A global daily seamless 9 km vegetation optical depth (VOD) product from 2010 to 2021

Abstract

Vegetation optical depth (VOD) products provide information on vegetation water content and correlate with vegetation growth status; these are closely related to the global water and carbon cycles. The L-band signal penetrates deeper into the vegetation canopy than the higher-frequency bands used for many previous VOD retrievals. Currently, there are only two operational L-band sensors aboard satellites, i.e., the Soil Moisture and Ocean Salinity (SMOS) satellite launched in 2010 and the Soil Moisture Active Passive (SMAP) satellite launched in 2015. The former has the limitation of a low spatial resolution of only 25 km, while the latter has improved this resolution to 9 km but has a shorter usable time range. Due to the influence of sensor and atmospheric conditions as well as the observation methods of polar-orbiting satellites (such as scan gaps and observation revisit times), the daily data provided by both satellites suffer from varying degrees of missing data. In summary, the existing L-band VOD (L-VOD) products suffer from the defects of missing data and coarse resolution of historical data. There is little research on filling gaps and reconstructing 9 km long-term data for L-VOD products. To solve this problem, our study depends on a penalized least-square regression based on a three-dimensional discrete cosine transform to firstly generate the seamless global daily L-VOD products. Subsequently, the nonlocal filtering idea is applied to spatiotemporal fusion between high-resolution and low-resolution data, resulting in a global daily seamless 9 km L-VOD product from 1 January 2010 to 31 July 2021. In order to validate the quality of the products, time series validation and simulated missing-region validation are used for the reconstructed data. The fusion products are validated both temporally and spatially and are also compared numerically with the original 9 km data during the overlapping period. Results show that the seamless SMOS (SMAP) dataset is evaluated with a coefficient of determination (R2) of 0.855 (0.947) and a root mean squared error (RMSE) of 0.094 (0.073) for the simulated real missing masks. The temporal consistency of the reconstructed daily L-VOD products is ensured with the original time series distribution of valid values. The spatial information of the fusion product and the original 9 km data in the overlapping period is basically consistent (R2: 0.926–0.958, RMSE: 0.072–0.093, and mean absolute error MAE: 0.047–0.064). The temporal variations between the fusion product and the original product are largely synchronized. Our dataset can provide timely vegetation information during natural disasters (e.g., floods, droughts, and forest fires), supporting early disaster warning and real-time responses. This dataset can be downloaded at https://doi.org/10.5281/zenodo.13334757 (Hu et al., 2024).