Developing unbiased estimation of atmospheric methane via machine learning and multiobjective programming based on TROPOMI and GOSAT data

Accurate monitoring of atmospheric methane concentration (XCH4) is relevant to improving carbon accounting and climate change attribution. Nevertheless, the commonly used full-physics carbon retrieval algorithm suffers from intensive computing burden and many algorithmic constraints. Aiming at providing a more efficient solution to advance global methane mapping, a novel XCH4 retrieval algorithm for monitoring atmospheric methane, that is, the UNbiased methane estimation with the aid of MAchine learning and MultiObjective programming (UNMAMO), was introduced. By taking advantage of a multiobjective programming approach, TROPOMI bands with apparent methane absorption features were first pinpointed via radiative transfer simulations, and band ratios were then calculated between methane sensitive and adjacent insensitive bands to enhance methane signal-to-noise ratio. Machine-learned prediction models were subsequently established using random forest by taking GOSAT XCH4 retrievals as the learning target with TROPOMI band ratios as the critical proxy variables. For demonstration, global XCH4 was mapped on a daily basis in 2021 with a grid resolution of 0.05 degrees. The validation results confirmed a better agreement of our XCH4 retrievals than the operational TROPOMI XCH4 product with ground-based TCCON methane observations, with a correlation coefficient of 0.91 and root mean square error of 17.16 ppb. Meanwhile, our XCH4 retrievals offered nearly twice as much spatial coverage relative to the operational product. Moreover, benefiting from the rationale of band ratios, surface albedo- and aerosol-related retrieval biases in the operational product were largely mitigated in our UNMAMO retrievals. Overall, UNMAMO provides a new way to map global XCH4 with higher accuracy and computing efficiency, making it better than the operational full-physics retrieval algorithms of its kind. The accuracy-enhanced methane retrievals enable us to better resolve global methane emissions from different sectors in support of global carbon accounting and sustainable development.