SCIENCE IN THE DIGITAL AGE: OVERCOMING UNCERTAINTY AND THE ADOPTION OF VOLUNTEERED
GEOGRAPHIC INFORMATION FOR SCIENCE
By
Shaun Arthur Langley

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Geography—Doctor of Philosophy
2014

ABSTRACT
SCIENCE IN THE DIGITAL AGE: OVERCOMING UNCERTAINTY AND THE ADOPTION OF VOLUNTEERED
INFORMATION FOR SCIENCE
By
Shaun Arthur Langley
With the advent of Web 2.0, the public is becoming increasingly interested in spatial data
exploration. The potential for Volunteered Geographic Information (VGI) to be adopted for
Science through collaborations between researchers and non-scientists is of special interest to
me. In particular, mobile devices and wireless communication permit the public to be more
involved in research to a greater degree. Furthermore, the accuracy of these devices is rapidly
improving, allowing me to address questions of uncertainty and error in data collections.
Cooperation between researchers and the public integrates themes common to VGI and PGIS
(Participatory Geographic Information) to bring about a new paradigm in GIScience. This
dissertation discusses VGI in the context of a new paradigm, eScience, and the broader
framework of Neogeography. I discuss current issues with data quality and uncertainty regarding
VGI and detail one approach to quality credibility of the data. Finally, the dissertation outlines
the framework for utilizing VGI in the context of case study in disease ecology for the purpose of
surveillance of tsetse flies, the primary vector of African Trypanosomiasis. My system allows for
two-way communication between researchers and the public for data collection, analysis, and
the ultimate dissemination of results. Enhancing the role of the public to participate in these
types of projects can improve both the efficacy of disease surveillance as well as stimulating
greater interest in science.

This work is dedicated to Sam, my little bundle of joy and the motivation to push myself, and to
Courtney whom I love and admire greatly. You inspire me to be a better person, and your neverending support for me gave me the strength to push on, even when it seemed difficult.

iii

ACKNOWLEDGEMENTS
I would like to acknowledge all of those individuals who helped me over the course of my tenure
as a PhD student. To the staff in Geography, particularly Sharon Ruggles, I am grateful for all you
have done to help me. I would have been perpetually lost without you.
I am grateful for the help I received while in Kenya. In particular, I want to acknowledge
Dr. Maitima for his guidance in pulling everything together while in Nguruman. I also want to
acknowledge the assistance of Joel Meiyponi in the collection of my flytrap data, and for assisting
with the translation of interviews that both he and I conducted in Nguruman. I'm also very grateful
for the healthcare workers that tended to me following my accident. I also want to acknowledge
the Elders in Ol'Kirmatian for their willingness to allow me to operate in the area.
I am immensely grateful for my committee members – Dr. Ashton Shortridge, Dr. Sue
Grady, Dr. Edward Walker, and my advisor Dr. Joseph Messina. Your support, guidance,
feedback, and collaborations were instrumental in getting me to this point.
I am grateful for the support of all my friends, in particular Dr. Hamm for pushing me to
finish this dissertation. The support of friends is critical to making it through Graduate School and
you all provided that support without precondition.
I am so thankful for the support of my Parents and family for being patient with me as I
am concluding my 14th year in college. Only sometimes did you comment that it was taking an
awful long time to finish. I hope that you feel it has been worth it!
Finally, and most importantly, I want to thank Courtney for your unwavering love and faith
in me. You have taught me more than you know; and to top it off, you've given me the most
wonderful gift of all, Sam. It is now my mission in life, over the next 60 years or so, to try and
repay you all the kindness and patience you've shown me.
iv

The work in this dissertation was supported in part through a grant from National Institutes
of Health, Office of the Director, Roadmap Initiative, NIGMS Award No. RGM084704A, from
three Graduate Office Fellowship awards, and from the US Government's Student Loan Program.

v

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................................. viii
LIST OF FIGURES .............................................................................................................................. ix
CHAPTER 1....................................................................................................................................... 1
INTRODUCTION ............................................................................................................................... 1
Conceptual Framework ............................................................................................................... 1
Data Quality ................................................................................................................................. 6
Spatial data management ........................................................................................................... 8
Concluding Thoughts ................................................................................................................... 9
Specific Aims.............................................................................................................................. 10
Outline of the dissertation ........................................................................................................ 11
CHAPTER 2..................................................................................................................................... 15
EMBRACING THE OPEN-SOURCE MOVEMENT FOR THE MANAGEMENT OF SPATIAL DATA: A
CASE STUDY OF AFRICAN TRYPANOSOMIASIS ............................................................................. 15
Abstract ..................................................................................................................................... 15
Introduction............................................................................................................................... 15
Case Study: A Model for African Trypanosomiasis in Kenya ................................................. 20
Data Holdings and Acquisitions ............................................................................................. 22
Development of a Spatial Database System ............................................................................. 26
Conceptual Model ................................................................................................................. 27
Database Standards ............................................................................................................... 31
SQL-Rule-Based Interactions and Scripting ........................................................................... 32
DBMS Interface – A Manager Perspective ............................................................................ 34
DBMS Interface – A User Perspective ................................................................................... 35
Statistical and Analytical Analysis .......................................................................................... 35
Implementation ......................................................................................................................... 36
Software Packages ................................................................................................................. 36
Initializing Postgres ................................................................................................................ 37
WKT Raster extension for PostGIS ........................................................................................ 38
Configuring GRASS ................................................................................................................. 39
Loading Data and interfacing GRASS with Postgres .............................................................. 40
Limitations and Future Expansion ............................................................................................. 43
Summary ................................................................................................................................... 43
CHAPTER 3..................................................................................................................................... 47
UTILIZING VOLUNTEERED INFORMATION FOR INFECTIOUS DISEASE SURVEILLANCE ................. 47
Abstract ..................................................................................................................................... 47
Introduction............................................................................................................................... 48
vi

Background................................................................................................................................ 49
Traditional Paradigm ............................................................................................................. 49
VGIS Paradigm ....................................................................................................................... 50
Disease Surveillance .............................................................................................................. 55
Case Study ................................................................................................................................. 56
Purpose .................................................................................................................................. 56
Site Description ..................................................................................................................... 57
Conceptual Model ................................................................................................................. 60
Data Collection and Interface ................................................................................................ 64
Information Reliability ........................................................................................................... 65
Utilizing Volunteered Information to Reduce Model Error and Uncertainty ....................... 69
Conclusion and Limitations ....................................................................................................... 72
CHAPTER 4..................................................................................................................................... 75
USING META-QUALITY TO ASSESS THE UTILITY OF VOLUNTEERED GEOGRAPHIC INFORMATION
FOR SCIENCE ................................................................................................................................. 75
Introduction............................................................................................................................... 75
Methodology ............................................................................................................................. 81
Results ....................................................................................................................................... 86
Discussion .................................................................................................................................. 95
CHAPTER 5................................................................................................................................... 103
SUMMARY AND CONCLUSION .................................................................................................... 103
Introduction............................................................................................................................. 103
Summary of Main Findings ...................................................................................................... 103
Theoretical Implications of this Dissertation .......................................................................... 109
Recommendations for Future Research ................................................................................. 110
APPENDICES ................................................................................................................................ 114
BASE CODE SIMULATION ..................................................................................... 115
SIMULATION 11 CODE.......................................................................................... 117
................................................................................ 122
REFERENCES ................................................................................................................................ 124
_Toc403810904

vii

LIST OF TABLES
Table 1.1: Components of data quality for spatial data .............................................................. 12
Table 2.1: A summary of our data library grouped by major theme ............................................ 23
Table 2.2: A sample of the subset of Kenya census data we hold from the 1990 National Census
....................................................................................................................................................... 46
Table 4.1: Reporter types and the criteria used to simulate their behavior ................................ 99
Table 4.2: Simulation results for simulated conditions. Values represent percent increase over
the base TED model .................................................................................................................... 100
Table 4.3: The percentage increase in the prevalence of tsetse over the base TED model for
simulations 8-12 .......................................................................................................................... 101
Table 4.4: The percentage increase in the prevalence of tsetse over the base TED model for
simulations 13-20........................................................................................................................ 102

viii

LIST OF FIGURES
Figure 2.1: Maximum extent of tsetse distribution predicted by the TED Model between the
beginning of 2002 and the end of 2009. As described by DeVisser et al. (2010), the TED Model
uses five scenes of MODIS 1km annual land cover, 207 scenes of MODIS 250 m Normalized
Difference Vegetation Index (NDVI), and 207 scenes of MODIS 1km day/night Land Surface
Temperature (LST) products to predict the fundamental niche of tsetse in Kenya, and a fly
movement model to predict tsetse distributions or realized niche of the tsetse species of interest.
The TED Model is written in Python scripting language and is run within ArcGIS 9.2 (or later
versions of ArcGIS). Other data sets used to construct the map include shapefiles of Kenyan major
roads, highways, cities, rivers, Kenyan water bodies, lakes, and African country political
boundaries. To construct the background topographic relief map, the Shuttle RADAR Topographic
Mission (SRTM) 90 m Digital Elevation Model (DEM) was used to create a gridded hillshade
product, and a dry season NDVI scene was combined with the SRTM DEM to create an
elevation/vegetation color scheme. ............................................................................................. 25
Figure 2.2: A conceptual model framework for the spatial DBMS. .............................................. 28
Figure 2.3: Flow of data through the model implementation as users interact with the system. It
should be noted that flows are one-directional, meaning that although users can interact with
the data to generate analysis, the results must be stored as a separate entity in the database.
This ensures that the underlying data cannot be changed. ......................................................... 29
Figure 3.1: Study Area ................................................................................................................... 57
Figure 3.2: This deployment diagram illustrates the interaction of the separate components of
the VGIS and the flow of information between each component ............................................... 62
Figure 3.3: We propose an iOS application (for iPhone or iPad) that allows for users to interact
with the VGIS, explore model predictions, volunteer data, or to contribute .............................. 63
Figure 3.4: To assess the reliability of volunteered information, a report is evaluated in the
context of a set of conditions. This figure presents a logical thought diagram for the application
of the computation of reliability (3-2) .......................................................................................... 67
Figure 3.5: Users may volunteer reports of tsetse presence under a range of scenarios. (A)
Illustrates the case where a report fills in a gap in a patch of tsetse, likely correcting an error in
TED model predictions. (B) Illustrates the case where a report establishes connectivity between
two isolated patches of tsetse. (C) Illustrates the case where a report of tsetse presence is
spatially isolated from the predicted distribution of tsetse. In each case, a user is presented with
a prediction of tsetse distribution from the TED model (Column 1). Users identify an error in the
model, observing tsetse in an area where they are not predicted to occur, and submit a report

ix

(Column 2 - black box). The report is submitted for reliability assessment; if deemed reliable, TED
model predictions are updated to reflect the new information (Column 3 – black box). ........... 71
Figure 4.1: A frequency plot representing the time-step in which reporters cluster into two
groups, for 100 replications of simulation 13 ............................................................................... 88
Figure 4.2: A frequency plot representing the time-step in which reporters cluster into two
groups, for 100 replications of simulation 10 ............................................................................... 89
Figure 4.3: A frequency plot representing the time-step in which reporters cluster into two
groups, for 100 replications of simulation 11 ............................................................................... 89
Figure 4.4: A frequency plot representing the time-step in which reporters cluster into two
groups, for 100 replications of simulation 12 ............................................................................... 90
Figure 4.5: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 10. Values represent the proportion of time-steps in the model where
tsetse were present; this is a rough approximation of the probability of tsetse occurrence. ..... 92
Figure 4.6: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 11. Values represent the proportion of time-steps in the model where tsetse
were present; this is a rough approximation of the probability of tsetse occurrence. ............... 93
Figure 4.7: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 12. Values represent the proportion of time-steps in the model where tsetse
were present; this is a rough approximation of the probability of tsetse occurrence. ............... 94
Figure 4.8: This figure overlays the scores of 100 reporters for simulation 8.............................. 96

x

CHAPTER 1
INTRODUCTION

Conceptual Framework
What does it mean to be a scientist? What does it mean to be a Geographer? These questions
have existed and been debated for as long as the discipline has existed. Geography is ripe with
tradition, having evolved many times to suit the consciousness of the day (Livingstone, 1992).
Early Geographers like Francis Bacon and John Locke were Empiricists, advocating the scientific
method as the means to explain the world. Plato was a proponent of realism, the philosophy that
states that truths are universal and have an objective or absolute existence. René Descartes,
arguably a geographer in name only, was a proponent of subjectivism, perceiving knowledge as
truly subjective. Under Descartes’ philosophy, there were no absolute, external, or objective
truths; rather, reality existed uniquely for each individual in their own way. Thomas Kuhn
transformed the discipline with a relativism paradigm, arguing that knowledge, truth, and
morality exist in relation to culture, society, or in a historical context; however, truths were not
absolute and thus conflicting theories weren’t necessarily the result of being right or wrong,
rather having arrived at a conclusion from different perspectives. Finally Karl Popper proposed
the critical rationalist philosophy, arguing that the falsifiability of science was paramount to the
generation of knowledge, and that all scientific theories should be rationally criticized and
subjected to tests of falsifiability.
As researchers began to acknowledge the importance of context in describing the
Geography of a place, citizens were engaged directly by researchers seeking to capture the “local

1

knowledge” that only a citizen can convey (Elwood, 2006a; Robbins, 2003). Participatory science
emerged in the 1990s as a context within which we can incorporate quantitative, qualitative, and
cartographic forms of data into a GIS environment (Elwood, 2006b; Warren, 1991). The term was
first used by Warren (1991) and Pickles (1995) to define the method of engaging and empowering
communities to address political, social, and environmental questions that directly concern and
impact them (Robbins, 2003). Warren (1991) notes, “Local knowledge is increasingly
acknowledged to be scientific in the sense that it evolves from experimental techniques of trialand-error conditions not entirely distinct from scientific practices”. He argues that this knowledge
should not be too quickly dismissed as it can inform us of the social, political, and economic
structure of communities. Furthermore, the use of cartographic and GIS tools can empower local
communities to use their own knowledge and understanding of their environment to better
manage their resources, even resolve standing disputes. The latter applications are commonly
associated with Participatory GIS (PGIS) in the work of Sarah Elwood (e.g. 2006a, 2006b) and
Michael McCall (2005). Turner and Hiernaux (2002) demonstrate the ability of cartographic
methods, informed by local knowledge, to develop more effective local management strategies
than can be achieved through a spatial modeling experiment alone. Finally, Robbins (2003)
extends the application of PGIS by Turner and Hiernaux (2002) indirectly in his notion of
“indigenous GIS”. It is the goal of PGIS to generate knowledge in collaboration with communities
to enable them to independently critique spatial data for their own purposes and benefits
(Elwood, 2006a). To quote Elwood (2006b),
“the existing literature demonstrates that the nature of the participatory
knowledge production in PPGIS (public participation GIS) is the result of
complex interactions of technological and social factors, but there have been
2

relatively few attempts to detail how these intersecting factors play out in and
are shaped by the daily choices and negotiations of PPGIS projects.”
Universal throughout the evolution of Geography and perceptions of science was the
distinction between citizens and scientists. There are those who can do science – experts in their
field, educated and trained, often residing in academic institutions, and producers of knowledge;
and there are those who cannot – citizens, laypeople, and consumers of knowledge. It is difficult
to identify a specific shift in consciousness; rather there has been a subtle recognition of the role
citizens can, and do, play in science. Beginning in the early 1900s, the Audubon Society began
enlisting the help of citizen scientists and volunteers in their annual Christmas Bird Watch. Still
today, these amateur ornithologists participate in the annual bird count; to-date they have
amassed more than 31 million records since its inception (Haklay, 2012; Silvertown, 2009). The
volunteers who participate do not do so at their own leisure though. They are specifically trained
in data collection methods and their data are subjected to at least a minimal amount of quality
control.
Haklay (2011) identifies four types of citizen science, which he distinguishes based on an
individual’s training and level of participation. The first type is “crowdsourcing”. Howe (2006)
defines it as engagement in the collection of data with a minimum level of effort. At this stage,
citizens are perceived as nothing more than data collection tools, or sensors, off which we can
glean information about their environment.
The second level is termed “Distributed intelligence”. Here, citizens are tasked with
slightly more complex tasks. They are usually given some amount of training in data collection
methods, and may be asked to perform simple interpretations of their environment. Citizens thus

3

become basic interpreters of the data they collect, rather than simply sensors of information
(Haklay, 2011).
The third level “participatory science” is marked by significant participation on the part of
citizens who collaborate with scientists directly at each stage in the project from inception to
execution and analysis (Haklay, 2011; Irwin, 1995). This mode of citizen science has dominated
the Geography literature and has garnered a great deal of attention for its ability to grant
empowerment to disenfranchised groups.
Finally, the fourth level of citizen science is “extreme”. It is at this stage that science
becomes truly collaborative as the citizen is fully integrated into the scientific process (Haklay,
2011). This can potentially allow for citizens to be scientists without the presence of the
professional or expert.
The last decade has seen dramatic technological and communications advances. With the
advent of Web 2.0 and the proliferation of new technologies and modes of communication,
exposure to geographical knowledge is increasingly pervasive in our society (O'Reilly, 2006, 2007)
(Goodchild, 2007a, 2007b). Utilizing web-based applications, citizens have developed an interest
in becoming active participants in describing their environment by volunteering geographic
information.
Whereas citizen science defines the actions of the individual in participating in the
scientific process, Neogeography reframes the context in which it occurs. Neogeography defines
the erosion of traditional roles and the blurring of the distinction between citizens and scientists
(Goodchild, 2009). Central to the idea is a reframing of the space in which knowledge generation
occurs. The literature points to the emergence of the Geoweb, the online environment where
4

geographical information is mashed with abstract data (Haklay, 2011). However, the activities of
individuals in the Geoweb have arisen organically, lacking the traditional oversight and training
in methodology given by scientists. Today there are literally thousands of web applications in
which individuals can actively volunteer information, often geographical in nature (Goodchild,
2009).
Volunteered geographic information (VGI) is a term that was coined by Michael Goodchild
to describe the phenomenon of citizens contributing, volunteering, and consuming geographical
information outside the purview of academia (Goodchild, 2007a). Perhaps the most commonly
cited examples of this phenomenon are Wikimapia1 and OpenStreetMap2; in these examples, the
citizens contribute or volunteer information and descriptions about the built environment
around them (Goodchild, 2007a; Haklay & Weber, 2008). However, the past decade has seen a
tremendous increase in the utilization of VGI (Elwood et al., 2011).
In this dissertation, I use the term Volunteered GIS (VGIS) to refer to a conceptualization
of GIS that emphasizes the interest of non-scientists to be active participants in the generation
of spatial knowledge (Flanagin & Metzger, 2008; Goodchild, 2007c, 2010b). A tradition previously
reserved for “professional” scientists, VGIS represents a shift from viewing science as having a
single authority (the scientist) to a model where authority is relative and expressed contextually.
Information abundance, repetition, and the information collective conveys credibility to itself
(Craglia et al., 2007). However, the issue of data quality has been debated extensively in the
literature, and represents a critical hurdle to the adoption of VGIS for Science.

1
2

http://wikimapia.org
http://www.openstreetmap.org

5

Data Quality
The quality of spatial data is the foundation for its utility in the scientific process. Quantifying
error and related uncertainty is the most critical evaluation of data that can be made. The
credibility of any information or data is directly related to an assessment of data quality. The
issue of data quality has a long tradition in the literature, particularly as it pertains to spatial data.
One of the most comprehensive lists of data quality metrics was outlined by van Oort (2005). In
his dissertation, he identifies 11 components of data quality that pertain to spatial data (Table
1.1). The assessment of data quality has objective and subjective components, which collectively
convey the reliability we have as to the truthfulness of the information. Although there are a
multitude of metrics, arguably the most important, and objective, components of spatial data
quality are: positional accuracy, attribute accuracy, and completeness (Haklay, 2010). Within the
context of traditional spatial data infrastructures (SDIs) these metrics facilitate an assessment of
the variability of the data with respect to the population being measured and allow for an
assessment of uncertainty. Assessing data quality using objective metrics allows for us to make
statements with regard to confidence in our results.
In the age of Neogeography, data are routinely volunteered without the corresponding
information needed to make an objective assessment of data quality (e.g. positional error, spatial
resolution, etc.). A subjective assessment of data quality, can serve as a means to determine a
dataset’s credibility and as such make a determination of its fitness-for-use (Grira et al., 2009).
As the determination of data quality is itself a subjective determination on the part of the
consumer, the communication of the metrics is critical. To this extent, the communication of
quality metrics is critical. Within metadata of a dataset, the user is provided with the information
6

needed to make a quality assessment. However, too often the communication to users (and
sometimes scientists) is done without regard for whether or not the receiver is capable of
understanding or perceiving it as is intended by the communicator (Grira et al., 2009). Reporting
data quality through metadata is therefore an ineffective means of communication and can result
in potential misuse of data (Comber et al., 2006; Grira et al., 2009). Subjective assessments are a
matter of perception. With regard to citizens and non-experts, the perception of quality is less a
matter of quality metrics, but rather relates to their perception of the communicator themselves
(Metzger et al., 2003).
The emergence of Citizen Science has generated volumes of data that are distinct from
traditional datasets, primarily because of the manner in which they were collected. There has
been a great deal of debate as to the best way to assess the credibility of volunteered data
(Elwood, 2006b; Flanagin & Metzger, 2008; Metzger et al., 2003). Many illustrations of this
assessment rely on a comparison of VGI against datasets of known quality (Koukoletsos et al.,
2012). But perhaps the most promising approach is to correlate the credibility of the
communicator with the quality of the data (Corbett, 2012).
The uncertainty concerning data quality for volunteered geographic information (VGI)
and crowdsourced data has resulted in poor adoption of citizen science initiatives for science,
whether experimental or analytical (Craglia et al., 2007; Elwood et al., 2013; Goodchild, 2009). A
2009 survey conducted by Elwood et al. (2011) looked at a range of projects identified to be
utilizing VGI in some form. Of the 99 studies they evaluated, only 3% were sponsored by an
academic institution, 7% were sponsored by a Government program, and 7% were affiliated with
an NGO. The majority of VGI initiatives are associated with for-profit institutions (63%). The
7

overwhelming element of for-profit companies utilizing VGI runs counter to the notion of VGI as
an expression of citizen science (Goodchild, 2007a). However, there is significant academic
interest in adopting VGI initiatives to enhance collaborations between researchers and the
communities (in which they operate), and to develop framework data (Craglia et al., 2007;
Elwood et al., 2011; Haklay, 2012). Framework data constitute the core of a spatial data
infrastructure (SDI) as it aims to represent core phenomena; it contains data on geodetic control,
orthoimagery, elevation, transportation, hydrography, governmental units, and cadaster (Craglia
et al., 2007; Elwood et al., 2011).
For VGI and crowdsourced data to be valuable for science in an academic context, we
must address the lingering questions of credibility (of the data) and uncertainty regarding data
quality (Craglia et al., 2007; Flanagin & Metzger, 2008; Haklay et al., 2010).

Spatial data management
The development of new communications and technology in the early 1990s created a need for
new data management strategies, and a fundamentally different standard data model — a set of
abstractions that define object, relation, and attribute types and their use (Devogele et al., 1998;
R. Groot & McLaughlin, 2000). Such data models form the basis for a spatial data infrastructure
(SDI) for the purpose of “facilitating coordination, production, access, and use of spatial data”
(Budhathoki et al., 2008; R. Groot & McLaughlin, 2000; I Masser, 2005). In 1994, President Clinton
signed an executive order establishing a national SDI (Coordinating geographic data acquisition
and access: The national spatial data infrastructure 1994). For more than a decade, the standard
SDI was sufficient for the data types available. However, with the emergence of the eScience
paradigm (Gray & Szalay, 2006) and Web 2.0 (O'Reilly, 2006, 2007), the types of spatial
8

information no longer fit neatly within in the constructs of the traditional data model (Rouse et
al., 2007). Thus, there were calls to modernize the standard SDI.
Next generation SDIs have implemented web services and broadened the types of data
supported; however the majority still perceive the user role as passive (Budhathoki et al., 2008),
thus rendering them ineffective for citizen science initiatives. This is a large part of the reason
that current SDIs are so vastly underutilized, resulting in widespread inconsistencies and a lack
of interoperability among systems (Budhathoki et al., 2008; Ian Masser, 2005).
It is not sufficient to simply define an alternate data model for use with VGI; such an
approach reduces the status of the information to a second tier, perpetuating the inequality of
roles and utility of the data. Users of SDIs tend to be expert organizations (Budhathoki et al.,
2008), while Neogeographers operate in a patchwork system of inconsistent authorities. Bringing
forward a unified data model will not only elevate the authority of VGI, but will improve overall
data quality by standardizing the communication of quality metrics.

Concluding Thoughts
The acquisition of data for science is arguably the single largest hurdle researchers face.
Information must be obtained in a controlled manner so as to assure the interpretation under
the prescribed theoretical model. Any uncertainty surrounding the quality or integrity of the
information obtained can have a crippling effect on the ability of the researcher assert
significance for any observed patterns (in a traditional statistical framework). Crowdsourced data
or VGI can provide significant contributions to science if questions of quality and integrity can be
addressed. Some have suggested that these uncertainties render such sources of information
useless; however I think we are approaching a new era in science that will dramatically change
9

our perception of such sources and bring about a new paradigm. I think the technological
advances of the past century and the advent of Web 2.0, combined with the proliferation of
mobile technologies and the increasing incorporation of smart sensors into these devices will
provide new opportunities for ordinary citizens to be active participants in science through the
collection, dissemination, and even analysis of information. Such opportunities are not only
beneficial for researchers to access new kinds of information, but also present a unique
opportunity for citizens to be more involved in science – reversing the disturbing trend of
scientific illiteracy we are witnessing in society today. In this regard, addressing the questions
posed in this dissertation become critical for ushering in a new era of scientific discovery and
opportunity for future generations.

Specific Aims
Objective 1: Address three recurring problems with spatial data management: scalability,
reliability, and security by:
1. Communicating a conceptual model for a comprehensive open-source computing
environment that promotes the efficient organization, storage and retrieval of disparate
data.
2. Extending the discussion of spatial databases by presenting a model framework for a
spatial DBMS that rigorously and consistently manages both spatial and nonspatial data.

Objective 2: Demonstrate the utility of VGI by:
1. Describing a prototype for the utilization of VGI to enhance disease surveillance
programs.
10

2. Articulating an approach for integrating VGI into a traditional species distribution model.

Objective 3: Address lingering concerns of credibility and data quality in VGI by:
1. Illustrating how to dynamically assess the reliability of reporters of VGI.
2. Assessing the impact of incorporating VGI of varying quality into a traditional species
distribution model.

Outline of the dissertation
Chapter 2 of the dissertation addresses specific aim 1 — issues of spatial data management that
are exacerbated by the volume of data generated through crowdsourcing and VGI. In this paper
I outline the framework for a spatial database management system (sDBMS) that facilitates the
efficient storage, query, and retrieval of spatially explicit data. The paper was published in 2011
in the Journal of Map & Geography Libraries and appears in this dissertation in its published form.
Chapter 3 of the dissertation answers specific aim 2 in illustrating the utility of VGI in
science. I utilize a case study in disease ecology to demonstrate how VGI can be incorporated
into a traditional species distribution model to enhance the output and relevance for managing
disease vectors more effectively. This chapter was published in 2013 in the International Journal
of Applied Geospatial Research, and appears in its published form.
Chapter 4 addresses specific aim 3 in which I directly discuss issues of credibility and
quality for VGI. I outline an approach in which the reliability of volunteers (or reporters) of VGI
can be computed, and how this measure can serve as a surrogate metric of data quality.

11

Table 1.1: Components of data quality for spatial data
Metric

Definition

Expression

Attribute accuracy

“an assessment of the accuracy of the identification
of entities and assignment of attribute values in the
data set.”

Khat or Cohen’s Kappa

Positional accuracy

“… to the degree to which the digital representation
of a real-world entity agrees with its true position on
the earth’s surface”

1
𝑅𝑀𝑆𝐸 = √ (∑ 𝛿𝑥𝑖2 + ∑ 𝛿 𝑦𝑖2 )
𝑛

“… the agreement between encoded and actual
temporal coordinates.”

Textual

Temporal accuracy

𝑛

𝑛

𝑖=1

𝑖=1

Semantic accuracy

∑𝜈∈𝑉 𝑤(𝜈) ∙ 𝑑(𝛾(𝜈), 𝛾 ′ (𝜈))
∑𝜈∈𝑉 𝑤(𝑣)

Completeness

“measurable error of omission observed between the
database and the specification.”

Omission: % of data missing relative to the
specification
Commission: % of data present that is not in
the current specification of dataset or extract
Coverage ratio: occurrences of one variable
per unit of another

Logical consistency

“refers to the absence of apparent contradictions in
the database”

textual
Continued on next page

12

Table 1.1 (cont’d)
Metric

Definition

Expression

Spatial resolution

“... the fineness of detail that can be observed.”

Integer

Temporal resolution

“… the minimum duration of an event that is
discernible.”

Integer

Thematic resolution

Categorical data: “resolution is defined in terms of
the fitness of category definitions.”
Quantitative data: “resolution is determined by the
precision of the measurement device.”

Integer

Fitness-for-use

“ability to use the dataset for a particular purpose or
situation.”

textual
Continued on next page

13

Table 1.1 (cont’d)
Metric

Definition

Expression

Lineage

“[Lineage] refers to source materials, methods of
derivation and transformations applied to the
database. This includes temporal information (data
that the information refers to on the ground), and is
intended to be precise enough to identify the sources
of individual objects (i.e. if the database was derived
from different sources, lineage information is to be
assigned as an additional action viewing objects in a
spatial overlay).”

textual

Meta-quality

Variability

“a measurement of the collective quality of the
data.”
“… the difference between expected measures and
actual values.”

14

𝜖 =𝑦−𝜇

CHAPTER 2
EMBRACING THE OPEN-SOURCE MOVEMENT FOR THE MANAGEMENT OF SPATIAL DATA: A CASE
STUDY OF AFRICAN TRYPANOSOMIASIS3

Abstract
The past decade has seen an explosion in the availability of spatial data not only for researchers,
but the public alike. As the quantity of data increases, the ability to effectively navigate and
understand the data becomes more challenging. Here we detail a conceptual model for a spatially
explicit database management system that addresses the issues raised with the growing data
management problem. We demonstrate utility with a case study in disease ecology: to develop
a multi-scale predictive model of African Trypanosomiasis in Kenya. International collaborations
and varying technical expertise necessitate a modular open-source software solution. Finally, we
address three recurring problems with data management: scalability, reliability, and security.

Introduction
The trans-disciplinary nature of modern research in disease ecology often requires and generates
vast quantities of data varying thematically and in structure. The data management challenge is
considerable and often cost prohibitively so. Data arise from a multitude of sources and occur in

3

This chapter appears in its published form. Minor adjustments have been made to the text to

address invalid references and/or grammatical errors. I address more substantially changes to
the code in Appendix C.

15

spatially explicit or aspatial forms with concomitant structures. Rarely are these data
accompanied by the ontologically coherent metadata necessary to facilitate cooperation and
collaboration. Recent discourse in the studies of infectious disease ecology have suggested a
need to emphasize the role of space and land cover change dynamics in describing the
interactions of diseases with environmental processes (Ostfeld et al., 2008). Therefore, effective
engagement in disease ecology research requires the ability to access, correctly interpret and
integrate these highly diverse data (Longley et al., 2005; Shekhar & Chawla, 2003; Watson et al.,
2004).
In the early 1990’s there was a great deal of research concerning the expansion of
traditional database management systems (DBMS) to incorporate functionality for handling
spatially explicit data types (Shekhar & Chawla, 2003; Michael Stonebraker & Moore, 1995).
Traditional DBMSs are self-describing in that the definitions of data are stored in a catalog, along
with the raw data, without the need to store separate descriptive files. This is an extremely
efficient means for managing and accessing large quantities of data. Databases improve our
abilities to interact with data through the construction and querying of indices, which serve as a
data roadmap stored on the physical media. Early database frameworks rarely interacted directly
with spatial data (Egenhofer, 1994; Michael Stonebraker & Kemnitz, 1991). Yet with technological
advances, particularly the development of satellite platforms for remote sensing, great quantities
of data were being generated, necessitating the development of DBMS capabilities specifically to
facilitate handling of spatial data.
SQL (standard query language) was formally adopted in 1986 by the American National
Standards Institute, as SQL-86, and is the most widely used database querying language (Lans,
16

2007). The SQL language framework provided a logical structure for querying information stored
in DBMSs. The first versions of SQL however, could not explicitly handle spatial data structures.
Several extensions of SQL were proposed including GEOQL, SAND, GEO-Kernel, and PSQL;
however the extension that proved most influential, and eventually adopted was Egenhofer’s
Spatial SQL (Adam & Gangopadhyay, 1997; Egenhofer, 1994). Spatial SQL extended the domain
to include spatial operators and attributes. Egenhofer (1994) further defined the Graphical
Presentation Language (GPL), a set of tools in which the results of a spatial query could be
manipulated. The Open GIS consortium (OGIS) promoted spatial functionality by recommending
a set of critical reforms in SQL, mainly the adoption of Egenhofer’s spatial abstraction model,
which introduced GEOMETRY as a base-class for spatial objects (Egenhofer, 1994; OGIS, 1999;
Shekhar & Chawla, 2003). These recommendations were fully adopted in 1999 with the release
of SQL3 (Lans, 2007; OGIS, 1999).
Data sharing is a critical consideration for our research group as we maintain
collaborations with institutions and researchers throughout much of the United States and East
Africa; yet the biggest problem we face as a group is the ability to share data and analysis. We
require a new medium that facilitates this flow of information without the need to physically
carry the data between institutions. Internet-based GIS and data servers are one solution we
employ to allow for simultaneous interaction and analysis of the data. Internet-based GIS
emerged with the release of two projects, GeoChange (Drew & Ying, 1996) and the Alexandria
Digital Library (Smith & Frew, 1995). GeoChange in particular set the standard in terms of
functionality and usability with an interface supporting batch processing, import/export, and
automatic metadata generation (Adam & Gangopadhyay, 1997). The Alexandria Data Library,
17

though spare in relation to GeoChange, became widely adopted by the University of California
system for in-house sharing of proprietary spatial data. Today, despite the improvements in
functionality, most spatial databases still lack the ability to efficiently handle the raster data
structure. Two ongoing projects promise to bring this functionality into spatial DBMSs (SDBMSs)
in 2010. Of particular interest to us is the WKTRaster project, a community supported opensource project by the Open Source Geospatial Foundation (OSGEO). WKTRaster extends the
functionality of the PostGIS library to implement the RASTER type class in the same way
GEOMETRY was done in order to support spatially explicit handling of vector data (Michael
Stonebraker & Moore, 1995). WKTRaster, when fully implemented, will facilitate consistency in
data handling of all spatial and non-spatial data types, similar to the manner in which spatial data
are managed with the georaster loader included in the Oracle Spatial 11g package.
While advances in scientific understanding and new technologies have helped to curb the
spread of infectious disease in the developed world, the same cannot be said about the
developing world. As the incidence of infectious diseases decreased in the developed world,
fewer resources and attention were devoted to combating those diseases in the developing
world, contributing to an increase in the prevalence of many illnesses there (Cohen, 2000). In
order to develop better strategies for combating disease, we need to enhance our understanding
of the underlying ecological conditions that contribute to the emergence of diseases and deliver
solutions practicable in developing world contexts (Ostfeld et al., 2008). Disease ecology is
generally not considered a discipline in itself but rather seeks to understand the relationship
between disease epidemiology and the landscape (climate, physical, and human) (Johnson &
Thieltges, 2010; Keesing et al., 2006; Sutherst, 2004; Tatem et al., 2006). The trans-disciplinary
18

nature of the field creates a unique set of problems, many of which pertain to the use of data
while maintaining the rigorous standards mandated by Institutional Review Boards, HIPAA, and
international research and privacy standards. Very little has been published that directly explores
these types of management problems within the disease ecology literature. Routinely, issues of
scalability, reliability, and security emerge that hinder the effective dissemination of federally
funded data and models. Storage of large quantities of data must at a minimum facilitate the
range of applications necessitated by the questions posed in disease ecology. Data scalability
speaks to the ability of researchers to address questions at multiple scales of spatial or temporal
resolution, depending upon the question being asked; the storage of such data must facilitate
the rapid, concurrent access and integration of the data across varying resolutions (Shekhar &
Chawla, 2003). Reliability requires mechanisms to ensure that data mismatches or inappropriate
analytical methods are identified or prevented (Devillers & Jeansoulin, 2006; Shi et al., 2002).
Furthermore, data reliability, particularly with concurrent usage and modification of the data,
necessitates mechanisms for ensuring the integrity of the underlying data over time (Olson, 2003;
Shi et al., 2002). Finally, security issues arise when interacting with individually identifiable
human data or sensitive community data stored or generated within the DBMS (Olson, 2003).
While institutional guidelines and privacy laws may restrict access of the data to pre-approved
users, the limitations should not preclude non-privileged users from asking broader questions,
which may interact with the underlying data when aggregated to remove identifiable data or
other information that may be restricted by institutional guidelines or laws (e.g. ethnic identity
at low densities in census block group data). Finally, privacy restrictions should be scalable,
changing dynamically with the user and scale of resolution requested.
19

In collaboration with the International Livestock Research Institute in Kenya (ILRI), we
have accumulated an extraordinary volume of data for Kenya. Irrespective of theme,
international collaborations often present unique problems in terms of the management,
sharing, and dissemination of data necessary to carry out analyses. Our framework for a data
management system is a novel solution for spatial modeling in disease ecology, and the use of
open-source software exclusively makes this a cost effective solution for sharing with
international collaborators and organizations with limited budgets. The entire suite of data and
models is designed to be packaged electronically or on a portable drive to facilitate electronic
transfer or physical transportation.
A framework for data management cognizant of these issues and flexible in the use of
restrictions is an ideal solution for working with data types characteristic of research in disease
ecology. As part of the National Institutes of Health “Roadmap” program and with the National
Institutes of Health General Medical Sciences support, we are developing a multi-scale predictive
model that defines the relationship between climate change, land use and land cover change,
social systems, and the distribution of tsetse flies and sleeping sickness across Kenya (Makido et
al., 2007). Here we present a case study for the implementation of a generalizable disease
ecology DBMS framework that provides scalability, reliability, and security to optimize
interactions between users and the data.

Case Study: A Model for African Trypanosomiasis in Kenya
African Trypanosomiasis (AT), or sleeping sickness, is a major threat to human health across
Africa, particularly among impoverished peoples (Brun et al., 2010; Gyapong et al., 2010).
Typically considered a disease of the past, its prevalence has increased in recent years,
20

particularly in East Africa, due to the declining emphasis on trapping and control, climate, and
anthropogenic factors (Batchelor et al., 2009; Bauer et al., 1992; WHO, 2005). Although
monitoring has improved, the extent to which AT impacts East Africa is largely unknown. Recent
contributions by foreign countries and aid organizations directed towards addressing AT have
declined dramatically in contrast to the increased attention towards AIDS, malaria, and other
diseases (Siringi, 2003; WHO, 2001). The WHO has responded by designating AT a neglected
tropical disease (Brun et al., 2010; Kennedy, 2005; WHO, 2006). Not fully understanding the
ecological processes that contribute to the spread of AT may result in the inefficient application
of control regimes and misallocation of resources; thus retarding the efforts of the African Union
to combat and control AT (Cox, 2004). As Trypanosomiasis has increased in prevalence, the
impact on human and animal populations has been considerable, resulting in severe economic
hardship for rural families throughout East Africa (Campbell et al., 2000; Campbell et al., 2004).
Tsetse flies, Glossinidae family, are the primary vectors for the cyclical transmission of
African Trypanosomiasis. The general distribution of tsetse has been demonstrated in terms of
the biophysical extent and the presence of suitable hosts (Cecchi et al., 2008; KETRI, 1996).
However, the precise limits, historical and contemporary, have not been formalized
experimentally (Wint, 2001). Furthermore, the current distribution belts reflect outdated data
and methodologies (Joint WHO Expert Committee and FAO Expert Consultation on the African
Trypanosomiases (1976: Rome), 1979; Muriuki et al., 2005; Wint, 2001). Through prior studies
our research group has been able to describe the inaccuracies of these distribution limits,
particularly in terms of seasonal changes (DeVisser & Messina, 2009; Moore & Messina, 2010).

21

Furthermore, global climate change is shifting tsetse habitats, though the degree to which this is
occurring is unknown (Sutherst, 2004).
To adequately understand the mechanisms behind the increasing incidence of AT, it is
important to consider a diverse range of inputs from social, physical, climatic, and even political
dimensions. The range of scientific disciplines and methodologies required necessitates an
extensive volume of data be created, collected, and maintained. The management of the volumes
and types of data, physically and logistically, has proven to be a significant challenge and the one
in which we address in this paper. Thus we present a conceptual model for a comprehensive
open-source, computing environment that promotes efficient organization, storage, and
retrieval of disparate data. Furthermore, we extend the discussion of spatial databases by
presenting a model framework for a spatial DBMS that rigorously and consistently manages both
spatial and non-spatial data.

Data Holdings and Acquisitions
We have collected all publicly available data and a significant portion of the known privately held
relevant AT disease ecology spatial data for Kenya. The data fall into a classification scheme
defined by topography, soils, vegetation, climate, ecological diversity, water resources, and
anthropogenic factors known to control or influence the ecological processes driving tsetse
distributions over time and space (see Error! Reference source not found. for a summary). Nonspatial data consist primarily of governmental and intra-agency reports obtained through private
libraries in Kenya. The reports collected focus on policies and governmental/community control
and eradication programs. Currently, these reports are neither catalogued nor indexed, limiting
efficient use of any information they may contain.
22

Table 2.1: A summary of our data library grouped by major theme
Biophysical
Precipitation totals
Monthly Temperatures
Evapotranspiration
Temperature Scenarios
Precipitation Scenarios
Historical Climate Data
Agro-Climatic Zones
Agro-Ecological Zones
Lithology
Soils
Landforms
Forest Range
Wetlands

Social
Population Density
Population Predictions
Historical Population
Town Locations
District Census Data
Livestock
Wildlife
Agricultural Production
Poverty
Various Cadastral data
Entomological
Tsetse Distribution
Tsetse Habitat Suitability

Geographical
Land Use / Land Cover
Land Use Projections
Satellite imagery
Lakes
Rivers
River Basins
Roads
National Parks
Railroads
National Boundaries
Administrative Districts
Elevation
Topographic Maps
(Countrywide)

Remotely sensed image data comprise the majority by physical file size of our data
collection. We possess the majority of the known publicly available aerial imagery for Kenya from
the Landsat, MODIS, PALSAR, and ASTER platforms, including a number of image products
summarizing land use classification (MODIS types 1-5 for 2001 to 2005 and Landsat MSS derived
1 km for 1980), land surface temperature (MODIS LST), precipitation (WorldClim 30yr average at
1 km), and vegetation indices4 (NDVI for 2001 to 2008 every 16 days, 250 m resolution).
Additional sources of land use and land cover information are provided with Africover as vector
or raster data types, GLC2000, CLIP cover1, and UMD Global Land Cover. We possess elevation
data from ASTER, SRTM, and digitized topographic maps (30 m, 90 m, and 250 m spatial
resolutions, respectively). With regards to the distribution of tsetse, we possess vector and raster

4

Data are publicly available
23

digitized distributions of fly belts for Kenya for 1967, 1973, 1996, and 20005. Finally, there are
rasterized estimates of livestock densities (# per 𝑘𝑚2 ) for 2007 (ILRI6). Utilizing these data,
DeVisser et al. (2010) developed the TED model to predict the distribution of tsetse in Kenya.
Figure 2.1 shows the predicted minimum range (the areas where tsetse persist all year) between
2002 and 2009. To effectively and efficiently recover and maintain the value of the data, we
require a solution that not only provides for efficient storage and retrieval of the data, but which
also allows for automated metadata generation.
We possess 103 spatial datasets of socio-economic and demographic assessments of
Kenya between 1971 and 2008. A portion of these data were provided by the Integrated Public
Use Microdata Series (IPUMS) International dataset for Kenya and originate from the Kenya 1989
and 1999 census collected by the Kenya National Bureau of Statistics. The IPUMS data are a
systematic sample of every twentieth household, which represented a sampling fraction of 5%
and expansion factor equal to 20; a long form questionnaire was implemented surveying
individuals within households. Location data for each individual is limited to respondent’s
province and district, the first and second administrative levels respectively, of five possible levels
each at an increasingly finer spatial scale of resolution. Data at finer spatial scales are not
available as part of the Kenya Bureau of Statistics’ effort to maintain privacy. A further
complication was a change in the number of districts in Kenya from 42 in 1989 to 69 districts in
1999; however, this change was made by subdividing existing districts allowing rough
comparisons between associated regions. The IPUMS sample provides 97 household and

5
6

Not all maps are publicly available
International Livestock Research Institute (Nairobi, Kenya)
24

individual variables, most importantly geographic information (urban-rural status, province,
district), utility (electricity, water supply, sewage type, and type of cooking fuel), and dwelling
(number of rooms, toilet type, floor, wall, and roof material). Other relevant data include

Figure 2.1: Maximum extent of tsetse distribution predicted by the TED Model between the
beginning of 2002 and the end of 2009. As described by DeVisser et al. (2010), the TED Model
uses five scenes of MODIS 1km annual land cover, 207 scenes of MODIS 250 m Normalized
Difference Vegetation Index (NDVI), and 207 scenes of MODIS 1km day/night Land Surface
Temperature (LST) products to predict the fundamental niche of tsetse in Kenya, and a fly
movement model to predict tsetse distributions or realized niche of the tsetse species of interest.
The TED Model is written in Python scripting language and is run within ArcGIS 9.2 (or later
versions of ArcGIS). Other data sets used to construct the map include shapefiles of Kenyan
major roads, highways, cities, rivers, Kenyan water bodies, lakes, and African country political
boundaries. To construct the background topographic relief map, the Shuttle RADAR
Topographic Mission (SRTM) 90 m Digital Elevation Model (DEM) was used to create a gridded
hillshade product, and a dry season NDVI scene was combined with the SRTM DEM to create an
elevation/vegetation color scheme.

25

individual variables describing household position, demographic characteristics, education,
employment, migration, and disability (see Table 2.2 for a sample). There are a total of 1,074,048
individual entries for the 1989 census and 1,407,597 for the 1999 census samples.
Decisions on data management often conflict during collaborative research, resulting in
the lack of a cohesive strategy for data management and the inability to share such data
effectively. While our colleagues in Kenya have the technological skills to work with spatial data,
the telecommunication network is insufficient to provide the necessary bandwidth or reliability
to acquire the data via direct transfer over FTP or other similar protocol. In order to overcome
this problem in the short term, it is necessary for us to carry the data into the country on physical
media, and to have it accompanied by a data management system that can facilitate interaction
with the large quantity of data. Thus, efficiency and portability are significant concerns.

Development of a Spatial Database System
Our solution for a spatial DBMS involves bridging a variety of software packages following the
basic framework as described by Câmara et al. (1996) for the development of the TerraLib GIS
Library and the integration considerations posed for the MurMur project (Parent et al., 2006).
First, we outline the conceptual framework, and second, the implementation of the design. Third,
we describe the development of routines, batch or other preconfigured shell scripts that can be
selected to run either from the command line or through an interactive GUI prompt, whereby a
user can add and recall raw data files, or query the database to return a mash up of spatial data
files or metadata. Finally, we develop a set of SQL triggers (code set to run when activated by a
defined action), to enforce data integrity.

26

Conceptual Model
Figure 2.2 demonstrates our conceptual model for a spatial DBMS. In contrast to previous
implementations of MySQL, Postgres, and other common spatial databases, modern DMBS
models facilitate raw data and metadata to be stored together, embedded in the database
(Elmasri & Navathe, 2004; Watson et al., 2004). The proposed spatial computing environment
uses open source, community supported software and standards, providing a solution to the data
management problem that is temporally extensible. The database is portable in so much that we
can copy the database and software binaries to a portable drive that can be carried to Kenya. Of
critical concern to us in the selection of software components was the interoperability of the
system, the ability of components to interface and work together. Our selected DBMS is
PostgreSQL (Postgres), an advanced, readily available, open source, object-relational database
management system. Utilizing standard SQL syntax, Postgres allows for complex query
capabilities, including spatial queries, and facilities strict rule and primary key enforcement.
Postgres is also extensible, allowing for the addition of new functionality (Michael Stonebraker &
Kemnitz, 1991; M Stonebraker & Rowe, 1986).

27

Figure 2.2: A conceptual model framework for the spatial DBMS.

PostGIS ("PostGIS," 2008) is an extension to the Postgres language that adds functionality
for the storage and retrieval of spatial data files. PostGIS is, at its core, a suite of tools that serve
as the backend for spatial functionality in Postgres. Of particular interest is the WKTRaster (Beta
0.1.6) project, which extends the ability of a Postgres database to store and index raster data, a
first of its kind. The project mirrors the inherent vector-based functions (of GEOMETRY type) for
raster data. The result is a single set of SQL functions that handle both spatial data types. This
extension has the potential to greatly enhance and facilitate the utilization of raster data by end
users. Figure 2.3 presents the conceptual model for the user interface to the Postgres DBMS. We
incorporate a variety of software packages, explained later, each of which provides the user with

28

statistical, visual, or geoprocessing capabilities; the user can interact with these packages through
a GUI or through a command line interface.

Figure 2.3: Flow of data through the model implementation as users interact with the system. It
should be noted that flows are one-directional, meaning that although users can interact with the
data to generate analysis, the results must be stored as a separate entity in the database. This
ensures that the underlying data cannot be changed.
GRASS (Geographic Resources Analysis Support System) is one of the few open source
options for the handling of raster GIS data. Developed by the Open Source Geospatial
Foundation, GRASS is an increasingly popular solution for the use and analysis of spatial data in

29

academic research. Like the many other components selected for the system presented here,
GRASS is interoperable with Python and Postgres and is extensible, allowing users to create and
add new functionality.
Although there are a plethora of statistical packages available, R is our preferred statistical
package in large part because it interfaces easily with GRASS and Postgres. We can do so without
needing to install any additional software components. R is an open-source solution, developed
by an international community of users, to create an alternative to the often expensive and
restrictive programs offered by statistical companies. Although R lacks a graphical interface, it
uses far less computer memory than most other comparable statistical packages. This allows us
to perform complex analyses with fewer hardware demands, an important consideration for
maintaining portability of the project.
Finally, Python ("Python," 2010) is an object-oriented programming language developed
by Guido van Rossum in 1991 and maintained, in large part, by the open source community.
Python is an efficient scripting language that facilitates interoperability between our database
(Postgres), statistical analysis software (R), and GIS (Grass) (Neteler et al., 2008). Python is a
highly intuitive, object-oriented programming language, easy enough to learn and use that it
makes for a good solution to bridge our project components. Furthermore, the open-source
nature of the language fits well with the other components, enabling us to incorporate additional
extensions written by the community.
The most critical consideration for long-term data storage is persistence and security
(Elmasri & Navathe, 2004; Watson et al., 2004). Digital data are inherently ephemeral in that over
time physical storage media will fail or degrade, thus requiring continuous rewriting on digital
30

media to ensure persistence. Though drive technology has progressed significantly in the past
decade, drive failure is not uncommon; the volume of data and frequency with which the data
are accessed puts immense stress on the physical mechanisms. Therefore, it is necessary to
employ a strategy that ensures the long-term viability of the data. To this extent, we employ a
RAID array (Redundant Array of Independent Disks) as our secondary storage medium, mirroring
data between groups of drives. This enables corrupted data resulting from disk failure to be
recovered in real-time, without the need to create tertiary backup regimes. Furthermore, we
enforce constraints to access of the data in making the data read-only, reducing the chance a
write error will occur or values inadvertently be changed. Finally, all files will have MD5
checksums (a cryptographic hash code generated from a file’s binary data) included as an
attribute of the file allowing us to verify the integrity of any data file (Rivest, 1992). The strategies
employed here will also enable us to protect against accidental user error, which may result in
inadvertent modification or deletion of data. Security restrictions, though a good first line of
defense, are frequently circumventable. The ability to rollback changes within Postgres, as well
as being able to restore data from the RAID, will ensure the long-term integrity of the data library.

Database Standards
Generic frameworks for database development, such as the one we outline here, should use
established ontologies in their component descriptions, which we satisfy by conforming to the
formal ontology described by the Open Geospatial Consortium (OGC). The OGC is an independent
group tasked with the purpose of developing and maintaining a set of standards for the
management of spatial data to promote both consistency and interoperability across GIS
platforms. GDAL (the Geographic Data Abstraction Library) stands as a single standard for
31

interoperability of raster data within the GIS community. As a library it facilitates the conversion
between data products. However, as is the case when working with proprietary data, conversion
between formats requires a commonly understood intermediate. The GDAL standards and
abstraction libraries utilized in our project facilitate this conversion between data formats by
providing a commonly understood intermediate. Custom GDAL libraries are largely used in our
implementation of GRASS as a mechanism to add support for a range of occasionally unsupported
data formats.
There are an inordinate number of disparate standards for metadata generation and
inclusion, none of which are remotely universal. Regardless, it is necessary for our own data
management that we accept and enforce a single standard for metadata management. Since our
DBMS facilitates the storage of raw data within the DBMS, it precludes the need to store
metadata separately as this information is included within the database as discrete variables (the
raw data are technically also included as an attribute value). However, it is foreseeable that we
may need, on occasion, to transfer data outside the realm of the database. Under this scenario,
it is necessary to have a mechanism to recreate the metadata files discarded earlier. Thus, we
incorporate custom scripts that can be called to assemble the necessary metadata precursor
information from the data table. The resulting metadata report meets the formatting and
content criteria specified by the OGC and ISO19115 specifications.

SQL-Rule-Based Interactions and Scripting
In an effort to ensure the integrity of the raw data and prevent accidental deletion or
modification, all users are restricted in the ways they can interact with the DBMS. In the majority
of cases, raw data are restricted to read-only access by users at the database level. One way we
32

achieve this is by restricting UPDATE and DELETE privileges to the database administrator account
only. This ensures that raw data cannot be changed or deleted through unverified scripting. It
further facilitates the simultaneous access and use of raw data by multiple processes. Figure 2.3
symbolizes the flow of data from the DBMS to the user. While the raw data are read-only, it is
useful to permit users to save the output of models to the database with the option to link to the
source data. Indexing these data sets together is useful when new users explore the data. We
will discuss this functionality further in the later section “A User Perspective: Interfacing with the
DBMS”. Metadata for these model outputs will vary, but will at a minimum include the user
identity and model code, as well as a summary of the model objective. A set of predefined rules
or triggers is loaded into the DBMS and provides enforcement of the desired constraints (e.g.
spatial/temporal mismatch encountered during scripted analysis). These rules serve to provide a
minimal standard of validity and consistency for model output and statistical analysis (Devillers
& Jeansoulin, 2006; Shi et al., 2002).
The storage of raw data within the DBMS precludes the need to regularly interact with
different data formats. Nevertheless, it may be useful to incorporate a mechanism whereby we
can convert data among data formats or export data into a range of formats. While the majority
of these tasks can be accomplished within the GRASS interface, we include in our library a set of
python scripts to extend our ability to move among data formats (particularly useful for the range
of raster imagery available). For example, our “data bank” includes raster imagery (*.img, .tif,
.hdf, among others), vector data (*.shp and others), text documents (*.doc, .docx, .txt, .pdf, and
others), metadata files corresponding to acquired imagery (*.xml, .pdf, .txt, and others), as well
as an assortment of other types not specifically mentioned here.
33

DBMS Interface – A Manager Perspective
Implicit in the design of our database are SQL rules (and triggers) that constrain the ways that
users can work with data in an effort to prevent common mistakes. Since the database holds data
at varying resolutions and extents, we constructed rules to check for common errors committed
by users in selecting data layers. In the event that data layers are deemed incompatible, the user
is alerted to the mismatch and encouraged, though not required, to restate their request. These
rule sets operate at the point of data retrieval and storage. When importing data, we check and
request the user define the relationship between spatial data and metadata. If metadata are not
available, the user is prompted to input known characteristics of the data file in an attempt to
force this paired relation. Common examples of information prompted include the timestamp of
data acquisition or generation, solar or view angle (for satellite derived imagery), or a definition
of codes that may occur within the data table (often the case with census data). Some of these
data can be retrieved from the raw data, such as extent, summary statistics, file type, and others.
Enforcing these relations at the time of storage will greatly reduce the number of unpaired spatial
data files and corresponding descriptors.
These rules sets also operate on a variety of demographic or other similar data types
stored in our database. In our specific implementation of the database, we have sample volumes
of census data collected by the Kenyan Government and acquired from IPUMS. As data is input
into the database, users are prompted as to the type of data being input and plain English
definitions of the variables (a long form description of the purpose of the data) included in the
dataset. We developed a means to query the data file for variables and return this list in memory

34

to the rule set. In the event the data file does not return the correct list of variables, the user is
prompted to specify the range of variables or create them.

DBMS Interface – A User Perspective
While many incantations of interfaces are possible, one means by which users can interact with
data is through a web browser, which connects to the database via an application developed for
Mac OS. Through the application, users have access to all the tools needed to query, utilize, and
analyze data from the DBMS. From a set of menus, the user is prompted to select a subset of
data. Next, they are given the opportunity of either selecting from a set of precompiled models
or statistical scripts that can analyze the data, or the data can be brought forward and
immediately visualized in the browser window. Finally, the user is provided with command line
tools that can augment their interaction with the data. This approach to data interaction lowers
the learning curve for new users and gets them instantly connected with the data.

Statistical and Analytical Analysis
We extend the functionality of our DBMS to assist users in browsing the library of data by using
the R package to automatically calculate descriptive statistics. These summaries are potentially
most valuable to users not involved with the production of the data, or who may not be familiar
with a region of interest. Furthermore, providing a means to compute descriptive statistics
automatically enforces consistency between files that is often a symptom of user error when files
are independently managed.
Perhaps the most common challenge users face in interacting with databases is retrieving
data both correctly formatted and appropriate for use in a particular analysis (Longley et al.,

35

2005). The scripted analysis tools help to bridge the gap in understanding for users not familiar
with the R statistical package. With a simple menu prompt, the user is able to specify which data
files and statistical methods are to be employed by the program. Though not a comprehensive
selection, it provides a means to conduct a range of simple statistical comparisons. Further
analyses can be performed though the R package directly linked to the database.

Implementation
To enable the reader to implement our DBMS model, we provide a general outline of the steps
needed to install and configure the software packages. We make no assumptions as to the
hardware on which the model is implemented. Our solution requires a number of software
components, including PostgreSQL, PostGIS (with GDAL and WKT Raster extensions), Python,
GRASS, and the R statistical package. In this section we will detail the required steps for the
implementation of each component within open source Debian distributions of Linux.

Software Packages
Debian distributions of Linux take advantage of the aptitude system for software distribution,
and therefore you can install all packages from the terminal. To install:
sudo apt-get install python postgresql
sudo apt-get install gdal-bin postgresql-8.4-postgis postgis grass rbase

You may also be interested in installing the Quantum GIS (QGIS) program, another convenient
GUI that interfaces with GRASS, which can also be installed with aptitude:
sudo apt-get install qgis

36

In order to allow interaction between GRASS and the GDAL libraries, you need to build and
configure the GRASS plugin for GDAL. Download and follow the installation instructions provided
by:
http://trac.osgeo.org/gdal/wiki/GRASS

Finally, you need to install optional extensions to the R package. Start R from a terminal by typing
R; then install packages with:
install.packages(“ctv”)
library(ctv)
install.views(“Spatial”)

Initializing Postgres
You will need to login to Postgres for the first time using the default “postgres” login. From the
terminal window you will then be able to create user accounts, set access restrictions, and the
database for the project data. First start Postgres from a terminal by typing:
psql -U postgres

You can now create your user accounts. In our implementation of the project, all users are
granted limited permissions by default. Additional rights can be given later depending on the
needs of the user.
CREATE USER <username>;
CREATEDB <dbname>;
GRANT ALL to <username> ON <dbname>;

It is necessary to configure the database to enable spatial functionality with PostGIS. The
instructions given are also available from the PostGIS documentation at:
http://postgis.refractions.net/documentation/

37

First, type:
CREATELANG plpgsql <dbname>

Now, copy the PostGIS spatial definition files into your new database and navigate to the PostGIS
installation directory specified during installation and type:
psql -d <dbname> -f postgis.sql
\q

#Quits the Postgres session

For additional security options that are available, you may refer to the Postgres documentation7.
You may now login to your new database from a terminal window by typing:
psql -U <username> <dbname>

Use caution when performing upgrades to PostGIS as you will likely have to rebuild support for
your database.

WKT Raster extension for PostGIS
WKT Raster is not yet included in the PostGIS pre-compiled binary as it is (as of 09/15/2010) still
in beta testing. Therefore, you will need to download and compile the extension. The following
instructions are available from the WKT Raster project documentation8. Depending on your
specific system configuration, it may be necessary to build the PostGIS libraries from source as
well as the required dependencies. Download the source code from:
http://www.postgis.org/download/

Unpack the tar file and navigate to the WKTRaster directory. Generate a configuration profile by
typing:
./autogen.sh

7
8

http://www.postgresql.org/docs
http://trac.osgeo.org/postgis/wiki/WKTRaster/Documentation01
38

Now run a configuration script. You will need to locate the installation directory for PostGIS:
./configure --with-postgis-sources=/thesrc/postgis-version

If no errors were generated, you can install WKTRaster. Since you are already in the installation
directory, simply type:
make & make install

Configuring GRASS
The first time you start GRASS, you will have to specify a location and path for your data. Choose
the data path that you setup earlier (it should already be the default entry). In order to create a
new location, it is easiest to have a georeferenced data file that spans the region of interest.
Although not necessarily required by the software, it greatly simplifies the process of adding a
spatial reference system to the data library to specify it right away. Start GRASS by navigating to
the icon or from a terminal:
grass64 –gui

Click on the Location wizard icon on the right hand side of the window. Follow the on-screen
instructions, selecting your georeferenced data file when prompted. Alternatively, you can create
a new location by entering grass with the default example “spearfish60”. From the GRASS
terminal navigate to the directory where your data is stored and type:
r.in.gdal i=<yourimage> -o <mapname> location=<newlocation>

The function r.in.gdal will parse the image file and automatically define the region based on
the extent of the selected image.

39

Loading Data and interfacing GRASS with Postgres
There are a great many possibilities for importing data and interfacing with GRASS. Here we
demonstrate several examples for importing data into Postgres and accessing those data within
GRASS. Landsat 7 data are downloadable from the USGS as georeferenced TIFF for individual
bands. To import them into a sample database “KENYA” utilize a python loader script9:
gdal2wktraster.py -r *.tif -t Landsat -s 4326 -k 100x100 -I >
Landsatloader.sql

The script does not directly load the data into the database; rather it creates the necessary SQL
script to do so. In this example, the -r option enables multiple files (selected with the asterisk) to
be imported simultaneously into a single table. The -t option specifies the table name the data
will be imported to. With the -s option, we specify the spatial reference system using SRID
numbers (Spatial Reference Identifier, OGC specifications), WGS84 in this example (Herring
2006). The -k option splits each raster into tiles that are 100x100 pixels. Finally, the -I option
requests that a spatial index file be created for each raster tile. Next pass the SQL loader for
processing with:
psql -U <username> -f Landsatloader.sql <yourdatabase>

If you do not want to create a spatial index or forgot to do so in the first step, you can easily
create one at the Postgres prompt by:
CREATE INDEX Landsat_SI ON Landsat USING GIST (ST_ConvexHull (rast));

In this example, the GIST (Generalized Search Trees) index is used, which has a balanced tree
index structure similar to a B-tree (Hellerstein, 1999; Kornacker, 1999). An example of a common

9

The loader is now called “raster2postgresql” and is installed by default with PostGIS.

40

format vector data type is the ESRI shapefile and an example is a digitized map of fly belt regions
FLY.shp. Postgres facilitates the importing of vector data from the terminal with:
shp2pgsql -s 4326 -I -D FLY.shp <yourdatabase>.flybelts > flybelts.sql

Here the -s flag specifies the spatial reference system using SRID codes. The -I option requests
the script initialize a GIST spatial index on the geometry column of the data. Finally, the -D option
creates a dump file (SQL loader) that can be imported into Postgres from the terminal, a faster
means of adding data to the database. Now pass the SQL loader to Postgres with:
psql -U <username> -f flybelts.sql <yourdatabase>

Alternatively, import vector data using a graphical interface by loading:
shp2pgsql –gui

To load the data into GRASS, initialize the Postgres driver from within the GRASS terminal. Start
GRASS specifying a work location (see previous instructions for the generation of LOCATION).
Now, load the driver defining the connection between Postgres and GRASS:
db.connect driver=pg database=“host=localhost, dbname=<yourdatabase>“
db.login user=<username>
db.connect -p
db.tables –p

After initializing the connection, it is now possible to query the database; for example, retrieving
the flybelts data described earlier:
v.in.ogr dsn=“PG:host=localhost dbname=<yourdatabase> user=<username>“
layer=????? output=flybelts type=boundary,centroid
v.db.select flybelts
v.info -t flybelts
d.vect flybelts

41

Data in the HDF formats can only be read with additional GDAL libraries, which are not included
with the standard distribution. However, the version of GDAL made available through the Ubuntu
repositories appears to support some limited functionality. If further interaction with HDF is
required, you will need to compile GDAL manually, inputting development files downloadable
from the HDF Group10. HDF formats are containers, and thus may hold multiple data sets. Header
data are accessed with:
gdalinfo sample.hdf

Sub dataset names are formatted as:
HDF4_SDS:subdataset_type:file_name:subdataset_index

A portion of the output reads:
SUBDATASET_8_NAME=HDF4_SDS:MODIS_L1B:GSUB1.A2001124.0855.003.200219309
451.hdf:7

SUBDATASET_8_DESC=[408x271] Range (16-bit unsigned

integer)

Detailed headers for this sub dataset can be viewed with:
gdalinfo
SUBDATASET_8_NAME=HDF4_SDS:MODIS_L1B:GSUB1.A2001124.0855.003.2002
19309451.hdf:7

Since the GRASS plugin was compiled with HDF support, image data, by individual band, are
directly imported into GRASS with r.in.gdal.
r.in.gdal
HDF4_SDS:MODIS_L1B:GSUB1.A2001124.0855.003.200219309451.hdf:7
out=hdfexample

10

http://www.hdfgroup.org

42

Once the data are loaded into the GRASS interface, they can be imported into Postgres either
directly with db.connect or by exporting the image as .TIF and importing with the
gdal2wktraster.py script.

Limitations and Future Expansion
The initial phase of our project is limited to the objectives addressing the misuse,
misrepresentation, and effective archiving of our data library. As advances and improvements
are made to the telecommunications infrastructure in Kenya, we will be able to share a common
easily accessible repository for data with our colleagues outside the United States. The DBMS
framework, including necessary software packages and data, can be packaged together and
distributed via portable hard drive. Future updates to our software model will include a webbased interface that will allow users to interact with the DBMS, including the suite of analysis and
visualization tools (R and GRASS), without the need to install and configure these programs
locally. This reduces the hardware requirements for working with the data, potentially allowing
a broader base of users to share in data access. This approach has been extensively applied in
many of the institutional projects referenced commonly in the literature (Câmara et al., 1996;
Parent et al., 2006).

Summary
Disease ecology is a trans-disciplinary field, exploring the complex interactions between diseases
and the environment. Computational and collaborative barriers inhibit meaningful advances in
the field. Major problems include data management schemas to facilitate the scalability (data are
re-sampled dynamically to avoid redundant storage), reliability (concurrent access to data

43

permitted while ensuring the raw data cannot be changed), and data security (the database
allows for a dynamic security access policy while meeting the HIPAA and IRB requirements). As
data become aggregated with decreasing spatial resolution, many of the privacy concerns
disappear but tracking and management problems proliferate. The DBMS must dynamically alter
the restriction rule-set to account for the aggregation and application challenges. Previous
implementations of data management systems required that multiple instances of the data be
stored, creating a problem of exponentially increasing data storage demands. Currently, no
framework for data management that addresses this set of integrated concerns exists.
Over the course of our research on African Trypanosomiasis, we have accumulated a large
library of data. Our database model utilizes open-source software so as to allow for flexibility and
extendibility to the model implementation. We take advantage of the new PostGIS extension,
WKTRaster, to allow for the storage of raster imagery within the database. This allows us to
enforce a single standard in the way all data formats, irrespective of the contents, are managed.
With this development, we are finally able to store the entirety of our data library, spatial and
non-spatial, explicitly within a single database implementation, and the restrictions and rule-sets
coded into the DMBS should ensure the long-term integrity and security of the data.

The overarching goal of this project is to create a multi-scale predictive model for the
tsetse and African Trypanosomiasis to provide a means whereby governments, communities, and
NGOs can make informed decisions for disease control or suppression that are spatially and
temporally aware. Given the variable background and technical expertise of the different groups,
our solution should be simple enough for the most basic user, yet powerful enough to be useful
44

for complex analyses by the most skilled. To maximize the utility of this system, we will utilize a
participatory design framework to develop Mac, iPhone, and Web applications for interfacing
with the models and data.
Open-source software is uniquely capable of rapid adoption of new technologies and
functionality due to the base of developers working on modular extensions to the software base.
Spatial databases are increasingly common. Mobile data applications are becoming increasingly
location-aware (e.g. Facebook, Twitter, Loopt, Turn-by-turn Navigation (Waze), among others),
with the purpose of providing users with context-specific information and opportunities. These
GIS technologies increasingly promote web-based interfaces to a spatial database. Perhaps the
most exciting vision for GIS is the adoption of mobile technologies, now incorporating GPS
technology, to provide for context-aware interaction with spatial database systems. The next
step in our project is to implement an efficient, two-way web-portal for the spatial DBMS that
will allow us to interact with data and model results from mobile devices in the field. Combined
with automated scripting and model execution, this would have the potential to dramatically
increase the amount of information shareable with local communities. While not yet possible
due to barriers in Kenya, we work towards the vision of achieving synchronicity between science
and practice.

45

Table 2.2: A sample of the subset of Kenya census data we hold from the 1990 National Census

1
2
3
4
5
6
7
8
9
10
11
12
13
14

cntry
404
404
404
404
404
404
404
404
404
404
404
404
404
404

year
1989
1989
1989
1989
1989
1989
1989
1989
1989
1989
1989
1989
1989
1989

sample
4041
4041
4041
4041
4041
4041
4041
4041
4041
4041
4041
4041
4041
4041

serial
1000
1000
1000
1000
2000
3000
3000
3000
3000
4000
5000
6000
6000
6000

persons
4
4
4
4
1
4
4
4
4
1
1
12
12
12

wthh
20
20
20
20
20
20
20
20
20
20
20
20
20
20

subsamp
26
26
26
26
76
2
2
2
2
92
81
5
5
5

46

gq
10
10
10
10
10
10
10
10
10
10
10
10
10
10

unrel
0
0
0
0
0
0
0
0
0
0
0
0
0
0

urban provke distke ownrshpd electrc
2
1
1010
216
2
2
1
1010
216
2
2
1
1010
216
2
2
1
1010
216
2
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
140
2
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
216
1
2
1
1010
216
1

CHAPTER 3
UTILIZING VOLUNTEERED INFORMATION FOR INFECTIOUS DISEASE SURVEILLANCE

Abstract
With the advent of Web 2.0, the public is becoming increasingly interested in spatial data
exploration. The potential for Volunteered Geographic Information (VGI) to be adopted for
passive disease surveillance and mediated through an enhanced relationship between
researchers and non-scientists is of special interest to the authors. In particular, mobile devices
and wireless communication permit the public to be more involved in research to a greater
degree. Furthermore, the accuracy of these devices is rapidly improving, allowing the authors to
address questions of uncertainty and error in data collections. Cooperation between researchers
and the public integrates themes common to VGI and PGIS (Participatory Geographic
Information), to bring about a new paradigm in GIScience. This paper outlines the prototype for
a VGI system that incorporates the traditional role of researchers in spatial data analysis and
exploration and the willingness of the public, through traditional PGIS, to be engaged in data
collection for the purpose of surveillance of tsetse flies, the primary vector of African
Trypanosomiasis. This system allows for two-way communication between researchers and the
public for data collection, analysis, and the ultimate dissemination of results. Enhancing the role
of the public to participate in these types of projects can improve both the efficacy of disease
surveillance as well as stimulating greater interest in science.

47

Introduction
Recent publications surrounding Volunteered Geographic Information (VGI) broadly represent
the belief among some in the academic community that non-scientists can be engaged in and
benefit from spatial data analysis (Connors et al., 2012; Flanagin & Metzger, 2008; Goodchild,
2007a, 2010a), a field previously reserved exclusively for academics. Focus on VGI represents a
paradigm shift from viewing science as having a single authority (the scientist) to a model where
authority is relative and expressed contextually. Abundance, repetition, and the collective
assessment of data (as well as the ability to correct) convey credibility to information that would
not necessarily exist otherwise (Connors et al., 2012). In this sense, a non-scientist plays a role in
validating data collected by others, and collectively assessing data quality (Connors et al., 2012;
Craglia, 2007).
The concept of Web 2.0 incorporates bi-directional collaborations in which users
collectively collate spatial data, stored in a central cloud repository and accessible by anyone for
whatever purpose deemed worthy. The Web 2.0 paradigm is represented widely through web
projects such as Wikimapia, OpenStreetMap, and even Google Earth. Within the context of these
volunteered GISystems (VGIS), users contribute information to develop a collective knowledge
base. Recent advances in mobile technology have furthered the applicability of Web 2.0 projects,
enabling easier access to the information, and even allowing for novel uses of crowd-sourced
information (Rosenberg, 2011). Sui (2008) extends the paradigm to include “the wikification of
GIS”, a notion which he defines as being the shift in perception that only people who are
specifically trained to “do GIS” should interact with spatial data and perform analysis. It is upon

48

this notion, specifically, that VGIS endeavors to enhance the role of the user in the collection and
analysis of spatial data.
The use of volunteered information for disease surveillance draws upon themes in the
participatory GIS (PGIS) literature in suggesting that GIS technologies can operate in concert with
volunteered information and local knowledge (S Boroushaki & Jacek Malczewski, 2010; Connors
et al., 2012; Elwood, 2010; Flanagin & Metzger, 2008). The key distinction between classical PGIS
methods and VGIS involves the role of the scientist. We refer here to McCall’s (2005) discussion
of good governance through improving dialogue, legitimizing and using local knowledge, the
redistribution of resources, access rights, and new skills training in geospatial methods. These
concepts support the idea that a PGIS or VGIS approach can contribute to the adoption of new
technologies for disease surveillance.

Background
Traditional Paradigm
The traditional paradigm in GIScience partitions individuals into experts versus non-experts. In
an academic context, this treats scientists as the experts and citizens as non-experts. Under this
traditional paradigm, public participation in the research process is hindered by a number of
factors. Most importantly, the traditional roles of experts (scientists) versus the public leaves
little room to consider alternative knowledge bases (i.e., local knowledge). Furthermore, there is
limited opportunity for citizens to become informed, equal participants, thereby limiting the
potential applicability of any results/understanding gleaned from the research process (Soheil
Boroushaki & Jacek Malczewski, 2010).

49

Under the traditional GIS model, technology and software are not readily accessible,
requiring either a specific skill set or simply being priced beyond the consumer market. Therefore,
citizens are relegated to operating as consumers of information exclusively, or as indirect
producers, mediated by communication to researchers in small group projects. Their interaction
with the data in this regard is strictly as a provider of information, not as producers of spatial
data products. Finally, the traditional GIS model treats data validation as achieved largely through
reputation (Flanagin & Metzger, 2008). Scientists and researchers are perceived as producers of
reliable data due to past training in data collection and analysis. Furthermore, the peer review
process adds credibility by requiring outside researchers to assess quality. Broadly though, data
collected by researchers is assumed to be reputable because it is collected within the context of
academic endeavors, and done by trained individuals. Information of this sort is generally
accepted to be true until shown to be otherwise. With few exceptions, the vast majority of GIS
data products are produced under the traditional GIS model. Citizens are largely excluded from
the process of data collection and analysis (Connors et al., 2012). We do not, however, suggest
that the traditional model must be replaced. Instead, we propose the standard model be
extended to facilitate collaboration between citizens and researchers.

VGIS Paradigm
Volunteered GIS represents a paradigm shift from viewing science as having a single authority
(the scientist) to a model where authority is relative and can be expressed contextually.
Information abundance, repetition, and the collective assessment of data convey plausibility to
data that would not otherwise necessarily exist. In this sense, a non-scientist plays a role in
validating data collected by others; collectively assessing data quality (Oreskes et al., 1994). This
50

concept is explored further by Craglia (2007) in his assessment of individuals as geosensors,
empowering them to validate global models using their own perceptions or impressions of the
data. Volunteered GIS therefore represents the broad interest by non-scientists to be engaged in
and to benefit from spatial data analysis.
Although Goodchild coined the term “volunteered GIS”, the movement towards a new
paradigm really began a decade earlier with the desire, on the part of scientists, to engage
citizens directly in the research process. Sara Elwood, through her work with PGIS, exemplifies
this desire and her work has been instrumental in the evolution of the VGIS paradigm (Elwood,
2006b). Other contributions have included work by Elmes et al. (2005) with their description of a
“community integrated GIS”, Turner’s “Neogeography” (2006), Balram and Dragicevic’s
“collaborative GIS” (2006), and Sieber’s “public- participation GIS” (2006). Collectively the work
of these individuals demonstrates the broader goal of direct community engagement in the
research process.
However, researchers have also made significant strides towards integrating components
of a VGIS into their own projects, including studies in environmental sensing, decision-making,
resource management, and community risk assessment. Project GLOBE, OakMapper, and
Audubon’s Christmas Bird Count (Connors et al., 2012; Goodchild, 2007a; House et al., 2001;
Yaukey, 2010) are long running projects for the purpose of monitoring spatial and temporal
distributions of resources and phenomena. By employing citizens to collect data, researchers are
able to more effectually analyze spatial processes by generating much larger quantities of data.
While data quality remains a concern, the large quantity of data collected diminishes the
influence of inaccurate data (Flanagin & Metzger, 2008).
51

The use of VGIS to answer questions of decision making draw upon the PGIS literature in
suggesting that GIS technologies and implementations can assist in conflict resolution and
multiple-criteria decision making (Soheil Boroushaki & Jacek Malczewski, 2010). Flanagin and
Metzger (2008) make reference to these types of questions in using GIS for collective community
efforts. The key distinction between classical PGIS methods and VGIS involves the role of the
scientist. PGIS seeks to improve dialogue between actors for the purpose of legitimizing and using
local knowledge, the redistribution of resources access and rights, and new skills training in
geospatial methods (McCall & Minang, 2005). However, the researcher plays a limited role as
teacher. VGIS builds on this by leveling the authority between actors; scientists and non-scientists
are viewed as having [almost] equal authority, allowing both actors to communicate more freely
with each other and to share expertise. Our prototype supports the idea that a PGIS or VGIS can
contribute to addressing questions of decision-making, and later resource management and
conflict resolution.
Volunteered GIS alters the standard GIS paradigm by substituting a producer-user model
instead of the traditional expert-user archetype. Under this framework, researchers and citizens
can act as either producers or consumers of spatial data, depending on the context within which
they are interacting with the data. Producers in this case need not necessarily be experts in all
areas of GIS or the broader research context. Rather the producer’s role is given to any individual
with information to contribute to the aggregate knowledge base (Soheil Boroushaki & Jacek
Malczewski, 2010; Flanagin & Metzger, 2008). User roles are given to individuals who consume a
spatial data product for any purpose. Under the new paradigm, roles are not fixed and not
exclusive.
52

Finally, under the traditional GIS model, data were perceived to be trustworthy because
of the perceived authority of the scientist. However, with changing roles, we need a new model
for data error assessment (Flanagin & Metzger, 2008; Goodchild, 2007a, 2007c). Possibly the
single largest barrier to the utilization of volunteered information is the uncertainty surrounding
its credibility (Connors et al., 2012; Flanagin & Metzger, 2008; McKnight et al., 2011). Under the
VGIS model, the credibility of volunteered information is achieved through volume. Intuitively,
we understand that if multiple individuals report similar information, the reports are likely
credible representations of the truth. The larger volume of data collected through a VGIS, albeit
repetitive, can achieve the same threshold for credibility as data collected under the traditional
model (Flanagin & Metzger, 2008).
Related, there’s a significant degree of uncertainty as to the nature of volunteered
information with respect to the types of error. McKnight et al. (2011) explore the relation of
volunteered information to assess spatial distribution of West Nile virus in Michigan. In their
analysis, they raised the issue of uncertainty with regards to types of error that influence the
data. For example, users may reliably report positive observations (i.e. in this case, observations
made of dead birds), but reports are likely to indicate the absence of data. Therefore,
volunteered information is heavily biased towards the observation of an outcome, and should
not be interpreted as a metric of prevalence. Utilization of volunteer information must be
cognizant of the nature of uncertainty.
The prevalence of mobile devices that have GPS capabilities, including cell phones, tablets
and laptop computers, has increased the accessibility of spatial data. Hardware is no longer
priced outside the realm of ownership for many people in the world, meaning that users can now
53

directly engage with spatial data in ways that they simply could not do before. The interaction of
the public with spatial data is now so prevalent that most users have developed sufficient
technological skills and spatial cognition (through interaction with online mapping tools) to
enable them to interact with spatial data in an intelligent manner, precluding the need for
training prior to participation in GIScience research. Paradoxically, it would appear that spatial
cognition is unrelated to global geographic awareness. Although people are able to position
themselves abstractly on the landscape, they remain illiterate as to the broader geographic
context in which they live.
Software interfaces fall into two broad classes: traditional desktop products and webbased applications. For the purposes of inter-acting with a VGIS, citizens are most likely to use a
web application since this does not require a specific platform or license to run. Desktop
applications, on the other hand, can be distributed to certain groups, allowing for a more
targeted interaction with the spatial data. The open-source software movement is most directly
credited with making GIS software accessible to the public, removing financial and hardware
restrictions for many GIS products. Most notable among these are GRASS and Quantum GIS, free
GIS packages modestly equivalent to ESRI’s ArcGIS®. Interfaces for spatial data analysis have been
developed with R and Python, interacting directly with Grass and Quantum GIS. Increasing
familiarity on behalf of the public in spatial tools, geospatial technologies, and mapping increases
the likelihood that they will be able to act as producers of high quality information. While the
increasing availability of mobile technologies has spurred public interested in GIS, the cost of
adopting new technologies remains a principle challenge, particularly in developing countries.

54

Disease Surveillance
Disease surveillance systems are established for the purpose of collation, analysis, and
dissemination of information so as to facilitate the allocation of resources in handling disease
outbreaks (Thacker et al., 1983). Broadly, surveillance programs are categorized as either passive
or active. Passive disease surveillance programs rely on reporting by healthcare providers to
public health authorities when specific signs and symptoms are observed, a diagnosis is made
and/or a diagnostic test is confirmed. Public health authorities collate these reports and assess
the need for a coordinated response. Upon making a determination, the authorities
communicate back to the local health care providers, and the necessary recommendations are
set forth to address the disease outbreak. An example of passive surveillance in the United States
involves the reporting of certain communicable diseases by health care workers after a diagnosis
is made. These reports are received by public health officials who are tasked with ensuring the
disease does not pose a threat to the welfare of the public.
Passive disease surveillance systems are hierarchical in nature with space and time
important factors. Knowledge of highly infectious diseases may be reported up and information
on the control of those diseases may be reported down the hierarchy very rapidly, whereas more
common diseases may be reported and intervened on slower schedules such as monthly or
semiannually. The management and coordination of communication within a passive surveillance
system therefore, needs to be agreed upon by all parties (i.e., levels within the hierarchy in order
to ensure the protection of population health). However, passive surveillance systems are widely
criticized for underreporting diseases (Thacker et al., 1983).

55

When passive surveillance systems break down and mandated reportable disease(s) are
not communicated from the local to central levels, there is a need to respond by implementing
an active surveillance program.
Active surveillance programs directly address the underreporting of disease by utilizing
teams to assess local conditions. Such programs begin with the recognition at the central level
that the expected communication in space and/or time has not been received and in response
actively reach out to that location for the information. During these visits, retrospective data are
collected and the management of the passive surveillance system is revived (e.g., manpower,
technology). The operations of disease surveillance systems are therefore highly dependent upon
the cooperation of all participants at level of the hierarchy. One well-cited example of active
surveillance is Snyder and Merson’s (1982) meta-analysis of diarrheal disease prevalence and
mortality throughout the developing world. Here they review 24 studies where data was actively
collected (either through home visits or other means) by trained personnel. In contrast to a
passive surveillance program, workers were employed for the sole purpose of collecting disease
prevalence data.

Case Study
Purpose
African Trypanosomiasis (AT) is a zoonotic disease transmitted by the tsetse fly. In Kenya, the two
most common forms of AT are Trypanosoma brucei (Nagana), the form of the disease that affects
cattle, and Trypanosoma rhodesiense (Sleeping Sickness) that affects humans. While sleeping
sickness is relatively rare in Kenya, Nagana is widespread and represents a major threat to the
livelihood of pastoralists (Baird et al., 2009; Tarimo-Nesbitt et al., 1999; Waller, 1990). The
56

prevalence of Nagana has increased in recent decades due to a decline in control regimes, climate
change, and anthropogenic factors (Batchelor et al., 2009; Bauer et al., 1992; WHO, 2005). Our
case study describes the prototyping of a VGIS for the purpose of surveillance of an infectious
disease vector.

Figure 3.1: Study Area

Site Description
Nguruman (Figure 3.1) is located at the base of the Rift Valley in southern Kenya, just east of the
Nguruman Escarpment. Formally, Nguruman is the local Maasai name for the settlement, which
57

occupies the area west of the Ewuaso-Nyiro River and the Kongo Forest to the Oloibortoto water
intake; it is bounded to the south by the Ol’Kirmatian Conservation Area and to the north by the
Oloibortoto River. Nguruman is also referred to locally as Oloibortoto. North of the Oloibortoto
River, and broadly included in our study area, is Entasopia, the largest settlement in the area. The
political “capital” of the Nguruman area is Ol’Kirmatian, a settlement 6.5km west of the EwuasoNyiro River, and home to the District office for the Kenya government as well as the office of the
local governor for the Ol’Kirmatian group ranch, the political arm of the Maasai in this area.
From the base of the Rift to Oloibortoto, the predominant land use is smallholder
agriculture. Streams dissect the region and are maintained by the community as means to irrigate
their farms. Dominant agricultural crops throughout the region include tomatoes, vegetables
destined for South Asian markets, and fruit trees (e.g., bananas, mangos) (Langley, 2010).
Southeast along the road from Oloibortoto to the Kongo forest, vegetation density rapidly
increases. The area is extremely rocky and dominated by herbaceous and woody shrub
vegetation, most abundant of which are Acacia tortillis, Salvandora persca (toothbrush tree),
Grewia tembensis, and Cordia sinensis (Maitima; Morris et al., 2009). Dominant grasses
throughout the region include Sporobolus spp., Setaria spp., and Cynodon dactylon (Morris et al.,
2009). The Kongo forest is the area of densely vegetated land between Oloibortoto and the
Ewuaso-Nyiro River. It is within this zone that we find abundant tsetse; unfortunately this zone
is often the only option available to the community for grazing their animals during the dry
season. Moving east from the Ewuaso-Nyiro River, the landscape dries quickly, resulting in a rapid
decrease in vegetation density. During the dry season, the area is devoid of most vegetation;
however after a short period of rains, the grasses in the area of Ol’Kirmatian re-emerge with
58

vigor. Across the entire region, these eco-zones are highly dynamic and respond rapidly to local
climatic shifts and the occurrence of precipitation.
Global climate change is dramatically influencing the local environment within our study
area (Moore et al., 2012). In past decades, annual precipitation in southern Kenya has remained
relatively constant despite significant increases in annual mean temperatures, however the
variance of the magnitude of precipitation events have increased and the seasonality of total
precipitation has become less predictable (Altmann et al., 2002; Messina et al., 2012; Moore &
Messina, 2010). The observed climate change and uncertainty in precipitation will undoubtedly
threaten the livelihood of farmers and pastoralists (Fischer et al., 2005). Indeed, these concerns
were conveyed to us in the course of our work; many farmers have already found it difficult to
determine the right time for planting due to changes in local weather and precipitation events
(Langley, 2010).
Trypanosomiasis (Nagana) in cattle is a major threat to the livelihood of Maasai
pastoralists in Nguruman. The risk of infection is chief among their concerns to the health and
well-being of their cattle herds. An important consideration for the community is the
management of grazing for cattle herds among the members of the group ranch. A committee of
elders, whose chief aim is to maximize utilization of the limited resources (while advocating
sustainability) for the benefit of the community, manages the patchwork of grazing areas. Of
particular interest to the grazing committee (as expressed through interviews) is the ability to
work with our research lab to incorporate predictions of the spatial and temporal trends in tsetse
populations and models of risk aversion.

59

DeVisser et al. (2010) developed a species distribution model for tsetse (TED) that predicts
tsetse presence/absence every 16 days based on the habitat requirements and movement rates
of the fly. The precision of the model predictions is limited spatially by the resolution of the inputs
(250m), and temporally by the availability of MODIS LST and NDVI data products (8 and 16 days
respectively). It is well established that tsetse are highly responsive to microclimatic conditions
supported by local variations in vegetation (Terblanche et al., 2008). The spatial resolution of the
TED model predictions limits consideration of such local configurations, thereby increasing the
likelihood of errors of omission. The TED model was designed to identify endemic tsetse and does
not model transient tsetse populations. By incorporating volunteered information from citizen
reporters, TED could better illustrate the distribution of the flies over space and time by reducing
errors of omission and reporting transient populations. Volunteered information may also be
used (to an extent) to confirm (Oreskes et al., 1994) the TED model predictions by giving us a
means to estimate model uncertainty.

Conceptual Model
Here we elaborate on the previously published framework for a VGIS (Langley & Messina, 2011)
by illustrating the construction and deployment of a working prototype. Furthermore, we discuss
potential challenges and limitations of our implementation and propose strategies to address
these issues. Figure 3.2 outlines the basic implementation of the proposed VGI and the methods
by which users will be able to interact with the spatial database (sDBMS), specifically through
mobile devices. The core of the proposed implementation is a Postgres database server that
stores both the spatial data as well as scripts to compute predictions of tsetse distributions,
process volunteered data (submitted first to a reliability assessment), and automatically retrieve
60

and process remotely sensed imagery as it becomes available. Users interact with the database
through an Apache server and an HTML interface. A MapServer implementation provides
functionality for visualization of spatial data. All components are open-source and platform
independent so as to convey maximum portability.
Our implementation of a VGIS seeks to achieve three goals: 1) facilitate user interaction
with the VGIS and model results so as to allow for the reporting of information that may correct
otherwise inaccurate data (defined as those predictions that contradict ground-level reports); 2)
assess the reliability of volunteered information; and 3) incorporate volunteered information to
calibrate a model of tsetse distribution and reduce errors of omission. To assess the functionality
of the VGIS to achieve these goals, we have developed a working prototype of the system to
illustrate our approach.

61

Figure 3.2: This deployment diagram illustrates the interaction of the separate components of the
VGIS and the flow of information between each component

We incorporate a variety of software packages, including GRASS and QGIS (for visual GIS
support), Python and R (for statistical and modeling tasks), each of which provides the user with
statistical, visual, and geoprocessing capabilities; the user can interact with these packages
through a GUI or through a command line interface. Our selected DBMS is PostgreSQL 9.1
(Postgres), an advanced, readily available, open-source, object-relational database management
system. Using standard SQL syntax, Postgres allows for complex query capabilities, including
spatial queries, and facilitates strict rule and primary key enforcement. Postgres is also

62

extensible, allowing for the addition of new functionality (Michael Stonebraker & Kemnitz, 1991;
M Stonebraker & Rowe, 1986). In contrast to previous implementations of MySQL, Postgres, and
other common spatial databases, modern DMBS models facilitate the combined storage of
spatially explicit data and corresponding metadata together in the database (Elmasri & Navathe,
2004; Watson et al., 2004). The proposed spatial computing environment uses open-source,
community-supported software and standards, providing a solution to the data-management
problem that is temporally extensible. Of critical importance to us is the improved functionality
available in PostGIS 2.0, which adds support for raster data types. PostGIS is an extension to the
Postgres language that adds functionality for the storage and retrieval of spatial data. PostGIS is,
at its core, a suite of tools that serves as the back end for spatial functionality in Postgres.

Figure 3.3: We propose an iOS application (for iPhone or iPad) that
allows for users to interact with the VGIS, explore model predictions,
volunteer data, or to contribute

63

Data Collection and Interface
Users and producers alike are able to interact with the VGIS in many ways, each according to
their own skills, interests, and available hardware. In our case study, we outline the mechanisms
whereby participants (either researchers or community members) can interact with the VGIS.
Figure 3.3 illustrates the broad deployment strategy for mobile device interaction.
Related to the deployment of a mobile interface, the web interface sports a comprehensive suite
of tools available to all participants from any web browser, with functionality dependent upon
the credentials supplied to the system.
Most participants will find themselves interacting with the VGIS primarily through their
mobile devices. For this purpose, we propose an iOS application that, for the most part, simply
employs an HTML wrapper allowing the user to interact with a MapServer application. Users are
able to query the database for specific, albeit limited, types of data, even define a specific range
of times over which to aggregate the data; the application is geographically aware, so it is able to
return information for a user’s specific coordinates by passing the current longitude and latitude
to the server. Most importantly, a user is able to volunteer information, with regards to the
distribution of tsetse, through the application. Simply, a user can use this function to report that
tsetse flies are present at their current location. A user’s unique device ID is logged with the
report and serves as a surrogate measure to distinguish between users.
Users are able to interact with the system in different environments, including a web
browser, a desktop application, and on a mobile device (Figure 3.3). Users ultimately will be able
to conduct a range of operations, such as obtaining spatially contextual information and model
predictions, defining new model runs, exporting data, and submitting volunteered information
64

or reporting map/model errors; however for the purpose of our prototype, functionality is limited
to the data querying, visualization, and reporting of tsetse occurrences.

Information Reliability
The traditional model of data reliability emphasizes the authority of the researcher and our belief
that trained individuals will generate reliable, trustworthy data (Craglia, 2007). Within the
context of a VGIS, we relax this assumption, instead qualifying reliability through data volume,
the idea being that credible information will tend to be generated independently by more than
one user (Flanagin & Metzger, 2008).
There are two fundamental approaches to assessing the reliability of crowd-sourced
information. In the simplest case, information is assumed either credible or not until confirmed
or rejected by a subsequent report. Under this model, all participants are treated equally with
respect to their prior knowledge/skills; reliability is assessed by their peers through the creation
of informal social networks [of trust] (M Bishr & Kuhn, 2007; Mohamed Bishr & Mantelas, 2008;
Flanagin & Metzger, 2008; Metcalf & Paich, 2005).
The second model takes a more nuanced perspective of the user, taking into account the
skill set of the person filing a report and their prior credibility. This approach is best approximated
as a Bayesian model of data quality where the reliability of a report is dependent on the prior
assessment of the user and previous reports made to the system (Crosetto & Rodriguez, 2001).
If a number of prior reports are rejected, the individual is given a low reliability score that may
lead to automatically rejecting any subsequent reports made (unless of course those reports are
later confirmed independently). However, if the user has a history of high quality submissions
that are routinely confirmed, they may be given a high credibility score, leading to automatic
65

accepting of the report into the database. Prior experience with the user is the crux of this model
approach to data quality. Conati (2004) demonstrated this approach in evaluation of models of
user affect; their study required that they be able to assess the reliability of self-reporting of
emotional states.

𝑈𝑠𝑒𝑟 𝑅𝑎𝑡𝑖𝑛𝑔 = 𝛼 + ∆

(3-1)

α = prior score
Δ = change in score output from Figure 3.4

In our case study, we employ a simple decision model (Figure 3.4), that integrates social
trust networks (e.g. Mohamed Bishr & Mantelas, 2008; Metcalf & Paich, 2005) and Bayesian
methods (e.g. Conati, 2004; Crosetto & Rodriguez, 2001) to assess the accuracy of data and the
reputability of volunteers who report on the presence of tsetse flies. To demonstrate our
approach, we evaluate the reliability of volunteered information under two scenarios. In the first,
a single reporter volunteers on multiple occasions. In this scenario, the reliability of the
information is determined and the rating (Equation 3-1) can be associated with the reporter’s ID.
Subsequent reports are evaluated on the merits of the information as well as the reliability score
of the reporter. In short, a reliable reporter is likely to submit reliable information. A record can
also be approved if a user is deemed trustworthy under the model. This value is calculated over
time as a measure of the number of reports that are confirmed versus contradicted.

66

Figure 3.4: To assess the reliability of volunteered information, a
report is evaluated in the context of a set of conditions. This figure
presents a logical thought diagram for the application of the
computation of reliability (3-2)

In the second scenario, several reporters each volunteer information only once. In this
case, a reliability score cannot be computed or used to evaluate the reliability of the information;
a report made under this scenario must be evaluated solely on the merits of the content. There
are two components in the report (in addition to the information itself) that are used to assess
reliability, context and authorship. In this scenario, authorship is of limited value since each
reporter submits only once; we cannot conceptualize an author profile. However, we can
evaluate the content of the information in the context of current predictions (of tsetse
distribution) as well as prior years’ predictions for the same period. A reliability score (Equation
3-2) is computed as a cumulative product of a user’s rating, the number of times a cell is occupied
in the previous time step in the current year, the number of times the cell is occupied on the
same date in previous years, and the number of neighbors occupied in the previous time step. A

67

report is deemed credible if the score exceeds a certain threshold. This threshold is initially set
to 5, but should be re-evaluated periodically to ensure data quality is maintained.

𝑅𝑒𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 𝜃 + 𝜌 +

𝜅
+𝛾
4

(3-2)

θ = user score
ρ = the number of times the cell was occupied previously, including the previous time step in the
same year and the same time step in the previous year (max = 2)

κ = number of neighboring cells that are occupied (max = 8)
γ = number of supporting reports

Volunteered information under both scenarios can also be evaluated in the context of
TED model predictions. Model predictions that take into account volunteered reports are
compared to predictions made 16 days prior as well as to the distribution of tsetse at the same
time in the previous year. We can reasonably assume that pockets of tsetse should maintain
connectivity. If a report is made of tsetse occurrence in an isolated area (as measured by number
of neighbors, κ) where no tsetse are predicted to occur, the probability of this report being
accurate is low. If we were to incorporate these data into the model, the resulting predictions
might dramatically impact the local reliability of the model outputs. By incorporating volunteered
information into our prediction of tsetse distribution, we can better represent fine scale
variability, particularly with regards to our ability to represent real-time distributions.

68

Utilizing Volunteered Information to Reduce Model Error and Uncertainty
Previously, we detailed our approach to assessing the reliability of volunteered information in
the context of our case study. To illustrate the performance of the VGIS in making this
determination, we simulate the reporting of tsetse occurrences across the study area. The
simulated reports are generated for each iteration of TED model prediction. Additionally, we can
illustrate reliability assessment under each of the two scenarios we detailed earlier; multiple
reports from a single user or single reports made from multiple users.
The TED model outputs a binary raster at 250 m pixels which represents the minimum
mapping unit for the predicted distribution of tsetse on the date the latest MODIS data product
was captured; the predictions are not real-time estimates (always 30-45 days past) of tsetse
distribution and are designed to underestimate the maximum distribution. Incorporating
volunteered data allows us to fill in the gap, providing more up-to-date predictions (Figure 3.5).
If the data reports are deemed reliable and differ from TED predictions, the cell represented in
the binary raster for the previous time step is updated to reflect tsetse presence. The next
iteration of TED will build on the ‘corrected’ raster.
Tsetse distributions expand and contract with seasonal climate. They achieve a minimum
distribution at the peak of the dry season; these regions of minimum tsetse distribution are
termed ‘reservoirs’ (DeVisser et al., 2010). Of relevance to our case study, we can use these
minimum distributions as opportunities to ‘reset’ the model predictions so as to reduce any
errors of omission that may have resulted over the previous season from incorporating
volunteered information. In doing so, we can ensure that TED model predictions are reliable
estimates of the minimum distribution of tsetse.
69

To test the functionality of the VGIS, we will simulate the implementation of the system
to test the evaluation of volunteered information and the integration of this information with the
DeVisser’s tsetse distribution model. These simulations will primarily explore the assessment of
volunteered reports of tsetse presence, under three scenarios. The first illustrates the case a
report fills in a gap in the predicted distribution of tsetse (Figure 3.5a). Presumably, this is a
product of error in the estimation of safety distribution. As stated previously, the TED model is
designed to minimize errors of commission at the expense of added errors of omission. The
assumption here is that the gap observed in the distribution is a product of this process. When a
report is made, occurring within the bounds of this, the probability of that report being accurate
is reasonably high. Therefore, our model should assign a high reliability score to that report.
The second case involves a report that connects two clusters of tsetse. In this case, we
assume that patches of tsetse distribution should, for the most part, maintain connectivity in
some way. If a report is made that establishes connectivity between patches (Figure 3.5b), there’s
a high likelihood that this report is reliable. Therefore, our model should assign a score that
reflects this likelihood.
Finally, we simulate the case in which a report is made which places tsetse in a region that
is isolated from predicted patches of tsetse distribution (Figure 3.5c). Since we have no prior
reason to believe tsetse occur in this region based on model projections, there is a low likelihood
that this report is true. Therefore, our model should assign a reliability score that emphasizes the
extreme nature of this report. Through these simulations, we can identify the effective threshold
for reliability.

70

Figure 3.5: Users may volunteer reports of tsetse presence under a range of scenarios. (A)
Illustrates the case where a report fills in a gap in a patch of tsetse, likely correcting an error in
TED model predictions. (B) Illustrates the case where a report establishes connectivity between
two isolated patches of tsetse. (C) Illustrates the case where a report of tsetse presence is spatially
isolated from the predicted distribution of tsetse. In each case, a user is presented with a
prediction of tsetse distribution from the TED model (Column 1). Users identify an error in the
model, observing tsetse in an area where they are not predicted to occur, and submit a report
(Column 2 - black box). The report is submitted for reliability assessment; if deemed reliable, TED
model predictions are updated to reflect the new information (Column 3 – black box).
71

Conclusion and Limitations
Communication barriers present one of the most prominent barriers to disease surveillance
programs. When communications between health care providers and regional health authorities
break down under passive surveillance systems, there is a need to make attempts to directly
collect disease incidence data directly. Yet, there are significant hurdles to implementing active
surveillance programs (e.g., costs and logistics). By adopting concepts of crowdsourcing, public
participation, and volunteered GIS, we can open the door for an intermediate solution for disease
surveillance. Such an intermediate solution employs citizens to collect surveillance information,
increasing the manpower available to collate the information. It may not then be necessary to
dispatch health professionals to procure the data directly. Targeted campaigns can also be
utilized to solicit participation on behalf of the public to assist in collecting surveillance data.
Finally, our approach to assessing the credibility of volunteered information increases the utility
and reliability of data obtained from these campaigns.
However, there are significant limitations to a full implementation of the VGI in our case
study. The region in Kenya in which we are working is remote; there are significant challenges in
terms of communication connectivity, reliable electricity, and necessary hardware; AMREF
(American Medical and Research Foundation) and the African Conservation Centre (AAC) have
made significant improvements to local infrastructure, but much more is required. Cost remains
the most substantial hurdle for regional implementation. Our utilization of open source solutions
mitigates, but does not eliminate this challenge. Absent assistance from international partners,

72

the likelihood of full implementation of the disease surveillance system will certainly remain in
the domain of our scientific and development collaborators.
Connors et al. (2012) draw attention to the potential value of incorporating additional
sources of information (e.g., Twitter, Flickr), aside from direct volunteering through a VGIS, to
allow for increased participation; however, in doing so we would be introducing new types of
uncertainty to the models. Our current design attempts to limit error exclusively to those of
omission (i.e., we have tried to ensure that TED model predictions are estimations of the
minimum area tsetse are distributed). In this way, we have greater confidence over the areas
TED predicts tsetse to occur. When users volunteer reports of tsetse occurrence, they do so by
providing GPS coordinates of their location (this is done in the background through the iOS
application). Incorporating Twitter feeds, geo-tagged Flickr photos, among others, would on the
one hand provide us with more information; however the cone of location uncertainty of that
information is much greater and far less tractable. These sources of volunteered information
represent important avenues for future development, particularly in the broader field of VGI, but
at this time are beyond the scope of what we believe to be possible to include in our project.
Critical to the success of VGIS for disease surveillance is adequate public participation.
Too few reporters can make it difficult to assess credibility and limits the conclusions that can be
drawn from the information collected; however if incentives for participation are carefully
considered, there can be a drive for individuals to accurately and reliably contribute to the
system. Integration of volunteered information for disease surveillance, especially in low-income
countries, can be used as an alternative to the high costs of active surveillance programs, which
are often implemented in rural areas to learn more about disease prevalence. The prototype for
73

a VGIS outlined in this study demonstrates how technology and participatory science can advance
passive disease programs to improve public health in needed parts of the world.

74

CHAPTER 4
USING META-QUALITY TO ASSESS THE UTILITY OF VOLUNTEERED GEOGRAPHIC INFORMATION FOR
SCIENCE

Introduction
The scientific paradigm has evolved many times over the last millennium --- empirical,
theoretical, and computational paradigms have dominated our identity as scientists. However,
we are standing on the apex of another transition as technological and communications barriers
are toppled (Elwood et al., 2013; Gray & Szalay, 2006), and the distinction between amateur and
professional scientist is eroded. Neogeography characterizes the “blurring of the distinctions
between producer, communicator, and consumer of geographic information”; the separation of
scientist and layperson, expert and novice, is obscured as citizens engage in the generation of
new knowledge (Goodchild, 2009). As citizens engage in Science, we need to reconsider our
traditional notions of authority, expertise, and purpose.
Neogeography, a type of citizen science, has garnered a great deal of attention in the
literature as we struggle to conceptualize the nature of “geographic expertise”; however, the
involvement of citizens in science has long been established (Goodchild, 2009; A. Turner, 2006).
Participatory science has sought to involve citizens directly in academic research and related
exploits (Elwood, 2006b; Haklay et al., 2008; Tulloch, 2008) on the premise that citizens are more
informed actors with respect to their local environment than researchers operating externally.
Citizens are perceived to hold authority through experience and status, and are acknowledged

75

for their capacity to convey unique understanding, or indigenous knowledge (Elwood, 2006b;
Elwood et al., 2013).
With the advent of Web 2.0 (O'Reilly, 2005, 2006) and the widespread availability of new
technologies (Corbett, 2012; Haklay et al., 2008), citizens are increasingly exposed to
geographical information. Citizens also increasingly volunteer spatially explicit (geographical)
information that is of relevance or interest to them, often integrating this information with
existing datasets, or mashups, utilizing it for their own gain (Miller, 2006; A. Turner, 2006).
Goodchild coined the term “volunteered geographic information” (VGI) to refer to spatial data
that is contributed by ordinary citizens, irrespective of their training in scientific methods
(Goodchild, 2007a). The notion of VGI grew out of recognition of the limitations of traditional
methodologies for adequately mapping and assembling spatial information around the world
that provided both good coverage and fine temporal resolution (Elwood, 2008b; Elwood et al.,
2011; Goodchild, 2008b). Goodchild further articulated the issue, drawing from the broader
social science literature, postulating that the problem of data coverage can be mitigated were
we to harness the “six billion sensors” on the earth (Goodchild, 2007a). He is of course referring
to the earth’s population of citizens, whom he notes begin to acquire spatial knowledge at a
young age. Combined with Web 2.0 technologies, he asserts citizens can use the tools available
to them to “volunteer” spatial knowledge of the world around them. While traditional spatial
data infrastructures (SDIs) represent one-way communication models (where users only receive
information from experts), VGI represents a model where communication flows in both
directions without consideration of the role of the individual (Goodchild, 2009). As a framework,
VGI encompasses citizen participation from a range of social classes and computing practices with
76

the express purpose of harnessing the collective intelligence (Connors et al., 2012; Elwood,
2006b); it builds on the notion that data can be shaped by social and political processes and an
individual’s expertise, context, and spatial awareness (Elwood, 2008a, 2008b; Elwood & Leitner,
2012; Harvey, 2012). Local knowledge is crucial to an accurate geographic description of
communities and social groups, involving the citizen in the process of data collection; this
understanding enables science to more accurately explain geographic phenomenon.
VGI in practice is now commonplace. Arguably one of the most successful, if not the most
widely cited, outlet for VGI has been Wikimapia (Elwood et al., 2011; Goodchild, 2007a). Here
individuals contribute knowledge of the physical, built environment around them in order to
create as accurate a representation as possible. Other prominent examples include
OpenStreetMap and Google Maps (utilizing the Google’s API or Maps Maker) (Haklay et al., 2008).
Recent events have also demonstrated the potential for VGI to assist in disaster response
(Goodchild, 2010a).
However, the utility of VGI remains limited. In the context of the broader GIS literature,
data quality has always been a concern (Elwood et al., 2011; Flanagin & Metzger, 2008). In the
case of VGI, this concern is exacerbated due to the lack of expertise, or credibility, of the
individual (Flanagin & Metzger, 2008). Given that VGI is user-generated information by nonexperts, there is no quality assurance of the data (Craglia, 2007). It is viewed as non-authoritative
data; as such, some have argued it should be informative, but not relied upon (Corbett, 2012).
Others have raised concerns over the motivations of the individual, whether data is volunteered
with an intent to inform or mislead, an act of digital vandalism (Tulloch, 2007).

77

Many approaches have been taken to assess the quality and reliability of VGI (e.g. Corbett,
2012; Elwood & Leitner, 2012; Flanagin & Metzger, 2008; Langley & Messina, 2013). The majority
of these approaches thus far have been conceptual in nature, with few implementations of
reliability assessments for VGI. The most common of these methods involves social trust
networks and reputation models (Corbett, 2012; Maué, 2007). Under this approach, data quality
is checked by other project participants for errors and inconsistencies. In this model, no single
expert is tasked with reviewing each volunteered report. Another approach recommended has
been to use existing data sets (collected using more authoritative methods) to check for
inconsistencies in data. However, quality is not absolute; a datasets fitness-for-use is contextual
and may have varying degrees of suitability for different users (Goodchild, 2008a). No single
metric can be used to determine whether a data set is suitable across all ranges of potential uses.
Thus, the context of a user's participation and interaction with VGI must be taken into account
when considering accuracy/quality of VGI.
Given the concerns raised over the uncertainty of data quality in VGI, there is significant
debate as to the utility of VGI for science. There are surprisingly few examples of academic
initiatives utilizing VGI. (2011) inventoried 99 projects utilizing VGI and found only 3% to have
academic affiliations. Perhaps one of the most prominent examples of VGI in science is the
Audubon Society’s Christmas Bird Count. This project has amassed a significant volume of
volunteered data; however despite attempts to train volunteers in data collection, lingering
questions of data quality, of reliability, have limited any analytical value and integration potential
with authoritative datasets (Wiersma, 2010).

78

Flanagin and Metzger (2008) frame the issue of data quality of VGI in terms of credibility,
and the primary components of trustworthiness and expertise (of the data source). The
credibility (or believability) of VGI can be described objectively by traditional measures of data
quality – the degree to which the information can be considered accurate, or as the subjective
perception on the part of the consumer (Flanagin & Metzger, 2008). Credibility is related to
notions of trust, reliability, accuracy, reputation, authority, and competence. However, for VGI
to be useful for science, it is the traditional, objective “credibility-as-accuracy” measure
demanded (Flanagin & Metzger, 2008). To fully quantify error in data, it is necessary to have a
measure or to make assumptions as to the nature of the population being measured, to compare
the distribution of data against the population as a whole. It is in this way we measure attribute
accuracy, completeness, thematic resolution, and variability, to name only a few.

Other

measurements rely on feedback from measurement equipment, such as positional accuracy,
temporal accuracy, spatial and temporal resolution, among others. As is often the case with VGI,
the individual either operates without measurement equipment, or does not volunteer the
additional metadata with their report. Participatory science and VGI Science (VGIS) often involve
datasets for which the nature of the population is not immediately known. Therefore, a direct
quantification of the error of VGI is only possible in a post-hoc analysis. However, it is the
immediate benefit VGI can provide us that is of interest here and so we must develop a
mechanism to evaluate the merits of VGI in real time (as it is contributed). In the absence of an
ability to directly measure error and uncertainty parameters of volunteered data, we can use a
surrogate measure, meta-quality, a measurement of the collective quality of the data (van Oort,
2005).
79

The objective of our work here is to improve the perceived value of VGI for science by
demonstrating a methodology for VGI data quality assessment. We accomplish this through a
mechanism to explicitly assess the reliability of reporters based upon their respective VGI
contributions.
To better illustrate our approach, we apply the methodology to a case study in disease
ecology where we model the distribution of the tsetse fly, the principle vector of African
Trypanosomiasis in sub-Saharan Africa. The “tsetse ecological distribution model” or TED, is
based on an assessment of environmental characteristics critical for the persistence of the fly
(DeVisser et al., 2010). The model simulates a 16 day cycle of the expansion and contraction of
the fly population as response to the changing fundamental niche given the movement potential
of the fly population (DeVisser et al., 2010). The model is a conservative estimation of the
population distribution specifically minimizing errors of commission; therefore, the TED model is
an estimation of the minimum extent of tsetse at each point in time. However, the model is
reliant on a static land cover classification and makes no adjustment for error intrinsic to the
model (DeVisser et al., 2010). The TED model produces estimates of the spatial distribution as
binary outputs indicating presence/absence of the fly for each time period.
Potentially the most important contribution to incorporating VGI into a species
distribution model of the kind here is the fact that we can explicitly address one component of
model error (omission) without contributing additional error.

TED was developed as a

conservative model of the minimum expected distribution of tsetse. By incorporating VGI into
the model results, we can effectively facilitate the population expanding over gaps of unsuitable
habitat, either due to actual conditions or poor input data. It is known that microclimates provide
80

refuge for tsetse in areas where the habitat would be otherwise unsuitable (Ford, 1971; Moore
& Messina, 2010). The spatial resolution of the underlying MODIS data miss these microsites and
therefore omit these cells in the estimated distribution. Allowing the distribution to be updated
based on the VGI, would allow us to more accurately reflect conditions as they exist reflecting
sub-pixel dynamic that otherwise would not be possible. Incorporating VGI into the model results
to expand the distribution can therefore reduce errors of omission without contributing
additionally to errors of commission, thereby reducing total error, and thus improving data
quality. Incorporating VGI into TED requires two distinct steps: 1) determine the reliability of the
reporter to assess whether the VGI meets the threshold for acceptance, and 2) update the tsetse
distributions by changing the binary tsetse presence/absence value for the cell (in which the
datum is located) to 1 – indicating presence of the fly. In cases where VGI reflects the predicted
distribution, no change is made.

Methodology
Here we undertake a series of experiments to illustrate the integration of VGI into a traditional
analytical model. First, we explore the characteristics of VGI and its impact on model results.
Second, we evaluate the sensitivity of the model to three types of error common to
crowdsourced data. Finally, we explore the importance of reliability, as measured by a reputation
score (Frew, 2007; Langley & Messina, 2013; Maué, 2007), in determining the threshold for
accepting the data for inclusion in the model, under both static (a pre-defined score) or dynamic
(a varying score) conditions.
To simulate the generation of VGI, we first consider the different kinds of reporters and
the characteristics of the data they might contribute (Table 4.1). We identify four basic types of
81

reporters: 1) “always right”, 2) “always, intentionally wrong”, 3) “random”, and 4) “normal”. The
“always right” reporter represents individuals who are judged, post hoc, to be highly reliable and
the data they contribute are of high quality, often promoted to the role of moderator in online
forums (Maué, 2007); there is no (or minimal) spatial or temporal error component to the data
they contribute.

The “always, intentionally wrong” reporter represents individuals who

consistently, and/or intentionally provide erroneous data (M Bishr & Kuhn, 2007; van den Berg
et al., 2011); these reporters are unreliable and the data they contribute should always be
rejected. The “random” reporter represents individuals who generate data, falling on a random
distribution, reporting tsetse, for example, at apparently random locations across the landscape
(whether or not they are actually present) ignorant of underlying habitat conditions (Chow, 2012;
D. Coleman & Sabone, 2010); due to the random nature of the reports, the data are therefore
unreliable. Finally, the “normal” reporter represents the typical individual who volunteers
information; the individuals have a high degree of credibility and the data are usually high quality
(Flanagin & Metzger, 2008), but there is a spatial and temporal error component to the data they
contribute. It is this type of reporter that we are most interested in evaluating reliability.
In the context of our case study, the simulated data for each reporter are based on habitat
suitability criteria. In a real scenario, it is not possible to assess the accuracy of any report by
itself; rather we can only assess the fitness-for-use of the data by placing it in application context
and asking whether it is plausible (Grira et al., 2009; R. T. A. d. Groot, 2012). We simulate this by
evaluating the data based on the likelihood of the data being correct given the underlying habitat
conditions. To simulate the data, we identify a set of conditions that would be consistent with
reports made for each reporter type, and use these conditions to identify points that can be used
82

in our sample data set. Table 4.1 fully describes the types of reporters and the set of conditions
used to simulate data. For completeness, we explore the impact on the predicted occurrence of
tsetse by simulating data, not only from the four reporter types but also from data generated
from all combinations of habitat suitability criteria. It is based, in part, on these simulations that
we identified the specific combination of criteria that would be used to render simulated VGI.
The simulated data are based on the underlying conditions present at each time step in
the model, but not necessarily on the predicted occurrence for that simulation. For each set of
criteria and combination thereof, we ran 100 simulations, identifying 100 points in each time step
to serve as mock reports. Pooling these data points together results in 10,000 potential locations
(some locations are represented more than once in the pool due to random selection in the
simulations) for reports for each time step from which we randomly draw from when simulating
reporters. This allows us to incorporate a minimum amount of stochasticity that would exist with
reporters in a real-world scenario.
The basic TED model was implemented in GRASS based on the methods outlined by
DeVisser et al. (2010) (see Appendix A for code). Building on our implementation of the TED
model, we model the predicted distribution of tsetse, incorporating VGI, and evaluate the
magnitude of the difference. Appendix B presents sample code for one simulation run (simulation
11). Each model was written in BASH, a UNIX shell-scripting language. The models were run on
the HPCC cluster at Michigan State University for a total of 9,321 simulations representing an
estimated 13,981 hours of computing time.
The normal reporter is defined as an individual who usually provides credible data, but
has the potential to submit erroneous data. Incorporating these inaccuracies into the data
83

stream produces some degree of error in the model output. In reality, it is not possible determine
the truthfulness of the data; therefore we must be able to determine the influence of error on
the model output. The standard “normal” reporter is assigned an error rate of 10% (an arbitrary
assignment); we measure the effects of this error by evaluating the impact on the resulting
distribution when the “normal” reporter is assigned an error rate of 50%. As the data are
constructed based on the combination of habitat suitability criteria, we evaluate introducing
error into the model in different ways. Erroneous data are simulated by selecting points in areas
of unsuitable habitat by shifting the location of the point (simulating positional error), or by
holding the data until the following time step (simulating temporal error). A z-score is computed
comparing each set of criteria against a simulation where points are selected at random, as well
as a test of significance against the output from the TED model alone (no VGI data incorporated).
An assessment of the reliability of the VGI requires us to first generate a dynamic history
for each reporter that reflects the plausibility of the data as determined by habitat suitability
criteria. Each reporter is assigned a score, a measurement of their reputation, which is a product
of these criteria (slightly modified from Langley and Messina 2013 to allow for negative changes
in reputation). The index returns an ordinal measurement of reliability; it is not constraint to a
particular range, rather is structured such that positive scores convey reliability. It is computed
as:
𝑅𝑒𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 𝜃 + 𝜌 +

𝜅
+𝛾
4

θ = reporter’s score
ρ = the number of times a cell was previously occupied (-1 if 0)
κ = the number of occupied cells in 4-cell neighborhood (-1 if 0)
84

(4-1)

γ = the number of supporting reports (-1 if 0)
We arbitrarily selected threshold scores of 5 and 8 for incorporation of the VGI into the
TED model results. A paired t-test is used to measure the significance of adjusting the threshold
and the potential importance the specific selection has on the resulting predicted occurrence. An
alternative approach to the arbitrary assignment of scores is to determine the threshold at which
reporter types can be distinguished from each other. We subject the history of reporter scores
to a k-means test; this analysis tries to iteratively place each reporter into one of two clusters
(we define these clusters to mean reporters of “plausible” or “erroneous” data). Cluster centers
were defined at random from the set of scores for each test. As reporter scores increase over
time, we expect it will take a certain number of model time steps before they will group properly.
The average reporter score (for the plausible group) from 100 iterations can be interpreted as a
reasonable threshold score under a static model.
Over time, the scores for reporters quickly exceed the small thresholds we set (reaching
values > 100 at the end of the simulation), which results in unqualified acceptance of the VGI into
the model. As such, we cannot detect or respond (within a reasonable time) to changing behavior
among reporters, reflecting the inability of arbitrary, static thresholds to capture potential
declining reliability and reputation of reporters over time. In the final set of simulations, we
explore the possibility of using a dynamic score model, where the threshold for acceptance is
drawn from the distribution of all reporter scores at each time step. For each simulation, we set
a threshold equal to the 1st quartile score, mean, or 3rd quartile score from the distribution of
all reporters’ scores at that time. This allows us to include only the most reliable reporters from
our total pool of participants, and the longer the model operates over time, the more reliable
85

our output becomes. The net benefit to the model should thus improve over time. Sets of paired
t-tests are used to measure the significance of the difference in predictions from the three
threshold models.

Results
In our case, the likelihood that tsetse are present in an area, the subject of the VGI in question,
is correlated with the habitat suitability as measured by land cover, land-surface temperature,
and NDVI. A reporter’s score is a measurement of their reputation, akin to eBay’s ratings system,
which quantifies the history of the individual to perform in a manner that is perceived positively
by their peers (Maué, 2007). We assume that if a reliable reporter contributes information that
confirms another’s data, the likelihood that datum being accurate is improved. However, this
method of confirmation by peers necessitates a set of reporters who have attained a data history.
Until a reporter attains a certain reputation, we do not have enough information to assess data
quality; however, we have seen that different reporters themselves quickly separate from each
other, allowing us to partition out individuals who are either reporting randomly (and thus
frequently inaccurately) or are simply providing erroneous data intentionally. Partitioning out
these two types of reporters alone immediately improves the quality of the contributed data.
Varying the criteria for spatially locating VGI greatly influences the overall impact on the
predicted occurrence of tsetse, however the impact varies markedly from year to year due to
environmental conditions and shifts in the habitat suitability (Table 4.2). Randomly locating
points results in an overall 9.81% (4.23% − 13.66% for individual model years) increase in the
number of cells in which tsetse are predicted to occupy over the time period in the model (recall
that incorporating VGI into the TED model can only increase the prevalence of tsetse). However
86

targeting specific locations where habitat is suitable and at least one neighbor is predicted to be
occupied (the criteria we assign to our normal reporter), yields an overall 0.03% (0.02% − 0.05%)
increase in occupied cells. Notably, selecting suitable habitat alone as our criteria influenced the
results the most, with an overall 14.06% (7.22% − 17.94%) increase in predicted occurrence.
Likely this speaks to the design goal of the TED model to minimize errors of commission.
Predictably, constraining report locations to only those cells in which tsetse are predicted to
occur (the condition for our “always right” reporter) yields no increase in the predicted
occurrence of tsetse over the base model. Selecting locations in which tsetse are not predicted
to occur or where habitat is unsuitable (conditions for the “wrong” reporter or a component of
error in the normal reporter, respectively) yields an overall 10.59% and 8.23% increase in the
predicted occurrence. All criteria tested yielded significantly different results over the random
model (p < 0.001 in each case).
In the static threshold score model, there was no significant difference in the overall
predicted occurrence of tsetse (p > 0.4). However, utilizing a dynamic threshold score model
resulted in significant differences between all three models (1st quartile, mean, and 3rd quartile)
with p-values < 0.001 in each case. The overall increase in predicted occurrence was 0.8%, 0.43%,
and 0.12% respectively; however, the results varied widely from year to year for both static and
dynamic threshold models (see Table 4.3). [Note: simulations 8 through 12 in the table consider
the cases for only normal reporters].

87

Figure 4.1: A frequency plot representing the time-step in which reporters cluster into two groups,
for 100 replications of simulation 13

The four types of reporters cluster into two groups — see simulations 13 and 14 (Table
4.4) for the cases where all reporter types are considered. The four reporters are not fully
distinguishable from each other at any time in our models (k-means with four clusters). Figure
4.1 presents the distribution curve (for all 100 replications) for the time step, at which point the
reporters can be distinguished using a k-means clustering approach. For simulation 13, where a
threshold score of 5 is used, the reporters can be separated, on average, in the 5th time step
(mean = 4.93, median = 5). The average reputation score in the 5th time step is 10.87 for the
“plausible” group. Reporters in simulation 14 (50% error rate) do not consistently cluster
together into two groups.

88

Figure 4.2: A frequency plot representing the time-step in which reporters cluster into two groups,
for 100 replications of simulation 10

Figure 4.3: A frequency plot representing the time-step in which reporters cluster into two groups,
for 100 replications of simulation 11.

89

Considering the dynamic score models, there were no significant differences in the time
needed for reporters to group together. For the 1st quartile threshold score (simulation 10),
reporters clustered into two groups, on average, in the 5th time step (mean = 4.61, median = 5).
The average score for the “correct” reporters in the 5th time step was 18.87 (Figure 4.2). In the
mean threshold score models (simulation 11), reporters clustered together in the 4th time step
(mean = 4.32, median = 4). The average reputation score for reporters in this time step was 15.06
(Figure 4.3). Finally, for the 3rd quartile threshold score model, reporters clustered together in
the 4th time step (mean = 4.21, median = 4) with an average reputation of 15.14 (Figure 4.4).

Figure 4.4: A frequency plot representing the time-step in which reporters cluster into two
groups, for 100 replications of simulation 12

The nature of error (positional vs. temporal) introduced into our models through
incorporating VGI did not appear to change the magnitude of the impact on predicted
occurrence. This was also true when varying the magnitude of the error, at least for the range
tested (5 – 25%). We did observe a significant increase in the predicted occurrence of tsetse
90

when the magnitude of the error introduced was 50% (where each reporter had a 50% chance of
contributing erroneous data); introducing error of any type, though, results in a significant
increase in the predicted occurrence compared to the case where no error is considered
(simulation 4). Therefore, at least in our case study, the error introduced from VGI is not
expected to a statistically significant effect on the prevalence of tsetse. This suggests that our
models are resilient to the introduction of some erroneous data. Adaptations of our model to
different studies will nevertheless necessitate an exploration of the role of introduced error from
VGI to assess the resiliency of scientific models.
While the analysis reveals significant differences in the predicted tsetse occurrence from
incorporating VGI into the TED model, global metrics are difficult to interpret given the
importance of spatial structure in the dataset. To this extent, visualizing the structure of tsetse
distribution patterns can lead to novel interpretations of the influence of VGI. Figure 4.5 - Figure
4.7 present the predicted distribution of tsetse over our study area (for simulations 10, 11, and
12 respectively); cell values indicate the proportion of time steps in the model (every 16 days
between 2004-2006) where tsetse are predicted to occur, averaged across 100 replications. The
distributions incorporating VGI closely mirror the base TED model with marked differences
between core tsetse areas. These maps illustrate specific areas where VGI is particularly
influential, likely due to the ability of tsetse populations to “jump” patches of unsuitable habitat.

91

Figure 4.5: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 10. Values represent the proportion of time-steps in the model where tsetse
were present; this is a rough approximation of the probability of tsetse occurrence.

92

Figure 4.6: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 11. Values represent the proportion of time-steps in the model where tsetse
were present; this is a rough approximation of the probability of tsetse occurrence.

93

Figure 4.7: The theoretical maximum and minimum extent (respectively) for the distribution of
tsetse for simulation 12. Values represent the proportion of time-steps in the model where tsetse
were present; this is a rough approximation of the probability of tsetse occurrence.

94

Time is a significant factor to consider when evaluating the results of our models. In
describing the output of TED model predictions, DeVisser et al. (2010) noted that tsetse
populations tended to reach their maximum extent at the end of the long rains (ending the
beginning of June). Populations tended to reach their minimum extent at the end of the cool dry
season (mid- to late-October). This interpretation of tsetse population distributions comports
with what is observed in my simulations, and is grounded in an ecological understanding of tsetse
population dynamics.

Discussion
Volunteered geographic information can make valuable contributions to science, enhancing
datasets from more authoritative sources. However, integrating VGI data necessitates assessing
the error and uncertainty of those data. Direct quantification of data quality in this context is
difficult; the traditional components (e.g. accuracy, precision, and variance) typically cannot be
ascertained for VGI. It is critical for us to at least be able to qualify data quality, as it serves as
the foundation from which we assess fitness-for-use. We have proposed using reputation or
reliability (of the reporter) as a surrogate measure of meta-quality. As an initial assessment,
meta-quality allows us to begin to break through the cloud of uncertainty inherent with VGI.

95

Figure 4.8: This figure overlays the scores of 100 reporters for simulation 8

We build on the power of the reliability/reputation assessment by considering a dynamic
threshold scoring model. While we considered three different criteria for establishing a threshold
(defined as the 1st quartile, mean, and 3rd quartile values in the distribution of reporter scores
in each time step), we did not find a significant difference between them – as measured by an
overall increase in the prevalence of tsetse in our models. In considering only those individuals
whose reliability exceeds the mean score for all reporters, we only incorporate VGI from a subset
of reporters we deem the most reliable. As scores improves for all individuals (regardless whether
we have incorporated their data into our models), the threshold for acceptance/inclusion in our
models also increases (approximately linearly in our models – Figure 4.8 shows the trend for one
simulation). Over time, the quality of VGI data that we incorporate will improve, and the impact
96

of any erroneous data we have included should decrease. Most importantly, a dynamic threshold
model facilitates detection of declining performance (of a reporter) and a rapid response to limit
the acceptance of poor quality data.
The potential value of a means to assess data quality of VGI is immense. The strongest
hurdle to fully utilizing VGI has been our inability to measure data quality and uncertainty. In
demonstrating a valuation system for VGI (based on the reputation of reporters themselves), we
have, in part, overcome this hurdle. To date, the utilization of VGI for science has been reserved
for those cases only where the performance of reporters is controlled through training and
guidance while closely monitoring the entire process from data collection to communication
(Elwood & Leitner, 2012; Tulloch, 2008; Wiersma, 2010). But this runs contrary to many of the
perceived strengths of VGI, the dissolution of traditional roles (Elwood et al., 2013; Goodchild,
2007c, 2009; Haklay et al.), and the establishment of a two-way communication model for
geographical information (Goodchild, 2007a). Projects that have tried to embrace VGI have done
so under the old model of participatory science, and thus are subject to all the perceived and
actual limitations (Elwood, 2006b; Miller, 2006). Many factors influencing quality remain difficult
to measure, including rates of participation and motivation to participate; the value of VGI cannot
be fully appreciated until we can reliably assess these factors and the role they play in
determining data quality.
It is our position that incorporating VGI into standard scientific models, particularly those
where available data are sparse, can significantly improve the performance of the models and
the predictive or explanatory power of the results. Consider the case of “Digital Earth”; first
conceived by then US Vice-President Al Gore, it represented a push to represent the planet in
97

high-resolution, multi-dimensional space for the primary purpose of improving our predictive
capabilities of Earth’s ecosystems (Craglia, 2007; Craglia et al., 2012). Twelve years later,
significant gaps still exist, particularly in terms of our capacity to collect certain types of data of
sufficient quality and resolution (Craglia et al., 2012). Harnessing the collective power of earth’s
citizens, the aggregate power of “six billion sensors”, we can make significant strides to improving
the predictive capacity of our models through incorporating new types of information
(Goodchild, 2007a). Therefore, it is critical we continue to explore ways to assess the credibility
of VGI, to embrace the new geographical traditions, while respecting the scientific paradigms of
the past.

98

Table 4.1: Reporter types and the criteria used to simulate their behavior
Reporter

Type

Model Criteria

1

Always right

Tsetse predicted

2

Always, intentionally wrong

Tsetse not predicted, habitat unsuitable

3

Random

Spatially random

4

Normal

Suitable habitat + one occupied neighbor

99

Table 4.2: Simulation results for simulated conditions. Values represent percent increase over the base TED model
Sim
1
2
3
4
5
6
7

Criteria
Random
Suitable habitat
One neighbor
Suitable habitat + one neighbor
Tsetse present
Tsetse not present
Habitat unsuitable

Overall
9.81
14.06
0.29
0.03
0
10.59
8.23

% Gain
2004
2005
4.23
13.66
7.22
17.94
0.17
0.39
0.02
0.04
0
0
4.78
14.57
3.46
11.75

100

2006
11.85
17.58
0.32
0.05
0
12.71
9.66

Overall
144.02
108.26
93.37
19.65
0.01
128.16
138.43

Variance
2004
2005
76.83 105.73
73.41
80.88
27.62
66.28
10.86
16.52
0.01
0.01
75.51
97.03
79.15 112.02

2005
106.62
77.11
65.48
13.97
0.01
91.45
109.58

Table 4.3: The percentage increase in the prevalence of tsetse over the base TED model for simulations 8-12
Sim
8
9
10
11
12

Score
5
8
st
1 quartile
Mean
3rd quartile

Overall
1.28
1.22
0.8
0.43
0.12

% Gain
2004 2005 2006
0.27 1.94 1.68
0.23 1.88
1.6
0.13 1.15
1.2
0.05
0.6 0.68
0 0.14 0.24

101

Overall
139.56
138.6
120.62
109.46
77.36

Variance
2004
2005
46.8 113.56
44.09 112.57
37.76
92.9
28.92 83.06
10.47 55.49

2006
100.52
99.1
95.07
83.27
66.11

Table 4.4: The percentage increase in the prevalence of tsetse over the base TED model for simulations 13-20
Sim
13
14
15
16
17
18
19
20

Error type
10%
50%
Spatial shift 5%
Spatial shift 10%
Spatial shift 25%
Temporal shift 5%
Temporal shift 10%
Temporal shift 25%

Overall
1.38
5.23
1.39
1.39
1.41
1.46
1.54
1.61

% Gain
2004 2005 2006
0.39 2.07 1.74
1.88 7.94 5.95
0.39 2.13
1.7
0.44 2.22 1.52
0.44 2.18 1.65
0.43 2.21 1.79
0.45
2.4 1.81
0.5 2.47
1.9

102

Overall
139.44
144.9
137.33
142.05
131.66
135.71
149.88
148.95

Variance
2004
2005
48.86 116.26
70.59 124.32
49.43 112.58
54.59 118.36
50.61 108.22
50.68 112.63
52.54 124.28
55.65 119.68

2006
95.18
99.84
92.83
97.97
95.31
97.89
97.73
97.55

CHAPTER 5
SUMMARY AND CONCLUSION

Introduction
With the emergence of new technologies, the availability of new web based tools, and the
prevalence of geographic information, the traditional distinctions between citizen and scientist,
consumer and producer of knowledge are eroding and the line between them is blurred. In this
era of eScience and Neogeography, citizens are embracing new freedoms to not only combine
datasets through mashups, but to engage in the generation of new knowledge. These activities,
traditionally the exclusive purview of the academy, necessitate new approaches for
management. The sheer volume of information being generated requires new approaches for
storage and retrieval. Furthermore, we need to devise new methods to assess the quality of
these data, particularly as traditional objective measurements of error are no longer
communicated through metadata, as has been the case with traditionally collected spatial
information. Data quality metrics are necessary for us to qualify the credibility of the information
before us to determine a dataset’s fitness-for-use in a particular analysis.

Summary of Main Findings
Objective 1: Address three recurring problems with spatial data management: scalability,
reliability, and security by:
1. Communicating a conceptual model for a comprehensive open-source computing
environment that promotes the efficient organization, storage and retrieval of disparate
data.
103

2. Extending the discussion of spatial databases by presenting a model framework for a
spatial DBMS that rigorously and consistently manages both spatial and nonspatial data.
Effective data management in the age of Neogeography is an emerging problem. Today’s data
management strategies not only need to account for the volume of information, but also the
modes by which it is acquired and disseminated. The rigorous standards to which we have
traditionally subjected data in terms of quality and completeness are often unfeasible. As citizen
science initiatives expand, volumes of spatial data are generated, much of which would
traditionally be considered incomplete, either because it is not accompanied with comprehensive
metadata or does not consistently report all attributes. Traditional DBMS afford a great deal of
flexibility, but they often do so through a lack of built-in checks for consistency and quality.
Modern spatial data infrastructures (SDIs) must address this changing nature of data.
Three recurring problems with data management are routinely cited in the literature.
Scalability refers to the ability of SDIs to manage interaction with data at multiple spatial,
temporal, and thematic resolutions (Shekhar & Chawla, 2003). Reliability speaks to the need for
mechanisms to protect against loss of data either through inadvertent changes or malicious
behavior; however, it must do so without impeding interaction with the data (Devillers &
Jeansoulin, 2006; Shi et al., 2002). Finally, security refers to the need to strictly control the
dissemination of various types of data (some of which may contain personally identifiable,
private information) without impeding the ability to use the information. Furthermore, it must
facilitate management of ownership; in particular, ownership of VGI is still a commonly cited
concern in the literature. Such safeguards are commonly required in academic or government

104

institutions. The implementation of the DBMS proposed directly addresses each of these
problems.
The aim of chapter 2 was twofold: first, outline the problem of spatial data management in
the age of Neogeography and the volume of data generated by citizens as VGI; second,
demonstrate an approach for the handling of spatial data of varying types in a manner that
facilitates efficient querying, storage, and retrieval. The ways in which we interact with spatial
data have changed. Thus, the goal here was to develop a DBMS that afforded flexibility and
extendibility to allow for the integration of future advances in technology.
It was important to utilize only open-source software solutions so as to aid its adoption
by communities and organizations seeking to incorporate VGI into their projects. The opensource software is also extendible, fulfilling one of the primary objectives. The DBMS facilitates
efficiency in the storage of spatial data by incorporating the WKTRaster module to the PostGIS
libraries11. WKTRaster extends the functionality of PostGIS to allow for the raster-based imagery
to be stored in the database just as has been possible for vector-based data. This makes it
possible to query imagery directly at multiple levels, and return it at the desired resolution and
extent. The decision to utilize PostgreSQL as the core of our DBMS also addresses the second and
third concerns raised: reliability and security (although the specific safeguards were not discussed
in detail until Chapter 3).

11

Since publication (Langley & Messina, 2011), WKTRaster has been fully implemented in PostGIS 2.0. It is now
referred to in the documentation simply as PostGIS Raster.

105

Objective 2: Demonstrate the utility of VGI by:
1. Describing a prototype for the utilization of VGI to enhance disease surveillance
programs.
2. Articulating an approach for integrating VGI into a traditional species distribution model.
The last decade has seen a dramatic rise in the availability of spatially explicit geographical
information and the frequency with which non-scientists are exposed to and utilize it. This is in
large part a result of new modes of communication and interaction via web based applications,
a phenomenon often referred to as Web 2.0 (Corbett, 2012; Haklay et al., 2008; O'Reilly, 2006,
2007). In Chapter 3, I made reference to two prominent and often cited examples of GIS in a
Web 2.0 environment; these include Wikimapia and OpenStreetMap (Goodchild, 2007a; Haklay
& Weber, 2008). Both sites allow for users to volunteer information about their environment.
These initiatives are distinctly spatial in nature and generate volumes of new geographical
knowledge, yet operate entirely outside the purview of the academy. I refer to these types of
initiatives (those devoted to the volunteering of geographic information) as a type of volunteered
GIS or VGIS.
The potential for scientists to tap into the VGI data collective, and to make use of new
tactics for data collection is attractive; Goodchild (2007c) makes reference to harnessing the
power of earth’s billions of sensors as a means of knowledge production. However, there are
significant concerns that have been raised with regards to the credibility of VGI (Flanagin &
Metzger, 2008). As such, there have been very few illustrations of VGI being utilized in scientific
research (Elwood et al., 2011).

106

The aim of Chapter 3 was to elucidate the value of VGI when incorporated into a
traditional species distribution model. Building off Chapter 2 (which outline the framework for
the SDI) we demonstrated the framework for incorporating the VGI. Our primary objective in
this chapter was to propose an approach to assessing the reliability of reporters of VGI which can
serve as an indicator of the quality of the data itself (Maué, 2007). We propose a quantitative
metric, Reliability, that is computed as the product of underlying model conditions; it is based in
part on the notion that VGI can be assessed by comparing it against datasets of known quality
(Koukoletsos et al., 2012). This approach has been cited in the literature as a possible method,
and implemented in part in several studies as a post-hoc analysis (D. J. Coleman & Georgiadou,
2009; Corbett, 2012; Frew, 2007; Maué, 2007). We propose a threshold be established below
which a reporter’s data are excluded/rejected.
Finally, we outlined our approach to integrating VGI into a traditional species distribution
model. In our case study, we build on a model developed by DeVisser et al. (2010) that models
the distribution of tsetse in Kenya on the basis of remotely sensed imagery as inputs. The
implementation in our study is such that the assessment of reliability is made at the moment
data is submitted; therefore immediate decisions can be made as to incorporating or excluding
the data from the communicated results.

Objective 3: Address lingering concerns of credibility and data quality in VGI by:
1. Proposing a method for dynamically assessing the reliability of reporters of VGI.
2. Assessing the impact of incorporating VGI of varying quality into a traditional species
distribution model.
107

In order to firmly establish the utility of VGI for science, it’s critical we be able to judge the
credibility of the information. The intent of Chapter 4 was to elucidate an approach for the
measurement of data quality for VGI. In reviewing the traditional metrics used to assess data
quality, it is noted that objective assessments are typically not possible for VGI (Flanagin &
Metzger, 2008); therefore we propose a subjective measure that uses the reliability of the
reporter (to produce credible data) as a surrogate measure to communicate data quality (Maué,
2007). This is drawn, in part, from the eBay model that references reputation as an indicator of
the probability of a positive future transaction.

We reference meta-quality (an officially

recognized metric) as the means by which we can communicate credibility of the VGI.
In the second part of this paper, we illustrate the impact VGI can have on the output of a
traditional species distribution model. Since all data are prone to include error, and it is our
assertion that credibility of the data communicates an estimation of error, we demonstrated the
impact different types of error can have and at varying magnitudes. Finally, we demonstrate that
the model outlined in Chapter 3 for assessing the reliability of reporters does in fact allow us to
partition out reporters perceived as credible and those who are not. We introduced the notion
of a threshold score model, which represents the point at which an individual is deemed reliable,
and have their data incorporated. We noted that the static model (one set threshold) did not
allow us to effectively account for changes in reporter behavior, leaving the system vulnerable to
“digital vandalism” (Tulloch, 2007). However, a dynamic model that adjusts the threshold to
include only the most reliable of reporters, allows the system to respond to changing behaviors,
mitigating the impact of malice and the incorporation of erroneous data.

108

Theoretical Implications of this Dissertation
As summarized in the main findings, this research has contributed to the discussion of
uncertainty of data quality for VGI. Numerous studies have articulated the issue, raising concerns
for utilizing VGI in the absence of effective data quality assessments (Corbett, 2012; Flanagin &
Metzger, 2008; Maué, 2007; Tulloch, 2007). In this dissertation I have demonstrated one
approach to evaluating VGI utilizing subjective measures of credibility. Numerous theoretical
studies have laid the framework for this approach, specifically correlating a reporter’s reputation
- their reliability to contribute credible information - with the quality of VGI and crowdsourced
data (D. J. Coleman & Georgiadou, 2009; Elwood et al., 2013; Maué, 2007). Other studies have
suggested that the credibility of individual datum is irrelevant as the collective contributions of
reporters, averaged together, can produce information that is highly credible (Elwood & Leitner,
2012; Haklay et al., 2010). An extension of Linus’ Law for Neogeography states that the number
of participants can be correlated with the credibility of the resulting information set (Haklay et
al., 2010). A popular expression of this law by Raymond Raymond (1999) says, “Given enough
eyeballs, all bugs are shallow”. The point here is that errors are less influential as the number of
participants in a project increases.
This dissertation specifically addresses issues raised by Flanagin and Metzger (2008),
Tulloch (2007), and Craglia (2007) regarding the lack of quality assurance available for VGI. I
further propose that as the reliability of VGI data quality metrics is recognized, Corbett’s (2012)
critique of VGI as non-authoritative should no longer be a barrier to utilizing VGI for science. It is
my assertion that the credibility of any information or data is directly related to an assessment
of data quality. While the data quality of VGI is most commonly communicated subjectively,
109

evaluating quality using objective methods allows for us to more directly make statements with
regard to confidence, particularly when evaluated in the context of more traditional data
collection methodologies.

Recommendations for Future Research
Future research should focus on a number of specific objectives that have been conveyed in this
dissertation.

We must further the development of web-based mapping capabilities to

incorporate new tools for efficient querying of spatially explicit information. National SDIs have,
to a limited extend, implemented web-based technologies; however their purpose is tailored to
the expert scientist for data analysis and visualization. The same functionality has not been
translated to the web-based GIS portals used by non-scientists (Craglia et al., 2012). As citizen
scientists engage in more advanced geospatial analyses, they demand the same functionality
afforded to academics. Craglia (2012) describes a vision in which farmers are able to use GIS
tools to monitor crop status and yield, perform long-term risk analysis on market prices and
trends, receive early warning of extreme weather and other environmental dangers, and overall
improve the management of their resources. However these capabilities require enhanced
communication and technological improvements not currently available to the general public.
The Digital Earth project is building off of the achievements in citizen mapping technologies to
improve access of non-scientists to spatial data and their ability to engage in more advanced
spatial tasks (Goodchild et al., 2012).
There has been significant discussion in the literature surrounding the lack of credibility
inherent in VGI (e.g. Elwood et al., 2013; Flanagin & Metzger, 2008; Goodchild & Li, 2012).
Numerous approaches have been proposed for assessing data quality, but most are still in the
110

theoretical stage (Goodchild & Li, 2012). Few studies have demonstrated the implementation of
quality assessments for VGI, and to my knowledge, all are post-hoc assessments. Linus’ Law is
often cited as justification for leniency in credibility assessments of VGI based on the correlation
observed between the number of participants in a project (or volunteers) and the aggregate
quality of the information (Goodchild & Li, 2012; Haklay, 2010; Raymond, 1999). However, this
approach precludes individual assessments of quality; if integration of VGI (such as was observed
in our case study) necessitates evaluating uncertainty of each datum, the approach is not
applicable.
Current approaches demonstrated in the literature assess VGI, not on its qualities, but
rather in comparison against datasets of known quality or on the bases of tertiary qualities
(Cipeluch et al., 2010; Girres & Touya, 2010; Koukoletsos et al., 2012). We need a methodology
that expedites an assessment of VGI objectively on its own merits, and in real-time. Despite its
limitations, VGI has the potential to serve as a means of acquiring high quality data for little cost
by harnessing the collective power of the crowdsourcing community. Therefore, we must
develop new approaches to objectively evaluate VGI on its own merits so we can quantify error
and uncertainty in the data.
The method proposed in this dissertation (Chapter 4) facilitates a credibility assessment
using a surrogate metric, meta-quality, which correlates the reporter’s reputation with the
credibility of the information they contribute.

However, the approach has only been

demonstrated theoretically; it must now be subjected to experimentation in real-world
applications to quantify the correlation between reporter reputation/reliability and the

111

credibility of the data they volunteer on an individual basis and not necessarily as a part of the
collective.
Finally, there is a need to explore the application of subjective valuations of VGI on
existing crowdsourced datasets so that the data can be subjected to the same scientific analyses
available to more authoritative data sets. This has been undertaken, to a limited extend, on the
Wikimapia and OpenStreetMap datasets (Haklay, 2010), but we do not have established
methodologies for post hoc valuations.
There is significant interest in the literature for methodologies that will allow
incorporating and utilizing VGI and other crowdsourced data for science (e.g. Craglia, 2007;
Goodchild, 2009; Haklay, 2010). In this dissertation, I broadly set out to advance the utility of VGI
for science. I articulated a data model for an SDI (Chapter 2) that is tailored to and supports the
integration of VGI. The model I have proposed takes steps to improve the communication of
data quality metrics that are commonly missing with crowdsourced data.

In Chapter 3, I

demonstrated how VGI could be integrated with a more traditionally authoritative dataset. Here
I expounded on the potential value this additional data stream can provide in the context of the
limitations with the existing datasets. Finally in Chapter 4, I responded to those who have raised
concerns with a lack of quality assurance in VGI by proposing a method to assess the credibility
of volunteered data. I did this by evaluating the context of the information and related this back
to the reliability of the reporter. Subsequent contributions by this reporter are then assessed in
a Bayesian model where the credibility of the data is a function of its context as well as the prior
performance of the reporter. Aggregately, the methods introduced and the points made with

112

regard the specific advancements (both theoretical and practical), all advance the affirmed goals
set out in this dissertation.

113

APPENDICES

114

BASE CODE SIMULATION
# Base TED simulation model
# Author: Shaun Langley
# Last Modified 3/25/2013
years="2003 2004 2005 2006"
times="001 017 033 049 065 081 097 113 129 145
209 225 241 257 273 289 305 321 337 353"

161

177

193

# The model will loop through each year and time value noted here.
g.mapset -c baseSim1 # create a seperate mapset for each simulation to
isolate output from each other to prevent possible data overwrite
g.mremove rast=* -f
g.region Kenya
r.mapcalc "distrib.tmp=initDistrib"
for year in $years; do
for i in $times; do
NDVImap=`g.mlist type=rast pat="bin${year}_${i}_NDVI"`
LSTDaymap=`g.mlist type=rast
pat="bin${year}_${i}_Day_LST_250m_Terra_16day"`
LSTNightmap=`g.mlist type=rast
pat="bin${year}_${i}_Night_LST_250m_Aqua_16day"`
LULCmap=`g.mlist type=rast
pat="bin${year}_LULC_Type_1_250m"`
r.mapcalc "suitable.$year.$i=($NDVImap * $LSTDaymap *
$LSTNightmap * $LULCmap)"
r.neighbors input=distrib.tmp output=distrib.grown.tmp
size=5 method=maximum --o
r.mapcalc "distrib.tmp=(distrib.grown.tmp *
suitable.$year.$i)"
g.copy distrib.tmp,distrib.$year.$i
done
done
g.remove distrib.tmp
g.remove distrib.grown.tmp
g.region zoom=studyarea
r.mask studyarea
r.series input=`g.mlist pat=distrib.200[4-6].* sep=,`
output=baseSim_average method=average
r.series input=`g.mlist pat=distrib.200[4-6].* sep=,`
output=baseSim_sum method=sum

115

r.series input=`g.mlist pat=distrib.2004.* sep=,`
output=baseSim_2004_average method=average
r.series input=`g.mlist pat=distrib.2004.* sep=,`
output=baseSim_2004_sum method=sum
r.series input=`g.mlist pat=distrib.2005.* sep=,`
output=baseSim_2005_average method=average
r.series input=`g.mlist pat=distrib.2005.* sep=,`
output=baseSim_2005_sum method=sum
r.series input=`g.mlist pat=distrib.2006.* sep=,`
output=baseSim_2006_average method=average
r.series input=`g.mlist pat=distrib.2006.* sep=,`
output=baseSim_2006_sum method=sum
r.mask -r
curl https://prowlapp.com/publicapi/add -F
apikey=2daeeaf780b2e94281d1089752c6698da81434a9 -F
application="GRASS" -F description="TED base simulation complete"

116

SIMULATION 11 CODE
##########################
# simulation 37 - all reporters with score tracking -- threshold=mean
# Author: Shaun Langley
# Last Modified 3/25/2013
g.mapset -c expSim37_$rep
g.mapsets addmapset=baseSim1
g.mremove rast=* -f
g.mremove vect=* -f
g.copy studyarea@PERMANENT,studyarea
g.region Kenya
r.mapcalc "distrib.tmp=initDistrib"
years="2004 2005 2006"
times=`seq -w 1 16 365`
ptslasttime=2003_353
lasttime=2003.353
lastyear=2003
for year in $years; do
for i in $times; do
r.neighbors input=distrib.tmp output=distrib.grown.tmp size=5
method=maximum --o
r.mapcalc "distrib.tmp=(distrib.grown.tmp * suitable.$year.$i)"
# Collect the first point
if [ `shuf -i 1-10 -n 1` -le 9 ]; then
v.extract input=sim4_pts_${year}_${i}@results output=pts_selected
random=1 --quiet --o
else
v.extract input=sim17_pts_${year}_${i}@results output=pts_selected
random=1 --quiet --o
fi
# each reporter has a 10% chance of being wrong. So pick points
individually until there are 100 points with each reporter
possibly geting it wrong.
while [ `v.info pts_selected | grep -e "Number of points:" | awk '{
print $5 }'` -lt 25 ]; do
if [ `shuf -i 1-10 -n 1` -le 9 ]; then

117

v.extract input=sim4_pts_${year}_${i}@results output=pts_selected_4
random=1 --quiet --o
v.patch input=pts_selected_4 output=pts_selected -ae --o
else
v.extract input=sim17_pts_${year}_${i}@results output=pts_selected_17
random=1 --quiet --o
v.patch input=pts_selected_17 output=pts_selected -ae --o
fi
done
g.remove vect=pts_selected_4
g.remove vect=pts_selected_17
# random reporter
v.extract input=sim1_pts_${year}_${i}@results output=pts_random
random=25 --o
# correct reporter
v.extract input=sim15_pts_${year}_${i}@results output=pts_correct
random=25 --o
# wrong reporter
v.extract input=sim17_pts_${year}_${i}@results output=pts_wrong
random=25 --o
v.patch input=pts_random,pts_correct,pts_wrong output=pts_selected -ae
--o
g.remove vect=pts_random
g.remove vect=pts_correct
g.remove vect=pts_wrong
# Add columns
v.db.addcol pts_selected columns="reporter integer, score double,
neighbors double, suitable double, occ_ly double, occ_lt double,
support double, delta double"
# Add reporter IDs
val=1
for v in `v.category pts_selected option=print`; do
v.db.update pts_selected column=reporter value=$val where="cat=$v"
val=$val+1
done
# Initialize score
if [ ${year}${i} -eq 2004001 ]; then
echo "loop1"

118

v.db.update pts_selected column=score value=0
else
echo "loop2"
for r in `seq 1 1 100`; do
v.db.update pts_selected column=score value=`v.db.select
pts_selected_$ptslasttime columns=score where="reporter=$r" -c`
where="reporter=$r"
done
fi
# Number of neighbors
# The default is to sum all 9-cells in the neighborhood. The
weights.txt file specifically excludes the middle cell from being
included. This is the way it was published, but I might want to
include this in the computation of scores
r.neighbors input=distrib.tmp output=neighbors_count size=3 method=sum
weight=weights.txt
r.mapcalc "neighbors_count=neighbors_count/4"
v.what.rast vector=pts_selected raster=neighbors_count
column=neighbors
v.db.update pts_selected column=neighbors value=-1 where="neighbors=0"
v.db.update pts_selected column=neighbors value=0 where="neighbors is
null"
g.remove neighbors_count
# Suitable Habitat -- not included initially in the published paper
v.what.rast vector=pts_selected raster=suitable.$year.$i
column=suitable
v.db.update pts_selected column=suitable value=-1 where="suitable=0"
v.db.update pts_selected column=suitable value=0 where="suitable is
null"
# Occupied in t-1
v.what.rast vector=pts_selected raster=distrib.$lasttime column=occ_lt
v.db.update pts_selected column=occ_lt value=-1 where="occ_lt = 0"
v.db.update pts_selected column=occ_lt value=0 where="occ_lt is null"
# Occupied in y-1
v.what.rast vector=pts_selected raster=distrib.$lastyear.$i
column=occ_ly
v.db.update pts_selected column=occ_ly value=-1 where="occ_ly=0"
v.db.update pts_selected column=occ_ly value=0 where="occ_ly is null"
# Supporting reports
v.distance from=pts_selected to=pts_selected upload=dist col=support
dmax=250 dmin=1

119

v.db.update pts_selected column=support value=-1 where="support is
null"
v.db.update pts_selected column=support value=1 where="support > 0"
# compute delta score
v.db.update pts_selected column=delta value="neighbors + suitable +
occ_lt + occ_ly"
# update the overall score
v.db.update pts_selected column=score value="score + delta"
# discard points that don't meet minimum standards
score=`v.univar pts_selected column=score type=point | grep -e "mean:"
| awk '{ print $2 }'`
echo "mean score is ${score}"
v.db.update pts_selected column=value value=0 where="score <=
${score}"
# output to raster
v.to.rast input=pts_selected type=point output=pts_selected use=attr
column=value
r.null pts_selected null=0
# update distribution
r.mapcalc "distrib.tmp=if(pts_selected == 1, 1, distrib.tmp)"
g.copy distrib.tmp,distrib.$year.$i
g.copy vect=pts_selected,pts_selected_${year}_${i}
g.remove rast=pts_selected
g.remove vect=pts_selected
g.remove distrib.grown.tmp
lasttime=$year.$i
ptslasttime=${year}_${i}
if [ $i -eq 353 ]; then
lastyear=$year
fi
done
done
g.remove distrib.tmp
g.region zoom=studyarea
r.mask studyarea
r.series input=`g.mlist pat=distrib.200[4-6]* sep=,
mapset=expSim37_$rep` output=sim37_run$rep\_average
method=average

120

r.series input=`g.mlist pat=distrib.200[4-6]* sep=,
mapset=expSim37_$rep` output=sim37_run$rep\_sum method=sum
r.series input=`g.mlist pat=distrib.2004.* sep=, mapset=expSim37_$rep`
output=sim37_2004_run$rep\_average method=average
r.series input=`g.mlist pat=distrib.2004.* sep=, mapset=expSim37_$rep`
output=sim37_2004_run$rep\_sum method=sum
r.series input=`g.mlist pat=distrib.2005.* sep=, mapset=expSim37_$rep`
output=sim37_2005_run$rep\_average method=average
r.series input=`g.mlist pat=distrib.2005.* sep=, mapset=expSim37_$rep`
output=sim37_2005_run$rep\_sum method=sum
r.series input=`g.mlist pat=distrib.2006.* sep=, mapset=expSim37_$rep`
output=sim37_2006_run$rep\_average method=average
r.series input=`g.mlist pat=distrib.2006.* sep=, mapset=expSim37_$rep`
output=sim37_2006_run$rep\_sum method=sum
r.mask -r
curl https://prowlapp.com/publicapi/add -F
apikey=2daeeaf780b2e94281d1089752c6698da81434a9 -F
application="HPCC" -F description="TED simulation 37
complete"

121

, run $rep

ADDENDUM TO CHAPTER 2
Since the publication of Chapter 2 (Langley & Messina, 2011), the process for standing up a
postgres database with PostGIS Raster support has changed substantially. This appendix serves
to provide updated code and instructions for installing the database in Ubuntu 13.10.
Overview of Requirements
Package Versions Required (May 2014):
Proj4 (4.8.0)
http://download.osgeo.org/proj/proj-4.8.0.tar.gz
GEOS (3.4.2)
http://download.osgeo.org/geos/geos-3.4.2.tar.bz2
GDAL (1.10.1)
http://download.osgeo.org/gdal/1.10.1/gdal1101.tar.gz
GRASS (6.4.3)
http://grass.osgeo.org/grass64/source/grass-6.4.3.tar.gz
Add postgres repository to the package manager (Ubuntu 13.10 Saucy)
sudo sh -c 'echo "deb http://apt.postgresql.org/pub/repos/apt/ saucypgdg main" >> /etc/apt/sources.list'
wget --quiet -O - http://apt.postgresql.org/pub/repos/apt/ACCC4CF8.asc
| sudo apt-key add -

Install packages from repository
sudo apt-get install <package>

-

build-essential
postgresql-9.3
postgresql-server-dev-9.3
pgadmin3
libxml2-dev
libjson0-dev
xsltproc
docbook-xsl
docbook-mathml

Compile and Install GEOS 3.4.X
PostGIS 2.1 is best used with GEOS >= 3.4 for several new features, however Ubuntu 13.10 only
has GEOS 3.3.3 available in packages, so it needs to be built from source. If you don't need the
new features, instead install the libgeos-dev package.
wget http://download.osgeo.org/geos/geos-3.4.2.tar.bz2

122

tar xfj geos-3.4.2.tar.bz2
cd geos-3.4.2
./configure
make
sudo make install
cd ..

Compile and Install PostGIS
wget http://download.osgeo.org/postgis/source/postgis-2.1.1.tar.gz
tar xfz postgis-2.1.1.tar.gz
cd postgis-2.1.1

Implement Basic configuration for PostGIS 2.1, with raster and topology support:
./configure
make
sudo make install
sudo ldconfig
sudo make comments-install

Lastly, enable the command line tools to work from shell:
sudo ln -sf /usr/share/postgresql-common/pg_wrapper
/usr/local/bin/shp2pgsql
sudo ln -sf /usr/share/postgresql-common/pg_wrapper
/usr/local/bin/pgsql2shp
sudo ln -sf /usr/share/postgresql-common/pg_wrapper
/usr/local/bin/raster2pgsql

Spatially enabling the database
Connect to the database. To add raster support:
CREATE EXTENSION postgis;

To create topology support:
CREATE EXTENSION postgis_topology;

123

REFERENCES

124

REFERENCES
Adam, N. R. A., & Gangopadhyay, A. A. (1997). Database Issues in Geographic Information
Systems. Boston: Kluwer Academic Publishers.
Altmann, J., Alberts, S. C., Altmann, S. A., & Roy, S. B. (2002). Dramatic change in local climate
patterns in the Amboseli basin, Kenya. African Journal of Ecology, 40(3), 248-251. doi:
10.1046/j.1365-2028.2002.00366.x
Baird, T., Leslie, P., & McCabe, J. (2009). The Effect of Wildlife Conservation on Local Perceptions
of Risk and Behavioral Response. History Teacher, 37, 463-474.
Balram, S., & Dragićević, S. (2006). Collaborative Geographic Information Systems. Hershey: Idea
Group Publishing.
Batchelor, N., Atkinson, P., Gething, P., Picozzi, K., Fevre, E. M., Kakembo, A., & Welburn, S. C.
(2009). Spatial Predictions of Rhodesian Human African Trypanosomiasis (Sleeping
Sickness) Prevalence in Kaberamaido and Dokolo, Two Newly Affected Districts of
Uganda.
PLoS
neglected
tropical
diseases,
3(12),
e563.
doi:
10.1371/journal.pntd.0000563
Bauer, B. O., Kabore, I., Liebisch, A., Meyer, F., & Petrich-Bauer, J. (1992). Simultaneous control
of ticks and tsetse flies in Satiri, Burkina Faso, by the use of flumethrin pour on for cattle.
Tropical Medicine and Parasitology, 43(1), 41-46.
Bishr, M., & Kuhn, W. (2007). Geospatial information bottom-up: A matter of trust and semantics.
In S. I. Fabrikant & M. Wachowicz (Eds.), The European Information Society: Leading the
Way with Geo-information (pp. 365-387). Springer.
Bishr, M., & Mantelas, L. (2008). A trust and reputation model for filtering and classifying
knowledge about urban growth. GeoJournal, 72(3-4), 229-237. doi: 10.1007/s10708-0089182-4
Boroushaki, S., & Malczewski, J. (2010). Measuring consensus for collaborative decision-making:
A GIS-based approach. Computers, Environment and Urban Systems, 34(4), 322-332. doi:
10.1016/j.compenvurbsys.2010.02.006
Boroushaki, S., & Malczewski, J. (2010). Using the fuzzy majority approach for GIS-based
multicriteria group decision-making. Computers and Geosciences, 36(3), 302-312.
Brun, R., Blum, J., Chappuis, F., & Burri, C. (2010). Human African trypanosomiasis. The Lancet,
375(9709), 148-159. doi: 10.1016/S0140-6736(09)60829-1

125

Budhathoki, N. R., Bruce, B. C., & Nedović-Budić, Z. (2008). Reconceptualizing the role of the user
of spatial data infrastructure. GeoJournal, 72(3-4), 149-160. doi: 10.1007/s10708-0089189-x
Câmara, G., Souza, R., Freitas, U. M., & Garrido, J. (1996). SPRING: Integrating remote sensing
and GIS by object-oriented data modelling. Computers & Graphics, 20(3), 395-403.
Campbell, D. J., Gichohi, H., Mwangi, A., & Chege, L. (2000). Land use conflict in kajiado District,
Kenya. Land Use Policy, 17(4), 337-348.
Campbell, D. J., Lusch, D., Smucker, T., & Wangui, E. E. (2004). Root causes of land use change in
the Loitokitok Area, Kajiado District, Kenya. Land Use Change Impacts and Dynamics
(LUCID) Project Working Paper, 19.
Cecchi, G., Mattioli, R., Slingenbergh, J., & de La Rocque, S. (2008). Land cover and tsetse fly
distributions in sub-Saharan Africa. Medical and Veterinary Entomology, 22(4), 364-373.
doi: 10.1111/j.1365-2915.2008.00747.x
Chow, T. E. (2012). “We Know Who You Are and We Know Where You Live”: A Research Agenda
for Web Demographics (pp. 265-285). Dordrecht: Springer Netherlands.
Cipeluch, B., Jacob, R., Winstanley, A., & Mooney, P. (2010, 20-23rd July 2010). Comparison of
the accuracy of OpenStreetMap for Ireland with Google maps and Bing maps. Proceedings
of the Proceedings, Ninth International Symposium on Spatial Accuracy Assessment in
Natural Resources and Environmental Sciences.
Cohen, M. L. (2000). Changing patterns of infectious disease. Nature, 406(6797), 762-767.
Coleman, D., & Sabone, B. (2010). Volunteering geographic information to authoritative
databases: Linking contributor motivations to program characteristics. Geomatica, 64(1),
27-40.
Coleman, D. J., & Georgiadou, Y. (2009). Volunteered Geographic Information: the nature and
motivation of produsers. International Journal of Spatial Data Infrastructures Research,
4(1), 332-358.
Comber, A. J., Fisher, P. F., Harvey, F., Gahegan, M., & Wadsworth, R. (2006). Using metadata to
link uncertainty and data quality assessments. In A. Riedl, W. Kainz & G. A. Elmes (Eds.),
Progress in Spatial Data Handling: 12th International Symposium on Spatial Data
Handling (pp. 279-292). Springer Berlin Heidelberg.
Conati, C. (2004). How to Evaluate Models of User Affect? In T. Kanade, J. Kittler, J. M. Kleinberg,
F. Mattern, J. C. Mitchell, O. Nierstrasz, C. Pandu Rangan, B. Steffen, M. Sudan, D.
Terzopoulos, D. Tygar, M. Y. Vardi, G. Weikum, E. André, L. Dybkjær, W. Minker & P.
Heisterkamp (Eds.), Affective Dialogue Systems (Vol. 3068, pp. 288-300). Berlin,
Heidelberg: Springer Berlin Heidelberg.
126

Connors, J. P., Lei, S., & Kelly, M. (2012). Citizen Science in the Age of Neogeography: Utilizing
Volunteered Geographic Information for Environmental Monitoring. Annals of the
Association
of
American
Geographers,
102(6),
1267-1289.
doi:
10.1080/00045608.2011.627058
Corbett, J. (2012). “I Don’t Come from Anywhere”: Exploring the Role of the Geoweb and
Volunteered Geographic Information in Rediscovering a Sense of Place in a Dispersed
Aboriginal Community (pp. 223-241). Dordrecht: Springer Netherlands.
Cox, F. E. G. (2004). History of sleeping sickness (African trypanosomiasis). Infect Dis Clin N Am,
18, 231-245. doi: 10.1016/j.idc.2004.01.004
Craglia, M. (2007). Volunteered geographic information and spatial data infrastructures: When
do parallel lines converge. Position paper for the VGI Specialist Meeting, Santa Barbara
13-14 December 2007.
Craglia, M., de Bie, K., Jackson, D., Pesaresi, M., Remetey-Fülöpp, G., Wang, C., . . . Woodgate, P.
(2012). Digital Earth 2020: towards the vision for the next decade. International Journal
of Digital Earth, 5(1), 4-21. doi: 10.1080/17538947.2011.638500
Craglia, M., Goodchild, M. F., Annoni, A., Câmara, G., Gould, M., Kuhn, W., . . . Parsons, E. (2007).
Next-generation digital earth. International Journal of Spatial Data Infrastructures
Research, 3, 146-167.
Crosetto, M., & Rodriguez, J. P. (2001). Uncertainty and sensitivity analysis: tools for GIS-based
model implementation. International Journal Of Geographical Information Science, 15(5),
415-437. doi: 10.1080/13658810110053125
Devillers, R., & Jeansoulin, R. (2006). Fundamentals of spatial data quality. Newport Beach, CA:
ISTE.
DeVisser, M., & Messina, J. P. (2009). Optimum land cover products for use in a Glossinamorsitans habitat model of Kenya. Int J Health Geogr, 8(39), 39. doi: 10.1186/1476-072X8-39
DeVisser, M., Messina, J. P., Moore, N. J., & Lusch, D. (2010). A dynamic species distribution
model of Glossina subgenus Morsitans: The identification of tsetse reservoirs and refugia.
Eco Soc America, 1(1), 1-21.
Devogele, T., Parent, C., & Spaccapietra, S. (1998). On spatial database integration. International
Journal Of Geographical Information Science, 12(4), 335-352.
Drew, P., & Ying, J. (1996). GeoChange: an experiment in wide-area database services for
geographic information exchange. Paper presented at the ADL '96 Forum Forum on
Research
and
Technology
Advances
in
Digital
Libraries.
http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=502512
127

Egenhofer, M. J. (1994). Spatial SQL: A Query and Presentation Language (PDF). IEEE Transactions
on Knowledge and Data Engineering, 6(1), 86-95.
Elmasri, R., & Navathe, S. (2004). Fundamentals of database systems (4th ed.). Boston:
Pearson/Addison Wesley.
Elmes, G. A., Dougherty, M., Challig, H., Karigomba, W., McCusker, B., & Weiner, D. (2005). Local
knowledge doesn’t grow on trees: Community-integrated geographic information
systems and rural community self-definition. Developments in Spatial Data Handling, 2939.
Elwood, S. A. (2006a). Critical issues in participatory GIS: deconstructions, reconstructions, and
new research directions. Transactions in GIS, 10(5), 693-708.
Elwood, S. A. (2006b). Negotiating Knowledge Production: The Everyday Inclusions, Exclusions,
and Contradictions of Participatory GIS Research. The Professional Geographer, 58(2),
197-208. doi: 10.1111/j.1467-9272.2006.00526.x
Elwood, S. A. (2008a). Volunteered geographic information: future research directions motivated
by critical, participatory, and feminist GIS. GeoJournal, 72(3-4), 173-183. doi:
10.1007/s10708-008-9186-0
Elwood, S. A. (2008b). Volunteered geographic information: key questions, concepts and
methods to guide emerging research and practice. GeoJournal, 72(3-4), 133-135. doi:
10.1007/s10708-008-9187-z
Elwood, S. A. (2010). Geographic information science: emerging research on the societal
implications of the geospatial web. Progress in Human Geography, 34(3), 349-357. doi:
10.1177/0309132509340711
Elwood, S. A. (2011). Geographic Information Science: Visualization, visual methods, and the
geoweb.
Progress
in
Human
Geography,
35(3),
401-408.
doi:
10.1177/0309132510374250
Elwood, S. A., Goodchild, M. F., & Sui, D. (2013). Prospects for VGI Research and the Emerging
Fourth Paradigm. In D. Sui, S. A. Elwood & M. F. Goodchild (Eds.), Crowdsourcing
geographic knowledge (pp. 361-375). Dordrecht: Springer Netherlands.
Elwood, S. A., Goodchild, M. F., & Sui, D. Z. (2011). Researching Volunteered Geographic
Information: Spatial Data, Geographic Research, and New Social Practice. Annals of the
Association of American Geographers, 102(3), 110809092640007. doi:
10.1080/00045608.2011.595657
Elwood, S. A., & Leitner, H. (2012). GIS and Community-based Planning: Exploring the Diversity
of Neighborhood Perspectives and Needs. Cartography and Geographic Information
Systems, 25(2), 77.
128

Fischer, G., Shah, M., N Tubiello, F., & van Velhuizen, H. (2005). Socio-economic and climate
change impacts on agriculture: an integrated assessment, 1990-2080. Philosophical
Transactions of the Royal Society B: Biological Sciences, 360(1463), 2067-2083. doi:
10.1098/rstb.2005.1744
Flanagin, A. J., & Metzger, M. J. (2008). The credibility of volunteered geographic information.
GeoJournal, 72(3-4), 137-148. doi: 10.1007/s10708-008-9188-y
Ford, J. (1971). The role of the trypanosomiases in African ecology. A study of the tsetse fly
problem. Oxford: Clarendon Pr.
Frew, J. (2007). Provenance and volunteered geographic information. Workshop on Volunteered
Geographic
Information.
Retrieved
June
30,
2013
from
http://www.ncgia.ucsb.edu/projects/vgi/docs/position/Frew_paper.pdf
Girres, J. F., & Touya, G. (2010). Quality assessment of the French OpenStreetMap dataset.
Transactions in GIS, 14(4), 435-459.
Goodchild, M. F. (2007a). Citizens as sensors: the world of volunteered geography. GeoJournal,
69(4), 211-221.
Goodchild, M. F. (2007b). Citizens as sensors: web 2.0 and the volunteering of geographic
information. Geofocus, 7, 8-10.
Goodchild, M. F. (2007c). Citizens as Voluntary Sensors: Spatial Data Infrastructure in the World
of Web 2.0. International Journal of Spatial Data Infrastructures Research, 2, 24-32.
Goodchild, M. F. (2008a). Assertion and authority: the science of user-generated geographic
content. Proceedings of the Proceedings of the Colloquium for Andrew U Frank's 60th
Birthday, Department of Geoinformation and Cartography, Vienna University of
Technology, Vienna.
Goodchild, M. F. (2008b). Commentary: whither VGI? GeoJournal, 72(3), 239-244.
Goodchild, M. F. (2009). NeoGeography and the nature of geographic expertise. Journal of
Location Based Services, 3(2), 82-96. doi: 10.1080/17489720902950374
Goodchild, M. F. (2010a). Crowdsourcing geographic information for disaster response: a
research frontier. International Journal of Digital Earth, 3(3), 231-241. doi:
10.1080/17538941003759255
Goodchild, M. F. (2010b). Twenty years of progress: GIScience in 2010. Journal of Spatial
Information Science, 1(1), 3-20. doi: 10.5311/josis.v0i1.32

129

Goodchild, M. F., Guo, H., Annoni, A., Bian, L., de Bie, K., Campbell, F., . . . Woodgate, P. (2012).
Next-generation Digital Earth. Proceedings of the National Academy of Sciences, 109(28),
11088-11094. doi: 10.1073/pnas.1202383109
Goodchild, M. F., & Li, L. (2012). Assuring the quality of volunteered geographic information.
Spatial Statistics, 1, 110-120. doi: 10.1016/j.spasta.2012.03.002
Gray, J., & Szalay, A. (May 29, 2006). eScience: The Next Decade Will Be Exciting. Lecture
presented at ETH, Zurich. Retrieved from http://research.microsoft.com/enus/um/people/gray/talks/ETH_E_Science.ppt
Grira, J., Bédard, Y., & Roche, S. (2009). Spatial data uncertainty in the VGI world: Going from
consumer to producer. Geomatica, 64(1), 61-71.
Groot, R., & McLaughlin, J. D. (2000). Geospatial data infrastructure : concepts, cases, and good
practice. Oxford: Oxford University Press.
Groot, R. T. A. d. (2012). Evaluation of a volunteered geographical information trust measure in
the case of OpenStreetMap. (MS), University of Münster, Germany. Retrieved from
http://run.unl.pt/handle/10362/8301
Gyapong, J., Gyapong, M., Yellu, N., Anakwah, K., Amofah, G., Bockarie, M., & Adjei, S. (2010).
Integration of control of neglected tropical diseases into health-care systems: challenges
and opportunities. The Lancet, 375(9709), 160-165. doi: 10.1016/S0140-6736(09)612496
Haklay, M. M. (2010). How good is volunteered geographical information? A comparative study
of OpenStreetMap and Ordnance Survey datasets. Environment and Planning B: Planning
and Design, 37(4), 682-703. doi: 10.1068/b35097
Haklay, M. M. (2011, 27 November, 2011). Citizen Science as Participatory Science. Retrieved
from
http://povesham.wordpress.com/2011/11/27/citizen-science-as-participatoryscience/
Haklay, M. M. (2012). Citizen Science and Volunteered Geographic Information: Overview and
Typology of Participation. In D. Sui, S. A. Elwood & M. F. Goodchild (Eds.), Crowdsourcing
Geographic Knowledge: Volunteered Geographic Information (VGI) in Theory and Practice
(pp. 105-122). Dordrecht: Springer Netherlands.
Haklay, M. M., Basiouka, S., Antoniou, V., & Ather, A. (2010). How Many Volunteers Does it Take
to Map an Area Well? The Validity of Linus’ Law to Volunteered Geographic Information.
Cartographic Journal, The, 47(4), 315-322. doi: 10.1179/000870410X12911304958827
Haklay, M. M., Singleton, A., & Parker, C. (2008). Web Mapping 2.0: The Neogeography of the
GeoWeb. Geography Compass, 2(6), 2011-2039. doi: 10.1111/j.1749-8198.2008.00167.x
130

Haklay, M. M., & Weber, P. (2008). OpenStreetMap: User-Generated Street Maps. IEEE Pervasive
Computing, 7(4), 12-18. doi: 10.1109/MPRV.2008.80
Harvey, F. (2012). To Volunteer or to Contribute Locational Information? Towards Truth in
Labeling for Crowdsourced Geographic Information (pp. 31-42). Dordrecht: Springer
Netherlands.
Hellerstein, J. (1999). The GIST Indexing Project. from http://gist.cs.berkely.edu
House, R., Javidan, M., & Dorfman, P. (2001). Project GLOBE: An Introduction. Applied
Psychology, 50(4), 489-505. doi: 10.1111/1464-0597.00070
Howe, J. (2006, June 2006). The rise of crowdsourcing. Wired magazine, 14.
Irwin, A. (1995). Citizen Science: A Study of People, Expertise and Sustainable Development. New
York, NY: Routledge.
Johnson, P. T. J., & Thieltges, D. W. (2010). Diversity, decoys and the dilution effect: how
ecological communities affect disease risk. The Journal of experimental biology, 213(Pt 6),
961-970. doi: 10.1242/jeb.037721
Joint WHO Expert Committee and FAO Expert Consultation on the African Trypanosomiases
(1976: Rome). (1979). The African Trypanosomiases: report of a joint WHO expert
committee and FAO expert consultation, Rome, 8-12 November, 1976. Geneva. World
Health Organization. Retrieved from http://www.who.int/iris/handle/10665/41349
Keesing, F., Holt, R. D., & Ostfeld, R. (2006). Effects of species diversity on disease risk. Ecology
Letters, 9(4), 485-498. doi: 10.1111/j.1461-0248.2006.00885.x
Kennedy, P. (2005). Sleeping sickness-human African trypanosomiasis. British Medical Journal,
5(5), 260-267.
KETRI. (1996). Tsetse Distribution in Kenya Showing Tsetse Belts and Conservation Areas. Kenya
Trypanosomiasis Research Institute.
Kornacker, M. (1999). High-Performance Extensible Indexing. Paper presented at the VLDB
Conference, Edinburgh, Scotland. ftp://ftp.u-aizu.ac.jp/ftp/pub/dbms/gist/hiperf-gist.pdf
Koukoletsos, T., Haklay, M. M., & Ellul, C. (2012). Assessing Data Completeness of VGI through an
Automated Matching Procedure for Linear Data. Transactions in GIS, 16(4), 477-498. doi:
10.1111/j.1467-9671.2012.01304.x
Langley, S. A. Developing a participatory GIS network for the mapping of African trypanosomiasis
in Kenya [Unpublished Interview Data]. (2010).

131

Langley, S. A., & Messina, J. P. (2011). Embracing the Open-Source Movement for Managing
Spatial Data: A Case Study of African Trypanosomiasis in Kenya. Journal of Map and
Geography Libraries, 7(1), 87-113. doi: 10.1080/15420353.2011.534693
Langley, S. A., & Messina, J. P. (2013). Utilizing Volunteered Information for Infectious Disease
Surveillance. International Journal of Applied Geospatial Research, 4(2), 54-70.
Lans, R. F. v. d. (2007). Introduction To Sql: Mastering The Relational Database Language (4th
ed.). Upper Saddle River, NJ: Addison-Wesley.
Livingstone, D. N. (1992). The Geographical Tradition: Episodes in the History of a Contested
Enterprise. Malden, MA: Blackwell Publishing Ltd.
Longley, P. A., Goodchild, M., Maguire, D. J., & Rhind, D. W. (2005). Geographic Information
Systems and Science (2 ed.). Hoboken, NJ: John Wiley & Sons.
Maitima, J. M. (N.d.). The variabilities in plant phenological activities in response to rainfall and
temperature variability in Nguruman. Unpublished manuscript.
Makido, Y., Shortridge, A. M., & Messina, J. P. (2007). Assessing Alternatives for Modeling the
Spatial Distribution of Multiple Land-cover Classes at Sub-pixel Scales. Photogrammetric
Engineering and Remote Sensing, 73(8), 935.
Masser, I. (2005, Oct. 14-16). The future of spatial data infrastructures. Proceedings of the ISPRS
Workshop on Service and Application of Spatial Data Infrastructure, Hangzhou, China.
Masser, I. (2005). GIS worlds: creating spatial data infrastructures (1st ed.). ESRI Press.
Maué, P. (2007). Reputation as tool to ensure validity of VGI. Position Paper for Specialist Meeting
on
VGI.
Santa
Barbara,
CA:
Retrieved
from
http://www.ncgia.ucsb.edu/projects/vgi/docs/position/Maue_paper.pdf
McCall, M. K., & Minang, P. A. (2005). Assessing participatory GIS for community-based natural
resource management: claiming community forests in Cameroon. Geographical Journal,
171(4), 340-356.
McKnight, K. P., Messina, J. P., & Shortridge, A. M. (2011). Using Volunteered Geographic
Information to Assess the Spatial Distribution of West Nile Virus in Detroit, Michigan.
International Journal of Applied Geospatial Research, 2(3), 72-85.
Messina, J. P., Moore, N. J., DeVisser, M., McCord, P. F., & Walker, E. D. (2012). Climate Change
and Risk Projection: Dynamic Spatial Models of Tsetse and African Trypanosomiasis in
Kenya. Annals of the Association of American Geographers, 102(5), 1038-1048. doi:
10.1080/00045608.2012.671134
Metcalf, S., & Paich, M. (2005). Spatial dynamics of social network evolution. Vol, 51, 61801.
132

Metzger, M. J., Flanagin, A. J., Eyal, K., Lemus, D. R., & McCann, R. M. (2003). Credibility for the
21st century: Integrating perspectives on source, message, and media credibility in the
contemporary media environment. Communication yearbook, 27, 293-336.
Miller, C. C. (2006). A Beast in the Field: The Google Maps Mashup as GIS/2. Cartographica: The
International Journal for Geographic Information and Geovisualization, 41(3), 187-199.
doi: 10.3138/J0L0-5301-2262-N779
Moore, N. J., Alagarswamy, G., Pijanowski, B., Thornton, P. K., Lofgren, B., Olson, J., . . . Qi, J.
(2012). East African food security as influenced by future climate change and land use
change at local to regional scales. Climatic Change, 110(3-4), 823-844. doi:
10.1007/s10584-011-0116-7
Moore, N. J., & Messina, J. P. (2010). A Landscape and Climate Data Logistic Model of Tsetse
Distribution in Kenya. PloS one, 5(7), e11809. doi: 10.1371/journal.pone.0011809
Morris, D. L., Western, D., & Maitumo, D. (2009). Pastoralist’s livestock and settlements influence
game bird diversity and abundance in a savanna ecosystem of southern Kenya. African
Journal of Ecology, 47(1), 48-55. doi: 10.1111/j.1365-2028.2007.00914.x
Muriuki, G. W., Njoka, T., Reid, R., & Nyariki, D. (2005). Tsetse control and land-use change in
Lambwe valley, south-western Kenya. Agriculture, Ecosystems and Environment, 106(1),
99-107.
Neteler, M., Beaudette, D. E., Cavallini, P., Lami, L., & Cepicky, J. (2008). GRASS GIS. In G. B. Hall
& M. G. Leahy (Eds.), (Vol. 2, pp. 171-199). Berlin, Heidelberg: Springer Berlin Heidelberg.
O'Reilly, T. (2005, September 30). What is Web 2.0.
Retrieved June 30, 2013, from
http://oreilly.com/web2/archive/what-is-web-20.html
O'Reilly, T. (2006, December 10). Web 2.0 Compact Definition: Trying Again. Retrieved June 30,
2013, from http://radar.oreilly.com/2006/12/web-20-compact-definition-tryi.html
O'Reilly, T. (2007). What is Web 2.0: Design Patterns and Business Models for the Next
Generation of Software. Communications & Strategies, 1(17).
OGIS. (1999). Open GIS Simple Features Specification for SQL (Revision 1.1). Open Geospatial
Consortium. Retrieved from www.opengeospatial.org
Olson, J. E. (2003). Data Quality: The Accuracy Dimension. San Francisco, CA: Morgan Kaufmann
Publishers, Elsevier Science.
Oreskes, N., Shrader-Frechette, K., & Belitz, K. (1994). Verification, Validation, and Confirmation
of Numerical Models in the Earth Sciences. Science (New York, NY), 263(5147), 641-646.
doi: 10.1126/science.263.5147.641

133

Ostfeld, R. S., Keesing, F., & Eviner, V. T. (2008). Infectious Disease Ecology: Effects of Ecosystems
on Disease and of Disease on Ecosystems. Princeton, NJ: Princeton University Press.
Parent, C., Spaccapietra, S., & Zimányi, E. (2006). The MurMur project: Modeling and querying
multi-representation spatio-temporal databases. Information Systems, 31, 733-769.
Pickles, J. (1995). Ground truth: The social implications of geographic information systems. The
Guilford Press.
PostGIS (Version 1.3.2) (2008) [Computer Program]: Refractions Research Inc. Retrieved from
http://postgis.net
Python (Version 2.6.5) (2010) [Computer Software]: Python Software Foundation.
Raymond, E. (1999). The cathedral and the bazaar. Knowledge, Technology & Policy, 12(3), 2349.
Rivest, R. (1992). The MD5 message-digest algorithm. MIT Laboratory for Computer Science and
RSA Data Security, Inc. Retrieved from http://www.faqs.org/rfcs/rfc1321.html#b
Robbins, P. (2003). Beyond ground truth: GIS and the environmental knowledge of herders,
professional foresters, and other traditional communities. History Teacher, 31(2), 233253.
Rosenberg, T. (2011, Apr 28). Crowdsourcing a Better World - NYTimes.com, Editorial. New York
Times.
Retrieved
from
http://opinionator.blogs.nytimes.com/2011/03/28/crowdsourcing-a-better-world
Rouse, L. J., Bergeron, S. J., & Harris, T. M. (2007). Participating in the Geospatial Web:
Collaborative Mapping, Social Networks and Participatory GIS. In A. Scharl & K.
Tochtermann (Eds.), The Geospatial Web. London: Springer.
Shekhar, S., & Chawla, S. (2003). Spatial databases: A tour. Prentice Hall.
Shi, W., Fisher, P., & Goodchild, M. F. (2002). Spatial Data Quality. New York, NY: Taylor & Francis.
Sieber, R. E. (2006). Public participation geographic information systems: A literature review and
framework. Annals of the Association of American Geographers, 96(3), 491-507.
Silvertown, J. (2009). A new dawn for citizen science. Trends in Ecology & Evolution, 24(9), 467471. doi: 10.1016/j.tree.2009.03.017
Siringi, S. (2003). Kenya rejects drugs deal. The Lancet Infectious Diseases, 3(6), 320-320. doi: doi:
DOI: 10.1016/S1473-3099(03)00641-8
Smith, T., & Frew, J. (1995). Alexandria digital library. Communications of the ACM, 38(4), 62.
134

Snyder, J. D., & Merson, M. H. (1982). The magnitude of the global problem of acute diarrhoeal
disease: a review of active surveillance data. Bulletin of the World Health Organization,
60(4), 605.
Stonebraker, M., & Kemnitz, G. (1991). The POSTGRES next generation database management
system. Communications of the ACM, 34(10), 78-92. doi: 10.1145/125223.125262
Stonebraker, M., & Moore, D. (1995). Object Relational DBMSs: The Next Great Wave. San
Francisco, CA: Morgan Kaufmann Publishers Inc.
Stonebraker, M., & Rowe, L. (1986). The design of POSTGRES. Proceedings of the SIGMOD
Conference.
Sui, D. Z. (2008). The wikification of GIS and its consequences: Or Angelina Jolie's new tattoo and
the future of GIS. Computers, Environment and Urban Systems, 32.
Sutherst, R. (2004). Global change and human vulnerability to vector-borne diseases. Clin
Microbiol Rev, 17(1), 136-173. doi: Doi 10.1128/Cmr.17.1.136-173.2004
Tarimo-Nesbitt, R., Golder, T. K., Dransfield, R., Chaudhury, M. F., & Brightwell, R. (1999).
Trypanosome infection rate in cattle at Nguruman, Kenya. Veterinary Parasitology, 81(2),
107-117.
Tatem, A. J., Rogers, D. J., & Hay, S. I. (2006). Global Transport Networks and Infectious Disease
Spread. Advances in Parasitology, 62, 293-343. doi: 10.1016/S0065-308X(05)62009-X
Terblanche, J. S., Clusella-Trullas, S., Deere, J. A., & Chown, S. L. (2008). Thermal tolerance in a
south-east African population of the tsetse fly Glossina pallidipes (Diptera, Glossinidae):
Implications for forecasting climate change impacts. Journal of insect physiology, 54(1),
114-127.
Thacker, S. B., Choi, K., & Brachman, P. S. (1983). The Surveillance of Infectious Diseases. JAMA,
249(9), 1181-1185. doi: 10.1001/jama.1983.03330330059036
Tulloch, D. L. (2007). Many, many maps: Empowerment and online participatory mapping. First
Monday, 12(2).
Tulloch, D. L. (2008). Is VGI participation? From vernal pools to video games. GeoJournal, 72(34), 161-171. doi: 10.1007/s10708-008-9185-1
Turner, A. (2006). Introduction to Neogeography. O'Reilly.
Turner, M., & Hiernaux, P. (2002). The use of herders’ accounts to map livestock activities across
agropastoral landscapes in Semi-Arid Africa. Landscape Ecology, 17(5), 367-385. doi:
10.1023/A:1021238208019

135

van den Berg, H., Coetzee, S., & Cooper, A. K. (2011, May 31 - June 2). Analysing commons to
improve the design of volunteered geographic information repositories. Proceedings of
the AfricaGEO 2011, Cape Town, South Africa.
van Oort, P. (2005). Spatial data quality: from description to application. (PhD), Wageningen
University. Retrieved from http://edepot.wur.nl/38987
Waller, R. D. (1990). Tsetse fly in western Narok, Kenya. The Journal of African History, 31(1), 81101.
Warren, D. M. (1991). Using indigenous knowledge in agricultural development. World Bank.
Watson, H. J., Wixom, B. H., & Goodhue, D. L. (2004). Data warehousing: the 3M experience. In
H. R. Nemati & C. D. Barko (Eds.), Organizational Data Mining: Leveraging Enterprise Data
Resources for Optimal Performance (pp. 202). Hershey, PA: Idea Group Pub.
WHO. (2001). Report on African Trypanosomiasis (sleeping sickness). Paper presented at the
Report of the Scientific Working Group meeting on African trypanosomiasis Geneva, 4-8
June, 2001. http://www.who.int/tdrold/cd_publications/pdf/aftryp_swg.pdf
WHO. (2005). Control of Human African Trypanosomiasis: A Strategy for the African Region. 1-9.
WHO.

(2006). African Trypanosomiasis.
Retrieved February
http://www.who.int/media/centre/factsheets/fs259/en/

10,

2007,

from

Wiersma, Y. F. (2010). Birding 2.0: Citizen Science and Effective Monitoring in the Web 2.0 World.
Avian Conservation and Ecology, 5(2).
Wint, W. (2001). Kilometre resolution Tsetse Fly distribution maps for the Lake Victoria Basin and
West Africa. Food and Agriculture Organization/International Atomic Energy Agency Joint
Division. Retrieved from http://ergodd.zoo.ox.ac.uk/tseweb/iaea/tse1kmrep.doc
Yaukey, P. H. (2010). Citizen Science and Bird–Distribution Data: an Opportunity for Geographical
Research. Geographical Review, 100(2), 263-273.

136