An inside look at choosing a snow forecast, what it means for the next storm

DETROIT – Detroit residents’ ears perk up when it snows in the future. It is then up to Local4Casters to provide top-notch weather forecasts. Computer models are an important forecast to analyze any prediction.

Since Tuesday evening, the snow is more than a day and a half. It may be soon for some. It’s an eternity when it comes to numerical models and their numerical predictions. Meteorologists use their training and experience to decipher the most relevant computer models to determine when snow will appear the day after tomorrow.

Related: Wind, flood warning, then winter storm watch in Metro Detroit: heavy snow ahead Thursday

Here’s a breakdown of this week’s computer models and the decision to opt for one over the other or combine elements from different models to help people in Southeast Michigan prepare without fear:

A d

A Tale of 2 Pattern Sets

Two computer models, the Global Environmental Multiscale System (GEM) model and the European Center for Medium-Range Weather Forecasts (ECMWF) model have a cold front developing and crossing the Motor City area Thursday morning. As it is, regular rain showers quickly turn into regular snow showers with little sleet or ice. Snow lasts much of the afternoon with this pattern as temperatures drop. Fast-flowing mercury and constantly falling snowflakes combine to provide a significant amount of snowfall for most of southeastern Michigan, from northern Oakland and Macomb counties south through Detroit and Ann Arbor to the Michigan-Ohio border. By the end of Thursday evening, these models predict 3 to 8 inches of snow, much of which will fall south of 8 Mile Road.

Two other computer models, the North American Mesoscale (NAM) model and the Global Forecast System (GFS) model have the same cold front but take its origin (called frontogenesis) and path in a very different way that results in results radically different snowy (or not so snowy). Both of these patterns have the cold front forming and remaining further west and north. It has the same drop in surface temperatures where people, cars, trucks, and houses are, but temperatures aloft (inside clouds) fluctuate near or above freezing. This results in a longer transition period from rain to snow on Thursday with a greater chance of freezing rain and ice pellets. This means the full transition to snow is occurring later than earlier in southeast Michigan. In fact, it snows earlier and longer, not just north of 8 Mile Road, but north of M-59 and in the Saginaw Valley. Thus, greater amounts of snowfall of 6 to 12 inches are possible in The Thumb and the Saginaw Valley according to these models. The forecast extension of wet and icy conditions reduces snow totals in southeast Michigan to 1 to 4 inches.

A d

The choice

Here’s what goes into choosing which models may make more sense at any given time.


In real estate, it’s location, location, location. In synoptic meteorology and forecasting its initialization, initialization, initialization. Initialization is the start time of any model when it is produced. The numerical models for the weather are a simulation. They simulate the weather in the future. At their initial point, they also simulate current conditions. For many weather forecasters, a key indicator of whether a weather model can make a good prediction is how well that model’s simulation of current conditions matches reality. The larger the match, the better this model can be at matching the future state of the weather.

Now, the GEM and ECMWF models start from environmental conditions that better match the field conditions at the time of their production. The NAM and GFS also don’t match.

A d


Consistency is another factor. Computer models are not produced once a day. Many are produced up to two, four or eight times a day. The High Resolution Rapid-Refresh (HRRR) pattern is produced hourly (up to 24 times per day). Each production is called a “run”.

IN BRIEF: A huge amount of computing power and time is required for each run of a weather model. Generally, the less frequent the run, the more a model can predict over time. Conversely, an hourly model like the HRRR cannot forecast as far back in time as the ECMWF, GEM, GFS or NAM models. Their races take place two to four times a day. Therefore, they can trade two to ten days in the future. The HRRR only looks 24-36 hours into the future and stops.

Usually, a weather model that produces similar forecasts for the same time period in each run can engender more confidence. In this week’s case, the four patterns and their progressive series have been pretty consistent. The ECMWF and GEM maintain higher snow totals farther south in southeast Michigan. The NAM and GFS maintain greater snow accumulations to the north and west of southeast Michigan.

A d

Conclusion and caveat

Therefore, with consistency being equal and initialization being the distinguishing factor, there is greater confidence in the GEM and ECMWF models which have a faster transition from rain to snow over Detroit and southeast Michigan. Thursday, resulting in an accumulation of 3 to 8 inches.

It is a conclusion but only for the moment. The caveat is that there is plenty of time until Thursday. The frontal system that will produce mild weather and rain turning to cold weather and snow and ice hasn’t even formed yet. Once this is done and as subsequent runs of the model are completed, a clearer picture is formed of when and how much snow will fall.

The latest model runs arrive when many are asleep, and we’ll be combing them through with a fine-tooth comb and having our latest predictions for you on TV and on our digital platforms before the alarm clock goes off.

So, as with all things media, stay tuned. Whatever the outcome, we’re here for you. We will provide the latest information with our latest forecast to help you and your family. Our goal is to make sure everyone is prepared and not afraid.

Copyright 2022 by WDIV ClickOnDetroit – All Rights Reserved.

Comments are closed.