Thoughts on Great Salt Lake Effect Forecasts from Computer Models

Great Salt Lake Effect forecasts from computer models have historically exhibited poor reliability for a few reasons.  

One is that many computer models have had insufficient resolution (or grid spacing) to adequately resolve the Great Salt Lake and surrounding mountain effects.  

Another is that the Great Salt Lake is simply very sensitive to small changes in the ambient flow and surface conditions, so when you look at a handful of model runs, they don't always capture the full range of possibilities.

Let's take a look at how these issues are manifest in current model runs, starting with the GFS.

The effective grid spacing of the GFS is 13 km.  At it's current level, the Great Salt Lake is about 25-km wide, meaning that in northwesterly flow, the GFS at best has two or three grid points that include lake characteristics.  In addition, the GFS is unable to distingish between the Stansbury, Oquirrh, and Wasatch Ranges, so it cannot separate mountain effects from lake effects.

Below is the GFS forecast for 1200 UTC (6 AM MDT) tomorrow.  The background lake and terrain are not from the GFS.  The lake is from an older, "good old days" lake level and the terrain is fairly realistic.  I use them for reference.  Note that the GFS is capable or producing lake-effect precipitation, but as an enormous blob covering a huge area.  While it is possible we will get lake effect, it's not going to look like this.  

The HRRR, on the other hand, has better resolution.  It has 3-km grid spacing and it better resolves the Great Salt Lake and various mountain ranges of northern Utah.  If we look at the HRRR forecast valid at the same time, we see considerably more structure due to this better resolution.  It produces a lake band that is more concentrated and it also produces orographic precipitation over the central Wasatch.  This is somewhat more realistic.  

That's good, but studies have shown that high resolution models still have problems. Several years ago John McMillen, a graduate student in my group, did retrospective forecasts of 19 lake-effect events using the Weather Research and Forecasting (WRF) model with 1.33 km grid spacing, even finer than the HRRR.  He found that those forecasts were "band happy."  While they produced banded forecasts of banded events, they also produced banded forecasts of non-banded events.  Basically, the model could not reliably distinguish events that were banded from non-banded.  So, when the HRRR generates a band, that's great, but you can't have a lot of confidence that Mother Nature will do the same.  

He also found that the bands were often in the wrong place and most commonly to the right (relative to the downstream flow direction) of the observed band position.  This is something that has also been identified in other model forecasts over other lakes, such as Lake Ontario.  

Another challenge for getting the lake-effect location right is sensitivity to the large scale flow.  The centers of the GFS and HRRR bands in the forecasts above are clearly different, but so are the flow fields.  Small changes in wind direction make a big difference, and current ensemble modeling systems can't paint out the full range of possibilities due either to insufficient resolution or too few members.  

So, many elements are in place for lake-effect tonight.  Many elements are also in place for orographic precipitation.  Either could materialize, but location and intensity are difficult to anticipate at this time.  

I look forward to seeing how this all plays out.