We have all seen them. A heavy rain event looms, and weather forecasters discuss how much precipitation is anticipated by various forecast models. A hurricane is approaching the Gulf of Mexico, and there is talk of where various models suggest the storm may head.
What are these “models,” and why do they sometimes miss the mark? Will they eventually predict such events so accurately that we have the certainty and time to truly prepare for them?
Forecast models are science-based tools used to predict future events. Only in the 1950s did meteorologists begin feeding data into computers, hoping to predict the state of the atmosphere. Unfortunately, the atmosphere is an extremely complex fluid that is always in motion and subject to many variables that can influence future weather.
To reflect this, weather scientists have developed computer-forecast models that use mathematical formulas to create what is now called “Numerical Weather Prediction.” Information about the current state of the atmosphere, such as temperature, pressure, moisture content, wind speed and direction, is entered into a computer, which uses equations to predict weather changes over time.
These efforts have led to gradual improvement in weather forecasts, as models such as the GFS (Global Forecast System), the ECMWF (European Center for Medium-range Forecasts) or the GEM (Global Environmental Multiscale Model), are used by forecasters to analyze and predict local, national and global weather.
This methodology has been adopted by hurricane specialists to improve forecasts of the tracks, intensity and storm-surge potential for tropical cyclones. Hurricane models such as the GFDL (Geophysical and Fluid Dynamics Laboratory), the HWRF (Hurricane Weather Research and Forecasting), and a U.S. Navy Version of the GFDL called the GFDN have been developed over the past couple of decades.
There is still room for improvement, however. One challenge is the impossibility of obtaining measurements of the entire atmosphere, leaving many gaps in data collection. Even slight errors or unknowns can lead to very different outcomes over time. A second issue has to do with computing power. More data fed into a computer requires more power to perform complex equations and develop accurate outcomes. A third challenge is that the exact weight or significance of each atmospheric component is not always certain, leading to different interpretations by various models.
In 2010, NOAA initiated a systematic 10-year program — the Hurricane Forecast Improvement Program — to markedly improve hurricane predictions, to reduce the average errors of hurricane track and intensity forecasts by 50 percent with a forecast period out to seven days. NOAA is adding computer capacity, and many models are adding data collection grids to provide better information on current atmospheric conditions.
Recently NOAA and Congress have cut funding to this project, despite the progress. The latest budget proposals are calling for further cuts, drawing protests from Florida’s two senators.
Improvements in hurricane forecasts that began in the 1970s have resulted in a steady decrease in average forecast track errors. The average forecast track error at 72 hours before landfall has decreased by more than 200 miles since the 1970s.
Despite this steady progress with landfall predictions, additional improvements may be more challenging to achieve, as will improvement in intensity and storm-surge forecasts.
There are, however, important caveats for the general public, regarding the use of computer models:
• Do not rely solely upon models. Weather and tropical cyclone forecasts reflect consideration of many factors, including all available model guidance, forecaster experience and other sets of data. Always heed the expertise of the forecasters when making decisions regarding major weather events.
• Be wary of forecasts based on only one model. The best forecasts reflect the results of many model outputs over time. While model ensembles (or “spaghetti” plots) are useful to trained meteorologists, they can be misleading for the general public, who may not understand the relative validity of the various models use to create the array of forecast tracks.
• Look at model trends rather than the latest model forecast alone. This is likely to provide a better idea of what the models reflect and forecasters are thinking.
• Remember that accuracy tends to increase over time. Even with improvements in prediction, there is likely to be considerable change between the 120-hour forecast and a subsequent 24-hour model outlook.
The good news is that improved models enable forecasters to make ever more accurate predictions. Progress has been made and the HFIP represents a decisive step in the right direction in hurricane prediction.