Spatial Analysis of Chimpanzee Habitat Quality Based on Vegetation Food Availability in Gombe National Park, Tanzania

Chimpanzee (Pan troglodytes) is an Endangered species listed in IUCN Red List with a most recently estimated total population ranging approximately from 172700 to 299700. In east Africa, habitat loss and degradation resulting from agriculture development is the major cause of population decline. Gombe National Park is one among few remaining habitats for chimpanzees in East Africa. The park has been well-protected and maintained high vegetation cover compared to its surrounding environment.

Given its importance to the chimpanzee’s diet, the vegetation cover defines the spatial variation of chimpanzee habitat quality. The aim of this project is to evaluate chimpanzee habitat quality for Gombe National Park by mapping the spatial variation of vegetation food availability, for the ultimate goal of supporting chimpanzee habitat conservation planning and studies of chimpanzee feeding behavior and ecology. Two sections of work were done to achieve this goal: vegetation species distribution modeling and vegetation cover mapping: Vegetation Species Distribution Modeling Vegetation food availability is a significant determinant of chimpanzee habitat quality. Different food resources can be of significant difference in indicating habitat quality due to chimpanzee’s feeding preference. Therefore, a species-level investigation of the forest cover is especially important for a small scale habitat in order to assess the spatial variation of chimpanzee habitat quality.

In this project, I used MaxEnt model to predict the distribution of 24 chimpanzee vegetation food species in Gombe National Park with a fine spatial resolution of 10 m. Furthermore, a chimpanzee habitat quality surface was generated based on vegetation food availability by overlaying the distribution extents of 24 important vegetation species after weighting each of them with chimpanzee’s feeding preference. Further, biological relevance validations which correlate habitat quality with chimpanzee feeding time using linear regression models where performed to evaluate how good the model outputs capture information that influence chimpanzee feeding behavior.

The overall performance of species distribution modeling is relatively high with an average AUC of 0.85 and an average overall accuracy of 0.78. Significant correlation between predicted chimpanzee habitat quality and chimpanzee feeding time also indicates high predication accuracy of the habitat quality. Overall, this project proves that MaxEnt is capable in predicting spatial variation of vegetation species distribution in a small area with a fine resolution of 10 m, which has seldom been investigated in precious studies, and proposes a method to evaluate vegetation food availability at the species level and to assess chimpanzee habitat quality based on vegetation food availability. Vegetation Cover Mapping Remote sensing technique is an alternative approach to characterize habitat quality by identifying vegetation cover in a broad scale. However, it is usually limited to general distinction of forest and non-forest or broad vegetation types rather than species level classification. To address this limitation, my goal of the second section of this project was to develop an innovative method of producing vegetation cover map which incorporates vegetation species composition information in each vegetation cover class, and to generate such a biologically meaningful vegetation cover map for Gombe National Park.

The method I used to achieve this goal is a combination of traditional remote sensing classification and ecological data mining technique. Specifically, I performed a cluster analysis of the vegetation survey data for the purpose of generating vegetation cover classes based on their similarity in vegetation species composition, and further used the clustered survey data to train a supervised classification algorithm – Maximum Likelihood Classification. Additionally, an innovative semi-automatic post-processing workflow was proposed and used to correct for misclassification resulting from spectral noises in high resolution images, which combines an automatic sieve and clump process, an automatic “buffer zone majority” correction, and a manual cloud and shadow correction. The result of this section of work is a vegetation cover map of Gombe National Park showing seven vegetation cover classes named after their dominant vegetation species with high spatial coherency.

The combination of cluster analysis and remote sensing technique advanced the vegetation cover classification to the species-level classification. Further, the cluster analysis method provides an alternative way of vegetation class schema design where little local knowledge of vegetation assemblages is needed. Moreover, the semi-automatic post-processing improved the efficiency and accuracy of post-processing of vegetation cover classification.

In conclude, this project produced 24 vegetation habitat suitability maps and distribution extent maps, an overall chimpanzee habitat quality surface map, and a vegetation cover map of Gombe National Park. These products provide useful spatial information to support vegetation studies, chimpanzee behavioral studies, and habitat evaluation and conservation planning in Gombe National Park.