This is the code used for study 'A Novel Finer Soil Strength Mapping Framework Based on Machine Learning and Remote Sensing Images'
Data extracted_gSSURGO_data.xlsx include the basic properties obtained from GSSURGO database to predict USCS soil classification, including sand, clay, silt, CEC, organic carbon, bulk density Data extracted_gSSURGO_data_with LL and PI.xlsx include measured liquid limit and plastic index for comparing with the criterion-based classification with estimated LL and PI by linear model Data SM_dataset_all_depth_FULLY.xlsx include the pixel information from Sentinel-2, SoilGrids, DEM model and estimated hydrodynamic properties for the soil mositure observation stations for model training that can be directly used to train the machine learning regressors. SoilGrids_Pred_USCS_15-30cm.xlsx is part of the prediction of the example study site based on SoilGrids pixels.
USCS_classfiers.py include 4 tree-based models (RF,LCE,XGBOOST,GBDT) for the prediction of USCS soil classification by using the data extracted_gSSURGO_data.xlsx evaluation metrics of accuracy, precision, recall, f1-score, kappa, confusion matrix are all included. SHAP interpretation method is also included in this file.
USCS_criterion_classification.py use the linear model to estimate LL and PI for USCS soil classification.
SM_regressors.py include 4 tree-based models (RF,LCE,XGBOOST,GBDT) for the prediction of soil mositure for multiple soil layers by using the data SM_dataset_all_depth_FULLY.xls mannual changes of directory, depth, models are required
SM_prediction_large_images.py are separate steps for generating the soil moisture map with high resolution considering the issue of limited CPU memory. Due to the large size of digital images, test data is not provided here.
RCI_calculation.py is used to generate the soil strength map on the study site, based on USCS soil classfication map and soil moisture map following the SMSP model.