Surface observations are crucial for understanding atmospheric pollutant sources and variability. However, interpreting these observations, particularly when comparing them to chemical transport model (CTM) simulations, remains challenging. A key difficulty is the spatial mismatch between model resolutions and the point measurements from surface stations. To mitigate this issue, we developed a deep learning (DL)-based method to quantify systematic discrepancies between surface carbon monoxide (CO) observations and GEOS-Chem model simulations across China during 2015–2022. Our method generated daily correction factors for adjusting modeled CO concentrations. Validation demonstrated good consistency between the observed-to-modeled (Obs/GC) concentration ratios and the derived correction factors, with correlation coefficients (R) ranging from 0.85 to 0.93. Our analysis reveals broadly uniform negative correlations between correction factors and observed CO concentrations across eastern China, suggesting that systematic discrepancies decrease with increasing local emissions. In contrast, positive correlations prevail in western China. Furthermore, significant temporal variability in systematic discrepancies was identified at seasonal scales, emphasizing the need for time-dependent dynamic corrections. Applying the DL-based correction approach to GEOS-Chem-simulated surface CO concentrations for 2015–2022 led to a significant improvement in model-observation agreement: R values increased from 0.30–0.43 to 0.63–0.70 (spatial consistency) and from 0.15–0.49 to 0.62–0.81 (temporal consistency). This work provides a novel data-driven approach for correcting systematic discrepancies in model/observation comparisons, which is important for more accurate interpretation of surface observations.
扫码关注我们
求助内容:
应助结果提醒方式:
