Date of Award

August 2018

Degree Type


Degree Name

Master of Science


Atmospheric Science

First Advisor

Paul J Roebber

Committee Members

Vincent E Larson, Clark Evans


cold season precipitation, HRRR-TLE, machine learning, post-processing


In this study we explore advanced statistical methods with the operational High-Resolution Rapid Refresh Model (HRRR) Time-Lagged Ensemble (TLE) to improve the prediction of cold season precipitation type. TLEs are a computationally efficient method to provide a slightly improved probabilistic forecast as the differences between model runs are an approximation of initial condition uncertainty. We apply evolutionary programming, weight-decay bias correction, and Bayesian Model Combination with fifteen HRRR forecast variables that potentially relate to precipitation type for station locations in the contiguous United States that are along and to the east of 100 W longitude to obtain probabilistic precipitation type forecasts. These methods are shown to provide improved probabilistic information for both the areal distribution of cold season precipitation and the timing and location of phase transitions.