Mark T.
I was aware of the truth of your comments before you posted them and agree with them in substance.

However, I was not discussing the employment functions of Mann (I tried to keep it broad with “Mann et al”). My concern was more with quality. I would sooner trust software that makes you think of Wernher von Braun than 20 year-old program versions stacked on versions. We have example after example of the problems Steve meets in reverse engineering old software. It’s quality that is the issue. Why not do the surveying exercise I sketched above and surprise yourself with the results?

Here are the inputs. What is the main Great Circle distance, to the nearest meter, from a famous Oxford University place at 51 deg 45 min 12.340 sec North; 1 deg, 15 min 14.460 sec West, to a nostalgic place at Sydney University, 33 deg 53 min 22.140 sec South; 151 deg 11 min 15.510 sec East? The WGS-84 geoid can be used.

Why not all run it routinely in any canned program you like and submit your results on the Message Board on November 1st, 2008, 1000 hours Greenwich Mean Time, so that people do not peek at each other’s answers first.

BTW, I do not concur that this exercise is irrelevant to climate work. If you use a routine to find nearest neighbours in a dense data field, some of the answers might be close (or equal in calculation)and you have to be sure which one to use. If you don’t carry enough figures, you can choose one place one time and another place another time, depending on the order your particular program defaults to. If you then try to match outcomes, you get a glitch and have to waste time tracking it down.

I actually bought my own professional copy for my MS thesis back in 1994. Back then, the student version sucked, but you could buy a pro version for $500 if you could prove academic intent. I could, and did. Now, the student version is identical to the pro version other than some nits about printing graphics and publishing stuff, but it still sucks because it doesn’t run on 64-bit platforms! I am fortunate that I have a copy from work and licenses are user based, not location based.

Mark

I would submit that R is more “latest” than Matlab

Is that an objective view based on extensive experience of using both there Steve?

Having used both, I find R and MATLAB are doing very similar things. To argue one is the latest and greatest seems a bit odd. Checking wikipedia (caveat emptor as usual) suggests both were formed in the second half of the 1970s (making each circa 30 years old) for more or less the same reason – to provide access to the extensive and powerful FORTRAN libraries through a simple and intuitive language more closely linked to how the maths would be written and a bit more divorced from the nuts and bolts of computer languages.

Both do this pretty well to be honest. I have limited experience in R but I can see some areas do seem better but other areas are stronger in MATLAB. The main difference is likely to be that one has a more distributed base and free, whilst the other is more centrally controlled and costs. To someone working at home, who wants to keep costs down to a minimum, R makes more sense. To a corporate environment, the support and consistency provided by a central design authority means MATLAB is probably going to be preferred. I also note that R is a bit stronger on the statistics side and MATLAB is a bit stronger on the engineering and control side (although each can play these roles).

PS. It is worth noting that there are a large number of “packages” available for MATLAB – take JEGs mapping library as an example. The Mathworks also have a pretty extensive upload site (although the quality can be variable! They do include reviews) Although if there is a built in option for MATLAB I usually prefer to use that on the basis of quality control, consistency etc.

Mark

Mark

Some people are early adopters, others are content to wait until version XYZ is about to expire. I’m not so close to it now as before, but I wonder aloud if CA people are keeping up with the latest – I suspect they are.

I would add, MATLAB is the latest in terms of scientific computing, though I’ve been using it over 20 years. There are many clones that are freely available, though they don’t come with the same level of maintenance (nor the depth of libraries) and may not always have the same code (Mathworks now compiles most of their m-files for both speed and copyright protection purposes, probably to keep the latest and greatest away from the clones).

Mark

In the more general sense, Mann et al use algoriths for estimation of great circle distances, they use lats and longs when searching for grid cells and points within, spherical coordinates, etc etc. There are many computational threads over the years on CA. There are other programs that do similar computations, but with different potential consequences for error and some are used by the same organisation, maybe with more attention to error. STK for example models oceanic columns for turbulence, stratification and other physical properties. I would have thought this relevant. (No, I’m not on the payroll, have never even spoken to them by email or any method).

Yes, I agree, but none of this requires the level of complexity contained within something like a satellite propagator, and ultimately, has nothing to do with his algorithm. In fact, the only reason “location” matters is whether or not his proxies relate to local temperatures, and what local effects need to be removed. Even with this, precision in location is not important. Now, claiming that the rainfall in Maine… ahem.

Mann isn’t doing anything that even closely resembles turbulence, oceanic, atmospheric or otherwise. Once he picks his proxies that provide enough spatial coverage of the globe, 3D implications no longer matter (and ultimately, it mostly boils down to a tree in Colorado from what I gather, but that’s another discussion, hehe).

The thing is that STK brings nothing to the table for doing proxy work. It’s like bringing a sledge-hammer when you have a screw.

Mark

Thanks for the response, Mark T.

In the more general sense, Mann et al use algoriths for estimation of great circle distances, they use lats and longs when searching for grid cells and points within, spherical coordinates, etc etc. There are many computational threads over the years on CA. There are other programs that do similar computations, but with different potential consequences for error and some are used by the same organisation, maybe with more attention to error. STK for example models oceanic columns for turbulence, stratification and other physical properties. I would have thought this relevant. (No, I’m not on the payroll, have never even spoken to them by email or any method).

When I learned machine language programming in 1970 I thought it was the bee’s knees until Basic appeared and I bought a machine with 8 K of ferrite core memory to run it. Then Fortran appeared, even better. Some people are early adopters, others are content to wait until version XYZ is about to expire. I’m not so close to it now as before, but I wonder aloud if CA people are keeping up with the latest – I suspect they are.