June 05, 2015

When forecasts disagree, who do you trust?

This weekend is going to be a warm and pleasant over western Washington State--you can  take that to the bank.   But the forecasts of various media outlets are not the same.

Let me show you, looking at one location, Seattle-Tacoma Airport, and the forecasts available at 8 PM Thursday.

The Seattle Times is going for 84F on Saturday, 87F on Sunday, and 81F on Monday



The National Weather Service?  83F on Saturday, 84 on Sunday and 78 on Monday

The Weather Channel?  Cooler on Saturday  (81),  with 83 and 80 on Sunday and Monday


In contrast, KING-TV is the same:

Accuweather is relatively cool (82F) on Sunday.

So the forecasts for Saturday range from 81 to 84F.  Sunday ranges from 82 to 87, and Monday from 78 to 81F.     This is a pretty typical range for such forecasts.

So noting that there are differences, who delivers the best forecast?   And why are they different?

One resource to check out is a web site:  ForecastAdvisor.com.   This compare the forecasts of a number of different outlets.   Can I guarantee the results from this website.  No.   But their results are qualitatively consistent with my experience.

So here are their results for last  and this year for Sea-Tac   Meteogroup, a European-based firm, does very well, but so does the Weather Channel and the firm they own:  the Weather Underground.   I often check Weather Channel predictions when I don't have the time to do it myself.  The National Weather Service is substantially less accurate.

Other locations?  Here is their analysis for Spokane and Yakima....quite similar.  Accuweather does decently as well.


Why does a group like the Weather Channel do so well?   Because they have a more sophisticated approach to weather forecasting than the National Weather Service.  But the National Weather Service has one big advantage that sometimes is important:  human forecasters.

All forecasts are based on numerical weather prediction models, with names such as the GFS, UKMET, ECMWF, WRF, and NMMB.  These models start with an observational description of the atmosphere and use the equations describing atmospheric physics to predict the future.

The National Weather Service runs several weather forecasting models, does simply statistic postprocessing to their output  (called MOS, Model Output Statistics), and gives the results to human forecasters, who actually make the predictions that are distributed.  Generally, US forecasters rely on a limited number of forecast models.

In contrast, the Weather Channel has pretty much taken humans out of the loop.   They take the forecasts from many modeling systems and then correct typical errors or biases based on past model performance.  Then they combine these corrected forecasts statistically in an optimal way, weighting the forecasts based on their contribution to making the best forecast.   It is very hard for humans to beat such a sophisticated combination of the best models and statistics, although there are situations when they can.

National Weather Service forecasters can sometimes beat automated (or "objective") forecasts during situations when many of the models fail, when something very unusual happens, when forecasts need to be rapidly updated, or when there are very local effects the models can't simulate and the statistics can't correct for.   Such situations are admittedly unusual, but can be critically important.  National Weather Service forecasters are stationed around the country and thus become experts in their local weather.    They can also consult and advice local governments and businesses--something the national forecast companies do not do.


And I should make something clear.   The commercial weather industry, like the Weather Channel, DEPEND on National Weather Service models and observational data.  To put it another way, the National Weather Service provides the essential weather infrastructure for the private sector..




6 comments:

  1. http://forecast.io/lines/

    What do you think of that page?

    ReplyDelete
  2. I admire the weather channel- but seriously Dr. Mass- those folks in Atlanta are NOT skilled when it comes to the unique features of our weather like the convergence zone or Hood Canal snow, just as two examples. They are not nearly as accurate as the Weather Service in these situations, because as you say, the Weather Service and its forecasters are experts for the local area.
    I can handle the weather service being a few degrees off in temperature prediction on a sunny day, but I cannot handle the weather channel predicting 38 degrees with rain when its 30 degrees and I have 6 inches of snow...

    ReplyDelete
  3. Basically, Cliff, I read your blog and listen to you every Friday. I sometimes look at other sources, mostly NWS here at home and Weather Channel when traveling, but you are my main source.

    ReplyDelete
  4. Thanks for making that last point clear, "The commercial weather industry, like the Weather Channel, DEPEND on National Weather Service models and observational data. To put it another way, the National Weather Service provides the essential weather infrastructure for the private sector."

    ReplyDelete
  5. I have been disappointed with the NWS precip forecasting for Port Angeles. I expect a few surprises as we are not far from the nebulous western edge of the rain shadow. But there is a dependable pattern in the 6-day forecast, of order-of-magnitude overstatement of precip amounts for several days, revised much lower in the last day or two, or not corrected at all. Been seeing this for several years, and it's worse in the dry season. However NWS is usually much closer on temp, wind, etc., than any of the commercial providers.

    ReplyDelete
  6. To answer your question: I trust the hotter forecast. Temps are normally warmer than predicted. In fact, we sometimes break record high temps on days when nobody mentions the possibility.
    It's now 87 degrees. "warm and pleasant"? Not by me.

    ReplyDelete

Please make sure your comments are civil. Name calling and personal attacks are not appropriate.