High-resolution Soil Moisture Retrieval Using Sentinel-1 Data for Evaluating Regenerative Agriculture: A feasibility study from Alentejo, Portugal

Abstract

Timely, reliable, and cost-efficient information about soil moisture is important for supporting agricultural practitioners in monitoring the impact of alternative agricultural practices. Regenerative agriculture is increasingly gaining traction; however, farmers lack easy access to information on key agricultural parameters such as soil moisture. Therefore, this study seeks to explore the feasibility of soil moisture estimation at high-resolution (around 10 m) using Sentinel-1 remote sensing radar data. A machine learning model was developed using a random forest regression algorithm with a combination of SAR-based, topography and Seninel-2 optical-based data as inputs. Through a k-fold cross-validation of the model, an average r-squared (R²) of 0.17, a root mean squared error (RMSE) of 3.51 (% VMC), and an mean absolute percentage error (MAPE) of 83.34, was achieved.