Bayesian Monte Carlo analysis applied to regional-scale inverse emission modeling for reactive trace gases

摘要

The purpose of this article is inverse modeling of emissions at regional scale for photochemical applications. The study is performed for the Ile-de-France region over a two summers (1998 and 1999) period. This area represents an ideal framework since concentrated anthropogenic emissions in the Paris region frequently lead to the formation of urban plumes. The inversion method is based on Bayesian Monte Carlo analysis applied to a regional-scale chemistry transport model, CHIMERE. This method consists in performing a large number of successive simulations with the same model but with a distinct set of model input parameters at each time. Then a posteriori weights are attributed to individual Monte Carlo simulations by comparing them with observations from the AIRPARIF network: urban NO and O_3 concentrations and rural O_3 concentrations around the Paris area. For both NO and O_3 measurements, observations used for constraining Monte Carlo simulations are additionally averaged over the time period considered for analysis. The observational constraints strongly reduce the a priori uncertainties in anthropogenic NOx and volatile organic compounds (VOC) emissions: (1) The a posteriori probability density function (pdf) for NOx emissions is not modified in its average, but the standard deviation is decreased to around 20% (40% for the a priori one). (2) VOC emissions are enhanced (+16%) in the a posteriori pdf's with a standard deviation around 30% (40% for the a priori one). Uncertainties in the simulated urban NO, urban O_3, and O_3 production within the plume are reduced by a factor of 3.2, 2.4, and 1.7, respectively.