So, how good are forecast models at predicting snow?

Snowstorms in Minnesota are like a box of Cracker Jack. There's always a free surprise inside.

Monday's snow system was a good study in the state of the science of snowfall forecasting. Let's take a look at what the forecast models did well, and not so well with Monday's system.

Storm lead time: Forecast model grade = A

The lead time for Monday's snow event was excellent. Forecast models and forecasters were alerting to the likelihood of sloppy snow three to five days in advance. That's actionable information well in advance of snowfall.

Precipitation totals: Forecast model grade = B-

The forecast for the amount of liquid precipitation with the storm was good to just OK. Many forecast models painted a large zone of about .75 inch to 1 inch of liquid with this system.

Here's what actually fell.

• St. Cloud 1.01 inches

• Chanhassen .71 inch

• MSP Airport .48 inch

So the notion of several inches of snow was well forecast for much of Minnesota, but overall precipitation amounts in the central and eastern Twin Cities fell short of model output.

Precipitation phase: Forecast model grade = C

In my analysis forecast model performance for precipitation phase was "just OK." The biggest uncertainty going into Monday's storm was where the rain snow line would set up, and the resulting precipitation type around the Twin Cities. That was well communicated by most forecasters. A layer of above freezing air kept precipitation-type rain to sleet in the Twin Cities through most of Monday morning. That kept snow totals lower in the downtowns.

Precipitation phase and precisely where the infamous "rain-snow line" will set up is often the most important element of a storm. It's also the most difficult to forecast with precision.

Precipitation Timing: Forecast model grade = C

As the storm approached most models ramped up precipitation Monday morning. That notion was correct, but again the precipitation type meant accumulating snow didn't arrive in most areas until Monday afternoon.

Experienced forecasters have an old saying: "Never rush a good snowstorm." Forecast models often bring snow in too early one to three days in advance. The last few model runs before the snow arrives often lock onto the slower arrival of snow systems.

Snowfall Totals: Forecast model grade = B-

In the days leading up to Monday's storm, most forecast models painted a large zone of potential 4- to 9-inch snowfall (with locally higher totals) across much of Minnesota and the Twin Cities.

Here's NOAA's GFS model forecast snowfall from the 12Z (6 a.m.) run Sunday.

While the overall coverage and notion of heavy snow was good, localized variations in snowfall were not well predicted by most forecast models.

Take a look at the actual snow totals. There were actually two different snowfall maxima. One was north, and the other just south of the greater Twin Cities.

The two snowfall maxima is interesting and somewhat unusual.

A 5- to 10-inch snowfall swath set up north of the Twin Cities in a zone from Detroit Lakes through Brainerd and Hinckley. A second snowfall zone of 5 inches to 8 inches set up in the southwest Twin Cities down to Mankato and Rochester.

In between there was a swath of lower snowfall totals around 3 inches to 4 inches favoring the central and eastern Twin Cities.

So the overall notion of significant to heavy snow was good, but most forecast models I saw did not pick up on the zone of lesser snow in the middle. And of course that developed right over parts of the highly populated Twin Cities.

State of the science

Monday's storm was a good example of why we give "snowfall ranges" for approaching snowstorms. The state of the forecast models and snowfall forecasting science is very good at giving us ample lead-time notice of an inbound snow system days in advance. But the models still struggle at times with the often highly complex temperature profile within the lowest mile of the atmosphere. Rain-snow lines continue to be difficult to pinpoint, and that leads to wide variations in eventual snowfall totals.

Numerical weather forecast models are getting better. We can tell you it's going to snow with high confidence with several days lead-time. We can tell you it will likely be plowable and impact your commute. But it's still nearly impossible to pinpoint snowfall down to the inch in your backyard a day or two in advance.