Hurricane models and accuracy

Is there anywhere on the web that shows predicted vs. actual paths for hurricanes according to model? Can I go to a site that will show me what the European model predicted for the last 24, 48, 72 hrs. etc as compared for the actual path? It seems the European model predicted Florence to turn south before the others and still differs by enough to make a pretty big difference to those likely to be impacted, at least wind direction-wise. While it doesn’t much matter if your roof blows off from the front or back, tides can be greatly impacted by wind direction.

It doesn’t show actual vs predicted, but it does show all the different model paths, in different variations of predictive maps. I’m guessing there are no maps showing where it went in history. Weather people probably don’t like to brag how right or wrong they were.

It’s a very busy website. Right now it’s five times as busy.

I still haven’t figured out the abbreviations of the different models. The best guess is in the green lines. Beware the black triangle line (XTRP). It’s only the current direction the hurricane is pointing to. I first noticed this when the other models were going up the coast, and XTRP was headed to the west coast.

Well, if all else fails, you can take a screen shot every day and save them as image files. Then, for example, you can compare the one you took five days ago to the one that they are displaying today.

Sorry, that should be the European models are the green lines.

A little digging turned up an article from 2014 stating that the U.S. computers needed to be updated and that that was going to take place over the next two years. Any idea if that ever happened? The European model out performed the U.S. model during Sandy and highlighted the need. If I were a weather guy and my model was consistently accurate, I’d be bragging about it.

Do you have a link to that article? I’m curious to see exactly what it said.

It’s important to realize that upgrading the computers themselves (which would let the meteorologists run their models more quickly) are different from updating or enhancing the actual prediction models. The meteorologists who create these models regularly fine-tune them, using past results to learn from what they got right, and what they didn’t get right.

So, if the European model got Sandy right (or other models got it wrong), they learn from that, and use that knowledge to create what they hope will be more accurate models going forward.

Also, as I understand it, no model is consistently “the best.” Some seem to have a better handle on certain types of storms than others (and these models are use for far more than hurricane forecasts). And, weather forecasting is simply a hugely complex task – our models are far better than they were in the past, but even the better ones are still only approximations, and there are variables that even the most skilled meteorologists can’t estimate. Just because the European model may have gotten one storm “right,” doesn’t mean that it’s going to be the one that accurately predicts the next storm.

Here’s a Mashable article from last year on the topic:

What I take away from this:

  • Yes, NOAA’s primary model, GFS, is known to generally be inferior to models from non-US agencies, including the European model, particularly when it comes to longer-range forecasts, and they’re in the midst of a multi-year program to upgrade their hardware and their model.

  • Part of why the US models aren’t as good as the European model is that NOAA is working with lower levels of funding and smaller staff than the European agency has, and does not have a centralized agency for atmospheric / meteorological models.

  • Even with this general statement, GFS did well in predicting Harvey’s track, while both GFS and the European model did poorly on Irma.

  • This is why forecasters try to never rely on a single model, but instead, develop “ensemble” forecasts based on what all of the models are saying.

Here’s the one I was talking about. 2013, not 2014.

Given the vulnerability of our coasts, you’d think they throw some more money at forecasting.

Yes, you would absolutely think so. But, as the Mashable article notes, the administration is actually trying to cut the forecast model budget.

It’s also worth noting that these are all probabilistic models, so it’s like saying that your prediction that two dice rolled together would most likely add up to 7 was wrong: Well, no, adding up to 7 is the most likely result, and any model that said otherwise would be wrong, but that doesn’t mean that every dice roll will add up to 7. And if there’s a model that says that it will add up to 6, and the dice actually add up to 6, that’s probably not a matter that this model was more accurate. If you try to match it, you’ll just be wrong more often.

That’s the smart thing to do- it’s the meteorological modeling equivalent of crowdsourcing.

Another thing to consider is that these models are really only accurate in fairly short time frames- hence the “cone” getting wider for times further out.

And I’d wager that the models’ accuracy varies with the time frame as well- for a hypothetical example, the GFS model could possibly be more accurate between days 3 and 5 than the European model, and less accurate between days 0 and 3, and equal outside of 5.

Exactly so. That’s one of the reasons why those “spaghetti maps” can be misleading to the layman – the “center line” of any model’s forecast storm path is the equivalent of the “7” in what you describe. It’s the most likely path that the model is predicting, and there’s a fairly high confidence that the storm will take a path that’s reasonably close to that line (particularly in the near term), but the actual output of the model is more like a cone, as uncertainty in any model increases the further out that they attempt to model.

The other reason that some meteorologists hate the “spaghetti map” (and forecasters who focus overly much on where the center of the storm is going), is that it can lead laymen into a false sense of security if they see that the forecast doesn’t put the storm center over their heads. Yes, a direct hit by the center of a hurricane can be devastating, but hurricanes are massive storms, and deal lots of damage to areas that are nowhere near the center.

It’s not model by model comparison, but is related.

This shows how the NWS forecast track for Florence has changed over time. There are archives of other storms if you dig around.