What is the Kilo-Member Ensemble?
The operational model guidance products used for tropical cyclone track prediction have drastically improved over the past few decades. Some of the global models have done exceptionally well over the past few seasons. Nevertheless, due to the chaotic nature of the atmosphere arising from instability and nonlinear interactions on many scales, small errors in the initial condition coupled with deficiencies in the model's representation of the atmospheric processes and dynamics result in a growth of errors in the model forecast. Thus, in time the model's forecast diverges from the evolution of the real atmosphere, sometimes slowly, and rapidly at other times. The uncertainty associated with a deterministic forecast can be predicted using ensemble techniques to sample the divergence of dynamical pathways from slightly different initial conditions. The information from the various members of a properly perturbed ensemble can be collected to yield a forecast that is reliably better than any single deterministic forecast.
The kilo-member ensemble uses a very simple, yet extremely efficient nondivergent multigrid barotropic model to produce over a thousand forecasts for each case. Each ensemble member starts from a slightly different initial condition (a perturbation) which simulates the uncertainties associated with the initial condition and subsequent flow evolution. Thus, the kilo-ensemble simulates a wide range of possible future atmospheric states, sampling the dynamical pathways available to the tropical cyclone, and (hopefully) producing an ensemble mean that is superior to a single control forecast.
The ensemble mean (a single moment statistic) is just one of many pieces of information that an ensemble may provide. The higher moment statistics such as ensemble spread (second moment) and skewness may be useful in estimating forecast uncertainty -- in other words, an ensemble with a large spread likely indicates an atmosphere that is very sensitive to small errors in the initial condition -- thus the forecast would be assigned a low confidence. Also, if the perturbations accurately simulate all of the growing modes of the atmosphere, and the ensemble is large enough, the ensemble envelope can be used as a prediction of the envelope of all possible storm tracks -- encompassing all possible outcomes.
The kilo-member ensemble seeks to accomplish these goals by perturbing across five classes in the parameter-phase space of a nondivergent barotropic model. Cross-multiplication across the five perturbation classes results in 1980 unique ensemble members. This research seeks to determine whether the above goals can be met using a very large ensemble.
Acknowledgements
This research has been conducted toward completion of my (Jonathan Vigh) Masters Thesis at the Department of Atmospheric Science (ATS) at Colorado State University. Much thanks goes to my advisor Dr. Wayne Schubert, and my committee members: Dr. Mark DeMaria (NOAA-NESDIS-CIRA), Dr. Bill Gray (ATS), and Dr. Gerald Taylor (CSU Mathematics Department). I am grateful to Scott Fulton for the use of the MUDBAR model, and for his modification to allow the use of model forecast wind fields for the time-dependent boundary conditions. I also appreciate the helpful interactions and assistance of other members of the Schubert Research Group including Rick Taft, Brian McNoldy, Chris Rozoff, Paul Ciesielski, and Gail Cordova. Finally, I'm grateful for the support of friends, family, and God through this long, yet rewarding process.
This work has been funded in part by a graduate fellowship from the American Meteorological Society and the U.S. Department of Energy, through fellowship support from the National Science Foundation through the Significant Opportunities in Atmospheric Research and Science (SOARS) Program, and by the following grants ??????.
