Comparison of fixed prior and geostatistical inverse methods for methane emission estimation from Central California

Inventory and model-based estimates of CH4 emissions are associated with large uncertainty due to poor quantification of factors controlling emissions. Inverse methods in conjunction with measurements of atmospheric composition have been used to improve quantification of CH4 emissions. However, atmospheric inverse methods are also affected by uncertainties such as the set-up of error covariance structure. To address a subset of the uncertainties limiting atmospheric inverse modeling of regional CH4 emissions from Central California, we investigate two inverse methods: (1) a Bayesian inverse method, and (2) a geostatistical inverse method. The first method uses a priori emissions with fixed spatial structure, while the geostatistical inverse method allows for a more flexible spatial distribution and may or may not use the fixed priors. To implement the methods, the atmospheric transport modeling is conducted by coupling the Weather Research and Forecast (WRF) model to the Stochastic Time-Inverted Lagrangian Transport (STILT) model. The inverse methods take a receptor-oriented approach to analyze one year of well-calibrated CH4 measurements made at a tall tower in Central California. For each inverse method, we estimate the error covariance structure differently and evaluate the impact of error covariance structure on CH4 emission estimates. We then determine whether a superior method exists for characterizing temporal and spatial variations in emissions. A sensitivity analysis of resolutions shows that the distribution of surface emissions is affected by both spatial and temporal resolutions.