 

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

 

DATE DUE DATE DUE DATE DUE

 

AUG 0 9252909 11

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/ClRC/Dale0ue.indd-p.1

 

 

 

 

ENHANCING RANGELAND SUSTAINABILITY WITH REMOTE SENSING AND
COLLABORATIVE INFORMATION EXCHANGE

By

H. Scott Butterﬁeld

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Plant Biology
Program in Ecology, Evolutionary Biology, and Behavior

2006

ABSTRACT

ENHANCING RANGELAND SUSTAINABILITY WITH REMOTE SENSING AND
COLLABORATIVE INFORMATION EXCHANGE

By

H. Scott Butterﬁeld

Remote sensing is a powerful tool for range management, but is used by only a
small fraction of private range managers. Factors limiting use of remote sensing may
include i) its cost and complexity, and ii) a dearth of means by which to quantify
senescent biomass. Senescent biomass is a signiﬁcant forage resource for livestock
during dry periods in grasslands. My research examines factors that inﬂuence the use of
remote sensing by private range managers and are associated with the use of remote
sensing for estimating rangeland biomass.

To understand what factors limit the use of remote sensing data by managers, I
conducted surveys with managers participating in a rangeland stewardship program in
California, in which they were provided with regular remote sensing-based analyses of
their properties. My work showed that managers of larger, commercially active ranches
found the experimental use of remote sensing to be a positive experience that convinced
them that this technology could help improve management. This suggests that the broad
use of remote sensing by managers of privately-held, commercial rangelands may be
limited in part by the simple lack of opportunity to test these technologies. Programs that
assist ranchers in obtaining appropriate products may thus be a cost-effective way to

enhance conservation on private rangelands. My ﬁndings suggest that voluntary self-

analysis by ranchers of the landscape dynamics of their own properties is likely to lead to
more engaged conservation efforts than top-down prescriptions.

Many grasslands experience a signiﬁcant annual dry period, during which
senescent biomass is the dominant canopy component. During these periods, remote
sensing indices such as the normalized difference vegetation index (N DVI) underestimate
total biomass (green and senescent), which can have signiﬁcant consequences for end-of-
the-season management decisions. Even though the general effect of senescence on the
NDVI-biomass relationship is well understood, no study has characterized this effect in
detail for annual grasses. To determine the period during which a single NDVI-biomass
relationship is useful, I grew annual grasses, and measured canopy properties weekly
from germination to the end of the season. NDVI underestimated biomass by
increasingly large amounts as the canopy transitioned from dominance by green to
senescent biomass. When the entire season was considered, the species-speciﬁc NDVI-
biomass equations were remarkably similar, suggesting that a single equation may be
robust to the structural and phenological differences that exist among annual grasses.

I used spectral measurements from these same grass stands to develop and
evaluate a vegetation index describing the mean PAR absorbed by the surface (MAPAR)
for senescent biomass estimates. Under some conditions, MAPAR was signiﬁcantly
correlated to total biomass throughout the season, including periods when senescent
biomass was dominant. However, the utility of MAPAR declined as the soil moisture or
organic matter content increased, and also when signiﬁcant stem lodging occurred.
These ﬁndings suggest that MAPAR, or a MAPAR-like index, may prove valuable as a

tool for evaluating senescent biomass in dry regions, such as California rangelands.

For my wife, Frances Knapczyk
And for my family, Lynn, Sue, Steve, Stacy, and Shelly Butterﬁeld

iv

ACKNOWLEDGMENTS

I would like to ﬁrst thank my advisor, Carolyn Malmstrom. Through the good
and bad times of graduate school, Carolyn has encouraged me to cultivate my strengths
and to work hard to overcome my weaknesses. She provided me the opportunity to work
on a fabulous rangeland stewardship program in California. Through involvement in this
program, I developed a passion for working with private range managers. I will have the
opportunity to continue working with managers as part of my new job with The Nature
Conservancy in California. I would also like to thank my guidance committee. They
have provided me with much needed insight and support during graduate school.

None of my work would have been possible without the inspirational range
managers that I worked with in California. I would also like to thank the staff at
Audubon-California, Jeanne Wirka, Judy Boshoven, Chris Rose, and Vance Russell, for
providing me, among many other things a place to sleep during long, hot summers in
California. I thank all of the ﬁeld assistants that helped me with my experiments in both
California and Michigan. Among those, Chatwin Hoe was integral in my dissertation
research. From countless conversations in the ﬁeld having to do with the virtue of
supporting Liverpool’s football club as opposed to Manchester United to sorting
countless bags of grass biomass, Chatwin was an amazing technician and ﬁiend. I also
thank Clifford Zehr and Brian Graff for their logistical support. Cliff was always only a

phone call away from helping me avoid a potential catastrophe in the ﬁeld.

Because I took the plunge to include a “non-traditional” outreach study as part of
my dissertation, I would like to thank all of the people who helped me develop this
chapter, and encouraged its development and publication! First, Carolyn was
instrumental in motivating me to start this chapter, and encouraging me to continue to
develop it. I am a ﬁrm believer that it was this chapter that landed me my dream job with
The Nature Conservancy. I thank Ann Austin, Diane Ebert-May, and Patricia Farrell for
taking time out of their busy lives to help with the survey and interview protocols as well
as with early drafts of the manuscript. The value of their advice cannot be measured with
words. Additionally, they introduced me to a wonderful group of people at MSU with
whom I otherwise would never have had the chance to interact.

Countless people provided needed insight and fresh eyes on draﬁs of all of these
chapters: Ann Austin, Mark Brunson, Gaylon Campbell, Jan Distlebrink, Diane Ebert-
May, Sarah Emery, Valerie Eviner, Patricia Farrell, Matt Germino, Bruce Kindel,
Frances Knapczyk, Joseph Kolowski, Lynn Lauerman, Peter Murphy, J iaguo Qi, Suzanne
Sippel, Stephen South, Craig Tweedie, Chuck Vaughn, Keith Weber, and Toshi Yoshida.

I would like to acknowledge my undergraduate advisor at Boston University,
Thomas Kunz, who inspired me to take on a career in ecology. I am forever indebted!

Finally, I want to thank my family for their love and support. They have been an
amazing source of inspiration and pride during this long process. I fear without my
wonderful wife I would not have made it through graduate school. She inspired me to
follow my dreams and passions. Words cannot describe what my family means to me —

this degree is as much for them as it is for me.

vi

TABLE OF CONTENTS

LIST OF TABLES ................................................................................... ix
LIST OF FIGURES .................................................................................. xi
CHAPTER I
INTRODUCTION .................................................................................... 1
Background ............................................................................................ 1
Organization of Dissertation ....................................................................... 1 1
CHAPTER 2
EXPERIMENTAL USE OF REMOTE SENSING BY PRIVATE RANGE MANAGERS
AND ITS INFLUENCE ON MANAGEMENT DECISIONS ................................ 14
Abstract ............................................................................................... 14
Introduction ............ . ............................................................................... 1 5
Methods ............................................................................................... 20
Growing season terminology .............................................................. 20
Remote sensing products ................................................................... 20
Surveys and interviews of land managers .............................................. 21
Analysis of responses ...................................................................... 22
Results ................................................................................................ 23
Cattle 1. Manager characteristics and experience ...................................... 24
Cattle 1. Inﬂuence of remote sensing on management decisions ................... 25
Cattle 2. Manager characteristics and experience ..................................... 27
Cattle 2. Inﬂuence of remote sensing on management decisions ................... 27
Sheep 1. Manager characteristics and experience ...................................... 29
Sheep 1. Inﬂuence of remote sensing on management decisions ................... 30
Sheep 2. Manager characteristics and experience ...................................... 31
Sheep 2. Inﬂuence of remote sensing on management decisions ................... 31
Discussion ........................................................................................... 32
Factors inﬂuencing the use of remote sensing by managers .......................... 32
Inﬂuence of remote sensing on decision making ...................................... 35
Management Implications .......................................................................... 37
CHAPTER 3
PHENOLOGICAL EFFECTS ON REMOTELY-SENSED BIOMASS ESTIMATES IN
ANNUAL GRASSLANDS ......................................................................... 39
Abstract ............................................................................................... 39
Introduction ........................................................................................... 40
Methods ............................................................................................... 43
Experimental design ........................................................................ 43
Measurements ............................................................................... 45
NDVI ................................................................................ 45

vii

Biomass .............................................................................. 46

LAI ................................................................................... 46
fAPAR ............................................................................... 47
Vegetation characteristics ........................................................ 48
Biomass equation analyses ....................................................... 49
Results ................................................................................................. 49
Seasonal canopy dynamics ................................................................ 49
Relationship between NDVI, fAPAR, and biomass ................................... 51
Generality of biomass equations ............................................................. 54
Discussion ............................................................................................. 60
CHAPTER 4
REMOTE SENSING-BASED ESTIMATES OF SENESCENT BIOMASS: COMMON
PROBLEMS AND A NEW APPROACH IN ANNUAL GRASSLANDS .................. 64
Abstract ............................................................................................... 64
Introduction .......................................................................................... 65
Methods ............................................................................................... 71
Index conceptualization in Michigan grass stands .............................................. 71
MAPAR ............................................................................ 71
Biomass .............................................................................. 73
Solar zenith angle tests ............................................................ 75
Soil background effects ........................................................... 76
Tests of MAPAR in California annual grasslands ...................................... 78
fAPAR, LAI, and biomass ........................................................ 78
Satellite-based MAPAR estimates .............................................. 81
Results ................................................................................................ 86
Tests ofMAPAR in Michigan grass stand586
fAPAR, LAI, and biomass ........................................................ 86
Solar zenith angle tests ............................................................ 90
Soil background effects ........................................................... 90
Tests of MAPAR in California annual grasslands ..................................... 92
fAPAR, LAI, and biomass ........................................................ 92
Satellite-based MAPAR estimates .............................................. 92
Discussion ............................................................................................ 96
MAPAR in Michigan grass stands97
MAPAR in California annual grasslands .............................................. 100
MAPAR for RDM estimates ............................................................ 101
CHAPTER 5
CONCLUSIONS ................................................................................... 105
BIBLIOGRAPHY ................................................................................. 1 09

viii

LIST OF TABLES

CHAPTER 2

Table 2.1. Motivation for involvement in the stewardship program. Responses were
given in survey 1, before managers were provided access to the remote sensing products,

on a Likert scale (1 = no motivation to 5 = high motivation) ................................. 18
Table 2.2. Management practices tested during the stewardship program .................. 19
CHAPTER 3

Table 3.1. ANCOVA results for differences in biomass equations among stand types.
Species, the independent categorical effect variable, include: A. fatua, B. hordeaceus, L.
multiﬂorum, and A. fatua-B. hordeaceus ........................................................................... 55

Table 3.2. Stand-speciﬁc biomass equations. NDVI equations are based on green
biomass data only. fAPAR and LAI equations are based on green plus senescent biomass
data. Note that these biomass equations are given for comparative purposes only, not as
deﬁnitive equations for all grassland situations. See Figure 3 for a graphical
representation of these equations .................................................................. 56

CHAPTER 4

Table 4.1. Phenological effects on MAPAR-based estimates in the Michigan Bromus
stands. Fractional cover values represent weekly means. Fractional soil cover was
calculated as the midpoint of Daubenmire classes (Daubenmire 1968). Fractional
senescent cover is the percentage of total vegetation cover. Biomass was only harvested
six times over the growing season. R2 values represent relationships with MAPAR.
Signiﬁcant at: p<0.001 = ***, p<0.01 = **, and p<0.05 = * ................................. 74

Table 4.2. The inﬂuence of soil type and condition on the ability of MAPAR to
discriminate vegetation from soil .................................................................. 77

Table 4.3. Results of the t-tests for differences between MAPAR-based vegetation and
soil values. The Michigan grass stand comparisons were made using green and senescent
biomass data (weeks 1—15). For the ground-based California annual grassland
comparisons, only green biomass was considered. The Landsat satellite-based California
annual grassland comparisons were made either in May, when senescent biomass was
dominant, or March, when green biomass was dominant ...................................... 93

Table 4.4. MAPAR-based RDM approaches. The Direct Approach quantiﬁes fall RDM
directly using data acquired before the ﬁrst fall rains. The Indirect Approach estimates
fall RDM using data acquired in late spring at maximum biomass together with the

ix

Bartolome et al. (2002) RDM algorithm. The Change Detection approach does not
quantify fall RDM directly, but instead can be used to identify pastures where RDM
levels are trending outside of the recommended range... ... .. ......................................... 103

LIST OF FIGURES

CHAPTER 3

Figure 3.1. Seasonal changes in A) NDVI and fAPAR; B) green (GRN) and senescent
(SEN) biomass; C) senescent vegetation cover; and D) LAI and height. Values are mean
weekly measurements averaged over all ﬁve stand types (N = 50) during early growth,
canopy maturation, and canopy decline. Weeks 5 and 9 (shaded columns) were
maximum greenness (MAX GRN) and maximum biomass (MAX B10). All ﬁve stand
types were combined because they displayed the same phenological trends except in
weeks 9 and 10; a severe storm caused stem lodging in the Avena and A vena-Bromus
plots in week 9, which impacted height and cover estimates in weeks 9 and 10 (see text
for details) ............................................................................................ 50

Figure 3.2. Seasonal relationships between biomass and A) NDVI, B) fAPAR and C)
LAI during early growth: germination (harvest 1: week 1) to maximum greenness;
canopy maturation: onset of senescence (harvest 3: week 6) to maximum biomass
(harvest 4: week 9); and canopy decline: maximum biomass to the end of the season
(harvest 6: week 15). Harvest 2 occurred in week 2 during early growth and harvest 5 in
week 10 during canopy decline. Values represent weekly means from A. fatua-B.
hordeaceus (open squares), A. fatua (open circles), B. hordeaceus (closed triangles), and
L. multiﬂorum (closed circles) stands. Numbers indicate biomass harvests, not weeks..52

Figure 3.3. Stand-speciﬁc biomass relationships. NDVI-biomass relationships represent
data for green biomass only. See Table 2 for equations, R2 and p values, and footnotes
detailing the range of values used to develop these equations. The top line is A. fatua, the
2"d line from the top is A. fatua-B. hordeaceus mixture, the 3“1 line from the top is B.
hordeaceus, and the bottom line is L. multiﬂorum ............................................. 57

Figure 3.4. Stand-speciﬁc biomass relationships. fAPAR-biomass relationships represent
data for green plus senescent biomass. See Table 2 for equations, R2 and p values, and
footnotes detailing the range of values used to develop these equations. The top line is A.
fatua, the 2nd line from the top is L. multiflorum, the 3'd line from the top is A. fatua-B.
hordeaceus mixture, and the bottom line is B. hordeaceus .................................... 58

Figure 3.5. Stand-speciﬁc biomass relationships. LAI-biomass relationships represent
data for green plus senescent biomass. See Table 2 for equations, R2 and p values, and
footnotes detailing the range of values used to develop these equations. The top line is B.
hordeaceus, the 2nd line from top is A. fatua-B. hordeaceus mixture, the 3rd line from the
top is A. fatua, and the bottom line is L. multiﬂorum .......................................... 59

CHAPTER 4

xi

Figure 4.1. Seasonal changes in surface reﬂectance in stands of A vena (high and low
density), Bromus, Avena-Bromus, and Lolium ﬁom week 1 (“germination”) to 5
(maximum greenness: maximum NDVI) to 9 (maximum biomass, fAPAR, and LAI) to
15 (canopy death). Values represent weekly means across all ﬁve stand types. Surface
reﬂectance was derived from UniSpec measurements. The blue (B: 450— 520 nm), green
(G: 520—600 nm), red (R: 630—690 nm), and near-infrared (N IR: 760—900 nm)
wavelength regions for Landsat satellite sensors are identiﬁed for reference ............... 70

Figure 4.2. Relationship between MAPARfu" and MAPAR,m (unweighted) in A) the
Michigan Bromus stands (N = 150) and B) the California annual grassland plots (N =
169). Relationships in the Bromus stands are shown for the green (GRN: weeks 1-7) and
senescent (SEN: weeks 8-15) periods as well as for the entire season (weeks 1-15) ...... 80

Figure 4.3. Relationships between A) MAPAR".1 (unweighted) (Eq. 4) and MAPARtmwcighwd)
(Eq. 5), and B) MAPARfun (Eq. 3) and MAPAR,m (weighted) in the Michigan Bromus stands
(N = 150); B) MAPARm (unweighted) and MAPARm(wcighwd) in the California annual

grassland plots (N = 169) ........................................................................... 83

Figure 4.4. Seasonal changes in A) NDVI and MAPARmn (Eq. 3) and B) fAPAR and
LAI; week 5 is maximum greenness and 9 is maximum biomass. Between weeks 9 and
10, there was a strong storm that caused signiﬁcant stem lodging and facilitated invasion
by green weedy vegetation; LAI decreased as the plots were weeded. Values represent
weekly means for the Michigan Avena (high and low density), Bromus, and Avena-
Bromus stands (11 = 40) ............................................................................... 87

Figure 4.5. Seasonal patterns in the relationship between MAPARfuﬂ (Eq. 3) and biomass
in A) all of the Michigan stand types combined (A vena high and low density, Bromus,
and Avena-Bromus), and B) Bromus stands alone. Relationships are shown for the green
(GRN: harvests 1—3; weeks 1, 3, and 6) and senescent (SEN: harvests 4—6; weeks 9, 10,
and 15) periods as well as for the entire season. In B), the solid line is for the GRN time
period and the dashed line is for the SEN time period ......................................... 88

Figure 4.6. Seasonal relationship between MAPARfun (Eq. 3) and A) fAPAR, and B)
LAI. Numbers (1-15) represent weeks, not harvests. Values represent means in the
Michigan Avena, Bromus, and Avena-Bromus stands (11 = 40) ............................... 89

Figure 4.7. Seasonal relationship between MAPARfun (Eq. 3) and solar zenith angle
(SZA, measured in degrees). MAPARfun values represent weekly means for the
Michigan Avena (high and low density), Bromus, and Avena-Bromus stands (11 = 40). . .91

Figure 4.8. Relationship between MAPAR!m (unweigmd) (Eq. 4) and A) LAI, B) biomass,
and C) fAPAR in California annual grassland plots during the period of maximum
greenness (28—29 March 2003). MAPARm (unweighted) was derived from UniSpec
measurements. LAI and fAPAR were derived from AccuPAR measurements ................ 94

xii

CHAPTER 1
INTRODUCTION

Background

Rangelands comprise approximately 315 million hectares, or some 40% of the
land mass in the United States (World Resources Institute 1996). Of this, approximately
one-half is privately-owned (USDA Natural Resources Conservation Service 2000) and
thus susceptible to pressures from urbanization (e.g., Heady and Child 1994, Smith 2000,
Haver 2001 , Jensen 2001). Rangelands are composed of grasses, forbs, shrubs, and trees,
and are predominantly used for livestock grazing in the western United States (Heady and
Child 1994). Rangelands can include natural grasslands, savannas, shrublands, many
deserts, tundra, alpine communities, coastal marshes, and wet meadows (Heady and
Child 1994, Barbour et a1. 1999).

Rangelands in the western United States provide a variety of ecosystem services,
including forage for grazing animals, clean air, and open spaces, and habitat for native
wildlife and plant species (Heady and Child 1994). Rangelands can also serve as high
quality watersheds if levels of residual dry matter (RDM)--the amount of dry plant
material on the ground at the beginning of the season--are maintained within
recommended ranges (Bartolome et al. 2002) and if riparian areas are properly managed
(e.g., fenced and either ungrazed or grazed only irregularly and by a small number of
grazing animals). In addition, rangeland ecosystems provide important economic
services, including supporting an approximately $80 billion beef and cattle industry as
well as private recreational (e.g., hunting) pursuits (World Resources Institute 1996,

Mitchell 2000)

Because of the collective importance of the services that rangeland ecosystems
provide, the “health” of rangelands often directly inﬂuences the economic well-being of
communities in the region, as is the case in the western United States (Heady and Child
1994, Mitchell 2000, O’Brien et a1. 2003). Deﬁning and monitoring rangeland health is a
complex process, and one for which a variety of state and federal agencies (e. g., the
Bureau of Land Management) as well as nonproﬁt organizations (e. g., the Central Coast
Rangeland Coalition) have developed their own unique systems (e. g., Bureau of Land
Management 1996, Ford et al. 2006). In these systems, rangeland health is often
evaluated by assessing a number of the following indicators: (i) degree of soil stability
and watershed function; (ii) integrity of nutrient cycles and energy ﬂows, including plant
productivity and the presence of desirable forage species; (iii) resistance and resilience to
unexpected or catastrophic disturbances, which is evaluated by measuring multiple
factors including age class distributions and plant vigor; (iv) maintenance of biological
diversity and habitat quality, and (iv) socio-economic sustainability of the rangeland and
ranching operation (The Nature Conservancy 2000, 2005, 2006, Ford et a1. 2006). Here,
I deﬁne socio-economic sustainability as the use of rangeland resources in such a way as
to meet present living costs while preserving natural resources and income sources for
future generations.

Enhancing rangeland “health” and achieving long-term sustainability are
important goals for many ranchers and conservation biologists, and thus motivating
factors for much work, including my own research. Because they are such broad goals,
however, my work has focused on the more immediate speciﬁc steps of evaluating and

expanding remote sensing use by land managers, with a particular focus on using remote

sensing to evaluate properties, such a biomass production, that are integral to analyses of
rangeland health, as described above. In the long-term, scientiﬁc advances in this area, in
combination with economic analyses, will likely help managers meet goals in rangeland
sustainability.

Broad questions about rangeland health and sustainability are important because
at present many rangeland ecosystems and the communities they support are threatened
by multiple factors, which include: (i) rising levels of invasive noxious weeds (e.g., Mack
1989, D’Antonio and Vitousek 1992, Vitousek et al. 1996, DiTomaso 2000); (ii) global
warming, which threatens to make rangelands more vulnerable to invasion by exotic
species (e.g., Alward et al. 1999); (iii) nitrogen deposition, which can favor the increased
dominance of exotic species (McLendon and Redente 1991, Field et al. 1992, Fenn et al.
1998, Koide et al. 1998, Weis 1999, Aber and Melillo 2001); and (iv) urbanization (e. g.,
Heady and Child 1994, Wilcove et al. 1998, Smith 2000, Haver 2001, Jensen 2001). In
combination, these threats are making it increasingly difﬁcult for private range managers
to maintain economically viable ranch operations (Leitch et al. 1994, World Resources
Institute 1996, Mitchell 2000). As economic margins decline while rangeland property
taxes and the value of land for urban development rise, more rangeland managers--even
many of whom who are part of multi-generation ranching families--consider selling their
properties (The Nature Conservancy 2005). When ranches are converted to urban and
suburban uses, large areas of open space are then lost, along with the ecosystem and
economic services that these rangelands provided (The Nature Conservancy 2000, 2005,
2006, Heady and Child 1994). Such sales therefore can have lasting detrimental

consequences, not only for the native species that live there but also for human

communities (World Resources Institute 1996, Roling and Wagemakers 1998, Mitchell
2000, The Nature Conservancy 2005).

In my research, I chose to work with private range managers, because they are
important contributors to the management and overall conservation of rangeland
ecosystems. Many of these managers are strongly committed to conservation in order to
preserve viable ranching operations for future generations of their families (Jensen 2001,
Butterﬁeld and Malmstrom 2006, The Nature Conservancy 2006). In addition, private
range managers possess a wealth of invaluable expert knowledge about ecological
dynamics in their regions, often described as “local ecological knowledge”(Berkes et al.
2000, Olsson and Folke 2001).

Interest in local ecological knowledge, or knowledge held by a speciﬁc group of
people (e. g., the western range management community in this case) about their local
ecosystems and the interplay among organisms and their environment (Olsson and Folke
2001), has grown recently within the scientiﬁc community. Researchers have shown that
such knowledge can contribute to the conservation of biodiversity (Gadgil et al. 1993),
the protection of endangered species (Colding 1998), the preservation of threatened
ecological processes (Alcom 1989), and increases in the overall sustainability of resource
use (Berkes et a1. 2000).

To meet the growing needs and pressures of management in rangeland
ecosystems, range managers have begun to evaluate new management practices designed
to decrease levels of noxious weeds and in the process increase total biomass production.
These practices include prescribed burning, short-duration, high-intensity grazing

(SDHI), and seeding of native perennial bunchgrasses (e. g., Thomsen et al. 1990, Taylor

and Ralphs 1992, Coppock and Birkenfeld 1999, Krueter et a1. 2001, Malmstrom et a1.
2004, DiTomaso and Johnson 2006). Noxious weeds, such as yellow starthistle
(Centaurea solstitialis L.), medusahead (T aeniatherum caput-medusae (L.) Nevski), and
leafy spurge (Euphorbia esula L.), are especially problematic in rangeland ecosystems in
the western United States because of their negative impact on livestock operations (Lusk
et al. 1961, Lym and Kerby 1987, Young 1992, Callihan et al. 1992, Callihan et al. 1995,
Vitousek et a1. 1996, DiTomaso 2000, DiTomaso and Johnson 2006). Noxious weeds
can reduce livestock carrying capacity (Lym and Kerby 1987, Heady and Child 1994,
Leitch et a1. 1994), and quickly decrease biomass production (Mooney et al. 1986, Mack
1989, D’Antonio and Vitousek 1992) as well as the quality (e. g., nutritional value) of
forage available for grazing animals (Lusk et al. 1961, Barry 1995). Some noxious
weeds (e. g., yellow starthistle) can even be toxic to livestock (Panter 1990). Once
noxious weeds have become established, the cost associated with the management of the
ecosystem as a whole increases greatly (Roche and Roche 1991, Leitch et al. 1994,
Randall 1996, DiTomaso 2000, Tu et al. 2001, DiTomaso and Johnson 2006).

To gain assistance in evaluating these new management practices, some managers
have formed watershed collectives with each other and forged relationships with non-
proﬁt conservation organizations and university scientists who can provide scientiﬁc
analyses (e.g., Haver 2001, Jensen 2001, Qi et al. 2002, Malmstrom et al. 2004, The
Nature Conservancy 2000, 2005, 2006). These collaborations provide managers the
unique opportunity to evaluate a variety of different practices in combination with one
another, and to collaborate in assessing which practices will decrease noxious weeds and

increase levels of desirable forage species most. In many cases, it is likely that managers

could increase the effectiveness of their ranch operations if they were able to monitor
rangeland vegetation in detail over their entire properties each year. This capability
would be especially helpful for identifying noxious weed infestations when they are small
and more easily controlled and/or eradicated (Everitt et al. 1995, Lass et al. 1996,
Boswell 2000, Lass et a1. 2002, Everitt et al. 2006, DiTomaso and Johnson 2006).
However, managers often cannot monitor their entire properties each year using ground-
based efforts alone because their ranches are large and ecologically complex (George and
Fulgham 1989, Rowan and Conner 1995, Tueller 1989, Coppock and Birkenfeld 1999,
Tueller 2001).

Remote sensing, however, can help managers conduct annual property-wide
assessments of range condition (Tueller 1989, Hunt et al. 2003, Wallace et al. 2003, Lass
et al. 2005, Washington-Allen et a1. 2006). For example, managers can use remote
sensing data to quantify the effect of their management practices on noxious weed spread
and biomass production from scales of individual pastures to entire watersheds. This
allows managers to quantify the impact of each practice within the context of non-
management induced variations, such as those related to topographic and soil differences
(Tueller 1982, Richardson and Everitt 1992, Pickup et al. 1994, Wessman et al. 1997,
Wallace et al. 2003).

Range managers can also use remote sensing data to expand the temporal scale of
their analyses. Many satellite sensors acquire data at least twice a month across western
United States rangeland ecosystems. Some of these sensors, like the Landsat series, have
been acquiring data for more than 20 years. For example, because of the time period over

which it has been acquired, Landsat satellite data can provide managers the unique

opportunity to analyze the effects of their current management practices within the
context of historical land use and land cover changes across their properties (Paruelo and
Golluscio 1994, Saltz et al. 1999, Wallace et al. 2003, Malmstrom et a1. 2004, Shaw
2005, Reeves et al. 2006, Washington-Allen et al. 2006). This capability, for example,
allows managers to determine which practices were most successful in the past at
decreasing noxious weed levels, and thus to focus future management efforts on those
which were most effective (Malmstrom et a1. 2004, Washington-Allen 2006). Remote
sensing data can also provide managers the unique ability to evaluate the impact of their
current management practices across temporal scales from months to decades (e.g.,
Tucker et al. 1983, Pickup et a1. 1994, Wessman et a1 1997, Wallace et al. 2003). This
capability enables managers to base management decisions on both short- and long-term
biomass and weed trends (e. g., Hunt et al. 2003, Lass et a1. 2005, Mustafa et al. 2005,
Everitt et al. 2006, Mundt et a1. 2006).

Despite their value, remote sensing data currently are used by only a small
fraction of range managers (Tueller 1982, 1989, Daberkow and McBride 2000, 2003).
The low use rates of innovative technologies, like remote sensing, often indicate either
that the technology has not been successﬁilly introduced to the end-user (i.e., the range
manager in this case) or that the end-user does not see its utility (F liegel 1993, Kreuter et
al. 2001, Daberkow and McBride 2003).

Another potential limitation is the current lack of remote sensing approaches for
quantifying total biomass during time periods when senescent biomass is present.
Senescent biomass is an important forage resource in grass-dominated rangeland

(grassland) ecosystems for a substantial portion of the year (e. g., George and Fulgham

1989, Richardson and Everitt 1992, Frank and Aase 1994, Qi et al. 2000). During dry
periods, senescent biomass is the only forage resource available for grazing animals
(George and Fulgham 1989, Richardson and Everitt 1992, Frank and Aase 1994, Qi et al.
2000). In these ecosystems, this can mean that senescent biomass is the only forage
resource available for up to six months during the year (George and Fulgham 1989,
Prince 1991, Saltz et al. 1999).

The amount of senescent plant material on the ground at the beginning of the
season, known as residual dry matter (RDM), is also an important indicator of rangeland
health (Bentley and Talbot 1951, Heady 1956, Bureau of Land Management 1996,
Bartolome et al. 2002, Guenther and Christian 2005, Ford et al. 2006). In grazed
grasslands, RDM is primarily composed of the foliage and stem biomass of grasses and
forbs from the current season (Bartolome et al. 2002, George et a1. 2006). However, in
ungrazed grasslands, RDM can also include dry tree leaves, woody debris, and grass and
forb litter from previous growing seasons (i.e., more than a year old). In western
rangeland ecosystems, properly managed RDM can reduce soil erosion, and thus increase
biomass production of desirable forage species and decrease levels of broadleaf noxious
weeds, such as yellow starthistle (Bartolome et al. 1980, McDougald et al. 1982, Heady
and Child 1994, George and Menke 1996, Molinar et al. 2001, Bartolome et al. 2002).
RDM measurements are used not only by range managers, but also by federal agencies,
such as the Bureau of Land Management, for compliance monitoring across federally
owned and/or managed grassland ecosystems (Bureau of Land Management 1996). In
addition, RDM assessments are used by range management specialists and conservation

organizations, such as The Nature Conservancy, for both compliance and effectiveness

monitoring across conservation easements (e.g., Guenther 1998, The Nature Conservancy
2000, 2005, 2006, Molinar et al. 2001, Guenther and Christian 2005).

Remote sensing data have been used for more than 30 years to quantify green
biomass in grassland ecosystems (Rouse et al. 1974). However, during periods when
senescent biomass is present, vegetation indices, such as the normalized difference
vegetation index (N DVI), underestimate total biomass. NDVI is calculated using surface
reﬂectance (R) values in the red (0.63—0.69 pm) and near infrared (N 1R) (0.76—0.9 um)
spectral regions as:

NDVI = (RNIR - Rred) / (RNIR+ Rm) [1311- 1]
NDVI is highly correlated to green biomass because it is calculated using the red spectral
region, which is sensitive to chlorophyll amount, and the NIR region, which is sensitive
to both leaf internal and canopy structure (Rouse et al. 1974, Tucker 1979). Thus, as
green biomass increases, NDVI increases. However, as vegetation senesces, chlorophyll
degrades and leaf internal structure declines (Tucker 1979). This results in NDVI values
more similar to that of the soil background (Huete et al. 1985), and causes NDVI to be a
poor predictor of total biomass during these time periods (e.g., Tucker 1979, Tucker et al.
1983, Gamon et al. 1995).

To assess grassland condition during time periods when senescent biomass is
present, researchers have thus developed a variety of approaches, including: 1) thermal
remote sensing data coupled to NDVI measurements (French et al. 2000); 2) vegetation
indices that use the shortwave infrared region (SWIR: 2000—2300 nm) (e. g., Qi et al.
2000); and 3) techniques such as spectral mixture analysis of the SWIR region (e. g.,

Gamon et a1. 1993). The SWIR region has been used because it contains cellulose (2090

nm and 2270 nm) and lignin (2130 nm and 2270 nm) absorption features that are masked
by water in green vegetation but become evident as vegetation senesces (Roberts et al.
1993). While each of these techniques has been used successfully in grassland
ecosystems to produce estimates of fractional senescent vegetation cover, [or the areal
proportion of the surface covered by senescent vegetation (White et al. 2000)], none can
be directly used to quantify biomass, either total (green plus senescent) or senescent. For
range management operations, fractional cover is not always in itself an adequate
indicator of rangeland condition because it is not a direct surrogate of productivity as
stand biomass is (Pickup et al. 1994).

To increase the utility of remote sensing data for biomass management in
rangeland ecosystems it is therefore imperative that we more clearly deﬁne the limits of
green vegetation indices (like NDVI) throughout the season as well as develop new
remote sensing approaches that can be used when senescent biomass is present. The
studies in this dissertation were motivated by the lack of detailed information about the
effect of senescence on the NDVI-biomass relationship for grass species and grass mixes
that dominate western grassland ecosystems. This detailed information is important
because it enables analysis of the error associated with the use of NDVI for biomass
estimates during time periods when senescent biomass is present.

The objectives of my dissertation were: 1) to identify limitations associated with
the use of remote sensing data by private range managers and to evaluate the impact of
the use of remote sensing data on manager decision-making, 2) to increase the accuracy

of NDVI-green biomass estimates in grassland ecosystems, and 3) to develop a new

10

remote sensing approach for the quantiﬁcation of biomass that can be used in grassland

ecosystems when senescent biomass is present.

Organization of the Dissertation

For my thesis research I examined three fundamental questions regarding the use
of remote sensing data by private range managers in California. First, what factors
inﬂuence the use of remote sensing data by the private range management community
and how does the use of remote sensing impact decision-making? Second, during what
part of the growing season can a single NDVI-biomass equation (i.e., biomass =
ﬂNDVI» be accurately used in annual grasslands with mixed species composition? And,
third, are there alternative means for quantifying senescent biomass from satellite data
that are accessible and affordable for private range managers?

In Chapter 2, Experimental use of remote sensing by private range managers and
its inﬂuence on management decisions (Butterﬁeld and Malmstrom 2006), I used case
study analyses to examine how the characteristics of range managers and their properties
inﬂuence the use of remote sensing data. I found that remote sensing data were most
likely to be used and invested in by range managers who had recently implemented other
new practices and who believed remote sensing products would help increase ranch
proﬁtability. In these cases, managers found that remote sensing data allowed them to
extend their intensive management efforts to a greater proportion of their properties and
to base their management decisions on multi-year forage and weed analyses.

In Chapter 3, Phenological eﬂects on remotely-sensed biomass estimates in

annual grasslands, I used annual grass stands planted in an agricultural ﬁeld on the

11

campus of Michigan State University in East Lansing, Michigan to examine in detail the
impact of canopy senescence on NDVI-biomass estimates. I used these data to determine
the phenological period during which a single NDVI-biomass equation could be used in
annual grasslands with mixed species composition. I found that in all stand types, NDVI
could be used to estimate green biomass throughout the season, regardless of the
proportion of senescent biomass present. Furthermore, all stand types displayed similar
phenological relationships between NDVI and biomass, reaching maximum NDVI and
maximum biomass simultaneously. Last, when the entire season was considered, the
species-speciﬁc NDVI-biomass equations were remarkably similar, suggesting that a
single equation may be robust to the structural and phenological differences that exist
among common grass species.

In Chapter 4, Remote sensing-based estimates of senescent biomass: Common
problems and a new approach in annual grasslands, I propose a new vegetation index for
senescent biomass estimates, MAPAR (the mean PAR absorbed by the surface), and
examine its potential for landscape-scale RDM estimates in western grassland
ecosystems. Across dry, light-colored sandy loam soils, MAPAR was correlated with
senescent biomass across a wide-range of conditions. However, the utility of MAPAR
declined where the soil was darker, either due to increased soil moisture or organic matter
content, and also when signiﬁcant stem lodging occurred. In most western grassland
ecosystems, such as those found in California, soil is often low in organic material (e.g.,
2—10 %). In addition, across grazed grasslands, stems often do not reach heights where

stem lodging would be a signiﬁcant issue. Therefore, these results suggest that MAPAR

12

may be a valuable approach for estimating senescent biomass, or RDM, in grazed
grassland ecosystems.
I conclude the theses with a summary of future research directions that were

inspired by my dissertation research in Chapter 5.

13

CHAPTER 2

EXPERIMENTAL USE OF REMOTE SENSING BY PRIVATE RANGE MANAGERS
AND ITS INFLUENCE ON MANAGEMENT DECISIONS

Abstract

Although remote sensing has many potential applications for range management,
its use by range managers has thus far been limited. To investigate the factors that
encourage use of remote sensing and to examine its inﬂuence on decision making by
individuals who manage privately owned rangeland, we evaluated the decision-making
processes of three ranch owners and one professional ranch manager who were
introduced to remote sensing while collaborating with us in a rangeland stewardship
program in California. Two of the participants had extensive ranching experience (11 to
> 20 years) and managed large cattle ranches (1000 to > 2000 ha) and two had less
experience and managed smaller sheep ranches (< 200 ha). During the ﬁve-year
program, the participants implemented a series of new management practices, including
prescribed burning, rotational grazing, and seeding of native grasses, with the aim of
reducing noxious weeds and increasing productivity. We used remote sensing to quantify
the effect of these practices and provided ranch-wide remote sensing analyses to each
manager on a password-protected website. Using case study methodologies, we found
that managers of larger, commercially active ranches found the experimental use of
remote sensing to be a highly positive experience that convinced them that this
technology could help address difﬁcult management situations and increase ranch
proﬁtability. This suggests that the broad use of remote sensing by managers of
privately-held, commercial rangelands may be limited in part by the simple lack of

opportunity to test these technologies. Programs that assist ranchers in obtaining

14

appropriate remote sensing products may thus be a cost-effective way to enhance
conservation on private rangelands. Our ﬁndings suggest that voluntary self-analysis by
ranchers of the landscape dynamics of their own properties is likely to lead to more

engaged conservation efforts than top-down prescriptions.

Introduction

Few range managers currently use remote sensing products to inform their
management decisions (Daberkow and McBride 2000, Hunt et al. 2003, Washington-
Allen 2006), even though remote sensing offers valuable means of assessing the
inﬂuence of management practices on forage production (e.g., Pickup et a1. 1994) and
invasive noxious weed spread (e.g., Lass et a1. 1996) across large range units. Several
studies have examined factors inﬂuencing the use of innovative technologies in general
(F liegel 1993, Rogers 1995, Roling and Wagemakers 1998, Daberkow and McBride
2003) and of speciﬁc range management technologies, such as cattle vaccines (Harris et
al. 1995) and prescribed burning (Kreuter et al. 2001). To our knowledge, however, no
study has sought to identify factors that promote the use of remote sensing technologies
by range managers or investigated how the use of remote sensing can inﬂuence manager
decision-making. With other innovative technology, low use rates have often been found
to indicate either that the technology has not been successfully introduced to the end user
(in this case, the range manager) or that the end user does not see its utility (Fliegel 1993,
Kreuter et al. 2001, Daberkow and McBride 2003). In the case of remote sensing, use

may also be limited by its cost and complexity or by the lack of opportunity to try it.

15

California’s rangelands are a good example of a system in which broad use of
remote sensing technologies could beneﬁt range managers by allowing them to assess
management techniques for weed control and forage improvement over large areas.
Since ﬁrst settled by European immigrants, California’s rangelands have been under
pressure from human activities, which have resulted in the conversion of this system from
one dominated by native vegetation including annual forbs and perennial bunchgrasses to
one dominated by introduced annual grasses (Heady 1977, D'Antonio and Vitousek
1992). These introduced annual grasses, which have relatively high forage value, have
supported an extensive ranching economy in the state for more than 150 years (George
and Fulgham 1989). Today, however, a wave of introduced noxious species, including
yellow starthistle (Centaurea solstitialis L.; all nomenclature follows Hickman (1993)),
medusahead (T aeniatherum caput-medusae (L.) Nevski), and barbed goatgrass (Aegilops
triuncialis L.), are spreading through the region, reducing rangeland productivity and
threatening the economic sustainability of established ranches (Maddox and Mayﬁeld
1985, Young 1992, Peters et al. 1996). Unlike the previously established exotic species,
these new invaders provide poor forage during most of the season (Bovey et al. 1961,
Lusk et al. 1961, Callihan et al. 1982, Callihan et al. 1995, Peters et al. 1996). To
manage the noxious weeds, range managers are testing a variety of new management
approaches, but it can be costly for them to assess the consequences of the new
approaches with on-the-ground surveys alone, given the extent of their properties.
Remote sensing offers an opportunity for managers to evaluate large areas more quickly

and cost-effectively.

16

To identify factors that promote the use of remote sensing by range managers and
to investigate the inﬂuence of remote sensing on range management decisions, we used
an in-depth, case-study approach to examine the experimental use of remote sensing
products by four individuals who manage private rangelands in the Western Sacramento
Valley foothills in California, as part of a ﬁve-year rangeland stewardship program
(1999-2004) (Malmstrom et al. 2004). Managers were involved in the program primarily
because they wanted to increase the productivity of their land and decrease noxious weed
levels (Table 2.1). To do this, they tested a series of new management practices (Table
2.2). None of these managers had speciﬁcally used remote sensing data to make
management decisions on his or her property prior to involvement in our study.

Here, we examine l) the ways in which managers’ ranching approaches and
previous ranching experiences inﬂuenced their interest in and use of our experimental
remote sensing products, and 2) the ways in which the remote sensing products in turn
inﬂuenced the managers’ decision-making. We produced a broad suite of remote sensing
products, which included a time series of spring forage estimates for the watershed, and a
map of noxious weed distributions. This information was presented to the managers
through an interactive website that allowed each to view his or her property as a whole or
on a ﬁeld-by-ﬁeld basis. Data were presented in graphs and as maps, which could be
selected to show estimated values for a given time period or patterns of change across
years. We conducted surveys and interviews with all four managers before and after they
worked with these remote-sensing materials and then used case study methodologies to

analyze their responses.

17

Table 2.1. Motivation for involvement in the stewardship program. Responses were
given in survey 1, before managers were provided access to the remote sensing products,

on a Likert scale (1 = no motivation to 5 = high motivation).

 

Cattle] Cattle2 SheeLl Sheep2

 

Increasing forage production 5 5 5 1
Decreasing noxious weeds 4 3 5 3
Establishing native bunchgrasses 5 l 2 1
Promoting landowner outreach 4 2 4 5
Increasing water quality 5 2 1 1

 

18

Table 2.2. Management practices tested during the stewardship program.

 

Cattle 1

 

1 unit totaling 12.1 ha

2 units totaling 157.8 ha
6 units totaling 157.8 ha
1 unit totalirglOl 1.7 ha

Prescribed burning, seeding of native bunchgrasses,
rotational grazing

Prescribed burning, rotational grazing

Rotational grazing

Fencing planned

 

Cattle 2

 

1 unit totaling 16.2 ha

Rotational grazing (spring only)

 

6 units totaling 550.4 ha Rotational grazing (fall only)
1 unit totalinLZ6.3 ha Prescribed burning, seeding of native bunchgrasses
Sheep 1

 

1 unit totaling 24.3 ha
3 units totaling 97.1 ha

2 units totaling 46.5 ha

Rotational grazing, prescribed burning, seeding of clover
Prescribed burning, seeding of native bunchgrasses,
rotational grazing

Prescribed burning, rotational grazing

 

Sheep 2

 

1 unit totaling 16.2 ha

1 unit totaling 16.2 ha

Prescribed burning, seeding of native bunchgrasses,
rotational grazing
Rotational grazing

 

l9

Methods
Growing season terminology

Most California rangelands experience a Mediterranean climate, distinguished by
a moderate fall-winter-spring growing season and a prolonged summer drought. We thus
consider time in terms of “biological” years that begin in fall at the end of the summer
drought (i.e., September 200N) and continue into August of the following calendar year
(200N + 1). We refer to September—November as fall, December—February as winter,
March—May as spring, and JunwAugust as summer. For most annual range grasses, the
growing season begins in fall with the ﬁrst rains, continues through the wet winter, and
reaches its peak in spring. By late May, most annual range grasses are senesced, but the

newer noxious weeds may remain green later into the summer.

Remote sensing products

To produce maps of green spring forage values, we used ﬁeld-calibrated
algorithms to estimate green forage biomass from NDVI (normalized difference
vegetation index) values derived from Landsat satellite imagery acquired in late March or
early April of each year from 1999 to 2004 (Malmstrom et al. 2004). We used Landsat
imagery both because of its availability and cost effectiveness to private range managers
as well as the appropriateness of its spatial scale (30 m) for rangeland forage analyses.
Forage estimates were made only at the peak of the growing season when green biomass
was dominant, because we found that remote sensing algorithms for quantifying
senescent biomass previously developed for southwestern systems (Qi et al. 2000) failed

to perform adequately in the California annual grasslands (Malmstrom et al. 2004). To

20

map the distribution of two dominant noxious weeds-medusahead and goatgrass—we used
a time series of ﬁne spatial resolution (1 ft) aerial photography acquired at key
phenological time points when these weeds showed reﬂectance patterns distinct from the
more valuable forage grasses (Malmstrom et al. 2004). To provide managers interactive
access to the remote sensing products, we built a password-protected website on which
each manager had access to information about only his or her own property. The website
offered managers whole-property and ﬁeld-by-ﬁeld access to spring forage maps from
1999—2004, forage change maps comparing differences across years, a quantitative 5-
year forage analysis, a map of noxious weed distribution in 2004, and land use history

and weather data.

Surveys and interviews of land managers

To gather baseline information about the managers’ experiences, practices, and
attitudes, we asked each manager to complete a survey in March 2004, prior to
introducing him or her to the remote sensing materials. This survey contained 25
multiple-choice Likert scale questions (Miles and Huberman 1994, Strauss and Corbin
1998) and ten short-answer ones. Among the subjects we explored were the managers’
1) assessment of the current range conditions at their property and the need for new
management approaches, and 2) previous experiences with using new management
practices such as prescribed burning for weed control. We interviewed each manager
individually to clarify his or her responses and to gather additional information about

management approaches used on his or her property and its land use history.

21

We then showed all four managers how to use the project website to access the
remote sensing products, and gave them ﬁve months (March—August 2004) in which to
explore the products and test their utility for management decision-making. In August,
we re-surveyed and interviewed the managers to determine to what extent they had used
the remote sensing products, how the products had inﬂuenced their decision-making, and
what improvements might increase their interest in using remote sensing in the future.
Like the ﬁrst survey, the second one contained 25 multiple-choice Likert scale and ten
short-answer questions, about half of which were identical to those asked in the ﬁrst
survey. Both surveys adhered to Dillman’s (1978) guidelines, except that we did not use

follow-up mailings because our group was small enough to contact by phone.

Analysis of responses

Because our study group was by necessity small, we used case study
methodologies to analyze manager responses (Yin 2003). Case studies provide the
opportunity to intensively examine the experiences and responses of a smaller group of
managers, who represent elements of a larger management community. While case
studies provide substantial insight into manager motivation, care must be taken when
extrapolating results from individual case studies to other situations. To construct the
case studies, we ﬁrst coded the survey and interview data to identify manager responses
in two broad categories of interest (Miles and Huberman 1994, Strauss and Corbin 1998):
1) Manager characteristics and experience and 2) Inﬂuence of remote sensing on
management decisions. Manager characteristics and experience included descriptive

information about each manager and his or her management operations; Inﬂuence of

22

remote sensing on management decisions included information about how each manager
used remote sensing data to evaluate his or her success in meeting his or her own
management objectives. During the coding process we sought illuminating quotations
from each manager that provided insight into his or her own unique story.

We used the managers’ responses from the second set of surveys and interviews
to determine their response to the remote sensing products. We deﬁned the use of remote
sensing during the study as the extent to which managers accessed the website and our
remote sensing products with the purpose of using the products to evaluate the impact of
their management efforts and to decide which approaches to use in the future. We also
quantiﬁed the managers’ self-reported planned intent to use these same remote sensing
products for management evaluations in the future and to invest in new remote sensing

data and products for their properties.

Results

We present each case study individually, with two sections within each: 1)
Manager characteristics and experience, which discusses the characteristics of each
manager and his or her management operations and motivations for involvement in the
stewardship program; and 2) Inﬂuence of remote sensing on management decisions,
which discusses how each manager used the experimental remote-sensing products to
evaluate his or her management strategies, and his or her suggestions about how the
products could be improved to enhance their value for private range managers such as

themselves or for long-term management planning. To protect their privacy, the

23

managers and case studies are referred to by code names: Cattle 1, Cattle 2, Sheep 1, and

Sheep 2.

Cattle 1

Manager characteristics and experience. At approximately 2000 ha, Cattle l was
the largest ranch studied. Cattle 1 has been family-owned for over 20 years and has been
used primarily for commercial cattle grazing during that time. The landowners’
motivations for involvement in the stewardship program were diverse but centered on the
desire to increase ranch productivity (Table 2.1). Both the current and former managers
(son and father) agreed that it was important to increase the value of their land, especially
given pressure from urbanization, governmental regulations, and global competition, so
that their family business “makes it to the next generation.” The father believed “grazing
alone maintained feed for cattle” and that “burning and seeding native perennials was too
expensive” to be a property-wide solution. However, his son believed that to “preserve
the ranching way of life in California” it was necessary to implement management
strategies that took advantage of new technologies and cooperative partnerships. His
outlook was evident in the diversity of management practices implemented during the
program (Table 2.2). Rotational grazing, the practice of alternating periods of grazing
and rest among two or more fenced pastures throughout the season, was an especially
distinctive change that occurred in the last 1—2 years of the program because the ranch .
had previously used set stocking rates. Cattle 1’s manager believed that “decreasing
stocking rates in the spring would allow more re-growth and, combined with burning in

the summer, would have a positive effect on forage.”

24

Inﬂuence of remote sensing on management decisions. Before working with the
remote sensing products, Cattle 1’s manager believed that there was a “place for remote
sensing in ranching,” but emphasized that it could never replace the skills of a rancher.
He believed that remote sensing would allow him to quickly visualize forage and weed
levels across his entire property, which would allow him to compare ﬁelds enrolled in the
program with those being managed with a traditional grazing approach.

After evaluating the remote sensing products on the project website, Cattle 1’s
manager concluded that he could use the weed map and the time series of forage
estimates to determine whether the programs’ restoration efforts had made impacts on
weed control and forage production that were large enough to justify their continued use.
During the second interview, for example, he discussed how his analysis of the remote
sensing maps had led him to conclude that “while burning increased forage levels the
year following the burn, two years later medusa was back and forage levels were where
they were before the burn.” Therefore, “burning is only beneﬁcial if used along with
seeding of good competitors, such as native perennial grasses.” He indicated that his
evaluation had also led him to conclude that his rotational grazing efforts had increased
forage and decreased weeds during the 2003—04 growing season. This conclusion was
signiﬁcant in a management context because “managing with cattle requires fewer
additional inputs of time and money compared to burning and seeding.”

Based on the value he gained from the remote sensing products we produced for
the lower portion of his ranch, Cattle 1’s manager requested that we also produce similar

products for the ranch’s upper portion. He found this upper 1000-ha portion to be more

25

difﬁcult to manage because of its ruggedness, size, and lack of fencing, and so had not
yet tried restoration efforts in it (Table 2.2). The additional forage maps led Cattle 1’s
manager to conclude that forage increases in the closely managed lower portions of the
property had been much greater than those in the upper portions. He believed that these
differences were due to his inability to “control cattle and visually inspect” the upper
portions of the property “throughout the season.” He believed that remote sensing data
would allow him to address both issues, and he planned to use the forage maps to
coordinate fencing of the upper portion in 2004—05 and to monitor forage thereafter
(Table 2.2). His ﬁrst-hand experience of remote sensing’s capacity to quantify
management effects and help enhance management efforts in remote terrain led Cattle 1’s
manager to increase the amount of money that he would be willing to invest annually in
these technologies from approximately $100 (survey 1) to $500 (survey 2), with the latter
value exceeding the cost of an entire Landsat Thematic Mapper (TM) scene ($425).

The outreach efforts at Cattle 1 involved a two-way exchange of information
between scientists and range managers. This exchange allowed us to produce remote
sensing products customized for the experience level and management needs of Cattle 1’s
manager. Cattle 1’s manager emphasized that the remote sensing did not change what
management strategies he believed were possible, but rather the means by which he could
assess their effectiveness: “This is what I meant when I said remote sensing was only a
tool. It helps me see the effects of management, but it cannot do them for me.” Voicing
an important common theme among the case study group, he indicated that he believed

that the increasing challenges of ranching were making cooperation between ranchers and

26

scientists even more important. This program conﬁrmed to him that such partnerships

“improve the chances of rancher survival.”

Cattle 2

Manager characteristics and experience. At approximately 1200 ha, Cattle 2 was
the second largest ranch in our sample. Cattle 2 is family-owned, commercially grazed
by cattle, and managed by a professional range manager with over 20 years of ranching
experience. Cattle 2’s manager participated in the program to increase ranch productivity
(Table 2.1). He felt rotational grazing was the best way to accomplish this because it

“did not require an additional investment of time and money.”

Inﬂuence of remote sensing on management decisions. Prior to evaluating the
project’s remote sensing products, Cattle 2’s manager was receptive to incorporating
remote sensing data into his management regime because he believed that there was a
“pressing need to increase the productivity and proﬁtability” of his ranch and that these
tools would allow him to “directly meet these needs.” Cattle 2’s manager felt remote
sensing would be useful for analyzing forage and weed levels before and after grazing
events. He believed that if monthly forage maps were available, he could make grazing
adjustments during the season, which would allow him to maximize the time his cattle
spent grazing while maintaining adequate forage for the following season. In addition, he
thought that using the forage and weed maps together would allow him to determine

whether there was “good feed in a ﬁeld or just medusa.”

27

After viewing the 2004 weed map, Cattle 2’s manager was surprised to see high
weed levels across select grazing ﬁelds. Even though he knew noxious weeds were a
signiﬁcant threat across his property, he believed that he had limited their impact with
grazing alone. He concluded that weed increases were due not to the ineffectiveness of
his grazing efforts, but rather to seasonal water limitations that restricted grazing to the
spring when the weeds were maturing and thus unable to be grazed effectively by cattle.
During the interview, he used the forage and weed maps together to show us that ﬁelds
grazed only in the fall had both increased forage (Table 2.2) and decreased weed levels.

Even though Cattle 2’s manager focused on grazing as a tool to manage his
property, the size of the ranch prevented him from tracking these effects “across every
ﬁeld at the same time.” After analyzing the time-series forage maps, he was encouraged
by the capacity of remote sensing to allow him to track forage levels across different
ﬁelds simultaneously without extensive ﬁeld work and to easily compare values from
growing seasons. Like Cattle 1’s manager, Cattle 2’s manager also believed that remote
sensing data would allow him to more intensively manage the upper portions of his
property, which would increase ranch productivity. Because of this potential, Cattle 2’s
manager increased the amount of money that he planned to invest annually in these
technologies from approximately $0 (survey 1) to $500 (survey 2).

The collaborative nature of the landscape analysis effort motivated the managers
of Cattle 1 and 2 to share their forage and weed species maps with one another, which
allowed Cattle 2’s manager to see the beneﬁts of practices he did not implement during
the program, such as the multi-year effect of burning and seeding with perennial grasses

(Table 2.2). While the results did not substantially change his mind about which

28

practices were most successful, they did convince him to try burning and seeding a 16-ha
ﬁeld heavily infested with medusahead and under seasonal grazing restrictions (Table

2.2), beginning in 2005.

Sheep 1

Manager characteristics and experience. Sheep 1 was the largest sheep ranch in
our sample, but at approximately 170 ha it was signiﬁcantly smaller than both cattle
ranches. The ranch was purchased by the current landowner in 1998 and subsequently
grazed by about 200 ewes. This manager initially had little commercial ranching
experience but is highly educated and made considerable efforts to increase her
knowledge of ranching during the program by enlisting the help of range managers and
scientists. In addition, because ranching is not her primary occupation, she has employed
a professional range manager to assist her. In this study, we interacted solely with the
landowner-manager herself. Her motivations for involvement in the program were
diverse, but she emphasized increasing ranch productivity and decreasing noxious weed
levels (Table 2.1).

When purchased in 1998, Sheep 1 was dominated by medusahead and yellow
starthistle. To restore it, half of the property was burned in the summer of 1999 and then
seeded with native perennial bunchgrasses in the fall of 2000. Unfortunately, bunchgrass
populations did not immediately take hold. During the 2000—01 growing season, the
landowner was thus forced to conﬁne her sheep to the unburned portion of her property,

which led to overgrazing. More ﬁelds were burned in 2000—03, with varying degrees of

29

success; the most success occurred in a 24-ha ﬁeld where the landowner used intensive

short—duration gazing followed by burning and seeding with clover (Table 2.2).

Inﬂuence of remote sensing on management decisions. Even before evaluating
the remote sensing products, Sheep 1’s manager felt that forage and weed maps would
provide her with an essential overview of her ranch and a means to “determine which
management practices worked and which ﬁelds needed to be attacked next.” Because
almost every hectare of Sheep 1 was involved in a restoration test, Sheep 1’s manager
had the unique opportunity to see the short- and long—term effects of a variety of
management efforts in combination with rotational gazing practices (Table 2.2).

Before analyzing the remote sensing products, Sheep 1’s manager did not believe
prescribed burning was a long-term solution for her property because of its “cost, danger,
and varied results.” After the unsuccessful restoration efforts of 1999—2000, Sheep 1’s
landowner expected this portion of her property to have low forage levels and large areas
of medusahead and goatgass. The remote sensing products supported these beliefs, and
contributed to her conclusion that prescribed burns alone were not effective enough to
offset their high cost. In addition, the time-series forage and weed maps led Sheep 1’s
manager to decide that rotational gazing in combination with other practices, such as
sowing good weed competitors, was the most effective strategy for increasing forage and
decreasing weeds on her property. Because remote sensing data allowed Sheep 1’s
manager to base her management efforts on multi-year forage trends, she increased the
amount of money that she planned to invest annually in these technologies from

approximately $100 (survey 1) to $500 (survey 2).

30

Sheep 1’s manager believed that the outreach efforts were an important part of the
stewardship progam. She felt these efforts were a model for how progams should be
carried out in the future and that “the in-person visits were essential components of
establishing trust and sharing knowledge.” Like other participants, she emphasized the
value of collaborative research, and she indicated that she believed that “California farm
land was being swallowed up by developers and that projects like this should serve as

models for how scientists and landowners can work together.”

Sheep 2

Manager characteristics and experience. At approximately 30 ha, Sheep 2 was
the smallest ranch in this study. The managers have owned the ranch for approximately
20 years but have never commercially managed the forage on their property. Sheep 2’s
managers were involved in the stewardship progam to increase collaborative interactions
with scientists and other participating landowners and to help conserve the gassland
habitat on their property (Table 2.1). Sheep 2’s managers did not manage the forage or
weed levels on their property during the progam, but they did allow project scientists

and other property owners to conduct restoration tests on their property (Table 2.2).

Inﬂuence of remote sensing on management decisions. After accessing the
website, Sheep 2’s managers responded that while the website “looked geat,” they had
“no need for weed maps or analyses of management practices.” They emphasized that if
their property were larger the remote sensing products would be helpful, but because it

was so small they could “walk across the property if they needed to see what was going

31

on.” Although Sheep 2’s managers did not believe that the website was useful on their
property, they did see the beneﬁt of time-series forage and weed maps, commenting that
“seeing how the land has changed allows us to see whether we are part of the problem or

part of the solution.”

Discussion
Factors inﬂuencing the use of remote sensing by managers

In this experimental test of the value of remote sensing for private ranch
managers, we produced and offered, at no cost to the managers, a suite of remote sensing
products tailored to the managers’ needs and worked individually with each manager to
ensure that any questions or concerns s/he had about how to access the products on the
project website could be addressed promptly. In doing so, we thereby removed or
reduced several barriers that might otherwise prevent private managers from
experimenting with remote sensing as a management tool, such as its cost and potential
uncertainty on the managers’ part about how to get appropriate products for their
properties. We then examined, among our case study goup, what other factors came into
play in determining the degee to which each manager was willing to “use” the remote
sensing products we provided to analyze them and draw conclusions from them about the
success of their management strategies. In addition, we evaluated the inﬂuence of the
experiment on the managers’ self-reported willingness to purchase remote sensing
products in the future.

In general, it has been found that several criteria need to be met for users to begin

to use and invest in new technology. Rogers (1995) and Somers (1998) concluded, for

32

example, that end users must ﬁrst believe that there is a relative advantage (e. g.,
ﬁnancial) to using the new technologies, and then have the opportunity to determine how
best to incorporate these technologies into their current practices. In our study, the three
managers whose properties are used for commercial livestock production (Cattle l, Cattle
2, Sheep 1) spent the most time analyzing and evaluating the remote sensing data and
were most interested in purchasing remote sensing products in the future. These three
managers concluded that remote sensing provided tools that could help maximize their
properties’ productivity and that they would like to cooperatively purchase additional
remote sensing products in the future. After experimenting with the remote sensing
products, all three managers of commercially active properties increased the amount of
money they indicated they would be willing to spend on remote sensing data.
Individually, their planned annual investments of approximately $500 would each be
enough to purchase an entire single Landsat TM scene (a 170 x 183 km scene costs
$425). If pooled, their planned annual investments of approximately $1500 would be
large enough to purchase multiple TM scenes and additional GIS data for their properties,
or additional aerial photogaphs for noxious weed mapping. While our expectations of
use and investment did not initially assume that cost-sharing would be necessary, from
this particular study, we would conclude that at least in the early stages of use
collaborations between managers increase the likelihood of investment by decreasing the
annual ﬁnancial obligation to any one manager. Unlike the other managers, the managers
of the smaller, commercially inactive property (Sheep 2), spent less time evaluating the

remote sensing data and were uninterested in purchasing products in the future.

33

Consistent with Rogers’ and Somers’ ﬁndings, it was evident that the managers
who used the remote sensing products most extensively during our study and who were
interested in purchasing products in the future were those who believed that the remote
sensing technologies could offer them management advantages. The managers of the
three commercially active ranches believed that their current management efforts were
not optimized and that there was thus a need for new management approaches on their
properties. These managers had participated in the stewardship pro gam in order to
increase forage production and decrease invasive noxious weed levels across their
properties (Table 2.1), and they were willing to test new management approaches to meet
these goals (Table 2.2). In other situations, Hanselka et al. (1990) and Kreuter et al.
(2001) likewise found that managers were more likely to use and invest in new range
management technologies if range conditions across their properties were poor. In
contrast, the managers of Sheep 2, while impressed by the remote sensing products, did
not see a need to test new management approaches and thus were less interested in
exploring the utility of remote sensing.

The managers most interested in using the remote sensing products not only felt
that there was a need to try new management approaches but also believed that using
remote sensing could effectively help them do it. Likewise, Kreuter et al. (2001)
concluded that “Brush Busters” management approaches were broadly used and invested
in across Texas rangelands because of their perceived effectiveness in decreasing brush
and increasing productivity. During our study, the managers of Cattle l, Cattle 2, and
Sheep 1 each had at least one experience that convinced them that they could increase the

productivity and proﬁtability of their property by using these particular remote sensing

34

products to inform their management decisions. For example, the managers of Cattle 1
and 2 concluded that forage and weed species maps provided them with the opportunity
to monitor their rotational gazing efforts more intensively during the season, which
allowed them to maximize the time cattle spent gazing while making sure adequate
forage was left for the following season. In addition, they were convinced that remote
sensing approaches would allow them to extend their rotational gazing efforts to the
upper portions of their properties, where increases in forage had been much smaller than
in their more intensively managed lower portions. Sheep 1’s manager had recently
implemented a series of strategies aimed at increasing forage production and decreasing
weed levels (Tables 2.1 and 2.2). She was able to use the multi-year forage analyses to
determine the effectiveness of these efforts and to develop a comprehensive adaptive
management strategy for her property. In contrast, the managers of Sheep 2 did not see
the need to explore whether remote sensing could help increase ranch productivity or

proﬁtability because their ranch was not commercially active.

Inﬂuence of remote sensing on decision-making

Range managers are skilled in reading and assessing landscapes and maps, and
our case study goup readily transferred these skills to interpreting remote sensing data
and incorporating it in to their management analyses. Among our manager goup, those
managing commercially active properties believed that remote sensing data allowed them
to base their decision-making process on multi-year forage trends across their entire
properties, rather than on one-year forage changes across individual pastures. The ability

to view multi-year forage trends allowed those managers who were actively managing

35

their land to quantitatively assess the forage impact of new management practices they
tested during the stewardship progam, and to determine whether these practices were
short-term ﬁxes or long-term solutions to problems such as increasing noxious weed
levels. For example, the multi-year analyses enabled the managers of both Cattle 1 and
Sheep 1 to conclude that they would only invest in prescribed burning in the future if it
was done in coordination with seeding of good competitive forage gasses like native
bunchgasses or clovers; otherwise, the positive impact of ﬁre on noxious weeds is too
temporary (1—2 years) for commercial range management operations to justify its
expense and potential hazard.

The remote sensing products used in this study were developed with the input of
our manager goup. Mutual discussion of the remote sensing products during the
progam helped us tailor the website and forage maps to the needs and experiences of
each manager, and also provided a forum in which to elicit information from managers
about historical land use and past management strategies. Historical land use data is an
invaluable resource for managers seeking to assess the long-term inﬂuence of
management strategies and other factors such as invasive species and climate change.
However, on many private ranches, including well managed ones, data on stocking rates
and other land use information are often not kept in a detailed or consistent manner.
Through involvement in this study, our manager goup was able to determine for
themselves the value of coordinating remote sensing analyses with on-the-gound
management data, and as a result, expressed increased commitment to keeping more

detailed management (e.g., gazing) records for decision-making in the future.

36

An important theme expressed throughout by the manager case study goup was
the importance of collaboration, among ranchers and between ranchers and scientists, in
ﬁnding ways to optimize rangeland management in regions facing pressure from forces
such as urbanization and invasive species. This sentiment contrasts with historical
expressions of enmity between conservation biologists and Western ranchers (e.g., Jensen
2001) and indicates the overwhelming need for innovation to protect remaining
rangelands as conservation and cultural resources (e. g., Weiss 1999). Our ﬁndings
suggest that voluntary engagement in collaborative rangeland analyses not only can
increase the success of stewardship progams like this one, but is also more likely to be
effective in supporting long-term efforts to improve rangeland conditions than top-down
prescriptions. Our work demonstrates that such collaborations can foster the
development and application of innovative management technologies and thereby

facilitate efforts to enhance rangeland sustainability.

Management Implications

Our ﬁndings suggest that one hurdle impeding the broad use of remote sensing by
managers of privately-held, commercial rangelands may simply be the lack of
opportunity to test it. When given this opportunity, all of the commercially active
managers in our case study goup responded very positively and found creative ways to
effectively use it to evaluate their management efforts. These managers chose to use our
remote sensing products during the study and indicated they intended to use them and

buy more products in the future because the experimental tests convinced them that such

37

remote sensing products would help optimize their management practices and increase
ranch productivity and proﬁtability.

Although the cost of some remote sensing products can be high, relatively low-
cost data have been traditionally available to the public through well established
progams such as the Landsat data acquisition progam. When such imagery are
available, use of remote sensing by private range managers can be economically feasible,
particularly if consortia of managers with properties falling within the same satellite
scenes can collaboratively cost-share and obtain technical support from local universities
or agencies. We hOpe that these ﬁndings will encourage more private range managers
and scientists to collaborate on efforts to incorporate remote sensing into commercial
range management and rangeland restoration efforts. Continued support of regular image
acquisition by reliable, appropriate-scale satellite systems with public data availability is

essential to this aim.

38

CHAPTER 3
PHENOLOGICAL EFFECTS ON REMOTELY—SENSED BIOMASS ESTIMATES IN
ANNUAL GRASSLANDS
Abstract
Remote sensing data can provide range managers the means to more efﬁciently
quantify the effects of their management efforts on biomass production across large range
units. Vegetation indices such as the normalized difference vegetation index (N DVI)
have been used for more than 30 years in rangelands to quantify geen biomass. In many
gass-dominated rangeland (gassland) ecosystems, however, there is a sigriﬁcant dry
period, during which time senescent biomass is the dominant forage resource for
livestock. During these periods, indices like NDVI underestimate total biomass and are
thus largely unhelpﬁJl as management tools. Both at the beginning and end of the
gowing season, forage can consist of a mix of senescent and geen material. Use of
NDVI to estimate total biomass during these periods must be carefully evaluated. Even
though the general effect of senescence on the NDVI-biomass relationship is well
understood, no study has speciﬁcally characterized this effect in detail for gass species
that are representative components of gassland systems globally. To examine the impact
of senescence on NDVI-biomass estimates and to determine the phenological period
during which a single NDVI-biomass equation (i.e., biomass =f(NDVI)) is useﬁJl for
estimating biomass in mixed species stands, I gew annual gass species in monoculture
and in mixtures in a common garden on the campus of Michigan State University in East
Lansing, Michigan, and measured a suite of stand canopy parameters weekly from

germination to the end of the season. In all stands tested, there was an approximately 40-

39

day lag between maximum NDVI and maximum biomass. During this period, there was
a simultaneous decrease in NDVI with increasing senescent vegetation cover. The result
was that NDVI values at maximum biomass in week 9 were similar to those in week 1,
even though biomass increased, on average, from 27.0 g/m2 in week 1 to 372.0 g/m2 in
week 9. When the entire season was considered the species-speciﬁc NDVI-biomass
equations were remarkably similar, suggesting that a single equation may be robust to
many of the structural and phenological differences that exist among gass species and
therefore can be used to estimate biomass in gassland ecosystems with mixed species

composition.

Introduction

Currently there are many remote sensing tools available that range managers
could use to help manage biomass on their properties (e. g., Tueller 1989, Hunt et al.
2003, Washington-Allen et al. 2006, Butterﬁeld and Malmstrom 2006). Range managers
are particularly interested in the potential for using remote sensing data to quantify
month-to-month variability in herbaceous biomass, the stem and foliage biomass of
gasses and forbs that is the primary forage resource for livestock (Malmstrom et al.
2004, Butterﬁeld and Malmstrom 2006). Remotely-sensed images with ﬁne spatial
resolution (e. g., 30 m) can be used to assay gassland conditions at the pasture level
(Wylie et a1. 2002, Malmstrom et al. 2004, Mustafa et al. 2005, Washington-Allen et al.
2006). In the western United States, range managers are experimenting with using

satellite-based geen biomass estimates to evaluate the effectiveness of their management

40

approaches, as a way to enhance ranch proﬁtability (Qi et al. 2000, Qi et al. 2002,
Malmstrom et al. 2004, Butterﬁeld and Malmstrom 2006).

Now that geen biomass estimates have proven valuable, managers are interested
in using similar remote sensing products to evaluate management effects during periods
when senescent biomass is present (Qi et a1. 2000, Qi et al. 2002, Butterﬁeld and
Malmstrom 2006). Senescent biomass is an important forage resource for livestock in
gass-dominated rangeland (gassland) ecosystems during dry periods (George and
Fulgham 1989, Richardson and Everitt 1992, Frank and Aase 1994, Maselli et al. 1998,
Qi et al. 2000, Hoare and Frost 2004). However, while there are many remote sensing
approaches for the quantiﬁcation of geen biomass (Rouse et al. 1974, Kauth and Thomas
1976, Tucker 1979, Huete 1988, Major et al. 1990, Qi et al. 1994), none allows for the
direct quantiﬁcation of senescent biomass.

Efforts to use remote sensing data to quantify senescent biomass in gasslands
have met with limited success. A major difficulty is associated with separating the
spectral proﬁle of senescent biomass from that of the soil backgound (Huete et al. 1985).
There has been some success in quantifying the fractional cover of senescent vegetation,
[or the areal proportion of the landscape occupied by senescent vegetation, not biomass
(White et al. 2000)], by employing approaches that take advantage of the shortwave
infrared region (SWIR: 2.0—2.3 pm), which is sensitive to changes in canopy water
content (Gamon et al. 1993, Qi et al. 2000, Asner and Heidebrecht 2002). The SWIR
region has been used because it contains cellulose (2090 nm and 2270 nm) and lignin
(2130 nm and 2270 nm) absorption features that are masked by water in geen vegetation

but become evident as vegetation senesces (Roberts et al. 1993). For example, NDSVI,

41

the normalized difference senescent vegetation index (Qi. et al. 2000), uses similar
principles to those associated with the calculation of NDVI, the normalized difference
vegetation index (Rouse et al. 1974), but substitutes reﬂectance in the NIR region with
reﬂectance in the SWIR region. In southwestern rangeland ecosystems, this allowed Qi
et al. (2000) to distinguish the fraction of the gound covered by senescent biomass from
that of bare soil. To estimate end-of-the-season biomass values, others have used time
integals of NDVI (Tucker et al. 1983, Tucker et al. 1985, Prince 1991, Wylie et al.
1991). However, these approaches do not satisfy range management needs for biomass
management because they either do not directly estimate biomass or can do so only for
some times of the year. These capabilities would be particularly important for assessing
gazing decisions during dry periods (George and Fulgham 1989, Pickup et a1. 1994,
Wessman et al. 1997, Saltz et al. 1999).

While research on the use of remote sensing for senescent biomass estimates
continues, the accuracy of geen biomass estimates may be increased by characterizing in
detail the impact of senescence on the relationship between NDVI and biomass in gass
species that are representative of gassland ecosystems globally. Vegetation indices, like
NDVI, were desigred to quantify geen biomass only, not total biomass (Tucker 1979).
Thus, most gassland researchers have limited their use of NDVI to parts of the season
when geen biomass is dominant (e.g., Gamon et al. 1995, Wylie et al. 2002, Malmstrom
et al. 2004). However, in many gassland ecosystems, the transition ﬁ'om dominance by
geen to senescent biomass varies from year-to-year, and for signiﬁcant portions of the

season the canopy is a mix of both biomass types (George and Fulgham 1989,

42

Richardson and Everitt 1992, Frank and Aase 1994, Maselli et al. 1998, Qi et al. 2000,
Hoare and Frost 2004).

Although the general effect of senescence on NDVI-biomass estimates is well
understood, especially in crop species (Tucker et al. 1980, Hatﬁeld 1983), no study has
speciﬁcally characterized this relationship for gass species or gass mixes. This detailed
information for gass species and gass mixes that are representative of gassland
ecosystems globally is important because it enables analysis of the error associated with
the use of NDVI for biomass estimates during time periods when senescent biomass is
present. Without detailed phenological information for representative gasses species and
gass mixes, estimates of biomass cannot be accurately made during the transition period
from dominance by geen to senescent biomass. This is especially problematic in range
management situations where within-season management decisions can have a large
impact on biomass production. The aim of this study was to quantify weekly variability
in the canopy properties of three annual gass species that are representative of annual
gasslands globally throughout an entire vegetation cycle, from germination to the end of
the season, to determine the time period that a single NDVI-biomass equation (i.e.,
biomass =f(NDVI)) could be used in gassland ecosystems with mixed species

composition.

Methods
Experimental Design
To examine the effect of canopy senescence on remotely-sensed biomass

estimates in different annual gasses, I established stands of three annual gass species:

43

Avenafatua L. (wild oats; all nomenclature follows Hickman (1993)), Bromus
hordeaceus L. (soft chess), and Lolium multiﬂorum Lam. (annual ryegass) in an
agricultural ﬁeld on the campus of Michigan State University in East Lansing, Michigan
in 2003. I established the plots in Michigan in order to have continuous access to
instrumentation. Use of a Michigan site for this experiment was considered acceptable
because the aim was to examine general temporal variation in the properties of annual
gass canopies in detail during a single gowing season, not to develop a site-speciﬁc
NDVI-biomass equation for any one gassland site.

I chose these species because all three are often common components of grassland
ecosystems in the western United States (Hickman 1993, http://plants.usda.gov/), and are
also representative of annual gasslands globally. Two of the species, A. fatua and L.
multiﬂorum, are naturalized in Michigan as weeds (Voss 2001), so I was conﬁdent that
they would complete their entire phenological cycle, from germination through
senescence, during the study period. In addition, I was intrigued by the structural
differences among these three species, including stem thickness, stem (stand) density, and
stem height (Crampton 1974, Hickman 1993, httmﬂplantsusdamﬂ, and the potential
impact these differences may have on NDVI-biomass estimates in mixed species stands
throughout the season. For example, A. fatua has thicker stems and on average gows
taller than B. hordeaceus and L. multiﬂorum; however, both B. hordeaceus and L.
multiﬂorum generally have geater stem densities than stands of A. fatua (S. Butterﬁeld,
personal observation). I used an experimental approach in order to reduce the impact of

two potential sources of variability in gassland ecosystems: l) backgound plant litter

44

from previous gowing seasons, and 2) changing gazing management regimes. To
ensure that no litter was present, the ﬁeld was tilled prior to the experiment.

To capture among-species variation in seasonal trajectories, I planted single-
species stands of high-density A. fatua (1,000 plants/m2) and low-density A. fatua (500
plants/m2), B. hordeaceus (6,000 plants/m2), and L. multiﬂorum (2,000 plants/m2), and
one mixed species stand containing A. fatua (500 plants/m2) and B. hordeaceus (4,000
plants/m2). I used these densities to broadly emulate those conditions found in annual
gasslands in the western United States (http://plants.usda.gov/). I chose the A. fatua-B.
hordeaceus mixture because these species are often co-dominants in California annual
gassland ecosystems (S. Butterﬁeld, personal observation). There were 10 replicates of
each stand type. I planted the stands in 1—m x 6—m plots in a randomized complete block
design. There were 6 different l—m2 plots within each l—m x 6—m plot. The soils in the

ﬁeld were Riddles-Hillsdale sandy loam 2—6%.

Measurements

ND VI. I measured surface reﬂectance in the experimental plots weekly for 15
weeks, from 11 June 2003 (approximately two weeks after seed germination) until 1
October 2003 (when most stands had senesced and begun to disintegate). To do this, I
used a UniSpec-DC ﬁeld hyperspectral radiometer with a 20°-ﬁeld-of-view fore optic (PP
Systems Inc., Amesbury, MA). The UniSpec-DC detects spectral intensity in 256 bands,
distributed between 0.3 and 1.1 pm, with a resolution of 3.7 nm. I made measurements
with the radiometer in nadir orientation, centered 1.43-m above each l-m2 plot on a boom

attached to a tripod. The gound resolution of these measurements was 0.25-m2. To

45

estimate the contribution of the soil backgound, I also measured surface reﬂectance in

plots of bare soil weekly. To minimize sun angle and shadowing effects, I collected

spectral data within one hour of solar noon, which was calculated using data found at

http://www.wunder;ground.com/. I converted spectral intensity to reﬂectance using a

Spectralon panel (Labsphere Inc., North Sutton, NH). To calculate NDVI, I used mean

reﬂectance (R) values in the red (0.63—0.69 pm) and NIR (0.76—0.9 um) regions as:
NDVI = (RNIR - Rred) / (RNIR + Rred) [EQ- 1]-

Biomass. To permit comparisons of NDVI and herbaceous biomass, I harvested
the stem and foliage biomass from 1--m2 plots within each l—m x 6—m plot at six time
points during the gowing season. For each harvest, 1 clipped biomass at gound level
from the 0.25-m2 portion of the l-m2 quadrat viewed by the radiometer. I then dried the
biomass at 65°C to constant mass, separated it into geen and senescent portions, and
weighed it. I harvested biomass within a few hours of the spectral measurements. I
chose the harvest dates to ensure that two samplings occurred within each of three canopy
stages: 1) early growth, germination to maximum geenness (deﬁned as maximum
NDVI); 2) canopy maturation, onset of senescence to maximum biomass (geen plus
senescent); and 3) canopy decline, maximum biomass to the end of the season.

LAI. To develop a set of nondestructive proxy data that could help me interpolate
the biomass values, I took indirect measurements of canopy leaf area index (LAI) weekly
using a sunﬂeck ceptometer (AccuPAR, Model No. PAR-80 Decagon Devices, Inc.,
Pullman, WA). The ceptometer uses broadband PAR (photosynthetically active
radiation) sensors that cannot readily distinguish between geen and senescent canopy

components (Gholtz et a1. 1991, Gamon et al. 1995, White et a1. 1997). Thus, I deﬁned

46

LAI as the total one-sided area per unit gound area of all abovegound canopy
components, including geen and senescent foliage and stems (Decagon Devices 2001).

AccuPAR-based LAI measurements require the determination of leaf angle
distribution and leaf absorptivity constants as well as measurements of the radiation
above and below the canopy. I calculated LAI as:

LAI=[(1—1/2K)fb - 1] In t / A (1 — 0.47 fb) [Eq. 2],

where K is the canopy extinction coefficient, fb is the fraction of incident PAR, r is the
fraction of transmitted PAR, and A is a function of leaf absorptivity in the PAR band
(Decagon Devices 2001). The canopy extinction coefﬁcient, K, is a function of solar
zenith angle, leaf distribution parameters, and sun conditions (Jones 1992). While the
leaf distribution parameter can vary substantially between crop species (e.g., 0.76 for
maize to 3.03 for strawberry), these differences are minimal for similar annual gass and
crop species (Jones 1992, Decagon Devices 2001). Because I did not have speciﬁc leaf
distribution values for any of the species used in this study, I used the same value, 1.0, for
all ﬁve stand types and took measurements within one hour of solar noon and as much as
possible under the same sun conditions. I selected this leaf distribution parameter
because it represented an average value for annual gass and crop species with similar
gowth habits and canopy structural characteristics in the Decagon manual (Decagon
Devices 2001). I took ﬁve measurements within each 0.25-m2 quadrat viewed by the
radiometer and calculated the mean.

fAPAR. fAPARm] canopy (hereafter referred to as fAPAR), which is used
interchangeably in the literature with jPAR, indicates the ﬁaction of PAR absorbed by

geen and senescent canopy components together (Gholtz et al. 1991, Gamon et al. 1995,

47

Asner and Wessman 1997, White et a1. 1997, Serrano et al. 2000a, Serrano et al. 2000b).
In gass stands, measurements of fAPARgreen, or the fraction of PAR absorbed by green
foliage, have traditionally been used to estimate photosynthetic rates (e. g., Monteith
1977, Gamon et al. 1995). However, measurements of fAPARgm.n and fAPAle canopy
have also been used to estimate gass stand biomass (Le Roux et al. 1997, Asner et al.
1998, Bremer et al. 2001). Thus, to permit comparisons between fAPAR and biomass, I
separately measured fAPAR each week using the AccuPAR ceptometer. The fAPAR
measurements require four readings per measurement: one each with the ceptometer
facing upward both above and below the campy and one each with the ceptometer facing
the gound above and below the canopy. fAPAR was then calculated as:
fAPAR=1-t—r+trs [Eq. 3],
where t is the fraction of incident radiation transmitted by the canopy, r is the fraction of
incident radiation reﬂected to the sensor above the canopy, and r5 is the reﬂectance of the
soil surface (Decagon Devices 2001). I took ﬁve sets of measurements within each 0.25-
m2 quadrat viewed by the radiometer and calculated the mean.

Vegetation characteristics. To compare remotely-sensed measures of changes in
canopy phenology with on-the-gound measures, each week I also measured mean
canopy height and estimated the fractional cover of geen vegetation, senescent
vegetation, and soil. I calculated mean canopy height using stems in six randomly
located circular plots, with a diameter of 9.0 cm, within each 0.25-m2 quadrat viewed by
the radiometer. I estimated cover using Daubenmire classes (Daubenmire 1968): 1 = 0—

5%; 2 = 5—25%; 3 = 25—50%; 4 = 50—75%; 5 = 75—95%, and 6 = 95—100%. In statistical

48

analyses, 1 used midpoint percentiles to represent cover class values: 1 = 2.5%, 2 = 15%,
3 = 37.5%, 4 = 62.5%, 5 = 85%, and 6 = 97.5%.

Biomass equation analyses. To analyze the relationships between biomass and
NDVI,fAPAR, and LAI, I used General Linear Model procedures in SYSTAT 10.2
(SYSTAT Software Inc., Richmond, CA). To compare the biomass equations (i.e.,
biomass =f(X), where X = NDVI,fAPAR, or LAI) among species types, I used Analysis
of Covariance (ANCOVA), in which species (Avena, Bromus, Avena-Bromus, and
Lolium) was the independent categorical effect variable, NDVI,fAPAR, or LAI were
covariates, and biomass was the response variable. The two Avena monoculture
treatments were not sigriﬁcantly different for the NDVI (p = 0.43 7), fAPAR (p = 0.054),
or LAI (p = 0.461) biomass equations, so I combined them for the ANCOVA analyses.
For the ANCOVA NDVI analyses, only geen biomass was considered because NDVI is
only used for estimates of geen biomass. For the NDVI and fAPAR analyses, biomass
values were transformed with the natural log to meet ANCOVA assumptions. In all

cases p 0.05 was considered signiﬁcant.

Results
Seasonal canopy dynamics

Three distinct phenological phases were evident in all of the annual gass species.
Early growth extended from germination until the canopy reached maximum geenness
in week 5 (Figure 3.1 a). Canopy maturation began in week 6, when senescent vegetation
cover became evident (Figure 3.1c) and NDVI began to fall (Figure 3.1a), and ended in

week 9 when maximum biomass (Figure 3.1b),fAPAR (Figure 3.1a), and height (Figure

49

Growth “ ‘ Decline Growth “ ' Decline
‘MAXGRN MAXBle ' ‘ C

 

 

' 3A: MAXGRN MAXBIO _._ NDVI

    

. cactuwu Cover (‘70)

  

 

 

 

Height (cm)

Biomass (g/m2)

 

 

1

 

 

 

 

U u‘ i I I i l i I 111i i i 1‘ i i
012 3 4 5 B 7 8 910111213141518 U 12 3 4 5 8 7 8 910111213141518
Time (week) Time (week)

Figure 3.1. Seasonal changes in A) NDVI and fAPAR; B) geen (GRN) and senescent
(SEN) biomass; C) senescent vegetation cover; and D) LAI and height. Values are mean
weekly measurements averaged over all ﬁve stand types (N = 50) during early growth,
canopy maturation, and canopy decline. Weeks 5 and 9 (shaded columns) were
maximum geenness (MAX GRN) and maximum biomass (MAX B10). All ﬁve stand
types were combined because they displayed the same phenological trends except in
weeks 9 and 10; a severe storm caused stem lodging in the Avena and Avena-Bromus
plots in week 9, which impacted height and cover estimates in weeks 9 and 10 (see text

for details).

50

3.1d) were reached. At this point, the canopy was almost entirely senescent (Figures 3.1b
and 3.1c), and NDVI values were similar to those from week 1 (Figure 3.1a). During
canopy maturation, there was an approximately 20% increase in biomass (Figure 3.1b).
Canopy decline began in week 10 and continued until week 15, during which time the
senesced gass canopies began to disintegate. In week 9, after measurements were
taken, a strong storm caused stem lodging in the Avena and Avena-Bromus stands, and
resulted in a height decrease at week 10 (Figure 3.1d). The subsequent canopy openings
allowed weed gowth in these plots, leading to a short-term increase in geen vegetation

cover (decrease in senescent cover), until weeds were removed (Figure 3.1c).

Relationship between ND VI, fAPAR, and biomass

During early growth, NDVI and fAPAR values increased in parallel with geen
biomass until the point of maximum geenness (deﬁned as maximum NDVI) (Figures
3.1a, 3.1b, 3.2a, and 32b). Because of the difﬁculty of predicting the date of maximum
geenness a priori, I did not harvest biomass exactly on the date but shortly thereafter.
Trends in LAI, the non-destructive proxy of biomass (R2 = 0.71, p < 0.001), suggest that
biomass increased throughout early growth from germination through maximum
geenness (Figures 3.1d and 3.2c). However, once the canopy started to senesce and lose
geenness (week 6), NDVI values began to fall and diverged from biomass and LAI,
which continued to increase until maximum biomass was reached four weeks later. Thus,
during canopy maturation and decline NDVI was a poor predictor of canopy biomass.
When the entire season was considered, however, NDVI was signiﬁcantly correlated to

geen biomass (R2 = 0.78, p < 0.001).

51

Figure 3.2. Seasonal relationships between biomass and A) NDVI, B) fAPAR and C)
LAI during early growth: germination (harvest 1: week 1) to maximum geenness;
canopy maturation: onset of senescence (harvest 3: week 6) to maximum biomass
(harvest 4: week 9); and canopy decline: maximum biomass to the end of the season
(harvest 6: week 15). Harvest 2 occurred in week 2 during early growth and harvest 5 in
week 10 during canopy decline. Values represent weekly means from A. fatua-B.
hordeaceus (open squares), A. fatua (open circles), B. hordeaceus (closed triangles), and

L. multiﬂorum (closed circles) stands. Numbers indicate biomass harvests, not weeks.

52

 

 

 

 

N DVI

4A8

 

 

 

\JUU

4oof

 

:nn

 

i

400-1

3+3

oo-
oo{

3 2
Huge mmmEEm

 

 

400{

00{
oo—

3 2
359 mm .835

100{ 1

 

Figure 3.2.

53

Unlike NDVI,fAPAR continued to climb with both biomass and LAI from week
5 to 9 (Figures 3.1 and 3.2). Thus, fAPAR was signiﬁcantly correlated to biomass
throughout the season (R2 = 0.82, p < 0. 001). Note that mean fAPAR values decreased
slightly between weeks 6 and 8. I did not sample biomass each week during this period.

However, trends in LAI suggest that biomass values leveled off (Figures 3.1a and 3.1d).

Generality of biomass equations

All of the stand types displayed similar phenological relationships among NDVI,
fAPAR, LAI, and biomass throughout the season (Figure 3.2). NDVI sigriﬁcantly
underestimated biomass in all of the stand types after harvest 3 in week 6, or the point in
the season when canopy dominance transitioned from geen to senescent biomass.
fAPAR increased in all of the stands along with biomass and LAI during canopy
maturation and reached its maximum value at harvest 4 in week 9 (Figures 3.2b and
3.2c). fAPAR then decreased, along with biomass and LAI, from harvest 4—6 (week 9 to
15) as the senesced canopies began to disintegate during canopy decline. Consequently,
unlike NDVI, fAPAR was signiﬁcantly related to biomass in all of the stand types from
harvest 1 through harvest 6.

The NDVI,fAPAR, and LAI biomass equations differed sigriﬁcantly among the
stand types tested (e.g., Species * NDVI, p < 0. 05) (Table 3.1). All of the stand types
reached maximum geenness and maximum biomass together (Figure 3.2). The Avena—
Bromus mixed stands had biomass equations intermediate between the Avena and
Bromus monocultures (Table 3.2; Figures 3.3, 3.4, and 3.5). In addition, when the entire

season was considered, the only pronounced differences in the biomass equations

54

Table 3.1. ANCOVA results for differences in biomass equations among stand types.
Species, the independent categorical effect variable, include: A. fatua, B. hordeaceus, L.

multiﬂorum, and A. fatua-B. hordeaceus.

 

 

 

 

 

 

 

Biomass = f(X) Df F P

X = NDVI

NDVI 1 669.76 < 0.001
Species 3 8.04 < 0.001
Species * NDVI 3 6.23 < 0.001
Error 213

X =fAPAR

fAPAR 1 1225.12 < 0.001
Species 3 17.16 < 0.001
Species *fAPAR 3 12.98 < 0.001
Error 216

X = LAI

LAI 1 1018.93 < 0.001
Species 3 1.38 0.15
Species * LAI 3 2.16 0.01
Error 226

 

55

Table 3.2. Stand-speciﬁc biomass equations. NDVI equations are based on geen
biomass data only'. fAPAR2 and LAI’ equations are based on geen plus senescent
biomass data". Note that these biomass equations are given for comparative purposes
only, not as deﬁnitive equations for all gassland situations. See Figure 3 for a gaphical

representation of these equations.

 

 

 

 

 

 

 

Biomass = f(X) Equation R2 P
x = NDVI

Avena y = 1.19e8-5” 0.91 < 0.001
Avena-Bromus y = 1.80e6'64" 0.86 < 0.001
Bromus y = 2.14e6"3" 0.87 < 0.001
Lolium y = 2.64e6'07" 0.93 < 0.001
x =fAPAR

Avena y = 12.45835" 0.91 < 0.001
Avena-Bromus y = 32.54e3'53" 0.90 < 0.001
Bromus y -= 50.65e2'86" 0.85 < 0.001
Lolium y = 27.78e3'78" 0.88 < 0.001
x = LAI

Avena y = 200.21x + 24.46 0.75 < 0.001
Avena-Bromus y = 196.15x + 47.86 0.73 < 0.001
Bromus y = 189.98x + 60.62 0.81 < 0.001
Lolium y = 179.07x + 50.94 0.76 < 0.001

 

 

' NDVI values used to develop these equations ranged from approximately 0.20 to 0.80.
2fAPAR values used to develop these equations ranged from approximately 0.05 to 0.80.

3 LAI values used to develop these equations ranged from approximately 0.05 to 3.0.

4 Biomass values used to develop these equations ranged from approximately 5 g/m2 to 565 g/m2.

56

 

Biomass (glm2)
3000 4000 5000 6000
1 l

2000

1000

 

0
1

 

 

 

_
—
—
‘
u
—

0.0 0.2 0.4 0.6 0.8 1.0
NDVI

Figure 3.3. Stand-speciﬁc biomass relationships. NDVI-biomass relationships represent
data for geen biomass only. See Table 2 for equations, R2 and p values, and footnotes
detailing the range of values used to develop these equations. The top line is A. fatua, the
2nd

line from the top is A. fatua-B. hordeaceus mixture, the 3"d line from the top is B.

hordeaceus, and the bottom line is L. multiﬂorum.

57

 

Biomass (gtm2)
1500 2000 2500
1 1

1000

500
1

 

 

 

0.0 0.2 0.4 0.6 0.8 1.0
fAPAR

Figure 3.4. Stand-speciﬁc biomass relationships. fAPAR-biomass relationships
represent data for geen plus senescent biomass. See Table 2 for equations, R2 and p
values, and footnotes detailing the range of values used to develop these equations. The
top line is A. fatua, the 2“d line from the top is L. multiﬂorum, the 3'1d line from the top is

A. fatua-B. hordeaceus mixture, and the bottom line is B. hordeaceus.

58

 

O
O—
(O /
_/
C')
O-—t
LO
ri—n O
N C)—
E V
x
9
m /
to o ’,
(D O— '
E (‘0
.9
II]
C)
C)—
N
C)
C)—
‘—

 

 

 

0.0 0.5 1.0 1.5 2.0 2.5 3.0
LAI

Figure 3.5. Stand-speciﬁc biomass relationships. LAI-biomass relationships represent
data for geen plus senescent biomass. See Table 2 for equations, R2 and p values, and
footnotes detailing the range of values used to develop these equations. The top line is B.
hordeaceus, the 2'“d line from top is A. fatua-B. hordeaceus mixture, the 3rd line from the

top is A. fatua, and the bottom line is L. multiﬂorum.

59

occurred in the Avena stands at NDVI and fAPAR values > 0.7 (Figures 3.3 and 3.4).
Thus, because biomass increases exponentially with increases in NDVI and fAPAR
(Figures 3.3 and 3.4), small stand-speciﬁc changes in NDVI or fAPAR above 0.7 could

lead to large differences in biomass estimates (Table 3.2).

Discussion

Although NDVI has been used for several decades to estimate geen biomass in
gassland ecosystems (Tucker et al. 1983, Richardson and Everitt 1992, Gamon et al.
1995, Qi et al. 2000, Wylie et a1. 2002, Mustafa et al. 2005), to my knowledge this is the
ﬁrst study to examine seasonal variability in NDVI-biomass relationships in annual gass
stands. My ﬁndings establish that, like perennial gasslands (e.g., Williams 1970, Tothill
1977, Veenendael 1996, Parihar 1999, Bremer et al. 2001), annual gass communities
experience a substantial lag between maximum geenness (maximum NDVI) and
maximum biomass, here approximately 40 days (Figures 3.1, 3.2a and 3.2b). During this
lag period, biomass increased on average by approximately 20% (Figure 3.2b) even
though NDVI decreased (Figure 3.1). Collectively, these ﬁndings demonstrate that a
signiﬁcant amount of above-gound production occurs during canopy maturation in
gassland ecosystems, which NDVI-like indices cannot measure. Consequently,
estimates of annual biomass production based on NDVI at maximum geenness will fail
to account for the additional biomass production that occurs as NDVI decreases. This is
a point of importance not only for range managers but also for carbon modelers who seek

to derive productivity estimates from vegetation indices, like those derived from the

60

MODIS NDVI (e.g., Huete et al. 1999) and/orfAPAR products (e.g., Knyazikhin et al.
1999)

In all stand types tested, fAPAR and biomass were sigriﬁcantly correlated
throughout the season, regardless of canOpy phenological and/or structural attributes
(Figure 3.2). This was the ﬁrst such ﬁnding for stands of annual gasses dominant in
western semi-arid annual gassland like those found in the Central Valley of California
(Malmstrom et a1. 2004, Butterﬁeld and Malmstrom 2006), although a strong season-long
correlation between fAPAR and biomass has also been found in other gassland
ecosystems, including African savannah (Le Roux et al. 1997), Kansas tallgass prairie
(Bremer et al. 2001), and dry Texas gassland (Asner et al. 1998). As in these other
gassland systems, fAPAR in this study increased during canopy maturation because
biomass and LAI increased (Figure 3.1) and so thus did the total amount of radiation
absorbed by geen plus senescent foliage and stems (Asrar et al. 1984, Asner et al. 1998,
Serrano et al. 2000b, J orgensen et al. 2003). Currently, fAPAR cannot be measured
directly from satellite data, so thus cannot be used to quantify stand biomass throughout
the season across large gassland landscapes. However, these ﬁndings do highlight the
potential importance of gound-basedePAR measurements for biomass estimates in
gassland ecosystems, particularly when senescent biomass is dominant and vegetation
indices such as NDVI cannot be used.

To my knowledge, this was the ﬁrst study to explicitly test the commonly held
assumption that remotely-sensed biomass equations can be applied in gasslands with
similar species composition or phenological traits. I found that there were signiﬁcant

differences in all three equation types tested (e. g., biomass =](NDVI)) (Table 3.1), most

61

likely as a result of differences in stand structural attributes (e. g., stem height, LAI, and
stem density, the number of stems per unit gound area). For example, the Bromus and
Lolium plots had geater NDVI per unit biomass than Avena plots (Figure 3.2a), most
likely because Bromus and Lolium stem densities and vegetation cover (data not shown)
were geater. It is also possible that this ﬁnding was a result of differing leaf nitrogen
contents, although I did not explicitly test this as part of this study. Hull and Mooney
(1990) found geater leaf nitrogen content in both B. hordeaceus and L. multiﬂorum than
in A. fatua. Leaf nitrogen content is positively correlated to total chlorophyll amount in
annual gasses (Gaborcik 2003) and thus also to total PAR absorbance and NDVI
(Tucker 1979, Gamon and Surfus 1999). Avena plots, on the other hand, most likely had
geaterfAPAR per unit biomass values than both Bromus and Lolium plots (Figure 3.2b)
because they had geater LAI, biomass, and height values (Asner et al. 1998).

Even though I found that all three equation types (e.g., biomass =f(NDVI)) were
sigriﬁcantly different among the species tested (Table 3.1), when the entire season was
considered there were remarkable similarities in these equations (Figures 3.3, 3.4, and
3.5). The most pronounced differences occurred in the Avena stands at NDVI and
fAPAR values > 0.7 (Table 3.2; Figures 3.3 and 3.4). NDVI values > 0.7, when they do
occur, are most likely to be limited to periods of maximum geenness (Figures 3.1 and
3.2), when both geen cover and geen stand biomass are geatest (Myneni and Williams
1994, Gamon et al. 1995, Bremer et al. 2001, Wang et al. 2001, Fensholt et al. 2004).

fAPAR values > 0.7, on the other hand, are most likely to occur from maximum biomass
to the point in the season when stand biomass starts to decline (e. g., from stem

disintegation). Avena stands with NDVI and fAPAR values > 0.7 are those which are

62

dense and have tall, thick stems. In these stand types, a “general” biomass equation
would likely underestimate stand biomass. However, tall, lush Avena stands are not
common in gazed gassland ecosystems (Butterﬁeld, personal observation). In addition,
in a number of gassland studies, including those in annual gasslands and tall gass
prairies, satellite-based NDVI values did not reach 0.7, even when the canopy reached
maximum geenness (Gamon et al. 1995, Wang et al. 2001, Wylie et al. 2001). Together,
these results suggest that a single equation will be most effective if developed in stands
with similar species composition. However, they also suggest that under most conditions
biomass equations are robust to the structural and phenological differences that exist
among common annual gasses and thus can be used to estimate stand biomass in

gasslands with mixed species composition.

63

CHAPTER 4
REMOTE SENSING-BASED ESTIMATES OF SENESCENT BIOMASS:

COMMON PROBLEMS AND A NEW APPROACH IN ANNUAL GRASSLANDS

Abstract
The value of remote sensing as a management tool in annual gasslands would be

enhanced if remote sensing could be used to directly quantify senescent herbaceous
biomass (dry forage). Globally, senescent biomass is the dominant forage resource for
livestock in gassland ecosystems during dry periods. Current approaches for quantifying
dry forage do not provide estimates of biomass, only of fractional senescent vegetation
cover, which is the areal proportion of the landscape occupied by senescent vegetation.
However, in range management operations, cover is not always an adequate indicator of
range condition, and is a less useful metric of forage production than is stand biomass. In
Chapter 3, I demonstrated that measurements of fAPAR, the fraction of PAR
(photosynthetically active radiation) absorbed by the canopy, could be used to quantify
senescent biomass. However, fAPAR is a gound-based measurement and cannot be
derived directly from satellite data, which limits its utility for large scale management.
To address this issue, I developed a new index, called MAPAR, which estimates the
mean PAR absorbed by the surface and which was based on a fAPAR index developed
by Asner et al. (1998) in gass leaves using hyperspectral reﬂectance data. Unlike
fAPAR, MAPAR can be derived directly from reﬂectance data, and so can be retrieved
both in the ﬁeld as well as from satellite sensor data. In this study, I examined whether
MAPAR could be used to estimate senescent biomass in annual gass stands. Because
MAPAR is a measure of surface absorbance, I also used these gass stands to examine

how changes in soil backgound conditions, such as cover, moisture, and organic matter

64

content, inﬂuenced the utility of MAPAR for biomass estimates. 1 found that across dry
sandy loam soils MAPAR was sigriﬁcantly correlated to senescent biomass. However,
the utility of MAPAR was reduced in stands where signiﬁcant stem lodging occurred and
under conditions where the soil backgound was darkened, either from increases in soil
moisture or organic matter content. In California annual gassland plots with canopy leaf
area index (LAI) values ranging from <0.2 to 6, MAPAR was sigriﬁcantly correlated to
biomass. In these same plots, the gound-based MAPAR measurements were
signiﬁcantly correlated to satellite-based ones, despite the difference in spatial scale (1 -
m2 vs. 900-m2) between the two. Collectively, these results suggest that MAPAR could
be used to estimate senescent biomass in annual gasslands under some conditions.
However, the application of MAPAR is most likely limited to situations where either

fractional soil cover is <15% or the soil is both dry and low in organic matter content.

Introduction

During dry periods in many gass-dominated rangeland ecosystems around the
world (hereafter referred to as gasslands), senescent herbaceous biomass, which consists
of the dry foliage and stems of gasses and forbs, is a dominant forage resource (e. g.,
Bentley and Talbot 1951, George and Fulgham 1989, Prince 1991 , Frank and Aase 1994,
Saltz et al.,1999, Qi et al. 2000). In addition, in these systems the amount of senescent
biomass on the gound at the beginning of the season, referred to as residual dry matter
(RDM) (Hedrick 1948, Bentley and Talbot 1951), is an important determinant of
rangeland condition. RDM quantities represent the combined effects of the previous

season’s production and of use by gazing animals of all types (Bartolome et al. 2002).

65

Some RDM measurements also include dry matter contributions from tree foliage, woody
debris, and gass and forb biomass more than one year old (e.g., Guenther 1998,
Bartolome et al. 2002). RDM is an especially important range indicator because of its
known impact on soil erosion, biomass production, and forage quality (Morrison et a1.
1993, Frank and Aase 1994, Heady and Child 1994, George and Menke 1996, Bartolome
et a1. 2002).

A remote sensing-based approach for the quantiﬁcation of senescent biomass
would provide range managers the means to assess annual gassland conditions and
gazing decisions throughout the season and to better optimize annual use of gassland
resources. Currently, such a tool does not exist: it is difﬁcult to discriminate senescent
biomass ﬁ‘om soil backgounds because senescent biomass lacks the unique signature of
geen biomass present in the visible and near-infrared spectral regions (Huete and
Jackson 1987, Streck et al. 2002). To assess gassland conditions during times when
senescent biomass is dominant, a variety of approaches have been deve10ped including:
1) thermal remote sensing data coupled to NDVI, the normalized difference vegetation
index (French et al. 2000); 2) spectral vegetation indices that use the shortwave infrared
(SWIR: 2000—2300 nm) region (McNaim and Protz 1993, van Deventer et al. 1997, Q1 et
al. 2000, Daughtry 2001, Nagler et al. 2003); and 3) spectral mixture analysis of the
SWIR region (e. g., Gamon et al. 1993). The SWIR region has been used because it
contains unique, relatively narrow cellulose (2090 nm and 2270 nm) and lignin (2130 nm
and 2270 nm) absorption features that are masked by water in geen biomass, but are
exposed as biomass senesces. All of these approaches have been used with some success

in gassland and cropping systems, but mainly to estimate fractional senescent vegetation

66

cover, or the areal proportion of the landscape occupied by vegetation (White et al.
2000), not biomass. These approaches are further limited by reliance on datasets to
which range managers typically do not have access.

In Chapter 3, I demonstrated that gound-based measurements of fAPAR', or the
total fraction of photosynthetically active radiation2 absorbed by the canopy, could be
used to quantify senescent biomass under some conditions in stands of annual gasses,
supporting the ﬁndings of Le Roux et al. (1997) in Aﬁican savannah, Asner et al. (1998)
in dry Texas gasslands, and Bremer et al. (2001) in Kansas tallgass prairie. fAPAR is
calculated as:

fAPAR= 1 —t—r + tr, [Eq- 11
where t is canopy transmittance, r is canopy reﬂectance, and r8 is soil reﬂectance. fAPAR
was measured in the above studies using a AccuPAR ceptometer (Decagon Devices, Inc.,
Pullman, WA).

fAPAR increases with biomass because the total amount of radiation absorbed by
foliage and stems increases (Asner et a1. 1998). Asner et al. (1998) concluded that
senescent gass stands can absorb as much PAR as geen gass stands, even though the
fraction that is used for photosynthesis (i.e.,fAPARm) decreases with increasing
senescence. Other authors have concluded that as a gass or crop canopy transitions from
dominance by geen to senescent biomass, PAR absorption is primarily dependent upon
LAI and the density of vegetation (e. g., Asrar et al. 1984, Serrano et al. 2000b, J orgensen
et al. 2003) rather than the total proportion of geen biomass. However, fAPAR cannot

be derived directly from satellite data because sensors cannot measure canopy

 

' fAPAR is used interchangeably in the literature with fPAR
2 PAR: 400—700 nm

67

transmittance or distinguish the contribution of soil to the overall reﬂectance signal. This
limits the applicability of fAPAR for large-scale range management.

Researchers interested in landscape-scale “fAPAR” measurements have
traditionally used NDVI to estimate fAPARgm, (Field et al. 1995, Knyazikhin et al. 1999,
Los et al. 2000), or the fraction of PAR absorbed solely by geen foliage, because there is
a strong linear relationship between NDVI and fAPARm (Asrar et al. 1984, Hatﬁeld et
al. 1984, Sellers 1985, Choudhury 1987). However, NDVI and fAPAR values for whole
canopies, including senescent fractions, are less correlated (Gamon et al. 1995, Asner and
Wessman 1997). Thus, fAPAR values for canopies with senescent elements must be
derived by other means (Asner et al. 1998). One such strategy proposed by Asner et al.

(1998) used hyperspectral instrumentation and an integating sphere to measure fAPAR:

k
(1.0 — rn — tn)

fAPAR = “:1 [Eq.2]

k

where k = # bands from 400—700 nm; rn = reﬂectance in band 11; and tn = transmittance in
band 11. However, this approach has limited ﬁeld applicability because it uses an
integating sphere to derive both reﬂectance and transmittance, which is not practical
with rangeland stands.

Motivated by the ﬁndings of Asner et al. (1998) and the strength of the season-
long relationship that I found between fAPAR and biomass in stands of annual gasses
(R2 = 0. 82, p < 0. 001) (Chapter 3), I analyzed hyperspectral reﬂectance data that was
taken simultaneously with these fAPAR measurements to explore whether radiometer-

based PAR reﬂectance trends were similar to ceptometer-basedePAR trends. During

68

the transition from dominance of geen to senescent biomass that occurred from
maximum NDVI in week 5 to maximum biomass in week 9 (Figure 4.1), I found that
reﬂectance in the PAR region decreased, and that mean PAR surface absorbance,
MAPAR (Eq. 3), increased with fAPAR, canopy leaf area index (LAI), and biomass
(Chapter 3). Because MAPAR development was motivated in part by the fAPAR
equation used in Asner et al. (1998) (Eq. 2), it is important to note the similarities and
differences between these equations; while both equations use reﬂectance values from the
entire PAR spectrum (400—700 nm), MAPAR (Eq. 3) does not include a transmittance
value in its calculation.

To determine whether MAPAR could be used to estimate senescent biomass, I
measured MAPAR and harvested biomass from stands of annual gass species planted in
an agricultural ﬁeld in East Lansing, Michigan (42° 44’ 10” N, 84° 28’ 59” W).
Because MAPAR is an index of surface absorbance, I also used these stands to examine
how changes in the soil backgound, such as cover, type, and condition, inﬂuenced the
utility of MAPAR for biomass estimates.

To determine the limitations of using gound-based MAPAR measurements in
annual gasslands, I investigated whether the index could be used to estimate geen
biomass in l-m2 plots located in Winters, California (38° 30’ 45” N, 121° 29’ 33” W)
with LAI values ranging from near zero to six. I was particularly interested in testing
MAPAR in annual gasslands where LAI values >5 because other studies have shown
that indices, such as NDVI, can saturate under these conditions in both annual crop
(Asrar et al. 1984, Hatﬁeld et a1. 1984) and gass species (Gamon et al. 1995).

Finally, based on these analyses, I examined whether MAPAR could be used to

69

0.5 .

 

,D
«b
Inn:

Surface Reﬂectance

0.1 -

 

 

 

16111111"IR
UUII'IIUI I.

I l l I l I I I T
400 500 600 700 800 900
Wavelength (nm)

0‘ .l.

 

Figure 4.1. Seasonal changes in surface reﬂectance in stands of Avena (high and low
density), Bromus, Avena-Bromus, and Lolium from week 1 (“germination”) to 5
(maximum geenness: maximum NDVI) to 9 (maximum biomass, fAPAR, and LAI) to
15 (canopy death). Values represent weekly means across all ﬁve stand types. Surface
reﬂectance was derived from UniSpec measurements. The blue (B: 450— 520 nm), geen
(G: 520—600 nm), red (R: 630—690 nm), and near-infrared (N IR: 760—900 nm)

wavelength regions for Landsat satellite sensors are identiﬁed for reference.

70

estimate landscape-scale RDM in gassland ecosystems in the western United States,
where LAI would typically range from <0.5 to 3 (Gamon et a1. 1995, Knyazikhin et al.
1999) depending upon the slope and aspect of the site (Bartolome et al. 2002) and the

management methods in use (Harris et a1. 2002).

Methods
Index. conceptualization in Michigan grass stands

For this study, I focused on four of the ﬁve stand types described in Chapter 3:
Avenafatua L. (all nomenclature follows Hickman (1993)) low density, A. fatua high
density, Bromus hordeaceus L., and A. fatua—B. hordeaceus mixed stands. I excluded
Lolium multiﬂorum Lam. because I wanted to focus on only species present in both
monoculture and mixed stands. There were 10 replicates of each stand type. The soils in
the Michigan gass stands (Chapter 3) were Riddles-Hillsdale sandy loam 2—6%. Prior to
the experiment, the ﬁeld was tilled, so there was no litter from the previous gowing
season present. This study considered the relationship between PAR absorption and
herbaceous biomass, the foliage and stem biomass of the gass species tested.

MAPAR. I measured surface reﬂectance in the Michigan gass stands using a
UniSpec-DC ﬁeld hyperspectral radiometer with a 20°-ﬁeld-of-view fore optic (PP
Systems Inc., Amesbury, MA). The UniSpec-DC detects spectral intensity in 256 bands,
distributed from 300 to 1100 nm, with a resolution of 3.7 nm. 1 made measurements with
the radiometer in nadir orientation, centered 1.43-m above the target portion (0.25-m2

gound resolution) of each plot. To minimize sun angle and shadowing effects, I

71

collected spectral data within one hour of solar noon. I converted spectral intensity in to
reﬂectance using a Spectralon panel (Labsphere Inc., North Sutton, NH).
To calculate MAPAR in the Michigan gass stands using the UniSpec data, I used

reﬂectance values in the PAR region as:

(1.0 — rn)
=1 We 31
k

MAPARfu" =

 

where k = # bands from 400—700 nm; and rn = reﬂectance in band 11.
In this case, MAPARfun uses the entire PAR spectrum, from 400 to 700 nm. There are 92
bands present in these wavelength regions in the UniSpec data. All radiation hitting a
surface must be reﬂected, absorbed or transmitted so that reﬂectance + absorbance +
transmittance = l (Bowers and Hanks 1965). For the MAPAR calculation, I assumed
that transmittance through the canopy was zero. Therefore, incoming radiation was either
reﬂected back to the sensor or absorbed by the surface (vegetation + soil). This
assumption was made partly by necessity, as the UniSpec-DC, like satellite sensors,
cannot measure canopy transmittance in the ﬁeld. For the whole canopy/soil system, this
assumption is met because reﬂectance is the sum of the direct reﬂected light from the
canopy plus the fraction of the transmitted light that is reﬂected from the soil or
understory and then re-transmitted through the canopy (H. Jones, personal
communication). This assumption is false for individual leaves or under conditions
where canopy LAI is low or vegetation is patchy (Goudriaan 1977, Jones 1992, Asner et
al. 1998, Asner et al. 2000). Thus, I assumed that a certain level of error would be

possible when calculating MAPAR early in the season when LAI was low, but that this

72

error would most likely decrease as the season progessed and fractional vegetation cover
and LAI increased (Chapter 3). I examined this assumption directly as the Michigan
gass stands progessed from germination in week 1 (low LAI and cover values) to
maximum biomass in week 9 (high LAI and cover) (e. g., Table 4.1).

I calculated MAPAR using mean PAR absorbance values rather than total PAR
absorbance values or the integal of PAR reﬂectance, as is the case for some
measurements of broadband albedo (the fraction of incident radiation reﬂected by a
surface) (e. g., Maurer et al. 2002). I chose to do this because I was interested in
developing a reﬂectance-based index that possessed a similar “fAPAR-like” ability to
estimate senescent biomass (Chapter 3) and Asner et al. (1998) had used a similar
equation to measure fAPAR in gass leaves using hyperspectral radiometer data (Eq. 2).
For this same reason, I initially calculated MAPAR in the Michigan gass stands using
the entire PAR spectrum (400 — 700 nm) rather than wavelength regions corresponding to
individual satellite sensors (e.g., Landsat: 450—520, 520—600, 630—690 run). I address the
impact of this decision and of scaling gound-based MAPAR measurements to satellite
scales in Tests of MAPAR in California annual grasslands. While I focused on mean
PAR absorbance in this study, the dynamic ranges of mean and total PAR are similar and
total PAR absorbance was also a signiﬁcant predictor of biomass season-long (R2 = 0.60,
p < 0.001). This suggests that these measures may be used interchangeably.

Biomass. I used 1) biomass data from six time points throughout the season, and
2) weekly LAI data (Chapter 3) to examine the seasonal ability of MAPAR to estimate
biomass, both geen and senescent. LAI was signiﬁcantly related to biomass throughout

the season in the Michigan gass stands used this study (R2 = 0. 71, p < 0. 001).

73

Table 4.1. Phenological effects on MAPAR-based estimates in the Michigan Bromus

stands. Fractional cover values represent weekly means. Fractional soil cover was

calculated as the midpoint of Daubenmire classes (Daubenmire 1968). Fractional

senescent cover is the percentage of total vegetation cover. Biomass was only harvested

six times over the gowing season. R2 values represent relationships with MAPAR.

Sigriﬁcant at:p<0.001 = ***,p<0.01 = **, and p<0.05 = *.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Week Soil Proportion of vegetation LAzl fAPAR Biongass
Fraction cover that was senescent (R) (R ) (R )
1 62.5 0 0.00 0.00 075*“
2 15 0 0.82*** 0.95***
3 15 0 0.71*** 0.72*** 0.94***
4 15 0 078*“ 0.79***
5 2.5 0 0.70*** 0.34
6 2.5 2.5 0.86*** 0.61* 0.11
7 15 15 0.71*** 0.70***
8 15 37.5 0.75*** 0.34
9 15 97.5 0.56* 0.48* 0.60**
10 15 85 0.72*** 0.70*** 0.76***
11 15 85 0.81*** 0.78***
12 37.5 97.5 0.58* 0.88***
13 37.5 97.5 0.73*** 0.77***
14 37.5 100 0.46* 0.56*
15 37.5 100 0.51"“ 0.56* 0.69***
Season 0.73*** 0.62** 0.91***

 

74

Therefore, I used LAI as a non-destructive surrogate for biomass.

I used General Linear Model (GLM) procedures in SYSTAT 10.2 (SYSTAT
Software Inc, Richmond, CA) to analyze the relationships between MAPAR and fAPAR,
LAI, and biomass. To compare the MAPAR-biomass equations (i.e., biomass =
f(MAPAR)) developed for the geen (harvests 1—3; weeks 1, 3, and 6) and senescent
(harvests 4—6; weeks 9, 10, and 15) time periods, I used Analysis of Covariance
(ANCOVA). In the ANCOVA analyses, time period (geen vs. senescent) was the
independent categorical effect variable, MAPAR was the covariate, and biomass was the
response variable. For these analyses, biomass values were natural-log-transformed to
meet ANCOVA assumptions. Relationships with p 0. 05 were considered sigriﬁcant.

Solar zenith angle tests. I analyzed weekly relationships 1) between mean
MAPAR values (Avena, Bromus, and Avena-Bromus) and solar zenith angle (SZA), a
signiﬁcant component of the BRDF; and 2) between MAPAR and LAI,/APAR, and
biomass in the Michigan Bromus stands. Solar zenith angle is the angle measured at the
earth’ surface between the sun and the zenith (Liang et al. 2002). I focused on the
Bromus stands because they provide a broad range of fractional vegetation and soil cover
ratios for evaluating the effectiveness of MAPAR for estimates of biomass (Table 4.1).
Across vegetated surfaces, NDVI and albedo have been shown to increase independently
of changes in canopy structure at SZA values >40° (Qi et al. 1995, Danaher 2002).
NDVI increases with SZA because of reduced illumination and increased shadowing of
the soil backgound, and increased illumination of the vegetated surface (Danaher 2002).
Using this logic, MAPAR, which is broadly inversely related to NDVI (and albedo),

values could be expected to decrease at SZA values >40°. This effect is reduced across

75

bare soil (Qi et al. 1995), although Idso et al. (1975) showed that albedo increased at SZA
values >40° as well. I calculated SZA using software available at
http://solardat.uoregon.edu/SolarPositionCalculator.htm1, together with sun rise and sun

set data found at http://www.wunderground.com/.

 

Soil background effects. Soil backgound effects are known to have a large
inﬂuence on remotely-sensed biomass estimates (e. g., Huete 1988). To estimate the
contribution of the soil backgound to the MAPAR measurements in the Michigan gass
stands, I measured surface reﬂectance in plots of bare soil (Riddles-Hillsdale sandy loam
2-6%) weekly, and compared MAPAR with weekly measurements of fAPAR, LAI, and
biomass in the Bromus and Avena stands. In these stands, I was able to examine the
ability of MAPAR to discriminate vegetation from backgound soils in plots with
fractional vegetation cover values ranging from 2.5% to 97 .5%. I also made additional
soil measurements in an adjoining agricultural ﬁeld at the Michigan State University
Plant Pathology farm located in East Lansing, Michigan with Houghton Muck organic
(8.0%) soils. In addition, I made measurements outside using pots with Riddles-Hillsdale
sandy-loam (dry/wet), sand (dry/wet), and Houghton Muck (dry/wet) soils, respectively.
I used repeated measures (paired) t-test procedures in SYSTAT 10.2 to analyze the
differences between MAPAR-based vegetation and soil values; mean weekly values were
calculated and used for each analysis. In all cases p 0. 05 was considered signiﬁcant.

To address the reliability of MAPAR for biomass estimates in different soil
backgound conditions than those used in this study, I compiled reﬂectance data from

both soil and vegetation on different soil types from the literature (Table 4.2). I broadly

76

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$2 atom E aims .3 5 Halawslealsz 02:6
32 BE E .0533 .3 b u 53 sea
52 come—03. was 80:2 N< .583. b u :33 3| :3
$2 815: can Eo3om mu £33582 :3 in; BBC in:
88 a a 055m engage be :5 salami?
who atom 7E 650% .3 331.6 Em «5305.5
82 min a as. 235m E .5525: 325 52 am aeoEoy
wba Lubom 7: .otownﬂs .3 “03%;. N30 0x0559—
052 .3 8 sum: <0 .momoEom 0x3 36 43—0 Eonom
wag worm 7E .ozuwﬂj .3 «03 Ewe. U"BU Eng—.202
$2 acme—02. can 225 N< £083. .03 gm 93:05:55
32 made: new EuBOm mm 53:22 33 “was 1.3qu
ME: seem E 6535 .3 b a 9% 8226 =8
$2 came—03 use 80:2 N< demosh b c team all EM:
5535 25 5.5230 9::on mar—om

 
 

83:58 «3:83 he ~33:

«3 2.39 8: «E .5 E 5 «E %:2 93:3 35:85:: :8...— m:o_:v:8=em

 

 

 

 

 

 

 

  

 

 

 

 

 

<0 5233 36 be xooysgﬂgumom 3.9:: .350
E gamma .m 399$ nice elated v.82 Egon
<0 .2253 a e be 884 “each =8
5 dead .m 86 a; mac. 32am 252::4232 f5
<0 5253 k: be as New 3&6
<0 .2253 as be so am 883.88%
<0 £253 23 5 Eu Begum
E .325 .m as he 58. a 2§=£ae§m :8
E .323 .m Eggs .235 saw Bus
85 2...; ﬁg: 538.50 9:5qu 81%

 
 

62.13.: 53:3 o> 2.33:8 .8613... .3389 o> =8.— ..8 ¢<A<E ” 2.: E: he «.3 an :38 enuEeEnaeE :5

.=8 80¢ couﬁowg oﬁﬁatoﬂv 3 M<m<2 mo 35% 05 so 8238 98 0%» :8 mo cocoscﬁ 2F .Né 03:.

77

categorized soils as light-colored or dark-colored. For those soils used in this study, I
deﬁned “light-colored” soils as those that had MAPAR values signiﬁcantly lower than
both green and senescent vegetation, and “dark-colored” soils as those with MAPAR
values greater than green and senescent vegetation. For example, I categorized the
Riddles-Hillsdale sandy-loam soil (2-6% organic matter content) used in the Michigan
grass stands as light-colored (Table 4.2) because the average MAPAR value ( 0.84) was
signiﬁcantly lower than values for both the green (0.93—0.97) and senescent (0.91—0.95)
vegetation throughout the entire season. By necessity, I included MAPAR values only
for the measurements that I took as part of this study. In all other cases, I broadly
categorized soils as “light” or “dark” based on their published reﬂectance curves as well

as their similarity to soils used in this study.

Tests of MAPAR in California annual grasslands

fAPAR, LAI, and biomass. To test the reliability of MAPAR for biomass
estimates in grassland ecosystems, I collected data ﬁ‘om 80 1.0-m2 plots randomly
distributed across annual grasslands located in Winters, California (Malmstrom et al.
2004, Butterﬁeld and Malmstrom 2006) from 28—29 March 2003; the period of
maximum greenness at this site. I determined the coordinates of each plot with a Trimble
Pro XRS GPS unit (Trimble Navigation Limited, Sunnyvale, CA) using real-time
differential correction. At each plot, I measured MAPAR,fAPAR, and LAI and assessed
Daubenmire fractional cover (Daubenmire 1968), and then clipped all of the aboveground

biomass. To facilitate comparisons between ground-based MAPAR measurements and

78

those from Landsat satellite data (see Satellite-based MAPAR estimates), 1 calculated

MAPAR from ﬁeld spectroradiometer data using the Landsat bandwidths as:

kblue kgreen kred
(1.0 — rum) + (1.0 — rngm) + (1.0 — rnwd)
nhltm= 1 ngrecn:l nred = 1

MAPARtm (unweighted) =

 

[EQ- 4]
kblue + l(green + kred

where kblue = # of bands from 450—520; kgreen = # of bands from 520—600;

kmd = # of bands from 630—690; rublue = reﬂectance in blue band n;

r = reﬂectance in green band n; rumd = reﬂectance in red band n;

"green
with UniSpec data, k1,.“c + kgrm + km, = 65;
and with Landsat data, kblue + kg...“ + km = 3
Note that when using Eq. 4 to calculate MAPAR there are regions within the PAR region
that are not included (e.g., 400 to 450 and 600 to 630). The relationship between
MAPARrun (Eq. 3) and MAPARm (unweightedﬁs represented in Figure 4.2 using the
Michigan Bromus stands and the California annual grassland plots. MAPARfuu was
greater than MAPAR.m (unweighted) under all conditions tested, including when green
biomass, senescent biomass, and the soil background were dominant. In the Michigan
Bromus stands and California annual grassland plots, this effect was reduced as MAPAR
(and biomass, LAI, and cover) increased.
I did not separate green and senescent biomass; therefore, biomass includes both
components. I dried biomass at 65°C to constant mass and weighed it. The soils in these
plots were either light-colored Sehom clay or Sehom-Balcom silty clay, or dark-colored,

Tehama loam (Table 4.2). I used GLM procedures in SYSTAT 10.2 to compare

79

 

0.98

llLll

0.96

q

111

5:

(D

A
I

d

.O

.0 (.0

(D M
Ill lllllll

.0

co

co
1

l

MAPARtm: Bromus

.0

00
c)
1

l

d

0mﬁ

GRN:y=1.0627x- 0.0624
SEN: y: 1.0323x- 0.0331
Season: y: 1.068x- 0.0669

El

El GRN
A SEN

A

 

 

002‘

0.82 0.84

11"]IIIVIIY‘IIIT—IUIjUUTTIrﬁl'ﬁ111lii

0.86 0.88 0.9 0.92 0.94 0.96 0.98
MAPARfuII: Bromus

 

 

000
007
000

6005

5094

( .

0.

£003
002
um

0.9

d]

' y=1.0341x- 0.0353

8

 

0.9 0.91 0.92 0.93 0.94 0.95 0.96 0.9?r 0.98

MAPARfUII: CA

Figure 4.2. Relationship between MAPARfuﬂ (Eq. 3) and MAPAR“m (unweighted) (Eq. 4) in

A) the Michigan Bromus stands (N = 150) and B) the California annual grassland plots

(N = 169). Relationships in the Bromus stands are shown for the green (GRN: weeks 1-

7) and senescent (SEN: weeks 8-15) periods as well as for the entire season (weeks 1-15).

80

MAPAR with fAPAR, LAI, and biomass.

To estimate the contribution of soil background to MAPAR measurements in the
California annual grassland plots, I measured surface reﬂectance in plots of bare soil
from all three soil types. In all cases, the soils were dry (Table 4.2) and data were
collected within the same +/- one hour of solar noon window as the vegetation
measurements. To analyze the differences between vegetation and soil values I used one-
sample t-test procedures in SYSTAT 10.2; p 0.05 was considered signiﬁcant.

Satellite-based MAPAR estimates. To examine whether I could use satellite-
based MAPAR measurements to predict on-the-ground conditions, I acquired a Landsat
Enhanced Thematic Mapper (ETM+) LlG geo-referenced image from 28 March 2003
that coincided with the ﬁeld measurements in the annual grassland in Winters, California.
Level 16 ETM+ imagery are delivered as calibrated digital numbers, which I converted
to top-of-the-atmosphere (TOA) reﬂectance using ERDAS Imagine 8.6 (Leica
Geosystems, Atlanta, GA). 1 then converted the TOA reﬂectance values to surface
reﬂectance with the SS radiative transfer model (V errnote and Roger 1996). I calculated
MAPARm (unweighted) values for each pixel using Eq. 4. When using actual Landsat data,
the summation terms in Eq. 4 are not needed because there is only one reﬂectance value
per pixel in the blue, green, and red bands, respectively. Because these bands are not
equal in size (e. g., 70, 80, and 60 nm, respectively), if values are not weighted by
bandwidth, the MAPARtm (unweighted) value (Eq. 4) may diverge somewhat from the
MAPARfu" values (Eq. 3) derived from UniSpec measurements under some conditions.
Differences may also exist because there are missing spectral regions in the MAPAR(m

(unweighted) calculation (Eq. 4) when compared to MAPARfuu (Eq. 3).

81

To analyze the effect of weighting vs. not weighting MAPAR,m by bandwidth, 1 used the
UniSpec data from the Michigan Bromus stands and California annual grassland plots. I
calculated MAPAR in the Bromus stands using the MAPARﬁ," (Eq. 3) and MAPAR(m
(unweighted) (Eq. 4) equations as well as a MAPARm (weighted) equation weighted for Landsat
bandwidth. (Note: I did not weight MAPARﬁm for UniSpec bandwidth differences
because the UniSpec bands are equal in size, 3.7 nm). The MAPAR(m (weighted) equation
has the same denominator as the MAPAR;m (unweighted) equation (Eq. 4) (i.e., km“e + kg":en +

km), but the numerator for MAPARtm (weighted) is different and is calculated as:

kbluc kgreen kred
[bw.,..*( 0.0—r1.....1)]+[bwm*( tl-o—rn...1)]+[bw~a*( [1.0—rub]
nblue=1 figmen=l nred =1

 

bwbluc + bwgreen + bwred

where kblue = # of bands from 450—520; k3,...” = # of bands from 520—600; km) = # of
bands from 630—690; rub,“ = reﬂectance in blue band 11; rum..." = reﬂectance in green band
n; rmd = reﬂectance in red band n; and bw = bandwidth.

I calculated MAPAR.lm in the California annual grassland plots using the
unweighted and weighted (Eqs. 4 and 5) equations. MAPAR“ (wcighwd) was greater than
MAPARtm (unweighted) under all conditions tested (Figure 4.3). This occurred because while
both equations have the same denominator (e. g., kblue + kgmm + km; = 65) (Eqs. 4 and 5),
the numerator for the MAPARtm (weighted) equation (Eq. 5) was always greater, often 3—4X
greater, than the numerator for the MAPAR,m (unweighwd) equation (Eq. 4). In the Michigan
Bromus stands this led to large differences in MAPAR-based biomass estimates; in all
cases, MAPAR:m (weightedybased biomass estimates (y = 6E-29 * EXP(71.97x)) were

smaller than MAPARtm (unweighmdybased estimates (y = 3E-08 * EXP(24.23x)) for a given

82

Figure 4.3. Relationships between A) MAPAR,m (unweighted) (Eq. 4) and MAPARtmwcighwd)
(Eq. 5), and B) MAPARfu" (Eq. 3) and MAPARm, (weighted) in the Michigan Bromus stands
(N = 150); B) MAPARtm (unweighted) and MAPARtmwcighted) in the California annual

grassland plots (N = 169).

83

 

 

 

 

 

 

 

 

 

 

 

 

0_gy_ /
0')
‘ m
E . 1: ma
2 . D
9‘. .
E 0.92-
1: ..
.9 .
a)
3, 4
g d
33 0.87— y: 0.3368x + 0.6633
a .
<(
E -l
‘ A
0.82 U I I I I T—T Iﬁ I I I 1 I I l
0.82 0.8? 0.92 0.97
MAPARtm (unweighted): Bromus
0.97.. M
03 ..
E 1:1 mil m
E D
‘F! .
E 0.92-
; .
.9
a)
3,
g ..
g 0'37: y: 0.3553x + 0.6448
c(
E
‘ B
0.82 I I I I l I I I I l I I 1 I I I
0.82 0.87 0.92 0.97
MAPARfuII (unweighted): Bromus
1 -
éa' ' u ‘3
9.
5096-
CD ..
3, .
g 0.94-
4: 1 y: 0.3338x + 0.6656
O.
< 1
F- 0.924
1 c
0.9 rTIIlIIII'IIII'IIII'IIII
0.9 0.92 0.94 0.96 0.98 1
MAPth (unweighted): CA
Figure 4.3.

84

MAPAR value. This occurred because as MAPAR.m (weighted) values got larger (vs. those
which were unweighted) stand biomass values remained the same. For this study, I chose
to calculate MAPAR with Landsat data using Eq. 4 because this allowed for the most
direct comparison with ground-based MAPAR measurements used in this study, and also
allowed for the greatest consistency between the California ﬁeld plot comparisons and
those from the Michigan grass stands. In addition, the MAPARm (unweighwd) dynamic
range (e.g., in the Bromus plots 0825—0955) was larger than the MAPARm, (weighted)
dynamic range (e. g., in the Bromus plots 0.945—O.985), which could allow for greater
discrimination of MAPAR and/or biomass values within a heterogeneous grassland.
However, end users should consider this weighting issue when deciding which MAPAR
equation is most appropriate for their own situation.

I compared the Landsat-based MAPAR;m (unweighted) values to those from the 80 l-
m2 quadrats. I compared quadrat values to those from Landsat pixels in which they were
nested. When more than one quadrat fell within a given 9OO-m2 Landsat pixel, the
quadrat values were averaged before comparison.

To examine whether satellite-based MAPAR measures could discriminate bare
soil from both senescent and green vegetation, I randomly extracted Landsat-scale (900-
m2) bare soil values from cultivated ﬁelds in Winters, California using a 9 May 2001
ETM+ image for the senescent period and the 28 March 2003 image for the green period.
I processed the May image using the same methodology described above. I extracted the
soil values from ﬁelds with dry light-colored Capay silty clay and dark-colored Tehama
loam soils (Table 4.2). I used these ﬁelds, instead of smaller bare soil patches, to

decrease the likelihood of mixed pixel effects. The ﬁelds had been harvested 2—4 weeks

85

before image acquisition, so the soil moisture conditions of bare ground were accurately
represented. I determined harvest dates using additional Landsat data acquired as part of
Malmstrom et al. (2004). One drawback of this approach is that cultivated land can be
texturally different than bare ground in a grassland ecosystem. However, this is likely
secondary to the sub-pixel mixing effects that would have occurred if I had used smaller
patches. I used t-test procedures in SYSTAT 10.2 to analyze differences. In all cases p

0.05 was considered signiﬁcant.

Results
Tests of MAPAR in Michigan grass stands

fAPAR, LAI, and biomass. MAPARfouAPAR, LAI, biomass, and NDVI
increased in all of the Michigan grass stands between week 1 and week 5. At week 5, all
plots achieved maximum greenness (deﬁned as maximum NDVI) (Figure 4.4). NDVI
decreased thereaﬁer due to increasing canopy senescence. In contrast, MAPARfuﬂ,
fAPAR and LAI continued to increase until week 9, when maximum biomass occurred.
Notably, MAPARﬁ," followed the phenological trends in all three canopy variables
through week 15, including the decreases from week 9 to 11. MAPARbeAPAR, and
LAI values did not change markedly from week 11—14 (Figure 4.4), thus causing the
MAPARfun-biomass (Figure 4.5a), fAPAR (Figure 4.6a), and -LAI (Figure 4.6b)
relationships to saturate around MAPARfu“ values of ~0.936. This phenomenon was
reduced somewhat in the Bromus stands (Figure 4.5b).

Across weeks and stand types, mean MAPARfuﬂ was strongly and signiﬁcantly

correlated to mean fAPAR, LAI, and biomass (Figures 4.5a and 4.6); this correlation was

86

 

 

 

 

 

MAPAR-based bare soil range: 0.80 - 0.64
I I

 

 

 

 

 

0-88 I Iﬁ fl I I I I I I 0-3
25 0.7
:B I

2i
1
1.5%
3 1
1+
0.55
-¢ LAI ;
0 ﬁ I I I I I I I I I I I I I .0
12 3 4 s 0 7 0 9101112131415

WEEK

Figure 4.4. Seasonal changes in A) NDVI and MAPARfuﬂ (Eq. 3) and B) fAPAR and
LAI; week 5 is maximum greenness and 9 is maximum biomass. Between weeks 9 and
10, there was a strong storm that caused signiﬁcant stem lodging and facilitated invasion
by green weedy vegetation; LAI decreased as the plots were weeded. Values represent
weekly means for the Michigan Avena (high and low density), Bromus, and A vena-

Bromus stands (11 = 40).

87

O)
O
D

 

GRN I

11: 25-07 EXP(22.56x)
R2 = 0.80, p < 0.001
SEN 0

NS (p > 0.05)

Season

y=1E-07 EXP(23.21x)
R2 = 0.75, p < 0.001

llllll

w h (J!

O O D

D O D
ll_lJ_LJlllll

M

D

D
111.1!

Biomass (gtm2): all stand types

..s

O

O
l

l I

 

 

 

0 ..l.—.!,l#',—...

0.75 0.8 0.85
MAPAR: all stand types

SOD-GRN . O

i y=1E-08 EXP(25.32X)
~ R2 = 0.90, p « 0.001 0
. SEN o

- 1;: 215-09 903050111) (é?
R2 = 0.43.10 < 0.01 0
Season

/
. y: 9507 EXP(25.58x) o o
3 R2 = 0.791 p < 0.001 0°

 

 

(.0
D
D
l l l l

m

D

o
r
O

Biomass (gim2): Bromus

 

 

 

 

1 I I I I U

' l
0.83 0.88 0.93
MAPAR: Bromus

Figure 4.5. Seasonal patterns in the relationship between MAPARﬁ.“ (Eq. 3) and
biomass in A) all of the Michigan stand types combined (Avena high and low density,
Bromus, and Avena-Bromus), and B) Bromus stands alone. Relationships are shown for
the green (GRN: harvests 1—3; weeks 1, 3, and 6) and senescent (SEN: harvests 4—6;
weeks 9, 10, and 15) periods as well as for the entire season. In B), the solid line is for

the GRN time period and the dashed line is for the SEN time period.

88

 

:y=8.85x-7.82 9
05.: 92 = 0.95 D
E p < 0.001 10:]

 

 

 

 

 

2.5
: y = 35-17 * EXP(40.79X)

: R2 = 0.92 9
2‘: p < 0.001 1g

1
1.5-:

 

 

 

 

Figure 4.6. Seasonal relationship between MAPARfun (Eq. 3) and A) fAPAR, and B)
LAI. Numbers (1—15) represent weeks, not harvests. Values represent means in the

Michigan Avena, Bromus, and Avena—Bromus stands (11 = 40).

89

also signiﬁcant in the Bromus stands alone (Table 4.1; Figure 4.5b). When data from all
four stand types were combined, MAPARfu“ was signiﬁcantly correlated to biomass
during the green period (week 1—7) and for the entire season (week 1—15) (Figure 4.5a).
However, during the senescent time period (week 8—15), MAPARfu“ and biomass were
not signiﬁcantly correlated (Figure 4.5a). In Bromus stands, MAPARfu" and biomass
were signiﬁcantly correlated during both the green and senescent time periods (Figure
4.5b). In addition, in Bromus stands the green and senescent biomass equations (Figure
4.5) were not signiﬁcantly different (MAPARfuﬂ * Time, df = 1, F = 0.068, p = 0.795).

Solar zenith angle tests. MAPARfuu increased with solar zenith angle (SZA) from
week 1 to 9 in the Michigan grass stands (Figure 4.7). During this time period, there was
a signiﬁcant positive relationship between MAPARfuﬂ and SZA (R2 = 0. 60, p < 0. 05).
However, this relationship lost its signiﬁcance as MAPARfun decreased ﬁ'om week 10 to
15, while SZA continued to increase (Figure 4.7). Solar zenith angles were greater than
40° in weeks 14 and 15. During this period, the decrease in MAPARﬁ." may have
occurred independently of the decreases in stand biomass and LAI (Figure 4.4).
MAPARfuﬂ'baSCd bare soil values ranged from 0.80—0.84, but were the same, 0.84, in
week 1 and 15. In the Bromus stands, MAPARmn was signiﬁcantly correlated to fAPAR,
LAI, and biomass within individual weeks, when SZA was constant (Table 4.1).
MAPARfuu was signiﬁcantly correlated to all three variables throughout the season,
except for weeks 1 (not signiﬁcant for fAPAR or LAI), 5 and 8 (fAPAR), and 6
(biomass).

Soil background eﬂects. The soil background was dominant in the Bromus stands

in week 1 (Table 4.1) and in the Avena stands in weeks 1—3 and 11—1 5 (fractional soil

90

 

SolarZenith Angle (SZA)

 

[3- B
9 '9‘“ .9- 32A _—0.89

 

 

 

15 I ﬁI I I I I I I I I I I I
1 2 3 4 5 6 7 8 9101112131415
Week

Figure 4.7. Seasonal relationship between MAPARfu" (Eq. 3) and solar zenith angle
(SZA, measured in degrees). MAPARfu“ values represent weekly means for the

Michigan Avena (high and low density), Bromus, and Avena-Bromus stands (11 = 40).

9|

cover: 62.5—97.5%). In the Bromus stands, MAPARfu" values in week 1 were correlated
only with biomass, not with fAPAR or LAI, which were difﬁcult to measure when
vegetation was short and cover was low. In the Avena stands, MAPARfu“ was
signiﬁcantly correlated to LAI in weeks 1—3, and 11, 12, and 14 (R2 = 0.49, 0.50, 0.50,
0.56, 0.62, 0.49; p < 0.05). In these same stands, MAPARfuu was signiﬁcantly correlated
to fAPAR in weeks 1—3 and 15 (R2 = 0.72, 0. 76, 0.55, 0.45; p < 0.05) and biomass in
weeks 1, 3, and15 (R2 = 0.66, 0. 76, 0.4 7; p < 0.05). Throughout the season in the
Michigan grass stands, MAPARfuﬂ values from dry soil were signiﬁcantly lower than
those from vegetation (Table 4.3). Values from dry Houghton Muck soils were similar
to, and in some cases larger than vegetation values (Table 4.2), although they were still
signiﬁcantly lower overall (Table 4.3). As soil moisture increased, soil values increased

and became similar to, or greater than, vegetation values (Table 4.2).

Tests of MAPAR in California annual grasslands

fAPAR, LAI, and biomass. In the California annual grassland plots, MAPAR,m
(unweighted) measured with ﬁeld instruments was signiﬁcantly correlated to LAI, biomass,
and fAPAR during the period of maximum greenness (Figure 4.8). Notably, unlike with
NDVI-like indices (Asrar et al. 1984, Hatﬁeld et al. 1984), the MAPAR-based
relationships did not appear to saturate at LAI values >5. Dry bare soil values for Sehom
clay, Sehom-Balcom silty clay, and Tehama loam soils were signiﬁcantly lower than
vegetation values (Table 4.3).

Satellite—based MAPAR estimates. Overall, in the California annual grassland

plots, the ground- and Landsat-based MAPAR(m (unweighted) values were signiﬁcantly

92

Table 4.3. Results of the t-tests for differences between MAPAR-based vegetation and
soil values. The Michigan grass stand comparisons were made using green and senescent
biomass data (weeks 1—15). For the ground-based California annual grassland
comparisons, only green biomass was considered. The Landsat satellite-based California
annual grassland comparisons were made either in May, when senescent biomass was

dominant, or March, when green biomass was dominant.

 

 

 

 

Soil Type Soil-Vegetation t df P
Comparison

Experimental plots

(Michigan)

Riddles-Hillsdale sandy loam Green and senescent -17.00 14 <0.00l

Houghton Muck organic Green and senescent -6.02 14 <0.00l

 

 

Natural grasslands - ground
(California)

 

Sehom clay/
Sehom Balcom silty clay/ Green -13.23 10 <0.00l
Tehama loam

 

 

Natural grasslands — Landsat

 

 

 

 

(California)

Camay silty clay— May Senescent -22.35 19 <0.00l
Tehama loam— May Senescent -l6.88 19 <0.001
Capay silty clay— March Green -26.86 19 <0.001
Tehama loam— March Green -25.47 19 <0.00l

 

93

Figure 4.8. Relationship between MAPARtm (unweighted) (Eq. 4) and A) LAI, B) biomass,
and C) fAPAR in California annual grassland plots during the period of maximum
greenness (28—29 March 2003). MAPAR:m (unweigmd) was derived from UniSpec

measurements. LAI and fAPAR were derived from AccuPAR measurements.

94

O)

 

 

  
  

 

 

 

 

 

 

 

 

  
  

 

 

 

g y = 3E-38 EXP(90.83x) U
5; R02 2 0.62
:p<0.01
84'.
c: 1
N i
63‘:
52%
1%
: _ . A
o .4..,_...ﬁ,....,.........
0.93 0.04 0.95 0.90 0.97 0.00
500.
5 y = 1E-17 EXP(46.37X)
m 4503 R02: 071 D
D 400': .
SE} jp<0001 091%
5 350:
A300“: 3
‘E
La3250-,
m 200‘:
09 1
g 150-;
1% 100;
50-5
2 B
n .r-.,....,.........,....
0.93 0.94 0.95 0.90 0.07 0.00
1
a y = 0.0269x + 0.9386
”9‘; R2 = 0.63 9"”
0.8g p «001 '3
307-3 ‘3
805.3
6 '
If1:05: [mg g
EDA-5 IUD
1.0.35
0.2-j a: i
01-? ﬂ,
2 C
n ,,.,
0.93 0.94 0.95 0.90 0.97 0.98

MAPAR: CA 2003

Figure 4.8.

95

correlated (R2 = 0.47, p < 0.001). In most cases, Landsat-based MAPARtm (unweighted)
values were lower than ground-based estimates, with differences ranging from 0.0004 to
0.0189. Landsat-based MAPARtmmweigmed) soil values for both light-colored Capay silty
clay and dark-colored Tehama loam soils (Table 4.2) were signiﬁcantly lower than

senescent and green vegetation (Table 4.3).

Discussion

A remote sensing-based approach for the quantiﬁcation of biomass that could be
used by managers regardless of the dominance of green or senescent biomass would
revolutionize biomass management in grassland ecosystems globally. Armed with this
tool, managers could assess grassland conditions, including RDM levels, throughout the
season and make real-time grazing decisions across large grassland units. One of the
main reasons such a tool has yet to be developed is the difﬁculty and expense associated
with quantifying senescent biomass (Huete et al. 1985, Huete and Jackson 1987, Streck et
al. 2002).

I tested the utility of the MAPAR index as a tool to estimate senescent biomass
because it behaved similarly to fAPAR throughout the season (Figure 4.4) and could be
directly derived from accessible satellite sensor data (i.e., Landsat). I found that, across
the dry sandy loam soils in this study, MAPAR closely followed the seasonal trajectory
of fAPAR, LAI and biomass (Figure 4.4; Chapter 3), demonstrating that in dry light-
colored soils (Table 4.2) MAPAR was sensitive to phenological changes that occur in

annual grass species.

96

It is important to deﬁne MAPAR in relation to albedo because albedo is a
common measurement used in the remote sensing literature (e. g., Qi et al. 1995, Liang et
al. 2002) as well as a current end-product of the MODIS satellite sensor (Lucht et al.
1998). It is intriguing to imagine a future situation where MAPAR values could be
extracted directly from MODIS data. MODIS, like the UniSpec data, acquires ground
reﬂectance data across the entire PAR spectrum using bands with approximately the same
bandwidth (10—1 5 nm in the case of MODIS). In general, MAPAR and albedo are
inversely related; as albedo increases, MAPAR decreases. However, this relationship can
vary depending on the surface type (and thus the magnitude of canopy transmittance) and
bidirectional reﬂectance distribution function (BRDF), as well as the instrument being
used to measure albedo. Ground-based albedo measurements, for example, are most
commonly acquired using an albedometer. Unlike a radiometer which measures
reﬂectance at a single view angle (e.g., nadir), an albedometer integrates the amount of
radiation reﬂected by a surface over all view angles. Thus, across non-Lambertian
surfaces, like the grass canopies measured in this study, reﬂectance and albedo can differ

signiﬁcantly depending on the BRDF (Liang et al. 2002).

MAPAR in Michigan grass stands

For range management decision-making, it would be ideal to use a single biomass
index all season long. In this study, when all of the Michigan stand types were grouped,
the season-long MAPAR-biomass relationship was signiﬁcant (Figure 4.5a). However,
the relationship appeared to saturate at MAPAR values near ~0.936. As a result, when

only the senescent period was considered, MAPAR was not signiﬁcantly correlated to

97

biomass (Figure 4.5a). This may indicate that the dynamic range of MAPAR is
insufﬁcient for late-season biomass estimates, either because of the amount of biomass
on the ground or because of the proportion of senescent biomass present.

However, it is also possible that the saturation phenomenon was an artifact of the
stem lodging event that occurred in the Avena and Avena-Bromus stands during week 10.
This conclusion is supported by the MAPAR-biomass relationship in the Bromus stands
as well as in the stands of Lolium multiﬂorum (Chapter 3), which were not impacted by
the stem lodging event. The radiation regime of a plant canopy is a ﬁinction of photon
scattering by leaves, stems, and soils (Jones 1992). The contribution of leaf, stem, and
soil optical properties to canopy PAR absorption is modulated by LAI, leaf angle
distribution, and foliage clumping (Asner et al. 1998), which determine the density and
optical depth of the canopy (Ross 1981). As grass stems lodge at the end of the season,
the density and optical depth of the canopy may be reduced (Ross 1981, Jones 1992,
Asner et al. 1998), thus decreasing the amount of PAR absorbed by the canopy
independently of biomass. In addition, stem lodging may cause openings in a grass
canopy, which in some instances could also decrease PAR absorption (Ross 1981 , Jones
1992, Asner et al. 1998) without parallel decreases in canopy LAI or biomass. In the
Michigan Bromus stands, the saturation phenomenon was reduced (Figure 4.5b) and
MAPAR was signiﬁcantly correlated to biomass during the senescent time period (Figure
4.5b; Table 4.1). Likewise, MAPAR was signiﬁcantly correlated to biomass during the
senescent time period in the Michigan Lolium stands (R2 = 0. 55, p < 0. 001). The

saturation phenomenon does highlight a potential limitation of MAPAR in grassland

98

ecosystems, where lodging is possible at the end of the season once vegetation has
senesced and stem integrity has declined.

A single species-speciﬁc biomass equation could be applied throughout the
season for Bromus (Figure 4.5b) and Lolium (y = 2E-06e20'40", R2 = 0. 91, p < 0. 001)
stands. Based on these results, I recommend using a single equation for the green and
senescent time periods assuming there is not a signiﬁcant amount of stem lodging.
Future efforts should be directed at testing the impact of canopy structural attributes on
the relationship between MAPAR and biomass throughout the season.

Like other reﬂectance-based indices, MAPAR measurements are inﬂuenced by
the BRDF, as a function of sun-surface-sensor geometry, canOpy architecture and optical
properties, soil background properties, and illumination conditions (Deering 1989).
While I tried to control for as many of these factors as possible in the Michigan stands, I
did not make any speciﬁc BRDF corrections. One important component of the BRDF
that was not controlled for was solar zenith angle (SZA). SZA increased in the plots
from ~20° in week 1 to 47° in week 15 (Figure 4.7). Studies designed to address BRDF
effects have shown that increases in SZA above 40° can cause reﬂectance-based indices,
like NDVI, as well as albedo to increase across vegetated surfaces (Qi et al. 1995).
NDVI increases because of reduced illumination and increased shadowing of the soil
background, and increased illumination of the vegetated surface (Danaher 2002).
Because MAPAR is broadly inversely related to reﬂectance-based indices like NDVI as
well as to albedo, it is possible that during weeks 13—15, when SZA was greater than 40°
(Figure 4.7), MAPAR values could have decreased independently of the decreases in

stand biomass and LAI (Figure 4.4). However, the results from this study suggest that

99

the SZA effect was most likely secondary to the relationship between MAPAR and
biomass. For example, in the Bromus stands, MAPAR was signiﬁcantly correlated to
biomass, as well as fAPAR and LAI, within individual weeks when SZA was constant
and therefore not impacting the predictive capacity of MAPAR (Table 4.1 ).

Because MAPAR is an index of surface absorbance, it is important to understand
how changes in the soil background inﬂuence the reliability of MAPAR. In general, dry
soils are usually more reﬂective than vegetation in the visible regions and less so in the
near infrared regions (Bowers and Hanks 1965). However, as moisture and organic
matter content increase and the color of the soil darkens, the amount of radiation
absorbed by soil increases and the differences in reﬂectance between soil and vegetation
decrease (Bowers and Hanks 1965) (Table 4.2). As a consequence, MAPAR-based bare
soil values increased, and the ability to discriminate vegetation decreased, when the light-
colored sandy loam soils used in the Michigan grass stands were wet. In addition,
MAPAR was not as effective for biomass estimates on dark-colored soils high in organic
matter content, such as Houghton Muck (80%) (Table 4.2). The soil background effect
would likely decrease as vegetation cover and LAI increase, and the canopy closes
(Hoffer 1978). Before using MAPAR, a manager should assess soil conditions across his
or her property. If both fractional soil cover on average is greater than 15% (canopy
cover is <85%) and the soil is either wet or has high organic matter content (Table 4.2), I

would not recommend using MAPAR for biomass estimates.

MAPAR in California annual grasslands

100

MAPAR derived from UniSpec measurements was signiﬁcantly correlated to
LAI, biomass, and fAPAR (Figure 4.8) at maximum greenness in California grassland
plots with LAI values from <0.2 to 6. These results, together with those from the
Michigan stands, suggest that across these soil types and conditions the dynamic range of
MAPAR may be sufﬁcient to predict a large range of fAPAR, LAI, and biomass values
that are naturally present during the period of maximum greenness.

Although it was outside the scope of this study to test MAPAR at the landscape-
scale throughout the entire season, I found that MAPAR measurements from l-m2 plots
were signiﬁcantly correlated to those from 900-m2 Landsat pixels. Combined with the
encouraging results from the Michigan grass stands and California annual grassland plots,
these results support testing MAPAR at the landscape scale using Landsat or MODIS
satellite data. Such tests would initially focus on identifying how sub-pixel level changes
in soil cover, type, and condition as well as vegetation condition (e. g., stem lodging)

impact the broad-scale applicability of MAPAR for biomass estimates.

MAPAR for RDM estimates

An important management application of the MAPAR index is landscape-scale
RDM estimates. Currently methods used by range managers are ground-based (e. g.,
Clawson et al. 1982) and time intensive, and have varying accuracy (Bartolome et al.
2002). While remote sensing-based RDM approaches have been developed for cr0p
systems (Daughtry 2001, Streck et al. 2002, Nagler et al. 2003), these approaches rely on
instruments and datasets to which private range managers usually do not have access. A

more accessible remote sensing-based RDM approach, which could be applied to Landsat

101

or MODIS data, for example, would likely increase the number of managers who use
remote sensing to monitor RDM levels. I did not speciﬁcally test the ability of MAPAR
to make RDM estimates in grassland ecosystems. However, MAPAR successfully
estimated biomass where conditions were like those a manager may encounter during the
period when RDM estimates are made: 1) senescent vegetation cover is dominant, 2) the
soil background and vegetation are dry, and 3) biomass values range from 900—2350
kg/ha (Figures 4.5 and 4.6; Table 4.1) (Bartolome et al. 2002). Because of the sensitivity
of MAPAR to soil background conditions, it may be difﬁcult to apply MAPAR in
situations where RDM is low (e. g., on ﬂat slopes and swales, <450 kg/ha), which is an
important issue to monitor in the future.

MAPAR, as calculated in this study, had two main limitations: the utility of
MAPAR for biomass estimates decreased as the soil darkened (Table 4.2) and when
grasses lodged (Figure 4.5a). There may be alternative MAPAR-based approaches that
decrease some of these soil background and vegetation condition limitations (Table 4.4).
Two possible approaches to test in the future are: 1) use MAPAR to quantify biomass at
maximum biomass and then estimate RDM using the ﬁeld-based guidelines developed by
Bartolome et al. (2002); and 2) use MAPAR for interannual RDM comparisons (i.e.,
RDM change detection). A change detection approach is promising because it would
provide the user the means to minimize the inﬂuence of the soil background (type,
condition) in the retrieval of biomass (and RDM) data (Mas 1999, Jensen 2000). In its
simplest form, change detection involves subtracting values (e. g., biomass) of one
satellite image from a second image that has been precisely registered to the ﬁrst (Singh

1989). Because images would be acquired at the same time of the year, soil background

102

 

 

 

 

88080880008 800.000 >880p8uw089 800» 008808 8080880088 800» 8000808 0_0>0_ 80008.?
20M 000.88 0 002 880 8880 0.00: 8003808 80:88 0808680018 08086800 80880m0> 20M :8 8 0008080886 80020025
:00 80.8 08800008 08028800 mom 08—030 05 80—80—00 0 8080
000805 0008808 8000800
888808 800.80 8030 000.800 M8880. 80% [>02 08uh8m308h8ﬁ08w 02 08808 383 ._0 00 080—080m
.20 08808 008800400 ”8080.86 M8808 00080 880—0008» :83: E080w0808h88080 8 8888030 06 3.08
. . . 80003.?
3808080008 2nd 8880 0888808 £08 3882 802.0880 02 038080808. Ea 008800 805 00.. 8
000.88 0 002 ".xcﬂv 8000 0. 008 800 ”000805 fammA 80>00 808000w0> 8028800 808000M0> 000803 888888 a :0 _
:00 08083 08080880 888808 00 808 080800 80808 80>00 808000w0> 00 000805 @8080
00 80:8: 3002 8002 000805 380223 08090 8 30— 880 .08 0=0m 8086800 mom
800E890 008 "8 8882 8080880 000080 880—0008w :02“? 0380800808.
8000 880 o\ m 8 v 8000 000803 80000800 8880— 080083820 02 8086800 808000w0> 080.. 808.“ 05 080.809 =0.« 8000897..
.00 080 0N 080. a 0380 0 .2880 800 fonwA 80>00 80830w0> . ..8>00 80m 0 0w0 .0 8 0A.: 8080 00.:
0_. 0 888 a 0 pm”. ”9808080008 80800 80808 80 _ “080 _0> 5 . End .0 a .n
:0 8. ._ _ 0:: E 20.x 00085 08090 8 30— 880 .00 mzom 8.8 _. m
~008§0>800t0 0.
00u0 80> 00. 0000 80> ~»_0>800=0 80 80 8 0.00000 0 008 . 80.: .000 .30
0 8 .n . 8< 00080—80335308055 8 u :0 00. 0. 8 E

809808 0 0008 0055

 

.0w808 80880880008 05 .«0 02380 32080.8 080 m_0>0_ End 0083 00.88008 @8802 08 80m:
0a 800 800008 85 050088 Sam :00 @8803 808 0008 8000850 80800809 08080 08,—. 858030 End 388 A0 00 080—080m
05 585 050mg 000805 888808 0 maﬁa 080— 8 8088000 808 w808 29M 20% 00808800 80008900 «00:85 05. .0808 :8

08¢ 05 00.808 80.88000 0008 M868 3000.88 2mm :8 0058080 8000.890. 00085 05. 0080008880 Sam 80009m<m<2 .0... 030,—.

103

conditions would likely be the same in both images. Thus, “change” would be due to
changes in biomass (or RDM) rather than soil cover, type, or condition (Hallum 1993).
Both of the RDM approaches detailed in Table 4.4 have certain advantages and
disadvantages. However, the ﬁndings from this study are sufﬁciently encouraging to

suggest that testing these approaches across grassland ecosystems would be worthwhile.

104

CHAPTER 5

CONCLUSIONS

Although my dissertation research has answered several questions about the use
of remote sensing in rangeland ecology, it has also led to many more. Having identiﬁed
factors that inﬂuence the use of remote sensing data by private range managers and
examined ways in which remote sensing use impacts manager decision-making, I am
now interested in investigating in more detail the economic impact of remote sensing
approaches on a diverse group of ranching operations. It would be valuable, for example,
to better understand whether there is a threshold sum that ranchers are willing to spend on
remote sensing data, as a function of their ranching operations’ size and diversity (as
measured in terms of acreage, head of cattle, diversity of land uses, etc.), as well as to
continue to adapt the development and delivery of remote sensing data to the unique
needs of individual range managers.

Although quantifying the economic impact of remote sensing data on ranching
operations may be challenging, such an assessment would be invaluable to managers
interested in using remote sensing data. For example, an economic assessment could
identify particular situations (e. g., a large ranch with many livestock and a widespread
weed infestation) in which investment in remote sensing data was likely to be
worthwhile. Alternatively, an economic analysis might conclude that remote sensing is
generally a cost-effective management tool in which land managers should more broadly
invest, but that its use is limited by availability and training. Without a full economic
analysis of the impact of remote sensing on ranch proﬁtability, it is likely that few range

managers will want to make long-term commitments to investing in remote sensing data.

105

 

My dissertation research examined a few of many possible applications of remote
sensing data for biomass and weed management in grassland ecosystems (e.g., Tueller
1989, Hunt et al. 2003). It is clear that remote sensing data can provide managers means
to evaluate management decisions across a variety of spatial and temporal scales and to
focus management practices on those which most efﬁciently increase rangeland health,
primarily by decreasing noxious weeds and increasing levels of desirable forage species.
It is less clear however, which remote sensing applications are most useful for private
range managers and therefore likely to be invested in by the managers themselves. My
study was the ﬁrst to speciﬁcally address the use of remote sensing data by private range
managers. Future efforts should be directed at evaluating similar questions of use and
impact across a broad and diverse group of range managers and ranching operations.
Such work would help answer questions concerning, for example, the importance of the
size of the ranching operation on the use and investment in remote sensing data. In
addition, as remote sensing data continue to be reﬁned and new applications are
developed which allow managers to evaluate measures of rangeland health at ﬁner spatial
and temporal scales (e.g., Lass et al. 2005, Mustafa et al. 2005, Everitt et a1. 2006, Mundt
et al. 2006), it is imperative that we evaluate how to best deliver these data to managers
and train managers in how to apply them to their individual management situations.

The range managers in California with whom I worked were interested in
investing in remote sensing data because they believed these data could help them
optimize their ranching operations and increase the protection of their ranches’ natural
resources (Butterﬁeld and Malmstrom 2006). However, this group of managers, unlike

most range managers, enjoyed free access to remote sensing data and training in its use

106

 

for ﬁve years. Without this support, it is likely that this manager group would have been
less interested in investing in remote sensing data. There are a number of reasons why
this may be the case, which should be investigated in the future, including: (i) the
perceived costs to individual managers and uncertainties about how to obtain appropriate

remote sensing products, and (ii) concerns that remote sensing data could be used by

_ _ I

government agencies or conservation organizations to regulate private grazing and/or
land management operations in an authoritarian manner (S. Butterﬁeld, personal

communication). Collaborations like the ones discussed in Butterﬁeld and Malmstrom

 

(2006) can help alleviate some of these difﬁculties by providing a forum in which l
scientists, managers, agency staff, and non-proﬁt conservation organizations can develop
cost-sharing collectives (e. g., for purchasing and processing remote sensing data) and
products tailored to different, often unique management situations. These collaborations
can also provide opportunities for exploring and addressing concerns managers may have
about the use of remote sensing data as a regulatory tool across their properties.

Such collaborations appear to be powerful means through which to develop
remote sensing products that are accurate, affordable, and easy to use in operational range
management contexts. Because of the pressures currently associated with maintaining
proﬁtable range management operations in the western United States (Roche and Roche
1991, Leitch et al. 1994, World Resources Institute 1996, Mitchell 2000, O’Brien et al.
2003) and because private lands often contain high-quality ecosystems and species of
conservation interest (e. g., The Nature Conservancy 2000, 2005, 2006), collaborations
like those described in Butterﬁeld and Malmstrom (2006) are likely to become more

common in the future.

107

My dissertation research also examined the properties of a new vegetation index,
MAPAR, which could be used under some conditions to quantify senescent biomass.
Future efforts will be focused on testing the applicability of MAPAR (and of associated
remote sensing-based approaches) at the landscape-scale and on identifying how sub-
pixel level changes (in soil cover, type, and condition as well as vegetation condition)
impact the use of MAPAR for senescent biomass estimates. An important application of
MAPAR to the range management and conservation communities is its use as an
indicator of residual dry matter (RDM). RDM is particularly interesting to conservation
organizations and their partners (e. g., The Bureau of Land Management 1996, Guenther
and Christian 2005, The Nature Conservancy 2006) because it can strongly inﬂuence
biomass production and noxious weed spread. Although I did not speciﬁcally test the
ability of MAPAR to make RDM estimates as part of my dissertation work, I did ﬁnd
that MAPAR successfully estimated biomass under conditions like those under which a
range manager would normally estimate RDM (i.e., senescent vegetation cover is
dominant, the soil background and vegetation are dry, and biomass values range from
900—2350 kg/ha). Because MAPAR is sensitive to changes in soil background
conditions, the application of MAPAR for RDM estimates is likely limited to situations
where ﬁ'actional soil cover <15% or the soil is both dry and low in organic matter
content. Future efforts should be directed at testing the limits of MAPAR across a wide
range of grassland conditions using ﬁne and moderate resolution satellite data, such as

that from Landsat TM and MODIS.

108

BIBLIOGRAPHY

Aber, J. D., and J. M. Melillo. 2001. Terrestrial ecosystems. Harcourt Academic Press:
New York, NY. 556 p.

Alcom, J. B. 1989. Process as resource. Advances in Economic Botany 7:63-77.

Alward, R. D., J. K. Detling, and D. G. Milchunas. 1999. Grassland changes and
nocturnal global warming. Science 283:229-231.

Asner, G. P., and K. B. Heidebrecht. 2002. Spectral unmixing of vegetation, soil and dry
carbon cover in arid regions: comparing multispectral and hyperspectral
observations. International Journal of Remote Sensing 23 :3939-3958.

Asner, G. P., and C. A. Wessman. 1997. Scaling PAR absorption from the leaf to
landscape level in spatially heterogeneous ecosystems. Ecological Modelling
103:81-97.

Asner, G. P., C. A. Wessman, and S. Archer. 1998. Scale dependence of absorption of
photosynthetically active radiation in terrestrial ecosystems. Ecological
Applications 8: 1003-1021 .

Asner, G. P., C. A. Wessman, C. A. Bateson, and J. L. Privette. 2000. Impact of tissue,
canopy, and landscape factors on the hyperspectral reﬂectance variability of arid
ecosystems. Remote Sensing of Environment 74:69-84.

Asrar, G., M. Fuchs, E. T. Kanemasu, and J. L. Hatﬁeld. 1984. Estimating absorbed
photosynthetic radiation and leaf area index from spectral reﬂectance in wheat.
Agronomy Journal 76:300-306.

Asrar, G., E. T. Kanemasu, R. D. Jackson, and J. P. J. Pinter. 1985. Estimation of total
above-ground phytomass production using remotely sensed data. Remote Sensing
of Environment 17:21 1-220.

Barbour, M. G., J. H. Burk, W. D. Pitts, F. S. Gilliam, M. W. Schwartz. 1999. Terrestrial
plant ecology. Addison Wesley Longrnan: New York, NY. 649 p.

Bartolome, J. W., M. C. Stroud, H. F. Heady. 1980. Inﬂuences of natural mulch on forage
production at differing California annual range sites. Journal of Range
Management 3314-8.

Bartolome, J ., W. E. Frost, N. K. McDougald, and M. Connor. 2002. California
guidelines for residual dry matter (RDM) management on coastal and foothill

109

annual rangelands. University of California, Division of Agriculture and Natural
Resources. Publication 8092. 8 p.

Barry, S. 1995. Cattle fecal fax. University of California Cooperative Extension. October
Report, Nov. 7, 1995.

Bentley, J. R., and M. W. Talbot. 1951. Efficient use of annual plants on cattle ranges in
the California foothills. USDA Circular 870. 52 p.

Berkes, F., J. Colding, and C. Folke. 2000. Rediscovery of traditional ecological F
knowledge as adaptive management. Ecological Applications 10:1251-1262.

Berkes, F., C. Folke, and M. Gadgil. 1995. Traditional ecological knowledge,
biodiversity, resilience, and sustainability. In: C. Perrings, K. G. Maler, C. Folke, _
C. S. Hollings, and B. O. Jansson [EDS.]. Biodiversity conservation. Kluwer
Academic Press, Dordrecht, The Netherlands. p. 281-299. i

 

Boswell, E. 2000. Guarding against invaders: remote sensing on the range. Rangelands
22: 14-17.

Bovey, R. W., D. Le Toumeau, and L. C. Erickson. 1961. The chemical composition of
medusahead and downy brome. Weeds 9:307-311.

Bowers, S. A., and R. J. Hanks. 1965. Reﬂection of radiant energy from soils. Soil
Science 100:130—138.

Bremer, D. J ., L. M. Auen, J. M. Ham, and C. E. Owensby. 2001. Evapotranspiration in a
prairie ecosystem: effects of grazing by cattle. Agronomy Journal 93:338-348.

Butterﬁeld, H. S., and C. M. Malmstrr'im. 2006. Experimental use of remote sensing by
private range managers and its inﬂuence on management decisions. In press
Rangeland Ecology and Management.

Callihan, R. H., L. W. Lass, C. W. Hunt, and G. Pritchard. 1995. Comparative
digestibility of yellow starthistle (Centaurea solstitialis L.). Western Society of
Weed Science, Research Progress Report. 36 p.

Callihan, R. H., R. L. Sheley, and D. C. Thill. 1982. Yellow starthistle: identiﬁcation and
control. University of Idaho, College of Agriculture: Current Information Series
No. 634. 4 p.

Choudhury, B. J. 1987. Relationships between vegetation indices, radiation absorption,
and net photosynthesis evaluated by a sensitivity analysis. Remote Sensing of
Environment 22:209-233 .

110

Clawson, W. J ., N. K. McDougald, and D. A. Duncan. 1982. Guidelines for residue
management on annual range. University of California, Division of Agriculture
and Natural Resources. Leaﬂet 21327. 3p.

Colding, J. 1998. Analysis of hunting options by the use of general food taboos.
Ecological Modelling l 10:5-17.

Coppock, D. L., and A. H. Birkenfeld. 1999. Use of livestock and range management
practices in Utah. Journal of Range Management 52:7-18.

Crarnpton, B. 1974. Grasses in California. University of California Press, Berkeley, CA.
178 p.

Daberkow, S. G., and W. D. McBride. 2000. Adoption of precision agriculture
technologies by US. farmers. In: P.C. Robert, R.H. Rust, and WE. Larson
(EDS.), Proc. 5th International Conference on Precision Agriculture. Madison,
WI: America Society of Agronomy.

Daberkow, S. G., and W. D. McBride. 2003. Farm and operator characteristics affecting
the awareness and adoption of precision agriculture technologies in the US.
Precision Agriculture 4: 163-177.

D'Antonio, C. M., and P. M. Vitousek. 1992. Biological invasions by exotic grasses, the
grass/ ﬁre cycle, and global change. Annual Review of Ecology and Systematics
23:63-87

Danaher, T. J. 2002. An empirical BRDF correction for Landsat TM and ETM+ imagery.
In: Proc. 11th Australian Remote Sensing and Photogrammetry Conference.
Brisbane, Australia. 12 p.

Daubenmire, R. F. 1968. Plant communities: A textbook of plant synecology. Harper &
Row, New York, NY.

Daughtry, C. S. T. 2001. Discriminating crop residues from soil by shortwave infrared
reﬂectance. Agronomy Journal. 93: 125-1 3 l .

Deering, D. W. 1989. Field measurements of bidirectional reﬂectance. In: G. Asrar [ED].
Theory and applications of optical remote sensing. New York, NY: John Wiley
and Sons. p. 14-65.

Decagon Devices. 2001. AccuPAR Model PAR-80 Operator’s Manual, Version 3.4.
Decagon Devices, Inc., Pullman, WA.

Demetriades-Shah, T. H., E. T. Kanemasu, and I. D. Flitcroft. 1992. Comparison of
ground- and satellite-based measurements of the fraction of photosynthetically

111

active radiation intercepted by tallgrass prairie. Journal of Geophysical Research
97: 1 8947-1 8950.

DiTomaso, J. M. 2000. Invasive weeds in rangelands: species, impacts, and management.
Weed Science 48:255-265.

DiTomaso, J. M., and D. W. Johnson [EDS.]. 2006. The use of fire as a tool for
controlling invasive plants. Cal-IPC Publication 2006-01. California Invasive
Plant Council, Berkeley, CA. 56 p.

Dillman, D. A. 1978. Mail and telephone surveys: the total design method. Wiley &
Sons, New York, NY. 375 p.

Everitt, J. H., G. L. Anderson, D. E. Escobar, M. R. Davis, N. R. Spencer, and R. J.
Andrascik. 1995. Use of remote sensing for detecting and mapping leaf spurge
(Euphorbia esula). Weed Technology 91599-609.

Everitt, J. H., C. Yang, R. S. Fletcher, and D. L. Drawe. 2006. Evaluation of high-
resolution satellite imagery for assessing rangeland resources in south Texas.
Rangeland Ecology and Management 59:30-37.

Fenn, M. E., M. A. Poth, J. D. Aber, J. S. Baron, B. T. Bormann, D. W. Johnson, A. D.
Lemly, S. G. McNulty, D. F. Ryan, and R. Stottlemyer. 1998. Nitrogen excess in
North America ecosystems: predisposing factors, ecosystem responses, and
management strategies. Ecological Applications 8:706-733.

Fensholt, R., I. Sandholt, and M. S. Rasmussen. 2004. Evaluation of MODIS LAI,
fAPAR and the relation between fAPAR and NDVI in a semi-arid environment
using in situ measurements. Remote Sensing of Environment 91:490-507.

Field, C. B., J. T. Randerson, and C. M. Malmstrom. 1995. Global net primary
production: combining ecology and remote sensing. Remote Sensing of
Environment 51:74-88.

Fliegel, F. C. 1993. Diﬂ’usion research in rural sociology. Greenwood Press, Westport,
CT. 138 p.

Ford, L. D., L. Huntsinger, and G. Hayes. 2006. Rangeland health indicators project and
rangeland health indicator monitoring system. Central Coast Rangeland
Coalition, San Francisco, CA. 34 p.

Frank, A. B., and J. K. Aase. 1994. Residue effects of radiometric reﬂectance

measurements of northern Great Plains rangelands. Remote Sensing of
Environment 49: 195-199.

112

 

French, A. N., T. J. Schmugge, and W. P. Kustas. 2000. Discrimination of senescent
vegetation using thermal emissivity contrast. Remote Sensing of Environment
74:249-254.

Gaborcik, N. 2003. Relationship between contents of chlorophyll (a + b) (SPAD values)
and nitrogen of some temperate grasses. Photosynthetica 41 :285-287.

Gadgil, M., F. Berkes, and C. Folke. 1993. Indigenous knowledge for biodiversity
conservation. Ambio 22:151-156.

Gamon, J. A., C. B. Field, M. L. Goulden, K. L. Grifﬁn, A. E. Hartley, G. Joel, J.
Penuelas, and R. Valentini. 1995. Relationships between NDVI, canopy structure,
and photosynthesis in three Californian vegetation types. Ecological Applications
5:28-41.

Gamon, J. A., C. B. Field, D. A. Roberts, S. L. Ustin, and R. Valentini. 1993. Functional
patterns in an annual grassland during an AVIRIS overﬂight. Remote Sensing of
Environment 44:239-253.

Gamon, J. A., and J. S. Surfus. 1999. Assessing leaf pigment content and activity with a
reﬂectometer. New Phytologist 143: 105-1 17.

George, M. R., J. Bartolome, N. K. McDougald, M. Connor, C. Vaughn, and G.
Markegrad. 2001. Annual range forage production. University of California,
Division of Agriculture and Natural Resources. Publication 8018. 9 p.

George, M. R., and K. O. Fulgham. 1989. Grazing systems and management of
mediterranean-type plant communities. In: Proc. XVI International Grasslands
Congress: Man and the Biosphere Symposium on Landscape Ecology. Nice,
France. p. 78-94.

George, M. R., and J. W. Menke. 1996. Soil surface management of annual rangelands.
Rangeland watershed program-fact sheet No. 30. Agronomy and Range Science
Department, University of California Davis. 5 p.

George, M. R., S. J. Barry, S. R. Larson, N. K. McDougald, T. A. Ward, J. M. Harper, D.
M. Dudley, R. S. Ingram, and E. A. Laca. 2006. Comparison of comparative yield
and stubble height for estimating herbage standing crop in annual rangelands.
Rangeland Ecology and Management 59:43 8-441 .

Gholtz, H. L., S. A. Vogel, W. P. Cropper, K. McKelvey, K. C. Ewel, R. O. Teskey, and
P. J. Curran. 1991. Dynamics of canopy structure and light interception in Pinus
elliottii stands, North Florida. Ecological Monographs 61 :33-51.

Goudriaan, J. 1977. Crop micrometeorology: a simulation study. Wageningen Pudoc. 249
p.

113

Guenther, K. S. 1998. Residual dry matter (RDM) monitoring photo guide. Keith
Guenther, Wildland Solutions. http://www.wildlandsolutions.com/.

Guenther, K. S., and C. Christian. 2005. Rangeland easement monitoring workshop. The
Nature Conservancy. Monterey, CA. 47 p.

Hallum, C. 1993. A change detection strategy for monitoring vegetative and land-use
cover types using remotely-sensed satellite data. Remote Sensing of Environment
43: 171-177.

Hanselka, C. W., A. McGinty, B. S. Rector, R. C. Rowan, and L. D. White. 1990.
Grazing and brush management on Texas rangelands: an analysis of management
decisions. College Station, TX: Texas Agricultural Extension Service Technical
Report. 22 p.

Harris, T. R., A. Bhattacharyya, W. G. Kvasnicka, and G. M. Veserat. 1995. What type of
rancher looks for new technology? Rangelands 17:217-221.

Harris, N. R., W. E. Frost, N. K. McDougald, M. R. George, and D. L. Nielsen. 2002.
Long-term residual dry matter mapping for monitoring California hardwood
rangelands. USDA Forest Service General Technical Report. PSW-GTR-184. p.
87-96.

Hatﬁeld, J. L. 1983. Remote sensing estimators of potential and actual crop yield. Remote
Sensing of Environment 13 :301-31 1.

Hatﬁeld, J. L., G. Asrar, and E. T. Kanemasu. 1984. Intercepted photosynthetically active
radiation estimated by spectral reﬂectance. Remote Sensing of Environment
14:65-75.

Haver, S. S. 2001. A case study in cooperation: one range management professional
reafﬁrms that cooperation works. Rangelands 23:16-17.

Heady, H. P., and D. Child. 1994. Rangeland Ecology and Management. Westview Press,
Boulder, CO. 519 p.

Heady, H. F. 1956. Changes in a California annual plant community induced by
manipulation of natural mulch. Ecology 37:798-812.

Heady, H. F. 1977. Valley grassland. In: M. G. Barbour and J. Major [EDS.]. Terrestrial
vegetation of California. Wiley & Sons, New York, NY. p. 491-514.

Hedrick, D. W. 1948. The mulch layer of California annual ranges. Journal of Range
Management 1:22-25.

114

 

Hickman, J. C. 1993. The Jepson Manual: Higher Plants of California. University of
California Press, Berkeley, CA. 1424 p.

Hoare, D., and P. Frost. 2004. Phenological description of natural vegetation in southern
Africa using remotely-sensed vegetation data. Applied Vegetation Science 7: 19-
28.

Hoffer, R. M. 1978. Biological and physical considerations in applying computer-aided
analysis techniques to remote sensor data. In: P. H. Swain and S. M. Davis
[EDS.]. Remote Sensing: The Quantitative Approach. McGraw-Hill, New York,
NY. p. 227-289.

Huete, A. R. 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of
Environment 25:295-309.

Huete, A. R., and R. D. Jackson. 1987. Suitability of spectral indices for evaluating
vegetation characteristics on arid rangelands. Remote Sensing of Environment
23:213-232.

Huete, A. R., R. D. Jackson, and D. F. Post. 1985. Spectral response of a plant canopy
with different soil backgrounds. Remote Sensing of Environment 17:37-53.

Huete, A. R., C. Justice, and W. van Leeuwen. 1999. MODIS vegetation index (MOD I3)
algorithm theoretical basis document. NASA Goddard Space Flight Center,
Greenbelt, MD 20771, USA. 129 p.

Hull, J. C., and H. A. Mooney. 1990. Effects of nitrogen on photosynthesis and grth
rates of four California annual grasses. Acta-Oecologia 11:453-468

Hunt, E. R., Jr., J. H. Everitt, J. C. Ritchie, M. S. Moran, D. T. Booth, G. L. Anderson, P.
E. Clark, and M. S. Seyﬁied. 2003. Applications and research using remote
sensing for rangeland management. Photogrammetric Engineering & Remote

Sensing 69:675-693.

Idso, S. B., R. D. Jackson, R. J. Reginato, B. A. Kimball, and F. S. Nakayama. 1975. The
dependence of bare soil albedo on soil water content. Journal of Applied
Meteorology 14:109-113.

Jensen, M. N. 2001. Can cows and conservation mix? BioScience 51:85-90.

Jensen, J. R. 2000. Remote sensing of environment: an earth resource perspective.
Prentice Hall, Upper Saddle River, NJ. 544 p.

Jones, H. G. 1992. Plants and microclimate: a quantitative approach to plant physiology,
2nd edition. Cambridge University Press, New York, NY. 428 p.

115

J orgensen, U., J. Mortensen, and C. Ohlsson. 2003. Light interception and dry matter
conversion efﬁciency of miscanthus genotypes estimated from spectral
reﬂectance measurements. New Phytologist 157:263-270.

Kauth, R. J ., and G. S. Thomas. 1976. The tasseled-cap - a graphic description of the
spectral-temporal development of agricultural crops as seen by Landsat.

Proceedings of the Symposium on Machine Processing of Remotely Sensed Data
41 -5 1 .

Knyazikhin, Y., J. Glassy, J. L. Privette, Y. Tian, A. Lotsch, Y. Zhang, Y. Wang, J. T.
Morisette, P. Votava, R. B. Myneni, R. R. Nemani, and S. W. Running. 1999.
MODIS leaf area index (LAI) and fraction of photosynthetically active radiation
absorbed by vegetation (FPAR) product (MODI5) algorithm, Theoretical Basis
Document. NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA.

Koide, R. T., L. F. Huenneke, S. P. Hamburg, and H. A. Mooney. 1988. Effects of
applications of fungicide, phosphorus and nitrogen on the structure and

productivity of an annual serpentine plant community. Functional Ecology 2:335-
344.

Kreuter, U. P., H. E. Amestoy, D. N. Ueckert, and W. A. McGinty. 2001. Adoption of
brush busters: results of Texas county extension survey. Journal of Range
Management 54:630-639.

Lass, L. W., T. S. Prather, N. F. Glenn, K. T. Weber, J. T. Mundt, and J. Pettingill. 2005.
A review of remote sensing of invasive weeds and example of early detection of
spotted knapweed (Centaurea maculosa) and babysbreath (Gypsophila
paniculata) with a hyperspectral sensor. Weed Science 53:242-251.

Lass, L. W., D. C. Thill, B. Shaﬁi, and T. S. Prather. 2002. Detecting spotted knapweed
(Centaurea maculosa) with hyperspectral remote sensing technology. Weed
Technology 16:426-432.

Lass, L. W., H. W. Carson, and R. H. Callihan. 1996. Detection of yellow starthistle
(Centaurea solstitialis) and common St. J ohnswort (Hypericum perforatum) with
multispectral digital imagery. Weed T echnology 10:466-474.

Lauver, C. L. 1995. Integrating remote sensing and vegetation analysis of shortgrass
rangeland for conservation planning in the high plains of western Kansas. In: 5th
International Rangeland Congress. Salt Lake City, UT. p. 306-307.

Le Roux, X. L., H. Gauthier, A. Begue, and H. Sinoquet. 1997. Radiation absorption and

use by humid savanna grassland: assessment using remote sensing and modeling.
Agricultural and Forest Meteorology 85:1 17-132.

116

Leitch, J. A., F. Leistritz, F. Larry, and D. A. Bangsund. 1994. Economic effect of leajy
spurge in the upper Great Plains: methods, models, and results. Agricultural
Economics Report No. 316. Fargo, ND. North Dakota State University,
Department of Agricultural Economics.

Liang, S., C. J. Shuey, A. L. Russ, H. Fang, M. Chen, C. Walthall, C. S. T. Daughtry, and
E. R. Hunt, Jr.. 2002. Narrowband to broadband conversion of land surface
albedo: II. Validation. Remote Sensing of Environment 84:25-41.

Los, S. 0., G. J. Collatz, P. J. Sellers, C. M. Malmstrbm, N. H. Pollack, R. S. DeFries, L.
Bounoua, M. T. Parris, C. J. Tucker, and D. A. Dazlich. 2000. A global 9-yr
biophysical land surface dataset from NOAA AVHRR Data. Journal of
Hydrometeorology 1 : 1 83-199.

Lowe, P. N., W. K. Lauenroth, and I. C. Burke. 2002. Journal of Range Management
55:94-98.

Lucht, W., C. B. Schaaf, A. H. Strahler, J. P. Muller, B. Hu, P. Lewis, Y. Liu, A. H.
Hyman, X. Li, M. J. Bamsley, N. C. Stugnell, and R. P. d'Entremont. 1998. At-
launch status of the MODIS BRDF/albedo algorithm: Implementation, AVHRR-
based prototyping and future plans. In: IGARSS 1998. Seattle, WA. p. 1289-
1291.

Lusk, W. C., M. B. Jones, D. T. Torell, and C. M. McKell. 1961. Medusahead
palatability. Journal of Range Management 14:248-251.

Lym, R. G., and D. R. Kerby. 1987. Cattle foraging behavior in leafy spurge (Euphorbia
esula) infested rangeland. Weed Technology 1:314-318.

Mack, R. N. 1989. Temperate grasslands vulnerable to plant invasions: characteristics
and consequences. In: Drake, J. A., H. A. Mooney, F. di Castri, R. H. Groves, F.
J. Kruger, M. Rejmanek, and M. Williamson [EDS.]. Biological invasions: a
global perspective. Wiley & Sons. New York, NY: p. 155-179.

Maddox, D. M., and A. Mayﬁeld. 1985. Yellow starthistle infestations are on the
increase. California Agriculture 39:10-12.

Major, D. J ., F. Baret, and G. Guyot. 1990. A ratio vegetation index adjusted for soil
brightness. International Journal of Remote Sensing 11:727-740.

Malmstriim, C. M. and H. S. Butterﬁeld with C. P. Barber, S. Bradenburg, C. C. Hughes,
J. Qi, T. Robinson, A. Schrotenboer, C. Stoner, P. Vamakovida, and T. Yoshida.
2004. Using remote sensing to assess forage dynamics in a California rangeland
restoration program. Project 01-N31 Final Report. Sacramento, CA: CALFED
Ecosystem Restoration Program. 79 p.

117

Mas, J. F. 1999. Monitoring land-cover change: a comparison of change detection
techniques. International Journal of Remote Sensing 20:139-152.

Maselli, F., M. A. Gilabert, and C. Conese. 1998. Integration of high and low resolution
NDVI data for monitoring vegetation in Mediterranean environments. Remote
Sensing of Environment 63:208-218.

Maurer, J ., A. F. H. Goetz, K. Steffen, and C. A. Wessman. 2002. Retrieval of surface
albedo from space. In: J. Maurer [ED.]. University of Colorado at Boulder,
Boulder, CO. http://cires.colorado.edu/~mauerj/albedo/albedo.htm. 13 p.

McDougald, N. K., W. J. Clawson, J. W. Bartolome, and W. E. Frost. 1991. Estimating
livestock grazing capacity on California annual rangeland. U.C. Davis Range
Science Report 29. 2 p.

McDougald, N. K., W. J. Clawson, and D. A. Duncan. 1982. Guidelines for residual dry
matter in California annual rangelands. University of California Cooperative
Extension, Division of Agriculture and Natural Resources Leaﬂet 2344. 4 p.

McLendon, T. and E. F. Redente. 1991. Nitrogen and phosphorus effects on secondary
succession dynamics on a semi-arid sagebrush site. Ecology 72:2016-2024.

McNaim, H., and R. Protz. 1993. Mapping corn residue cover on agricultural ﬁelds in
Oxford County, Ontario, using Thematic Mapper. Canadian Journal of Remote
Sensing. 19:152-159.

Miles, M. B., and A. M. Huberman. 1994. Qualitative data analysis: an expanded
sourcebook, 2nd Edition. Sage Publications, Thousand Oaks, CA. 337 p.

Mitchell, J. E. 2000. Rangeland resource trends in the United States: A technical
document supporting the 2000 USDA Forest Service RPA Assessment. RMRS-
GTR-68. US. Department of Agriculture, Forest Service, Rocky Mountain
Research Station. 84 p.

Molinar, R. F., D. Galt, and J. Holechek. 2001. Managing for mulch. Rangelands 23:3-7.

Monteith, J. L. 1977. Climate and the efﬁciency of crop production in Britain.
Philosophical Transactions of the Royal Society of London Series B 281 :277-294.

Mooney, H. A., S. P. Hamburg, and J. A. Drake. 1986. The invasions of plants and
animals into California. In: Mooney, H. A. and J. A. Drake [EDS.]. Ecology of

biological invasions of North America and Hawaii. Springer Verlag: New York,
NY: p. 250-272.

Morrison, J. E., Jr., C. Huang, D. T. Lightle, and C. S. T. Daughtry. 1993. Residue cover
measurement techniques. Journal of Soil & Water Conservation. 48:479-483.

118

Mundt, J. T., D. Streutker, and N. F. Glenn. 2006. Mapping sagebrush distribution via
fusion of hyperspectral and LiDAR classiﬁcation. Photogrammetric Engineering
& Remote Sensing 72:47-54.

Mustafa, M., J. E. Norland, R. L. Crabtree, and M. E. Biondini. 2005. Hyperspectral one-
meter-resolution remote sensing in Yellowstone National Park, Wyoming: 11.
Biomass. Rangeland Ecology and Management 58:459-465.

Myneni, R. B., and D. L. Williams. 1994. On the relationship between FAPAR and
NDVI. Remote Sensing of Environment 49:200-21 1.

Nagler, P. L., Y. Inoue, E. P. Glenn, A. L. Russ, and C. S. T. Daughtry. 2003. Cellulose
absorption index (CAI) to quantify mixed soil-plant litter scenes. Remote Sensing
of Environment. 87:310-325.

O'Brien, R.A., C.M. Johnson, A.M. Wilson, and V.C. Elsbemd. 2003. Indicators of
rangeland health and functionality in the Intermountain West. RMRS-GTR-104.
US. Department of Agriculture, Forest Service, Rocky Mountain Research
Station. 13 p.

Olsson, P., and C. Folke. 2001. Local ecological knowledge and institutional dynamics
for ecosystem management: a study of Lake Racken Watershed, Sweden.
Ecosystems 4:85-104.

Padgett, D. E., and E. B. Allen. 1999. Differential responses to nitrogen fertilization in
native shrubs and exotic annuals common to Mediterranean coast sage scrub of
California. Plant Ecology 144:93-101 .

Panter, K. E. 1990. Toxicity of knapweed in horses. Knapweed 4:3-4.

Parihar, S. S., A. Kak, D. Saba, and S. K. Rei. 1999. Flowering phenology, dispersal unit
and germination in eight tropical perennial grasses. Tropical Ecology 40:27-32.

Paruelo, J. M., and R. A. Golluscio. 1994. Range assessment using remote sensing in
Northwest Patagonia (Argentina). Journal of Range Management 47:498-502.

Peters, A., D. E. Johnson, and M. R. George. 1996. Barb goatgrass: a threat to
California's rangelands. Rangelands 18:8-10.

Pickup, 6., G. N. Bastin, and V. H. Chewings. 1994. Remote-sensing-based condition
assessment for nonequilibrium rangelands under large-scale commercial grazing.
Ecological Applications 4:497-517.

Price, J. C. 1994. How unique are spectral signatures? Remote Sensing of Environment
49: 1 8 1-1 86.

119

Prince, S. D. 1991. Satellite remote sensing of primary production: comparison of results
for Sahelian grassland 1981-1988. International Journal of Remote Sensing
12:1301-1311.

Qi, J ., A. Chehbouni, A. R. Huete, Y. H. Kerr, and S. Sorooshian. 1994. A modiﬁed soil
adjusted vegetation index. Remote Sensing of Environment 48:1 19-126.

Qi, J ., M. S. Moran, F. Cabot, and G. Dedieu. 1995. Normalization of sun/view angle
effects using spectral albedo-based vegetation indices. Remote Sensing of
Environment 52:207-217.

Qi, J ., R. Marsett, and P. Heilman. 2000. Rangeland vegetation cover estimation from
remotely sensed data. In: Proc. 2nd International Conference on Geospatial
Information in Agriculture and Forestry. Lake Buena Vista, FL. p. 243-252.

Qi, J ., R. Marsett, P. Heilman, S. Biendenbender, M. S. Moran, D. C. Goodrich, and M.
Weltz. 2002. RANGES improves satellite-based information and land cover

assessments in Southwest United States. EOS Transactions of the American
Geophysical Union 83:60] -606.

Randall, J. 1996. Weed control for the preservation of biological diversity. Weed
Technology 1 02370-3 83.

Reeves, M. C., M. Zhao, and S. W. Running. 2006. Applying improved estimates of
MODIS productivity to characterize grassland vegetation dynamics. Rangeland
Ecology and Management 59:1-10.

Richardson, A. J ., and J. H. Everitt. 1992. Using spectral vegetation indices to estimate
rangeland productivity. Geocarto International. 7:63-69.

Roberts, D. A., M. O. Smith, and J. B. Adams. 1993. Green vegetation,
nonphotosynthetic vegetation, and soils in AVIRIS data. Remote Sensing of
Environment 44:255-269.

Roche, B., and C. Roche. 1991.1dentification, introduction, distribution, ecology, and
economics of Centaurea. In: L. James, J. Evans, M. Ralphs, and R. Child [EDS.].
Noxious range weeds. Westview Press, Boulder, CO. p. 274-291.

Rogers, E. M. 1995. Diffusion of innovations, 4th Edition. The Free Press, New York,
NY. 518 p.

Rdling, N. G., and M. A. E. Wagemakers. 1998. Facilitating sustainable agriculture.
Cambridge University Press, Cambridge, UK. 318 p.

120

Ross, J. K. 1981. The radiation regime and architecture of plant stands. Kluwer
Publishing, Boston, MA. 392 p.

Rouse, J. W., R. H. Haas, J. A. Schell, and D. W. Deering. 1974. Monitoring vegetation
systems in the great plains with ERTS. In: Proc. 3'd NASA ERTS Symposium,
NASA SP-351. p. 309-317.

Rowan, R. C., and J. R. Conner. 1995. Small-acreage livestock operators and resource
management. Rangelands 17:167-171.

Saltz, D., H. Schmidt, M. Rowen, A. Karnieli, D. Ward, and I. Schmidt. 1999. Assessing
grazing impacts by remote sensing in hyper-arid environments. Journal of Range
Management 52:500-507.

Sellers, P. J. 1985. Canopy reﬂectance, photosynthesis and transpiration. International
Journal of Remote Sensing 6: 1 335-1372.

Serrano, L., I. Filella, and J. Penuelas. 2000a. Remote sensing of biomass and yield of
winter wheat under different nitrogen supplies. Crop Science 40:723-731.

Serrano, L., J. A. Gamon, and J. Penuelas. 2000b. Estimation of canopy photosynthetic
and nonphotosynthetic components from spectral transmittance. Ecology 81 :3 149-
3162.

Shaw, D. R. 2005. Translation of remote sensing data into weed management decisions.
Weed Science 53:264-273.

Singh, A. 1989. Digital change detection techniques using remotely-sensed data.
International Journal of Remote Sensing 10:989-1003.

Smith, M. A. 2000. Management of rangeland and grazing-determination of stocking
rates. Rangelands 22:49-51.

Somers, N. 1998. Learning about sustainable agriculture: the case of Dutch arable
farmers. In: N. G. Roling and M. A. E. Wagemakers [EDS.]. Facilitating
sustainable agriculture. Cambridge University Press, Cambridge, UK. p. 125-
133.

Strauss, A., and J. M. Corbin. 1998. Basics of qualitative research: techniques and

procedures for developing grounded theory, 2"d Edition. Sage Publications,
Thousand Oaks, CA. 311 p.

Streck, N. A., D. Rundquist, and J. Connot. 2002. Estimating residual wheat dry matter

from remote sensing measurements. Photogrammetric Engineering and Remote
Sensing. 68:1193-1201.

121

Taylor, C. A., and M. H. Ralphs. 1992. Reducing livestock losses from poisonous plants
through grazing management. Rangelands 4529-12.

The Nature Conservancy. 2000. Conserving the landscapes of San Luis Obispo County.
Final report. San Francisco, CA. 93 p.

The Nature Conservancy. 2005. The Nature Conservancy national easement study:
issues, insights, and tips for practitioners. F inal report. San Francisco, CA. 39 p.

The Nature Conservancy. 2006. California Central Coast ecoregional plan update. San
Francisco, CA. 301 p.

Thomsen, C. D., W. A. Williams, and M. R. George. 1990. Managing yellow starthistle
on annual range with cattle. Knapweed 4:3-4.

Tothill, J. C. 1977. Flowering phenology of some native perennial tropical grasses from
north-central Australia. Austral Ecology 2:199.

Tu, M., C. Hurd, and J. M. Randall. 2001. Weed control methods handbook. The Nature
Conservancy. http://tncweeds.ucdavis.edu. Version: April 2001.

Tucker, C. J. 1979. Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sensing of Environment 8: 127-150.

Tucker, C. J ., B. N. Holben, J. H. Elgin, and J. E. McMurtrey III. 1980. Relationship of
spectral data to grain yield variation. Photogrammetric Engineering & Remote
Sensing 46:657-666.

Tucker, C. J ., C. Vanpraet, E. Boerwinkel, and A. Gaston. 1983. Satellite remote sensing
of total dry matter production in the Senegalese Sahel. Remote Sensing of
Environment 13 :46 1 -474.

Tucker, C. J ., C. L. Vanpraet, M. J. Sharman, and G. V. Ittersum. 1985. Satellite remote
sensing of total herbaceous biomass production in the Senegalese Sahel: 1980-
1984. Remote Sensing of Environment 17:233-249.

Tucker, C. J ., and P. J. Sellers. 1986. Satellite remote sensing of primary production.
International Journal of Remote Sensing 7: 1 395-1416.

Tueller, P. T. 1982. Remote sensing for range management. In: C. J. Johannsen [ED.].
Remote sensing for resource management. Soil Conservation Society of America.
Madison, WI. p. 125-140

Tueller, P. T. 1989. Remote sensing technology for rangeland management applications.
Journal of Range Management 42:442-453.

122

 

Tueller, P. T. 2001. Remote sensing of range production and utilization. Journal of Range
Management 54:A77-A89.

United States Department of Agriculture, Natural Resources Conservation Service. 2000.
Summary Report: 1997 Natural Resources Inventory. Natural Resources
Conservation Service, Washington, DC. 89 p.

United States Department of Interior, Bureau of Land Management. 1996. Utilization
studies and residual measurements. Interagency Technical Reference 1734-3.
Bureau of Land Management, Washington, DC.

van Deventer, A. P., A. D. Ward, P. H. Gowda, and J. G. Lyon. 1997. Using Thematic
Mapper data to identify contrasting soil plains and tillage practices.
Photogrammetric Engineering and Remote Sensing 63 :87-93.

Veenendael, E. M., W. H. O. Ernst, and G. S. Modise. 1996. Reproductive effort and
phenology of seed production of savanna grasses with different growth form and
life history. Plant Ecology 123291-100.

Verrnote, E., and J. C. Roger. 1996. Radiative transfer modeling for calibration and
atmospheric correction. In: G. D'Zouza [ED.]. Advances in the use of NCAA
A VHRR data for land applications. Brussels. p. 49-72.

Vitousek, P. M., C. M. D'Antonio, L. L. Loope, and R. Westbrooks. 1996. Biological
invasions as global environmental change. American Scientist 84:468-478.

Voss, E. G. 2001. Michigan Flora. Part 1: Gymnosperms and Monocots. Cranbrook
Institute of Science, Bloomﬁeld Hills, MI. 488 p.

Wallace, 0. C., J. Qi, P. Heilma, and R. Marsett. 2003. Remote sensing for cover change
assessment in southeast Arizona. Journal of Range Management 56:402-409.

Wang, J ., K. P. Price, P. M. Rich. 2001. Spatial patterns of NDVI in response to
precipitation and temperature in the central Great Plains. International Journal of
Remote Sensing 22:3 827-3 844.

Washington-Allen, R. A., N. E. West, R. D. Ramsey, and R. A. Efroyrnson. 2006. A
protocol for retrospective remote sensing-based ecological monitoring of
rangelands. Rangeland Ecology and Management 59:19-29.

Weiss, S. B. 1999. Cars, cows, and checkerspot butterﬂies: nitrogen deposition and

management of nutrient-poor grasslands for a threatened species. Conservation
Biology 13:1476-1486.

123

Wessman, C. A., C. A. Bateson, and T. L. Benning. 1997. Detecting ﬁre and grazing
patterns in tallgrass prairie using spectral mixture analysis. Ecological
Applications 72493-51 1

White, J. D., S. W. Running, R. Nemani, R. E. Keane, and K. C. Ryan. 1997.
Measurement and remote sensing of LAI in Rocky Mountain montane ecosystem.
Canadian Journal of Forest Research 27: 1714-1 727.

White, M. A., G. P. Asner, R. R. Nemani, J. L. Privette, and S. W. Running. 2000.
Measuring fractional cover and leaf area index in arid ecosystems: digital camera,

radiation transmittance, and laser altimetry methods. Remote Sensing of
Environment 74: 45-57.

Wilcove, D. S., D. Rothstein, J. Dubow, D. Philips, and E. Losos. 1998. Quantifying
threats to imperiled species in the United States. BioScience 48:607-615.

Williams, G. B. 1970. Population dynamics of two perennial grasses in Australian semi-
arid grassland. Journal of Ecology 58:869-875.

World Resources Institute. 1996. World Resources [996-9 7. Oxford University Press,
New York, NY. 384 p.

Wylie, B. K., J. A. Harrington, S. D. Prince, and I. Denda. 1991. Satellite and ground-
based pasture production in Niger: 1986-1988. International Journal of Remote
Sensing 12:1281-1300.

Wylie, B. K., D. J. Meyer, L. L. Tieszen, and S. Mannel. 2002. Satellite mapping of
surface biophysical parameters at the biome scale over the North American

grasslands. A case study. Remote Sensing of Environment 79:266-278.

Yin, R. K. 2003. Case study research: design and methods, 3rd Edition. Sage
Publications, Thousand Oaks, CA. 181 p.

Young, J. A. 1992. Ecology and management of medusahead (T aeniatherum caput-
medusae ssp. asperum [simk.] melderis). The Great Basin Naturalist 52:245-252.

124

 

UN ERSITY LIB

Millililllllllllg

3 1293 02845