Jump to content
  • Member Statistics

    17,606
    Total Members
    7,904
    Most Online
    ArlyDude
    Newest Member
    ArlyDude
    Joined

NWS Product Debate


Deck Pic

Recommended Posts

ers - I just left my feedback on the NWS site.  One thought was to take the table of "chance of snow accumulation" and turn it into a graph.  So, for the last storm, you would probably have something like below (that isn't drawn by a 3 year old).  I think that would communicate exactly what you are going for, that the best chances are 6-9", with lower probabilities, but still a chance at higher or lower values.  You could do a point/click on a map, or on certain dots on a map, to make the graph.

 

attachicon.gifUntitled.png

 

I like this. Coming from a GIS background I love the point and click idea, even if it has to be county based to start. Personally, standard deviation is more useful than percentile.

 

I think the public gets confused with "percentage vs. percentile". Now, I know the product we're talking about here is mainly for EM customers and not the general public so they should know the difference. But with social media now and these images getting around well.. let's just say it has the potential to cause confusion.  If images were strictly made for the public however perhaps those 10,50,90 percentile maps should read something like "At least, Most Likely, At most". That way the map is explaining qualitatively the chances, instead of being purely quantitative. If I remember correctly, public dissemination of products should be at a 3rd or 4th grade level? If that's the case, then they are not going to get "percentage vs. percentile" but they will understand "at least, most likely, at most".

Link to comment
Share on other sites

  • Replies 247
  • Created
  • Last Reply

I like this. Coming from a GIS background I love the point and click idea, even if it has to be county based to start. Personally, standard deviation is more useful than percentile.

I think the public gets confused with "percentage vs. percentile". Now, I know the product we're talking about here is mainly for EM customers and not the general public so they should know the difference. But with social media now and these images getting around well.. let's just say it has the potential to cause confusion. If images were strictly made for the public however perhaps those 10,50,90 percentile maps should read something like "At least, Most Likely, At most". That way the map is explaining qualitatively the chances, instead of being purely quantitative. If I remember correctly, public dissemination of products should be at a 3rd or 4th grade level? If that's the case, then they are not going to get "percentage vs. percentile" but they will understand "at least, most likely, at most".

Good points. On the front page it does say... Min (expect at least this much), Max (prepare for possibility it could be this much) and Most Likely.

Link to comment
Share on other sites

When that sort of range pops up, it is an indication of

1. lack of model support; either the models are fluctuating from run to run or are not converging on a solution

2. there is enough precipitable water vapor forecast to move into the area and enough lift to ring out 16" but confidence is low because of volatility in the models.

3. good old fashioned synoptic meteorology is in cognitive dissonance with the models. For instance, a meteorologist may examine all the standard levels of the atmosphere and then evaluate which troughs and which ridges may strengthen/weaken; all this based upon what is taught in meteorology school and then reinforced through on the job experience. So what does a meteorologist do if they feel confident in their synoptic analysis and their analysis clashes with numerical modeling? The meteorologist calculates the minimum snow that the system can deliver and the maximum that the system can deliver. Such a range of potential snowfall doesn't imply that the meteorologist doesn't know what they are doing. It implies that at that particular time those predictions are conceivable. Note that forecaster has every intention of tightening the upper and lower limits as the storm approaches. Note that with the last system, LWX went with a minimum of 3.4" for DCA. Those meteorologists must have been particularly confident that 3.4" would be an achievable minimum. They must have been really confident.

Well...probabilistic forecasting allows the meteorologist to communication what he or she thinks will probably occur and also allows them to express their confidence level.

Maybe...LWX can put some thought into defining each winter storm event by the specific forecast lead times for that system. Simply put, some systems have model convergence way out in front and LWX could share with the public that it is appropriate for that particular storm to give a preliminary detailed forecast 36 hours before the start with fine tuning 24 hours out. Before anyone balks at the idea, meteorologists can state with conviction that for a well-modeled storm, the start of the art and science governs such a lead time. For a tricky system with poor model convergence, LWX reserves the right to state that they are watching the system closely and a draft forecast is available at 36 hours with a preliminary detailed forecast at 24 hours and fine tuning 16 hours before the start. If those lead times are state of the art, it is integrity in science to express those lead times to the public. People need to accept that some systems don't show their hand until 12 to 16 hours before the start, (and sometimes shorter).

Good Stuff!

Link to comment
Share on other sites

even though I think  the concerns raised about the hi/lo maps are valid, i am really a fan of probability forecasts, and I think movement in that direction is progress.

We've already discussed it ad nauseam, so prob not worth it at this pt. I prob shouldn't have posted it. They did well with the last one so perhaps this is a magical formula.. I still have some trouble believing that a goose egg is impossible IMBY but that's a debate for another time probably.

Link to comment
Share on other sites

We've already discussed it ad nauseam, so prob not worth it at this pt. I prob shouldn't have posted it. They did well with the last one so perhaps this is a magical formula.. I still have some trouble believing that a goose egg is impossible IMBY but that's a debate for another time probably.

It is worth bringing up and discussing but there is also a thread ( http://www.americanwx.com/bb/index.php/topic/43219-nws-product-debate/) for this if anyone wants to discuss further. I personally am enjoying the debate.

 

 

Edit: Not picking on anyone in particular (Ian) just chose the last in a long string of entries on the subject of LWX to highlight.

On to the discussion of white rain for Tuesday.

Link to comment
Share on other sites

It will take a long time for proper education to all sectors. Years

The use of the maps isn't really the debate here. Obviously they are going to be used. I don't personally believe that 0.7" is the absolute minimum "unlikely" scenario IMBY.. though with SREF backing off I suppose that will drop too now. Given the pt and click is 1-2 possible I'm not sure it makes sense there either. Same issue with the last storm even if it verified.  Clearly those numbers are very heavily based off computer guidance with likely minimal human interaction.  I know you say they get looked over but it's still unclear how much.  In a case like we see this week I'm not sure how anyone can say a minimum of 0 is not possible. 

Link to comment
Share on other sites

The use of the maps isn't really the debate here. Obviously they are going to be used. I don't personally believe that 0.7" is the absolute minimum "unlikely" scenario IMBY.. though with SREF backing off I suppose that will drop too now. Given the pt and click is 1-2 possible I'm not sure it makes sense there either. Same issue with the last storm even if it verified.  Clearly those numbers are very heavily based off computer guidance with likely minimal human interaction.  I know you say they get looked over but it's still unclear how much.  In a case like we see this week I'm not sure how anyone can say a minimum of 0 is not possible. 

Is the .7 inches the absolute minimum as per this tool, or the 10th percentile?  I may have misunderstood, I thought the two scenarios reflected the 10th and 90th percentile of possible snowfall.  

Link to comment
Share on other sites

Is the .7 inches the absolute minimum as per this tool, or the 10th percentile? I may have misunderstood, I thought the two scenarios reflected the 10th and 90th percentile of possible snowfall.

I think it's 10th percentile but it's sold as 'expect at least this much' or 'if the forecast fails you'll get this much' or something. If it's confusing to me I can only imagine how it is to people who don't follow weather as closely.
Link to comment
Share on other sites

Not sure you get the point. Maybe there is no point. I'm about to give up caring lol.

 

the way the product is designed, it is going to give a 10th percentile number that almost always will be too high.....esp from 48 hrs + out....I understand that giving planners a number of 0" every storm is not useful, but as it appears the NWS is advertising this product to the public, it will need some various tweaking.  Right now, I have a 60% chance of receiving 2"...that is obviously too high....problem with reliance on the SREFS or even any models, is that they often will advertise snow and then take it away....so the product will run counter to a good forecasting model which is to increase snow amounts as the event approaches....I like the ambition...It is experimental...We have covered a lot in this thread,....one thing I would add is these guys at LWX are talented mets..I'd like to see them have liberal opportunity to put their human imprint on them..this has been a good winter to have a bullish product....if we go back to an 8-12" winter next year, the product will not work as well as intended

Link to comment
Share on other sites

the way the product is designed, it is going to give a 10th percentile number that almost always will be too high.....esp from 48 hrs + out....I understand that giving planners a number of 0" every storm is not useful, but as it appears the NWS is advertising this product to the public, it will need some various tweaking.  Right now, I have a 60% chance of receiving 2"...that is obviously too high....problem with reliance on the SREFS or even any models, is that they often will advertise snow and then take it away....so the product will run counter to a good forecasting model which is to increase snow amounts as the event approaches....I like the ambition...It is experimental...We have covered a lot in this thread,....one thing I would add is these guys at LWX are talented mets..I'd like to see them have liberal opportunity to put their human imprint on them..this has been a good winter to have a bullish product....if we go back to an 8-12" winter next year, the product will not work as well as intended

Yes, I agree pretty much entirely. Everyone on social media is selling it as the absolute bottom of the potential which is incorrect.

Link to comment
Share on other sites

Yes, I agree pretty much entirely. Everyone on social media is selling it as the absolute bottom of the potential which is incorrect.

 

This has been an absolutely exceptional winter... I think IAD is 4th all time? and could move into 3rd....When we go back to reality and have another 7.6" winter, these products aren't going to be as successful...

Link to comment
Share on other sites

This has been an absolutely exceptional winter... I think IAD is 4th all time? and could move into 3rd....When we go back to reality and have another 7.6" winter, these products aren't going to be as successful...

Yup, only 1.1" off 02-03 there now.  DC is at #23, #10 at DCA.

Link to comment
Share on other sites

I'm curious to know how much human intervention that there was in the probabilities for snow in this morning's update (latest time stamp *for all of the data listed below* is 7:44am).

This is the data for the grid point that is in the middle of Montgomery County, MD:

10th percentile (expect at least this much): 1.4"

Most likely: 1-2"

90th percentile (potential for this much): 6.1"

And below those three maps:

Chance that snow accumulation will be greater than...

>0.1": 87%

>1": 87%

>2": 50%

>4": 30%

>8": 0%

-----

An 87% of >1" indicates that there's a 13% chance of 1" or less, yet the 10th percentile snowfall is 1.4". Those numbers don't add up. Yes, it's experimental, but it would be nice to have consistency amongst products that are supposedly derived from the same thing.

I think those percentiles are pretty bullish for daytime snowfall in late March without crazy rates, but I'll wait for the verification before digging in :P I know that this stuff is initially computer-generated, but I'd think in this case there would be more human tweaking to bring the range lower (unless the forecasters are really confident in these higher numbers...). Regardless of the forecast, it'll be interesting to see just how that band sets up tomorrow and if it will be strong enough to produce that widespread 1-2"+ snowfall.

EDIT:

Also, what's up with the 0% chance of 0.1-1.0"?

post-96-0-55610100-1395671437_thumb.png

Link to comment
Share on other sites

I'm curious to know how much human intervention that there was in the probabilities for snow in this morning's update (latest time stamp *for all of the data listed below* is 7:44am).

This is the data for the grid point that is in the middle of Montgomery County, MD:

10th percentile (expect at least this much): 1.4"

Most likely: 1-2"

90th percentile (potential for this much): 6.1"

And below those three maps:

Chance that snow accumulation will be greater than...

>0.1": 87%

>1": 87%

>2": 50%

>4": 30%

>8": 0%

-----

An 87% of >1" indicates that there's a 13% chance of 1" or less, yet the 10th percentile snowfall is 1.4". Those numbers don't add up. Yes, it's experimental, but it would be nice to have consistency amongst products that are supposedly derived from the same thing.

I think those percentiles are pretty bullish for daytime snowfall in late March without crazy rates, but I'll wait for the verification before digging in :P I know that this stuff is initially computer-generated, but I'd think in this case there would be more human tweaking to bring the range lower (unless the forecasters are really confident in these higher numbers...). Regardless of the forecast, it'll be interesting to see just how that band sets up tomorrow and if it will be strong enough to produce that widespread 1-2"+ snowfall.

EDIT:

Also, what's up with the 0% chance of 0.1-1.0"?

20140324_MoCoSnow.png

Good discussion here. I like it. Our goal is to narrow the goal posts as the event draws near. If we are saying the minimum is 1.0 then we are expecting at least that much. We should be advertising that because we have advisories out for 80% confidence.

Link to comment
Share on other sites

One of the real solid points I have taken from here is we ingest low models which are the 10th percentile but if we manually edit the product it is not fair to say it is still the 10th percentile. Main thing is to not follow percentiles but a range of possible solutions... Communicating uncertainty

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.

×
×
  • Create New...