Published: Jan. 23, 2015

Conditional Extreme Value Analysis for Severe Storm Environments

Eric Gilleland

Research Applications LaboratoryNational Center for Atmospheric Research (NCAR)

Date and time: 

Friday, January 23, 2015 - 3:00pm

Location: 

ECCR 265

Abstract: 

A severe storm environment indicator is a climate model-scale variable that informs about atmospheric states that are conducive to severe, and potentially destructive, weather phenomena, many of which occur at very fine scales (e.g., tornadic storms) that cannot be resolved by the relatively coarse scale climate models.  The product of maximum updraft velocity (m/s) and 0 - 6 km vertical wind shear (m/s) is considered, here, because it has been shown to be a relatively good discriminator of various categories of severe storm environments.  Although high values of this product (henceforth, WmSh) are of interest, extreme WmSh values are perhaps even more important because of the potential for very destructive storms involving hail, tornadoes, severe wind speeds, etc.  While techniques for analyzing extremes in one dimension are well established, extreme value analysis (EVA) in a spatial setting is an active area of research.  Here, we employ the conditional modeling framework, introduced by Heffernan and Tawn in 2004, in order to model the processes at each point of a grid from the National Center for Atmospheric Research (NCAR)/National Center for Environmental Prediction (NCEP) reanalysis data set over North America conditioned on the existence of high spatial WmSh energy.  The conditional EVA model excels at capturing the spatial patterns of WmSh in the presence of extreme energy in space, and physically meaningful environments are recovered (e.g. tornado alley in the spring).  The model gives a potentially useful means for analyzing future climate conditions beyond the reporting of means.  A review of univariate statistical EVA will be given along with background on the conditional extreme value model, as well as an initial look into an alternative and promising method for parameter estimation for all of the parameters in the model.