Downscaling AOD

We develop a two-stage hybrid downscaling algorithm to generate high-resolution, seamless aerosol optical depth (AOD) maps for China: (1) Ensemble learning models are employed to simultaneously fill gaps in MODIS AOD and correct MERRA-2 AOD at the pixel level, creating reliable training labels and spatially consistent input features, while (2) the downscaling stage utilizes an enhanced Downscale-UNet model incorporating Convolutional Block Attention Module (CBAM) and Depthwise Separable Convolution (DSC) to automatically extract spatial features and generate fine-scale outputs. This integrated approach effectively reconciles the complementary strengths of satellite and reanalysis data while overcoming their respective limitations of missing values and low resolution, ultimately providing more reliable AOD products for regional-scale analysis. In addition to the algorithm, we produce a nationwide AOD dataset covering China at a 1 km spatial resolution for the period 2014–2023. The dataset provides 550 nm AOD maps in GeoTIFF format with WGS84 geographic coordinates, offering valuable resources for long-term aerosol monitoring and climate studies.