Sensitivity Analysis Of Probabilistic Multi-Model Ensemble Forecasts Of Wintertime Fronts Over Northwestern Nevada
AuthorYoung, Amanda Katherine
AdvisorKoracin, Darko R
StatisticsView Usage Statistics
Probabilistic ensemble forecasting has become an essential tool to numerical weather prediction. With the chaotic nature of the atmosphere, decisions made by operational meteorologists are made with imperfect weather models. These deterministic numerical weather forecasts can be complemented with the use of regional ensemble predictions incorporating enhanced probabilistic, statistical analysis tools. The challenge is providing better statistical information using ensemble probabilistic information forecasts of mesoscale frontal features to better characterize frontal precipitation fields, intensity, and direction of movement. The purpose of this study was aimed at drawing attention to certain probabilistic distribution patterns for specific mesoscale circulations when physical parameterizations and/or initial conditions are varied for specific ensemble forecast members. A statistical sensitivity error-trend analysis of multi-model (MM5, COAMPS, and WRF) ensemble prediction system (EPS) was conducted to provide insight into how inherent changes to model parameterizations, i.e. PBL, convection, radiation, and microphysics can manifest intrinsic variability to ensemble predictability. Most studies in ensemble prediction used a single model in an ensemble mode, using variations in model initial conditions as the basis to produce simulation ensemble members and in most cases the total ensemble members were limited to 6-10. A total of 153 ensemble members with a horizontal resolution of 36 km were evaluated for this study using three state of the art regional-mesoscale models. Its focus was directed towards the use of a multi-model EPS to measure the statistical sensitivity of a sequence of three winter-time fronts observed over western Nevada during the period of 12-27 December 2008. The corresponding analysis and evaluation underscored a process through which 500 hPa thermal field dataset temperature differences, as it applied to rank data calculated for the three cold frontal systems observed over the period of the 15 day simulation, can also be applied to ensemble model spread and error trend analysis. This study enabled the extension of the forecast simulation period to two weeks, which is the assumed predictability limit for atmospheric simulations. Therefore, it became apparent that the use of statistical rank data error trends and ensemble model spread can improve predictability of certain aspects of frontal activity based on COAMPS smaller (high a priori forecast accuracy) ensemble simulation spread as compared to MM5 and WRF larger (low a priori forecast accuracy) ensemble spread.