Transformer-Based Inverse Radiance Transfer Framework for Global Carbon Monoxide Retrieval Using Metop-B/IASI

Abstract

Carbon monoxide (CO) plays a significant role in both climate and atmospheric chemistry by modulating hydroxyl radical concentrations. Accurate and timely global monitoring of CO concentrations is therefore crucial for assessing its climatic and environmental impacts. This study proposes a transformer-based approach called inverse radiative transfer transformer (IRTT) for retrieving global CO concentrations from Level-1C radiance data acquired by the infrared atmospheric sounding interferometer (IASI) instruments aboard Metop-B satellite. We developed an encoder-only transformer to establish the mapping between radiance and CO columns using a dataset simulated by a radiative transfer model (RTM), achieving a coefficient of determination (R2) greater than 0.999 on the validation set. The proposed model was subsequently applied to IASI-observed spectra to derive global CO vertical column densities (VCDs) from January to December 2022. Compared with the IASI products, the retrieved annual mean CO VCDs yield a correlation coefficient (R) of 0.963. In four representative regions, the model achieves high R values exceeding 0.85 with relative biases ranging from 2.42% to 6.57%. Compared with network for the detection of atmospheric composition change (NDACC) FTIR observations, IRTT yields correlation coefficients ranging from 0.77 to 0.96 across five sites with sufficient data coverage. Across 20 total carbon column observing network (TCCON) sites with sufficient data, IRTT achieves R values ranging from 0.72 to 0.94. These results are consistent with those obtained from the official Level-2 products. The model also offers high computational efficiency, requiring only ~10 s per orbit. The proposed approach shows strong potential for extension to other trace gases and satellite missions.