881 Application of a Subsetting Ensemble Postprocessing Method on the HWRF-Based Ensemble Prediction System

Tuesday, 14 January 2020
Hall B (Boston Convention and Exhibition Center)
Zhan Zhang, EMC, College Park, MD; and W. Wang, L. Zhu, B. Liu, K. Wu, A. Mehra, and V. Tallapragada

The Hurricane Weather and Research Forecasting (HWRF) is one of the operational hurricane model systems at NCEP. It has undergone yearly upgrades and consistent improvements since its operational implementation at 2007. HWRF based Ensemble Prediction System (HWRF-EPS) has been improved along with its deterministic version, and has been running in real time parallel for storms in the North Atlantic basin since 2014 in the support of Hurricane Forecast Improvement Project (HFIP).

In this study, the statistical characteristics of HWRF EPS is first evaluated and analyzed based on past 6-year ensemble track and intensity forecasts through posterior analysis. The analysis revealed several interesting features of HWRF EPS including: 1). the average of the intensity probability density function (PDF) over all storms/cycles produced by HWRF EPS is Gaussian alike, except for the lowest and highest ends of the spectrum, indicating un-predictability of the system on some particular storms and cycles; 2). each ensemble member has about equal probability to be closest to the observed values in the best track; 3) HWRF EPS is under-dispersed in terms of intensity forecasts.

A sub-setting ensemble mean method to post-process the track and intensity forecasts from HWRF-EPS is then described and proposed. The verification shows that the track and intensity skills produced by sub-setting post-process are improved about 10% before 48 hours of forecasts on top of the ensemble mean of all members.

An overall verification is also provided, comparing two sets of HWRF EPS track and intensity forecasts, ensemble mean of all members and ensemble mean of subset members, with that from deterministic version for each individual year and all year combined Atlantic storms. The results demonstrated both track and intensity from all member ensemble mean always have lower forecast errors than the one from un-perturbed deterministic system.

Key words: HWRF, Ensemble, Posterior analysis, Hurricane Track/Intensity

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner