Friday, December 23, 2016

Merry Christmas to all

And a viable New Year (as our CFO used to wish)

As a Christmas gift, I've been sparring at WUWT with folk who insist that raw temperature data is hidden/deleted/not available. I point them, of course, to GHCN Daily. But there was much talk of BoM, and I must admit that the data page isn't as obvious as it was. But it is there. And not only can you get an unadjusted daily file for just about any station they have, back to start, but you can download zipped csv files for max and min (but not together, unfortunately). And there is also extensive metadata.

On to interesting times. Here is Sou on Mike Mann's court victory. And Eli.


  1. Hi Nick,

    I just received you mail while sitting in front of the PC :-)
    Thanks for the wishes, 100% replied from Germany...

    Regards, J.-P.

  2. Merry Christmas!

    I hope that is not a picture of the future Arctic!!

  3. Nick,
    As a long time lurker, I would like to thank you very much for your blog and wish you a safe, prosperous and happy New Year!

  4. Happy new year, Nick. And thanks heaps for all the work you do, particularly demystifying temperature records.

    1. Thanks, Sou,
      and thanks for all your work too. It's going to be an interesting year. We'll need you. Best wishes for 2017

  5. Nick, hope you and all the visitors here have a great new year. I've been playing around comparing reanalysis data over the last week and came to the conclusion that yes indeed, the NCEP CFSR/CFSV2 does stick out like a sore thumb. I compared the NCEP/NCAR R1, NCEP CFSR/CFSV2, ERAI unadjusted, and ERAI recently adjusted global surface temperature anomaly estimates. The recent ERAI adjustment brings it into better alignment with the NCEP/NCAR R1 which I understand has not had any modifications since its inception (and therefore should serve as a good guide for longer term trends). So I took a crack at adjusting the CFSR to also bring it into alignment here:

    The simple adjustment I made does bring the global surface temperature anomaly estimates into a nice alignment, although I recognize that fully adjusting the CFSR for a better match is a much more complex and difficult task. However, I still believe that the reanalysis approach in general is the way of the future for tracking both weather and climate changes.

    1. Bryan,
      Best for the NY to you too. Sorry to be a bit late replying, but we went away for a couple of days.

      I think the main problem with reanalysis is that it is not homogeneous. It is assimilating information from a wide range of instruments, and as the mix shifts over time, so does the result. It's like a big version of the buoy/ships issue with SST (which required the Karl adjustments). The trouble with reanalysis is that the mix is so varied that I don't think they can correct. So stuff drifts. The CFSR had a lot of that happen in one version change.

      That's why I think NCEP/NCAR is good for comparing days in a month, and months in a year, and maybe a few years, but not for decades.

    2. Nick, the NCEP/NCAR R1 was run back to 1948. I would guess it has data from all the stations in the GHCN-based estimates, plus more. But it uses 4 times a day instead of min/max for weather stations. I compared the monthly estimates with NCEI and they compare quite well back to the 1960's, but show about 0.2C higher in the 1950's. Maybe that points to issues with NCEI in the 1950's? When you have two watches with different readings (reanalysis versus GHCN-based) how do you tell which one is better?

    3. One further thought. I suspect the main advantage of the reanalysis approach is that it likely does a better job at filling in data sparse regions since it uses a NWP model to guess ahead 9? hours and blends that with observations. It would be interesting to see some testing in that regard by withholding a few remote weather stations from both approaches and then comparing how well they do in estimating temperatures for those stations.

    4. Bryan,
      There is a summary of reanalysis strengths and weaknesses here and a more detailed paper here. For NCEP/NCAR back to 1948, I see three disticnt stages:
      1. From about 1994-now, when they are using currently collected NWP datasets
      2. From about 1980-1993, when I expect they are using archived NWP data, which probably wasn't as good anyway when collected, and probably has data gone missing
      3. From 1948-about 1980, when they have to try to put data retrospectively. It wasn't nearly as good in the first place, missing is worse, and they don't havethe arrangements for filling gaps that I'm sure you would develop when collecting regularly.

      So I'm pretty doubtful about early years. And the great thing about post-1994 NCAR is that there is nothing missing. That makes integration a breeze. Before that, it isn't so easy.

    5. Nick, thanks for the links. I will check them out.