Monday, October 5, 2020

Improving Medium Range Forecasts

In late September, the National Weather Service upgraded its Global Ensemble Forecast System, GEFS, which should lead to improved medium range forecasts for ski season.  

The upgrades include the following:

  • GEFS is now based on the new "FV3" dynamical core
  • Horizontal grid spacing decreased from 33 to 25 km
  • Members increased from 21 to 31 members
  • Use of new techniques to produce initial conditions for the members
For more details, see this NOAA Press Release or this Service Change Announcement.  

Time will tell if these upgrades lead to significant improvements, but I'm optimistic they will help some.  The increase in membership and new techniques they are using to initialize those members should help improve the spread of the ensemble, providing better probabilities of precipitation thresholds.  

What we don't know is how these changes are going to affect GEFS biases at individual locations and, in turn, how those biases will affect our downscaled forecasts.  It will take a few storms (maybe more) to start getting a handle on this. 

We will have our first chance Sunday and Sunday night when the first major fall system of the season looks to push through northern Utah.  Our downscaled NAEFS product, which includes all 31 of the new GEFS members, indicates an ensemble mean of nearly an inch of water equivalent in "favored" Wasatch Mountain locations through 0000 UTC 12 October (6 PM MDT Sunday) and 6-12 inches of snow for upper elevations around the Cottonwoods.  

For Alta-Collins, the GEFS mean is just under 12 inches.  Curiously, that's quite a bit higher than the Canadian Ensemble mean, which is a bit unusual.  There are three major clusters of GEFS members, one that produces two inches or less, a second that produces about 6-12 inches, and a third that is in the 18-28 inch range (see lower-left panel).  One very excited member produces more than forty inches ;-).  

A big question that I have at this point is whether or not these forecasts exhibit more bias than they did last year.  Are they skewed toward higher amounts, for example, or is the model and downscaling reasonably calibrated.  Large spread in this instance doesn't surprise me based on the characteristics of the trough, which is digging and closing off over the western U.S.  Such situations typically exhibit significant uncertainties for local snowfall accumulations since much depends on if, where and how the low closes off.  

I look forward to cooler weather, but would rather have the snow hold off a bit longer, especially if it's only going to be a modest event.  If it goes big, I guess I'm more interested, but even in that situation, it will likely rot on north aspects and melt elsewhere, so let's save it until a bit later in the season.  

No comments:

Post a Comment