FOCUS: A Noise-Aware Geospatial Learning Framework for PFAS Contamination Mapping

Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants with significant public-health impacts, yet large-scale monitoring remains severely limited due to the high cost and logistical challenges of field sampling, and the difficulty of simulating their spread. As a result, scientific understanding of PFAS transport in surface waters is incomplete. At the same time, rich geospatial and satellite-derived data describing land cover, hydrology, and industrial activity are widely available, creating an opportunity for AI to integrate sparse observations with large-scale environmental context. We introduce FOCUS, a geospatial deep learning framework for PFAS contamination mapping that learns from sparse point measurements propagated over satellite-based raster data while explicitly accounting for the resulting label noise. Rather than assuming known governing equations, FOCUS incorporates priors derived from hydrological connectivity, land cover, source proximity, and sampling distance to model uncertainty in supervision. These priors are integrated into a principled noise-aware loss, yielding a robust training objective under label noise. Across extensive ablations, robustness analyses, and real-world validation, FOCUS consistently outperforms baselines including sparse segmentation, Kriging, and pollutant transport simulations, while preserving spatial coherence and scalability over large regions. Our results demonstrate how AI can support environmental science by combining large-scale geospatial data with sparse, uncertain measurements to enable reliable PFAS contamination screening in the absence of complete physical models.