Date for "Climate Aridity Amplifies Global Riverine Nitrate Beyond Human Inputs"

This repository contains:

1. "predicted_nitrate_concentrations_global.txt" - global dataset with ensemble averaged long-term mean ntirate concentration (mean_nit, mg/l), their standard deviation (sd_nit, mg/l), coefficient of variation (cv_nit, %), median (p50_nit, mg/l), 10th (p10_nit, mg/l) and 90th (p90_nit, mg/l) percentile of predicted nitrate concentrations across 1000 models in 2895895 global river segments (MERIT Hydro COMID used as identifiers)
2. “global_river_segment_classes” – global dataset indicating the landuse, aridity, income, rural-urban and stream order classes along with directly draining area and population residing there for 2895895 global river segments.
3. "trained_models.RData " – 1000 XGBoost models along with their preprocessing models used to predict global nitrate concentrations.
4. “BRT_input_data_global.txt” – global dataset of catchment characteristics used to drive the model for predicting global nitrate concentrations.
5. “trained_model_results_summary.zip” – it contains following files:
1. “BRT_model_performance.txt”: Model performance metrics for 1000 models across training and testing data
2. “BRT_mean_importance.txt”: Mean importance of catchment characteristics across 1000 models
3. “BRT_predictions_7213sites.txt”: Model predictions from 1000 models for 7213 sites with data used in training the model
4. “SHAP_model_338.txt”: SHAP values for catchment characteristics for the model with performance closest to the median model performance across 1000 models.
5. “SHAP_interaction_model_338.txt”: SHAP interaction values for catchment characteristics for the model with performance closest to the median model performance across 1000 models.