Here are the results for the Moyhu NCEP/NCAR reanalysis index for August. It looked for a while to be a record breaking month, with steady warmth. But then there was a sudden late cool spell, apparently mostly Antarctica, and then an equally sudden recovery to warmth. The end result was 0.306°C, a big rise from July's 0.164. That is just slightly cooler (in this record) than May 2014 at 0.315°C. But it is the warmest for 2015 so far.
It makes a slight difference month-month what anomaly base period is used, and so the Moyhu table gives results also on the 1951-80 base (for GISS) and 1961-90 (NOAA Mlost). So the comparable GISS-base number would be 0.87°C. But as mentioned in earlier posts, the NCEP index, being air temperature, has been running rather cool relative to the land/ocean indices which using the warm current SST. So I would not be surprised if GISS were even higher - maybe even 0.9°C. The record GISS anomaly is Jan 2007 at 0.96°C.
ps This post is a little later than usual. The volatility meant that I wanted to go right to the end of month, and the last day of NCEP/NCAR was posted a little late at NOAA.
Al Gore: Climate Barking at COP29
4 hours ago
Posts
Nick,
ReplyDeleteI've been following your daily anomaly numbers for maybe a year now. A big part of the fun of this site is getting a daily update as to whether it's trending towards a particularly hot month or cooler than usual. However, I have seen that your reanalysis number is not as easily comparable to the number GISS finally publishes for the month, and I think you are alluding to this in your current pot. Can you (or have you already) provide the aproximate relationship between your index and the GISS monthly anomalies. Is it a straightforward linear relationship? Is it a different calculation for each month? I would just like to use your data to make a prediction of what GISS (and the other groups that put out monthly global temp. anomalies) will have as the anomaly at the end of the month. I know that it is only a prediction and you have described at least one possible reason above, but understanding the relationship between what your index predicts and what the data providers finally will make it possible for us to play from home instead of just waiting for the numbers at the end of the month.
Thanks for all you do,
Robert
Thanks, Robert. I have a table below the monthly averages (scroll down and on left) which shows the NCEP on the GISS base (1951-90) rather than its own base (1994-2013).. So for August it is 0.87. You can see the pattern of recent months to see how well that would actually match GISS. Since NCEP doesn't have reliable data in 1951-80, I have to use GISS data to bridge the gap.
DeleteThe other way I look at it is to calculate differences. If it rose .14 this month, a gues sis that GISS would rise the same. Or maybe go back to a more similar month, say May, where NCEP is only 0.06 higher (this way, the end result is rather different in this case).
OK, thanks - that clears up a lot for me. I had overlooked the estimate of GISS for the month to date at the bottom of the window, I'll certainly keep an eye on that moving forward. I had thought that I could just look at index and expect similar values of the GISS anomaly for similar values of the index, but I think there has been enough variance in the past that I wasn't trusting that approach (just my own misunderstanding - expecting too high of a correlation between the two data sets).
ReplyDeleteBased on this, I took your index values from Jan. 2014 to present and obtained the matching monthly GISS anomalies from http://data.giss.nasa.gov/gistemp/tabledata_v3/GLB.Ts+dSST.txt and ran a linear regression in Excel. Excel gave me the equation GISS Anomaly = (NCEP/NCAR reanalysis surface temp anomaly) * 121.86 + 51.467 (with r squared = .809). Plugging in the index value of .306 for August predicts an August GISS anomaly of 89. So now I can look at this number and the value you predict at the bottom of the page to see how the month is going as well as seeing how the final prediction at the end of the month compares to the value published by GISS. I'll have to redo the linear regression each month as many of the monthly GISS anomaly values change each month (due to re-homginastion each month, I am told). It will be interesting to see if the monthly variations make much of a difference in the predicted GISS value. Now I have what I'm looking for. Thanks again for your help.
There is a wrinkle that I described here. Anomalies are calculated monthly, and so just adding a constant amount over a year doesn't quite work. If Junes were warm in 1951-80 (GISS, relative to later Junes), and cool in 1994-2013, then NCEP will tend to show June 2015 anomaly warmer than GISS. It's a small effect - I gave a table of numbers that can be used to correct. The extra table uses that correction.
DeleteAnomalies may rise to +1,3°C in march 2016 according to french website global-climat.com, http://bit.ly/1PQIshW. Do you think it is possible ?
ReplyDeleteKC,
DeleteWell, they're predicting 0.86 for August, and I think Aug might be even higher. The first test is probably the 1.00 for September. But yes, I think it is possible. In 1997/8, GISS rose from 0.42 in Aug to 0.89 in Feb, a rise of 0.47. They have 1.28 in Feb 2016; which if this Aug is 0.88 would be a rise of 0.4.
Thanks. Indeed 97/98 showed à big climb. Gmao also predicts big anomalies but based on 3.5 el Nino (!!) in the 3.4 région. Ncep is based on a 2.5 el Nino. Antarctica might be à key component of this forcast as it shows à big rise there.
DeleteOne more thing : it seems ncep anomalies reported on your web site differ from those of weatherbell web site and i don't think it 's à matter of base years. Why ?
KC,
DeleteThere are some differences. I'm not sure which Weatherbell page you are referring to, but I see mainly NCEP CFSR. I'm using NCEP/NCAR, an older version. The main practical reason is availability - the NCAR version comes on a 8 Mb file that I can download every day. CFSR is much bigger; maybe paid-up users get something better. I thin k the public version is also less current. Another difference is that it is 2m altitude. I'm using sig995, which is about 40 m (basically the midpoint of the surface cells). For the data I can access, it is better quality than surface.
It could be also anomaly base. You need to use a daily base, and that brings in quite a lot of noise. I use a 20 year period (about all you can get without gaps). So if 20 instances of 4 Sep turned out to be warmer than 20 of 5 Sep, then the differences is embedded in the anomaly plot. And for a different base period, the difference is different.
All that said, looking here and allowing for smoothing I think the match isn't bad.
Indeed, I'm referring to pages from Weatherbell using NCEP CFSR (base 1981-2010)... but didn't realise yet that was a more recent version. See this page :
ReplyDeletehttp://models.weatherbell.com/temperature.php
There is only month to date anomalies on weatherbell but you may can get different data on a paid version, I don't know. Indeed, the anomalies are quite similar to those you're using, I mean NCEP/NCAR. There are just small differences depending on the months.
NCEP CFRS is nice when compared to GISS from last months. As you made on your site it is possible to provide an estimation of GISS based on NCEP. Sometimes differences in the polar regions explain small gaps between NASA and NCEP.
The trouble comes when you want to compare months of 2015 to those of 2010 or 1998 for example. According to NCEP CFSR, 2010 was warmer than 2015 and 1998 was nearly cold, which does not make sense.
KC,
ReplyDeleteI plotted the NCAR monthly index in this post; the third graph is user-draggable, and you can take it back to 2010 or 1998. It tracks along reasonably well with the other indices.
Yes, I think the differences are often due to polar.
Thank you, I saw on your nice graph that in 2010 and 1998 NCAR was quite similar to other major datasets. But that's not the case for NCEP CFSR, not shown on your graph. For example, according to NCEP CFSR, jan 1998 is colder that jan 1999, wich can't be. Jan 2007 is really warm compared to jan 98. Other datasets show a very warm jan 2007 but Giss doesn't climb as much even though jan 2007 is still the biggest anomaly ever recorded.
Delete