Predicting soil organic carbon by integrating Landsat 8 OLI, GIS and data mining techniques in semi-arid region

Soil organic carbon (SOC) is an important indicator for soil quality and environmental health. It also plays a key role in the semi-arid region. The aims of this study were to derive models for SOC prediction using Landsat 8 OLI data in dry and wet months of a semi-arid region. To this end, the SOC contents were measured in 165 points from agricultural soils (0–15 cm depth) based on a stratified random sampling method. The measured data were divided randomly into a calibration data-set (75%) and validation data-set (25%). The multiple linear regression (MLR), support vector machine (SVM), and artificial neural network (ANN) models were then employed to predict SOC contents by using two Landsat 8 OLI images acquired in dry (June 2019) and wet (February 2019) months. The accuracy of developed models was evaluated by applying the ME (mean error), R² (coefficient of determination), and RMSE (root mean square error) indices. The results indicated that the derived ANN model performed better than the developed MLR and SVM models for predicting SOC contents in both dry and wet months. Overall, the best result for SOC contents prediction was obtained by the ANN model in dry month (ME = -0.055, RMSE = 0.163 and R² = 0.743). It was concluded that using Landsat 8 OLI image in the dry month brings higher accuracy for SOC prediction.