Ball State UniversityLocal Weather Alerts
There are currently no active weather alerts.

Model uncertainty, and how it affects a forecast

If you’ve explored the weather side of social media today, you would’ve likely seen some news about severe weather potential across parts of the Mississippi Valley, with Ball State included, Tuesday and/or Wednesday of next week. However, when the event is not expected for another 3+ days, a lot of uncertainty is in the air. The Storm Prediction Center has a Slight risk (level 2 of 5) for both Tuesday and Wednesday, but not a lot of uncertainty with the system.

But, what is uncertainty in the weather world, and how does that pertain to a forecast?

Model uncertainty
Uncertainty in models refer the difference of outcomes over a certain time. For example, one piece of data from a model brings thunderstorm chances at some time, but another model is completely dry for that same time period. The discontinuity there is the uncertainty of the thunderstorm potential.

What are models?
Models in the weather world don’t use a catwalk. They are data outputs from various sources using different formulas and historical data. They drive almost every weather forecast that you see. Most common models are the European Centre for Medium-Range Forecasts (ECMWF, or Euro), Global Forecast System (GFS), North American Model (NAM), Rapid Refresh (RAP), and so many more models! They use current surface and upper-air data to build an extended model forecast output. As time goes on in a model forecast, data between other models don’t line up as well, and become noticeably different. The outcome becomes less clear. The image below shows GDP projection. Although not weather related, it is a model that shows GDP projection, but the upper and lower bounds grow as time goes on. For weather, model data even a couple days out could be very uncertain and have multiple different outcomes.

Source: statmodeling.stat.columbia.edu

So if models are uncertain, meaning that output data from different models are showing different results for a certain time, how would that mess up a forecast? Well, if a ton of models are all in disagreement, our forecasts that are made risk being inaccurate as time gets closer to the event. If x, y, and z are all different results, and a forecaster choose x, there is still the possibility of y or z happening. If y or z actually occur, then the result the forecaster chose becomes incorrect. When forecasting, it’s always important to take uncertainty into account, because being accurate is vital!

If you ever wonder why winter weather can be so hard to pinpoint, this is due to model uncertainty, as snow relies on where there will be enough moisture available, along with other things, and most often coincides with a weather system. The track of the weather system is a defining factor of snowfall for most of our snow, so any movement of a system will have an impact. If model data shows different tracks, it will show different snowfall amounts across an area expecting snow.

What type of weather events come with uncertainty? Well, everything! It all depends on how well models mend together. For example, the severe weather potential for February 27-28 this upcoming week has some uncertainty regarding where the low pressure center will be, when exactly this system pushes through, and if the atmospheric dynamics will actually be favorable for severe weather, and if so, what hazards are likely? This is where ensemble forecasts come in.

Ensemble Forecast Models
Ensemble forecast models are certainly beneficial to the weather world. Ensemble models use the same models as mentioned before, but they spice things up a bit, changing the formulas ever-so-slightly, or they change the surface temperatures, etc. and shows all of the different results of that, or averages it into one graphic. Whenever you see the “spaghetti tracks” for hurricanes, that is from ensemble forecasts. Especially in the winter time, ensemble forecasts are great at showing the average expected outcome, and also shows the expected bounds of the event, which is key. Ensembles show the uncertainty by showing the bounds of such an event.
Another example: We believe that there will be thunderstorms on Tuesday next week, but we don’t know where exactly the low pressure center will go, since models don’t have a specific point. This is where an ensemble forecast can help to show the potential low pressure track outcomes, so I can build my forecast around that and explain any uncertainties, which is key in material such as thunderstorm potential and accumulating snowfall.

However, ensemble products don’t always agree either, so mixing in the outcomes of an event, and keeping an eye on the weather forecast is important as time goes on. On the upcoming severe weather potential, ensemble models are disagreeing on the location and intensity of the system Tuesday afternoon, but they have a good idea of the placement of this system and when it moves through.

Source: pivotalweather.com

Model uncertainty is a nuisance, but knowing the potential outcomes of an event is important, which where ensembles shine. The farther away an event is, the more uncertain the model data will be. As time goes on, models do a much better job of agreeing. So when you hear uncertainty in the weather world, don’t think it is because we don’t know, but rather, the possibility of multiple different outcomes are on the table.

Cardinal Weather Service Forecaster Lance Huffman

Related Post