A Statistical Model for Multisource Remote-Sensing Data Streams of Wildfire Aerosol Optical Depth.

This article focuses on a physics-informed statistical spatiotemporal model designed to estimate and predict wildfire aerosol propagation by integrating heterogeneous multisource remote-sensing data streams with the underlying advection-diffusion physical process. The model addresses challenges such as differing measurement biases, missing data, and errors across data sources, specifically applying to aerosol optical depth (AOD) data from the GOES-16 and GOES-17 satellites during the 2020 California Glass Fire. It incorporates a bias correction mechanism to account for model approximations and truncation errors, enabling improved accuracy and interpretability over purely data-driven or single-source approaches. Comparative analyses demonstrate the model's superior predictive performance and robustness against observational noise relative to existing data fusion and data-driven methods. The approach is applicable to other environmental and engineering processes governed by advection-diffusion dynamics with multisource observations.