I've read or skimmed most of the Climate Science info in the cp.net webpage. It seems quite naive, if you ask me. Mostly atmospheric hydrodynamics, with condensation modeling, and albedo of land and cloud surfaces determining energy input -- radiation. I was wondering about seasons -- why is summer (assuming Oxford summer) hotter on the mean than winter for the whole globe? Greater water surface in the southern hemisphere?

Anyhow -- I would assume that parameter validation would look at local conditions -- perhaps historical data for a few hundred grid squares, with RMS variance from the model used for validation. Rain in the Sahara, or low wind velocity in the North American Steppes, for instance, would seem to indicate an invalid set of parameters.

And -- snow albedo being such an important component of the model, some glitch allowing snow retention in prolonged 20C weather would throw off the model regardless of the parameters. Also -- does snow accumulate over liquid water, or only atop ice and land? Snow code -- it would seem -- is inadequate in the model and ought be modified. Here is a picture from my current run

http://www.geocities.com/bob_corey/Late_May_Snow_Anomoly.gif

---

The model seems to be just starting phase 1, some models take a bit to settle down. Phase 1 is for calibration and spinup, so don't worry, if the model has really unrealistic parameters, it will be unstable in phase 2 and so won't be counted in the final results.

cpdn have not said what their model evaluation criteria are for experiment 2 yet. However, they are cooperating with the Hadley Centre on it and judging by the Hadley Centre's model evaluation (the skill of the model) the criteria used will be pretty sophisticated.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

I went to the Hadley website after posting this. It's some interesting stuff there. They have the makings for some pretty exhaustive modeling that considers biosphere changes and the like. I tried to understand from the abstracts and summary information there what's all involved in cloud physics which seems to play a big role in cp.net. I gather much effort has been made at empirical parameterization -- taking actual measurements. Perhaps snow melt is involved in the pertubations -- I saw that melt had finally occured by mid-july. I can see this leading to positive feedback/instability. We'll see when it gets to Phase II.

So -- perhaps not a glitch in the model itself, but in the parameters I'm running. I hope that's the case.

---

&gt; I've read or skimmed most of the Climate Science info in the cp.net webpage.
&gt; It seems quite naive, if you ask me.

Which stuff is naive? If you mean the basic science outlined on the cp.net site, then sure. We've always intended to have layers to that, but never really got round to writing deeper materials. If you mean the actual code in the model, I'd be interested in what you're comparing it to, in order to arrive at "naive".

&gt; Mostly atmospheric hydrodynamics, with
&gt; condensation modeling, and albedo of land and cloud surfaces determining
&gt; energy input -- radiation.

Our description focusses on cloud formation because this is a major source of uncertainty in climate models. There are loads of processes we don't describe on our site but are in the model (6 different sorts of boundary layer types; radiation absorbed by various different cases; 4 sorts of convection...). The pages on our site were initially built for general participants, and later for students, too. Though I'd like to see some deeper material available for people to browse as they get more into the project, I'm not sure quite when it would happen (or who would do it - probably has to be a labour of love for someone, rather than something we could get funding for).

&gt; I was wondering about seasons -- why is summer
&gt; (assuming Oxford summer) hotter on the mean than winter for the whole globe?
&gt; Greater water surface in the southern hemisphere?

Yeah - the land surface responds more vigorously than the oceans to changes in radiation, so the Northern Hemisphere responds more sharply in summer than the Southern (even though we're nearer the sun in January than we are in July).

&gt; Anyhow -- I would assume that parameter validation would look at local
&gt; conditions -- perhaps historical data for a few hundred grid squares, with RMS
&gt; variance from the model used for validation. Rain in the Sahara, or low wind
&gt; velocity in the North American Steppes, for instance, would seem to indicate
&gt; an invalid set of parameters.

We're currently working on building a method of developing a (reasonably comprehensive "model skill score" (based on work undertaken by the guys at CMIP-2 and our friends at QUMP in the Hadley Centre). The idea is to take the sorts of data we currently have available and use these to constrain the models in their control phase. We know it doesn't rain in the Sahara at pre-industrial or modern levels of CO2, but we're open to different futures. [In the limit, if a model does a sterling job of predicting most climate variables under present and recent past conditions, but predicts something quite different in the future, then that's an interesting result, and one that current data wouldn't reject out of hand.]

Dave

---

Thanks for the post Dave

One thing not covered by your answer is what you think of this:
<img src="http://www.geocities.com/bob_corey/Late_May_Snow_Anomoly.gif">

Is this likely to be due to strange parameters,
something wrong with the modelling of snow,

a third possibility is a display glitch with the model working fine,
or something else?

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

Bob one thing you may want to try is to download MartinSykes CPview. See

<a href="http://www.users.globalnet.co.uk/~sykesm/cpdn.html">Martin Sykes web site</a>

This may give us a better idea of whether this may just be a display glitch.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

&gt; We know it doesn't rain in the Sahara at
&gt; pre-industrial or modern levels of CO2, but we're open to different futures.

Strangely enough, for a few thousand years after the last ice age the Sahara was wetter than it is now at approximately the same global temperatures, but slowly dried out. See the wikipedia <a href="http://en.wikipedia.org/wiki/Sahara">entry</a> for an overview.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

&gt; Which stuff is naive?

I'm neither a climate scientist nor a code geek -- I'm talking mostly from a viewpoint of what gives confidence in a model to a non-expert. And stability is certainly a criteron. Some resemblance to regional patterns in historic setting seems what a "consumer" of the model would like to see before trusting its conclusions about the future. I would want to know the phase III variation of that subset of stable phase II runs (20%) which meet some broad criterion for resembling the world we observe. Without such a cull, I don't place any faith in especially the more alarmist outcomes of Phase III calculations. They may indeed be valid predictions of possible futures, but stability alone is not validity.

&gt; I'd like to see some deeper material available for
&gt; people to browse as they get more into the project,

I must blushingly admit that I enjoyed SimEarth from MAXIS on the NES, maybe SNES many years ago. You speak of what I imagine are quite refined models of energy flux, at phase boundaries, by spectral slices, with spectral variations of response by various materials. I saw some of that at Hadley and am duly impressed. I keep picturing biosphere changes as essential, though. Snow cover is not the only seasonal variation in spectral response. Vegetation has its seasonal cycles. It also has a climatic response, which understandably need not be considered in a first look. I wonder too, in the hydrodnamic modelling what consideration is made of orography on such a coarse scale.

This really doesn't address the question of the content of the "Climate Science" section of the site, which I did find informative, I assure you. Links to some monographs could help those who wish to look deeper. With all the emphasis on cloud formation in the model, though -- some basic stuff on condensation dynamics, entrainment, rain formation, etc, and how the parametric pertubations reflect different assumption would be nice to see. I don't understand the process much at all. Maybe I missed it in my browsing.

Thanks for clarifying the seasonal variation of the mean. And your last paragraph well addresses the concerns of the first paragraph in this reply. Stronger validation criteria for historical runs will indeed give more confidence is predictive runs, while one may still see considerable variation.

I notice that Hadley posts a map predicting the desertification of Guyana. I didn't check out the fate of Southern NZ -- but a friend in Christchurch says its unwise to go outdoors without a hat. You seem to know especially well the impact of man on global conditions, like the Antarctic ozone hole.

---

I've noticed "snow" before in warmer areas of a run and always assumed it was an oddity of the large grid squares. For example, right now I'm on the southern California coast, it's about 80 degrees F on the beach but I can see the mountains (within my "grid cell") and they are covered with thick snow. So if you "averaged out" that would be some snow in this "grid cell." Since your pic is the Appalachians I imagine that could be an answer (note I'm a computer geek and not a climate/atmos physics geek :-)

&gt; So if you "averaged out" that would be some snow in this "grid cell." Since
&gt; your pic is the Appalachians I imagine that could be an answer (note I'm a
&gt; computer geek and not a climate/atmos physics geek :-)
&gt;

Likely not Carl. At least not at the end of May as the date shows. It also goes along the Gulf Coast into Texas with no discernable topography in that area. I also wonder how well the elevation of the Appalachians are resolved by the grid spacing for the climate model. Seems more like to be a display glitch, or some quirk with the parameters associated with snow. Would have been interesting to see the surface temperature at the same timestep.

As I read this, snow has accumulated in the same are at the end of March 1814. Temperatures seem to vary from 0 to 15C I've not seen them dip below 0 (green color) south of the Carolinas any of 3 consecutive nights.

I wonder -- given 1000 models, and 20 "judges" from different areas of the world, how many models would pass the scruitiny of judges saying "Hey! it's not like that where I live!" Perhaps it's just my prediliction to cull results, But if 50 pass the scruitiny of 18 judges, I'd put my computing resources into finer grid/time scale or more computationally complex variants of those 50. Maybe that gets past what distributive computing can do. And maybe I'm naive. I'm learning that sometimes "tweaking" adjustments of models or fine tuning of algorithms can produce large differences using the same assumptions (parameter suites).

---

I have the impression that at the moment they want to keep lots of models in. In later experiments they will want to cull model versions. Preferably this should be done in an objective manner so that a probability distribution of the remaining models can be considered objective.

Have you read the paper in nature? (see news from home page).

They have tried a RMSE approach and most models were nearly as good as each other and in gereral better than the CMIP models. Yours does seem strange, but does this make it an exception?, or are most models equally as bad?, or is this just a display bug with the model actually calculating sensible values?

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

&gt; I wonder -- given 1000 models, and 20 "judges" from different areas of the
&gt; world, how many models would pass the scruitiny of judges saying "Hey! it's
&gt; not like that where I live!" Perhaps it's just my prediliction to cull
&gt; results, But if 50 pass the scruitiny of 18 judges, I'd put my computing
&gt; resources into finer grid/time scale or more computationally complex variants
&gt; of those 50.

ClimatePrediction.net is currently producing about 300 models a day. To evaluate a model an judge would have to look at the temperature, pressure, rainfall and cloud cover for the last few years of phase 2, a process that could not realistically take less than 10 minutes. That is 30 evaluations a day, assuming the judge would do other things for a few hours (meetings, read e-mail, etc.) So it would take 200 judges just to keep up with the current number of models generated. With the current surge I would expect the number of models a day to reach 500 before long. There are 3,000,000 different combinations of parameters so it would take over 6,000 man-years of work to evaluate them all.

After the first few days it would be mind-numbing work; believe me, I've done similar evaluations for speech retrieval experiments. The judges would require a high level of expertise and would be expensive to employ - probably over $20M/year.

Worst of all human judges would not apply consistent standards, if a judge saw a once in a thousand year event in the results should he throw it out? What about a once in a million years event? **** The results from such a study would be appropriate for a lesser known social science journal, but not for Nature.


**** I estimate that there is a probability greater than 50% that each "one in a million" weather event will be seen in at least one model purely by chance. So is May snow on the gulf coast a one in a million years weather event? I don't know and it is probably very difficult to compute accurately what its probability is, but if it is then you are possibly just "lucky" to have seen it.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

I guess with 30,000 trickling machines, 300/day sounds about right. Lacking human 'judges,' by the constraints put forward -- lacking any validation of variant models, I'd be inclined to accept maybe .5 sigma of the indicated variation, not the whole range or 2 sigma interval. Still, a 3C - 5C range of temperature rise, while not so alarmist as 10C, is cause for concern, that I wish my government would take seriously. My countrymen obviously have other priorities, such as being ready for the rapture, so they can watch from heaven the rest of us deal with circumstances which experience to date leaves us unprepared.

I'm idly reading some of other papers reference from this site and around the internet.... for one thing I see that late/slow snow melt has been an issue with GCM's in the past. I still don't know if what I'm observing is parametric or systemic. It seems modeled moisture entrainment was affected as much as surface albedo by the anomaly.

I get indications that physical measurements of cloud physics are indeed uncertain -- even indicated cloud density seems dependent on measurment technique. I'm reading a report now on GCM validation....

http://www.cru.uea.ac.uk/~mikeh/research/mhfinal.pdf

If I understand what CMIP means, I don't trust it. How are models to be validated by comparison to one another? I probably don't understand the concept.

I see that climate study must recognize the possibility of 1000 year events, that aren't necessarily invalidating. My theory of cull, as an engineer, has always been to recognize that a selection of any number of choices is likely to reject an equal number of valid or even superior choices, while retaining a large number of dead end choices. But an engineer has different objectives, which are not necessarily unmet by failure to recognize the best possible option.

---

Hi Bob

By now your model must have done a few more years of phase 1. Is it showing more usual weather for spring/summer of subsequent years?

---

__________________________________________________

&gt;
&gt; If I understand what CMIP means, I don't trust it. How are models to be
&gt; validated by comparison to one another? I probably don't understand the
&gt; concept.
&gt;

Sorry I have probably confused you there. There may be methods of validating models by comparison to other models. I do not understand such techniques either.

What I was trying to say was that on a RMSE comparison to actual observations, the climateprediction models were in general better than the CMIP II models. The graph in the paper says relative to unperturbed model but I think this is just indicated that the score of all models has been changed by a factor to make the score of the unperturbed model equal to 1.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

Yes -- 4 years into Phase I, I still see anomalous?? snow accumulation and retention in the SE USA.

My questions regarding validation are answered in a more careful re-reading of the Nature paper, largely in Figure 2b. Indeed, Phase II results were compared with observed values, and, as crandle points out, are quite sound through the whole ensemble relative to a variety of GCMs. Though local resolution isn't explicitly examined -- it it validated by the RMSE calculation.

And on second thought, I do see value in intercomparative validation of models -- to see if reducing the computational complexity (by time and grid scales, simplified algorithms, etc) still retains utility in its conclusions resembling those of the 'definitive' model. Anyhow -- I've put my feet in my mouth quite enough. I'll put them in snowshoes -- gonna need them through the month of May in sunny, warm, snow-bound Virginia.

---

Hi again Bob

You seem to be running two models in parallel on the same machine, and they're both at a similar stage in phase 1. Is the other model showing more usual weather?

When the phpbb forum is up and running again, you will see that there is a special section for people to flag up unusual runs, presumably to let the researchers know. You will have to post yours - I don't think that this May snow in Carolina has been seen before.

---

__________________________________________________

Just a thought, are there unusually low temperatures on the eastern coasts of any of the other continents?

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

I'm sure most of us might find things in the models we're running that strike us as odd. I notice that Central Asia gets a snow cover through most of Russia and half of China, but not the pacific coast or Europe. What bothered me initially was also snow cover on oceans at what I would think are low latitudes for ice. I did notice the same snow anomoly -- not so pronounced, around Bangaladesh. Not much happening in the southern hemisphere. And yes, both models show the SE US getting much snow, and retaining through much of the spring months... An example from the other model... it's not low ground temperatures at issue...

http://www.geocities.com/bob_corey/July_Snow_Anomaly.gif
http://www.geocities.com/bob_corey/July_Snow_Anomaly_Temp.gif

There is a 9 hour difference between the shots, but I assure you that temperatures through the night did not dip below 10C, probably not 15C.

I'm convinced, since no one is responding to this thread with similar observations that the cause is parametric, not systemic. But what about snow on the sea off the coast of Portugal?

---