Model Intercomparison and Resolution Dependence in Real-Time Numerical Air Quality Forecasting over North China

High-resolution air quality models (AQMs) are critical for real-time air quality forecasting and exposure assessment, although their computational costs increase cubically with resolution. Quantifying model sensitivity to resolution is therefore crucial for developing effective forecasting systems. This study conducts a systematic model intercomparison of three widely used AQMs (CAMx, CMAQ, NAQPMS) under identical input conditions at 45, 15, and 5 km resolutions to forecast PM2.5 and O3 in the North China Plain during 2021. Results indicate distinct, model-dependent responses to grid refinement. NAQPMS achieves the optimal PM2.5 forecasting performance at 5 km, with improvements in nearly all evaluated statistics. CMAQ excels in O3 prediction at 5 km resolution, with RMSE reducing 6.48 μg/m3 relative to the coarsest grids. We also found that terrain complexity significantly influences these resolution-dependent biases, leading to a substantial 19.51% reduction in NMB in the CAMx PM2.5 simulation over mountain areas. Moreover, the evaluation of 10-day forecasting accuracy suggests that a high-resolution setting is recommended for NAQPMS and CMAQ, whereas a coarser resolution is sufficient for CAMx. These findings underscore that optimizing real-time forecasting strategies requires a critical investigation of inter-model physicochemical discrepancies rather than universally pursuing higher resolution.