Deep learning and data fusion to estimate surface soil moisture from multi-sensor satellite images

Area of Science:

• Earth Science
• Remote Sensing
• Artificial Intelligence

Background:

• Surface soil moisture is a critical variable in hydrological and ecological processes.
• Accurate soil moisture estimation is vital for various applications, including agriculture and climate modeling.
• Satellite remote sensing offers a promising approach for large-scale soil moisture monitoring.

Purpose of the Study:

• To develop and evaluate a novel Artificial Neural Network (ANN) model for estimating surface soil moisture.
• To assess the performance of the ANN model using satellite-derived features on the Kosi River alluvial fan.
• To compare the ANN model's accuracy against ten other machine learning algorithms.

Main Methods:

• Extracted nine features from Sentinel-1, Sentinel-2, and SRTM DEM satellite products using linear data fusion.
• Employed a fully connected feed-forward ANN architecture for soil moisture estimation.
• Conducted field campaigns for in-situ soil moisture measurements using a TDR probe at 224 locations.

Main Results:

• The ANN model achieved a high correlation coefficient (R = 0.80) and low RMSE (0.040).
• The ANN model demonstrated superior performance compared to ten benchmark machine learning algorithms.
• Feature importance analysis confirmed the impact of different satellite-derived features on soil moisture prediction.

Conclusions:

• The proposed ANN model provides an accurate and robust method for estimating surface soil moisture from satellite imagery.
• The findings support the use of integrated satellite data and advanced machine learning for hydrological applications.
• This study advances the capability for large-scale soil moisture monitoring in complex terrains like the Himalayan Foreland.