Real-Time Monitoring and Prediction of Evapotranspiration and Organic Carbon in Soil using IoT Sensors

<p dir="ltr">The Ph.D research presents a comprehensive, step-by-step investigation aimed at improving precision agriculture and environmental monitoring by proposing a cost-effective, scalable, and self-sustainable system to replace expensive commercialised tools. The system utilizes IoT-enabled smart sensing technologies integrated with advanced machine learning models to predict water loss through Evapotranspiration and measure soil organic carbon. The study addresses significant gaps in measuring spatial and temporal resolution of soil water data, particularly within the agricultural context of New South Wales (NSW), Australia. The first phase focused on mitigating uncertainty in existing water monitoring networks. For this, a multidepth-embedded, microcontroller-based smart sensor node was developed using low-cost, low-power sensors and essential electronic components. The system was validated in controlled environments before open-field deployment, where it collected real-time data on soil moisture (SM), soil temperature (ST), rainfall precipitation (P), and environmental parameters such as, temperature (T), humidity (RH), wind speed (u2), and net radiations (Rn). Regression and correlation analyses of the field data revealed strong dependencies: SM varied significantly with P and Rn, while ST was largely influenced by T. Such findings not only validated the system’s monitoring capabilities but also demonstrated its potential for developing irrigation strategies and predictive models for groundwater recharge. Building on this foundation, the second phase introduced an extension of the smart sensor node to develop predictive models for evapotranspiration (ET) using advanced machine learning techniques. Real-time environmental data, including air temperature, barometric pressure, wind speed, relative humidity, rainfall, solar exposure, as well as soil temperature, and changes in soil moisture, were collected from an agricultural farm where the developed nodes were deployed. Three machine learning models, Support Vector Machine (SVM), Extreme Learning Machine (ELM), and M5P Regression Tree, were developed and benchmarked against empirical models (Penman-Monteith and Soil Water Balance) and climate ET data from the Australian Bureau of Meteorology. Among the models, SVM showed the highest accuracy (R² = 0.97, RMSE = 0.19 mm/d, MAE = 0.14 mm/d), followed by ELM and M5P. These models effectively captured complex interactions among variables, offering high-resolution insights into soil-water dynamics under different climatic scenarios. In the final phase, a novel electromagnetic sensor was introduced for detecting soil organic carbon (SOC), a key indicator of soil quality. The sensor was designed using a combination of interdigital and spiral inductance-capacitance geometries on an FR-4 printed circuit board. Initially, the sensor impedance characterization was correlated with the SOC percentage content determined by the Walkley–Black (WB) method. Subsequently, a sensor calibration and integration using an AD5933 Impedance analyzer setup was conducted to be able to test the sensor in real-world conditions, demonstrating good overall performance (R²=0.92) with minimal errors (MAE=0.45) in both low and high carbon soils. All soil samples collected from pastoral regions around Sydney were utilized to train and validate the model. The sensor demonstrated an accuracy of ±0.5% across 1000 runs, confirming its sensitivity and reliability. The model successfully predicted SOC content while accounting for confounding environmental factors, such as temperature, humidity, and soil moisture, making it suitable for field deployment in diverse conditions. The PhD project study collectively demonstrates the viability and impact of integrating low-cost smart sensors with advanced machine learning analytics for robust soil-environmental monitoring and predictive modeling, providing scalable solutions for precision agriculture and sustainable water resource management.</p>