Global 20-year time series of XCO2 concentrations derived from satellite observations and interpolations to analyze XCO2 anomalies caused by wildfires

Abstract

Accurate, long-term records of atmospheric CO${}_2$ concentrations are crucial for understanding carbon cycle dynamics and evaluating the impacts of natural and anthropogenic emissions. In this study, we present a global dataset of column-averaged CO${}_2$ concentrations (XCO${}_2$) covering the period 2003–2022, generated using an enhanced version of the XCO2SAT+ method. This approach integrates satellite observations from four missions (ENVISAT, GOSAT, GOSAT, GOSAT-2 and OCO-2) harmonized through the EMMA v4.5 product, and applies a novel spatio-temporal interpolation algorithm to estimate monthly mean XCO${}_2$ values at 23 958 locations worldwide ($\approx$95 km grid spacing). The dataset provides XCO${}_2$ estimates over global land surfaces and adjacent coastal regions. Validation against 28 TCCON sites yielded a mean absolute error (MAE) of $0.76ppm$, comparable to NOAA’s CarbonTracker model ($\mathrm{MAE}=0.71ppm$). The method reliably reconstructs XCO${}_2$ time series even in regions with sparse satellite coverage, while preserving spatial and seasonal variability. As a demonstration, we identify XCO${}_2$ anomalies linked to major wildfire events, including North American mega-fires in 2018 and recurring fire activity in tropical northern Africa, with localized monthly anomalies exceeding 2 ppm–4 ppm. Comparison with recent machine learning (ML) products confirms that the XCO2SAT+ method achieves higher accuracy across multiple TCCON sites and time frames. This semi-empirical dataset provides a valuable alternative to assimilation-based models, supporting climate research, emission monitoring, and the detection of CO${}_2$ perturbations from extreme events at high spatial and temporal resolution.