Been loking at he CFS runs as they come out and it really is struggling ot produce anything wintry for a decent period through the months of DJF

JFF JFF JFF JFF

Well it is a model that has a case of over-progressiveness that makes GFS FI look sensible and restrained

I have long wondered how it ever manages to produce multi-run averages indicating high latitude blocking, and from what I've seen over the past year, it appears that even when it sends the jet well to the south south, it keeps it very active, producing vicious storms which in turn support monstrous blocking highs that span huge stretches of the northern hemisphere.

That's why when it does show a bit of high-lat. blocking, it tends to be sensational, leading to those charts that leave people dreaming of raging blizzards and all the rest.

What's more, CFS loves to project a very flat jet profile (fits with the tendency to ramp up the intensity), so on those occasions where that jet is fired due east, CFS can end up showing an entire month of blizzard after blizzard . Conversely, if the jet is fired NE, the blocking lasts no more than a few days.

Thanks to the habit of developing huge storm systems, CFS favours the NE jet, which IMO makes those occasions when it shows more than a transient period of blocking a more significant occurance; it's working against even greater odds than reality to project that.

I have given this a lot of thought over the past 12 months. The tendency to develop extensive blocking only occasionally, but to then take that to the extreme, explains a lot of what I've observed, but what about others on this forum? I'm very interested in your opinions on this

Been loking at he CFS runs as they come out and it really is struggling ot produce anything wintry for a decent period through the months of DJF

JFF JFF JFF JFF

David Icke is more likely to be right than any of these CFS charts. I look at as a bit of fun but you cant take anything it produces (trends or otherwise) seriously, same as any long range view

Hence the repeated JFF

What's more, CFS loves to project a very flat jet profile (fits with the tendency to ramp up the intensity), so on those occasions where that jet is fired due east, CFS can end up showing an entire month of blizzard after blizzard . Conversely, if the jet is fired NE, the blocking lasts no more than a few days.

IMO, this is a significant cause of the different climate regimes of multi decadal warming and cooling. The Azores high, affected by the ozone changes caused by the different output by the sun at certain UV levels, either generally, deflects the jet NEwards - or it doesn't.

Makes a lot of sense - perhaps the people behind CFS should focus on very accurate modelling of the Azores High. Trouble is, the overcooking of storm systems and resultant steering of the jet energy NE often leads to more ridging of the Azores High northeastward than is realistic.

When CFS was persistently predicting an extreme March in terms of blocking at the high latitudes, it could be that there was a very strong signal for the Azores High to be supressed or absent, something that no amount of northward bias can overcome.

To be fair on CFS, GFS can be just as bad in the 192-384h range, and ECM is sometimes not much better for 168-240h.

I believe the problem the computer models have is that they transfer energy too efficiently, this being a result of assumptions made in the programming, which were required partly to prevent the modelling becoming so complex that a single run takes more than a day, and partly because some of those smaller details remain poorly understood.

As an example, the models often see all of the energy from a strong thermal gradient (clash of airmasses) being bundled neatly into a large and powerful storm system which then retains a lot of it's energy and makes a lot of progress in whatever direction the steering pattern allows for. Yet the reality is often far less tidy, with energy either forming a more unstable system that breaks apart (disrupting trough) or forming numerous seperate disturbances which interact in a complex manner, making for tricky forecasting. In both cases, progression is usually far more limited as energy is dispersed - it has a lot less 'shove'.

My entire theory is now here for all to see... make of it what you will

What model is good after 168??

In mobile conditions I should add

Not bad snow cover for late September across Eastern Siberia - more than this time last year (scroll down for Eurasia):

http://www.climate4you.com/SnowCover.htm

And latest:

http://www.natice.noaa.gov/pub/ims/ims_gif/DATA/cursnow_asiaeurope.gif

Will be interesting to see if that builds up further.

Not bad snow cover for late September across Eastern Siberia - more than this time last year (scroll down for Eurasia):

 

http://www.climate4you.com/SnowCover.htm

 

And latest:

 

http://www.natice.noaa.gov/pub/ims/ims_gif/DATA/cursnow_asiaeurope.gif

 

Will be interesting to see if that builds up further.

 

 

Looks like a lot more ice as well around the North pole. I wonder what effect if any that will have on the NH winter.

What's more, CFS loves to project a very flat jet profile (fits with the tendency to ramp up the intensity), so on those occasions where that jet is fired due east, CFS can end up showing an entire month of blizzard after blizzard . Conversely, if the jet is fired NE, the blocking lasts no more than a few days.

IMO, this is a significant cause of the different climate regimes of multi decadal warming and cooling. The Azores high, affected by the ozone changes caused by the different output by the sun at certain UV levels, either generally, deflects the jet NEwards - or it doesn't.

Makes a lot of sense - perhaps the people behind CFS should focus on very accurate modelling of the Azores High. Trouble is, the overcooking of storm systems and resultant steering of the jet energy NE often leads to more ridging of the Azores High northeastward than is realistic.

When CFS was persistently predicting an extreme March in terms of blocking at the high latitudes, it could be that there was a very strong signal for the Azores High to be supressed or absent, something that no amount of northward bias can overcome.

To be fair on CFS, GFS can be just as bad in the 192-384h range, and ECM is sometimes not much better for 168-240h.

I believe the problem the computer models have is that they transfer energy too efficiently, this being a result of assumptions made in the programming, which were required partly to prevent the modelling becoming so complex that a single run takes more than a day, and partly because some of those smaller details remain poorly understood.

As an example, the models often see all of the energy from a strong thermal gradient (clash of airmasses) being bundled neatly into a large and powerful storm system which then retains a lot of it's energy and makes a lot of progress in whatever direction the steering pattern allows for. Yet the reality is often far less tidy, with energy either forming a more unstable system that breaks apart (disrupting trough) or forming numerous seperate disturbances which interact in a complex manner, making for tricky forecasting. In both cases, progression is usually far more limited as energy is dispersed - it has a lot less 'shove'.

My entire theory is now here for all to see... make of it what you will

im by no means a techy but are the models programmed to 'learn' or are they purely predictive based on some set of complex algorithms

I agree with what you say but as we see the same thing year on year, across most models, do they not incorporate some level of predcictive analysis based on whats actually transpired and factor that into future modelling?

http://modeles.meteociel.fr/modeles/cfs/run3/cfsnh-0-2382.png?12

JFF JFF JFF JFF

A New Years Day easterly

I suppose it's a question of scale. The surface area of a sphere the size of the Earth is vast, making computer-calculations of the small-scale events which disrupt troughs impractical, unless you're prepared to wait 168+ hours for the model run to finish.

If an attempt were made to cause the model to learn not to allow huge, organized storm systems to develop so readily, I guess you would have to programme it to generate some sort of random disruption, which would render the rest of the run equally random, and God doesn't play dice with the Model Output thread.

Good points there I've not seen any evidence of the models learning from what's transpired and I can't really see that sort of thing being possible without requiring unjustifiable computor resources.

The learning has to be done by humans through research, which can then be used to upgrade the model workings. A very useful area of research in the past few decades has been one that involves taking a complex relationship between variables and developing a way of achieving very good approximations for the outcome using fewer variables. Unfortunately, that leads to assumptions which then cause the problems seen in the current models.

With that in mind, it's going to be hard to produce more accurate forecasts in the future through any other means than faster computers. Thankfully, there are promising developments in computing technology at the moment, including atomic-level manufacturing (early days on that though) and quantum-computing which could prove very useful indeed for weather forecasting.

I hope computer models aren't able to 'learn' just yet.  I'd like a few more years before Skynet takes over

Good points there I've not seen any evidence of the models learning from what's transpired and I can't really see that sort of thing being possible without requiring unjustifiable computor resources.

The learning has to be done by humans through research, which can then be used to upgrade the model workings. A very useful area of research in the past few decades has been one that involves taking a complex relationship between variables and developing a way of achieving very good approximations for the outcome using fewer variables. Unfortunately, that leads to assumptions which then cause the problems seen in the current models.

With that in mind, it's going to be hard to produce more accurate forecasts in the future through any other means than faster computers. Thankfully, there are promising developments in computing technology at the moment, including atomic-level manufacturing (early days on that though) and quantum-computing which could prove very useful indeed for weather forecasting.

The 'big data' approach and the ability to store and process huge data sets in new ways (Hadoop etc) might be the answer. After all weather data must be a great source for this type of approach and then layer predictive analysis tools on top of it

Indeed you wonder if the more predictive analysis approach might overtake the tradiotional algorithm led 'super computer' approach at some point?

http://www.wired.com/insights/2013/02/how-big-data-can-boost-weather-forecasting/

http://arstechnica.com/business/2012/03/how-ibms-deep-thunder-delivers-hyper-local-forecasts-3-12-days-out/

http://www.forbes.com/sites/alexknapp/2013/06/13/forecasting-the-weather-with-big-data-and-the-fourth-dimension/

http://modeles.meteociel.fr/modeles/cfs/run2/cfsnh-0-2958.png?06

JFF JFF JFF JFF

An Easterly late on in January from CFS , but that is about all.

Moscow will be 10 degrees below average by Tuesday with a max of only 2C  not sure how this bodes for our weather down the line >>> http://www.wunderground.com/cgi-bin/findweather/getForecast?query=Moscow,%20Russia&wuSelect=WEATHER

And look at those uppers too  http://www.wetterzentrale.de/pics/MT8_Moskau_ens.png

Stockholm, Helsinki and Vilnius are the first European capitals to fall below freezing this autumn. Let the descent into winter begin.

Stockholm, Helsinki and Vilnius are the first European capitals to fall below freezing this autumn. Let the descent into winter begin.

Looks like you missed one, Oslo was another http://classic.wunderground.com/history/airport/ENGM/2013/9/26/DailyHistory.html?req_city=NA&req_state=NA&req_statename=NA

Moscow will be 10 degrees below average by Tuesday with a max of only 2C  not sure how this bodes for our weather down the line >>> http://www.wunderground.com/cgi-bin/findweather/getForecast?query=Moscow,%20Russia&wuSelect=WEATHER

And look at those uppers too  http://www.wetterzentrale.de/pics/MT8_Moskau_ens.png

Gavin highlights this well in his reccent video, some decent cold coming down to Eastern Europe early

The 'big data' approach and the ability to store and process huge data sets in new ways (Hadoop etc) might be the answer. After all weather data must be a great source for this type of approach and then layer predictive analysis tools on top of it

Indeed you wonder if the more predictive analysis approach might overtake the tradiotional algorithm led 'super computer' approach at some point?

 http://www.wired.com/insights/2013/02/how-big-data-can-boost-weather-forecasting/

http://arstechnica.com/business/2012/03/how-ibms-deep-thunder-delivers-hyper-local-forecasts-3-12-days-out/

http://www.forbes.com/sites/alexknapp/2013/06/13/forecasting-the-weather-with-big-data-and-the-fourth-dimension/

 

Thanks for those informative linsk Seems to be a lot of promise in that, although we don't want the models becoming so accurate that we can't enjoy discussing the probability of various charts verifying unless there at 240h+

On a serious note, there's still good progress to be made for locallised forecasts even at just a day's range or two - that's where efforts should be focused for now IMO

Before long we'll have another winter unfolding, with various models gaining and losing credence as it progresses

The 'big data' approach and the ability to store and process huge data sets in new ways (Hadoop etc) might be the answer. After all weather data must be a great source for this type of approach and then layer predictive analysis tools on top of it

Indeed you wonder if the more predictive analysis approach might overtake the tradiotional algorithm led 'super computer' approach at some point?

 http://www.wired.com/insights/2013/02/how-big-data-can-boost-weather-forecasting/

http://arstechnica.com/business/2012/03/how-ibms-deep-thunder-delivers-hyper-local-forecasts-3-12-days-out/

http://www.forbes.com/sites/alexknapp/2013/06/13/forecasting-the-weather-with-big-data-and-the-fourth-dimension/

 

Thanks for those informative linsk Seems to be a lot of promise in that, although we don't want the models becoming so accurate that we can't enjoy discussing the probability of various charts verifying unless there at 240h+

On a serious note, there's still good progress to be made for locallised forecasts even at just a day's range or two - that's where efforts should be focused for now IMO

Before long we'll have another winter unfolding, with various models gaining and losing credence as it progresses

http://modeles.meteociel.fr/modeles/cfs/run4/cfsnh-0-1554.png?18

CFS gives us a chilly end to November

http://modeles.meteociel.fr/modeles/cfs/run4/cfsnh-0-2184.png?18

Santa brings a bitter blast with him

http://modeles.meteociel.fr/modeles/cfs/run4/cfsnh-0-2694.png?18

Middle of January sees a Easterly blow

All JFF JFF JFF of course