson.ars.ag.gov/agwa. Accessed in June 2003.

Miller, S. N., Semmens, D. J ., Goodrich, D. C., Hernandez, M., Miller, R. C., Kepner, W.
G, Guertin, D. P. 2006. The Automated Geospatial Watershed Assessment Tool. J.
Environmental Modeling and Software. 22:365-377

Mitas, L., Mitasova, H., 1998, Distributed erosion modeling for effective erosion
prevention. Water Resources Research Vol. 34, No. 3, pp. 505-516.

207

Mitasova, H., J. Hoﬁerka, M. Zlocha, L.R. Iverson, 1996, Modeling topographic
potential for erosion and deposition using GIS. Int. Journal of Geographical Information
Science, 10(5), 629-641.

Molich, R. and J. Nielsen. 1990. Improving a humancomputer dialogue. Communications
of the ACM. Vol. 33(3). pp. 338 — 348.

Mueller, K., 2005. GAMA: Grid Account Management Architecture. GridSphere and
Portals Workshop, February 4-5 2005, Baton Rouge, LA.

National Research Council. 1999. New strategies for America’s watersheds. National
Academy Press, Washington DC.

Newman, S., T. Lynch, and AA. Plummer. 2000. Success and failure of decision support
systems: Learning as we go. Proc. Am. Soc. Anim. Sci., 1999. Available at:
http:/lwww.asas.org/jas/symposia/proceedings/0936.pdf. Accessed January 2002.

Newson, Malcolm. 1992. Water and sustainable development: The "turn-around decade"?
J. Envir. Planning and Management 25(2): 175-183.

Nickum, James E., and K. William Easter. 1990. Institutional arrangements for managing
water conﬂicts in lake basins. Natural Reources Forum 14(3): 210-220.

Nikolaidis, N.P., Heng, H., Semagin, R., Clausen, J .C., 1998. Nonlinear response of a
mixed land use watershed to nitrogen loading. Agricult. Ecosyst. Environ. 67, 251-265.

Novoty, V., 1980. Delivery of Suspended Sediment and Pollutants from Nonpoint
Sources During Overland Flow. Water Resources Bulletin 16(6): 1057-1065.

NSDI: National Spatial Data Inﬁastructure, http:/lwww.fgdc.gov/nsdi/nsdi.htrnl last
checked 09/14/2008.

NSDI future directions (2004) http://www.ucgis.org/EVENTs/NSDI_04l 52004.htm

Nyakatawa, E. Z., K. C. Reddy, J. L. Lemunyon, 2001. Predicting soil erosion in
conservation tillage cotton production systems using the revised universal soil loss
equation (RUSLE). Soil and Tillage Research, Volume 57, Issue 4, January 2001, Pages
213-224.

Nyakatawa, E. Z., V. Jakkula, K.C. Reddy, J .L. Lemunyon, B.E. Norris Jr., 2007. Soil
erosion estimation in conservation tillage systems with poultry litter application using
RUSLE 2.0 model. Soil and Tillage Research, Volume 94, Issue 2, June 2007, Pages 410-
419.

O'Callaghan, J. F. and D. M. Mark, (1984), "The Extraction of Drainage Networks From
Digital Elevation Data," Computer Vision, Graphics and Image Processing, 28: 328-344.

208

OGIS, 2002. http://www.opengis.org. Last checked 3/10/02.

Ouyang, D., and J. Bartholic, 1997. Predicting sediment delivery ratio in Saginaw Bay
watershed. In: Proceedings of the 22nd Annual Conference of the National Association of
Environmental Professionals. 19-23 May 1997.

Ouyang, D., and J. Bartholic. 2001. Web-based application for soil erosion prediction. An
International Symposium - Soil Erosion Research for the 2lst Century. Honolulu, HI.
January 3-5, 2001.

Pahl-Wostl, C., M. Craps, A. Dewulf, E. Mostert, D. Tabara, and T. Taillieu. 2007. Social
learning and water resources management. Ecology and Society 12(2): 5.

Palmer J W, and Speier C, 1997. A typology of virtual organizations: an empirical study,
Proceedings of the Association for Information Systems 1997 Americas conference,
August 15-17, 1997, Indianapolis, USA.
http://hsb.baylor.edu/rarnsower/ais.ac.97/papers/palm_spe.htm. last checked 3/10/02.

Parker, R, Letcher, R., Jakeman, A., et al., 2002. Progress in integrated assessment and
modelling. Environmental Modelling & Software 17, 209-217.

Parson, E., 1995. Integrated assessment and environmental policy making. Energy Policy
23 (4), 463-475.

Peng, Z.-R., and M.-H. Tsou, 2003, Internet GIS: Distributed Geographic Information
Services for the Internet and Wireless Networks. Published by John Wiley and Sons, Inc.
Hoboken, New Jersey. Page 5-33.

Prato, T. and S. Hajkowicz. 1999. Selection and sustainability of land and water resource
management systems. Journal of the American Water Resources Association, Vol. 35(4),
pp 739-752.

Purcell, P., 2006. Networked Neighbourhoods: The Connected Community in Context.
Published by: Springer 2006.

Quinn, P., K. Beven, P. Chevallier and O. Planchon, (1991), "The Prediction of Hillslope
Flow Paths for Distributed Hydrological Modeling Using Digital Terrain Models,"
Hydrological Processes, 5: 59-80.

R Development Core Team (2005). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-
07-0, URL http:/lwww.R-project.org [Last Checked 08/05/08].

Rao, M., Guoliang Fan, J. Thomas, G. Cherian, V. Chudiwale and M. Awawdeh, 2007. A
web-based GIS Decision Support System for managing and planning USDA's

209

Conservation Reserve Program (CRP). Environmental Modelling & Software,
22(9): 1270-1280.

Rattray, N. 2006. A User-Centered Model for Community-based Web-GIS. Urban and
Regional Information Systems Association (URISA), Web article:
http://www.urisa.org/publications/joumal/articles/a_user__centered_model [Last checked
8/5/08]

Rees, A., J. Alroy, C. Scotese, A. Memon, D.B. Rowley, J .T. Parrish, D.B. Weishampel,
E. Platon, M.A. O'Leary, M.A. Chandler, 2007. Phanerozoic Earth and Life: The
Paleointegration Project. Abstracts, GSA Geoinformatics Division, San Diego, CA, May
2007

Reichert, P., Borsuk, M., Hostrnann, M., Schweizer, S., Sporri, C., Tockner, K., Truffer,
B., 2007. Concepts of decision support for river rehabilitation.Environmental Modelling
& Software 22(2), 188-201.

Reimold, Robert J ., 1998. Watershed Management: Practice, Policies, and Coordination.
McGraw-Hill Professional, 1998.

Renard, K.G. and VA. F erreira. 1993. RUSLE model description and database
sensitivity. Journal of Environmental Quality, 22(3):458-466.

Renard, K. G., J. M. Laﬂen, G. R. Foster, and D. K. McCool. 1994. The Revised
Universal Soil Loss Equation. In: Lal, R. (ed). 1994. Soil Erosion Research Methods.
Ankeny, IA: Soil and Water Conservation Society.

Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.K., Poder, DC, 1997. Predicting
soil erosion by water—a guide to conservation planning with the Revised Universal Soil
Loss Equation (RUSLE). Agriculture Handbook: 703. USDA-Agricultural Research
Service, Washington DC. 384 pp.

Renfro, G.W.: 1975, Use of erosion equations and sediment delivery ratios for predicting
sediment yield, in “Sediment yield workshop: Present and prospective technology for
predicting sediment yield and sources” Proceedings USDA Sedim. Labor., USA,
Publication ARS-S-40, pp. 33-45.

Rewerts CC. and Engel B.A., 1991, ANSWERS on GRASS: Integrating a watershed.
simulation with a GIS. ASAE Paper No.91-2621.

Riedemann, C. and W. Kuhn, 1999. What Are Sports Grounds? Or: Why Semantics
Requires Interoperability. In: A. Vckovski, K. Brassel, and H.-J. Schek (eds.)
International Conference on Interoperating Geographic Information Systems, Berlin,
Springer. Lecture Notes in Computer Science 1580: 217—229.

210

Ritter, William F ., Shirmohammadi, Adel. 2000. Agricultural Nonpoint Source Pollution:
Watershed Management and Hydrology. Published by: CRC Press.

Rob, P., Carlos Coronel, 2006. Database Systems: Design, Implementation, and
Management. Course Technology, 2006.

Roehl, J .E.,l962. Sediment source areas, delivery ratios and inﬂuencing morphological
factors, Int. Ass. Sci. Hydro]. Pub]. 59, 202-213, 1962.

Saghaﬁan, B., 1993, Implementation of a distributed hydrologic model within
Geographic Resources Analysis Support System (GRASS), Proceedings of the Second
International Conference on Integrating Environmental Models and GIS, Breckenridge,
CO.

Salayandia, L., Y. Huang, A.Q. Gates, S. Roach, 2006. GeoNet: Use of Grid
Technologies in Geoinforrnatics for the Transition Zone between the Colorado Plateau
and the Basin and Range Province. in, Sinha, A. K. (ed), Geoinfonnatics: Data to
Knowledge: Geological Society of America, Special Paper 397, p. 183-194, 2006.

Sankar, 08., RN. Ford, and M. Bauer. 1995. Decision support systems. Vol. 13. pp. 93-
104.

Sarewitz, D., Pielke, R.A., 2007. The neglected heart f science policy: reconciling supply
of and demand for science. Environmental Science & Policy 10, 5-16.

Savory, A., Butterﬁeld, J ., 1999. Holistic Management: A New Framework for Decision
Making. Published by: Island Press.

Schiettecatte, W., L. D’hondt, W.M. Comelis, M.L. Acosta, Z. Leal, N. Lauwers, Y.
Almoza, G.R. Alonso, J. Diaz, M. Ruiz, D. Gabriels, 2008. Inﬂuence of landuse on soil
erosion risk in the Cuyaguateje watershed (Cuba) CATENA, Volume 74, Issue 1, 15 June
2008, Pages 1-12

Schiller, A., Hunsaker, C.T., Kane, M.A., Wolfe, A.K., Dale, V.H., Suter, G.W., Russell,
C.S., Pion, G., Jensen, M.H., Konar, V.C., 2001. Communicating ecological indicators to
decision makers and the public. Conservation Ecology 5 (1), 19.
http://www.consecol.org/volS/issl/art19/.

Schramm, Gunter. 1980. Integrated river basin planning in a holistic universe. Natural
Resources Journal 20: 787-805.

Seaber, P.R., Kapinos, RR, and Knapp, G.L., 1987, Hydrologic unit maps: US.
Geological Survey Water-Supply Paper 2294, 63 p.

211

Seber, D., R. Keller, K. Sinha, C. Baru. 2003. GEON: Cyberinﬁ‘astructure for the
Geosciences. Abstract submitted to American Geophysical Union, Fall Meeting, 2003,
San Francisco, CA.

Segal, Nathan 2003. How Grid Computing Can Improve Database Performance.
http://www.gridcomputingplanet.com/opinions/article.php/ 1 5771 71

Selvage, M., Wolfson, D., Zurek, B., Kahan, E., 2006. Achieve semantic interoperability
in a SOA http://www.ibm.com/developerworks/webservices/library/ws-soa-
seminterop.htrnl

Shao YP, Liao SY, Wang HQ, 1998. A model of virtual organizations, Journal of
Information Science, 24(5), 305-312.

Shen, Z.F., Luo, J .C. and Zhou, CH, 2004. Architecture design of grid GIS and its
applications on image processing based on LAN. Information Sciences. vl66. 1-17.

Shekhar S, Ravada S, Kumar V, Chubb D, and Turner G, 1996. Parallelizing a GIS on a
Shared-Address Space Architecture, IEEE Computer, 29(1), 42-48

Shi, Y, Shortridge, A. M., and Bartholic, J. F. 2002. Grid computing for real time
distributed collaborative geoprocessing. 10th International Symposium on Spatial Data
Handling (SDH'02), Ottawa, Canada, July, 2002.

Shu,Yanfeng, Zhang, Jack Fan, Zhou,Xiaofang, 2006. A Grid-enabled Architecture for
Geospatial Data Sharing. APSCC '06. IEEE Asia-Paciﬁc Conference on Services
Computing, 2006.

Sieber, P., J. Griese (eds.), Organizational Virtualness, Proceedings of the VONet-
Workshop, April 27-28, 1998, Simowa Verlag, Bern, 1998.

Siepen, G.L., Westrup, J ., 2002. Communicating vegetation management science to land
managers and other stakeholders. Rangeland Journal 24 (1), 170-181.

Singh, V.P. (ed.) 1995. Computer models of watershed hydrology. Water Resources
Publications. Highlands Ranch, Co.

Sinha, A.K., C. Smyth, A. Zendel, B. Brodaric, C. Barnes, A. Snoke, B. Ludaescher, D.
Seber, C. Baru, 2003. Igneous Rocks, Terranes and GEON. Abstract submitted to GSA
Annual Meeting in Seattle, WA, 147-9 2003

Sinha, A.K., K. Lin, R. Raskin, C. Barnes, D. McGuinness, J. Najdi 2005. Geo-

Ontology: empowering new discoveries in Earth sciences. American Geophysical Union,
Annual Meeting, v.87

212

Sinha, A.K., A. Zendel, B. Brodaric, C. Barnes, J. Najdi, 2006. Schema to Ontology for
Igneous Rocks. in, AK. Sinha, editor, Geoinforrnatics, Special Paper 397, Geological
Society of America ,p. 169-182 2006.

Skidmore, AK. 1990, Terrain Position as Mapped from a Gridded Digital Elevation
Model. International Journal of Geographical Information Systems, 4(1), 33-49.

Smith, R., Sheahen, A., 2008. Creating an Extensible GIS with SOA and Web Services.
Jan 2008, Directions
Magazinehttp://www.directionsmag.com/article.php?article_id=2656&trv=1

Soille, P., Gratin, C., 1994. An Efﬁcient Algorithm for Drainage Networks Extraction on
DEMs. J. Visual Commun. Image Represent. 5 (2), 181-189.

Sondheim,M., Gardels, K., Buehler, K., 1999 GIS Interoperability. Geographical
Information Systems, 1999.

Sorokine, A., Daniel, J ., Liu, C., 2005. Parallel visualization for GIS applications. In:
Proceedings GeoComputation 2005. {http://igre.emich.edu/geocomputation2005/ [Last
checked 8/5/08]}.

Spector A Z. 2003. Decades of Distributed Computing, CMU Distinguished Lecture.

Sprague, R.H., Jr., and EC. Carlson. 1982. Building effective decision support systems.
Prentice Hall, Englewood Cliffs, NJ.

Srinivasan, R. and Engel, B.A., 1994. A spatial decision support system for assessing
agricultural nonpoint pollution. Water Resources Bulletin 30 3, pp. 441—452.

Srinivasan, R. and J .G. Arnold. 1994. Integration of a Basin-scale water quality model
with GIS. Water Resources Bulletin. Vol. 30(3), 453-462.

Sterck, Hans De, Zhang, Chen, Papo, Aleks, 2007. Database-Driven Grid Computing
with GridBASE. Advanced Information Networking and Applications Workshops, 2007,
AINAW '07. let International Conference on. Publication Date: 21-23 May 2007.
Volume: 1, On page(s): 696-701. Location: Niagara Falls, ON, Canada.

Steyaert, Louis T., 1993. A Perspective on the State of Environmental Simulation
Modeling. Environmental Modeling with GIS. Eds. Michael F. Goodchild, Bradley 0.
Parks, Louis T. Steyaert. New York: Oxford University Press, 1993. 16-30.

Stockinger,H. 2007. Deﬁning the grid: a snapshot on the current view. The Journal of
Supercomputing. 42 (1), 3-17.

Stonebreaker M, 1995. An overview of the Sequoia 2000 Project. Digital Technical
Journal 7(3) 39-48.

213

http://www.research.compaq.com/wrl/DECarchives/DTJ /DTJ J 03/DTJJ 03PF .PDF , last

checked 3/10/02.
Strader,T.J., F.R. Lin and M.J. Shaw, Information infrastructure for electronic virtual
organization management. Decision Support Systems 23 1 (1998), pp. 75—94.

Sugumaran, R. 2002. Development of an integrated range management decision support
system. Computers and Electronics in Agriculture. Vol. 37, 199-204.

Sui, D.Z. and RC. Maggio. 1999. Integrating GIS with hydrologic modeling: Practices,
problems, and prospects. Computers, Environment and Urban Systems. Vol. 23, 33-51.

Tabara, J .D. 2003. Sustainability cultures. In: Governance for Sustainable Development.
Barcelona: International Institute on governance and government of Catalunya. Papers de
Sostenibilitat, 2: 53-85.

Tang Z., B.A. Engel, J .Y. Choi, K. Sullivan, M. Sharif, K.J. Lim, 2004. A Web-based
DSS for erosion control structure planning. Applied Engineering in Agriculture,
20(5):707-714.

Tarboton, D. G., (1997), A New Method for the Determination of Flow Directions and
Contributing Areas in Grid Digital Elevation Models, Water Resources Research, 33(2):
309-319.

Tarboton, D. G., (2008) website:
http://www.neng.usu.edu/cee/faculty/dtarb/tarsoftware.htrnl

TNM: The National Map, http:/lnationalmapgov/ last checked 09/14/2008.

Turcotte et al., 2001. Determination of Drainage Structure of a Watershed Using a
Digital Elevation Model and a Digital River and Lake Network. Journal of Hydrology.
V240. 225-242.

Turner, 2002. EL. Turner, 11 , Contested identities: human—environment geography and
disciplinary implications in a restructuring academy. Annals AAG 92 (2002), pp. 52—74.

UCGIS (1996) RESEARCH PRIORITIES FOR GEOGRAPHIC INFORMATION
SCIENCE

UCGIS (1998) Distributed Computing, Revised UCGIS White Papers. Online:
http://www.ucgis.org/research_white/distrib.html, last checked 10/15/01 .

UCGIS (2002) Distributed Computing

UCGIS (2002) Pervasive Computing

214

UNESCO 1993, Integrated Water Resource Management: Meeting the sustainability
challenge. IHP Humid Tropics Programme, Series 5, Paris: UNESCO Press.

Uran, O. and R. Janssen. 2003. Why are spatial decision support systems not used? Some
experiences from the Netherlands. Computers, Environment and Urban Systems. Vol. 27.

USEPA Region 7 and Iowa Dept. of Natural Resources 2006, Comprehensive Report of
Ambient Water Quality Monitoring Programs in Iowa.

USGS, (2002), Physiographic provinces data download link:
http://water.usgs.gov/GIS/dsdl/physio.e00.gz [Last Checked 08/05/08].

USGS, (2003), The National Map Seamless Data Distribution Server, EROS Data Center,
http://seam1ess.usgs.gov/ [Last Checked 08/05/08].

USGS, 2007 Hydrologic Unit Codes (HUCs) Explained http://nas.er.usgs.gov/hucs.asp/
[Last checked 8/5/08]

USGS, 2008 About USGS http:/lwww.usgs.gov/aboutusgs/ / [Last checked 8/5/08]

van der Sluijs, J .P., 2007. Uncertainty and precaution in environmental
managementzlnsights ﬁ'om the UPEM conference Environmental Modelling & Software
22 (5), 590-598.

Vieux, B. E., Bralts, V. F., Segerlind, L. J ., Wallace, R. B., 1990, Finite Element
Watershed Modeling: One-Dirnensional Elements, J. of Water Resources Planning and
Management, Vol. 116, No. 6, p803-819.

Vogt JV, Colombo R, Bertolo F (2003) Deriving Drainage Networks and Catchment
Boundaries. A New Approach Combining Digital Elevation Data and Environmental
Characteristics. Geomorphology 53 :28 1—298.

Voinov & Costanza 1999 Watershed management and web, Journal of Env. Management
(1999) 56, 231-245

Walsh, M. R. 1993. Towards Spatial Decision Support Systems in Water Resources.
Journal of Water Resource Planning and Management, Vol. 119(2), pp. 158-169.

Walters, B., 2003. Organizing a watershed alliance, Land and Water,
November/December 2000, 43-44.

Wang, 8., Armstrong, M. P., Ni, J ., Liu, Y. 2005. GISolve: A grid-based problem solving
environment for computationally intensive geographic information analysis. Proceedings
of the 14th International Symposium on High Performance Distributed Computing
(HPDC-l4) Challenges of Large Applications in Distributed Environments (CLADE)
Workshop, IEEE Press, pp. 3-12.

215

Wang, S., Bennett, D. A., Armstrong, M. P., Rajagopal, R., and Brands, E. 2002. Using
grid-enabled teleirnmersive spatial decision support systems (TIDSS) to visualize
uncertainty for water quality protection in agroecosystems. Proceedings of the
International Conference of Geoinfonnatics 2002 for Global Change Studies and
Sustainable Development, 1-3, June, 2002, Nanjing University, Nanjing, China.

Wang, L, and Q. Cheng, 2007. Design and implementation of a web-based spatial
decision support system for ﬂood forecasting and ﬂood risk mapping, Geoscience and
Remote Sensing Symposium, 2007. IGARSS 2007. IEEE International, pp. 45 88-4591.

Wang X., M. Homer, S.D. Dyer, C. White-Hull and C. Du, 2005. A river water quality
model integrated with a web-based geographic information system. Journal of
Environmental Management, 75(3):219-228

Wang, S., Zhu, X., 2008. Coupling Cyberinfrastructure and Geographic Information
Systems to Empower Ecological and Environmental Research. BioScience, Volume 58,
Issue 2, pp. 94-95.

Ward, Andrew D. & Stanley W. Trimble & M. G. (FRW) Wolman 2003. Environmental
Hydrology. CRC Press, 2003.

Watermolen, DJ. 2008. Land cover data and tools for analyzing alternative futures: one
state's lessons learned. Chapter 16, pp. 283-320. In North American Land Cover Summit:
A Special Issues of the Association of American Geographers. (In press).

WATERS Network, 2008. http://www.watersnet.org/index.html. last checked 09/14/08.

Watson, L.T., Vinod K. Lohani, David F . Kibler, Randel L. Dymond, Naren
Ramakrishnan,and Clifford A. Shaffer, 2002. Integrated Computing Environments for
Watershed Management. J. Comp. in Civ. Engrg., Volume 16, Issue 4, pp. 259-268.

Webler, T., Seth Tuler 2001. Public Participation in Watershed Management Planning:
Views on Process from People in the Field, Human Ecology Review, Vol. 8, No. 2, 2001

Williams, D.N., David E. Bemholdt, Ian T. Foster, and Don E. Middleton, 2007. The
Earth System Grid Center for Enabling Technologies: Enabling Community Access to
Petascale Climate Datasets, CTWatch Quarterly 3 (2007), invited article.

Wilson, J .P., and J .C. Gallant (ed.) 2000. Terrain Analysis: Principles and Applications.
J. Wiley & Sons. New York.

Wilson J. P., Mitasova H. and Wright D. S., 2000. Water Resource Applications of
Geographic Information Systems, URISA Journal 12(2): 61-79.

216

Wischmeier, W.H., Smith, DD, 1978. Predicting rainfall erosion losses: a guide to
conservation planning. Agriculture Handbook, vol. 537. USDA, Washington DC, USA.

58 pp.

Wood, M., F emer, C., Brown, J. 2004. Toward a Graphical User Interface for Grid
Services. SoutheastCon, 2004. Proceedings. IEEE Publication Date: 26-29 Mar 2004 On
page(s): 316- 324.

Woodhill A.J. 2003. Dialogue and transboundary water resources management: towards a
framework for facilitating social learning. In: S. Langaas and Timmennan, J .G.,
(eds.),The role and use of environmental information in European transboundary river
basin management. IWA Publishing, London.

Woolhiser, D. A., Smith, R. E. and Goodrich, D. C. 1990. KINEROS: A kinematic runoff
and erosion model: documentation and user manual, USDA Agricultrual Research
Service ARS-77.

Xie, H. & P.D.Yapa., 2006. Developing a Web-Based System for Large-Scale
Environmental Hydraulics Problems with Application to Oil Spill Modeling, J. Comp. in
Civ. Engrg., Volume 20, Issue 3, pp. 197-209

Xu, X.-P., Booij, M.J., Mynett, A.E., 2007. An appropriateness framework for the Dutch
Meuse decision support system. Environmental Modelling & Software 22 (11), 1667-
1678.

Xue, Y., Cracknell, A. P. and Guo, H. D., 2002, Telegeoprocessing: The Integration of
Remote Sensing, Geographic Information System (GIS), Global Positioning System
(GPS) and Telecommunication. International Journal of Remote Sensing, Vol. 23, No. 9,
pp1851-1859.

Yoon, J. 1996. Watershed-scale nonpoint source pollution modeling and decision support
system based on a model-GIS-RDBMS linkage. In: Proceedings of the AWRA Annual
Symposium, GIS and Water Resources. American Water Resources Association,
Hemdon, Virginia, TPS-96-3, 482 pp.

Youn, C., T. Kaiser, C. Santini, D. Seber, 2006. Developing Metadata Services for Grid
enabling Scientiﬁc Applications. ICCS 2006, University of Reading, UK, May 28-31,
2006, Proceedings, LNCS v. 3991, pp. 379-386, 2006.

Zhang, Tong, Tsou, Ming-Hsiang, 2005. The Integration of Grid-enabled Internet
GIServices and Geographic Semantic Web Technologies Geographic Information
Sciences Vol. 11, No. 1, June 2005.

Zhang, J ., Sven Erik Jargensen. 2005. Modelling of point and non-point nutrient loadings
from a watershed. Environmental Modelling & Software 20 (2005) 561-574

217

Zhang, W., Wang, X.J., 2002. Modelling for point-non-point source effluent trading:
perspective of non-point sources regulation in China. Sci. Total Environ. 292 (3), 167-
176.

Zhou, Q. and Liu, X., 2002. Error Assessment of Grid-based Flow Routing Algorithms
Used in Hydrological Models, International Journal of Geographical Information
Science, 16(8): 819-842.

Zimmerman, A. & Nardi, B. (2006): Whither or Whether HCI: Requirements Analysis
for Multi-Sited, Multi-User Cyberinfrastructures. In Extended Abstracts of SIGCHI
conference of Human Factors and Computing, 1601-1606.

Zlatanova, S. and Jantien Stoter, 2006. The role of DBMS in the new generation GIS
architecture. in Frontiers of Geographic Information Technology. Springer, 2006.

218

3