Half the World Is Drying, but Not as Models Predict: New Global Soil Moisture Study
31 Mar 2026
A research team led by Prof. WANG Shudong at the State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), has made significant progress in reconstructing long-term global surface soil moisture variations and improving future projections through observation-constrained climate modeling. The study, Quantifying Historical and Future Surface Soil Moisture Drying Using Deep Learning and Remote Sensing, was recently published in Earth's Future.
Surface soil moisture is a crucial link between the atmospheric, hydrological, and ecological processes. It plays an essential role in agricultural production, drought monitoring, heatwave and wildfire risk assessment, and the characterization of land-atmosphere interactions. Understanding how surface soil moisture evolves and whether drying will persist in the future, is a key challenge in global change research. However, uncertainties in satellite observations and limitations in Earth system models which represent land-atmosphere feedback processes have led to substantial discrepancies in estimates of long-term soil moisture trends and future projections.
To address this challenge, Prof. Wang's team developed a framework integrating satellite observations, deep learning, and Earth system model simulations. Deep learning was first applied to fill gaps in global microwave satellite-derived surface soil moisture datasets from 1983 to 2020, producing a more complete and temporally consistent observational record. Outputs from 23 CMIP6 Earth system models were then incorporated to link simulations with observations.
This allowed the reconstruction of historical soil moisture variations for 1901–1980 and correction future projections for 2021–2100. Compared to conventional statistical bias-correction approaches that rely on linear assumptions and fixed distribution relationships, the method provides a flexible and physically consistent pathway for predicting soil moisture under complex land–atmosphere coupling processes.
Results show that the framework significantly improves the completeness and reliability of global surface soil moisture observations. Observational coverage increased by approximately 15%, and independent validation achieved a coefficient of determination of about 0.9. Data from 465 in situ observation sites worldwide further demonstrated strong agreement with reconstructed satellite products. Observation-constrained model simulations also showed closer agreement with satellite observations than original model outputs across most regions, enhancing confidence in historical reconstructions and future projections.
The study reveals several new insights into global soil moisture evolution. Over the past four decades, approximately half of the global land surface experienced drying. However, the observation-constrained simulations indicate that drying trends in climate transition zones and monsoon marginal regions may be weaker than previously model estimates, indicating that current models overestimate future drought intensification due to insufficient representation of soil moisture-atmosphere feedbacks.
Importantly, the study challenges the "dry-get-drier, wet-get-wetter" paradigm. Only about one-third of global land areas strictly follow this pattern, while many regions exhibit more complex or even opposite responses. Future soil moisture changes are shaped evapotranspiration, precipitation changes, and regional land-atmosphere, highlighting strong spatial heterogeneity and nonlinear responses. These results provide new observational evidence for understanding global soil moisture dynamics in a changing climate.
Changes in surface soil moisture and drought frequency during the past four decades. (Image by AIRCAS)
Mechanism and interpretation of soil moisture-atmosphere feedback. (Image by AIRCAS)
Research News
Half the World Is Drying, but Not as Models Predict: New Global Soil Moisture Study
A research team led by Prof. WANG Shudong at the State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), has made significant progress in reconstructing long-term global surface soil moisture variations and improving future projections through observation-constrained climate modeling. The study, Quantifying Historical and Future Surface Soil Moisture Drying Using Deep Learning and Remote Sensing, was recently published in Earth's Future.
Surface soil moisture is a crucial link between the atmospheric, hydrological, and ecological processes. It plays an essential role in agricultural production, drought monitoring, heatwave and wildfire risk assessment, and the characterization of land-atmosphere interactions. Understanding how surface soil moisture evolves and whether drying will persist in the future, is a key challenge in global change research. However, uncertainties in satellite observations and limitations in Earth system models which represent land-atmosphere feedback processes have led to substantial discrepancies in estimates of long-term soil moisture trends and future projections.
To address this challenge, Prof. Wang's team developed a framework integrating satellite observations, deep learning, and Earth system model simulations. Deep learning was first applied to fill gaps in global microwave satellite-derived surface soil moisture datasets from 1983 to 2020, producing a more complete and temporally consistent observational record. Outputs from 23 CMIP6 Earth system models were then incorporated to link simulations with observations.
This allowed the reconstruction of historical soil moisture variations for 1901–1980 and correction future projections for 2021–2100. Compared to conventional statistical bias-correction approaches that rely on linear assumptions and fixed distribution relationships, the method provides a flexible and physically consistent pathway for predicting soil moisture under complex land–atmosphere coupling processes.
Results show that the framework significantly improves the completeness and reliability of global surface soil moisture observations. Observational coverage increased by approximately 15%, and independent validation achieved a coefficient of determination of about 0.9. Data from 465 in situ observation sites worldwide further demonstrated strong agreement with reconstructed satellite products. Observation-constrained model simulations also showed closer agreement with satellite observations than original model outputs across most regions, enhancing confidence in historical reconstructions and future projections.
The study reveals several new insights into global soil moisture evolution. Over the past four decades, approximately half of the global land surface experienced drying. However, the observation-constrained simulations indicate that drying trends in climate transition zones and monsoon marginal regions may be weaker than previously model estimates, indicating that current models overestimate future drought intensification due to insufficient representation of soil moisture-atmosphere feedbacks.
Importantly, the study challenges the "dry-get-drier, wet-get-wetter" paradigm. Only about one-third of global land areas strictly follow this pattern, while many regions exhibit more complex or even opposite responses. Future soil moisture changes are shaped evapotranspiration, precipitation changes, and regional land-atmosphere, highlighting strong spatial heterogeneity and nonlinear responses. These results provide new observational evidence for understanding global soil moisture dynamics in a changing climate.
Changes in surface soil moisture and drought frequency during the past four decades. (Image by AIRCAS)
Mechanism and interpretation of soil moisture-atmosphere feedback. (Image by AIRCAS)