Has this been tried for weather forecasting?
So in one of my random brain spasms, I wondered about bayesian learning.
Many spam filters use it - it starts off crap, but as you teach it (by clicking "this is not spam" or "this is spam" when it gets it wrong), it "learns" the characteristics of your email, and after a couple of days it can filter with a very low false positive rate. Could this be applied to weather? Or has it already been done? So far as I'm aware from sort of general knowledge, current weather forecasting relies on networks of sensors and satelite data, which is then plugged into mathematical models, squeezed through the Met's latest supercomputer, and prettified for our viewing pleasure. The obvious downside to that is that you only really know the conditions where the sensors have sampled; anything else is extrapolation (guesswork), and small differences scaled up can mean the forecast for 3 days down the line is totally screwed. Would it be possible to somehow combine the models with the bayesian techniques, so that the system could "learn" to spot when the weather is likely to do something unexpected that the rigid maths won't pick up? Just a bit of a musing really. |
Re: Has this been tried for weather forecasting?
All I know is they break up forecasting into grids. The smaller the grids, the more accurate the prediction. But the smaller the grids, the more intensive it is computing wise. Which is why the have super computers.
|
Re: Has this been tried for weather forecasting?
Yeah, the trouble with the grids is that however small you make them, the contents of the grid is an average, so the only way to ensure absolute accuracy is to have infinately many infinately small cells (ie. it can't be done).
I was just thought it might be interesting if bayesian learning could sort of "spot" the innacuracies caused by the grid system. |
Re: Has this been tried for weather forecasting?
Weather forcasting is just a classification problem like any other, albeit one which is exceptionally difficult due to the large number of relevant features, and the extreme sensitivity to initial conditions. Hence most standard statistical classification techniques will be possible (Bayseian inference is just one of many), and I'd be very surprised if there werent already weather forecasting models which incorporated Bayesian methods.
Bear in mind how much more difficult weather forecasting is than spam detection though. The reason spam filters have a higher degree of success that weather forecasters isnt primarilly because they are using magical bayesian techniques - its because they are solving a far easier problem. |
Re: Has this been tried for weather forecasting?
Fair enough - like I say, it was just a "hmm, i wonder" moment. To be honest, I'd be amazed if it hasn't been tried, tested and either incorporated as "a good thing" or rejected due to being ineffective.
I've just never heard of any learning techniques in weather forecasting, not that that means there aren't any. |
Re: Has this been tried for weather forecasting?
Quote:
Obviously you wouldnt ever find such a simple "necessary and sufficient" criteria for spam in the real world though, which is why statistical methods are useful. But the idea of learning rules from a training set is fairly universal. |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
What I wondered is whether (given that the atmosphere is just a big ongoing calculation of the Navier-Stokes equations) it's possible to do computation by setting up appropriate wind conditions and then doing some later measurement, and then whether maybe you could upload your consciousness into the wind and somehow download it later by meditating on a warm day.
|
Re: Has this been tried for weather forecasting?
chaos :(
a mountain is just continually solving dv/dt = 0 with initial conditions v = 0 but you couldnt upload your mind to it :( |
Re: Has this been tried for weather forecasting?
|
Re: Has this been tried for weather forecasting?
Quote:
What is this if not a description of the human mind? As for the mountain, that just proves there are systems not supporting complex phenomena. There certainly are differential systems supporting computation - just take the equations governing electric potential in the circuit of an electronic calculator say. |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
Hmmmm, I thought they already used the data collected over the past 100 years to predict the weather.
Something along the lines of "Oh, high and low pressure here and there, tempuratures like that, about the same as on the 20th of Februari in 1990, and it rained the day after, so we predict rain for tomorrow". Roughly said... Ok, it's not learning, but comparing. But anyway, I don't think this learning thing exists because of the huge amounts of data that needs to be processed, and the lack thereof in certain areas. The weather system is very complex and the condictions around the entire globe influence the weather on a single spot. Even if you knew the weather conditions of every place on the planet, their interactions, how they influence eachother, are in my opinion too complex to be defined by a set of finite rules. |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
Quote:
Quote:
Quote:
edit: although if you just knew the weather conditions on this planet then you wouldnt know everything, so your predictions could still be screwed up by external influences like meteors flying past earth. |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
Quote:
Even the guys at the Enterprise couldn't predict the weather. Mr Data: Sensors indicate that the storm will reach the area in about two hours ... and hour later ... Picard: The storm is catching up, we can't take off, I thought it would hit in two hours Data: well... buggery <insert exciting episode about crew members hiding in caves and running around through a storm with stroboscope lightning and without getting wet> Hanibal: I love it when a plan comes together The End |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
Quote:
PS Chaos theory! |
Re: Has this been tried for weather forecasting?
Chaos theory hurts my head. I read a book about it once and went "muuuuuuuuuuhhhhh!" for days afterwards. Lots of stuff about ratios and "e" and how it crops up in nature a lot - i remember there being something about the factor a river bends being directly equal to it, or as near as makes no difference.
Fractals and water turbulence and, yes, some stuff about weather predictions figured too - in fact what I was saying about the problem with the weather grid system came from that book, now I think about it; there was a section on a weather simulation system giving different results for each run and things. I might try reading it again sometime and see if I grasp it any better. [edit] I'm almost certain it's this one: http://www.amazon.co.uk/exec/obidos/...464901-4231613 It's at my parents house so I can't tell for sure. |
Re: Has this been tried for weather forecasting?
Quote:
When I kick a football, the path it travels depends on a lot of factors, such as how hard I kick it, the wind speed, and so on. But a small change in one of these factors will not have much effect on the ball - if I kick it a bit harder, it will travel slightly further, but not much. If the wind is a bit stronger, it will curve a little extra, and so on. So this system isnt chaotic. But for something like the weather, even a tiny change in the starting conditions can produce huge changes to the weather. Therefore forecasting becomes difficult, because even if you correctly measure an initial condition to (say) 3 decimal places, the rounding errors you make in the lower decimal places can multiply rapidly and screw up your predictions. edit: for instance, I'm currently trying to get a recurrence relation to work in C, but its messing up because precision is being lost in the calculation of floating point numbers. The values should be 0.333333 for every memebr of the sequence, but instead I'm getting Quote:
|
Re: Has this been tried for weather forecasting?
Yeah, that's about as far as I already understand it - it's the specifics which intrigue me, yet I know I'm not really mathematically minded enough to "get" it. It's a bit like thinking about the universe - you see how all these mathematical "special" numbers somehow crop up in nature and go "woah".
Incidentally, you should be able to reduce the problem in your C example by using a double float (or a long float, not sure which is valid off the top of my head) but of course there will always be rounding errors. |
Re: Has this been tried for weather forecasting?
Quote:
|
Re: Has this been tried for weather forecasting?
You're not the only one. The only possibility I can think of is that float uses a different/better rounding system - maybe it's rounding something down that was previously rounded up, or vice versa.
Beyond that and without having the C/C++ doc on me, I don't know as I can't remember the differences between the data types. |
Re: Has this been tried for weather forecasting?
There are GCMs running that use bayesian calculations but I'm not sure of the specifics or whether it's applied how you were suggesting.
I don't know what journals people will have access to so it's probably easiest to post a google scholar link: http://scholar.google.com/scholar?hl...22&btnG=Search The ECMWF site probably has more useful info if you're really interested. |
Re: Has this been tried for weather forecasting?
without wishing to dumb down the thread as im enjoying it (and tried to pos rep nod for *i think* the first time ever):
**** me its ragnarak! |
Re: Has this been tried for weather forecasting?
Quote:
Try googling for pentium float error, or maybe pentium float divide error (that is, if theres a divide in it). I think thats where I found some info. Curse drugs and there amazing ability to destroy memory.... |
Re: Has this been tried for weather forecasting?
Quote:
(I've been in France for 5 weeks on an atmospheric sciences course) |
All times are GMT +1. The time now is 03:34. |
Powered by vBulletin® Version 3.8.1
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Copyright ©2002 - 2018