.
.414 .

l x .. .
(civil r16,
.. unu¢.1sv..€ “.1.ch l I)
ll: .1: .3 i .

fidfl.

i‘

.62!

 

hilt¢n¢~3p

i :Q

3110! I?"
W": i
.30; I

 

z,
.5 int.- x

.v IN.
or. I“

 

a u. 1 V , 1.x £3.
rum,” Eng fig?

:. 3 .3
(or-.1 - {.4 Jib Y; a KI
r: .i ;, ‘ ,

Pg

~

 

 

"”" LIBRARY I
Michigan State
University

 

 

 

This is to certify that the
thesis entitled

A CLIMATICALLY DRIVEN SPACE TIME MODEL OF
GLOSS/NA-MORSITANS HABITAT AND DISTRIBUTIONS:
THE IDENTIFICATION OF DRY SEASON RESERVOIRS

presented by

Mark H. DeVisser

has been accepted towards fulfillment
of the requirements for the

Masters of

Science degree in Geography

 

 

_’ )

WW

451' rofessor’s Signature

/4AU607

 

Date

MSU is an Affirmative Action/Equal Opportunity Employer

 

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

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:/Proj/Aoc&Pres/ClRC/DateDuerindd

A CLIMATICALLY DRIVEN SPACE TIME MODEL OF
GLOSSINA-MORSITANS HABITAT AND DISTRIBUTIONS:
THE IDENTIFICATION OF DRY SEASON RESERVOIRS

By

Mark H. DeVisser

A THESIS

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

MASTERS OF SCIENCE
Geography

2009

ABSTRACT

A CLIMATICALLY DRIVEN SPACE TIME MODEL OF
GLOSSINA—MORSITAN S HABITAT AND DISTRIBUTIONS:
THE IDENTIFICATION OF DRY SEASON RESERVOIRS

By
Mark H. DeVisser

Tsetse flies are the primary vector for African trypanosomiasis, a disease that
affects both humans and livestock across the continent of Africa. In 1973 tsetse were
estimated to inhabit 22% of Kenya; by 1996 that number had risen to roughly 34%.
Efforts to control the disease are hampered by a lack of information and costs associated
with the identification of infested areas. To aid control efforts I have explored possible
fluctuations over space and time of tsetse distributions in Kenya by constructing the
Tsetse Ecological Distribution Model (TED Model). The TED Model is a 250m
resolution raster based spatially explicit dynamic model that predicts tsetse distributions,
based on habitat suitability and fly movement rates, at 16 day intervals over a 4 year
period between 1/1/2001 and 12/31/2004. Using the TED Model I have identified where
and when fly populations should be constrained by unfavorable ecological conditions to
parcels of suitable habitat. My research shows that tsetse are most constrained at the end
of the cool dry season, on approximately October 15‘h of each year. The parcels of land
that I have identified as having tsetse present at the end of the cool dry season have been

mapped and labeled as dry season reservoirs.

© Copyright by
MARK H DeVISSER
2009

DEDICATION

To everyone in the past that has told me at one time or another that
“you are not living up to your potential.”

May I never hear those words directed at me again.

iv

Acknowledgments

First and foremost I offer my sincerest gratitude to my advisor Dr. Joseph
Messina. Your guidance has led me to become a better student, writer, and scientist. I
look forward to our continued collaboration in the years to come as I stay at Michigan
State University and work towards obtaining my PhD. Additionally I would like to thank
Dr. David Lusch and Dr. Nathan Moore for their input on my research and for serving on
my thesis committee. Their insights on avenues of possible model improvement were
especially constructive and will guide my future work both tsetse and non-tsetse related.

I am also grateful to Dr. Joseph Maitima whose experience and knowledge of tsetse led to
the basic concepts behind my thesis. Furthermore I am indebted to the National Institutes
of Health, Office of the Director, Roadmap Initiative, and NIGMS: award #
RGM084704A for funding my research, without this finical support this body of work
would not have been possible.

Although there have been numerous professors who have helped me get where I
am today, I need to thank three in particular who have greatly influenced the direction of
my research interests and in part this body of work. First on my list is Dr. Andrew
Fountain, who hired me as an undergraduate research assistant at Portland State
University digitizing glaciers off of aerial photography. Although the work was far from
glamorous, you solidified my desire to do research and continue on for an advanced
degree. Another Portland State University Professor that deserves recognition is Dr.

J iunn-Der (Geoffrey) Duh, who first exposed me to GIS and its possibilities. Third is Dr.
James Millington who taught ‘Systems Modeling & Simulation,’ the class where I

developed the prototype Tsetse Ecological Distribution Model.

I owe my deepest gratitude to my family, in particular my parents Don and Pam
DeVisser, and Grandmother Helen Bronkhorst, who never gave up on me. Without your
support over the years (a few more than you had hoped for I am sure), I doubt I would be
where I am today. You never once showed any doubt that I could achieve this level of
success, and for that I will be eternally grateful.

I am also indebted to my lovely, beautiful, and intelligent wife Liz. Not only did
you ‘convince’ me to finish my bachelors degree, but then you also moved to Michigan
despite more than one misgiving so I could get my Masters. The words that describe how
much you mean to me have yet to be invented. I know the next few years will be
difficult. Although I may not know exactly what the future holds, I do know you're there
with me and that you’re all I need.

Finally there are too many friends and fellow MSU grad students to list who have
supported me during the last two years. Just know that you will not be forgotten. In
closing I would like to thank one last person, a person whom I have never met, but whose
influence in writing this thesis must be acknowledged. Thank you Dr. Joan Fallon, the

inventor of Strattera®.

vi

TABLE OF CONTENTS

LIST OF TABLES 0000000000 OOOOOOOOOOOOOOOOOOOOO ...... 0.000... ..... .00... ......... O OOOOOOOOOOOOOOOOOOOOOOO x
LIST OF FIGURES ................................................. . ........................... xi
LIST OF ACRONYMS ............................................................................. xiv
CHAPTER 1
TRYPANOSOMIASIS AND THE TSETSE FLY ................................................. 1
1.1 INTRODUCTION ......................................................................................................... 1
1.2 STATEMENT OF PROBLEM ........................................................................................ 3
1.3 PURPOSE OF THIS STUDY .......................................................................................... 4
1.4 KENYA AND THE TSETSE FLY ................................................................................... 4
CHAPTER 2
RELEVANT RESEARCH ................................................................................ 10
2.1 TSETSE FLY ECOLOGY ........................................................................................... 10
2.1.1 Land Cover ..................................................................................................... 10
2.1.2 Climate ............................................................................................................ 12
2.1.3 Hosts ................................................................................................................ 14
2.1.4 Seasonal Distributions ................................................................................... 14
2.2 HUMAN INTERACTIONS WITH TSETSE FLY POPULATIONS ................................... 16
2.2.1 Traditional Societies and the Tsetse Fly ...................................................... 17
2.2.2 Colonialism and the Tsetse Fly ..................................................................... 19
2.2.2.1 Disease and Depopulation ...................................................................... 19
2.2.2.2 Colonial Control Efforts ......................................................................... 21
2.2.3 Modern Control Methods .............................................................................. 23
2.2.3.1 Insecticides ............................................................................................... 24
2.2.3.2 Tsetse Fly Bait Technology .................................................................... 25
2.2.3.4 Biological Agents ..................................................................................... 27
2.2.3.5 Sterile Insect Technique ......................................................................... 28
2.2.3.3 Trypanocidal Drugs ................................................................................ 29
2.2.3.6 Eradication versus Control .................................................................... 30
2.3 MODELING OF THE TSETSE FLY ............................................................................ 33
2.3.1 Remotely Sensed Data and Mapping Vector-Borne Disease ..................... 33
2.3.2 Past Modeling and Mapping of Tsetse Fly .................................................. 33
2.4 CURRENT TSETSE FLY RESEARCH ......................................................................... 35
CHAPTER 3
METHODS ...................................................................................................... 38
3.1 THE TSETSE ECOLOGICAL DISTRIBUTION MODEL ............................................... 38

Vii

3.1.1 Habitat Suitability Model .............................................................................. 38

3.1.2 Fly Movement Model ..................................................................................... 40
3.2 MODEL INITIALIZATION ......................................................................................... 40
3.2.1 The Identification of the Optimal Land Cover Product ............................. 40
3.2.1.] Data .......................................................................................................... 41
3.2.1.3 Suitability Classification ......................................................................... 46
3.2.1.4 Validation ................................................................................................ 49
3.2.1.5 Mapcurves Goodness of Fit Test ........................................................... 52
3.2.2 Temporal Initialization .................................................................................. 56
3.2.3 Starting Distribution ...................................................................................... 56
3.3 TED MODEL DATA AND PARAMETERS VALUES ................................................... 58
3.3.1 Normalized Difference Vegetation Index ..................................................... 58
3.3.1.] Data .......................................................................................................... 59
3.3.1.2 Suitability Classification ......................................................................... 59
3.3.2 Land Surface Temperatures ......................................................................... 60
3.3.2.1 Data .......................................................................................................... 60
3.3.2.2 Suitability Classification ......................................................................... 60
3.3.3 Fly Movement Rates ...................................................................................... 61
3.4 OUTPUTS OF THE TED MODEL .............................................................................. 62
3.5 TED MODEL VALIDATION ..................................................................................... 63
3.5.1 Ground Truth Comparison ........................................................................... 63
3.5.2 Sensitivity Analysis ........................................................................................ 65
3.5.2.1 Sensitivity Index ...................................................................................... 65
3.5.2.2 The Spatial Location of Potential Tsetse Expansion ........................... 67
CHAPTER 4
INITIALIZATION OF THE TED MODEL:
RESULTS, DISCUSSION, AND CONCLUSION.... ....................................... .....68
4.1 LAND COVER .......................................................................................................... 68
4.1.1 Results ............................................................................................................. 68
4.1.2 Discussion of Which Land Cover Product to Use in the TED Model ....... 71
4.1.3 Additional Points of Interest in the Land Cover Analysis ......................... 72
4.1.3 Conclusion ...................................................................................................... 73
4.2 STARTING DISTRIBUTIONS ..................................................................................... 74
4.2.1 Results ............................................................................................................. 74
4.2.2 The Determination of the Tsetse Distribution used to initialize the TED
Model ........................................................................................................................ 76
4.2.3 Conclusion ...................................................................................................... 77
CHAPTER 5
THE TED MODEL OUTPUTS, VALIDATION, AND UNCERTAINTY:
RESULTS AND DISCUSSION ........ ........................... . ............. .....79
5.1 THE TED MODEL OUTPUTS ................................................................................... 79
5.1.1 Individual Scenes ........................................................................................... 79
5.1.1.] Results ...................................................................................................... 79

viii

5.1.1.2 Discussion ................................................................................................ 79

5.1.2 Percent Probability Map Results and Discussion ....................................... 93

5.1.3 Tsetse Reservoirs ............................................................................................ 95

5.1.3.1 Results ...................................................................................................... 95

5.1.3.2 Discussion ................................................................................................ 97

5.2 VALIDATION .......................................................................................................... 101

5.2.1 Ground Truth Data Comparison ............................................................... 101

5.2.1.1 Results .................................................................................................... 101

5.2.1.2 Discussion .............................................................................................. 101

5.2.2 Sensitivity Analysis ...................................................................................... 104

5.2.2.1 Moisture Sensitivity Index Results ...................................................... 104

5.2.2.2 Maximum Temperature Sensitivity Index Results ............................ 106

5.2.2.3 Minimum Temperature Sensitivity Index Results ............................. 108

5.2.2.4 The Spatial Location of Potential Tsetse Expansion Results ............ 110

5.2.2.5 Sensitivity Analysis Discussion ............................................................ 113

5.3 UNCERTAINTY ....................................................................................................... 117

5.3.1 Spatial Resolution ........................................................................................ 117

5.3.2 Fly Movement ............................................................................................... 118

5.3.3 Data ............................................................................................................... 119

5.3.3.1 NDVI ...................................................................................................... 119

5.3.3.2 Hosts ....................................................................................................... 119

5.3.3.2 Land Cover ............................................................................................ 120
CHAPTER 6

CONCLUSION .. ........ ....... 122

APPENDIX.... ............... ........ ........................... ....... 130

REFERENCES ..... . ...................... ...... ..... . ...... .............. 132

ix

Table 3.1:

Table 3.2:

Table 3.3:

Table 4.1:

Table 4.2:

Table 5.1:

Table 5.2:

Table 5.3:

Table 5.4:

Table 5.5:

Table 5.6:

Table 5.7:

LIST OF TABLES

The land use land cover data sets that are publicly available for Kenya ........ 42
MODIS Type 1 LULC classes and their suitability classification .................. 47
Sensitivity classes adapted from Lenhart et al. (2002) ................................... 66
The amount of woody vegetation and suitable tsetse habitat (when combined

with environmental variables) predicted by the LULC binary maps .............. 68
Results of the Mapcurves GOF analysis between the LULC binary suitable
tsetse habitat maps and the combined F AO / IAEA distribution map and the
1996 fly belt map ............................................................................................ 70

The 69 scenes produced by the TED Model, separated by year, and ranked by
surface area predicted to have tsetse present .................................................. 82

Annual land use land cover conversion with regards to tsetse land cover

suitability ......................................................................................................... 84
Mean surface area predicted by the TED Model to have tsetse present,

temperatures, and NDVI between 2002 and 2004 .......................................... 86
Ground truth comparison GOF scores .......................................................... 101
Moisture sensitivity analysis results ............................................................. 105
Maximum temperature sensitivity analysis results ....................................... 107
Minimum temperature sensitivity analysis results ........................................ 109

 

LIST OF FIGURES

Images in this thesis are presented in color

Figure 1.1: Geographic location and topography of Kenya ........................ * ....................... 5
Figure 1.2: The 1996 KETRI fly belts map ....................................................................... 8
Figure 2.1: Woody savannah suitable for tsetse .............................................................. 12
Figure 2.2: A comparison of relative moisture during the wet and dry seasons .............. 15
Figure 2.3: A Nguruman Tsetse Trap .............................................................................. 26
Figure 3.1: The Tsetse Ecological Distribution (TED) Model structure ......................... 39
Figure 3.2: A comparison Of MODIS Type 1 Global Land Cover at 500m and 1km
resolutions ..................................................................................................... 44
Figure 3.3: Annual precipitation data from WorldClim classified to create a binary tsetse
precipitation suitability map ......................................................................... 48
Figure 3.4: The binary suitability maps created when Afn'cover and MODIS 1km Type 1
were combined with classified elevation and precipitation data .................. 49
Figure 3.5: The Food and Agriculture Organization of the United Nations / International
Atomic Energy Agency combined moristans tsetse species group distribution
map ................................................................................................................ 51
Figure 3.6: An example a cross tabulation matrix that would be calculated in the first
step Of the Mapcurves GOF analysis ............................................................ 54
Figure 3.7: An example of a weighted ratio comparison matrix for the calculation of a
Mapcurves GOF score .................................................................................. 54
Figure 3.8: An example of a cumulative ratio frequency distribution and integration table
for the calculation of a Mapcurves GOF score ............................................. 55
Figure 3.9: A comparison of tsetse distributions in 1973 and 1996 ................................ 57
Figure 3.10: A graphical representation of a fly front ..................................................... 62
Figure 4.1: Mapcurves GOF scores for each LULC data set when compared to the FAO

/ IAEA combined distribution map ............................................................... 70

xi

Figure 4.2: A comparison of the three data sets used to initialize the TED model .......... 75

Figure 5.1: A graph of the surface area predicted to have tsetse present by the TED
Model between 2002 and the end of 2004 .................................................... 80

Figure 5.2: A graph of the intra annual variability of the TED Model predicted tsetse
distributions ................................................................................................... 8 1

Figure 5.3: A graph of the mean surface area of the TED Model predicted tsetse
distributions, maximum temperature, and minimum temperature ................ 85

Figure 5.4: A graph of the mean surface area of the TED Model predicted tsetse

distributions and NDVI ................................................................................. 88
Figure 5.5: The location of suitable and unsuitable moisture regimes at the beginning of

the hot and cool dry seasons ......................................................................... 92
Figure 5.6: The TED Model tsetse percent probability map ............................................ 94
Figure 5.7: The amount of surface area predicted to have tsetse present by the TED

Model between 2002 and the end of 2004 .................................................... 96
Figure 5.8: The location of the tsetse cool dry season reservoirs .................................... 98
Figure 5.9: The location of the tsetse hot dry season reservoirs ...................................... 99

Figure 5.10: The location of the tsetse dry season reservoirs compared to the TED Model
percent probability map .............................................................................. 100

Figure 5.11: A graph of the NDVI thresholds and the corresponding surface area state
variable used in the sensitivity analysis ...................................................... 104

Figure 5.12: A graph of the maximum temperature thresholds and the corresponding
surface area state variable used in the sensitivity analysis ......................... 106

Figure 5.13: A graph of the minimum temperature thresholds and the corresponding
surface area state variable used in the sensitivity analysis ......................... 108

Figure 5.14: The spatial location of potential expansion of tsetse distributions if an
ecological variable was removed from the TED Model ............................. 111

Figure 5.15: The location of potential tsetse expansion based on minimum temperature in
relation to the Kenyan Highlands ............................................................... 112

Figure 5.16: A graph of the combined and standardized parameter threshold values used
in the sensitivity analysis. ........................................................................... 114

xii

 

Figure 6.1: The TED Model percent probability map compared to the 1996 fly belts
..................................................................................................................... 127

xiii

AHP
AIDS
ASALS
AU-IBAR
AVHRR
Avia-GIS
CDC
CLIP
CPDD

DDT
DFID
DFMO
DNA
DSR
ERGO
EST
FAO

F PAR
GDP
GLC2000
GLCC
GOF
GPS
HIV
IAEA
ICIPE
IDW

LIST OF ACRONYMS

Animal Health Programme

Acquired Immune Deficiency Syndrome

Arid and Semiarid Lands

African Union Inter-African Bureau for Animal Resources
Advanced Very High Resolution Radiometer

Agriculture and Veterinary Information and Analysis

Centers for Disease Control and Prevention

Climate Land Interaction Project at Michigan State University
University of North Carolina at Chapel Hill Consortium for Parasitic Drug
Development

DichlorO-Diphenyl-Trichloroethane

Untied Kingdom’s Department for International Development
Difloromethylomithine

Deoxyribonucleic Acid

Dry Season Reservoirs

Environmental Research Group Oxford

Expressed Sequence Tags

Food and Agriculture Organization of the United Nations
Fraction of Photosynthetically Active Radiation

Gross Domestic Product

Global Land Cover 2000

Global Land Cover Classification

Goodness of Fit

Global Positioning System

Human Immunodeficiency Virus

International Atomic Energy Agency

International Centre of Insect Physiology and Ecology

Inverse Distance Weighted

xiv

IGBP
IGGI
ILRI

International Geosphere-Biosphere Programme
International Glossina Genome Initiative

International Livestock Research Institute

IRD-CIRRAD Centre International de Recherche De'veloppement sur l’Elevage en zone

ITC
ITCZ
KARI-TRC
KETRI
LAI
LCCS
LST
LULC
MODIS
MRT
NASA
NDVI
NIR
NPP
PAAT
PAATIS
PAR
PATTEC
RIKEN
RNA

SI

SIT
TALA
TED Model
UMd
UN

Subhumide

International Trypanotolerance Centre

Intertropical Convergence Zone

Kenya Agricultural Research Institute Trypanosomiasis Research
Kenyan Trypanosomiasis Research Institute

Leaf Area Index

Land Cover Classification System

Land Surface Temperature

Land Use Land Cover

Moderate Resolution Imaging Spectroradiometer

MODIS Reprojection Tool

National Aeronautics and Space Administration

Normalized Difference Vegetation Index

Near-infrared

Net Primary Production

Programme Against African Trypanosomiasis

Programme Against Afiican Trypanosomiasis Information System
Photosynthetically Active Radiation

Pan African Tsetse and Trypanosomiasis Eradication Campaign
Rikagaku Kenkyfisho

Ribonucleic Acid

Sensitivity Index

Sterile Insect Technique

Trypanosomaiasis and Land-use in Africa Research Group at Oxford
Tsetse Ecological Distribution Model

University of Maryland

United Nations

XV

USGS United States Geological Survey
WHO ' World Health Organization

xvi

CHAPTER 1
TRYPANOSOMIASIS AND THE TSETSE FLY

1.1 Introduction

African trypanosomiasis, otherwise known as sleeping sickness in humans and
nagana or sura in cattle or pigs, is a neglected tropical disease (Yamey, 2002; Hotez et al.,
2006 / 2007; WHO 2009). Over the past 20 years considerable attention has been
devoted to particular diseases (e. g., malaria, tuberculosis, HIV / AIDS), while other
infectious diseases have been relatively ignored (Yamey, 2002; Hotez et al., 2006).
These ignored diseases have come to be known as neglected tropical diseases, and are
often associated with poverty stricken and marginalized human populations living in
remote rural areas, urban slums, or conflict zones (WHO, 2009). The impacts of
neglected tropical diseases are chronically underestimated, with some claiming over 1
billion people affected by and roughly 530,000 deaths annually attributed to one of the 14
diseases (Sachs & Hotez, 2006; WHO, 2009). The number of new sleeping sickness
cases reported annually is roughly 13,000, however, as a neglected tropical disease many
cases are not recognized or reported and the actual human infection rate is likely higher
(CDC, 2008a / 2008b). In addition to the problems associated with being a neglected
tropical disease, trypanosomiasis poses the unique problem of infecting both humans and
domesticated livestock populations, thus increasing the total number of humans affected
by the disease.

African trypanosomiasis is caused by a single celled protozoan, a trypanosome,

most Often transmitted by the tsetse fly (genus Glossina). It is considered one of the most

I l A. “1-l

important economically debilitating diseases in Sub-Saharan Africa (Olet, 2003; 0100,
2006). Three major epidemics have occurred in the past hundred years, one between
1896 and 1906, and the other two in 1920 and 1970 (WHO, 2006). In 1986,
approximately 70 million people were estimated to be at risk of exposure to tsetse (WHO,
2006). A decade later, it was estimated that at least 300,000 cases of sleeping sickness
were underreported due to lack of surveillance capabilities, diagnostic expertise, and
health care access (Kennedy, 2005; WHO, 2006). In 2001 as a response to these
limitations, the World Health Organization (WHO), with public and private partnerships,
initiated a new surveillance and elimination program (WHO, 2006), during which
approximately 25,000 new cases were reported annually (Weekly Epidemiological
Record, 2006). In some areas, sleeping sickness symptoms were misdiagnosed as
malaria, and therefore masked the overall number Of new cases (Kennedy, 2005 / 2006).
Nagana and sura also indirectly affects the lives of people in Sub-Saharan Africa
because it can decimate domesticated animal populations thus impacting nutrition and
livelihoods. It is estimated that livestock productivity decreases by 20% to 40% in tsetse
infested areas (Hursey, 2001; Rogers and Randolph, 2002). In Kenya where livestock
production accounts for approximately 12% of the Gross Domestic Product (GDP) (FAO
& AGAL, 2005), the economic burden of trypanosomiasis is felt at both local and
national scales (Welbum et al., 2006). Because the trypanosomes use long lived wild
vertebrate populations as natural hosts, most modern efforts to control the spread of
trypanosomiasis have focused on tsetse populations (i.e., the vector) (Jordan, 1986;

Grant, 2001).

Adult tsetse are between 6-14 mm in length, and have been described as “rather
dull in appearance, varying in color from a light yellowish-brown to a dark blackish-
brown” (Laird, 1977). They are a diurnal biting fly, which naturally feeds on wild
ungulate and ruminant populations (Pollock, 1982a/ 1982b). The geographic distribution
of tsetse is limited to Sub-Saharan Africa, where they infest 8.5 million km2 in 37
countries (Allsopp, 2001). This study focuses on Kenya, where in 1973 tsetse were
estimated to inhabit 22% (129,229 kmz) of the country (Ford and Katondo, 1977). By
1996 the amount of Kenya infested with tsetse had risen to roughly 34% (202,774 km?)

(KETRI, 2008).

1.2 Statement of Problem

Tsetse control efforts have been chronically hampered by identification of
legitimate infested areas, reinvasion of tsetse into previously cleared regions, and
substantial costs in conjunction with limited resources. Given changing land cover and
climate conditions, a spatially explicit model that can predict tsetse distributions
temporally based on habitat suitability is critical to efforts aimed at controlling the
disease. Current models predict tsetse distributions at a particular moment in time and do
not inform control efforts on any possible fluctuations in tsetse distributions. If tsetse
distributions do fluctuate over space and time, then a dynamic‘spatially explicit model
that predicts when and where fly distributions are constrained due to local availability of
suitable habitat could be used to maximize the limited resources available for tsetse

control.

 

1.3 Purpose of this Study

Tsetse require ecologically suitable habitat, which includes proper moisture
levels, temperature regimes, and land cover. My research tests the notion that in Kenya
ecologically suitable tsetse habitat fluctuates over space and time. Fluctuations in
ecologically suitable tsetse habitat occur at both an intra-annual (seasonal) and inter-
annual temporal resolutions. In addition to testing the temporal resolution of fluctuating
tsetse habitat, I will identify the season, approximate date, and geographic location when
tsetse distributions should be constrained and occupy the least amount of surface area.
Throughout the rest of this paper, the hypothetical parcels of suitable tsetse habitat that
potentially exist at sometime during the year will be referred to as tsetse reservoirs. I also
examine the respective roles of the individual climate variables (i.e., moisture and
temperature) on the potential establishment of tsetse reservoirs. By constructing a
spatially explicit dynamic model using remotely sensed climate and land cover data
combined with fly movement rates, any spatial and temporal fluctuations in tsetse
distributions can be tracked, the identification tsetse reservoirs in Kenya is possible, and
the role Of the various ecological variables on influencing tsetse distributions can be

explored.

1.4 Kenya and the Tsetse Fly

This study focuses on the country of Kenya, which lies on the equator in East
Africa (Figure 1.1). Kenya has an official area of 582,650 kmz, roughly twice the size of
the US. state of Nevada. With the proximity of the Indian Ocean, and including the

Great Rift Valley, Lake Victoria (i.e., the largest lake in Africa), vast expanses of

 

flatland, and Mount Kenya (i.e., the second highest peak in Africa) Kenya boasts a
variety of physiographic features. The diversity of physiographic features in Kenya
contributes to multiple regional, meso, and micro scale climate regimes (e.g., cool moist

highlands above 1,500m elevation, near desert conditions in the north, a warm humid

coastal plain).

 

 

I.

 

o .59 1'06; .1 210° . . .
Figure 1.1: Geographic location and topography of Kenya.

 

 

 

 

Seasonal climate conditions in Kenya are primarily driven by the Intertropical
Convergence Zone (ITCZ). As the ITCZ oscillates to the north and south over the
equator, Kenya experiences four distinct seasons (the long and short rains, and the hot
and cool dry seasons). The long rains coalesce as the ITCZ moves over the equator
(March equinox) heading to the north of Kenya, and last from early March to late May
(Gatebe et al., 1999). The long rains produce on average half of the precipitation
received by Kenya (Awange et al., 2008), and are generally reliable in the timing of their
onset. Following the long rains is the cool dry season, which lasts from early June to late
October (Gatebe et al., 1999). The cool dry season starts just before the June solstice,
when regionally the coolest temperatures can occur, and ends just afier the September
equinox, when regionally the warmest temperatures can occur. The cool dry season ends
with the onset of the short rains in late October or Early November. The short rains
coalesce as the ITCZ moves to the south of Kenya, approximately one month after the
September equinox, and can last until late December (Gatebe et al., 1999). The short
rains can produce one third of the precipitation in Kenya (Awange et al., 2008).
Following the short rains is the hot dry season, named for having average temperatures
warmer than that of the longer cool dry season. The warmer temperatures that occur
during the hot dry season are in part due to the drier and warmer continental air that
frequently comes from the Sahara at this time (Gatebe et al., 1999), and the lower amount
of precipitation produced in the short rains to mitigate temperatures before returning to
dry conditions. At the end of hot dry season in late February, the long rains return and
the seasonal cycle begins anew.

Although the general description of the four seasons in Kenya portray the timing,

duration, and climate conditions as being constants, in actuality there is Often a high
degree of inter-annual variability leading to possible drought or flood conditions
(Awange et al., 2008). The variability in the climate conditions in conjunction with the
diverse landscapes present in Kenya, poses a significant challenge when modeling any
physical or ecological phenomenon. However, the diversity also allows for the
examination of tsetse in wide variety of physiographic and climatic scenarios.

In Kenya, tsetse populations exist across these varied habitats; however, their
populations are concentrated in six distinct zones: North and South Rift Valley, Arid and
Semiarid Lands (ASALS) North of Mt. Kenya, Central Kenya, Coastal, Transmara—
Narok-Kajiado, and the Western Kenya & Lake Victoria belts (Figure 1.2) (Muriuki et
al., 2005; KETRI, 2008). These tsetse infested zones, commonly referred to as fly belts,
can contain one or more tsetse species with boundaries set by a variety of physical,
biological, and anthropogenic barriers.

Tsetse (genus Glossina) are divided into three sub-genus groups, all of which are
found in Kenya. The sub-genus Austem'na, also referred to as the fusca group, are
commonly considered forest tsetse species, with the notable exception Of G. longipennis,
which lives in sparsely vegetated arid regions (Cecchi et al., 2008a). Three species
within the fusca group are found in Kenya: G. brevipalpis, G. fuscipleuris, and G.
Iongipennis (Pollock, 19823). The sub-genus Nemorhina or palpalis group, a riverine
species group, with only one species, G. F uscipes, is also present in Kenya (Pollock,
1982a). The third sub-genus Glossina or morsitans group is considered a woody
savannah tsetse species. Four species of the morsitans group are found within Kenya: G.

austeni, G. morsitans, G. swynnertoni, and G. pallidipes.

1996 Tsetse Fly Belts

 

       

1) North 8. South Rift Valley Belt
2) ASALa North of Mount Kenya
3) Central Kenya Belt

4) Coastal Belt

5) Tranamara-Narok-Kajlado Belt
6) Weetern Kenya 5 Lake Victoria Belt

Kilian-tore
' 0 50 100 200

Figure 1.2: The 1996 KETRI fly belts map.

For the purposes of this study the morsitans group was selected as the primary
focus. The moristans group has the greatest spatial distribution in Kenya; with palpalis
only located along the shore of Lake Victoria and the Ugandan, and fixsca, whose
distributions tend to overlap that of moristans, found in isolated pockets of forest and

along the Tanzanian border (Pollock, 1982a; Wint & Rogers, 2000). Also, in Kenya, four

Of the eight tsetse species belong to the moristans group. Finally, one Of the four species
within the moristans group, G. pallidipes, is considered the fly species most responsible
for transmitting trypanosomiasis in Kenya. Hereafter when I use the general

nomenclature “tsetse,” I will be specifically referring to the morsitans group.

CHAPTER 2
RELEVANT RESEARCH

2.1 Tsetse Fly Ecology

Like other living organisms, tsetse rely on suitable environmental conditions for
survival. Suitable habitat for tsetse must include acceptable land cover types, climate
conditions, and food sources (Pollock, 1982b). All three tsetse habitat requirements vary
over space and time, and have been individually theorized to be the primary driver
influencing observed changes in tsetse distributions.
2.1.1 Land Cover

Until the early 19603 it was widely accepted that particular plant Species (e. g.,
Brachystegia, Fuirena, Combretum, Adansonia, Diplorhynchus) were required to
consider an area suitable for tsetse (Swynnerton, 1921; Austen & Hegh, 1922; Ford,
1971). Since then, researchers have come to realize that it is not a specific plant Species,
but rather the vegetation geometry (i.e., height, diameter, and texture) which creates
suitable environmental conditions that attract tsetse to certain types Of land cover (Leak et
al., 2008). The vegetation geometry required by tsetse consists of woody plant material
greater than 1 to 3 cm in diameter, a minimum of l to 4 meters in height, and with a
coarse surface (e. g., rough / loose bark), hereafter referred to as woody vegetation
(Austen & Hegh, 1922; Jordan, 1986).

Woody vegetation land cover can be found in sufficient quantities in several types
Of ecosystems including riparian, tropical dry forest, some agricultural zones, and most

importantly woody savannah (Pollock, 1982b; Cecchi et al., 2008a / 2008b; Leak et al.,

10

2008). Riparian ecosystems considered suitable for tsetse are generally seasonal flood
plains and swamps, where woody shrubs are able to grow when standing water is not
present (Cecchi et al., 2008a). Tropical dry forests are considered forests with distinct
wet and dry seasons, and are particularly important to G. austeni along the Kenyan coast
(Pollock, 1982b). Agricultural zones with woody crops, e. g., shrub crops, palms, and tree
plantations, can support tsetse populations (Leak et al., 2008). Although agricultural
zones are not considered primary habitat due to the lack of natural hosts and a high rate
of disturbance, they are important environments given the increased likelihood of human
and tsetse interaction (Cecchi et al., 2008a). Woody savannah, the mosaic of grass,
shrubs, and trees (Figure 2.1), is arguably the most important ecosystem for tsetse due to
the presence of abundant woody vegetation, preferred host species, and suitable climate
conditions. Cecchi et al. (2008b) estimated that 64.6% of the land in sub-Saharan Africa
infested with tsetse is some form of woody savannah (e.g., tree savannah, shrub
savannah, shrubland, woodland).

Researchers have estimated that tsetse only leave woody vegetation resting sites
for no more than one hour each day to feed, breed, or locate a more suitable resting site
(Rogers & Randolph, 1985; Jordan, 1986; Leak, 1999). The remaining 23 hours a day
are spent at resting sites waiting for a potential host to pass by, digesting a blood meal,
and taking advantage of the micro habitat / climate provided by woody vegetation (Laird,
1977). These micro habitats (e.g., loose bark, underside of branches / logs, hollows in
tree trunks or logs) mitigate high temperatures and provide preferred moisture levels to
prevent against starvation and desiccation as discussed in the next section (Austen &

Hegh, 1922; Pollock, 1982a/ 1982b).

11

 

 

Figure 2.1: Woody savannah suitable for tsetse in Nguruman, Kenya. (Photo by
author)

2.1.2 Climate

Extensive research has shown that a significant relationship exists between tsetse
and climate conditions (e.g., Nash, 1933; Bursell, 1957; Hargrove, 1980; Williams et al.,
1990; Hargrove, 2001). Climate greatly influences the activity levels and rate at which
both water and fat stores are consumed physiologically by tsetse (Leak, 1999). As
temperatures increase the rates of fat and water consumption increase, requiring the fly to
either seek out a host on which to feed or risk dying of starvation or desiccation
(Hargrove, 2001). Tsetse populations are generally found in regions with mean annual
temperatures between 19—28°C, and thrive when temperatures are between 21-26°C
(Pollock, 1982b). However, when temperatures rise above ~32°C tsetse seek out woody
vegetation refuges (Pilson & Pilson, 1967), which can be up to 45°C cooler than ambient
air temperatures (Torr & Hargrove, 1999; Muzari & Hargrove, 2005). At first this

behavior seems counterintuitive; tsetse opt to rest instead of finding a blood meal when at

12

risk Of dying due to starvation or desiccation. However, when in flight tsetse greatly
increase the rate at which both fat and water are used (Loke & Randolph, 1995). This
increased use of fat and water reserves, in conjunction with the increased rates
experienced during periods of high temperatures, drive tsetse to seek shelter rather than
attempt to locate food when temperatures rise above ~32°C. The probability of survival
drops to 50% when tsetse are exposed to temperatures greater than ~36°C for three hours
(Terblanche et al., 2008), and temperatures greater than 40°C are considered lethal
(Knight, 1971; Torr & Hargrove, 1999).

Low moisture levels compound the threat of high temperature by increasing the
rate of water consumption leading to desiccation in adult tsetse (Leak, 1999). A
significant negative correlation has been reported between fly populations and saturation
deficits (i.e., a measure of the drying power of the atmosphere based on temperature and
relative humidity, see Randolph & Storey, 1999) (Nash, 1933; Rogers, 1979, Hargrove,
2001). However, the question of whether or not tsetse in dry conditions die from
starvation or desiccation has been debated for some time (see Nash, 1937; Buxton, 195 5 ;
Bursell, 1961 / 1963; Rogers, 1979/ 1990; Hargrove, 1980 / 2001; Rogers & Randolph,
1986 / 1991). Regardless, it is clear that low moisture levels do impact tsetse
populations, with optimum saturation deficits between 4.5 — 13 mm Hg (Rogers, 1979).

Unlike high temperatures and low moisture levels, low temperatures slow tsetse
physiology and induce a “chill coma” (Terblanche et al., 2008). The chill coma effect
sets in when temperatures drop below 17-20°C, preventing tsetse from flying, carrying
out normal life activities, and leading to starvation (Mellanby, 1936/ 1939; Knight, 1971;

Pollock, 1982b; Hargrove, 1980). A further drop in temperature will kill tsetse outright,

13

with the probability of survival dropping to 50% when exposed to temperatures below
~10°C for 3 hours (Terblanche et al., 2008).
2.1.3 Hosts

Tsetse ingest a blood meal from a host on average every 2-3 days, which provides
both food and water (Randolph et al., 1991; Schofield & Torr, 2002). Tsetse feed on a
variety of host species, the majority of which are wild ungulates and ruminants (e.g.,
warthogs, buffalo, bushbuck) (Pollock, 1982a / 1982b). With regards to domesticated
livestock, cattle are favored over donkey, sheep, or goats (Leak, 1999). However,
preferred host species may vary from place to place based on host availability, host’s
tolerance Of being bitten, and the digestibility of the hosts blood by tsetse (W eitz &
Glasgow, 1956).
2.1.4 Seasonal Distributions

Seasonal fluctuations in the distribution of tsetse have long been Observed and
discussed (e.g., Austen & Hegh, 1926; Nash, 1933; Bursell, 1956; Leak, 1999; Hargrove,
2001; Bett et al., 2008). Three primary theories have been suggested to explain seasonal
fluctuations in tsetse populations: 1) hosts distributions, 2) moisture availability, and 3)
suitable temperatures. The theory of host species determining tsetse distributions focuses
on the notion that tsetse follow migrating wild host populations as they move to take
advantage of fresh grth during the wet seasons (Austen & Hegh, 1926). The moisture
level theory cites the high correlation between tsetse and saturation deficits first reported
by Nash (1933). The third theory argues that temperature is the primary driver of

seasonal tsetse distributions (Bursell, 1963).

14

 

 

 

 

 

 

 

Figure 2.2: Map A is the relative available moisture in Kenya during the wet
season; Map B is the relative available moisture during the dry season. In both
maps Nguruman is highlighted to demonstrate the possible development of a fly
reservoir due to a drop in moisture levels.

Each of the three theories are plausible. For example, without host populations
tsetse would die of starvation or desiccation, leading to the conclusion that tsetse would
need to follow migrating host populations. The possibility of starvation or desiccation is
also increased by seasonal changes in climate. For instance, seeking out locations with
suitable moisture levels during the dry season (e.g., Figure 2.2) would in turn increase the
chance of survival.

No single theory has been universally accepted. It is likely that a combination of
the three theories drive any observed changes. For example, as the wet season begins

new vegetation grth encourages wild host populations to migrate to new feeding

locations. At the same time the temperature drops and the saturation deficit lowers,

allowing tsetse to expand out to feed on themigrating host populations and take
advantage of the newly available habitat. As the cooler wet season ends and distributions
Of host populations change, locations with suitable habitat for tsetse start to become
limited.

The combination of the three theories on seasonal fluctuations in suitable tsetse
habitat has the advantage allowing the role of each driver to change based on current
conditions (i.e., no one driver is dominate at all times). For instance, the host populations
might migrate before climate conditions become unsuitable for tsetse, requiring tsetse to
leave the area or starve. In the following year or in a nearby area a drought could occur
and despite suitable temperatures and the presence of host species and force tsetse to
retreat or risk starvation or desiccation. Alternatively high temperatures could force
tsetse to retreat to dry season reservoirs despite proper moisture levels and host species.

In Kenya, where a variety of physiographic landscapes and climate conditions
exist, no one theory on what drives seasonal fluctuations should be considered dominate.
Thus a spatial model that predicts the seasonal location and timing of tsetse distributions
in Kenya should include a variable that represents each theory (i.e., hosts, moisture, and

temperature).

2.2 Human Interactions with Tsetse Fly Populations

Humans living in the savannahs of East Afi‘ica have historically attempted to limit
their own exposure and that of their livestock to tsetse (Lambrecht, 1964). Limited
exposure to tsetse can be accomplished in two ways, through avoidance or by controlling

fly populations. While avoiding areas where tsetse are present may be the simplest

16

approach to dealing with the fly (Jordan, 1986), it hampers where and when humans can
live and travel in East Africa. In lieu of expanding human populations and the relative
ease when compared to dealing with the disease and host Species, control of the vector
has become a more widely adopted approach to avoiding trypanosomiasis (Grant, 2001).
In the next three sections I discuss control strategies employed at various times in East
Africa. First I review the traditional methods for controlling tsetse populations. Next I
examine the effect colonialism had on tsetse distributions and the control methods used
during this time period. The third section describes four control strategies implemented
post colonialism, and I finish with a brief summary of the debate on tsetse control versus
eradication.
2.2.1 Traditional Societies and the Tsetse Fly

In East Africa prehistoric man may have relied solely on avoidance of tsetse
infested regions as evident by the trypanosome parasite not evolving to include humans
as a natural host species (Lambrecht, 1980). However, with expanding human
populations and the introduction of non-indigenous livestock around 6,000 B.P., human
populations would have increased interactions with tsetse and trypanosomiasis
(Lambrecht, 1964). Under these circumstances, limiting exposure to sleeping sickness
and nagana would have only been possible through controlling fly populations. The only
methods available to indigenous societies for controlling tsetse would have been various
landscape modification techniques, which decrease the amount of suitable tsetse habitat
by altering land cover and host population dynamics (Ford, 1971).

One widely used traditional method of landscape modification across East Africa

has been field burning. Communities would use fire for a variety Of purposes including:

17

bush clearing, pasture regeneration, hunting, weed control, charcoal production, and
general shaping of the environment (Eriksen, 2007). With regards to tsetse, field burning
practices would physically kill the fly, and more importantly, destroy suitable tsetse land
cover (Knight, 1971; White, 1995).

Fire alone would not entirely eliminate the threat of trypanosomiasis. Once the
landscape was cleared of the undesirable land cover through field burning, various forms
of agriculture were implemented to ensure that woody vegetation and tsetse did not return
(Ford, 1971). In the case of crop agriculture, cleared lands are planted and maintained,
preventing the regrowth of woody land cover in and around settlements (Giblin, 1990).
Another traditional agricultural method in East Africa is pastoralism, which is the semi-
nomadic herding of domesticated animals in the dry grasslands and savannahs where
unpredictable precipitation limits crop agriculture (Smith, 1984). By using a combination
of field burning and grazing, pastoralist herders can markedly decrease the amount of
woody vegetation on the landscape (Guy, 1989), especially when goats are used to clear
some of the woodier vegetation on the periphery of tsetse habitats (Ford, 1971).

Increased pastoralism and crop production also diverted water resources away
from natural plant communities. Although not intended, this further contributed to the
control tsetse populations by decreasing the amount of woody vegetation on the
landscape (Ford, 1971). Human water use in conjunction with the other previous
mentioned land use practices of field burning, crop production, and pastoralism puts
stress not only on tsetse, but also on the local non-domesticated animal populations. The
net effect is to drive Off both the hosts and the flies into regions infrequently used by

humans (Giblin, 1990). As the human populations grew, more land was put under

18

production, until only marginal shrubland and forest ecotones with low moisture levels
relative to human needs for agriculture and pastoralism were left for both host and fly to
live (Ford, 1971).

Eventually a balance was reached between man’s need for arable land, grazing
pasture for domesticated livestock, wildlife populations, and tsetse (Adams & McShane,
1992). Through the use of the previously mentioned land use practices, the East African
savannah landscape was modified and humans could more easily avoid tsetse and
trypanosomiasis. Waller (1990) documented that before the disruption of traditional East
Africa savannah societies by European explorers in the early 18005, seldom is there any
mention of problems associated with tsetse by indigenous populations.

2.2.2 Colonialism and the Tsetse Fly

Prior to 1880 roughly 90% of Sub-Saharan Africa was controlled by indigenous
populations (Stock, 2004). In the winter of 1884-1885 a conference was held in Berlin
where various European nations divided up the African continent into colonial territories
and ushered in the age of Colonialism in East Africa. Colonialism had a wide range of
consequences for the African continent, one of which was to alter the traditional methods
of dealing with tsetse. In this section I will discuss the major impacts that colonialism
had on human, wildlife, and tsetse populations, followed by a description of the three
methods (i.e., host destruction, land clearing, and resettlement) that the European colonial
powers implemented in the attempt to control the threat posed by trypanosomiasis.
2.2.2.1 Disease and Depopulation

European Colonialism had many impacts on the ecology of Eastern African

savannah, though by far the largest was the introduction of the rinderpest plague that

19

occurred in the late 19th and early 20‘h centuries. Rinderpest, a virus that infects both
wild and domestic ungulates, originated in Asia and was introduced in Africa by the
Italian Army in 1889 (Ofcansky, 1981). The airborne disease is easily transmitted
through close or indirect contact between infected animals, and was especially virulent to
the ungulate populations of East Africa that had never previously been exposed to the
virus. Estimates of 80 to 90 percent of undomesticated ungulates (Mack, 1970) and 90 to
95 percent of all domesticated livestock died between the initial introduction and the
early 19005 (Rogers & Randolph, 1988). Pastoral communities living in the East Afiican
savannah were devastated. An estimated two thirds of the Maasai in and around the
Maasai Mara died as a result Of starvation during the initial outbreak Of rinderpest
(Nelson, 2003).

The loss of game populations only exacerbated food insecurities, and entire
communities left traditional lands in an attempt to locate food (Musere, 1990).
Widespread emigration, malnourishment, and starvation occurring in East Africa
contributed to the proliferation of human diseases, such as small pox, that mirrored the
rinderpest plague occurring in livestock herds (Ford, 1971). By 1900, missionaries in the
region remarked that the local populations in East Africa were half of their precolonial
levels (Giblin, 1990). With the onset of World War I in 1914 skirmishes between British
and German colonies in Eastern Africa soon became common. Both sides demanded
requisitions in the form of grain, meat, and labor from local populations at a time of
regional drought causing further depopulation (Hoppe, 2003).

With a significant drop in landscape modification by humans, livestock, and wild

grazing animals, woody vegetation was able to grow unabated for over 10 years during

20

the initial riderpest epidemic (Ford, 1971). In the early 20th century the newly created
habitat allowed for wildlife populations to rebound quickly. As the wildlife spread across
the East African savannahs, tsetse followed (Nelson, 2003).

2.2.2.2 Colonial Control Efforts

In the 19205 colonial Officials set forth to control tsetse, as the fly was deemed the
greatest menace to the development of tropical Africa (Rogers & Randolph, 1988). The
British, who had gained control of the majority of the East Africa, initiated efforts to
control tsetse by using three primary methods: control of host populations, land clearing,
and human resettlement.

The approach of controlling fly populations through the control of host
populations was based on the idea that it might be easier to manipulate the host as
compared to the vector (e. g., similar to the idea that it is easier to control tsetse as
compared to the trypanosomes) (Jordan, 1986). One method of controlling of wild host
populations was to hire hunters to eradicate specific tsetse host species, depriving the fly
of food (Hoppe, 2003). Another approach was to create animal proof buffer zones using
a combination of fences and patrolling hunters in an effort to confine host and fly
populations to “wild” areas (Jordan, 1986). Both methods were eventually abandoned
since the cost of maintaining the fences and hiring full time hunting patrols
overshadowed the initial success (Ford, 1971).

Hunting was also often used in conjunction with large scale land clearing projects
that cleared all of the woody vegetation over vast areas. As the land was cleared by
workers, hunters would shoot any wildlife to prevent them from resettling in other

locations (Jordan, 1986). This approach received mixed reviews from various groups

21

within the white community and was labeled a reckless and wonton destruction of the
environment (Rogers & Randolph, 198 8). In an effort to reduce costs and appease the
public, the colonial governments attempted selective and discriminative land clearing
methods (Hoppe, 2003). This type Of land clearing involves removing only the habitat
used by tsetse (i.e., the understory) leaving the tallest trees intact (Jordan, 1986).
Although initially successfiil in ridding an area of tsetse, land clearing strategies only
achieved long term success if woody vegetation was re-cleared on a regular basis (Hoppe,
2003)

An alternative method to re-clearing woody vegetation was the use of human
settlements to suppress the growth Of undesirable vegetation (Ford, 1971). However,
most of the resettlement projects were not voluntary and employed forced labor to clear
the land (Hoppe, 2003). The interference by colonial officials with the communities did
not stop with the initial resettlement, but rather extended into traditional agricultural
practices. One such interference was the use of fire, which was traditionally started in the
late dry season as the fire burned much hotter, incinerated more woody vegetation,
destroyed tsetse breeding sites, and promoted denser grass growth during the upcoming
wet season (White, 1995). Europeans promoted early season field burning, which
promotes more vigorous vegetation grth and was considered not as destructive as late
season fires (Eriksen, 2007). However, the early season fires actually favored the growth
of woody vegetation, which in turn created more tsetse habitat and suppressed grass
growth (Knight, 1971). Colonial governments also initiated the hunting of elephants,
which were deemed a hazard to the resettled communities (White, 1995). Elephants not

only consume green flora as food, they also decrease the amount and density of existing

22

woody vegetation due to their large size and tendencies to damage shrubs and trees (Guy,
1989). Soon elephant populations started to decrease, with a noticeable effect on the
savannah ecosystem. Ultimately the majority of the resettled communities were unable
to maintain the cleared land, which the colonial powers blamed on the indigenous
populations inability to cultivate properly and occupy the land effectively (Hoppe, 2003).

In the 19505 European colonial powers started to relinquish control of their
territories and independent countries run by African leaders were formed. A review of
colonialism and the impact it had on tsetse populations shows that European Officials had
a lack of understanding about the historical balance that existed between traditional
societies and the natural environment. By altering the traditional land use practices of the
indigenous populations, colonial officials expanded the distribution of the tsetse in East
Afiica.
2.2.3 Modern Control Methods

Coinciding with the end of colonialism was the introduction of new methods to
control tsetse in East Africa. Modern methods for managing fly populations include the
development of insecticides, tsetse baits (i.e., traps, targets, and treated cattle), biological
agents, and the sterile insect technique. In addition to new methods of controlling tsetse,
control of the disease rather than the vector though the use of trypanocidal drugs has been
introduced. With a variety of methods at the disposal of management Officials,
eradication versus control has become a hotly debated issue, with fervent supporters on

both sides.

23

2.2.3.1 Insecticides

From its introduction shortly following World War 11 until the early 19905, the
use of insecticides was the most popular choice in combating tsetse in Africa (Allsopp,
2001). At first, persistent organochlorine insecticides (e.g., Dichloro-Diphenyl-
Trichloroethane a.k.a. DDT) were applied indiscriminately and in large dosages in an
attempt to minimize the number of sprayings needed and to maintain cost effectiveness
(Grant, 2001). Environmental concerns about the use of these highly persistent
insecticides were first raised by Du Toit (1954) and later Carson (1962). Further studies
spm'red the switch in the 19705 to less persistent organochlorine insecticides and non-
residual synthetic pyrethroids. These slightly more ecological friendly insecticides were
applied in smaller dosages, a more selective manner, and in sequential spray treatments
timed to kill emerging flies before they were allowed to reproduce (Rogers & Randolph,
1988; Douthwaite, 1992).

Two general methods exist for the application of insecticides: ground spraying
and aerial spraying. Ground spraying, either from a human carried pressurized knapsack
sprayer or vehicle mounted fogger, was at first the most widely used application
technique (Allsopp, 2001). The ground spraying approach allowed managers to either
indiscriminately spray or target specific tsetse refuges depending on the particular goals
of the campaign (Grant, 2001). Aerial spraying is basically crop dusting of savannah
ecosystems. An insecticide dispensing sprayer is attached to the aircrafi, either an
airplane or a helicopter, which then flies low to the ground to insure proper deposition of

the chemicals on the vegetation below (Jordan, 1986).

24

The use of insecticides in the management of tsetse has some distinct advantages.
First and foremost, insecticides immediately reduce fly populations without apparent loss
in ecosystem services available to local human populations. Another distinct advantage
of insecticides is the low monetary cost to area treated ratio, especially when aerial
spraying is used. Therefore the use of insecticides gives management Officials a
relatively inexpensive method of killing a large number of tsetse quickly over huge areas.
Despite the advantages of using insecticides, in the late 19805 / early 19905 their use was
greatly diminished due to environmental concerns, lack of finding, and availability of
new tsetse trap technology (Douthwaite, 1992; Alsop, 1994; Allsopp, 2001; Grant, 2001;
Schofield & Maudlin, 2001; Hargrove, 2003). However, recent advancements in Global
Positioning Systems (GPS) have increased both the accuracy and cost effectiveness of
aerial spraying campaigns, spurring renewed interest in insecticide use in large scale
eradication campaigns (Hargrove, 2003; Kgori et al., 2006; Childs, 2008).
2.2.3.2 Tsetse Fly Bait Technology

The development of baits to attract tsetse was brought about by the desire to have
an efficient way to Obtain an accurate fly census (Vale, 1993). The use of bait technology
as a means of tsetse management was suggested after an improved understanding of
tsetse reproductive rates, which are quite low at roughly one pupa every 9-10 days per
adult female (Leak et al., 2008). This low reproductive rate means that a decline in
isolated fly populations will occur if management officials are able to reduce the female
population by only 2% per day (Langley & Weidhaas, 1986). Tsetse baits that attract
then kill flies have been shown to meet and often exceed the 2% per day reduction of

females, and thus diminish the over all fly population (Vale et al., 1988).

25

Stationary baits commonly use black and blue colors to visually lure the fly from
longer distances and then employ odors to draw the fly in closer (Figure 2.3) (Vale,
1993). Once the fly has reached the device it is either deceived into entering a
containment apparatus (i.e., trap) or it lands on insecticide treated cloth or electrified
metal screen and is killed (Vale & Hargrove, 1979). The odors used to bait tsetse traps
and targets were originally urine collected fiom livestock or tsetse host species, with later
experiments identifying several artificial odor—baits, of which acetone was the best
performer (Vale, 1993). Since Odor-baits are designed to simulate the natural smells that
host species emit, rather than waste resources on the stationary targets, in areas of high
livestock production, cattle have commonly been substituted as tsetse bait (Hargrove,
2003). Insecticide dips and sprays are applied to the skin of cattle, creating a mobile
tsetse target, which has an added benefit of protecting the livestock from other blood

feeding insects (e.g., ticks) (Peter et al., 2005).

 

Figure 2.3: A Nguruman Tsetse Trap. The blue attracts tsetse from long distances,
while the black lures the fly directly underneath the actual trap. When a fly
discovers no actual food is present on the black cloth, it flies up and into the trap
(i.e., the white mesh at the top of the trap). (Photo by Dr. Joseph Messina)

26

The use of bait technologies to attract and then either kill or contain tsetse has
become increasingly popular for being both effective and ecologically friendly (Muzari,
1999; Allsopp, 2001; Schofield & Maudlin, 2001; Hargrove, 2003). Odor-baited traps
have been shown to reduce tsetse populations by 95 to 99% annually, with only 10% Of
the total number of insects killed being non target species (Hargrove & Vale, 1979; Vale
et al., 1988). More than other methods, tsetse baits also offer the Opportunity to involve
local communities in smaller scale management efforts (Alsop, 1994). By training local
human populations to maintain the traps and use livestock insecticide dips, management
officials utilize local labor resources to help sustain cost effectiveness (Schofield &
Maudlin, 2001). Despite the advantages of using baits to manage fly populations, the
technology has not been widely implemented. In part, this is due to thefi of cloth traps
and targets (Hargrove, 2003), lack of external funding (Schofield & Maudlin, 2001), and
local herders using their limited resources to purchase trypanocidal drugs rathervthan
participate in community eradication or suppression campaigns (Barrett & Okali, 1998;
Torr et al., 2005).
2.2.3.4 Biological Agents

The use of biological agents to control tsetse populations implies the use of living
organisms to reduce fly populations (Drew, 1982). One approach to biological control is
to utilize natural predators to kill the target species. Tsetse have several types of potential
vertebrate and arthropod predators that will prey upon both adult and larval tsetse
including: baboons, birds, reptiles, arachnids, ants, and wasps (Austen & Hegh, 1922;

Laird, 1977; Leak, 1999). However, the manipulation of these species seems unlikely to

27

bring about a sustained and cost effective reduction in fly populations without the
introduction of non-native predatory species (Van der Vloedt, 1991).

The use of pathogens has been suggested as a possible means of managing tsetse
(Laird, 1977; Poinar et al., 1979), with fiingi, both sprayed at tsetse breeding Sites and
used in place of insecticide on targets, producing the best results (Kaaya & Munyinyi,
1995). However, using fungi as a biological agent against tsetse results only in a
reduction of fly populations equal to that of insecticide treated targets (Maniania &
Takasu, 2006). Yet the technology suffers from all of the drawbacks associated with
insecticides treated targets, and the additional complication of producing the flingi
(Maniania, 1998), making it a less cost effective method. Further research is needed to
explore the potential of non—native predators and pathogens as methods of managing
tsetse with the understanding that a high degree of caution must be implemented before
introducing any foreign species into a new environment (e. g., Boettner et al., 2000).
2.2.3.5 Sterile Insect Technique

The sterile insect technique (SIT) is similar to the use of biological agents to limit
the growth Of tsetse populations, but rather than manipulating a non-target species the
fly’s own reproduction system is used against it. SIT involves raising a large number of
male tsetse within a laboratory, sterilizing them using low levels of radiation, and then
releasing them to mate with the wild female flies (Vreysen, 2001). The technique relies
on female flies mating relatively few times in their lifespan, and when they do mate it is
with a sterile male, thus ensuring no Offspring will be produced (Leak, 1999).

SIT has the distinct advantage, compared to any of the previously mentioned

tsetse population reduction methods, Of only effecting tsetse populations and not any

28

other species (Offori, 1982). Although SIT itself does not directly affect non-target
species, use any chemical inputs, and is considered environmentally benign, it is
commonly associated with an initial tsetse population suppression phase (e. g., aerial
spraying of insecticides, targets, traps) which does impact the local ecology (Allsopp,
2001; Vreysen, 2001). In addition to the initial suppression phase, SIT also has the
disadvantage of being costly (WHO, 2009). This makes SIT only feasible for isolated
tsetse populations where the goal of the campaign is eradication (Hargrove, 2003).
2.2.3.3 Trypanocidal Drugs

Unlike the previously mentioned methods aimed at controlling tsetse,
trypanocidal drugs or trypanocides (i.e., drugs designed to kill the trypanosomes) attempt
to cure rather than prevent infection. The use of trypanocidal drugs has become the
predominate method of controlling trypanosomiasis in most African countries (Leak,
1999; Van den Bossche & Connor, 2000). Trypanocides come in two classes: human
sleeping sickness drugs (e. g., suramin, melarsoprol, pentamidine, nifurtimox, eflomithine
/ difloromethylomithine or DFMO) and those used on livestock (e.g., diminazene,
homidium bromide, quinapyramine, isometamidium chloride / samorin) (Anene et al.,
2001; Docampo & Moreno, 2003; Fairlamb, 2003; Barrett et al., 2007; Delespaux &
Koning, 2007). Both human and livestock drugs act in Similar manners attempting to kill
of the trypanosomes via creating a toxic environment or limiting particular enzymes
needed by the parasite (Barrett & Barrett, 2000; Anene et al., 2001). Unfortunately the
trypanocidal drugs are also considered highly toxic to mammalian cells, several of which

are also used as human cancer treatments (Barrett & Barrett, 2000; Legros et al., 2002).

29

Despite the negative side effects, the dominance of trypanocides was brought
about by the decreased number or ineffectiveness of large-scale control campaigns (e. g.,
aerial spraying in the early 19905 and traps / targets used at inappropriate scales) and the
relatively inexpensive cost of trypanocidal drugs (Torr et al., 2005; Delespaux & Koning,
2007). The cost to treat cattle is estimated at US$1 per dose (Holmes et al., 2004), and
several trypanocidal drugs have been provided to impoverished individuals diagnosed
with sleeping sickness free of charge by the WHO since 2000 (Barrett et al., 2003;
Barrett et al., 2007), leading to reliance on drugs to treat rather than prevent infection.
The prolonged (>50 years) wide spread use of trypanocides has lead to the evolution of
trypanosomes resistant to particular drugs (e.g., melarsoprol, diminazene, isometamidium
chloride, quinapyramine) (Fairlamb, 2003; Delespaux & Koning, 2007). To combat
trypanosome resistance, combinations of drugs or sanative pairs and higher drug doses
have been implemented in some locations (Anene et al., 2001; Keiser et al., 2001;
Kennedy, 2004; Barrett et al., 2007). However, in lieu of increased trypanosome
resistance, the long term use of trypanocidal drugs to control the spread trypanosomiasis
will depend on the development of new drugs (Anene et al., 2001; Fairlamb, 2003;
Kennedy, 2004).
2.2.3.6 Eradication versus Control

The question of whether or not to eradicate or control tsetse is relatively new.
Traditional East African societies lived in an established balance with tsetse (Adams &
McShane, 1992). By modifying the landscape they imposed the first methods of
controlling fly populations, limiting tsetse to marginal shrubland and forest ecotones

(Ford, 1971). With the introduction of European colonial governments in the early 19005

30

came policies aimed at eradicating rather than just controlling tsetse (Rogers &
Randolph, 1988). The colonial policies primarily involved large scale land clearing, host
destruction, and resettlement campaigns with some local successes (Ford, 1971).

The dawn of the pesticide era ushered in renewed calls for tsetse eradication
(Rogers & Randolph, 1988). Large-scale campaigns using insecticides were
implemented across Africa with multiple claims of regional scale success (e.g., parts of
Nigeria, Zululand in South Africa, the Okavango Delta in Botswana, Lambwe Valley in
Kenya) (Du Toit, 1954; Davies, 1964; Davies, 1971; Turner & Brightwell, 1986;
Hargorve, 2003; Kgori et al., 2006). Yet by the end of the 19805, tsetse had not been
eliminated from Afiica, which gave rise to increased questioning Of eradication strategies
as a whole (see Matthiessen & Douthwaite, 1985 ; Jordan, 1986; Rogers & Randolph,
1988; Giblin, 1990). By the mid 19905 the Spraying of insecticides was greatly reduced
(Allsopp, 2001), and the majority of groups involved in tsetse management no longer
viewed eradication as achievable or desirable (Barrett & Okali, 1998; PATTEC, 2001).

The debate was renewed in the late 19905 with the successful eradication Of tsetse
populations using a combination of targets and SIT on the Tanzanian island of Zanzibar
(Vreysen et al., 2000). The success on Zanzibar helped to create the Pan African Tsetse
and Trypanosomiasis Eradication Campaign (PATTEC), an offshoot of the Programme
Against African Trypanosomiasis (PAAT) international alliance, with such members as
the Food and Agriculture Organization of the United Nations (FAO), International
Atomic Energy Agency (IAEA), WHO, and the African Union lnter-Afiican Bureau for
Animal Resources (AU—IBAR) (Taveme, 2001). PATTEC’S goal, unlike PAAT whose

initial policy was one of control, was to facilitate area-wide (i.e., all of Afiica) tsetse

31

eradication (PATTEC, 2001; Feldmann, 2006). Since the inception of PATTEC, the
majority of tsetse related management programs in Africa have followed eradication
policies, the result of which has been two countries (Botswana and Namibia) claiming
tsetse-free status and an additional ten countries with active eradication campaigns
(PAAT, 2008). Despite the apparent success in PATTEC’S eradication efforts, the debate
over the long term feasibility of tsetse eradication versus control policies continues
(Matima personal communication, 2008).

The current debate over eradication versus control is centered around two key
issues; economics and scale. Proponents of eradication point to the high economic cost
tsetse exact on the human populations affected by trypanosomiasis (Kabayo, 2002;
Kamuanga, 2003), thus framing eradication efforts as a campaign against poverty (UN
Wire, 2002). They also claim in the long run the cheapest way of controlling the disease
is to eradicate the vector (Hocking et al., 1963), and unlike control efforts that require a
continuous source of money and resources, eradication only requires a one time input if
successful. Eradication Opponents counter with arguments stating that the scale at which
eradication efforts are being implemented at, some pockets of tsetse are bound to be
missed / survive and populations will regenerate (UN Wire, 2002). Reinvasion into
tsetse-free areas will follow, just as observed in the past (e. g., Nigeria and Botswana)
(Kgori et al., 2006; Oluwafemi et al., 2007), and the investment in tsetse eradication will

have been wasted.

32

2.3 Modeling of the Tsetse Fly
2.3.1 Remotely Sensed Data and Mapping Vector-Borne Disease

Vector borne diseases in much of the world occupy places difficult to access
using in situ data collection or they operate across spaces too large to easily or effectively
sample. Satellite based sensors allow for synoptic coverage and the routine collection Of
data over these sites and situations. Curran et al. (2000) outline three underlying
premises to justify the use of remotely sensed data in the modeling of vector borne
diseases: 1) remotely sensed data can be used to provide information on land cover /
climate and by association the habitat of species (Innes & Koch, 1998), 2) the spatial
distributions of vector-bome diseases are related to the habitat of the vector (Pavlovskii,
1966), and 3) if these are true, then remotely sensed data can be used to provide
information on the spatial distribution of vector-borne diseases (Hay et al., 1997). For
this reason, remotely sensed data have been used as descriptors in multiple vector-bome
disease modeling research studies (see e.g., Kitron et al., 1996; Wint & Rogers, 2000;
Gilbert et al., 2001; Kristensen et al., 2001; Wint, 2001; Estrada-Pena, 2002; Hendrickx
et al., 2002; Levine et al., 2004; Omumbo et al., 2005; Goodin et al., 2006).
2.3.2 Past Modeling and Mapping of Tsetse Fly

Population density models characterized most early attempts at modeling tsetse.
These early models tended to use linear regression to predict fly population densities
based on climate variables highly correlated with tsetse survival (e.g., Nash, 1933; and
Bursell, 1956). More advanced dynamic population models that expanded upon the
simpler linear regression models were first developed in the late 19705 (Rogers, 1979),

with the most recent population model (i.e., Tsetse Muse) available for downloading at

33

Tsetse.org (Vale & Torr, 2005). These new mathematical density dependant models built
upon the expanding knowledge of tsetse behavior and ecology, and have been used to
help control efforts estimate the needed level of fly suppression to attain the level of
control desired (Artzrouni & Gouteux, 2001 / 2003 / 2006; Hargrove, 2003). However,
these models assumed that suitable habitat and tsetse are present at the modeled locations,
and thus, in effect, only predicted fly population densities in known locations (i.e., they
produce no information on the spatial distribution of tsetse).

The first widely accepted tsetse distribution maps were created in 1977 by Ford
and Katondo (1977). These species maps were based on field work and intimate
knowledge of the African landscapes and tsetse ecology. In 1986, Rogers and Randolph
(1986) at the University of Oxford constructed the first predictive tsetse distribution maps
based off of climate maps of temperature and saturation deficit. By the early 19905,
Rogers and Randolph, along with others from Oxford, were using remotely sensed land
cover data in conjunction with climate data to further refine the ability to identify suitable
tsetse habitat (Rogers & Randolph, 1991; Williams et al., 1992; Rogers & Williams,
1994; Robinson et al., 1997).

In 1993, Rogers, Randolph, and others helped form the Trypanosomaiasis and
Land-use in Africa (TALA) Research Group at the University Of Oxford. In the late
19905, TALA partnered with the Environmental Research Group Oxford (ERGO) and
PAAT to create a spatially explicit model named PAAT — Information System
(PAATIS). PAATIS used discriminant analysis and maximum likelihood statistics on
remotely sensed environmental variables (e.g., NDVI and climate), socioeconomic data,

and the Ford & Katondo (1977) distribution maps, to predicted tsetse distributions at a

34

5km spatial resolution across Afiica (Gilbert et al., 2001). The PAATIS model was later
refined by ERGO, the FAO, and IAEA employing logistic regression on the same
remotely sensed variables as PAATIS and produced 1km spatial resolution percent
predicted tsetse presence for the continent of Africa (Wint, 2001).

Similar to PAATIS and the subsequent refinement by ERGO / FAO / IAEA, the
Agriculture and Veterinary Information and Analysis (Avia-GIS) company produced a
tsetse model for parts Of South Africa in 2002. The model used comparable data (e. g.,
NDVI, temperature, physiographic) and employed Kriging and binary logistic regression
to produce binary presence-absence maps for Kwa—Zulu Natal, South Africa (Hendrickx
et al., 2002). Unlike PAATIS and the ERGO / FAO / IAEA models, the Avia-GIS model
allowed for temporal analysis by incorporating time series remotely sensed data, but no
mention of identifying intra-annual or inter-annual fluctuations was made in the Official

summary report (Hendrickx et al., 2002).

2.4 Current Tsetse Fly Research

In this next section I will briefly summarize current tsetse and trypanosomiasis
research. There are four generalized categories: modeling, genetics, tsetse-trypanosomes
interactions, and control technology. The most recently produced models related to tsetse
have been produced by TALA, ERGO, Avia-GIS, and the Natural Resources Institute
(NR1) at the University of Greenwich (the creators of Tsetse Muse), and are covered in
more detail in the previous tsetse fly modeling section.

Research on genetics focuses on the complete mapping and expressed sequence

tags (EST) of both tsetse and trypanosomes genes. Genetic research is currently being

35

carried out by multiple institutions (e. g., University of Bristol Trypanosomiasis Research
Group, University of Glasgow Department of Clinical Neuroscience, Wellcome Trust
Sanger Institute, Christoffels Lab at the University Western Cape South African National
Bioinformatics Institute, Rikagaku Kenkyfisho (RIKEN) Advanced Science Institute, and
University of Cape Town), several of which contribute to the VectorBase invertebrate
vector genomics project and the International Glossina Genome Initiative (IGGI). The
goal of tsetse and trypanosome genetic research is a complete knowledge of the DNA and
RNA of the host and microbe so possible transgenic control efforts can be explored.

Similar to the genetic research being preformed is the work being done examining
the relationship between tsetse and the trypanosomes. This research focuses on the
evolution of symbiosis and host-pathogen interactions, with the hope that a method of
blocking the development of trypanosomes within tsetse can be discovered. The majority
Of the institutions performing tsetse-trypanosomes interaction research work in
conjunction with the IGGI include the Aksoy Lab at Yale University School of Public
Health, the Lehane Lab at the Liverpool School of Tropical Medicine, the Institut de
Recherche pour le Développement / Centre International de Recherche Développement
sur l’Elevage en zone Subhumide (IRD-CIRAD) mixed research unit, the Van Den
Abbeele Lab at the Institute of Tropical Medicine Antwerp, Christoffels Lab at the South
African National Bioinformatics Institute, the Rio Lab at West Virginia University
Department of Biology, and the Animal Health Programme (AHP) within the Untied
Kingdom’s Department for International Development (DFID).

The last area Of active research involves the development Of new control

technologies. One such avenue of research being investigated by the International Centre

36

of Insect Physiology and Ecology (ICIPE) focuses on the development of tsetse repellants
developed fi'om animals that are naturally avoided by tsetse (e.g., Zebra and Waterbucks).
Another is the breeding of trypanotolerant cattle is being investigated by International
Trypanotolerance Centre (ITC) and Donnan Laboratories within the University of
Liverpool’s School of Biological Sciences in conjunction with the International Livestock
Research Institute (ILRI). The hope is to develop a breed of cattle tolerant of
trypanosomiasis similar to wild host populations. One promising avenue of research into
control technologies is the development of new a trypanocidal drug (pafuramidine) by the
University of North Carolina at Chapel Hill Consortium for Parasitic Drug Development
(CPDD) (Barrett et al., 2007). The CPDD includes partners from Georgia State
University, Swiss Tropical Institute, London School of Hygiene and Tropical Medicine,
Ohio State University, University of South Florida, University of Glasgow, Gorgas
Memorial Institute (Panama), Kenya Agricultural Research Institute, and Immtech
Pharmaceuticals Inc., and is funded by the Bill & Melinda Gates Foundation (Tidwell,
2006). The Bill & Melinda Gates Foundation also funds research into a trypanosomiasis

vaccine (Ratajczak, 2005), but to date no vaccine has been developed.

37

CHAPTER 3
METHODS

3.1 The Tsetse Ecological Distribution Model

The Tsetse Ecological Distribution (TED) Model is a 250m resolution, raster-
based, spatially explicit dynamic model that predicts the presence or absence of tsetse
flies at 16 day intervals over a 4 year period between 1/1/2001 and the end of 12/31/2004,
based on habitat suitability and fly movement rates. At its simplest, the TED Model can
be described in two separate parts: 1) a spatially explicit model that identifies suitable
tsetse habitat and 2) a fly movement model that utilizes tsetse distributions and fly
movement rates (Figure 3.1).
3.1.1 Habitat Suitability Model

Based on the ecological needs of tsetse, a spatially explicit model that predicts the
location and timing of suitable tsetse habitat should include variables that represent land
cover, hosts, moisture, and temperature. However, in Kenya no accurate intra-annual
data on host distributions currently exist. For this reason, host populations are not
included for in the TED Model habitat model.

The habitat suitability model uses four remotely sensed data sets: 1) day land
surface temperature (LST), 2) night LST, 3) Normalized Difference Vegetation Index
(N DVI) as a surrogate for available moisture (e.g., Williams et al., 1992), and 4) a Land
Use Land Cover (LULC) data set. Each of the four habitat variables was classified to a

66199

binary suitable vs. unsuitable classification scheme, with a value of assigned to

suitable areas and a value of “0” assigned to unsuitable areas. The four binary suitability

38

maps based on the habitat variables were then combined to create an overall suitable
tsetse habitat map every 16 days. The determination of suitable versus unsuitable
threshold values for each variable will be explained firrther when the parameterization of

each variable is discussed.

The Tsetse Ecological Distribution (TED) Model

—

lepow hiquarns 1'4!qu

 

 

 

lepow rueureaow 11),)

Figure 3.1: The Tsetse Ecological Distribution (TED) Model structure. The flow
chart represents one run of the TED Model, with blue circles representing driving
variables, yellow squares representing conversions and calculations, green ovals are
derived auxiliary variables, the red oval is the model output for the scene. The
dotted line represents the starting tsetse distribution, which was only used during
model initialization.

39

3.1.2 Fly Movement Model

Although the first part of the model will determine the locations of suitable tsetse
habitat, it does not address where the flies are actually located at various times of the
year. As the Kenyan landscape changes, so should the amount of suitable tsetse habitat.
However, the expansion of suitable habitat might be greater than tsetse movement rates,
and therefore not all suitable habitats will have tsetse present.

To identify the location of tsetse distributions in response to changes in suitable
habitat, a simple fly movement model was coupled to the habitat suitability model. The
fly movement model expands the previous tsetse distributions by an assigned fly
movement rate. The expanded tsetse distributions are then combined to the new suitable
habitat model output to identify where tsetse distributions could expand if suitable habitat
is present. Due to the starting distribution being externally or artificially generated, a one
year initialization (i.e., 2001 to 2002) was run before any predicted tsetse distributions

from the TED Model were used for analysis purposes.

3.2 Model Initialization
3.2.1 The Identification of the Optimal Land Cover Product

The TED Model uses an externally generated LULC product to account for land
cover in the tsetse habitat model. Since fifteen LULC products are publicly available for
Kenya, I analyzed each product to identify which should be used before the TED Model
could be run. Rather than relying on reported accuracy assessments, not always available
for each LULC product and expensive or impossible to perform post-production, I

developed a generalizable method for identifying the optimum land cover products to be

40

used in the TED Model. The first step in my method involves creating a binary habitat
suitability map for each of the LULC products being examined. This entails identifying
the land cover classes that contain woody vegetation and overlaying them with
environmental variables (i.e., elevation and precipitation) to create a binary suitability
map. The binary suitability maps are then compared to some form of species distribution
ground truth data using a modified Mapcurves Goodness of Fit (GOF) test (see Hargrove
et al., 2006). Once the GOF scores are calculated, a simple t-test is used to identify
LULC products with significant scores.

3.2.1.1 Data

Fifteen public LULC products (Table 3.1) are available from sources including
National Aeronautics and Space Administration (NASA), International Geosphere-
Biosphere Programme (IGBP), the FAQ, The Global Environment Monitoring Unit at the
University of Maryland (UMd), and the Climate Land Interaction Project (CLIP) located
within the Center for Global Change and Earth Observations at Michigan State
University. All Of the LULC products used in this analysis were originally in or
converted to a raster format with a spatial resolution Of 1km or 500m, and cover the
entire country of Kenya. Each LULC data set is unique based on its production methods,
classification scheme, temporal acquisition date, and intended use.

The IGBP DISCover land cover product produced by the United States
Geological Survey (USGS) Land Cover Working Group in 1995 was created using the
Advanced Very High Resolution Radiometer (AVHRR) NDVI 10 day composites from
April 1992 to May 1993 (Hansen & Reed, 2000). The land cover classes were

determined using unsupervised classification on the AVHRR NDVI data on a continental

41

scale (Loveland et al., 2000). The accuracy of the IGBP DISCover land cover product
has been estimated at 66.9% for overall area weighted accuracy, and an accuracy range of

40% to 100% for individual classes (Scepan, 1999).

Table 3.1: The land use land cover data sets that are publicly available for Kenya
(each type MODIS of product is sub divided into 500m and 1km data sets).

 

 

 

. Classes Temporal
Data Set Resolution Classrfication Scheme in Kenya Range Platform
Africover 1:200,000 Regional FAO LCCS 29 1995 LANDSAT
Combination of 1995 /
CL'vae' 1”" GLC2000 and Africover 43 1999 — 2000 NA
GLCZOOO 1km FAO LCCS 22 1999 - 2000 SPOT 4
IGBP
DISCover 1km IGBP 16 1992 — 1993 NCAA
UMd .
GLCC 1km UMd modified IGBP 11 1992 - 1993 NCAA
Produced
1km IGBP 16 Annually “frag?
MODIS 2001 - 2004
Type 1 Produced MODIS
500m IGBP 17 Annually Terra &
2001 — 2005 Aqua
Produced
1km UMd modified IGBP 14 Annually “$22?
MODIS 2001 — 2004
Type 2 Produced MODIS
500m UMd modified IGBP 14 Annually Terra &
2001 — 2005 Aqua
Produced
1km LAI / FPAR 9 Annually "£2?
MODIS 2001 — 2004
Type 3 Produced MODIS
500m LAI / FPAR 11 Annually Terra &
2001 — 2005 Aqua
Produced
1km Net Primary Production 9 Annually “4'22?
MODIS 2001 — 2004
Type 4 Produced MODIS
500m Net Primary Production 9 Annually Terra &
2001 — 2005 Aqua
Produced
1km Plant Functional Type 11 Annually “£2?
MODIS 2001 — 2004
Type 5 Produced MODIS
500m Plant Functional Type 12 Annually Terra &
2001 - 2005 Aqua

 

42

 

The Global Land Cover Facility at the UMd produced the UMd Global Land
Cover Classification (GLCC) LULC data set utilizing the same underlying remotely
sensed AVHRR NDVI data as the IGBP DISCover land cover product, but employed a
decision tree classification method resulting in a different classification scheme (Hansen
et al., 2000). As explained in Hansen and Reed (2000), the major difference between the
IGBP DISCover and the UMd GLCC classification schemes is the exclusion of
permanent wetlands, cropland / natural vegetation mosaic, and ice / snow in the UMd
GLCC product. No formal accuracy assessment has been performed on the UMd GLCC
product, though the reported agreement between the UMd GLCC product and the IGBP
DISCover is 74% (Hansen & Reed, 2000).

This study also used all five types of Moderate Resolution Imaging
Spectroradiometer (MODIS) Global Land Cover products in both 500m and 1km spatial
resolutions (MCD12Q1 & MOD12Q1), which are publicly available from NASA.
Although both spatial resolutions of each type of MODIS Global Land Cover product are
produced using the same classification method and scheme, the resulting 500m and 1km
data sets are quite different in the patterns of land cover classes that they display (Figure
3.2). For this reason, each resolution of each type of MODIS LULC product is

considered a separate data set in our analysis.

43

 

Figure 3.2: 2001 MODIS Type 1 Global Land Cover 500m (A) and 1km (B)
resolution. The classification scheme is simplified to highlight the differences
between the two data sets despite the same classification methods. The “Woody
Vegetation” (depicted in the color red) is a generalized class comprised of mixed
forest, shrubland, and savannah land cover, which are considered most suitable
tsetse land covers.

The MODIS Global Land Cover products were produced annually from 2001 to
2004 for the 1km data, and 2001 to 2005 for the 500m data. Only the 2001 data are
analyzed here as they are the closest match to the production dates of the validation data.
The MODIS Type 1 product is produced using MODIS NDVI data and the same IGBP
global vegetation classification scheme as the IGBP DISCover land cover product
(Friedl, 2002). MODIS Type 2 uses the UMd modified IGBP scheme and methodology
and the same MODIS NDVI data used to create the MODIS Type 1 land cover product
(Zhan, 1999). The MODIS Type 3 land cover product is derived from known

relationships between estimated leaf area index (LAI) and fraction of photosynthetically

active radiation (FPAR) (Tian et al., 2000). The MODIS Type 4 land cover product is

44

derived from the net primary production (NPP) MODIS products, which measure the
growth of the terrestrial vegetation. The MODIS Type 4 classification scheme is
primarily geared towards the identification of forest types, such as deciduous broadleaf
vegetation and evergreen broadleaf vegetation. The MODIS Type 5 land cover product
was designed to be used in the Community Land Model for the purposes of climate
modeling, and focuses on classifying land cover type based on the plant functional type
or plant biome.

The Global Land Cover 2000 (GLC2000) product was produced by the Joint
Research Centre Global Vegetation Monitoring Unit and using a 14 month (November,
1999 to December, 2000) of VE GE T A TION imagery from the SPOT-4 satellite (Torbick
et al., 2006). The classification scheme used by GLC2000 was the Land Cover
Classification System (LCCS) designed by the FAO (Di Gregorio & Jansen, 2000).
Mayaux et al., (2006) have estimated that the overall global accuracy of the GLC2000
data set is 68.5 i 5 %. In this study, a 26-class African version of GLC2000 was used; 22
of those classes are found within Kenya.

The LCCS was originally developed to aid in the production of the Africover
product (Di Gregorio & Jansen, 1998). Africover was created by combining both
computer based unsupervised classification and an expert system supervised
classification performed by visual interpretation of mid-19905 era Landsat images by
local experts (Torbick et al., 2006). Several Africover products exist; for this study the
spatially aggregated Kenya-specific product was used. The original'Kenya-specific
Africover product is in vector format, with 105 LULC classes, and has a nominal scale of

l:200,000. The 105-class Africover vector data set was converted into a raster data

45

structure with a 1km spatial resolution, using the highest maximum combined area of all
LULC classes found within a grid cell to determine the final raster cell class. To deal
with the mixed LULC classes frequently found within the Afiicover product, the LCCS
Code 1 class (i.e., the predominate LULC class in each polygon) was assigned as the
overall polygon class. This method reduced the number of LULC classes found in Kenya
from 105 to 95, eliminating particular classes not often found or with small surface areas
(e.g., snow).

The final LULC data set examined in this study is CLIP Cover produced by the
CLIP project at Michigan State University. The CLIP Cover LULC product is a hybrid
of GLC2000 and Africover land cover products and essentially uses Africover
agricultural data where available and GLC2000 non-agricultural land cover data (Torbick
et al., 2006). CLIP Cover was only produced for East Afiica.
3.2.1.3 Suitability Classification

The determination of whether or not a class in a LULC data set contained both the
correct type and quantity of woody vegetation suitable for tsetse was based on the
methods outlined in Cecchi et al. (2008), which entails examining class descriptions
found in the LULC product’s metadata or user manuals and comparing it to published
land cover requirements (Table 3.2). Once suitable tsetse land cover classes were

identified, the LULC data sets were classified into binary suitable vs. unsuitable maps.

46

Table 3.2: MODIS Type 1 LULC classes and their suitability classification.

 

 

 

Suitable Area of
Cl'gss 3:: Class Description Tsetse Ken a
Land Cover (km )
0 Water Fresh or saline water body No 12,825
Evergreen .
A landscape dominated by
1 needleleaf trees more than 2 meters tall Yes 503
forest
Evergreen .
A landscape domlnated by
2 broadleaf trees more than 2 meters tall Yes 15’6”
forest
Deciduous .
A landscape domlnated by
3 needleleaf trees more than 2 meters tall Yes 1
forest
Deciduous .
A landscape domlnated by
4 broadleaf trees more than 2 meters tall Yes 899
forest
Mixed A landscape dominated by
5 forests trees more than 2 meters tall Yes 716
Closed A landscape dominated by woody
6 shrublands vegetation no more than 2 meters tall Yes 20’998
Open A landscape dominated by woody
7 shrublands vegetation no more than 2 meters tall Yes 207’803
Woody .
8 savanna s A mosaic of grass, trees, and shrubs Yes 42,972
9 Savannas A mosaic of grass, trees, and shrubs Yes 122,514
10 Grasslands P'ima'y ”99‘3““ is grass No 97 005
or grass-like plants ’
11 Permanent A permanent mosaic of water, Yes 436
wetlands herbaceous, and woody vegetation
Lands primarily used for
12 Croplands agricultural purposes No 18,536
13 ”Pa." and Human built environment No 1,295
bUIlt-up
Cropland/
natural A mosaic of cropland, trees,
14 vegetation shrubs, and grasslands No 13’461
mosaic
Barren or . .
Any land surface Wlth little or no
16 sparsely . No 31 .448
v e 99 tate d vegetatlon (e.g., sand / rock / salt pans)

 

47

 

For the purpose of identifying the optimum LULC product to be used in the TED
Model, the LULC binary suitability maps were combined with environmental variables.
Following Leak (1999), I used 500mm as a proxy for the minimum annual level of
precipitation for tsetse survival (Figure 3.3), and a maximum elevation of 2200m as'a
surrogate for minimum temperature. The resulting binary suitability maps based on land
cover, elevation, and precipitation suitability were then combined to create an overall

suitability map for each of the fifteen LULC data sets (Figure 3.4).

B

   

Figure 3.3: Map A is a 1km resolution annual precipitation data set from
WorldClim for the year 2000 (Hijmans et al., 2005). The WorldClim precipitation
data set was classified to create Map B, a binary precipitation suitability map.

48

Binary Suitability Maps Based on Environmental Variables

MODIS 1km type 1

   

Figure 3.4: The binary suitability maps created when Africover and MODIS 1km
Type 1 were combined with classified elevation and precipitation data.
3.2.1.4 Validation

The lack Of publicly available country-wide tsetse census data meant that an
alternative ground truth data set had to be identified. The first source of ground truth data
used in this study was a 1996 fly belts map (see Figure 1.2) produced by the former
Kenyan Trypanosomiasis Research Institute (KETRI), now known as Kenya Agricultural
Research Institute Trypanosomiasis Research Centre (KARI-TRC) (Muriuki et al., 2005;
KETRI, 2008). This map represents the most recent field data on tsetse distributions and
shows the general location of tsetse fly belts across Kenya.

A previous study performed by Cecchi et a1. (2008) used the 5km PAATIS maps
as best available tsetse distributions data. I chose to use the higher spatial resolution 1km
FAO / IAEA distribution maps as a second source of ground truth data. The FAO / LAEA

tsetse species distributions maps were produced using logistic regression models, with

49

variables such as NDVI, land surface temperature, infrared reflectance, vapor pressure
deficit, air temperature, surface rainfall, elevation, slope, and potential evapotranspiration
(Wint, 2001). The classification scheme of each map displays the predicted percent
probability of a particular fly species being found at any given time. Here, the
distribution maps of the four Glossina sub-genus species (austem', morsitans, pallidipes,
and swynnertoni) were combined to create a morsitans group distribution map for all of
Kenya. The combined F AO / IAEA morsitans distribution map was produced using the
mosaic tool in ArcGIS, with the maximum mosaic method to ensure that species with the
greatest probability would be reported as the pixel probability (Figure 3.5).

The use of the combined FAO / IAEA morsitans distribution map as validation
data did pose the problem of using a classification scheme dissimilar to the fifteen LULC
binary suitability maps (i.e., percent probability versus binary habitat suitability).
Although the classification schemes appear to be different, the variables used to produce
the FAO / IAEA distribution maps were ecological suitability variables, and therefore the
probability of presence is based on habitat and a direct comparison is reasonable. In
order to account for the differences posed by the percent classification scheme of the
combined F A0 / IAEA distribution map and the binary LULC suitability maps, an
extension of the Mapcurves GOF method of comparison was developed (see Hargrove et

aL,2006)

50

Combined FAO / IAEA Distribution Map

      

{Us

3

. ..;,,;
g.) r ‘
it"

s,

 
    

  
  
    
   

‘1 .3 ‘

'l

 

Probability of Tsetse
Fly Presence

'100%
» 0%

Figure 3.5: The Food and Agriculture Organization of the United Nations /
International Atomic Energy Agency combined moristans tsetse species group
distribution map.

Kilometers

0 50 100 200

51

3.2.1.5 Mapcurves Goodness of Fit Test

The Mapcurves GOF score is a measure of the degree of spatial concordance
between classes of categorical maps with higher Mapcurves GOF scores indicating
higher agreement between classes (Hargrove et al., 2006). The calculation of a
Mapcurves GOF score is not limited by differences in resolution, number of classes, or
data format, but rather that two maps being analyzed overlap spatially and that the
amount of spatial overlap can be measured. One method for calculating the amount of
Spatial overlap between the classes of two data sets is through the creation of a cross
tabulation matrix (Foody, 2007). The tabulation matrix displays the classes of one data
set as rows in the table, and the classes of the other data set as columns (Pontius &
Cheuk, 2006), and therefore the matrix is comprised of the degree of spatial overlap
between the individual classes of the two data sets being compared. Figure 3.6 shows
two example categorical data sets and the cross tabulation matrix that constructed in the
first step of the Mapcurves GOF analysis.

The resulting cross tabulation matrix table is used to create a weighted ratio
comparison matrix. The weighted ratio comparison matrix is constructed by taking the
area of two intersecting categories divided by the total area of the Map 1 category, which
is then multiplied and weighted by the intersecting area divided by the total area of the
Map 2 category. By weighting the proportion of spatial overlap for Map 1 by the
proportion of spatial overlap of Map 2, distortion caused by the presence of large, but
minimally intersecting categories, is prevented (Williams et al., 2008). Each cell within
the matrix displays the GOF ratio for the intersecting Map 1 and 2 categories in the

associated rows and columns; this information can later be used to determine the best

52

reclassification scheme depending on which map is identified as the reference map. The
summing of the rows and columns of the weighted ratio comparison matrix will yield the
GOF score of each class category contained in both Map 1 and Map 2 (Figure 3.7). This
information can be used to determine the degree of concordance between categories of
the two maps, and is used to create a cumulative ratio frequency distribution.

The overall Mapcurves GOF score is the integration of a cumulative ratio
frequency distribution (Figure 3.8). Hargrove et a1. (2006) used 0.02 as the threshold for
the cumulative ratio frequency distribution, and 0.02 was used here as well. The
cumulative ratio frequency distribution shows the declining ratio of map categories on the
y-axis that still satisfy a GOF Mapcurves score on the x-axis. Once the cumulative ratio
frequency distribution has been created, a simple integration will yield the Mapcurves
GOF score. This process is then completed for both directions in order to determine
which direction has the higher Mapcurves GOF sCore and therefore the correct direction
for reclassification if the reference map has yet to be determined. The direction that
yields the highest Mapcurves GOF score is considered to be the best mathematical fit and
is considered the reference map. A Mapcurves GOF score of 1.00 represents 100%
agreement between the two maps being analyzed (i.e., they are the same map); a low
Mapcurves GOF score (e.g., 0.10) is indicative of a high degree of disagreement between

the maps.

53

ism—«E. hOO mot—.232 2: he no: 3...: 2: E 18.2.5.3 2. 2.5.5

:2: 5.5.2: Sta—=83 $98 2: he 5535832 a a. coma—am... 2:3 2:. 45:59:23 EU 2:
55...» 323.5 5.53: 552.55 “8.5 2: no 55358.39. .25? a a. _= 29:53— .:onoana "—00
$3333). 05 aim: 3.59:3 on 3 £3. 33. 32.53qu 29:83 9:. = can _ 29:83— 66 Ping

 

 

 

 

 

 

 

 

 

 

. 60..< .88 .88

as... .88 . a. a banana .6 w . o. 0 bases 3 w as... .38
.38

m ”3an .358. o .25 .1558. u can m
Go w m be no.“ mascots“... m no 33 9.3823...
_

< “was .9353. a can 1.3.9.65 o was <
h o w < Co 3.6 93322:. < Go 3.3 9.83.3...

as... .58 o o r388

.=_ .= .—

 
  

 

 

 

 

54

6.88 .mOO 83:99.:
a .3 5523—3 2: .5. 5.52: EEEQEeo 3:: SEE?» a me 29:38 .2. "En Bani

 

 

 

 

 

o bemoan“. ._o w 0 Demons... .0 w 9.00m “.06
m e328 1.993.258 . .swso . Sassom. m
.o w .. . 39m .3258 . .. . . .sm .s...s...0m .
< roman... .sso .15....9.‘ . 1.90.3953 v <
.ow ...ss<.§efo<. Leafstfoi
28m now a 0 2.0.8.8

 

 

 

 

 

55

6.88 hOU mot—8:32 a .8 5523—3 2: .8.—
933 5:232:15 :otsatE—v 5:259... 3:: 952.559 « .3 29:38 =< £6. 25E

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N as: .o . an: .o
5:353... 5.3535 n 23m ".09 3:32.22
be 5.6 he .55
cod cod ood cod 8..
.m:_m> :5 .908 “.00 mod
9: S um__q_=:E m5 m>onm .0 am
mmtommfio 9m 55 mmtommfio
amE ho 03mm amE 8 25¢ mod
Nod «06 co; co; ood
23m
as an an an
N s. _. .2 u s. v .2 ”.00
22353:. 5:252“.

 

 

 

56

Once Mapcurves GOF scores have been calculated for both directions, and the
reference map has been determined, then the weighted ratio comparison matrix (Figure
3.7) is used to determine the GOF between individual classes and how to best reclassify
the target map based on the classification scheme of the reference map. The
reclassification of the target map is implemented by first identifying the highest
Mapcurves GOF score in each category’s associated row or column in the weighted ratio
comparison matrix, then that category is reclassified based on the corresponding class in
the reference map.

For this study, the fifteen LULC binary suitability maps were compared to the
combined FAO / IAEA distribution map and the 1996 fly belts map using the Mapcurves
method. To facilitate the comparison between the F AO / IAEA distribution map and the
binary suitability maps, the percent classification scheme of the FAO / IAEA map was
classified into a categorical map. A bin value of 0.02 was selected, reclassifying the FAO
/ IAEA distribution map into a SO-class categorical map (i.e., 2% percent probability per
class). A one tail t-test, with a significance level ofO. 1, was used to detect if any of the
LULC data sets had significant levels of agreement with the two ground truth maps.
3.2.2 Temporal Initialization

The TED Model predicts the spatial distribution of tsetse within Kenya at a 16
day temporal resolution between the beginning of 2001 and the end of 2004 (i.e., 92
unique tsetse distribution maps). The 23 scenes produced during the first year (i.e., all of
2001) were used only for model initialization in an attempt to establish potentially viable

tsetse populations that could survive any fluctuations in suitable habitat observed. The

57

remaining 69 scenes, produced between the beginning of 2002 and the end of 2004, were
used for analysis purposes (e. g., identifying tsetse reservoirs and model validation).
3.2.3 Starting Distribution

Three starting distributions were considered for the initialization of the TED
Model. The first was the 1973 fly belts map produced by Ford & Katondo in 1976. This
distribution map is considered by some (e. g., Wint, 2001, Hendrickx et al., 2002) to be
the most detailed tsetse distribution map produced; however, it is over 30 years old and
has not been updated to reflect any possible changes in tsetse distributions.

The second distribution data set available for initializing the TED model was the
1996 KETRI fly belts map. The use of the 1996 KETRI fly belts as the initial tsetse
distribution in the TED Model seemed to be the most logical choice since it was the most
recent data available. When compared to the Ford & Katondo 1973 fly belts maps, the
1996 KETRI map shows general expansion of tsetse distributions across all of Kenya,
especially around Lake Victoria in Western Kenya (Figure 3.9). However, the 1996 fly
belts map excludes several pockets in Northern and Southern Kenya and along the Tana
River that were present according to the Ford & Katondo’s 1973 map and are possibly
still in existence. The exclusion of the Tana River tsetse populations are of particular
interest since in 2002 it was estimated that 24% of all cattle in that region were infected
with Nagana (Catley, 2002).

The third option was to initialize the TED Model with all of Kenya as being
infested with tsetse. Doing this is obviously not an accurate representation of current
reality, but has the potential to identify any tsetse populations that could have been

established in the five years following the creation of the 1996 fly belts map. All three

58

starting distributions have both advantages and disadvantages. To identify which starting
distribution should be used to initialize the TED Model, the affect that each starting
distribution had on the predicted tsetse distributions was analyzed for the inclusion /
exclusion of particular tsetse populations (e. g., around the Tana River) and a

reexamination of my research objectives in using the TED Model.

The Difference Between
1973 and 1996 Tsetse Fly Belts

  
  

 
 
  

 

C
\Noflhem /'

Pockets

 

 
 
 
  

- Ford 8- Katondo 1913
- KETRI ms Klimt."

' O 50 100 200

Figure 3.9: Tsetse distributions in 1973 (Ford & Katondo, 1977) and 1996 (KETRI,
2008). A comparison between the two tsetse distributions shows a general expansion
of tsetse throughout Kenya, especially in Western Kenya around Lake Victoria,
during the intervening 23 year period. A closer examination of the 1996 fly belts
shows that several pockets in the Northern and Southern Kenya and along the Tana
River that were present in 1973 have since disappeared or were excluded during the
1996 map production.

59

3.3 TED Model Data and Parameters Values
3.3.1 Normalized Difference Vegetation Index

Available moisture, especially saturation deficit, has been highly correlated to
tsetse survival (Nash, 1933; Bursell, 1956; Rogers, 1979; Hargrove, 2001).
Representing moisture in the TED Model is problematic since no publicly available in
situ, remotely sensed, or modeled measure of available moisture with a temporal
resolution greater than 16 days currently exists for Kenya. Predicted humidity data sets
based on precipitation and temperature data can be derived, however, a study by
Williams et al. (1992) demonstrated that remotely sensed NDVI data used as a surrogate
outperformed predicted humidity data with regards to modeling tsetse. NDVI, which is a
measure of reflectance / absorption of Photosynthetically Active Radiation (PAR) and
Near-infrared (NIR) wavelengths, essentially measures the presence and condition of
green vegetation (Lillesand et al., 2004). If green vegetation is present and healthy,
remotely sensed NDVI values will be higher than if vegetation is not present or unhealthy
(e.g., brown or dormant). The findings of Williams et al. (1992) that NDVI can be used
as a surrogate for moisture data logically follows since healthy green vegetation generally
requires water to exist. For this reason the TED Model uses NDVI as a surrogate for
available moisture when predicting suitable habitat.
3.3.1.1 Data

NDVI data were acquired from NASA in the form of the MODIS Terra NDVI
Vegetation Indices 250m (MOD13Q1) product. The MODIS NDVI product is available
at 16 day increments. Since the study covered a four year period including the one year

initialization (i.e., the beginning of 2001 to the end of 2004) and the MODIS data are on

60

16 day repeat cycle, 138 scenes of the MODIS NDVI from January of 2001 to December
of 2004 were downloaded, compiled, reprojected, and clipped to the borders of Kenya
using the MODIS Reprojection Tool (MRT) and ArcGIS 9.2.
3.3.1.2 Suitability Classification

Multiple models have used NDVI as a variable to predict suitable tsetse habitat in
the past (e.g., Rogers & Williams, 1994; Gilbert et al., 2001; Wint, 2001), but these
models use NDVI as a land cover descriptor and not as a surrogate for climate variables.
Therefore, the NDVI threshold values reported by these models are not considered
appropriate for the TED Model. The previously mentioned study by Williams et a1.
(1992) that used NDVI as a surrogate for humidity reported a dry season NDVI threshold
value of 0.39 based on several statistical tests including a linear regression, a non-linear
regression, discriminate analysis, k — nearest neighbor analysis, and experiments using a
neural network. Thus, the TED Model was parameterized using 0.39 as the threshold
value for NDVI suitability.
3.3.2 Land Surface Temperatures
3.3.2.] Data

Temperature data were acquired from the NASA in the form of the MODIS Terra
Day and Night LST 1km (MOD11A2) V004 products. The MODIS LST products are
available at 8 day increments; however, to match the same temporal resolution as the
MODIS NDVI product (i.e., 16 days) only scenes with the same date as the NDVI were
used. The LST products were downloaded and processed in the same manner as the
MODIS NDVI product, with an additional step of filling “no data” gaps caused by the

presences of clouds. These gaps were filled by using the inverse distance weighted

61

(IDW) interpolation (Li, 2004). Although IDW may not produce the most accurate
results when compared to using other methods of interpolation (e. g., spline, kriging,
regression, etc.), IDW was used to both reduce the amount of method introduced
uncertainty into the LST data and processing time. Three scenes (year 2002, ordinal date
81, LST day; year 2002, ordinal date 113, LST night; year 2003, ordinal date 113, LST
night) were determined to have too large of data gaps to perform an accurate interpolation
on. For these three dates, an average of the scenes 8 days before and after was used to
account for the missing data.
3.3.2.2 Suitability Classification

It is generally accepted that temperatures above 36°C and below 17°C will greatly
hinder tsetse normal activity (Mellanby, 1936 / 1939; Knight, 1971; Hargrove, 1980;
Pollock, 1982b; Leak, 1999; Terblanche et al., 2008). However, tsetse have adopted
several behaviors to cope with temperature extremes (e. g., seeking out of micro habitats).
For this reason a 4°C buffer was added to the maximum day and night temperature
thresholds to account for this behavior (Torr & Hargrove, 1999; Muzari & Hargrove,
2005). The threshold for night time minimum temperatures was also altered from the
accepted 17°C to a threshold of 10°C to account for the diurnal nature of tsetse. At night
a fly can tolerate lower temperatures since it has no need to be active at this time,
although below 10°C tsetse start to experience detrimental biological effects and much
reduced survival rates (Nash, 1933; Ford, 1971; Terblanche et al., 2008). The end result
was the classification of the LST data to suitable if a pixel had a day LST value between

17°C and 40°C, and a night LST value between 10°C and 40°C.

62

3.3.3 Fly Movement Rates

The rate of tsetse movement can be described in two ways: 1) maximum daily
movement rate of an individual fly or 2) advancement of tsetse populations as a fly front
(i.e., areas of high tsetse population densities with lower population densities preceding
the fly front) (Figure 3.10) (Vale & Torr, 2005). An individual fly can move up to 800m
per day (Vale et al., 1984), while fly fronts tend to advance at a slower rate of 11.6km per
year or roughly 31m per day (Hargrove, 2000). Since the TED Model is designed to
predict tsetse distributions as a whole, the potential movement rate of an individual is less
important than the movement rate of the overall population. Thus a movement rate of

31m per day, or in the case of the TED Model 500m every 16 days, was used.

 

 

 
 
  

 

 

 

 

um .
.33 E
U) .
C .
o :
o :
c :
.2 : ‘c‘
:«3 er
3 i A
a. : ~
0 g I“
9- :
0 :
U) .
uh! I
a) u
m :
I- :
Low . _
Distance

Figure 3.10: A graphical representation of a fly front adapted from Vale & Torr
(2005). The fly front advances more slowly than individual pioneering flies, leading
to lower tsetse population densities preceding the fly front.

63

3.4 Outputs of the TED Model

The TED Model produces 92 unique tsetse distribution maps or scenes that
predict the spatial distribution of tsetse within Kenya, at a 16 day temporal resolution
between 1/1/2001 and 12/31/2004. Due to the one year initialization period (i.e.,
1/1/2001 to 12/31/2001), only the 69 scenes between 1/1/2002 to 12/31/2004 are used for
any analysis purposes. The 69 tsetse distribution maps were combined into a percent
probability of tsetse presence map. The percent likelihood of tsetse presence was created
by summing all of the 16 day distributions for the three year period and dividing by the
number of scenes (69). This map could then be used to compare to existing ground truth
data and quantify the sensitivity of each variable in the TED Model.

The identification of tsetse reservoirs was accomplished by examining the area
predicted by each of the 69 tsetse distribution maps sequentially and determining the
scenes in each of the three years that had the smallest tsetse distributions (i.e., minimum
annual surface area). The timing of these three scenes was then analyzed to see if they
coincided with a particular season and an approximate date. The three minimum annual
surface area scenes were then combined and spatial locations where tsetse were present

during all three years were identified to construct a tsetse reservoir map.

3.5 TED Model Validation

Validation of the TED Model was accomplished in two ways, an accuracy
assessment of the model outputs and a sensitivity analysis of the model parameters. The
accuracy assessment is performed to determine how accurate the TED Model percent

probability map is compared to ground truth data. While the sensitivity analysis

64

examines how changes in parameter values affects the TED Model’s outputs. Using
these two validation methods I will be able to both judge how accurate the tsetse
distributions predicted by the TED Model are, and report what variables are the most
important in predicting said distributions.

3.5.1 Ground Truth Comparison

As mentioned in section 3.2.1.4, the lack of country-wide tsetse census data meant
that an alternative ground truth data was used. The first source of ground truth data used
in this study was a 1996 fly belts map (Figure 1.2), and the second was the combined
FAO / IAEA tsetse distribution map (Figure 3.5). The TED Model percent probability
map was compared to both ground truth data sets using the Mapcurves GOF test
explained in section 3.3.4.

To facilitate the calculation of a Mapcurves GOF scores both the FAO / IAEA
tsetse distribution map and the TED Model percent probability map were" classified into a
ten classes (i.e., each class represented an increase.10% probability of tsetse presences).
Since the combined FAO / IAEA distribution map and the TED Model percent
probability map had the same classification scheme, a traditional kappa coefficient was
calculated. A kappa coefficient is a rescaled proportion of agreement between two data
sets, and can be calculated by taking the observed accuracy minus the chance accuracy
divided by one minus the chance accuracy as shown below in Equation 3.1 (Lillesand et

aL,2004;Foody,2006)

Observed Accuracy- Chance Accuracy

 

Kappa =
1 - Chance Accuacy
Equation 3.1

65

A kappa coefficient of 1.00 indicates perfect agreement between the two maps.
The kappa coefficient and the Mapcurves GOF not only test the agreement between the
combined FAO / IAEA distribution map and the TED Model percent probability map, but
also explored the level of agreement between the different statistical methods. In
addition to the accuracy assessment performed using the TED Model percent probability
map, a Mapcurves GOF score was calculated between the 1996 fly belts map and the
F AO / IAEA distributions to assess the level of agreement between these two products.
3.5.2 Sensitivity Analysis
3.5.2.1 Sensitivity Index

NDVI, maximum temperature, and minimum temperature were all analyzed in an
effort to assess their sensitivity in and their effect on the TED Model. Land cover was
excluded from the sensitivity analysis since the Mapcurves analysis identified the
optimum LULC product to be used in the TED Model. All four variables used in the
TED habitat model have the potential to both limit (i.e., changes in threshold values
cause predicted distributions to decrease in surface area) and expand (i.e., changes in
threshold values cause predicted distributions to increase in surface area) tsetse
distributions.

To rank the ability of the three variables to limit or expand tsetse distributions a
sensitivity index (SI) was calculated. A SI is calculated by adjusting each variable’s
parameter by i 2% increments (e.g., 0.02 for NDVI, and 1°C for minimum and
maximum temperatures), and measuring the change in the TED Model (Equation 3.2)

(see Hamby, 1994; Lenhar et al., 2002).

66

( AA
ABaseIine

l
1
1
Sensitivity Index = l
I
l

(-—“’ )
PBaseIine

- r ,- . r L_ r, _ ”no Equation3.2

 

Where A is the area of the TED Model state variable, P is the parameter of interest, and
Baseline refers to the standard TED Model run used to identify potential dry season
reservoirs and create the percent probability map. The state variable in the sensitivity
analysis for the TED Model was the amount of surface area in the TED model percent
probability map above 50%, which was selected to match the recommendation by Wint
(2001) for turning the FAO / IAEA combined percent probability map into a binary
presence -— absence map. The resulting SI calculated for each parameter was then
classified into one of four classes from insensitive to extremely sensitive as described in
Lenhar et al. (2002) (Table 3.3). A parameter was deemed insensitive if the SI was
between 0.00 and 0.05, implying that a change in the parameter resulted in little to no
change in the state variable. Moderate and high sensitivity was assigned to a SI between
0.05 to 0.20 and 0.20 to 1.00 respectively. A calculated SI greater than 1 would imply a
proportional change in the state variable greater than the proportional change in the

parameter, and was considered to be extremely sensitivity.

Table 3.3: Sensitivity classes adapted from Lenhart et a1. (2002).

 

 

 

Sensitivity Index Class Sensitivity
0.00 5 SI < 0.05 I lnsensitive
0.05 5 SI < 0.20 11 Moderate
0.20 5 SI < 1.00 111 Highly
SI 2 1.00 1V Extremely

 

 

67

To identify thresholds of interest a relative SI was created (Equation 3.3), which
calculates the model’s sensitivity to changing the parameter relative to the previous

parameter and surface area, rather than the Baseline parameter surface area.

( ABaseline)— (AAA ABaseline

(P AP2 )_ (P APJ )
PBaseline PBaseline

Where A1 and P. are the surface area and parameters of the previous model run (e. g.,

 

Relative Sensitivity Index -

Equation 3.3

model run #2 which raises the parameter by 2% from the baseline), and A2 and P2 are the
surface area and parameters of the current model run (e.g., model run #3 which raises the
parameter by %4 from the baseline). The advantage of using the relative SI is that the
user can track the changes in the SI and identify a threshold where a parameter becomes
insensitive (i.e., no change in the state variable) relative to the previous state variable
rather than the baseline model run.
3.5.2.2 The Spatial Location of Potential Tsetse Expansion

In an effort to discern the spatial location of where each variable in the tsetse
habitat model (i.e., NDVI, maximum temperature, minimum temperature, and land cover)
constrains tsetse distributions from expanding, the TED Model was run with each
variable removed one at a time. The baseline model run was then subtracted from the
resulting four distributions, leaving only areas of tsetse expansion associated with the
particular variable removed from the habitat model. Each resulting distribution was then

classified into a binary presence — absence map using the amount of surface area in the

68

 

above 50%, which was selected to match the recommendation by Wint (2001) for turning
the F AO / IAEA combined percent probability map into a binary presence — absence
map. The four maps of tsetse expansion were then overlaid, with the resulting map
showing the spatial location where tsetse distributions would expand if each variable was

removed from the TED Model.

69

CHAPTER 4

IN ITIALIZATION OF THE TED MODEL:
RESULTS, DISCUSSION, AND CONCLUSION

4.1 Land Cover
4.1.1 Results

The fifteen LULC data sets vary widely in the amount of woody vegetation
predicted to be in Kenya from roughly 45,000 km2 to 523,000 km2 (Table 4.1). With the
addition of environmental variables and the creation of the binary habitat suitability maps
the amount of suitable tsetse habitat decreases for each data set and ranges from roughly
31,000 km2 to 205,000 kmz, still a wide range. The overall decrease in suitable habitat
range is primarily caused by low precipitation in the northern parts of Kenya, which
creates inhospitable moisture regimes for both tsetse adults and pupae.

Table 4.1: The amount of woody vegetation and suitable tsetse habitat (when
combined with environmental variables) predicted by the LULC binary maps.

 

 

 

 

 

Data Set Amount of Woody Vegetation Predicted Suitability

Area km2 % of Kenya Area km2 % of Kenya
Africover 515,518 88 205,864 35
CLchover 324,896 55 163,340 28
GLC2000 217,938 37 143,683 24
IGBP DISCover 523,527 89 191,849 33
UMd GLCC 280,451 48 _ 149,603 25
MODIS 1km 412,459 70 178,669 30
Type 1 500m 364,527 62 126,326 22
MODIS 1km 412,403 70 178,647 30
Type 2 500m 387,720 66 146,710 25
moms 1km 412,319 70 178,626 30
Type 3 500m 387,750 66 146,740 25
moons 1km 79,768 14 58,741 10
Type 4 500m 45,209 8 31,409 5
moons 1km 296,386 50 90,609 15
Type 5 500m 311,623 53 80,611 14

 

 

70

The Mapcurves GOF between the 1996 fly belts map and the combined FAO/
IAEA distribution map to the LULC binary suitability maps resulted in similar levels of
agreement between the fifteen data sets, with a range between 0.52 —— 0.59 and 0.53 - 0.65
respectively. When the weighted ratio comparison matrix between {each of the binary
suitability maps unsuitable class and the 1996 fly belts map and the FAO / IAEA
distribution map was examined, high levels of agreement were observed with a range
between 0.70 — 0.95 and 0.75 - 0.95 respectively. These observations lead to the
conclusion that the high level of agreement between one of the two classes inflated the
overall GOF score for each LULC data set, creating a false confidence in the results. For
this reason only the GOF between the suitable class of the binary suitability maps and the
two maps used as ground truth were examined to determine which data set had the
highest level of agreement of tsetse presence.

The comparison of the binary suitability maps to the 1996 fly belts resulted in
Africover, CLIPcover, IGBP DISCover, and MODIS 1km Type 1, 2, and 3 products
having significant levels of agreement based on the calculated 1 scores. The comparison
of the binary suitability maps to the F AO / IAEA combined distribution map resulted in
Africover, IGBP DISCover, UMd GLCC, and MODIS 1km type 1, 2, and 3 products

having significant levels of agreement (Table 4.2 and Figure 4.1).

71

Table 4.2: Results of the Mapcurves GOF analysis between the LULC binary
suitable tsetse habitat maps and the combined FAO / IAEA distribution map and
the 1996 fly belt map. Significant levels of agreement in bold.

 

 

 

 

 

 

Ma curv s
LULC Data Set col: Scoere ‘s°°'° P va'”
Fly Belts FAO/IAEA Fly Belts FAO/IAEA Fly Belts FAO/IAEA
Africover 0.45 0.53 5.53 5.40 0.00 0.00
CLIPcover 0.37 0.39 2.71 1.30 0.01 0.11
GLC2000 0.31 0.36 0.78 0.59 0.22 0.28
IGBP DISCover 0.40 0.49 3.84 4.20 0.00 0.00
UMd GLCC 0.33 0.43 1.19 2.44 0.13 0.01
MODIS 1km 0.36 0.45 2.26 3.09 0.02 0.00
Type 1 500m 0.26 0.29 -1.04 -1.51 0.63 0.85
MODIS 1km 0.38 0.45 2.88 3.09 0.01 0.00
Type 2 500m 0.31 0.34 0.54 0.04 0.30 0.48
MODIS 1km 0.38 0.45 2.88 3.06 0.01 0.00
Type 3 500m 0.31 0.34 0.54 0.05 0.30 0.48
MODIS 1km 0.12 0.16 -5.86 -5.10 1.00 1.00
Type 4 500m 0.06 0.08 -7.89 -7.45 1.00 1.00
MODIS 1km 0.19 0.20 -3.66 -4.05 1.00 1.00
Type 5 500m 0.16 0.16 -4.69 -5.15 1.00 1.00

 

Ho: p Value 2 0.10
Ha: p Value < 0.10

 

I FAO / IAEA Map
l Fly Belts

9
or
o

9
a
o

P
w
a

0. 20

9
.s
O

 

Mapcurves GOF Scone

 

0. 00

Africover

MO DIS t1 1km
MO DIS t2 1km
MO DIS t3 1km
UMd GLCC
CLIPcover
GLC2000
MODIS t3 500m
MODIS t2 500m
MO DIS t5 1km
MO DIS t4 1km
MODIS t5 500m
MODIS t4 500m

IGBP DISCover
MODIS t1 500m

Figure 4.1: Mapcurves GOF scores for each LULC data set when compared to the
FAO / IAEA combined distribution map. Data sets are sorted in order from highest
to lowest GOF with the F A0 / IAEA combined distribution map.

72

 

4.1.2 Discussion of Which Land Cover Product to Use in the TED Model

Based on the results of the Mapcurves GOF analysis, the top LULC data set for
use in predicting suitable tsetse land cover was Africover. Possible reasons for the
Africover product out performing the other LULC products include the higher spatial
resolution data used in the creation of the product, local knowledge in the initial
classification, and country specific classes. Africover coincidently predicts the largest
area of suitable tsetse habitat out of the fifteen LULC products examined. The possibility
that this contributed to Africover out performing the other products was explored;
however, the apparent relationship between high amount of predicted suitable tsetse
habitat and high GOF scores does not display a proportional change between percent
suitable and GOF scores. If one examines the percent difference of predicted suitability
(Table 4.1) between Africover and IGBP DISCover, a 2% difference is observed,
compared to fly belt GOF score difference of 0.05 and FAO / IAEA GOF score
difference of 0.04. Similar comparisons between the other LULC data sets GOF scores
and percent predicted suitability show no direct proportional relationship, and the general
relationship between predicted suitability and GOF scores was considered a negligible
result of examining the suitability GOF rather than the overall GOF of each data set.

Although Africover was identified as the top performer, one goal was to identify
multiple LULC products that could be used to model tsetse in Kenya. To that end IGBP
DISCover and MODIS type 1, 2, and 3 Global Land Cover at 1km resolution products
were also determined to be strong performers. The decision on which of the five LULC
products to use in the construction of a tsetse habitat model can now be made using

factors not directly examined in this study (e. g., accuracy assessments, temporal

73

resolution, data availability). With regards to constructing a model that could predict
changing tsetse habitat based on land use, land cover, and climate, the ability to perform
an analysis of LULC change is beneficial.

Unlike Africover, all of the lkrn MODIS land cover products, in addition to the
2001 data used in this analysis, were produced annually for 2002, 2003, and 2004.
Further examination of the three MODIS products shows that the GOF scores were
nearly identical. To tease out the most favorable product type the results of other LULC
data sets that employed the same classification methods and schemes as the MODIS type
1, 2 and 3 products were examined. MODIS type 1 was determined to be the optimal
MODIS product since it is constructed using the IGBP DISCover classification scheme
and method, which had the second highest GOF of all fifteen data sets examined.
4.1.3 Additional Points of Interest in the Land Cover Analysis

An unexpected result of the GOF analysis was that of the 500m MODIS LULC
products when compared to the 1km MODIS LULC products. Although the 500m data
has four times the spatial resolution of the 1km MODIS products, all five of the products
were calculated to have insignificant GOF scores. It is my belief that the lower GOF
scores are due in part to the over estimation of grassland in southern Kenya. For
example, the 500m type 1 product contains roughly 6% more grassland than the 1km type
1 product (see Figure 3.2).

The low level of agreement between the binary suitability maps and the available
ground truth data may be partly attributable to the way the ground truth data were
constructed. When considering the existing FAO / IAEA products it is important to note

that they have not been through peer review, nor do they have a published accuracy

74

statement, thus the low level of agreement may simply be an artifact of accumulative
uncertainty. The 1996 fly belts may also have a high degree of uncertainty due to their
apparent generalized locations when compared to the more detailed 1973 fly belts
produced by Ford & Katondo (1977) (see Figure 3.9).

Despite the low level of agreement between the binary suitability maps when
compared to the FAO / IAEA map and the 1996 fly belts, the GOF method does identify
the LULC products that best predict land cover required by tsetse. The GOF method can
be used to differentiate between various LULC products and be applied to any such
research when there is a known relationship between a species and land cover. The
importance of performing this type of analysis can be observed in the results of the GOF
scores produced by GLC2000 when compared to Afiicover. A previous comparison of
GLC2000 and Africover performed by Torbick et al. (2005) concluded that GLC2000 out
performed Africover for predicting natural land cover such as grassland, savannah, and
forest. However, in my analysis I found that the Africover out performs GLC2000 for
identifying suitable tsetse land cover classes. This discrepancy epitomizes the
importance of evaluating the available LULC products and not relying on mere accuracy
assessments.

4.1.3 Conclusion

Each LULC product is different often having a distinct production method,
classification scheme, and intended use. For this reason it is important to assess the
ability of each LULC product to predict the land cover of interest rather than just relying
on mere accuracy assessments. Africover was identified as the best predictor of suitable

tsetse land cover. However, since no plans currently exist to produce another LULC

75

product similar to Africover, and based on the need to model tsetse over time, this
product is not considered to be the best choice. The MODIS global land cover products
were produced annually from 2001 to 2004. In total, use of a MODIS product would
provide an updateable account of land cover change and match the temporal resolution of
the TED Model, making one of the MODIS products the preferred data set to construct a
tsetse habitat model. MODIS type 1 was determined to be the optimal MODIS product

and was selected to be the land cover data set in the TED Model.

4.2 Starting Distributions
4.2.1 Results

The results of initializing the TED Model with the each of the three tsetse
distributions (i.e., 1973 fly belts, 1996 fly belts, and all of Kenya as infested) are
summarized in Figure 4.2. Running the TED Model with the 1973 fly belts as the
starting distribution resulted in the northern pockets of possible tsetse infestation
disappearing; however, tsetse populations in the south and along the Tana River remained
calling into question the accuracy of the 1996 fly belts map. The use of the 1973 fly belts
also reinforced the notion of possible tsetse expansion since the maps construction in
1977. The most obvious expansion of tsetse distributions from the 1973 distributions as
predicted by the TED Model occurred around southern Lake Victoria (areas denoted in

dark grey around Lake Victoria Figure 4.2).

76

The Differences In Predicted Tsetse Distributions
Caused by Varying Starting Distributions

 

 
  

  

L’ _-

Predicted tsetse presence
based on the tsetse distributions
used to Initialize the TED Model

82
fit;

1“;
; ri

we
- 1996 Fly BeIts

- Tsetse Present in all of Kenya

- Locations of Mutual Agreement 200

Figure 4.2: The differences between the three data sets used to initialize the TED
model.
0 Yellow and dark grey represents the areas where tsetse flies are found
when the 1973 fly belts are used to initialize the TED model.
0 Green and dark grey represents the areas where tsetse flies are found
when the 1996 fly belts are used to initialize the TED model.
0 Red, yellow, green, and dark grey represents the areas where tsetse
flies are found when all of Kenya is considered infested with tsetse
flies with regards to initializing the TED model.

77

 

Running the TED Model with the 1996 fly belts as the starting distribution
resulted in further expansion of tsetse populations in Western Kenya around Lake
Victoria and several other locations (areas denoted in green in Figure 4.2). However,
when compared to the TED Model output produced by initializing with the 1973 fly
belts, using the 1996 fly belts as the starting distribution resulted in tsetse populations in
Southern Kenya and along the Tana River being excluded (areas denoted in yellow in
Figure 4.2).

The initialization of the TED model using all of Kenya as having tsetse present
resulted in all the fly populations of concern (i.e., Tana River, Lake Victoria, and pockets
in Southern Kenya) as expected were present in the model output (areas denoted in
yellow and green in Figure 4.2). In addition to the tsetse populations included in the
model runs using 1973 / 1996 fly belts to initialize the TED Model, new potential
expanded tsetse populations were identified in Northwestern / Southern / Central Kenya,
around Lake Victoria, on the Tana River, and along the Somali border (areas denoted in
red in Figure 4.2).

4.2.2 The Determination of the Tsetse Distribution used to initialize the TED Model

The determination of which tsetse distribution to initialize the TED Model with
was greatly influenced by the exclusion of particular areas, the apparent expansion of
tsetse in certain locations, and the ultimate goal of using the TED Model in my research.
The use of the 1996 fly belts to initialize the TED Model resulted in the exclusion of
Tana River tsetse populations, which were documented to exist in 2002 (Catley, 2002).

Initializing the TED Model with the 1973 fly belts did result in the inclusion of Tana

78

River and Southern Kenya tsetse populations, but didn’t allow for tsetse populations to
fully expand into potentially suitable habitat, especially around Lake Victoria.

Initializing the TED model with all of Kenya as having tsetse present is the least
logical scenario. By using all of Kenya as the initial tsetse distribution the amount of
land where tsetse are predicted as present is surely over estimated. However, running the
TED Model in this manner potentially identifies unknown sustainable fly populations.
For example, on a trip to Kenya during August, 2008 I was told of unconfirmed rumors
of tsetse populations expanding along the Somali border (Matima personal
communication, 2008). When the TED Model is initialized with all of Kenya as having
tsetse present the expansion of tsetse along the border and other locations is observed. In
addition to identifying unknown fly populations, the TED Model if initialized with all of
Kenya as infested with tsetse has the potential to identify suitable tsetse habitat currently
uninhabited but that could support tsetse populations if introduced.
4.2.3 Conclusion

Ultimately the goal of using the TED Model was to test the hypothesis that fly
populations fluctuate seasonally, and if they do, identify tsetse reservoirs. To this end the
most accurate and up to date tsetse distribution should be used to initialize the TED
model. However, the 1996 fly belts map excluded known fly populations, drawing into
question its overall accuracy. Although the 1973 fly belts map did include the tsetse
populations excluded by the 1996 map, it had the drawback of preventing tsetse
populations from fully expanding and was created over 30 years ago. In lieu of the of the

problems associated with both the 1.973 and 1996 distributions, and keeping in line with

79

the goals of constructing the TED Model, I initialized all subsequent model runs with all

of Kenya infested with tsetse.

8O

CHAPTER 5

THE TED MODEL OUTPUTS, VALIDATION, AND
UNCERTAINTY: RESULTS AND DISCUSSION

5.] The TED Model Outputs
5.1.1 Individual Scenes
5.1.1.1 Results

The TED Model produced 69 unique binary tsetse distribution maps displaying
the predicted location of high tsetse population densities (i.e., edge of the fly front) every
16 days from 1/1/2002 — 12/19/2004. The spatial distributions of tsetse produced by the
TED Model fluctuate both at intra-annual and inter-annual temporal resolutions (Figure
5.1), with a maximum surface area of 67,378 km2 occurring during the 2002 long rains
(i.e., 5/25/2002) and a minimum surface area of 33,201 km2 at the end of the 2003 cool
dry season (i.e., 10/ 16/2003). The tsetse distribution data were then combined to create
the percent probability map, and separated into years and ranked to identify the tsetse
reservoirs (Table 5 .1).
5.1.1.2 Discussion

The fluctuations in tsetse distributions predicted by the TED Model correspond
with seasonal weather patterns (Figure 5.1; Figure 5.2). Fly populations expand with the
onset of the long rains (roughly beginning of March), contract at the start of the cool dry
season (roughly beginning of June), expand again with the commencement of the short
rains (roughly beginning of November), and contract during the hot dry season.
However, a closer examination the TED Model predicted tsetse distributions shows that

fly populations expand with the onset of the short rains in November (which corresponds

81

.maeuaembmmu 3.83 6838.5 2: 3 626.8%
«E382— ?eegg 2: «Ema—i=5 3 @8286 one 55 2: me .858.— 35 2888 Se.— 2; .25" he .25

a... one 33 .8952. .252 am: 2: .3 3:08.:— 888 «>2. 3 683—8.:— 33 Satan 2:. “—.m flaw:—

 

 

 

 

33>
«8.4. 88 «SN «ecu
. . .. . i u. . l e
mm mm . mm ._ .. m m a
w u u m . m w .
E" S. . . E. 8. er... E. S. M. .. so 3
w. m" w m m“ m a"
. u u. u.
. . . . . . W n . u §.°N
. . . . _ . u .
sm m m u s" m n m em " n
. . . . . _ . . . u . .
u n u u n n u n u . u . e8 8
u m m m . m .2 m m m m
u . u n u u . u n r 08.3
. . . . . .
. . . . . .
u n n . n
. n u . . 80.8
. m
.
n . 86.8
. 30.2.
as... be 23¢ be 88: be 88: be 8.8 be 852 be
.35 boo 9.3 .6: team 80 as... .6: team .80 9.3 3:
08.8

«some... 083»... e>az 9 .262 omh
05 3 38.095 no.2 moatam

(aunt) 901v pawns

82

2:23:2— _a=8a8 2: «Ema—i=8 B 632.2. 558.. 55 we:
388... 5?» $533283 885 83:39:. .252 Gm: 2: he .E—Eetg REE“ «.55 2:. and Earn

 

 

 

 
 

 

flooodm

23.6520
8n 8... SN 8w 8. 8. on a
. e . _ x . . _ e . . x . :eosl o
m _m mm m 38 4
u .l w . 2:
.w ".m. :w E O 8
w .2... mm a mum
U
m “m "J m 68.8
“m. u
m .9. _ n
u u .83”
n u
n 4 58.8
n
4
O

   

 

 

u
a e a .
. ooodw
$8.2.
«Set teem A5 300 at?! 934 ED 3:
8nd»

 

.28: em: 55 3 8.285
8:258me 3333... be 8.2 eomtam :85.

83

(run) wv psalms

Table 5.1: The 69 scenes produced by the TED Model, separated by year, and

ranked by surface area predicted to have tsetse present from lowest to highest. The
lowest (i.e., Rank #1) scene for each year is considered to be the dry season
minimum, and was used to identify_the location of the drxseason reservoirs.

 

 

 

 

 

 

Rank Standard Date Ordinal Date Area (ka)
2002 2003 2004 2002 2003 2004 2002 2003 2004
1 10/16 10/16 10/15 289 289 289 39,417 33,201 35,141
2 9/30 9/30 9/29 273 273 273 41 ,036 36,396 37,043
3 2/18 3/6 9/13 49 65 257 42,643 38,805 37,764
4 1 1/1 1 1/1 10/31 305 305 305 43,497 40,425 39,802
5 11/17 3/22 8/28 321 81 241 46,006 43,841 40,974
6 9/14 9/ 14 3/21 257 257 81 46,197 44,033 41,160
7 3/6 11/17 3/5 65 321 65 48,317 44,073 41,969
8 1/1 12/3 8/12 1 337 225 48,637 45,235 43,333
9 12/3 12/19 11/16 337 353 321 48,738 46,172 44,837
10 8/29 8/13 7/27 241 225 209 49,195 47,423 45,993
1 1 2/2 8/29 1/1 33 241 1 49,532 47,708 46,242
12 1/17 4/7 4/6 17 97 97 50,532 48,147 46,258
13 8/13 2/18 12/2 225 49 337 50,627 49,104 48,267
14 12/19 7/28 7/1 1 353 209 193 50,695 49,412 48,474
15 3/22 4/23 2/18 81 113 49 51,438 51,212 49,655
16 7/28 5/9 2/2 209 129 33 53,336 51 ,724 50,605
17 4/7 7/12 4/22 97 193 113 54,002 53,298 51,191
18 4/23 5/25 6/25 1 13 145 177 59,307 53,829 51,941
19 5/9 1/1 12/18 129 1 353 63,065 53,829 52,083
20 7/12 6/26 1/17 193 177 17 63,515 55,071 52,226
21 6/26 2/2 5/8 177 33 129 65,049 55,239 53,851
22 6/10 6/10 6/9 161 161 161 65,237 55,505 54,489
23 5/25 1/17 5/24 145 17 145 67,378 57,498 55,190

 

84

 

to scene 305), but continue to expand during the hot dry season for an additional two
scenes (scenes 1 and 17). Due to the decreased precipitation afier the short rains, and that
the hot dry season is commonly when the warmest temperatures occur in Kenya, it was
surprising to see an expansion of fly populations during this period. This apparent
discrepancy led to the conclusion that the onset of the wet and dry seasons alone do not
account for the expansion and contraction of suitable tsetse habitat.

Seeing that one of the goals of this research was to explore the respective roles of
the individual ecological variables used in the TED Model, I have attempted to ascertain
the effect that each variable has on tsetse distributions and try to explain the expansion of
tsetse distributions at the beginning of the hot dry season. Due to the size, variety of
physiographic landscapes, and natural variability of climate, no one variable should be
considered the primary driver in all locations. However, by using the country wide mean
values of the variables used in the TED tsetse habitat model in each scene, the general
trend can be ascertained.

The exception to this is land cover, which is only updated once per year (i.e.,
scene 1) in the TED Model. A change in LULC has the potential to cause a net gain or
loss of suitable tsetse habitat. A simple LULC change analysis for each year showed that
on average 55,986 km2 were converted from unsuitable to suitable, and on average
62,853 km2 were converted from suitable to unsuitable tsetse land cover (Table 5.2). The
conversion of more surface area to unsuitable land cover should cause a net decrease in
fly populations in the first scene of each year (i.e., scene 1). However, the spatial
location of decreases in suitable tsetse land cover could have been occurring in regions

with no tsetse present, and the increase in suitable land cover could have been occurring

85

 

in locations where fly populations existed nearby. Thus, despite a net decrease in suitable
tsetse land cover, fly populations could have expanded at the first of the year due to the
conversion of previously unsuitable land cover to suitable in areas neighboring existing
fly populations. To test whether or not this was contributing to the expansion of tsetse in
the first scene of each year, the TED Model was run with static LULC (i.e., only using
the 2001 LULC data rather than updating it annually) to see if the increase still occurred.
When the increase at the first of the year was observed to still occur, it was determined

that annual change in suitable land cover was not the root cause of this phenomenon.

Table 5.2: Land use land cover conversion with regards to tsetse land cover

 

 

 

suitability.
Temporal Conversion to Conversion to
Period Suitable Land Cover Unsuitable Land Cover
2001 - 2002 82,028 87,375
2002 - 2003 47,157 52,947
2003 - 2004 38,772 48,238
Mean 55,986 62,853

 

 

Minimum temperature was also determined to have no apparent effect on intra
annual fluctuations in tsetse distributions, nor does it explain the continued expansion of
fly populations at the beginning of the hot dry season. The mean minimum temperature
never dropped below 20°C, thus never approaching the tsetse survival threshold of 17°C
(Figure 5.3; Table 5.3). Although this eliminates minimum temperature as a significant
driving variable for the majority of Kenya, as discussed later in section 5.2.2, it does

affect distributions in locations with higher elevations.

86

 

62.3 I .55 9.59.358 .5555: ER .90.. I 382v 23.22.53 5255::
.99.” I 3.23 2.32::me 9303 36:5:— Eua—Z Gm: 2: he so...“ 35...; :3:— 2; "rem 2&3

 

 

 

8.5 .8520
omn can own 8N cm... 8' on w
c D b D I I I O
£5. 10:
m 32 no: . 88.2
«0.. III
2. <
. Soda
N me
n .
u on . So 8
I.
O .
w u
w on . 8o 8
m on . Sadr.
m an 1‘
. 08.8
9.
at I 08.2
33¢ :25 ha 300 «Sam 98.. be .0:
cm 98.8

 

ooSfiEan... coo—2 3 neon—coo
.3239me oflom... ho no: oomtzw cams.

(awn) wv aoeuns

87

Table 5.3: Mean surface area predicted by the TED Model to have tsetse present,
temperatures, and NDVl between 2002 and 2004.

 

 

 

Date Mean Area Mean Temperature (C°) Mean
Standard Ordinal (kmz) Maximum Minimum NDVI
1/1 1 49,569 33.8 22.0 0.400
1/17 17 53,419 35.0 21.8 0.386
2/2 33 51,792 37.9 22.6 0.349
2/18 49 47,134 39.3 23.4 0.322
3/6 65 43,031 38.3 22.6 0.315
3/22 81 45,480 37.1 23.5 0.331
4/7 97 49,469 34.8 22.6 0.376
4/23 1 13 53,903 32.2 22.0 0.444
5/9 129 56,213 32.0 21.3 0.475
5/25 145 58,799 32.7 21 .4 0.447
6/10 161 58,410 32.0 20.2 0.397
6/26 177 57,354 31.8 20.2 0.368
7/12 193 55,095 32.9 20.1 0.346
7/28 209 49,580 33.6 20.1 0.324
8/13 225 47,128 33.9 20.9 0.320
8/29 241 45,959 36.1 21.4 0.317
9/14 257 42,664 36.0 21.5 0.308
9/30 273 38,159 38.6 21.8 0.299
10/16 289 35,920 36.2 21.4 0.315
11/1 305 41,241 32.8 21.0 0.395
11/17 321 44,972 33.0 21.3 0.440
12/3 337 47,413 31.1 20.9 0.459
12/19 353 49,650 34.7 21.4 0.441

 

 

Investigations into maximum temperatures effect on tsetse distributions showed
as expected that higher temperatures were associated with a drop in surface area
predicted to have tsetse present (Figure 5.3). When mean maximum temperatures were

rising (e.g., beginning of the dry seasons) and went above ~36°C the mean surface area

88

 

predicted to have tsetse present decreased rapidly (e. g., scenes 33 and 241). During
periods of decreasing or with low mean maximum temperatures, tsetse distributions
generally expanded (e. g., during the long and short rains). However, due to the strong
negative correlation with NDVI and moisture and that mean maximum temperature never
exceeded the 40°C threshold for tsetse survival (over half of Kenya was always
considered suitable habitat for tsetse with regards to maximum temperature), an
examination of moisture conditions during periods of tsetse expansion and contraction is
necessary before reaching any conclusions about maximum temperatures effect on tsetse
distributions.

NDVI of the four variables used in the tsetse habitat model visually matched the
intra annual expansion and contraction of tsetse distributions the best (Figure 5.4), and
was the only variable to have its calculated mean scene value drop below the threshold
used in the TED Model. During both the hot and cool dry seasons when NDVI values
dropped below 0.35, the surface area predicted to have tsetse present decreased rapidly
(i.e., scenes 33 and 193). When NDVI values started to rise with the beginning of the
long and short rains, the surface area predicted to have tsetse present immediately
increased (i.e., scenes 81 and 305). A drop below the threshold used in the TED Model
(e.g., 0.39 for NDVI) means at that time less then half of the country was suitable tsetse
habitat. Since NDVI is the only variable that had mean values below the threshold used
in the TED Model, I determined that moisture has the potential to have the greatest effect
on limiting tsetse distributions. That being said, moisture does not appear to explain the
expansion of tsetse distribution at the beginning of the hot dry season since NDVI values

are decreasing at that time.

89

 

IACIN "'9“

mod

2.9

26

and

mud

and

mad

36

mvd

and

can

83 .255

8» SN 8N 8. 8.

.2823 :52
.55 $0..» I no.5; “53:55:. @308 @8269... Eco—z Gm; 2: he no...“ coat—a =3:— 2: "1m unsure

 

 

2.?! team

”Sat 93..

be Boo

_>QZIOI

u2<Ill

be 8:

I 80.0..

u 80.94.

v 9.5.8

I 80.3

r 86.3.

v 898

v 89.2.

 

 

502 cows. on cocoa E00
23:39me mun-om... ho moi oomtzm cams.

Soda

(gum) 981v aoayns

90

NDVI of the four variables used in the tsetse habitat model visually matched the
intra annual expansion and contraction of tsetse distributions the best (Figure 5.4), and
was the only variable to have its calculated mean scene value drop below the threshold
used in the TED Model. During both the hot and cool dry seasons when NDVI values
dropped below 0.35, the surface area predicted to have tsetse present decreased rapidly
(i.e., scenes 33 and 193). When NDVI values started to rise with the beginning of the
long and short rains, the surface area predicted to have tsetse present immediately
increased (i.e., scenes 81 and 305). A drop below the threshold used in the TED Model
(e.g., 0.39 for NDVI) means at that time less then half of the country was suitable tsetse
habitat. Since NDVI is the only variable that had mean values below the threshold used
in the TED Model, I determined that moisture has the potential to have the greatest effect
on limiting tsetse distributions. That being said, moisture does not appear to explain the
expansion of tsetse distribution at the beginning of the hot dry season since NDVI values
are decreasing at that time.

The examination of moisture did reveal that maximum temperature has an
independent effect on tsetse distributions during the cool dry season. Between scenes
209 and 289 mean NDVI values stabilize around 0.30 :t 0.01, however, the surface area
predicted to have tsetse present continues to decrease. During this same time period
maximum temperatures are rising from 336°C to 386°C, providing evidence that
maximum temperature, although highly correlated with NDVI (e. g., a Pearsons
correlation coefficient of -0.831 was calculated between mean annual NDVI and Day
LST), does effect tsetse distributions independently of NDVI. However, as observed

with moisture, maximum temperatures are rising at the beginning of the hot dry season

91

and should have the effect of restricting fly populations. Since this is not occurring, it
was concluded that similar to the three other variables, maximum temperature does not
explain the expansion of tsetse distributions at the beginning of the hot dry season.

An explanation of the apparent discrepancy occurring at the beginning of the hot
dry season might lie in the duration of the short rains, which is roughly one month shorter
than the long rains. The extra 30 days (two model scenes) during the long rains could
allow tsetse to fully expand into all available suitable habitat before the beginning of the
cool dry season. When the cool dry season begins tsetse are unable to expand further,
even though climate conditions have yet to dramatically limit their distributions. The
only slight negative slope and larger surface areas occupied by tsetse during the first two
scenes (i.e., scenes 161 and 177) of the cool dry season could be possible evidence of this
scenario. Thus the expansion of fly populations at the beginning of the hot dry season
would be a result of still suitable climate conditions allowing tsetse to continue to expand
into suitable habitat until conditions become either too hot or dry.

To test the validity of this hypothesis the TED Model fly movement rates were
accelerated. This would allow tsetse to expand faster, and if tsetse did indeed reach their
maximum distribution at the end of the long rains, no substantial change should be
observed in the surface area of distributions at that period of time. However, when the
test was performed, fly distributions in both the hot and cool dry seasons expanded past
the baseline model distributions by nearly the same proportional amount. This is
evidence that tsetse do not inhabit all suitable land by the start of the dry seasons, and that
baseline tsetse distributions could expand firrther in the long rains if unsuitable

environmental conditions during the cool dry seasons did not restrict them. The extended

92

 

duration of the long rains does not explain the lack of tsetse expansion at the beginning of
the cool dry season, nor provide any insight into why tsetse distributions continue to
expand in the hot dry season as climate conditions should be restricting fly populations.

The spatial locations of the fluctuations in tsetse distributions during the hot dry
season were also examined in an effort to postulate a reason for the phenomenon. The
tsetse distribution maps were analyzed for any regional contraction or expansion of fly
populations, with one discemable pattern being observed. At the start of the hot dry
season fly populations in Southern Kenya were still expanding, while at the start of the
cool dry season fly populations in Southern Kenya were contracting. All three dynamic
ecological variables (i.e., minimum temperature, maximum temperature, and moisture)
were examined, with conclusion that during the start of the cool dry season Southern
Kenya was experiencing a decrease in moisture, while at the start of the hot dry season it
was experiencing an increase in moisture levels (Figure 5.5).

The explanation for this increase of moisture during the beginning of the hot dry
season lies in the positioning of the ITCZ, which is to the south of Kenya following the
December solstice. As the ITCZ heads south and passes over Kenya the short rains
occur, and the last region to receive precipitation is southern Kenya. When the long rains
occur the ITCZ is moving to the north, and Southern Kenya is the first region to receive
precipitation, and the first to experience seasonal drying. Since a large portion of tsetse
in Kenya are found in the south, populations are able to expand at the beginning of the
hot dry season because moisture conditions have not become unsuitable due to the timing

of the short rains in this region.

93

.282. 2.3.5.:— o.: 85.. an... 99.: 238...: —>Dz 2: 95.... 68.8.9... .8 .322—
..8332. .25... 9::— «a.: «no... 83%... 2.3 x26 .3 :39... be 5 23.5. 6.523.. F6 38 E...
:2— 2: me 33:33 2: an v.25»... 9.50.3... 0335...... can 033...... me 5:32 2E. "Wm 9...»...—

2318. .35...» I
2312. 4.8.3.... E
23...... 2.828. I

2333. 2.828.. I

     

cocoon >5 .80 .333 to «o...

2.038 to 2: ho 95.503 2.:
«a E=na=§< 233.05. 083...

94

 

Regardless of the reason for the expanding tsetse distributions at the beginning of ”r”
the hot dry season, the seasonal patterns displayed by the predicted tsetse distributions are
quite evident. Tsetse distributions predicted by the TED Model expand out at the onset
of the short and long rains, and reach a peak surface area of tsetse present at the end of or
shortly after both wet seasons. During the dry seasons fly populations become
constrained due to unsuitable habitat conditions, primarily due to a lack of moisture
followed by high temperatures, reaching a minimum annual distribution before the onset
of the long and short rains. Further discussion of the dry season minimum tsetse
distributions can be found in section 5.1.3 entitled dry season reservoirs.

5.1.2 Percent Probability Map Results and Discussion

The individual TED Model binary tsetse distribution maps were summed and
divided by 69 (i.e., the number of scenes) to produce the percent probability map. Due to
the use of the fly front in the TED tsetse movement model, the TED Model percent
probability map displays the percent likelihood of encountering high tsetse population
densities at any time between the beginning of 2002 and the end of 2004 (Figure 5.6).
Since the TED Model only identifies areas of high population densities, in reality one
should expect to find tsetse in lower densities outside the predicted TED Model
distributions. However, the exact fly population densities and distance from the edge of

the TED Model predicted tsetse distributions will vary from locatiOn to location.

95

 

The TED Model
Percent Probabllity Map

 

    
  
 

Percent Probability
of Tsetse Presence

100%
Kilometers

0% so 100 200

Figure 5.6: The TED Model percent probability of tsetse presence map.

96

5.1.3 Tsetse Reservoirs
5.1.3.1 Results

The identification of the annual tsetse reservoirs was performed by identifying the
scene with the minimum annual surface area predicted to have high tsetse population
densities present for each year of the TED Model (i.e., 2002, 2003, and 2004) (Table 5.1).
Using the surface area table the scene that displayed the minimum annual area was
identified, which occurred each year during the cool dry season on the ordinal date of 289
(i.e., October 15th/ 16“) (Figure 5.7). The three scenes were then combined to produce a
map showing the potential spatial location of the tsetse reservoirs, which due to the
occurrence during the cool dry season will be forthwith referred to as dry season
reservoirs (DSR) (Figure 5.8).

Predicted constrained tsetse distributions during the hot dry season were also
examined to see if fly populations inhabited the same regions during both dry seasons.
The hot dry season minimum scenes occurred on ordinal dates 49, 65, and 81 (i.e.,
February 18‘“, 2002; March 6th, 2003; March 215‘, 2004). The three scenes were then
combined to produce a map showing the potential spatial location of the hot DSRs in

Kenya (Figure 5.9).

97

 

Ana... 2.3.3.3.. .832. b... 2.. 3.52.3 2 .52. @238 2.. so... 2.52.5.3...

8.03 3.2.5.... 5255... :8an b... .25.... 2.. 5.3 5.......5.a.9. 52.35.2125 2 682.2.

9.... .85....0 $552.8 .2... .25.... 22.. .553... be .25 2.. .e ”5.5. 2.... .33 .e .28 2.. .2... «:3
.5252. .25.). DH... 2.. .3 28.8.... 8.8. 9:... 2 622.5... 3.... 35...... .e 252.... 2.... ”hm Paw...

 

 

 

 

3;
mg 30" noou NOON
. . . . o
.m... m. m
.m. .m. u.
a.
m m" m . 8......
m. m.
.25.... noon . . 08.3
M..." s" .35... «8.. s"
_ .
u . u
u .. 89.3
v 80......
9:35.93: .
.352 38 . c8 3
r 80;:
83... be 8.3.. be 8.3.. be
:28 .80 .33 38 team 38
08.8

.82... am: 2.. B 8.28.... 2.2.35»...
once»... 53.555. .352 .0 GEE...

(awn) mv Gowns

98

5.1.3.2 Discussion

The scenes used to identify the tsetse cool DSRs all occurred roughly one month
after the ITCZ passed over the equator heading south (i.e., the September equinox), at the
end of the cool dry season, and at the beginning of the short rains. Compared to the cool
DSRs, the hot DSRs were less consistent in their timing and occurred on three separate
dates. When mapped the cool DSRs were more compact than the hot DSRs (Figure 5.8
and 5.9). The surface area predicted to have tsetse present in all three cool dry season
minimum scenes is 14,009 kmz, compared to 17,650 km2 infested with tsetse in all three
hot dry season minimum scenes. This led to the conclusion that the hot dry season is too
short to fully constrain tsetse populations into parcels of suitable habitat. For this reason
the hot DSRs were not used, rather just the minimum annual surface area (i.e., cool
DSRs) as originally intended, to identify the location of the annual DSRs. The
identification of the tsetse DSRs demonstrates that according to the TED Model tsetse are

seasonally constrained by the availability of suitable habitat (Figure 5.10).

99

 

L’U

 

 

The Location of Tsetse
Cool Dry Season Reservoirs

 

        

The number of years
when tsetse were
predlcted present at
the end of the

cool dry season

- One year

- Two years
- All three years

Kilometers

50 100 200

Figure 5.8: The location of the tsetse cool dry season reservoirs as predicted by the
TED Model for Kenya.

100

 

The Location of Tsetse
Hot Dry Season Reservoirs

 

a.»

“l. ‘j V. ‘
i

    

The number of years
when tsetse were
predicted present at
the end of the

hot dry season

- One year
- Two years
- All three years

Figure 5.9: The location of the tsetse hot dry season reservoirs as predicted by the
TED Model for Kenya.

101

The Location of Tsetse
Dry Season Reservoirs

 

      

Predicted Probability
of Tsetse Presence

100%

0% Kilometers
- Dry Season Reservoirs 50 100 200

Figure 5.10: The location of the tsetse dry season reservoirs compared to the TED
Model percent probability map.

102

5.2 Validation
5.2.1 Ground Truth Data Comparison
5.2.1.1 Results

The Mapcurves GOF method performed on the TED Model percent probability
map and the 1996 fly belts map resulted in a GOF score of 0.560 (Table 5.4). The
comparison between the TED Model and the FAO / IAEA distributions resulted in a
Mapcurves GOF score of 0.126 and a Kappa GOF score of 0.122. The Mapcurves GOF
method performed on the F AO / IAEA distribution map and the 1996 fly belts map

resulted in a GOF score of 0.650.

Table 5.4: Ground truth comparison GOF scores.

 

 

 

Maps Compared GOF Score
TED Model 8. ,,
1996 Fly Belts 0560
FAO / IAEA & .
1996 Fly Belts 055°
TED Model &

FAO , IAEA 0.126* / 0.122"

 

 

* indicates a Mapcurves GOF score
# indicates a Kappa Coefficient

5.2.1.2 Discussion

The Mapcurves GOF analysis and the Kappa coefficients show that the level of
agreement between all three maps is very low. However, the low level of agreement does
not assert that any one map is more or less accurate than another. The low level of
agreement between the three distribution maps analyzed in this study shows that the maps

do not agree where tsetse are located. The disagreement between the three maps might

103

 

be attributable to the differences in how each map was constructed (i.e., spatially explicit 'é
and dynamic fly habitat / movement model, logistic regression, and general fly capture
sites) and what the three maps display. The use of the fly front rather than individual fly
movement rates in the TED Model means that the TED Model identifies locations with
high tsetse population densities rather than anywhere a fly can be found. Since both the
FAO / IAEA and 1996 fly belts maps make no distinction between high or low fly
population densities and attempt to identify anywhere tsetse are present within Kenya, a
great deal of disagreement between these products should be expected.

The lack of fly movement in the FAO / IAEA distributions maps could lead to an
over estimation of tsetse distributions. The logistic regression model used to construct
the FAO / IAEA distribution maps does not take into account fly movement rates and
thus might be predicting tsetse in locations where suitable habitat is only present
seasonally. If neighboring tsetse populations are too far away for invasion to occur, then
portions of the FAO / IAEA map could identify tsetse distributions that are unable to
exits when intra-annual fluctuations in suitable habitat occur.

The 1996 fly belts map, although based on actual field data, may also have a high
degree of inaccuracy due to generalization (see section 3.2.3). Furthermore the 1996 fly
belts map displays the distribution of all eight tsetse species in Kenya. Both the TED
Model and FAO / IAEA distribution maps only focus on the morsitans group. Although
the distributions of the other four tsetse species is either relatively small in size or
overlapping regions where morsitans species are found, the addition of the other species

may lower the level of agreement in any comparison performed.

104

 

Although the F A0 / IAEA combined distribution map and the 1996 fly belts map
represent the best available distribution maps, a direct comparison to the TED Model
percent probability map is not possible since each tsetse distribution is constructed in
unrelated ways. Due to the different methods used to construct the three maps, a low
GOF score should be expected. That being said, if the premises described by Curran et
al. (2000) and outlined in Section 2.3.1 are correct, along with the parameters and
remotely sensed data used in the TED Model being accurate, then the TED Model should
be correctly identifying the spatial and temporal location of tsetse distributions. If the
TED Model is accurately predicting tsetse distributions, then the low GOF scores
produced in the ground truth comparison highlight the inaccuracy of the existing FAO /
IAEA and 1996 fly belts maps.

The ground truth comparison between the TED Model percent probability map
and the F AO / IAEA combined distribution map did allow for a comparison of
Mapcurves and Kappa GOF scores. Both methods displayed only a small difference
between the two calculated levels of agreement (0.004). The similar GOF score
calculated by the kappa coefficient and the Mapcurves GOF methods shows that
Mapcurves is a viable method of assessing agreement between two maps (DeVisser &

Messina, 2009).

105

5.2.2 Sensitivity Analysis
5.2.2.1 Moisture Sensitivity Index Results

The sensitivity index calculated for moisture parameter (i.e., NDVI) showed that
as the NDVI threshold was increased (i.e., tsetse required higher levels of moisture) the
amount of suitable tsetse habitat decreased (Figure 5.11). The SI calculated for moisture
ranged from 0.90 to 3.37 (Table 5.5), with only one SI below 1.00 allowing for moisture
to be labeled as having class IV sensitivity (i.e., extremely sensitive to any parameter
value change). Examination of the relative SI showed that at roughly 0.19 and 0.62 the
SI was just above the 1.00 sensitivity class threshold and at 0.15 and 0.67 the moisture

sensitivity class dropped from IV to 111 (i.e., from extremely sensitive to highly

 

 

 

 

 

sensitive).
120,000
100,000 4
E 00,000 ..
E
< 60,000 -
§
‘3 40,000 -
a
0)
20,000 -
0 u u u
0.00 0.20 0.40 0.60 0.80 1 .00
NDVI Threshold

Figure 5.11: NDVI thresholds and the corresponding surface area state variable
used in the sensitivity analysis. The larger light grey point represents the baseline
TED Model run.

106

 

Table 5.5: Moisture sensitivity analysis results. The A parameter and A area refers

to the change compared to the baseline model run (i.e., a threshold parameter of

 

 

 

0.39).
Threshold . . . .
Surface Sensltmty Relatlve

Pa(l';'a[l)n\z?r Area (kmz) A Parameter A Area I n d ex SI
0.00 111,105 1.00 1.64 1.64 0.09
0.15 109,621 0.62 1.60 2.61 0.43
0.19 107,774 0.51 1.56 3.04 1.10
0.21 105,395 0.46 1 .50 3.26 2.42
0.23 100,181 0.41 1.38 3.36 3.31
0.24 96,614 0.38 1.30 3.37 3.52
0.25 92,810 0.36 1.20 3.36 3.41
0.26 89,127 0.33 1.12 3.35 3.38
0.27 85,474 0.31 1.03 3.35 3.48
0.28 81,717 0.28 0.94 3.34 3.76
0.29 77,654 0.26 0.84 3.29 3.82
0.30 73,526 0.23 0.75 3.24 3.97
0.31 69,236 0.21 0.64 3.14 3.75
0.32 65,192 0.18 0.55 3.06 3.76
0.33 61,131 0.15 0.45 2.94 3.38
0.35 53,827 0.10 0.28 2.72 2.87
0.37 47,627 0.05 0.13 2.56 2.56
0.39 42,092 na na na na
0.41 37,578 -0.05 -0.11 2.09 2.09
0.42 35,805 -0.08 -0.15 1.94 1.64
0.43 34,192 -0.10 -0.19 1.83 1.49
0.45 31,409 -0.15 -0.25 1 .65 1 .29
0.47 28,781 -0.21 -0.32 1.54 1.22
0.49 26,266 -0.26 -O.38 1.47 1.17
0.51 23,808 -0.31 -0.43 1.41 1.14
0.52 22,192 -O.33 -0.47 1.42 1.50
0.53 21,270 -0.36 -0.49 1 .38 0.85
0.55 18,672 -0.41 -0.56 1.36 1.20
0.57 16,212 -0.46 -0.61 1.33 1.14
0.59 13,479 -0.51 -0.68 1.33 1.27
0.62 10,105 -0.59 076 1.29 1.04
0.67 5,382 -0.72 -0.87 1 .21 0.88
0.72 2,158 -0.85 -0.95 1.12 0.60
0.82 164 -1.10 -1.00 0.90 0.18

 

107

 

 

 

5.2.2.2 Maximum Temperature Sensitivity Index Results

The sensitivity index calculated for the maximum temperature parameter showed
that as the threshold was increased (i.e., tsetse could survive in higher temperatures) the
amount of suitable tsetse habitat increased (Figure 5.12). Sensitivity indices calculated
for maximum temperature ranged from 0.44 to 3.90 (Table 5.6), with only three below
1.00 allowing for maximum temperature to be labeled as having class IV sensitivity.
Examination of the relative SI showed that at roughly 29°and 44° the SI was just above
the 1.00 sensitivity class threshold and at 28° and 45° the maximum temperature

sensitivity class dropped from IV to III.

 

 

60,000 1 '

.3

SummAmmn’)
:3 .8
§ § §

10,000 -

 

 

 

 

01 I I I I I I

20 25 30 35 40 45 50 55 60

Maximum Temperature Threshold (C°)

Figure 5.12: Maximum temperature thresholds and the corresponding surface area
state variable used in the sensitivity analysis. The larger light grey point represents
the baseline TED Model run.

108

-.““l

 

 

Table 5.6: Maximum temperature sensitivity analysis results. The A parameter and
A area refers to the change compared to the baseline model run (i.e., a threshold

parameter of 40°C).

 

 

Threshold Surface

Sensitivity Relative

 

 

Par(ag)eter (12:?) A Parameter A Area In d ex SI
25 125 -0.375 -0.997 2.66 0.26
28 932 -0.300 -0.978 3.26 0.84
30 2,695 -0.250 -0.936 3.74 2.31
31 5,126 -0.225 -0.878 3.90 5.12
32 10,518 0200 -0.750 3.75 4.52
33 15,276 0175 -0.637 3.64 3.77
34 19,244 0150 -0.543 3.62 3.89
35 23.335 -0.125 -0.446 3.56 3.85
36 27.387 -0.100 -0.349 3.49 3.76
37 31,340 -0.075 -0.255 3.41 3.67
38 35,203 0050 -0.164 3.27 3.38
39 38,757 -0.025 -0.079 3.17 3.17
40 42,092 na na na na
41 45,032 0.025 0.070 2.79 2.79
42 47.522 0.050 0.129 2.58 2.37
43 48,988 0.075 0.164 2.18 1.39
44 50,051 0.100 0.189 1.89 1.01
45 50,662 0.125 0.204 1.63 0.58
46 51,037 0.150 0.213 1.42 0.36
47 51,208 0.175 0.217 1.24 0.16
48 51,312 0.200 0.219 1.10 0.10
49 51,346 0.225 0.220 0.98 0.03
50 51,359 0.250 0.220 0.88 0.01
60 51,381 0.500 0.221 0.44 0.00

 

 

109

 

5.2.2.3 Minimum Temperature Sensitivity Index Results

The sensitivity index calculated for the minimum temperature parameter showed
that as the threshold was decreased (i.e., tsetse flies could survive in colder temperatures)
the amount of suitable tsetse habitat increased (Figure 5.13). Sensitivity indices
calculated for minimum temperature ranged from 1.39 to 2.32 (Table 5.7), with no SI
below 1.00 allowing for minimum temperature to be labeled as having class IV
sensitivity. Examination of the relative SI showed that at roughly 4° / 11° the SI was just
above the 1.00 sensitivity class threshold and at 3° / 10° the minimum temperature

sensitivity class dropped from IV to III.

90,000 *
80,000
70,000 -

 

 
   
     

Surface Area (km?)
3 3 _8 S 8
§ § § § 5

10,000 -

 

 

 

 

0/7 5I12 10I17 15I22 20I27 25I32
Minimum Nightl Day Temperature Threshold (C°)

Figure 5.13: Minimum temperature thresholds and the corresponding surface area
state variable used in the sensitivity analysis. The larger light grey point represents
the baseline TED Model run.

110

 

 

Table 5.7: Minimum temperature sensitivity analysis results. The A parameter and
A area refers to the change compared to the baseline model run (i.e., a threshold
parameter of 10°C at night and 17°C during the day).

 

 

 

 

Threshold Surface . . . .
Parameter (C°) Are? Parafneter A Area 85?:521" Relgtive
Night Day (km )
0 7 78,575 0.59 0.87 1.47 0.21
1 78,065 0.53 0.85 1.61 0.39
2 9 77,094 0.47 0.83 1.77 0.69
3 10 75,386 0.41 0.79 1.92 0.74
4 11 73,552 0.35 0.75 2.12 1.23
5 12 70,512 0.29 0.68 2.30 2.19
6 13 65,093 0.24 0.55 2.32 2.44
7 14 59,060 0.18 0.40 2.28 2.73
8 15 52,306 0.12 0.24 2.06 2.15
9 16 46,992 0.06 0.12 1.98 1.98
10 17 42,092 na na na na
11 18 38,363 -0.06 —0.09 1.51 1.51
12 19 34,419 -0.12 -0.18 1.55 1.59
13 20 30,357 -0.18 -0.28 1.58 1.64
14 21 26,884 -0.24 -0.36 1.54 1.40
15 22 23,779 -0.29 -0.44 1 .48 1 .25
16 23 21,317 -0.35 -0.49 1.40 0.99
17 24 17,998 -0.41 -0.57 1.39 1.34
18 25 11,263 -0.47 -0.73 1.56 2.72
19 26 1,872 -0.53 -0.96 1.80 3.79
20 27 217 -0.59 -0.99 1.69 0.67

 

111

5.2.2.4 The Spatial Location of Potential Tsetse Expansion Results

The result of identifying the spatial location of where each of the four variables in
the TED Model habitat model constrains tsetse distributions from expanding was Figure
5.14. Maximum temperature when removed produced the smallest predicted expansion
of tsetse distributions, with only 4,276 km2 of added tsetse habitat. Minimum
temperature produced the second smallest expansion of tsetse distributions when
removed, with 33,833 km2 of added tsetse habitat. The majority of expanded tsetse
distributions due to the removal of minimum temperature was isolated in and around the
Kenyan highlands (Figure 5.15), making it of relatively localized importance. The
removal of land cover from the TED Model expanded tsetse distributions by 35,082 kmz,
the second highest variable with regards to potential tsetse expansion. The variable that
constrains tsetse distributions the most in the TED Model was NDVI (i.e., moisture). The
removal of moisture as a limiting ecological variable in the model resulted in fly
populations expanding 63,828 kmz, which would constitute a 150% increase in surface

area infested with tsetse compared to the baseline model run.

112

 

 

The Spatial Location of Potential Tsetse Expansion

 

       

- Predicted Tsetse Distribution
The Variable Limiting Tsetse ‘
Distributions in the TED Model
- Moisture

- Maximum Temperature
- Minimum Temperature
- Land Cover

B More Than One Variable

Kilometers

0 50 100 200

Figure 5.14: The spatial location of potential expansion of tsetse distributions if an
ecological variable was removed from the TED Model.

113

The Location of Potential Tsetse Expansion
Due to Increases in Minimum Temperature

 

 

 
 

Ethiopia

 

 

- Highlands
1? Nairobi

 
  
 

Areas where minimum
- temperature prevents
tsetse expansion

Predicted Probability
Of Tsetse Presence

100% Tanzania

0%

Figure 5.15: The location of potential tsetse expansion based on minimum
temperature in relation to the Kenyan Highlands. The TED Model percent
probability map is included to show the proximity of other fly populations in and
around the highlands.

114

5.2.2.5 Sensitivity Analysis Discussion

The sensitivity analysis explored the impact that each variable had in limiting or
expanding the predicted tsetse distributions produced by the TED Model. Maximum
temperature produced the highest SI values (i.e., 3.90 and 5.12), followed by NDVI and
minimum temperature. This could be interpreted as high temperatures having the
greatest impact on tsetse distributions; however, the majority of the SI values calculated
for maximum temperature occurred when the threshold was lowered (i.e., tsetse require
cooler temperatures to survive). By raising the maximum temperature threshold (i.e.,
tsetse can survive in warmer temperatures), little change in tsetse distributions is
observed. This is best exemplified in Figure 5.12, which shows very little expansion of
tsetse when maximum temperature is removed from the TED Model.

In some models this might be cause to remove the variable in an effort to increase
parsimony, but not in the case of the TED Model. A comparison of each parameter’s
percent change relative to the predicted tsetse distributions surface area (Figure 5.16)
shows that when the maximum temperature threshold is raised after a few degrees no
significant change would be observed, which is consistent with it lack of potentially
expanding tsetse distributions. However, when the maximum temperature threshold is
lowered, the steepest slope of any of the variables is observed. Thus changes in
maximum temperature would have only a slight ability to expand and a high potential to

limit tsetse distributions.

115

 

 

 

 

    

 

 

 

   

120,000
+ NDVI
100,000 4 -o— Max Temp
—-— Min Temp
.. 80,000 -
“a
5.
,, 60,000 -
2
<
40,000 «
20,000 .
o 1 V
-1 00% -50% 0% 50% 1 00%

Change in Threshold Parameter (%)

Figure 5.16: Combined and standardized parameter threshold values and the
corresponding surface area state variable used in the sensitivity analysis.

In addition to having the highest potential to limit tsetse populations, removal of
maximum temperature would increase the TED Model’s sensitivity to moisture.
Evidence for this can be found in the regions depicted in Figure 5.12 that show more than
two variables effecting tsetse distributions (i.e., when a variable is removed another
variable still constrains tsetse populations in that location). For example on and around
Mount Kenya tsetse distributions are constrained by both minimum temperature and
unsuitable land cover (e.g., bare rock). When either minimum temperature or land cover
is removed from the TED Model, the other variable still prevents tsetse expansion into
that location. Maximum temperature and moisture, which as previously mentioned, are

highly correlated and act in a similar way in northern Kenya. When either variable is

116

 

 

removed, the other still makes that location unsuitable habitat for tsetse. Removal of
maximum temperature from the TED Model would greatly increase the TED Model’s
sensitivity to moisture and decrease the model’s robustness.

NDVI, which has the second highest SI values, has the opposite relationship as
compared to maximum temperature. The highest SI values are calculated when the
NDVI threshold is lowered (i.e., tsetse require less moisture to'survive). Raising the
NDVI threshold (i.e., tsetse require more moisture to survive) only gradually limits the
predicted surface area when compared to maximum and minimum temperature.
Therefore changes in moisture have least potential to limit and the greatest potential to
expand tsetse distributions.

Minimum temperature, which has the lowest SI values, has the potential to both
limit and expand tsetse populations. If the threshold is raised (i.e., tsetse require warmer
minimum temperatures), then tsetse distributions are limited, but not to the same degree
as lowering the maximum temperature threshold. If the threshold is lowered (i.e., tsetse
can survive in colder temperatures), then tsetse distributions are able to expand, but not to
the same extent that lowering the NDVI threshold would expand fly populations.

Natural systems can display different patterns and relationships at different scales
(Pigozzi, 2004). The moderate sensitivity displayed by minimum temperature is likely
linked to the broad scale at which the SI was calculated (i.e., all of Kenya). In southern
Africa minimum temperature is considered to be the most influential climate variable
limiting tsetse distributions (Leak, 1999). In Kenya, which is located on the equator,
minimum temperature only affects tsetse in the highland areas around Mount Kenya and

the rim of the rift valley (Figure 5.15). However, if the SI was calculated only for the

117

 

highland areas rather than all of Kenya, minimtun temperature would have the highest SI
and be the most important ecological variable with regards to both limiting and
expanding tsetse distributions.

Land cover, which did not have a calculated SI, had the second highest potential
to expand tsetse distributions. However, as discussed in section 5.1.1.2 (Table 5.2)
suitable tsetse land cover was replaced by unsuitable land cover by an average of 6,867
km2 annually between 2002 and 2004. If this trend continues, then land cover conversion
has little potential to expand tsetse distributions. Some claim (see Reid et al., 2000;
Bourn et al., 2001; Hargrove, 2003) that human conversion of land cover from suitable to
unsuitable, similar to the land modification practices of traditional east African societies,
is the only way to effectively control tsetse populations in the future.

One unavoidable driver of change in both LULC and tsetse distributions is
climate change. Some climate change scenarios predict possible increased precipitation
during the rainy seasons and warmer temperatures across East Africa (Boko et al., 2007).
Since moisture and minimum temperature are considered the most sensitive variables to
possible tsetse expansion in the TED Model, the possibility of increased precipitation and
warmer temperatures should be of concern. Figure 5.14 affords a look into potential
changes in future tsetse distributions by examining the changes resulting from removing a
variable from limiting fly populations. Using minimum temperature as an example, if the
climate became warmer in the highland areas (i.e., essentially the same as decreasing the
minimum temperature threshold), then tsetse distributions could expand into regions
previous unsuitable (areas denoted in blue in Figure 5.14 and 5.15). However, if

temperatures rise not only in the highlands, but the rest of Kenya as well, then maximum

118

 

temperature would also limit tsetse distributions in some regions. The locations of these
areas would depend on the level of warming, but due to the sensitivity of maximum
temperature in the TED Model, tsetse distributions could be greatly impacted if warming

does occur.

5.3 Uncertainty

George Box (1976) said “since all models are wrong the scientist must be alert to
what is importantly wrong. “It is inappropriate to be concerned about mice when there
are tigers abroad.” In the following sections I will identify a few of the “tigers” (i.e.,
major underlying assumptions and sources of uncertainty) within the TED Model.
5.3.1 Spatial Resolution

Arguably the largest source of uncertainty is the spatial resolution of the data used
in the TED Model. All outputs from the TED Model are at a spatial resolution of 250m,
however, only the NDVI data are collected at that spatial resolution. The temperature
and LULC data are both at a coarser spatial resolution of 1km. Regardless of the exact
spatial resolution of the data used in the TED Model, the underlying assumptions in using
remotely sensed data within the TED Model are that conditions are uniform within a
pixel, and that all suitable tsetse habitat can be identified using the MODIS products at
250m and 1km resolutions. These assumptions are surely violated when one considers
that a particular plant species (e.g., Cordia sinensis) can provide tsetse with suitable
habitat and be much smaller that 250m2, thus conditions within each pixel are not
uniform and some suitable habitat is unable to be detected.

However, the goal of the TED Model was not to identify the spatial location of all

suitable tsetse habitats, but rather test that intra-annual and inter-annual fluctuations in

119

suitable tsetse habitat occur. To that end, if the conditions within each pixel are close
approximations to the average conditions in situ, then the TED Model can be used to test
the existence of fluctuations in tsetse distributions at various temporal resolutions.
Nevertheless, the second part of my research was to identify the geographic location of
tsetse reservoirs, which is influenced by the exclusion of habitats below the spatial
resolution of the remotely sensed data is TED Model. Ultimately, any use of the TED
Model will need to be accompanied with the understanding that the model may exclude
tsetse distributions that rely on reservoirs below 250m spatial resolution.
5.3.2 Fly Movement

In addition to the TED model not being able to identify suitable tsetse habitat
below the spatial resolution constraints imposed by the remotely sensed data, the TED
Model does not identify the spatial location of tsetse that move at a faster rate than the
general fly populations (i.e., the fly front). The use of the fly front has the potential
introduce uncertainty into the TED Model by not allowing tsetse to expand out, albeit at
lower population densities, to their maximum extent. In some situations this might
prevent the TED Model from predicting invasion or reinvasion of areas that were
previously unable to sustain tsetse do to a particular event or anomaly captured in the
remotely sensed data (e.g., a drought or a change in human land use practices).

However, in constructing the TED Model I was less concerned about accurately
identifying everywhere on the landscape tsetse might exist, but rather modeling seasonal
fluctuations in suitable tsetse habitat and the timing and geographic location of tsetse
reservoirs. That being said, the use of the fly front was the most parsimonious way to

represent tsetse distributions.

120

. .1

 

5.3.3 Data
5.3.3.1 NDVI

NDVI essentially measures the presence and condition of green vegetation
(Lillesand et al., 2004), and was used as a surrogate for moisture in the TED Model
(Williams et al., 1992). The use of NDVI in this manner is based on the assumption that
healthy green vegetation requires water to be present. Although this assumption is
generally accurate, in East Africa the relationship between NDVI and precipitation is less
straightforward (Eklundh, 1998) and is often associated with a lag period between the
height of a precipitation event and the peak NDVI values (Martiny et al., 2006).
However, tsetse do not require free standing water to be present in the environment as
they obtain water from the taking of blood meals. Rather adult tsetse require moisture to
be present in the atmosphere to avoid desiccation, similar to that of plants. When green
vegetation starts to die off (i.e., lower NDVI values), generally both soil moisture and
atmospheric moisture are decreasing (Sandholt et al., 2002), and thus the environment is
less suitable for tsetse. The lag between NDVI and precipitation may affect the timing at
which tsetse could expand out in the TED Model, since after a precipitation event and
before the green-up of vegetation measured by NDVI occurs, suitable moisture levels
could exist. Although the lag between precipitation events and NDVI does not affect the
identification of fluctuations in suitable habitat and the establishment of tsetse reservoirs,
it is an avenue of possible model improvement in the future.
5.3.3.2 Hosts

In section 2.1.4 I suggest that a spatial model should include a variable that

represents each theory of what causes seasonal fluctuations in tsetse distributions (i.e.,

121

 

hosts, moisture, and temperature). However, in section 3.1.1 I state that in Kenya no
accurate intra-annual data on host distributions currently exist and therefore host
populations are not accounted for in the TED Model habitat model. The assumption that
food for tsetse is present uniformly throughout Kenya at all times in the TED Model is a
source of uncertainty since both wild and domestic hosts migrate at various time scales
for a variety of reasons (e. g., Homewood et al., 2001).

A primary factor in animal migration is search for food (Eloff, 1959), which in the
semi-arid regions of Kenya can be considered a limited resource. The use of NDVI as a
surrogate for moisture, could also be used a surrogate for vegetation fodder for animals.
If vegetation fodder for animals is limited in Kenya, then the presence of healthy green
vegetation would attract animals, which in turn would mean hosts would be available for
tsetse to feed from. Thus, the use of NDVI in the TED Model could possibly be
considered to function as both a surrogate for moisture and host species in the TED
Model.
5.3.3.2 Land Cover

The use of land cover data in the form of MODIS type 1 Global Land Cover
product introduces uncertainty in the TED Model in two ways: 1) the assumption that
MODIS type 1 LULC is correctly identifying suitable tsetse land cover types and, 2) the
temporal resolution of the data. The assumption that MODIS type 1 correctly identifies
suitable tsetse land cover stems from the fact that MODIS type 1 was not specifically
designed to accomplish this task. Although through the use of the modified Mapcurves

GOF test MODIS type 1 was identified as a top performer in identifying suitable tsetse

122

 

land cover, in an ideal situation a land cover data set would be specifically designed for
this purpose rather than adapting a global LULC product.

In addition to possible classification inaccuracies with regards to tsetse, MODIS
type 1 LULC product is only produced annually. Thus, in the TED Model land cover is
relatively invariant when compared to the temporal resolution of the climate data used in
the TED habitat suitability model. Since LULC change is a dynamic process and is
constantly occurring on the Earth’s surface, the use of an essentially static LULC data set
means that the variable is a departure from reality. Future work in modeling of tsetse
may require the construction of both a tsetse specific land cover data set and LULC

change model to account for change in land cover at a more appropriate temporal scale.

123

 

CHAPTER 6
CONCLUSION

African trypanosomiasis is a neglected tropical disease with an estimated 13,000
new human cases in Sub-Saharan Africa annually (CDC, 2008a / 2008b). In addition to
the problems associated with being a neglected tropical disease, trypanosomiasis poses
the unique problem of infecting both humans and domestic livestock populations, thus
increasing the total number of humans affected by the disease. Currently no vaccines to
prevent infection exist, and trypanocidal drugs are toxic and potentially lethal to the
patient. Since there is no way to vaccinate against infection, and the means to cure the
disease are problematic, most efforts to control trypanosomiasis have focused on the
tsetse fly.

Efforts to control tsetse have been chronically hampered by identification of
legitimate infested areas, reinvasion of tsetse into previously cleared regions, and
substantial costs in conjunction with limited resources. Current models predict tsetse
distributions at a particular moment in time and do not inform control efforts on any
possible fluctuations in tsetse distributions. If tsetse distributions do fluctuate over space
and time, then a dynamic spatially explicit model that predicts when and where fly
reservoirs occur, could be used to maximize the limited resources available for tsetse
control. To track any possible fluctuations in tsetse distributions over space and time,
and identify the spatial location and timing of tsetse reservoirs in Kenya, 1 have
constructed the Tsetse Ecological Distribution (TED) Model.

The TED Model is a 250m resolution raster based spatially explicit dynamic

model that predicts tsetse distributions at 16 day intervals over a 4 year period between

124

 

the beginning of 2001 and the end of 2004, based on habitat suitability and fly movement I! -*
rates. At its simplest the TED Model can be described in two separate parts: 1) a
spatially explicit model that identifies suitable tsetse habitat using remotely sensed land
cover and climate data and 2) a fly movement model that utilizes the notion of a fly front
and tsetse movement rates. The habitat suitability model uses four remotely sensed data
sets: 1) day LST, 2) night LST, 3) NDVI as a surrogate for available moisture, and 4) a
publicly available LULC data set. Since fifteen LULC products are publicly available for
Kenya, I analyzed each product using a modified Mapcurves GOF test to identify which
should be used the TED Model. My analysis showed that although Afiicover produced
the highest Mapcurves GOF scores, the 11011 MODIS type 1 Global Land Cover product
should be used due to its ability to account for annual LULC change and align temporally
with the remotely sensed climate data used in the TED Model.

Provided the data used in the TED Model are accurate, and the parameters used to
identify suitable versus unsuitable habitat are correct, then the TED Model is able to track
intra-annual and inter-annual fluctuations in tsetse distributions, and identify the season,
approximate date, and geographic location of constrained tsetse distributions. In addition
to tracking seasonal fluctuations in tsetse distributions, the TED Model has the ability to
provide information on the drivers of any observed fluctuations. All four variables used
in the TED Model habitat model have the potential to both limit (i.e., changes in
threshold values cause predicted distributions to decrease in surface area) and expand
(i.e., changes in threshold values cause predicted distributions to increase in surface area)
tsetse distributions. A parameter sensitivity analysis on the four variables was performed

to quantify the degree to which each variable influenced seasonal fluctuations in tsetse

125

 

distributions. The sensitivity analysis was performed via calculating both a standard
sensitivity index and the newly developed relative sensitivity index, along with
identification of regions of potential tsetse expansion.

By using the TED Model I have confirmed that tsetse distributions do fluctuate at
intra-annual (i.e., seasonally) and inter—annual temporal resolutions, and are constrained
to tsetse reservoirs at the end of both dry seasons. The fly populations predicted with the
TED Model expand with the onset of the long rains (roughly beginning of March),
contract at the start of the cool dry season (roughly beginning of June), expand again with
the commencement of the short rains (roughly beginning of November), and contract
during the hot dry season. Lack of moisture followed by maximum temperatures were
determined to have the greatest impact on tsetse distributions within the TED Model, and
were determined to be the root cause of the establishment of dry season reservoirs.

The sensitivity analysis showed that maximum temperature had the greatest
potential to limit fly populations, with a 1°C increase in maximum temperatures resulting
in a ~8% decrease in the surface area of tsetse distributions. Moisture was calculated to
have the greatest potential to expand tsetse distributions, with a 5% increase in moisture
resulting in a ~13% increase in tsetse distributions. Minimum temperature produced
moderate sensitivity indices for both limiting and expanding tsetse distributions.
However, analysis of where tsetse would expand if low temperatures did not limit their
distributions showed that minimum temperatures greatly influences fly populations in the
highland areas around Mount Kenya and the rim of the rift valley. Since a large portion
of the population of Kenya lives in this area, a slight warming of minimum temperatures

in this locale could have a significant impact.

126

Some climate change scenarios predict possible increased precipitation during the
rainy seasons and warmer temperatures across East Africa (Boko et al., 2007). Since
moisture and minimum temperature are considered the most sensitive variables to
possible tsetse expansion in the TED Model, the possibility of increased precipitation and
warmer temperatures should be of concern. However, if temperatures rise not only in the
highlands, but the rest of Kenya as well, then due to the sensitivity of maximum
temperature in the TED Model tsetse distributions could be greatly constrained in some
regions.

The final variable in the TED Model habitat model is land cover. Land cover did
not have a calculated SI due to the optimum LULC product being identified via the
modified Mapcurves GOF analysis, but did have the second highest potential to expand
tsetse distributions. However, the ability for LULC to expand tsetse distributions was
estimated by removing land cover as a variable in the TED habitat suitability model. As
this scenario is unrealistic, the use of the TED Model to estimate the impact of land cover
conversion on tsetse distributions may be inappropriate. However, suitable tsetse land
cover was replaced by unsuitable land cover by an average of 6,867 km2 annually
between 2002 and 2004. If this trend continues, then land cover conversion has little
potential to expand tsetse distributions. Some claim (see Reid et al., 2000; Bourn et al.,
2001; Hargrove, 2003) that human conversion of land cover from suitable to unsuitable,
similar to the land modification practices of traditional east African societies, is the only
way to effectively control tsetse populations in the future.

Unfortunately in its current state the TED Model can only make broad

generalizations about future changes in tsetse distributions. In order for the TED Model

127

 

 

to predict more specific future fly distribution scenarios it would need to be coupled with
other models (e.g., LULC or climate change models). Another drawback of the TED
Model is the lack of fly population densities, which is often of particular interest to
control efforts. Incorporating a population density model into the TED Model would
pose a significant challenge. However, the benefit to management officials to know what
control technique to use based on the estimated numbers of flies and the optimal location
in which deploy their limited resources, makes this avenue of research highly promising.

Although the potential exists to improve the TED Model, the model in its current
form is quite informative. In particular the confirmation that tsetse distributions should
fluctuate over space and time based on the availability of ecologically suitable habitat
advances the understanding and fly population dynamics. The identification of potential
dry season reservoirs has the potential to inform to future tsetse control efforts of where
and when to target tsetse populations. The percent probability map produced from the 69
individual tsetse distributions generated by the TED Model also advances the current
level of knowledge surrounding tsetse distributions by identifying possible
undocumented expansion of fly populations (e.g., along the Kenya / Somali border). The
populations along the border are of particular interest since they are rumored to exist
(Maitima personal communication, 2008), but are not displayed on most recent

distributions maps, modeled or otherwise (Figure 6.1).

128

 

The TED Model Percent Probability Map
Compared to the 1996 Fly Belts

i
40°!

 

   
    
   
  
 

I
35':

Ethiopia

 

 

Predlcted Probability
Of Tsetse Presence
100%

0%

- 1996 Fly Belts Tanzania

 

 

Figure 6.1: The TED Model percent probability map and 1996 fly belts maps. Of
particular interest are the TED Model predicted fly populations that do not fall
within the boundaries of the fly belts.

129

 

111111

 

Furthermore the TED Model percent probability map can be considered the most as
up to date map of tsetse distributions in Kenya, similar to the modeled distributions
produce by PAATIS and the FAO / IAEA at the time they were created. However, just
like the PAATIS and the FAO / IAEA maps, the TED Model outputs (i.e., percent
probability map along with the spatial location and timing of the dry season reservoirs)
need to be validated with some form of field validation. Ideally this would entail using
fly traps to collect data on tsetse presence using transects that pass through predicted dry
season reservoirs, extending out past the maximum distance at which tsetse are predicted
to spread during the intervening wet seasons. This form of field validation would test the
establishment, location, and timing of dry season reservoirs.

To perform a full model accuracy assessment a complete fly census for all of
Kenya would be ideal. Unfortunately the cost and scope of such an endeavor prohibits
efforts by organizations like PATTEC from collecting such data at this time (ICIPE
personal communication, 2008). In place of a complete fly census smaller study sites
could be used, preferably in a location predicted by the TED Model to support tsetse
populations previously undocumented (e.g., the previously mentioned fly populations
along the Kenya / Somali border, though due to the inherent instability in this region
another site would be preferable).

Despite the lack of field validation, the TED Model does show that tsetse
distributions fluctuate over space and time at intra-annual and inter-annual time scales.
These fluctuations in fly populations can be tracked using remotely sensed data, allowing
for the identification of tsetse reservoirs. Since the tsetse reservoirs occur during the dry

season and fly populations are most vulnerable at that time (see Schofield & Maudlin,

130

 

2001; Peter et al., 2005), this research will maximize the limited resources available and

increase the likelihood of success for future tsetse control campaigns.

131

 

l
’7

 

APPENDIX

The Tsetse Ecological Distribution Model Script
#
# The Tsetse Ecological Distribution (TED) Model.py

# Created By: Mark DeVisser

# Created on: 2/22/09

# The TED Model is designed to be run in ArcGIS 9.2 in Batch Mode
#

 

 

# Import system modules
import sys, string, 05, arcgisscripting

# Create the Geoprocessor object
gp = arcgisscripting.create()

# Set Cell Size to 250m for all outputs
gp.cellSize = "250"

# Variables List:

# MODIS NDVI Data, file name is the ordinal date of all data
NDVI = gp.GetParameterAsText(0)

# Base Name with Extension
DateExt = os.path.basename(NDVl)

# Base Name (ordinal date)
Date = string.split(DateExt,".")[0]

# File Directory Path
TED_Dir_NDVl = os.path.dirname(NDVl)
TED_Dir = string.split(TED_Dir_NDVI,"N")[0]

# MODIS Day Land Surface Temperature Data
LST_Day = TED_Dir + "\\LST_Day\\" + Date + "_LST_Day_1km.img"

# MODIS Night Land Surface Temperature Data

LST_Night = TED_Dir + "\\LST_NightW' + Date + "_LST_Night_1km.img"

# Classified Day LST
Day_LST_RC = TED_Dir + "\\Model_Data\\Day_LST_RC"

# Classified Night LST
Night_LST_RC = TED_Dir + "\\Model_Data\\Night__LST_RC"

# Classified NDVI
NDV|_RC = TED_Dir + "\\Model_Data\\NDVl_RC"

# Suitability Based on Day and Night LST
LST_Suit = TED_Dir + "\\Model_Data\\LST_Suit"

132

 

 

 

# Suitability Based on Day LST, Night LST, and NDVI L..—--'
Climate_Suit = TED_Dir + "\\Model_Data\\Climate_Suit"

# Suitability Based on Day LST, Night LST, NDVI, and Land Cover
, Total_Suit = TED_Dir + "\\Model_Data\\Total_Suit"

# Tsetse Distribution on the julian date in the file name
Current_Dist = TED_Dir + "\\Output\\" + Date + "_Tsetse_Dist"

# MODIS Land Cover Type 1 suitability data
MODIS_LC_RC = TED_Dir + "\\Model_Data\\MODlS_t1_2001_1km_suit.img"

# Expanded Tsetse Distribution based on fly movement rate.
# This data set also functions as the Initial Distribution and must be "reset” before each model run
Max_Dist = TED_Dir + "\\Model_Data\\Max_Dist"

# Processing Steps:

# Reclassify NDVI
gp.Reclassify_sa(NDVl, "Value", "-3 0.22 0;0.22 3 1", NDVI__RC, "DATA")

# Reclassify Day LST
gp.Reclassify_sa(LST_Day, "Value", "-30 17 0;17 40 1;40 100 0", Day_LST_RC, ”DATA")

# Reclassify Night LST
gp.Reclassify_sa(LST_Night, "Value", "-30 10 0:10 40 1;40 100 0", Night_LST_RC, "DATA")

# Reclassified Day LST Multiplied by Reclassified Night LST to create LST Suitability
gp.Times_sa(Day_LST_RC, Night_LST_RC, LST_Suit)

# Reclassified NDVI Multiplied by Temperature Suitability to create Climate Suitability
gp.Times_sa(NDVl_RC, LST_Suit, Climate_Suit)

# Reclassified Land Cover Multiplied by Climate Suitability to create Total Suitability
gp.Times_sa(Ciimate_Suit, MODIS_LC_RC, Total_Suit)

# Total Suitability Multiplied by Expanded Tsetse Distribution to create the Tsetse Distribution on
# the ordinal date in the file name
gp.Times_sa(Total_Suit, Max_Dist, Current_Dist)

# Expand Current Tsetse Distribution by 500m to simulate potential fly distribution expansion
# under optimal conditions
gp.Expand_sa(Current_Dist, Max_Dist, ”2", ”1")

# Delete extraneous variables before next iteration
gp.delete (NDVI_RC)

gp.delete (Day_LST_RC)

gp.delete (Night_LST_RC)

gp.delete (LST_Suit)

gp.delete (Climate_Suit)

gp.delete (Total_Suit)

133

 

___ J

 

REFERENCES

Adams, J. S., & McShane, T. O. (1992). The myth of wild Africa: conservation without
illusion. University of California Press, Berkeley.

Adlington, D., Randolph, S. E., & Rogers, D. J. (1996). Flying to feed or flying to
mate: gender differences in the flight activity of tsetse (Glossina palpalis).
Physiological Entomology, 21(2), 85-92.

Afrane, Y. A., Zhou, G., Lawson, B. W., Githeko, A. K., & Yan, G. (2007). Life-table
analysis of Anopheles arabiensis in western Kenya highlands: Effects of land

covers on larval and adult survivorship. The American Journal of Tropical
Medicine and Hygiene, 77(4), 660-666.

Allospp, R. (1985). Variation in the rates of increase of Glossina morsitans centralis and
their relevance to control. Journal of Applied Ecology, 22(1), 91-104.

Allsopp, R. (2001). Options for vector control against trypanosomiasis in Africa.
Trends in Parasitology, 1 7(1), 15-19.

Allsopp, R., & Baldry, D. A. (1972). A general description of the Lambwe Valley area
of South Nyanza District, Kenya. Bulletin of the World Health Organization,
47(6), 691-697.

Alsop, N. (1994). A Review of Recent Approaches to Sustainable Control and an
Analysis of the Potential of Modern Techniques for Large Scale Use. Food and
Agriculture Organization of the United Nations.

Anene, B. M., Onah, D. N., & Nawa, Y. (2001). Drug resistance in pathogenic African
trypanosomes: what hopes for the future? Veterinary Parasitology, 96(2), 83-100.

Artzrouni, M., & Gouteux, J. P. (2001). Population Dynamics of Sleeping Sickness: A
Microsimulation. Simulation & Gaming, 32(2), 215-227.

Artzrouni, M., & Gouteux, J. P. (2003). Estimating tsetse population parameters:
application of a mathematical model with density-dependence. Medical and

Veterinary Entomology, 1 7(3), 272-279.

Artzrouni, M., & Gouteux, J. P. (2006). A parity-structured matrix model for tsetse
populations. Mathematical Biosciences, 204(2), 215-231.

134

 

 

Austen, E., & Hegh, E. (1922). T setse-flies; Their characteristics, distribution and
bionomics: with some account of possible methods for their control (p. 188).
London, UK: The Impererial Bureau of Entomology.

Awange, J. L., Ogalo, L., Bae, K. H., Were, P., Omondi, P., Omute, P., et al. (2008).
Falling Lake Victoria water levels: Is climate a contributing factor? Climatic
Change, 89(3), 281-297.

Baldry, D. (1972). A history of Rhodesian sleeping sickness in the Lambwe Valley.
Bulletin of the World Health Organization (WHO), 47(6), 699-718.

Barrett, K., & Okali, C. (1998). Partnerships for tsetse control — community
participation and other options. World Animal Review, 90, 11.

Barrett, M. P., Boykin, D. W., Brun, R., & Tidwell, R. R. (2007). Human African
trypanosomiasis: pharmacological re-engagement with a neglected disease.
British Journal of Pharmacology, 152(8), 1 155 -1 171.

Barrett, M. P., Burchmore, R. J. S., Stich, A., Lazzari, J. 0., Frasch, A. C., Cazzulo, J.
J ., et al. (2003). The trypanosomiases. The Lancet, 362(9394), 1469-1480.

Barrett, S. V., & Barrett, M. P. (2000). Anti-sleeping sickness drugs and cancer
chemotherapy. Parasitology Today, 16(1), 7-9.

Berrang-Ford, L., Odiit, M., Maiso, F ., Waltner-Toews, D., & McDermott, J. (2006).
Sleeping sickness in Uganda: revisiting current and historical distributions.
African Health Sciences, 6(4), 223-231.

Bett, B., Irungu, P., Nyamwaro, S. O., Murilla, G., Kitala, P., Gathuma, J., et al. (2008).
Estimation of tsetse challenge and its relationship with trypanosomosis incidence
in cattle kept under pastoral production systems in Kenya. Veterinary
Parasitology, 155(3-4), 287—298.

Biesemans, J ., Hendrickx, G., & Van Camp, N. (2002). T setse presence-absence
prediction model for Glossina austeni and Glossina brevipalpis in KwaZulu
Natal—South A fiica; Part 2: Modeling tsetse populations. Methodological
approaches used. (pp. 20-53). Technical Report, Agriculture and Veterinary
Intelligence and Analysis.

Boettner, G. H., Elkinton, J. S., & Boettner, C. J. (2000). Effects of a biological

control introduction on three nontarget native species of saturniid moths.
Conservation Biology, 14(6), 1798-1806.

135

A“

Boko, M., Niang, 1., Nyong, A., Vogel, C., Githeko, A., Medany, M., et al. (2007). ’-"
Africa. Climate Change 2007: Impacts, Adaptation and Vulnerability.
Contribution of Working Group II to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change, ML Parry, OF Canziani, JP
Palutikofi PJ van der Linden and CE Hanson, Eds. Cambridge University Press,
Cambridge UK.

Bourn, D., Reid, R., Rogers, D., Snow, B., & Wint, W. (2001). Environmental change
and the autonomous control of tsetse and trypanosomosis in Sub-Saharan Africa.
Oxford, UK: Information Press Limited.

Box, G. E. (1976). Science and statistics. Journal of the American Statistical Association,
71(356), 791—799.

Brun, R., Schumacher, R., Schmid, C., Kunz, C., & Burri, C. (2001). The phenomenon
of treatment failures in Human African Trypanosomiasis. Tropical Medicine &
International Health, 6(11), 906-914.

Bursell, E. (1957). The Effect of Humidity on the Activity of Tsetse Flies. Journal of
Experimental Biology, 34(1), 42-51.

Bursell, E. (1961). Starvation and desiccation in tsetse flies (Glossina). Entomologia
Experimentalis et Applicata, 4(4), 301-310.

Bursell, E. (1963). Tsetse-fly physiology: a review of recent advances and current aims.
Bulletin of the World Health Organisation, 28, 703-709.

Buxton, P. (1955). The Natural History of Tsetse Flies: An Account of the Biology of the
Genus Glossina (Diptera) (p. 816). London, UK: H. K. Lewis and Co. Ltd.

Carson, R. (1962). Silent Spring. Boston, MA: Houghton Mifflin.

Catley, A., Irungu, P., Simiyu, K., Dadye, J ., Mwakio, W., Kiragu, J ., et al. (2002).
Participatory investigations of bovine trypanosomiasis in Tana River District,
Kenya. Medical and Veterinary Entomology, I 6(1), 55-66.

CDC. (2008a). East Afiican Trypanosomiasis Fact Sheet. Centers for Disease Control
and Prevention, Division of Parasitic Diseases.

CDC. (2008b). West African Trypanosomiasis Fact Sheet. Centers for Disease Control
and Prevention, Division of Parasitic Diseases.

Cecchi, G., Mattioli, R. C., Slingenbergh, J., & De La Rocque, S. (2008). Land cover
and tsetse fly distributions in sub-Saharan Africa. Medical and Veterinary
Entomology, 22(4), 364-373.

136

 

Cecchi, G., Mattioli, R. C., Slingenbergh, J ., & de La Rocque, S. (2008). Standardizing
land cover mapping for tsetse and trypanosomiasis decision making. PAAT
Technical and Scientific Series 8 (pp. 1-97). PAAT Information Service
Publications, Rome: Food and Agriculture Organisation of the United Nations;
IAEA Joint Division, International Atomic Energy Agency.

Challier, A. (1982). The ecology of tsetse (Glossina spp.)(Diptera, Glossinidae): a
review (1970-1981). Insect Science and its Application, 3, 97-143.

Cherenet, T., Sani, R. A., Speybroeck, N., Panandam, J. M., Nadzr, S., & Van den
Bossche, P. (2006). A comparative longitudinal study of bovine trypanosomiasis
in tsetse-free and tsetse—infested zones of the Amhara Region, northwest Ethiopia.
Veterinary parasitology, 140(3-4), 251-258.

Cockerell, T. D. A. (1917). A fossil tse-tse fly and other Diptera from F lorissant,
Colorado. Proceedings of the Biological Society of Washington, 30, 19-22.

Curran, P. J ., Atkinson, P. M., Milton, E. J ., & Foody, G. M. (2000). Linking remote
sensing, land cover and disease. Advances in Parasitology, 4 7, 37-80.

Davies, H. (1964). The eradication of tsetse in the Chad river system of Northern
Nigeria. Journal of Applied Ecology, 387-403.

Davies, H. (1971). Further eradication of tsetse in the Chad and Gongola River systems
in northeastern Nigeria. Journal of Applied Ecology, 8(2), 563-578.

De La Rocque, S., Lefrancois, T., Reifenberg, J. M., Solano, P., Kabore, I., Bengaly, Z.,
et al. (1998). PCR Analysis and Spatial Repartition of Trypanosomes Infecting
Tsetse Flies in Sideradougou Area of Burkina Faso a. Annals of the New York
Academy of Sciences, 849(1), 32-38.

Delespaux, V., & de Koning, H. P. (2007). Drugs and drug resistance in African
trypanosomiasis. Drug Resistance Updates, 10(1-2), 30-50.

DeVisser, M. H., & Messina, J. P. (2009). Optimum land cover products for use in a
Glossina-morsitans habitat model of Kenya. International Journal of Health
Geographics, 8, 39.

Di Gregorio, A., & Jansen, L. J. M. (1998). A new concept for a land cover
classification system. The Land, 2(1), 55-65.

Di Gregorio, A., & Jansen, L. J. M. (2000). Land Cover Classification System: LCCS:

Classification Concepts and User Manual (p. 91). Rome: Food and Agriculture
Organization of the United Nations.

137

 

 

Docampo, R., & Moreno, S. N. (2003). Current chemotherapy of human African
trypanosomiasis. Parasitology Research, 90, 10-13.

Douthwaite, R. (1992). Trypanosomiasis and tsetse — Africa’s disease challenge: Non-
target effects of insecticides used in tsetse control operations
. World Animal Review, 70—71.

Dransfield, R. D., & Brightwell, R. (2001). Trap efficiency for Glossina pallidipes
(Diptera: Glossinidae) at Nguruman, south-west Kenya. Bulletin of
Entomological Research, 91(06), 429-444.

Drew, D. R. (1982). A system dynamics model for the integrated control of tsetse flies
and african trypanosomiasis. In Proceedings of the 14th conference on Winter
Simulation- Volume I (pp. 111-125). Winter Simulation Conference.

Du Toit, R. (1954). Trypanosomiasis in Zululand and the control of tsetse flies by
chemical means. Onderstepoort Journal of Veterinary Research, 26(31), 317-
387.

Eklundh, L. (1998). Estimating relations between AVHRR NDVI and rainfall in East
Africa at lO-day and monthly time scales. International Journal of Remote
Sensing, 19(3), 563—570.

Eriksen, C. (2007). Why do they burn the'bush'? Fire, rural livelihoods, and
conservation in Zambia. Geographical Journal, I 73(3), 242-256.

Estrada-Pena, A. (2002). Increasing habitat suitability in the United States for the tick
that transmits Lyme disease: a remote sensing approach. Environmental Health
Perspectives, 110(7), 63 5-640.

Fairlamb, A. H. (2003). Chemotherapy of human African trypanosomiasis: current and
future prospects. Trends in Parasitology, 19(11), 488-494.

F AO, & AGAL. (2005). Livestock Sector Brief. Food and Agriculture Organization of
the United Nations & Livestock Information, Sector Analysis and Policy Branch;

F eldmann, U. (2006). Pan-African Tsetse and Trypanosomosis Eradication Campaign
(PATTEC). Presented at the F AO / IAEA Interregional Training Course: Use of
the Sterile Insect and Realted Techniques for the Integrated Area-wide
Management of Insect Pests, Power Point, Gainesville, FL.

F evre, E. M., Odiit, M., Coleman, P. G., Woolhouse, M. E. J ., & Welbum, S. C.

(2008). Estimating the burden of rhodesiense sleeping sickness during an
outbreak in Serere, eastern Uganda. BMC Public Health, 8(1), 96-107.

138

 

Fevre, E. M., Picozzi, K., Fyfe, J ., Waiswa, C., Odiit, M., Coleman, P. G., et al. (2005).
A burgeoning epidemic of sleeping sickness in Uganda. The Lancet, 366(9487),
745-747.

F évre, E. M., Tilley, A., Picozzi, K., Fyfe, J ., Anderson, I., Magona, J. W., et al. (2006).
Central point sampling from cattle in livestock markets in areas of human
sleeping sickness. Acta Tropica, 97(2), 229-232.

F oody, G. M. (2006). What is the difference between two maps? A remote senser’s
view. Journal of Geographical Systems, 8(2), 119-130.

Foody, G. M. (2007). Map comparison in GIS. Progress in Physical Geography, 31(4),
439-445.

Ford, J. (1971). The Role of T rypanosomiases in Afi'ican Ecology (pp. 1-568). Oxford,
UK: Clarendon Press.

Ford, J ., & Katondo, K. (1977). Maps of tsetse fly (Glossina) distribution in Africa.
Bulletin of Animal Health and Production in Africa, I 5, 187-193.

Ford, L. B. (2007). Civil conflict and sleeping sickness in Afiica in general and Uganda
in particular. Conflict and Health, I (6), 6-16.

Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y., Muchoney, D., Strahler,
A. H., et al. (2002). Global land cover mapping from MODIS: algorithms and
early results. Remote Sensing of Environment, 83(1), 287-302.

Gatebe, C. K., Tyson, P. D., Annegam, H., Piketh, S., & Helas, G. (1999). A seasonal
air transport climatology for Kenya. Journal of Geophysical Research, 104(D12),
14,237-14,244.

Gatebe, C. K., Tyson, P. D., Annegam, H. J., Helas, G., Kinyua, A. M., & Piketh, S. J.
(2001). Characterization and transport of aerosols over equatorial eastern Africa.
Global Biogeochemical Cycles, 15(3), 663-672.

Giblin, J. (1990). Trypanosomiasis Control in African History: An Evaded Issue?”.
Journal of Afi'ican History, 31(1), 59-70.

Gilbert, M., Jenner, J., Pender, D., Rogers, D., Slingenbergh, J ., & Wint, W. (2001). The
Programme Against African Trypanosomiasis Information System (PAATIS). In

World Class Parasites: The African T rypanosomes (Vol. 1, pp. 11-24). Boston,
MA: Kluwer Academic Publishers.

Gilioli, G., & Baumgartner, J. (2007). Adaptive Ecosocial System Sustainability
Enhancement in Sub-Saharan Africa. EcoHealth, 4(4), 428-444.

139

Glasgow, J. P., & Wilson, F. (1953). A Census of the Tsetse-fly Glossina pallidipes
Austen and of its Host Animals. Journal of Animal Ecology, 22(1), 47-56.

Glover, P. E., Le Roux, J. G., & Parker, D. F. (1958). The extermination of Glossina
palpalis on the Kuja-Migori river systems with the use of insecticides (pp. 331-
342). Presented at the Proceedings of the International Scientific Committe for
Trypanosomiasis Research 7th Meeting, Brussels.

Goodin, D. G., Koch, D. E., Owen, R. D., Chu, Y. K., Hutchinson, J. M., & Jonsson,
C. B. (2006). Land cover associated with hantavirus presence in Paraguay.
Global Ecology and Biogeography, 15(5), 519-527.

Gouteux, J. P., Artzrouni, M., & Jarry, M. (2007). A density-dependent model with
reinvasion for estimating tsetse fly populations (Diptera: Glossinidae) through
trapping. Bulletin of Entomological Research, 91 (03), 177-183.

Grant, I. F. (2001). Insecticides for tsetse and trypanosomiasis control: is the
environmental risk acceptable? Trends in Parasitology, 1 7(1), 10-14.

Guy, P. R. (1989). The influence of elephants and fire on a Brachystegia-Julbemardia
woodland in Zimbabwe. Journal of Tropical Ecology, 5(2), 215-226.

Hansen, M. C., DeFries, R. S., Townshend, J. R. G., & Sohlberg, R. (2000). Global
land cover classification at 1 km spatial resolution using a classification tree
approach. International Journal of Remote Sensing, 21(6), 1331-1364.

Hansen, M. C., & Reed, B. (2000). A comparison of the IGBP DISCover and
University of Maryland 1 km global land cover products. International Journal
of Remote Sensing, 21 (6), 1365-1373.

Hargrove, J. W. (1980). The effect of ambient temperature on the flight performance of
the mature male tsetse fly, Glossina morsitans. Physiological Entomology, 5(4),
397-400.

Hargrove, J. W. (1999). Reproductive abnormalities in tsetse flies in Zimbabwe.
Entomologia Experimentalis et Applicata, 92(1), 89-99.

Hargrove, J. W. (2000). A theoretical study of the invasion of cleared areas by tsetse
flies (Diptera: Glossinidae). Bulletin of Entomological Research, 90(3), 201-209.

Hargrove, J. W. (2001). Factors affecting density-independent survival of an island
population of tsetse flies in Zimbabwe. Entomologia Experimentalis et Applicata,
100(2), 151-164.

140

 

 

Hargrove, J. W. (2003). T setse eradication: sufliciency, necessity, and desirability (pp.
1-134). Research Report, University of Edinburgh, UK: Animal Health
Programme, Center for Tropical Veterinary Medicine.

Hargrove, J. W., & Vale, G. A. (1979). Aspects of the feasibility of employing odour-
baited traps for controlling tsetse flies (Diptera: Glossinidae). Bulletin of
Entomological Research, 69, 283-290.

Hargrove, W. W., Hoffman, F. M., & Hessburg, P. F. (2006). Mapcurves: a
quantitative method for comparing categorical maps. Journal of Geographical
Systems, 8(2), 187-208.

Hay, S. I., & Lennon, J. J. (1999). Deriving meteorological variables across Afiica for
the study and control of vector-home disease: a comparison of remote sensing and
spatial interpolation of climate. Tropical Medicine & International Health, 4(1),
58-7 1 .

Hay, S. 1., Packer, M. J ., & Rogers, D. J. (1997). Review article: The impact of remote
sensing on the study and control of invertebrate intermediate hosts and vectors for
disease. International Journal of Remote Sensing, 18(14), 2899-2930.

Hendrickx, G., Biesemans, J ., & Van Camp, N. (2002). T setse presence-absence
prediction model for Glossina austeni and Glossina brevipalpis in KwaZulu
Natal-South Africa. Technical Report, Agriculture and Veterinary Intelligence
and Analysis (Avia-GIS).

Hendrickx, G., de La Rocque, 8., Reid, R., & Wint, W. (2001). Spatial trypanosomosis
management: from data-layers to decision making. Trends in Parasitology, 1 7(1),
35-41.

Hendrickx, G., Napala, A., Dao, B., Batawui, D., De Deken, R., Vermeilen, A., et al.
(2007). A systematic approach to area-wide tsetse distribution and abundance
maps. Bulletin of Entomological Research, 89(03), 231-244.

Hide, G. (1999). History of sleeping sickness in East Africa. Clinical Microbiology
Reviews, 12(1), 112-125.

Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., & Jarvis, A. (2005). Very
high resolution interpolated climate surfaces for global land areas. International
Journal of Climatology, 25(15), 1965-1978.

Hocking, K. S., Lamerton, J. F., & Lewis, E. A. (1963). Tsetse-fly control and
eradication. Bulletin of the World Health Organization, 28, 811-823.

Holmes, P. H. (1997). New approaches to the integrated control of trypanosomosis.
Veterinary Parasitology, 71 (2-3), 121-135.

141

Holmes, P. H., Eisler, M. C., & Geerts, S. (2004). Current chemotherapy of animal
trypanosomiasis. The T rypanosomiases, 431-444.

Hoppe, K. (2003). Lords of the fly: Sleeping sickness control in British East Africa,
1900-1960 (pp. 1-203). Westport, CT: Praeger Publishers.

Hotez, P., Ottesen, E., Fenwick, A., & Molyneux, D. (2006). The neglected tropical
diseases: the ancient afflictions of stigma and poverty and the prospects for their
control and elimination. In Advances in Experimental Medicine and Biology

(Vol. 582, pp. 23-33).

Hotez, P. J ., Molyneux, D. H., Fenwick, A., Kumaresan, J., Sachs, S. E., Sachs, J. D.,
et al. (2007). Control of neglected tropical diseases. The New England Journal of
Medicine, 35 7(10), 1018-1027.

Hursey, B. S. (2001). The programme against African trypanosomiasis: aims,
objectives and achievements. Trends in Parasitology, 1 7(1), 2-3.

Hutchinson, O. C., F évre, E. M., Canington, M., & Welburn, S. C. (2003). Lessons
learned from the emergence of a new Trypanosoma brucei rhodesiense sleeping
sickness focus in Uganda. The Lancet Infectious Diseases, 3(1), 42-45.

Hutchinson, O. C., Picozzi, K., Jones, N. G., Mott, H., Sharma, R., Welburn, S. C., et
al. (2007). Variant Surface Glycoprotein gene repertoires in Trypanosoma brucei
have diverged to become strain-specific. BMC Genomics, 8, 234-244.

ICIPE. (2002). Trapping the tsetse fly: Kenya. Science and Technology, I, 162-173.

Indeje, M., Ward, M. N., Ogallo, L. J ., Davies, G., Dilley, M., & Anyamba, A. (2006).
Predictability of the Normalized Difference Vegetation Index in Kenya and

Potential Applications as an Indicator of Rift Valley Fever Outbreaks in the
Greater Horn of Africa. Journal of Climate, 19(9), 1673-1687.

Innes, J. L., & Koch, B. (1998). Forest biodiversity and its assessment by remote
sensing. Global Ecology and Biogeography Letters, 7(6), 397-419.

Jordan, A. (1986). Trypanosomiasis Control and African Rural Development (pp. 1-
357). New York, NY: Longman Singapore Publishers Ltd.

Kaare, M. T., Picozzi, K., Mlengeya, T., Fevre, E. M., Mellau, L. S., Mtambo, M. M.,
et al. (2007). Sleeping sickness—A re-emerging disease in the Serengeti? Travel
Medicine and Infectious Disease, 5(2), 1 17-124.

Kaaya, G. P., & Munyinyi, D. M. (1995). Biocontrol potential of the entomogenous

fungi Beauveria bassiana and Metarhizium anisopliae for tsetse flies (Glossina
spp.) at developmental sites. Journal of Invertebrate Pathology, 66(3), 237-241.

142

Kabayo, J. P. (2002). Aiming to eliminate tsetse from Africa. TRENDS in
Parasitology, 18(11), 473-475.

Kamuanga, M. (2003). Socio-economic and cultural factors in the research and control
trypanosomiasis. Food and Agriculture Organization of the United Nations:
PAAT Technical and Scientific Series 4 .

Keiser, J ., Burri, C., & Stich, A. (2001). New drugs for the treatment of human Afiican
trypanosomiasis: research and development. Trends in Parasitology, 1 7(1), 42-
49.

Kennedy, P. G. E. (2004). Human African trypanosomiasis of the CNS: Current issues
and challenges. The Journal of Clinical Investigation, 113(4), 496-504.

Kennedy, P. G. E. (2005). Sleeping sickness-human Afiican trypanosomiasis. British
Medical Journal, 5(5), 260-267.

Kenya Meteorological Department. (2008a). Review of the weather in June - July -
August (JJA) 2008 seasons and the outlook for the October-November—December
2008 “Short Rains” season. Republic of Kenya, Ministry of Environment and
Mineral Resources, Kenya Meteorological Department .

Kenya Meteorological Department. (2008b). Weather outlook for the “Long-Rains”
(March-May) season 2008. Republic of Kenya, Ministry of Environment and
Mineral Resources, Kenya Meteorological Department .

Kenya Meteorological Department. (2009). Climate outlook for March-April-May 2009
("Long Rains" season). Republic of Kenya, Ministry of Environment and Mineral
Resources, Kenya Meteorological Department .

KETRI. (2008). Tsetse distribution in Kenya showing tsetse belts and conservation
areas. Kenya Trypanosomiasis Research Institute (KETRI).

Kgori, P. M., Modo, S., & Torr, S. J. (2006). The use of aerial spraying to eliminate
tsetse from the Okavango Delta of Botswana. Acta T ropica, 99(2-3), 184-199.

Kioy, D., & Mattock, N. (2005). Control of sleeping sickness—time to integrate
approaches. The Lancet, 366(9487), 695-696.

Kitron, U., Otieno, L. H., Hungerford, L. L., Odulaja, A., Brigham, W. U., Okello, O.
O., et al. (1996). Spatial analysis of the distribution of tsetse flies in the Lambwe
Valley, Kenya, using Landsat TM satellite imagery and GIS. Journal of Animal
Ecology, 65(3), 371-380.

143

 

Knight, C. G. (1971). The Ecology of African Sleeping Sickness. Annals of the
Association of American Geographers, 61(1), 23-44.

Kristensen, T. K., Malone, J. B., & McCarroll, J. C. (2001). Use of satellite remote
sensing and geographic information systems to model the distribution and
abundance of snail intermediate hosts in Africa: a preliminary model for
Biomphalaria pfeifferi in Ethiopia. Acta T ropica, 79(1), 73-78.

Laird, M. (1977). T setse the future for biological methods in integrated control (pp. 1-
220). Ottawa, Canada: International Development Research Center.

Lambrecht, F. L. (1964). Aspects of Evolution and Ecology of Tsetse Flies and
Trypanosomiasis in Prehistoric African Environment. Journal of African History,
5(1), 1-24.

Lambrecht, F. L. (1980). Palaeoecology of Tsetse flies and sleeping sickness in Africa.
Proceedings of the American Philosophical Society, 124(5), 367-85.

Lamprey, H. F., Glasgow, J. P., Lee-Jones, F., & Weitz, B. (1962). A simultaneous
census of the potential and actual food sources of the tsetse fly Glossina
swynnertoni Austen. The Journal of Animal Ecologi, 31(1), 151-156.

Langley, P. A., & Weidhaas, D. (1986). Trapping as a means of controlling tsetse,
Glossina spp.(Diptera: Glossinidae): the relative merits of killing and of
sterilization. Bulletin of Entomological Research, 76(1), 89-95.

Leak, S. G. A. (1999). T setse Biology and Ecology. New York, NY: CABI Publishing.

Leak, S. G. A., Ejigu, D., & Vreysen, M. J. .. (2008). Collection of baseline
entomological baseline data for tsetse area-wide integrated pest management
programmes. Food and Agriculture Organization of the United Nations.

Legros, D., Ollivier, G., Gastellu-Etchegorry, M., Paquet, C., Burri, C., & Jannin, J.
(2002). Treatment of human African trypanosomiasis - Present situation and
needs for research and development. Lancet Infectious Diseases, 2(7), 437-440.

Levine, R. 8., Peterson, A. T., & Benedict, M. Q. (2004). Geographic and ecologic
distributions of the Anopheles gambiae complex predicted using a genetic
algorithm. The American Journal of Tropical Medicine and Hygiene, 70(2), 105-
109.

Lillesand, T., Kiefer, R., & Chipman, J. (2004). Remote Sensing and Image
Interpretation (5th ed., p. 763). Hoboken, NJ: John Wiley & Sons, Inc.

Lindquist, D. A. (1984). Atoms for pest control. IAEA Bulletin, 26(2), 22-25.

144

Loke, H., & Randolph, S. E. (1995). Reciprocal regulation of fat content and flight
activity in male tsetse flies (Glossina palpalis). Physiological Entomology, 20(3),
243-247.

Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. 0., Zhu, Z., Yang, L., et al.
(2000). Development of a global land cover characteristics database and IGBP
DISCover from 1 km AVHRR data. International Journal of Remote Sensing,

21(6), 1303-1330.

Mack, R. (1970). The great African cattle plague epidemic of the 18903. Tropical
Animal Health and Production, 2(4), 210-219.

MacLean, L., Chisi, J. E., Odiit, M., Gibson, W. C., Ferris, V., Picozzi, K., et al. (2004).
Severity of Human African Trypanosomiasis in East Africa Is Associated with
Geographic Location, Parasite Genotype, and Host Inflammatory Cytokine
Response Profile . Infection and Immunity, 72(12), 7040-7044.

MacLean, L., Odiit, M., MacLeod, A., Morrison, L., Sweeney, L., Cooper, A., et al.
(2007). Spatially and Genetically Distinct African Trypanosome Virulence
Variants Defined by Host Interferon-y Response. The Journal of Infectious
Diseases, 196, 1620-1628.

Magona, J. W., Walubengo, J ., Odiit, M., Okedi, L. A., Abila, P., Katabazi, B. K., et al.
(2005). Implications of the re-invasion of Southeast Uganda by Glossina
pallidipes on the epidemiology of bovine trypanosomosis. Veterinary
Parasitology, 128(1-2), 1-9.

Maitima, J. (2008). The Framework and guidelines for assessing environmental and
socioeconomic impacts of tsetse fly interventions. International Livestock

Research Institute.

Maniania, N. K. (1998). A Device for Infecting Adult Tsetse Flies, Glossina spp., with
an Entomopathogenic Fungus in the Field. Biological Control, 11(3), 248-254.

Maniania, N. K., & Takasu, K. (2006). Development of microbial control agents at the
international centre of insect physiology and ecology. Bulletin of the Institute of
Tropical Agriculture, Kyushu University, 29(1), 1-9.

Martiny, N., Camberlin, P., Richard, Y., & Philippon, N. (2006). Compared regimes of
NDVI and rainfall in semi-arid regions of Africa. International Journal of

Remote Sensing, 27(23), 5201—5223.

Matthiessen, P., & Douthwaite, B. (1985). The impact of tsetse fly control campaigns on
African wildlife. Oryx, 19(4), 202-209.

145

Mattioli, R. C., Feldmann, U., Hendrickx, G., Wint, W., Jannin, J ., & Slingenbergh, J.
(2004). Tsetse and trypanosomiasis intervention policies supporting sustainable
animal-agricultural development. Journal of Food, Agriculture & Environment,

2(2), 310-314.

Matzke, G. (1979). Settlement and sleeping sickness control--a dual threshold model of

colonial and traditional methods in East Africa. Social Science & Medicine.
Medical Geography, 13(4), 209-14.

Mayaux, P., Eva, H., Gallego, J ., Strahler, A. H., Herold, M., Agrawal, S., et al. (2006).
Validation of the global land cover 2000 map. IEEE Transactions on Geoscience
and Remote Sensing, 44(7 Part 1), 1728-1739.

McDermott, J. J ., & Coleman, P. G. (2001). Comparing apples and oranges—model-
based assessment of different tsetse-transmitted trypanosomosis control strategies.
International Journal for Parasitology, 31(5-6), 603-609.

McKelvey, J. J. (1973). Man against tsetse: Struggle for Africa. Ithaca and London:
Cornell University Press.

Mdachi, R. E., Thuita, J. K., Kagira, J. M., Ngotho, J. M., Murilla, G. A., Ndung'u, J.
M., et al. (2009). Efficacy of the Novel Diamidine Compound 2, 5-Bis (4-
Amidinophenyl)-Furan-Bis-O-Methlylamidoxime (Pafuramidine, DB289) against
Trypanosoma brucei rhodesiense Infection in Vervet Monkeys after Oral
Administration? Antimicrobial Agents and Chemotherapy, 53(3), 953-957.

Mellanby, K. (1936). Experimental work with the tsetse-fly, Glossina palpalis, in
Uganda. Bulletin of Entomological Research, 27, 611-632.

Mellanby, K. (1939). Low temperature and insect activity. Proceedings of the Royal
Society of London. Series B, Biological Sciences (1934-1990), 12 7(849), 473-
487.

Minchin, E. A. (1905). Report on the Anatomy of the Tsetse-Fly (Glossina Palpalis).
Proceedings of the Royal Society of London. Series B; Containing Papers of a
Biological Character, 7 6(5 12), 531-547.

Mugisha, A., McLeod, A., Percy, R., & Kyewalabye, E. (2005). Strategies,
Effectiveness and Rationale of Vector-borne Disease Control in the Pastoralist

System of South-westem Uganda. Tropical Animal Health and Production,
3 7(6), 479-489.

Muller, G., Grébaut, P., & Gouteux, J. P. (2004). An agent-based model of sleeping

sickness: simulation trials of a forest focus in southern Cameroon. Comptes
rendus-Biologies, 327(1), 1-1 1.

146

Muriuki, G. W., Chemuliti, J., Changasi, R., Maichomo, M., & Ndung’u, J. (2005). The
Impact of Changing Landscapes on Tsetse Distribution in the ASALS North of
Mt. Kenya. The International Scientific Council for Trypanosomiasis Research
and Control.

Muriuki, G. W., Njoka, T. J ., Reid, R. S., & Nyariki, D. M. (2005). Tsetse control and

land-use change in Lambwe valley, south-westem Kenya. Agriculture,
Ecosystems and Environment, 106(1), 99-107.

Musere, J. (1990). African sleeping sickness (pp. 1-208). Lewiston, NY: The Edwin
Mellen Press.

Muzari, M. O. (1999). Odour-baited targets as invasion barriers for tsetse flies (Diptera:
Glossinidae): a field trial in Zimbabwe. Bulletin of Entomological Research,
89(1), 73-77.

Muzari, M. O., & Hargrove, J. W. (2005). Artificial larviposition sites for field
collections of the puparia of tsetse flies Glossina pallidipes and G. m. morsitans
(Diptera: Glossinidae). Bulletin of Entomological Research, 95(3), 221-229.

Nash, T. A. M. (1933). A statistical analysis of the climatic factors influencing the
density of tsetse flies, Glossina morsitans Westw. Journal of Animal Ecology,

2(2), 197-203.

Nash, T. A. M. (1937). Climate, the vital factor in the ecology of Glossina. Bulletin of
Entomological Research, 28, 75-127.

Nash, T. A. M. (1948). Tsetse Flies in British West Africa (p. 77). Colonial Office by
HM Stationery Office.

Ndegwa, P. N., & Mihok, S. (1999). Development of odour-baited traps for Glossina
swynnertoni (Diptera: Glossinidae). Bulletin of Entomological Research, 89(3),
255-261.

Nelson, R. H. (2003). Environmental Colonialismz" Saving" Africa from Afiicans. The
Independent Review, 8(1), 65-86.

Odulaja, A., Baumgiirtner, J., Mihok, S., & Abu-Zinid, I. M. (2001). Spatial and
temporal distribution of tsetse fly trap catches at Nguruman, southwest Kenya.
Bulletin of entomological research, 91(3), 213-220.

Ofcansky, T. P. (19.81). The 1889-97 rinderpest epidemic and the rise of British and

German colonialism in eastern and southern Afiica. Journal of Afiican Studies,
8(1), 31-38.

147

 

Offori, E. D. (1982). The scourge of the tsetse. International Atomic Energy Agency
Bulletin, 23(3), 43-46.

Olett, P. (2003). Historical review of successive tsetse control activities and policies in
Western Kenya, in reference to impact on natural resources. Farming in Tsetse
Controlled Areas. The International Livestock Research Institute (ILRI), Nairobi,
Kenya.

0100, F. (2006). Literature survey on unpublished records on environmental and socio-
economic impacts assessment on tsetse and trypanosomiasis interventions in
Kenya. The International Livestock Research Institute (ILRI), Nairobi, Kenya.

Oluwafemi, R. A. (2009). The impact of African animal trypanosomosis and Tsetse fly
on the livelihood and well-being of cattle and their owners in the BICOT study
area of Nigeria. The Internet Journal of Veterinary Medicine, 5(2), 4.

Omumbo, J. A., Hay, S. 1., Snow, R. W., Tatem, A. J., & Rogers, D. J. (2005).
Modelling malaria risk in East Africa at high-spatial resolution. Tropical
Medicine & International Health, 10(6), 557-566.

Ouma, J. O., Marquez, J. G., & Krafsur, E. S. (2006). Microgeographical breeding
structure of the tsetse fly, Glossina pallidipes in south-western Kenya. Medical
and Veterinary Entomology, 20(1), 138-149.

PAAT. (2008). Programme Against African Trypanosomiasis tsetse and
trypanosomiasis information 2008. Food and Agriculture Organization of the
United Nations.

PATTEC. (2001). Pan African tsetse and trypanosomosis eradication campaign
(PATTEC) action plan: Enhancing Africa's prosperity. Organization of Afiican
Unity.

Pavlovskii, E. N. (1966). Natural nidality of transmissible diseases, with special
reference to the landscape epidemiology of zooanthroponoses. (p. 261). Urbana,
111: University of Illinois Press.

Pepin, J ., & Milord, F. (1991). African trypanosomiasis and drug-induced
encephalopathy: Risk factors and pathogenesis. Transactions of the Royal Society
of Tropical Medicine and Hygiene, 85(2), 222-224.

Peter, R. J ., Van den Bossche, P., Penzhom, B. L., & Sharp, B. (2005). Tick, fly, and

mosquito control—lessons from the past, solutions for the future. Veterinary
parasitology, 132(3-4), 205-215.

148

 

Picozzi, K., Fevre, E. M., Odiit, M., Canington, M., Eisler, M. C., Maudlin, 1., et al.
(2005). Sleeping sickness in Uganda: a thin line between two fatal diseases.
British Medical Journal, 33 1 (7527), 1238-1241.

Picozzi, K., Tilley, A., Fevre, E. M., Coleman, P. G., Magona, J. W., Odiit, M., et a1.
(2002). The diagnosis of trypanosome infections: applications of novel

technology for reducing disease risk. Afi'ican Journal of Biotechnology, 1(2), 39-
45.

Pigozzi, B. W. (2004). A hierarchy of spatial marginality through spatial filtering. The
Professional Geographer, 56(4), 460-470.

Pilson, R. D., & Pilson, B. M. (1967). Behaviour studies of Glossina morsitans Westw.
in the field. Bulletin of Entomological Research, 57(2), 227-257.

Poinar, G. O., Wassink, H. J ., Leegwater-van der Linden, M., & van der Geest, L. P.
(1979). Serratia marcescens as a pathogen of tsetse flies. Acta T ropica, 36(3),
223-227.

Pollock, J. N. (1982a). Training Manual for T setse Control Personnel. Vol. 1: T setse
biology, systematics and distribution, techniques. Food and agricultural
Organization of the United Nations, Rome.

Pollock, J. N. (1982b). Training Manual for T setse Control Personnel. Vol. 2:
Ecology and Behaviour of T setse (pp. 1-105). Food and agricultural
Organization of the United Nations, Rome.

Pontius, R. G., & Cheuk, M. L. (2006). A generalized cross-tabulation matrix to
compare soft-classified maps at multiple resolutions. International Journal of
Geographical Information Science, 20(1), 1-30.

Randolph, S. E., & Storey, K. (1999). Impact of microclimate on immature tick-rodent
host interactions (Acari: Ixodidae): implications for parasite transmission.
Journal of Medical Entomology, 36(6), 741-748.

Randolph, S., Rogers, D., & Kiilu, J. (1991). The feeding behaviour, activity and
trappability of wild female Glossina pallidipe in relation to their pregnancy cycle.
Medical and Veterinary Entomology, 5(3), 335-350.

Ratajczak, A. (2005). US F raunhofer Center receives Grant from the Bill & Melinda
Gates Foundation. Fraunhofer-Gesellschaft.

Reid, R. S., Kruska, R. L., Deichmann, U., Thornton, P. K., & Leak, S. G. A. (2000).

Human population growth and the extinction of the tsetse fly. Agriculture,
Ecosystems and Environment, 77(3), 227-236.

149

Reid, R. S., Wilson, C. J ., Kruska, R. L., & Mulatu, W. (1997). Impacts of tsetse
control and land-use on vegetative structure and tree species composition in
south-western Ethiopia. Journal of Applied Ecology, 34(3), 731-747.

Robinson, T., Rogers, D., & Williams, B. (1997). Mapping tsetse habitat suitability in
the common fly belt of southern Afiica using multivariate analysis of climate and

remotely sensed vegetation data. Medical and Veterinary Entomology, I I (3),
23 5-245.

Rogers, D. (1977). Study of a natural population of Glossina fuscipes fuscipes Newstead
and a model of fly movement. Journal of Animal Ecology, 46(1), 309-330.

Rogers, D. (1979). Tsetse Population Dynamics and Distribution: A New Analytical
Approach. The Journal of Animal Ecology, 48(3), 825-849.

Rogers, D. J ., & Packer, M. J. (1993). Vector-bome diseases, models, and global
change. Lancet, 342(8882), 1282-1284.

Rogers, D. J ., & Randolph, S. E. (1985). Population Ecology of Tsetse. Annual
Reviews in Entomology, 30(1), 197-216.

Rogers, D. J., & Randolph, S. E. (1986). Distribution and Abundance of Tsetse Flies
(Glossina Spp.). Journal of Animal Ecology, 55(3), 1007-1025.

Rogers, D. J., & Randolph, S. E. (1988). Tsetse flies in Afiica: bane or boon?
Conservation biology, 2(1), 57-65.

Rogers, D. J ., & Randolph, S. E. (1991). Mortality rates and population density of
tsetse flies correlated with satellite imagery. Nature, 351 (6329), 739-741.

Rogers, D. J ., & Randolph, S. E. (2002). A response to the aim of eradicating tsetse
fi'om Africa. Trends in Parasitology, 18(12), 534-536.

Rogers, D. J ., & Williams, B. G. (1994). Tsetse distribution in Africa: seeing the wood
and the trees (pp. 247-271). Oxford, UK: Blackwell Scientific Publications.

Sachs, J. l D., & Hotez, P. J. (2006). Fighting tropical diseases. Science, 31 1 (5767),
1521.

Sandholt, I., Rasmussen, K., & Andersen, J. (2002). A simple interpretation of the
surface temperature/vegetation index space for assessment of surface moisture

status. Remote Sensing of Environment, 79(2), 213—224.

Scepan, J. (1999). Thematic validation of high-resolution global land-cover data sets.
Photogrammetric Engineering and Remote Sensing, 65(9), 1051-1060.

150

 

Schofield, C. J., & Maudlin, I. (2001). Trypanosomiasis control. International Journal
for Parasitology, 31(5-6), 615-620.

Schofield, S., & Torr, S. J. (2002). A comparison of the feeding behaviour of tsetse and
stable flies. Medical and Veterinary Entomology, 16(2), 177-185.

Shanks, G. D., Biomndo, K., Hay, S. I., & Snow, R. W. (2000). Changing patterns of
clinical malaria since 1965 among a tea estate population located in the Kenyan
highlands. Transactions of the Royal Society of Tropical Medicine and Hygiene,
94(3), 253—255.

Shaw, A., Hendrickx, G., Gilbert, M., Mattioli, R., & Wint, W. (2003). Mapping the
benefits: first steps in developing a new decision tool for tsetse and
trypanosomiasis interventions. In 27 th Meeting of the International Scientific
Council for T rypanosomiasis Research and Control (ISC T RC), South Afi'ica.

Shaw, A., Hendrickx, G., & Wint, W. (2003). Mapping the Benefits: First steps in
developing a new decision tool for tsetse and trypanosomiasis interventions (pp.
1-9). International Scientific Council for Trypanosomiasis Research and Control.

Smith, A. B. (1984). Environmental limitations on prehistoric pastoralism in Africa.
Afiican Archaeological Review, 2(1), 99-1 11.

Sombroek, W. 6., Braun, H. M. H., & Pouw, B. J. A. V. D. (1982). Exploratory
Soil Map and Agro-climatic Zone Map of Kenya, 1980 (pp. 1-60). Nairobi,
Kenya: Ministry of Agriculture; National Agriculture Laboratories.

Stock, R. F. (2004). Africa South of the Sahara (Second Edition, p. 477). New York,
NY: Guilford Press.

Strahler, A., Friedl, M. A., Schaaf, C. L. B., Zhang, X., & Hodges, J. C. F. (2003).
Validation of the consistent-year v003 MODIS land cover product (p. 13).
Center for Remote Sensing, Department of Geography, Boston University.

Sutherst, R. W. (2004). Global Change and Human Vulnerability to Vector-Bome
Diseases. Clinical Microbiology Reviews, 1 7(1), 136-173.

Swynnerton, C. F. M. (1921). An Examination of the Tsetse Problem in North
Mossurise Portuguese East Afiica. The Imperial Bureau of Entomology.

Symeonakis, E., Robinson, T., & Drake, N. (2007). G18 and multiple-criteria evaluation
for the optimisation of tsetse fly eradication programmes. Environmental
Monitoring and Assessment, 124(1), 89-103.

Taveme, J. (2001). PAAT and PATTEC: differences and synergy. Trends in
Parasitology, 1 7(7), 310-31 1.

151

 

Terblanche, J. S., & Chown, S. L. (2007). The effects of temperature, body mass and
feeding on metabolic rate in the tsetse fly Glossina morsitans centralis.
Physiological Entomology, 32(2), 175-180.

Terblanche, J. S., Clusella-Trullas, S., Deere, J. A., & Chown, S. L. (2008). Thermal
tolerance in a south-east African population of the tsetse fly Glossina pallidipes
(Diptera, Glossinidae): Implications for forecasting climate change impacts.
Journal of Insect Physiology, 54(1), 1 14-127.

Tian, Y., Zhang, Y., Knyazikhin, Y., Myneni, R. B., Glassy, J. M., Dedieu, G., et al.
(2000). Prototyping of MODIS LAI and FPAR algorithm with LASUR and
LAN DSATdata. IEEE transactions on geoscience and remote sensing, 38(5 Part
2), 2387-2401.

Tidwell, R. (2006). UNC-led consortium for parasitic drug development. DNDi
Newsletter, 1.

Torbick, N., Lusch, D., Qi, J ., Moore, N., Olson, J ., & Ge, J. (2006). Developing land
use/land cover parameterization for climateland modelling in East Africa.
International Journal of Remote Sensing, 27( 19), 4227-4244.

Torbick, N., Qi, J ., Ge, J ., Olsen, J ., & Lusch, D. (2005). An assessment of Africover
and GLC2000 using general agreement and videography techniques. In

Geoscience and Remote Sensing Symposium, 2005. IGARSS'05. Proceedings.
2005 IEEE International (Vol. 7, pp. 5005-5008).

Torr, S. J ., & Hargrove, J. W. (1999). Behaviour of tsetse (Diptera: Glossinidae)
during the hot season in Zimbabwe: the interaction of micro-climate and
reproductive status. Bulletin of Entomological Research, 89(4), 365-379.

Torr, S. J ., Hargrove, J. W., & Vale, G. A. (2005). Towards a rational policy for
dealing with tsetse. Trends in Parasitology, 21 (1 1), 537-541.

Torr, S. J ., Maudlin, I., & Vale, G. A. (2007). Less is more: restricted application of
insecticide to cattle to improve the cost and efficacy of tsetse control. Medical
and Veterinary Entomology, 21, 53-64.

Turner, D. A., & Brightwell, R. (1986). An evaluation of a sequential aerial spraying
operation against Glossina pallidipes Austen(Diptera: Glossinidae) in the Lambwe
Valley of Kenya: aspects of post-spray recovery and evidence of natural
population regulation. Bulletin of entomological research, 76(2), 331-349.

Uilenberg, G. (1998). A field guide for the diagnosis, treatment and prevention of

African animal trypanosomosis. Food and agricultural Organization of the United
Nations, Rome.

152

 

UN Wire. (2002, November 13). IAEA Pursues Controversial Plan to Kill Tsetse Fly.
United Nations Foundation and National Journal Group Inc., Washington D. C.,
USA.

Vale, G. A. (1993). Development of baits for tsetse flies (Diptera: Glossinidae) in
Zimbabwe. Journal of Medical Entomology, 30(5), 831-842.

Vale, G. A., & Hargrove, J. W. (1979). A method of studying the efficiency of traps
for tsetse flies (Diptera: Glossinidae) and other insects. Bulletin of Entomological
Research, 69, 183-193.

Vale, G. A., Hursey, B. S., Hargrove, J. W., Torr, S. J., & Allsopp, R. (1984). The use
of small plots to study populations of tsetse(Diptera: Glossinidae): Difficulties
associated with population dispersal. Insect Science and Its Application, 5(5),
403-410.

Vale, G. A., Lovemore, D. F ., Flint, S., & Cockbill, G. F. (1988). Odour-baited targets
to control tsetse flies, Glossina spp.(Diptera: Glossinidae), in Zimbabwe. Bulletin
of Entomological Research, 78(1), 31-49.

Vale, G. A., & Torr, S. J. (2005). User-friendly models of the costs and efficacy of
tsetse control: application to sterilizing and insecticidal techniques. Medical and
Veterinary Entomology, 19(3), 293-305.

Van den Bossche, P., De Deken, R., Brandt, J ., Seibou, B., & Geerts, S. (2004).
Recirculation of Trypanosoma brucei brucei in cattle after T. congolense
challenge by tsetse flies. Veterinary Parasitology, 121(1-2), 79-85.

Van den Bossche, P., Doran, M., & Connor, R. J. (2000). An analysis of trypanocidal
drug use in the Eastern Province of Zambia. Acta T ropica, 75(2), 247-258.

Van den Bossche, P., Ky-Zerbo, A., Brandt, J ., Marcotty, T., Geerts, S., & De Deken, R.
(2005). Transmissibility of Trypanosoma brucei during its development in cattle.
Tropical Medicine & International Health, 10(9), 833-839.

Van der Vloedt, A. (1991). The Integration of Biological Techniques into Tsetse
Control Programmes. Programme for the Control of Afiican Animal
Trypanosomiasis and Related Development: Ecological and technical aspects-
Proceedings of the FAO Panel of Experts, Harare, Zimbabwe .

Vanderplank, F. L. (1948). Studies of the behaviour of the tsetse-fly (Glossina

pallidipes) in the field: the attractiveness of various baits. Journal of Animal
Ecology, 13(1), 39-48.

153

Vreysen, M. J. B. (2001). Principles of area-wide integrated tsetse fly control using the
sterile insect technique. Medecine T ropicale, 61 (4-5), 397-410.

Vreysen, M. J. B., Mebrate, A., Menjeta, M., Bancha, B., Woldeyes, C., Music, K., et
al. (1999). The distribution and relative abundance of tsetse flies in the Southern
Rifi Valley of Ethiopia: preliminary survey results. In 25 th meeting of the

International Scientific Council for T rypanosomiasis Research and Control (pp.
202-213). Mombasa, Kenya.

Vreysen, M. J. B., Saleh, K. M., Ali, M. Y., Abdulla, A. M., Zhu, Z. R., Juma, K. G.,
et al. (2000). Glossina austeni (Diptera: Glossinidae) eradicated on the island of

Unguja, Zanzibar, using the sterile insect technique. Journal of Economic
Entomology, 93(1), 123-135.

Waiswa, C., Picozzi, K., Katunguka-Rwakishaya, E., Olaho-Mukani, W., Musoke, R. A.,
& Welburn, S. C. (2006). Glossina fuscipes fuscipes in the trypanosomiasis
endemic areas of south eastern Uganda: Apparent density, trypanosome infection
rates and host feeding preferences. Acta T ropica, 99(1), 23-29.

Waller, R. D. (1990). Tsetse Fly in Western Narok, Kenya. Journal of African History,
31(1), 81-101.

Weekly Epidemiological Record. (2006). Human African trypanosomiasis (sleeping
sickness): epidemiological update. Weekly Epidemiological Record, 81(8), 69-80.

Weitz, B., & Glasgow, J. P. (1956). The natural hosts of some species of Glossina in

East Africa. Transactions of the Royal Society of Tropical Medicine and Hygiene,
50, 593—612.

Welburn, S. C., Coleman, P. G., Maudlin, 1., Fevre, E. M., Odiit, M., & Eisler, M. C.
(2006). Crisis, what crisis? Control of Rhodesian sleeping sickness. Trends in
parasitology, 22(3), 123-128.

White, L. (1995). Tsetse visions: narratives of blood and bugs in colonial Northern
Rhodesia, 1931-9. Journal of Afi'ican History, 36(2), 219-245.

WHO. (2006). African trypanosomiasis (sleeping sickness) fact sheet. World Health
Organization.

WHO. (2009a). Vector Control. World Health Organization (WHO) Human African
Trypanosomiasis. Retrieved April 5, 2009, .

WHO. (2009b). Neglected Tropical Diseases Frequently Asked Questions & Fact File.
World Health Organization.

154

Williams, B., Brightwell, R., & Dransfield, R. (1990). Monitoring tsetse fly
populations: II. The effect of climate on trap catches of Glossina pallidipes.
Medical and Veterinary Entomology, 4(2), 181-193.

Williams, B., Dransfield, R., & Brightwell, R. (1990). Monitoring tsetse fly populations.
1. The intrinsic variability of trap catches of Glossina pallidipes at Nguruman,
Kenya. Medical and Veterinary Entomology, 4(2), 167-179.

Williams, B., Dransfield, R., & Brightwell, R. (1992). The control of tsetse flies in
relation to fly movement and trapping efficiency. Journal of Applied Ecology,
29(1), 163-179.

Williams, B., Rogers, D., Staton, G., Ripley, B., & Booth, T. (1992). Statistical
modeling of georeferenced data: mapping tsetse distributions in Zimbabwe using
climate and vegetation data. In Modeling Vector-Borne and Other Parasitic
Diseases (pp. 267-280). Nairobi, Kenya: The International Laboratory for
Research on Animal Diseases.

Williams, C. L., Hargrove, W. W., Liebman, M., & James, D. E. (2008). Agro-
ecoregionalization of Iowa using multivariate geographical clustering.
Agriculture, Ecosystems and Environment, 123(1-3), 161-174.

Wint, W. (2001). Kilometre resolution tsetse fly distribution maps for Lake Victoria
Basin and West Africa (pp. 1-15). Consultants' Report, Vienna, Austria: Food
and Agriculture Organisation of the United Nations; IAEA Joint Division,
International Atomic Energy Agency.

Wint, W., & Rogers, D. (2000). Predicted Distribution of T setse in Afiica (pp. 1-62).
Consultants' Report, Rome: Food and agricultural Organization of the United
Nations.

Yamey, G. (2002). The world's most neglected diseases: Ignored by the pharmaceutical
industry and by public-private partnerships. BM], 325, 176-177.

Zhan, X., De Fries, R. S., Hansen, M. C., Townshend, J ., Di Miceli, C. M., Sohlberg,
R., et al. (1999). MODIS enhanced [and cover and land cover change product,
algorithm theoretical basis documents (A T BD), version 2.0 (p. 93). University of
Maryland, College Park, MD: Department of Geography and Institute for
Advanced Computing Studies.

155