Antonio Belchior Anciaes Ferraz, “From Earth Observation to Biodiversity Action: Where Foundation Models Can Help—and Where the Growing Pains Lie”
Abstract
Abstract: Geospatial foundation models are well positioned to learn from the vast, multimodal, spatially explicit, and temporally dense observations provided by satellite remote sensing. So far, early community efforts have focused largely on generic pixel-level geospatial tasks—classification, segmentation, and interpolation—such as land cover mapping or canopy height estimation, often at limited time steps. Biodiversity science and conservation action require a demanding leap: mapping landscapes through key ecological dimensions such as structure, composition, and function, and translating these into biologically meaningful products such as species and community distribution, abundance, phenology, disturbance, habitat integrity, and connectivity. Beyond mapping, the field also needs models that can detect ecologically meaningful change and, more aspirationally, forecast biodiversity trajectories under anthropogenic pressures and management interventions. In this talk, I will argue that significant progress will depend on three priorities: (1) integration of existing and emerging satellite remote sensing with high relevance to biodiversity, including imaging spectroscopy, thermal data, and animal movement tracking; (2) stronger biological grounding through in situ observations and biodiversity databases; and (3) closer collaboration between biodiversity scientists and machine learning researchers to couple foundation models with ecological process understanding, in order to develop interpretable, transferable models and maps that can earn trust in conservation practice.
Antonio Ferraz is a Scientist at NASA’s Jet Propulsion Laboratory, where he leads the Strategic Thrust Area on Biodiversity Change, serves as Project Scientist for NASA’s Internet of Animals project, and represents JPL in the NASA FireSense. He focuses on how multimodal satellite time series can advance biodiversity science and ecological conservation across scales, from landscapes to global systems. He leads community-driven work to inform priorities for NASA’s Earth Science Decadal Survey by assessing end-user needs, with the goal of defining future satellite missions and data analysis tools. His work emphasizes cross-disciplinary collaboration across remote sensing, biodiversity, wildfire science, machine learning, while engaging with technologists and practitioners developing emerging tools such as acoustic sensors, animal tracking tags, camera traps, and eDNA.