Hurricane Katrina: The meteorology

By Chris Davis, NCAR

Almost ten years ago, on August 25, 2005, a category 1 hurricane named Katrina made landfall in Miami, Florida. The storm produced significant rains but not much damage or fanfare compared to other storms that have hit Miami.  Forecasters were anticipating what Katrina would do after it passed over the Everglades and back over open water in the Gulf of Mexico.  A turn to the right was expected, but how soon? How intense would the storm get?  Each successive advisory delayed the right turn, and in doing so, placed the storm over the Gulf of Mexico for a longer period of time, over increasingly high ocean heat content.  The advisory issued at 11 PM on August 26 portrayed a major hurricane reaching eastern Louisiana on Monday afternoon, August 29.  Hurricane warnings were posted on the evening of August 27.  The storm came ashore on Monday morning, somewhat earlier than expected, but very close to the track anticipated more than two days earlier.

The forecast of Katrina was deemed a major success.  We all know that was, tragically, only part of the story.  The levees collapsed, New Orleans was flooded and more than 1000 Errors in tropical cyclone forecast for different lead timespeople perished despite a good forecast.  But was the forecast really that good? The actual position error in the official 60-hour forecast of Katrina was about 120 nautical miles.  That was close to the average error for all 60-hour forecasts made in 2005 (Figure at right).  The figure, adapted from the National Hurricane Center’s web page, shows errors for 48 h and 72 h forecasts (green and yellow curves), so we have to do a little interpolation to arrive at this conclusion.  We can debate whether an error in the location of landfall is worse than the timing of landfall. It probably is, but the standard operational verification of hurricane position forecasts makes no distinction between errors along the track (i.e., when landfall occurs) and errors across the track (where landfall occurs).  From the metric of position error, the 60 h forecast of Katrina was not exceptional for 2005, and longer-range forecasts were considerably off the mark.

Let’s place the typical forecast errors from 2005 in the context of today’s forecasts.  The figure shows how errors in forecasts of different lead times (corresponding to different colors) have decreased during the past 15 years or so.  Errors at long lead times have decreased more rapidly than errors at short lead times.  That is not surprising because it is very difficult to improve upon forecasts with small errors, such as 24-h forecasts.  Today, a similar forecast accuracy is attainable at 84 hours that was possible at only 60 hours in 2005.  From this perspective, the forecasts of Katrina made at that time would be considered poor by today’s standards.

Does the improvement of hurricane forecasts since Katrina make a qualitative difference, or is it incremental?  I believe the former.  We have reached a fundamental threshold in hurricane forecasting skill.  The horizontal black line in the figure represents the average radial extent of gale force winds from the center of all Atlantic storms since 1988. This extent is about 125 nautical miles.  Nearly all the destruction near the coast occurs within this radius when the storm comes ashore.   Storms come in different sizes, so this is just an average. Where the black line intersects the downward sloping lines tells you the lead time at which the position error and storm size are comparable.  If a given colored line lies below the black line, then at that lead time the error in predicting storm location is smaller than the extent of the region experiencing hazardous conditions.  At that lead time, on average, NHC can produce accurate forecasts of where the strong winds will occur along the coast, and with those winds comes the storm surge that often causes most of the damage. 

By this benchmark of accuracy, where the position error is comparable to the storm size, forecasts in 2014 were accurate to almost four days, but forecasts in 2005 were only accurate to 2.5 days.  This difference is important.  It takes several days to evacuate people from many locations.  Accurate forecasts four days in advance not only give emergency managers (EMs) time to evacuate people, accurate forecasts also give EMs greater certainty about who NOT to evacuate.

What would have happened in Katrina if the forecast was as accurate at 4 days as it was at 2.5 days?  Would it have mattered?  For the levees, certainly not, and some people would have stayed no matter what.  But consider this: The first forecast that placed New Orleans squarely in the cross hairs of Katrina was issued at 11 PM on a Friday night.  The hurricane warning came out at 10 PM on Saturday night.  Even though this was just ten years ago, few people had smart phones, and few forecast products were available on those phones that existed.  I would also expect that fewer people were paying attention to forecasts on the weekend than during the week.  Perhaps they did not notice the forecast until Sunday, just 24 hours (or less) before landfall.  By then it might have been too late to move.

Consider a four-day lead-time for Katrina.  The first accurate forecast would have come out Thursday morning.  There would have been nearly two full workdays to inform people and make preparations.  We will never know how many lives would have been saved with the extra lead-time.   But it is easy to imagine that more people would have been warned in time to take action, or warned for a longer time so that the message would have gotten through.  The outcome would have been positive, perhaps dramatically so.

The greatest hazard in coastal regions posed by hurricanes is storm surge, not wind.  With greater accuracy of forecasts, potentially useful for explicit hazard prediction out to four days in advance, EMs will have more information to make critical decisions.  But with the emphasis on predicting hazards, we need to reconsider how we assess the hazard posed by a given storm.  It turns out that while the hurricane intensity is measured by the maximum wind, the maximum wind is not the most important factor affecting the surge.  The most important factor is the size of the storm.  Of course the wind speed matters, but it is the wind over an extended area, not the wind at a point, that really matters.  We as a community have never focused on making and evaluating predictions of storm size.  It is likely that we can make some significant improvements in this area that will translate directly into improved hazard prediction.  We should use our improved prediction capabilities to make our forecasts more specific to the actual threat, and work to communicate that threat effectively.