Wednesday, April 29, 2015

Don't Trust the GFS Precipitation in the Central Wasatch

This winter, NCEP upgraded the Global Forecast System (GFS), with improvements in both resolution and physics.  While it is likely improved forecasts of fronts, cyclones, and other major weather systems the precipitation forecasts over the central Wasatch are simply abysmal.

I noticed something was up immediately following the upgrade.  NCEP began to provide direct model output (known as BUFR) for the model grid point closest to Alta.  I began downloading and plotting this data [see (GFS at bottom) and] and noticed immediately that the GFS was generating far too much precipitation.

I then began downloading the native (13-km) resolution GFS grid from NCEP and examining the precipitation, which showed extreme amounts of precipitation over many mountain areas.  Here's an example from 23 January through 22 Feb (courtesy of a colleague).  During this period, the Alta-Collins and Alta-Top Cecret observing sites received a total of about 65 mm of precipitation (snow-water equivalent, grey and black lines)).  The ECMWF forecasts (blue) tracked this pretty well, but the GFS (maroon-I think!) produced nearly twice as much precipitation as was observed.  The NAM-4km (magenta) is even worse, but that's a story for another day.
On the plus side, NCEP smooths the GFS forecast to a 0.25ºx0.25º latitude–longitude grid before distributing it to NWS forecasters.  On this smoothed grid, the forecast looks better and closely tracks the ECMWF, falling just behind observed.
As we move into the spring, we're still seeing massive overprediction.  In the loop below you can see the problem.  Nearly nonstop precipitation over the central Wasatch and a few other mountain areas.  Note how these precipitation  bull's-eyes don't move and, like the Energizer Bunny, just keep going.

If we extract the time series for Alta (lower left), we see that the 84-hour forecast shown above (ending 6 PM Saturday) generates almost 1.5 inches of water equivalent.  If we look at the forecast through the middle of next week, Alta gets a whopping 4 inches of water equivalent.  Booyah!

We're entering into a spring pattern and we may see some rain or snow showers from time to time over the next few days, but barring the formation of an intense thunderstorm right over Alta, we're not going to see anything near 1.5 inches of water equivalent.

The bottom line is that the GFS precipitation forecasts are way overdoing precipitation in the central Wasatch, as well as some other mountain areas.  Let's hope this issue is addressed by NCEP in the future as the GFS would be even more valuable if it provide more reliable precipitation forecasts.


  1. It would be interesting to to see if the precipitation is originating from the large scale precipitation parameterization or if it is coming from the convective parameterization. To me, the former would imply that it was caused by the change in the dynamics (eulerian to semi-lagrangian), while the latter may be related to increased sensitivity of the convective scheme to the higher resolution grid spacing. Of course, these are just thoughts of a numerical weather prediction armchair quaterback, :)

  2. Unless the models get down below 10 km grid spacing (which will present its own issues), the models should predict lower totals than the site total at Alta for most events. The ECMWF time series looks reasonable for its resolution, but the 0.25 degree by 0.25 degree GFS total should probably be lower.