A multilevel downscaling model for enhancing nocturnal aerosol optical depth reanalysis from CAMS over the Beijing-Tianjin-Hebei region, China

Nocturnal aerosol optical depth (AOD) serves a critical indicator for investigating the diurnal aerosol’s climatic and environmental effects. However, the nocturnal AOD product is lacking due to that current daytime AOD retrieval algorithms are inapplicable to nighttime. Despite important contribution of spatiotemporal continuous global reanalysis datasets to producing atmospheric composition forecasts and analyses, its feasibility for the characterization of nocturnal aerosol variation over small scales is still a major challenge due to its coarse resolution. In this study, a multilevel two-stage downscaled (TSD) model by integrating a linear mixed effect (LME) and a geographic weight regression (GWR) model is proposed to improve the spatial resolution of Copernicus Atmosphere Monitoring Service reanalysis (CAMSRA) nocturnal AOD. The multilevel downscaled model is referred to as the M_TSD model. The M_TSD model is employed over the Beijing-Tianjin-Hebei (BTH) region in China for the years from 2018 to 2022. Cross-validation of the retrieval results versus original CAMSRA data shows good performance of the M_TSD model with a determination coefficient (R 2 ) of daily nocturnal AOD of 0.9569, root mean squared prediction error (RMSE) and the relative prediction error (RPE) of 0.0939 and 15.1 %, respectively. The inter-annual average nocturnal AOD indicate significant spatial variation with high value in southeastern plain and low value in the northwestern mountainous and plateau areas of BTH. Meanwhile, the nocturnal AOD is subject to seasonal variability. The M_TSD model may serve as a valuable reference to provide nocturnal AOD data with high spatial resolution for small scale. • M_TSD model was proposed to downscale the nocturnal AOD reanalysis. • The model accurately captures spatiotemporal variation of AOD-variables relationship. • The downscaled nocturnal AOD is highly correlated with AERONET observations.