Compound dry-hot events intensify widespread lake deoxygenation in China

This dataset supports the research on "Compound dry-hot events intensify widespread lake deoxygenation in China", encompassing three core components to enable reproducible analysis of lake dissolved oxygen (DO) dynamics and climate change impacts.### 1. Historical DO Estimation Dataset (Lake_surface_DO_CHN (2000-2020).csv)This file contains monthly surface DO concentrations (mg/L) and DO percent saturation (DO%sat, %) for 11,571 Chinese lakes over the 2000–2020 period. Derived from machine learning (Random Forest) reconstruction integrating multi-source data (CNEMC monitoring observations, hydro-climatic variables, lake morphometry, and anthropogenic activity metrics), the dataset includes key variables: lake identifiers (Hylak_id), geographic coordinates (Lon, Lat), annual/monthly DO/DO%sat values, and auxiliary environmental predictors (e.g., elevation, lake area).<br>### 2. Future DO Projection Dataset (Lake_surface_DO_CHN (2021-2099).csv)This dataset provides future projections of monthly DO and DO%sat under three Shared Socioeconomic Pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5) from 2021 to 2099. Projected using the optimized machine learning model and CMIP6 climate outputs, supports assessments of long-term deoxygenation risks across emission trajectories.<br>### 3. R Codes for Machine Learning Construction (R Codes for ML Construction.text)This script details the full workflow for DO/DO%sat reconstruction and projection: (1) data preprocessing (integration, standardization, missing value handling); (2) machine learning model development (comparison of RF, SVM, XGBoost, etc., recursive feature elimination, hyperparameter tuning via 10-fold cross-validation and bootstrapping); (3) model validation (MAE, RMSE, R²) and interpretation (permutation importance, SHAP analysis).