Bristol, UK, 16 May 2014; Updated 19 May 2014
The study finds significant differences between the three assessments and also finds that the independent assessments of forcing and climate sensitivity within AR5 are not consistent if one assumes the simple energy balance model to be a perfect description of reality.
The overall innovation of the manuscript is very low, as the calculations made to compare the three studies are already available within each of the sources, most directly in Otto et al.
The finding of differences between the three "assessments" and within the assessments (AR5), when assuming the energy balance model to be right, and compared to the CMIP5 models are reported as apparent inconsistencies.
The paper does not make any significant attempt at explaining or understanding the differences, it rather puts out a very simplistic negative message giving at least the implicit impression of "errors" being made within and between these assessments, e.g., by emphasising the overlap of authors on two of the three studies.
What a paper with this message should have done instead is recognising and explaining a series of "reasons" and "causes" for the differences.
- The comparison between observation based estimates of ECS and TCR (which would have been far more interesting and less impacted by the large uncertainty about the heat content change relative to the 19th century) and model based estimates is comparing apples and pears, as the models are calculating true global means, whereas the observations have limited coverage. This difference has been emphasised in a recent contribution by Kevin Cowtan, 2013.
A careful, constructive, and comprehensive analysis of what these ranges mean, and how they come to be different, and what underlying problems these comparisons bring would indeed be a valuable contribution to the debate.
Thus I would strongly advise rejecting the manuscript in its current form.
- The casting of ECS in the odd units of K/(W/m**2) is completely unnecessary and not only is confusing, but makes it difficult to check some of the numerical values reported. ECS should be reported in K since it is a temperature change in response to 2xCO2 forcing. Instead of equation 1, simply write ECS = F(2xCO2) * delta-T/(F – N).
- The present manuscript is unacceptably unclear about exactly what values are used in constructing fig. 1. For the 4 cases considered (AR4, AR5, Otto et al., and the CMIP5 models), you should construct a table (source of info. vs. value of each parameter) providing all the values (or range of values) used for ECS, delta-T(2xCO2), F(2xCO2), F, and N. Indicate which values (if any) were not reported by the referenced study itself, but were adopted for use from some other source. With this information you might be able to convince the reader that there are in fact differences and inconsistencies in each of the studies. As the manuscript stands, I am left wondering whether the apparent discrepancies might actually be explained more by differences in ocean heat uptake values used as opposed to uncertainty/differences in forcing and ECS. I note that many of the discrepancies disappear if AR5 assumed a somewhat larger value for N than in the other studies.
- For clarity (and strict correctness), log-log plots should show the relationship between non‐dimensional quantities only (because taking the log of a dimension yields nonsense). So fig. 1 should show log(ECS/delta-T) as a function of log( (F-N) / F(2xCO2) ). This will also make it easy for the reader to understand the meaning of the numerical values plotted: the ordinate indicates by what factor the GMST equil. response exceeds the temperature difference between some perturbed state and the control (preindustrial) states (in this case warming since pre‐industrial times). On the abscissa, F-N would appear normalized by 2xCO2 forcing. An equilibrium climate with 2xCO2 will by construction be plotted at the origin (i.e., ECS/delta_T(2xCO2) = 1 ).
- In the current manuscript, important assumptions are that uncertainty in delta‐T (obs) and N(obs) is negligible and that the values should be the same for use in all 4 studies. I’m not sure this is valid, since the estimates of “present‐day” forcing are for different time periods (I think). Moreover, it does not seem consistent to evaluate N over the period from 1971 to 2010 and GMST change from 1850‐1900 to 2003‐2012. For this to be an appropriate comparison, you must assume the rate of ocean heat uptake is the same during 2003‐2012 as it is during 1971-2010. You also must assume that in the period 1850‐1900 the system is in equilibrium (with N=0 during that period). I note that Otto et al. assume heat uptake of 0.08 +‐ .03 W/m**2 for their reference period (1860-1879). I suspect for 1850-1900, the comparable number might be somewhat larger, which would reduce your N by a non‐negligible fraction. In any case N is highly uncertain, and you should discuss how this affects your results. Similarly, you need to consider uncertainty in GMST. Although this quantity is reasonably well measured over the historical period, we cannot expect it on short time‐scales to necessarily exactly be related to net radiative flux by a constant. These and other uncertainties lead to the range of values shown, for example, in Otto et al. (for the decade 2000‐2010 values range around delta‐T = 0.7 K with a standard error estimate of about 20%, at least as best I can determine from their fig. 1). Again how these uncertainties affect each of the 4 studies you consider should be discussed; it’s possible that the discrepancies could disappear if different studies used different values of N and GMST change within the accepted uncertainties.
- One way to better indicate uncertainties on your graph would be to replace your log‐log plot with a plot of (F-N)/F(2xCO2) along the ordinate and 1/ECS on the abscissa, perhaps labeled non-linearly with values of (1/6, 1/5, ¼, ½, and 1) so the reader could directly read the temperature. On this plot you could then indicate the region compatible with temperature observation uncertainty by plotting a couple of lines emanating from the origin with slope equal to different values of delta-T [nb. delta-T /ECS = (F‐N)/F(2xCO2)]. You could also indicate how the uncertainty in N affects your projection lines corresponding to the current diagonal lines in your fig. 1 by plotting at their central point a vertical error‐bar line (vertical because recall I’ve put F‐N on the y-axis). This figure would resemble fig. 1 of Otto et al., but with their obs. change in GMST replaced by 1/ECS and their shaded diagonal lines of ECS replaced by GMST. You would probably only have to display 2 diagonal lines indicating the uncertainty in obs. GMST change.
- In your current discussion you imply that differences in F‐N across different studies are attributable to differences in F, but N could also be responsible.
- The study would be much more valuable if it attempted to also begin to address the 4 questions posed in the conclusions. I suspect the answers are really quite mundane, although the tone of the discussion implies otherwise.