Enhancing soil moisture retrieval in semi-arid regions using machine learning algorithms and remote sensing data

Soil moisture is an essential quantitative characteristic in hydrological processes and agricultural production. Satellite remote sensing has been extensively used to estimate topsoil moisture. However, gathering Soil Moisture Content (SMC) data with high spatial resolution in diverse watersheds takes a lot of work and money to maintain. In this research, a significant soil moisture retrieval analysis in a semi-arid region of Pakistan was done to investigate the potential use of machine learning algorithms in the agricultural field. Various machine learning algorithms, i.e., Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Elastic Net Regression (EN), were applied to retrieve soil moisture using Landsat 8 thermal and optical sensors. As a result, enhancing retrieval from remote sensing data is critical, which is vital for land resource planning and management. Many techniques for estimating soil moisture content in various geographical and climatic circumstances based on satellite-derived vegetation indices have been established. Machine learning, statistical approaches, and physical modeling techniques were used to retrieve soil moisture. Compared to other ML models, it shows a Nash-Sutcliffe efficiency of 1.9, an index of agreement 2.08 for predicted SMC for the RF model. According to the data analysis, the RF technique showed superior performance with the maximum Nash-Sutcliffe Efficiency value (0.73) for soil moisture retrieval across all land-use categories sound reflectivity, and supplemental geographical data can be combined with the outputs of this research to give more helpful insight for estimation of SMC having precise agricultural applications.