Tag Archives: Forecast Post-Mortem


An interesting snowstorm, with some aspects forecast correctly and others not.

On the plus side—

  • Precipitation amounts (QPF) were reasonably accurate and timing was pretty good.
  • Snow accumulations are always difficult to predict in March, but there were some official reports of snow totals in the 9-12 inch range at the upper end.  (Official NWS map shows totals were closer to forecast than I realized.)
  • Temperatures were accurately predicted at 32-33.

Interestingly, the Kuchera algorithm NAM snow totals and axis from last night was probably the most accurate!

What was off on the forecast—

  • The heavy snow axis was closer to I-95, not the far north and west as originally thought.  (This was captured by the NAM last night.)
  • Total snowfall was less than the 10:1 (16 inches max prediction)
  • Winds were much lighter (luckily) than had been expected.

I’m ready for spring!



The snow totals exceeded everyone’s expectations, including my own.   So what happened?

I can only speak for myself— I was very reluctant to predict snow totals based only on the raw model numbers for a storm in March.  Indeed, I felt I over-estimated last night when I said 2-4 inches and reneged this morning.

But if I had gone by the raw model numbers, the forecast would have been more on target.   So I can’t say the models got it wrong, except for perhaps surface temperatures, which showed to be colder than forecast. They apparently didn’t get the ‘dynamic cooling’ factored in at the surface.

Indeed, as said in my blog post yesterday, if this had been  a storm in January, I would have predicted 7+ inches.   But, I felt that warm surface temperatures and radiant energy through clouds would severely reduce accumulations and over-ride the raw numbers.

But the raw model numbers (at least based on the NAM FOUS) had enough QPF and low enough temperatures to account for what we have received.   So the models did well, based on my usual criteria.  I just didn’t trust it explicitly for the forecast.

I use the NAM FOUS data, an odd, tabular set of numbers that often provides everything I need for a good snow forecast; I rarely use the preconfigured snow totals of some of the model outputs.  

BTW, the NAM from this afternoon shows snow ending shortly after 7PM, although current radar would suggest otherwise.   We’ll see.



We had about 2.5 inches snow just northwest of the city.  I had predicted zero to 1 around here, so what happened?

First, it was always what I called a boundary conditions forecast, since the forecast parameters for snow vs. sleet were so on the borderline of different outcomes.   I wrestled with the forecast for days.

Here’s what I got wrong- temperatures at two critical levels of the atmosphere below 7000 feet were forecast to stay at or below freezing.  I ignored these temperatures, since the other critical level of the atmosphere, “the thickness level 1000-500 mb”, (a measure of the average temperature below 18,000 feet) has always served me well and it was expected to warm up.  It didn’t.

It turns out the earlier NAM’s predicted 500 mb thickness level was wrong.   It had been forecast earlier in the day to rise to 5440 meters by 7 PM.  (Too warm for snow usually.)

However,  the 7 PM  NAM initial conditions showed the thickness level to be 5420 and not increasing.  (Snow vs rain occurs if  the thickness level is at or below 5400 in our area.)    So the mid levels of the atmosphere never warmed as expected.

This “critical thickness” level error was made by both the GFS and NAM models.   Since the “predicted thickness” level was so close to the “critical thickness”,  any error was going to blow the forecast.

Indeed, the NWS had to issue a Winter Storm Warning last night as late as 6:30 PM.  So they were also caught by surprise.

That wasn’t all that went wrong with the models forecast-  All models, including the National Blend of Models had the storm intensifying and lasting through at least midnight.  The NBM had the chance of precip of near 100%  at midnight in PHL.  But the storm was a faster mover, didn’t develop as expected, and was long out of here by 10 PM!

So that’s what happened — Parameters on the borderline weren’t accurately predicted and the storm speed and development wasn’t modeled properly!

Hey, bad modeling by engineers designing  bridges causes them collapse;  this bad forecast is just going to just melt away today!