EAGER: Artificial Intelligence (AI) and Machine Learning (ML)-Enhanced Multi-Molecular Proxies for Tracing Fire-Climate-Ecosystem Feedbacks

Wildfires are becoming larger, more frequent, and more destructive across western North America in response to drought and other changes in the regional environment. Although ancient organic molecules preserved in geological archives contain valuable information on past wildfire activity, scientists can currently interpret only a small fraction of these data. This project will pioneer the development of new artificial intelligence (AI) and machine learning (ML)-enabled methods to decode the vast range of naturally occurring molecules that can be used to reconstruct ancient wildfires and their ecological context. This research serves the national interest by providing insights into prevalence and extent of natural hazards such as wildfires and droughts, thereby accelerating the development of approaches to reduce vulnerabilities and increase community resilience to such threats. The project will also provide interdisciplinary training of the geoscience workforce at the intersection of AI/ML techniques with Earth science and analytical chemistry. This project will test whether AI/ML-driven molecular similarity networks can recover meaningful relationships among thousands of untargeted organic molecules and preserve predictive information about ecosystem processes and fire regimes. High-resolution mass spectrometry datasets generated from vegetation, soils, waters, and mineral archives spanning fire-prone landscapes in the western US will be integrated to represent a continuum in which organic molecules are produced by plants, altered by wildfire, transformed through biological and geochemical processes, transported across interconnected environments, and ultimately preserved in geological archives. Molecular features will be organized into families to extract information about vegetation composition, combustion products, wildfire activity, and post-fire ecosystem and preservation processes, enabling interpretation of previously uncharacterized compounds. The resulting molecular networks will be evaluated to determine whether integrated multi-molecular signatures provide more robust reconstructions of fire-environment-ecosystem interactions than conventional organic biomarker approaches. This research will establish the first scalable analytical and computational framework for linking molecular production, transformation, and preservation, opening new opportunities for reconstructing ecosystem dynamics, wildfire history, and Earth system processes from a wide range of natural archives. 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.