evaluation and development of ensemble prediction system for the operational hwrf model zhan zhang,...

Post on 20-Jan-2016

217 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Evaluation and Development of Ensemble Prediction Evaluation and Development of Ensemble Prediction System for the Operational HWRF ModelSystem for the Operational HWRF Model

Zhan Zhang, V. Tallapragada, R. Tuleya, Q. Liu, Y.

Kwon, S. Trahan, J. O’Connor, and, W. M. Lapenta

65th Interdepartmental Hurricane Conference Miami, FL. Feb. 28 – Mar. 3

OutlineOutline

Uncertainties in Hurricane Forecasts;Uncertainties in Hurricane Forecasts; Experiment Design;Experiment Design; Results:Results: - Track Forecasts- Track Forecasts - Intensity Forecasts- Intensity Forecasts Probabilistic Products;Probabilistic Products; Summary & Future Works.Summary & Future Works.

Possible Uncertainties in Hurricane Possible Uncertainties in Hurricane Model ForecastsModel Forecasts

Initial Large Scale Flows;Initial Large Scale Flows; Lateral Boundary Conditions;Lateral Boundary Conditions; Initial Storm Structure;Initial Storm Structure; Model Physics.Model Physics.

Ensemble Ensemble Member IDMember ID

Input DataInput Data RMW RMW PerturbationPerturbation

Convection Convection SchemeScheme

PBL SchemePBL Scheme

M00 – M20M00 – M20 GEFS (T190L28)GEFS (T190L28) NoNo SASSAS GSF PBLGSF PBL

M21 (CTRL)M21 (CTRL) GFS (T574L64)GFS (T574L64) NoNo SASSAS GFS PBLGFS PBL

M22 M22 GFS (T574L64)GFS (T574L64) NoNo Kain-Fritsch Kain-Fritsch GFS PBLGFS PBL

M23-M24M23-M24 GFS (T574L64)GFS (T574L64) RWM Plus/minus RWM Plus/minus

25%25%

SASSAS GFS PBLGFS PBL

M25-M26M25-M26 GFS (T574L64)GFS (T574L64) RWM Plus/minus RWM Plus/minus

25%25%

Kain-FritschKain-Fritsch GFS PBLGFS PBL

M27-M47M27-M47 GEFS (T190L28)GEFS (T190L28) NoNo Kain-FritschKain-Fritsch GFS PBLGFS PBL

M48M48 GFS (T574L64)GFS (T574L64) NoNo SASSAS MYJ PBLMYJ PBL

M49M49 GFS (T574L64)GFS (T574L64) NoNo Kain-FritschKain-Fritsch MYJ PBLMYJ PBL

Experiment Design

List of Experiment:

• ControlControl: HWRF V3.2 Baseline, : HWRF V3.2 Baseline, M21M21• Large Scale Flow & LBC Perturbations: - M00-M20, 21 members - M27-M47, 21 members• Initial Storm Structure Perturbations: - M23-M26, 4 members• Physics-Based Perturbations: - M21, M22, M48, M49, 4 members

Total: 50 ensemble membersTotal: 50 ensemble membersHurricane Earl: 2010082512 -2010090412Hurricane Earl: 2010082512 -2010090412

Track ForecastsTrack Forecasts

More than ~15% improvement in track forecasts

Track forecasts are improved by all sub-sets of ensembles;

Ensembles have less impacts on the track forecasts before 48h;

GEFS-SAS GEFS-KF

Perturbed initial structure

Physics-based

Northeast bias

West bias at late stage

Relatively narrow track spread

Intensity ForecastsIntensity Forecasts

No clear intensity improvement from all sub-sets of ensembles.

GEFS-SAS slightly better

Positive bias for weaker storm

Negative bias for stronger storm

For Earl, there are overall strong negative sample bias.

Init intensity=75kts

Init intensity=35kts

Init intensity=50kts

Ranked Ensemble members

Rel

ativ

e F

requ

ency

(%

)Ranked Histogram for 10m Max Wind Speed

Hurricane Earl, 2010

No sample bias at initial time

Strong negative sample bias

00h 24h 48h

72h 120h All time

Equal weights

Intensity forecasts are improved with weighted ensemble mean at all time levels

~ 17% Improvement

Hurricane Probabilistic ProductsHurricane Probabilistic Products

Ensemble Member-based Storm Strike Probability Forecast for 120h

obs.

ens. mean

Blue: obs

Yellow: ens. mean

Ensemble Spread (along/cross) Based Storm Strike Probability Forecast

Ensemble Intensity Forecasts

Probability Forecasts of 10m Wind Speed greater than 30m/s Earl, 2010082700

Double probability max centers

Max centers for wind speed and probability are not co-located

Contour: Predicted 10m wind speed isotach from CTRL exp.

Shading:

Probability Forecast of 10m wind speed greater than 30m/s

Danielle

Earl

120h Forecast of Strike Probabilities for Wind Speed greater than 20m/s, Earl, 2010082700

Summary & Future WorksSummary & Future Works

Storm track forecasts are improved in all sub-sets of Storm track forecasts are improved in all sub-sets of ensembles;ensembles;Model-based ensemble sample bias can be corrected by Model-based ensemble sample bias can be corrected by applying weights to ranked ensemble members;applying weights to ranked ensemble members;Storm intensity forecasts are improved by weighted Storm intensity forecasts are improved by weighted ensemble average;ensemble average;HWRF is not very sensitive to storm initial radius of HWRF is not very sensitive to storm initial radius of maximum wind;maximum wind;Physics-based ensemble reduces model-based storm Physics-based ensemble reduces model-based storm intensity bias;intensity bias;

Future works:Future works:Sensitivity test for storm initial positions;Sensitivity test for storm initial positions;Optimum combination of ensembles;Optimum combination of ensembles;

Storm Initial Position Uncertainties

(All 2010 Storms)

ATL: lat

ATL:lon

EP: lat

EP: lon

Storm Initial Position PFD

All 2010 Samples

Gaussian–like distribution around zero

top related