Thursday, November 12, 2015

DIY Forecast Tools, Part I

I know this blog is read by numerous DIY weather forecasters and others looking for the next big storm.  I thought I would take a few moments today to discuss some of the unique products that we have on the Utah Weather Center web site ( that might be useful for their efforts. 

1. NAEFS-Experimental Precipitation Forecasts

 NAEFS is an acronym for the North American Ensemble Forecast System, a combination of medium-range ensemble forecasts produced by the US National Weather Service and the Meteorological Service of Canada.  Ensembles are great for weather forecasting, but the NAEFS lacks sufficient resolution to capture the detailed effects of the mountains of western North America.  

The NAEFS-experimental precipitation forecasts takes forecasts from the NAEFS and downscales them to higher resolution.  The downscaling method is quite simple and based on monthly high-resolution precipitation climatologies for the western U.S.  The good news here is that the resulting forecasts are much better than those produced directly by the NAEFS.  The bad news is that this method does not account for other unresolved effects (e.g., lake effect) and when you downscale using climatological precipitation-altitude relationships, you don't account for situations that depart dramatically from those relationships.  For example, frontal passages feature weaker enhancement over the mountains compared to climatology, whereas some pre- and post-frontal periods can produce much larger enhancement.

Despite these warts, the NAEFS products we produce are quite useful.  We generate plots of 7-day downscaled quantitative precipitation forecasts (water equivalent) and snowfall for several regions.  Below is an example for northern Utah.  

We also do plumes for many locations including Alta-Collins, Canyons (mid mountain), Park City Base, Brighton, Sundance, Snowbasin, etc.  Below is the latest for the base of Park City. 

2. HRRR forecasts

The HRRR is the High-Resolution Rapid Refresh, a 3-km model run hourly at the National Centers for Environmental Prediction that is used for short-range (< 18 hour) forecasting.  It can be quite useful for anticipating the timing of frontal passages, whether or not snow will develop in a few hours, etc.  We provide plots of (1) surface wind and simulated radar reflectivity and (2) surface wind and 1-h accumulated precipitation.  Below is a forecast of the former for later this afternoon showing just a few showers along the UT-ID border.

Because I like to look at long loops of past data for context, the loops on the Utah Weather Center include a large number of analysis frames showing the past weather.  I suppose that chews up a lot of data on your mobile phone plan, but hey, you get what you pay for!

3. NAM-4km/GFS-13-km

For those of you who think resolution is everything, we provide forecasts from the high-resolution NAM nest, which has 4-km grid spacing compared to the regular 12-km NAM.  We also provide high resolution forecasts from the GFS at 13-km grid spacing, which is as close to native resolution as you can get. 

I don't use these much for precipitation because they have an overforecast problem over the mountains.  They always forecast large precipitation amounts and about 1 in 10 times they get it right.  Unfortunately, they give you a false alarm the other 9 out of 10 times and false alarms are extremely damaging for ones meteorological reputation!

4. NCAR 3-km Ensemble (Coming Soon)

The National Center for Atmospheric Research (NCAR) runs a 10 member high-resolution (3-km grid spacing) ensemble once a day.  It produces forecasts out to 48 hours.  We're working with it offline right now and hope to put it up on the Utah Weather Center sometime this winter.  I don't know how skillful the forecasts are, but the horizontal plots sure look awesome.  Here's an example from a couple of days ago.  

We've been working on plume and box-and-whisker plots to indicate accumulated precipitation and periods when heavier precipitation rates are likely.  

5. Acknowledgements

Special thanks to Trevor Alcott, Tom Gowan, NCEP, and NCAR for their coding expertise or models!  

I'll try and cover some more tools in a future post as I couldn't get to them all today.  


  1. I've been hoping you would make a post like this for a long time now, thank you!

  2. really interesting stuff! I am very much an amateur when it comes to meteorology and find it fascinating and entertaining to try to predict snowfall.

    Could you explain what the resolution means for the different models. I have a loose understanding I think, but when you say, "which is as close to native resolution as you can get." I don't know what that means, which makes me realize that maybe I don't understand it as well as I think. Thanks!

    1. Numerical models can't solve the equations governing the atmosphere directly. They need to use approximations, sometimes referred to as numerics. The easiest to conceptualize is a grid point model where calculations are done at grid points that spaced some distance apart. The smaller that distance, the higher the resolution of the model, meaning it can resolve smaller scale atmospheric features.

      Native resolution means that you are looking at data on the model grid. Close to native means you are looking at a grid that is very close to the native grid.

      The GFS is a bit more complicated because it uses a spectral model and not all calculations are done on a grid. In the interest of time, "close to native resolution" is a way of saying you're getting as good as you can get.


  3. Great news about the NCAR 3-km Ensemble that is coming! I have yet to find anything that can forecast local thermal winds in Utah. I really wish the NWS or the U of U would run a 2km WRF-GFS.

  4. Thanks much Jim! What I find most useful is your own experience and insights into the pros & cons of the various models. I knew that resolution is a problem for us in the Wasatch, but without your explanation I couldn't figure out why the NAM4k is almost always optimistically off (assuming one wants snow..), and the 12K that one would expect to average out the mtn range effects is often more reflective of actual snowfall.

    Again, very helpful to us snow hunters/hopers. Best regards.