.png){kind=small}
Monitoring the great wildebeest migration from space using deep learning
Ms WU Zijing
( Supervisor: Prof Peng Gong )
Abstract:
The great migration of over 1 million wildebeest in the Serengeti-Mara Ecosystem in East Africa is an iconic phenomenon that attracts many tourists worldwide every year. The movement of massive herds in search of forage and water is essential in sustaining the trophic interactions and biodiversity in both the terrestrial and aquatic ecosystem and underpins the local economy. But recent climate changes have introduced extreme weather changes to the system, altering the vegetation structure and threatening many animal species. Human disturbances like intensive livestock grazing and the changed fire regimes also bring uncertainty to the future of the wildebeest migration and the abundance and diversity of the wildlife. Monitoring the dynamics of wildebeest migration and the landscape changes is critical to understanding how the animal migration interacts with the environment so as to project their responses to such climatic and anthropogenic changes and improve species conservation planning.
Previous studies on wildebeest migration highly rely on Global Positioning System collar-tracked movement data of a rather small number of animals and small-scale aerial surveyed or camera trapped herd data. Meanwhile, information on landscape dynamics such as vegetation structure and fire burned area are also mostly constrained by the coarse spatial resolution, which can neglect small-scale changes that are connected to animal movement patterns. Larger-scale distribution data of wildebeest populations and finer-detailed landscape dynamics mapped in different aspects are still absent for more precise monitoring as found in the literature. Deep learning techniques combined with higher-resolution satellite images have proven successful in detecting smaller ground objects and mapping more detailed vegetation characteristics from space. This research aims to exploit the potential of satellite-based techniques to monitor the wildebeest migration and landscape changes at unprecedented scales and details in the Serengeti-Mara ecosystem in the past decade; and to understand the interactions between wildebeest and the resources of the environment. Specifically, the research will focus on the following objectives: 1) to map the distribution of migratory wildebeest herds in Serengeti-Mara ecosystem using submeter-resolution satellite imagery in multiple years; 2) to map the spatial and temporal changes in vegetation composition and structure together with fire burned area using meter-resolution multispectral satellite imagery; 3) to investigate the connections between wildebeest distribution patterns and spatio-temporally varied landscapes.
Comments