Temperature leading sea level

Posted: September 23, 2011 by tchannon in climate, Ocean dynamics
jason-uah-1

Figure 1

Recent Internet talk about lead/lag, ocean and so on tripped me into doing a quick rework of the temperature vs. sea level finding. This is of course all conjecture.

I’ve extracted what some might call a principle wave component from both datasets and these are very similar. As models I can time shift trivially. All the data shown has been normalised to the sea level data so they plot as one.

In hindsight I could have shown some earlier data. This is quite interesting and can involve Geosat and Church & White, all pretty much in agreement. Sea level did wobble during the 1980/90 but longer term the whole thing is what I call iffy.

If there is a temperature lead this is nice because it means we have a clue on what comes next with sea level.

Timing?

I think lead time is very vague. Previously I put this at about 4 years but that was based on a slightly different temperature dataset. Here is is a little over 1.5 years. None is a surprise for slow changes on short data.

Not detailing, all r2 > 0.9

The result is worse than previously, perhaps why is explained below.

As usual what ought to be trivial turned out to be not so simple.

A few of you might recall I found the satellite based sea level data predictive, complete with spikes and seeing the flattening off before it became obvious.

There is a tale in how Colorado started withholding data and fiddling with the various software versions. Ten day data six months late when you expect to see a fall? This happened to coincide with the launch and commissioning of JASON2, a very correct thing to do, the replacement satellite in service before the old one fails. They had to reorbit JASON1 and it seems there were problems when the two were producing data, as well as posing data problems with the changed orbit.

Given what I had been seeing for some time I had already started archiving the data since old data is not archived by the publishers. It is a mess. Doesn’t matter if it is put right in the end, just that you can see how they are struggling.

I knew the web site had changed. On digging I discover they no longer provide the data version I was using, which was the least statistics meddled with. A quick check on the data and the software here is showing distress, doesn’t like the new data. Until this is sorted out the only solution is minimise analysis, the rest is junk.

I’ve asked for the other version. Probably ought to look and seen how old/new has been changed. Revision of historic published data usually shows incompatibility, happens often in science. Meantime the above result will have to do.

Colorado sea level

UAH TLT no link given. Was produced from gridded data here and will match published. Ask if you want it. For those who hate that version, RSS is the same, difference is immaterial.

Tim

Comments
  1. Ray Tomes says:

    I assume that the correlation r^2 > 0.9 is for the smoothed curves. Such low frequency component will not have very high significance for such a short period and any determined lag will be very uncertain. It really does need much longer time period for low frequency signal.

  2. tchannon says:

    I’m not disagreeing. Unfortunately direct r2 between the two datasets is an awkward problem, sea level data is irregularly sampled, scattered around with plenty of missed samples. I never wrote the active code to resample this kind of problem, experimentally yes.
    fs=36.900369
    fs=36.764706

    Reminding myself daily MSU data is available, then the documentation is missing, however I spent ages some time ago discovering the column headings, came from a blog post.
    If anyone is interested daily uah msu data is here and data column headings, usual warnings about might be wrong.
    IYR IDY IDY78 ACHA MGL MSH MNH MTR (tropics)

    Ought to be possible to construct a sea level sample match set from that, I think easier than oversampling or decimating the other data. Any volunteers?

    RMSD is a better measure for this kind of stuff but means nothing outside of context. I’ve never come across a good general measure, ‘cos I don’t think there is one.

    They are modelled curves, filtered would be similar but with unavoidable end effects.

  3. Michele says:

    Solar OT !

    Rog, Solar fux 175

    Bendandi prof !

    http://daltonsminima.altervista.org/?p=16112

  4. Roger Andrews says:

    Hi Tim:

    Last year I constructed a relative global sea level rise series between 1900 and 2010 from scratch using 328 unadjusted tide gauge records from the PSMSL data base. When I compared it with the unadjusted global ICOADS SST series I got a remarkably close match (R^2 = 0.92 for annual means, with a 1C rise in SST corresponding to a 100mm rise in relative SLR). There was, however, no evidence that SST leads sea level. I could send you the data if you’re interested.

  5. tchannon says:

    Had a quick look at new data vs. archived old here.

    Think I’d better wait to see what Colorado have to say.

    I am quite sure they have changed all the data yet again including significantly changing the trend, but it is a different dataset class. (really shouldn’t be by much)

  6. tchannon says:

    Roger, if you succeeding in reconstructing you did well.

    I’ve tried various exercises with tide gauge data but always hit non-availability or bad data. One I wanted to do needed high resolution, try and figure out the sequence of events for the 1983 el nino, including LoD data. This needs good data on opposites sides of oceans but there are essentially no pairs of stations, let alone with high res and online.

    Yes I like to look at data. (sent you an email)

  7. tchannon says:

    Roger, had a look at the data. Nice to see you have retained plenty of digits, which tends to take Shannon violation out of the equation. I do wish stats people would junk what they are taught when it comes to PCM data. There is a literal trade of time and amplitude precision.

    As you say the basic shape is a good match.

    On running a quick analysis of both data it suggests there is little else in common. However a more careful look is suggestive there are common terms but these are not exact. All bit vague.

    If I had to declare lead lag I’d say there are plenty of hints temperature leads. All the rough matches are that way, the rest seems noise.

    I don’t trust SST. If I recall at least one dataset is a composite of very different sources.

  8. Roger Andrews says:

    Tim:

    I don’t see any evidence that SST leads SLR, but maybe I’m missing something.

    Could you post a plot of the SLR vs. SST data? It might be helpful if others could see what we’re talking about.

  9. tchannon says:

    I doubt you have missing anything, I was looking at some data internals, not the kind of thing for casual showing.

    What I think I’ll do is make a post of the data if that’s okay with you. Part of the reason is offline composing is much easier, some of the messy bits are handled automagically, then upload.

    Using normalised data makes more sense. Spectra tells little. Thinking about how to best handle lead/lag, annual data is rather coarse.

    Likely I need to get rid of the long term slope, we cannot see the ends so it is no help. One way is take first difference. An oddity appears, try rsqr at 15 year SST lag. Comparing the data shows there are then features lined up but I doubt this is cause and effect.

  10. Roger Andrews says:

    Tim;

    Fine by me if you want to make a post of the data.