High-resolution anthropogenic emission inventories with deep learning in northern South America

Air quality in northern South America faces significant challenges due to insufficient high-resolution emission inventories and sparse atmospheric studies. This study addresses these gaps by developing a novel framework that integrates high-resolution nighttime light data from SDGSAT-1 and multisource remote sensing datasets with deep learning techniques to downscale emission inventories. The refined inventories are coupled with meteorological inputs into the Weather Research and Forecasting (WRF-Chem) model, enabling precise simulation of pollutant dynamics. Validated against ground measurements from Colombia's SISAIRE monitoring network, demonstrates significant improvements in spatiotemporal accuracy, particularly for particulate matter (PM) and nitrogen dioxide (NO₂) with error reductions of 22–30 % and correlation coefficients increasing from 0.68 to 0.85. These findings underscore the critical role of satellite-enhanced inventories in resolving localized emission patterns and seasonal variability, such as dry-season PM₁₀ spikes (150 % increase from wildfires). The framework provides policymakers with actionable insights to prioritize mitigation in rapidly urbanizing regions and manage transboundary pollution. By bridging data scarcity gaps, this replicable methodology offers transformative potential for global air quality management and public health protection, advocating for expanded ground monitoring networks and real-time satellite data integration in future applications.