Update (March 9th) – Dr. Roy Spencer just gave an interesting response:
“yes, we have been aware of some spurious warming over land versus over the ocean after approximately 2000. Our version 6 dataset (now close to completion) will have most of that removed, although it looks like some of it is genuine.”
I guess we all just have to wait and see …
I have earlier noted a rather curious blocklike shift up in the UAH tlt (lower troposphere temperature) timeseries occurring abruptly some time in 2005. (There is most likely a similar – only downward – step at the same time in the RSS tlt timeseries; however, this post will not address this one.)
The 2005 shift seems very much to originate in the land portion of the UAH dataset. The shift can readily be seen here, but not at all in the oceanic portion, a situation which is quite unprecedented in the record – global land temps simply do not by any known natural mechanism all of a sudden jump out of step with the global ocean temps and then remain elevated high above thereafter:
Figure 1. As you can see, something quite out of the ordinary happens in the UAH land curve in 2005.
Figure 2. Zooming in on the final part of Figure 1, from 1997 onwards. The land shift occurs abruptly in the latter half of 2005, and the blue land curve stays firmly elevated all the way down to 2014, when the ocean curve actually catches up once again. This circumstance will be discussed later …
How can we resolve this? It sure doesn’t look natural.
Do we in fact know that it’s the land portion having a problem here, and not the oceanic part?
It should be fairly easy to find out.
By comparing with the surface temps.
Global tropospheric temps, after all, very much mimic (though amplify) the surface signal, as one would expect, knowing how the solar heat generally moves through the Earth system: Mostly propagating first down to the surface; the surface absorbs the heat and warms as a result; in turn, the surface transfers its absorbed solar energy as heat to the troposphere above it; the troposphere absorbs this heat and warms also; finally, the energy is emitted as heat back out to space, from the atmosphere (that’s the OLR through the ToA).
The lead-lag correlation is obvious.
First we check the oceanic portion.
Here’s global sea surface temperature anomalies (SSTa) 1997-2014/15 as represented by HadSST2, HadISST1, ERSST.v3b and Reynolds (NOAA) OI.v2:
Figure 3. The HadSST2 is the oceanic part of the HadCRUt3 dataset. It exhibits larger amplitudes than the other three datasets, especially after 2002/03, but is still consistent overall with their mean progression through time.
Since the four datasets in Figure 3 appear to be in such close general agreement, I choose their average as my ‘official’ rendition of global sea surface temperature anomalies from Jan 1997 to Jan 2015:
Let’s compare this one, then, with our UAH oceanic curve from Figures 1 and 2:
Figure 5. Note that the tropospheric curve (turquoise) is downscaled (its data divided by 2) to fit with the surface curve (green). Watch also how the tropospheric curve clearly seems to lag the surface curve. The lag is more distinct (bigger) during proper ENSO warm and cool events.
The surface signal is strongly amplified (~doubled) in the troposphere above the ocean. This situation is very different from the one over land, as we shall see.
Anyway, there appears to be no problem with the oceanic portion of the UAH tlt dataset. Its progression agrees rather perfectly with that of the ocean surface.
So what about the land portion?
We will compare it with CRUTEM3 gl (the land part of the HadCRUt3 dataset). If you feel this particular choice is a tad reactionary, seeing how there are so many newer, updated (hence, of course, by implication better/improved) land datasets out there, here’s CRUTEM3 vs. CRUTEM4 (to the left) and BEST (to the right):
Plus CRUTEM3 vs. GISTEMP land, 250 km:
All that the newer and ‘better’ datasets do is simply adjust the major 1997/98 El Niño signal down and the minor 2006/07 El Niño signal up to make it seem like there’s a steady rise between them, so if anything, the ‘plateau’ only starts in 2007, when all those La Niñas started affecting the trend anyway. The CRUTEM3 thus, in my opinion, portrays the situation in a much more honest fashion, and, not to say the least, in much closer agreement with the sea surface temps: The ‘plateau’ clearly starts with a lofty 1997/98 El Niño and levels off already from 2001/02.
This is why I won’t average the four land datasets as I did with the sea surface datasets. I will simply use CRUTEM3 after having made sure (above) that the other three don’t deviate in any meaningful way from its general course, significantly, that no blockwise step up around 2005 is to be seen in either of them:
Animation 1. Here you can see how there’s a clear step up in the tropospheric curve (blue, UAH) in 2005 that simply does not exist in the surface curve (yellow, CRUTEM3). As it turns out, it is a blockwise upward shift of ~0.17 degrees. Note, the troposphere only amplifies (somewhat) the surface signal during the biggest El Niño events (1997/98 and 2009/10). Otherwise the amplitudes are comparable to the surface ones. This is very different from the situation over the ocean.
Let’s adjust the UAH land portion curve in Animation 1 down by 0.17 degrees from Sep’05 on and see how it looks:
Gone is the step. And the agreement with the surface so much better.
How, then, does this ‘new’ UAH land curve fit together with the UAH oceanic curve (recalling back to Figure 2, when they didn’t fit at all)?
Clearly a much better (and more natural-looking) fit!
This is simply how the two curves should track each other. And it’s how they do track throughout the record from 1979, except around and after that abrupt shift in 2005. That we have now eliminated …
(We still note, however, the conspicuous divergence between the two curves occurring in 2014. We will address this at the end of this post.)
So from this (with the adjustment of the land curve) we compute a new (and ‘improved’) total UAH 5.6 tlt gl curve (land (x0.29) + ocean (x0.71)):
And then finally, we overlay the (downscaled and lagged) NINO3.4 SSTa curve for a neat comparison:
Two notes at the end.
1) The global troposphere carries much more of the ocean signal in it than what the surface does.
As we observed from Figures 5 and 8 above, the troposphere doubles the sea surface signal, but does not amplify the land surface signal at all (except a little bit during large El Niños). So the ocean part of the surface signal is much more prominent in the total tropospheric signal than it is in the total surface signal itself.
This is why especially the great ENSO events (warm and cool) stand out so much more in the tropospheric (satellite) temperature timeseries than in the surface (instrumental) ones. The tropospheric column is affected much more by oceanic processes than by land processes. The convective coupling between the surface and the troposphere above is simply much stronger over the ocean (‘deep moist convection’ being the key process).
2) The great SSTa peak of 2014.
Noted under Figures 2 and 9, the oceanic portion of the UAH tlt dataset diverges strongly from the land portion in 2014. This happens as the SST anomalies in the extratropical NE Pacific soar, starting in 2013, but peaking impressively in 2014. The striking divergence between the oceanic and the land portions of the dataset during this development is not exclusive to the troposphere. You can see the very same thing in the surface datasets – compare the ocean surface in Figure 4 with the land surface in Figures 6 and 7.
The reason for this global ‘decoupling’ is most likely the fact that the 2014 SSTa peak originates in a restricted region outside the tropics, the normal centre of such action, so that worldwide propagation via convection>atmospheric circulation>teleconnections is much restricted.
There is no evidence whatsoever to suggest that anything similar to this rather unprecedented situation occurred also in 2005 – only the direction reversed (!?) – to explain the semi-permanent decoupling between ocean and land as is observed in the UAH dataset.