The purpose of this post is to give people a basic knowledge of how to read, interpret and use snowfall forecast maps from the various weather models. This includes basic rules of thumb, snowfall calculations and the advantages and disadvantages different models have when producing snowfall maps.
The first rule of thumb for model snow maps is to never take what they say verbatim. There will almost always be some major caveat to the maps that makes them consistently less reliable than almost every map produced by a human forecaster.
Most model snow maps are produced very simply; by taking the amount of snow/frozen QPF (forecast precipitation liquid equivalent) from the models and multiplying that by 10 for the snowfall. This is flawed for several reasons.
One of the most glaring reasons is that many times the snowfall does not come with these 10:1 ratios. If snow is falling when surface/low-level temperatures are near or above freezing, this will often result in lower ratios, which means that it takes more QPF to produce any given amount of snowfall compared to what the models show. When the atmosphere is fairly cold (say low to mid 20s or lower at the surface and aloft), the snowfall:liquid ratio often goes above 10:1, and can be as high as 30+:1 in extreme cases like in lake-effect snow. Very few model output sites go beyond this basic 10:1 ratio to calculate the model snowfall maps, but even after accounting for some of the inaccuracies, these model maps are still far from perfect.
Model snow maps also rely on how the models handle the boundary layer (near surface atmosphere) when temperatures may be/will be above freezing. Some models are better at handling this delicate boundary layer than others, and one is not always better than another in every situation. It takes a good understanding of the models and experience to determine which model may best handle the boundary layer conditions and when to favor one particular model over another, which is why defaulting to one model for its snow forecast can be dangerous.
One of the areas human forecasters will almost always beat the models is where the rain/snow/mix areas fall. Very small detail changes in the vertical temperature profile can mean big changes to the snowfall. The ECMWF has a particularly difficult time with this area, in which it highly favors snowfall when there is a layer of above freezing temperatures somewhere between the ground and the cloud tops. This often results in much greater snowfall in areas that may actually receive little to no snow! It is difficult to say just how the European model arrives at its snow/frozen QPF output since I am not fortunate enough to have that information on hand, but since most model sites just take that output and multiply it by the standard 10:1, it’s hard to go much deeper than what I have written.
While a certain amount of QPF may fall as snow, models also struggle to determine just how much of that snow will actually accumulate on the surface and not just melt away. In most cases, they do fine when the surface temperature is below freezing and has been for awhile, but they will often struggle with snowfall totals if the surface temperature is at or above freezing. It is entirely possible to get accumulating snow when it is above freezing, but the snow needs to be falling at a high enough rate to achieve it. If the ground is already wet before the snow starts, it will almost always take longer for snow to start accumulating. Some surfaces may see faster/earlier snow accumulation than others. If the ground is well above freezing, it will take longer for the ground to cool to a low enough point for the snow to accumulate, though this may not always be true when heavy snow is falling. The caveats in this paragraph is the other big area in which human forecasters can beat the model output.
Mesoscale events such as lake-effect snow are difficult for the global models like the GFS or ECMWF to accurately pick up on, meaning that their snowfall can be significantly lower than what actually happens since it cannot accurately resolve these smaller scale events. Higher resolution models are often better for forecasting lake-effect snow, but you should be weary when using these models past 48 hours as the models’ skill diminishes. These higher resolution models may also provide added value in areas that receive mesoscale banding within a synoptic system. While the location of the band of heavier snow may not be accurate, the presence of a band in the hi-res models that are portrayed as weaker or non-existent in the global models can be a good hint that some locally higher snowfall is likely.
Another thing is to make sure what kind of snowfall map you are looking at. While most sites offer how much QPF they think falls as snow, other maps feature how much snow the models think will actually be on the ground. When using the maps that show snow on the ground, if an area ends up with 20 inches of snow on the ground on the map, but the event starts with 10 inches on the ground, you must remember to subtract the pre-storm amount to get how much new snowfall the model actually expects to be on the ground. This can be more useful sometimes because at least some models/web sites do take melting/compaction into account with these maps. These maps are not terribly common compared to the typical snowfall forecast maps, but it is something to keep in mind so you are aware of exactly which kind of map you’re looking at.
While that was a bit of a long read, I hope that you will use all of this information when looking at a model snowfall map so you can better gauge just how much snow to actually expect in your area.