Spatio-Temporal reconstruction to fill spatial gaps in global satellite vegetation optical depth products

Vegetation Optical Depth (VOD) serves as a crucial tool for monitoring vegetation characteristics and plays a vital role in terrestrial ecosystems. The vegetation optical depth dataset from AMSR-E/2 using multi-channel collaborative algorithm (MCCA-AMSR VOD) possesses a longer time series spanning from 2002 to 2022. However, due to the limitations of satellite orbital scanning gap and retrieval algorithms, existing VOD datasets do not achieve complete coverage in global land. To address the need for a VOD dataset with high spatio-temporal coverage across different bands, we reconstruct the MCCA-AMSR VOD dataset using the 3D partial convolutional neural network. Three evaluation methods are employed to further assessthe reconstructed VOD dataset. The reconstructed VOD achieves a higher average spatial coverage (95.01 %) compared with the original dataset (72.55 %). The correlation coefficients (R) between original and reconstructed VOD dataset for all five simulated missing regions exceeded 0.99, indicating that the reconstructed VOD has a very high correlation with the original dataset at the regional scale. Furthermore, comparison with optical vegetation indices indicates that the reconstructed VOD can also capture vegetation water contentwell. Therefore, reconstructed VOD can provide valuable support for research on drought and vegetation monitoring.