Friday, November 12, 2021

New GFS Derived Guidance for Upper Little Cottonwood

I've been posting some new GFS-derived forecast guidance products for upper Little Cottonwood Canyon in posts over the past few weeks and just added a link for them to weather.utah.edu.  You can find them in the top navigation bar by clicking on LCC GUIDANCE (NEW).  The old guidance is still available there too, but eventually I'll get rid of it.  

The old guidance includes tabular data from the 12-km NAM and the 3-km NAM, but these modeling systems are aging, so I'm not spending any more time on them.  I'm going to start deemphasizing them for the most part.  weather.utah.edu is also horribly old and ancient, but I simply don't have the time to rewrite everything, so we'll keep limping along.  

The new LCC Guidance page is pretty basic and includes forecast meteograms of several variables at the top and a window with scrollable tabular output.  Yeah, it's not mobile friendly, but you get what you pay for here at the Wasatch Weather Weenies :-).  

Here's a summary of the products available and how they are produced:

  • Mt. Baldy Temperature and Dewpoint: Based on a straightforward machine learning approach based on two cool seasons (October to April) of training with past GFS forecasts (multiple levels and variables) and Mt. Baldy observations.  Given that temperature and dewpoint vary diurnally, we have unique equations for each model initialization time and forecast hour, the latter at 3-h intervals.  For hourly forecasts, we use the GFS forecast valid at that time, but the equation for the nearest 3-h forecast hour.
  • Mt. Baldy Wind Speed, Wind Gust, and Direction: Same as above, but for wind.  
  • Alta-Collins Snow Ratio: This one is more difficult because a two year record doesn't have a lot of snowfall events, even at Alta.  Thus, we train over a longer period of time not to GFS forecasts, but to atmospheric analyses and then apply to the GFS. 
  • Alta-Collins 1H/3H/Accumulated Precipitation (a.k.a. QPF): Comes directly from the nearest GFS grid point to Alta.  No downscaling is applied.  I've elected not to apply downscaling because the GFS tends to be a fairly wet model despite not fully resolving the terrain.  So far this year, for larger storms, it has held up pretty well.
  • Alta-Collins 1H/3H/Accumulated Snow: Based on hourly or 3 hourly totals multiplied by the snow ratio at similar intervals.  
  • Wet-bulb zero level: The wet-bulb temperature is the temperature the air would have if it were cooled through evaporation to saturation (i.e., 100% relative humidity).  Typically the snow level is a bit below this level.  We calculate this level based on temperature and humidity at the GFS gridpoint nearest to Salt Lake City.  The reason for this if if we use the gridpoint nearest to Alta, we can't determine the wet bulb zero level when it is below about 8000 feet.  Thus, we use an upstream profile.  
These techniques produce better average errors than using the GFS directly for temperature, dewpoint, and wind on Mt. Baldy.  The snow-to-liquid ratio also produces better average errors than any other product we've compared to, including that used for the old Little Cottonwood product.  However, no forecast is perfect, and here are a few biases I've either noticed or I think may be evident moving forward:
  • Afternoon temperatures may be too low in the shoulder seasons (October, March, April)
  • Wind speeds and gusts will be too low during high wind event periods (e.g., gusts > 80 mph) and forecasts at longer lead times will not stray too far from mean values.  There are ways around the latter, but the reality is that if you compare GFS wind speed forecasts for Mt. Baldy to observations at 168 hours it looks like buckshot.  You simply don't have a reliable forecast if you use a single model.  The best path forward is to use an ensemble, but we haven't had time to deal with this.  
  • You're not going to see many snow ratios < 7:1 (unless Collins is below the wet-bulb zero) or > 20:1.  This is a tradeoff for forecast reliability.  If we use a technique that will go for big SLR, like 30:1, there's going to be a lot of false alarms.
I hope you find this project useful.  For the latest, click here.  

Special thanks to past and current Steenburgh group members for their contributions to this effort.  

3 comments:

  1. I'm just a rando who reads your blog, so take this for whatever it's worth, but I love the weather.utah.edu page despite (and possibly because of) its horrible age and ancientness. Thanks for providing this information!

    ReplyDelete
  2. Is the GFS really more reliable than the 12km NAM? I know from your own research that 3km suffers many terrain induced biases, but I thought the 12km NAM was a near second to the HRRR for mountain QPF.

    ReplyDelete
    Replies
    1. I said they were aging, which means that the weather service is no longer doing major upgrades. They will eventually be replaced. While the NAM is still useful, I need to put my time and energy into the future.

      Delete