Prediction of Soil Organic Carbon in Unscientific Coal Mining Area Using Landsat Auxiliary Data

Unscientific coal mining is affecting soil attributes, deteriorating soil health and crop productivity, and is reflected by the soil organic carbon (SOC) content. The quantification of SOC is challenging with limited resource availability; however, satellite covariance is the alternate source of SOC determination with minimum labour requirement and limited laboratory facilities. An attempt is made to estimate the SOC using Landsat data and a model is developed by evaluating stepwise and enter/removal regression approaches. Fourteen predictor variables were used to build models and evaluate the prediction accuracy. Results showed that the SOC ranges 0.81–2.41% under unscientific coal mine affected sites; NDVI and BSI range 0.16–0.61 and −0.35–0.12 with mean 0.32 and −0.03, respectively. SOC is correlated with RI (r −0.33) and GRVI (r 0.34). The enter (all variables in a block enter in a single step) approach linear regression model (Model 3) using multiple variables can explain only 43% SOC (RMSE 0.66 and R2 0.43). The stepwise linear regression model (Model 2) and Model 3 predicted the SOC as 6.94 and 3.78% higher than the actual SOC data. The model performance is increased by using multiple variables which may subside the less number of soil samples.