Impact of Local-Scale Effects in Methane (CH₄) Inversions on Model-Observation Discrepancies

Accurate estimates of greenhouse gas emissions are critical for determining the effectiveness of mitigation strategies under the Paris Agreement. These estimates are commonly derived by atmospheric inversion frameworks, which combine atmospheric transport models with in situ observations to obtain greenhouse gas fluxes. However, regional inversions are often challenged by local-scale signals in atmospheric measurements, that are insufficiently represented by the models. If not properly accounted for, these can introduce biases in inverse flux estimates undermining the reliability of emission estimates.

To address this limitation, observational data has typically been filtered for local influences before being used in inversion simulations, based on assumptions such as stable boundary conditions or wind speed. To make full use of the available dataset, we implemented an observation-dependent model-data uncertainty in the inversion optimisation process, allowing local signals to be explicitly considered. This approach has been applied to CH₄ inversions over Europe using the mesoscale Jena CarboScope-Regional (CSR) system at 0.25° × 0.25° resolution.

To determine the time varying model-data uncertainty based on the local influence signal, a leave-one-out cross validation was performed for ground based in situ data of 47 atmospheric stations, excluding one station per inversion simulation. By determining the difference between modelled and observed concentrations, a model-data mismatch was estimated across station categories defined by surrounding land type. These estimates were then combined with local signal features, resulting from low wind speeds, atmospheric stability, and concentration spikes using a multivariate regression. The derived model-data mismatch function was applied to adjust the data weighting in the inversion enabling the inclusion of the observational dataset without discarding any measurements.

In this presentation, we demonstrate the potential of this novel approach to improve the robustness of regional CH₄ inversions and to reduce the bias from local-scale signals.