Building reliable global flora and fauna distribution models incorporating climate change, human pressures, and biotic interactions

Species distribution models (SDMs) predict the spatial distribution of individual species based on ecological and environmental data. The complexity of the task often leads to incomplete use of available information, e.g., records of species observations (Global Biodiversity Information Facility, GBIF), remote sensing data such as high-resolution satellite imagery, and other data on anthropogenic impacts such as habitat fragmentation and land-use change, including deforestation and the expansion of agriculture. Recently, however, machine learning methods are used to improve the performance and accuracy of SDMs. 

This project aims to construct accurate spatial and temporal distribution models for the world's flora and fauna by integrating complex multiscale data in AI models. When moving from species level to community level, biotic interactions are crucial for building an accurate distribution model. Starting with the most widely spread species, predictions for more localized species will successively be made taking interactions with wider-spread species into account by applying and adapting techniques used for joint-species distribution models (JSDMs). To quantify and reduce model uncertainty, multiple model variants will be combined into ensemble models.