Predictive Analysis of Soil Organic Matter and Moisture Content Using Image-Based Modeling

Abstract

Soil organic matter (SOM) and soil moisture content (SMC) are critical indicators of soil health, yet their measurement using conventional methods is often prohibitive due to high time, labor, and financial costs. To address these challenges, a novel model employing image processing techniques has been developed to predict SOM and SMC based on soil colour features. This model utilizes stepwise multiple linear regression (SMLR) to correlate soil colour attributes, such as colour moments, Gray Level Co-occurrence Matrices (GLCMs), and various colour models, with the moisture and organic content of the soil. Field samples were systematically collected at defined intervals to represent continuous variation in soil properties. The ground truth for model calibration was established using the loss of ignition method. The efficacy of the model was validated externally, with the selection of 34 initial and then 6 optimal predictor variables, yielding an R2 of 0.67 and a Root Mean Square Error (RMSE) of 0.76 for SOM prediction, and an R2 of 0.77, RMSE of 0.55, and a Ratio of Performance to Inter-Quartile (RPIQ) of 1.07 for SMC. These results demonstrate the potential of using image -based modeling as a robust tool for soil property analysis.