Uncertainty Quantification
Because atmospheric models attempt to predict future behavior, we can never be certain of their 100 percent accuracy. To address this we use uncertainty quantification—statistics to identify parameters in a model that lead to uncertainties in simulations. Our team works to quantify uncertainties in commonly used atmospheric modeling tools. This is necessary to identify potential sensitivities and inconsistencies that may occur in a model.
Flow chart of applied sensitivity analysis framework. Enlarge Image
Applying Uncertainty Quantification
Our team has focused on the identification of parameters within the commonly used Mellor-Yamada-Nakanishi-Niino boundary-layer and revised Yonsei University surface-layer parameterizations. In this case, a total of 26 different parameters were selected and allowed to vary over a finite range. This range was determined using values appearing in the scientific literature. In total, over 500 simulations were conducted and analyzed as part of this study.
A sample time series, with a different line for each simulation, shows the simulated range in hub-height wind and the associated range in power production from a hypothetical wind turbine. Based on this analysis, our team identified the largest sensitivities are related to the surface roughness, turbulence kinetic energy (TKE) dissipation rate, TKE length scales, and the Prandlt number.