32
Dynamical Seasonal Hurricane Hindcast Simulations Tim LaRow Y.-K. Lim, D.W. Shin, E. Chassignet and S. Cocke CDPW Meeting – October 23, 2007- Tallahassee email: [email protected]

Dynamical Seasonal Hurricane Hindcast Simulations

Embed Size (px)

DESCRIPTION

Dynamical Seasonal Hurricane Hindcast Simulations. Tim LaRow Y.-K. Lim, D.W. Shin, E. Chassignet and S. Cocke CDPW Meeting – October 23, 2007- Tallahassee email: [email protected]. Outline. Motivation Previous Studies Detection/Tracking Algorithm Experimental Design Results - PowerPoint PPT Presentation

Citation preview

Dynamical Seasonal Hurricane

Hindcast Simulations

Tim LaRowY.-K. Lim, D.W. Shin, E. Chassignet and S. Cocke

CDPW Meeting – October 23, 2007- Tallahassee

email: [email protected]

Outline

•Motivation

•Previous Studies

•Detection/Tracking Algorithm

•Experimental Design

•Results

•Atlantic Domain

•Summary/Conclusions/Future

Motivation – Part 1

Can We Simulate Interannual Variability?

1997 Observed Tracks 2005 Observed Tracks

Motivation – Part 2: Can We Simulate Intensity?

http://www.nhc.noaa.gov/pastall.shtml

Previous Studies

•Current climate models can simulate many of the features of

observed tropical cyclones that have spatial scales resolvable

by such models. These include: •Warm core structure (upper-tropospheric anticyclonic

circulation above cyclonic low-level circulation) •Existence of strong upward motion andHeavy precipitation accompanying the storm

•Geographical distributions, intraseasonal, and interannual

variability of simulated storms are similar to observed

(Manabe et al. 1970, Manabe 1990, 1992; Wu and Lau 1992

Haarsma et al. 1993; Vitart et al. 2006, 2007;

Bengtsson et al. 1982, 1995,2007;

Camargo et al. 2005; Knutson et al. 2007)

Summary from Previous Studies

• Modeled tropical cyclones tend to be:• too weak• tracks too short and some have a poleward bias• storms too large and • lack of genesis in certain regions.

Problems in part due to low resolution models used. O(200-400km) –

although not the complete story.

Detection Algorithm

•Local relative vorticity maximum greater than 4.5x10-5 s-1 is located at

850hPa.

•Next, the closet local minimum in sea level pressure is detected and defines

the center of the storm. Must exist within a 2°x2° radius of the vorticity

maximum.

•Third, the closest local maximum in temperature averaged between 200hPa

and 500hPa is defined as the center of the warm core. The distance from the

warm core center and the center of the storm must not exceed 2°. The

temperature must decrease by at least 6K in all directions from the warm

core center within a distance of 4°.

Max/Min are located and gradients calculated using bicubic splines which

allow for higher precision than the model resolution.

Tracking Algorithm

After the data base of storm snapshots are collected a check is performed to

see if there are storms within 200km during the next 6 hours.

If no, the trajectory is stopped. If yes, the closest storm to the previous 6

hours storm's trajectory is picked. If more than one storm is identified,

preference is given to storms which are west and poleward of the given

location.

•Trajectories must last more than 2 days, have lowest model level wind

velocity within a 8° radius circle centered on the storm center greater than 17

m s-1 during at least 2 days (does not have to be consecutive days).

Experiments

Experimental Design

•Atlantic hurricane season (June-November) hindcast simulations from 1986 to 2005 (20 years).

•Weekly updated observed SSTs (Reynolds et al. 2002).

•FSU/COAPS global spectral model – T126L27 resolution ~ 100km

•4 member ensembles for each year. Time lagged ECMWF atmospheric initial conditions centered on 1 June of the respective year. A total of 80 experiments.

•RAS Convective Scheme (Hogan and Rosmond 1991) - Control

NCAR (Zhang and McFarlane 1995) Convection Scheme•6 hourly output

Storm Composite

Ensemble Model Results

Atlantic “HTV” Interannual Variability

r=0.78

EL EL EL EL EL EL ELLA LA LA LA LA

ACE Index - Atlantic Domain

(units 104 kt2)

r=0.85

Model and Observed TC Tracks 1986-2005Control

Ensemble 1

500hPa Streamlines – Control Experiment

“HTV” Landfalls 1986-2005 - Control

“Gates” Ensemble 1 Ensemble 2 Ensemble 3 Ensemble 4 HURDAT

Texas 8 14 10 4 35

Louisiana-Miss. 3 3 2 5 12

Florida-Georgia-Al. 23 17 17 15 32

Mid-Atlantic 2 4 5 2 13

New England 2 4 5 2 13

Ensemble Summary

Observations Ensemble 1 Ensemble 2 Ensemble 3 Ensemble 4 Ens. Mean

Total # ofStorms

245 242 234 234 249 240

Correlation 0.76 0.51 0.71 0.62 0.78

Variance 25.25 20.2 12.96 18.01 14.89 12.55

Sensitivity to Convection Scheme

r=0.78

r=-0.01

Sensitivity to Convection - continued

NRL NCAR

Control Experiment NCAR Convection Scheme

Sensitivity to Convection Scheme - Continued

NCAR Convection

500hPa Streamlines NCAR Convection

“HTV” Landfalls 1986-2005 – Sensitivity to Convection

Scheme

“Gates” NCAR Convection Scheme HURDAT

Texas 21 35

Louisiana-Miss. 6 12

Florida-Georgia-Al. 25 32

Mid-Atlantic 12 13

New England 12 13

Intensity Issue

Model and Observational Wind-Pressure Relationship-

Atlantic DomainKnutson et al. 2007, BAMS

18km Non-Hydrostatic Model

FSU/COAPS T126 Model(All 80 Ensemble Members)

1980-2005 1986-2005

Wind Pressure Relationship

Min

slp

(h

Pa

)

Wind Speed (m/s)

Lowest Pressure 936hPa

Model Atlantic/Pacific Basin Summary

Atlantic Pacific

Avg. Duration 7.5 days 8.3 days

(8.9 days) (9.6 days)

Avg. Central Pressure 995.9hPa 990.7hPa

850hPa Wind Max 60m/s 66m/s

Max Intensity 936hPa 926hPa

Avg. Number of Storms 11.7 15.9

Conclusions/Summary/Future Work

Summary/Conclusions

Ensemble hindcast results from a relatively high resolution atmospheric model

(T126L27) have been presented for 20-years of the Atlantic Basin hurricane

season using 2 different convection schemes.

Linear correlation of the interannual variability of the tropical storm frequency

against observation was found to be high (0.78) using the Hogan and Rosmond

convection, less so for the Zhang and McFarlane convection scheme.

Large sensitivity in track locations, storm numbers and interannual variability

was found between the two convection schemes and choice of diffusion coefficient

(not shown).

Model appears to simulate the ENSO-Atlantic covariation well.

Summary/Conclusions - cont.

•The model with the best interannual variability was NOT the best in simulating

land falling storms along the east coast of the U.S. and Gulf of Mexico. In part due

to the atmospheric large-scale response to the model's convection and the resulting

large-scale steering flow.

The model's surface wind-pressure relationship was found to be similar to a

20km global model (JMA not shown) and also an 18km non-hydrostatic model

(GFDL). All models fail to produce sufficient CAT3-5 level storms in terms of

surface winds.

Present Work

Better understanding of the sensitivity of tracks and intensity to the

choice of convection, diffusion coefficients and tracking algorithm.

Use selective years from the hindcast experiments and run the

FSU/COAPS regional spectral model to study higher horizontal

resolution impacts on hurricane seasonal statistics.

HOPE-OM1 Ocean Grid

Questions