The study is important because XCO₂ satellite products are not directly interchangeable. Passive instruments such as OCO-2, OCO-3, GOSAT and GOSAT-2 infer carbon dioxide from sunlight reflected by Earth’s surface or clouds in near-infrared and shortwave-infrared absorption bands. They therefore require daylight and are filtered by clouds, aerosols, viewing geometry and surface-reflectance conditions. ACDL instead measures differential absorption along the two-way path between paired laser pulses and a hard target, usually the ground or an optically thick cloud top. Although the lidar signal is range-resolved for locating that target, its carbon-dioxide measurement is path-integrated rather than a set of independent layer-by-layer concentrations. Its retrieval therefore uses a vertical weighting function to describe the effective column sensed by the instrument.

Researchers addressed those physical and sampling differences with a unified framework built on a global 2° by 2° grid. Candidate cells had to exhibit weak spatial carbon-dioxide gradients and coherent short-term variation in the Copernicus Atmosphere Monitoring Service, or CAMS, global inversion product. A second score required sufficient, reasonably even coverage by DQ-1, OCO-2, OCO-3, GOSAT and GOSAT-2, including occasions when at least two sensors sampled the same cell on the same day. The combined score was based on the geometric mean of the coherence and sampling components, and its 90th percentile was selected separately over land and ocean. This conservative rule retained 1,297 cells, including 540 over land and 757 over the ocean, primarily in open Southern Hemisphere waters and regions such as North Africa–Arabia, Europe, Central Asia, East Asia and Australia.

Within each selected cell and day, quality-controlled soundings were averaged rather than matched one by one, and a satellite pair was retained only when the difference between its mean observation times was no more than four hours. The researchers also calculated the theoretical carbon-dioxide difference caused by different observation operators. CAMS model-level profiles were passed through each instrument’s pressure weights, averaging kernels or lidar weighting function, creating a common estimate of what the same atmospheric profile would yield under each product’s effective column definition. The resulting theoretical offset was then added to the evaluated satellite’s observed XCO₂. This model-assisted step addresses column-definition differences, but it does not correct instrumental or retrieval errors, nor does it make CAMS an observational truth.

Independent comparisons with the Total Carbon Column Observing Network, or TCCON, provided a check using ground-based Fourier transform spectrometers that measure direct solar absorption. Daytime DQ-1 retrievals had a correlation coefficient of 0.92 with TCCON, a mean bias of 0.21 parts per million and a root-mean-square error of 1.49 parts per million. The bias confidence interval was −0.08 to 0.48 parts per million, while the RMSE confidence interval was 1.25 to 1.74 parts per million. Those results were comparable to GOSAT, which achieved a correlation of 0.91, a bias of −0.28 parts per million and an RMSE of 1.45 parts per million. OCO-2 and OCO-3 had slightly lower RMSE values of 1.09 and 1.12 parts per million. DQ-1 collocations were fewer because its narrow ground-track geometry less often coincided with TCCON sites, but they were distributed across 24 sites, with no single site contributing more than 11.4% of the matches.

Over the selected comparison regions, DQ-1 was most consistent with the OCO pair. Correlations for DQ-1 against OCO-2 and OCO-3 ranged from 0.86 to 0.93 across land and ocean, mean biases ranged from 0.06 to 0.49 parts per million and RMSE values ranged from 1.26 to 1.71 parts per million. The products also reproduced similar large-scale patterns, with higher XCO₂ over much of the Northern Hemisphere and lower values over Southern Hemisphere oceans. Zonal mean DQ-1 XCO₂ was approximately 420.15 parts per million versus 419.74 for OCO-2 and 419.17 versus 419.00 for OCO-3. Most grid-cell differences lay within about ±1 part per million, with no persistent global bias. The closest agreement appeared over coherent, well-sampled cells, so the results do not imply identical performance in emission hotspots, coastal transition zones or rapidly changing weather.

Regional and ocean diagnostics showed why apparently small residuals should not automatically be attributed to a single sensor failure. Over Australia, DQ-1 was higher than OCO-2 and OCO-3 by 0.76 and 0.57 parts per million, respectively, and it departed from the CAMS-derived diagnostic reference in the opposite direction from the OCO products. In the tropical Pacific, by contrast, the larger scatter appeared on the DQ-1 side of comparisons with the common CAMS reference. Shortwave-infrared reflectance from the MODIS instrument was associated with ocean residuals, but the direction depended on the comparison: DQ-1-minus-OCO-2 bias and RMSE increased with reflectance over ocean, while the corresponding DQ-1-minus-OCO-3 trends decreased. This pair dependence means reflectance was not a universal control on the differences. Likewise, residual statistics declined in groups of ocean cells with higher ERA5 10-metre wind speed, although relationships at individual-cell level were weak and wind speed only indirectly represented roughness and sun-glint geometry.

ACDL’s clearest potential advantage appeared outside passive sensing conditions. Because its laser does not depend on sunlight, DQ-1 can collect observations at night and at high latitudes. The researchers evaluated nighttime retrievals against CAMS-derived XCO₂ under the ACDL column definition, explicitly treating this as a model-reference comparison rather than independent observational validation. Across daily 2° samples, nighttime DQ-1 and CAMS had a correlation of 0.91, a mean difference of 0.04 parts per million and an RMSE of 1.33 parts per million; land RMSE was 1.18 parts per million and ocean RMSE 1.50 parts per million. A zonal diagnostic for June and December 2022 found no persistent daytime-low signal: differences between the models’ ascending-minus-descending changes were +0.09 parts per million in June and −0.05 in December, each with an RMSE near 0.18 parts per million.

Monthly regional comparisons demonstrated the practical complementarity of active and passive observing systems. In ten representative regions, DQ-1, OCO-2, OCO-3, GOSAT and empirically corrected GOSAT-2 generally captured similar seasonal changes and the overall rise in regional XCO₂. The sampling contrast was stark where passive observations were restricted by clouds, aerosols, geometry or daylight. Under a requirement of at least five valid observation days per month, DQ-1 supplied valid monthly means in all 31 months in every one of the ten regions. OCO-3 had none in the Boreal Forest or Antarctica, 20 valid months in India and 22 in the Amazon. In Antarctica, OCO-2 provided 10 valid months, GOSAT-2 two, and GOSAT none, whereas DQ-1 covered all 31. A prolonged OCO-3 interruption associated with storage on the International Space Station reduced its sampling, but comparison across 21 common full months showed that the instrument still supplied only about 51% as many valid observations as OCO-2 in the affected analysis.

Several limitations define what the study can establish. CAMS contributes to region selection, column-definition harmonization and the nighttime comparison, making the inter-satellite results model-assisted consistency estimates rather than fully model-independent measurements of absolute error. Daily 2° aggregation and a four-hour matching window reduce noise and sampling imbalance but can smooth genuine carbon-dioxide gradients near strong sources, coasts, complex terrain or rapidly transported air. Product algorithms, quality filters, versions and bias-correction status also differ, and the official GOSAT-2 correction used TCCON as its reference, limiting the independence of that particular comparison. Sensitivity tests using less restrictive 80th- and 70th-percentile spatial thresholds and temporal windows tightened to one, two or three hours produced only small changes in correlations, biases and RMSE values. Within those boundaries, the evidence supports DQ-1 ACDL as a quantitatively comparable active component that can strengthen temporal continuity and expand coverage in joint satellite carbon-dioxide analyses.

Subject of Research: Unified evaluation of Da Qi-1 active-lidar and passive-satellite XCO₂ measurements

Article Title: Assessing the consistency of XCO2 products from DQ-1 ACDL and passive satellites using a unified comparison framework

Article References: Wang, Y., Han, G., Zhang, H., Huang, Y., Zhou, K., & Gong, W. (2026). Assessing the consistency of X CO 2 products from DQ-1 ACDL and passive satellites using a unified comparison framework. Atmospheric Chemistry and Physics, 26(20), 14793-14822. https://doi.org/10.5194/acp-26-14793-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-14793-2026

Keywords: DQ-1 satellite, Aerosol and Carbon Dioxide Lidar, XCO₂, carbon dioxide monitoring, OCO-2, OCO-3, GOSAT, TCCON, active remote sensing, CAMS, Assessing, consistency