News & Blogs
Hurricane Spaghetti Models: Four Things You Need to Know to Track Storms Like the Pros
Published: August 6, 2019
There's a delicious-sounding term that's about to make its way back into the weather forecasting lexicon as hurricane season ramps up, but it has nothing to do with food.
Spaghetti weather models, also known as spaghetti plots, are a simplistic way of conveying a lot of tropical information quickly, but there can also be downfalls to relying on these plots.
(EXPLAINED: What is the Cone of Uncertainty?
1. Spaghetti Plots Do Not Portray Any Impacts
Although most models show possible impacts, to present many models succinctly on a single chart, meteorologists generally produce spaghetti plots that usually only show the “where” and a loose representation of “when” for tropical systems.
To get to this level of brevity, meteorologists must only focus on the center point of a tropical system, which may or may not be accurate. We’ll get to more on that limitation later, but for now, let’s focus on the lack of impacts.
These plots do not speak to whether a storm will bring rainfall, hurricane-force winds, surge, or other data; they just contain information about the center of a storm's future track.
There are a few cases where spaghetti models are essentially useless.
One instance is with a developing tropical system. Tropical storms in the end of their formative stage are often still trying to wrap thunderstorms around to their left-front side, especially if they are gaining latitude. This is typically the weakest side of a tropical storm since winds and forward speed are opposite.
(MORE: Hurricane Central
Throw in wind shear and/or dry air from one side of the system, and almost all of the impacts are felt on the other side of the storm and, sometimes, well away from some of those skinny strands of spaghetti that make up the spaghetti plot.
Now, put a landmass on the left side of that tropical storm. You'd probably think having a tropical storm 10 to 50 miles off the east coast of, say, Florida or the Carolinas would be a bad thing. But go back to the scenario above, and all of the thunderstorms and higher winds are now in the Atlantic, even with a storm very close to shore.
Did that strand of spaghetti really convey any useful information for anyone but, perhaps, the history books?
An additional limitation spaghetti models have is that they don't show any representation of intensity or size of a particular storm. These are represented on different charts, usually for individual storms.
2. Each Model Has a Slightly Different Purpose ... and You’re Probably Reading Them Wrong
Most models have the goal to be the very best, but each one has a different way of getting to that result.
Some weather models are built on statistics, some on atmospheric dynamics, others are built on other models and others yet are built entirely on climatology and persistence of the current atmosphere.
(MORE: What is a Tropical Wave?
Two of these models, called the CLP5 (the CLImatology and PERsistence model) and the XTRP (Extrapolated), seem to always get found on model plots, but neither contains any useful information about the forecast.
The CLP5 uses past weather situations, or analogs, to diagnose what similar storms have done in the past. A "bad model" is one that does worse than the CLP5. The XTRP simply extends the storm’s recent motion out to five days and is always a straight line. This batch of models is often called the pure statistical models.
|Model name||Model Type||Main Use|
|American GFS||Dynamical||Global Model|
|ECMWF or Euro||Dynamical||Global Model|
|UKMET or EGRR||Dynamical||Global Model|
|GEFS/AEMN||GFS Ensemble/Consensus||Global Models/|
Estimate of Model Confidence
Estimate of Model Confidence
|TVC#||Track Consensus||Estimate of Forecast Confidence|
|ICON||Intensity Consensus||Estimate of Forecast Confidence|
|TABs||Trajectory||Estimate of Shear|
Some models just follow the winds, and they are collectively called the TABs (or Trajectory and Beta models).
These three models — shallow, medium and deep — are slightly more useful because the closer they are together, they indicate that there is less wind shear in the atmosphere. On the contrary, if they are spread out, this is indicative that there's more wind shear and the system will likely stay weak. A weak system should not be monitored using the deep version of the TABs — called the TABD — since those systems do not usually tap the upper portions of the atmosphere.
The statistical-dynamical weather models are a little more complex. These models combine statistics such as storm location, time of year and what hurricanes of the past have done with simple dynamics such as steering flow. This suite includes the SHIPS and LGEM models, which are largely intensity models.
The most complex are the dynamical weather models, which take into account the current state of the atmosphere using observations from the ground, ocean and air, as well as complex physics equations, to forecast the atmosphere. This suite of models includes the American Global Forecast System (GFS), and the hurricane models (HWRF and HMON), among many others.
Which model was the best in 2018? According to the National Hurricane Center, a corrected version of the European model did the best in track forecasts while corrected versions of the GFS and HWRF weren't far behind in the short term while UKMET performed well beyond 3 days. Models that use weighted averages of several corrected global scale and statistical models did even better than the Euro by itself. Two weighted averages that did particularly well were the HCCA and the TVCA.
In the intensity department, a blend of other intensity models called the IVCN model, the Florida State Superensemble, a corrected ensemble of several other models, and the HCCA did the best in 2018. The HWRF and GFS did the bet as far as individual models go.
One major advantage spaghetti models have is when most of the models overlap, this is a big confidence booster for forecasters because most of the models have the same idea, even if they are getting to it different ways.
Another confidence booster is consistency between forecast model runs. When numerous runs show similar ideas and stay consistent with those ideas, it can be helpful for forecasters. When models change from run to run, this means that either the atmosphere is changing or the model does not have a good idea about what's happening, and it is usually the latter.
Models usually run every six hours.
3. Forecast Models Are Limited By Human Imagination and Bounded By Weather Data
In many cases, an educated imagination comes into play when picking a starting point for these spaghetti models. These cases include the formative stages of tropical cyclones that incorporate invests, tropical depressions and tropical storms, where picking out the center of circulation — the point where models must latch onto — can be difficult.
The image below, for instance, shows the model track forecasts for July 2016's Invest 97L. Half of the problem here is that we included both "early" and "late" models in the graphic. By early and late, we are talking about how early or late models run respective to when the National Hurricane Center produces their official updates.
(MORE: What Is an Invest?
The other half of the problem is that even within one batch of models (i.e. early vs. late or a single model run many times, called ensembles), the origin points are not always the same. Look at the big variation in where the green models (AP## or GEFS) begin. In a case where this is close to land, that can mean the difference between having a tropical system over land or in the water, which can have drastic repercussions as little as 12 hours into the future.
Another case where forecasts may not be as good is over the open ocean, since the amount of land-based and even ocean-based observations drop.
The model is usually most accurate at the point of origin, and model accuracy decreases over time. Without this point being accurate, the repercussions end up being a rather inaccurate model.
Over the years, the amount of data going into our models has continued to grow in order to make them more accurate. Of course, bad data, such as a bad point of origin, depletes this accuracy.
What do we do to fix this? In short, we make more data. When a tropical system threatens, the Hurricane Hunters fly into the storm, more weather balloons are released and satellites are turned on rapid-scan mode to collect as much information as possible.
4. Looking at Ensembles May Be the Way to Go, Especially Days in Advance
There is also a second flavor of models that can be especially helpful 3-7 days in advance called an "ensemble."
Think for a second about a musical orchestra with dozens of musicians. This orchestra represents the entire suite of musical opportunities can take the audience in one direction or another even as some instruments move up-tempo or down a note or two.
This is analogous to the entire suite of models that we as meteorologists have to come up with a forecast, often shown in the typical spaghetti plots. This suite can be full of more than 50 weather models with varying levels of correctness and experience.
But let's just back into the orchestra with only with the flutes this time. Again, each one should sound roughly the same for the big performance, but each one will actually sound ever so slightly different based on the instrument itself and the experience of the musician playing.
This is roughly analogous to an ensemble suite of one model. The most well-known models – the Euro, GFS, Canadian, and others – all have ensembles. An ensemble is a collection of forecasts all valid at the same forecast time.
For instance, the GFS is run 21 times with slightly varying initial conditions and physics to get the Global Ensemble Forecast System (GEFS). The GEFS's members are expected to vary somewhat due to their differences in how they are started and run.
Ensembles should be leaned on in the medium to long-term forecast realm to see all of the possibilities for a given period.
Ensemble systems can be helpful in multiple ways.
Firstly, if these ensembles are tightly packed close together in 3 to 7 days, the confidence in a forecast is higher, but it still should be checked against other ensembles like the European or Canadian. Remember that each ensemble member is still buying into the main member's ideas, and it will go roughly where that main member goes.
Secondly, if a model's ensemble is tightly packed but still diverges from other models like the Euro or the hurricane models, it could be either very arrogant or likely to be correct. Figuring out which of these possibilities is correct comes with forecaster experience.
Finally, if this ensemble's members are spread apart within two to four days, you know that model has less confidence or that the overall forecast is a highly uncertain forecast.
The Weather Company’s primary journalistic mission is to report on breaking weather news, the environment and the importance of science to our lives. This story does not necessarily represent the position of our parent company, IBM.