Remote Sensing in Ecology and Conservation Nature and Landscape Conservation
Mapping Middens: Multimodal Imagery Illuminates Rhino Space Use in Savanna Landscapes
Abstract
ABSTRACT Ecosystem engineering by many animal species creates distinct features across landscapes that shape biodiversity and ecosystem structure and serve as key signatures of animal space use. Remote sensing has been used to map the spatial distributions of such features across landscapes, enhancing our ability to characterize the biotic factors that create resource heterogeneity. However, some animal‐engineered features are challenging to reliably detect even with high‐resolution remotely sensed imagery, limiting our understanding of how ecosystem engineering shapes ecological processes and animal spatial behavior. In African savannas, communal dung middens created by rhinos are central to olfactory communication and nutrient redistribution, but their spatial distribution and the environmental factors shaping midden placement have not been quantified. We used coincident high‐resolution thermal, RGB, and LiDAR data to detect and map rhino middens in Kruger National Park, South Africa, and demonstrate that active learning using multimodal imagery can efficiently detect middens across large areas. We then analyzed midden spatial patterning and used remotely sensed habitat metrics to quantify drivers of midden placement across spatial scales. Although we found weak and inconsistent deviations from complete spatial randomness in midden spatial patterning, midden placement was not random but strongly associated with proximity to or density of animal paths, proximity to water, and low levels of vegetation cover. These midden sites likely offer ease of access in areas of high rhino traffic that further facilitates the broadcasting of social cues. We found no relationship between midden placement and terrain attributes, suggesting that middens are placed throughout the topographic landscape, providing new insights into rhino space use. We demonstrate how using multimodal remote sensing data can facilitate a scalable approach to detecting cryptic animal‐made features in heterogeneous landscapes, providing a foundation for studying broad‐scale impacts of overlooked ecosystem engineering features that can be used to inform conservation strategy.