Digital Mapping of Soil Equivalent Calcium Carbonate Using Landsat 8 Satellite Images and Environmental Data by Machine Learning Models in Badr Watershed, Kurdistan Province

The calcium carbonate equivalent (CCE) in soil is one of the most important soil properties. Predicting the amount of calcium carbonate equivalent in soil is essential for sustainable soil fertility management. The present study aimed to digitally map calcium carbonate equivalent using auxiliary environmental variables, Landsat 8 satellite images, and predictive models and to present the best models in the Badr watershed in the south of Qorveh district. In the first phase, a geomorphologic map was created using a geologic map and based on the ZINC method in a geographic information system environment. In the second phase, the location of 125 survey profiles was determined using the Latin hypercube technique, and the calcium carbonate equivalent of the soil horizons was measured by acid titration. The auxiliary variables included derivatives of the digital elevation model, remote sensing indices from the Landsat 8 satellite, and a geopedological map. The principal component analysis (PCA) method was used to select suitable auxiliary variables. In the third phase, the modeling was carried out, digital maps of the soil classes and properties were created, and the models were evaluated. Two different cases were investigated in this study to estimate the calcium carbonate equivalent of the soil. In the first case, artificial neural network models, decision tree analysis, random forest, and the K-nearest neighbor model were used for prediction. The multiple linear regression model was also used to combine the results of the models. Among the models used to predict the equivalent amount of calcium carbonate using the 10-fold cross-validation method, the multiple linear regression (MLR) model had the highest prediction accuracy with a coefficient of determination of 0.796 and a mean square error of 6.514. In the 5-fold cross-validation method, the K-nearest neighbor (KNN) model had the highest predictive accuracy with a coefficient of determination of 0.9845 and a root mean square error of 2.1258. Due to the spatial nature of the 10-fold cross-validation method, the use of this method is preferable to the 5-fold cross-validation method. In addition, the most important auxiliary variables in order of importance to predict the calcium carbonate equivalent in soil were the carbonate index, slope direction, geomorphology, the base level of the catchment network, and the slope of the catchment.