Enhancing spatial resolution of satellite soil moisture data through stacking ensemble learning techniques.

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Soil moisture (SM) is crucial for environmental processes but often lacks sufficient spatial resolution from traditional satellite sensors for local studies.
• Water scarcity in regions like Urmia basin necessitates improved SM monitoring for effective resource management.

Purpose of the Study:

• To develop a novel stacking ensemble learning framework to enhance the spatial resolution of satellite-derived SM data to 1 km.
• To improve the accuracy and spatial detail of SM estimations for local-scale environmental and agricultural applications.

Main Methods:

• Integrated in-situ SM measurements (TDR), SMAP and AMSR2 SM products, MODIS LST and vegetation indices, precipitation, and topography data.
• Employed a stacking ensemble learning approach, selecting top-performing base models (Random Forest, Gradient Boosting, XGBoost) using COPRAS.
• Utilized SHapley Additive exPlanations (SHAP) to analyze model contributions.

Main Results:

• The ensemble model significantly improved SM estimation accuracy and spatial resolution compared to individual models.
• XGBoost and Gradient Boosting meta-models achieved high accuracy (ubRMSE: 1.23% m3/m3, R2: 0.97) in testing.
• SHAP analysis confirmed synergistic benefits from combining diverse machine learning models.

Conclusions:

• Ensemble learning offers a powerful approach to enhance the spatial resolution and accuracy of satellite-derived SM data.
• This study sets new benchmarks for soil moisture monitoring, providing valuable insights for water-stressed regions.
• The developed framework supports improved environmental science research and agricultural planning.