Jump to content
  • Member Statistics

    17,609
    Total Members
    7,904
    Most Online
    NH8550
    Newest Member
    NH8550
    Joined

February mid to long range threats discussion


Ji

Recommended Posts

Really?  I hope you're joking here somewhat, but fear you are being serious.  Models are programmed, they are objective guidance and calculate parameters based upon the governing equations of motion, thermodynamics, physics, etc., within the limits of our ability to estimate those things and within our current computing ability.  I'm over-simplifying here for sure and others more in the know can add more.  But there's a lot of complicated interaction involved and to say they have not produced a viable or valuable product this winter season is pure bunk.  It's pretty ridiculous and laughable to think you can code in something to say "hmmm, SNE is getting hammered this year, so let's make this low go north 100 miles to screw the Mid Atlantic and give Boston another foot because that's the seasonal trend!"

A model is only as good as the data that goes into it. However, we will never be able to sample precisely every layer of the atmosphere at every location point across the entire globe, thus a CPU that discovers where it is most often inaccurate, could start improving assumptions over sparse data environments and improve those assumptions with time based on results and verification. 

Link to comment
Share on other sites

  • Replies 2k
  • Created
  • Last Reply

Artificial intelligence.  Probably too costly for NWS to experiment with, but that's the wave of the future, CPUs that improve themselves.

 

not quite -- NWP is not purely an optimization problem -- our issues are not finding a better way to do it (which is what AI does) -- we need better observations, more CPU power, better parameterizations (and the eventual move to more discrete physics for convection/microphysics/etc..), better DA but ultimately we still have predictability limits

 

just to put this in context -- you can run the same code, same inputs, same everything on two systems with slightly different system architectures and just from very small rounding differences you can end up with  5 to 10 day forecasts that are vastly different -- just from the propagation of non-linear error and its growth.... 

Link to comment
Share on other sites

A model is only as good as the data that goes into it. However, we will never be able to sample precisely every layer of the atmosphere at every location point across the entire globe, thus a CPU that discovers where it is most often inaccurate, could start improving assumptions over sparse data environments and improve those assumptions with time based on results and verification. 

 

this makes a bit more sense, so I see what you're trying to say --- while this might be possible -- another angle to look at is finding areas where new observations could provide the most impact -- on some level we already attempt to do this, we have special experiments which attempt to assess the impact of future sensors on NWP

Link to comment
Share on other sites

I call bull****.

We average enough, DCA, certainly IAD, and BWI that we can expect good snow events.

There's probably some sort of complicated cycle of weather that leads to us not cracking climo back to back in nearly 30 years. IAD hasn't done it since 81-82/82-83.

Going back to 2000, IAD beat climo 3 times but didn't crack 10" 4 times.

I don't think we can expect good snow events in any year. The risk for a dud is always present. And coming off a big year made it higher.

Link to comment
Share on other sites

There's probably some sort of complicated cycle of weather that leads to us not cracking climo back to back in nearly 30 years. IAD hasn't done it since 81-82/82-83.

Going back to 2000, IAD beat climo 3 times but didn't crack 10" 4 times.

I don't think we can expect good snow events in any year. The risk for a dud is always present. And coming off a big year made it higher.

 

My baseline is one area-wide WSW/Warning event a year. Just the one. I don't think that is an unreasonable baseline. You can get that and still be below average. I would expect with our climo, one a year is doable, at the least.

Link to comment
Share on other sites

There's probably some sort of complicated cycle of weather that leads to us not cracking climo back to back in nearly 30 years. IAD hasn't done it since 81-82/82-83.

Going back to 2000, IAD beat climo 3 times but didn't crack 10" 4 times.

I don't think we can expect good snow events in any year. The risk for a dud is always present. And coming off a big year made it higher.

Nobody has missed climo, yet.

Link to comment
Share on other sites

There's probably some sort of complicated cycle of weather that leads to us not cracking climo back to back in nearly 30 years. IAD hasn't done it since 81-82/82-83.

Going back to 2000, IAD beat climo 3 times but didn't crack 10" 4 times.

I don't think we can expect good snow events in any year. The risk for a dud is always present. And coming off a big year made it higher.

 

I think the major disappointment is that all of the fall indicators were there for a great winter. The Enso modelling fell through and that seemed to doom a lot of forecasts from the get go. It's not over yet for sure. But I dont see any of us reaching climo this winter.

Link to comment
Share on other sites

I think the major disappointment is that all of the fall indicators were there for a great winter. The Enso modelling fell through and that seemed to doom a lot of forecasts from the get go. It's not over yet for sure. But I dont see any of us reaching climo this winter.

Yep. It was a unanimous met and weenie long range bust. Yea, sne managed to score but not because the lr forecasts verified.

I was worried we were in pretty big trouble after December and then first half of Jan not flipping in the blocking dept made it easy to let go.

We have about 33 days of decent snow climo left. Let the chips fall.

Link to comment
Share on other sites

not quite -- NWP is not purely an optimization problem -- our issues are not finding a better way to do it (which is what AI does) -- we need better observations, more CPU power, better parameterizations (and the eventual move to more discrete physics for convection/microphysics/etc..), better DA but ultimately we still have predictability limits

 

just to put this in context -- you can run the same code, same inputs, same everything on two systems with slightly different system architectures and just from very small rounding differences you can end up with  5 to 10 day forecasts that are vastly different -- just from the propagation of non-linear error and its growth.... 

I was thinking something similar. Also, love the putting it into context piece! That really is wild. I want to ask so many questions, but the one that makes me wonder is , what goes into finally deciding on the model code that that becomes the "operational" model. We look at the Ensembles which shows us many different iterations, what makes these wrong the production one right? Is it based on error measuring? I suspect I would have to read up on it. I am amazed that we get a model output the way we do. We have come so far. The issue is that we expect perfection from a model.. there is meteorology involved too! People can look at other things! You can see that there is no Negative NAO, no true 50/50 low, a poor placement of a ridge, or even the latitude at which a storm is forming and make educated predictions.. the models help us recognize these things as well, and so specific issues with R/S line and temp profiles and ultimate true Low placement are not going to be certain.. that certainty does change as we get closer to the event.. or it is more certain if we have stable and predictable patterns (like last year).. I think people expect too much! I know I do!

Link to comment
Share on other sites

I noticed that for every single winter in the past 13 years that I hit climo, my area got a MECS.

Very hard to reach without a big storm.

It is hard but it does occasionally happen. I reached average in 04/05 without an event over 5". And Matt has posted a thread before on reaching 20" at DCA (well above average) without a 6" storm-- it happened three times: 62/63, 67/68, and 77/78. 

Link to comment
Share on other sites

not quite -- NWP is not purely an optimization problem -- our issues are not finding a better way to do it (which is what AI does) -- we need better observations, more CPU power, better parameterizations (and the eventual move to more discrete physics for convection/microphysics/etc..), better DA but ultimately we still have predictability limits

 

Given that there will always be errors in the models for reasons you describe above, using AI to predict (and correct for) the delta between the models and reality is not a bad idea.  I suspect that good mets are able to identify systematic errors and trends in the models and correct for them.  It's not that the mets have more sensors than the models do, or that they are doing math on a finer grid -- it's just pattern recognition.  If humans can do that, we should be able to get a machine to do it too.  

Link to comment
Share on other sites

Given that there will always be errors in the models for reasons you describe above, using AI to predict (and correct for) the delta between the models and reality is not a bad idea.  I suspect that good mets are able to identify systematic errors and trends in the models and correct for them.  It's not that the mets have more sensors than the models do, or that they are doing math on a finer grid -- it's just pattern recognition.  If humans can do that, we should be able to get a machine to do it too.  

 

yeah AI could eventually replace (or do a better job than) human forecasters -- as data mining NWP output is an optimization problem that AI could excel at 

Link to comment
Share on other sites

Really?  I hope you're joking here somewhat, but fear you are being serious.  Models are programmed, they are objective guidance and calculate parameters based upon the governing equations of motion, thermodynamics, physics, etc., within the limits of our ability to estimate those things and within our current computing ability.  I'm over-simplifying here for sure and others more in the know can add more.  But there's a lot of complicated interaction involved and to say they have not produced a viable or valuable product this winter season is pure bunk.  It's pretty ridiculous and laughable to think you can code in something to say "hmmm, SNE is getting hammered this year, so let's make this low go north 100 miles to screw the Mid Atlantic and give Boston another foot because that's the seasonal trend!"

 

yea, you can't code models using the gambler's fallacy.

Link to comment
Share on other sites

There's probably some sort of complicated cycle of weather that leads to us not cracking climo back to back in nearly 30 years. IAD hasn't done it since 81-82/82-83.

Going back to 2000, IAD beat climo 3 times but didn't crack 10" 4 times.

I don't think we can expect good snow events in any year. The risk for a dud is always present. And coming off a big year made it higher.

I am hoping every year we will overperform. It is getting old now. I have lived in the region over 30 years. I'd even take an ice storm a la '94 right now.

Link to comment
Share on other sites

I've read both and feel that they did not cover the topics properly.

I agree.  Incredible information can be put on the head of a pin but cannot get weather models to reflect that in a given winter, for example, that 9 out of every 10 low pressure systems end up 200-300 miles further north.

Link to comment
Share on other sites

So, eventually (next year?) the elusive -NAO will reappear.  But, this is no guarantee there is any meaningful cold air around.  2009-10 was pretty average, temperature wise overall.  

 

As some were skeptical (in hindsight, rightly so) about DC experiencing 2 back to back +SN winters during the previous pre-season, it now looks as if next year we could be looking for a 3rd consecutive BN winter temperature wise.  Or, at least a 3rd consecutive season that is not materially AN temperature wise (anticipating the DCA crew to descend and point out BN for DCA is far from a seasonal lock this year).  Pretty sure you can still get 33 and rain with a -NAO in DC and the surrounding areas.

 

I am equally skeptical on our chances to pull off a 3rd consecutive "nice" summer, temperature wise.

Link to comment
Share on other sites

So, eventually (next year?) the elusive -NAO will reappear. But, this is no guarantee there is any meaningful cold air around. 2009-10 was pretty average, temperature wise overall.

As some were skeptical (in hindsight, rightly so) about DC experiencing 2 back to back +SN winters during the previous pre-season, it now looks as if next year we could be looking for a 3rd consecutive BN winter temperature wise. Or, at least a 3rd consecutive season that is not materially AN temperature wise (anticipating the DCA crew to descend and point out BN for DCA is far from a seasonal lock this year). Pretty sure you can still get 33 and rain with a -NAO in DC and the surrounding areas.

I am equally skeptical on our chances to pull off a 3rd consecutive "nice" summer, temperature wise.

2009-10 was a very cold winter in the south... mainly because of both the El Niño and amazing blocking. Temps were only moderately cold at our latitude that winter, but the overall setup was perfect for us to get heavy snow.

I doubt that next winter will be cold... maybe it's my inner Ian but the east coast is probably way overdue for a mundane winter without any KU that is also somewhat mild. Hope I'm wrong.

Link to comment
Share on other sites

2009-10 was a very cold winter in the south... mainly because of both the El Niño and amazing blocking. Temps were only moderately cold at our latitude that winter, but the overall setup was perfect for us to get heavy snow.

I doubt that next winter will be cold... maybe it's my inner Ian but the east coast is probably way overdue for a mundane winter without any KU that is also somewhat mild. Hope I'm wrong.

I'm just looking for some snow justice!  :wacko2:

Link to comment
Share on other sites

yeah AI could eventually replace (or do a better job than) human forecasters -- as data mining NWP output is an optimization problem that AI could excel at 

Thanks for your replies.  I admit that the models, and physics, and data assimilation that goes into these complex models is beyond the limits of my comprehension.  I'm sure there are very few people that understand how a model works from the very basic level to the most complex algorithm, and I truly have loads of respect for those who work in this field.  The fact that a model can be even remotely accurate at 5+ days is astounding to me - 100 mile difference in the location of a low pressure system is really small given the surface area of the globe.  As you pointed out, a simple rounding difference in a calculation can have enormous impacts downstream, especially beyond 3, 5, or even 10 days.  I'm very impressed with what we have currently.  The guys at NWS do a fantastic job.  And honestly, I know very little about AI, but from what I've read the possibilities are extremely intriguing, and something that could benefit the forecasting field in some shape or form in the long term.

Link to comment
Share on other sites

IMO- this +pna pattern is here to stay. It's going to flex and relax and not be static where we are on the back side of the trough for days on end.

The big hp next week is going to happen in some fashion and will likely squish things for a time. Our best chance at a big storm is something gliding under the ridge and teaming up with a ns shortwave. Otherwise, we can get clipped or just get lucky with a big ns dig. There are ways to do it. All we can do is hope something breaks our way.

The +pna ridge could last for a month on the means. If we can't get a single accum event I'd be kinda surprised.

Link to comment
Share on other sites

Archived

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

Guest
This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.

×
×
  • Create New...