Gridded post-processing of air quality predictions based on the Community Multi-scale Air Quality (CMAQ) model

Systematic and random errors remain in operational predictions of fine particulate matter (PM_2.5) and ozone (O₃) from the Community Multiscale Air Quality (CMAQ) model. Current post-processing methods at the National Air Quality Forecast Capability (NAQFC) apply the analog ensemble (AnEn) technique only at observation sites, followed by spatial interpolation to produce gridded bias-corrected fields. This approach, however, can degrade performance in regions with sparse monitoring or discontinuous bias patterns, such as during wildfire smoke events.

This study evaluates a novel framework in which AnEn is applied directly at every CMAQ grid point, leveraging bias-corrected “ground truth” fields derived by merging AirNow observations with either Copernicus Atmosphere Monitoring Service (CAMS) analyses or CMAQ analyses using the Satellite-Enhanced Data Interpolation (SEDI) technique. Year-long evaluations for 2021 across the contiguous United States demonstrate that all bias-corrected methods substantially improve PM_2.5 and O₃ forecasts relative to raw CMAQ. Among them, AnEn driven by bias-corrected CMAQ analyses consistently yields the best overall performance across metrics, while AnEn with CAMS inputs performs best during extreme PM_2.5 episodes associated with wildfire smoke. For ozone, bias correction improves daytime forecasts, though challenges remain at night due to nonlinear chemistry and sparse observations. These findings highlight the effectiveness of gridded AnEn post-processing for improving operational CMAQ forecasts, particularly for PM_2.5, and emphasize the need for pollutant-specific strategies. The proposed framework provides a pathway toward more accurate and spatially complete air quality information to better support public health protection.