Major Forecast Failure

Weather prediction has become hugely more skillful during the past several decades, but there are still some failure modes.

This week in western Washington, we had a master class in forecast failure due to our local weather nemesis:  low clouds.

On Monday afternoon, the National Weather Service forecast for the high temperature in Seattle on Tuesday was 76F.   For Wednesday afternoon, 72F (see below).

The actual highs?  66 F and 57F.  

HUGE errors on the cold side (10F and 15F!)

Instead of a continued warm period, temperatures were well below normal (see a comparison to climatology below).  The brown color shows the normal range. 

The vaunted UW high-resolution ensemble of many model forecasts at Seattle (below) was highly accurate on Monday, but too cool on Tuesday and Wednesday (yellow dots are observations).

In contrast, forecasts of the National Weather Service and the UW model for the Washington Coast and Eastern Washington were excellent.

So what is going on?   

This event (and others like it) demonstrates a great weakness of most weather prediction systems:   an inability to predict or maintain a cool, shallow cloud layer.

Most models tend to produce too much low-level mixing, which destroys shallow, cold, saturated layers.

Let me show you the proof.  

There are visible satellite images for 2 PM on Tuesday and Wednesday.  Lots of low clouds over the western lowlands.  Clouds associated with cool surface temperatures.

In contrast, the predicted clouds, the UW WRF model, or the National Weather Service GFS models had no low clouds over western Washington.

Several of us are working on this problem, but in the meantime, either the National Weather Service forecasters need to more actively intervene in the forecasts, or we need to more effectively apply statistical corrections during these low-cloud periods.

This is a serious problem for Western Washington since low clouds are quite frequent during late spring and early summer.