Monday, February 8, 2010
Humility and Forecast Busts
Weather prediction is a business that promotes humility, and nothing brings one down to earth like a forecast failure--known as a "forecast bust" in the business. Major forecast failures are less frequent than even a decade ago...mainly due to a huge increase in satellite data and better computer models... but they still occur with discomforting frequency!
The most failure-prone forecasts are the most long-ranged ones...such as seasonal predictions done months in advance. Want a primo failure? In the fall of 2008, we were in a neutral ENSO (El Nino-Southern Oscillation) year and the long-range prediction was for a stormy, but normal winter. In December, there was an abrupt, unexpected turn to La Nina, with a change to cold and snow in the NW that took down our erstwhile mayor. Complete failure....embarrassing! (for both forecasters AND the mayor).
In contrast, this year everything went smoother. Moderate El Nino...with expectations in the NW for warmer and drier than normal conditions after January 1. Almost on schedule it turned warmer, but initially wetter. January was the warmest on record (much more than we expected). On the other hand the weather pattern of the last 3 weeks have been classic El Nino, with a split flow (jet stream going into AK and CA) and southern CA getting pummeled by strong storms. Passable grade?
Sometimes the most difficult forecasts are when there are weak systems floating around and the flow is modest. Like this weekend.
On Friday, both the NWS and I were going for essentially the same forecast. Mediocre Saturday as a weak system was coming in and a much better (dry, sun) on Sunday. The truth was almost reversed. Saturday became warm and sunny and Sunday was cloudy with rain in the morning.
Below are three computer forecasts for 24-h precipitation ending 4 PM on Sunday.
The first was made Friday morning. No rain except for the coast!
The second was made Saturday morning. Rain much more extensive.
The third was made Saturday night. Even more rain.
So you can understand how we went wrong on Friday. And you can see how the forecasts tended to get better as we got closer in time.
But my forecast failure was completely unnecessary (and yes so was the NWS's)
There is another way of forecasting...the right way...using a collection of forecasts..called ensembles. Run and use many forecasts all starting a bit differently or with slighly different model physics. If ALL of them are dry, then you go dry. If half of them are wet, 50% might be a good forecast. Etc.
Well, I sheepishly looked at the UW emsemble forecast system Sunday and sure enough some of the ensemble members on Friday were much wetter (see the ensemble member based on the Canadian model). And putting 8 members together the probability of precipitation for the two 12-h periods covering 5 PM on Saturday to 5 PM on Sunday (see graphics) had significant probablities of rain (at least 50%). So I SHOULD have forecast a significant chance of rain on Sunday. No more rushed forecasts for me. Until next time.
So the moral of the story...don't look at a single forecast model output. Look at many! Think probabilistically. Even though this technology is under development here at the UW, old habits are slow to die!
Posted by Cliff Mass at 8:39 PM