I unfortunately own property in Tampa, FL (or at least I do as of the time of this writing anyway). In following Hurricane Milton and trying to assess what might happen, I realized I had no idea how good (or bad) we actually are at predicting either the path or intensity (specifically with respect to wind) of hurricanes. So, I started digging. Note that all of the below data is dealing specifically with the Atlantic Basin (most relevant for the United States).
So what's the answer? Unfortunately that probably depends on who you ask. What defines a "good" forecast, after all? Does it need to predict the path within a mile? Surely not. Is 100 miles sufficiently accurate? Unclear. One thing is for sure though: we're definitely getting better at predicting both the path and intensity of hurricanes, and that is very encouraging.
Below are some charts showing the tracking errors for both the path of the storm (retroactively described as the hurricane's "best track") and the intensity of the storm (generally measured by winds). Perhaps not surprisingly, our forecasting accuracy is inversely correlated to the time since we made the forecast (to use an extreme example, a storm is likely to deviate the least from a forecast that was made 5 minutes ago compared to one made 5 days ago). Simple geometry makes this quite reasonable (if you miss your forecast by 5 degrees, every day that goes by that error is compounded).
What really stood out to me when I looked at the data was how much better we're getting at forecasting at each time interval. The improvements have been so noticeable that forecasters have actually started forecasting tropical cyclones further in advance. Note that a "tropical cyclone" is "a generic term used by meteorologists to describe a rotating, organized system of clouds and thunderstorms that originates over tropical or subtropical waters and has closed, low-level circulation." Once a cyclone reaches sustained winds of 74 MPH, it becomes classified as a hurricane, typhoon or tropical cyclone, depending on where it is in the world. "In the North Atlantic, Central North Pacific, and eastern North Pacific, the term 'hurricane' is used," the National Oceanic and Atmospheric Administration tells us.
The first recorded 24 hour forecasts came about in 1954, and then extended to 48 hours in 1961, and both 12 and 72 hour forecasts were introduced in 1964. 36 hour forecasts were introduced in 1988. Then in 2001, 72 hour (3 day) forecasts were introduced. Finally in 2003, 96 (4 day) and 120 hour (5 day) forecasts were introduced.
Here's what the tracking error charts look like for each of these intervals. For sake of chart clarity, I'm going to break all this up into two groups. To be clear, what these charts show is how many (nautical) miles away hurricanes tend to be, on average, at a given time after a forecast is made. So for example, in 2023, the average tracking error for forecasts made 12 hours ago was 24.5, indicating that on average, the eye of the hurricane was 24.5 nautical miles off from where it was forecasted to be. Similarly, for forecasts made 36 hours ago, the average margin for (tracking) error is more like 40-50 (nautical) miles.
2023 aside (which seems to have been a tougher year for forecasters), notice the downward trajectory in each of the lines in both of the charts above. This indicates that we're getting better at predicting the path these hurricanes are taking (indicated by the number of miles off our forecasts are). Why does this matter? It reduces the number of people who might need to evacuate, spend extra money taking what might otherwise end up being unnecessary precautions to protect their houses (including evacuation costs), etc. Better forecasts save society time, energy and money.
Before we turn to wind, a closer look at some of the charts did catch my eye. While we seem to be getting better at forecasting further in advance, we don't actually seem to be much better at forecasting at sooner intervals. While this should in theory have less of an impact (ideally you want a more accurate forecast further in advance to give people more time to take precautions, and / or evacuate), it's interesting that the improvements in forecasting at further out intervals (say 3-5 days) aren't also translating into improved forecasts at 0-2 days. Consider the following:
Here we have clear trend improvement, absent 2023, which as noted above appears to be a rougher year for forecasters. Now here's the 4 and 3 day forecast charts. Also notice the clear downward trend even recently, absent the spike in 2023.
Now let's compare these to the 0-2 day charts. Notice how much more those charts have plateaued in the last couple of decades.
While it's less noticeable in the 48 hour chart, now look at 24, 12 and 0 hour charts, where forecast errors seem to have even more so generally plateaued for the better part of a decade (and even more for the 0 hour forecasts).
Why this happening is not entirely clear, but it's worth investigating further.
What about the intensity of hurricanes (best measured by wind)? The same general trends are occurring, except here we don't seem to have the same plateauing effect for forecast errors in nearer dated forecasts. 2023 was also a tough year for forecasters from a wind perspective as well, but consider the below chart, and again notice the downward sloping lines across almost all forecast periods.
So in summary, we're getting better at forecasting both the path, and intensity of hurricanes, and particularly so at further out intervals (or said differently, further in advance). Our progress in predicting their path at sooner intervals, however, seems to have stalled out, but this is less important than getting the forecasts more correct further in advance. We've come a long way from Air Force pilots flying rogue into the center of a hurricane on a bet (this is an absolutely fascinating, must read story for a variety of reasons), but hopefully we can continue to improve as AI and other computing advances improves our hurricane models further.
Unfortunately for us, one other thing to consider here is that the societal benefits of this improvement are probably somewhat of a counter-factual. You might ask, for example, what is the evidence this improvement in forecasting is doing anything for keeping damage costs down, when it seems like every year the damage estimates get worse and worse? While this is true, there's several reasons for this:
First, inflation. Higher building costs each year is going to make damage estimates go up, all other things being equal.
Third, and probably most importantly, more and more people continue to move, populate and prop-pulate (a word I made up to describe more property development) the parts of the country that are getting impacted by these storms. Thus, more people and more property on a per square mile basis just makes it that much more likely that for each storm, on an apples-to-apples basis strength wise, property and total damages are going to increase.
The counter-factual point here is that absent these better forecasts, things almost assuredly would be worse than they've turned out. This is, of course, difficult if not impossible to prove, but it seems far more likely than not.
For those interested in this stuff, whether out of sheer curiosity, or because you have loved ones or property in the southeast, the National Oceanic and Atmospheric Association National Hurricane Center website has a ton of super interesting data on this stuff. You can go look at hurricane summaries as far back as 1995, get initial damage estimates, track data, etc. The National Hurricane Center Hurricane Tracker App is also awesome, free, and has the added benefit of not giving you all the fear that you'll get from watching TV (having watched more of the Weather Channel in the last two days than I have in the last two years, it's very clear how much they hedge against providing good news or good updates compared to providing bad ones). Additionally, that app it will give you all of the updates from the NOAA almost instantly from when they come out, and again, devoid of opinion.
Best of luck to those impacted.
Comments