92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Thursday, 26 January 2012
Ensemble Forecasting of Hurricane Intensity based on Biased and non-Gaussian Ensemble Sampling
Hall E (New Orleans Convention Center )
Zhan Zhang, NOAA/NWS/NCEP/EMC, Camp Springs, MD; and V. Tallapragada and R. Tuleya
Manuscript (97.5 kB)

This study investigates all possible uncertainties that are caused by either initialization process or model physics using the NCEP operational HWRF system. Two ensemble generation methods have been used in this study: single model with multi-initial condition and multi-model, multi-physics. The ensemble one is designed to study the impact of uncertainties from large scale flow, which uses the GEFS data (resolution T190L28) as input data to initialize HWRF domain. It contains three sub-sets that use three different cumulus convection schemes. The second set of ensemble is intended to understand the impact of uncertainties in multi-model and multi-physics, which include the operational HWRF with various physics and GFDL model.

The intensity error probability density function (PDF) is then estimated from the ensemble members. The PDF shows that the predicted hurricane intensity errors are biased and non-Gaussian distributed, which indicates that arithmetic mean among the ensemble members will not improve the intensity forecasts.

Two post-process methods are introduced in this study: 1. to correct the model bias through ensemble PDF; 2. to find most probable mode through PDF kernel density estimation method. The first method is applied to the single model with multi-initial condition ensembles while the second is applied to the multi-model, multi-physics ensembles.

The results showed that hurricane intensity forecast skills are greatly improved by implementing ensemble post-processes.

Supplementary URL: