Advanced Machine Learning Techniques for Soil Moisture Monitoring Using SAR and Optical Remote Sensing Data

Abstract

Soil moisture is a crucial variable in various environmental and agricultural processes. Accurately estimating soil moisture content using remote sensing techniques has been a longstanding challenge. This research paper investigated the effectiveness of various machine learning models in retrieving soil moisture content using satellite data. In-situ soil moisture data was collected using HydraGo probes, while Sentinel-1 and Sentinel-2 satellite images were acquired. Different combinations of input parameters, including VH, VV, LIA, DPRVI, RVI, and NDVI, were tested for ANN, ANN-PCA, LSTM, and ELM models. The results showed that the best combination for ANN, LSTM, and ELM was VH, VV, LIA, DPRVI, and RVI, while for ANN-PCA, NDVI, VH, and VV performed best. Overall, the ANN model with the VH, VV, LIA, DPRVI, and RVI combination achieved the highest accuracy (R-squared = 0.5032) for soil moisture prediction. The study highlights the potential of machine learning techniques for accurate soil moisture retrieval from satellite data, which can be valuable for agricultural and water resource management.