Ensemble Forecasting of Hurricanes
FY 2001 Proposal to the NOAA HPCC Program

August 1,2000

| Title Page | Proposed Project | Budget Page |

Principle Investigator: Sim D. Aberson

Line Organization: OAR
Routing Code: R/AOML
                  AOML/Hurricane Research Division
                  4301 Rickenbacker Causeway
                  Miami, FL  33149

Phone: (305) 361-4334
Fax: (305) 361-4402
E-mail Address: Sim.Aberson@noaa.gov

Co-Investigator: Nirva Morisseau-Leroy
E-mail Address: nirva.morisseau@noaa.gov

Proposal Theme: Disaster Monitoring or Response

Funding Summary:    FY 2001 $62,100

Sim D. Aberson, Meteorologist
AOML/Hurricane Research Division
4301 Rickenbacker Causeway
Miami,FL 33149
Hugh E. Willoughby, Director
AOML/Hurricane Research Division
4301 Rickenbacker Causeway
Miami,FL 33149
Kristina Katsaros, Director
4301 Rickenbacker Causeway
Miami,FL 33149


Ensemble Forecasting of Hurricanes

Proposal for FY 2001 HPCC Funding

Prepared by: Sim D. Aberson

Executive Summary:

The current proposal is an effort to communicate the highly effective techniques of ensemble forecasting to the tropical cyclone community using an easy-to-use and flexible interface to display the large amount of resultant data.  Though ensemble forecasting systems have been used operationally since 1993 to forecast mid-latitude weather around the globe, these techniques have not yet been transferred to tropical areas, and more specifically to tropical cyclones.  Further, none of the current display systems are sufficiently flexible or simple to allow for easy access to the output.  The proposal is an effort to initiate an effective ensemble forecasting system for the North Atlantic basin, and to display the forecasts in an easy-to-use and flexible manner.

Problem Statement:

Since the atmosphere is a chaotic dynamical system, small errors in model initial conditions may lead to growing errors in the forecast, eventually leading to a total loss of any predictive information.  The rate of error growth, and hence the lead time at which predictability is lost, depends on a number of factors, estimates of which may be obtained by running a numerical weather prediction model from a number of initial conditions lying within the estimated uncertainty surrounding the control analysis, the best estimate of the true state of the atmosphere.  Ideally, all forecasts should be made from such an ensemble of initial conditions, since, in any nonlinear dynamical system, this approach offers the best possible forecast with the maximum information content.  Averaging the forecasts from the different ensemble members provides a forecast more reliable than any of the single forecasts, including that started from the control analysis.  Additionally, from the spread of the ensemble the reliability of the predictions can be assessed, and, with a sufficiently large number of realizations, any forecast quantity can be expressed in terms of probabilities, thus allowing for the conveyance of all the information available regarding future weather.

Tropical cyclones cause an average annual loss of near $5 billion in the United States.  Though timely warnings can lead to the saving of lives and property, the increasing population in coastal areas increases the necessary warning lead time so that necessary precautions and evacuations can take place.  Increasing the lead time of forecasts requires ensemble forecasting, so that forecasters and emergency management officials have available increasingly accurate forecasts with which they can assess the various possibilities, and the likelihood of each, in determining their response to potential disaster.  Such ensemble forecasts have been produced by a number of national meteorological centers since 1993.  However, these techniques have been developed for forecasting of mid-latitude weather.  To date, no operational tropical cyclone ensemble forecasting systems is available, though a number of feasibility tests have been made.

One of the most challenging aspects of ensemble prediction is condensing the vast amounts of model output and information into an operationally relevant and useful form.  One could simply display every product for each of the numerous forecasts, but this is extremely cumbersome and difficult to comprehend.  A satisfactorily flexible, comprehensive, and easy-to-use method to display the essential information must be developed.

An effective tropical cyclone ensemble forecasting system will require a model of adequate reliability in forecasting the important diagnostic (eg. track), an adequate number of ensemble members so that probabilistic forecasts can be made, an adequate display format so that the vast amount of critical information can be easily accessed by those in need, all delivered in an adequately timely manner to increase the effectiveness of NOAA operational hurricane forecasts and of emergency management in stricken areas.  The current proposal is an effort to initiate and provide such a system.  

Proposed Solution:

The proposed solution is in two parts:

  1. the development of the ensemble forecasting systems
  2. the development of a web-based application using Java component architectures to publish the results over the internet and AOML intranet.

Barotropic models provide skillful forecasts out to five days in the Atlantic basin, and provide the best, or second-best, forecasts on average, depending on lead time, of the currently available suite of forecast models.  As barotropic models are relatively simple, inexpensive, and accurate, a prototype orthonormalized bred-mode ensemble forecasting system utilizing VICBAR, a model developed at NOAA s Hurricane Research Division and run in real time since 1998 in deterministic mode, has already been developed.  However, current resources prohibit an ensemble forecasting system with an adequate number of members from being available in a timely manner due to the relatively large amount of computation.  The first part of the proposal involves transferring the system to an adequate computer system to allow for timely and sufficient results.

The second part of the proposal involves four parts:

  1. user requirement elicitation and database analysis and design
  2. storage of the ensemble output data using a modern commercial Database Management System (DBMS)
  3. development of a friendly user interface in Java, a web-based component to dynamically and efficiently query the output data stored in the database
  4. development of Java components that allows the query results to interact with the plotting mechanism of a meteorological graphics package to display the data.


A number of systems for the display of the output from ensemble forecasting systems are currently available.  None, however, are satisfactory.  Some, such as the Canadian system, allows for display of the many individual ensemble members using only a preset list of diagnostics; users therefore must look at a large number of very small maps, and may need to divine features of interest from what is presented.  The public system at the National Centers for Environmental Prediction is very cumbersome, though it allows for presentation of more diagnostics in an easier-to-read format than the Canadian system; one can look at single maps of ensemble means, and diagrams in which a feature of all ensemble members are plotted on one map (usually called spaghetti diagrams), though all the diagnostics are pre-set.  In all these systems, the graphics are pre-created, and only specific diagnostics are presented.

The proposed system would allow for presentation of any available diagnostic at any available level in the atmosphere, with a variety of presentation options, such as the ensemble mean, the ensemble spread, spaghetti diagrams, and probabilistic forecasts, and animations of these.  Since the barotropic model forecasts only wind and mass fields in one layer of the atmosphere, the early versions of this system created specifically for VICBAR could be augmented to allow for presentation of more sophisticated tropical cyclone (or even regional or global model) ensembles once they become available, therefore potentially benefitting the entire forecasting community.

Performance Measures:

All deliverables will depend upon time relative to the busy hurricane season. 



Project-related References

Budget Summary:  
Contracts University Cooperative Agreement
Contracts  Software Support
Capital  Costs Computer and Software for ensemble runs
Capital Costs Oracle8I Parallel Query Option
Capital Costs Oracle8I Partition Option
Capital Costs 1G RAM memory for database
Capital Costs 20G hard drive
Travel/Training _________$5,000
Total Requested:


The University Cooperative Agreement grants are for a part-time contract Data-Base Administrator and application developer, and for a part-time student contractor for application development and maintenance, and overhead for both.

Software support is technical support for the database software.

Capital costs include a Unix machine for running of the actual ensembles with a speed of at least 20Specfp95, necessary querying and partitioning software to augment the current AOML database, RAM memory for the database, and a hard drive to store the large amounts of output data from the ensemble forecasting system.

Travel and training costs include travel to conferences and for database training.

  FY2001 request                                                  $62100
  FY2002 possible recurring                                 $40000
  FY2003 possible recurring                                 $30000

Administrative Officer: Cathy Steward

Phone:  (305) 361-4303
E-mail Address:  Cathy.Steward@noaa.gov

FMC Number:  940