Q&A of the Day – The Improvement in Hurricane Forecasting & Accuracy of Models
Each day I feature a listener question sent by one of these methods.
Email: brianmudd@iheartmedia.com
Gettr, Parler & Twitter: @brianmuddradio
iHeartRadio: Use the Talkback feature – the microphone button on our station’s page in the iHeart app.
Today’s entry: Hi Brian, I have a question which perhaps you could address in a future Q&A. It seems many forecasters have their own preferences for specific hurricane models which impacts the way they cover the threat from a potential hurricane, in this case Ian. Are there models that have consistently been the best?
Bottom Line: Yes, models for tropical systems are a lot like political polls. Subject to frequent change, but there are some that have proven to be more accurate than others. Before getting into the specifics about which models have performed best, here’s what’s changed as of the start of this year’s hurricane season.
- 2021 was a record year for hurricane forecast accuracy
- Today’s five-day forecasts from the National Hurricane Center are as accurate as the two-day forecasts 30 years ago
- Forecast accuracy has improved by 75% over the past 30 years
- Every forecast model has improved with time
If only political polls more closely resembled the progress with hurricane forecast modeling. Anyway, the point is there has been steady improvement generally with improvements in technology over time. That said, yes just as you often see vast differences with spaghetti models, there are vast differences in overall accuracy. Yale completed a study earlier this year on this topic and there’s a pretty clear picture of overall model performance.
Two models have consistently outperformed the others. The GFS, which is the NOAA’s Global Forecast System model and the Euro, which is the European Center for Medium-range Weather Forecasting global forecast model. To give you an idea of just how much variance there is in the success of the models. The performance of the GFS model has been about 10% more accurate than the performance of the European model and 48% better than even the fifth best model by way of accuracy. It helps paint the picture as to why getting carried away with a bunch of spaghetti models doesn’t make a whole lot of sense and why even looking at “most of the models” may be highly misleading as well. It’s frankly no surprise to see NOAA’s model is consistently a top performer and that’s why you’ll often hear that one cited above others. But there is one model which trumps them all, including the GFS model. It’s the National Hurricane Center’s model.
There’s a reason why the National Hurricane Center exists within the National Oceanic and Atmospheric Administration. The experts in the NHC are the best in the world at forecasting tropical systems. The reason the GFS model, which as mentioned is a NOAA model, isn’t the official NHC forecast, is due to the human factor of weighing models with other atmospheric data. What we’ve seen is that at all stages of storm forecasting the NHC consistently performs better than all models independently. Their expertise provides an official forecast path, or the cone, which is anywhere from 3% to 5% more accurate than the GFS model at all stages of forecasting. That’s why it’s unwise to engage in model speculation prior to an official NHC forecast having been released, as was commonly the case with what eventually became Ian. There’s also one other dynamic in play as well. The timeline.
Studies have shown reliability to be excellent two days out with current forecasting tools and decent five days in advance. There’s a reason why the National Hurricane Center doesn’t offer official projections beyond five days. Even with the improvement in technology and models over the years they’re not reliable enough to warrant consideration, yet last Thursday many were attempting to do just that when Ian was still just a disturbance. As always there are two sides to stories and one side to facts. The fact of the matter is the National Hurricane Center’s official guidance is more accurate than any of the models independently and their guidance remains the guidance we should pay attention to.