Dense greenhouse-gas (GHG) observations, particularly from satellites, require fast atmospheric transport models that link surface emissions to these observations. We introduce GenGHG, a generative machine learning (ML) model that emulates footprints from the Stochastic Time-Inverted Lagrangian Transport (STILT) model. These footprints estimate how a unit of emissions would alter a downwind atmospheric measurement. Unlike existing deterministic ML emulators, which give a single prediction, GenGHG predicts an ensemble of plausible footprints under given meteorological forcing conditions to represent stochastic atmospheric transport. GenGHG retains ML-level computational efficiency, with each member generated in less than 2 s on a single GPU, and we evaluate its generative advantage over deterministic ML across a benchmark of 60 urban areas worldwide. Results show that GenGHG preserves footprint total mass substantially better than deterministic ML, with a mean relative bias of +2.67 % compared with −22.38 %, and also better preserves the footprint-value distribution. Grid-level accuracy also shows that GenGHG better captures spatial footprint patterns, with larger gains in more dispersive transport regimes. We further test GenGHG’s advantage in a synthetic inverse modeling experiment, where transport from GenGHG yields methane emission estimates that closely follow the STILT transport reference and outperform deterministic ML. Finally, we highlight GenGHG’s compatibility with any global meteorology product at a 0.25° resolution. This flexibility, combined with its low computational cost, makes it straightforward to run footprints using multiple meteorology products and subsequently evaluate the possible effects of meteorological uncertainties.

Methane (CH4) flux estimates from high-latitude North American wetlands remain highly uncertain in magnitude, seasonality, and spatial distribution. In this study, we evaluate a decade (2007–2017) of CH4 flux estimates by comparing 16 process-based models with atmospheric CH4 observations collected from in situ towers. We compare the Global Carbon Project (GCP) process-based models with a model inter-comparison from a decade earlier called The Wetland and Wetland CH4 Intercomparison of Models Project (WETCHIMP). Our analysis reveals that the GCP models have a much smaller inter-model uncertainty and have an average magnitude that is a factor of 1.5 smaller across Canada and Alaska. However, current GCP models likely overestimate wetland fluxes by a factor of two or more across Canada and Alaska based on tower-based atmospheric CH4 observations. The differences in flux magnitudes among GCP models are more likely driven by uncertainties in the amount of soil carbon or spatial extent of inundation than in temperature relationships, such as Q10 factors. The GCP models do not agree on the timing and amplitude of the seasonal cycle, and we find that models with a seasonal peak in July and August show the best agreement with atmospheric observations. Models that exhibit the best fit to atmospheric observation also have a similar spatial distribution; these models concentrate fluxes near Canada's Hudson Bay Lowlands. Current, state-of-the-art process-based models are much more consistent with atmospheric observations than models from a decade ago, but our analysis shows that there are still numerous opportunities for improvement.

Dense greenhouse-gas (GHG) observations, particularly from satellites, require fast atmospheric transport models that link surface emissions to these observations. We introduce GenGHG, a generative machine learning (ML) model that emulates footprints from the Stochastic Time-Inverted Lagrangian Transport (STILT) model. These footprints estimate how a unit of emissions would alter a downwind atmospheric measurement. Unlike existing deterministic ML emulators, which give a single prediction, GenGHG predicts an ensemble of plausible footprints under given meteorological forcing conditions to represent stochastic atmospheric transport. GenGHG retains ML-level computational efficiency, with each member generated in less than 2 s on a single GPU, and we evaluate its generative advantage over deterministic ML across a benchmark of 60 urban areas worldwide. Results show that GenGHG preserves footprint total mass substantially better than deterministic ML, with a mean relative bias of +2.67 % compared with −22.38 %, and also better preserves the footprint-value distribution. Grid-level accuracy also shows that GenGHG better captures spatial footprint patterns, with larger gains in more dispersive transport regimes. We further test GenGHG’s advantage in a synthetic inverse modeling experiment, where transport from GenGHG yields methane emission estimates that closely follow the STILT transport reference and outperform deterministic ML. Finally, we highlight GenGHG’s compatibility with any global meteorology product at a 0.25° resolution. This flexibility, combined with its low computational cost, makes it straightforward to run footprints using multiple meteorology products and subsequently evaluate the possible effects of meteorological uncertainties.

Dense greenhouse-gas (GHG) observations, particularly from satellites, require fast atmospheric transport models that link surface emissions to these observations. We introduce GenGHG, a generative machine learning (ML) model that emulates footprints from the Stochastic Time-Inverted Lagrangian Transport (STILT) model. These footprints estimate how a unit of emissions would alter a downwind atmospheric measurement. Unlike existing deterministic ML emulators, which give a single prediction, GenGHG predicts an ensemble of plausible footprints under given meteorological forcing conditions to represent stochastic atmospheric transport. GenGHG retains ML-level computational efficiency, with each member generated in less than 2 s on a single GPU, and we evaluate its generative advantage over deterministic ML across a benchmark of 60 urban areas worldwide. Results show that GenGHG preserves footprint total mass substantially better than deterministic ML, with a mean relative bias of +2.67 % compared with −22.38 %, and also better preserves the footprint-value distribution. Grid-level accuracy also shows that GenGHG better captures spatial footprint patterns, with larger gains in more dispersive transport regimes. We further test GenGHG’s advantage in a synthetic inverse modeling experiment, where transport from GenGHG yields methane emission estimates that closely follow the STILT transport reference and outperform deterministic ML. Finally, we highlight GenGHG’s compatibility with any global meteorology product at a 0.25° resolution. This flexibility, combined with its low computational cost, makes it straightforward to run footprints using multiple meteorology products and subsequently evaluate the possible effects of meteorological uncertainties.

Dense greenhouse-gas (GHG) observations, particularly from satellites, require fast atmospheric transport models that link surface emissions to these observations. We introduce GenGHG, a generative machine learning (ML) model that emulates footprints from the Stochastic Time-Inverted Lagrangian Transport (STILT) model. These footprints estimate how a unit of emissions would alter a downwind atmospheric measurement. Unlike existing deterministic ML emulators, which give a single prediction, GenGHG predicts an ensemble of plausible footprints under given meteorological forcing conditions to represent stochastic atmospheric transport. GenGHG retains ML-level computational efficiency, with each member generated in less than 2 s on a single GPU, and we evaluate its generative advantage over deterministic ML across a benchmark of 60 urban areas worldwide. Results show that GenGHG preserves footprint total mass substantially better than deterministic ML, with a mean relative bias of +2.67 % compared with −22.38 %, and also better preserves the footprint-value distribution. Grid-level accuracy also shows that GenGHG better captures spatial footprint patterns, with larger gains in more dispersive transport regimes. We further test GenGHG’s advantage in a synthetic inverse modeling experiment, where transport from GenGHG yields methane emission estimates that closely follow the STILT transport reference and outperform deterministic ML. Finally, we highlight GenGHG’s compatibility with any global meteorology product at a 0.25° resolution. This flexibility, combined with its low computational cost, makes it straightforward to run footprints using multiple meteorology products and subsequently evaluate the possible effects of meteorological uncertainties.