Marko Scholze

Atmospheric inverse modelling and ecosystem data assimilation are two complementary approaches to estimate CH₄ emissions. The inverse approach infers emission estimates from observed atmospheric CH₄ mixing ratio, which provide robust large scale constraints on total methane emissions, but with poor spatial and process resolution. On the other hand, in the ecosystem data assimilation approach, the fit of an ecosystem model (e.g. a Dynamic Global Vegetation Model, DGVM) to eddy-covariance (EC) flux measurements is used to optimize model parameters, leading to more realistic emission estimates. Coupled data assimilation frameworks capable of assimilating both atmospheric and ecosystem observations have been shown to work for estimating CO₂ emissions, however ecosystem data assimilation for estimation CH₄ emissions is relatively new. Kallingal et al. (2024a) developed the GRaB-AM data assimilation system, which performs a parameter optimization of the LPJ-GUESS against eddy-covariance estimation of CH₄ emissions. The optimization improves the fit to EC data, but the validity of the estimate at large scale remained to be tested. In this study, we confronted CH₄ emissions optimized using the GRaB-AM system to atmospheric CH₄ observations and to emission estimates from the LUMIA regional atmospheric inversion system (Monteil and Scholze, 2021). We found that the two approaches lead to very consistent corrections to the prior emission estimate from natural wetlands, with roughly a halving of the annual total compared to the LPJ-GUESS prior. Our findings confirm the interest of the GRaB-AM approach to constrain the contribution of natural ecosystems to the total methane budget, which is difficult to achieve for atmospheric inversions outside regions where emissions from natural ecosystems clearly dominate the emission budget.