Dropout Forecasts

Every now and then, the forecast skill of a computer modeling system plummets, leading to a brief period of poor performance.  Such periods are sometimes referred to as dropout forecasts.

I've identified some examples in the plot below, which presents one measure of forecast skill, the standardized anomaly correlation of the 5-day (120-hour) sea level pressure forecast relative to the National Weather Service Weather Prediction Center sea level pressure analysis for the contiguous United States.  Without getting into details, the higher the standardized anomaly correlation, the better the forecast.  The graph presents the standardized anomaly correlation over the past 91 days from several models including the GFS (red circles), UK Met Office Model (brown diamonds), and ECMWF Integrated Forecast System (green Y).  

Source: NCEP

One can see that model skill varies daily, but there are times when one or more models suddenly produce much lower standardized anomaly correlations values.  These are the dropouts.  The UK Met Office Model seems to have done this several times during the period (I've identified five such forecasts in red circles).  Several model forecasts on the 19th of August were also quite poor and I've identified these with a purple circle.  

The causes of these dropouts is an important research topic.  They strongly impact forecast reliability.  If your forecast model gets it right 90% of the time, but it fails spectacularly the other 10%, it doesn't inspire confidence!

An example of a regional forecast dropout is about to unfold in the western U.S. over the next few days.  Below is the 180 hour GFS forecast from 12 UTC 11 October valid 0000 UTC 19 October (6 PM MDT Monday).  Ridging dominates the western U.S. with not a single drop of precipitation to be found in the western states (I've never considered Texas to be a western state, so ignore the little blob of green there).  There is evidence of a weak trough over the lower Colorado River Basin.    

A day later, only two of the 52 NAEFS members suggested any precipitation at Alta-Collins in the period around 0000-1200 UTC 19 October.  

Then the forecasts shifted.  The forecast from this morning, valid 0000 UTC 19 October, features a much deeper closed low centered right over Utah.  And there's precip in northwest Nevada, northern Utah, southeast Idaho, and western Wyoming!

If that verifies, we'll get a little mountains now out of the system, which would be a blessing, but for our area, this is a pretty big dropout.  No real hint of the closed low at 6-7 days lead time.  

An example of gremlins in the vast machine that forecasts our weather.