Predictive Analysis for Soil moisture content and Soil organic matter from Soil Images

Abstract

– Soil organic matter (SOM) and soil moisture contents (SMC) are two main properties in defining soil health. It is a challenge to measure organic matter and moisture content in soil, as the conventional methods are time, labor and money consuming. In order to overcome these challenges, various image processing based models have been proposed to predict SOM and SMC. Proposed model uses a stepwise multiple linear regression (SMLR) method to predict these properties on basis of soil color features like color moments, GLCMs and different color models as well, since, soil moisture and organic content are influenced by soil color. Multiple soil samples from field are collected with a certain distance in order to simulate continuous variation in the contents. Loss of ignition method is used to generate the ground truth to feed the model. For produce and compare the result first 34 and then 6 optimal predictor variables are used in model. The output results in external validation for SOM prediction were: R 2 = 0.07, RMSE = 0.76, RPIQ = 1.00 and that for SMC were: R 2 = 0.77, RMSE = 0.55, RPIQ = 1.07.