MODELING THE DISTRIBUTION OF AFRICAN SAVANNA ELEPHANTS IN KRUGER NATIONAL PARK: AN APPLICATION OF MULTI-SCALE GLOBELAND30 DATA

• Main Authors: W. Xu, B. Hays, R. Fayrer-Hosken, A. Presotto
• Format: Article
• Language: English
• Published: Copernicus Publications 2016-06-01
• Series: The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
• Online Access: https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLI-B8/1327/2016/isprs-archives-XLI-B8-1327-2016.pdf
• ISSN: 1682-1750; 2194-9034

The ability of remote sensing to represent ecologically relevant features at multiple spatial scales makes it a powerful tool for studying wildlife distributions. Species of varying sizes perceive and interact with their environment at differing scales; therefore, it is important to consider the role of spatial resolution of remotely sensed data in the creation of distribution models. The release of the Globeland30 land cover classification in 2014, with its 30 m resolution, presents the opportunity to do precisely that. We created a series of Maximum Entropy distribution models for African savanna elephants (<i>Loxodonta africana</i>) using Globeland30 data analyzed at varying resolutions. We compared these with similarly re-sampled models created from the European Space Agency’s Global Land Cover Map (Globcover). These data, in combination with GIS layers of topography and distance to roads, human activity, and water, as well as elephant GPS collar data, were used with MaxEnt software to produce the final distribution models. The AUC (Area Under the Curve) scores indicated that the models created from 600 m data performed better than other spatial resolutions and that the Globeland30 models generally performed better than the Globcover models. Additionally, elevation and distance to rivers seemed to be the most important variables in our models. Our results demonstrate that Globeland30 is a valid alternative to the well-established Globcover for creating wildlife distribution models. It may even be superior for applications which require higher spatial resolution and less nuanced classifications.