Substantial Discrepancies Across Global Satellite XCO2 Products: A Systematic Evaluation

Zeyu Ma, Jiani Tan, Ruixun Xia, Yang Liu, Andrew P. Morse, Qing MuRemote 2026 · DOI

TL;DR

Accurate monitoring of atmospheric CO2 is crucial for climate research, but different satellite products show significant discrepancies. This study systematically evaluates twelve global satellite XCO2 products, finding that the ACOS-OCO-2 product offers the best overall performance. The findings provide essential guidance for selecting appropriate XCO2 data for climate and sustainability research.

What Problem Does This Paper Address?

Atmospheric carbon dioxide (CO2) is a major greenhouse gas, and precise monitoring of its concentration is vital for understanding and addressing climate change. Satellite remote sensing is the primary tool for this global monitoring, providing broad coverage. However, a multitude of satellite missions and retrieval algorithms have led to numerous distinct XCO2 (column-averaged CO2) products. The challenge is that these products often exhibit substantial discrepancies, making it unclear which ones are most reliable and suitable for specific research applications. This lack of a systematic, comprehensive evaluation hinders informed selection and application of these valuable datasets for climate and sustainability research.

Methods

The researchers conducted a comprehensive evaluation of eleven individual satellite XCO2 products and one ensemble product (EMMA). They employed a systematic approach:

  • Spatiotemporal Matching: A crucial step was establishing robust matching criteria between satellite data and ground-based Total Carbon Column Observing Network (TCCON) measurements. This involved a sensitivity test of collocation criteria, ultimately settling on a 1.5° Chebyshev distance for spatial matching and a ±1 hour window for temporal matching.
  • Evaluation Metrics: Three standard statistical metrics were used to assess product performance against TCCON:
  • R² (Coefficient of Determination): Measures the proportion of variance in TCCON data explained by the satellite product, indicating the strength of the linear relationship.
  • RMSE (Root Mean Square Error): Quantifies the magnitude of errors, with higher values indicating larger deviations.
  • MAE (Mean Absolute Error): Provides a robust measure of the average bias magnitude.
  • Bias Analysis: Scatter plots were used to visualize biases, with data points above the 1:1 line indicating overestimation and points below indicating underestimation.
  • Ensemble Mean Distribution: To analyze inter-product deviations, an ensemble mean was computed from all available products, serving as a reference. Global and seasonal deviations were then calculated for each product relative to this ensemble mean.

Datasets

The study utilized data from the following sources:

  • Satellite XCO2 Products:
    • Envisat: BESD-Envisat, WFMD-Envisat
    • GOSAT: ACOS-GOSAT, OCFP-GOSAT, SRFP-GOSAT, NIES-GOSAT
    • GOSAT-2: NIES-GOSAT-2, SRFP-GOSAT-2
    • OCO-2: ACOS-OCO-2
    • OCO-3: ACOS-OCO-3
    • TanSat: OCFP-TanSat
    • Ensemble: EMMA-Ensemble
  • Ground-Based Reference Data:
    • TCCON (Total Carbon Column Observing Network): Data from 31 stations, processed using the GGG2020 software, served as the "ground truth" for validation.

Key Insights and Ideas

  • No Single "Best" for All: While ACOS-OCO-2 emerges as the top performer globally, the study highlights that product performance varies significantly with latitude, season, and even specific geographical locations. This underscores the need for careful, context-specific product selection.
  • Newer Isn't Always Better: Surprisingly, newer satellite products (like GOSAT-2) did not consistently show superior performance compared to older ones, sometimes even underperforming due to factors like uncorrected biases or shortened retrieval windows.
  • Ensemble Benefits and Drawbacks: The EMMA-Ensemble product effectively reduces overall bias and improves data consistency. However, it can inherit localized high biases from its constituent products, indicating that data fusion is not a magic bullet and requires careful consideration of input data quality.
  • Tropical Challenges: The tropics (20°S–20°N) consistently show the largest uncertainties, likely due to complex atmospheric conditions, high aerosol loading, and spatially heterogeneous land cover, which challenge retrieval algorithms.
  • High-Latitude Accuracy: In contrast, high-latitude regions (60–80°N) generally exhibit the highest accuracy, particularly during summer, suggesting more favorable conditions for satellite retrievals in these areas.

Key Results

  • Top Performer: The ACOS-OCO-2 product demonstrated the best overall performance globally, achieving the highest R² (0.96) and the lowest RMSE (1.35 ppm) and MAE (1.03 ppm) when compared to TCCON.
  • Data Availability: OCO satellites generally provide the highest data availability, significantly exceeding that of GOSAT satellites.
  • Latitudinal Variations: Accuracy is generally higher in high latitudes (60–80°N) during summer, with optimal RMSE < 1.0 ppm. The tropics exhibit the largest uncertainties (RMSE > 2 ppm).
  • Seasonal Biases: Systematic biases are most pronounced in winter, with MAE increasing by 0.3–1.0 ppm compared to other seasons.
  • Specific Product Performance:
    • BESD-Envisat performed significantly better than WFMD-Envisat.
  • Among GOSAT products, ACOS-GOSAT was the best.
  • NIES-GOSAT-2 showed the poorest performance among the GOSAT series.
  • OCFP-TanSat showed moderate performance but had limited temporal coverage.
  • ACOS-OCO-3 showed increased overall bias and higher frequency of larger errors compared to OCO-2, potentially due to ISS orbital constraints.

Limitations and Future Work

  • TCCON Network Imbalance: A primary limitation is the geographical imbalance of TCCON stations, with a heavy concentration in the 20–40°N latitude band. This means that while statistical results are robust at station locations, they may not fully represent broader continental-scale conditions, especially over oceans and in regions like Africa.
  • Seasonal Data Gaps: In some high-latitude regions, matched data for winter were unavailable, limiting evaluation for those periods.
  • Temporal Mismatch: The varying operational periods of satellites (e.g., OCFP-TanSat's limited record) can introduce biases related to inter-annual variability.
  • Attribution of Biases: While specific biases are identified, their attribution to distinct physical mechanisms (e.g., aerosol models) is beyond the scope of this study. Future work should focus on disentangling these algorithmic elements to improve retrieval accuracy.
  • Future Ensemble Development: The study suggests that future XCO2 ensemble products need to address inherited errors from constituent datasets to ensure more robust data fusion.

Summary updated

Abstract

Accurate monitoring of atmospheric carbon dioxide (CO2) is critical for addressing climate change, as CO2 is one of the dominant greenhouse gases. Satellite remote sensing remains the primary method for monitoring column-averaged CO2 (XCO2), yet different satellite missions and retrieval algorithms generate distinct XCO2 products. Thus, recommendations for selecting appropriate XCO2 products remain unclear due to a lack of systematic evaluation of XCO2 products. Here, we present a comprehensive evaluation of eleven XCO2 products from major satellite missions—including the Environmental Satellite (Envisat), Greenhouse Gases Observing Satellite (GOSAT/GOSAT-2), Orbiting Carbon Observatories (OCO-2/OCO-3), and TanSat—alongside one ensemble product based on the ensemble median algorithm (EMMA). We assess their spatiotemporal coverage and performance using Total Carbon Column Observing Network (TCCON) measurements as reference, evaluating both at global and regional scales across seasons. Our results reveal distinct latitudinal and seasonal variations in the evaluation results. Most products show the highest accuracy at 60–80°N in summer (optimal root mean square error &lt; 1.0 ppm), while the largest uncertainties appear in the tropics (20°S–20°N; root mean square error &gt; 2 ppm). Furthermore, systematic biases are most pronounced during winter, with mean absolute error increasing by 0.3–1.0 ppm compared to other seasons. Among the twelve satellite XCO2 products, the Atmospheric CO2 Observations from Space-Orbiting Carbon Observatory-2 (ACOS-OCO-2) product shows the best overall performance globally. These results provide practical guidelines for the informed selection and application of satellite-derived XCO2 products in climate research.
BibTeX
@inproceedings{ma2026substantial,
  author = {Zeyu Ma and Jiani Tan and Ruixun Xia and Yang Liu and Andrew P. Morse and Qing Mu},
  title = {{Substantial Discrepancies Across Global Satellite XCO2 Products: A Systematic Evaluation}},
  booktitle = {Remote},
  year = {2026},
  doi = {10.3390/rs18020371},
}
Figure