Update 2/7/17: I used the wrong GISS data (land only) in the comparison as pointed out by @cce! Agreement now is reasonably good for all temperature series.
This is my crib sheet for comparing temperature anomalies across ground and satellite data. Temperature anomalies are usually (but not always) defined relative to a baseline 30 year ‘climatology’. This simply means that a 30-year average temperature for each month is calculated at each location. The anomaly is then the difference of the mean temperature from that monthly ‘normal’. First here is a table showing which baseline each group uses.
Group | Baseline |
NASA GISS | 1951 – 1980 |
Berkeley (BEST) | 1951 – 1980 |
Hadcrut4.5 | 1961 – 1990 |
NOAA | 1971 – 2000 |
UAH | 1979 – 2010 |
RSS | 1979 – 1984 |
This means that you can only compare temperature anomalies once they have all been normalised to the same baseline. In order to do that you must first calculate the 30 year average monthly temperature for the new baseline and then subtract it. These are the normalisation ‘offsets’. The UAH baseline 1979-2010 is the only one where all datasets have overlapping values. These are the offsets you need to plot all series together.
Group | Offset for 1979 – 2010 baseline |
NASA GISS | |
Berkeley (BEST) | 0.3635 |
Hadcrut4.5 | 0.2907 |
NOAA | 0.1838 |
H4-ST | 0.315 |
RSS | 0.0925 |
UAH clearly has no offset and H4-ST is my own Spherical Triangulation of Hadcrut4.5 stations. The offsets above should be subtracted from all anomaly values in each series. Here is an animation of the results.
The agreement across groups. is good except for GISS. The ‘warming’ observed by GISS is far greater than any other temperature index and renormalisation does not change the slope. It is an outlier.
Shown below is the table of offsets needed to normalise all series to the same baseline as Hadcrut4.5. The satellite offsets have been deduced by simply using the negative of the average Hadcrut4.5 anomaly in each of their respective baseline periods.
Group | Offset for 1961 – 1990 baseline |
NASA GISS | |
Berkeley (BEST) | 0.0643 |
NOAA | -0.1216 |
RSS | -0.1982 |
UAH | -0.2907 |
Here is the comparison for all series plotted on the Hadcrut4.5 baseline. GISS does show the largest overall warming trend, but it is not an outlier.
You can check my values and derive new ones using this spreadsheet. I also cannot understand why there is not already some agreed IPCC renormalisation. However, I am pretty sure it would be the same as mine.
Note: This post was prompted after a twitter exchange with Victor Venema. 😉
Clive,
I agree about the utility of normalising. I’d note though that the UAH period is actually 1981-2010, which is also the current WMO recommendation.
I post an updated normalised graph here using 1981-2010. It doesn’t really show that GISS is an outlier. Showing at the bottom fringe in early 20 Cen. Here is a plot with GISS in blue for contrast:
I don’t think it is a good idea to extend UAH back with HADCRUT – they are pretty different. If there is a complete choice of reference period, it might as well be one for which all have data.
Nick,
I think maybe the old normalisation period was 1979-2010 because I picked that up from from somewhere. But it looks like you are right from John Spencer’s web page and UAH now use 1981-2010. I will correct it but it should make only a small difference.
Your plot looks like it is annual data or a 12 month running average. What I see is that GISS consistently records the hottest peak months from 2000 onwards, yet the coldest months before 1978.
That graph of GISTEMP can’t be correct. Are you sure you aren’t using their “Met Only” series?
Yes, I checked your spreadsheet and that is what you are doing. You need this data: https://data.giss.nasa.gov/gistemp/tabledata_v3/GLB.Ts+dSST.csv
I am using this data https://data.giss.nasa.gov/gistemp/tabledata_v3/GLB.Ts.csv
I had to rearrange the time axis into a 12 monthly partition, but that is all.
Yes, that’s their legacy series derived from land-based meteorological stations only. You need to use the series I linked which includes SST.
Thanks – that looks like a bit of a screw up on my part – fairly normal I admit! However I can’t help thinking that the dataset should really self describe exactly what it is !
I will fix it
🙂
Their official names are: Combined Land-Surface Air and Sea-Surface Water Temperature Anomalies (Land-Ocean Temperature Index, LOTI) and Global Mean Estimates Based on Land-Surface Air Temperature Anomalies Only (Meteorological Station Data, dTs)
OK you were right! GISS agrees with everyone else (approximately), and I got hold of the wrong Land only data. Their metadata is however cryptic !
🙂
Clive: When you want to discuss the difference between to series, a difference plot would make a lot of sense. There supposed is almost twice as much warming in the upper atmosphere (not sure which record) than at the surface during the 97/98 El Nino. There is a hot-spot associated with warming then.
Frank,
Here is the difference between UAH and Hadcrut4.5 – the blue signal (righthand y-axis). As you can see there are two effects. First the UAH global warming trend is about 0.01C/y less than H4.5. Secondly the UAH El Nino warming peak 1998 is much stronger than H4.5 while a smaller effect is observed in 2016.
What is the effect of limiting all datasets to the coverage of UAH, i.e. excluding the poles? This avoids the need of arctic extrapolation also.
I thought UAH did cover the poles. If you restrict coverage to where there are measurements then you more or less get the same as Hadcrut4.
UAH stops at +-82.5°;. RSS doesn’t go so far, probably with justification. But you’ll notice that people never quote UAH as a limited region. They quote it as a global figure. And I guess if challenged they would say – well, it’s only a small area and it can’t be too different from the rest, can it?
And that shows the effect of leaving it out. Effectively, you assign to it the average global temperature, which is not very good. What interpolation does is to assign the temperature of the nearby region, which is surely a better estimate.
Excuse my ignorance but those satellites are in polar orbit so how come they can’t measure the poles ?
Because they are not completely over the pole
See this post with orbit graphs
https://wattsupwiththat.com/2013/11/14/curry-on-the-cowtan-way-pausebuster-is-there-anything-useful-in-it/
Clive, one more detail concerning baselines: RSS’ is 1979-1998.
Excuse my ignorance but those satellites are in polar orbit so how come they can’t measure the poles ?
It is not ignorance but simply believing what after all is written by Roy Spencer himself and accessible to everybody. Let us look at the bottommost line in
http://www.nsstc.uah.edu/data/msu/v6.0/tlt/uahncdc_lt_6.0.txt
There you see:
NoExt 20N-90N, SoExt 90S-20S, NoPol 60N-90N, SoPol 90S-60S
But in fact the UAH people don’t give us valuable data below 82.5S nor above 82.5N: that you easily can see when looking at their 2.5° grid data.