Impact of Grid Resolution on Chemical Transport Modeling in the Southeast US: GEOS-Chem Model Constrained with Observations from SEAC4RS Aircraft Campaign
Wednesday, 17 December 2014
Atmospheric processes span a wide range of spatial and temporal scales. Combined with the inherent non-linearity of many chemical processes, this presents difficulties for accurate numerical modeling of atmospheric composition. Earlier works have found that with coarse resolution models, ozone production tends to be overestimated in urban environments due to NOx being diluted within large grid cells that artificially mix both urban and non-urban air. The impact of grid resolution on modeling the chemical regime of the Southeastern US, which is largely affected by biogenic VOCs, has not been previously studied. In this study, the sensitivity of various species to resolution is examined by running the GEOS-Chem chemical transport model at 0.25x0.3125 degree (~25 km), 2x2.5 degree, and 4x5 degree resolution and quantifying model errors as constrained by observations from the SEAC4RS aircraft campaign. Species respond differently to grid resolution depending on its lifetime and chemical non-linearity. We find that higher resolution models are better able to capture variability present in the observations, leading to better representation of extreme values, even if correlations between model and observations do not always improve with increasing resolution. Coarse resolution models also induce bias – the 4x5 and 2x2.5 degree models were unable to accurately simulate the low-NOx, high-isoprene environment found in parts of the Southeastern US, tending to underestimate isoprene and overestimate NOx, which leads to overestimates of ozone in these regions. At higher resolution, a smaller fraction of the isoprene hydroxy-peroxy radicals react with NO, possibly due to better segregation of high-NOx and low-NOx environments. Given the computational cost of running high-resolution simulations, it is important to understand what kind of biases and differences occur when running a model at different resolutions.