SG: Species Distribution Modeling on the A.I. frontier: Deep generative models for powerful, general and accessible SDM

Globally, millions of plant and animal species exhibit unique geographic distributions, influenced by their distinct yet intertwined biology and habitat needs. This pioneering initiative seeks to enhance predictive understanding of these patterns by employing advanced generative Artificial Intelligence (AI) methods, supported by comprehensive global environmental and species occurrence data. The transformative potential of generative AI has already been established with tools that can answer questions with human-like responses or create stunning images based on textual descriptions. The project will apply similar AI techniques to the task of predicting how species are distributed across environments on Earth. This is crucial for biodiversity conservation, and could help identify priority areas for protection and management, and predict species responses to climate change. The heart of the project is the creation of a foundation model - a versatile AI model that can be used by scientists, conservationists, and educators to better understand and protect the natural world. By training this model on a vast array of data, it will learn to mimic the patterns of species distributions, making predictions of where different species are likely to be found. Once developed, the researchers will release the model to the public, allowing for widespread use and continuous improvement. By empowering scientific communities with this tool, the project can collectively contribute to the preservation of biodiversity and the well-being of the planet. Leveraging a generative AI method called diffusion models – known for handling complex conditional probabilities – the researchers aim to develop a model capable of understanding complicated species-niche relationships within a high-dimensional environmental space, demonstrating versatility across a broad spectrum of species. To support this aim, the project will also create an extensive training dataset of unprecedented size using cleaned occurrence records from global databases, encompassing a wide range of birds, mammals, amphibians, and reptiles. This novel approach will enhance Species Distribution Models (SDMs), a traditional tool in ecology, evolutionary biology, and conservation, providing a version that shares information between species during fitting and can be fine-tuned for new datasets without retraining, ensuring accuracy even with minimal input. This generative AI-driven approach to SDMs promises to advance understanding of biodiversity, enabling accurate predictions and visualizations of species distributions across various landscapes and conditions, including data-scarce regions. Beyond conservation, this tool will serve educational purposes and foster public engagement with the natural world. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.