Disturbance off of Africa now #AL92; how to think about #Hurricane track models

The disturbance off of Africa is now organized enough for the computer models to start to get a lock on it, and it has been assigned the temporary ID of AL922023 (as noted in the previous post, you may see it as AL92 or 92L). It’s still not looking like much on satellite, but the main global models show it spinning up into a tropical cyclone of some kind early next week. NHC give it a 70% chance as of this morning at 8am.

Now that the models are running tracks, you know what that means: spaghetti time! (Of course, everyone knows that the spaghetti harvest is in spring, as this classic BBC documentary from 1957 describes.) Here is the current (Saturday morning) map with a few key models highlighted …

Click to embiggen any image …

For being a disturbance that isn’t really a storm yet, this is actually a very consistent picture. Part of the reason for that is the “steering” forces (middle to upper level winds) are fairly strong and well behaved right now.

What follows is a longer discussion of track models, what to look for, and a bit about how they work in the forecast process. If all you care about is planning your next week, you can stop here, and check back in a couple days to see if anything interesting has happened, but if you’d like to know more read on!

Before discussing track models and forecasts, I want to make one thing very clear: for the vast majority of people and applications, the only track and forecast that matters is the official forecast from the National Hurricane Center and the subsequent local forecasts from your local weather service office (for the US at this link, click on the map for your local info) and, if actions are called for, local emergency management. If you want to understand what goes in to forecasts or find tracking interesting, fine, but unless you are one of the limited number of actual professional tropical cyclone specialists in the world (and that doesn’t include the TV and internet forecasters for the most part – not criticizing here, but this is specialized work), you shouldn’t be trying to sort out this stuff and make decisions based on the raw track models (in other words, don’t eat the raw spaghetti). So be careful who is cooking your spaghetti, and who is serving it as well! If you use an app or source of some kind make sure the data is timely, and not being filtered or distorted in order to keep you engaged and clicking. I’m suspicious of most apps, even ones that claim they are “free” really aren’t, they are tracking you and selling your data for marketing purposes. Just use the NHC/NWS web sites. They don’t need to track you (the CIA does it for them 😛 ).

In this case, AL922023 does not have an official forecast track – just the general area in which the storm may form and move as shown on the seven day outlook. And since there are no “magic words” nothing to do or worry about as of yet.

So, you’ve been warned, and understand what follows is for educational purposes. Great! Tropical cyclone research and science is really interesting stuff. It is first of all important to understand the various classes of models. The simplest are the “climatology and persistence” models. One of the oldest and most famous is actually called “CLIPER,” for CLImatology and PERsistance. Despite being simple, these models actually do a fair job out to three days or so in many cases. They also don’t need very much information, in some cases only the current location and where the storm was 12 hours ago. This map shows four of these models:

The cyan line is persistence. In other words, if the storm keeps going in a straight line at the same speed it has been moving over the last 12 hours, this is where it will go. The yellow line is from a model known as HURAN, an acronym from HURricane ANalogs. This was one of the first statistical models used by NHC. While it is no longer in use at NHC, I run an updated version of it “in-house” on storms around the world so that I have a single global reference model. The greenish line is the CLIPER model run by NHC. This is their reference model for computing forecast accuracy. Again, the only thing CLIPER “knows” is where the storm is, where it has been, and how past storms behaved (computed from a regression model). From a forecast standpoint, if you can’t do better than that you are wasting your time. The dark line is a modified version of CLIPER that NHC runs, the Trajectory CLIPER.

Statistical/Persistence models are a good reference as to how hard the storm track is to forecast. If the more sophisticated models are close to these tracks, that means the storm is probably being “well behaved” – in other words, it is acting like past storms. If they deviate, it is a good indication there is something “different” about his particular storm.

The second class of models are those based on the global dynamical weather models that are used in every day forecasting. The US model is known as the Global Forecast System (GFS). For historical reasons it is labeled “AVNO” on most track maps (it used to be known as the global Aviation model). The European Meteorological Centre’s Integrated Forecast System model is commonly called the European Model (EMC), likewise, there are Canadian (CMC) models, for Pacific storms the Japan Aviation model is used, and so forth. These models are complex dynamical models that, in the past, were too coarse to fully resolve hurricanes. However, over the last 20 years, they have become detailed enough to become fairly good at simulating hurricanes. In addition, these models can be run in “ensemble mode” to create probabilistic models. Here is what this morning’s GFS run looks like.

The primary, “deterministic” run is the blue line. This is the run based on the “best estimate” of where the storm is, and what the atmosphere around it is like. The “cloud” of thin gray lines are various probabilistic runs. These runs are based on trying different initial positions for the storm – especially for a weak, disorganized system, we may not really know where the center of the storm is! Other variables are upper level winds, etc. The orangish line is the mean of the probabilistic runs. By looking at the details of these runs one can figure out what the storm track is most sensitive to, and how good the “primary” run is . HOWEVER these are not “stand alone” tracks – they can only be evaluated in the context of the primary run! So here is a key warning: not all lines on these spaghetti maps are of equal weight or validity!

By comparison, here is what the Canadian model shows for AL92:

The third major class of models are detailed, dedicated, hurricane specific models such as HWRF and successors. I’ll discuss them when we have a more mature storm with some examples to look at, since this post is already too long …

In terms of overall accuracy, each year NHC produces a report discussing how accurate the official and model forecasts are. Generally the official subjective forecast is the best – and is very rarely “worst”, which is something people don’t think about. A good, even great model can get spoofed and create a monumentally bad forecast – humans can often catch those kinds of mistakes. Human interpretation is still needed – so best to stick with the official forecast. But the track and intensity models are the “raw material” that a forecaster uses to create their forecast, and understanding them can help you understand what the forecaster is facing in terms of uncertainty.

Leave a Reply