Sensitivity Analysis of Mesoscale-Coupled Large-Eddy Simulations for Wind Energy Applications

The performance of wind plants is affected by a wide range of scales of atmospheric motion, spanning from mesoscale variability to fine scales of turbulence. A comprehensive approach to predictive simulation requires the linkage of models for the mesoscale (i.e. numerical weather prediction models) with models for the microscale (i.e. large eddy simulation [LES] models). To represent their unresolved scales of motion, each of these model types employs different kinds of parameterizations, which make different assumptions and rely on different sets of generally uncertain parameters. Uncertainties in both types of models must be taken into account when attempting to quantify the overall uncertainty in mesoscale-microscale coupled simulations. However, the large number of parameters plus the relatively high computational cost of LES precludes a naive strategy of sampling the range of all uncertain parameterization inputs. Rather, it is necessary to first identify the most critical parameters in each of the mesoscale and LES closures before undertaking a combined analysis. Past work has examined the sensitivity of hub-height winds in the Columbia Basin region of Oregon and Washington to parameters of planetary boundary layer and surface layer schemes in conventional mesoscale runs of the Weather Research and Forecasting (WRF) model. These studies found that much of the variability in predicted wind speed can be traced to just a few of the uncertain parameters. Here we revisit the complex terrain of the Columbia Basin and evaluate the sensitivity of boundary layer winds and turbulence predicted in mesoscale-coupled LES to parameters of a 1.5-order, turbulent kinetic energy based subgrid-scale (SGS) turbulence closure and of a surface flux scheme. We sample a range of parameter values to generate an ensemble of coupled mesoscale-microscale model runs using a nested WRF/WRF-LES computational approach. This set of WRF/WRF-LES model runs is then used to create a statistical meta-model that can be efficiently sampled. Note that coefficient values employed in SGS closures are either theoretically derived using strongly simplifying assumptions such as isotropy of small-scale turbulence or arrived at heuristically. In either case, considerable uncertainty arises in specifying appropriate values for use in highly non-idealized cases. Therefore, observations from the Wind Forecast Improvement Project in Complex Terrain (WFIP2) are used extensively to ensure that relevant ranges of parameter values are sampled by our WRF/WRF-LES simulation ensemble.