Sensitivity of Deterministic Prediction to Model Configuration and Initial Uncertainty

University Of Utah, Salt Lake City UT

Investigators

Abstract

This research addresses predictability of short to medium range weather evolution. The first research component will examine reasons for the relatively small benefits of high resolution limited area models compared to more crudely resolved global models revealed in the Principal Investigator's (PI) recent studies. These studies suggest that high resolution limited area models are superior to more coarse resolution global models at 12 and 24 hours, but lose their superiority by 36 hours. The PI hypothesizes that this short period of higher accuracy may be explained by incompatibilities between the global model and limited area model that are coupled at one-way imposed lateral boundaries of the latter, and by dynamical inconsistencies between these models particularly in regard to orographic flow-blocking. It is also hypothesized that the proximity of the validation domain to the data sparse Pacific may limit accuracy of all models. The relevance of these hypotheses will be examined by repeating prior limited area experiments with a global, variable resolution model that focuses its high-resolution window upon the entire conterminous United States. This approach allows two-way global interaction between the highly resolved domain and the rest of the world and will be validated over both the western and eastern United States. The second portion the research will examine fundamental predictability issues regarding saturation of uncertainty growth, the spectral distribution of uncertainty evolution, and possible inferences for the spacing of atmospheric observations. This work follows recent studies in which initial state uncertainty is determined from the difference of two separate, equally credible atmospheric analyses obtained from the United States and European operational numerical forecast models. Those results indicate that initial state uncertainty from the larger scales contribute more to error growth than do uncertainties at the smaller scales. These results, however, were based upon studies that used relatively coarse simulations of only five days duration that did not attain uncertainty saturation. The PI will extend these investigations to higher resolution models with integrations up to two weeks in order to obtain more complete estimates of saturation of uncertainty, the spectral distribution of uncertainty evolution, and observation/analysis requirements. Successful completion of this research could lead to improved weather forecasts and provide guidance for optimal siting of observational instruments.