Statistical verification of Short Term Explicit Prediction (STEP) Program precipitation and streamflow forecasts for the 2013 Colorado Front Range flooding event

The Short Term Explicit Prediction (STEP) program is a multi-laboratory program at NCAR aimed at improving short-term forecasting of high impact weather events, with a goal of achieving such societal benefits as reduced fatalities and injuries, reduced property damage, and improved efficiency for industry, transportation, and agriculture. One facet of this effort is to improve prediction of heavy rainfall and flash flooding events across the Colorado Front Range. To this end, team members are developing a comprehensive quantitative precipitation estimation, forecasting, nowcasting, and streamflow prediction system, and applying a variety of statistical verification methods to evaluate contributions of the various components of the system.

This presentation demonstrates the benefits of applying a wide range of verification approaches to evaluate different aspects these forecasts that are relevant for both urban and rural flooding prediction. The analysis includes WRF 3DVAR forecasts, AutoNowCaster (ANC) nowcasts, and streamflow predictions of the 11-13 Sept 2013 flash flooding case in the Colorado Front Range, which resulted in nine fatalities, evacuations of canyons and cities with residents instructed to travel on foot, thousands of homes damaged or destroyed, widespread washouts of roads and bridges, and the near-destruction of several mountain communities. The Model Evaluation Tools (MET) software is used to produce traditional contingency table statistics on the forecasts compared to Quantitative Precipitation Estimate (QPE) products at various precipitation and scale thresholds. A neighborhood approach is used to evaluate forecast skill at various spatial scales through application of the Fraction Skill Score (FSS) method. The Method for Object-based Diagnostic Evaluation (MODE) is also used to provide object-based verification of precipitation, addressing concepts of object area, orientation, displacement, and intensity of precipitation objects that may be more relevant for assessing user-relevant aspects of forecast performance.