Real-time soil moisture mapping using scalable RF sensor networks

Soil moisture mapping is critical for precision irrigation, crop health, and water management, yet traditional approaches are limited by sparse sampling and lack of adaptability in large or dynamic fields. This paper presents a real-time soil moisture mapping framework that leverages scalable RF sensor networks, Gaussian Process Regression (GPR), and a cost-function-based optimization scheme. The system dynamically calculates optimal pixel sizes and positions, enabling smooth transitions between soil moisture maps as the underlying network topology changes due to node failures, communication dropouts, or reconfigurations. GPR is employed to filter noisy Received Signal Strength Indicator (RSSI) values and interpolate missing data, while the cost function balances mapping accuracy with consistency across RSSI-to-moisture projections and probe measurements. The proposed approach was validated through simulation and field trials in Cobbitty, NSW, demonstrating adaptability to asynchronous data streams, scalability with network size, and reliable accuracy, achieving a mean absolute error of 1.28% and a mean bias of -0.277% compared to ground-truth probes. These results highlight the potential of this framework to provide robust, real-time soil moisture monitoring for precision agriculture and large-scale field deployment. • A real-time framework is designed to calculate the optimal positions and sizes of moisture pixels and smoothly transform soil moisture maps across variable dimensions. • A cost function is proposed that takes into account mapping errors, ground-truth measurements, and RSSI-to-moisture projections. By minimizing the cost function with forgetting factors, the optimal pixelated moisture values are obtained in real time. • The GPR method is used to smooth the RSSI and probe streams, and the integral errors are leveraged to obtain real-time estimation with maximal credibility.