Model Selection and Inference for Atmospheric Vortices
Lynn Greenleaf University of Oklahoma
(lgreenleaf@math.ou.edu)
ABSTRACT
The focus of this talk concerns valid statistical inferences from tangential wind measurements on intense atmospheric vortices arising in dust devils, waterspouts, tornadoes, mesocyclones and tropical cyclones when the analysis depends on a parametric model of the information in the data. The mathematical data analysis methodology selected is sufficiently general to allow measurements from Doppler radar velocity signatures, numerical model output, or Monte-Carlo simulation data. In order to achieve this, a set of a priori candidate tangential velocity models of atmospheric vortices that each possess a small number of parameters are selected based on scientific principles and are ranked as to their ability to capture information in the data. The candidate models include versions of the Wood-White tangential wind profile as well as the idealized Vatistas, Rankine, Burgers-Rott and Sullivan vortex models. A model is selected from this list of candidate models to seek the model that loses as little information from the data as possible using Akaike's Information Criterion (AIC). AIC is a measure of the Kullback-Leibler information loss when a model is used to approximate the tangential wind profile of an atmospheric vortex. Other criteria are also discussed in this context. The observations from Doppler radar contain information on the actual or true tangential winds as well as noise from other sources and instrumentation. The goal of this mathematical analysis is to assess each model's ability to capture information about the true tangential winds contained in observations. The model selection method will address two approaches. The first approach will involve ranking the models based on capturing the information content of the tangential velocity profile in the data. The second approach will involve optimization of the predictions of the radial and vertical components of velocity in a vortex given a model tangential velocity profile. This will allow predictions of the vertical and radial velocity profiles along with statistical measures of the uncertainty in these predictions.
Preference: Oral presentation
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner