Dr. Klotzbach, I read your new paper (in press at JGR-A) with someAnd here is the response of Klotzbach et al.
interest. In it you make use of the expected amplification of the MSU-LT
data over surface temperature data by a factor of about 1.25. This number
comes from global calculations across the AR4 models reported in CCSP and,
as you know, is related mainly to the expected tropical amplification of
surface warming over the oceans.
However, I am puzzled by your claim in the paper that the same
amplification number holds for the metrics calculated over land only. The
reference for that is a personal communication from Ross McKitrick, who is
(surprisingly) a source for the behaviour of the GISS model (that I run).
Prof. McKitrick is not one of our collaborators (as far as I am aware) and
has no privileged access to the model output. Since MSU diagnostics were
not part of the CMIP3 archive, I would be highly surprised if he were able
to have calculated these diagnostics himself. (They are not complicated,
but it does take some effort).
It is possible that he is using some supplemental data I placed online (in
relation to Schmidt, 2009;
http://pubs.giss.nasa.gov/abstracts/inpress/Schmidt.html ). However, this
SAT and MSU data are a particular sub-sample of points from the GISS model
and were very restricted in scope and purpose. In fact, I do not think
that these data can be used to calculate the diagnostic you want.
In a transient simulation with land temperatures rising faster than the
global mean, the moist adiabat in the tropics is tied mostly to the ocean
temperatures. Noting also the fact that most of the land is not in the
tropics, I would have expected the amplification to be substantially less
over land than globally.
To test this, I took the GISS-ER results from 1979-2005 (20C3M runs, five
ensemble members) and calculated the global, ocean and land averages
(using the model's landmask) for the surface air temperature and the
pseudo-MSU-LT diagnostics. 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). For the annual values (as you use in your
paper), I then calculated the expected amplification using a linear
Amplification factors (MSU/SAT) (linear reg./ann data/ 95%conf)
global ocean land
run_a 1.25+/-0.07 1.44+/-0.11 0.95+/-0.07
run_f 1.27+/-0.09 1.49+/-0.10 0.96+/-0.08
run_g 1.28+/-0.08 1.45+/-0.10 0.99+/-0.07
run_h 1.24+/-0.07 1.47+/-0.11 0.92+/-0.07
run_i 1.26+/-0.09 1.49+/-0.12 0.95+/-0.07
The global average amplification is indeed near 1.25, but the value over
ocean is significantly higher, and the value of land significantly less.
Indeed, there is no expected amplification at all!
I attach two figures - one showing the transient behaviour of these
measures in a particular simulation, and the second emulating your figure
1, but using the model diagnostics.
Possible reasons for the discrepancy with Prof. McKitrick's
comunication might lie in what land mask is being used or some issue
related to area weighting or the sampling. However, my calculation above
is certainly more complete and I think more relevant.
Given the potential importance of this for your paper, I thought it best
to notify you as soon as possible. If you would like to check these
calculations on your own, please let me know and I will place the raw data
on our ftp server. If you would prefer a calculation that might be more
specifically tied to the land mask you are using for your averaging,
please let me know what that is and I will update my calculation
Thank you very much for your thorough and informative note that you sent us on Friday. We appreciate the comments.
We first note that your comments relate specifically to the amplification factors currently present in several realizations of a version of the NASA GISS model. In your comments you do not dispute our main conclusion that the there is significant disagreement between the observational satellite and surface temperature datasets, especially over land areas (which obviously is independent of uncertainties within or across models) and that sampling the temperature near the ground, as a means to estimate temperature trends through a deeper layer of the atmosphere, introduces a bias in that context. The use of a global average surface temperature trend that includes that surface data, therefore, overstates the magnitude of climate system heat changes. Your comments provide a welcome confirmation that our analysis is robust to model uncertainties.
Thanks for giving us the newly-calculated amplification factors. We have repeated all of our calculations using the amplification factors that you provided. Although it changes the magnitude of the linear trends, the statistical significance of the differences in trends is only minimally altered. The significance of the trend over land between the Hadley Centre and RSS is no longer statistically significant at the 95% level, however, all other differences between ocean and global trends are now significant using the amplification factors that you provided. The new numbers that you have given us provide additional evidence that there are issues remaining to be resolved associated with the reconciliation of the surface ocean and satellite tropospheric ocean measurements. However, as your analysis helps show, this issue goes well beyond the scope of our paper, which focused on temperatures over land.
Table 1 provides the linear trends using the amplification factors that you provided on Friday along with the original amplification factors in our “in press” paper. Figures 1-6 summarize the temperature trends and compare them with the new amplification factors, in a similar manner to the way that we made the calculations in our “in press” paper.
Table 1. Global, land, and ocean per-decade temperature trends over the period from 1979-2008 for an assumed 1.25 amplification factor over the globe, an 0.95 amplification factor over land and a 1.47 amplification factor over the ocean. Included in parentheses are global, land, and ocean per-decade temperature trends over the period from 1979-2008 for an assumed 1.2 amplification factor as calculated in our “in press” paper. Differences are calculated for the NCDC surface analysis – UAH lower troposphere analysis, for the NCDC surface analysis – RSS lower troposphere analysis, for the Hadley Centre surface analysis – UAH lower troposphere analysis and for the Hadley Centre surface analysis - RSS lower troposphere analysis. Trends that are statistically significant at the 95% level are highlighted in bold face.
Figure 1: HadCRUT3v versus satellite analysis using 1.25 amplification factor for the entire globe.
Figure 2: NCDC versus satellite analysis using 1.25 amplification factor for the entire globe.
Figure 3: CRUTEM3v versus satellite analysis using 0.95 amplification factor for land.
Figure 4: NCDC versus satellite analysis using 0.95 amplification factor for land.
Figure 5: HadSST2 versus satellite analysis using 1.47 amplification factor for ocean.
Figure 6: NCDC versus satellite analysis using 1.47 amplification factor for ocean.
We have also spoken to Ross McKitrick with regards to the calculations he supplied us, using the model output that you had earlier provided. Specifically, he made his calculations from the five runs and ensemble mean that were released with your IJOC paper that you referenced in your email of Friday morning. He calculated these ratios over the 440 grid cells that were available from the period between 1979-2002 which match with the data available from the HadCRUT3 dataset (Figure 7). We accept your offer to put your raw data on an FTP site. In order to replicate your calculations we will need the monthly temperatures since 1979 for all grid cells at the surface and lower troposphere levels. Please send us the FTP address as soon as these have been posted. Thank you very much.
Figure 7: Grid cells for which data is available from CRU over the period from 1979-2002.
Also, have any other GCM groups computed similar land/ocean/entire globe amplification factors? I think a comparison of your result with that of other modeling groups would certainly be of interest to the climate community at large, especially to identify differences across models. The information that you provided reminds us that there remains a very wide range of possible observations that might be judged to be consistent with the very large range of outputs from even a single family of GCMs.
Because the analysis that you have provided represents a useful extension of our original analysis, and strongly shows that it is robust to large model uncertainties, we invite you to join us as a co-author on a short piece along the lines of this response that integrates your initial comments with the additional material presented here.
Thanks again for providing these comments.
Phil Klotzbach, Roger Pielke Sr., Roger Pielke Jr., John Christy, Dick McNider