Number 33, April 2002

Guest Editorial

Tropical Cyclone Landfall Forecasting: Making Research Relevant

We've recently published research that calls into question some long-held beliefs about hurricane forecasting and research. We conclude that hurricane research can be made more relevant.

Hurricane forecasters do not focus their forecasts and warnings on the exact location of the storm center because damaging winds cover a large region surrounding the actual center. Also, since forecasters issue new storm path predictions every 6 h, the geographic region expected to be impacted by damaging winds and storm surge also varies.

How are these forecasts used in practice? Anyone with Internet access can plot the forecast track and see where it crosses the coast and how the predicted landfall location changes every 6 hours. There are even commercial software products that provide full color animation of the forecast wind field at landfall. Though probabilistic landfall forecasts are issued by the National Hurricane Center (NHC) in both text and graphical format, the general public is typically focused on when and where the storm will hit land. So in practice, even if they are looking at the center of a cone of uncertainty in a graphical product, some portion of the informed public, including the emergency managers, are really paying attention to the "landfall forecast," but this forecast is neither explicitly issued nor routinely verified.

The 12, 24, 36, 48, and 72 h (and perhaps eventually 96 and 120 h) storm positions are contained in the official forecast and these are compared to the observed location of the storm when compiling the annual error statistics. These data show 1-2% annual improvements in Atlantic basin tropical cyclone forecasts over the last three decades. While this improvement is commendable, only 15% of the forecasts issued from 1976-2001 were for storms predicted to cross the U. S. coastline, and 2% of these 15% were for storms that ended up remaining at sea. Unless the 24, 48, and 72 h positions happened to coincide with the coastline, there is no available information to suggest that the overall improvement trend applies to landfall forecasts, or that it can be projected into the future. Nevertheless, goal-setting documents from the U. S. Weather Research Program, the National Weather Service, and the Executive Office of the President have recently made this assertion. By not verifying the most important part of the forecast in terms that the public can readily understand (when and where it crosses the coastline), we may be missing an opportunity to inform the public about the capabilities and limitations of our predictions.

In our research we asked: Does the improvement trend apply to landfall forecasts? To examine this question we followed a simple approach first used by Charlie Neumann and Joe Pelissier at NHC back in 1980: compare the observed landfall position and time to where and when the officially forecast track of the storm first crosses the coastline. Our findings (recently published in the December issue of the AMS Bulletin) suggest:

• The improvement trend does not apply to landfall position forecasts. Position forecasts do not show a significant improvement (or degradation) trend since 1976, although time of landfall does show significant improvement at the 24 h period.
  
• Within 30 h of the coast, forecasts tend to predict landfall too early.
  
• The early bias and lack of improvement are consistent with a "least regret" forecast and warning strategy to account for possible storm accelerations and intensity changes.
  
• Uncertainty in the time of landfall is similar when the storm is 24 h or 36 h offshore (8h +/- 11 h), suggesting that it may be possible to extend warnings to 36 h (although the length of warned coastline would need to increase).
  
• Landfall position and time forecasts are skillful. To be skillful the forecast has to do better than a simple model based on extrapolating the current motion (persistence) and considering how past storms have moved (climatology).
  
• Current hurricane warning areas imply a 5% chance that a storm might hit outside the warned area; this is ~200 km smaller than a warning area based on reducing the risk to 1%, suggesting an inherent value of $70 million per warning episode.
  

Landfall time and position are never specifically forecast yet several research goal and policy- related documents have been published that aspire to improve landfall forecasts by applying the well- documented improvement in basin-wide hurricane forecasting. The National Weather Service strategic plan "Vision 2005" seeks to "Increase the average lead time for hurricane landfall forecasts from 19 h to beyond 24 h with no increase in warned area." In a document entitled "Effective Disaster Warnings" published by the Subcommittee on Natural Disaster Reduction with the seal of the Executive office of the President on the cover, it was stated, "Prediction of hurricane landfalls is improving...For the next four year period, forecasts for landfalling storms should improve an additional 20% due to use of better models and data." Even the U. S. Weather Research Program lists a goal of "reducing landfall track and intensity forecast errors by 20%."

Our results suggest these goals may not be achievable and the perception of accuracy improvement may be leading to unrealistic expectations of both scientists and end users alike. If such documents are calling for improvements in landfall forecasts, then we need to understand why these forecasts have not improved despite evidence of improvement of forecasts throughout the Atlantic basin. One possibility is that the forecasts near land are already pretty good (taking advantage of the proximity of the observation network) and therefore difficult to improve. Recent work by Aberson suggests that landfall position forecasts meet or exceed an estimate of the predictability limit in the Atlantic basin at 36, 48, and 72 h. Further examination of objective guidance and official forecasts is necessary to determine whether a "course of least regret" diverges from one of continuous improvement. Researchers should examine landfall forecasts of numerical weather prediction models to determine errors and whether improvement trends exist. In addition, they should examine landfall intensity prediction errors of official forecasts and model guidance.

If the goal is to improve landfall forecasts, let's focus on that small number of cases that really produce the impact. Our results suggest that landfall forecasts are already pretty good and difficult to improve, but we're not really sure why. The tropical cyclone community and USWRP should examine this problem in more detail and then come back with research and operational goals for forecasts that will communicate uncertainty, yet focus on what the user community wants to know: where and when is that storm going to strike, how strong will it be when it hits, how much rain will we get, and what areas will be flooded by surge, waves, and rain?

Mark Powell and Sim Aberson
Hurricane Research Division,
Atlantic Oceanographic and
Meteorological Laboratory
Mark.Powell@noaa.gov
Sim.Aberson@noaa.gov