Soil moisture mapping using radio signal strength and gaussian process regression

Abstract Background Soil moisture is a key variable in precision agriculture, affecting irrigation management, crop productivity, and water-use efficiency. However, field-scale soil moisture monitoring remains challenging because conventional sensing approaches are often costly and provide limited spatial coverage. This study proposes a simulation-based framework for soil moisture mapping that integrates near-ground radio-frequency (RF) sensing, Gaussian Process Regression (GPR), and adaptive sample selection Methods Received signal strength (RSS) measurements obtained from paired RF sensors were used as path-averaged proxies for soil moisture. GPR was employed to reduce measurement noise and reconstruct spatial soil moisture distributions from sparse ground-truth observations. An error-driven adaptive sampling strategy was developed to identify locations for additional soil moisture measurements under constrained sampling budgets. The framework was evaluated using numerical simulations based on synthetic soil moisture fields with varying levels of spatial variability and measurement noise. Results Simulation results demonstrated that the proposed framework reduced soil moisture estimation error compared with random sampling and conventional GPR interpolation when only a limited number of samples were available. The adaptive sampling strategy improved the efficiency of data collection by prioritizing regions with higher prediction uncertainty. Reconstruction accuracy was influenced by both soil moisture variability and measurement noise, highlighting the importance of sensor quality and field heterogeneity in system performance. Conclusion The study demonstrates the potential of combining RSS-based RF sensing, probabilistic regression, and adaptive sampling for low-cost and scalable soil moisture mapping in precision agriculture. Although the framework was evaluated only under idealized simulation conditions and no field validation was conducted, the results provide proof of concept and support future development, field testing, and extension of the approach for practical agricultural applications.