Forecast Failure

Although forecasting skill is dramatically better than even 20 years ago, sometimes predictions are less skillful.

Such a situation occurred on Friday and Saturday over western Washington.

The National Weather Service forecast on FRIDAY MORNING for the high temperature that day in Seattle was for highs in the upper 70s to low 80s (see below).  For Saturday, mid-70s to lower 80s.

What were the actual highs at SeaTac airport?   73 on Friday and 70 on Saturday.

Ouch.   Major errors.

The NWS errors reflected problems with the numerical forecast models.  Here are the temperature forecasts made Thursday morning from the UW ensemble (many forecasts) prediction system. The solid black line is the average of many forecasts (which is usually very skillful), and the orange dots are the observed temperatures.

The forecasts were excellent on Thursday,  but FAR too warm on Friday and Saturday. Other modeling systems had similar errors.

So what went wrong?

The models did not have sufficient clouds, which are potent sources of cooling, since they reflect the sun's radiation back to space.

Here is the simulated (infrared) satellite image (from the forecast) on Saturday at 2 PM.   Just some scattered clouds over western Washington.

Reality?   A potent band of relatively thick clouds moving around the offshore low.  Not good.

On Friday afternoon at 2 PM, the model had a few areas of clouds.

Reality was a relatively thick band of clouds over western Washington.  Much more cooling.

Relatively narrow cloud bands like this are a substantial forecast challenge.  Models have to get their location and timing correct as they swing around the low.   

There are also issues of simulating them correctly, with model descriptions of cloud processes (called cloud microphysics) having substantial problems, something I am working on.

Plenty of work for meteorologists over the next few years to address these smaller-scale cloud physics issues.   Or we can hope Machine Learning solves it all.😊