A high-resolution daily CO₂ dataset for China (2016-2020)

Area of Science:

• Atmospheric Science
• Earth System Science
• Environmental Monitoring

Background:

• High-resolution column-averaged dry-air CO₂ mole fraction (XCO₂) data are crucial for understanding carbon cycles and supporting climate policies.
• Satellite CO₂ retrievals face limitations due to cloud cover and aerosols, resulting in fragmented and discontinuous data.

Purpose of the Study:

• To develop a robust method for reconstructing spatiotemporal XCO₂ dynamics across China.
• To improve the accuracy and continuity of CO₂ measurements for carbon research and climate policy.

Main Methods:

• Utilized an XGBoost model optimized via Bayesian optimization (XGBoost-BO).
• Incorporated SHAP (SHapley Additive exPlanations) for enhanced model interpretability.
• Mapped atmospheric XCO₂ concentrations using multi-source auxiliary parameters.

Main Results:

• Achieved high consistency with OCO-2 satellite observations (R²=0.98, RMSE=0.58 ppm, MAPE=0.07%).
• Validated against TCCON ground-based measurements in China (Hefei: R²=0.92, Xianghe: R²=0.70).
• Demonstrated physically meaningful spatiotemporal XCO₂ dynamics reconstruction.

Conclusions:

• The XGBoost-BO model with SHAP provides a reliable method for reconstructing high-resolution XCO₂ data.
• The reconstructed dataset enhances the characterization of carbon sources and sinks in China.
• This approach supports climate policy goals like carbon peaking and neutrality through improved data continuity and accuracy.