Citation If you use the data from this repository in your research, please cite our paper: Zhao, J., Liu, Y., Li, G., Zuo, H., 2026. Thresholds in the controls of denudation rates: A global analysis of tectonic, climatic and biological factors based on machine learning. Earth and Planetary Science Letters 674, 119750. https://doi.org/10.1016/j.epsl.2025.119750 Description This repository contains the primary dataset and key results from our global analysis of denudation rates. In our study, we developed a machine learning model to predict denudation rates based on a compilation of ~4,000 river basins worldwide. This dataset serves as the foundation for the model and its final outputs. The data provided here allow for the full reproduction of our model evaluation and the visualization of our main findings. The model itself was trained and applied using Google Earth Engine (GEE) due to the large scale of the global predictor datasets, but the core training data is provided here for transparency and re-use. Files in this Repository This archive contains three key files: basin_denudation_predictors.csv (Input Data) Description: This is the core dataset used to train and validate our machine learning model. It contains data for 3,967 river basins. Format: Comma-Separated Values (CSV). Content: Each row represents a single river basin. The columns include: log_denud: The natural logarithm of the observed denudation rate (in mm/kyr). This was the target variable for the model. 14 Predictor Variables: Basin-averaged values for the environmental predictors used in the study, including topographic (e.g., Slope, Elevation), climatic (e.g., MAT, MAP), and biological (e.g., NDVI, EVI) factors. Usage: This file can be used by other researchers to train their own models, replicate our model's performance, or conduct further analyses on the relationships between denudation rates and environmental controls. model_predictions_vs_observations.csv (Model Evaluation) Description: This file contains the results of our model's performance on the validation dataset. It directly compares the observed denudation rates with the rates predicted by our trained Random Forest model. Format: Comma-Separated Values (CSV). Content: Each row corresponds to a basin from the original dataset. Key columns include: Denudation_OCTOPUS: The observed denudation rate from OCTOPUS database. Denudation_RandomForest: The predicted denudation rate from our model. Denudation_Zondervan: The predicted denudation rate calculated from the map of Zondervan et al. (2023). Denudation_Larsen: The predicted denudation rate calculated from the map of Larsen et al. (2014). Usage: This file is essential for reproducing the model validation statistics (e.g., R²) and the scatter plots (like Figure 7B-D in our paper) that demonstrate the model's accuracy. Global_denudation_map_1km.tif (Primary Output) Description: This is the main scientific product of our study—a global map of predicted denudation rates at a 1-km spatial resolution. Format: GeoTIFF, a standard raster format compatible with GIS software (e.g., QGIS, ArcGIS) and programming languages (e.g., R, Python). Content: Spatial Resolution: ~1 km (0.008333 degrees). Coordinate Reference System (CRS): WGS 84 (EPSG:4326). Units: The pixel values represent the predicted denudation rate in millimeters per thousand years (mm/kyr). Coverage: Global land surface, with areas covered by major ice sheets (Antarctica, Greenland) and permanent snow/ice masked out. Usage: This map can be used for large-scale geomorphological analysis, as an input for global biogeochemical models, or for identifying global hotspots of erosion and sediment production. Methodology Summary The .csv files were generated by compiling observed denudation rates and extracting predictor values for each basin polygon using Google Earth Engine (GEE). A Random Forest regression model was trained on these data within GEE. The Global_denudation_map_1km.tif was then produced by applying this trained model to global raster layers of the 14 predictor variables. License The data in this repository are made available under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to share and adapt these materials for any purpose, provided you give appropriate credit by citing our original publication.