by Nic Lewis
In a recent article here, which summarised a longer piece at ClimateAudit, I discussed the December 2015 Marvel et al.[1] paper, which contends that estimates of the transient climate response (TCR) and equilibrium climate sensitivity (ECS) derived from recent observations of changes in global mean surface temperature (GMST) are biased low.
Marvel et al. reached this conclusion from analysing the response of the GISS-E2-R climate model in simulations over the historical period (1850–2005) when driven by six individual forcings, and also by all forcings together, the latter referred to as the ‘Historical’ simulation. Using their ‘iRF’ measure of forcing, I can largely replicate Marvel et al.’s results. However, there are doubts over whether the calculated iRF values for the Historical simulation did in fact include land use change (LU) forcing. See my related longer article at ClimateAudit, here, for an examination of this issue.
I want to concentrate here on exploring further the response to LU forcing, which in these simulations primarily represents increased albedo. Areas cleared for cultivation and grazing are typically brighter than the forest or other natural vegetation they replace. Marvel et al. found LU iRF forcing to have an extremely high ‘transient efficacy’. The model’s GMST response to the (negative) LU iRF forcing over the historical period, relative to what it would have been if the same forcing had been caused by changing the CO2 concentration (its transient efficacy), was almost four. This very high LU efficacy appears to account for at least half the apparent low bias in TCR estimation from the Historical simulation when using iRF. As I wrote originally, Marvel et al. used an unphysical regression model to estimate iRF efficacies. That accounts for most, but not all, of the large excess over one of their efficacy estimate for LU iRF. Here I want to examine the explanation for the remaining excess.
Ensembles of five simulation runs were carried out for LU and each other individual forcing. Marvel et al.’s estimates were based on averaging over the relevant simulation runs; undertaking multiple runs and taking the ensemble average reduces the impact of random variability. I commented in my original article that LU run 1 showed a much stronger negative GMST response than any of runs 2 to 5, from the middle of the 20th century on (see Figure 5 in the original article). The iRF transient efficacy calculated from this run is extraordinarily high: 9.5.
I conjectured that run 1 might be a rogue. Excluding it as well as using a physically-consistent estimation method for estimating iRF transient efficacy for iRF would reduce the estimate from 3.89 to, depending on the exact method used, 0.7 to 1.1. Excluding also the LU run with the lowest GMST response, to balance excluding the run with the highest GMST response, would cause little further change in the efficacy estimate.
Figure 1 shows the changes simulated by GISS-E2-R in response to LU forcing for each of simulation runs 2 to 5. They compare mean temperatures over 1976–2000 with those over 1850–75. Note that the temperature scale in Figure 1 is double that in subsequent figures.
Figure 1. Simulated surface temperature change (1850–75 to 1976–2000 mean) driven by land use change forcing only: maps a to d show results from runs 2 to 5 respectively.
As the forcing and resulting temperature changes are small, internal variability has a significant effect on simulated changes even when comparing 25 year means, with changes varying in sign over some land areas and most of the ocean. Figure 2, which shows the average of all the plots in Figure 1, confirms that temperature changes for the runs 2 to 5 mean are small everywhere. Globally averaged, there is a cooling of 0.04°C. Although land use forcing in GISS-E2-R is very geographically concentrated (Figure 4c of Miller et al 2014 shows 2000 values), its effects on surface temperature are widely distributed, and generally not linked to where the forcing takes place. The generally greater cooling in land masses than over the ocean is mainly due to temperatures over land generally being more sensitive to global forcing, not to LU forcing being located in land masses.
Figure 2. Average of temperature changes for runs 2 to 5, as plotted in Figure 1.
However, the temperature changes simulated in run 1, shown in Figure 3 are very different from the average of the other runs. The global average cooling is 0.35°C, nearly nine times as high as for the average of runs 2 to 5. And the next highest cooling, in run 3, is only 30% as large.
Figure 3. LU run 1 surface temperature change from 1850–75 to 1976–2000 mean.
In run 1, a very cold anomaly develops in the ocean south of Greenland. In most of the dark blue patch, the temperature has dropped by over 3°C, and by over 5°C in the centre of this area. Additional cooling to that shown in runs 2–5 has developed almost everywhere, apart from in the ocean around Antarctica, where some areas have warmed strongly and others cooled strongly. This all seems to point to some major change in the ocean overturning circulation having occurred in run 1, resulting in the cold ocean anomaly south of Greenland and substantial surface cooling in most areas.
Whatever the exact cause of the massive oceanic cold anomaly developing in the GISS model during run 1, I find it difficult to see that is has anything to do with land use change forcing. And, whether or not internal variability in the real climate system might be able to cause similar effects, it seems clear that no massive ocean temperature anomaly did in fact develop during the historical period. Therefore, any theoretical possibility of changes like those in LU run 1 occurring in the real world seems irrelevant when estimating the effects of land use change on deriving TCR and ECS values from recorded warming over the historical period. Hence I think there is a good case for excluding LU run 1 when estimating LU forcing efficacies.
Interestingly, I have very recently discovered (from Chandler et al 2013) that there was an error in the ocean model in the version of GISS-E2-R used to run the CMIP5 simulations, which may be relevant to the development of the ocean anomaly in LU run 1. Chandler et al. write, in a paper about simulating the Pliocene climate using GISS-E2-R:
“We discuss two versions of the Pliocene and Preindustrial simulations because one set of simulations includes a post-CMIP5 correction to the model’s Gent-McWilliams ocean mixing scheme that has a substantial impact on the results – and offers a substantial improvement, correcting some serious problems with the GISS ModelE2-R ocean.”
Chandler et al. say that a miscalculation in the isopycnal slopes in the mesoscale mixing parameterisation led to spurious heat fluxes across the neutral surfaces, resulting in an ocean interior generally too warm, but with southern high latitudes that were too cold.
Interestingly, they also write, comparing results from the corrected (GM CORR) and uncorrected model (GM UNCOR):
“One of the most significant differences of the Pliocene GM CORR simulations, compared with those of the uncorrected model, is the characteristic of the meridional overturning in the Atlantic Ocean. In GM UNCOR the Atlantic Meridional Overturning Circulation (AMOC) collapsed and did not recover, something that was expected to be related to problems with the ocean mixing scheme. Although we hesitate to state that this is a clear improvement (little direct evidence from observations), it seems likely that the collapsed AMOC in the previous simulation was erroneous.”
It occurs to me to wonder whether this error in the GISS-E2-R ocean mixing parameterisation, which gave rise to AMOC instability in the Pliocene simulation, might possibly account for the model’s behaviour in LU run 1. It looks to me as if something goes seriously wrong with the AMOC in the middle of the 20th century in that run, with no subsequent recovery evident.
Schmidt et al. wrote, in their paper about the GISS ModelE2 contributions to the CMIP5 archive:
“Since the simulations were completed, we discovered an error in the implementation of the skew flux Gent-McWilliams parameterization which had the effect of making the eddy-related diffusion more horizontal than intended. We are currently exploring the impact of this error, and the results will be reported elsewhere.”
Although the ocean error was identified at least three years ago, I have not found any publication presenting the results of the investigations into its impacts. It is of course possible that GISS has concluded that the impacts are negligible or perhaps small except where an instability arises. If this is the case it would be good to have a clear statement from GISS confirming such.
The single forcing simulations were part of the CMIP5 design. Although it is possible that some or all of them were run after the ocean correction was implemented, I can see no evidence of that being the case and I think it is unlikely. A commentator on the ClimateAudit thread has asked Gavin Schmidt, in a comment submitted to RealClimate, whether temperature and net flux data for GISS-E2-R available via the CMIP5 portals and KNMI Climate Explorer are based on a model corrected to fix the ocean heat transport problem. It would be helpful and refreshing to get a clear response to this question – and also to an earlier question asking what are the iRF and ERF forcing values in GISS-E2-R attributable to a doubling of CO2 concentration, upon which all the results in Marvel et al. depend.
[1] Kate Marvel, Gavin A. Schmidt, Ron L. Miller and Larissa S. Nazarenko, et al.: Implications for climate sensitivity from the response to individual forcings. Nature Climate Change DOI: 10.1038/NCLIMATE2888. The paper is pay-walled, but the Supplementary Information (SI) is not.Filed under: climate models, Sensitivity & feedbacks
