Key dataset and code used in the paper "Abrupt surface water decline during 2023–2024 record warming"

Summary This repository contains the comprehensive dataset and source code supporting the study "Abrupt surface water decline during 2023–2024 record warming." The dataset integrates global-scale streamflow simulations, satellite-derived surface water storage variations, and hydroclimatic attribution analyses. It combines a machine learning framework for hydrological modeling with Landsat/GSW-based surface water monitoring to assess global water dynamics from 1991 to 2024. Dataset Structure The data is organized into two primary directories under the Data folder: 1. Lakes_and_meteorological_driving Code (01_Code): Scripts for processing Landsat/GSW data to calculate surface water storage changes. Hydroclimatic Data (02_Hydroclimatic_Data): Annual time-series (1991–2024) for Streamflow (Q), Precipitation (P), Temperature (TA), Leaf Area Index (LAI), and Water Storage (derived from Landsat & GSW). SHAP Analysis (03_SHAP_Analysis): Machine learning attribution results identifying the drivers of streamflow change (importance of P, TA, and LAI). Geospatial (04_Shapefile): Basin boundaries (HydroSHEDS Level 4) with Aridity Index attributes. 2. Streamflow Simulation Framework: A transfer-learning approach using an ensemble of five machine learning models (Random Forest, GBM, SVM, NN, GPR) weighted by Bayesian Model Averaging (BMA). Coverage: Monthly streamflow simulations for 986 HydroSHEDS basins globally. Contents: Inputs: Trained model objects, BMA weights, and global basin geometries. Outputs: Final simulated monthly streamflow and performance visualizations. Code: MATLAB scripts for model global extrapolation/mapping (step4). Temporal Coverage: 1991 – 2024 Spatial Coverage: Global (HydroSHEDS Level 4 basins) Key Identifiers: Basins are linked across all files via the HYBAS_ID.