January 26, 2012

The Truth About the Pacific "Data Void"

How often do you hear statements like this?

Forecasting skill is bad on the West Coast because there are few weather observations over the Pacific.

We don't know what is coming from the west, so it is impossible to forecast in the Northwest.

In reality these statements are really not true.  We have a huge amount of information over the Pacific and that is probably the main reason why weather forecasts have improved so much over the past 30 years.

So what observing assets do we have out there?  Be ready to be impressed!

We start with weather buoys, as shown by this map (red diamonds indicate stations that have not reported during the past 8 hrs)


Buoys are good, but the problem is that they often are put out of commission with the first big storm.  The big ones are anchored, and a bunch of them just drift around. Here is what one of the fixed buoys looks like (46005, roughly 330 miles west of Aberdeen, WA).


Then there weather observations from commercial ships.  As part of the Volunteer Observing System (VOS), mariners take observations every six hours.  The light blue dots on this chart show where ships were reporting at one particular time:

For some reason there are less ship reports in the middle of night.  And as forecasts get better, the ships avoid the areas we really need data---in the middle and near major storms.  When forecasts were bad, ships would get trapped in dangerous conditions--now they can get out of the way.

   To forecast the atmosphere we need data aloft, since the atmosphere is completely 3D.  A relatively new source of information (few decades) is from commercial planes.   A number of wide-body aircraft provide important weather data (temperature and winds, mainly).  Most of this information is at flight level (typically 30-40 thousand feet), but we also get nice vertical soundings when they land and take off.  Here is an example for three-hours on January 7th of this year.  Lots of data for flights going to Hawaii and to Asia.  But there are holes...big ones.
Well, the sources shown above only represent a few percent of the observations we get.  The bulk are from weather satellites. For example, some satellites can figure out the winds aloft by tracking clouds at various levels.  Or they can sense variations in water vapor and track them.   We get tens of thousands of wind reports per day in this way.  Here is a small sample.   Hundreds of wind reports!


 But it gets better than that.  You remember Star Trek when they orbited a new planet and Mr. Spock scanned the atmosphere and told Captain Kirk about the structure and composition of the atmosphere below?   We can do that, perhaps even better than the folks on the U.S.S. Enterprise.
Getting ready to check the atmosphere of planet Beta-5.

Here is an example of a vertical sounding (temperature and humidity structure with height) at a location over the Pacific from a National Weather Service GOES satellite.



Satellites can do much more. By sending microwave radiation down to the ocean surface and measuring how much is scattered back, some satellite can tell us the wind speed and direction.  Don't believe me?  Check this out!


And there is MUCH, MUCH more.  The bottom line is that meteorologists now have massive amounts of information from around the globe and this information is used to create a 3D initialization for numerical weather prediction models.  We may have some holes and gaps at times, but we are no longer blind over the oceans and polar regions--there are huge amounts of data everywhere.

There is one figure that reveals the impact of all this data on forecasts so beautifully that I have to show you...check the figure below.  This tells you about forecast skill in the northern and southern hemisphere from 1981 to now.  It is about the skill at 500 hPA--around 18,000 ft, a good middle level.  The higher the number (anomaly correlation) the better--100 would be a perfect forecast. You see the forecasts for various forecast ranges (3, 5, 7, and 10 days, varying color).  For each color the top line in for the northern hemisphere and the bottom line for the southern hemisphere.

 It is clear that forecast skill for both hemisphere has greatly improved over time.

But there is something even more profound. Back in the 80s, the northern hemisphere, with more land area and far more observations, had much higher skill.  But that seemed to change in the 90s and the early part of this century.  Why?  The answer is the satellites.   During that period major advances in using satellite data in weather prediction were made.  Since there is about as much satellite data in the southern hemisphere as the northern, why should the northern hemisphere have better forecasts?

I was really stunned when I saw this figure for the first time.  Few graphs illustrate so concisely how far we have come.


With more satellite data sets coming on line during the next decade, and improvements in the technology of data assimilation, expect further improvements in weather forecasts.

18 comments:

  1. "Since there is about as much satellite data in the southern hemisphere as the northern, why should the northern hemisphere have better forecasts?". Uhh, land data disparity? Please tell me we haven't discounted land based sources entirely. I mean I've heard a lot of buzz about some new radar thingy somewhere on the coast. Guessing we have some other land based assets as well. Sorta just kidding, but (and tell me if I've missed something here), seems to be the obvious variable.

    ReplyDelete
  2. This is why I love this blog. I don't really care about forecasts, but the science of forecasting.

    I leave comments all the time, usually negative (devils advocate), because I agree with you 98% of the time.

    But....I've been noticing the awesome improvement of the models 72hr to 180 hrs out. Huge strides, kudos. But I can't say the same within 24hrs (relatively speaking). Its similar to the hurricane forecasting dilemma you posted.

    My life motto, applies to weather too. Get today correct before you worry about tomorrow.

    ReplyDelete
  3. There is definitely an east coast bias towards weather data. Maybe because we only have five large cities between three coastal states? I don't know but it does seem like we are a foot note when it comes to weather prediction.

    ReplyDelete
  4. Another great post! What is clear is the challenge of digesting all this data and plugging it into the models.

    ReplyDelete
  5. Cliff, 3 day forecasts are near 98% Can they get any closer to 100? Seems like there is room for improvement with the 5 and 10 day, but that the 3 day has just about maxed out due to technological/mathematical limitations and chaos.

    ReplyDelete
  6. Hey Cliff, it would be awesome if you posted links to any publicly available resources you talk about during your posts. I live on the east coast now (grew up in seattle), but like smokejumper, I read your blog for the meteorological explanations. I'm sure other enthusiasts like myself would love to be able to get to some of your source material as well if it's available on the web. Thanks!

    ReplyDelete
  7. The irony of this is that the closer you get to 100%, the greater the outrage will be when the forecast is wrong. Rather than be happy with how much better the forecasting is, people develop a sense of entitlement, as if someone must pay when we can't predict the weather, EXACTLY, 100% of the time.

    A prophet is never welcome in his own town, Cliff. ;-)

    ReplyDelete
  8. @Andrew: 3-day and shorter forecasts still do have room for improvement, as the recent forecast failure on the ice storm shows.

    ReplyDelete
  9. "For some reason there are less ship reports in the middle of night." HAHAHAA...having filled out countless VOS reports for NOAA, including being on board a ship that won an award for being the most reliable reporting station (S/S KAUAI on a Seattle->Oakland->Honolulu run, back in 1999).

    I can tell you making observations at night was an exercise in futility. You could get temp, dew point, rough wind speed/direction, and that's about it.

    ReplyDelete
  10. Awesome post. The only other good comparison I can think of is just watching a tv newscast from 1980 vs. now:
    http://www.youtube.com/watch?v=iBqI5GsldMQ

    ReplyDelete
  11. what does the langley hill radar add to information from satellites

    ReplyDelete
  12. Excellent explanation blog about why forecasts have improved. While I take issue with the ~97% accuracy for the Northern Hemisphere it does seem that forecasts are improving. I wish that NOAA would publish a post-mortem on daily forecasts with respect to accuracy. I live in Bellingham and the accuracy is about 75%.

    ReplyDelete
  13. Hi Cliff, While skill is much much better due to satellites, there are still patterns that are more skillful than others. This winter has been an example. Also, when you look at 500 mb skill across the hemisphere, you have to look at regional skill. I think you'll find the Pacific Northwest skills this year to be rather bad for several important forecast periods.

    This grading process, while it gives you some information, doesn't address regional problems nor does it address specific patterns that bring significant weather -- like to our area.

    ReplyDelete
  14. Chris, blackcap others regarding accuracy...

    Cliff posted a graphic for the improving 'anomaly correlation' for the ECMWF 500mb height--approaching 98% for a day three forecast. That is a far different thing than choosing freezing rain versus rain, or snow, predicting a likely amount, and assigning a probability to things down here at sea level. We don't live at 500mb. Of course the 500mb chart is a great forecast, and that the models have indeed improved a huge amount I don't dispute. But getting the forecast right for snow or freezing rain in the Western Washington lowlands is going to fall far short of 98% for a day three forecast. Obviously. Depending on what you measure, the error bars are pretty big. There is room for a lot of discussion regarding forecast error, accuracy vs. precision, probabilities and all of that.

    ReplyDelete
  15. A wealth of information.Great web-site.

    ReplyDelete
  16. This is fantastic Dr Mass...thank you for explaining this to us. And I thought I sounded so smart when I would spout off saying 'well, the reason why people get the forecasts wrong is due to lack of data'. In the future, I think I'll say the reasons people get forecasts wrong is that it's a really really hard problem. I think you'd agree? Thanks again for providing this for people.

    ReplyDelete
  17. I love how the Anomaly Correlation graph represents the change in several variables in a simple, easy to understand fashion. It reminds of me Minard's statistical graphic of Napoleon's Russian campaign: http://en.wikipedia.org/wiki/File:Minard.png
    (and yes, I'm a dork)

    ReplyDelete
  18. I've always wondered about this. THank you for the great explanation.

    ReplyDelete

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

Should I Let AI Write My Blogs?

I spend a lot of time thinking about and writing my blogs.   Could AI apps do the work for me?   Could any of you tell the difference? Well,...