Improving high-resolution soil moisture mapping in China via dense ground-based observations and multi-source data fusion

Soil moisture (SM) is a vital element of the terrestrial hydrological cycle and is essential for fields like agriculture and ecology. While satellite-based microwave remote sensing is a common method for monitoring soil moisture, the resulting data often have a coarse spatial resolution that is not suitable for fine-grained applications such as drought monitoring and irrigation management. This study presents a new two-step framework for creating high-precision soil moisture data with both 1-km spatial resolution and daily continuity, which combines multi-source ground data from satellites and a dense network of ground stations in China. The proposed method includes a satellite-satellite soil moisture downscaling module and a satellite-ground soil moisture calibration module. The downscaling module uses a random forest model and variables like vegetation and temperature to downscale coarse-resolution data (25 km) from the ESA CCI Combined SM dataset to high-resolution data (1 km). Meanwhile, this step also improves the temporal continuity and usability of the data, which is only 71.2% valid in its original form and seamless in the downscaled SM. The calibration module then uses a large volume of in situ measurements from the China Meteorological Administration’s network to further correct the downscaled data. The evaluation results show that the calibrated data has an impressive correlation coefficient (CC) of 0.772 and a root mean square error (RMSE) of 0.073 m³ ∙m⁻³ . This marks a notable improvement of 70% in CC and 33% in RMSE compared to the original ESA CCI Combined SM data. Soil moisture in China exhibits a spatial pattern of high values in the southeastern coastal areas and low values in the northwestern inland regions. The study concludes that the accuracy improvement primarily stems from the satellite-ground calibration module, while the downscaling module provides the necessary high-resolution, continuous data for this calibration.