EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Selective calibration of direct OCO-2 level-1B XCO2 retrieval across held-out TCCON stations

Wei Feng

Frontiers in Earth Science · 20.11.2026 · 14

DOI: 10.3389/feart.2026.2017776 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Frontiers in Earth Science, online erschienen am 20.11.2026. Verfasst von Wei Feng. Die Arbeit nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Introduction

Prescreening, retrieval failure, and quality filtering leave many Orbiting Carbon Observatory-2 (OCO-2) soundings without usable estimates of the column-averaged dry-air mole fraction of CO 2 (XCO 2 ).

Methods

We developed a selective calibration framework to retrieve XCO 2 directly from Level-1B spectra and improve its accuracy at locations outside the training domains. A neural network ensemble learns from quality-filtered Level-2 labels. Predictions with low disagreement across training checkpoints are accepted for residual calibration against measurements from the Total Carbon Column Observing Network (TCCON). We evaluated the framework at seven TCCON stations using leave-one-station-out evaluation and independent annual training for 2020 and 2021. Each target station was withheld from all model development.

Results

Screening retained 86.7% of matched soundings in 2020 and 89.1% in 2021. On the accepted observations, calibration reduced root mean square error (RMSE), averaged equally across stations, from 2.666 to 1.907 ppm in 2020 and from 3.390 to 2.508 ppm in 2021. RMSE decreased at all seven held-out stations in both years, and the average improvement persisted across all 12 spatial and temporal matching windows. Calibration also improved estimates for soundings without matched Level-2 records, which comprised approximately half of the accepted observations in each year.

Discussion

These results establish that residual corrections learned at source stations can improve direct retrieval at unseen stations, extending externally evaluated XCO 2 estimates to observations outside the conventional Level-2 population.

Zitieren

APA

Feng, W. (2026). Selective calibration of direct OCO-2 level-1B XCO2 retrieval across held-out TCCON stations. Frontiers in Earth Science, 14. https://doi.org/10.3389/feart.2026.2017776

Vancouver

Feng W. Selective calibration of direct OCO-2 level-1B XCO2 retrieval across held-out TCCON stations. Frontiers in Earth Science. 2026;14. doi:10.3389/feart.2026.2017776

BibTeX
@article{Feng2026,
  author = {Feng, Wei},
  title = {Selective calibration of direct OCO-2 level-1B XCO2 retrieval across held-out TCCON stations},
  journal = {Frontiers in Earth Science},
  year = {2026},
  volume = {14},
  doi = {10.3389/feart.2026.2017776}
}
Kostenlos starten →