Jump to content
Will it snow? 7 day forecast Weather Radar Will it be cold?
Snow?
Local
Radar
Cold?

February1978

Members
 View Profile  See their activity

  • Posts

    114

  • Joined

  • Last visited

 Content Type 

Posts posted by February1978

    • Popular Post
    • Popular Post

    Model Output Discussion - Spring Has Sprung

    in Forecast Model Discussion

    Posted

    Good to see a southward trend of the models, much better for longevity through February.

    Meanwhile, whilst we’re all wondering where the ‘snow line’ is going to be on Thursday, it’s worth remembering that in these global models 50-100 miles is quite a small distance - but for us it could be a huge deal.

    Expect ‘small’ changes right until the end & enjoy the ride……!

    • Like 14

    Model Output Discussion - Spring Has Sprung

    in Forecast Model Discussion

    Posted

    1 hour ago, Catacol said:

    Coming back on this one. The probability of the unexpected. A genuinely relevant and interesting question in itself. Some will groan and grind their teeth at the tendency for anything that can derail a cold pattern from coming to pass almost routinely. How many times do we read curses around "short waves" messing up the pattern, for example?! I dunno. Some of this is micro stuff, and when we get to the micro analysis of tiny shifts of 50 miles or low pressure waves popping up - I begin to glaze over. We don't have the computing power yet to be able to forecast these with any accuracy and I think they are beyond the wit of the human brain to be able to interpret from an analytical perspective. So the unexpected micro development will be a thorn in our side, and a thorn that arrives on the scene frequently, until resolution and computer modelling gets better.

    Agreed, will it ever be possible to model these small scale features at more than a handful of days? There will always be an error in the initial state, whether it’s instrument error, lack of coverage in certain parts of the atmosphere, topology or the like and that’s before any physics assumptions are made. Due to the chaotic nature of meteorology, there is often not a ‘steady state solution’ & errors grow, irrespective of how fine the resolution or accurate the dynamical equations are. Rubbish in, rubbish out……..so I think NWP will always struggle beyond a few days.

    On a slightly different tack - the (off-the-) scale of EAMT anomaly highlighted in Tamara’s post is really something - must surely have an impact in due course….?

    • Like 2

    Model Output Discussion - Colder but how cold and for how long?

    in Forecast Model Discussion

    Posted

    1 minute ago, kold weather said:

    I think the other thing that warrants it is there are ensemble members and the occasional operational model still making it into at least parts of the south. Interestingly the ARPEGE ensembles have swung northwards considerably in its most recent run.

    Its a classic low chance but high impact scenario, and I'm willing to bet the met office already have a yellow/amber warning ready made should things seem to be north of where currently expectations are.

    Lets just say I wouldn't be surprised if there is a small line of 5-10cms on the ground by the end of Wednesday. I also wouldn't be surprised if the front never leaves the French coastline. Both and everything inbetween is still on the table.

    ECM keeps it away from the south coast on 12z (except Cornwall) on this one. 

    But watching brief until it happens as you and saintkip say........

    • Like 1

    Model Output Discussion - Colder but how cold and for how long?

    in Forecast Model Discussion

    Posted

    2 minutes ago, kold weather said:

    Man that front is close to the south coast, or even on the south coast on Wednesday on the 12z runs.

    Honestly its well within the margin of error still. I recall a similar rain event which progged to only brush the south coast but ended up producing a pretty signifncant rain event back in December. 

    Thats not to say I think its the likely outcome, I think its not. But its still close enough call that it will need very close watching right through to Wednesday, we are talking literally 10-20 miles making the difference between basically nothing and 3-5cms into southern coastal counties. 

    I agree with this, 50 miles is not far in NWP, would make a massive difference to us. I've seen rain extend quite a bit further north than forecast in this type of set-up before.

    That said, there is pretty good consensus between the models for it to stay at coast or south.

    • Like 2

    Model Output Discussion - Colder but how cold and for how long?

    in Forecast Model Discussion

    Posted · Edited by February1978

    7 minutes ago, Derecho said:

    It's an interesting take on the models MIA and the questions from me weren't necessary a criticism, more a position of intrigue.

    There are two main aspects to the model uncertainty argument:

    1) The assumptions aren't fully correct - This is why weather modelling is so difficult (chaos). It makes up the majority of the errors in longer range forecasting as you say.
    2) Lack of data coverage: - Bodes into the "junk in, junk out" argument as well. Without a full understanding of the surface state the weather model is already wrong.

    Just out of interest here is the uncertainty at T0 on the GFS and EC

    image.thumb.png.46f3fd3eb783e6c798c3eb9e5959740d.pngimage.thumb.png.080eebff13ba35d16b18ba67b13a994f.png

    Both the key areas of uncertainty at T0 are with regards to the Azores low and over Greenland.

    image.thumb.png.43221eb3402aaf10fd2226b37e9b1bb7.pngimage.thumb.png.fa9d72ebf1683de806c87ad59a105a0c.png

    You can see from the two plots how much quickly the uncertainty increases first around areas where the understanding of the initial surface state isn't as good.

    So I believe that if we had a correct surface state the 'butterfly effect' would kick in later.  More data would result in less error but there is a limit into how much we can improve due to incorrect assumptions. 

    How much can we learn about these assumptions if we have more data though?

    Interesting topic.

    ...and the pressure of the Greenland High has largest uncertainties/differences.....

    • Like 1

    Model discussion highlights

    in Forecast Model Discussion

    Posted

    41 minutes ago, Midlands Ice Age said:

    Derecho...

    30 years of modelling, programming , analysis, technology design and implementation has left its scars on me. 

    Twenty years ago we needed more machine power.

    Now, unless we sort out the assumptions  and get them analysed correctly into mathematics  we will only magnify the errors as we push ever deeper into being able to look at ever more local detail.

    Tamara is showing where some of the current assumptions are wrong, or at least give a totally different view. .

    People on here showed how the stratosphere could affect things greatly. (10 years ago) This was incorporated into the models (though I believe is now possibly over responsive (see below)).

    The only people to gain by increasing the computing power right now are 1) computer salesmen, and 2) system design specialists who have no further ideas as to how to improve the current models.

    Incorrect assumptions ALWAYS increase volatility in output. Increased  computing power will only serve to show up any design errors.

    I appreciate that increasing the spread by the changing the input criteria, is one method of reducing any errors. But the initial 'errors' are still present, and will only serve to produce more chaos.

    I am talking about from where the models are standing today (which is much better than they were 5-10 years ago).  Spending hundreds of millions on more computing power right now will not give the same rewards as ensuring some of the 'unknowns' are fully researched and included  as regards the accuracy of the models..

    1)     Extra notes.... Take today and this graphic taken from the above (to be saved)....

     image.png.2601d1ad27281e20ce541275cf1e7ad3.png 

    The whole of even  the western hemisphere has been impacted by the the slope and shape of the high forecast to develop  over  the Bering Ocean.  Changing its inclination or intensity will dramatically change the forecast for the whole Northern Hemisphere via mechanisms previously discussed on here. Both the ECM and GFS have shown this happening over the last week. Both had to withdraw from their positions . Something is not 'spot on'.

    Will they this time? Can you tell me? I assume that this situations has come about as to their incorrectly handling the totality of the 'telecomms' signals. 

    2) As to whether the models are accurate from a scientific pov.    I cannot make any comments  (you are correct).  However they have now released documentation on the 'CC' models, which were based upon forecast models  (I  am told),  and for the previous release of them there were still around 25 assumptions which were not yet proven. If they were missing in the 'CC' models then I assume they were not in the forecast models.  - In the CC models they substituted parameters. 

    Sorry to mention the XX word banned on here  - MODS.

    So IMO opinion,   (FWIW)  there is still quite a way to go before we can say that all the assumptions (and therefore the absolutely correct science) built into the models, means that models are correct.

    Until that happens  putting in more computing power will increase chaos (not decrease). that is what my 30 odd years of experience has taught me.

    MIA

    As someone also in a similar scientific software business, I agree with the sentiment of this.

    Initial conditions are a big thing in my opinion. In an inherently chaotic system, if you don't get the initial state correct you won't get an accurate answer. Rubbish in ---> rubbish out. 

    Areas lacking data are at altitude, over the oceans and near the poles where there are no humans to record and report data, so satellite and aircraft data are used I believe.

    Then you have the solution method which always has a truncation error and probably has efficiencies to make sure it is a) robust and b) quick enough to reach a solution. Developers make compromises/assumptions which always have impacts in due course. For example, mountainous areas will be a problem - how do you model the boundaries there? You can't directly model each feature (mountain), it's too small a part of the domain, but they are important for (eg) torque as discussed at length by others previously – how do we know we are accurate enough?

    All numerical models have inherent assumptions and weaknesses, much of the skill of those using them is to understand where those limitations are. It has been really fascinating to see the experts here comment on the output, what they trust and what they don't. We're in one of those spells where they are not convinced by the outputs they are seeing and giving their opinions, much respect for them sharing their knowledge from me – thanks!

    Apologies if this is slightly off piste, but it is relatively quiet….


    Original post: https://community.netweather.tv/topic/99760-model-output-discussion-colder-but-how-cold-and-for-how-long/?do=findComment&comment=5008639
    • Like 2
    • Popular Post
    • Popular Post

    Model Output Discussion - Colder but how cold and for how long?

    in Forecast Model Discussion

    Posted

    41 minutes ago, Midlands Ice Age said:

    Derecho...

    30 years of modelling, programming , analysis, technology design and implementation has left its scars on me. 

    Twenty years ago we needed more machine power.

    Now, unless we sort out the assumptions  and get them analysed correctly into mathematics  we will only magnify the errors as we push ever deeper into being able to look at ever more local detail.

    Tamara is showing where some of the current assumptions are wrong, or at least give a totally different view. .

    People on here showed how the stratosphere could affect things greatly. (10 years ago) This was incorporated into the models (though I believe is now possibly over responsive (see below)).

    The only people to gain by increasing the computing power right now are 1) computer salesmen, and 2) system design specialists who have no further ideas as to how to improve the current models.

    Incorrect assumptions ALWAYS increase volatility in output. Increased  computing power will only serve to show up any design errors.

    I appreciate that increasing the spread by the changing the input criteria, is one method of reducing any errors. But the initial 'errors' are still present, and will only serve to produce more chaos.

    I am talking about from where the models are standing today (which is much better than they were 5-10 years ago).  Spending hundreds of millions on more computing power right now will not give the same rewards as ensuring some of the 'unknowns' are fully researched and included  as regards the accuracy of the models..

    1)     Extra notes.... Take today and this graphic taken from the above (to be saved)....

     image.png.2601d1ad27281e20ce541275cf1e7ad3.png 

    The whole of even  the western hemisphere has been impacted by the the slope and shape of the high forecast to develop  over  the Bering Ocean.  Changing its inclination or intensity will dramatically change the forecast for the whole Northern Hemisphere via mechanisms previously discussed on here. Both the ECM and GFS have shown this happening over the last week. Both had to withdraw from their positions . Something is not 'spot on'.

    Will they this time? Can you tell me? I assume that this situations has come about as to their incorrectly handling the totality of the 'telecomms' signals. 

    2) As to whether the models are accurate from a scientific pov.    I cannot make any comments  (you are correct).  However they have now released documentation on the 'CC' models, which were based upon forecast models  (I  am told),  and for the previous release of them there were still around 25 assumptions which were not yet proven. If they were missing in the 'CC' models then I assume they were not in the forecast models.  - In the CC models they substituted parameters. 

    Sorry to mention the XX word banned on here  - MODS.

    So IMO opinion,   (FWIW)  there is still quite a way to go before we can say that all the assumptions (and therefore the absolutely correct science) built into the models, means that models are correct.

    Until that happens  putting in more computing power will increase chaos (not decrease). that is what my 30 odd years of experience has taught me.

    MIA

    As someone also in a similar scientific software business, I agree with the sentiment of this.

    Initial conditions are a big thing in my opinion. In an inherently chaotic system, if you don't get the initial state correct you won't get an accurate answer. Rubbish in ---> rubbish out. 

    Areas lacking data are at altitude, over the oceans and near the poles where there are no humans to record and report data, so satellite and aircraft data are used I believe.

    Then you have the solution method which always has a truncation error and probably has efficiencies to make sure it is a) robust and b) quick enough to reach a solution. Developers make compromises/assumptions which always have impacts in due course. For example, mountainous areas will be a problem - how do you model the boundaries there? You can't directly model each feature (mountain), it's too small a part of the domain, but they are important for (eg) torque as discussed at length by others previously – how do we know we are accurate enough?

    All numerical models have inherent assumptions and weaknesses, much of the skill of those using them is to understand where those limitations are. It has been really fascinating to see the experts here comment on the output, what they trust and what they don't. We're in one of those spells where they are not convinced by the outputs they are seeing and giving their opinions, much respect for them sharing their knowledge from me – thanks!

    Apologies if this is slightly off piste, but it is relatively quiet….

    • Like 4
    • Insightful 9
×
×
  • Create New...