Latest Supercomputers Enable High-Resolution Climate Models, Truer Simulation of Extreme Weather
Berkeley Lab researcher says climate science is entering a new golden age.
Not long ago, it would have taken several years to run a high-resolution simulation on a global climate model. But using some of the most powerful supercomputers now available, Lawrence Berkeley National Laboratory (Berkeley Lab) climate scientist Michael Wehner was able to complete a run in just three months.
What he found was that not only were the simulations much closer to actual observations, but the high-resolution models were far better at reproducing intense storms, such as hurricanes and cyclones. The study, “The effect of horizontal resolution on simulation quality in the Community Atmospheric Model, CAM5.1,” has been published online in the Journal of Advances in Modeling Earth Systems.
“I’ve been calling this a golden age for high-resolution climate modeling because these supercomputers are enabling us to do gee-whiz science in a way we haven’t been able to do before,” said Wehner, who was also a lead author for the recent Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). “These kinds of calculations have gone from basically intractable to heroic to now doable.”
Using version 5.1 of the Community Atmospheric Model, developed by the Department of Energy (DOE) and the National Science Foundation (NSF) for use by the scientific community, Wehner and his co-authors conducted an analysis for the period 1979 to 2005 at three spatial resolutions: 25 km, 100 km, and 200 km. They then compared those results to each other and to observations.
One simulation generated 100 terabytes of data, or 100,000 gigabytes. The computing was performed at Berkeley Lab’s National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science User Facility. “I’ve literally waited my entire career to be able to do these simulations,” Wehner said.
The higher resolution was particularly helpful in mountainous areas since the models take an average of the altitude in the grid (25 square km for high resolution, 200 square km for low resolution). With more accurate representation of mountainous terrain, the higher resolution model is better able to simulate snow and rain in those regions.
“High resolution gives us the ability to look at intense weather, like hurricanes,” said Kevin Reed, a researcher at the National Center for Atmospheric Research (NCAR) and a co-author on the paper. “It also gives us the ability to look at things locally at a lot higher fidelity. Simulations are much more realistic at any given place, especially if that place has a lot of topography.”
The high-resolution model produced stronger storms and more of them, which was closer to the actual observations for most seasons. “In the low-resolution models, hurricanes were far too infrequent,” Wehner said.
The IPCC chapter on long-term climate change projections that Wehner was a lead author on concluded that a warming world will cause some areas to be drier and others to see more rainfall, snow, and storms. Extremely heavy precipitation was projected to become even more extreme in a warmer world. “I have no doubt that is true,” Wehner said. “However, knowing it will increase is one thing, but having a confident statement about how much and where as a function of location requires the models do a better job of replicating observations than they have.”
Wehner says the high-resolution models will help scientists to better understand how climate change will affect extreme storms. His next project is to run the model for a future-case scenario. Further down the line, Wehner says scientists will be running climate models with 1 km resolution. To do that, they will have to have a better understanding of how clouds behave.
“A cloud system-resolved model can reduce one of the greatest uncertainties in climate models, by improving the way we treat clouds,” Wehner said. “That will be a paradigm shift in climate modeling. We’re at a shift now, but that is the next one coming.”
The paper’s other co-authors include Fuyu Li, Prabhat, and William Collins of Berkeley Lab; and Julio Bacmeister, Cheng-Ta Chen, Christopher Paciorek, Peter Gleckler, Kenneth Sperber, Andrew Gettelman, and Christiane Jablonowski from other institutions. The research was supported by the Biological and Environmental Division of the Department of Energy’s Office of Science.
# # #
– See more at: http://newscenter.lbl.gov/2014/11/12/latest-supercomputers-enable-high-resolution-climate-models-truer-simulation-of-extreme-weather/#sthash.HIQAHanC.dpuf
This model has a 2.5% error in energy transfer by water evaporation from tropical seas, http://judithcurry.com/2013/06/28/open-thread-weekend-23/#comment-338257
It’s the only way they could keep the water vapor feedback negative.
Dammit… I meant positive.
Water vapor feedback is whatever they say it is at that moment.
Nothing more and nothing less.
On the other hand it means that predictions become unreliable after 2 days.
on an initial state, iterate a 0.025 error factor several hundred times over many thousands of grids. what is output state? Garbage.
Curious George: This model has a 2.5% error in energy transfer by water evaporation from tropical seas, http://judithcurry.com/2013/06/28/open-thread-weekend-23/#comment-338257
that’s not the half of it. You only refer here to the latent heat of vaporization. What is necessary for accurate modeling is the rate of evaporation, and the change in that rate as temperature increases. (followed by accurate modeling of cloud cover, et seq.)
It seems obvious that if they set the latent heat too high, and they have the energy input rate correct, then they must have the evaporation rate too low. Getting from there to a quantitative assessment of how much error, accumulated across the distribution of regions and temperatures, looks intractable now. And if the temperature response to a doubling of CO2 concentration is less than 1% of the baseline temperature, on average, it would seem that a 2.5% error in the latent heat is a non-negligible error.
But hasn’t water vapor been declining for many decades?
Do they also pull most of the energy out of the water, which lowers its temperature, as it actually happens in the real world.
I particularly loved the closing line of the graphic: “Data is freely available”. Oh boy. I’ll just go and get my old desktop out and give it a run….
This is not going to make any advances in the prediction of climate because this product like all the others before it will once again be based on incomplete , inaccurate ,as well as missing data.
This is a waste of time ,money and effort which is the norm when it comes to climate science.
The blind leading the blind and that is essentially what climate science is at this juncture.
You said incomplete, inaccurate, and missing data. This seems to imply no intent. I think fraudulent data more accurately describes much of the intent of so called climate science.
More like the cunning leading the blind.
“Clouds aren’t important unless you have money to give us to study them then they are very important. But we will need a bigger computer…”
“…models will help scientists to better understand how climate change will affect extreme storms.”
That seems exactly backwards to me. It’s like suggesting models can help better understand how forests cause trees.
No no. Climate change is not the same as a changing climate. Climate change is code for man’s Satanic gases.
Why do they need the new computer models when the old ones were 100% accurate?
Simple. There was still more funding in the kitty.
There are two fundamental problems with this PR.
First, to even attempt to model convection cells (clouds, Tstorms) the grid resolution has to be 10km or less, not 25. Better is still not near good enough. This run still took 3 months to simulate 26 years to 2005. So a single run at 25km grid would take about 15 months to reach 2100 for ‘CMIP 7’. (And still not have the physics, only parameterizatioms.) Still incomputable, let alone for ensembles. IPCC AR5 said this also, using the PR example here of clouds. WG1 7.2.1.2.
Second, finer resolution just increases the initial value problem. So the nonlinear dynamic model results just diverge more and more rapidly. And attempting to convert that mathematical fact to a boundary envelope problem via ensembles is impossible owing to the computational constraints. See problem 1.
Both issues illustrated in essays Cloudy Clouds and Models all the way Down in Blowing Smoke.
Give the supercomputer to the NWS to do real weather forecasting out a few days on regional fine scales. See Cliff Mass blog. Stop wasting taxpayer money on shiny toys that are inherently not fit for purpose.
Why do we spend ever more money on supercomputers and other methods to attempt to predict the unpredictable. So far the models have produced no better then a guessed line or curve. The funniest part is that the line following the 5000 year trend tends to be closest. Apart from that there is far too short a range in all accurate data the satellite record doesnt even manage 50 years yet and doesnt span the warming/cooling periods experienced since the LIA, the predictions from all these data sets is therefore flawed right from the start.
If that wasnt inaccurate enough how can you take a piece of new data like the greater depth readings from buoys and with absolutely no historical data to speak of use that information to predict the future.
In 100 years time maybe we may make some headway into how all the forcings/feedbacks fit together, even then I doubt prediction of the future weather will improve much – its just too chaotic.
Shouldnt all this money be used in streamlining the human race by improving energy efficiency and embarking on a mild depopulation programme worldwide – now that would have benefits!! not some mega terabyte super ruler
Y mean it takes them years to produce one load of crap? I can do it in about six hours
More likely to be the Iron Pyrite Age of Climate Modeling.
A thousand years ago, there was big money in alchemy. Apparently everything really does come back around.
You have to admit, they do a good job of turning lead into gold.
(RoHS compliant, I assume?)
And in the end this matters how? Another stupor-computer climate model that does what in comparison to what? This is nothing more than a shameful waste. I can see starting a climate model out with real world data collected with some semblance of conformity and consistency but that data would be, what, maybe twenty years old? The earth is how many years old? Twenty years worth of usable data means what.
Same case with green house gas emissions, folks have thirty years worth of data on a feature of earth’s atmosphere that we have no idea how old it could possible be. I think that scientific study in the end is a good thing and developing models is of paramount importance but the idea of attempting to regulate society with so little data and such poor tools and systems of measurement is nothing more than hysterical political tripe.
In all the time that model ran it never got dark, the sea ice never budged, and as noted above the snow never fell. Reflects observation? I think not.
See how nice it is, living in the land of models??? 🙂 Don’t like something?…just delete it.
I think this is just wind current modeling over a static surface map.
One thing that seems missing is circumpolar currents, but I’m not really certain what they were trying to visualize.
” “In the low-resolution models, hurricanes were far too infrequent,” Wehner said.”
How infrequent is infrequent?
The last major hurricane to strike the United States made landfall on Oct. 24, 2005.
Far too infrequent as in most low resultion models can’t spontaneously generate hurricanes.
High-resolution model bias takes a lot of money and computing power.
Look, faster, better computers are always nice. The trouble is order complexity of algorithms. A problem which can’t be solved in polynomial time is basically intractable, doesn’t really matter how powerful the computer is trying to solve it. Simulations are always going to be an approximation that are going to diverge from reality by the nature of the math involved and the parameterization of very complicated things, running on a better computer isn’t going to alter that fundamental reality.
Computer science majors learn this as freshmen or sophomores.
+1
“Gee-whiz science”? More like Cheez Whiz science. Lots of fluff and pretty, but no substance to it.
As a plod who has the odd three decades of coding and systems experience, I wonder what they are ACTUALLY running on multi-billion dollar massively parallel Beowulf cluster, because the latest GISS model (GCM ModelE) 5.x doesn’t support such an architecture. See, being a pedantic software engineer of the olde school, I downloaded and compiled the beast, and read ALL of the VERY CLUNKY FORTRAN. Where has the $50 BILLION gone?
From the system specifications:
“Note that the parallelization used in the code is based on the OpenMP shared memory architecture. This is not appropriate for multi-processing on a distributed memory platform (such as a Beowlf cluster). For machines that share a number of processors per board, the OpenMP directives will work up to that number of processors. We are moving towards a domain decomposition/MPI approach (which will be clear if you look at the code), but this effort is not yet complete or functional.”
http://www.giss.nasa.gov/tools/modelE/HOWTO.html#part0_3
Anyone interested you can look at the code on my site. I put it all in one convenient place.
https://github.com/addinall/GISS_climate_model
Thinking about re-writing it in a sensible language. Reverse engineering, at a glance, will show how trivial this model is.
The code will run correctly on something like a SunFire with 16-32 CPUs on the same MOBO, but it WILL NOT make use of a parallel architecture.
“ModelE uses OpenMP application program interface. It consists in a set of instructions (starting with C$OMP) which tell the compiler how to parallelize the code. To be able to run the model on multiple processors one has to compile it with enabled OpenMP. This is done by appending MP=YES to the compile line, i.e.
gmake gcm RUN=E001xyz MP=YES
It is important to keep in mind that one can’t mix OpenMP objects with a non-OpenMP compilation. This means that one has to do gmake vclean when switching from OpenMP to non-OpenMP compilation. The option MP=YES can be set in ~/.modelErc file (see Part 0.2). In that case one can skip it on the command line.
In the setup stage, the number of processors is defined by the relevant environmental variable $MP_SET_NUMTHREADS. However, when using runE, it is similar to the way it is used for the non-OpenMP model, except that one has to specify the number of processors as a second argument to runE. For instance if one wants to run the model E001xyz on 4 processors one starts it with
runE E001xyz 4
” […]
Now, I am not impressed by “I can run me program an it produces 12 zillion petabytes of random numbers, all falling within pre-ordained upper and lower bounds” when the data for the priming of the model is sparse indeed.
“Boundary and initial conditions for the AR4 version can be downloaded from fixed.tar.gz (191 MB). This is a large amount of data due to things like transient 3-D aerosol concentrations etc.”
http://www.giss.nasa.gov/tools/modelE/
Boundaries and initial conditions for the WHOLE PLANET can squeeze into 191 MB???? Amazing!
My temperature start data for one automatic lathe under Statistical Process Control is about thirty times more complex than the WHOLE PLANET! Me must be doing it wrong!
Can I please have the super-computer to do some REAL WORK. Proteomics and Epigenetics could stand a nice new shiny machine(s).
Shhhh, you’re going to ruin it for them…
Is the GISS Model E same as CAM5.1? I thought that this post was about CAM5.1.
They are not the same. Every time I hear about the “gold standard” of models, GISS is trotted out. That is what peaked my curiosity initially. The code is bloody old and awful, and will run OK on an XBOX.
They have altered the run from CAM4.x to CAM5.1 so that now CAM5.1 can copy (close enough) the real observed data. Warming over the 20th century has dropped from 0.84C mean to 0.35 mean (both meaningless numbers).
CAM5.1 can make use of parallel operation:
“CAM makes use of both distributed memory parallelism implemented using MPI (referred to throughout this document as SPMD), and shared memory parallelism implemented using OpenMP (referred to as SMP). Each of these parallel modes may be used independently of the other, or they may be used at the same time which we refer to as “hybrid mode”. When talking about the SPMD mode we usually refer to the MPI processes as “tasks”, and when talking about the SMP mode we usually refer to the OpenMP processes as “threads”. A feature of CAM which is very helpful in code development work is that the simulation results are independent of the number of tasks and threads being used.”
But not very well.
The software was not written with parallelism in mind as an architectural model. The existing shared memory serial code has been hacked around. That is probably why a run is taking months!
They would be better served (sic) writing some decent software rather than buying shiny new boxes.
No. CAM 5.1 and GISS Model E are two different models. CAM 5.1 is actually well done as far as computer models go. Great documentation and implementation. Hats off to NCAR.
GISS Model E is a piece of old FORTRAN junk that should not even be compared with CAM 5.1 or other state of the art climate models. Gavin doesn’t have time to document it properly so no one really knows what’s in it…
Mosher: “The latest GISS model has exactly Nothing to do with this post. good engineer. not.”
Are you implying that the REAL model code is “classified” so that it is safe from programmers with 20-30 years experience who might find something wrong with it?? Shades of climategate. And yes I saw the list of lame excuses for doing this.
The latest GISS model has exactly Nothing to do with this post.
good engineer. not.
Maybe it’s about time ‘Theoretical Climate Science’ was treated as a distinctly separate discipline to ‘Actual Climate Science’.
My experience with numerical weather prediction models is that the higher-resolution models (both in the horizontal and vertical) are not necessarily any more accurate with their projections than the lower-resolution NWP models.
ECMRWF was days ahead of NWS on Hurricane Sandy. Go to the Cliff Mass blog for lots of specifics and details. Point is not that finer resolution models are perfect (that darned Lorenz nonlinear dynamics effect), but often are practically better. Essay Models… In Blowing Smoke uses a concrete example for an Arizona thunderstorm weather front. Real data from an event with both radar and weather models of varying grid fineness.
I agree that the ECMWF model is a good model (better than the GFS), but I’ve still seen it (ECMWF model) depict totally anomalous features at 240 hours that never actually materialize.
The new NAM model has a relatively fine horizontal resolution (4 km), but is really no better than the earlier 12 km NAM, based on my experience with working with it on essentially a daily basis.
Gigi – just a whole lot faster. One thing I know is if they ever get a model together that comes near to reality I’ll hear of it from WUWT.
I believe the “golden age” he is speaking about is the grant money deposits in their bank accounts.
You have it in one!
Golden age of stupid, he means.
I think its cute in a really scary way that anyone would claim climate models are gaining accuracy when we do not even know what drives ENSO yet, which caused rates of change far surpassing the ones we are supposed to fear. We have talk of an energy im balance yet cant explain, or even highlight a source for a massive transfer of energy.
We totally understand the climate! Except the predicted rate of change needed for the dangerous ends of the claims isnt even vaguely coming into fruition, and major energy movements happen without us even have an inkling of what triggers it. Science is settled! Well unless you look at the published work where basically every variable is still up for debate. hmm. no more questions!
The video shows Australia as not having ANY cyclones for the entire period.
The Oz BOM show so many cyclone tracks over the same period that it looks like the spaghetti graphs typically shown for climate mpdel runs.
Ah. That must be the LOW-Res model you’re looking at. The one with not enough storms.
A computer model, no matter how powerful, is no better than the programming and the input data. A computer is not a magical machine.
I believe that a computer IS a magical machine. A program with input data but without a computer is only a dead model.