Friday, January 29, 2010

New version of CRUTEM3 and HADCRUT3

There's a new version of the Met Office land surface temperature record out with lots and lots more stations.

Plus it includes corrections for all the problems I found with the data (they didn't make good on their promise to acknowledge me, sadly).

But my handiwork is shown by the points in green:


My two corrections: A and B.

I'll run these through my own programs to see what they produce.

10 comments:

iv6 said...

Hello,

I'm also working on reproducing CRUTEM3 record.

It was relatively easy to reproduce temperature anomalies, but I'm having trouble with error estimates. Have you been able to reproduce CRUTEM3 station errors, too?

iv6 said...

So... no, then? Please note that I was talking not about coverage bias estimated via NCEP/NCAR 40-Year Reanalysis Project. You showed it earlier. I was talking about station errors (red band on figures in Brohan et al.), CRUTEM3_station_error and CRUTEM3_sampling_error.

John Graham-Cumming said...

I have not had the time to work on the other errors

iv6 said...

Sorry for insisting, then. I still couldn't figure why I'm getting narrower estimates for station errors than in CRUTEM3_station_error and this bothers me. Everything matches perfectly when there is only one station in the grid-box, but combining two and more station errors yields smaller estimate than in CRUTEM3_station_error.

Do you plan to work on other errors, if time permits?

John Graham-Cumming said...

I'd suggest publishing your code somewhere and asking for comments. Document clearly how you are coming up with the error estimate.

iv6 said...

Okay, here. Actually, calculations are very simple and no code is needed.

Could you take a look at it, when you would have time?

John Graham-Cumming said...

It's possible that this is simply because the Met Office has not released _all_ the data that underlies CRUTEM3 so you probably can't reproduce it exactly.

iv6 said...

No, I excluded that possibility. I checked CRUTEM3_nobs, and there are only 2 observation records in this grid-box, both published.
Besides, if there would be more stations, combined error estimates would be even smaller, as they decrease with number of stations.

John Graham-Cumming said...

I will take a look at it. That seems very odd and would be very good to get to the bottom of it.

John Graham-Cumming said...

iv6: I have found out why but I am about to board a plane to the UK and will need to go into it tomorrow.

It appears that the normal error calculation is using the wrong number of years (15 instead of 30). I don't know why this is, it could be a bug, but I need to read the paper again to make sure I am not doing something silly.

If this is a bug we need to reproduce it for all of CRUTEM3 and make sure we report it to the Met Office. Can you send me an email so we can talk about this?