Reconciling the Global Ethane Budget

Atmospheric Science
GEOS-Chem
Bayesian Inversion
PSU CCAR
First-authored, NSF-funded atmospheric research using global GEOS-Chem simulations and an analytical Bayesian inversion to explain why standard inventories underestimate atmospheric ethane — optimizing emission fluxes across 35 source regions to account for the post-2010 rebound.
Published

December 1, 2022

The puzzle

Atmospheric ethane is the most abundant non-methane hydrocarbon, co-emitted with methane and a driver of tropospheric ozone. There was a stubborn problem with it: standard bottom-up emission inventories failed to reproduce what instruments actually measured in the atmosphere, especially a puzzling rebound in Northern-Hemisphere ethane after 2010. My research, conducted at Portland State’s Center for Climate and Aerosol Research and funded by NSF Atmospheric & Geospace Sciences Grant 1950702, asked why — and whether the gap could be closed.

Establishing the gap

Using the GEOS-Chem high-performance model (GCHP v13.0.2, with MERRA-2 meteorology, run on the COEUS HPC cluster) and observations from roughly 16 NOAA Global Monitoring Laboratory flask sites, I first quantified the discrepancy. The baseline model underestimated observed ethane by as much as 500–1000 pptv across the Northern Hemisphere, against a strong hemispheric gradient and clear seasonal cycle. The conclusion was unambiguous: the standard inventories were missing or under-counting real sources.

This baseline comparison became my first-author REU research manuscript (2021).

Closing it: the Bayesian inversion

Diagnosing the gap is one thing; correcting it rigorously is another. I built an analytical Bayesian emission inversion to optimize regional ethane fluxes directly from the observations. The technical core:

  • A state vector of regional emission strengths and an observation vector of NOAA concentrations, linked by a Jacobian (H-matrix) constructed from tagged-tracer model runs that map each source region to its fingerprint at each measurement site.
  • An analytical posterior solution with prior- and observation-error covariances, where the spatial and temporal prior structure was built as a Kronecker product of exponential kernels.
  • Because the system is underdetermined (more unknowns than independent constraints), I solved it via a singular-value-decomposition pseudo-inverse rather than a naive matrix inversion that would have been numerically unstable.

Comparing 2007–2008 against 2019–2020 and optimizing fluxes across 35 source regions, the inversion offered a quantitative account of the decade-scale ethane trend. I presented this work at the AGU Fall Meeting in 2021 and 2022.

Why it matters

Ethane and methane are co-emitted by fossil-fuel activity, so getting the ethane budget right sharpens our understanding of where hydrocarbon emissions truly originate — the foundation of any credible mitigation policy. This was also where I learned that good environmental science lives or dies on rigor: a beautiful result from an unstable inversion is worthless, and the discipline of doing the linear algebra correctly is the science.

Documents