Wednesday, August 19, 2009

Curiouser and curiouser

Down and down into the Pielkeian rabbit hole we go...

Recall that Klotzbach et al observe that the expected (model-based) tropospherical amplification factor of 1.2 is not easily found in a comparison of land surface observations and satellite measurements. They suggest that this is due to a change in lapse rate on calm nights, such that the surface (1.5-2m) temp is warmer than expected, relative to the satellites. There is an underlying series of papers (eg Lin et al 2007, Pielke and Matsui 2005) which discuss boundary layer effects over land at night.

Now I hadn't noticed on my first reading, but the claim that the 1.2 factor applies over land is actually attributed to analysis of the GISS model output by one Ross McKittrick, who one might consider a curious if not altogether dubious authority for this statement.

Up pops Gavin Schmidt a couple of days ago, pointing out that in fact the GISS model output, when correctly analysed, has an amplification factor of just under 1 over land. This result appears to be supported by simple physical understanding and thus probably holds for other models too.

This correction immediately knocks off half of the missing amplification effect that Klotzbach et al was explaining. In the original paper, the difference in trend over land (assuming a 1.2 amplification factor) was estimated as 0.07-0.21C/decade (mean of 0.14), for the 4 combinations of {HadCRU,NHDC} vs {UAH,RSS}. With the correct "amplification" of 1 (it should more precisely be 0.95 according to Gavin's figures) the difference is 0.01-0.13 (mean of 0.07).

Of course in Pielke-speak this "confirm the robustness of our findings to model uncertainties". Actually, in numerical terms it halves the magnitude of the effect, but perhaps this interpretation is just me playing semantic games. It has been interesting to see how much of their result has collapsed, and how rapidly it did so, once their paper saw the light of day. RPSr's original "conservative estimate" of this boundary layer effect being worth 0.21C/decade over land (meaning 0.06C/decade in the global trend, about a third of the overall observed value) now appears to be an overstatement by a factor of three, according to these updated results. Nevertheless, if the remainder of their analysis is valid, this would still be a worthwhile contribution towards improving the compliance of the satellite and ground-based observations, which would bring them more comfortably within each others' respective error bounds.

(I'm still half-expecting someone to pop up and say the whole idea is wrong from start to finish, though.)

8 comments:

Deep Climate said...

Gavin Schmidt has pointed out that essentially all strong amplification is found over ocean, at least in the GISS model. (Of course, this makes intuitive sense as soon as it's raised).

Now if you look at K et al's response incorporating estimated factors from GISS, you see better agreement between HadCRU (amplified) and sats over land than over ocean (and of course globally). So not only is the effect greatly reduced, it's hard to argue that the discrepancy between expected and observed amplification is due to "bias" over land.

The large discrepancy between UAH and RSS should also give pause. It's difficult to draw any conclusions about the genesis of sat-surface discrepancies while that equally large discrepancy remains unresolved. And other estimates of tropospheric temperature (Fu, Vinnikov) are even higher than RSS.

Finally, not only did K et al not use a reasonable amplification factor for land, but they also did not account for the inherent uncertainty in the estimate of amplification. A more cogent analysis would follow the lead of past analyses by Santer et al and compare model mean amplification with observed, properly taking into account the various uncertainties.

Deep Climate said...

There is one large discrepancy between data sets that is not discussed in Klotzbach et al, but may be the most interesting aspect to emerge from the analysis.

The land-ocean differential in NCDC is remarkable over 1979-2008: 0.31C/decade (land) vs. 0.11C/decade (ocean), for a trend difference of 0.20C/decade. HadCRU has 0.22C/decade (land) vs. 0.14C/decade (ocean), with a differential of less than half that of NCDC, i.e. 0.08C/decade.

So an interesting question for further study would be: where in this wide range of land-ocean differential do climate models generally fall? And what are possible explanations for the wide discrepancies (especially over land)?

A first thought that occurs to me is that the latitudinal masks may be different in the various data sets - certainly something to watch out for.

James Annan said...

The land-ocean warming ratio covers a range of about 1.4-1.8 in models. Here's a paper (Sutton et al GRL 2007)

Note, however, that they used 1%pa CO2 runs and realistic aerosol loading will (I expect) reduce this ratio by primarily cooling over land.

bobunk said...

GISTEMP (1979-2008):
Land: +0.28°C/decade
Ocean: +0.12°C/decade

bobunk said...

oh, and:
Global: +0.16°C/decade

Deep Climate said...

Actually I've reread Gavin's letter to Klotzbach a little more carefully and he gives the GISS-ER figures:

As might be expected, the land temperatures rise faster than the global mean or ocean values (0.26 deg C/dec vs. 0.17 deg C/dec and 0.14 deg C/dec)

So land to ocean trend ratios are:
GISS-ER: 1.9 (1979-2005)

SAT obs (1979-2008)
HadCru: 1.6
GisTemp:2.3
NCDC: 2.8

James Annan said...

Sutton et al didn't say which model had which ratio, but over that recent interval, maybe the aerosol forcing has actually reduced (this factor is fingered for the recent rapid warming over Europe at least). Maybe there's also a chunk of natural variability.

Deep Climate said...

NCDC land/ocean ratio is actually 2.6, not 2.8. The difference appears to be due to update in the NCDC (land is now .300°C/dec., not.31°C/dec. with ocean at 0.114°C/dec.), as well as possible rounding error.

Also worth noting, the ocean warming has dropped off a little in NCDC. 1979-2005 was 0.132°C/dec., resulting in land/ocean ratio of 2.3 in that time frame.