Unprecedented growth in global atmospheric CO2 revealed by daily global XCO2

This repository contains the scripts and notebooks used for global column-averaged CO₂ (XCO₂) reconstruction, including data preprocessing, feature engineering, model training, prediction, and evaluation. The workflow consists of the following main components: Resampling and preprocessing.All scripts related to spatial and temporal resampling are provided in the Resample_codes directory. Feature engineering.Feature_Engineering.ipynb processes resampled feature variables and target variables, performs feature engineering, and converts the original NetCDF (.nc) files into NumPy (.npy) format for efficient model training and analysis. The outputs include processed XCO₂ spatiotemporal feature datasets and corresponding .npy files. Correlation analysis.plot_corr_figure.ipynb performs correlation analysis on non-spatial datasets generated after feature engineering. Model training.XCO2_training.ipynb trains an XGBoost-based regression model for XCO₂ reconstruction using the engineered non-spatial samples and feature datasets. The trained model is saved as xgb_xco2.json. Prediction and reconstruction.Prediction.ipynb applies the trained XGBoost model to generate global XCO₂ predictions. Monthly batch processing is implemented to prevent memory overflow. The outputs include global daily XCO₂ datasets and monthly .npy files covering the global domain. Product evaluation and comparison.Compare_with_other_products.ipynb compares the reconstructed XCO₂ product with TCCON observations and CAMS reanalysis data. Visualization and analysis.plot_xco2_scatter_detail_map_figure.ipynb generates global XCO₂ maps before and after reconstruction and provides regional zoom-in analyses.plot_co2_seasonal_year.ipynb analyzes seasonal cycle characteristics of global XCO₂.plot_co2_trend_time.ipynb evaluates long-term temporal trends. Data format conversion.data_format_transfer.ipynb converts reconstructed XCO₂ data from NumPy (.npy) format back to NetCDF (.nc) format for further analysis and data dissemination. This codebase supports reproducible research for global XCO₂ reconstruction. Users who apply this code in academic work are requested to cite the associated publication.