By Nic Lewis
This article concerns the paper “Globally resolved surface temperatures since the Last Glacial Maximum” by Matthew Osman et al.[2] (hereafter Osman 2021) published by Nature in November 2021.
Introduction
Readers may recall my articles in 2018 about statistical flaws in a Nature paper that claimed to show ocean warming was greater than generally thought[1]. That paper was subsequently retracted. Nature tends to publish papers that use novel approaches and/or provide newsworthy results, which have not yet stood the test of time. Given that background, and that peer review often fails to spot problems with methods or calculations, I read Nature papers with particular care. Helpfully, Nature makes peer review files available, unlike most journals. Although Nature does not also make the submitted version available, unlike the EGU Open Access journals, authors are increasingly posting open-access preprints when submitting manuscripts to journals, thus revealing changes made during peer review or after acceptance.
This article concerns the paper “Globally resolved surface temperatures since the Last Glacial Maximum” by Matthew Osman et al.[2] (hereafter Osman 2021) published by Nature in November 2021. Its Abstract reads:
Climate changes across the past 24,000 years provide key insights into Earth system responses to external forcing. Climate model simulations and proxy data have independently allowed for study of this crucial interval; however, they have at times yielded disparate conclusions. Here, we leverage both types of information using paleoclimate data assimilation to produce the first proxy-constrained, full-field reanalysis of surface temperature change spanning the Last Glacial Maximum to present at 200-year resolution. We demonstrate that temperature variability across the past 24 thousand years was linked to two primary climatic mechanisms: radiative forcing from ice sheets and greenhouse gases; and a superposition of changes in the ocean overturning circulation and seasonal insolation. In contrast with previous proxy-based reconstructions our results show that global mean temperature has slightly but steadily warmed, by ~0.5 °C, since the early Holocene (around 9 thousand years ago). When compared with recent temperature changes, our reanalysis indicates that both the rate and magnitude of modern warming are unusual relative to the changes of the past 24 thousand years.
Matthew Osman kindly sent me a copy of the paywalled paper; the submitted version preprint is available here. The significance of this topic relates both to the use of Last Glacial Maximum (LGM) to preindustrial global temperature change to estimate climate sensitivity, and the magnitude and causes of temperature variability since the LGM, including in particular how preindustrial global temperature compares with that in the early Holocene.
Osman 2021 reconstructed time-series of global, spatially-resolved surface temperatures from the LGM period some 20,000 or so years ago to the preindustrial late Holocene, through extending data-assimilation methods developed by its second author, Jessica Tierney, and used by her in a 2020 paper to spatially reconstruct LGM temperatures[3] (hereafter Tierney 2020).
Although Tierney 2020 states that best estimates of the LGM global mean surface air temperature (GMAT)[4] change relative to preindustrial ranged from −1.7°C to −8.0°C, most were between −3.0°C and −6.0°C. For what it is worth, in the two most recent global climate model (GCM) generations simulated preindustrial-to-LGM GMAT change varied from –2.7°C to −5.4°C and from −3.3°C to −7.2°C.[5] Tierney 2020 estimated the preindustrial to LGM GMAT change (from the stable period 19,000 to 23,000 years ago) to be −5.9°C, with a tight −6.3°C to −5.6°C 95% uncertainty range. Osman 2021 go further, estimating that change as −6.8 ± 1.0 °C (95% uncertainty range)[6]. Surprisingly, given the very similar methodology and proxies they used, their two estimates are almost statistically inconsistent.[7]
Both Osman 2021 and Tierney 2020 used a large selection of four types of sea surface temperature (SST) proxies; no land or deep ocean proxies were used. It appears that the 539 proxies used by Osman 2021 – which sought long-record, time resolved proxies to enable reconstruction over the whole LGM to preindustrial period – were largely a subset of the 955 LGM and 880 late Holocene proxies used in Tierney 2020. Proxy coverage in the central Pacific ocean, which was very limited in Tierney 2020 with one proxy near the equator and one each in the north and south, is non-existent in Osman 2021. It is unclear why those three proxies, which were common to the LGM and late Holocene periods, were not used in Osman 2021.
Osman 2021’s proxy-only reconstruction
In addition to their main data-assimilation reconstruction (“LGMR”), Osman 2021 produced a proxy-only reconstruction. I will start by examining that. They estimate local SST from proxy data, a process subject to substantial uncertainty and possible bias.[8] By binning these local SST estimates by age-range and latitude-band, they derive mean 60°S−60°N SST at 200 year resolution and scale this to give an estimated GMAT time series.[9] This well established procedure inevitably involves uncertainty and possible bias. However, unlike with the data-assimilation method, the resulting GMAT estimates are not dependent on the spatial and temporal accuracy of paleoclimate simulations by a single GCM, which may well be poor.
Fig. 4 of their Nature paper shows the Osman 2021 proxy-only reconstruction, over 0–22,000 yr BP[10] in panel a and, enlarged, over 500–11,000 yr BP in panel b. Helpfully, Nature makes the underlying plot data available. Unfortunately, there are at least two errors in panel b. First, the dotted lines connecting the time-axes of panels a and b are incorrect; they wrongly imply that panel b covers 0–10,600 yr BP. Secondly, Fig. 4b show a value of −0.46°C for 11,000 BP but no such data point exists[11], while the Fig. 4b plot data is shown as starting at 500 yr BP but actually starts at 100 yr BP.
Of greater concern, the Fig. 4a full period proxy-only reconstruction in Nature differs in shape from the version in their preprint. Figure 1 compares the two versions, matched at age 0, with identical SST to GMAT scaling factor of 1.90.[12] (The actual mean scaling factor used in the preprint was much higher at 2.44, but that was obviously inappropriate.[13]) The Nature version shows a larger temperature change between the LGM and the early Holocene, and less pronounced short term fluctuations. For example, the dip between 12,000 and 13,000 yr BP is much smaller, and the temperature decline between 5,000 and 1,000 yr BP is much more even. This may indicate added smoothing, but that can’t be the whole explanation. It is unclear from the Methods descriptions in the preprint and Nature versions of the paper why their proxy-only reconstructions differ.[14] I wrote to Matthew Osman asking about this but received no reply. It seems surprising that the authors changed their proxy-only reconstruction without making any mention of doing so in any of their peer review responses.
Fig 1. Osman 2021 Nature and preprint Fig. 4a proxy-only reconstructions compared
In the absence of a satisfactory justification for these differences, I give more credence to the preprint proxy-only reconstruction. As the two versions of Osman 2021 Fig. 4a show, that matches a scaled version of the well-established Shakun-Marcott Curve (SMC) reconstruction considerably more closely between 14,000 and 1,000 yr BP than does the Nature version.
When applying the 1.90 mean 60°S−60°N SST scaling factor, as per the Nature paper, to the preprint proxy-only reconstruction, the mean GMAT over the stable period 19,000 to 23,000 years ago (commonly used to represent the LGM) is 5.0°C cooler than that over the last 600 years, which appears to be a suitable measure of preindustrial GMAT[15].
However, the 1.90 scaling factor appears excessive. It reflects the mean ratio of GSAT to 60°S−60°N SST in PMIP2 and PMIP3 GCM simulations of the LGM.[16] PMIP2 simulations are very dated. The corresponding mean from PMIP3 simulations alone was 1.84.[17] Moreover, published results from the more recent PMIP4 simulations, involving a larger ensemble of more advanced models with updated estimates of ice sheet configurations and other boundary conditions, show a mean ratio of GSAT to 60°S−60°N ocean surface air temperature of 1.64, 6% lower than that for PMIP3.[18] That points to reducing the PMIP3 derived 1.84 ratio of GSAT to 60°S−60°N SST to 1.73. Applying that PMIP4 ratio to the Osman 2021 preprint proxy-only reconstruction would reduce the preindustrial to LGM GSAT change to −4.6°C, as compared with −4.85°C using the PMIP3 ratio of 1.84. All these estimates are of course subject to substantial uncertainty.
In GCMs, ocean surface air temperature changes more than SST, however the best estimate per the IPCC AR6 report is that changes in SST and in ocean surface air temperature are the same[19]. On that basis, the PMIP4 derived scaling ratio of 1.64 from 60°S–60°N ocean surface air temperature to GMAT changes can be applied directly to SST changes. Applying that ratio to the Osman 2021 preprint proxy-only reconstruction would reduce the preindustrial to LGM GSAT change to −4.3°C.
Osman 2021’s data-assimilation reconstruction (LGMR)
The data-assimilation method used involves, in essence, simulating the global climate at suitable dates using a GCM and then modifying the simulation values at each spatial grid location, statistically interpolated to each 200 year reconstruction time step period[20], by reference to the difference between all proxy values relevant to that time step and the interpolated simulation values at the proxy locations. Their method implements a ‘Kalman filter’ update.
The extent of the modification made at each grid point depends on the covariance across an ensemble of ‘model priors’, representing means of 50 year GCM simulation time slices, of temperatures at that grid-point with those at all sites where proxy values at or close to the same date exist. It also depends, inversely, on sum of the covariance of the simulated values and scaled-down uncertainty in the proxy values. Unfortunately, Osman 2021 does not show the ensemble-mean model prior, nor the difference between it and the LGMR reconstruction, either as a global mean time series or as a map at the LGM. It is therefore not possible to tell to what extent the model prior has been modified in producing the reconstruction.[21]
As well as, like the proxy-only reconstruction, depending on the accuracy of the proxy-derived local SST estimates, the accuracy of the LGMR data-assimilation reconstruction therefore depends critically on the realism of the GCM simulations used and of the spatial covariance structure that they produce. The covariances used appear to be applicable to multidecadal internal variability in the GCM used, and may be different to those applicable to longer term forced climate change.[22]
Data assimilation is a technique suited to cases where observational evidence is abundant and reasonably accurate, with modest uncertainty, and where in addition the model simulations used are known to represent spatiotemporal changes and their covariance reasonably well. In such cases the reconstruction should be dominantly determined by information from the observations. However, none of those conditions are satisfied at the LGM.
When observational evidence is limited and highly uncertain and/or model covariances are low, a data assimilation reconstruction will largely represent model simulation values. It is therefore relevant to note that the model used by Osman 2021, iCESM1.2[23], has a particularly high preindustrial to LGM change in GMAT. The main, CESM1.2, PMIP4 model cooled by 6.8°C, over 2°C more than the PMIP4 average.
It is moreover not entirely clear that the iCESM1.2 model used had been adequately equilibrated before the simulations were carried out: the two preindustrial simulation ensembles listed in Extended Data Table1 had mean GMSTs with ranges that were narrow (14.03–14.25°C and 13.22–13.33°C) but differed by almost 1°C[24].
The Osman 2021 preprint Extended Data Figure 6 gives some indication of the LGMR reconstruction’s dependence on the model-simulation prior. It shows that in many cases best-estimate LGMR GMAT changes exceed those per both the model simulation prior and the proxy-only reconstruction, even before correcting the erroneously high 2.44 60°S−60°N SST to GMAT scaling factor used for the preprint proxy-only reconstruction. Since that scaling factor is far lower in the iCESM1.2 model used, larger LGMR changes than for either the model prior or the proxy-only reconstruction suggest strong dependence of the LGMR GMAT on model-simulated SST changes at the proxy locations generally being lower than their average for the latitude involved (so that proxy changes are larger than those per the model simulation at the same location). However, no evidence is provided as to why that would be the case.
Worryingly, the LGMR results presented in Nature differ from those in the preprint, with cooling at the onset of deglaciation 0.2°C greater in the Nature version. Moreover, the spatial cooling between 9 kyr and 2 kyr shown in the two Osman 2021 Figure 2 versions differs substantially. I could not spot any mention in the peer review files of changes being made.
Validating the LGMR data-assimilation reconstruction
It is impossible to properly verify the spatiotemporal accuracy of the model simulations used, however Osman 2021 do carry out some, albeit limited, validation tests of their LGMR reconstruction. Their ‘external validation’ consists of comparing the LGMR (posterior) Δδ18Op values, and also the model simulation (prior) values, at the proxy locations, with independent ice core and cave speleothem proxy record data. This is similar to the validation test undertaken by Tierney 2020.
Osman 2021 claim that their external validation test indicates that LGMR substantially improves over the model prior, with the 62% of the variance in the independent records explained by LGMR against 37% by the model prior (simulation values). That is, the R2 improves from 0.37 to 0.62, as shown in Figure 2. This is the final, Nature, version. Oddly, it is not identical to the preprint version, without any mention in the peer review file of any changes being made.
Fig 2. Reproduction of Osman 2021 Extended Data Fig. 3(i)&(j). Triangles represent ice core values and circles represent cave speleothem values, relative to preindustrial, for each Differencing period (simulation time slice). ‘Prior’ refers to the model simulation values. ‘Posterior’ refers to LGMR reconstruction values.
However, Δδ18Op values for different differencing periods are not independent, since changes from preindustrial to any time slice except the final, 3 kyr BP, period will include changes since later time slices. An even more serious concern is that combining the ice core and cave speleothem proxies when estimating the explained variance will artificially increase the proportion of variance explained, since they/their locations evidently have very different sensitivities to temperature changes between preindustrial and the LGM.[25] Osman 2021’s claimed model prior R2 of 0.37, and higher LGMR posterior R2 of 0.62, are spurious, misleading figures.
A fairer test would be to consider changes only over the longest differencing period, that between preindustrial and 21 kyr BP[26], and to analyse the ice core and speleothem datasets separately. To clarify the data concerned, Figure 3 shows 21 kyr BP minus preindustrial differences for the model simulations (Prior: blue symbols) and the LGMR reconstruction (Posterior: red symbols), plotted against the co-located proxy (Observed) values.
Fig 3. Ice core values (triangles) and cave speleothem values (circles) for the 21 kyr differencing period, digitized from Osman 2021Extended Data Fig. 3(i)&(j). Blue symbols show model simulation (Prior) values. Red symbols show LGMR reconstruction (Posterior) values. Observed values are in both cases shown along the x-axis. The blue and red lines show the best fit for Observed values when predicted by (regressed on) respectively the model Prior and the LGMR reconstruction Posterior values. The black lines join Prior and Posterior points with the same ordering of Observed (x) digitized values in Extended Data Fig. 3(i) and Fig. 3(j) respectively.
Ice core proxy validation results
For the ice core records, the ordered Observed (x) values from Extended Data Fig. 3(i) (Prior) and Fig. 3(j) (Posterior) are all identical within digitization error (max 0.05‰), with the black lines joining the Prior and Posterior y-values being vertical, as should be the case. The data-assimilation method changes the model simulation values, but it cannot change the observed proxy values.
The best fit when predicting observed values from the model prior ice core values (blue line) explains a negligible proportion of variance (R2 = 0.09; adjusted R2 < 0). The best fit when predicting observed values from the LGMR posterior ice core values (red line) is better, but it is not statistically significant (p >0.05) and it only explains a minority of the variance (R2 = 0.32; adjusted R2 = 0.25). Moreover, the slopes of the best fits are in both cases far from a 1-to-1 relationship through the origin (green dashed line) , albeit less so for the LGMR posterior.
A large majority of the model prior values, and of the LGMR posterior values, show larger changes than the observed values. The mean observed Δδ18Op value is −5.4‰, while the model prior mean is a third larger at −7.2‰ and the LGMR posterior mean is almost as large at 7.0‰. Thus, both the model prior and the LGMR posterior substantially overestimate mean LGM cooling at the ice core locations, by respectively 33% and 31%[27]. Interestingly, if the −6.8°C preindustrial to LGM change in GMAT per the LGMR posterior were scaled down by 1/1.31, it would become −5.2°C, quite close to the change implied by the proxy-only reconstruction.
Cave speleothem proxy validation results
Observed cave speleothem values have a negligible correlation with either prior or posterior values (R2 < 0.001). That provides no support for the validity of either the model simulations or the LGMR reconstruction.
Moreover, a majority of the ordered cave speleothem proxy record values in Extended Data Fig. 3(j) do not match those in Fig. 3(i) within 0.1‰, double the digitization error, as shown by the black lines joining the circles not being vertical in some cases. The most negative Observed cave speleothem (leftmost dark blue circle in each of panels (i) and (j) of the original figure ) values are glaringly different: −5.65‰ and −6.15‰: a discrepancy of 0.5‰ , ten times digitization error. This cannot possibly be correct. Something has evidently gone seriously wrong here, and a correction by Osman and co-authors appears to be required.
Other tests
Osman 2021 also show results for what they call ‘internal validation testing’ using values for 20% of their proxies randomly withheld at each time step. However, this appears to be a weak test; most of their proxies have nearby other proxies. Moreover, as they say, the U-shaped (rather than flat) rank histogram indicates a lack of structural variance in the model prior. Tierney 2020 carried out a similar analysis, but did not claim that the results validated their LGM reconstruction. Indeed, Osman 2021 seemingly admit as much in a response to a reviewer, writing “our contention that LGMR improves upon our model priors is not based on any particular metric of prior vs. posterior ensemble spread, but rather our external validation tests”.
Conclusions
I do not consider Osman 2021’s main LGMR, data-assimilation reconstruction, which estimates 6.8°C mean LGM cooling, to be reliable. It is highly dependent on the spatiotemporal accuracy of LGM simulations by a single GCM. The external validation tests, when analysed properly, show no significant skill by either the model simulation prior or the LGMR posterior in predicting observed LGM changes in independent cave speleothem records. Moreover, those tests show that both the model prior and the LGMR posterior substantially overestimate observed LGM changes per independent ice core records. While the LGMR reconstruction is innovative, I am doubtful that the accuracy of model simulations and the spatial coverage and accuracy of proxy data over the period extending back to the LGM are adequate for a data-assimilation approach to give unbiased GMAT estimates. In my view, the method employed by Annan and Hargreaves (2013)[28], which scales model-simulation cooling patterns to best match proxy data (giving an estimate of 4.0°C LGM cooling in their case), is more suitable.
In my view, Osman 2021’s preprint proxy-only reconstruction appears credible, although a lower SST multiplier than they used is appropriate. I consider it to be more credible than the version of their proxy-only reconstruction published in Nature, as no explanation was given for that being different from the reconstruction in their preprint, and it matches the shape of the de facto preexisting standard SMC reconstruction time series less well than the preprint version. Using a multiplier based on the mean from the most recent, PMIP4 models, estimated average LGM GMAT was ~4.6°C cooler than preindustrial per the preprint proxy-only reconstruction. If, unlike in GCMs, SST actually changes as much as marine air temperature, as assessed by the IPCC in AR6, then the implied GMAT cooling would instead be ~4.3°C.
Nicholas Lewis April 2022
Addendum
Since writing this article a new Holocene study, Thompson et al. (2022), has been published in Science Advances (open access). Using the same GCM as Osman 2021, it shows that adding forest cover in the Sahara and mid-latitudes (partial at 3 kyr BP) and (except at 3 kyr BP) in the Arctic, to match pollen records substantially increases simulated GMAT at 3, 6 and 9 kyr BP. The difference at 6 kyr BP is 0.72°C, with the resulting GMAT being well above the preindustrial level. Osman 2021’s simulations incorporated this greening only partially.[29] Even with all these regions greened, the model simulation (prior) 6 kyr BP GMAT was still well below that of the Osman 2021 proxy-only reconstruction. These facts suggest that Osman 2021’s model prior may be particularly unsatisfactory, spatially as well in the global mean, during the Holocene.
[1] L. Resplandy, R. F. Keeling, Y. Eddebbar, M. K. Brooks, R. Wang, L. Bopp, M. C. Long, J. P. Dunne, W. Koeve & A. Oschlies, 2018: Quantification of ocean heat uptake from changes in atmospheric O2 and CO2 composition. Nature, 563, 105-108. https://doi.org/10.1038/s41586-018-0651-8
[2] Osman, M.B., Tierney, J.E., Zhu, J., Tardif, R., Hakim, G.J., King, J. and Poulsen, C.J., 2021. Globally resolved surface temperatures since the Last Glacial Maximum. Nature, 599(7884), pp.239-244. https://www.nature.com/articles/s41586-021-03984-4 Preprint available at https://eartharxiv.org/repository/object/2219/download/4584/
[3] Tierney, J.E., Zhu, J., King, J., Malevich, S.B., Hakim, G.J. and Poulsen, C.J., 2020. Glacial cooling and climate sensitivity revisited. Nature, 584(7822), pp.569-573. https://www.nature.com/articles/s41586-020-2617-x
[4] GMAT is referred to as GMST in Osman 2021.
[5] Kageyama, M., et al.., 2021. The PMIP4 Last Glacial Maximum experiments: preliminary results and comparison with the PMIP3 simulations. Climate of the Past, 17(3), pp.1065-1089. Note that of the four PMIP4 simulations with a GMAT exceeding 5.3°C, three are by a variants of HadCM3, an old model that dates back three climate model generations, and one is by the base-version model (CESM1-2) of that used by Osman 2021, for which it was subsequently found that the simulated LGM climate is very sensitive to treatments of cloud microphysical processes.
[6] They also estimate cooling of 7.0°C at the point of deglaciation onset, but it is standard to take the mean over a period of several thousand years. Moreover, their only GCM simulations circa the LGM were at 18,000 and 21,000 yr BP
[7] Taking the uncertainty distributions to be normal, the distribution of their difference has only a little over 5% of its probability in the region where the two estimates are compatible.
[8] The estimated relationship between proxy data and local SST together with other relevant climate variables is reflected in ‘forward models’ of proxy values. SST estimates can then be derived using the proxy data and estimates of the other climate variables involved, via Bayesian ‘inverse models’. SST estimates therefore depend on the forward models, on the related Bayesian inverse models, and on the ancillary climate variable estimates used in addition to the SST proxy data. Much of the resulting uncertainty is unavoidable, however the use of Bayesian inverse models introduces a further source of uncertainty and possible bias, in addition to that (including as to age calibration) arising directly from the nature of the proxies.
[9] The 60°S–60°N mean SST to GSAT scaling factor used in the final Nature paper was 1.90, a mean value originally derived from PMIP2 and PMIP3 simulation data.
[10] BP: before present
[11] The 200-yr resolution reconstruction values are all at odd multiples of 100 years BP.
[12] The plotted lines are accurately digitized copies, with the preprint version rescaled from 2.44 to 1.90 times mean 60°S−60°N SST to GMAT to match the 1.90 scaling in the Nature version. The preprint line has also been shifted by −0.25°C to rebase it to match the Nature version (which takes preindustrial from the means of up to the last 4,000, rather than 1,000, years proxy values). Note that the Fig. 4a version in Nature appears to have been stretched slightly in the time dimension and shifted it by 100 years, relative to the published data. The preprint version very likely did the same, since its turning points coincide with those in the Nature version.
[13] The 2.44 mean scaling factor was taken from a study involving a proxy-derived estimate of the change in mean temperature of the ocean interior, not of the change in SST. Bereiter, B., Shackleton, S., Baggenstos, D. et al., 2018. Mean global ocean temperatures during the last glacial transition. Nature 553, 39–44.
[14] One potentially relevant difference is that only the Nature version mentions including data from those proxies for which an estimate of preindustrial temperature could not be derived from core-top values, with instrumental reconstruction values being used instead. It is not obvious why that should lead to a larger LGM to preindustrial temperature change in the Nature version, but it is possible since SST estimates from the affected proxies might generally be negatively biased and/or the .instrumental reconstruction values might have positive biases.
[15] This is after increasing the warning from the 19,000–22,000 yr BP period, which the proxy-only reconstruction does not extend beyond, by the 0.015°C by which Osman 2021’s LGMR 19,000-23,000 yr BP mean GMAT was cooler than its 19,000-22,000 yr BP mean. Mean PI temperature is almost the same whether the mean of the last 600, 400 or 200 years is used. Note that the mean scaling factors used in the Nature version of Osman 2021 have been used, with uncertainty in them, and other uncertainties, being ignored.
[16] Snyder, C.W., 2016. Evolution of global temperature over the past two million years. Nature, 538(7624), pp.226-228. https://climate.fas.harvard.edu/files/climate/files/snyder_2016.pdf
[17] Friedrich, T., Timmermann, A., Tigchelaar, M., Elison Timm, O. and Ganopolski, A., 2016. Nonlinear climate sensitivity and its implications for future greenhouse warming. Science Advances, 2(11), p.e1501923.
[18] Using the more recent PMIP4 data, from simulations by much more advanced models and incorporating updated ice-sheet etc. boundary conditions I derive a mean scaling factor of 1.64 from 60°S–60°N ocean surface air temperature to GSAT. (Kageyama et al, 2021.The PMIP4 Last Glacial Maximum experiments- preliminary results and comparison with the PMIP3 simulations, Clim.Past, https://doi.org/10.5194/cp-17-1065-2021) Kageyama et al, 2021 does not give values for SST.
[19] Although in GCMs ocean surface air temperature changes more than SST, the IPCC AR6 report concluded that evidence for this, either from theory or observations, was poor, and that GCM behaviour might reflect a common model bias arising from use of the same parameterization. AR6 therefore assessed, as its best estimate, no difference between SST and ocean surface air temperature changes.
[20] Since there are 2 or 3 kyr gaps between the GCM simulations it is difficult to see that the true time resolution of the reconstruction can be anywhere near as good as 200 years.
[21] Previously developed code packages that implement the Kalman filter and the proxy forward models have been made publicly available, but not the code that Osman 2021 developed to produce their reconstructions using those packages, nor the model simulation data that they used. This makes replicating their results and evaluating the effect of varying their assumptions exceedingly difficult if not impossible.
[22] It is not completely clear to me from the Methods descriptions in Osman 2021 and Tierney 2020 how exactly the model priors and covariances were calculated, or the down-weighting applied, but that is not directly relevant to the issues raised here. Note that to limit spurious relationships between proxies and far away regions, a localization weighting was applied that downweights covariances between far distant points, with strong downweighting where their separation exceeds ~10,000 km.
[23] iCESM1.2 is a water isotope-enabled variant of the main CESM1.2 model. For two simulations, Osman 2021 used the similar iCESM1.3 variant in addition to iCESM1.2.
[24] The former range is much closer to the value of 14.2°C for CESM1.1 per Figure 1 of Bacmeister et al., 2020. CO2 Increase Experiments Using the CESM: Relationship to Climate Sensitivity and Comparison of CESM1 to CESM2. JAMES12, e2020MS002120. https://doi.org/10.1029/2020MS002120
[25] If one combines two sets of paired (x,y) values, each with zero correlation between its x and y values, but the two sets have different mean x and y values, the x and y values of the merged data set will be correlated.
[26] 21 kyr BP is the mid-point of the stable 19-23 kyr BP period usually taken as representing the LGM.
[27] Regression forced through the origin gives a similar value for LGMR posterior overestimation of LGM cooling.
[28] Annan, J.D. and Hargreaves, J.C., 2015. A perspective on model-data surface temperature comparison at the Last Glacial Maximum. Quaternary Science Reviews, 107, pp.1-10. https://dx.doi.org/10.1016/j.quascirev.2014.09.019
[29] None of Osman 2021’s 3kyr BP simulations partially greened the Sahara and northern hemisphere mid-latitudes. Only two thirds of their 6 kyr BP simulations greened the Sahara and the Arctic (north of 50°N); none greened mid-latitudes. None of their 9 kyr BP simulations greened either the Arctic or mid-latitudes. Moreover, their 6 kyr BP simulations that did green the Sahara and Arctic had a mid-range GMAT only 0.27°C warmer than those that didn’t do so, whereas per Thompson et al. in such simulations the average difference was 0.57°C.
Originally posted here, where a pdf copy is also available
Pingback: Osman et al. 2021: a flawed Nature paleoclimate paper? – Climate- Science.press
Pingback: Osman et al. 2021: a flawed Nature paleoclimate paper? - News7g
The worst of this to me is the misguided conception of climate change as global mean annual temperature.
Were winter temperatures to decrease, but summer temperatures to decrease, annual mean change might be zero, but with a vastly different climate.
Similarly, were winter temperatures to increase, but summer temperatures to decrease, again, zero annual temperature change, but vastly different climate.
Poles versus tropics, land versus ocean, winter versus summer – these matter.
We are stoopid.
Seems like it was around 2009 that I read an article by Judith that discussed the pitfalls of using a global mean average temperature for measuring climate.
Since that time for me it has become crystal clear that the only utility of a global average mean temperature (which I consider arbitrary) is to support a poorly supported theory, and nothing else.
The global mean temperature informs everyone of nothing. It means nothing to anyone or anyplace at anytime.
How would global average wind speed, rainfall, cloud cover, barometric pressure, humidity, or color inform anyone of anything?
But yet people still think global mean average temperature is a measure of climate change. When scientists look at meaningful climate change, we are not looking at temperatures. We are looking at rainfall since that is a more robust indicator of biological production, which is all that matters with regards to climate change; will the plants today be here tomorrow?
Climate change is an undefined term. To measure what is undefined isn’t science.
I am stoopid – should read:
2. Were winter temperatures to decrease, but summer temperatures to increase, annual mean change might be zero, but with a vastly different climate.
Another aspect which I found troubling is the acknowledged resolution was over 150-200 years due to dating isusues with the proxies, secondly, the oceans by their very nature take considerably longing to reflect changes in atmospheric temperatures, (20+-50+ years by some estimates).
I do recall a tweet from Tierney oct/nov 2021 time from admitting the resolution was as robust as needed. (my apologies for not being able to locate the tweet at this time)
need an edit button – the tweet from Tierney was to the effect that the resolution of the proxies was NOT as robust as should be. Again my apologies for not bookmarking the tweet for later reference and posting.
The paper:
https://eartharxiv.org/repository/view/2219/
There are several active sunspots in the northern solar hemisphere that produce M-class flares. Low activity in the southern hemisphere, indicating a lack of synchronization of southern and northern magnetic field activity.
https://i.ibb.co/prGhRwW/SDO-HMIIF-1024.jpg
What are the best estimates of GMST at the LGM and the difference from Pre-Industrial?
Some estimates and sources are:
GMST at LGM (19-23 ka) and difference from Late Holocene (0-4 ka)
Source: Tierney et al., 2020, ‘Glacial cooling and climate sensitivity revisited’
https://www.nature.com/articles/s41586-020-2617-x
LGM, °C 0-4 ka, °C Δ, °C
GMST 6.1
GSST 3.1
How cold was the ice age? Researchers now know
https://www.sciencedaily.com/releases/2020/08/200826141405.htm
Scientists have figured out just how cold the last Ice Age was. Here’s why it matters
https://www.weforum.org/agenda/2020/09/last-ice-age-global-temperature-scientist-predict
7.78 °C
Pre-industrial – LGM temps (source: IPCC AR5 WG1 Ch 5 Section 5.3.3.1, Table 5.2)
https://www.ipcc.ch/site/assets/uploads/2018/02/WG1AR5_Chapter05_FINAL.pdf
90% Conf. Int. Avg Methods Reference and remarks
4.4 7.2 5.8 Single-EMIC ensemble with microfossil assemblage derived tropical Atlantic SST Schneider von Deimling et al. (2006)
4.6 8.3 6.5 Single-EMIC ensemble with multi-proxy derived tropical SST Holden et al. (2010a)
1.7 3.7 2.7 Single-EMIC ensemble with global multi-proxy data Schmittner et al. 2011
3.9 4.6 4.3 Multi-proxy Shakun et al. (2012); for the interval 17.5–9.5 ka
3.4 4.6 4.0 Multi-AOGCM ensemble with global multi-proxy data Annan and Hargreaves (2013)
3.1 5.9 4.5 Multi-AOGCM ensemble PMIP2 and PMIP3/CMIP5
3.5 5.0 4.3 Average
BEST Land+Ocean anomalies and GMST are published here:
http://berkeleyearth.lbl.gov/auto/Global/
http://berkeleyearth.lbl.gov/auto/Global/Land_and_Ocean_summary.txt
What are the best estimates of GMST at the LGM and the difference from Pre-Industrial?
Some estimates and sources are:
GMST at LGM (19-23 ka) and difference from Late Holocene (0-4 ka)
Source: Tierney et al., 2020, ‘Glacial cooling and climate sensitivity revisited’
https://www.nature.com/articles/s41586-020-2617-x
LGM, °C 0-4 ka, °C Δ, °C
GMST 6.1
GSST 3.1
How cold was the ice age? Researchers now know
https://www.sciencedaily.com/releases/2020/08/200826141405.htm
Scientists have figured out just how cold the last Ice Age was. Here’s why it matters
https://www.weforum.org/agenda/2020/09/last-ice-age-global-temperature-scientist-predict
7.78 °C
Pre-industrial – LGM temps (source: IPCC AR5 WG1 Ch 5 Section 5.3.3.1, Table 5.2)
https://www.ipcc.ch/site/assets/uploads/2018/02/WG1AR5_Chapter05_FINAL.pdf
90% Conf. Int. Avg Methods Reference and remarks
4.4 7.2 5.8 Single-EMIC ensemble with microfossil assemblage derived tropical Atlantic SST Schneider von Deimling et al. (2006)
4.6 8.3 6.5 Single-EMIC ensemble with multi-proxy derived tropical SST Holden et al. (2010a)
1.7 3.7 2.7 Single-EMIC ensemble with global multi-proxy data Schmittner et al. 2011
3.9 4.6 4.3 Multi-proxy Shakun et al. (2012); for the interval 17.5–9.5 ka
3.4 4.6 4.0 Multi-AOGCM ensemble with global multi-proxy data Annan and Hargreaves (2013)
3.1 5.9 4.5 Multi-AOGCM ensemble PMIP2 and PMIP3/CMIP5
3.5 5.0 4.3 Average
BEST Land+Ocean anomalies and GMST are published here:
http://berkeleyearth.lbl.gov/auto/Global/
http://berkeleyearth.lbl.gov/auto/Global/Land_and_Ocean_summary.txt
–In North America and Europe, the most northern parts were covered in ice and were extremely cold. Even here in Arizona, there was big cooling,” Tierney said. “But the biggest cooling was in high latitudes, such as the Arctic, where it was about 14 C (25 F) colder than today.”–
So, Arizona averages about 16 C, and 20,000 years ago it was like say, Idaho which averages about 6.5 C?
Pingback: Osman et al. 2021: a Flawed Nature Paleoclimate Paper? – Watts Up With That?
Pingback: Osman et al. 2021: a Flawed Nature Paleoclimate Paper? |
Pingback: a Flawed Nature Paleoclimate Paper? – Watts Up With That? - Lead Right News
These comments are addressing more than the subject paper and apply to climate science peer review in general.
When I go to the doctor I feel comfortable that the physician meets certain standards for knowledge and expertise. That is in no small measure because of state boards and the rigor of evaluating the prospective doctors. I have no such confidence in the qualifications of the reviewers or level of rigor exhibited in the peer review process. In particular, I’ve wondered about the level of expertise in statistics possessed by the authors and the reviewers.
While my qualms are of a layman with much less sophistication than that possessed by these authors, I share their impressions.
“.We have the impression that the discussion about statistical methodology in the climate sciences is generallynot very deep and that straightforward craftsmanship is pursued in many cases. As a consequence, much ofthe statistical practice in climate science is of a home-grown nature. It is often ad hoc,………..We feel that the cooperation between the statistical and climate sciences does not function as well as-that between, for example, statistical and biomedical science. Among other reasons, this is due to the two particularities mentioned above: confounded dynamics with a large number of degrees of freedom, so that it is not always easy to identify isolated problems amenable to statistical analysis, and the impossibility ofgenerating additional independent data in experiments (apart of numerical experiments with physically basedmodels). Thus, better communication between statisticians and climatologists requires a better understandingby statisticians of the specifics of climate science, and a greater effort by climatologists to communicate the specifics of open problems to statisticians.
The link
https://www.researchgate.net/publication/227550255_On_the_role_of_statistics_in_climate_research
CKid, your comment is at the heart of the issue. Your link is to Hans Von Storch whom I know had a major role in the McIntyre and McKitrick expose’ of the Mann et al (1998) hockey stick data torture.
Nic, thanks for your continued diligence on a matter above my head and apparently also above Osman’s and most other climate science practitioners, (as climateaudit.org exposed for many years).
Nick …I’m curious, have you submitted this to Nature? Do you anticipate a retraction and resubmittal by Osman, as was the case with Resplandy? Thank you for posting your work.
Thanks, Bill, and apologies for the delayed reply – for some reason I had difficulty posting replies.
I sent a copy to Nature and to Osman, but I haven’t had a response from either. I think a correction, of the affected figures at least, is more likely than a retraction. But maybe neither Osman or Nature are fussed about leaving such errors uncorrected.
Nic … sorry for the misspelling of your name.
> But maybe neither Osman or Nature are fussed about leaving such errors uncorrected.
I hope not. They should thank you, actually.
Another fantastic piece of work … and always the gentleman.
Can never give us (at best perhaps) anything more than a clue as to what the sun was doing, 20,000 years ago?
Alternatively, using Leftist logic – If humanity had industrialized to the planet 20,000 years ago, there would be there would have been no life on the planet 100 years later…
Ha, in the published paper he has a version of Mike’s Nature Trick. He splices the modern instrument record onto the end of the computer model reconstruction (the light grey line) and then covers the splice with a big orange dot.
Oops, I was commenting of Thompson 2022 (which Nick talks about at the end) not Osman 2021
Pingback: a Flawed Nature Paleoclimate Paper? – Watts Up With That? - News7g
Pingback: 결함이 있는 자연 고기후 종이? – 그걸로 왓츠업? – Blog Ciencia
The problem today is not that we have too few data, but that we have too much data, which seduces researchers into ransacking it for patterns that are easy to find, likely to be coincidental, and unlikely to be useful.
https://www.bloomberg.com/opinion/articles/2022-04-26/bad-uses-of-big-data-in-academic-studies-shake-belief-in-science
Thanks for the link, jim2. Here is a truly relevant paragraph from it:
“Real researchers don’t correlate random numbers but, all too often, they correlate what are essentially randomly chosen variables. This haphazard search for statistical significance even has a name: data mining. As with random numbers, the correlation between randomly chosen, unrelated variables has a 5% chance of being fortuitously statistically significant. Data mining can be augmented by manipulating, pruning and otherwise torturing the data to get low p-values.”
One day we may actually crack the randomness problem but until then …
“Below we have a special graphic, that shows the Spring temperature impact of a La Nina phase for the United States. We can see that the cold north/warm south pattern extends into the Spring season.”
https://www.severe-weather.eu/long-range-2/la-nina-update-cooling-warm-cold-season-forecast-fa/?fbclid=IwAR0h8tZqh-xJZh0h86PpWXBO2Oyf8_ysZqF8D3Z-RrMOdwwyfrUgUGMO-g8
Thanks Nic for this comprehensive post. The Osman et al (2021) paper has some impact on present estimations of the Climate sensitivity: As more it cooled from the PI to LGM as higher is the sensitivity, in this case ECS. In the end is a cooling of 6.8K as ist’s “estimated” in Osman’s “reanalysis” ( a wild mixture of the proxy-reconstruction and a model) a try to establish a high ECS from the paleo climatologgy. Therefore the rebuttal of this paper is so important because it seems to me it includes the described flaws with some intention.
Thanks, Frank
Amazing this makes news….”We demonstrate that temperature variability across the past 24 thousand years was linked to two primary climatic mechanisms: radiative forcing from ice sheets and greenhouse gases; and a superposition of changes in the ocean overturning circulation and seasonal insolation. ”
But then articles using paleoclimate data (speleothems) and which showed that climate change (over the past 24k) is decoupled from greenhouse gasses goes completely ignored?
Funny, and I thought orbital control drove the increased heating of northern latitudes and changes is circulation.
So this means that greenhouse gasses also control the tilt of the planet.
This new science is soooooo amazing.
\sarcasm off\
Nic, thanks as always for your interesting analyses. I was struck by the following comment:
I’ve noted that oddity in the CMIP6 models, and it has always mystified me. I’ve never been able to come up with even a far-fetched physical reason why that might be true, much less a feasible physical reason.
So I was glad to see your footnote [19] which says:
It’s particularly good to see that the IPCC realizes that the “physics” of ALL the GCMs might be wrong in the same fundamental way in the same locations over an entire simulated period of time …
Keep up the good work, give our best to your good lady,
w.
Willis, thanks for your comment. As per my reply to Ken, I imagine that the model paramterizations follow the treatment in
The Role of Ocean-Atmosphere Interactions in the CO2 Climate Problem (1981) by Ramanathan.
Like you, I was encouraged by the IPCC AR6 treatment of this issue.
Thank you also for the interesting articles that you post on climate blogs, and give my best wishes to your ex-fiancee.
N
1. Earth’s Without-Atmosphere Mean Surface Temperature calculation
Tmean.earth
So = 1.361 W/m² (So is the Solar constant)
S (W/m²) is the planet’s solar flux. For Earth S = So
Earth’s albedo: aearth = 0,306
Earth is a smooth rocky planet, Earth’s surface solar irradiation accepting factor Φearth = 0,47
(Accepted by a Smooth Hemisphere with radius r sunlight is S*Φ*π*r²(1-a), where Φ = 0,47)
β = 150 days*gr*oC/rotation*cal – is a Rotating Planet Surface Solar Irradiation INTERACTING-Emitting Universal Law constant
N = 1 rotation /per day, is Earth’s axial spin
cp.earth = 1 cal/gr*oC, it is because Earth has a vast ocean. Generally speaking almost the whole Earth’s surface is wet. We can call Earth a Planet Ocean.
σ = 5,67*10⁻⁸ W/m²K⁴, the Stefan-Boltzmann constant
Earth’s Without-Atmosphere Mean Surface Temperature Equation Tmean.earth is:
Tmean.earth= [ Φ (1-a) So (β*N*cp)¹∕ ⁴ /4σ ]¹∕ ⁴ (K)
Τmean.earth = [ 0,47(1-0,306)1.361 W/m²(150 days*gr*oC/rotation*cal *1rotations/day*1 cal/gr*oC)¹∕ ⁴ /4*5,67*10⁻⁸ W/m²K⁴ ]¹∕ ⁴ =
Τmean.earth = [ 0,47(1-0,306)1.361 W/m²(150*1*1)¹∕ ⁴ /4*5,67*10⁻⁸ W/m²K⁴ ]¹∕ ⁴ =
Τmean.earth = ( 6.854.905.906,50 )¹∕ ⁴ = 287,74 K
Tmean.earth = 287,74 Κ
And we compare it with the
Tsat.mean.earth = 288 K, measured by satellites.
These two temperatures, the calculated one, and the measured by satellites are almost identical.
Conclusions:
The planet mean surface temperature equation
Tmean = [ Φ (1-a) S (β*N*cp)¹∕ ⁴ /4σ ]¹∕ ⁴ (K)
produces remarkable results.
The calculated planets temperatures are almost identical with the measured by satellites.
Planet…….Tmean….Tsat.mean
Mercury…..325,83 K…..340 K
Earth……….287,74 K…..288 K
Moon………223,35 Κ…..220 Κ
Mars………..213,21 K…..210 K
–
The 288 K – 255 K = 33 oC difference does not exist in the real world.
There are only traces of greenhouse gasses.
The Earth’s atmosphere is very thin. There is not any measurable Greenhouse Gasses Warming effect on the Earth’s surface.
There is NO +33°C greenhouse enhancement on the Earth’s mean surface temperature.
Both the calculated by equation and the satellite measured Earth’s mean surface temperatures are almost identical:
Tmean.earth = 287,74K = 288 K
https://www.cristos-vournas.com
A Simple Theorem, but a very important Theorem.
“… For the Planet Earth without-atmosphere the (N*cp) product is (N*cp = 1) and it is 150 times higher than the necessary condition of (N*cp = 1/150) .
Consequently, Earth’s effective temperature Te the numerical value cannot be equal to Earth’s without-atmosphere mean surface temperature… not even close.”
Link: https://www.cristos-vournas.com/449683314
The T = ( J /σ )¹∕ ⁴ is a mistake !
Stefan-Boltzmann emission law doesn’t work vice-versa !
The old convincement that the Stefan-Boltzmann emission law works vice-versa is based on assumption, that EM energy obeys the 1st Law of Thermodynamics (1LOT). That assumption was never verified, it was never been confirmed by experiment.
Let’s see:
The Stefan-Boltzmann emission law states:
J = σ*Τ⁴ (W/m²) EM energy flux (1)
The mathematical ability to obtain T, for a given J led to the misfortunate believe that the Stefan-Boltzmann emission law formula can be used vise-versa:
T = ( J /σ ) ¹∕ ⁴ (K) (2) as the surface (vise-versa) radiative emission temperature “definition”.
But the
T = ( J /σ ) ¹∕ ⁴ (K) (2) as the irradiated surface (vise-versa) radiative emission temperature “definition”… is utterly unacceptable, because it has not a physical analogue in the real world.
That is why we should consider planet effective temperature
Te = [ (1-a) S /4σ ]¹∕ ⁴ (K)
as a mathematical abstraction, which doesn’t describe the real world processes.
https://www.cristos-vournas.com
As usual Nic Lewis is too in depth for me, my problem not his, and his analysis of this paper where I can understand it, touches on relevant concerns.
His main criticisms seem to be of the sparse data availability, the disjoint between two unexplained versions of the graph, the variation in temperatures arrived at by using different variation factors and the failure to establish consistency of any sort between the two proxy methods used.
Fair enough.
–
My take on it comes from a different angle.
If we had a GM 24000 years ago this implies a rise to another intraglacial before that.
Presumably at the same sort of warming that has occurred since.
The authors write as tough this was all due to” two primary climatic mechanisms: radiative forcing from ice sheets and greenhouse gases”.
Yet when it cooled before from the previous intraglacial which was warmer these ice sheets had to form from what causation, while GHG could have done nothing since CO2 is supposed to have been stable for much longer than that.
The mechanisms for these changes are sadly left unexplained [Milankovich does not cut it].
–
The radiative output of the sun or the type of cloud cover and albedo over these periods is a much more likely causation of temperature change than that postulated.
With 4 -6 C of temperature change ocean out gassing of CO2 should have meant an increase of CO2 and a much lower CO2 than postulated in the past.
Mind you a larger sea volume would imply greater overall uptake of CO2 as a mechanism that might explain relatively stable CO2 with large changes in sea /ocean volumes.
Pingback: Osman et al. 2021: a Flawed Nature Paleoclimate Paper? – TrendRadars
Nic, I have not yet read in detail your recently posted analysis but was glad to see you return to posting an analysis here. Your analyses always provoke thought and promote learning – something of an ideal for the benefits of the internet.
I was struck with this excerpt from your footnote (19) on my first read through of your post:
Thanks, Ken. I imagine that the model parameterization of ocean surface air temperature, derived from SST (actually that of the top ocean model layer, typically 5m or 10m thick, I believe) and ancillary variables, follows the analysis in Ramanathan’s 1981 paper “The Role of Ocean-Atmosphere Interactions in the CO2 Climate Problem”. But, as AR6 says, theoretical physical understanding of this issue is in fact still limited.
Nic, the modeled tos-tas to actual (SST-SAT) divergence problem is I think worthy of a separate future post. I would love to see you summarize the history of what Huang (2015) did to the GISTEMP through Karl (2015) and the probable biases it introduced.
Steve McIntrye did a great post on this topic in 2017 here. A lot of us commented there.
And, in another post perhaps we could look at why the models project
-Wikipedia
Wikipedia link
Which section of AR6 discusses the differences in SST and SAT warming?
If F_2x were 3.6 W/m2, ECS were 3.6 K/doubling, and there were no SWR feedback, then the planet would be radiating an additional 1 W/m2 per degK of surface warming (a feedback parameter of 1 W/m2/K). If ECS were 1.8 K/doubling, it would be 2 W/m2/K.
However, saturation water vapor pressure just above the surface of the ocean increases at 5.6 W/m2/K (0.7%/K * 80 W/m2 of latent heat) and some sensible heat is transported at the same time. It is impossible for at least 5.6 W/m2/K of latent heat (and more sensible heat) to leave the surface of planet with warming and only 1-2.5 W/m2/K escape to space.
Since a 1%/K change in albedo is only 1 W/m2/K in the SWR channel, a large reduction in SWR reaching the surface with warming (SWR feedback) is unlikely to compensate for this large increase in latent heat flux.
One explanation is that vertical transport of latent (and sensible) heat must slow with surface warming. This is normally described as a slowing of the overturning of the troposphere. Much of this slowing may occur at the interface between the top ocean grid cell (TOS) and the lowest atmospheric grid cell (TAS), but climate models lack to resolution to properly represent this process.
A very thin layer of air immediately above the ocean is saturated with water vapor, so the rate limiting step in “evaporation” is actually transport from this thin layer to the bulk air above. (The same is true for sensible heat transport.) The rate of evaporation is proportional to wind speed and the undersaturation (1-RH) of the bulk air above the ocean. Since the relative humidity of the air immediately above the ocean is typically 80%, a 1% increase in relative humidity is a 5% reduction in undersaturation. A reduction in the surface winds that are part of the Hadley circulation would be expected to be part of a reduction in “atmospheric overturning” that is associated with global warming.
Therefore, in my amateur opinion, the TOS-TAS temperature difference should be expected to shrink with warming and grow with cooling. However, AOGCMs don’t have the resolution to properly represent this process.
Best wishes, Frank
Frank, “Which section of AR6 discusses the differences in SST and SAT warming?” I mentioned the source where you can find the sentence which is most imortant in this case in my comment below ( April 29th).
Frank, I have had an interest in the differences between tos and tas trends in the climate models and observed -if any- and thus I was attempting on first reading of your conjecture to understand where the heat you talk about was generated in excess and by the amounts that I think you are indicating. The heat of condensation and evaporation are identical but of opposite sign. It appears to me that you are describing a process for changing the amount of heat going into the ocean with a warming world – but I do not follow.
Kenneth, in the model world there is a parameter “tos”, it’s equivalent to SST in the real world. This “tos” has a lower slope over the time with ongoing warming then “tas” over the oceans, which is the equivalent to T2m. BUT: This is a model bias: ” Because of the strong interactions between the ocean and the atmosphere, the sea surface temperature (SST) is very close to the temperature of the air above it.” (see a classic textbook for climate) The AR6 concluded: “The GSAT metric is 2 m air temperature over all surfaces and is the diagnostic generally used from climate models. Changes in GMST and GSAT over time differ by at most 10% in either direction (high confidence),…Therefore, long-term changes in GMST/GSAT are presently assessed to be identical…“ Page TS 27 WG1. Therefore all “blended” ( tas for land and tos for the oceans ) approaches to estimate the warming trends from GCM are not physically.
frankclimate, it is good that you have given a more detailed explanation of the difference between tos and tas for climate models and what is observed. I was aware of this difference in models from work I have done and reading some papers published on the subject and some that tended to use the model difference to correct for SST for observations. The observed data at the time I was looking was sparse and made comparing observed and with model differences difficult to make with any certainty. Since then I have seen Nic Lewis make some better comparisons.
What struck me was the IPCC AR6 stand on the matter as there being no difference. I was expecting that given historical stands on matters such as these the IPCC might have pointed to the model difference and indicated that the observed data was inconclusive. As I recall, the AR5 models did show a range of differences between tos and tas and some showing no difference.
Ken: Thanks for you interest in my comment. My analysis was not initially directed towards understanding the tas/tos difference, but when it is complete, it can be applied to this problem. In short, my argument is:
1) The temperature difference in the top grid cell in the ocean (tos) and the lowest grid cell in the atmosphere (tas) is the intial step in transporting heat absorbed by the ocean to space. With their huge grid cells, climate models can’t properly represent the physics at the interface between ocean and atmosphere. However, they need to parameterize it correctly.
2) It is well known that the rate of overturning of the troposphere is expected to slow upon global warming. For example, climate models predict an increase of rainfall of about 2% per degK of warming, not the 7%/K one would expect if water vapor transport from the saturated layer adhering to the surface of the ocean remained constant.
3) The rate of vertical flux of heat in the troposphere depends on the temperature gradient driving that flux. This is obviously true for radiation and conduction of sensible heat. For the portion of the troposphere near the moist adiabatic lapse rate on the average, this should also to be true. More below.
4) Therefore, I predict that the tos/tas difference will shrink.
——————————————————-
Why does turnover of the atmosphere slow (rather than speed up as intuitive suggests) in response to global warming.
For a high ECS of 3.6 K/doubling and an F_2x of 3.6 W/m2, the climate feedback parameter must be -1 W/m2/K. For low ECS of 1.8 K/doubling, -2 W/m2/K. Most likely the net increase of radiation across the TOA at equilibrium after forced global warming is near -1 to -2 W/m2/K.
For a steady state to exist, the increase in net heat flux across the surface (between tos and tas) must also be near -1 to -2 W/m2/K.
Using MODTRAN, you can calculate that the net LWR flux (OLR-DLR) at constant relative humidity doesn’t change appreciably. Most DLR photons are emitted from near the surface where surface warming and atmospheric warming are similar.
SWR feedback at the surface is probably small, between -1 W/m2/K and 1 W/m2/K. This is because albedo is only 100 W/m2, and a +/-1%/K change in albedo is a -/+1 W/m2/K change at the surface. Most clouds (except MBLs) form where air is rising. Rising air must come down somewhere else (where there will be no clouds.) Most climate models predict SWR feedback of +0.5 W/m2/K and only some of that SWR is absorbed by the surface. So the net flux of SWR from the atmosphere to the surface probably doesn’t change much either.
If the transport of water vapor from the ocean to the atmosphere increased with saturation vapor pressure (7%/K), the flux of heat would increase at -5.6 W/m2/K. Most would come from the ocean. We can’t have an additional 5.6 W/m2/K of latent heat leaving the surface during global warming and only an additional 1-2 W/m2/K escaping to space.
This is why turnover of the atmosphere must slow. The mechanism of slowing involves slower vertical convective transport of water vapor out of the boundary layer, which increases relative humidity in the boundary layer and near the surface of the ocean, and a slowing of wind speed, which slows turbulent vertical transport of water vapor in the boundary layer. The rate of evaporation is proportional to wind speed and undersaturation of the air just above the ocean. Wind speed and vertical transport are linked in the Hadley circulation.
Slowing overturning is likely associated with a narrowing of the tos/tas difference.
frankclimate, thanks for the detailed explanation on the tos and tas model difference versus the observed no difference. I am aware of this difference from my own analysis and read papers that tended to accept it as a difference for the observed. What I was struck about was the IPCC current stand that there is no difference in the observed tos and tas. Historically I might have expected the IPCC to point to the model difference and then noting that the observed data on the matter being inconclusive.
As I recall the AR5 models had a range of differences in tos and tas temperature trends and with some showing no difference.
Frank: Nice to hear from you and thank you for the reference. Did my comment above make any sense to you? I’ve tried to make this point several times before, without drawing comments from anyone. For heat to flow from the surface of the ocean to the upper troposphere, tos needs to be higher than tas (one the average).
This difference is maintained by convection of heat into the upper troposphere. So, if atmospheric overturning slows, the tos – tas gap should shrink. All of this seems to be due to the fundamental basics of heat transfer. It is certainly correct for heat transfer by conduction and by radiation. Heat transfer by convection, of course, depends on turbulence (wind speed). Both latent and sensible heat are co-transported by bulk motion, and it is clear that latent heat (evaporation) can’t increase at 7%/K.
Off to AR6.
Hi Frank, also glad to hear from you again! I tried it with the observations of Tair over the ocean, I selected parts of the IPWP (30S…30N; 120E …180E) , I used ICOADS v2.5 Tair and compared the trends 2000…2020 with SST ( NCEP OI v2 ) and both datasets got the same slopes: 0.0117 K/a. This said to me: the difference in the warming is near zero. Perhaps it gets smaller with ongoing warming :-) (With a little help of the KNMI Climate explorer)
best Frank
I’m putting this here because it is another breathless extinction paper published by “Science.” It is pay walled, so can’t see detail. Perhaps someone can get a copy.
Global warming threatens marine biota with losses of unknown severity. Here, we quantify global and local extinction risks in the ocean across a range of climate futures on the basis of the ecophysiological limits of diverse animal species and calibration against the fossil record. With accelerating greenhouse gas emissions, species losses from warming and oxygen depletion alone become comparable to current direct human impacts within a century and culminate in a mass extinction rivaling those in Earth’s past. Polar species are at highest risk of extinction, but local biological richness declines more in the tropics. Reversing greenhouse gas emissions trends would diminish extinction risks by more than 70%, preserving marine biodiversity accumulated over the past ~50 million years of evolutionary history.
https://www.science.org/doi/10.1126/science.abe9039
jim2, From the open introduction: “They found that under business-as-usual global temperature increases…”
From the open SI: 1) a high-emissions scenario,in which CO2 emissions accelerate throughout much of the 21st 31 century and reach a nominal
radiative forcing of 8.5 W/m2 in 2100 CE…”
As usual I stop reading a paper when I find the words “b.a.u. and RCP 8.5” in the text. I did so. ;-)
I notice that Osman et al did not cite a 2014 Nature article by Lachniet, “Orbital control of western North America atmospheric circulation and climate over two glacial cycles.” This is most accurate reconstruction I know of for western North America. Why would it not be referenced, other than it does not fit the political religious narrative? Why use models when you have reality?
‘This repeating cycle of 100,000-year glaciations and 10,000 to 20,000 year interglacials has been fairly consistent over the past 2.6 million years. The planet has trundled through the entire cycle dozens of times. If the pattern holds, we are due for another major glaciation sometime in the next several thousand years…’ ~Mario Loyola
We have discovered the Planet Surface Rotational Warming Phenomenon.
The Planet Surface Rotational Warming Phenomenon states: Planets’ mean surface temperatures relate (everything else equals) as their (N*cp) products’ sixteenth root.
The discovery has explained the origin of the formerly observed the planets’ average surface temperatures comparison discrepancies.
Earth is warmer than Moon because Earth rotates faster than Moon and because Earth’s surface is covered with water.
What we do in our research is to compare the satellite measured planetary temperatures. We call it “The Planets’ Surface Temperatures Comparison Method”.
A faster rotating planet accumulates much more solar energy, than a slower rotating one.
https://www.cristos-vournas.com
The Earth’s surface is warmer than the Moon’s because the variance between day and night temperatures (the diurnal temperature range) is much greater on the Moon. Thus the Moon re-emits the daytime energy much more efficiently due to much more powerful blackbody spectrum of extremely high daytime surface temperature.
The faster the rotation the more evenly the surface is heated and the less efficiently it re-emits energy.
On Earth the CO2 is dissolved in oceanic waters. That is why there is not a runaway greenhouse effect on Earth’s surface.
The closer to sun Venus couldn’t have liquid water. The water vapor vanished in space, since it is a lighter gas.
As a result on Venus the entire CO2 is in atmosphere – thus the very strong runaway greenhouse effect.
https://www.cristos-vournas.com
“On Earth the CO2 is dissolved in oceanic waters. That is why there is not a runaway greenhouse effect on Earth’s surface.”
I like big picture thinking but this is funny logic, because in the next sentence you say: “The closer to sun Venus couldn’t have liquid water. The water vapor vanished in space, since it is a lighter gas.”
Well, the logic could just as easily be trimmed by Occam’s razor to say: “The closer to sun Venus is hotter than Earth.”
It seems many times climate science produces paper after paper that use the greenhouse effect as one of the assumptions in the logic that concludes there is in fact a greenhouse effect. I don’t doubt there are plenty that we want to explain Mars, Earth and Venus through the lens of CO2.
In The Inconvenient Truth Al Gore used ice cores scientist Lonnie Thompson as the movie’s science consultant to prove that the glacial cycle was controlled by CO2. They overlaid the matching charts to prove it. However, by the movies release scientists had known for years that atmospheric CO2 concentration lagged temperature by hundreds of years, showing that temperature controlled CO2 (due to ocean absorption), not the other way around.
Gaseous planets have similar atmospheric gases content.
I have the gaseous planets at 1 bar level the satellite measured temperatures comparison in relation to the gaseous planets’ rotational spins.
Gaseous planets (Jupiter, Saturn, Uranus, Neptune) have similar atmospheric gases content. The more close the content is the better the satellite measured temperatures relate in accordance to the Rotational Warming Phenomenon.
Link:
https://www.cristos-vournas.com/445559910
Judy,
Nic has a serious post, with a lot of work in it. It’s not impenetrable, easier to read than the typical McIntyre stuff, for sure. But still hard to engage.
If you look at the comments, very few people have been able to engage in the content. Part of this is because the stuff is hard. But the other part is the presentation. Yes, even with decent writing (word choice, sentence length, para length).
1. I wish that he would tell us the major findings up front. It is very tedious to read through this length of an article, not knowing where it is headed and what to watch out for. I finally jumped to the conclusions, something I do often with science papers. But not something I should have to do with a blog post, with an explanatory article. [Tell em what you’re gonna tell em. Tell em. Tell em what you told em.]
2. In addition, tell us the structure up front (the major divisions of the blog post, article). Note that (1) and (2) are basic writing advice.
3. The comments about the previous retraction are unfortunate. I don’t think this was malicious. But I was also confused if he was trying to assert the same level of problems with this piece as earlier or just citing what results he’d achieved previously. An up front caveat (in text) would have been appropriate. Not even for concern to the reviewed paper, but just for reader clarity. (Less mystery, more clarity.)
4. The paragraph lengths are good. And sentences are not too long. But I still think stronger structure within the paras would help. Less of a narrative and more of a logical explanation. This applies to a certain extent to the overall paper as well. (Minto Pyramid Principle or just basic writing advice.)
5. On the content, I have to admit to not reading it all. (I suspect very few commenters did either, but I, at least, admit it.) The stuff is hard…which might have stopped me, even with a better presentation. But the presentation also slowed me down. And I say this as someone who would enjoy having learned something, to reading an easy to understand review of the paper, even to following the mild drama of an amateur analyzing an expert and finding issues.
6. I guess the one semisubstantive comment I have is about the discussion of the change in preprint to published proxy reconstruction. I don’t think it’s that unusual to see changes/improvements/evolution from preprint to published. And I doubt most authors or even critics would expect an explanation of the evolution. After all, the published paper sort of stands on its own. Also, I’m not clear that “being closer to the previous work” is the best criteria for judging if the preprint graph or published graph is most valid. This isn’t to say it’s not interesting to look at preprint versus published. Or even that the preprint isn’t more correct. Just that it’s rather tenuous to make any strong assertion here.
Good points. Optimal blog level somewhere is between Paul Dirac and Kamala.
Disagreement here.
My view is that the Steve McIntyre articles typically on Climate Audit blog are a rather easy read for people who are used to reading scientific material. I’d personally rank them among the easiest to read and I would not like to see people deterred from reading them by anonymous reporting “easier to read than the typical McIntyre stuff, for sure” There is a good deal of wit and veiled criticism in Steve’s writing at CA. Geoff S
Pingback: AWED NEWLETTER: We cover COVID to Climate, as well as Energy to Elections. - Dr. Rich Swier
Pingback: The Media Balance Newsletter: 9/5/22. - Australian Climate Sceptics blog
Pingback: Energy and Environmental Review: May 9, 2022 - Master Resource
The Rotating Planet Surface Solar Irradiation Interacting-Emitting Universal Law is based on a simple thought.
It is based on the thought, that physical phenomenon which distracts the “black body” surfaces from the instant emitting the absorbed solar radiative energy back to space, warms the “black body” surfaces up.
In our case those distracting physical phenomena are the planet’s sidereal rotation, N rotations/day, and the planet’s surface specific heat, cp cal/gr oC.
What we have discovered is the ROTATING PLANET SURFACE SOLAR IRRADIATION INTERACTING-EMITTING UNIVERSAL LAW:
Jemit = 4πr²σΤmean⁴ /(β*N*cp)¹∕ ⁴ (W)
Planet Energy Budget:
Jnot.reflected = Jemit
πr²Φ*S*(1-a) = 4πr²σTmean⁴ /(β*N*cp)¹∕ ⁴ (W)
Solving for Tmean we obtain the PLANET MEAN SURFACE TEMPERATURE EQUATION:
Tmean.planet = [ Φ (1-a) S (β*N*cp)¹∕ ⁴ /4σ ]¹∕ ⁴ (K)
https://www.cristos-vournas.com
Documents translation services http://www.translationservicesworld.com/japanese_translation/ here.
Pingback: Osman et al. 2021: a flawed Nature paleoclimate paper? - Trueorleans News
This article concerns the paper “Globally resolved surface temperatures since the Last Glacial Maximum” by Matthew Osman et al.[2] (hereafter Osman 2021) published by Nature in November 2021.