High quality soil moisture (SM) estimation is crucial for various applications such as drought monitoring, environmental assessment, and agricultural management. The advancement of remote sensing technology has made it possible to use active and passive sensors to retrieve near real-time soil moisture on the Earth's surface. However, the European Space Agency's Climate Change Initiative (CCI) SM product combines data from multiple sensors, but sacrifices spatiotemporal resolution and coverage due to satellite orbit limitations and retrieval algorithms. To address this issue, a SM reconstruction method based on a conditional variation autoencoder model was developed by utilizing the high spatial resolution of SMAP L4 data and the accuracy of CCI fusion products in different land cover types. Through this method, a global three-day SM product with a temperature range of 0.0625 ° was created, spanning from 2015 to 2021.
The European Space Agency's CCI SM product integrates multiple SM data from active and passive microwave sensors, creating three types of products: active products, passive products, and integrated products that utilize active and passive microwave data. We collected the combined microwave products of CCI as the basic data source for SM product reconstruction, and readjusted the entire second level observation data to a pattern based general climatology. This dataset spans 40 years with a spatial resolution of 0.25 ° and a time span from November 1978 to December 31, 2021.
In this study, we collected the SMAP Level 4 and global 9-kilometer EASE Grid surface and root zone soil moisture analysis datasets with a spatial resolution of 9 kilometers updated by the National Snow and Ice Data Center (NSIDC).
We propose a new approach to reconstruct HR SM products on a global scale. Using the comprehensive European Space Agency CCI SM dataset and SMAP assimilation products, we employed mutation inference and conditional mutation autoencoder (CVAE) to seamlessly merge these datasets and improve the accuracy and coverage of SM estimation. This method overcomes challenges related to missing values, limited use of multi-source observations, and spatial distribution differences. By integrating the advantages of two datasets, we have obtained globally consistent high-quality SM products.
The reconstructed SM product has undergone rigorous validation by global core SM sites and sparse observation networks. The evaluation used multiple indicators, including global unbiased root mean square error (ubRMSE) and correlation coefficient (CC). The verification results show that the ubRMSE values of the core SM site and sparse observation network are approximately 0.029 and 0.071 m3/m3, respectively, and the CC values are approximately 0.863 and 0.743, respectively. Compared to existing benchmarks, this reconstruction product has global coverage and higher accuracy.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.ZENODO.DB6519.2024
10.5281/zenodo.8000601