Evaluation of satellite soil moisture products (AMSR2, SMAP L3/L4) across mainland China using in situ data (2020–2024)

This study evaluated the soil moisture monitoring performance of three satellite products (AMSR2, SMAP L3, and SMAP L4) across mainland Chinese from January 1, 2020 to June 30, 2024. On-site soil moisture observations from 3293 China Meteorological Administration stations were utilized as reference data to validate satellite estimates. By analyzing the impact of site representativeness errors on the validation results, performance was evaluated across multiple times (annual, seasonal, and monthly) and spaces (seven geographical regions), and further analyzed by land cover type to clarify how different underlying surfaces affect accuracy. Four statistical metrics—Bias, Root Mean Square Error (RMSE), unbiased Root Mean Square Error (ubRMSE), and correlation coefficient (R)—are used to evaluate accuracy, and Monte Carlo simulations are conducted to analyze trends and ranges of variation, providing statistical robustness through random sampling and uncertainty quantification. The results indicate that SMAP L4 performs the best, followed by SMAP L3, while AMSR2 has the largest error, and this ranking is not affected by site distribution. Among all time scales and geographical regions, SMAP L4 consistently demonstrates the highest accuracy, particularly in winter (Bias = 0.0049 m3/m3) and humid areas (such as Central China, R = 0.4891). In contrast, AMSR2 displayed dry bias, significantly underestimating soil moisture across all temporal and spatial scales (RMSE = 0.2075–0.2265 m3/m3), with pronounced vegetation interference (summer ubRMSE increased to 0.1548 m3/m3). Land cover analysis showed that SMAP L4 performed well on homogeneous underlying surfaces such as agricultural land (R = 0.4763) and grassland, and maintained good accuracy in forest areas, while AMSR2 yielded significantly increased errors in densely vegetated areas. High-heterogeneity areas such as settlements and wetland posed challenges to all products. Monte Carlo simulations further confirmed that SMAP L3 and L4 products maintained stable error distributions (SMAP L4 R-value fluctuation was merely 0.0002), indicating superior accuracy and reliability. This study reveals the mechanism driving performance differences in satellite products from the perspective of land cover types, providing a scientific basis for the application and algorithm optimization of soil moisture remote sensing products in China. As a crucial physical parameter in hydrological research, soil moisture significantly influences global water cycles, ecosystem health, and agricultural water management.