Assessing uncertainty in simulated atmospheric transport and dispersion using a WRF-3DVAR cycling multiphysics ensemble

This study examines the degree of uncertainty that occurs in atmospheric transport and dispersion (ATD) modeling due to using different parameterizations of physical processes such as shortwave and longwave radiative processes, precipitation formation, and atmosphere-surface interactions. An ensemble approach is taken to examine how this one aspect of meteorological model uncertainty affects subsequent ATD simulations. Differences in these Weather Research and Forecasting (WRF) physics configuration strategies are investigated through their effects on offline HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) ATD simulations of three October 2010 tracer releases conducted in a region of complex terrain. The WRF model consists of three nested domains, with the innermost (highest resolution) domain having a horizontal grid spacing of 2 km. A twenty member ensemble of input meteorological data for the ATD simulations is generated through 6 h WRF cycling runs that employ three-dimensional variational data assimilation (3DVAR) on the outer two domains, but not on the innermost domain. Each of the twenty members uses a different combination of WRF physics parameterizations. Characteristics of several aspects of the 3DVAR system are presented. Single observation tests reveal increments that reflect physical balances imposed during the data assimilation process. Consistent RMS error reduction by 3DVAR improves the quality of information provided at the lateral boundaries of the innermost domain (Delta x = 2 km). RMS errors of 10-meter u and v winds are generally reduced by 0.4-0.5 m s-1 , and RMS errors of 2-meter temperature are reduced by 1.5-3 K throughout the cycling runs. Considerable spread in PBL height, a key ATD variable, is produced in the innermost domain, especially during the daytime hours (sigma = 141-200 m). HYSPLIT dispersion simulations using the ensemble members show a wide spread of ground-level concentration fields, as well as varying degrees of vertical transport, that are a consequence of the choice of model physics configurations, with plume areas sometimes varying by over 1000 km 2 . Verification metrics quantitatively illustrate the concentrations at the ground truth field study measurement stations that can be obtained by varying the model physics parameterizations. Ensemble-based probabilities provide a useful method of describing the likelihood that a given concentration will be exceeded.