Satellite remote sensing offers several advantages, including contactless measurements, wide observation range, high sampling frequency, excellent spatiotemporal continuity, and low cost per measurement. Satellite-based observations of atmospheric carbon dioxide (CO₂) are crucial for China’s major strategic goals of “carbon neutrality and carbon peaking,” and they also support the current global “carbon inventory” task. Several foreign carbon-monitoring satellites, such as Japan’s Greenhouse Gases Observing Satellite (GOSAT)-1/2 and the American Orbiting Carbon Observatory (OCO)-2/3, have achieved operational high-precision detection of atmospheric CO₂ and provide internationally recognized data products. Since the launch of China’s first domestic carbon-monitoring satellite, TanSat, in December 2016, retrieving global atmospheric CO₂ concentrations with high precision from TanSat Level 1B (L1B) data has been a major research focus. However, the instability of its spectral performance-due to factors such as cosmic radiation exposure, launch vibrations, and changes in environmental temperature and pressure-has significantly affected the retrieval success rate and hindered accuracy improvements. High-quality spectra are essential for accurate CO₂ retrieval, but many previous studies have overlooked this requirement. In this study, we quantify and correct wavelength and radiance inaccuracies in TanSat’s spectra to enhance spectral quality, aiming for more accurate and reliable CO₂ detection compared to existing studies.

As the key original calibration parameters measured in the ground laboratory seem unsuitable for TanSat’s on-orbit measurements, we continually adjust the wavelength shift and squeeze until we obtain an optimal solution between TanSat’s direct solar spectra and the high-resolution, high-reliability Kurucz solar spectrum. Then, we quantify the wavelength shift and correct it with high time frequency. After that, we choose the region of 15°N-20°N and 0°W-15°W, which is located in the Sahara Desert and has a similar surface albedo, and perform simulation experiments of radiative transfer (Fig. 3). Under the low-cloud and aerosol scenario, we construct a simulated optical environment by using the parameters of aerosol optical density, albedo, volume fraction of CO₂, atmospheric profiles, and geometric angle of satellite observations provided by multisource data. We use the libRadtran model to simulate the spectra that should be obtained by observing TanSat from the Earth’s surface. Then, we derive radiometric calibration coefficients from the simulated and measured spectra, which serve as a basis for evaluating and correcting radiance distortions, optical structure, and other issues. As shown in Fig. 4, we develop a scheme to invert O₂ and CO₂ vertical column density synchronously for TanSat X_CO2 retrieval based on the iterative maximum a posteriori differential optical absorption spectroscopy (IMAP)-DOAS algorithm, a forward model developed specifically for inverting near-infrared absorbing gases, characterized by direct nonlinear iterative fitting of the optical density spectrum. Moreover, we optimize the configuration of the retrieval algorithm by reconstructing the solar irradiance spectrum, constructing a priori reference spectral database with high spatiotemporal resolution, updating the slit function, and building air mass factor lookup lists. Finally, we evaluate the accuracy of our retrieved X_CO2 data by verifying our results against global ground-based TCCON sites. In addition, to quantify the difference between our results and other similar satellite products, we implement a cross-comparison among TanSat, GOSAT, and OCO-2.

Our on-orbit recalibrations reveal that TanSat’s L1B spectra in the O2A, WCO2, and SCO2 channels experience significant wavelength shifts since launch. As shown in Fig. 5, the wavelength shifts across the nine footprints (FPs) exhibit similar trends over time. Initially, the wavelength shift in the O2A and WCO2 channels reaches approximately 10% and 30% of the resolution, respectively. After June 2018, the shifts increase rapidly, causing notable spectral instability. The wavelength shift in the SCO2 band is particularly severe, reaching up to about 3.75 times the spectral resolution. On-orbit radiometric recalibrations identify inherent optical structure and radiance biases due to variations in instrument performance. For the O2A band, radiometric deviations are within ±5% initially, whereas significant instability is observed between November 2017 and January 2018, with deviations exceeding 5% for most FPs (Fig. 6). The WCO2 channel shows more intense radiance deviations, reaching ±10% at most wavelengths (Fig. 7). Deviations at the wavelength edges of FPs are even greater with some exceeding 15%, and worsening over 20% as instrument performance deteriorates. Sensitivity experiments demonstrate that correcting orbital wavelength and radiance can significantly improve inversion results, optimizing the success rate, root mean square (RMS) of fitting, and uncertainty by 24%, 15%, and 30%, respectively (Fig. 8). Using the recalibrated spectra and our retrieval algorithm, we obtain TanSat X_CO2 results for approximately one and a half years (March 2017 to September 2018). Validation with TCCON data confirms a global detection accuracy of 1.24×10⁶, with an average bias of only 5×10^-8 (Fig. 9). TanSat’s detection accuracy is better than 2×10^-6, and the average bias is within ±1×10^-6 near most global TCCON sites (Table 3). Cross-comparison with GOSAT and OCO-2 show that TanSat’s X_CO2 product reliability is comparable to current international standards (Fig. 10).

Our study reveals that the long-term instability in TanSat’s on-orbit spectral performance is a crucial factor affecting retrieval success and accuracy. On-orbit recalibration significantly improves retrieval quality, with the optimized retrieval algorithm achieving accuracy better than 1.3×10^-6. In the future, TanSat-2, China’s new-generation carbon-monitoring satellite, is expected to improve instrument performance and hardware parameters, including larger orbit widths, shorter revisit periods, and higher spatial coverage, potentially improving CO₂ detection accuracy to within 1×10^-6. However, TanSat-2 will be placed in large elliptical orbits, resulting in substantial distance variations from the receiving station and significant signal amplitude fluctuations. Maintaining the optimal instrument conditions and achieving high spectral quality will be key challenges. The spectral correction and inversion scheme developed in our study provides a new solution for addressing similar issues that TanSat-2 might encounter.