Real-time monitoring of soil moisture and nutrients using an integrated dielectric sensor and an underground wireless network

Abstract

Continuous, accurate measurements of soil moisture and nutrients enable data-driven decisions that reduce water and fertilizer waste, improve yields, and minimize environmental impact. However, no practical sensor system currently provides these measurements simultaneously with the required accuracy. Dielectric spectra sensors offer a promising solution, but significant challenges, including system reliability, field validation, and identifying frequency responses specific to individual soil properties, remain. This dissertation addresses these challenges by developing and validating an integrated system for accurate measurement of volumetric water content (VWC), nitrate-nitrogen (NO3-N), and ammonium-nitrogen (NH4-N), with data transmitted through wireless underground networks, to advance sustainable agricultural management. A multi-layer soil propagation model was introduced to predict signal attenuation in complex soil profiles. Extensive field tests across different soil types were conducted to validate this model and demonstrate the feasibility of wireless underground communication using various transceivers. The proposed model significantly improved attenuation estimation, and field results confirmed reliable underground data transmission at depths of up to 50 cm. LoRa-based technology was selected as the preferred wireless communication method due to its robust performance across soil conditions. A dielectric sensor system capable of measuring both magnitude and phase responses from near DC to 1 GHz was designed and implemented. These measurements provide information on soil permittivity, which can be used to estimate key soil properties. The system integrates circuits for signal generation, conditioning, processing, data acquisition, storage, and wireless transmission, all controlled by an STM32F407 microcontroller with efficient power management. System performance was thoroughly tested and validated. An integrated software platform with a user-friendly graphical interface was developed for real-time data visualization and interpretation. Robust communication protocols were established for seamless connectivity among wireless sensor nodes and user devices. The interface supports data collection setup, real-time monitoring, sensor geolocation, and integration with weather data. Laboratory and field experiments were performed to evaluate sensor performance for soil property measurement. Multiple experimental datasets were analyzed to identify the most responsive frequencies for detections of VWC, NO3-N, and NH4-N using variable selection methods, including competitive adaptive reweighted sampling (CARS), minimum redundancy maximum relevance (MRMR), and successive projections algorithm (SPA), in conjunction with artificial neural network (ANN) models. The system achieved good predictive performance, with selected frequencies enabling the development of final models yielding test set R2 values of 0.80, 0.66, and 0.76 for VWC, NO3-N, and NH4-N, respectively. In summary, this dissertation introduces a comprehensive soil monitoring system that delivers accurate, real-time data to advance sustainable agricultural practices.