Hurricane Forecasting: Part 2 – Dumpster Diving, Squashed Spiders, & The Flying Spaghetti Monster

This is the second installment of a look at hurricane forecasting by Ryan Wade from UAHuntsville. The first part is here on the blog at: Hurricane Forecasting: Part 1.

Hurricane Forecasting: Part 2 – Dumpster Diving, Squashed Spiders, & The Flying Spaghetti Monster

When it comes to hurricanes, it seems like everyone becomes a meteorologist with their own opinion on where the storm will make landfall and how fast the maximum winds will be. No other weather forecasts are as closely watched by the general public and come under as much scrutiny as hurricane forecasts (except maybe a 10 year-old watching the weatherman forecast a snow day). We live in a glorious information age where detailed weather observations/radar/satellite/models/forecasts are available at our fingertips like never before in history. There was a time not so long ago when budding meteorologists (meteorology students and/or weather hobbyists) would comb through the trash cans at a local National Weather Service Office (with permission of course) trying to get a glimpse of Hurricane Hunter data from inside the eye, a grainy satellite photo, or even a forecast made by a primitive computer model. Now all it takes for someone to make a forecast is a personal computer and an internet connection.

Which is how it should be.

The general public should be educated users of weather information & forecasts, but at the same time should also be skeptical consumers. I’m NOT advocating that someone should ignore forecasts/warnings made by the local broadcast meteorologist, the national weather service, or private weather companies. Rather, John Q. Public should have a basic understanding about some of the limitations of the state of the forecasting science, as well as some of the benefits & pitfalls of the tools meteorologists use to produce those forecasts. Let’s explore some of the computer models used in hurricane forecasting.

Computer models first ingest weather observations and satellite data before using the equations of motion of the atmosphere to step ahead through time to make a forecast. Some of the computer models used in hurricane forecasting are designed only to simulate/forecast the global weather patterns, some are designed to use climatology/statistics, while others attempt to simulate the hurricane itself and how it interacts with the larger weather pattern. The National Hurricane Center has an excellent summary of many of these computer models here:

http://www.nhc.noaa.gov/modelsummary.shtml

You might be asking yourself: “How do I know which model to use?”

The answer is: “It’s complicated and it just depends.”

-It depends on which model is best handling the overall weather pattern.
-It depends on whether the storm is weak/disorganized or intense.
-It depends on whether the storm is large or small.
-It depends on which model is best handling the intensity of the storm.
-It depends on whether you are making a short-term forecast (next 24 hrs) or a long-term forecast (3-5 days).
-It depends, it depends, it depends.

Spaghetti with Skepticism

The spaghetti plots showing several computer model forecasts on one image really got thrust into the public eye during the record-breaking 2005 hurricane season (see satellite montage image near the top). These plots can be quite useful in determining whether or not there is good agreement between several computer model forecast tracks. A forecaster must first have a good handle on the current structure/intensity of a tropical cyclone, understand the overall weather pattern, and have a good working knowledge of the different models used on these spaghetti plots. If you remember in Hurricane Forecasting: Part 1 we discussed that you cannot separate track and intensity. Well, the BAMS (shallow BAM) uses steering much lower in altitude than the BAMM (medium BAM) or the BAMD (deep BAM) to forecast the track of a storm. The GFS may have a better handle on the overall weather pattern than the CMC for a particular day. The HWRF may better simulate the internal structure/dynamics of a storm better than the GFDL. The CLIPER5 is based primarily on climatology, statistics, & persistence.

The problems with using these plots by themselves are 1) that we have no idea about the structure/intensity of the storm, 2) we have no idea about the overall weather pattern, and 3) each of these models are designed to simulate something different. Thus, the spaghetti plots should not be used by themselves! The spaghetti plots above for Tropical Storm Isaac should be used with extreme caution because Isaac is still a disorganized/weak system. While there may be good agreement between the models in the spaghetti plot, this agreement may just give a false sense of security because they may be agreeing for the wrong reasons. It is still too early to tell what the intensity will be in a few days (it’s battling dry air and it might be battling land/mountains as well), so we just don’t know which winds at what altitude will have the most impact on steering Isaac.

The two images above of Ophelia (2005) and Debby (2012) show that the spaghetti model plots of weak/sheared storms often look like “squashed spiders” with legs going all over the place. Q: How do you get any meaningful forecast information out of confusing plots such as these? A: By understanding the weather pattern, the structure of the storm, and each one of these models. For Debby, if you thought it would either be a weaker storm or that the high pressure ridge over the Southeast U.S. would build in quicker, then you go with the models forecasting a westward track into TX or LA. If you thought if would be a stronger storm or the low pressure trough would dig down into the Southeast U.S., then you would forecast a northeastward track into FL.

The spaghetti plots for Hurricane Katrina below (click the image for an animated loop) show very close agreement for the forecast track after Katrina entered the Gulf of Mexico. This is not true when Katrina was still a weak/developing system to the east of FL while over the Bahamas. In the animation you can clearly see that the models initially do not agree until Katrina becomes a well-developed/intense hurricane over the central Gulf of Mexico. The same reasoning applies here as I’ve described above: The models are still trying to figure out the steering currents for weak storms, thus the large spread in the forecast tracks. For strong/intense storms, the steering currents are generally much better know, thus the forecast tracks are generally much more accurate.

Use of the spaghetti plots by meteorologists and other forecasters have become almost a religion, like the Flying Spaghetti Monster. When it comes to computer generated forecasts, you can only have so much faith in the accuracy of these forecasts. It is imperative (as previously stated above) that forecasters have a thorough understanding of the structure/intensity of the storm, the overall weather pattern interacting with the storm, and a knowledge of the nuts and bolts of the models used in the spaghetti plots.

So, go out there and ask a meteorologist/forecaster to provide at least a little bit of an explanation/reasoning to go along with the spaghetti plots. If he/she knows what they are talking about, then they will be more than happy to provide that information to you!

Ryan Wade
PhD Candidate
Severe Weather & Radar Group
University of Alabama in Huntsville
Follow us on Twitter: @uahsevere