Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Natural and anthropogenic influences on elephants and other ungulates in the Congo forest Beyers, Rene 2008

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-UBC_2008_fall_beyers_rene.pdf [ 4.26MB ]
Metadata
JSON: 24-1.0058315.json
JSON-LD: 24-1.0058315-ld.json
RDF/XML (Pretty): 24-1.0058315-rdf.xml
RDF/JSON: 24-1.0058315-rdf.json
Turtle: 24-1.0058315-turtle.txt
N-Triples: 24-1.0058315-rdf-ntriples.txt
Original Record: 24-1.0058315-source.json
Full Text
24-1.0058315-fulltext.txt
Citation
24-1.0058315.ris

Full Text

NATURAL AND ANTHROPOGENIC INFLUENCES ON ELEPHANTS AND OTHER UNGULATES IN THE CONGO FOREST by  RENE BEYERS  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE STUDIES (ZOOLOGY)  THE UNIVERSITY OF BRITISH COLUMBIA (VANCOUVER)  JUNE 2008  © RENE BEYERS, 2008  ABSTRACT In Central Africa, wildlife populations are increasingly influenced by humans, even in protected areas. This raises the question how spatial patterns of wildlife abundance are affected by human activities and habitat and how these patterns change over time. I address these questions by developing spatial models combined with line transect survey data in two forest sites in Central Africa. In the Odzala National Park in the Republic of Congo, I examine elephant dung abundance data in relation to human threats and protection. In the Okapi Faunal Reserve in the Democratic Republic of Congo (DRC), I developed spatio-temporal models for elephants and other forest ungulates to examine temporal changes in their densities as a result of changes in human impact in the context of a civil war that took place in the region between the two surveys. Covariates related to human influences dominated the observed patterns at both sites. In Odzala, elephant dung densities were mainly influenced by protection. They were higher inside the protected area and closer to anti-poaching patrol routes. In the Okapi Faunal Reserve, populations of all ungulate species declined severely between the two survey periods. Declines in elephant abundance were relatively higher closer to the park boundary and areas of intense human activity. After the war, elephant densities were higher in a small area in the centre of the park that may have acted as a refuge. Forest duikers also declined, but the spatial pattern of their decline was different than that of elephants. Densities dropped more in the southern part of the reserve, probably due to pre-exisisting higher levels of hunting there. Besides explaining spatial patterns of abundance, spatial modeling was shown to be useful in improving the precision of density estimates and in predicting densities across a surface in the Odzala National Park. In summary, humans overwhelmingly determined the distribution and abundance of ungulates in both sites. The civil war in DRC led to a dramatic increase in elephant poaching for ivory which caused a major decline in elephant populations. It aggravated the bushmeat hunting of duikers whose populations also declined sharply.  - ii -  TABLE OF CONTENTS ABSTRACT .................................................................................................................ii TABLE OF CONTENTS .............................................................................................iii LIST OF TABLES .....................................................................................................viii LIST OF FIGURES ......................................................................................................x ACKNOWLEDGEMENTS .........................................................................................xii CO-AUTHORSHIP STATEMENT .............................................................................xiv CHAPTER 1: Introduction .........................................................................................1 1.1. General Introduction ........................................................................................1 1.2. References.......................................................................................................6 CHAPTER 2: Spatial modeling of elephant abundance in and around the Odzala National Park in the Republic of Congo ..................................................................9 2.1. Introduction ......................................................................................................9 2.2. Methods .........................................................................................................11 2.2.1. Study site ................................................................................................11 2.2.2. Study design and data collection ............................................................12 2.2.3. Spatial environmental data .....................................................................13 2.2.4. Law enforcement monitoring data ..........................................................14 2.2.5. Data analysis ..........................................................................................17 2.2.5.1. Distance analysis .............................................................................17 2.2.5.2. Spatial modeling ..............................................................................18 - iii -  2.2.5.3. Generalized Linear Models ..............................................................19 2.2.5.4. Multicollinearity ................................................................................20 2.2.5.5. Model simplification .........................................................................21 2.2.5.6. Spatial modeling at different spatial scales ......................................21 2.2.5.7. Generalized Additive Models ...........................................................22 2.2.5.8. Spatial autocorrelation .....................................................................24 2.2.5.9. Precision and variance estimation ...................................................25 2.2.5.10. Predicting density over the area ....................................................25 2.3. Results ...........................................................................................................26 3.3.1. Law enforcement monitoring ..................................................................26 3.3.2. Elephant dung densities .........................................................................27 2.3.2.1. Distance analysis .............................................................................27 2.3.2.2. Spatial modeling ..............................................................................28 2.3.2.3. Spatial autocorrelation .....................................................................30 2.3.2.4. CV of encounter rates ......................................................................31 2.3.2.5. Prediction of densities over a surface ..............................................31 2.4. Discussion......................................................................................................31 4.4.1. Environmental and human influence on elephants .................................31 4.4.2. Scale and spatial autocorrelation............................................................35 4.4.3. Coefficient of variation and implications for monitoring ..........................35 4.4.4. Predicted elephant densities...................................................................37 2.5. Tables ............................................................................................................39 2.6. Figures ...........................................................................................................45 2.7. References.....................................................................................................55 CHAPTER 3: Spatio-temporal modeling of wildlife abundance in the context of a war in the Okapi Faunal Reserve, DRC .................................................................63 3.1. Introduction ....................................................................................................63 3.2. Methods .........................................................................................................66 - iv -  2.2.1. Study area ..............................................................................................66 2.2.2. Mammal surveys.....................................................................................67 2.2.3. Spatial covariates ...................................................................................69 3.2.3.1. Human-related covariates ................................................................69 3.2.3.2. Habitat and slope .............................................................................71 2.2.4. Data analysis ..........................................................................................72 3.2.4.1. Estimates of densities and changes in densities .............................72 3.2.4.2. Proportion of change in densities ....................................................73 3.2.4.3. Spatial modeling ..............................................................................74 3.3. Results ...........................................................................................................76 3.3.1. Declines in dung densities ......................................................................76 3.3.2. Proportional declines in densities ...........................................................77 3.3.3. Spatial and spatio-temporal modeling of human influences and habitat.77 3.3.3.1. Elephants .........................................................................................77 3.3.3.2. Okapi ...............................................................................................78 3.3.3.3. Duikers .............................................................................................78 3.4. Discussion......................................................................................................79 4.4.1. Elephants ................................................................................................80 3.4.1.1. Elephant population decline as a result of the war ..........................80 3.4.1.2. Area of refuge ..................................................................................82 4.4.2. Okapi ......................................................................................................83 4.4.3. Duikers....................................................................................................83 3.4.3.1. Bushmeat hunting ............................................................................84 3.4.3.2. Mining ..............................................................................................85 3.5. Conclusions ...................................................................................................85 3.6. Tables ............................................................................................................87 3.7. Figures ...........................................................................................................96 3.8. References...................................................................................................100  -v-  CHAPTER 4: General Conclusions ......................................................................108 4.1. Spatial Modeling ..........................................................................................108 4.2. Conservation ecology: impact of humans on wildlife ...................................109 2.2.1. Human influence on elephants .............................................................109 2.2.2. Duikers and bushmeat hunting .............................................................110 2.2.3. Impact of wildlife depletions on ecosystems .........................................112 4.3. Conservation ecology: recommendations ....................................................113 3.3.1. Protect large roadless areas with a high area/edge ratio .....................113 3.3.2. Invest in law enforcement and patrols ..................................................114 3.3.3. Ensure continued investment in park staff and protection of "safe zones" during periods of political turmoil ...............................................115 3.3.4. Monitor wildlife populations and illegal activities using appropriate scientific methods ...........................................................................................116 3.3.5. Clamp down on illegal ivory trade and sales ........................................117 3.3.6. Increase global financial support ..........................................................117 4.4. References...................................................................................................119 APPENDICES .........................................................................................................123 Appendix 2.1. Correlation matrix for selected variables at the sampling site level for the entire dataset (a) and the park dataset only (b) in the Odzala National Park ....124 Appendix 2.2. Correlation matrix for selected variables at the segment level for the entire dataset (a) and the park dataset only (b) in the Odzala National Park ....127 Appendix 2.3. Comparison of the empirical Cumulative Distribution Function (CFD) of elephant dung encounter rates (dung piles/km) with the normal and poisson Cumulative Distribution Functions ...........................................................................130 Appendix 2.4. GAM plots: Results of the Generalized Additive Model fits to elephant dung encounter rates within the entire study area (model MS3) (a) and the National Park only (model MP4) (b) in the Odzala National Park ..........................................131  - vi -  Appendix 3.1. Selected DISTANCE analysis models (models for okapi and duikers used pooled data across survey periods) ................................................................134 Appendix 3.2. Amount of forest loss per quarter degree grid cell between 1990 and 2000 in and around the Okapi Faunal Reserve .......................................................135 Appendix 3.3. Ecozones (habitat types) in the Okapi Faunal Reserve ...................136 Appendix 3.4. Slope in degrees in the Okapi Faunal Reserve (from SRTM data) ..137 Appendix 3.5. Predicted density maps (animals per km2) for each ungulate species in 1995 and in 2006 in the Okapi Faunal Reserve using Kriging ...............138 Appendix 3.6. Gam plots of the effect of each smoothed variable on estimated dung densities conditional on other variables included in the model ......................144 Appendix 3.7. Conflict timeline. A Chronology of Military Occupation, Elephant Poaching, and ICCN Control in the RFO (with permission from Dr. John Hart) ......153  - vii -  LIST OF TABLES Table 2.1: Candidate variables to be included in the spatial models ........................39 Table 2.2: Collected law enforcement data in the Odzala National Park in 1999 and in the Lossi area in 2000 ....................................................................................40 Table 2.3: Fitted Generalized Linear Models of elephant dung at the sampling site level  ......................................................................................................................40  Table 2.4: Fitted Generalized Linear Models of elephant dung at the segment level  ......................................................................................................................41  Table 2.5: Fitted Generalized Additive Models of elephant dung at the segment level  ......................................................................................................................42  Table 2.6: Moran's I (index of spatial autocorrelation) with associated z-statistic and p-value of original dung encounter rates ............................................................43 Table 2.7: Moran's I (index of spatial autocorrelation) with associated z-statistic and p-value of the residuals after modeling ..............................................................43 Table 2.8: Coefficient of variation (CV) of modeled encounter rates for fitted models ......................................................................................................................44 Table 3.1: Number of sampling locations and transects for each wildlife survey (1995 and 2006) in the Okapi Faunal Reserve .........................................................87 Table 3.2: Candidate covariates included in the spatial models for 1995 and 2006 ...................................................................................................................88 Table 3.3: Survey effort, encounter rates and dung densities of different ungulates in the RFO from the data subset used for spatial and spatio-temporal models ........89 Table 3.4: Change in ungulate dung densities (per hectare) between 1995 and 2006 in the whole reserve, the Green and the Red Zone .........................................90 Table 3.5: Differences in ungulate dung densities (per hectare) between the Green and the Red Zone in the Okapi reserve in 1995 and 2006 ........................................91 Table 3.6: Fitted GAM's of ungulate dung densities in the Okapi reserve in 1995, 2006 and in both time periods combined ........................................................92 - viii -  Table 3.7: Fitted GAM's of ungulate dung densities in the Okapi reserve in 1995, 2006 and in both time periods combined, including ecozone covariates ........94  - ix -  LIST OF FIGURES Figure 2.1: Map of the Odzala National Park in the Republic of Congo and Survey blocks ......................................................................................................................45 Figure 2.2: The design of the sampling unit in Odzala ..............................................46 Figure 2.3: Encounter rates of all human activities (a) and hunting signs only (b) in Odzala National Park and Lossi. Encounter rates are given as number of observations per day .................................................................................................47 Figure 2.4: Catch per Unit Effort (CPUE) index (see text) for hunting signs in relation to distance from the nearest village (a) in Lossi. Elephant dung encounter rate in relation to distance from the nearest road (b) and Catch per Unit Effort for hunting signs (c) in Lossi ......................................................................................................................48 Figure 2.5: Conditioning plot of elephant dung encounter rate in relation to distance from the nearest road, given distance from the nearest patrol in the park ................49 Figure 2.6: Conditioning plot of elephant dung encounter rate in relation to distance from the nearest forest clearing given distance from the nearest patrol route ..........50 Figure 2.7: Elephant dung encounter rates in relation to distance from the nearest patrol route (a), distance from the nearest road (b), slope (c) and distance from the nearest forest clearing (d) within Odzala National Park ............................................51 Figure 2.8: Boxplots for all types of vegetation and aggegated classes in relation to elephant dung encounter rates ..................................................................................52 Figure 2.9: Predicted density map for elephants (animals per km2) for the entire study area using model MS2 .....................................................................................53 Figure 2.10: Predicted density map for elephants (animals per km2) for the National Park area using model MP2 ......................................................................................54 Figure 3.1: Map of the Okapi Faunal Reserve in the Democratic Republic of Congo with sampling locations .............................................................................................96  -x-  Figure 3.2: Mean ungulate dung density estimates in 1995 and in 2006 in the entire area of the Okapi Faunal Reserve and in the Green and Red Zone within the reserve ......................................................................................................................97 Figure 3.3: Change in ungulate dung density between 1995 and 2006 versus ungulate density in 1995 ...........................................................................................98  - xi -  ACKNOWLEDGEMENTS I would like to express my sincere gratitude to my supervisor, Professor Dr. Anthony Sinclair for making this thesis project possible and for his dedicated support and guidance during all these years. I'm also grateful to the members of my thesis committee who were very helpful in giving me advice and reviewed my results and writing: Dr. Brian Klinkenberg, Dr. Peter Arcese, Dr. Diane Srivastava and Dr. Jacob Goheen. I want to express special thanks to the late Dr. Jamie Smith who continued to offer his help despite his illness. My thesis work would not have been possible without Dr. John Hart. He shared all the survey data for the Okapi Faunal Reserve and provided indispensable information, scientific support and motivation. I admired his enthusiasm and his positive attitude in face of the incredible difficulties of working in the Democratic Republic of Congo during the periods of civil unrest. I also thank his wife Dr. Terese Hart for her support and hospitality. I'm also very grateful to Falk Grossman who organised, with Dr John Hart, the surveys in the Okapi Reserve in 2004-2006 and who was indispensable in getting the data organised. I sincerely thank the courageous field people that collected the data in the Democratic Republic of Congo, especially the field team leaders Simeon Dino, Faustin Kahindo and Chryso Vyahavwa. Paulin Tshikaya also helped with the field work and the data management for the survey from 1994 - 1996. I'm grateful to Jean-Marc Froment and ECOFAC (Conservation et Utilisation Rationelle des Ecosystèmes Forestiers d'Afrique Centrale), who provided the logistical support and field staff that made our wildlife surveys in the Odzala National Park possible. Also many thanks to the field teams, in particular, Viktor Mbolo, Hilde Vanleeuwe and Saturnin Ibata who worked hard to collect data under difficult field conditions. I want to thank Professor Steve Buckland, Dr. Len Thomas and Dr. Fiona Underwood from the University of St Andrews who consulted on the design of the surveys in Odzala and the analysis of elephant survey data for the MIKE Pilot Program (see be- xii -  low). I'm indebted to Dr. Samantha Strindberg from the Widlife Conservation Society, who gave valuable advice on my analysis of the data. I thank Dr. Will Cornwell who helped me with programming in R statistical software. I would like to thank Erik Lindquist from the South Dakota State University who provided data on forest cover in the Okapi Reserve. I thank Dr. Nadine Laporte from the Woods Hole Research Center who gave us remote sensing data for the Okapi Wildlife Reserve and Professor Philippe de Mayer from the University of Ghent who provided digital cartographic data for the Okapi reserve. Part of the data that I used in this thesis was collected under the MIKE (Monitoring Illegal Killing of Elephants) Pilot Program for Central Africa. I acknowledge CITES (Convention on International Trade in Endangered Species of Wild Fauna and Flora) and WCS (Wildlife Conservation Society) for whom I helped conducting the initial MIKE pilot surveys in Central Africa. Field work for the MIKE surveys was funded by the United States Fish and Wildlife Service (USFWS) and WCS. WCS also provided funding for the field surveys in 1995-1996 and 2006-2007. I acknowledge the Institut Congolais pour la Conservation de la Nature (ICCN) of the Democratic Republic of Congo and the Ministere des Eaux et Forêts of the Republic of Congo who gave the authorization and administrative support to conduct the surveys. And last but not least, I am very indebted to my wife and my four children for their encouragement and their patience for the odd times that I worked, especially on evenings and weekends. I hope that my work will inspire my children to care about our living world and environment.  - xiii -  CO-AUTHORSHIP STATEMENT I am the main author of all four chapters in this dissertation and prepared the entire manuscript. For Chapter 2, I was responsible for the design of the research and the field sampling. I coordinated the data collection, trained and hired people and was also personally involved in the field sampling. I have done all the data analysis and statistics, written the entire text and created all tables, figures and maps. For Chapter 3, I received data from Dr. John Hart from a field survey done in 1994-1995 and from Dr. John Hart and Falk Grossman from a field survey done in 2005-2006. However, the analysis of the data was my work and responsibility. I also wrote the entire text, created all tables and figures.  - xiv -  CHAPTER 1. Introduction 1.1 General Introduction Until recently, little was known about the population sizes of large mammals in Central African forests. In contrast to the extensive research conducted in the savannas in Eastern and Southern Africa, wildlife surveys and monitoring have been rare and scattered in both space and time. Surveys in the dense equatorial forests were hindered by difficult access, limited visibility, political instability and absence of funding. It was long believed that the rainforests of Central Africa still sheltered large numbers of animals from human onslaught because of their vastness and difficult access. Even elephants were thought to be more protected from hunters in those dense forests than in open savannah where they were being depleted outside of protected areas (Douglas-Hamilton, 1987). As more surveys were conducted during the last 10-20 years, it has become apparent that humans have played a key role in determining the distribution and abundance of most animals in forest habitats, at least as much as in savannas. Barnes et al. (1991) has established a relationship between the distribution of elephants and humans in northeast Gabon showing that heir numbers increased with increasing distance from roads and human settlements. This general relationship was also observed in other places across Central Africa (Fay and Agnagna 1991, Barnes et al. 1997, Buij et al. 2007). The same impact of roads on population numbers was found with other mammals, such as forest antelopes (Laurance et al. 2006) and certain monkeys (Lahm et al. 1998). In all of those studies, human impacts outweighed habitat in determining the distribution of wildlife. Recent surveys have also found that densities of some large mammal species were much lower in forest habitat areas than expected. For example, a survey in 2003 in Salonga in the Democratic Republic of Congo, an area of 36000 km2 that was once thought to be one of the last strongholds for African forest elephants, showed that not more than 2000 elephants remained in the entire area (Blake et al. 2007). Many wildlife  -1-  populations in Central Africa are decreasing, although little is known about the magnitude of this decline. This trend is accelerating as more areas are being developed and opened up by logging companies and development projects. However, it is also clear that habitat loss alone is not the main cause of the declines of large-bodied forest mammals. Hunting for meat is attaining massive proportions across the region (Fa et al. 2005, Milner-Gulland and Bennett 2003). This is a result of growing human populations (Wilkie and Carpenter 1999), increased access to hitherto remote areas, improved hunting methods, the greater availability of modern weapons, and civil strife (Draulans and Van Krunkelsven 2002, Dudley et al. 2002). Besides hunting for meat, the illegal killing of elephants for ivory has also been a major cause of decline for forest elephants (Barnes et al. 1995). Diseases are another factor. Gorillas are being decimated by Ebola in north-east Gabon and large areas of the Republic of Congo (Walsh et al. 2003). The Ebola virus has also been found in chimpanzees and duikers, but it is unknown how it has affected populations of these species. Ebola outbreaks in humans in the region were triggered by handling carcasses of infected animals (Rouquet et al. 2005). Despite these dramatic changes in the last few decades, reliable information and quantitative data about spatial and temporal trends in wildlife populations is still lacking in most parts of Central Africa. It was, therefore, timely that CITES (Convention of International Trade in Endangered Species of Wild Fauna and Flora) initiated a monitoring program of elephants, called MIKE (Monitoring Illegal Killing of Elephants). I was involved in setting up the pilot project in Central Africa to develop a site-based monitoring program of elephants and law enforcement (Beyers et al. 2001). My PhD research originated from experience with this program and the need to address several questions about how to improve monitoring designs for elephants, how to respond to emerging threats and how to maximize the rate of information gain to help park and wildlife staff manage these threats. For my PhD research project I have been particularly interested in the spatial and spatio-temporal patterns of the abundance of elephants and other ungulates as a func-  -2-  tion of human and environmental covariates. In particular, I have focused on answering the following key questions: (1) How does animal abundance respond spatially to human threats and habitat at the scale of a National Park. (2) What is the influence of protection and law enforcement on these patterns? (3) How do relationships with human factors change over time? My study sites were the Odzala National Park in the Republic of Congo (Chapter 2) and the Okapi Faunal Reserve in the Democratic Republic of Congo (Chapter 3). These are two forest sites, although Odzala also contains some savannah habitats in the southern part of the park. In Chapter 2, I look at spatial patterns of elephant abundance inside and outside the Odzala National Park and try to explain these as a function of human influences and habitat. Besides variables associated with negative human impacts I focus specifically on the impact of law enforcement. I develop spatial models of elephant densities as a function of spatial covariates by combining estimates of elephant dung collected on line transect surveys with multivariate generalized modeling techniques. I obtain many covariates from GIS sources and Remote Sensing (satellite images and radar images). These sources of information are relatively cheap and easily accessible. This is important for the sustainability of a long-term monitoring program, especially in Central Africa where financial and logistical constraints of field work are very challenging. Spatial analysis of wildlife data has become popular in recent years, with advances in the development of Geographic Information Systems, powerful statistical software, and an increase in access to spatial data. However, most studies have focused on modeling and/or predicting species occurrence rather than modeling species abundance (Scott et al. 2002). Combining density estimates from line transects with spatially explicit models was originally developed for marine mammal surveys (Hedley et al. 2004, Williams et al. 2006). This approach has only been recently applied in Central Africa in a regional study of the relationship between elephants and distance to roads (Blake et al. 2007). Besides ecological questions I address a few methodological issues in this chap-3-  ter. First, I compare the performance of 2 statistical modeling techniques, Generalized Linear Models (GLM) and Generalized Additive Models (GAM), in terms of their ability to explain patterns of elephant densities. Generalized linear models are an extension of least squares regression models. They allow for more flexibility with regard to the relation between independent and dependent variables and are not restricted to data with a normal error distribution. Generalized Additive Models go even further by modeling nonlinear relations through a smooth function of the independent variables. Second, spatial modeling has a potentially interesting application in monitoring population trends. The statistical power of detecting trends in animal populations depends on the precision of the estimated abundance and will be higher with increased precision. If spatial models can explain spatial variation in the data, they will improve the precision of abundance estimates and thus they increase the power to detect trends (Augustin et al. 1998). This could be particularly important for small populations and populations in decline, because precision is usually inversely related to abundance and will therefore be smaller in small populations (Barnes 2002). I examine if there were any improvements in precision with the spatial models that I created. Third, I use the models to predict densities across a surface in the national park. Since we conducted surveys in sub-zones within the park, this predictive model allows me to obtain an estimate of the elephant population for the whole park as well as for the subzones. The focus of the third chapter of this thesis is entirely on spatial and temporal trends of distribution and abundance of elephants and other mammals in response to humans and the civil war in the Democratic Republic of Congo. As far as I am aware, this is also the first study in Central African forests where a quantitative change in wildlife populations can be established from a comparison between two surveys that used exactly the same design and methodology. It is also one of a few studies that looks at trends of multiple species rather than of a single species. I obtain line transect estimates of dung densities of elephants, okapi and small forest antelopes. I then analyse changes in den-4-  sities of two time periods, and expect changes to be significant because of an episode of civil unrest between the two surveys. The Democratic Republic of Congo went through a devastating war between 1996 and 2006 and the Okapi reserve was located in one of the most turbulent regions in the country. Two thirds of the reserve was not accessible to wildlife staff during several years and there were reports of large scale ivory poaching in this area. I analyse line transect data from a survey before the conflict (1995-1996) and after the conflict (2006-2007). I compare densities and develop spatiotemporal models relating these to survey period and time specific factors related to human impact. I also model each survey separately to see if animals responded differently in each time period and I compare responses between different species or species groups.  -5-  1.2 References Augustin, N. H., D. L. Borchers, E. D. Clarke, S. T. Buckland, and M. Walsh. 1998. Spatiotemporal modelling for the annual egg production method of stock assessment using generalized additive models. Canadian Journal of Fisheries and Aquatic Sciences 55: 2608-2621. Barnes, R. F. W., A. Blom, and M. P. T. Alers. 1995. A review of the status of forest elephants Loxodonta africana in Central Africa. Biological Conservation 71: 125-132. Barnes, R. F. W., K. L. Barnes, M. P. T. Alers, and A. Blom. 1991. Man determines the distribution of elephants in the rain forests of northeastern Gabon. African Journal of Ecology 29: 54-65. Barnes, R. F. W., K. Beardsley, F. Michelmore, K. L. Barnes, M. P. T. Alers, and A. Blom. 1997. Estimating forest elephant numbers with dung counts and a geographic information system. Journal of Wildlife Management 61: 1384-1393. Barnes, R. F. W. 2002. The problem of precision and trend detection posed by small elephant populations in West Africa. African Journal of Ecology 40: 179-185. Beyers, R., Thomas, L., Hart, J., and S. Buckland. 2001. Monitoring Illegal Killing of elephants (MIKE), Central African Pilot Project, Technical Report No 2: Recommendations for ground-based survey methods for elephants in the central african forest region. http://www.cites.org/eng/prog/MIKE/pilot/tech_rep4.shtml Blake, S., S. Strindberg, P. Boudjan, C. Makombo, I. Bila-Isia, O. Ilambu, F. Grossman, L. Bene-Bene, B. de Semboli, V. Mbenzo, D. S'hwa, R. Bayogo, L. Williamson, M. Fay, J. Hart, and F. Maisels. 2007. Forest elephant crisis in the Congo Basin. PLOS Biology 5: 0001-0009.  -6-  Buij, R., W. J. McShea, P. Campbell, M. E. Lee, F. Dallmeier, S. Guimondou, L. Mackaga, N. Guisseougou, S. Mboumba, J. E. Hines, J. D. Nichols, and A. Alonso. 2007. Patch-Occupancy Models Indicate Human Activity As Major Determinant Of Forest Elephant Loxodonta Cyclotis Seasonal Distribution In An Industrial Corridor In Gabon. Biological Conservation 135: 189-201. Douglas-Hamilton, I. 1987. African elephants population trends and their causes. Oryx 21: 11-24. Draulans, D., and E. Van Krunkelsven. 2002. The impact of war on forest areas in the Democratic Republic of Congo. Oryx 36: 35-40. Dudley, J. P., J. R. Ginsberg, A. J. Plumptre, J. A. Hart, and L. C. Campos. 2002. Effects of war and civil strife on wildlife and wildlife habitats. Conservation Biology 16: 319-329. Fa, J. E., S. F. Ryan, and D. J. Bell. 2005. Hunting vulnerability, ecological characteristics and harvest rates of bushmeat species in afrotropical forests. Biological Conservation 121: 167-176. Fay, J. M., and M. Agnagna. 1991. A population survey of elephants in Northern Congo. African Journal of Ecology 29: 177-187. Hedley, S. L., S. T. Buckland, and D. L. Borchers. 2004. Spatial distance sampling methods. In: Advanced Distance Sampling (Buckland et al.), Oxford University Press. Lahm, S. A., R. F. W. Barnes, K. Beardsley, and P. Cervinka. 1998. A method for censusing the greater white-nosed monkey in northeastern Gabon using the population density gradient in relation to roads. Journal of Tropical Ecology 14: 629-643.  -7-  Laurance, W. F., B. M. Croes, L. Tchignoumba, S. A. Lahm, A. Alonso, M. E. Lee, P. Campbell, and C. Ondzeano. 2006. Impacts of Roads and Hunting on Central African Rainforest Mammals. Conservation Biology 20: 1251-1261. Milner-Gulland, E. J., and E. L. Bennett. 2003. Wild meat: the bigger picture. Trends in Ecology and Evolution 18: 351-357. Rouquet, P., J.-M. Froment, M. Bermejo, A. Kilbourn, W. Karesh, P. Reed, B. Kumulungui, P. Yaba, A. Delicat, P. E. Rollin, and E. M. Leroy. 2005. Wild animal mortality monitoring and human Ebola outbreaks, Gabon and Republic of Congo, 2001-2003. Emerging Infectious Diseases 11: 283-290. Scott, J. M., P. J. Heglund, M. L. Morrison, J. B. Haufler, M. G. Raphael, W. A. Wall, and F. B. Samson. 2002. Predicting species occurrences: issues of accuracies and scale. Island Press Walsh, P. D., K. A. Abernethy, M. Bermejo, R. Beyers, P. De Wachter, M. E. Akou, B. Huijbregts, D. I. Mambounga, A. K. Toham, A. M. Kilbourn, S. A. Lahm, S. Latour, F. Maisels, C. Mbina, Y. Mihindou, S. N. Obiang, E. N. Effa, M. P. Starkey, P. Telfer, M. Thibault, C. E. G. Tutin, L. J. T. White, and D. S. Wilkie. 2003. Catastrophic ape decline in western equatorial Africa. Nature 422: 611-614. Walsh, P. D., L. J. T. White, C. Mbina, D. Idiata, Y. Mihindou, F. Maisels, and M. Thibault. 2001. Estimates of forest elephant abundance: Projecting the relationship between precision and effort. Journal of Applied Ecology 38: 217-228. Wilkie, D. S., and J. F. Carpenter. 1999. Bushmeat hunting in the Congo Basin: an assessment of impacts and options for mitigation. Biodiversity & Conservation 8: 927-955. Williams, R., S. L. Hedley, and P. S. Hammond. 2006. Modeling distribution and abundance of Antarctic baleen whales using ships of opportunity. Ecology & Society 11:1 -8-  CHAPTER 2. Spatial modeling of elephant abundance in and around the Odzala National Park in the Republic of Congo 1 2.1 Introduction Management and conservation of wildlife species requires reliable and frequent information about trends in their distribution and abundance over space and time. Wildlife managers are also interested in the impact of threats and their own management policies on wildlife. There is a long tradition of monitoring large mammals in East and Southern Africa (Jachmann 2001), but systematic standardized surveys in Central Africa are fairly recent and have been obstructed by difficult access, dense canopy cover, political instability and other major logistical constraints. However, surveys and monitoring of species like forest elephants, great apes and other megafauna are increasingly needed as these are becoming highly threatened in many parts of Central Africa. For example, a recent survey (2003) in the Salonga National Park in the Democratic Republic of Congo, an area of 36000 km2 that was once thought to be one of the last strongholds for African forest elephants because of its remoteness, showed that no more than 2000 elephants remained in the entire park (http:/ /www.wcs.org/international/Africa/africanelephants/MIKE/MikeDRC). Gorillas are also declining rapidly in parts of Gabon and the Republic of Congo as a result of Ebola and bushmeat hunting (Walsh et al. 2003). Hunting of wildlife for meat, including elephants and large apes, has increased dramatically in many parts of Central Africa (Fa et al. 2005; Milner-Gulland and Bennett 2003). This is a result of a rising demand due to the growing human populations (Wilkie and Carpenter 1999), increased access to hitherto remote areas, improved hunting  1.  A version of this chapter will be submitted for publication. Beyers, R., Sinclair, A.R.E., Hart, J. 2008. Spatial modeling of elephants in the Odzala National Park, Republic of Congo.  -9-  methods, the greater availability of modern weapons, and civil strife (Draulans and Van Krunkelsven 2002; Dudley et al. 2002). Ivory hunting has also been a major cause of decline for forest elephants (Barnes et al. 1995). Habitat loss also plays a role in the decline of wildlife species in Central Africa, but second only to hunting (Wilkie and Carpenter 1999). From 1999 to 2001, I was involved in setting up a pilot project in Central Africa to develop a site-based elephant monitoring programme, called MIKE (Monitoring Illegal Killing of Elephants) which was mandated by CITES (Convention of International Trade in Endangered Species of Wild Fauna and Flora) (Beyers et al. 2001). Using data from the programme, I investigate in this thesis the factors that influence the spatial distribution and abundance of elephants in the Odzala National Park in the Republic of Congo, one of the MIKE pilot sites. To address this question, I analyse line transect data of forest elephant dung as an indirect indicator of elephant abundance, using a spatial modeling approach. Line transect surveys of animal signs are the most commonly used method to estimate densities of forest mammals in Central Africa (Plumptre 2000; Koster and Hart 1988; Barnes et al. 1995). By modeling dung abundance as a function of spatial variables I try to explain patterns of elephant abundance as a function of human threats, protection and habitat. I specifically look at variables associated with human presence and access (roads, settlements), protection and law enforcement (national park boundary, field patrols) and habitat (vegetation, forest clearings, slope). Besides questions related to the conservation biology of elephants, I also examine some methodological issues. First, I compare the performance of Generalized Linear Models and Generalized Additive Models to model patterns of elephant abundance. Second, I evaluate these models in terms of their ability to improve precision of global abundance estimates of dung density. Third, I use the models to predict elephant abundance across a surface in and beyond the study area to obtain global estimates for subzones of the study area and the national park which extends beyond the study area.  - 10 -  2.2 Methods 2.2.1 Study site The study area is located in and around the Odzala National Park in the Republic of Congo (figure 2.1). The diverse vegetation in Odzala includes dense rainforest with a closed canopy and more open forest dominated by Marantaceae in the understorey, open savannah, and swamp forest (Maisels et al. 1997; Lejoly 1996). A special feature in the park is the existence of forest clearings, also locally called "bais", or forest clearings which attract gorillas (Gatti et al. 2004), elephants (Querouil et al. 1999) and a host of other animals. The National Park boasts a large number of animal species and contains the highest diversity of diurnal primates in the Central African rainforest basin. It has some of the largest populations of lowland gorillas in the world (Bermejo 1999) but these have recently been declining rapidly due to Ebola (Walsh et al. 2003; Bermejo et al. 2006). Other large mammal species include duikers, bongo and chimpanzee in forest; elephant, forest buffalo and bushbuck in forest and savannah; and lion and Grimm's duiker which are only found in savannah (Ecofac, http://www.ecofac.org). Odzala National Park has a humid tropical climate with a mean annual rainfall above 1500mm, 2 wet and 2 dry seasons (Querouil et al. 1999). The area around the park is bordered by villages and roads in the south, east and north. The National Park was created in 1935 and expanded in 2001 from 2848 km2 to about 13545 km2 (Aveling 2001). The area was virtually unprotected until 1992 when the European Union launched its regional conservation programme, ECOFAC (Conservation et Utilisation Rationelle des Ecosystèmes Forestiers d'Afrique Centrale - Conservation and rational use of forest ecosystems in Central Africa). ECOFAC has since been carrying out a variety of conservation activities including development of park infrastructure and tourist facilities, training of guards and small development projects around the reserve to discourage local people from hunting (Ecofac, http://www.ecofac.org). Prior  - 11 -  to ECOFAC, hunting in the park was severe, and elephants were decimated during the 1980's (Aveling 2001).  2.2.2 Study design and data collection We conducted surveys of elephants and other large wildlife species inside and outside the Odzala National Park in 2000 as part of the MIKE Pilot Project. Field teams led by two Congolese biologists, Saturnin Ibata and Viktor Mbole and one expatriate biologist, Hilde Vanleeuwe, carried out the surveys. I was responsible for the overall supervision of the field work together with Dr. John Hart and I also participated in some of the field sampling. The local team leaders were trained in distance sampling, navigation and basic field biology skills during a regional field training program from September to November 1999 in the Nouabale Ndoki National Park in northern Congo. The training was coordinated by Dr. Lee White and I was one of the four trainers. A follow-up training in basic data management, analysis and reporting was led by Dr. John Hart and myself at the Limbe botanical gardens in Cameroon in December 2000. Logistical and financial constraints, and safety concerns in the northern part of the National Park, led us to select three blocks within a larger study area rather than attempting to survey the entire area (figure 2.1). Within the survey blocks, starting points of the sampling units (transects, see below) were systematically placed on intersections of the gridlines of a grid (which had a cell size of 2'30" by 2'30" or 4.6 by 4.6 km). They were 13.8 km apart in an east - west direction and 9.2 km in a north - south direction. We opted for a systematic survey design as opposed to a purely random design because the former provided more equal coverage of the area and captured better the range of values of spatial variables that we were interested in. A systematic design usually yields better precision of abundance estimates and "there is no compelling reason to use a random design, unless the design coincides with a regular natural pattern" (Buckland et al. 2001, page 231). The survey blocks represented different environmental conditions in the area. The most northern block ("North"), with an area of 2367 km2, comprised 13 sampling units and was situated in a - 12 -  more remote part of the park where human influence was known to be very low. The southern block ("South"), with an area of 1931 km2, had 9 sampling units and was located in an unprotected area outside the park without much prior knowledge of the conditions that existed in the area. The middle block ("Middle"), with an area of 3563 km2, had 22 sampling units and was placed in the southern part of the park, closer to human habitation, but also where most of the park facilities were located (headquarters, tourist camp and air strip). We used indirect methods to count elephants because, despite their large size, they are rarely visible in forest. Line transects and counts of dung or other signs are the most practical method to count wildlife in dense forests of tropical Africa (Barnes and Jensen 1987; Fay 1991; Plumptre 2000). Each sampling unit was 5 km long and consisted of 5 short straight line transects of 200 m each, interspersed with 4 reconnaissance trails (recces) of 1 km each (figure 2.2). The start of each transect was determined in advance from the sampling design and located in the field with a GPS to an accuracy of 30m. Transects were cut in a straight line following a predetermined compass bearing. Reconnaissance walks, also called "recces", followed the "path of least resistance" toward a fixed direction, but not deviating more than an arbitrary 45 degrees. They often followed animal or human trails and circumvented obstacles like thick vegetation and rocks and required much less effort than cutting straight line transects. The data collected on recces was biased because they did not adhere to a random design. However, in combination with transects, recce data could improve overall precision of abundance estimates (Beyers et al. 2001; Walsh et al. 2001). We recorded signs of animals and humans on both transects and recces and measured the distance along the line. On transects we also measured the perpendicular distance of the observed object to the transect line.  2.2.3 Spatial environmental data I collected data on candidate covariates, representing both habitat and human influences in the study area (table 2.1). Because this study was designed to become part of a regular monitoring cycle, I used variables that could be measured at the scale of the - 13 -  study area in a short period of time and that could be monitored on a regular basis using satellite photos and limited field input. I used ArcGIS (ESRI) to store, retrieve and manipulate spatial data. I georeferenced all spatial data to a common framework (UTM, WGS 94 spheroid) that had the same datum and projection as the one that was used for georeferenced field data. I calculated locations of sampling segments and derived values of candidate explanatory variables for each segment using GIS. I measured distances from linear and point features (roads, villages, patrol routes, park infrastructure, rivers, forest clearings) to the middle of each sampling unit or segment. Vegetation data were obtained for the national park only from 2 classified SPOT satellite images from 1994 (Maisels et al., 1997). I either calculated the proportion of each vegetation class within a buffer around the sampling unit, or identified the major vegetation class surrounding a sampling unit (see below). Slope was derived from a digital elevation model (DEM) produced by the Space Shuttle Radar Topography Mission (SRTM Seamless Data Distribution System, Earth Resources Observation and Science (EROS), http:/ /seamless.usgs.gov). The DEM had a cell size of 90 meters. I calculated the average slope in a buffer area of 0.9 km on each side of the sampling segments.  2.2.4 Law enforcement monitoring data I used law enforcement monitoring data, collected on patrols, to derive a measure of protection and to gain information about the relationship between spatial variables that I was interested in modeling and human impact on wildlife. Specifically, I used the location of patrol routes and the distance traveled on them as a measure of patrol effort. I used observations of human activities as a measure of human impact in the patrolled area. Patrol staff employed handheld computers connected to a GPS and a software package, called "Cybertracker" (http://www.cybertracker.co.za/), to collect data on their movements and observations. This allowed them to record an accurate position (accurate within a radius of 30 meters) for each observation and to avoid double counting on subsequent patrols. They recorded data on transport used (foot, car, pirogue), regular waypoints of the followed route, signs of human activities (hunter camps, snares, am-  - 14 -  munition, footprints, machete cuts and encounters with hunters) and signs of animals (live observations, tracks, dung and nests). There were two types of patrol activities that I used data from. The first type was part of a regular patrol routine within the park boundaries. The main purpose of these patrols was to deter people from hunting and conducting other illegal activities inside the park and to monitor illegal incidents. Most patrols were on foot, but some patrols made use of a car or boat to get quickly to more distant areas. I plotted all patrol waypoints and connected these to create linear patrol routes. For each sampling location of the wildlife transect survey, I calculated the distance to the nearest patrol route and included this as a measure of protection effort in the spatial models. Staff collected data on 38 patrol days during their monthly patrols in April, June, July and August 1999. Although they represented only about half of all patrols that were typically carried out within this timeframe, the routes corresponded well with the area of the park that was actively and regularly patrolled (Jean-Marc Froment, Ecofac project manager, personal communication 2003). The second patrol type was a one-off 'reconnaissance patrol' conducted during 40 days in the months of April, May, June and July 2000 in an area near the village of Lossi (Lossi gorilla sanctuary), southwest of the Odzala National Park. In this area, gorillas had been studied and habituated to humans since 1995 (Bermejo 2004). The objective of the reconnaissance patrol was to assess human impact on wildlife in a normally unpatrolled area. Since this was a one-off effort of limited duration, it is unlikely that this mission would have had an effect on elephants during our wildlife surveys. Therefore, I did not include the routes from this patrol in my analysis of the elephant transect data. I used data from both patrol types to look at signs of hunting and how they related to spatial covariates, such as location of roads and villages. To do this, it was necessary to normalize the number of observations made by patrol members as a function of effort. I calculated a simple Catch Per Unit Effort Index (CPUE) analogous to that which has been used in fisheries research. Hilborn et al. (2006) also used a CPUE to monitor - 15 -  poaching intensity in Serengeti National Park. The expected or probability of catch (i.e. the number of observations of illegal activities) was related to the available number of cases and search effort (equation 2.1)  C = NβE  (2.1)  where, C is expected catch, N is the number of cases available to be caught, β is a constant relating catch to effort, and E is effort (Jachmann and Beyers 2003). I used a simple measure of effort, namely the distance traveled along a section of the patrol route. I plotted patrol routes on a georeferenced map in a GIS and overlaid this with a UTM grid with squares measuring 5x5 km. I then used the distance traveled along the patrol route in a square as the unit of effort for each square. I calculated the CPUE index by dividing the number of observations of an illegal event (the "catch") by the distance travelled. I used a correction factor for different visibilities along the patrol route due to vegetation as proposed in Jachmann and Beyers (2003). The number of people in a patrol also influences the probability of catch (Arcese et al. 1995). However, I did not include it in the CPUE as these data were not available, but from personal observations I knew it to be quite constant and typically not below five. The "catch" included all signs of activities associated with hunting, such as direct encounters with poachers, poacher camps, elephant carcasses, found firearms, found ivory and snares. I used GLMs in a spatial modeling framework to relate CPUE to spatial covariates, as explained below. I lumped all signs of hunting into one index because the sample size of each sign on its own was too small to model them separately. In the Lossi area I also looked at the relationship between encounter rates of elephant dung observed on the reconnaissance patrol and distance from the nearest road and from the nearest village.  - 16 -  2.2.5 Data Analysis 2.2.5.1 Distance analysis I estimated densities of elephant dung from the line transects using the software programme DISTANCE 4.0 (Research Unit for Wildlife Population Assessment, http:/ /www.ruwpa.st-and.ac.uk/distance/). A short introduction to Distance is given by Thomas et al. (2002). Buckland et al. (2001) describe it in detail. The underlying assumption of line transect sampling is that all objects on the transect line are detected, but that the probability of detection declines with increasing distance from the line. This probability of detection is mathematically modeled as a detection function. The area under the detection function curve is called the effective half-strip width, which corresponds with the width of a strip if all objects would have been detected within that strip. Since this applies to one side of the transect line only, the half-strip width should be multiplied by 2 to get the 'Effective Strip Width' (ESW) which covers both sides of the transect. Density is calculated from the number of detected objects ( n ) divided by the length of the transects ( L ) and the effective strip width (ESW): Dˆ =  n ESW * L  (2.2)  In order to model the detection function, it is usually necessary to truncate the data at some distance from the transect line, thereby dropping observations that are beyond that distance. There are several functions (uniform, half-normal, hazard rate and negative exponential) available in Distance to model detection probabilities and adjustment terms can be added to modify the shape of the curve to obtain a better fit (Thomas et al. 2002). I explored several options and selected the model with the best fit, using Akaike's Information Criterion (AIC). Detection probabilities can vary with observer, habitat, and other covariates that affect visibility. I stratified the data and fitted separate detection  - 17 -  functions to the three survey strata and to forest versus savannah habitat and tested for differences in the detection function. To increase sample size and the area covered by sampling, I combined dung encounter rates on recces with encounter rates on transects. To reduce the above mentioned bias that occurs with recce data, I calibrated encounter rates on recces with data from nearby transects using a ratio estimator as described in Beyers et al. (2001).  2.2.5.2 Spatial modeling I fitted spatial models to encounter rates of elephant dung employing Generalized Linear Models (GLM) and Generalized Additive Models (GAM), using the statistical software package R (http://www.r-project.org). Cumberworth et al. (1996), Hedley et al. (1999) and Williams et al. (2006) successfully used GAMs in combination with line transect sampling to model abundance of marine mammals as a function of spatial covariates. Details on the concept and techniques of this approach can be found in the above mentioned publications and in Hedley et al. (2004). I employed this model-based approach for the following purposes: 1. to explain variation in abundance of elephants (in this case as indicated by elephant dung) across space, and identify factors influencing abundance, 2. to improve precision of the global abundance estimate, and 3. to create predicted density maps of elephants across the study area. With a successful model, one can predict abundance in subzones or strata of any size and shape within the study area. This contrasts with a traditional design-based approach where it is necessary to fix strata in advance and implement a random or systematic sampling design containing enough sampling locations within each stratum to get estimates for each stratum (Hedley et al. 2004).  - 18 -  2.2.5.3 Generalized Linear Models Generalized linear models (GLM's) are a flexible extension of least-squares linear regression models. They allow some of the rather limiting assumptions of the linear models to be relaxed (Efron and Tibshirani 1993; Mc Cullagh 1989) with regard to: - the linear relationship between response variable and explanatory variables, - the normal probability distribution, and - the constant variance independent of the mean. In least-square regression, the response variable is directly related to explanatory variables through a linear relationship. In a GLM, the dependent variable is related to a linear combination of explanatory variables through a link function: g(E(Y )) = α + β1 x1 + ... + β p x p + ε  (2.3)  where g(E(Y )) is the link function of the expected value of explanatory variable Y, α is the intercept, x p is the explanatory variable, β p is the regression coefficient (one for each explanatory variable) and ε is the error. The link function ensures that the data are linear and that they are constrained within a range of possible values (Guisan et al. 2002). In this way negative predictions can be avoided, which is important with count data (Jones et al. 2002). For count data a log link is often appropriate because of Poisson errors. When developing the models I tried different transformations of the independent variables and checked if these improved model fit. GLM's are flexible in that they can accommodate a variety of probability distributions and are, therefore, more suitable for modeling most ecological data. With least squares regression, the response variable is assumed to follow a normal distribution. However, with wildlife count data, this is almost never the case. Count data are usually highly skewed and often follow a Poisson frequency distribution with many observations of small counts and few observations of large counts. Unlike in a normal distribution, which assumes a constant variance, the variance in a Poisson is equal to the mean and thus increases with the mean. When the variance is higher than the mean, the distribution is - 19 -  said to be overdispersed (Vincent and Haworth 1983). This is often the case with count data that have many zeros (absences). In forest elephants, for example, Walsh et al. (2001) found that the variance/encounter rate for elephant dung followed a power law. One can choose the wrong model and misinterpret the results if overdispersion of the Poisson model is not accounted for. I checked for overdispersion by looking at the ratio of the residual deviance over the remaining degrees of freedom in a fitted model. A value of higher than one indicated overdispersion (Crawley 2002). I corrected for overdispersion using a Quasipoisson distribution instead of a Poisson distribution in R. With a Quasipoisson distribution, the dispersion parameter is not fixed as it is in a Poisson distribution (where it is the mean), but is an estimate of dispersion and is thus allowed to vary (Potts and Elith 2006).  2.2.5.4 Multicollinearity I checked for muticollinearity between indpendent variabls by plotting a matrix diagram that showed correlation coefficients and scatterplots of all variables against each other. If two or more variables were linearly correlated with each other and had a correlation coefficient of more than 0.75, I kept the covariate that was best correlated with the dependent variable. Distance from the nearest village was strongly correlated with distance from the nearest road (Appendix 2.1) and I excluded it from the models. Distance from the park boundary was also correlated with distance from the nearest patrol but this was due to a linear correlation between these two variables for data points outside the park because all patrols were located inside the park. The same was true for the relationship between distance from the nearest forest clearing and distance from the park boundary, since all forest clearings were also inside the park. I therefore included interaction terms for park and patrol and for park and forest clearing in the initial models. To examine the effect of patrol routes and forest clearings within the national park only, I developed separate models for the entire study area and for the park. Within the park, the linear relationships between patrol and park and between forest clearing and park disappeared (Appendix 2.1). However, the same strong correlation existed be- 20 -  tween distance from the nearest road and distance from the nearest village so, consequently, distance from the nearest village was excluded from this analysis. Also, a strong linear correlation between distance from park boundary and distance from the nearest road appeared. The latter is not surprising as the road runs parallel to the park over much of the length of the park boundary. I ran separate models including either road or park in order to see which one would be the better predictor of elephant abundance.  2.2.5.5 Model simplification I fitted an initial model, called the maximal model, that included all candidate covariates and their interactions. I then simplified the model by eliminating redundant terms (i.e. variables) to achieve the simplest model possible following the principle of parsimony. I used a backwards deletion procedure recommended by Crawley (1993). This comprised the removal of non-significant independent variables and interaction terms between them and, in case of vegetation classes, the grouping of factor levels that were not significantly different from each other. I started by removing the variable with the least significant effect (highest p-value) first. Then, I tested if the removal of the variable caused a significant increase in deviance compared to the previous model using Analysis of Variance (ANOVA) and the F-test. If deviance did not increase significantly, I dropped the variable. Otherwise, I put it back into the model and removed the second least significant term. I continued doing this until there were only significant terms in the model. This was the minimum adequate model.  2.2.5.6 Spatial modeling at different spatial scales I fitted Generalized Linear Models to the data at 2 different spatial scales. The first level ("sampling location level") was at the scale of each sampling location whereby the entire 5km recce-transect combination was taken as one data point. Abundance estimates were obtained for the whole sampling unit and values of covariates were measured for the mid-point of the sampling unit. For vegetation in the park, I calculated  - 21 -  the proportion of each vegetation type inside a 900 m buffer area area along the sampling unit. The proportion of each vegetation type was thus included in the models as a continuous variable. The second level ("segment level") was at a finer scale of sections of the sampling unit for which combined encounter rates were obtained. I calculated encounter rates for 4 sections of 1200 m each (one recce plus one adjacent transect). This provided 4 times more data than the sampling location level. I measured covariate values at the center of each segment. For vegetation in the park, I selected the major vegetation type surrounding the segment and included it in the analysis as a factor. I first looked at the effects of vegetation on elephant abundance separately using ANOVA. I simplified the ANOVA model by aggregating vegetation classes and performing a F-test to see if the aggregation was justified; that is, if it didn't contribute significantly to an increase in deviance. If I found a significant effect of certain vegetation classes or aggregated classes on elephants, I included them in the GLM together with the other covariates.  2.2.5.7 Generalized Additive Models A Generalized Additive Model (GAM) is a non-linear extension of a GLM. As with a GLM, a link function relates the dependent variable to explanatory variables. However, covariates are assumed to influence the dependent variable through additive smooth functions instead of through simple linear parameters (Hastie and Tibshirani 1990). This permits modeling of truly non-linear phenomena that cannot be transformed into linear relations. As with a GLM, the data may conform to a variety of distributions (poisson, binomial, gamma, etc.) A GAM has the following general form: g(E(Y )) = α + s1 (x1 ) + ... + s p (x p ) + ε  (2.4)  where s p (x p ) is a smooth function of explanatory variable (x p ) and the other terms are the same as in a GLM. Degrees of freedom in a GAM are real numbers, not integers.  - 22 -  They are a measure of the degree of complexity of the smoothing curve. In contrast with GLMs, interactions between variables can not be directly modeled, as terms are assumed to be additive. GAMs are sometimes viewed as "data-driven" as opposed to "model-driven" parametric models such as linear models or GLMs (Guisan et al. 2002). Instead of trying to fit the data to the constraints of a parametric model, it is largely the data that define the model in a GAM. This flexible approach may be more in tune with ecological phenomena, which are often difficult to coax into constraining theoretical linear functions. Model simplification in a GAM is more challenging than in a GLM. Besides having to decide which variables to drop from the model, the optimal smoothing parameter for each variable has to be determined. This can be a very tedious process involving testing of many different combinations of variables and smoothing parameters. A balance must be sought between having too much smoothing and consequently the loss of degrees of freedom and having too little smoothing with little explanatory power. I used mgcv GAMs in R which makes model selection much more convenient. The mgcv package provides automatic selection of smoothing parameters using Generalized Cross Validation (Wood 2001). Note that variables in the model should still be removed manually. I dropped a variable if it met the following three criteria (Wood 2001): - the estimated degrees of freedom of the variable were close to 1, - zero was within the confidence band for the variable, and - the GCV score was lower for the model without the variable than for the model including the variable. If the estimated degrees of freedom were close to one, but there was no increase in GCV when dropping it, nor was zero within in the confidence band, the variable was left in the model as a linear term without a smoothing function. The term for which the confidence band included zero, was dropped first (Wood 2001). I fitted GAMs to the data only at the finer scale of the segment, because only this level provided enough degrees of freedom for this type of modeling. In order to compare - 23 -  the minimum adequate GAMs with the minimum adequate GLMs, I used the same mcgv algorithm on the GLMs, but without any smoothing functions. I then compared GCV scores, pseudo-R values and deviance explained of all models for the same dataset.  2.2.5.8 Spatial autocorrelation Spatial autocorrelation occurs when observations at different locations are more correlated to other closer observations than to more distant observations. In other words, if spatial autocorrelation exists, pairs of observations at a certain distance are more similar or less similar than would be expected from randomly associated variables (Legendre 1993). Spatial autocorrelation in animal abundance can be caused by a spatial trend due to external (environmental) factors that influence abundance or it can be 'true' spatial autocorrelation, perhaps caused by social behaviour, for example herding in a gregarious species. Spatial models, using GLM or GAM, that incorporate locational covariates driving those spatial trends are often able to remove spatial autocorrelation in the residuals (Augustin et al. 1998). Spatial autocorrelation terms can also explicitly be included in a GLM or a GAM and these models may better predict spatial distributions of animals. However, ordinary models are preferred when estimating global characteristics of wildlife distributions (Augustin et al. 1998). Since I was more interested in the latter and especially in explaining patterns of abundance as a function of spatial covariates, I preferred to use ordinary GLMs and GAMs that would explain spatial autocorrelation rather than model it explicitly (Hedley et al. 2004). I used global Moran's I index in the S-Plus for Arcview GIS SpatialStats module (Anonymous 1998) as a measure of spatial autocorrelation. Moran's I statistic ranges from -1 to +1, with zero indicating there is no spatial autocorrelation. Increasing or decreasing values indicate stronger positive or, respectively, negative autocorrelation (Kitron and Kazmierczak 1997). I compared Moran's I in the raw encounter rates of elephant dung to Moran's I in the residuals of the models to detect any improvement after modeling. - 24 -  2.2.5.9 Precision and variance estimation In estimating densities using DISTANCE, encounter rates constitute typically the highest proportion of the total coefficient of variation (CV). If I can explain some of this variation with spatial models, I can decrease the overall CV of the density estimate. I predicted encounter rates at each sampling location and then used predicted values instead of the original values to calculate CV. The overall mean remains the same, but the CV could be lower if some of the variance has been successfully explained. To calculate CV at the level of the sampling locations, a parametric estimation of variance could be used because the sampling units were independent. However, to estimate CV of encounter rates at the segment level I used bootstrapping (Efron and Tibshirani 1993) since these segments were not independent (Buckland et al. 2001). The total CV for density included the other sources of variance associated with detection probability and the ratio estimator. I summed all variance components using the delta method (Buckland et al. 2001; Seber 1982) as follows: CV 2 (D) = CV 2 (E) + CV 2 (ESW ) + CV 2 (R)  (2.5)  where D = density, E = combined adjusted encounter rate, ESW = effective strip width, R = ratio estimator.  2.2.5.10 Predicting density over the area The expected density for each sampling segment can be calculated by dividing encounter rates by the effective area (which is the effective strip width times length) using equation 2.2. From selected models I predicted elephant density for a larger area encompassing the surveyed strata and the area between them. This was reasonable given the broad coverage of the sampling design. I also predicted density for the national park. I overlaid the area with a grid with a cell size of 1 by 1 km and predicted the density for each cell using equation 2.2. From this predicted density map, it is possible to de-  - 25 -  rive estimates of density for a specific subzone or sector (Hedley et al. 2004; Royle et al. 2002). I calculated densities for the three survey zones and the national park. To convert dung densities to actual elephant densities, I used the following formula (Plumptre 2000): T=DP/Q  (2.6)  where T = density of elephants per km2; D = density of dung per km2; P = mean rate of dung decay; Q = mean number of dung piles produced per day (defecation rate). Logistical and financial constraints did not permit us to conduct dung decay or dung defecation rates in Odzala, so I used estimates from the literature with a defecation rate of 19.77 piles per day (CV 1.2%) (Tchamba 1992) and an average daily dung decay rate of 0.0233 (CV 9.44%) (Barnes et al. 1995). To obtain the total CV of animal density, CVs of both defecation and decay rate can be combined with the CV of dung density using the above described delta method.  2.3 Results 2.3.1 Law Enforcement Monitoring From May to August 1999, 92 observations of illegal human activities, of which 58 related to hunting, were recorded on 33 days of patrolling in the Odzala National Park. In Lossi, 39 days of reconnaissance patrols yielded 198 observations of human activities, of which 145 were related to hunting. Encounter rates of all indicators of hunting combined were also higher in Lossi that inside the national park (figure 2.3). The area over which these observations were made was in both sites roughly the same. In each site, patrols traversed 24 UTM squares of 5 km2 each, which amounts to 600 km2 per site. In the Lossi sanctuary, the minimum adequate GLM retained distance from the nearest village as the best predictor for the CPUE index of hunting (R2 = 0.398, F = 13.05, p  - 26 -  = 0.002) (figure 2.4a). More hunting occurred closer to human habitation in the northern part of the area. I excluded the village of Lossi from the model because this was a village with an active conservation programme and people were unlikely to hunt close to the village. Indeed, in a univariate model, the relationship between hunting and distance from the nearest village (which was Lossi) was not significant (R2 = 0.003, F = 1.14, p = 0.302). In the Odzala National Park, distance from the park boundary was the best predictor for the CPUE index of hunting (R2 = 0.273, F = 30.176, p < 0.001). There were fewer or no signs of hunting deeper inside the park. In Lossi, I also analysed encounter rates of elephant dung as a function of spatial variables. The minimum adequate GLM showed a significant inverse relationship between elephant dung abundance and distance from the nearest road (R2 = 0.163, F = 4.846, p = 0.043 ) (figure 2.4b). Not surprisingly, there was a significant negative correlation between elephant density and hunting (R2 = 0.394, F = 19.311, p <0.001 ) (figure 2.4c). These results indicated that distance from the nearest village, distance from the park boundary and distance from the nearest road would be good candidate variables to include in a spatial model of elephant abundance as proxies for human impact.  2.3.2 Elephant dung densities 2.3.2.1 Distance analysis I recorded 2014 elephant dung piles at the 44 sampling locations, 450 of which were on the transect portions of the recce-transects. Exploratory analysis in DISTANCE showed that truncation of the data was necessary and a cut-off at 400 cm on each side of the transect was found to be sufficient. This left 418 dung piles or 93 % of the transect observations in the dataset. I also found that grouping the data into 50 cm intervals provided the best option for modeling the detection function. I selected a half-normal key function with cosine adjustments as the final model. Estimated density of elephant dung for the whole area was 25.9 dung piles per hectare (CV 15.69%). The encounter rate was 9.5 dung piles per km (CV 14.49%). The effective strip width was 366 cm (CV  - 27 -  6.02%). Stratifying into the three survey strata or into forest/savannah did not produce any differences in the detection function. Therefore, I assumed that the same detection function could be applied to the entire area. Elephant dung counts at sampling locations showed a very skewed distribution as expected. Although the distribution was closer to a Poisson than a normal distribution (Appendix 2.3), it was still significantly different from a Poisson distribution and overdispersed (Kolmogrov-Smirnov test, ks=0.5227, p<0.001). The ratio estimator for calibrating recce encounter rates with encounter rates was close to one (1.02, CV = 7.79%) and encounter rates on transects were not significantly different from encounter rates on nearby calibration segments of recces (Wilcoxon signed rank test, Z = 1.4547, p = 0.1457).  2.3.2.2 Spatial modeling At the scale of the sampling location, the mean encounter rate of elephant dung was 11.28 dung piles/km in the national park and 0.97 dung piles/km outside the park. This difference was highly significant (Kruskal-Wallis, chi-squared = 18.4258, df = 1, p = 1.767e-05). Protection was the most important factor influencing elephant density (table  2.3). More elephant dung was found near patrol routes, even within the park only. For the entire area, interactions between park and forest clearings and between park and patrols were also significant. Because of these interactions, both park and forest clearing were also included in the model as main effects although individually their coefficients were not significant and they contributed very little to a reduction in deviance. Inside the park, more elephants were found closer to the nearest road but this was mainly observed in the southern part of the park where most patrols were also found. A conditional plot (figure 2.5) shows that more elephants occurred near roads where roads were close to patrol routes. Beyond 15km from patrol routes, this relationship showed signs of reversing, which may suggest that where protection was low, more elephants were found at greater distances from the nearest road. This was confirmed by dividing the data into a subset within 15 km from the nearest patrol route and a subset - 28 -  beyond 15 km. I then tested each subset for distance from the nearest road alone and found a negative relationship between dung abundance and distance to the nearest road in the first (F=6.0708, p=0.02067) and a positive relationship (F=17.973, p= 0.008174) in the second subset. Slope was another important variable. Elephants tended to avoid areas with steeper slopes, even if the difference in slope between sites was very modest. None of the vegetation types had a significant impact on elephants. In addition to the models that included vegetation with the other covariates, I also ran a model using vegetation only but, even here, I did not find a significant effect. At the finer scale of the sampling segment, minimum adequate GLMs retained distance to patrol routes and slope as the most significant variables (table 2.4). The analysis of the entire dataset showed that distance from the nearest road was a better predictor than distance from the park and the interaction between roads and patrol routes became apparent. In the park data subset, the interaction between distance from roads and patrol routes that was demonstrated at the scale of the sampling location was confirmed. Forest clearings and an interaction between forest clearings and patrols was also significant. Elephants appeared to be more abundant near forest clearings and patrol routes. Univariate GLMs confirmed that dung encounter rate was inversely correlated with both distance to the nearest forest clearing (F = 6.6746, p = 0.011) and distance to the nearest patrol route (F = 11.090, p = 0.001). There was some geographic overlap between the location of patrol routes and forest clearings, which may have confounded the relationships. A conditional plot (figure 2.6) of the effect of distance from forest clearings given different distances from the nearest road showed, however, that even further away from patrols, the correlation between elephants and forest clearings persisted. Scatterplots of the 4 main significant variables in relation to elephants are shown in figure 2.7. Minimum adequate GAMs (table 2.5) performed well with 72.4% of deviance explained for the whole dataset and 62.1% for the park dataset only. Estimated smooth - 29 -  plots of the covariates that were retained in the models are shown in Appendix 2.4. Distance from patrols, distance from park boundary and slope were all significant as was longitude which might have been a proxy for some unmeasured spatial covariate. Distance from the nearest forest clearing was not retained as a predictor in contrast with the GLMs. However, in the GLMs, it's main contribution was through an interaction with distance from patrols which was not modeled in the GAMs. GAMs performed better than GLMs (more deviance explained, lower GCV scores and higher adjusted R2 values, table 2.4 and 2.5). A Kruskal-Wallis test showed that vegetation as a categorical variable alone had a significant effect on elephant densities (Kruskal-Wallis chi-squared = 13.9283, df = 6, pvalue =0.03045). After aggregating vegetation classes, I ended up with 2 groups that were significantly different from each other (t=-3.233, p=0.0015) (figure 2.8) . Elephant densities were higher in open forest (Marantaceae forest, swamp forest) and old secondary forest and savannah (both wooded and grass). They were lower in mature closed forest and forest recolonising savannah. A GLM only containing these 2 vegetation groups confirmed that their effect on elephants was significantly different. However, when they were included as factors in a model together with the other covariates, they were not retained in the minimum adequate model. There was, however, a significant relationship between slope and vegetation. Mature closed forest was more often found on steeper slopes and the open forest types and savannah on flatter terrain (ANOVA: F=30.909, p= 1.440e-07).  2.3.2.3 Spatial autocorrelation Original dung encounter rates showed significant spatial autocorrelation at both the sampling location and segment levels (table 2.6). I found considerably less spatial autocorrelation in the residuals of the fitted values (table 2.7). At the sampling location level, most spatial autocorrelation was successfully removed by the modeling and Moran's I was not significantly different from the null model of no spatial autocorrelation. At the  - 30 -  segment level, there was still residual autocorrelation in all the models, expect for the GAM in the park (model MP3).  2.3.2.4 CV of encounter rates The CVs of modeled dung encounter rates (table 2.8) were smaller than the CVs of unmodelled encounter rates (which were 14.63% for the whole area, 13.16% for the park). Predicted CVs at the segment level were slightly better than predicted CVs at the sampling level.  2.3.2.5 Prediction of densities over a surface I used model MS2 to predict densities of elephants for all three surveyed blocks and the areas between them (figure 2.9). Other models produced similar results. From this model, I calculated predicted densities for the three survey blocks which were 3 elephants per km2 in the northern block, 3 elephants per km2 in the middle block and only 0.29 elephants per km2 in the southern block outside the reserve. From the predicted density map it can also be seen that densities were highest inside the park and that there was a decreasing gradient of density away from patrol routes. I used model MP2 to create a predicted density surface for the whole park (figure 2.10). From this model, I predicted a mean density of 1.75 elephants per km2 for the whole park which corresponded with 23703 animals (confidence interval: 17851 31115).  2.4 Discussion 2.4.1 Environmental and human influence on elephants The results of the analysis show that at all levels, variables related to protection were the most important factors determining distribution and abundance of elephants. The difference in dung densities between the Odzala National Park and the area outside the park was overwhelming. Elephants were 35 times more numerous in the protected part of the study area than in the unprotected part. Even within the park, the ac- 31 -  tively patrolled area seemed to determine elephant densities and counter a potential negative effect of proximity to villages or roads. During the 1980's, there was large scale poaching of megafauna in Odzala, and elephants especially were decimated. As recently as 1996, hundreds of elephant carcasses, of which at least 80 were recent, were found at the Mouandji clearing in the far northern part of the park near the Kuokoua river (Vanleeuwe et al. 1998). Active management and protection of the park only started in 1992 with the installation of the ECOFAC project by the European Union. ECOFAC invested in the development of infrastructure, research and gradually built up a surveillance force (Aveling 2001). Although it is possible that elephant populations were at similar levels before ECOFAC and although there haven't been any similar systematic surveys before this study, anecdotal information (Jean Marc Froment personal communication, Conrad Aveling personal communication) suggests that protection has been increasingly effective. If this is indeed the case, then it remains unknown whether elephants recovered locally and/or whether they moved from other hunted areas into a 'safe zone'. I did not find the same clear correlation between elephant densities and distance from the nearest road or village that others have found in Central Africa (Barnes et al. 1991; Fay and Agnagna 1991; Barnes et al. 1997; Laurance et al. 2006; Michelmore et al. 1994) or that elephant densities were higher in more remote areas (Hall et al. 1997). One explanation could be that the road running South - North, east of the park was only used on foot and not accessible to cars along the southern 2/3rd of the park boundary. This reduced access may have limited the degree of poaching on that side of the park. Nevertheless, the interaction with patrol routes in the park suggests that where there was less or no active protection, elephant densities did increase with distance from roads, including the ones only accessible on foot. The analysis of elephant dung encounter rates from the reconnaissance patrol near Lossi outside the park seemed to confirm this, as more elephant signs were found further away from the nearest road and village. At the same time there were more signs of hunting closer to villages. Not sur-  - 32 -  prisingly elephant dung encounter rates were inversely correlated with human signs. It is well known that roads provide access to hunters in Central Africa (Wilkie et al. 2000). Indeed, Barnes et al. (1995) found that elephants even avoided roads and villages when they were not hunted. I observed an inverse relationship between dung densities and distance to the nearest forest clearing, at the finer (1km) scale, within the park. Elephants have long been known to be attracted to salt licks to supplement their diet with minerals. They often have a shortage of sodium when eating fruits, bark or leaves, which prompts them to eat soil with high mineral content. This behaviour is known in both savannah (Holdo et al. 2002) and forest elephants (Turkalo and Fay 1995; Turkalo 1996; Blake and Inkamba-Nkulu 2004). Salt licks are social hotspots and in Odzala they can attract hundreds of individuals. Querouil et al. (1999) for example observed 3314 elephants over 8 months in Maya forest clearing in the northern part of the park and identified 629 elephants. Visits to these salt licks by elephant varies over time according to dietary needs and mineral availability (Blake and Inkamba-Nkulu 2004). As a result, some clearings become active and are enlarged by elephants clearing the vegetation and others are deserted and disappear slowly by overgrowing vegetation. In this study I did not differentiate between active and non-active clearings nor did I take seasonality into account, as I lacked sufficient detailed information, but even so, the effect of these clearings became apparent. Vegetation seemed to have some influence on elephant densities. At a finer scale in the park, densities were higher in more open 'secondary' forest (including Marantaceae, secondary and swamp forest) and in savannah than in mature closed forest. This is in line with what Barnes et al. (1991) and Merz (1986) found, that elephants prefer open secondary forest over closed primary forest. However, when combined with other covariates, vegetation wasn't retained in the final models, so its influence is unclear. There may be several reasons for this.  - 33 -  First, elephants respond to certain aspects of vegetation that require more spatial and temporal detail than what can be captured from satellite photos. Forest elephants eat a lot of fruit and its availability may be patchy and seasonal. For example, White (1994) found that elephants in Lope, Gabon moved to local stands of Sacoglottis gabonensis forest during its fruiting period. Short (1983) found similar correlations between patches of fruiting trees and elephants in Bia, Ghana. Blake and Inkamba-Nkulu (2004) also found higher densities of fruit trees near elephant-made trails, but pointed out that this resource is patchy in both space and time. Obtaining information on the spatial occurrence and densities of fruit trees and the period of fruiting is difficult and time-consuming, especially at the scale that this study was conducted on. Second, human factors may override the influence of vegetation. Thus Theuerkauf et al. (2001) and Fay and Agnagna (1991) found that human presence was the main determinant of spatial distribution of elephants and then vegetation structure. Third, slope was a strong determinant of elephant densities and it also correlated well with vegetation. Open forest and savannah were mainly associated with flatter terrain where more elephant dung was found than mature forest which occurred in steeper terrain. Could it be that slope was, at least partially, mimicking the effect of vegetation? To test this, I ran a multivariate model, including vegetation classes without slope, but vegetation still didn't come out as a significant variable in combination with the other covariates that remained in the model. It is very possible, if not likely, that broad vegetation patterns did influence spatial patterns of elephant abundance, but it remains unclear to what extent and how. Elephants may actively avoid slopes because of the energy cost associated in walking uphill. Wall et al. (2006) found an exponential decrease of elephant density with increasing slope. The biggest decrease occurred between 1 and 4 degrees, which is quite similar to what I observed. However, slopes in Odzala were generally very gentle and it seemed more likely that the effect of slope on elephant densities was at least partly due to its association with vegetation.  - 34 -  2.4.2 Scale and spatial autocorrelation I found differences in spatial autocorrelation depending on the scale at which I was modeling. Qi and Wu (1996) found that scale influences indices of spatial autocorrelation such as Moran's I. In their study, Moran's I decreased for biomass and elevation at increasing spatial scale. They argued that it is not useful to use measures of spatial autocorrelation at a single spatial scale. Although they were looking at immobile landscape variables, it may be more general and apply to mobile organisms as well. Beyers et al. (2001) showed that for the Odzala dataset, there was considerable spatial autocorrelation between short (200m) sections of the recce-transects up to 2 km, where it started to decline. This explains why Moran's I for unmodelled encounter rates at the segment level was much higher than at the sampling level. Spatial modeling at the coarser 5 km scale was more successful at removing spatial autocorrelation than at the finer 1.2 km scale of the segment level. This could indicate that most of the remaining spatial autocorrelation was due to finer scale, local variables. It is hard to know from the data whether this was due to unmodelled local environmental covariates (e.g. finer scale vegetation structure or composition) or because of some behavioural component of elephants. Elephants may prefer or avoid certain vegetation types indistinguishable from satellite images. Forest elephants are also known to use established trails ('elephant boulevards') for traveling and foraging (Vanleeuwe and Gautier-Hion 1998) and these trails may have crossed a recce or transect several times at short distances from each other. Survey teams even followed these trails on recces for short stretches, enhancing the chance of counting more dung on those stretches. Social behaviour may have also increased spatial autocorrelation of dung piles although group sizes of forest elephants are much smaller (2-3 individuals) than in Savannah elephants (White et al. 1993).  2.4.3 Coefficient of variation and implications for monitoring The CVs of modeled dung encounter estimates were in all cases considerably lower than the original CV of the unbiased estimate of transect data. This gain in precision  - 35 -  means that we would be able to detect a significant change in the population in a followup survey faster with a modeled than with a unbiased estimate. Following Walsh and White (1999) and Plumptre (2000), we can calculate the percent change that we can detect in population abundance between two surveys using the following equation: ˆ Δ ≈ 2.77CV ( D)  (2.7)  ˆ is the coefficient of variawhere Δ is percent change that can be detected and CV ( D) tion of the density or encounter estimate. This simple calculation assumes that: - there is the same amount of sampling effort for the 2 surveys, - the CV of the estimate is the same in both survey periods, and - the difference between the estimates is normally distributed. For example, if we take the CV of the encounter rate of the global model with the lowest CV (MS3, gam model at the segment level), which is 8.38%, we would be able to detect a significant change in dung population encounter rates of 23.21 %. Compare this to the CV of the unbiased dung encounter estimate of 14.63% where we would only detect a change if it would be greater than 40.52%. In other words, with an unbiased estimate, we might lose 40% of the population before we would detect a change versus 23% with the modeled estimate. This has important consequences for management. Managers would be able to pick up significant changes faster and respond more timely to mitigate population declines. To calculate the total CV of the density estimate using the delta method, we should include the CV of the detection function and the ratio estimator, as mentioned above. For example, the total CV of the modeled density estimate using model MS2, would then be 12.93% as opposed to 15.82% for the unbiased estimate. In terms of change detectable, this would amount to 35.81% for the modeled versus 43.82% for the unbiased estimate. The difference between the two is now much smaller than when we con- 36 -  sidered encounter rates only. This is partly due to the relative high CV of the ratio estimator (7.79%). However, since the ratio estimator was close to 1 and encounter rates on transects were not significantly different from encounter rates on nearby calibration segments of recces, we could assume encounter rates between transects and nearby recces to be the same and combine them without calibration. Total CV of the modeled density estimate without the ratio estimator is now considerably lower, 10.32% with a detectability of change of 28.58%.  2.4.4 Predicted elephant densities Maps showing predicted densities of elephants within the surveyed area are useful for wildlife managers (figure 2.9). They capture the essence of the relationships between elephant abundance and their environment. Managers can see where numbers of elephants are predicted to be the greatest and where elephants are threatened. Analysing and presenting data of repeated surveys would of course be even more informative, as it would show the changes in elephant populations and threats over time. The models can also predict elephant densities in areas beyond the survey zone (figure 2.10). However, predictions such as these should be considered with caution. Local unmeasured variables may influence population densities and will not be reflected in the results of the models. Outside the survey zone, predicted model parameters may attain values that are well outside their range in the original model and this may result in the prediction of unrealistic numbers. The predicted population density for Odzala National Park was 23703 elephants (confidence interval 17851 - 31115) with a mean density of 1.75 elephants per km2. Blake et al. (2007) reported a much lower population of 14000 animals (mean density of 1 elephant per km2) for the entire park in 2005, but this difference was largely due to differences in dung decay rates that were used to convert dung densities to actual elephant densities. I used a mean dung decay rate of 43 days as reported by Barnes et al. (1995) for an area in North East Gabon, which is the closest area to Odzala where dung decay studies were carried out. Blake et al. used a much lower mean dung decay rate of 90 days for all their surveys in Central Africa. - 37 -  Care should be taken when comparing absolute estimates of animal abundance based on indirect methods of population surveys such as dung counts as dung decay and defecation rates can be an important source of error (Barnes 2001). Assuming that decay rates were similar between the two surveys, comparing dung density estimates directly showed that my predicted model estimate of 15.85 dung piles per ha (confidence interval:12.87 - 19.49) was in fact very close to Blake et al.'s estimate of 17.75 dung piles per ha (confidence interval 13.93 - 22.61), (Steve Blake personal communication) and that there was no significant difference between the two surveys. This also suggests that the results of my model were realistic.  - 38 -  2.5 Tables Table 2.1. Candidate variables to be included in the spatial models. Variable  Code  Source  Indication  distance to the nearest road road outside the protected area  GPS waypoints field teams MIKE human access / ECOFAC and 1:200 000 topographic map, IGN  distance to the nearest "non-conservation" village  village  GPS waypoints field teams and human presence and 1:200 000 topographic map, IGN human access (Institut Geographique National, Paris)  distance to the park boundary  park  ECOFAC, ministère des Eaux et Forêts, Brazzaville, République du Congo  distance to the nearest patrol route  patrol  GPS waypoints ECOFAC Patrols protection  distance to the nearest Guard Post  post  GPS waypoints ECOFAC  protection  distance to park headquarters  infra  GPS waypoints ECOFAC  protection  SPOT image classification  habitat habitat  vegetation type  protection / human access  distance to the nearest forest clearing ("bai")  forest clearing  GPS waypoints ECOFAC  distance to the nearest river  river  1:200 000 topographic map, IGN habitat (Institut Geographique National, Paris)  slope  slope  Digital Elevation Model created from SRTM data (Shuttle Radar Topography Mission)  longitude  long  GPS waypoints field teams MIKE and 1:200 000 topographic map, IGN (Institut Geographique National, Paris)  latitude  lat  GPS waypoints field teams MIKE and 1:200 000 topographic map, IGN (Institut Geographique National, Paris)  - 39 -  habitat / topography  Table 2.2. Collected law enforcement data in the Odzala National Park in 1999 and in the Lossi area in 2000.  Year  Area patrolled  No of patrol days  Patrol months  No of Animal waypoints signs  Human signs  1999  Odzala National Park  38  April, June, July, August  5245  3760  97  2000  Lossi  40  April, May, June, July  5462  3627  200  Table 2.3. Fitted Generalized Linear Models of elephant dung at the sampling site level. The "+" sign denotes a positive correlation, the "-" sign denotes a negative correlation. Model  Model predictors  regres- residual F value sion deviance deviance  p-value  GCV  R-sq (adj)  Deviance explained  3368  1049  16.44  <0.0001  39.128  0.595  68.9%  1971  799.94  11.04  <0.0001  30.941  0.526  59.4%  Whole study area1 MS1  -patrol -slope +park +forest clearing +park:forest clearing -park:patrol  National Park only2 MP1  -patrol -road -slope +road:patrol  - 40 -  Table 2.4. Fitted Generalized Linear Models of elephant dung at the segment level. The "+" sign denotes a positive correlation, the "-" sign denotes a negative correlation. Model Model  regresresidual F value sion deviance deviance  p-value  GCV  R-sq (adj)  Deviance explained  16830  8400  32.517  <0.0001  66.806  0.416  50.1%  12077  6983  15.25  <0.0001  58.616  0.357  42.2%  Whole study area1 MS2  -patrol -road -slope +road:patrol  National Park only2 MP2  -patrol -road -slope +forest clearing +road:patrol -forest clearing:patrol  - 41 -  Table 2.5. Fitted Generalized Additive Models of elephant dung at the segment level. Model  Predictors  GCV  R-sq (adj)  Deviance explained  0.61  72.4%  0.52  62.1%  Whole area MS3  s(lat, 8.14) 43.89 s(long, 6.69) s(park, 8.93) s(patrol, 8.93) slope  National Park Only MP3  s(lat, 6,64) 49.97 s(long, 7.11) s(park, 3.76) s(patrol, 7.28) slope  - 42 -  Table 2.6. Moran's I (index of spatial autocorrelation) with associated z-statistic and pvalue of original dung encounter rates. Model spec.  Moran's I  Z  P  Entire area sampling location level  0.4866  3.88  1.043e-4  Park sampling location level  0.274  2.123  0.03374  Entire area segment level  0.7179  8.39  4.872e-17  Park segment level  0.5346  5.615  1.961e-8  Table 2.7. Moran's I (index of spatial autocorrelation) with associated z-statistic and pvalue of the residuals after modeling. Model spec.  Model  Moran's I  Z  P  Entire area sampling location MS1 level  -0.0346  -0.073  0.942  Park sampling location level  MP1  -0.0193  0.090  0.928  Entire area segment level  MS2  0.2695  3.122  0.002  Park segment level  MP2  0.2006  2.154  0.031  Entire area segment level (GAM)  MS3  -0.2928  -3.250  0.001  Park segment level (GAM)  MP3  -0.1916  -1.911  0.056  - 43 -  Table 2.8. Coefficient of variation (CV) of modeled encounter rates for fitted models. Model spec.  Model  CV  Entire area sampling location level  MS1  10.99  Park sampling location level  MP1  9.57  Whole area segment level  MS2  8.75  Park segment level  MP2  8.35  Entire area segment level (GAM)  MS3  8.38  Park segment level (GAM)  MP3  8.71  - 44 -  2.6 Figures Figure 2.1. Map of the Odzala National Park in the Republic of Congo and Survey blocks.  - 45 -  Figure 2.2. The design of the sampling unit in Odzala. The sampling unit consisted of 5 short transects of 200 m each (T1..T5) interspersed with 4 reconnaissance trails (recces) of 1 km each. The start of the transect ("Start") was located in the field using a GPS and the bearing of a compass was set in the direction of the predetermined end-point of the sampling unit ("Stop"). Transects were walked in a straight line, recces followed the "path of least resistance", but were not allowed to deviate more than 45 degrees from the predetermined compass bearing.  - 46 -  Figure 2.3. Encounter rates of all human activities (a) and hunting signs only (b) in Odzala National Park and Lossi. Encounter rates are given as number of observations per day. (a) Encounter rate per day of human signs in Lossi and Odzala Lossi Odzala NP  Hunting  Gathering  Fishing  Other  (b) Encounter rate per day of hunting signs in Lossi and Odzala Lossi Odzala NP  Camps  Hunters  Bullets  - 47 -  Line traps  Other traps  Figure 2.4. Catch per Unit Effort (CPUE) index (see text) for hunting signs in relation to distance from the nearest village (a) in Lossi. Elephant dung encounter rate in relation to distance from the nearest road (b) and Catch per Unit Effort for hunting signs (c) in Lossi.  25 20 15 10 5 0  Catch per unit effort hunting  30  (a)  15  20  25  30  Distance from the nearest village (km)  150 100 50 0  50  100  150  Elephant dung encounter rate (dung/km)  (c)  0  Elephant dung encounter rate (dung/km)  (b)  2  4  6  8  10  12  14  0  Distance from the nearest road (km)  500  1000  1500  2000  2500  Catch per unit effort hunting  - 48 -  3000  Figure 2.5. Conditioning plot of elephant dung encounter rate in relation to distance from the nearest road, given distance from the nearest patrol in the park. The horizontal bars in the top box represent different ranges for distances from the nearest patrol route for which a plot is given below showing elephant dung encounter versus distance from the nearest road (from bottom left to the upper right). Elephant densities were higher closer to a road where distance from the nearest patrol was small, but this relationship was reversed with increasing distance from the nearest patrol route. (a)  Elephant dung encounter rate (dung/km)  Given: distance from the nearest patrol route (km)  Distance from the nearest road (km)  - 49 -  Figure 2.6. Conditioning plot of elephant dung encounter rate in relation to distance from the nearest forest clearing given distance from the nearest patrol route. The horizontal bars in the top box represent different ranges for distances from the nearest patrol route for which a plot is given below showing elephant dung encounter versus distance from the nearest forest clearing (from bottom left to the upper right). Elephant densities are higher closer to a forest clearing at both smaller and bigger distances from the nearest patrol route. (b) Given: distance from the nearest patrol route (km) 5  6  8  20  10  25  2  4  6  8  10 60  4  15  60  • •• •  •  •• •• • •  • •• •• •  • • ••  • •  • •  •• •  •  •• •  •  • • •• ••  •  •  • •  •  • ••  • • • •  • • • •  •  •  •• • • • •  • •  • •  • •  •  • • • ••  • • •• •  • • • •• •  • • •• • •  • • •  •  • ••  •  •  40  • • •  20  • •  20  • • •• • ••  •  40  •  •  0  Elephant dung encounter rate (dung/km)  •  •  •  •  • • •  •  • • • • •• • •• • •  •  • •• •• •  •  •  • •• •  ••  •  • •  •  •  •• •• •  • •• • •  2  • • • • • • •• •  •  • • • • •  • • •  ••  •  4  6  •  • •  • • •  •  8  • •• • •  •• •• • ••  • • • •• • • •  • • •• •• ••  • • •  • ••  •  • •  • •  •  10  Distance from the nearest forest clearing (km)  - 50 -  0  2  10  Figure 2.7. Elephant dung encounter rates in relation to distance from the nearest patrol route (a), distance from the nearest road (b), slope (c) and distance from the nearest forest clearing (d) within Odzala National Park.  50 40 30 20 0  10  Elephant dung encounter rate (dung/km)  50 40 30 20 10 0  Elephant dung encounter rate (dung/km)  60  (b)  60  (a)  0  5  10  15  20  25  30  10  Distance from nearest patrol (km)  20  40  50  60  Distance from nearest road (km)  50 40 30 20 0  0  10  20  Mike2000_Surveys/OdzEle_2RS ~/Documents/_DATA_AFRICA/Site_Odzala/6_Analysis/  10  30  40  50  Elephant dung encounter rate (dung/km)  60  (d)  60  (c) Elephant dung encounter rate (dung/km)  30  1  2  3  4  5  6  2  4  6  8  10  Distance from the nearest forest clearing (km)  Slope (degrees)  - 51 -  Figure 2.8. Boxplots for all types of vegetation and aggegated classes in relation to elephant dung encounter rates. CHS = 'Marantaceae forest' (canopy cover of less than 40%), DHS = 'classical' dense mixed-species rainforest, FM = 'swamp' forest, FR = forest colonising savanna within savanna blocks, SA = old secondary forest, SB = wooded savannah, SH= grass and  60  60  shrub savannah.  ●  50  50  ● ●  40  40  ●  30  ●  0  0  10  10  20  20  30  ● ● ●  CHS DHS  FM  FR  SA  SB  SH  DHS_FR  Boxplot all vegetation classes  SB_SH_CHS_SA_FM  Boxplot aggregated vegetation classes  - 52 -  Figure 2.9. Predicted density map for elephants (animals per km2) for the entire study area using model MS2.  - 53 -  Figure 2.10. Predicted density map for elephants (animals per km2) for the National Park area using model MP2.  - 54 -  2.7 References Anonymous. 1998. S-plus for ArcView GIS. User's Guide. Version 1.1. Mathsoft. Data Analysis Division, Mathsoft, Inc. Seattle, Washington. Arcese, P., Hando, J. and L. Campbell. 1995. Historical and present-day antipoaching efforts in Serengeti. Pages 506-533 in Sinclair, A.R.E. and P. Arcese (editors). Serengeti II, Dynamics, Management, and Conservation of an Ecosystem. The University of Chicago Press. Augustin, N.H, Mugglestone, and S.T. Buckland. 1998. The role of simulation in modelling spatially correlated data. Environmetrics, 9:175-196. Aveling, C. 2001. L'extension du parc national d'Odzala. Une opportunité de développement. Canopee 20: 13-15. Barnes, R. F. W. 2001. How reliable are dung counts for estimating elephant numbers? African Journal of Ecology, 39:1-9. Barnes, R. F. W., A. Blom, and M. P. T. Alers. 1995. A review of the status of forest elephants Loxodonta africana in Central Africa. Biological Conservation 71: 125-132. Barnes, R. F. W., K. L. Barnes, M. P. T. Alers, and A. Blom. 1991. Man determines the distribution of elephants in the rain forests of northeastern Gabon. African Journal of Ecology 29: 54-65. Barnes, R. F. W., A. Blom, M. P. T. Alers, and K. L. Barnes. 1995. An estimate of the numbers of forest elephants in Gabon. Journal of Tropical Ecology 11: 27-37. Barnes, R. F. W., K. Beardsley, F. Michelmore, K. L. Barnes, M. P. T. Alers, and A. Blom. 1997. Estimating forest elephant numbers with dung counts and a geographic information system. Journal of Wildlife Management 61: 1384-1393.  - 55 -  Barnes, R. F. W., and K. L. Jensen. 1987. How to count elephants in forests. IUCN African Elephant & Rhino Specialist Group Technical Bulletin 1, 1-6 Bermejo, M. 2004. Home-range use and intergroup encounters in western gorillas (Gorilla G. Gorilla) at Lossi Forest, North Congo. American Journal of Primatology 64: 223-232. Bermejo, M. 1999. Status and conservation of primates in Odzala National Park, Republic of the Congo. Oryx 33: 323-331. Bermejo, M., J. D. Rodriguez-Teijeiro, G. Illera, A. Barroso, C. Vila, and P. D. Walsh. 2006. Ebola outbreak killed 5000 gorillas. Science 314: 1564. Beyers, R., L. Thomas, J. Hart, and S. Buckland. 2001. Monitoring Illegal Killing of Elephants (MIKE) Central African Pilot Project. Technical Report No 2. Recommendations for Ground-Based Survey Methods for Elephants in the Central African Forest Region. Blake, S., and C. Inkamba-Nkulu. 2004. Fruit, minerals, and forest elephant trails: Do all roads lead to Rome? Biotropica 36: 392-401. Blake, S., S. Strindberg, P. Boudjan, C. Makombo, I. Bila-Isia, O. Ilambu, F. Grossman, L. Bene-Bene, B. de Semboli, V. Mbenzo, D. S'hwa, R. Bayogo, L. Williamson, M. Fay, J. Hart, and F. Maisels. 2007. Forest elephant crisis in the Congo Basin. PLOS Biology 5(4): 0001-0009. Buckland, S. T., K. P. Burnham, and N. H. Augustin. 1997. Model selection: an integral part of inference. Biometrics 53: 603-618. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to Distance Sampling. Estimating abundance of biological populations. Oxford University Press. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. B. Laake, D.L., and L. Thomas. 2004. Advanced distance sampling. Oxford University Press.  - 56 -  Crawley, M. J. 1993. GLIM for ecologists. Blackwell Scientific Publications. Crawley, M. J. 2002. Statistical Computing. An Introduction to Data Analysis using S-Plus. John Wiley & Sons. Cumberworth, S. L., S. T. Buckland, and D. L. Borchers. 1996. A spatial modelling approach for the analysis of line transect data. Paper SC/48/012 presented to the IWC Scientific Committee, June 1996 (unpublished) Draulans, D., and E. Van Krunkelsven. 2002. The impact of war on forest areas in the Democratic Republic of Congo. Oryx 36: 35-40. Dudley, J. P., J. R. Ginsberg, A. J. Plumptre, J. A. Hart, and L. C. Campos. 2002. Effects of war and civil strife on wildlife and wildlife habitats. Conservation Biology 16: 319-329. Efron, B., and R. Tibshirani. 1993. An Introduction to the Bootstrap. Chapman and Hall, New York, London. Fa, J. E., S. F. Ryan, and D. J. Bell. 2005. Hunting vulnerability, ecological characteristics and harvest rates of bushmeat species in afrotropical forests. Biological Conservation 121: 167-176. Fay, J. M. 1991. An elephant (Loxodonta africana) survey using dung counts in the forests of the Central African Republic. Journal of Tropical Ecology 7: 25-36. Fay, J. M., and M. Agnagna. 1991. A population survey of elephants in Northern Congo. African Journal of Ecology 29: 177-187. Gatti, S., F. Levrero, N. Menard, and A. Gautier-Hion. 2004. Population and group structure of western lowland gorillas (Gorilla gorilla gorilla) at Lokoue, Republic of Congo. Primatology 63: 111-123. Guisan, A., E. Jr,T.C, and T. Hastie. 2002. Generalized linear and generalized additive models in studies of species distributions: setting the scene. Ecological Modelling 157: 89-100. - 57 -  Hall, J. S., B.-I. Inogwabini, E. A. Williamson, I. Omari, C. Sikubwabo, and L. J. T. White. 1997. A survey of elephants (Loxodonta africana) in the Kahuzi-Biega National Park lowland sector and adjacent forest in eastern Zaire. African Journal of Ecology 35: 213-223. Hastie, T. J., and R. J. Tibshirani. 1990. Generalized Additive Models. Monographs on statistics and applied probability, Chapman & Hall. Hedley, S. L., S. T. Buckland, and D. L. Borchers. 1999. Spatial modelling from line transect data. Journal of Cetacean Research & Management 1: 255-264. Hedley, S. L., S. T. Buckland, and D. L. Borchers. 2004. Spatial distance sampling methods. In: Advanced Distance Sampling (Buckland et al.), Oxford University Press. Hilborn, R., Arcese, P., Borner, M., Hando, J., Hopcraft, G., Loibooki, M., Mduma, S. and A.R.E. Sinclair. 2006. Effective enforcement in a conservation area. Science, 314 (5803):1266-1270. Holdo, R. M., J. P. Dudley, and L. R. McDowell. 2002. Geophagy in the African elephant in relation to availability of dietary sodium. Journal of Mammalogy 83: 652-664. Jachmann, H. 2001. Estimating Abundance of African Wildlife, an aid to adaptive management. Kluwer. Jachmann, H., and R. H. V. Bell. 1984. The use of elephant droppings in assessing numbers, occupancy and age structure. A refinement of the the method. African Journal of Ecology 22: 127-142. Jachman, H. and R. Beyers. 2003. MIKE Law Enforcement Analysis Strategy at the Site Level. Unpublished report, CITES - MIKE programme. Jones, M. T., G. J. Niemi, J. M. Hanowski, and R. R. Regal. 2002. Poisson regression: a better approach to modeling abundance data? In Predicting species occur-  - 58 -  rences: issues of accuracies and scale ed. Scott, J.M., Heglund, P.J., Morrison, M.L., Haufler, J.B., Raphael, M.G., Wall, W.A., Samson, F.B. 411-418. Island Press. Kitron, U., and J. J. Kazmierczak. 1997. Spatial analysis of the distribution of Lyme disease in Wisconsin. American Journal of Epidemiology 145: 558-566. Koster, S. H., and J. A. Hart. 1988. Methods of estimating ungulate populations in tropical forests. African Journal of Ecology 26: 117-126. Laurance, W. F., B. M. Croes, L. Tchignoumba, S. A. Lahm, A. Alonso, M. E. Lee, P. Campbell, and C. Ondzeano. 2006. Impacts of roads and hunting on Central African rainforest mammals. Conservation Biology 20: 1251-1261. Legendre, P. 1993. Spatial autocorrelation: trouble or new paradigm? Ecology 74: 1659-1673. Lejoly, J. 1996. Synthese regionale sur la biodiversite vegetale des ligneux dans les 6 sites du projet ecofac en afrique centrale. Project Ecofac - cellule de coordination. Maisels, F.G, V. Mbolo, S. Gulick, M. Fay and K. Peirsman. 1997. Cartographie de la végétation du Parc National d'Odzala. Projet ECOFAC - Composante Congo. AGRECO-CTFT. Mc Cullagh, J. A. N. 1989. Generalized linear models. Chapman & Hall. Merz, D. 1986. Counting elephants (Loxodonta africana cyclotis) in tropical rain forests with particular reference to Tai National Park, Ivory Coast. African Journal of Ecology 24: 61-68. Michelmore, F., K. Beardsley, R.-F.-W. Barnes, and I. Douglas-Hamilton. 1994. A model illustrating the changes in forest elephant numbers caused by poaching. African Journal of Ecology 32: 89-99. Milner-Gulland, E. J., and E. L. Bennett. 2003. Wild meat: the bigger picture. Trends in Ecology and Evolution 18: 351-357.  - 59 -  Plumptre, A. J. 2000. Monitoring mammal populations with line transect techniques in African forests. Journal of Applied Ecology 37: 356-368. Potts, J. M. and J. Elith. 2006. Comparing species abundance models. Ecological Modelling, 199:153-163. Qi, Y., and J. G. Wu. 1996. Effects of changing spatial resolution on the results of landscape pattern analysis using spatial autocorrelation indices. Landscape Ecology 11: 1-64. Querouil, S., F. Magliocca, and A. Gautier-Hion. 1999. Structure of population, grouping patterns and density of forest elephants in north-west Congo. African Journal of Ecology 37: 161-167. Royle, A. J., W. A. Link, and J. R. Sauer. 2002. Statistical mapping of count survey data. 625-638 pages. Seber, G. A. F. 1982. The estimation of animal abundance and related parameters. pages. Short, J. C. 1983. Density and seasonal movements of forest elephant (Loxodonta africana cyclotis) in Bia National Park. African Journal of Ecology 21: 175-184. Tchamba, M. N. 1992. Defaecation rates by the African forest elephant (Loxodonta Africana Cyclotis) in the Santchou Reserve, Cameroon. Mammalia 56: 155-158. Theuerkauf, J., H. Ellenberg, W. E. Waitkuwait, and a. M. Mühlenberg. 2001. Forest elephant distribution and habitat use in the BossematiéForest Reserve, Ivory Coast. Pachyderm 30: 37-43. Thomas, L., S. T. Buckland, K. P. Burnham, D. R. Anderson, J. L. Laake, D. L. Borchers, and S. Strindberg. 2002. Distance sampling. pp 544-552 in Encyclopedia of Environmetrics, Edited byAbdel H. El-Shaarawi and Walter W. Piegorsch John Wiley & Sons, Ltd, Chichester.  - 60 -  Turkalo, A., and J. M. Fay. 1995. Studying forest elephants by direct observation: preliminary results from the Dzanga clearing, Central African Republic. Pachyderm 20: 45-54. Turkalo, A. 1996. Studying forest elephants by direct observation in the Dzanga clearing: an update. Pachyderm 22: 59-60. Vanleeuwe, H., S. Cajani, and A. Gauthier-Hion. 1998. Forest clearings and the conservation of elephants in the Northeast of Congo republic. Pachyderm 24: 46-52. Vanleeuwe, H., and A. Gautier-Hion. 1998. Forest elephant paths and movements at the Odzala National Park, Congo: The role of clearings and Marantaceae forests. African Journal of Ecology 36: 174-182. Vincent, P. J., and J. M. Haworth. 1983. Poisson regression models of species abundance. Journal of Biogeography 10: 153-160. Wall, J., I. Douglas-Hamilton, and F. Vollrath. 2006. Elephants avoid costly mountaineering. Current Biology 16: R527-R529. Walsh, P. D., K. A. Abernethy, M. Bermejo, R. Beyers, P. De Wachter, M. E. Akou, B. Huijbregts, D. I. Mambounga, A. K. Toham, A. M. Kilbourn, S. A. Lahm, S. Latour, F. Maisels, C. Mbina, Y. Mihindou, S. N. Obiang, E. N. Effa, M. P. Starkey, P. Telfer, M. Thibault, C. E. G. Tutin, L. J. T. White, and D. S. Wilkie. 2003. Catastrophic ape decline in western equatorial Africa. Nature 422: 611-614. Walsh, P. D., and L. J. T. White. 1999. What it will take to monitor forest elephant populations. Conservation Biology 13: 1194-1202. Walsh, P. D., L. J. T. White, C. Mbina, D. Idiata, Y. Mihindou, F. Maisels, and M. Thibault. 2001. Estimates of forest elephant abundance: Projecting the relationship between precision and effort. Journal of Applied Ecology 38: 217-228.  - 61 -  White, L. J. T., C. E. G. Tutin, and M. Fernandez. 1993. Group composition and diet of forest elephants, Loxodonta africana cyclotis Matschie 1900, in the Lope Reserve, Gabon. African Journal of Ecology 31: 181-199. White, W. T. 1994. Sacoglottis gabonensis fruiting and the seasonal movements of elephants in the Lopé Reserve, Gabon. Journal of Tropical Ecology 10: 121-125. Wilkie, D., E. Shaw, F. Rotberg, G. Morelli, and P. Auzel. 2000. Roads, development, and conservation in the Congo basin. Conservation Biology 14: 1614-1622. Wilkie, D. S., and J. F. Carpenter. 1999. Bushmeat hunting in the Congo Basin: an assessment of impacts and options for mitigation. Biodiversity & Conservation 8: 927-955. Williams, R., S. L. Hedley, and P. S. Hammond. 2006. Modeling distribution and abundance of Antarctic baleen whales using ships of opportunity. Ecology & Society 11:1 Wood, S. 2001. mgcv: GAMs and Generalized Ridge Regression for R. R News 1/2: 20-26.  - 62 -  CHAPTER 3. Spatio-temporal modeling of wildlife abundance in the context of a war in the Okapi Faunal Reserve, DRC 1 3.1 Introduction Since 1996, a devastating civil war has been raging through the Democratic Republic of Congo (DRC), spilling over from the genocide in Rwanda in 1994. The war affected 50 million people and close to 4 million have died, most of them indirectly through starvation and disease (International Rescue Committee, http://www.theirc.org/). Many different ethnic groups, the army, rebel factions and other militias, and 6 countries became involved, making it the largest regional war in modern African history (Global Security, http://www.globalsecurity.org/military/world/war/congo.htm). The desire to exploit the great wealth of mineral resources in Congo contributed substantially to this conflict and to the involvement of so many parties, which also included private overseas companies that helped finance the war (United Nations 2002). Despite the relative stability in most of the country, ensuing from a peace agreement in 2002 and the withdrawal of foreign troops, instability persisted in the Eastern Congo and military clashes and massacres continued up to the time of this writing. The war caused immense human suffering but had serious ecological consequences as well. Habitat degradation and major declines in wildlife have been documented in Virunga National Park (Kalpers 2001), Kahuzi-Biega National Park (Hart et al. 2007) and Garamba National Park (Hart and Mwinyihali 2001) in Eastern DRC. In this study, I examined the effects of the war on wildlife species in the Okapi Faunal Reserve (Réserve de Faune d’Okapi, RFO), which is also located in Eastern Congo. First, I predict an overall decline in populations of elephants, okapi and forest antelopes (duikers) as a result of the conflict. I used dung counts of these species as indi-  1. A version of this chapter will be submitted for publication. Beyers, R., Hart, J., Grossmann, F., Sinclair, A.R.E. 2008. The impact of the civil war (1996-2005) on elephants and other ungulates in the Okapi Faunal Reserve in the Democratic Republic of Congo. - 63 -  cators of their population densities and compared dung densities before and after the war. I also expect declines to be different in areas with different levels of security during the war. I compared changes in densities in two zones, a Green Zone which was relatively safe and accessible to park staff and biological monitoring teams during most this period and a Red Zone which was inaccessible and dangerous from the start of the conflict in 1996 until late 2006. The RFO, as a forest reserve, is unique in DRC or even in the whole Congo Basin, in that wildlife surveys have been carried out in two different time periods (before and after the war) using exactly the same transect sampling design. This allowed comparisons of pre- and post conflict population densities without a design bias. Second, I predict declines to be greater in those areas where initial densities of widlife species were higher, because poachers tend to seek out areas with higher wildlife densities to maximize their return (Marks 1996). I examined proportional declines of dung densities across all sampling locations in the reserve to test this prediction. Third, I predict spatial patterns of wildlife densities and changes in densities to be predominantly influenced by humans. I expect densities to be lower in areas that were more accessible or closer to human settlements and I expect this avoidance of humans to be greater after the war. I developed spatial models using distance from the nearest road as a proxy for human access. I used covariates such as distance from human settlements and distance to and the size of deforested areas as proxies for human presence. I also expect densities to be higher and declines to be lower in areas that were better protected. As indicators of protection, I included covariates such as distance from the reserve boundary and distance from guard posts. I expect habitat to be a less important predictor in determining spatial patterns of abundance. Indeed, published work elsewhere in Central Africa shows that humans have overwhelming effects on the spatial distributions of wildlife populations. Distribution and abundance of elephants have been correlated with distance from roads and human settlements as proxies for human  - 64 -  impact (Buij et al. 2007, Blake et al. 2007, Barnes et al. 1997, Barnes et al. 1995). Human related variables were often more important than habitat (Chapter 2, Barnes et al. 1991) or preferred food availability (Buij et al. 2007). Similar relationships have been observed for other species in Central Africa (Laurance et al. 2006, Lahm et al. 1998, Blom et al. 2005). To examine influences of humans and habitat on spatial patterns of dung densities, I employed spatial modeling techniques using line transect density estimates. These techniques have been successfully used in marine animal surveys (Swartzman et al. 1992, Hedley et al. 2004) and elephant surveys (Chapter 2, Blake et al. 2007). The line transect survey is the most commonly used method for counting forest mammals in Central Africa (Plumptre 2000, Barnes et al. 1995). This type of survey usually adheres to a random or systematic design and provides an estimate of abundance in a fixed study area or stratum. Spatial modeling, however, does not require a strictly random or systematic design. Indeed, these methods were developed in situations where it was difficult to implement any kind of randomized design and where "platforms of opportunity" such as ferries, merchant navy vessels, etc. were used to asses populations of large marine mammals using non-random line transect data (Buckland et al. 2004). Few studies have yet tried to quantify trends in populations of wildlife in Central Africa and what is available does not always provide detailed information or does not allow precise estimates of change (Blanc et al., 2007). In particular, no studies have quantified temporal trends in abundance of multiple species in relation to potential drivers of trends that vary spatially. And very few studies in Central Africa have quantified spatial patterns of more than one species simultaneously (Laurance et al. 2006, Blom et al. 2005, De Merode et al. 2000).  - 65 -  3.2 Methods 3.2.1 Study area The Okapi Faunal Reserve is located in the Ituri forest in North-Eastern Congo between 1º and 2º 30' N and 27º 30' and 29º 30' E and has a surface area of 13700 km2 (figure 3.1). It belongs to the North Eastern Congolian forest block (Wilkie et al. 1988). More than 90% of the reserve is covered by dense tropical forest consisting of either humid mixed evergreen forest with dominant canopy species such as Cynometra alexandri, Julbernardia seretii (Hart et al. 1996) and Brachystegia laurentii, or monodominant forest dominated by almost pure stands of Gilbertiodendron dewevrei (called Mbau). Mbau occurs in small patches to large blocks of several tens of square kilometers (Hart et al. 1989) and is found mainly in the southern and western parts of the reserve. Besides these two main vegetation types, there are smaller patches of other types such as swamp forest along rivers where there is poor drainage, and drier forest and shrub on granite inselbergs in the northern part of the reserve (Hart et al. 1994). In the north, the forest borders a mosaic of forest-savannah. Secondary forest occurs mainly on abandoned agricultural clearings that were made by shifting cultivators in the vicinity of settlements. Much of the secondary forest along the road is heavily dominated by Musanga cecropioides (Thomas 1991). Elevation in the reserve ranges from 500m to 1500m. Rainfall is high with an annual mean of 1700 mm recorded between 1987 and 1993. There is a distinct dry season from January to March and, outside this period, rainfall is variable throughout the year (Hart et al. 1994). A national road (RN4), connecting eastern to western Congo, bisects the southern half of the reserve. A second road runs along the eastern boundary of the reserve and a third one is at some distance from its western boundary (figure 3.1). Most stretches of these roads have been barely maintained since the 1960's and are seriously degraded. Along them are numerous, small villages where people live on subsistence agriculture and hunting. Around these villages, small fields are cleared in secondary forest (Wilkie - 66 -  et al. 1998, Mulley 2004). Four larger towns exist around the reserve (Mambasa, Niania, Wamba and Mungbere). The reserve headquarters, at Epulu, is located along the national road in the central part of the reserve. Besides these settled populations, the forest harbors some of the few remaining hunter-gatherer groups (Mbuti and Efe) in the world. These people hunt duikers (forest antelopes), primates and rodents, and also gather medicinal plants, tubers and other forest products (Wilkie and Finn 1988). The Okapi Faunal Reserve was created in May 1992 and was recognized as a World Heritage Site in December 1996. It contains possibly one of the largest remaining elephant populations in the Congo and numerous other forest mammals including Okapi (Okapi johnstoni), which is endemic to DR Congo (Mubalama and Mapilanga 2001).  3.2.2 Mammal surveys I compared data from two surveys, one from before the conflict and one from after the conflict. The pre-war survey (further referred to as the “1995 survey”) covered the whole area of 13700 km2 and was carried out during several months between March 1994 and November 1995 by local field teams led by Dr. John Hart. The post-war survey (the “2006 survey”) was conducted under the WCS-Congo (Wildlife Conservation Society Congo) Inventory and Monitoring Unit program. This survey was carried by teams supervised by Falk Grossman and Dr. John Hart. I was involved in the survey design and the training of some of the team leaders in wildlife sampling techniques during a training workshop in the Nouabale Ndoki National Park in the Republic of Congo from September to November 1999. The survey was performed in two stages. An area of 5600 km2 in the centre of the reserve was surveyed from April to June 2005. At that time, only this area (referred to as the “Green Zone”) was secure and accessible to wildlife staff and monitoring teams. Reserve headquarters at Epulu and two scientific study areas, Edoro and Lenda were located within this zone (figure 3.1). As military forces began to withdraw, the remainder of the reserve (the “Red Zone”) became accessible again and was surveyed between November 2006 and May 2007.  - 67 -  Another survey (the “2000 survey”) was done in the middle of the war during 12 months between May 2000 and October 2001 under the CITES MIKE (Monitoring Illegal Killing of Elephants) program. I did not use the data of this survey because they only covered a limited area of the reserve. Estimates of elephant densities for this sub-zone are presented in Beyers et al. (2001). Both the 1995 and 2006 surveys had the same basic design and used the same sampling locations. Original sampling locations in 1994-1995 were laid out on a map and located in the field (figure 3.1). The design was not entirely random because it was often difficult to find the exact position in the field due to the lack of good maps, satellite images and GPS, which were not available in 1995. The actual sampling locations were well marked during the first survey and, with the help of the original pygmy trackers, it was possible to relocate almost all transects and survey locations in subsequent surveys. Coordinates were then recorded using GPS. At each location, one to a maximum of four transects were established in different compass directions, all starting from the same departure point. The minimum angle between 2 transects was 45 degrees. The length of each transect was between 2.5 to 5 km. An overview of the number of locations and transects for each survey is given in table 3.1. Standard line transect methodology was employed in order to record all observations of mammals or their signs such as dung and nests (Buckland et al. 1993, White and Edwards 2000). A straight line was cut through the forest following a fixed compass bearing. Observers walked slowly on the transect line and used a hip-chain to measure the distance traveled. When an animal or an animal sign was detected at any distance from the transect line, the perpendicular distance from the transect line to each observation was measured to the nearest cm. The surveys were conducted by several teams, each team consisting of 5-6 people who received similar training in wildlife survey and transect methodology. Several observers who took part in the pre-war surveys also participated in the post-war surveys.  - 68 -  Perpendicular distances were measured to dung of the following species: forest elephant (Loxodonta african cyclotis), buffalo (Syncerus caffer), okapi (Okapia johnstoni), blue duiker (Cephalophus monticola), Bates’ antelope (Neotragus batesi), bay duiker (Cephalophus dorsalis), white-bellied duiker (Cephalophus leucogaster), black-fronted duiker (Cephalophus nigrifrons), Weyn’s duiker (Cephalophus weynsi), water chevrotain (Hyemoschus aquaticus) and yellow-backed duiker (Cephalophus silvicultor). Because it is difficult to distinguish between droppings of different forest antelopes, observations were grouped in three classes according to weight: “small” duikers (blue duiker (4.7 kg), Bates’ antelope (2.25 kg)); “red” duikers (bay (22 kg), white-bellied (16.7 kg), blackfronted (13.9 kg), Weyn’s duiker (17.7 kg), water chevrotain (11.2kg)) and yellowbacked duiker (68 kg) (Hart 2000, Kingdon 1997). I did not analyze buffalo dung data because the small sample sizes (50 dung piles in 1995 and 17 piles in 2006) made distance and spatial modeling not feasible.  3.2.3 Spatial covariates 3.2.3.1 Human-related covariates I predicted that patterns of wildlife densities would be correlated with spatial covariates associated with humans and habitat (table 3.2). I used proxies for human access (roads), human presence (all human settlements, major towns, deforestation), protection (park boundary, guard posts), security during the war (security zones) and habitat (slope and ecozones). Formulating a-priori hypotheses guided the selection of covariates that I included in the spatial models. Landsat images were georeferenced using GPS waypoints (Beyers and Hart 2001). They were subsequently used as a background to digitize roads and rivers (work performed by the Department of Geography, University of Ghent (http://geoweb.ugent.be/ sygiap/)). Geographic coordinates for villages and guard posts were obtained using GPS in the field (Beyers and Hart 2001). Guard posts were different in 1995 and 2006. In 1995 there was only one active guard post in Epulu where the reserve headquarters  - 69 -  was located. In 2006 there were 6, 5 of them positioned along the road going through the reserve. Reserve boundaries were taken from the database of the Central African Regional Program for the Environment (CARPE, http://carpe.umd.edu/) and adjusted to the natural (rivers) and human-made (roads) limits that defined the reserve. I measured the shortest distance from each geographic linear or point feature to the mid-point of each transect using ArcMap 9.2 (ESRI Inc.). These values were subsequently used in the spatial models. Because human demographic data were not available for all of the survey periods, I used a composite “deforestation index” of non-forested land (mostly agriculture and urban development) and distance to non-forested land as an indicator for the extent and intensity of human activity. De Merode et al. (2000) also used agricultural land as a proxy for human presence in Garamba National Park and they calculated NDVI values from Landsat TM images to classify agriculture. I obtained forest cover maps from the Carpe Decadal Forest Change Mapping project (CARPE Decadal Forest Change Mapping (DFCM) Project, http://carpe.umd.edu/resources/dfcm) and South Dakota State University through Erik Lindquist. These maps show per-pixel estimates of forest likelihood in 1990 (time 1) and in 2000 (time 2) at a spatial resolution of 60m, derived from Landsat and Modis satellite images. I constructed a grid with a grid cell size of a quarter degree and measured the area of non-forested land at time 1 and time 2 in each cell. This allowed me to quantify the amount of deforestation from 1990 to 2000 per grid cell. I clipped the grid to an area with a buffer zone of 15 km around the reserve (Appendix 3.2). I created two grids, one including the reserve and the buffer zone and one with the buffer zone but excluding the reserve. For each time period, I calculated a composite index representing the ‘deforestation environment’ at each transect based on distance from the transect to each cell of the grid and amount of deforestation in each cell (equation 3.1).  - 70 -  1  It = ∑ ai i=1  1 di  (3.1)  where It is the index at time t, ai the extent of agriculture and other non-forested land in grid cell i, di the distance from the middle of the transect to grid cell i. A high index represented large areas of deforestation (deforestation ‘hotspots’) close to the transect. I hypothesized that higher densities of animals were correlated with a low index representing less human activity and vice versa. In the GAM models I used the grid excluding the cels within the reserve because the index based on this grid was less correlated with other covariates in the model (especially distance from the nearest road) than the index based on the grid that included the reserve and because some areas in the northern part of the reserve were classified as non-forest while this was not anthropogenic.  3.2.3.2 Habitat and slope I classified habitat into the following ‘ecozone’ categories: mixed hill forest, rocky outcrops (inselbergs), savannah-forest ecotone, mixed forest, mono-dominant forest consisting of Gilbertiodendron dewrevei, swamp forest and non-forested area (Appendix 3.3). A classification of 2 sets of Landsat images (1994 and 2002) for vegetation did not produce satisfactory results because several bands (1,2 and 3) of the acquired images contained very little variation in their reflectance, possibly due to data corruption. Another reason was that we had to stop an attempt to ground-truth the vegetation prematurely because of an intensification of the war. Hill forest habitat types and savanna-forest ecotones were digitized directly from the Landsat images by Dr. John Hart who had an intimate knowledge of the area. Nonforested land was obtained from the above mentioned Decadal Forest Change Mapping. The remaining area was considered forest and classified as either mixed, monodominant or swamp forest on a grid basis using proportions of the different vegetation types found on the survey transects. I calculated proportions of each vegetation type for each quarter degree grid cell of the park. - 71 -  I constructed a digital elevation model (30m resolution) using data produced by the Space Shuttle Radar Topography Mission (SRT Seamless Data Distribution System, Earth Resources Observation and Science (EROS), http://seamless.usgs.gov). I derived slope from this model and calculated average slope in a strip 90m wide along each transect line using ArcMap 9.2 and Spatial Analyst (ESRI Inc.).  3.2.4 Data analysis 3.2.4.1 Estimates of densities and changes in densities I estimated densities of elephants, okapi and duikers for all surveys using the software program DISTANCE 5.0 (Research Unit for Wildlife Population Assessment, http:/ /www.ruwpa.st-and.ac.uk/distance/). A short introduction to Distance or line transect sampling is given by Thomas et al. (2002), while Buckland et al. (2001) describe it in detail. The underlying assumption of Distance sampling is that all objects on the transect line are detected, but that the probability of detection decreases with increasing distance from the line. This probability is mathematically modeled as the detection function. After fitting a detection model to the data, the area under the curve is calculated. This is called the effective half-strip width and corresponds with the width of a strip if all objects would have been detected within that strip. Since it applies to one side of the transect line only, the half-strip width should be multiplied by 2 to get the 'Effective Strip Width' (ESW) which covers both sides of the transect. Object density ( Dˆ ) is calculated from the number of detected objects ( n ) divided by the length of the transects ( L ) and the effective half-strip width ( ESW ): Dˆ =  n ESW * L  (3.2)  It is usually necessary to truncate the data at a certain perpendicular distance from the transect line to obtain a good model of the detection function. This distance is called the “truncation distance”.  - 72 -  To model the detection function, I used data from all locations at each time period. I explored several models that were available in DISTANCE and selected the model with the best fit using Akaike’s Information Criterion (AIC) (Appendix 3.1). To estimate densities, I grouped data from all transects within a sampling location to obtain independent sampling units. I only used observations from those transects and transect segments that overlapped between 1995 and 2006. Thus, I minimized any design bias due to placement and differences in transect length when comparing both survey periods. For okapi and all duiker species, I pooled the observations for each species from both surveys. Sample sizes for okapi and yellow-backed duiker were too small in each survey alone to model the detection function. The data for small and red duikers showed a disproportionately large number of perpendicular distance measurements at zero in 1995, due to incorrectly measuring small perpendicular distances as zero cm from the transect line. This caused problems for modeling the detection curve, but after combining the 1995 and 2006 data, I obtained acceptable models. By pooling the data, I assumed that detection probabilities were similar between the two time periods. This was a reasonable assumption since field protocols were identical and many of the same observers were involved in both surveys. For elephants, I obtained good models with the data from each survey separately. To analyze changes in wildlife densities between 1995 and 2006, I used z-tests to test the null hypothesis that the difference in density between both surveys was zero (Thomas et al. 2002). To study the effect of secure areas on wildlife during the war, I analyzed changes in density between 1995 and 2006 in the Green and Red Zone separately. I also compared mean density estimates between the Green and Red Zone in each time period.  3.2.4.2 Proportion of change in densities I correlated the amount of absolute change in densities versus initial densities in 1995 using linear least squares regression (changes in density were normally distributed). I also examined how much densities had changed proportionally relative to their - 73 -  initial densities. For this, I ran a simulation in the software program R (http://www.rproject.org) that randomized changes in density in proportion to initial densities across all sampling locations. After 400 simulations, I calculated the mean linear regression slope and confidence intervals of the simulated data versus initial densities. This slope corresponded to expected changes in density if proportional changes would have been the same at all locations. I then compared the slope of the observed data to the simulated slope to see if changes were higher or lower than expected.  3.2.4.3 Spatial modeling To test the a priori formulated hypotheses presented in table 3.2, I developed spatial regression models using Generalized Additive Modeling (GAM). Cumberworth et al. (1996), Hedley et al. (1999) and Williams et al. (2006) successfully used GAMs in combination with line transect sampling to model abundance of marine mammals as a function of spatial covariates. I fitted density to spatial covariates in a GAM with the following form: q  di = exp{α + ∑ f (xij )}  (3.3)  j =1  where di is the density estimated on transect i , α is the intercept and f (xij ) is a smooth function of covariate x on the i th transect. Because the influence of a covariate is modeled using a smooth function instead of a linear function as is the case with linear and generalized linear regression, the relationship between an independent and the dependent variable can include truly non-linear shapes that cannot be transformed into linear forms (Hastie and Tibshirani 1990). To avoid over fitting the models, I restricted the maximum degrees of freedom for fitting each parameter to three. Errors showed signs of an over-dispersed Poisson distribution and I modeled these using a Quasipoisson distribution for over-dispersed data (Chapter 2, Potts and Elith 2006). This distribution  - 74 -  does not assume a fixed dispersion parameter, as the Poisson distribution does, but estimates it during modeling. I examined multicollinearity between variables using scatterplots and Pearson correlation tests. I kept only those covariates with a correlation coefficient of less than 0.6, an arbitrary threshold which ensured that my a priori formulated hypotheses in terms of human presence, human access, protection and habitat were still represented. Distance from the nearest village was not included as a covariate because of its tight linear correlation with distance from the nearest road. Distance from the nearest major town was correlated with deforestation index, which was expected since major towns were surrounded by extensive deforestation. It was also excluded from the models. I used "mgcv" (multiple smoothing parameter estimation by Generalized Cross Validation) gams (Wood 2006) in R (http://www.r-project.org) which provide automatic selection of smoothing parameters for each covariate using Generalized Cross Validation (GCV). "Mgcv" gams also give a good fit of data with many zeros if a Quasipoisson distribution is used (Wood 2004). Model simplification and model selection was carried out by the process of backward deletion. I starting from an initial model comprising all candidate covariates and then dropped terms sequentially. Each time a term was dropped, I checked plots and GCV scores (equivalent to AIC) to see if the deletion was warranted (Chapter 2, Wood 2001). I developed spatio-temporal models in which I included time as a covariate (Buckland and Underwood, 2000). For those covariates that changed over time, I used separate values for each time period. I also fitted individual models for each survey in 1995 and 2006, which enabled me to compare the relative impact of each covariate between surveys. I tested models with human-related covariates first and then models that included habitat covariates as well, to see if the latter could explain additional variation in density patterns. I interpreted the results of the GAM’s using a visual presentation of densities. I used "Ordinary Kriging" in ArcGIS geostatistical analyst (Johnston et al. 2001, ESRI, http:/ - 75 -  /www.esri.com/) to generate a continuous surface of animal densities for each species on a map (Appendix 3.5). This method is based on the assumption that objects closer to each other are more similar than objects further apart. Kriging assigns a weight to any particular point of a surface based on the measured values of neighbouring sampling locations, the distance to those locations and the overall spatial arrangement of the data points. I used this technique to inspect spatial patterns and to interpret the results of the spatial models, not to predict densities in unsampled areas.  3.3 Results 3.3.1 Declines in dung densities Dung densities of all species declined significantly between 1995 and 2006 (table 3.3 and 3.4, figure 3.2). Elephant dung density in the reserve decreased by 48%. This was the case in both the Green Zone (45%) and the Red Zone (51%) (table 3.4), but declines in these strata alone were not statistically significant. This may have been due to the small sample sizes in each stratum and the high variance in encounter rates. Elephant dung densities were not significantly different between the Green and Red Zone in either survey period (table 3.5). Okapi dung densities declined by 43% overall, mainly due to a decline of 58% in the Red Zone. Densities remained stable in the Green Zone. In 1996, densities were higher in the Red than the Green Zone, but this difference disappeared in 2006. Small duiker dung densities did not decline significantly in the reserve as a whole. However, they declined steeply in the Green Zone (77%), while there was no significant decline in the Red Zone. In 1995, small duiker dung was more common in the Red than in the Green Zone, and this difference became larger by 2006. Dung of red and yellow-backed duikers declined significantly in the reserve, mainly due to large declines in the Green Zone (85% for red duikers and 84% for yellowbacked duiker). In both surveys, they were more abundant in the Red Zone than in the Green Zone and that difference also became greater in 2006. - 76 -  3.3.2 Proportional declines in densities Elephant dung densities declined proportionally more at sampling locations with higher initial densities in 1995 than in locations with lower initial densities (figure 3.3). This suggests that relatively more animals may have disappeared in sites of higher abundance than in sites of lower abundance through perhaps hunting, emigration or other reasons. In 12 locations, densities actually increased (figure 3.3) and 8 of these were located in the Green Zone. All these locations, except for one, had initial low densities. A similar pattern was observed for okapi, blue and yellow-backed duikers but not for red duikers.  3.3.3 Spatial and spatio-temporal modeling of human influences and habitat. 3.3.3.1 Elephants A spatio-temporal model of elephant dung confirmed the observed decline between the two survey periods and the higher densities in the Green Zone across both surveys (table 3.6). The deforestation index in the buffer zone around the reserve was the most significant variable in 1995, predicting higher densities on transects that were farther away from areas with high deforestation (Appendix 3.5 and 3.6). After the conflict, densities were also higher farther from deforestation hotspots, but other covariates became more important. Elephant dung densities were higher at a larger distance from the park boundary and lower in the Red Zone. Elephant dung was more abundant toward the centre of the park in the Green Zone. The negative relationship between abundance and distance to the nearest road can be attributed to the section of the road going through the middle of the Green Zone inside the reserve where elephants appeared to be more common and also to some extent to a few transects with higher dung densities near the road in the north-east of the reserve. Guard posts seemed to have little effect in protecting elephants as dung densities increased with distance from the nearest guard post. However, this correlation was mainly influenced by guard posts located at  - 77 -  the periphery of the reserve. In an univariate model with Epulu as the only guard post, higher densities were found closer to Epulu in the centre of the Green Zone (F = 11.36, p = 0.001). When habitat covariates were included, the 1995 model (table 3.7) improved only slightly (6% more deviance explained). Deviance explained in the 2006 model improved by 19%. Elephant dung densities were positively correlated with hill forest in both years and also with swamp forest in 2006.  3.3.3.2 Okapi Spatial model fits for okapi were poor (table 3.6). The 1995 model suggested some correlation between okapi dung densities and deforestation index in the buffer zone and distance from the nearest guard post, but the model fit was poor (only 12.95 deviance explained). No significant correlations emerged from the 2006 model. The inclusion of ecozones added some improvement to the model, especially in 2006, but it was difficult to draw any conclusions about associations with particular habitats because different habitat types were selected in 1995 than in 2006.  3.3.3.3 Duikers Survey year was the most important covariate in the spatio-temporal model for red duiker dung, essentially confirming the observed decline between 1995 and 2006. Contrary to my prediction, red duiker dung densities were lower near the guard post in Epulu in 1995. In 2006 this was also the case but, in the final model, security zone was a better predictor and guard post was dropped from the final model. Obviously security zone and guard posts are related since all guard posts are located within the Green Zone. Distance from the nearest road and deforestation index in the buffer zone were also significant covariates in both surveys. However, contrary to expectations, densities in 2006 were positively correlated with deforestation index. This could be attributed to a relative shift of densities toward the northern part of the reserve, which brought them relatively closer to adjacent deforestation hotspots (Appendix 3.5). Models including - 78 -  habitat outperformed models without it only slightly and associations with habitat types were inconclusive. Small duiker dung densities followed a similar pattern of correlations with spatial covariates to that of red duikers except that survey year was not a significant factor (table 3.6). Small duiker dung was scarcer close to roads and guard posts and was more common deeper inside the reserve (table 3.6, Appendix 3.5 and 3.6). In 2006, densities were significantly lower in the Green Zone. In the 2006 model they were correlated with forest-savannah ecotone and mixed forest, but this added only 9% to the explained deviance. Yellow-backed duiker dung densities also followed a similar pattern to that of red duikers, with an overall decline in the reserve but more so in the Green Zone (table 3.6, Appendix 3.5). This was confirmed by the spatio-temporal model in which survey year and security zone were the most significant predictors. Roads, guard posts, park boundary and deforestation index were all significant covariates in the 1996 model. In 2006, densities were lower in the Green Zone, and were correlated with distance from the park boundary and distance from guard posts. Model fits were poorer than for other duikers, which may be explained by smaller sample sizes and the fact that the large majority of the transects had no dung observations (75 out of 110 in 1995, and 93 out of 110 in 2006).  3.4 Discussion My results suggest that populations of all studied species declined substantially in the reserve between 1995 and 2007. The exceptions were small duikers which may have declined only in the Green Zone and not in the Red Zone.  - 79 -  3.4.1 Elephants 3.4.1.1 Elephant population decline as a result of the war Elephant dung densities declined from 4.09 to 2.13 dung piles per hectare. Using a dung defecation rate of 19.77 dung piles per day (Tchamba 1992) and a mean estimated dung decay rate of 44 days from a dung study carried out in the reserve by Hart and Hall (1996), and assuming that dung decay rates were similar for both survey periods, this corresponds to a decline of actual animal densities from 0.47 to 0.24 elephants per km2. If we consider this estimate to be representative for the entire reserve, the loss in elephants amounts to 3151 animals in the last decade, from 6439 individuals to a current population of 3288. These absolute figures should be treated with caution because defecation rates may be different in the RFO and seasonal, climatological and habitat related effects that affect decay rates of dung were not taken into account (Barnes 2001, Nchanji and Plumptre 2001). Nevertheless, my results clearly suggest that elephants suffered a serious decline that coincided with the conflict in the region. Before 1996, there was little poaching of elephants in the RFO, because poaching had halted after the CITES ban on ivory trade in 1989 (Hart and Hall 1996). In 1996 (Appendix 3.7), military and rebel factions moved into the area, looted park headquarters, disarmed park guards, brought in hunters and opened markets around the reserve for bushmeat and ivory. These militias were replaced by Ugandan backed rebels in 1999. The killing of elephants was widespread in 2000, and military deserters set up large poaching camps to the southeast, southwest and west of the reserve as well as inside the northeastern part (Hart and Mwinyihali, 2001). After intense diplomatic negotiations with the Congolese Rally for Democracy/Liberation Movement (RDC/ML), the main rebel group in the region, a joint military/park staff anti-poaching operation under the name “Tango” was launched (Mubalama and Mapilanga 2001). During this operation, 117 kg of ivory and 215 kg of elephant meat was recovered and 20 poachers were apprehended. At least 41 dead elephants were reported, but this was believed to be a fraction of the total number of animals killed as the operation only covered one third of - 80 -  the reserve in and around the Green Zone. Elephants in the region were killed not just for ivory but also to feed armed forces between Bunia and Kisangani (Plumptre et al. 2000). At that time, killing of elephants in the Green Zone was probably limited. Indeed, surveys carried out for the MIKE (Monitoring Illegal Killing of Elephants) program from 2000-2001 did not find a significant decline in elephants since 1995 in this area. The MIKE analysis compared only a subset (14) of transects of the 1995 surveys with the same transects in 2000 (Beyers et al. 2001). The worst of the killing, however, happened between 2002 and 2004 when rebel militias clashed in areas of high elephant density. In 2002, undercover operations discovered that 7 tons of ivory had left the reserve and adjacent areas over a period of 12 months. Assuming an average 6.9 kg of ivory per elephant (Hunter et al. 2004), this corresponded to 1014 dead elephants. Between June and December 2004, ICCN staff documented that about 17 tons of ivory (equivalent to 2463 elephants) was smuggled out of the area (Amboya 2004). This brings the total to 3477 elephants killed, remarkably close to the decline in the elephant population that I estimated between 1995 and 2007. It has to be noted though, that not all documented ivory came from the RFO but also from surrounding areas, mainly to the east of the reserve. On the other hand, it is also highly likely that not all killing was reported. Ivory was shipped to neighboring countries. For example, Milliken et al. (2006) reported that, around that time, ivory sales were booming in Angola, most of it originating in DRC. Hunter et al. (2004) estimated that about 4000 elephants were needed each year to supply both the illegal international (mainly Asian) and the internal African ivory market and most of this ivory was believed to have come from Central Africa, especially eastern DRC. The result of my analysis, combined with the above information on ivory seizures, suggests that the Ituri region may have been one of the most important global sources of ivory in the 2000-2004 period.  - 81 -  3.4.1.2 Area of refuge The spatial models and kriging maps (Appendix 1) suggest that elephants near the end and immediately after the conflict were more abundant toward the centre of the reserve than elsewhere. Dung densities were particularly higher in the middle third of the Green Zone, away from the boundaries of the park. All of the sampling locations in the Green Zone where densities had increased compared to 1995 were also located here. During the conflict, despite the overall decline, this area may have provided a refuge for elephants. This is supported by the fact that throughout the period of the worst elephant killing between 2002 and 2004, there was some active protection in the area. ICCN staff and two NGO partners, the Wildlife Conservation Society (WCS) and Gilman International Conservation (GIC), with financial support of the UNESCO conservation in crisis programme, were able to deploy some antipoaching patrols mainly around headquarters in Epulu and the research zones (Grossman et al. 2006). The post-war spatial model suggested that elephant densities were higher closer to the nearest road, which could, for the most part, be attributed to the road that runs through the reserve. Several large-scale studies of multiple sites in Central Africa have shown a negative relationship between distance from roads and elephant densities (Blake et al. 2007, Barnes et al. 1991, Barnes et al. 1995, Laurance et al. 2006). However, as shown in this analysis, this relationship with roads is not always valid at the smaller spatial scale of an individual site and may be confounded by other factors such as protection and habitat (Chapter 2, Blom et al. 2004). The estimated mean density of elephants in both survey periods was within the range of what Hall et al. (1997) found in 1996 in Kahuzi-Biega National Park and adjacent areas in Kivu. However, a recent survey in the lowland forest sector of this park found that these populations have been virtually eliminated during the conflict (Hart et al. 2007). The current RFO elephant population is well above the extremely low density of 0.05 elephants per km2 that Blake et al. (2007) recently reported in the Salonga NP in Central Congo, where virtually no active protection existed. Thus, despite a serious de-  - 82 -  cline, and probably thanks to some protection in a very limited area, the RFO remains one of the most important areas for elephants in the Democratic Republic of Congo.  3.4.2 Okapi Okapi dung densities declined significantly in the Red Zone. Several killed okapi were found by field teams during the war (John Hart, personal communication). In the Green Zone, populations remained relatively stable between 1995 and 2006. Okapi appeared to have a patchy distribution (Appendix 3.5). Only the 1995 model explained some spatial variation in dung densities in terms of human covariates (table 3.6). However, the results were contrary to my expectations because Okapi densities were found to be higher further from the guard post at Epulu and closer to deforestation hotspots. Habitat appeared to have a greater influence in the models, but it was difficult to draw any conclusions about preferences of okapi for certain habitats, possibly because the resolution at which the vegetation data was collected was not fine enough.  3.4.3 Duikers All duiker species appeared to have declined during the war, but overall, small duikers suffered a smaller decline than the larger red duikers or yellow-backed duiker. This may reflect both the preference of hunters for larger bodied species and the higher reproductive rates in small duikers (Fa et al. 2002). The dung data suggest that declines of duikers were much larger in the Green Zone than in the Red Zone. Densities were also significantly lower in the Green Zone than in the Red Zone in both surveys, except for yellow backed duiker in 1995 where the difference was not significant. Most of the red zone occupied the northern part of the reserve, while the green zone was located in the southern part. The results seem to confirm earlier studies by Wilkie and Finn (1990) who found higher duiker densities in an area in the north-east than those found by Koster and Hart (1988) in the south. The spatial models suggest that humans had a significant influence on the distribution and abundance of duikers both before and after the conflict. All species appeared - 83 -  generally more abundant further away from roads and the park boundary. This was especially true for the central area of the reserve (Green Zone) where the presence of the road and settlements along the road had a negative impact on duiker dung densities. The spatio-temporal models also confirmed that duiker dung densities were lower and declines were relatively larger in the Green Zone. Thus, the hypothesis that the Green Zone provided relative security for wildlife species during the war did not hold for duiker species. It is interesting to ask why this was so?  3.4.3.1 Bushmeat hunting It is possible that the observed differences in duiker densities and declines between the Green Zone and the Red Zone can be attributed to differences in hunting levels related to geography (north - south) and cultural practices rather than to differences associated with a temporary and arbitrary security gradient. Duikers are traditionally hunted by Mbuti Pygmies in the center and south of the reserve (Hart 2000) and by Efe Pygmies in the northern half (Vansina 1990). Hunting strategies of these two groups are very different. Mbuti Pygmies hunt in groups of 10-40 individuals with nets with which they catch duikers almost exclusively. Efe Pygmies hunt as individuals or in small groups with bow and arrow and target primates and small carnivores as well as duikers. Mbuti net hunters are much more geared to catching larger quantities of animals than Efe archers. Their subsistence economy is more dependent on large scale hunting and they generally go deeper inside the forest to hunt, while Efe pygmies tend to hunt closer to villages (Wilkie et al. 1998). These different levels and patterns of hunting might explain geographic differences in duiker densities, besides possible but unknown ecological factors. They may also account for the impact of human settlements along the road in the Green Zone on duiker densities. Pre-existing differences in hunting pressure might also explain why duikers declined more in the Green Zone than in the Red Zone during the war. Hunting had already been increasing before the war because of an increase in the demand for bushmeat by growing human populations in and around the reserve (Hart 2000). Since the outbreak of the - 84 -  civil war, bushmeat hunting suddenly soared due to anarchy and a total breakdown of control by reserve staff, while at the same time demand for bushmeat increased. Much of the meat was still consumed or traded locally but commercial markets also opened up near the periphery of the reserve to supply the army, rebel groups and general commercial trade (Hart and Mwinyihali 2001). The centre and the southern part of the reserve were more affected than the north, possibly because of higher hunting levels that existed in the south.  3.4.3.2 Mining Another but much more local factor adding to the hunting pressure was mining for coltan and gold. Coltan is a metallic ore containing columbium and tantalum and is used in the production of capacitors for cell-phones and other electronic devices. Eastern Congo has large deposits of easily extractable coltan and widespread extraction has been having large socio-political and ecological impacts (Hayes and Burge 2003, United Nations 2002). Between 1999 and 2001 coltan miners were abundant in the southern half of the RFO including Green Zone. They hired bushmeat hunters to supply them with meat (Hart and Mwinyihali 2001) and medium-sized mammals around mining camps were often completely eliminated. The coltan mining period was brief but intense.  3.5 Conclusions The civil war in the Democratic Republic of Congo had a major impact on all studied ungulates in the Okapi reserve but the impact was different for different species. Elephants suffered a major decline overall but declines were not uniform in the reserve. An important part of the population appears to have survived, mainly in the centre of the reserve, probably partly thanks to the dedication and efforts of local wardens, with support from international organizations. Without these efforts, the decimation might have been even greater, as was observed in nearly all other areas in Eastern Congo (Hart et al. 2007, Hart and Mwinyihali 2001, Kaboza and Debonnet 2005, Kalpers 2001, Associa- 85 -  tion nationale pour l'évaluation environmentale 2004). Spatial patterns of decline were not the same for forest duikers as for elephants which may be attributed to different types and levels of hunting pressure. This implies that monitoring programs should target several species to understand the impacts of humans on wildlife as a whole. Despite the war, the future of the reserve and its wildlife will more likely be determined by road development, growing human populations and demands for bushmeat and other resources. As of this writing, elephant poaching has declined substantially but the bushmeat hunting continues unabated (John Hart, personal communication). The latter shows that even traditional hunting methods can spin out of control and lead to a local depletion of wildlife. It is not enough to protect just charismatic species like elephants. Sustainable conservation should cover the whole range of species, including those that are exploited in a context of traditional lifestyles. Approaches and potential pitfalls to developing sustainable harvest programs of bushmeat species are described by Robinson and Bennett (2000) and Bennett et al. (2007).  - 86 -  3.6 Tables Table 3.1. Number of sampling locations and transects for each wildlife survey (1995 and 2006) in the Okapi Faunal Reserve. 1995  2006  Locations  Transects  Locations  Transects  All locations  54  123  51  123  Used in comparative analysis  51  110  51  110  - 87 -  Table 3.2. Candidate covariates included in the spatial models for 1995 and 2006.  Name  Covariate  Source  Hypotheses  Landsat ETM image  Roads provide access to hunters and wildlife densities are lower closer to the nearest road.  HUMAN RELATED COVARIATES Roads  Distance from the nearest road  Villages  Distance from the nearest village GPS waypoints  Wildlife is less abundant near human habitation and wildlife densities are lower closer to the nearest village.  Main towns  Distance from the nearest major GPS waypoints town  Big towns have proportionally more impact on wildlife than smaller settlements and wildlife densities are lower closer to the nearest larger town.  Park  Distance from the park boundary CARPE database  The park boundary acts as a protective barrier against hunters and animal densities are higher further inside the park.  Guard posts  Distance from the nearest guard GPS waypoints post  Guard posts provide protection to wildlife and there is a negative relationship between wildlife densities and distance to the nearest guard post.  Deforestation index  Composite index of deforestation Landsat and extent and distance from each Modis images, transect to all deforestation sites GIS on a predefined grid  Wildlife abundance is negatively associated with proximity to and extent of deforestation as a proxy for human population density  Security zones in 2006  Green (secure and accessible) Zone versus Red Zone (insecure and inaccessible) zone during the civil war from 2000 - 2005  Digitized on RFO basemap created from Landsat ETM image  Secure areas during the war had a positive influence on wildlife abundance compared to insecure areas.  Slope influences abundance of elephants and other wildlife species either directly or through different types of vegetation that are associated with a different topography.  HABITAT RELATED COVARIATES Slope  Average slope within a 100 meter buffer along transect  SRTM  Habitat  Ecozones  Digitized from A specific animal species prefers Landsat satellite certain habitats over others. images and field data.  - 88 -  Table 3.3. Survey effort, encounter rates and dung densities of different ungulates in the RFO from the data subset used for spatial and spatio-temporal models. CV (n/L) = coefficient of variation for dung encounter rates (dung piles per km), ESW = effective strip width, D = dung density per hectare, CV (D) = coefficient of variation for dung density per hectare, CI = confidence interval for dung density per hectare. Species  Survey Samples Total effort (km)  No obs.  n/L (per km)  CV (n/ ESW D L) (m) (per ha)  CV (D)  Elephant  1995  51  280  460  1.64  18.63  2.01  4.09  19.07 2.83-5.93  2006  51  280  286  1.02  28.08  2.40  2.13  28.70 1.22-3.70  1995  51  280  91  0.33  17.04  1.53  1.06  18.73 0.74-1.52  2006  51  280  51  0.18  18.79  1.53  0.60  20.34 0.40-0.88  1995  51  280  226  0.81  19.69  0.99  4.08  23.09 2.61-6.37  2006  51  280  167  0.60  28.83  0.99  3.01  31.25 1.66-5.48  1995  51  280  398  1.42  17.65  1.19  5.97  18.12 4.20-8.49  2006  51  280  231  0.83  26.00  1.19  3.47  26.33 2.08-5.76  1995  51  280  70  0.25  20.47  1.61  0.78  21.79 0.51-1.88  2006  51  280  29  0.10  32.64  1.61  0.32  33.49 0.17-0.61  Okapi Small duiker Red duiker Yellow-backed duiker  - 89 -  CI  Table 3.4. Change in ungulate dung densities (per hectare) between 1995 and 2006 in the whole reserve, the Green and the Red Zone.  Whole area  Green zone  Red zone  Elephant  Okapi  Small duiker  Red duiker Y-backed duiker  Density 1995  4.09  1.06  4.08  5.97  0.78  Density 2006  2.13  0.60  3.01  3.46  0.32  Change in density (%)  -1.96 (-48%)  -0.466 (-43%)  -1.06 (-26%)  -2.51 (-42%)  -0.456 (-58%)  z  1.978*  2.005*  0.799  1.770*  2.267*  p-value  0.024  0.023  0.212  0.038  0.012  Density 1995  5.25  0.705  2.18  3.81  0.543  Density 2006  2.82  0.591  0.49  0.56  0.087  Change in density (%)  -2.43 (-46%)  -0.114 (-16%)  -1.74 (-77%)  -3.25 (-85%)  -0.456 (-84%)  z  1.648  0.450  1.69*  3.126*  2.397*  p-value  0.050  0.326  0.004  <0.001  0.008  Density 1995  2.88  1.432  6.06  8.24  1.026  Density 2006  1.40  0.597  5.66  6.52  0.570  Change in density (%)  -1.48 (-51%)  -0.835 (-58%)  -0.41 (-7%)  -1.72 (-21%)  -0.456 (-44%)  z  1.223  2.213*  0.158  0.650  1.264  0.437  0.257  0.103  p-value 0.111 0.013 z-test on difference, * is significant at 0.05 alpha level  - 90 -  Table 3.5. Differences in ungulate dung densities (per hectare) between the Green and the Red Zone in the Okapi reserve in 1995 and 2006. Elephant  Okapi  Small duiker  Red duiker Y-backed duiker  2.37  -0.73  -3.88  -4.43  -0.48  z  1.581  1.872*  2.112*  2.027*  2.124  p-value  0.057  0.031  0.017  0.021  0.077  1.42  -0.01  -5.16  -5.964  -0.48  1.200  0.024  2.712*  3.276*  2.124*  0.003  <0.001  0.017  Green vs Red Difference in Zone in 1995 density  Green vs Red Difference in Zone in 2006 density z  p-value 0.115 0.491 z-test on difference, * is significant at 0.05 alpha level  - 91 -  Table 3.6. Fitted GAM's of ungulate dung densities in the Okapi reserve in 1995, 2006 and in both time periods combined. The "+" sign denotes a positive correlation, the "-" sign denotes a negative correlation.  Species  Survey  Model covariates  R2 (adjusted)  Deviance explained (%)  GCV  Elephant  both  year red zone s(defor) s(park) +  0.0687  20.2  0.8057  1995  post s(defor) -  0.0863  17.9  0.8685  2006  red zone s(defor) s(park) + s (post) + s (road) -  0.26  47  0.5317  both  post + s(defor) -  0.0317  4.53  0.2004  1995  post + s(defor) -  0.0861  12.9  0.22296  2006  null model  both  green zone s(road) -/+ s(post) + s(park) +  0.394  47.6  0.50931  1995  post+ s(road) + s(defor) s(park) +  0.526  52.4  0.42295  2006  green zone s( post) -/+ s(road) -/+ s(park) +  0.518  62.8  0.34806  both  year s(post) + s(park) + s(road) +  0.346  45.6  0.52338  1995  post + s(park) + s(road) + s(defor) -  0.436  47.6  0.53196  Okapi  Small duiker  Red duiker  - 92 -  Species  Survey  Model covariates  R2 (adjusted)  Deviance explained (%)  GCV  Red duiker  2006  green zone defor + park+ s(road) -/+  0.485  58.9  0.37806  Yellow-backed duiker  both  year green zone s(park) +  0.118  20.9  0.15669  1995  post + s(park) + s(defor) -  0.18  26.9  0.18858  2006  green zone post s(park) + s(defor) +  0.307  33.6  0.08999  - 93 -  Table 3.7. Fitted GAM's of ungulate dung densities in the Okapi reserve in 1995, 2006 and in both time periods combined, including ecozone covariates. The "+" sign denotes a positive correlation, the "-" sign denotes a negative correlation. Species  Survey  Model covariates  R2 (adjusted) Deviance explained (%)  GCV  Elephant  all  year red zone s(defor) swamp forest + hill forest + ecotone + s(mixed forest) +  0.162  28.6  0.74254  1995  post s(defor) slope s(hill forest) +  0.125  23.8  0.83787  2006  defor red zone road s(park) + s(post) + s(swamp forest) + s(hill forest) -/+  0.768  66  0.35962  all  post + defor swamp forest + mixed forest + s(ecotone) + s(fhill forest) +  0.036  9.22  0.19891  1995  post + s(defor) mixed forest + ecotone +  0.137  18.2  0.21758  2006  swamp forest + non forest s(mono forest) s(ecotone) +/ -  0.154  23.2  0.13741  all  green zone post + s(park) + s(road) -/+  0.422  45.8  0.46349  Okapi  Small duiker  - 94 -  Species  Survey  Model covariates  R2 (adjusted) Deviance explained (%)  GCV  Small duiker  1995  post + slope + s(road) + s(park) + s(defor) -  0.561  54.3  0.41409  2006  green zone s(post) -/+ s(road) -/+ s(park) + s(ecotone) + s(mixed forest) +  0.68  71.6  0.2865  all  year s(post) + s(park) + s(road) + s(non forest) -  0.386  48  0.51023  1995  post + s(road) + s(defor) s(park) + s(mono forest) + s(non forest) -  0.51  54.2  0.50004  2006  green zone defor + park + ecotone + mixed forest + non forest + s(swamp forest) + s(slope) + s(hill forest) +  0.534  66.8  0.34834  all  year green zones(park) +  0.118  20.9  0.15669  1995  post + s(park) + s(defor) -  0.18  26.9  0.18858  2006  green zone s(park) + s(defor) +/slope -  0.321  39.3  0.082939  Red duiker  Yellow-backed duiker  - 95 -  3.7 Figures Figure 3.1. Map of the Okapi Faunal Reserve in the Democratic Republic of Congo with sampling locations. Mungbere  ± !  !  ! !  Wamba  ! !  !  !  Red zone  ! !  ! ! ! !  western road  eastern road ! !  ! !  Green zone  !  ! !  Niania  !  Legend Roads  ! !  !  Epulu  Headquarters  !  Sampling locations  !  !  Lenda  ! !  Mambassa  !  !  !  !  !  !  ! !  Major town  $  !  $ !!  ! !  !  !  Edoro  ! !  !  ! !  ! !  Red zone  Green zone 0  Park boundary Nigeria  Location of the Okapi Faunal Reserve (RFO) in Central Africa  25  50  75  100 Kilometers  Chad  Central African Republic  Ethiopia  Sudan  Cameroon  RFO Equatorial Guinea Gabon  Uganda Kenya Congo Rwanda Democratic Republic of the Congo  Burundi  Tanzania  Angola  Mozambique  Zambia  Namibia  - 96 -  Zimbabwe  - 97 -  Survey 1995  Species and area  Survey 2006 d  ke  ac  -b  er  ik  du  ed  "R  en  re  "G  d  ke  ac  -b er  ik  du  ed  "R  en  re  ed  R "G er  ik w  lo  Ye l d  ke  ac w  lo  Ye l  -b  w  lo  Ye l  du  er  ik  du  ed  R  ed  R  ed  R  r"  en  re  " "  ne  zo  ne  zo  "  "  er  ik  du  ne  zo  ne  zo  "  "  r  er  ik  du  ne  zo  ne  zo  ke  ui  "  "  ne  zo  ne  zo  pi  ka  O  "  ne  zo  "  ne  ea  ar  zo  ld  al  Sm G  r"  ke  ui  ld  al  Sm  ke  ui  ld  al  Sm  ed  "R  en  re  "G pi  ka  O  pi  ka  O  ed  "R  en  re  le  ho  w  "G nt  ha  ep  El  nt  nt  ha  ha  ep  El  ep  El  Dung density per ha  Figure 3.2. Mean ungulate dung density estimates in 1995 and in 2006 in the entire  area of the Okapi Faunal Reserve and in the Green and Red Zone within the reserve.  9  8  7  6  5  4  3  2  1  0  Figure 3.3. Change in ungulate dung density between 1995 and 2006 versus ungulate density in 1995. The solid line is the observed regression and the dashed line is the simulated regression when all sites would have undergone proportionally the same change in density. Plus signs represent sampling locations in the Red Zone, dots represent sampling locations in the Green Zone. The dotted line corresponds with zero change in density.  (b) Okapi  0.2  1.0  (a) Elephants  ●  ●  ● ●● ● ●● ●  ●  ●  !1.0  ●  ●  ● ●● ●  ●  0.1 0.0  0.0  ●  ● ● ● ● ● ●  ●  ● ●  ●●  ●  ● ●  ●  ●  !0.2  Change in density  0.5  ●  ●  ●  ●  !0.4  ●  !2.0  Change in density  ● ●  0.0  0.5  1.0  1.5  0.0  2.0  Density in 1995  0.1  0.2  0.3  Density in 1995  - 98 -  0.4  (d) Red duikers  0  (c) Small duikers  ● ● ● ● ● ● ● ● ● ● ●● ● ●  ●  ● ● ● ● ●● ● ●● ●●  ● ●  ● ● ●  0.5  1.0  !2 !5  1.5  2.0  2.5  0.2 ●  ● ● ●  ●  !0.2  ●  ●  ●  ●  !0.4  ●  ●  0.0  0.1  0.2  0.3  1  2  3  4  Density in 1995  (e) Yellow backed duiker  ● ● ● ●  ●  0  Density in 1995  0.1  ●  !6  !2 0.0  0.0  ● ● ●  ● ●  Change in density  ●  !3  0  ●  ●●  !4  Change in density  ●  !1  Change in density  1  !1  ●  0.4  Density in 1995  - 99 -  5  6  3.8 References Association nationale pour l'évaluation environmentale (ANEE). 2004. Actes de l'atelier sur les impacts et les enjeux environmentaux des conflits armés en République Démocratique du Congo. Kinshasa, RDC, 26-27 Octobre 2004. Barnes, R. F. W., A. Blom, and M. P. T. Alers. 1995. A review of the status of forest elephants Loxodonta africana in Central Africa. Biological Conservation 71: 125-132. Barnes, R. F. W., K. L. Barnes, M. P. T. Alers, and A. Blom. 1991. Man determines the distribution of elephants in the rain forests of northeastern Gabon. African Journal of Ecology 29: 54-65. Barnes, R. F. W., A. Blom, M. P. T. Alers, and K. L. Barnes. 1995. An estimate of the numbers of forest elephants in Gabon. Journal of Tropical Ecology 11: 27-37. Barnes, R. F. W., K. Beardsley, F. Michelmore, K. L. Barnes, M. P. T. Alers, and A. Blom. 1997. Estimating forest elephant numbers with dung counts and a geographic information system. Journal of Wildlife Management 61: 1384-1393. Barnes, R. F. W. 2001. How reliable are dung counts for estimating elephant numbers? African Journal of Ecology 39: 1-9. Bennett, E. L., E. Blencow, K. Brandon, D. Brown, R. W. Burn, G. Cowlishaw, G. Davies, H. Dublin, J. E. Fa, E. J. Milner-Gulland, J. G. Robinson, J. M. Rowcliffe, F. M. Underwood, and D. S. Wilkie. 2007. Hunting for consensus: reconciling bushmeat harvest, conservation, and development policy In West and Central Africa. Conservation Biology, 21: 884-887. Beyers, R., and J. Hart. 2001. Monitoring Illegal Killing of elephants (MIKE), Central African Pilot Project, Technical Report No 4: Site base maps, Data bases and Mike Site Reports. http://www.cites.org/eng/prog/MIKE/pilot/tech_rep4.shtml.  - 100 -  Beyers, R., L. Thomas, J. Hart, and S. Buckland. 2001. Monitoring Illegal Killing of Elephants (MIKE) Central African Pilot Project. Technical Report No 2. Recommendations for Ground-Based Survey Methods for Elephants in the Central African Forest Region. Blake, S., S. Strindberg, P. Boudjan, C. Makombo, I. Bila-Isia, O. Ilambu, F. Grossman, L. Bene-Bene, B. de Semboli, V. Mbenzo, D. S'hwa, R. Bayogo, L. Williamson, M. Fay, J. Hart, and F. Maisels. 2007. Forest elephant crisis in the Congo Basin. PLOS Biology 5: 0001-0009. Blanc, J.J., R.F.W. Barnes, J.C. Craig, H.T. Dublin, C.R. Thoules, I. DouglasHamilton, J.A. Hart. 2007. African Elephant Status Report 2007. An update from the African Elephant Database. Occasional Paper of the IUCN Species Survival Commission No. 33. Blom, A., R. van Zalinge, E. Mbea, I. M. A. Heitkonig, and H. H. T. Prins. 2004. Human impact on wildlife populations within a protected Central African forest. African Journal of Ecology 42: 23-31. Blom, A., R. van Zalinge, I. M. A. Heitkonig, and H. H. T. Prins. 2005. Factors influencing the distribution of large mammals within a protected central African forest. Oryx 39: 381-388. Buckland, S. T., D. R. Anderson, K. P. Burnham, and J. L. Laake. 1993. Distance sampling. Estimating abundance of biological populations. Chapman & Hall. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to Distance Sampling. Estimating abundance of biological populations. Oxford University Press. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. B. Laake, D.L., and L. Thomas. 2004. Advanced distance sampling. Oxford University Press. Buckland, S. T., Underwood, F. 2001. MIKE Pilot Analysis, first interim report In: Beyers, R., Thomas, L., Hart, J., Buckland, S. Monitoring Illegal Killing of Elephants - 101 -  (MIKE) Central African Pilot Project. Technical Report No 2. Recommendations for Ground-Based Survey Methods for Elephants in the Central African Forest Region. Buij, R., W. J. McShea, P. Campbell, M. E. Lee, F. Dallmeier, S. Guimondou, L. Mackaga, N. Guisseougou, S. Mboumba, J. E. Hines, J. D. Nichols, and A. Alonso. 2007. Patch-occupancy models indicate human activity as major determinant of forest elephant (Loxodonta cyclotis). Seasonal distribution In an industrial corridor In Gabon. Biological Conservation 135: 189-201. Cumberworth, S. L., S. T. Buckland, and D. L. Borchers. 1996. A spatial modelling approach for the analysis of line transect data. Paper SC/48/012 presented to the IWC Scientific Committee, June 1996 (unpublished). De Merode, E., K. Hillman-Smith, A. Nicholas, A. Ndey, and M. Likango. 2000. The spatial correlates of wildlife distribution around Garamba National Park, Democratic Republic of Congo. International Journal of Remote Sensing 21: 2665-2683. Douglas-Hamilton, I. 1987. African elephants population trends and their causes. Oryx 21: 11-24. Grossman, F., J. Hart, and S. Dino. 2006. Réserve de Faune à Okapi. Post conflict faunal baseline surveys 2005 Central Sector "Zone Verte". IMU Technical Report Nº 5. Hall, J. S., B.-I. Inogwabini, E. A. Williamson, I. Omari, C. Sikubwabo, and L. J. T. White. 1997. A survey of elephants (Loxodonta africana) in the Kahuzi-Biega National Park lowland sector and adjacent forest in eastern Zaire. African Journal of Ecology 35: 213-223. Hart, J., and J. Hall. 1996. Status of eastern Zaire's forest parks and reserves. Conservation Biology 10: 316-327. Hart, J., M. Carbo, F. Amsini, F. Grossmann, and C. Kibambe. 2007. Parc National de Kahuzi-Biega, Secteur de Basse Altitude: Inventaire préliminaire de la grande faune avec une évaluation de l’impact des activités humaines et la situation sécuritaire  - 102 -  2004 - 2007. Inventory and Monitoring Unit, Rapport No7, Novembre 2007, Widlife Conservation Society, Democratic Republic of Congo. Hart, J. A. 2000. Impact and sustainability of indigenous unting in the Ituri Forest, Congo-Zaire: a comparison of unhunted and hunted duiker populations. In: Robinson, J. G, Bennet, E. L. 2000. Hunting for sustainability in tropical forests. New York: Columbia University Press. Hart, J. A., M. Katembo, and K. Punga. 1996. Diet, prey selection and ecological relations of Leopard and Golden cat in the Ituri forest. African Journal of Ecology 34: 364-379. Hart, T., and R. Mwinyihali. 2001. Armed conflict and biodiversity in sub-saharan Africa: The case of the Democratic Republic of Congo (DRC). Washington, D.C.: Biodiversity Support Program. Hart, T. B., J. Hart, R. Deschamps, M. Fournier, and M. Ataholo. 1994. Changes in forest composition over the last 4000 years in the Ituri Basin, Zaire. In The Biodiversity of African Plants, Procedings XIV AETFAT Congress, Wageningen, August 1994, Dordrecht, Kuluwer Academic Press. Hart, T. B., J. A. Hart, and P. G. Murphy. 1989. Monodominant and species-rich forests of the humid tropics: causes for their co-occurrence. American Naturalist 133: 613-633. Hastie, T. J., and R. J. Tibshirani. 1990. Generalized Additive Models. Monographs on statistics and applied probability, Chapman & Hall 43. Hayes, K., and R. Burge. 2003. Coltan mining in the Democratic Republic of Congo: how tantalum - using industries can commit to the reconstruction of the DRC. Fauna and Flora International. Hayward, M. W., P. Henschel, J. O'Brien, M. Hofmeyr, G. Balme, and G. I. H. Kerley. 2006. Prey preferences of the leopard (Panthera Pardus). Journal of Zoology (London) 270: 298-313. - 103 -  Hedley, S. L., S. T. Buckland, and D. L. Borchers. 2004. Spatial distance sampling methods. In: Advanced Distance Sampling (Buckland et al.), Oxford University Press. Hedley, S. L., S. T. Buckland, and D. L. Borchers. 1999. Spatial modelling from line transect data. Journal of Cetacean Research & Management 1: 255-264. Hofmann, T., and H. Roth. 2003. Feeding preferences of duiker (Cephalophus maxwelli, C. rufilatus, and C. niger) in Ivory Coast and Ghana. Mammalian Biology 68: 65-77. Hunter, N., E. B. Martin, and T. Milliken. 2004. Determining the number of elephants to supply the current unregulated ivory markets in Africa and Asia. Pachyderm 36: 116-128. Johnston, K., J. M. Ver Hoef, K. Krivoruchko and N. Lucas. 2001. Using ArcGIS Geostatistical Analyist. ESRI. Redlands, CA. Kaboza, Y., Debonnet, G. (editors). 2005. Promouvoir et préserver le patrimoine congolais. Lier diversité biologique et culturelle. Actes de la conférence et des ateliers. UNESCO, 13-17 septembre 2004. World Heritage Series n°17. Kalpers, J. 2001. Volcanoes under siege: impact of a decade of armed conflict in the Virungas. Washington, D.C.: Biodiversity Support Program. Kingdon, J. 1997. The Kingdon field guide to African mammals. Academic Press. Koster, S. H., and J. A. Hart. 1988. Methods of estimating ungulate populations in tropical forests. African Journal of Ecology 26: 117-126. Lahm, S. A., R. F. W. Barnes, K. Beardsley, and P. Cervinka. 1998. A method for censusing the greater white-nosed monkey in northeastern Gabon using the population density gradient in relation to roads. Journal of Tropical Ecology 14: 629-643.  - 104 -  Laurance, W. F., B. M. Croes, L. Tchignoumba, S. A. Lahm, A. Alonso, M. E. Lee, P. Campbell, and C. Ondzeano. 2006. Impacts of roads and hunting on Central African rainforest mammals. Conservation Biology 20: 1251-1261. Marks, Stuart A.. 1996. Local hunters and wildlife surveys: an assessment and comparison of counts for 1989, 1990 and 1993. African Journal of Ecology, 34: 237-257. Milliken, T., A. Pole, and A. Huongo. 2006. No peace for elephants. Unregulated domestic ivory markets in Angola and Mozambique. (TRAFFIC Online Report Series No. 11).Cambridge: TRAFFIC International. Mubalama, L., and J. J. Mapilanga. 2001. Less elephant slaughter in the Okapi Faunal Reserve, Democratic Republic of Congo, with Operation Tango. Pachyderm 31: 36-41. Mulley, B. 2004. La Réserve de Faune à Okapis: agricultural zoning project, Interim Report. IMU-RFO Rapport No 1. Nchanji, A. C., and A. J. Plumptre. 2001. Seasonality in elephant dung decay and implications for censusing and population monitoring in south-western Cameroon. African Journal of Ecology 39: 24-32. Nunez-Iturri, G., and H. F. Howe. 2007. Bushmeat and the fate of trees with seeds dispersed by large primates in a lowland rain forest in western Amazonia. Biotropica 39: 348-354. Plumptre, A. J., T. Hart, A. Vedder, and J. Robinson. 2000. Support for Congolese conservationists. Science 288: 617. Plumptre, A. J. 2000. Monitoring mammal populations with line transect techniques in African forests. Journal of Applied Ecology 37: 356-368. Potts, J. M. and J. Elith. 2006. Comparing species abundance models. Ecological Modelling, 199:153-163.  - 105 -  Robinson, J. G. and E.L. Bennet. 2000. Hunting for sustainability in tropical forests. New York : Columbia University Press. Stoner, K. E., K. Vulinec, S. J. Wright, and C. A. Peres. 2007. Hunting and plant community dynamics in tropical forests: a synthesis and future directions. Biotropica 39: 385-392. Swartzman, G., C. H. Huang, and S. Kaluzny. 1992. Spatial analysis of Bering Sea groundfish survey data using generalized additive models. Canadian Journal of Fisheries and Aquatic Sciences 49: 1366-1378. Tchamba, M. N. 1992. Defaecation rates by the African forest elephant (Loxodonta Africana Cyclotis) in the Santchou Reserve, Cameroun. Mammalia 56: 155-158. Thomas, L., S. T. Buckland, K. P. Burnham, D. R. Anderson, J. L. Laake, D. L. Borchers, and S. Strindberg. 2002. Distance sampling. pp 544-552 in Encyclopedia of Environmetrics, Edited byAbdel H. El-Shaarawi and Walter W. Piegorsch John Wiley & Sons, Ltd, Chichester. Thomas, S. C. 1991. Population densities and patterns of habitat use among anthropoid primates of the Ituri forest, Zaire. Biotropica 23: 68-83. United Nations. 2002. Final report of the Panel of Experts on the Illegal Exploitation of Natural Resources and Other Forms of Wealth of the Democratic Republic of the Congo. UN Doc: S/2002/1146, 16 October 2002. Vansina, J. 1990. Paths in the rainforest. Madison, Wisconsin. The University of Wisconsin Press. Wang, B. C., V. L. Sork, M. T. Leong, and T. B. Smith. 2007. Hunting of mammals reduces seed removal and dispersal of the afrotropical tree Antrocaryon klaineanum (Anacardiaceae). Biotropica 39: 340-347. White, L., and A. Edwards (editors). 2000. Conservation research in the African rain forests, a technical handbook. The Wildlife Conservation Society, New York 444.  - 106 -  Wilkie, D. S. 1989. The impact of roadside agriculture on subsistence hunting in the Ituri forest of notheastern Zaire. American Journal of Physical Anthropology 78: 485-494. Wilkie, D. S., and J. T. Finn. 1988. A spatial model of land use and forest generation in the Ituri forest of Northeastern Zaire. Ecological Modelling 41: 307-324. Wilkie, D. S., B. Curran, R. Tshombe, and G. A. Morelli. 1998. Modeling the sustainability of subsistence farming and hunting in the Ituri forest of Zaire. Conservation Biology 12: 137-147. Wilkie, D. S., and J. F. Carpenter. 1999. Bushmeat hunting in the Congo Basin: an assessment of impacts and options for mitigation. Biodiversity & Conservation 8: 927-955. Wilkie, D. S., and J. T. Finn. 1990. Slash-burn cultivation and mammal abundance in the Ituri forest, Zaire. Biotropica 22: 90-99. Williams, R., S. L. Hedley, and P. S. Hammond. 2006. Modeling distribution and abundance of Antarctic baleen whales using ships of opportunity. Ecology & Society 11. Wood, S. 2001. mgcv : GAMs and Generalized Ridge Regression for R. R News 1/2: 20-26. Wood, S. N. 2006. Generalized Additive Models: an introduction with R. Chapman and Hall/CRC Press. Wood, S. I., N. 2004. Stable and efficient multiple smoothing parameter estimation for generalized additive models. Journal of the American Statistical Association 99: 673.  - 107 -  CHAPTER 4. General Conclusions In this dissertation I examined the question of what anthropogenic and natural factors determined the distribution and abundance of elephants and other ungulates at the scale of a protected area in the forests of Central Africa. I analyzed elephant census data from the Odzala National Park in the Republic of Congo in relation to spatial factors (Chapter 2). In the Okapi Forest Reserve (RFO), I examined the effects of a civil war on the temporal and spatial patterns of ungulate abundance (Chapter 3). I used spatial modeling to try to answer these questions and, as well, I examined a few methodological issues associated with this approach. In section 1, I highlight some of the aspects of the spatial modeling techniques that I employed. In section 2, I summarize the results of the two studies of Chapter 1 and 2 and its implications for conservation ecology. In section 3, I make some recommendations for conservation that relate to the findings of this study.  4.1 Spatial Modeling I applied spatial and spatio-temporal models that combined line transect methodology, GIS and multivariate generalized modeling regression to successfully document and explain the spatial distribution and abundance of elephants in the Odzala National Park and of elephants and other ungulates in the Okapi Faunal Reserve. I obtained density estimates of dung of these species through Distance Sampling and related these to human and environmental variables using Generalized Linear Models (GLM) and/or Generalized Additive Models (GAM). I found GAMs to perform better than GLMs in explaining the spatial variation in abundance. For both the GLMs and GAMs, it was important to select an appropriate error distribution and to account for over-dipsersion. Overdispersion was mainly due to the existence of many zeros in the data, which commonly occurs with abundance data (Barry and Welsh 2002). A modeling approach such as this had several advantages over conventional survey estimates. - 108 -  First, it provided spatial information on threats that influenced wildlife abundance, information which could support management decisions and operations. Second, in a spatio-temporal context (RFO) it provided information on trends in abundance in relation to changing human impacts. Threats identified through spatial analysis could be monitored on a regular basis using relatively easily obtainable data from remote sensing and limited field work. Third, it could be used to predict population densities across a surface within the study area and provide estimates of abundance for sub-zones within the area. Fourth, it improved the precision of the estimates of abundance. Fifth, It could be used to analyze data from biased designs, including Law Enforcement Monitoring data, provided that these adhered to high data collection standards.  4.2 Conservation ecology: impact of humans on wildlife 4.2.1 Human influence on elephants Human influences dominated the observed patterns of distribution and abundance of elephants in the Odzala National Park (Chapter 2) and the Okapi Faunal reserve (Chapter 3). In Odzala, the most important factors were related to protection. Far more elephants were found inside than outside the National Park. In the park, densities were higher where there were more anti-poaching patrols. Elsewhere, densities increased with distance from the nearest access road. Historical and anecdotal information supports the hypothesis that a conservation program by the European Union program, ECOFAC, which commenced in 1992, has been effective in protecting elephants in Odzala. In the RFO, elephants were also more abundant further away from human settlements. Populations were halved as a result of the civil war. After the war, densities were correlated with distance from the park boundary and elephants were more abundant  - 109 -  deeper inside the park. Their densities were relatively higher in a small area in the centre of the park, where there was some protection during the war. The relationship between abundance or occurrence of elephants and distance to the nearest road or human settlement has been well documented elsewhere in Central Africa (Barnes et al. 1991, Fay and Agnagna 1991, Laurance et al. 2006, Michelmore et al. 1994, Barnes et al. 1997, Blake et al. 2007), but few studies examined the impact of protection or the interaction of protection and roads. Blake et al. (2007), in their regional analysis of elephant densities in relation to distance from roads, did acknowledge that there were additional site-specific factors, either human or ecological, that were not captured by the relationship with roads. Laurance et al. (2006) also observed a much stronger effect of roads on elephants outside than inside a logging concession, which had presumably a protective effect on elephants. Roads fragment otherwise continuous tracts of forest and provide access for hunters to previously remote areas that were available as de facto wildlife sanctuaries. Blake et al. (2007) stated that in Central Africa there is only 21 845 km2 of forest left that is over 50 km from the nearest road, and 65% of all forest is within 10 km of a road. With areas getting smaller and more accessible, elephant populations are likely to decline further, unless there they are protected.  4.2.2 Duikers and bushmeat hunting Duiker densities in the RFO were higher in remote areas of the reserve and further away from roads in both 1995 and 2006. Contrary to what I observed for elephants, their density was lower near the road, villages and guard posts that were located inside the park. The relative protection that a small area in the center of the park provided for elephants during the war did not benefit duikers. On the contrary, declines of duikers in this area were greater than elsewhere. This reflects a different hunting dynamic for duikers than for elephants. While elephants were mainly hunted by professional poachers with rifles or automatic weapons for ivory and meat, duikers were killed by local people with nets, snares or bow and arrow for local consumption and peripheral markets. New bushmeat markets opened up during the war to supply the army, rebel groups and com- 110 -  mercial trade. Coltan and gold miners, operating in the southern part of the reserve, were also supplied with bushmeat by Mbuti pygmies. Bushmeat hunting, which was already at high levels, especially in the centre and south of the park, flourished because of a total collapse of control. Hunting for meat is probably a far more important threat to wildlife and ecosystems in Central Africa than logging, deforestation or agricultural development (NOSS 1998, Wilkie et al. 1998). Bushmeat hunting is much greater in this part of the world than in any other tropical region, with extraction rates 20-50 times higher per km2 than, for example, in the Amazon basin (Robinson et al. 1999). This extraction rate was estimated by Fa et al. (2002) to be 6 times higher than the sustainable rate. The majority of households in Central Africa depend on meat of ungulates and primates for their primary source of protein (Wilkie and Carpenter 1999), and selling bushmeat also provides extra sources of revenue. A fast growing demand from urban centers has led to a booming commercial trade, that is often organized by people coming from outside the area. Demand is likely to increase further because of human population growth, which is at 3.39% per year in the DRC and at 2.6% in the Republic of Congo (CIA The World Factbook,  https://www.cia.gov/library/publications/the-world-factbook).  The  area  around the RFO has also seen a steady immigration of people during the last decade. The bushmeat problem is being aggravated by logging companies that provide access into previously roadless areas. Hunting associated with logging is often more destructive than the logging itself (Robinson et al. 1999). Employees and their families living in logging camps also hunt for food. In one example in the Republic of Congo, 648 people harvested 124 tons of meat a year (Auzel and Wilkie 2000). Logging causes changes in nearby communities, giving them greater access to wildlife and better infrastructure to transport bushmeat to markets. More than 600 000 km2 or 30% of the forest in the Congo basin is currently divided up in logging concessions and Laporte et al. (2007) detected 51 916 km2 of new logging roads on satellite images between 1976 and  - 111 -  2003. Their study also found new expansions of logging in DRC, an area that contains 63% of the rain-forest in Africa. The combination of fragmentation of the forest and human population growth may cause local extinctions even in areas where the forest cover still looks intact from the air. Roadless forest blocks may thus become smaller islands surrounded by roads and settlements from which continuous hunting pressure emanates. Brashares (2001) found that reserve size and human population density around the reserve explained almost all the variation in extinction rates of mammal species between reserves in Ghana. As hunting increases, there is a high risk that there may be a sudden collapse in wildlife populations after initial good harvests. Simulation models of over-harvesting showed similar trends for duikers as has been shown for the fisheries where heavily exploited species collapsed beyond recovery (Barnes 2002).  4.2.3 Impact of wildlife depletions on ecosystems What is especially worrying is that bushmeat hunting is not targeting just a few species, but the whole range of medium- to large-sized animals ranging from rats and snakes to forest buffalo and elephant. Usually the bigger species go first because of their lower reproductive rates, and because they provide more return for hunting effort. Many hunting methods, however, such as snaring and trapping, which are common in Odzala and the RFO (where drive hunts with nets are also used), catch a wide range of species and can be indiscriminate as to the species that get killed. However, Arcese et al. (1995) found snares to be quite selective in the Serengeti National Park. The less abundant species that favored woodland and thickets, where snares were set, were caught more often than the more common grassland species. Potentially, complete communities could disappear from the forest ecosystem, leading to the "empty forest syndrome", whereby superficially intact looking forests are completely devoid of medium- and large-sized mammals (Redford 1992). The impact of the depletion or extinction of these species on the forest ecosystem is largely unknown. Most studies date from the last few years. In 2002, there were globally - 112 -  only 12 studies that compared plant communities in hunted forest versus non-hunted forest (Stoner et al. 2007). Duikers and primates are mainly frugivurous and play a role in the dispersal of fruit trees (Hofmann and Roth 2003). This may have an effect on reproductive success of certain tree species. Depletion of duikers may further reduce numbers of predators such as leopard and golden cat, but no data are yet available to test this hypothesis. Elephants also contribute to the seed dispersal and germination of several tree species (Walsh and Blake 2002, Yumoto et al. 1995). Seedlings from certain trees that passed through an elephant gut had shorter germination times and higher growth rates (Nchanji and Plumptre 2003). Elephants are known to alter vegetation in southern and eastern Africa (Laws 1970, Whyte et al. 1999, Jacobs and Biggs 2002) and determine the state of ecosystems (Dublin et al. 1990). Much less is known about their physical impact on forest ecosystems in Central Africa, but they probably play a role in maintaining forest clearings that are important ecological features in Odzala and other forests (personal observations, Vanleeuwe et al. 1998).  4.3 Conservation ecology: recommendations 4.3.1 Protect large roadless areas with a high area/edge ratio In the absence of effective law enforcement, big unpopulated roadless areas with a high area to edge ratio offer the best protection for wildlife. The findings of this study and others (Brashares et al. 2001) puts into question the viability of narrow unprotected parks or wildlife corridors where hunters have easy access. It may be better to reserve scarce conservation resources for setting aside large blocks of roadless forest. Particular attention should be paid to roads and settlements inside a protected area, where (bushmeat) hunting and access should be strictly regulated and monitored. Logging companies should adopt and enforce policies to limit or prevent bushmeat hunting in their concessions and surrounding areas affected by their operations.  - 113 -  4.3.2 Invest in law enforcement and patrols Frequent law enforcement patrols can mitigate hunting pressures originating from the edges of protected areas or roads and villages located inside a reserve. Hilborn et al. (2006) showed that an increase in anti-poaching patrols and park budgets in the Serengeti National Park in Tanzania in the mid-1980's led to a steep decline in poaching. Populations of elephants, buffalos and black rhinos which were decimated by rampant poaching before have rebounded as a result of this. In Luangwa, a savannah park in Zambia, a law enforcement staff density of one person per 20 km2 was found to be necessary to eliminate all poaching of elephants and rhinos (Leader-Williams et al. 1990). In the Virungas in Rwanda, 1 guard per 2-6 km2 was deployed to effectively protect mountain gorillas. There are no calculations available for effective staffing density needed to protect elephants in dense tropical lowland rain-forest, but presumably it would have to be somewhere in between these two figures. Protection of smaller mammals may require even higher patrol staff densities. On the other hand, poachers are limited in their movements by the density of the vegetation and they use the same trails that animals and patrol guards use. This may increase the area that a patrol can effectively cover. Obviously such high staffing densities would require a large number of people and may be prohibitively expensive. Not all areas within a reserve have to be equally well patrolled however. Patrols should concentrate on the periphery of the reserve where there is hunting access and/or concentration of human settlements. Access roads and rivers can be controlled by strategically locating guard posts and deploying random mobile checkpoints on access roads, which is commonly done in Central Africa. Spatial analysis of wildlife densities and illegal activities in relation to threats help to determine priority areas for patrolling and thus optimize the use of resources. Continuous law enforcement monitoring (LEM) and spatial analysis of LEM data is useful to monitor illegal activities and adaptively respond to threats.  - 114 -  4.3.3 Ensure continued investment in park staff and protection of "safe zones" during periods of political turmoil In the RFO during the war, elephants were protected to some degree in a small area in the center of the reserve from the onslaught elsewhere. This emphasizes the importance of 'safe zones', at least for this charismatic species, in times of political turmoil. The protection provided would not have been possible without the commitment of highly motivated field staff supported by NGO's (WCS and the Gilman Foundation) and UNESCO's Conservation in Crisis programme. Field staff in the RFO had received longtime support from above-mentioned NGO's even before the war and were well trained. This was important as they had to rely on themselves when official institutions collapsed during the conflict. The park staff continued to receive salaries which helped them and their families through the war, and this was also an important factor in their motivation to stay despite the high risks involved in protecting the park. In periods of war or political instability, bilateral and multilateral donors usually retract from financial and other aid to a country, including conservation projects. The lack of ability to adapt to a changing political context shown by donor-driven projects poses a major problem for biological conservation, which can be irreversibly damaged during this time (Kalpers 2001). In the case of DRC, this has global consequences as this country has more species of mammals and birds than any other African country and it includes several endemic mammal and bird species (Hart and Mwinyihali 2001). Unlike governmental organizations that are often unable or unwilling to provide support to politically unstable countries, NGO's don't always have these constraints and they can have a very positive impact on conservation during times of instability, as has been shown in the RFO and in Garamba and Virunga national parks (Kalpers 2001, Hart and Mwinyihali 2001). Given internal conflicts or political instabilities that currently exist in several biodiversity rich countries, especially in Africa, international emergency funds for conservation should receive ample support from donor countries. Funds could be funneled through dedicated NGO's with experience in those countries bypassing inflexible or dysfunctional government bureaucracies. - 115 -  In areas of civil strife, protection of wildlife often requires a military solution and it may be necessary for wildlife managers and other people involved in conservation (including international organizations) to negotiate with the military or militias who's personnel is often involved in poaching. Operation Tango in the RFO was an example of such a negotiation, which was initially successful in reducing elephant poaching (Mubalama and Mapilanga 2001).  4.3.4 Monitor wildlife populations and illegal activities using appropriate scientific methods Management decisions and planning for conservation should be based on good science and information, and this had been lacking in Central Africa until quite recently (Blake and Hedges 2004). Wildlife surveys should be conducted using standard techniques that are comparable between different sites and over time. The CITES MIKE program is an example of how surveys of elephants are standardized and implemented across Africa and Asia (Hedges and Lawson 2006). Estimates for the surveyed area could be obtained using traditional methods, but spatial modeling can considerably enhance the information contained in the data. I recommend that these methods are standardized, used more broadly and applied to other areas for conservation. Law enforcement monitoring (LEM) is a useful addition to formal surveys and can be conducted on a continuous basis. LEM provides valuable information on patrol and protection effort that can be related to indicators of illegal activities and wildlife populations (Arcese et al. 1995, Jachmann 1998). Data collected routinely on law enforcement monitoring patrols comes at little or no extra cost compared to more expensive dedicated wildlife surveys. While difficult to quantify in absolute numbers, LEM can provide insight in the relative abundance and distribution of illegal activities, provided that the data is accurately collected in a way that allows analysis using reliable methods. It may be particularly valuable for detecting hunting pressures on smaller species that may otherwise go undetected. - 116 -  4.3.5 Clamp down on illegal ivory trade and sales Despite a CITES ban on international ivory sales, ivory continues to be traded illegally and DRC has been one of the main suppliers of global ivory in recent years. Urgent action is required to bring this trade under control from source to destination. Intelligence information about ivory trade routes and shipments is extremely valuable in leading to successful seizures, and much of this information can be obtained through undercover operations in the areas of origin (Amboya 2004). Because a substantial amount of ivory is sold domestically on internal markets in Africa, governments should clamp down on this mainly unregulated trade (Hunter et al. 2004, Milliken et al. 2006).  4.3.6 Increase global financial support It has been estimated that an investment of about 1 billion US dollars is needed over ten years to set up a protected area network that would be representative for ecosystems in the Niger delta - Congo forest basin, as proposed by WWF (Blom et al. 2001). After this investment it would cost an additional 87 million US dollars annually to maintain the system (Blom 2004). The total annual budget for conservation in the entire Congo basin (738 038 km2) was about 30 million dollars in 2004 or 40 dollars per km2. This is only slightly more than the 28 million dollar annual budget for Yellowstone National Park (8983 km2) in the USA in 2004 (Blake and Hedges 2004), where 3076 dollars was spent per km2. It is clear that conservation in Central Africa is grossly underfunded, resulting in a rapid loss of wildlife populations, including those in national parks and other protected areas. Most countries in the region are heavily indebted, lack incentives, and have a very low capacity and few resources to put toward conservation. Furthermore, most people are extremely poor and operate in a survival economy. They depend largely on the extraction of natural resources, including bushmeat, for their livelihoods. If the global community wants a future where big mammals like elephants and apes and the rich biodiversity of Central Africa continue to exist, it will have to commit to large financial, technical and political efforts in order to support biodiverstiy conservation and promote - 117 -  alternative incomes for people in this part of the world. Both the Odzala and the RFO case show, however, that even limited support can help local managers to meet at least some conservation objectives under very difficult circumstances.  - 118 -  4.4 References Amboya, C. 2004. Rapport sur le braconnage à l’Eléphant et sur la commerce de l’ivoire dans et à lapériphérie de la Réserve de Faune à Okapis. Inventory and Monitoring Unit, Rapport No 3, December 2004, Widlife Conservation Society, Democratic Republic of Congo, 33 pp. Auzel, P., and D. S. Wilkie. 2000. Wildlife use in Northern Congo: hunting in a commercial logging concession. In Robinson, J. G., Bennet, E. L. Hunting for sustainability in tropical forests. New York : Columbia University Press. Barnes, R. F. W., K. L. Barnes, M. P. T. Alers, and A. Blom. 1991. Man determines the distribution of elephants in the rain forests of northeastern Gabon. African Journal of Ecology 29: 54-65. Barnes, R. F. W., K. Beardsley, F. Michelmore, K. L. Barnes, M. P. T. Alers, and A. Blom. 1997. Estimating forest elephant numbers with dung counts and a geographic information system. Journal of Wildlife Management 61: 1384-1393. Barnes, R. F. W. 2002. The Bushmeat Boom And Bust In West And Central Africa. Oryx 36: 236-242. Barry, S. C., and A. H. Welsh. 2002. Generalized additive modelling and zero inflated count data. Ecological Modelling 157: 179-188. Blake, S., S. Strindberg, P. Boudjan, C. Makombo, I. Bila-Isia, O. Ilambu, F. Grossman, L. Bene-Bene, B. de Semboli, V. Mbenzo, D. S'hwa, R. Bayogo, L. Williamson, M. Fay, J. Hart, and F. Maisels. 2007. Forest elephant crisis in the Congo Basin. PLOS Biology 5: 0001-0009. Blake, S., and S. Hedges. 2004. Sinking the flagship: the case of forest elephants in Asia and Africa. Conservation Biology 18: 1191-1202.  - 119 -  Blom, A. 2004. An estimate of the costs of an effective system of protected areas in the Niger Delta - Congo Basin Forest Region. Biodiversity and Conservation 13: 2661-2678. Blom A., Kamdem-Toham A., D’Amico J., O’Hara D., Abell R. and Olson D. (eds). 2001. Assessment of biological priorities for conservation in the Niger Delta – Congo Basin Forest. Wildlife Fund, Washington, DC. Brashares, J. S., P. Arcese, and M. K. Sam. 2001. Human demography and reserve size predict wildlife extinction in West Africa. Proceedings of the Royal Society London 268: 2473-2478. Fa, J. E., C. A. Peres, and J. Meeuwig. 2002. Bushmeat exploitation in tropical forests: an intercontinental comparison. Conservation Biology 16: 232-237. Fay, J. M., and M. Agnagna. 1991. A population survey of elephants in Northern Congo. African Journal of Ecology 29: 177-187. Hart, T., and R. Mwinyihali. 2001. Armed Conflict and Biodiversity in Sub-Saharan Africa: The Case of the Democratic Republic of Congo (DRC). Washington, D.C.: Biodiversity Support Program. Hedges, S., and D. Lawson. 2006. Dung survey standards for the MIKE programme. Monitoring the Illegal Killing of Elephants. Mike Tecnical Report. Hilborn, R., P. Arcese, M. Borner, J. Hando, G. Hopcraft, M. Loibooki, S. Mduma, and A. R. E. Sinclair. 2006. Effective enforcement in a conservation area. Science 314: 1266-1269. Hofmann, T., and H. Roth. 2003. Feeding preferences of duiker (Cephalophus maxwelli, C. rufilatus, and C. niger) in Ivory Coast and Ghana. Mammalian Biology 68: 65-77.  - 120 -  Hunter, N., E. B. Martin, and T. Milliken. 2004. Determining the number of elephants to supply the current unregulated ivory markets in Africa and Asia. Pachyderm 36: 116-128. Jachmann, H. 1998. Monitoring Illegal Wildlife Use and Law Enforcement in African Savanna Rangelands. The Wildlife resource Monitoring Unit, Lusaka. Kalpers, J. 2001. Volcanoes under Siege: Impact of a Decade of Armed Conflict in the Virungas. Washington, D.C.: Biodiversity Support Program. Laporte, N. T., J. A. Stabach, R. Grosch, T. S. Lin, and S. J. Goetz. 2007. Expansion of industrial logging in Central Africa. Science 316: 1451. Laurance, W. F., B. M. Croes, L. Tchignoumba, S. A. Lahm, A. Alonso, M. E. Lee, P. Campbell, and C. Ondzeano. 2006. Impacts of roads and hunting on Central African rainforest mammals. Conservation Biology 20: 1251-1261. Leader-Williams, N., S. D. Albon, and P. S. M. Berry. 1990. Illegal exploitation of black rhinoceros and elephant populations: Patterns of decline, law enforcement and patrol effort in Luangwa Valley, Zambia. Journal of Applied Ecology 27: 1055-1087. Michelmore, F., K. Beardsley, R.-F.-W. Barnes, and I. Douglas-Hamilton. 1994. A model illustrating the changes in forest elephant numbers caused by poaching. African Journal of Ecology 32: 89-99. Milliken, T., A. Pole, and A. Huongo. 2006. No peace for elephants. Unregulated domestic ivory markets in Angola and Mozambique. (TRAFFIC Online Report Series No. 11).Cambridge: TRAFFIC International. Mubalama, L., and J. J. Mapilanga. 2001. Less elephant slaughter in the Okapi Faunal Reserve, Democratic Republic of Congo, with Operation Tango. Pachyderm 31: 36-41.  - 121 -  Nchanji, A. C., and A. J. Plumptre. 2003. Seed germination and early seedling establishment of some elephant-dispersed species in Banyang-Mbo Wildlife Sanctuary, South-Western Cameroon. Journal of Tropical Ecology 19: 229-237. Noss, A. J. 1998. Cable snares and bushmeat markets in a central African forest. Environmental Conservation 25: 228-233. Redford, K. H. 1992. The Empty Forest. (Cover Story). Bioscience 42: 412. Robinson, J. G., K. H. Redford, and E. L. Bennet. 1999. Wildlife harvest in logged tropical forest. Science 284: 595-596. Stoner, K. E., K. Vulinec, S. J. Wright, and C. A. Peres. 2007. Hunting and plant community dynamics in tropical forests: a synthesis and future directions. Biotropica 39: 385-392. Vanleeuwe, H., S. Cajani, and A. Gauthier-Hion. 1998. Forest clearings and the conservation of elephants in the Northeast of Congo republic. Pachyderm 24: 46-52. Walsh, P. D., and S. Blake. 2002. Elephant ranging and tree distribution in a Congo forest. Ecological Society of America Annual Meeting Abstracts 87: 292. Wilkie, D. S., B. Curran, R. Tshombe, and G. A. Morelli. 1998. Managing bushmeat hunting in Okapi Wildlife Reserve, Democratic Republic of Congo. Oryx 32: 131-144. Wilkie, D. S., and J. F. Carpenter. 1999. Bushmeat hunting in the Congo Basin: An assessment of impacts and options for mitigation. Biodiversity & Conservation 8: 927-955. Yumoto, T., T. Maruhashi, J. Yamagiwa, and N. Mwanza. 1995. Seed-dispersal by elephants in a tropical rain forest in Kahuzi-Biega National Park, Zaire. Biotropica 27: 526-530.  - 122 -  APPENDICES  - 123 -  Appendix 2.1. Correlation matrix for selected variables at the sampling site level for the entire dataset (a) and the park dataset only (b) in the Odzala National Park. Correlation coefficients are shown in the top right half. Bivariate plots and fitted smooth regression lines are shown in the bottom left half.  - 124 -  20 40 60  0  !40 0 20 60  10 30 50  !  !  !  !! ! !  !  !  ! ! !!  !  ! ! ! !  !  !! ! !  ! ! ! !! !  !  ! ! ! ! ! ! !! ! ! !  !! ! !! ! !  !  !  !  !  ! ! !  !  !  !  !  ! ! !  !  !  !  !  !  !  !  0  10  ! ! !! ! ! ! ! ! ! !! ! ! ! ! ! !! ! ! !!  !  !  !  !  ! ! ! !! ! ! ! ! ! ! ! !! ! ! !! ! ! ! ! ! !  ! !  !  !  ! ! !! ! ! ! ! !!!! ! ! !! !!! ! !! !! !! !!!  !  !  !!  !  !  !  !! ! ! !! ! ! ! ! ! ! ! !! ! ! ! !! !! ! !!  !  ! ! !! !  !  ! ! ! !! ! ! !  ! !  ! ! ! ! ! !  !  !  !  ! ! ! ! ! ! !  !  !  !  !  !  !  !  !  !  ! !!  !  !  !  !  !  !  !  !  20  !  !  !  ! !! !  !  !  !  30  !  !  !  !  !  !  40  !  !  !  !  !  !  ELEPHANT DUNG  !  !  ! ! !  !  !  ! !  !  !  !  !  ! !  ! ! ! ! ! ! ! ! !!  ! ! ! !  ! !  ! !  !  !  !  !  ! ! ! ! ! !! ! !! ! ! ! ! ! ! ! ! ! !! !  !  !  !  !  !  !  !  !  ! ! ! ! ! ! ! !!! ! !! ! !! !! !! ! !! ! ! ! !! !! !  ! !! !! ! ! ! ! ! ! ! ! ! ! ! !! ! !! !!! !! ! ! ! !!  !  !  !  !  !  !  !  !  !  !  patrol  0.53  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  ! !  !  !  !  !  ! ! !  ***  !  !  !  !  ! !  ! !  ! !  !  !  !  !  !  10 20 30 40 50 60  ! ! ! ! ! ! !! ! ! ! ! !! ! !!! ! !! ! ! ! ! ! !! !  !  0  ! ! !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  ! !  !  ! !  !  !  !  !  !  !  ! !  !  !  !  !  ! !  ! !  !  !  !  !  !  !  !  !  ***  **  !  !  !  !  0 10 20 30 40 50 60 70  !!  !! !  !  !  !  ! ! ! !  !  !  !  !  ! !  !  !  !  ! ! !  ! !! !  !!  ! !!  !  !  !  ! !! ! ! !  ! !  !!  !  !  ! !! ! ! ! !! !! !! ! ! ! ! ! !!!!!! ! !!! ! ! !!  ! !! !! ! ! ! ! ! ! ! ! ! ! !! ! !! ! ! ! ! ! !! ! !! ! !  !  infra  0.68  0.39  !  ! ! ! ! ! !! ! !  !! ! !  ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !  !!  !!  !!  !  !!  ! ! ! ! ! ! ! ! ! !! !  ! !! ! ! !! ! ! ! ! ! !! !! ! !  ! ! ! !! !! !! ! ! !!!! !! ! ! ! ! ! ! ! ! ! !! !  !  !  !  !  !  !  !  !  !  !  !  !  !  !!  !  !  !  !!  !  ***  ***  *  !  !  !  !  !  !  20 40 60  park  0.62  0.84  0.38  !40 !20 0  (a) Entire dataset  0  !!  !  ! !  !  ! ! !!! !!! ! !!  !!  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  10 20 30 40 50 60  !  !  ! ! ! ! ! !!! ! !!  !!  ! !  ! ! ! ! ! !!!! ! ! ! !! !! ! ! !! ! !!! ! ! !! !!  ***  ***  ***  **  forest clearing  0.86  0.67  0.83  0.42  !!  ! !  ! !  ! !  !  ! !  !  ! !! ! ! !  !  ! ! ! !  !  ! ! ! !  village  0.17  0.55  0.035  0.19  0.051  !  !!! !! !  !  ***  ! ! !  !  10 20 30 40 50 60  0  ***  *  ***  **  10 20 30 40 50 60  road  0.92  0.33  0.72  0.16  0.44  0.11  0 10 20 30 40 0 20 40 60 20 40 60 0 20 40 60 0  - 125 -  0 5 10 20  !20  !40  10 30 50  !  0  !  !  !  !  !!  !  !  !! !  !! ! !  !  !!  !  ! ! !  !!  !  !  !  !  !  !  !  !  !  !  !  !  !  10  !  ! ! !! ! ! ! ! !!  ! ! !  !  !  !  !  ! !  !  ! ! ! !! ! ! ! !  ! !  !  !  !!  !  !!  !!  ! ! !  !  !  ! !  !!  !  !  !  ! ! ! !  !  ! ! ! !  !  ! !  !  !  !  !  !  ! !  !  !  !  ! ! ! ! ! ! !! ! ! !  !! ! !! ! !  ! ! !!  ! !!  !  !!  !  ! ! ! ! !! ! !! !  !  !!  !  ! !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !!  !  !  !  ! !!  !  20  !  !  !  !  !  !  !  30  !  !  !  !  !  !  !  !  !  !  !  !  40  ELEPHANT DUNG  !  !  5  ! ! !  !  !  !  !  ! ! ! !  !  !  !  !  !  !  ! !  ! !  ! !  !  !! !  !  !  !  !  !!  !!!  !  !  !!  !  !  !  !  !  !  !  !  ! !  !  !  !  !  !  !  !  !  !!  !  !  !  !  !  ! !  !  !  ! !  !  !  !  !!  !  !  !  !  !  !  !  !  !  !  !  !  !  !!  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  ! !  !  !  !  !  !  patrol  0.29  !  !  !  !  ! !  !  !  !  !  !!  !!  !  .  10 15 20 25  ! ! ! ! ! !! !  !!!  !  !!! !!! ! ! ! ! !!  !  0  !  !  !  !  !  !  !  !  !! !  !  ! !  !  !  !  !  !  !  !  !  !  !  !  ! ! !  !  !  !  !  !  !  !  !  !!  !  !  !  !  !  !  !  !  0 10 20 30 40 50 60 70  !!  !  !  ! ! !  ! !  !  !  !  ! !! !  !!  ! !!  !  !! !  ! ! !  ! !  ! ! ! !  !  !  !! ! !  ! !  !  ! !! ! ! !  ! !  !!  !  !  ! !!  !  ! !  !  !!!!! ! ! !  !  !  !! !! !!  !! ! ! !  ! !  !  !!  !  !!  infra  0.15  0.12  !  !  !  !  ! !  !  !  ! ! ! ! ! !  ! !  ! ! ! ! ! ! ! ! !  !  !  !  !  !  ! !  !  ! !  ! !  !  ! !! !!  !  !  ! ! !  ! ! !  !  !  !!  !  ! !  !  !  ! !  ! !  !  !  park  0.016  0.17  0.07  !  !  !  !  !  !  ! ! ! !  ! !  !  ! !  ! !!  !  !  !  !  !  !40 !30 !20 !10  (b) Park only  !!  !  !!  !  !  !  !  !  !  !  !  2  !  !  !  !  !  !  !  !  !  !  !  ! ! !  ! !  ! ! !  !  !  4  !  !  ! ! !  !  !  ! !  ! ! !  !  !  !  ! ! !  !  !  !  ! ! !  !  !  6  !  !  ! !  !  !  ! !  8  forest clearing  0.16  0.13  0.09  0.22  !  !  ! !  10  !  !  !!  ! !  ! !  ! !  !  ! !! !  ! ! ! !  !  ! ! !  village  0.19  0.88  0.28  0.022  0.18  !!! !! !  !  ***  .  ! ! !  !  10 20 30 40 50 60  ***  ***  10 20 30 40 50 60  road  0.97  0.13  0.91  0.26  0.01  0.16  0 10 20 30 40 0 20 40 60 2 4 6 8 10 10 30 50  - 126 -  Appendix 2.2. Correlation matrix for selected variables at the segment level for the entire dataset (a) and the park dataset only (b) in the Odzala National Park. Correlation coefficients are shown in the top right half. Bivariate plots and fitted smooth regression lines are shown in the bottom left half.  - 127 -  20 40 60  0  !40 0 20 60  10 30 50  ! !!  !!  !  !  !  !  !  !  !  !  !  !  !  !  ! !  !  !  !  !  !  !  !  !  !  ! !  !  !  !  !  !  ! ! !  !  !  !  !  !  ! ! ! ! !! ! !  !! !  !  !  !  !  !  ! !! ! !  !  !  !  !  !  !  ! !  !  ! !  !  ! !!  !  ! !  ! !  !  ! ! !  !! !!!!! ! ! !! !!! !!!! ! !! !!! !! ! ! ! !! !! !! ! ! !! !! ! !! !! ! ! !!! !!!!!! ! ! ! !! ! ! ! ! ! !!!! ! ! ! ! ! ! !! !!!! ! ! ! !! ! ! ! !! ! ! !! !!!! !! ! !! !! ! ! !! ! !  !! ! ! ! ! ! !  !! !!! !! ! ! !! !!! !!!! !! !!!! ! ! ! !! ! ! ! ! ! !! !! !! !!! ! ! ! ! ! ! !! ! !! !! ! ! ! !! ! !!! ! ! !! !! ! ! ! ! ! !! !! ! ! !!!! ! ! ! ! !! ! ! !! !! ! !! ! ! ! ! !! !! ! ! ! ! !!  ! !! ! ! ! !  !! ! ! !!! !!! ! ! !!!!! ! !!! ! ! ! ! !!! ! ! !!! ! !!! ! ! !! ! ! !! !! !! !! ! !! ! ! !! !! !! ! !! ! !! ! !! !!! !!!! ! ! !! !! !! !!! !  !! !! ! !! ! !! !! ! !!!!!!!! !! !! !! ! !!! !!!!!! !! !! !!!! !!!! ! !!! !! !!! !!!! !!! !! !! ! !!!! !!! !! ! !!! !!! !!  ! ! !!! !!! !!!! ! ! !!! ! ! !!! ! !! ! !!!!! !! !! !! ! !! !! ! ! !!! ! ! ! ! ! !! ! !!! !! ! !! ! !! ! !!! ! !!!! !! ! ! !! ! ! ! ! ! !  !!!!  ! !  ! !  ! ! ! !  ! !  ! !  patrol  0.41  !!!  !!!  !! !!  !! ! !  !! !!  ! !!!  !!!  !!!  ! ! ! !  ! ! ! !  ! ! ! !  ! !  ! ! !!! !! !!  !! !!  !! !!  !! ! !  !! !!  !! !! !! !!  !!! !!!  !! ! !!  ***  !! !!  !! !!  !! !!  !!!  !!!  10 20 30 40 50 60  !!! ! !! ! !!! !! ! !!! !!! ! !!!! !  0 100 200 300 400 500 600  ! ! ! ! !! ! ! !! ! ! ! !! !! ! !! ! ! !!!!!!! ! ! !! ! !! !! ! ! ! !! ! ! !! ! !! ! ! ! ! ! ! ! !! !! !!! ! !! ! !! !!!! ! !!!!! ! ! ! ! ! ! !! ! ! !! ! !!!!! ! ! ! !! ! ! !! ! !! ! ! ! ! !!! ! !! ! ! ! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! !  ! ! !! ! ! !! ! ! ! !! ! ! ! !! ! ! ! ! !!!!!!! ! ! ! ! ! !!! ! !! ! ! ! ! !! ! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! ! !! ! ! ! !! !! ! ! ! ! ! ! ! ! ! !! ! !! ! ! ! ! ! !!! !! ! ! ! ! !! ! ! ! !! ! ! ! !! ! ! ! !!! ! ! ! !! ! !! ! ! ! ! ! ! !! ! ! ! !  !!!! !!! !!! ! ! ! ! ! ! ! ! ! !! ! !! ! ! !!!! ! !! ! ! !! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! !! !!!! ! ! ! !! ! ! ! !! !! ! !! ! !! ! ! ! !! !! ! !! ! ! ! !! ! ! !! !!  ! ! ! ! !!!  ! ! !! ! !! ! ! ! ! ! ! !!  !! ! !! ! !! ! ! !! ! !!! ! !!! !!!! ! !!! !! !! ! !! !! !! !! !! ! ! ! !! !!! !! !!!! ! ! !! !!!!!! !! ! ! ! ! ! ! ! !!! ! !! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! !!!! !!! !!!! !  !! ! !! !  ! !! !! ! !! ! !  !! !! ! ! ! ! ! ! ! !! ! ! ! !! !! ! ! !! !!! ! ! !!! !! !!! !! ! ! !! ! ! !!! ! !!! !!! ! ! ! ! !! ! ! ! !! ! !!! !!! ! ! !!!!! ! ! !! !! ! ! !! ! ! ! !! !!!!! ! ! ! !!!! ! !! ! ! !! ! ! ! ! !! ! ! ! !! !! ! !!! ! ! ! !!  !!!  ! ! ! !!!! ! ! ! ! !!!! ! !!! ! ! ! ! ! !!! ! ! ! !! ! ! !! ! !! !! ! ! ! !! ! ! !! !!! !!! ! ! ! !!!! ! ! ! !! !!!! !! !!! ! !! ! ! ! ! !! ! ! !!!! ! ! ! ! !!! ! ! !! ! !!!  !  !! ! ! ! !! ! !! ! !! ! ! !!  ELEPHANT DUNG  0  !! ! !!  !!! !!  !!!  ***  ***  ! ! !!  ! !! ! !! ! !  !! !!  ! !!  ! ! ! ! ! !! !  ! ! ! ! !! !! ! !! !  !! !!  ! ! !!  0 10 20 30 40 50 60 70  ! !! ! ! ! ! ! !! !! ! !! ! !! ! !! ! !! ! !!! !!!! ! ! !! !! ! ! !! ! !! !! !! ! ! !! ! ! ! ! ! ! ! ! ! !! ! !! ! ! ! ! ! ! ! !! !!! !! ! !! !! !!! !! !! !!! !! ! !! !  !!! ! !!  !! !  ! ! ! ! ! !! ! ! !! ! ! ! ! ! ! !! !! ! !! ! ! !! ! ! ! ! ! ! ! !! ! ! !!! ! ! ! ! ! ! ! ! ! ! ! !! !! ! ! !! !! ! ! ! ! !! ! ! ! ! ! ! ! ! ! !!! ! ! !!!! ! ! ! ! !! !! !!! ! !!! ! ! ! !! ! ! !! ! ! !!!! ! ! !!! !! !! !! ! ! !! !!! !! !!! !!!! !! !  !! !! !!!! ! ! ! !!! ! ! !! !! ! !! ! ! ! !!!! ! !!! !! !!! !!!! ! !! !! ! !!!!! !! !! ! !! ! !! !! !! ! ! ! ! ! !! ! ! ! !! !! !!! !! !!! !! !! !! !! !!! !! !!! ! !! !!! !!! !!!  ! ! !!  ! !!  !! !!!! !!! !! !! !!!! ! !!!! !!!! !!! ! !!!! !! !! !! !!! !!! !! !! !! ! !!! ! !! !!! !!!! !!! !! ! !! ! ! !! !! !! ! ! !! !!! !! ! !! !!! ! !  !!  infra  0.42  0.30  ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! !! ! ! ! ! ! !! ! ! ! ! ! ! !! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! !  ! ! ! ! ! ! ! !! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! !! ! ! ! ! !! ! ! ! ! !! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !  ! ! ! ! ! ! ! ! ! !! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! !! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! !! ! !! ! ! ! ! ! ! ! ! ! !  ! !  ! ! ! !  ! ! ! !  park  ! ! ! ! ! ! !  ! ! ! !  ! ! ! !  ! ! ! ! ! ! ! !  ! ! ! ! !! !  ! ! ! ! ! ! !!  ! ! ! !  ! ! ! !! ! ! !  ! ! ! ! ! ! ! !  ! ! ! !! ! ! ! ! ! ! !  ***  ***  20 40 60  0.10  0.84  0.29  !40 !20 0  (a) Entire dataset  !! !!  !! !! !!!!  !!!!  !!!  !! !!  !! !!  !! ! !  ! !!! ! ! !!  ! ! !!  ! ! !!  0 10 20 30 40 50 60  ! ! ! ! !! !!! !! ! ! ! !!! !! !!! !!!! !!! !! ! !! !! ! !! ! !! !! ! ! !! ! ! !! !! ! !! !! ! ! ! ! ! ! ! !! ! ! ! !!!! !! ! !! ! ! !!! !!! !! !! !! ! !! !!! !!! !! ! ! ! ! !!  ! ! !! ! !!! ! ! !!!! ! !!! !! !!! !!! ! !! !!! ! !! !!!!! !!! !! ! !! ! !! ! !!! !! !! ! ! !!! !! ! ! !! ! !! ! ! ! !!! !! !!! ! !! ! ! ! !! !!! !!! !! ! ! ! !! !!! ! ! ! ! !! !!  ***  ***  ***  ***  forest clearing  0.87  0.36  0.87  0.34  ! !! !!! !! ! !!! ! !!!  *  ***  ***  *  !!! ! !! !!!! !!!! !! ! ! !! ! !!!!! ! ! !! !! !!! !! !! !! ! ! ! !! ! !! !! !! !!! !! !! ! !! !! !! !!!!! !!! !! !!! ! ! !! !! !! !!! !! ! !!! ! !! !! !! ! ! ! !!!!  village  0.18  0.56  0.50  0.17  0.038  10 20 30 40 50 60  0  ***  ***  ***  ***  ***  10 20 30 40 50 60  road  0.94  0.34  0.70  0.42  0.37  0.054  0 200 400 600 0 20 40 60 20 40 60 0 20 40 60 0  - 128 -  0 5 10 20 30  !40 !20  10 30 50  ! ! ! ! ! ! !! ! ! ! ! ! ! !!! !! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! !! ! !! !! ! ! ! !! !! !! ! ! ! ! ! !!! ! ! ! !! ! ! ! !!! !! ! !! ! ! !! ! ! ! ! ! ! !!! ! ! !! ! !! ! ! ! ! ! ! ! !!! ! !!  ! !! !  ! !!!!! ! ! !  !  !  ! !!  !  !  !  !  !  !  ! !  !  !  !  !  !  ! ! !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  !  ! !  ! !  !  ! !  !  !  !  !  ! !  !  ! !  !  !  !!!  !!!  ! !!  ! ! !!!  !!!!  patrol  0.25  !! ! ! ! ! ! !  !!!  !! !! !!  ! !!!  ! !!  !  ! !  !  !!  !!!  !!  !  !  ! ! !!!  !!!!  !  !!  !  !!!  !!  !!! !  !! !!!!!  !!!!  !!  ! !  !!!! ! !! ! ! ! ! !! !! !! !!! !! !! !! !! ! ! !!!!!! ! ! ! !!!! ! ! ! ! !! ! !! ! !!!! ! ! ! !! ! ! ! !! ! ! !!!! !! ! !! ! ! ! !! !  ! ! ! ! ! ! ! !  ! ! !! !  ! !! ! !! !! ! !  !!!! ! ! ! ! !! !!! !!! !! ! !! ! !!!!! !!! !! ! ! ! ! ! ! ! ! ! !! !! !! ! !! !! !! ! ! ! !! !! ! !! !! ! !! !  ! !! ! ! ! !  !  !  !  !!!  !  !!!  !!!!  !!!  !!!  !!  !!!!!!!!  !!!!!!!!!  !!  !!  !  !!! ! !!!!! ! !!!! ! ! !!  !!  ! ! !! ! ! !! ! !! !! ! ! ! ! ! ! !! !! !! !!!! ! ! ! !! !!! !! ! ! !!! ! ! ! !!!!! !! ! !! !! ! ! ! ! ! ! ! ! ! ! !! !! !!  ! !!  !!  !! ! !  !!!  !!  !! ! !  !!  !!  !!!! !  !!!  !  ! !!!  ! !!  !  !!  !!!  **  !  !  !  !  !  !!  !!!  !!  !!  !!  !!!!  10 15 20 25 30  !!!!  !!!  !!!  5  ! ! !!! ! !!!!! !! ! !!!! !! !! !! !! !! ! !! ! !! !! ! ! !! !!! ! ! !! ! ! ! ! !! !!! !!!! ! !! !! ! !! ! ! ! !! !  ! !  ! !! !  0 100 200 300 400 500 600  ! ! ! ! !! ! ! !! ! ! ! !! !! ! !! ! ! !!!!!!! ! !!! !! ! ! ! ! !! !! ! ! !! ! !! ! ! ! ! ! ! ! !! !!! ! !! !! !! !! ! !! ! ! ! !! ! ! !! !! !! !! ! ! ! !! ! !! !! ! ! ! ! !!! ! !! ! ! ! !!! ! ! ! ! ! ! !  ! ! !! ! ! !! ! ! ! !! ! ! ! !! ! ! ! ! !!!!!!! ! ! ! ! !!! ! ! ! ! !! ! ! ! ! !! !! ! !! ! ! !!! ! ! !! ! !! ! ! ! !! !! !! !! !! ! ! !! !! ! ! ! ! ! ! !! !! ! !! !! ! ! ! ! ! !!! ! ! ! ! ! ! ! ! !! ! ! ! !  ! ! !!!!!  ! !! !! !! !! !!  !! !! ! !! !! !! ! !! ! !! ! ! !!  !! ! ! !!  ! ! ! ! ! ! !!! ! !!!  ! ! ! ! !!! ! !  ! !  !  !  !  !!  !  ! !  !  !  !  !  !  !  !!  ! ! !!  !  !  !  !  !  !  !  !! !  !  ! !!  !  !  !  !  !! ! !  ! ! ! ! !!!! !! !! ! !!! ! ! ! ! !! ! ! !! !!!! !! ! ! !! !  ! ! !! ! !!!! !! ! !! ! ! ! !!! !! ! !! !! !! !! ! !! !! ! ! !! ! ! ! ! ! ! ! !!!! ! ! ! ! ! !!! !!!! ! !! ! ! !! ! ! ! ! ! ! ! ! !! !! !  !!! ! !! !! ! ! ! ! ! ! ! !! !! !! ! !! ! ! !! !!! !! ! ! ! ! ! !!!! !!!! ! ! ! ! !! ! !!! !!! ! ! !!!!! ! ! !! !! ! ! !! ! ! ! !! !!!!! ! ! ! !!!! ! !! ! ! !! ! ! ! ! !! ! ! ! !! !! ! !!! ! ! ! !!  !  !  ELEPHANT DUNG  0  !!  !  !! !  !! !!! !! !  ! ! ! !!! !!!  !  ! !  ! ! !!!! !  ! ! ! !  !!  !  !!  !!!  !! !!  !!  !  ! ! ! ! !! ! !  !! !!  !! !!  !!!  !! ! ! !!  ! ! ! !  !! !!  ! !! !  ! ! ! !  ! ! ! !  ! ! !  !  ! !!  !!  !! ! ! !! ! !  ! !!  ! ! !  !  !!  ! ! ! !  ! ! ! !  ! !! !! !  !  !  !!!  ! ! ! ! ! ! !  **  *  ! !  !!  !! !  !!  !  ! ! !! ! !!  !! ! !  !! !! ! !! !  ! !! !! ! !! !  !  !  !!  ! ! !!  0 10 20 30 40 50 60 70  ! !! ! ! !! !! ! !! ! !! !! ! !! !!! !!!! ! ! !! ! !! ! !! ! !! !! !! ! !!!! ! !! ! ! ! ! ! ! !! !!! !! ! !! !!! !! !!! !!  !!! ! !!  !!! !!!!  ! ! ! ! ! !! ! !!! ! ! ! ! ! ! ! ! !! !! ! ! !! ! ! ! !! ! !!! !!! !! !! ! ! ! !! !! !! !!! !!! ! ! !! ! !! ! !!!! ! !!!  !  !!! ! ! ! !! !  !  ! !! !!  !! ! ! ! !  !! !!  ! ! !!  ! !!!! !  !!  !!!  ! ! ! !!!  ! !!  !!! ! ! ! !! !! ! ! !!! !! ! !!!! ! ! !! !!! ! ! ! ! ! ! !!! !! !! ! ! ! !! !!! ! ! !! !! !!  !! !!! !! ! !! !  infra  0.23  0.21  !!  !  !!  ! !! !!! !! !! !!!! ! !!! !! !! !!! !! !!!! !!! !! !!  ! !! !!! !! !! !! !! ! ! !!! !! !!!!!!! !! !! !! !  !!  !  !!! !!!  !!!  ! !!!!  !!!!  !  !  ! ! !  ! ! !  ! !!  ! !!  !  !!! !!!!!  !  ! !  !!!  ! !!! !! !!  !!! !! !! ! ! !! !!! ! ! !! ! ! ! !!!! !! !!  !!!! !! !  ! !!! ! !! !! !!!! !! ! !! !!! ! ! ! !! ! !! ! !  !!  !!  !!  !! !  *  ! !  !  !  ! !!!  !! !! !! !!  !!! ! ! ! ! !!! !! ! ! ! !! !! !! !! !! ! ! !!! ! !!!  !! !!! ! !! !! ! !! !! ! ! ! !! ! ! ! !! !! ! ! !! ! !! !! ! ! !!!!!! !! ! !! !! !!!  !!  !!!!!  !  !!  !  !!!!  !  !!  !!!  !  !  !! !!!  !!  ***  ! !  !  !  ! ! !!  park  0.53  0.17  0.04  !40 !30 !20 !10  (b) Park only  !  !  ! ! !  !  !  !  !  ! ! !  !  ! !  !  !  ! ! ! ! ! ! ! ! !  2  ! ! ! ! ! ! ! ! !  ! ! ! ! ! ! ! ! !  !  !  ! ! ! ! ! ! ! ! ! !  ! ! !  ! ! ! ! !  ! ! ! ! ! ! !  ! ! !  !  ! ! ! !  !  ! ! ! ! ! ! ! ! ! ! ! ! !  ! ! !  *  *  4  ! !  !  ! ! ! ! ! ! ! ! ! !  ! !  ! ! ! !  ! ! ! ! ! ! ! ! ! ! ! ! !  ! !  ! ! ! !  ! ! ! !  ! ! ! ! ! ! !  ! ! ! ! !  ! !  ! ! ! ! ! ! ! ! !  ! ! !  6  ! !  !  ! ! !  ! !  ! !  ! !  ! ! ! ! !  ! !  !  ! !  ! !  ! !  !  ! ! ! ! !  ! !  8  !  ! !  ! ! !  forest clearing  0.077  0.053  0.17  0.21  ! ! ! !  !  !  ! ! !  10  ! !  !  ! !  !  !! !! ! !  !! !!  !! !! ! !!!  ! !! !!! !! ! !!! ! !!!  ***  ***  !!! ! !! !!!! !!!! !! ! ! !! ! !!!!! ! ! ! ! !! !!! !! !! ! ! !! !! !! !! !!! ! !! !! ! ! ! !!!!! !! !  village  0.11  0.88  0.67  0.044  0.12  10 20 30 40 50 60  ***  ***  ***  10 20 30 40 50 60  road  0.97  0.06  0.91  0.69  0.012  0.11  0 200 400 600 0 20 40 60 2 4 6 8 10 10 30 50  - 129 -  Appendix 2.3. Comparison of the empirical Cumulative Distribution Function (CFD) of elephant dung encounter rates (dung piles/km) with the normal and poisson Cumulative Distribution Functions.  normal CDF  empirical CDF  poisson CDF  - 130 -  Appendix 2.4. GAM plots: Results of the Generalized Additive Model fits to elephant dung encounter rates within the entire study area (model MS3) (a) and the National Park only (model MP4) (b) in the Odzala National Park. The plots show the smooth function of dung encounter rate estimates (solid line) and confidence intervals (dashed lines) conditional on each covariate. The X axis shows the covariate name and the Y axis is labeled (cov, edf) where cov is the covariate name and edf is the estimated degrees of freedom for the smooth function. A rug plot just above the X-axis indicates the density of observations  - 131 -  s(LAT,8.14)  s(PARK,8.85)  100  0.0  !40  0.2  !20  0.4  0 PARK  LAT  0.6  20  0.8  40  1.0  60  1.2  0  14.4  10  (a) Model MS3 (entire study area) 100 50 0  50  0  !50  !100  100  50  0  !50  !100  s(LONG,6.69) s(PATROL,8.93)  !50 !100 100 50 0 !50 !100  - 132 -  14.6  20  PATROL  30  LONG  14.8  40  15.0  50  15.2  60  s(LAT,6.64)  4  2  0  !2  0.4  !40  0.6  !30  0.8  PARK  LAT  !20  1.0  !10  1.2  !4  !6  4  2  0  !2  !4  !6  s(PARK,3.76)  !4 !2  0  2  4  !6  !2  0  2  4 0  14.7  (b) Model MP4 (park only)  s(LONG,7.11) s(PATROL,7.28) !4 !6  - 133 5  14.8  10  14.9  15.0  PATROL  15  LONG  20  15.1  25  15.2  30  Appendix 3.1. Selected DISTANCE analysis models (models for okapi and duikers used pooled data across survey periods). Species  Survey  Model  Truncation distance (m)  Elephant  1995  half-normal, cosine adjustment order 2  4.5  2006  half-normal, cosine adjustment order 2  4.5  1995  half-normal, cosine adjustments order 2  4  2006  half-normal, cosine adjustments order 2  4  1995  half-normal, cosine adjustment orders 2,3,4  2.4  2006  half-normal, cosine adjustment orders 2,3,4  2.4  1995  half-normal, cosine adjustment order 2  2.5  2006  half-normal, cosine adjustment order 2  2.5  Yellow Backed 1995  half-normal, no adjustments  1.6  2006  half-normal, no adjustments  1.6  Okapi  Small duikers  Red duikers  - 134 -  Appendix 3.2. Amount of forest loss per quarter degree grid cell between 1990 and 2000 in and around the Okapi Faunal Reserve.  Legend Park Boundary  Road  .  Forest loss (in %) 0.00 - 0.17  2.46 - 3.67  0.18 - 0.56  3.68 - 5.24  0.57 - 1.06  5.25 - 8.71  1.07 - 1.65  8.72 - 13.37  1.66 - 2.45  13.38 - 20.20  30  15  - 135 -  0  30 Kilometers  Appendix 3.3. Ecozones (habitat types) in the Okapi Faunal Reserve.  Legend  .  Park Boundary Road Mixed forest, Mbau, Swamp forest Outcrops Hill forest Forest / Savannah ecotone Non forest  20  10  - 136 -  0  20 Kilometers  Appendix 3.4. Slope in degrees in the Okapi Faunal Reserve (from SRTM data).  .  Legend Park Boundary Road  Slope High : 28.931923 Low : 0.000000  20  10  - 137 -  0  20 Kilometers  Appendix 3.5. Predicted density maps (animals per km2) for each ungulate species in 1995 and in 2006 in the Okapi Faunal Reserve using Kriging. Sampling locations are shown on both the 1995 and 2006 maps. Increase, decrease and no change in densites in 2006 compared to 1995 are shown on the 2006 maps.Color codes for densities are on same scale for each species in 1995 and 2006 but they are on a different scale between species.  Legend Park boundary  !  Sampling locations 1995 Sampling locations 2006  Green Zone  –  Road $  ( !  Guard post 2006  E  Decrease Stable Increase  Density  Zero Low  High  –E––Zone – Red – E – EGreen – E–– Zone$ – – Epulu $E $– $E ––– –$ $– Red Zone ! (  ! ( !! !  (! !! ( ! !! ! (! ! (! ! ! !  (! ( (! ! !( ! ! ( ( ! ! ! ! ( ! ( (! ! ( ( ! ! ! ( ! ( ! ! ( ! !! ( ( ! (! ! ( ! !! ! ! ! ( ( ( !  - 138 -  - 139 -  !  !  !  !  !  ! ! !  !  !  !  !  $  !  !  ! ! !  !  !  !  !  !  !  !  !  ! !  Epulu  !  !  !  Red Zone  !  !  !  !  Green Zone  !  !  !  Red Zone  !  1995  !  !  !  !  !  !  !  !  !  !  ! !  !  (a) Elephant  $!(  – – –  ( ! ( ! ( !  –E – –  ( !  $  E  –  E  ( !  –  ( !  ( !  ( !  E  ( !  ( !  –$ –– E–$–E –  E  –  ( !  Epulu  –  ( !  Red Zone  ( !  Green Zone  –  –  ( !  Red Zone  ( !  2006  ( !  ( !  $!(  –  ( !  ( !  E  –  ( !  $ ( !  - 140 -  !  !  !  !  !  ! ! !  !  !  !  !  !  $  !  !  ! ! !  !  !  !  !  !  !  !  !  ! !  Epulu  !  !  !  Red Zone  !  !  !  Green Zone  !  !  !  Red Zone  !  1995  !  !  !  !  !  !  !  !  !  !  ! !  !  (b) Okapi  $!(  ( !  – – ( !  –  E ( !  –  ( !  $  –  ( !  ( !  ( !  –  ( !  ( !  EE E E  $ –!( $ E  – ( !  –– – E E  ( !  Epulu  E –  Red Zone  –  ( !  ( ! ! (  Green Zone  ( !  ( !  E  Red Zone  ( !  2006  ( !  E  $  E  E  E!(  ( !  E$ –  - 141 -  !  !  !  !  !  !  !  !  !  !  !  !  $  !  !  ! ! !  !  !  !  !  !  !  !  !  ! !  Epulu  !  !  !  Red Zone  !  !  !  !  Green Zone  !  !  Red Zone  !  !  1995  !  !  !  !  !  !  !  !  !  !  ! !  !  – $  (c) Small duikers  ( !  ( !  – ( !  –  ( !  –  –  ( !  $  ( !  –  ( !  ( !  –  ( ! $–  ( !  ( !  ( !  ( !  ( !  ( !  E $ EE  ( !  ( !  –– –  EE  Epulu  E –  Red Zone  ( !  ( !  –  Green Zone  –  –  Red Zone  E  E  2006  ( !  $!(  ( !  –  ( !  ( !  – –$ –  - 142 -  !  !  !  !  !  !  !  !  !  !  !  !  !  $  !  !  ! ! !  !  !  !  !  !  !  !  !  ! !  Epulu  !  !  !  Red Zone  !  !  !  Green Zone  !  !  Red Zone  !  !  1995  !  !  !  !  !  !  !  !  !  !  ! !  !  $!(  (d) Red duikers  –  ( !  ( !  –  – –  –  $  –  ( !  ( !  $  –  ( !  ( !  ( !  ( !  ( ! ! (  –  ( !  $  ( ! ! ( ( !  ( !  ( !  –– –  EE  Epulu  – –  Red Zone  –  ( !  ( !  ( !  Green Zone  E –  –  –  Red Zone  ( !  ( !  2006  ( !  $!(  ( !  – –  ––  –  $ ( !  - 143 -  !  !  !  !  !  ! ! !  !  !  !  !  $  !  !  ! ! !  !  !  !  !  !  !  !  !  ! !  Epulu  !  !  !  Red Zone  !  !  !  !  Green Zone  !  !  !  Red Zone  !  1995  !  !  !  !  !  !  !  !  !  !  ! !  !  $!(  (e) Yellow-backed duiker  ( !  E  E  – –  ( !  E – – –  $  –  ( !  ( !  $  ( !  ( !  ( !  ( ! ! (  ( !  ( !  ( !  ( !  – $ E  –  E  ( !  –– –  E ( !  Epulu  – –  Red Zone  –  ( !  ( !  Green Zone  ( !  –  ( !  Red Zone  ( !  2006  ( !  –  –  $!(  ( !  ( !  ( !  ( !  $ ( !  Appendix 3.6. Gam plots of the effect of each smoothed variable on estimated dung densities conditional on other variables included in the model. Estimates are shown by the solid line and confidence intervals by the dashed lines. The X-axis represents the value of the covariate and the Y-axis the smooth function. A rug plot just above the X-axis indicates the density of observations. Elephant (a) Elephant dung, 1995  - 144 -  (b) Elephant dung, 2006  (c) Elephant dung, 1995 / 2006  - 145 -  Okapi (a) Okapi dung, 1995  (b) Okapi dung, 1995 / 2006  - 146 -  Small duikers (a) Small duiker dung, 1995  - 147 -  (b) Small duikers dung, 2006  - 148 -  (c) Small duikers dung, 1995 / 2006  - 149 -  Red duikers (a) Red duikers dung, 1995  (b) Red duikers dung, 2006  - 150 -  (c) Red duikers dung, 1995 / 2006  - 151 -  Yellow-backed duiker (a) Yellow-backed duiker dung, 1995  (b) Yellow-backed duiker dung, 2006  (c) Yellow-backed duiker dung, 1995 / 2006  - 152 -  Appendix 3.7. Conflict timeline. A Chronology of Military Occupation, Elephant Poaching, and ICCN Control in the RFO (with permission from Dr. John Hart).  - 153 -  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0058315/manifest

Comment

Related Items