It looks like we’ll get a touch of snow on Friday afternoon/night – the WRF Model is calling for two inches, the Canadian Model is calling for one inch, and the American Model is calling for a dusting.
This weekend looks to be partly cloudy to mostly sunny with highs right around freezing. And we’ll have the usual Front Range patrol zone breeze, probably more so on Sunday than Saturday. The American Model is calling for an inch of snow on Sunday, but the Canadian Model barely even has clouds forecasted for Sunday.
Our next shot of snow looks to come in on Tuesday to Wednesday, with both the American and Canadian Models calling for an inch.
Note on Google’s GenCast AI Forecasting Model:
And a quick side note on GenCast as everyone seems to be talking about it in light of the recent article in Nature. Probabilistic weather forecasting with machine learning | Nature. For those who haven’t read the article, GenCast is a machine learning weather model made by Google, which Google claimed in a recent article in Nature beats the European Model 97.2% of the time in medium range forecasting. (The European Model is usually considered the gold standard in weather forecasting, though it certainly is not that great in our patrol zone for predicting snow, as we’ve seen from years of these forecasts.)
Before anyone gets too excited about GenCast and the recent press, a few words of caution. First, in the Nature article, GenCast’s ability to predict tropical cyclones was mentioned 23 times. Its ability to predict snowfall was not even mentioned once. Like many weather models, it seems to be primarily designed not for what we care about, i.e., snowfall, but rather (and understandably) for extreme weather events.
Second, GenCast’s grids are still very large – about the size of the American Model’s grids and are significantly larger than the Canadian Model’s grids. If anything, part of the issue with snow prediction is getting the subtilities of topography, which a ¼ degree latitude American Model grid obviously does not do. A 200-foot vertical rise can impact snow amounts, and so the 17 plus mile per side grid used by GenCast suffers from the same issues as other forecast models.
Third, the comparison in the study was about predicting 2019’s weather, which is obviously in the past. The Nature article unwittingly was following the Yogi Berra line: “It’s tough to make predictions, especially about the future.”
In the coming decades, I have no doubt that machine learning models will overtake and supplant our current numerical weather prediction models – and will produce far better forecasts in far less time. However, let’s all not get too excited that this will start happening tomorrow.
On a more practical point, it’s yet to be seen whether GenCast will be free. If it becomes free, and its data is easy enough to interpret, I’ll certainly try to add it to my forecasts, so we can all judge it for ourselves. Anyhow, that’s just my two cents, and I’m sure others on the patrol may have more insightful thoughts than me on GenCast.
Cheers.
-Jordan (Thursday 12/12 morning)
Geeky Notes:
References to the American Model are to the American (GFS) Model grid including Brainerd Lake with an average elevation of 9,439’. References to the Canadian Model are the Canadian (GDPS) Model grid including Brainerd Lake with an average elevation of 10,253’. References to the WRF Model are the CAIC WRF Hi-Res Model point forecast for Eldora Ski Area with an elevation of 9,189’. References to the European Model are to the European (ECMWF) Model on a point with my cursor at my best estimate of Eldora Ski Area. For big picture overviews, I tend to rely on the American Model, not because I think it is the most accurate, but because (i) it is free and (ii) I like its interface.
