Data-Driven Soil Water Content Estimation at Multiple Depths Using SFCW GPR

This paper provides a cost-effective solution to Soil Water Content (SWC) estimation at multiple root-zone depths using Ground Penetrating Radar (GPR) and Machine Learning (ML) based on an extensive measurement campaign conducted at Worcester Polytechnic Institute (WPI). SWC characterization is critical for optimal industrial farming irrigation and, in turn, impacts water conservation and the mitigation of soil quality degradation. Accurate prediction of the water table and SWC of the root-zone soil is invaluable for precision farming. High-resolution modeling of SWC at varying sub-surface depths can potentially increase irrigation efficiency and the yield of crops such as maize, which has a massive water footprint upwards of 768 billion cubic meters and accounts for an estimated 5% percent of the world’s daily calorie intake. Traditional methods of subsurface soil characterization by subsurface probes are invasive, costly, and labor-intensive. Our approach generates an accurate and precise characterization of the soil water content of loamy soil at multiple root level depths using Signal Processing principles and ML applied to a small dataset of size 51 of real field measurements collected between October 20th to 30th 2022. We applied ML algorithms to the preprocessed data collected by a Stepped Frequency Continuous Wave (SFCW) GPR signal and extracted the most relevant features related to SWC prediction at multiple depths. We used these extracted features to achieve a mean absolute percentage error as low as 6% across the four root-zone depths of our field data. This study was conducted within the 0.4 to 2.0 GHz frequency range, and provides an analysis of frequencies key to root-zone SWC characterization.