Sixth International Workshop on Tropical CyclonesSixth International Workshop on Tropical Cyclones

21 – 30 November 2006, San Jose, Costa Rica

Topic 3.1 :

Advances and Requirements for Operational Tropical Cyclone Track Prediction

WG Members :

Wes Browning, Philippe Caroff, Eun-Jeong Cha, Duong Lien Chau, Cuiyin

g Tian, Julian Heming, Alan Radford, Prisco Nilo, Mannoji Nobutak, David

Richardson, Ali Shareef, S.K. Subramanian

T.C. Lee (Rapporteur)

1. Recent developments in operational TC track prediction techniques and tools

2. Improvements in operational TC track forecasts

3. Roadblocks

4. Future opportunities and recommendations

5. Summary

Content

Advances and Requirements for Operational Tropical Cyclone (TC) Track Prediction

1. Recent developments in operational TC track prediction techniques and tools

(a) Improvement in skill of NWP model (JMA & UKMO)

Figure 3.1.1(a) 5-year running mean tropical cyclone forecast errors of JMA global model for western NorthPacific and South China Sea from 1995 to 2005 (Data Source : RSMC-Tokyo)

0

100

200

300

400

500

600

700

800

900

1000

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year

5-ye

ar m

ean

erro

r (k

m)

24-hr F/C

48-hr F/C

72-hr F/C

96-hr F/C

120-hr F/C

JMA Global Model

(i) 72-hr F/C error in 2001-2005 < 48-hr F/C error in 1991-1995

(ii) 5-year mean F/C error : 10-15% reduction between 1996-2000 and 2001-2005

(b) Availability of NWP products and operational forecasting tools

(i) More NWP products for National Met. and Hydrological Services (NMHSs) via GTS and/or Internet through bilateral or regional arrangements (e.g. JMA, ECMWF, NCEP, etc)

(ii) Official TC warnings from Severe Weather Information Centre (SWIC) website of the WMOhttp://severe.worldweather.wmo.int/

(iii) JMA implemented a password-protected "Numerical Tropical C

yclone Prediction Web Site" - displays TC forecast tracks from major NWP centres for access by the Members of the ESCAP/WMO Typhoon Committee

(c) Developments in multi-model consensus TC track forecasts/EPS

(i) The multi-model consensus approach has been adopted by several warning centres as the primary guidance for TC track forecasts in recent years.

(ii) Selective consensus method(Carr et al. 2001), weighted consensus(Weber 2003), and statistical/linear regression schemes (Vijaya Kumar et al. 2003; Zhang 2006) have been developed to reduce the error of the consensus track.

(iii)A consensus prediction error product (Goerss 2004, 2006) :- provide a measure of confidence in the consensus forecasts.

(iv)Ensemble Prediction Systems (EPS) outputs are made available to NMHSs (e.g. Strike probability maps of ECMWF)

(d) Development of tools for operational TC track forecasting

Developed tools to assimilate and display track forecast information and to interpolate model tracks to facilitate the formulation of forecasts and warning strategies

Tropical Cyclone Information Processing System (TIPS) operated in HK, China

2. Improvements in operational tropical cyclone track forecasts

(i) The accuracy of 24-, 48-, and 72-h track forecasts for 8 centres from 1990 to 2005 was reviewed. (data provided by working group members or available online)

- Regional Specialized Meteorological Centre-Tokyo (RSMC-Tokyo); - Regional Specialized Meteorological Centre-La Reunion (RSMC-La Reunion);- National Weather Centre of China Meteorological Administration (CMA); - Joint Typhoon Warning Center of U.S. Department of Defense (JTWC); - National Hurricane Center of U.S.A. (NHC);- Central Pacific Hurricane Center of U.S.A. (CPHC); - Vietnam National Center for HydroMeteorological Forecasting (NCHMF); and- Hong Kong Observatory (HKO)

(ii) 5-year running mean error : smooth the interannual variations in track errors and the number of TC forecasts. Thus illustrate the long-term trend.

Figure 3.1.5(a) 5-year running mean error of 24-hour tropical cyclone track forecasts by different operationaltropical cyclone warning centres (1990-2005)

100

150

200

250

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year

5-y

ear

ru

nnin

g m

ean

err

or

(km

)

HKO

CMA

RSMC-Tokyo

NHC-EPC

NHC-Atlantic

JTWC-WNP

JTWC-NIO

JTWC-SH

RSMC-La Reunion

NCHMF

CPHC

24-hour forecast errors

Figure 3.1.5(b) 5-year running mean error of 48-hour tropical cyclone track forecasts by different operationaltropical cyclone warning centres (1990-2005)

150

200

250

300

350

400

450

500

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year

5-y

ea

r ru

nnin

g m

ea

n e

rro

r (k

m)

HKO

CMA

RSMC-Tokyo

NHC-EPC

NHC-Atlantic

JTWC-WNP

JTWC-NIO

JTWC-SH

RSMC-La Reunion

CPHC

48-hour forecast errors

Figure 3.1.5(c) 5-year running mean error of 72-hour tropical cyclone track forecasts by different operationaltropical cyclone warning centres (1990-2005)

250

300

350

400

450

500

550

600

650

700

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year

5-y

ea

r ru

nn

ing

me

an

err

or

(km

)

RSMC-Tokyo

NHC-EPC

NHC-Atlantic

JTWC-WNP

JTWC-NIO

JTWC-SH

CPHC

72-hour forecast errors

(iii) Significant improvements have been achieved by practically all of the centres

(iv) In general, the 24-, 48- and 72-hour TC track forecast errors have been reduced to around 150 km, 250 km, and 350 km, respectively, during 2001-2005.

(v) 10% to 35% reductions in the 5-year mean 24- and 48-hour errors in the two consecutive 5-year periods of 1996-2000 and 2001-2005

(vi) Reductions in errors for some NMHSs and RSMCs are noticeably higher than the 10% to 15% of individual models.

The use of the multi-model consensus and the EPS products add values to the model TC track forecasts

2. Improvements in operational tropical cyclone track forecasts

3. Roadblocks

(i) Lack of observations

- lack of observations over data-sparse areas hinders the numerical and operational TC track prediction

- temporal coverage of polar orbiting satellites does not meet all the ne

ed for real-time operations

(ii) Diversity of NWP guidance in forecasting TC track

- simple consensus forecasts usually do not work when models gives very diverse forecasts

- very large forecast errors could be resulted if forecasters could not identify and remove the erroneous model/cluster

(iii)Difficulty in forecasting landfalling TCs

- for landfall point forecast, an average 24-hour error of about 100-150 km is not sufficiently accurate for the protection of a city/place

(a small deviation in F/C track will make a lot of difference)

(iv)Limited accessibility of NWP products and lack of operational TC track forecasting tools

- Some NMHSs only have access to the outputs of no more than one or two NWP models. Multi-model consensus tracks thus cannot be constructed.

- Those NMHSs already using multi-model consensus method require more NWP products to identify the optimal combination of forecast members

- Some NMHSs also lack resources and experience to develop operational tools for assimilating and displaying TC forecast information from advanced centres.

(v) Lack of training

- Many forecasters and meteorologists lack the appropriate training to make good use of NWP products from advanced centres.

3. Roadblocks

4. Opportunities and recommendations

(i) More observations for determining TC position and intensity

- NPOESS satellites to be launched in next few years

- NMHSs are encouraged to disseminate weather obs. data and radar fix (RADOB) as much as possible

(ii) Development and utilization of EPS products

- post-processing of EPS TC tracks to improve track prediction

- integrating the spread information from consensus or the EPS track forecasts with the operational forecast track to indicate the confidence (rather than fixed climatological average errors)

balance the lead-time versus accuracy trade-off evaluate investments to reduce lead times or create flexibility in preparations

4. Opportunities and recommendations

(iii) Development of mesoscale modeling techniques

- High-resolution mesoscale models incorporating local climate features and topography may help to improve TC track forecasts near landfall

(e.g. North American regional model of UKMO and GRAPES of CMA)

(vi) Availability of TC forecast information and tools

- The JMA "Numerical Tropical Cyclone Prediction Web Site" is an encouraging example. It is recommended that similar initiatives be developed for other TC basins

- NWP centres should disseminate TC track forecasts and EPS products in a standardized digital / alphanumeric format via GTS and internet.

4. Opportunities and recommendations

(v) Collaboration and training

- more training for forecasters: the use of multi-model consensus and EPS products in operational TC track forecasting.

- Enhance interaction between researchers and forecasters on Operational Tropical Cyclone Forecasting. (both regional and International)

Enable : NMHSs to provide feedback on the usefulness of the NWP models; forecasters to share experiences on operational TC forecasting; model developers and researchers to better understand the needs and diffic

ulties of the forecasters/NMHSs; and the identification of research needs and how research results could be adopt

ed for operational use.

5. Summary

Considerable progress in operational TC track forecasting in the past decade (Improved NWP models guidance and the use of the multi-model consensus forecasts and EPS products)

The 5-year mean of the 24-, 48- and 72-hour forecast errors have been reduced to around 150 km, 250 km, and 350 km respectively during 2001-2005, roughly 10% to 35% less than those during 1996-2000.

Require more reliable objective forecasting techniques (divergent track scenarios and short-range forecasting of the landfall point)

NMHSs need more TC track guidance from NWP centres. (timely and convenient access and in a standardized format)

Forecasters require software tools and appropriate training to fully utilize the improved NWP-based TC forecast information in the operational environment.

Thank you

Figure 3.1.1(b) 5-year running mean tropical cyclone track forecast errors of UKMO global model forNorthern Henisphere from 1992 to 2005 (Data Source : UKMO)

100

200

300

400

500

600

700

800

900

1000

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Year

5-y

ear

mea

n e

rror

(km

)

24-hr F/C

48-hr F/C

72-hr F/C

96-hr F/C

120-hr

UKMO Global Model (Northern Hemisphere)

ECMWF strike probability maps for Hurricane Katrina for EPS forecasts initialized on 26 August at 00 UTC (left) and 12 UTC (right).

Strike probability maps for Typhoon Songda based on JMA EPS run

initialized at 12 UTC on 2 September 2004

Strike probability maps for Typhoon Songda based on JMA EPS run initialized at 12 UTC on 2 September 2004 conditioned by three actual TC observed positions (at 12, 18, 00 UTC) between the EPS run initial time and data availability time (around 00-01 UTC) at HKO, with actual TC track in black