Spatiotemporal Reconstruction of MODIS Normalized Difference Snow Index Products Using U-Net with Partial Convolutions

Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover product is one of the prevailing datasets for global monitoring, but cloud obscuration leads to discontinuity ground coverage information in spatial and temporal. To solve this problem, a novel spatial-temporal missing reconstruction model based on U-Net with partial convolutions (PU-Net) proposed recover gaps MODIS Normalized Difference Snow Index (NDSI) products. Taking Yellow River Source Region as study case, which characterized by shallow, fast-changing complex heterogeneity, NDSI 2018–2019 season reconstructed, accuracy validated simulated mask situ depth (SD) observations. The results show that under scenario, mean absolute error (MAE) reconstructed pixels from 4.22% 18.81% different scenarios patch ratio applied mask, coefficient determination (R2) ranges 0.76 0.94. validation SD observations at 10 sites shows good consistency, overall increased 25.66% 49.25% compared Aqua-Terra combined product, its value exceeds 90% 60% observation stations.