Quantifying subgrid pollutant variability in Eulerian air quality models

In order to properly assess human risk due to exposure to hazardous air pollutants or air toxics, detailed information is needed on the location and magnitude of ambient air toxic concentrations. Regional scale Eulerian air quality models are typically limited to relatively coarse grid resolutions when simulating mean pollutant concentrations for each grid cell volume, and subgrid pollutant extremes are not represented. Continual improvements in computing power and refinement of nested grid techniques have allowed the regional air quality models to simulate down to grid spacings on the order of one kilometer. Concurrent developments in neighborhood scale modeling using computational fluid dynamics (CFD) and coupled large-eddy simulation (LES) with photochemistry techniques have permitted simulations with grid spacings of meters to tens of meters for domains limited to several kilometers. Our goal is to develop a methodology to utilize available fine resolution gridded model results to produce statistical analyses of pollutant subgrid variability applicable to the coarser grid resolutions, thereby somewhat bridging the gap between regional scale and neighborhood scale air quality models.

Exploratory data analysis (EDA) statistical techniques are being employed in the development of a post-processing software tool which systematically analyzes fine resolution gridded model results in order to objectively determine best-fit univariate distributions representing subgrid pollutant concentration variability. Specific probability density functions (pdfs) for the selected distributions are then produced, which can subsequently be used as input data for hazardous air pollutant human exposure models.

Initial application of the pdf analysis software has been on fine resolution results from a Community Multiscale Air Quality (CMAQ) modeling system urban case study. Statistical analyses of selected trace gas concentrations from both an urban and a rural area will be compared. Additional analyses applied to the entire model domain will illustrate how the derived distributions vary spatially. As expected, the fitted pdfs are shown to be functions of pollutant, time, location, and overall meteorological-photochemical scenario characteristics. The potential for creating improved parameterizations of subgrid pollutant variability will be explored based on the fields of pdfs and their associated location, variation, and/or shape parameters. This new pollutant subgrid variability analysis package can be applied to any data set, whether originating from air quality model output or monitoring network observations.