This graduate seminar surveys recent advances in developing and applying foundation models across the environmental and life sciences, including geospatial foundation models such as AlphaEarth, biodiversity foundation models such as BioCLIP, and genomic foundation models such as Evo 2. Through guided discussions of contemporary research papers, students will explore how these models learn representations of the geosphere and biosphere from diverse sources of data, including satellite imagery, geographic locations, species observations, and genomic sequences. Students will examine the underlying approaches to model development, the environmental and biological data used for training, and their emerging applications in environmental science, ecology, and biodiversity research. Particular emphasis will be placed on critically evaluating what these models learn, how their capabilities are assessed, and how training data, modeling assumptions, and observational biases constrain their scientific utility.

The seminar welcomes graduate students and advanced undergraduates from environmental science, ecology, biology, data science, computer science, and related disciplines. No prior programming or machine-learning experience is required, although students should be comfortable reading and discussing scientific literature and be willing to engage with research outside their primary discipline. Through weekly reading discussions, student-led paper presentations, and a final synthesis essay, students will develop the ability to explain the principles underlying these foundation models, critically evaluate their scientific capabilities and limitations, and synthesize emerging opportunities for foundation models in geospatial and ecological research.