As we work later into winter, the phrase “computer models” keeps getting thrown around on social media what seems like 24/7/365 anytime there’s snowstorms/ extreme cold. Our social media inboxes become flooded with messages saying “you said 9.3 inches for Chicago but I only got 6.1, what happened????” When you start to see these maps floating around online, what exactly are you looking at and where are they coming from?
What is a computer model?
A weather computer model is simply put (they are rather complex) a supercomputer that is “fed” current weather observations and then spits out a simulation of how it sees the weather playing out based on equations & parameters that the model uses (physical laws of motion/momentum…etc.).
Why do models differ?
If we (meteorologists) had 30 different models that were all “built” the same, all models would output the same thing- that’s not helpful at all. So, each model is “tweaked” in some way. Imagine 30 people baking a cake- they are all given the same general guidelines, but each person’s cake is going to be different. One may add a touch more sugar, another one may bake it for 5 more mins…etc. The same is true with models, each one varies slightly so that we can see different situations play out given slightly different initial parameters.
How do they differ?
The words “computer model” are rather vague in the world of weather because it really depends on the scale you are talking about. There are climate models that generate outputs for the overall state of the climate going hundreds of years out while there are also models that forecast the weather only for the US for as little as 18 hours out. Without going into too many details, some of the ways models differ include allowing ocean-atmosphere coupling, allowing for convection, and breaking the atmosphere into different amounts of “layers”. Some see the atmosphere as one big layer while others see tens of hundreds of layers.
How many models are there?
A lot! It depends on what you are looking at. If we are looking at forecasting over the next day or two, there’s “short range” models like the HRRR, NAM 12km, NAM 3km, RAP, NSSL, RGEM, ARW, NMMB, NBM, HRDPS (to name a few). If you are looking more “long term” (days 4-10), there are models such as GFS, ECMWF, ICON, UKMET, and GDPS- among others. There are also other “groups” of models such as seasonal models that forecast out months in advance. As you may expect, short range models are usually limited to forecasting a “small area” (the US) and cannot forecast that far out due to their high resolution. Seasonal models are used for global patters months down the road and cannot tell you what is going to happen tomorrow in Ohio.
How often do models run?
It depends on which model you are talking about. Short range models like the HRRR run every hour, while others like seasonal models run as little as once a month. The most used model guidance for your typical 7 day forecast *usually* runs 2 or 4 times every 24 hours.
So, which model is right?
There is no answer here- no model is “right”. Some models generally do better than others (ECMWF over the GFS), but that does not mean there aren’t times where the GFS (American) outperforms the ECMWF (Euro). While we can’t speak for all meteorologists, at BAMWX we look at all models and see which models are performing “best” vs the current conditions to give us an idea as to which model to base our forecasts on. There are plenty of days where we do not use the more reliable models because they are not performing as well as other models. How does a usually reliable model turn unreliable? It goes back to the “why do models differ?” section. Based on how each model is “built”, some are able to pick up on certain patterns better than others.
Why do snowfall totals from models seem to change so much in the winter?
To put it simply, precipitation totals and temps have a MAJOR impact on totals, so slight variations will have big impacts. In the summer, for most of us there is no noticeable difference whether we get 0.70 inches of rain or 1.0 inches of rain. It also doesn’t matter if its 78 or 85 when its raining, its going to come as rain. However, things are totally different in the winter. The difference between 0.70 inches of liquid vs 0.90 inches of liquid can be a difference of over 3+inches of snow. How so? Well, it depends on temps. For example, if we get 0.5″ of liquid that falls from the sky, this could equate to ~4-5 inches of snow if temps are closer to freezing vs ~9+ inches if temps are near 0. Same amount of liquid falling, but temps will greatly affect the snow ratio (light and fluffy vs heavy and wet). So, next time you see your county moved from 4-8″ of snow to 2-4″ of snow, it’s not because meteorologists love messing with totals (or your emotions)- it could literally be a model changing total precip by a tenth of an inch or temps going up/down a few degrees.
How accurate are models?
Once again, there is not a true answer to this. While certain aspects of certain models are tracked to see their performance, it is not like a test where the model is graded and then is given a score like “87% accurate”. One thing is certain, NO MODEL IS EVER 100% ACCURATE. Models will never have the ability to be 100% accurate due to the complexity of our atmosphere. Another thing we want to stress- MODELS ARE NEVER MEANT TO BE TAKEN VERBATIM. Why? Because models are simply simulations as to what could play out. In the image below, Cities A, B, C, and D will not see exactly 2.5, 4.2, 5.7, and 4.7 inches of snow, respectively. This should be used as a guideline, not an exact forecast. There are too many dynamics going on in the atmosphere and models do not have the capability to forecast down to a backyard- meaning there will not be one single model that nails every part of a forecast.
If models are not 100% accurate, why use them?
As mentioned, models are not meant to be used verbatim. If that was the case, meteorologists would not have a job. Again, we can’t speak for all meteorologists but we use this data as guidelines for a forecast, not just pulling data from one model and calling it a day. There is a lot more that goes into a forecast than just “model X has 3 inches and model Y has 7 inches so it’ll snow 5 inches”. Models are simply a simulation, not the exact scenario that will play out
Should I trust a snowfall map from a model I see online?
Dig a little into the map to see what exactly the map is showing. If a map is showing snowfall 7 days out, be very cautious because forecasts can change *a lot* over the course of 7 days. Below is an example, note how the model goes from widespread 8-15” for WI (bottom image) to 3-6” in just one model run!! If the map is for the next 12-48 hours, chances are the model is much more accurate than what it’s showing for 7 days out. Also, pay attention to the model being shown. Some models like the GFS like to overplay cold and snow in the extended range (5+ days) so totals are usually a lot higher than what is observed.
Make sure you are also checking who is posting the snowfall map! A Facebook/Twitter account with 85 followers that posted a snowfall map for the next 2 weeks is certainly NOT a reason to hop in the car and go stock up on bread and milk because one run of one model has 3 feet of snow for your city. If you are part of an operation that is impacted by weather in any way, check out BAMWX! We provide daily, state forecasts for Midwest states as well as long range forecasting across the county. We want to show you how much time and money you will save by having accurate, detailed weather information. Our clients range from landscape companies, MLB teams, commodity traders, snow removal companies, and many more! Inquire today: [email protected]