Date of Award

May 2013

Degree Type

Thesis

Degree Name

Master of Science

Department

Mathematics

First Advisor

Allen C. Evans

Second Advisor

Paul J. Roebber

Committee Members

Vincent E. Larson

Keywords

Convection Initiation, Numerical Weather Prediction

Abstract

Improvements in numerical forecasts of deep, moist convection have been notable in recent years and are in large part due to increased computational power allowing for the explicit numerical representation of convection. Accurately forecasting the timing and location of convection initiation (CI), however, remains a substantial forecast challenge. This is attributed to the inherently limited intrinsic predictability of CI due to its dependence on highly non-linear moist physics and fine-scale atmospheric processes that are poorly represented in observations. Because CI is the starting point of deep, moist convection that grows upscale, even small errors in initial convective development can rapidly spread to larger scales, having potentially significant impacts on downstream forecasts. This study investigates the practical predictability of CI using the Advanced Research Weather Research and Forecasting (WRF-ARW) model with a horizontal grid spacing of 429 meters. A unique object-based method is used to evaluate very high-resolution model performance for twenty-five cases of CI across the west-central High Plains of the United States from the 2010 convective season. CI objects are defined as areas of higher observed or model simulated radar reflectivity that develop and remain sustained for a sufficient period of time. Model simulations demonstrate an average probability of detection of 0.835, but due to significant overproduction of CI, an average false alarm ratio of 0.664 and bias ratio of 2.49. The average critical success index through all simulations is 0.315. Model CI objects that are matched with observed CI objects show, on average, an early bias of about 7 minutes and distance errors of around 62 kilometers. The operational utility and inherent biases of such high-resolution simulations are discussed.

Share

COinS