2020 ESA Annual Meeting (August 3 - 6)

COS 14 Abstract - Broad-scale spatial distribution of African bush elephant (Loxodonta Africana) using combination of machine learning algorithms

Lei Song and Lyndon Estes, Graduate School of Geography, Clark University, Worcester, MA
Background/Question/Methods

African bush elephants (Loxodonta africana) play a critical role in maintaining the biodiversity of African savannah ecosystem. The elephant population in the past few decades keep dropping rapidly because of poaching, human-elephant conflicts, and habitat loss, even with growing national and international efforts. To assess the conservation status and propose future strategies, this study estimates the spatial distribution of African bush elephants and analyzes the effectiveness of conservation areas across the African continent. We combined two presence-only species distribution models (SDMs): Maxent, and Isolation forest to estimate the spatial distribution and analyze its driving forces. Maxent is a density-based outlier detector using the maximum entropy algorithm, which is widely used as a SDM. Isolation forest is an anomaly-detection machine learning algorithm designed to use incomplete information from a dataset to calculate the probability of being anomalous, such that it has higher tolerance of imperfect detections and sampling bias as a SDM. This study also applies continuous Boyce index (CBI) to evaluate the modeled results and provide objective thresholds to further classify the probability map to categories: optimal, suitable, marginal and unsuitable habitat.

Results/Conclusions

Maxent provides estimated habitats that most closely approximate the realized niche, and Isolation forest gives results that moves more in the direction of the potential niche. Furthermore, the results show that Maxent has higher modeling error over the estimated suitable areas, while Isolation forest has higher modeling errors over the estimated unsuitable areas. Thus, the complementary combination of these two methods increases the modeling reliability. The estimated spatial distribution is consistent with the extent map of IUCN red list, yet further reveals an increase in elephant population in Southern and Western Africa and an decrease in Eastern Africa, especially in Tanzania and Mozambique. Based on the importance of environmental variables, transportation construction, precipitation and its seasonal variation are the major reasons for the population decrease. The future conservation efforts should focus on the temporal and spatial changes in precipitation and the human disturbances, even if it is necessary to consider several factors to conserve the elephants, such as habitat connectivity.