Global soil moisture dynamics since 1980: datasets biases, trends, and science-informed selection

Soil moisture is critical for climate prediction, ecological management, and disaster warning. However, multi-source datasets show spatiotemporal inconsistencies and uncertain regional applicability due to algorithmic and observational limitations. We assess the statistical performance and spatiotemporal variations of 23 global surface soil moisture datasets (1980–2023) from reanalysis, land surface models, and microwave remote sensing across global and regional scales (classified by Köppen climates and IPCC land uses). Results show a slight long-term (1980–2023) global surface soil moisture decline (−4.30×10 –4 m 3 m –3 a –1 ), with some datasets indicating short-term wetting (7.17×10 –4 m 3 m –3 a –1 ) post-2010 (2010–2023). A dual-validation against 992 and a filtered subset of 483 highly representative in situ stations shows that most products perform moderately well (Pearson R ≈ 0.5–0.7). Microwave remote sensing products, especially those based on SMAP, consistently demonstrate superior performance in capturing temporal dynamics ( R ≈ 0.7). Our analysis demonstrates that spatial representativeness error can mask true performance, with validation in the tropics improving dramatically after site filtering (mean R increase of 0.41). The findings highlight product-specific strengths and weaknesses, underscoring the necessity of a science-informed, application-specific approach to dataset selection for robust hydrological and climatic research.