The accuracy of track forecasts of tropical cyclones has improved significantly over the last 39 years. The National Hurricane Center (NHC) official 72-hour average forecast track errors have decreased from about 450 nautical miles (nm) to less than 150 nm since 1970. Beginning in 2003, the NHC began to issue official 96 hour and 120-hour forecasts. Although the record for these extended range forecasts is short, they also appear to be improving with time.
Within this general trend in the average forecast track errors, there are particular storms that are more difficult than average to forecast accurately. We have begun a project to investigate the forecasts for these troublesome storms, concentrating on the performance of the models the NHC uses in their forecast process. A subjective list of storms which showed larger track errors than average was compiled by the Hurricane Specialist Unit of the NHC. The model forecasts for these storms were plotted, together with the best-track locations for verification. The resulting images were examined to look for trends in the forecast errors, and to determine if there were particular points in the storms' lifetimes where track errors tended to occur. The model initializations were then examined and compared, particularly for the model forecasts that began at the times of larger than average track errors. The models examined were the Global Forecast System (GFS) from the U.S. National Centers for Environmental Prediction, the Unified Model (UKMET model) from the United Kingdom's National Weather Service, the global atmospheric model from the European Center for Medium-Range Forecasting (ECMWF model), the Global Forecast Model from the Canadian Weather Service (GEM), and the Navy Operational Global Atmospheric Prediction System (NOGAPS). These models were selected because they are either examined directly by the NHC, or their initializations and forecasts are used for regional or special purpose forecast models, such as the Hurricane version of the Weather Research and Forecasting model (HWRF model), which uses the GFS for initialization and time-dependent boundary conditions.
The model initializations were compared with each other, and with whatever upper air observations were available, including satellite imagery. Water vapor imagery was especially helpful to judge the amplitude and location of upper air troughs and ridges. Using these analyses, it was possible to identify defects in the analyzed storm environment. The Advanced Research version of the Weather Research and Forecasting model (ARW) was used to understand the impact of varying initial conditions on the subsequent track of the storm. Model runs were made using initial conditions from different models but with a very large outer domain for the simulation, to try to minimize the influence of the boundary conditions on the model simulation. Thus, differences in subsequent storm tracks would be due entirely to the differences in initial conditions.
The attached Figure shows the 500 hPa initial analyses from the GFS (green contours) and the NAM (yellow contours) at 00 UTC, November 7, 2009, when Tropical Storm Ida was still well south of the domain shown. Both analyses show a trough over northeastern Mexico, but the trough in the NAM analysis has a much higher amplitude than that in the GFS analysis. The resulting steering flows are very different from each other, and will have very different influences on the subsequent track of a tropical system embedded in them.
Our goal for this project is to quantify the impact of the initial analysis on a model forecast, and to determine practical ways in which the quality of the initial model analyses can be used in real time to aid the forecasters. In this example, water vapor imagery showed that the sharper trough featured in the NAM initial analysis was correct.
Figure. Gempak depiction of the 500 mb initial analysis height field (in m) for the GFS (green) and the NAM (yellow) for 00 UTC, November 7, 2009.