Evaluation and Applications of the Prediction of Intensity Model Error (PRIME) Forecasting System

Wednesday, 20 April 2016: 2:15 PM
Ponce de Leon C (The Condado Hilton Plaza)
Kieran Bhatia, Univ. of Miami/RSMAS, Miami, FL; and D. Nolan, A. B. Schumacher, and M. DeMaria

Forecasters and end users of tropical cyclone (TC) intensity forecasts would greatly benefit from a reliable expectation of model error to counteract the lack of consistency in TC intensity forecast performance. As a first step towards producing error predictions to accompany each TC intensity forecast, Bhatia and Nolan (2013) investigated the relationship between synoptic parameters, TC attributes, and forecast errors. In this study, we build on previous results of Bhatia and Nolan (2013) by testing the ability of the Prediction of Intensity Model Error (PRIME) model to forecast the absolute error and bias of four leading intensity models available for guidance in the Atlantic and East Pacific basin. PRIME forecasts are independently evaluated at each 12-hour interval from 12 to 120 hours during the 2007-2015 hurricane seasons. The absolute error and bias predictions of PRIME are compared to their respective climatologies to determine their skill. In addition to these results, we will present the performance of the operational version of PRIME run during the 2015 hurricane season.

PRIME verification results show that it can reliably anticipate situations where particular models excel, and therefore could lead to a more informed protocol for hurricane evacuations and storm preparations. These positive conclusions suggest that PRIME forecasts also have the potential to lower the error in the original intensity forecasts of each model. As a result, two techniques are proposed to develop a post-processing procedure for a multimodel ensemble based on PRIME. The first approach is to inverse-weight models using PRIME absolute error predictions (higher predicted absolute error corresponds to lower weights). The second multimodel ensemble applies PRIME bias predictions to each model's intensity forecast and the mean of the corrected models is evaluated. The forecasts of both of these experimental ensembles are compared to those of the equal-weight ICON ensemble, which are currently the most reliable intensity forecasts in the Atlantic basin.

References Bhatia, K. T., and D. S. Nolan, 2013: Relating the Skill of Tropical Cyclone Intensity Forecasts to the Synoptic Environment. Wea. Forecasting, 28, 961–980.

- Indicates paper has been withdrawn from meeting
- Submission entered in competition