Advancing Finely Resolved Adjoint Analysis of Global Premature Mortality From Ambient Exposure to Fine Particulate Matter

Abstract The adjoint of chemical transport models is an effective tool for linking emissions to air quality‐related premature mortality and formulating mitigation strategies. Due to substantial computational demands, however, adjoint simulations are constrained to coarse resolutions. This introduces potential biases in health risk assessments that are resolution‐sensitive. Here, we conducted a comprehensive evaluation of resolution impacts on health‐focused adjoint assessments of ambient fine particulate matter pollution by analyzing variations in the GEOS‐Chem Adjoint model results across different resolutions and spatial scales. We observed that enhancing the model resolution from 4° × 5° to 2° × 2.5° increased the adjoint‐based estimates of global ambient PM 2.5 ‐related premature deaths by 8.0%, with nearly half of the countries and grid cells showing discrepancies exceeding 20% and 30%, respectively. Further resolution refinement in eastern China to 0.25° × 0.3125°, based on nested simulations, revealed an additional 7.1% increase in estimated mortality. These resolution‐induced disparities predominantly arise from a more pronounced spatial covariation between emissions and forcing (i.e., the partial derivative of the cost function with respect to concentration) at finer resolutions. To mitigate this limitation, we developed a process‐based hybrid approach that dynamically models the inverse transport and convection processes at fine resolutions while simulating chemical and removal processes at coarse resolutions. Compared with direct interpolation methods, this hybrid approach reduces the root mean squared error by more than 40%, accurately reproducing finely resolved adjoint sensitivities and offering a promising avenue for enhancing both forward and adjoint global health assessments at higher spatial resolutions.