Earth Simulator Now Predicting Hurricanes? 167
GeoGreg writes "The BBC is reporting that the Japanese Earth Simulator supercomputer is producing results showing that it is possible to model climate down to the level of severe weather events such as hurricanes. This computer has been discussed on Slashdot previously, and it sounds like at least some of the hype around this beastie was justified."
Earthsim do cool things (Score:5, Informative)
The conclusion was basically that Japan would be f***'ed if such was to happen, but that's rant for another day.
So, earthsimulator simulates a lot of things. I am surprised that they don't model nuclear blasts on them, because it certainly CAN. Or at least we just don't know about it.
One thing is for sure, though - I will attest that NEC definitely made a bundle over this =)
btw, for ppl who are in japan, you can schedule tours to the place. I havn't tried yet, but in case anyone is interested... (now that I think about it, wasn't there a story about this a while back?) but here is a link just for fun: visitor information [jamstec.go.jp].
and if you are brave enough for the same page in japanese, click here [jamstec.go.jp]. (The japanese page has a japanese map, which shows station names in kanji. I always found kanji station names to be more help, but that might be just me...
Re:Output (Score:5, Informative)
There is a technique called ensemble forecasting, whereby you run multiple instances of the model with slightly disturbed initial conditions and/or slightly tweaked model parameters. You can then examine the statistics of the ensemble to try and obtain information a deterministic forecast might not be able to give you.
Note that the goal in this particular case is not hurricane forecasting as such. The newsworthy information is that this is the first time that a climate model can be run at a resolution high enough that hurricanes become possible within the simulation. Short term models used for the daily weather forecast do this reasonably well already.
Re:Earthsim do cool things (Score:5, Informative)
Re:Model (Score:4, Informative)
dictionary.com gives me...
Insure - To make sure, certain, or secure. (See Usage Note at assure)
Ensure - To make sure or certain; insure: Our precautions ensured our safety. (See Usage Note at assure.)
I'm still not sure who is correct here. Please don't make me diagram the sentence.
Davak
Not really predicting per se... (Score:2, Informative)
Re:No pussy-footing for NEC (Score:2, Informative)
Re:10km resolution (Score:4, Informative)
Therefore 10 cubic kilometers is the volume of 10 such cubes. For example, a volume of 10km*1km*1km is 10 cubic kilometers.
If you want a cube of 10 cubic kilometers, it would have a height (and width and depth, of course) of [cubic-root of 10]km, which is about 2.15km.
Re:10km resolution (Score:4, Informative)
Re:Nonsense in article (Score:1, Informative)
The tracks in the UK aren't cheap and lightweight - the defining characteristic that causes them to buckle is that the rail sections are longer than those on the continent, and have less in the way of inter-section spacing for the rails to expand into when they heat up.
These measures were, as I understand it, introduced to reduce maintenance time on tracks and carriages, improve ride quality and increase efficiency generally.
The decision to base the tolerances round a (relatively) low maximum temperature is the mistake, if there is one. That said, rail buckling problems due to ambient temperature are quite predictable, and in the grand scheme of things, all it meant was some delays in hot weather, comparable to those we already had on the roads.
Not the end of the world, I think, and it's not entirely fair to imply rail buckling is a negligent planning issue (cheap and lightweight).
Now, on all the other negligence, you will get significantly less argument from me...
Re:Model (Score:2, Informative)
NCEP Skill Scores [noaa.gov]
If this doesn't convince you that much research is currently being done to improve weather prediciton, here is the fields most recent effort, the WRF model, a collaborative Operational/Research model.
WRF Model [wrf-model.org]
Keep in mind, the model can only resolve a solution near that of the actual resolution of the input data, i.e. observed conditions. This is why weather prediction is still an evloving science. Not only must these supercomputers solve the non-linear multivariate/ multidimenional equations governing the atmosphere, but scientist must also devise methods to quickly, and more importantly, accurately input the most recent data so the products can be made in a reasonable time for the public.
Re:Model (Score:4, Informative)
First of all, that is already happening with current weather models...those are the ones that predict hurricane paths and such. There were already predictions that this would be an unusually heavy hurricane season before it started - those were due to climate models that showed the ocean area responsible would be warmer than normal.
Predicting where hurricanes will appear and where they will go ahead of time (that is without looking at the current weather patterns while it is happening) involves that pesky chaos thing and good luck with that.
Perhaps what the person was trying to say is that this is the first time researchers have been able to run 10 km. (or 5, or 1) resolution models on a global scale all at once - and that is quite an achievement if so.
BTW, the point of all this is not to predict individual hurricanes or tracks. It is primarily to identify long-term climate trends. From the article:
"This means that we potentially have the capability to predict whether storms like Hurricane Isabel will be on the increase in future." - Professor Julia Slingo. (Hmmm, I guess she's from Soviet Russia;)
Re:Who needs a supercomputer? (Score:3, Informative)
small differences in the initial data manifest themselves as large-scale phenomena down the road.
A good point to emphasize when people are blindly clamoring for more computer power.
Lyapunov exponents [mpipks-dresden.mpg.de] for portions of phase space for a nonlinear system will cause this divergence.
So, yes, no matter how many bits of floating point mantissa you carry, or how precise your measure your initial conditions, exp(at) will inevitably grow if a > 0.
And, just in case anyone's proud of their accurate code and precise initial conditions, there's still Heisenberg to prevent you from measuring too much accurate information at the same time.
IIRC, someone once determined that a pencil, balanced on its point, would always fall within 22 seconds because of the inherent uncertainty in position and momentum that could be established initially.
Americans DO know the least about the climate (Score:3, Informative)