Monday, April 26, 2010

Spatial Temperature distributions in TempLS v2

This is another trailer for v 2. It can fit temperatures for individual years, and I expect even months, though I haven't tried that yet. The model is that data depends on local effects as before, plus just the temperature of the specified year, over space. Of course that doesn't make a good predictor overall, but the aim is just to get the temp distribution for that year that gives the best fit.

A technical snag which I'm still working on is that this model does not generate the info needed to adjust to a base period. It may be necessary to add that as another fitting parameter. However, since I'm currently comparing to GISS, I just add an offset to match the GISS average anomaly for the relevant year.

I've compared these with the GISS plots. I've also gone back to match the trands of the previous post with the GISS trends for the same period, since GISS produces better maps. BY request, I've also plotted the trend for the period 1900-1945 and compared with GISS. All below the jump.

The last big task remaining for V2 is a mechanism for doing all this regionally, where spherical harmonics would be inefficient.

GISS comparisons

GISS is of course using somewhat different underlying data - all mine are based on GHCN raw and HADSST2. GISS adds other data, particularly Antarctic, and uses a different HAD SST set to 1981, and an OI set beyond. GISS has recently posted a new explanatory paper.

As I mentioned above, the program now does individual years, so I'll show here 2009 and 2008. Because of lacking GHCN data, I've rubbed out above 80N and below 65S. I've more scientifically matched the GISS colors and levels (some GIMP magic).

Here's 2009:
And 2008


Again, remember that GISS is using somewhat different underlying data. Part of the usefulness of TempLS is that it enables you to try the effects of different data. Remember too that the match of colors and levels is still not exact. Despite what the titles say, my output is not here in C/decade. I've rescaled to match what GISS do. They calculate the expected temperature difference over the time period due to the trend. They call it just a temperature difference, but that is what they mean. So I've multiplied the trend by the elapsed time (in decades). So, 1901-2005:
and 1979-2005
and finally, 1901-1945


  1. Well, I think you finally managed to convince me to learn R. :-p

    Neat results.

  2. Yes, I hadn't used R much (although I used S way back) but it makes very easy to manage array programming.

  3. Take a look at the latitudes comparing the maps from 1901-1945 and 1979-2005.

    Could you also do the SST's from 1901-1945 and 1979-2005? Take a look at the latitudes too.

    Thank you, Ibrahim

  4. Ibrahim,
    What are you noticing in the latitudes? I can't see it.

    I have a new post about issues that arise when global orthogonal functions are applied to regions with sections of missing data. That would apply to SST's only.

    But you'd also expect that an SST analysis, if it worked, would give a similar result. Over the oceans, it is minimising the same sum of squares.

  5. Zeke

    Make sure to join the R-help list. Godsend. answers in minutes.