Page 1 of 3 1 2 3 LastLast
Results 1 to 25 of 69
  1. #1
    Join Date
    Oct 2003
    Posts
    1,037

    Winter Science in the PNW

    I broke my back speedriding in Chamonix last spring. Prior to that event I had focused on improving weather forecasting for the avalanche and ski community, but had yet to coalesce the pieces into a unified product. That break gave me the time to do so. You can see the preliminary results here:

    http://winterscience.com/about/

    The UI is a bit thrown together, but the data being produced is very rich. It's a work in progress, but wanted to get it out there for anyone who is interested.

    Obligatory photo of my last run over in Chamonix before everything went south:

    Name:  747_10154166648147847_1404323974259694558_n.jpg
Views: 1665
Size:  137.3 KB

  2. #2
    Join Date
    Dec 2003
    Location
    Seattle
    Posts
    33,546
    Thanks Chris.
    Quote Originally Posted by Downbound Train View Post
    And there will come a day when our ancestors look back...........

  3. #3
    Join Date
    May 2009
    Location
    inpdx
    Posts
    20,197
    subscribed

    gonna read thru it later

  4. #4
    Join Date
    Jun 2010
    Location
    Powder Mountain
    Posts
    841
    Solid talk at NSAW this year. Way more pertinent than many of the others

  5. #5
    Join Date
    Aug 2011
    Location
    panhandle locdog
    Posts
    7,836
    Very cool. So what your projected opening for PNW ski areas?

  6. #6
    Join Date
    Aug 2008
    Location
    PNW
    Posts
    1,633
    nice. bookmarked

  7. #7
    Join Date
    Sep 2007
    Location
    vernon
    Posts
    2,977
    This is nice, was playing with it during the dump up here yesterday, the predictions were pretty much bang on with the timing and placement of precip.
    www.skevikskis.com Check em out!

  8. #8
    Join Date
    Feb 2008
    Location
    In a parallel universe
    Posts
    4,755
    This is cool, thank you.

    Hope your convalescence is or has been going well.

  9. #9
    Join Date
    Jun 2005
    Location
    Alpental, WA
    Posts
    221
    Very handy

  10. #10
    Join Date
    May 2007
    Location
    Mt. Baker
    Posts
    1,754
    Quote Originally Posted by BeardedClam View Post
    Solid talk at NSAW this year. Way more pertinent than many of the others
    Ditto! Stoked to see how this evolves.

  11. #11
    Join Date
    Sep 2016
    Posts
    141
    Great start! I like the simplicity of the site and it has a solid goal.

    But homie, something is dreadfully wrong with your WRF physics selections (or something else in your model setup; no information is given so I can't help you) if you are underpredicting snow water equivalent by the magnitudes you are over the past 24-72 hours (this is at 8:20pm PST 16 November 2016). 4.63" observed but 0.76" predicted, yikes! Timberline is even worse!

    Click image for larger version. 

Name:	Screen Shot 2016-11-16 at 8.11.24 PM.png 
Views:	153 
Size:	182.0 KB 
ID:	192736
    Click image for larger version. 

Name:	Screen Shot 2016-11-16 at 8.32.01 PM.png 
Views:	149 
Size:	214.4 KB 
ID:	192739Click image for larger version. 

Name:	Screen Shot 2016-11-16 at 8.32.18 PM.png 
Views:	156 
Size:	213.4 KB 
ID:	192740
    Last edited by 2_1_3; 11-16-2016 at 09:33 PM. Reason: Add a few more stations for WRF verification

  12. #12
    Join Date
    Oct 2003
    Posts
    1,037
    Quote Originally Posted by 2_1_3 View Post
    Great start! I like the simplicity of the site and it has a solid goal.

    But homie, something is dreadfully wrong with your WRF physics selections (or something else in your model setup; no information is given so I can't help you) if you are underpredicting snow water equivalent by the magnitudes you are over the past 24-72 hours (this is at 8:20pm PST 16 November 2016). 4.63" observed but 0.76" predicted, yikes! Timberline is even worse!
    Thanks for taking the time to give it a look. This is an example of bad UI and not a bad WRF. Those predictions are for the NEXT 72 hours, and the measurements are for the PREVIOUS 72 hours.

    There's quite a bit in between the WRF runs and the output on the page. It depends on the case but the machine is using over a million learned points from historical WRF runs to create each hourly forecast. I've focused most of my time on that process as it has not been done before. If you're a WRF person I'd be interested in talking more about methods to optimize the start. Interestingly enough if it's actually better for me to keep perpetuating small errors in the model itself as the machine has learned to forecast for them, changes require rerunning years of WRF sims for re-learning to take place.

  13. #13
    Join Date
    Nov 2003
    Location
    Portland
    Posts
    17,475
    Bueno!
    Damn shame, throwing away a perfectly good white boy like that

  14. #14
    Join Date
    Oct 2003
    Posts
    1,037
    Per the example here's the last 72 or so hours at Baker, Forecast vs what actually happened at the NWAC station. This one happens to be impossibly accurate, can't say it always is this good but it's promising.
    Click image for larger version. 

Name:	Dashboard 3.png 
Views:	134 
Size:	160.3 KB 
ID:	192743

  15. #15
    Join Date
    Nov 2014
    Posts
    344
    very cool. i'm fascinated by machine learning. is applying it to weather forecasting a pretty new thing? i wonder how well it will predict relatively rare anomalies, since i can imagine there might not be a ton of these in the training data set.

    may i ask what software/language you used to make the GUI?

  16. #16
    Join Date
    Sep 2016
    Posts
    141
    Word! Now that is some verification! These kinds of plots, with the next 72 hour forecast included, would be great to have on the site (perhaps instead of the big tables of values). As people see how well the model does overall they should be more inclined to utilize it and share it around or point out interesting events where it didn't verify as well.

    I'll email you about some other suggestions/thoughts that will hopefully help improve to the products. Keep up the great work!

    Quote Originally Posted by cmor View Post
    Per the example here's the last 72 or so hours at Baker, Forecast vs what actually happened at the NWAC station. This one happens to be impossibly accurate, can't say it always is this good but it's promising.
    Click image for larger version. 

Name:	Dashboard 3.png 
Views:	134 
Size:	160.3 KB 
ID:	192743

  17. #17
    Join Date
    Sep 2007
    Location
    vernon
    Posts
    2,977
    Question for you. On the forecasted accum precip map (west coast) what is the measurement? .01 to 6.0

    Thanks.
    www.skevikskis.com Check em out!

  18. #18
    Join Date
    Sep 2010
    Location
    Cascades
    Posts
    666
    Quote Originally Posted by 2_1_3 View Post
    Word! Now that is some verification! These kinds of plots, with the next 72 hour forecast included, would be great to have on the site (perhaps instead of the big tables of values). As people see how well the model does overall they should be more inclined to utilize it and share it around or point out interesting events where it didn't verify as well.

    I'll email you about some other suggestions/thoughts that will hopefully help improve to the products. Keep up the great work!
    I second this - That chart is very helpful! And clearly your model runs on a sub 12-hr basis (that chart looks like timesteps of 1hr?) - would it be possible to provide such data, rather than just 12,24,48 and 72 hrs?

    On the snow forecast page: does this take the results from your SWE forecast page and regress against some relationship for (forecasted) temperature to produce snowfall in inches? If so, these don't appear to line up....i.e. currently the Paradise 72 hr SWE forecast=0.97" (on average ~10-12" of snow) but the snow forecast shows 0. (The snow forecast button is also gone, but clicking the area where it should be still links, so perhaps this is broken).

    Also, wouldn't mind a few more graduations on the Wx Maps (more than just the min and max). Interpreting values between can be difficult. Also, would help clear up if the scale bar is linear (which I would assume).

    VERY COOL PRODUCT though! Excited to see where this goes.

  19. #19
    Join Date
    Sep 2010
    Location
    Cascades
    Posts
    666
    Quote Originally Posted by el hefe View Post
    Question for you. On the forecasted accum precip map (west coast) what is the measurement? .01 to 6.0

    Thanks.
    the scale bar indicates in. (inches)

  20. #20
    Join Date
    Jan 2006
    Location
    Vanity Fair
    Posts
    2,720
    Quote Originally Posted by TripleT View Post
    On the snow forecast page: does this take the results from your SWE forecast page and regress against some relationship for (forecasted) temperature to produce snowfall in inches? If so, these don't appear to line up....i.e. currently the Paradise 72 hr SWE forecast=0.97" (on average ~10-12" of snow) but the snow forecast shows 0.
    Where is this? I can't find forecasted new snow heights. In the forecast plots there is a grey "measured" button, which makes a block of shading appear when turned on, for past time steps. Is everything in the shaded block actually measured or is this the output of past model runs? Should there be an extra graph showing measured AWS data? Am I missing something here? (Also having a hard time with the weird units )

    There is a typo on the starting page, you are thanking the "Northest" Avalanche Center.

    Really, really cool site and project. Nice to see SNOWPACK output somewhere for everyone to look at.
    Ich bitte dich nur, weck mich nicht.

  21. #21
    Join Date
    Oct 2003
    Posts
    1,037
    Quote Originally Posted by raypruit View Post
    very cool. i'm fascinated by machine learning. is applying it to weather forecasting a pretty new thing? i wonder how well it will predict relatively rare anomalies, since i can imagine there might not be a ton of these in the training data set.

    may i ask what software/language you used to make the GUI?
    It's in Django with bootstrap.

    On the snow forecast page: does this take the results from your SWE forecast page and regress against some relationship for (forecasted) temperature to produce snowfall in inches? If so, these don't appear to line up....i.e. currently the Paradise 72 hr SWE forecast=0.97" (on average ~10-12" of snow) but the snow forecast shows 0. (The snow forecast button is also gone, but clicking the area where it should be still links, so perhaps this is broken).
    The snow forecast page has got some issues. It runs off the SNOWPACK model, which is producing accurate output for the available models on each forecast page. However, one of the data feeds in the raw output has some sort of error I've yet to completely troubleshoot. That's why I took the link down from the top bar, cus it's garbage. I've got a few open questions out to my swiss friends to get this sorted.

  22. #22
    Join Date
    May 2007
    Location
    Mt. Baker
    Posts
    1,754
    Quote Originally Posted by cmor View Post
    Thanks for taking the time to give it a look. This is an example of bad UI and not a bad WRF. Those predictions are for the NEXT 72 hours, and the measurements are for the PREVIOUS 72 hours.
    I actually like the UI. Everything I need in one chart.

  23. #23
    Join Date
    Jun 2006
    Location
    Seattle
    Posts
    1,991
    Fantastic work. I agree with Gunder, everything needed in one chart.

    One wish...add Mt Bachelor to the available areas.

  24. #24
    Join Date
    Oct 2003
    Posts
    1,037
    Thanks for the feedback. I did reverse the order of the measured SWE vs the FX swe, hopefully that reduces the confusion there. I also redid the table as I thought it was hard to read.

    Quote Originally Posted by puma View Post
    Fantastic work. I agree with Gunder, everything needed in one chart.

    One wish...add Mt Bachelor to the available areas.
    I hope to eventually expand to a wider network, but right now it's just NWAC sites. I think SNOTEL would be the next logical move, but it's really work intensive to put all the parts together.

  25. #25
    Join Date
    May 2009
    Location
    inpdx
    Posts
    20,197
    I find it very hard to read. Enough so that I have a hard time getting what it's supposed to show.

    (others seem to be reacting well to it, so maybe it's just me...i'm no scientist)

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •