Le Duc National Center for Hydro-Meteorological Forecast

A short range ensemble prediction system

applied in TC forecast

Le DucNational Center for Hydro-Meteorological Forecast

Jeju, 05 - 2009

Motivation

•The success of SREPS in other centers (SREPS in NCEP, COSMO-LEPS in ECMWF, SREPS in INM, …):

•The useful information of EPS in storm movement forecast (ECMWF EPS, NCEP EPS, JMA EPS)

•SREF can detect the occurrence of extreme phenomena like heavy rainfall, heat wave, …

Method

•Breeding of growing mode (NCEP)

•Singular vectors (ECMWF)

•Observation perturbations (CMC)

•Ensemble transform Kalman filter

•Ensemble transform

We take the multi-model, multi-analysis approach.

SREPS description

•4 times per day, 72h forecast (00Z, 12Z), 48h forecast (06Z, 18Z)

•output format: netcdf (interpolated to a common area)

•parallel post processing (graphics)

•access through intranet

Computational resources

PC Cluster 16 nodes, 4 cores per node, 8G RAM per node:•BOLAM: 3 nodes•Eta: 4 nodes•HRM: 3 nodes•WRFNMM: 6 nodes

Dell 2 CPUs, 4 cores per CPU, 16G RAM: pre and post processing

Website

Stamp map: storm tracks

A member forecast

Strike probability maps

Point accumulated strike probability charts

Example: TC Higos

Stamp map: 06h precipitation

Precipitation probability maps

EPSgram

Largest value

Upper quartile

Lower quartile

Median

Smallest value

Interpretationof boxplots

Imageof PDF

Future work: new website

Future work: NAEFS

Future works

•Verification

•Post-processing: bias correction, BMA or NGM

•A specific SREPS for TC forecast: 5 models BoLAM, BRAMS, HRM, MM5, WRF-ARW, storm target domain

•Clustering

Thank you