Depth-specific soil moisture estimation in vegetated corn fields using a canopy-informed model: A fusion of RGB-thermal drone data and machine learning

Abstract

Accurate soil moisture estimation is fundamental for optimizing irrigation strategies, enhancing crop yields, and managing water resources efficiently. This study harnesses time-series RGB-thermal imagery to assess soil moisture throughout various growth stages of corn, emphasizing depth-specific soil moisture estimation and time-series analysis of canopy information such as canopy structure and canopy spectral across growth stages. By integrating a comprehensive dataset that covers the full spectrum of the growing season from early to late stages. we evaluated soil moisture at multiple depths including 10, 20, 30, and 40 cm. Sophisticated regression models such as Gradient Boosting Machines (GBM), Least Absolute Shrinkage and Selection Operator (Lasso), and Support Vector Machines (SVM) were employed to analyze the effects of spectral indices, land surface temperature (LST), and structural canopy variables on soil moisture estimation accuracy. Our results reveal that thermal variables, particularly LST, exhibit significant correlations with soil moisture at shallower depths, especially in non-irrigated plots where moisture variability tends to be greater. The GBM model performed exceptionally well, achieving a coefficient of determination (R²) of 0.79 and a root mean square error (RMSE) of 1.86 % at a depth of 10 cm, showcasing its precision in moisture prediction. At a depth of 30 cm, the GBM model still demonstrated robust performance with an R² of 0.69 and an RMSE of 3.38 %, adapting effectively to different canopy densities and soil conditions. As canopy density increased, the effectiveness of LST in predicting soil moisture decreased, underscoring the dynamic interaction between plant growth stages and moisture estimation accuracy.