Study on spatio-temporal simulation and prediction of regional deep soil moisture using machine learning

Area of Science:

• Hydrology
• Remote Sensing
• Ecology

Background:

• Deep soil moisture (SM) is vital for vegetation restoration in semi-arid regions.
• Existing SM products lack the necessary spatio-temporal resolution and soil depth for eco-hydrological studies.

Purpose of the Study:

• To develop a random forest model for predicting SM at various depths.
• To integrate SMAP satellite data for regional-scale SM estimation.
• To create a continuous, high-resolution SM data product for arid and semi-arid areas.

Main Methods:

• Utilized an international SM network dataset to identify SM drivers and vertical correlations.
• Developed a random forest prediction model for SM at different soil depths.
• Integrated SMAP satellite data to upscale point-scale SM estimates to regional coverage.

Main Results:

• Precipitation-SM correlation evolves into inter-layer interactions with depth; shallow SM response to precipitation is rapid (1-day lag).
• Key drivers of SM include slope, land use, clay content, leaf area index, potential evapotranspiration, and land surface temperature.
• The random forest model achieved high prediction accuracy at both site and regional scales, outperforming CLDAS products in spatial detail.

Conclusions:

• The developed model provides a reliable method for acquiring deep SM data in arid/semi-arid regions.
• The resulting SM data product offers improved spatio-temporal resolution for eco-hydrological research.
• This study offers technical support and new insights for deep SM monitoring and vegetation restoration efforts.