Observation-driven dynamic correction of urban CO2 simulations in WRF-Chem for high-resolution carbon monitoring

Introduction Accurate simulation of near-surface CO 2 concentrations in urban areas is essential for quantifying city-scale carbon emissions. However, the strong spatiotemporal variability of urban emissions and boundary-layer dynamics introduces large uncertainties into atmospheric transport simulations. Methods To address this issue, this study develops an observation-driven dynamic correction framework to improve urban CO 2 simulations. First, the WRF-Chem model was configured using the ODIAC fossil-fuel CO 2 emission inventory, NCEP FNL meteorological fields, and GlobeLand30 land-cover data to provide a physically consistent baseline simulation for urban CO 2 transport. Subsequently, high-temporal-resolution CO 2 observations obtained from a network of photoacoustic spectroscopy (PAS) instruments were used to quantify systematic discrepancies between model simulations and measurements over a 24-h cycle. Based on these discrepancies, hourly correction factors were derived at each monitoring site and extended to the entire study domain through spatial interpolation to construct a continuous correction field. Results The correction factors were closer to unity during nighttime and early morning, implying smaller relative correction requirements during these hours, whereas substantially lower values occurred during daytime and evening periods, indicating stronger systematic overestimation in the original simulations. After applying the dynamic correction, the simulated CO 2 concentrations showed improved temporal evolution and spatial consistency relative to the PAS observations. Leave-one-site-out cross-validation indicated that the framework performed consistently across independent validation sites. Across the ten validation runs, the mean bias changed from 92.68 ± 3.89 ppmv to 0.14 ± 2.50 ppmv, and the RMSE decreased from 128.05 ± 7.03 ppmv to 10.44 ± 2.71 ppmv, corresponding to an average RMSE reduction of 91.9 ± 2.0%. Discussion These results indicate that integrating high-frequency ground observations with atmospheric transport models provides an effective approach for improving urban CO 2 simulations and supporting high-resolution carbon monitoring.