Model evaluation results and Python scripts for ensemble machine-learning inversion of suspended sediment concentration in high-sediment rivers

This repository provides model evaluation results and Python scripts supporting the study entitled "Ensemble machine-learning inversion of suspended sediment concentration in high-sediment rivers using multi-sensor remote sensing data." The repository contains the results of 100 repeated random training–validation experiments and Python scripts for evaluating model performance and generalization under different temporal, seasonal, spatial, and suspended sediment concentration (SSC) conditions. ## File descriptions **01_SSC_Model_Comparison.xlsx** Contains the results of 100 repeated random training–validation experiments comparing four base machine-learning models (SVR, RF, XGBoost, and MLP) and three stacking ensemble models. - `100_Runs_Summary`: Mean and standard deviation of model performance metrics across 100 experiments.- `Detailed_Runs`: Performance metrics for each individual experiment, including R², RMSE, MAPE, Bias, and Ridge regression coefficients.- `RunInfo`: Experimental settings and configuration information. **02_SSC_Model_Comparison_100_Repeated_Random_Splits.txt** Python script for evaluating the performance and stability of different machine-learning models through 100 repeated random training–validation partitions (70% training and 30% validation). **03_SSC_Model_Temporal_Generalization_Experiments.txt** Python script for evaluating model generalization between earlier and later observation periods through bidirectional training–validation experiments. **04_SSC_Model_Seasonal_Generalization_Experiments.txt** Python script for evaluating model generalization between flood-season and non-flood-season conditions. Samples are ranked according to their temporal distance from July 31, the reference date representing the center of the flood season. **05_SSC_Model_Spatial_Generalization_Experiments.txt** Python script for evaluating model generalization between upstream and downstream river sections through bidirectional spatial training–validation experiments. **06_SSC_Model_Concentration_Generalization_Experiments.txt** Python script for evaluating model generalization between relatively high-SSC and low-SSC conditions. Samples are ranked according to the mean SSC of their corresponding hydrological stations.