Accurately monitoring the spatiotemporal distribution of atmospheric CO₂ is essential for understanding the carbon cycle, formulating effective emission reduction strategies, and achieving carbon neutrality. However, research in this area is constrained by a lack of high-quality carbon monitoring data. While satellite remote sensing technologies can provide atmospheric CO₂ data with a high spatial resolution and broad coverage, inherent limitations often result in substantial data gaps. Addressing these gaps and generating high-resolution, gap-free CO₂ concentration datasets have thus become a critical research focus. This study utilizes column-averaged dry air CO₂ mole fraction (XCO₂) data retrieved from the OCO-2 satellite (2021–2022) as its observational input. It integrates XCO₂ data from the coarse-resolution CarbonTracker (CT) reanalysis ( 3° × 2°) and multiple environmental variables as the predictive inputs to develop and optimize an extreme random tree (ET) model. The goal is to generate a daily, high-resolution (0.01°) XCO₂ dataset with full spatial coverage across China. Spatiotemporal cross-validation demonstrates the model's high accuracy and stability, yielding an R² of 0.93 and an RMSE of 0.75 ppm. Independent validation using data from the TCCON and WDCGG sites further confirms the model’s effectiveness in capturing atmospheric CO₂ dynamics. This approach not only bridges critical gaps in the existing observational networks but could also enhance carbon cycle analyses and related research. Additionally, it can be extended to longer time series and broader regions, providing robust scientific support for policymakers in climate decision-making.
Previous Article in event
Next Article in event
Next Article in session
Daily High-Resolution XCO2 Mapping across China Using OCO-2 Data and Machine Learning Model
Published:
30 May 2025
by MDPI
in The 7th International Electronic Conference on Atmospheric Sciences
session Air Pollution Control
Abstract:
Keywords: XCO₂ ;OCO-2 satellite;Full spatial coverage;Machine learning
Comments on this paper
