Sunday, April 3, 2016

Global NCEP/NCAR down by 0.057°C (updated); still hot

Update: Commenter Steven D has pointed out a leap year error in calculating anomalies. The March anomaly is now 0.783°C, not 0.755 as I had earlier calculated. So the drop is 0.057, not 0.085.

The NCEP/NCAR index that I calculate from their reanalysis was down by 0.085°C in March, from 0.84°C to 0.755 (anomaly base 1994-2013). February was very hot indeed, so March is still the second hottest in the record, and of course the hottest March. The early part was part of a warm peak, over 1°C. Around 9 March it dropped to about the levels of January, and pretty much stayed there, slowly bringing down the average.

You can see the regional story on the active sphere map. In the date selector on right, just click the black square at the head of the March column. In N America, the W and mid US were cool, but a band of warmth around the East coast, and through Canada to Alaska. The Arctic was warm, of course, and also a band through East to central Asia, although E Europe was cold. El Nino is still there, and elsewhere mixed.

There are other similar results. Karsten's GFS is down by just 0.02°C; UAH is down about 0.07°C. Ryan Maue reports a CFSR drop of 0.07°C. In all cases, March is still close to second hottest month ever. And the BoM tells us that Australia had the hottest March on record, even including the hottest day (for the country) on March 2 (outside summer!).

I expect to have useful TempLS surface measures in a week or so.

The pattern of a Feb peak, with drop in March and then some recovery, was also shown in various records in 1998. In fact, the similarity is striking, though I think probably coincidental in detail.


  1. Monthly mean for march (base 1981-2010) calculated with sig995 gave me 0,9, down 0,02 from january.I noticed big divergence between your index and the data I used a the end of march. It seems to me it can't be explained by the basis only.

    1. It could be the anomaly base; I looked at cases here. However, the fact that 1994-2013 overlaps 1981-2010 would tend to counter that.

      Did you mean down 0.02 from February?

    2. Sorry, I meant 0,02°C down from february. With sig995 (using Panoply)
      Feb : 0,92
      March : 0,90
      "Surface" measurement from ESRL gives 0,92 also. March is not released yet but it should be about the same as sig995, like in february, I guess ?
      The basis comparison surely explain a drop of about 0,1°C between 1981-2010 and 1994-2013 at most.
      Your monthly march index reports a 0,75°C which I found quite far from 0,90°C obtained with sig995 and panoply, even if it is for 1981-2010. At the end of march, between the 29 and the 31, I saw anomalies reaching 0,95°C on average with sig 995.
      So it's difficult to guess what GISS could show. The dataset I used could be far from GISS cause UAH showed a 0,1°C drop, closer to your index variation.

  2. Karsten's NCEP/NCAR for March is ready now, down by 0.03 from Feb. So we have a spread of minus 0.02-0.07 in the reanalyses.
    Giss-guess for March 1.30 C..?

    1. I would say GISS March 1.21 or 1.24°C. With the later, my regressing forecast for the year 2016 runs towards +1.1°C instead +1.0°C the previous month.

  3. Well, Surface on ESRL is +O,89°C in march, down O,O25°C (1981-2010)...

  4. Nick, You say that the NCEP/NCAR index for March 2016 is 0.755°C relative to baseline 1994-2013. However, from the ESRL website I calculate that that number is 0.782°C (rather than 0.755°C). I think something is wrong with the recent daily values of your NCEP/NCAR index, perhaps due to a leap year issue, starting on 29 February 2016 (Leap Day)?

    1. I'll check. I don't think there is an issue with daily values, which seemed quite continuous over the leap day. But maybe the month alignments have drifted.

    2. I checked the daily file. It seems to have handled Feb 29 correctly, and I manually did the average for March, which was OK. I'll try to follow your ESRL path.

    3. Steven D,
      Yes, I got the ESRL PSD data, and using surface temp, you're right. It matches past months almost exactly, except for that March discrepancy. I'm using their surface temp, rather than sig995 (which they don't seem to give as monthly av), but that shouldn't make much difference - certainly past matches have been good.

      I'm checking.

    4. I've continued checking. I think the program is OK with leap year. I think it may be an unusual discrepancy between temp surface and sig995. I use sig995 because it has better data continuity in mid '90s. The discrepancy is fairly small, so I think I'll wait till April is available. If that is discrepant too, then further investigation.. Meanwhile, it may well be that March should be 0.782, which would mean a drop of about 0.06, instead of 0.085.

    5. Nick, using the 1000 mbar temps from ESRL and the 1994-2013 base, I got 0.759 C for March, down 0.092 from February (0.851).
      It looks like the decrease Feb-Mar increases with height (isn't 1000 mbar (adjusted to sea level?)) a little higher than sig 995?

      Another issue is if there are minor discrepancies due to different methods of area weighting. As I understand this can be done crudely by the cosine of gridcell midpoint, or exactly by sine of the northern limit minus sine of the southern limit. I think such errors grow larger when one approaches the poles.
      Please correct me if I'm wrong, I'm not the mathematician here...

    6. Nick, I tried to check the daily values of your NCEP/NCAR index. (I've been using surface air temperature from this webpage, downloading the gridded data and then calculating the area-weighted global average). This seems to confirm that the problem started on leap day, 29 February 2016. The daily values that I calculate before that date are in very good agreement with your daily NCEP/NCAR index. But the daily values that I calculate after that date are systematically higher than yours, by about 0.02°C or 0.03°C or so.

      Are you sure that after 29 February 2016, your daily data are still aligned correctly with the 1994-2013 climatology? Climatologically, global temperature increases during March at a rate of about 0.02°C to 0.03°C per day, so that could explain the discrepancy?

    7. Steven D,
      Thanks, I think you've solved it. Yes, it is the climatology. I use a 365 day cycle. Each year, any given day drifts by 1/4 day relative to its climatology (however set) and is adjusted back in a leap year. That's going to happen anyway; how it's done is a matter of convention. I hadn't thought about it too much, and was just going to duplicate a day at the end of the cycle, making the leap day Dec 31. But of course the regular convention is Feb 29. I'll fix that.

      There are in principle better ways. April 1, say, could have 4 different climatologies depending on where it sits in the leap year cycle (with 2000 being a nuisance). But I guess the best thing is just to use the matching convention.

    8. Olof,
      I think Steven D has solved the mystery - I'll recalculate using conventional leap year treatment of climatologies. As to how to actually do the area weighting, the methods you describe are both good. They are second order schemes; one is the midpoint rule, one is effectively trapezoidal. I use midpoint, but I've checked the difference, and IIRC is small. My reason for mid-point is that trapezoidal gives zero weight to the pole value. For surface work with stations that is serious, because the South Pole has important data. It could be said that mid-point overweights the pole, but, well, something has to give.

    9. Steven D,
      I recalculated using the conventional leap day for climatology, so Feb 29 is the same as Feb 28, and Mar 1 is Mar 1. And it now agrees with your calc (well, .783 vs .782). Thanks much for that. The code should now display that right, including daily values. I'll write a blog post some time today.

      In checking the code, I notice that I do for this use trapezoidal rather than mid-point. I'm not sure why I changed. The pole issue doesn't matter so much with gridded, since there is averaging over the grid cell anyway.

  5. Let me add to the confusion..
    At KNMI Climate explorer NCEP/NCAR 2m for March is up 0.02 C from Feb (everything on the 1951-1980 base).
    NCEP/NCAR sea is down by 0.08 C whereas the land share is up 0.27 C.
    However, ERSSTv4 March is up 0.02 C from Feb.
    Based on this a mixed SST+land 2m index should go up about 0.10 C from February.
    The coming weeks will be interesting... Can we have an early March indication from TempLSmesh soon?

    1. Olof,
      TempLS starts to report when it has 3500 stations. That is usually SST + US plus just a few more countries (Europe). Anything less is fairly meaningless. Last run (5 Apr local time) had 3426, so next run should give results.

    2. More details - of those 3426 stations, 2351 were SST, 1075 land, 613 US and 462 ROW.