Adaptive Reference Spectrum-Based Spectral and Radiometric Calibration of GF-5B GMI Using Physics-Based Radiative Transfer Simulation

Accurate spectral and radiometric calibration is essential for reliable greenhouse gas retrieval from satellite hyperspectral observations. However, long-term in-orbit operation may introduce wavelength shifts and radiometric response inconsistencies, particularly within narrow atmospheric absorption bands, reducing the reliability of Level-1 radiance products. This study proposes an observation-dependent adaptive reference spectrum framework for spectral and radiometric calibration of the Greenhouse Gas Monitoring Instrument (GMI) onboard the GaoFen-5B (GF-5B) satellite. The proposed framework integrates the SCIATRAN radiative transfer model with the instrument line shape (ILS) of GMI to generate physically consistent reference spectra under varying observation conditions. Sensitivity analysis was performed to identify dominant parameters affecting simulated radiance spectra, and an adaptive reference spectrum library was constructed by considering surface reflectance, solar zenith angle, aerosol conditions, and geographic characteristics. The generated reference spectra were subsequently used to estimate spectral shifts and correct radiometric response deviations between observed and simulated spectra. GF-5B GMI observations acquired from September 2021 to December 2024 over Dunhuang, Golmud, and Railroad Valley (RRV) were used for validation. Results demonstrate that the proposed framework effectively reduces spectral offsets, improves radiometric consistency, and enhances spectral agreement for the O2, CO2-1, and CO2-2 bands. The multi-site validation confirms the robustness of the proposed approach under different surface and atmospheric conditions. This study provides a physically constrained and transferable calibration framework for improving the long-term spectral and radiometric consistency of greenhouse gas hyperspectral observations and supporting more reliable atmospheric composition retrievals.