Remark by Cowtan & Jacobs on Lewis & Curry 2018 and Reply: Half 1

From Dr. Judith Curry’s Local weather And so on.

Posted on December 16, 2019 by niclewis | 52 Feedback

By Nic Lewis

A touch upon LC18 (current paper by Lewis and Curry on local weather sensitivity)  by Cowtan and Jacobs has been revealed, together with our response.

Introduction

In an earlier article right here I mentioned the Lewis and Curry (2018) paper “The influence of current forcing and ocean warmth uptake knowledge on estimates of local weather sensitivity” (LC18) and set out its outcomes.

The LC18 evaluation used a world power funds mannequin to estimate the planetary equilibrium local weather sensitivity (ECS) and transient local weather response (TCR). ECS and TCR are estimated from modifications (Δ) in international imply floor temperature [T], efficient radiative forcing (ERF) [F] and the planetary radiative imbalance[1] [N] between a base and a last interval, as:

ECS = F2×CO2  × ΔT /(ΔF – ΔN)  and  TCR = F2×CO2 × ΔT /ΔF

the place F2×CO2 is the ERF for a doubling of atmospheric CO2 focus.

The primary LC18 estimates for ECS and TCR had been as per Desk 1. The details of word are that they lie close to the underside finish of the IPCC AR5 ‘doubtless’ ranges for ECS and TCR, and that they’re each much less unsure and barely decrease than these given within the predecessor examine, Lewis & Curry (2015) when utilizing HadCRUT4 international floor temperature knowledge. The LC18 greatest estimates primarily based on the quicker warming infilled Cowtan & Method Had4_krig_v2 temperature dataset are similar to the HadCRUT4-based leads to Lewis & Curry (2015).

Desk 1 (primarily based on Desk three in LC18) Finest estimates (medians) and uncertainty ranges for ECS and TCR utilizing the bottom and last intervals indicated. Values in roman sort compute the temperature change concerned (ΔT) utilizing the HadCRUT4v5 dataset; values in italics compute utilizing the infilled, globally-complete Had4_krig_v2 (Cowtan & Method) dataset. The popular estimates are proven in daring. Ranges are said to the closest zero.05 Okay. Additionally proven are the comparable outcomes (utilizing the HadCRUT4v2 dataset) from LC15 for the primary two interval combos given in that paper. ECS estimates assume that efficient local weather sensitivity doesn’t change with time elapsed since imposition of forcing.

Abstract of the Remark and Reply

A Touch upon LC18 by Kevin Cowtan and Peter Jacobs, and a Reply from myself and Judith Curry, have simply been revealed by Journal of Local weather. A replica of the Reply is out there right here.

The Remark (known as CJ20, because it  seems within the 1 January 2020 problem) is arguably extra a critique of observational sea floor temperature (SST) datasets than of the strategies and outcomes of LC18. Its summary reads as follows:

Lewis and Curry (2018) (hereafter LC18) current a way for the estimating the transient local weather response (TCR) of the local weather system from the temperature change between two time home windows – an early baseline interval within the 19th century, and a contemporary interval primarily within the 21st century. The outcomes recommend a decrease worth of TCR than estimates from local weather mannequin simulations. Earlier research have recognized uncertainty within the historic forcings, the influence of the time evolution of the forcing on temperature response, and observational points as contributory components to this disagreement. We examine an additional issue: uncertainty within the bias corrections utilized to historic sea floor temperature knowledge. This uncertainty can significantly influence the estimation of variables on decadal timescales, and due to this fact influence the estimation of TCR utilizing the window technique in addition to estimates of inner variability. We display that use of the entire historic file can mitigate the impacts of working with quick time home windows to some extent, significantly with respect to the early a part of the file.

Initially, CJ20 asserted that the bottom and last intervals – what they name early and late home windows – chosen in LC18 – which had been matched as regards volcanic forcing and affect from multidecadal inner variability – led to decrease values of TCR (CJ20 didn’t handle the LC18 ECS estimates). They subsequently eliminated that declare, which the evaluation in our submitted Reply disproved. The ultimate model of CJ20 focuses on the potential influence of utilizing home windows fairly than all of the historic knowledge, particularly the influence – primarily based on evaluating warming in CMIP5 (present technology) local weather fashions and in observations – of the selection of various dates for the home windows, and on uncertainty in bias corrections to historic SST knowledge. CJ20 give attention to use of the HadCRUT4 temperature file, however – as LC18 made clear –  it’s acceptable to make use of a globally full file for comparability with local weather mannequin outcomes. We accordingly used solely Kevin Cowtans’s infilled model of HadCRUT4, Had4_krig_v2, in our Reply.

The summary for my and Judith Curry’s Reply to CJ20 reads as follows:

Cowtan and Jacobs assert that the strategy utilized by Lewis and Curry  in 2018 (LC18) to estimate the local weather system’s transient local weather response (TCR) from modifications between two time home windows is much less strong – particularly towards sea floor temperature bias correction uncertainty – than a way that makes use of all the historic file. We display that TCR estimated utilizing all knowledge from the temperature file is intently consistent with that estimated utilizing the LC18 home windows, as is the median TCR estimate utilizing all pairs of particular person years. We additionally present that the median TCR estimate from all pairs of decade-plus size home windows is intently consistent with that estimated utilizing the LC18 home windows, and that incorporating window choice uncertainty would make little distinction to complete uncertainty in TCR estimation. We discover that when variations within the evolution of forcing are accounted for, the connection over time between warming in CMIP5 fashions and observations is according to the connection between CMIP5 TCR and LC18’s TCR estimate, however fluctuates resulting from multidecadal inner variability and volcanism. We additionally present that numerous different issues raised by Cowtan and Jacobs have negligible implications for TCR estimation in LC18.

In a nutshell, we refuted all factors of substance made in CJ20. I plan to cope with the variations between noticed and CMIP5 model-simulated historic warming, which shaped the premise of CJ20’s numerical evaluation, in a subsequent article. On this article, I’ll elaborate on our refutation of factors within the the rest of CJ20.

Window choice associated uncertainty

Concerning the declare by CJ20 regarding uncertainty induced by window selection, that is what we needed to say within the Reply, having examined the results of random choice of home windows from a decade upwards in size,[2] all of which led to median TCR estimates very near LC18’s 1.33 °C [= 1.33 K]:

For estimates with the best (2.zero Wm−2) minimal forcing improve, that are most related to LC18’s TCR estimate, the 5–95% TCR uncertainty vary arising from random window choice is 1.08–1.54 Okay, or 1.20–1.59 Okay utilizing zero.55-scaled volcanic forcing. The width of those ranges – zero.103 and zero.073, respectively, in fractional customary deviation phrases[3] –  displays the truth that lots of the window combos contain mismatched influences from inner variability and/or volcanism. These window choice uncertainty ranges don’t indicate that LC18 underestimated uncertainty in international temperature change: the 1σ fractional uncertainty in LC18’s most popular TCR estimate attributable to temperature change uncertainty (together with that from inner variability) alone was zero.103.[4] Furthermore, even when no allowance is made for double counting of temperature change uncertainty, estimated general TCR uncertainty would improve little if window choice uncertainty had been added. Including (in quadrature) the zero.103 or zero.073 1σ fractional uncertainty in TCR from window choice to the 1σ fractional uncertainty of the popular LC18 TCR estimate, would solely improve it to 1.13⤬ its unique degree, or to 1.07⤬ that degree if utilizing zero.55-scaled volcanic forcing.[5]

This exhibits that uncertainty in TCR estimation arising from window choice is minor even when no allowance is made for double counting of temperature uncertainty, and negligible if allowance is made for such doubling counting.

Utilizing knowledge from all the historic file

CJ20 suggest use of knowledge from all the historic file. In truth, LC18 examined doing so, by the standard regression technique, however discovered mismatching volcanic affect made estimation delicate to the scaling issue used for volcanic forcing. With out cutting down volcanic forcing the TCR estimate from regression over the entire historic interval is much decrease than that from utilizing the home windows technique. That is what we mentioned within the Reply:

When AR5 volcanic forcing is scaled by zero.55, regression of median annual-mean temperature on forcing over 1850–2016 provides a 1.27 Okay Had4_krig_v2-based TCR estimate, marginally decrease than LC18’s 1.33 Okay two-window primarily based most popular estimate. Regressing pentadal means (over 1852–2016) considerably improves the match (to an R2 of zero.92) and offers a TCR estimate of 1.33 Okay. Utilizing such pentadal-mean regression on every of the 500,000 pairs of samples of temperature and forcing time collection provides a 5–95% TCR vary of zero.91–1.84 Okay, marginally decrease and narrower than the LC18 most popular estimate vary.

So, the outcomes of TCR estimation utilizing knowledge from all the historic file is intently consistent with these utilizing LC18’s window technique and chosen home windows, supplied the volcanic forcing is scaled down as per LC18’s advice. Nonetheless, the uncertainty induced by having to estimate the suitable volcanic forcing scaling issue arguably makes utilizing knowledge from the total historic file a much less passable method than utilizing the home windows technique.

Points with historic sea floor temperature knowledge

There may be certainly vital uncertainty as to the accuracy of the worldwide SST file. Nonetheless, CJ20 didn’t present that the LC18 TCR estimates had been materially affected by any recognized errors in SST bias corrections. Nor did they present that uncertainty within the SST file was better than that estimated by the suppliers of the datasets utilized in LC18.

CJ20 make the purpose that protection of the ‘water hemisphere’ was nearly non-existent within the 1860s. Nonetheless, the 1869–82 major early window utilized in LC18 avoids the 1860s (save for 1869, when protection was higher), and offers barely larger protection within the (land-sparse) southern hemisphere than within the northern hemisphere.

CJ20 additionally state that nineteenth century temperatures are depending on giant ‘bucket corrections’ to sea floor temperature (SST) observations, nonetheless CJ20 themselves recommend that the change from wood buckets to poorly insulated canvas buckets requiring a big bias correction occurred primarily throughout 1890–1910. Bucket corrections had been comparatively small throughout 1869–82, the LC18 early window.

Potential misestimation of forcings

That is what we wrote within the Reply regarding two forcing estimation points raised in CJ20:

CJ20 declare that earlier research have recognized variations in inferred forcings and within the temperature influence of historic versus transient forcing modifications as potential explanatory components for current observational energy-budget TCR estimates being decrease than common local weather mannequin TCR values. Not one of the three supporting research that they cite helps both rivalry.

and

CJ20 declare that comparability of modeled and noticed temperatures for late home windows beginning after 2005 is affected by overestimation of forcings in fashions. Since LC18 didn’t make any comparisons of modeled and noticed temperatures over the historic interval, the one problem of relevance to LC18 is whether or not it misestimated current forcing. Not one of the three supporting research that CJ20 cite point out that LC18 misestimated current forcing.

In truth, a extra complete examine[6] discovered, of their CMIP5-specification historic simulations, that because the mid-2000s underestimation of modifications in different forcing brokers greater than counteracted overestimation of modifications in photo voltaic and volcanic forcing. Furthermore, not one of the research cited in CJ20 addressed the actual drawback, of bias in CMIP5 mannequin forcing that already existed a number of many years in the past (resulting from principally to extreme aerosol forcing); none of their analyses began earlier than 1980.

Ocean and air floor temperature in fashions and observations

In CMIP5 fashions near-surface marine air temperature warms greater than the ocean floor temperature area (‘tos‘). CJ20 state that “Lewis and Curry argue that this area [tos] isn’t the highest layer of the majority ocean floor temperature” (to which measured SST broadly corresponds). Nonetheless, this straw man argument, which CJ20 disprove, was by no means made in LC18. Because the reply states:

CJ20’s declare that LC18 “argue that this area [tos] isn’t the highest layer of the majority ocean floor temperature” is wrong. Quite, LC18 argued that the tas/tos warming distinction displays the model-simulated warming distinction between tas and ocean pores and skin temperature, which can heat in another way from SST.

There are theoretical causes for anticipating air simply above the ocean floor to heat barely quicker than the ocean pores and skin temperature. Nonetheless, the extent of the distinction is determined by many components and is unsure, as is the distinction between the warming charges of SST and of ocean pores and skin temperature. LC18 due to this fact centered on observational fairly than CMIP5 mannequin proof on this space. We are saying within the Reply:

LC18 (part 7e) concluded from observational and reanalysis proof that in the actual local weather system, tas warmed at most a number of per cent greater than a mix of tas and tos (mannequin high ocean layer temperature), a considerably smaller distinction than that claimed by CJ20. Certainly, the 1979-onwards ERA-interim reanalysis globally-complete floor air temperature file, adjusted for inhomogeneities of their SST supply (Simmons et al. 2017), exhibits barely decrease warming over 1979–2016 than does Had4_krig_v2.

Additionally it is value noting that in CMIP5 fashions tas, not like tos, is a diagnostic fairly than a prognostic variable – it’s a parameterised extraneous variable, not a variable that includes within the primary mannequin physics.

Conclusion

Not one of the criticisms of LC18 within the Reply stand as much as examination. I depart examination of variations between noticed and CMIP5 model-simulated historic warming, which shaped the premise of CJ20’s numerical evaluation, to a subsequent article. Suffice to say right here that such variations, when correctly analysed within the gentle of variations in forcing evolution, are absolutely according to the LC18 TCR estimate.

Nicholas Lewis    December 2019

[1] N is estimated from its counterpart, the speed of local weather system warmth uptake, which is principally by the ocean.

[2] Since small inter-window forcing will increase present poor TCR estimation, minimal required inter-window forcing will increase, starting from 1.zero to 2.zero Wm−2, had been imposed. (The better the forcing improve the decrease the relative uncertainty, as regards each forcing and the change in temperature that it causes. The home windows used for LC18’s important ECS and TCR estimates gave a forcing improve of two.52 Wm−2.) There have been over 11,000 decade plus lengthy window combos giving a forcing improve of two.zero Wm−2 or extra. For computational tractability, early and late home windows had been specified to be of equal size. When utilizing LC18’s instructed zero.55 scaling of volcanic forcing the median TCR estimates had been even nearer to 1.33 Okay in any respect ranges of required forcing improve, and had decrease uncertainty ranges, than when utilizing unscaled volcanic forcing.

[3] In order to have the opportunity readily to mix uncertainties, we work with 1 customary deviation fractional uncertainties, right here derived by scaling from 17-83% ranges and medians in Desk 1

[4] Scaling from the 5-95% vary and median for Had4_krig_v2 ΔT in Desk 2 of LC18. If temperature uncertainty alone is included, the fractional uncertainty in TCR equals that in ΔT.

[5] Scaling from the 17-83% vary in Desk three of LC18, giving a fractional customary deviation of zero.193 for the popular LC18 TCR estimate. Uncertainties are taken to be usually distributed and impartial for the needs of deriving their customary deviations and mixing them. Including in quadrature a fractional customary deviation of zero.103 (zero.073) to the unique degree of zero.193 will increase it to zero.219 (zero.207).

[6] Outten, S., Thorne, P., Bethke, I. and Seland, Ø., 2015. Investigating the current obvious hiatus in floor temperature will increase: 1. Development of two 30‐member Earth System Mannequin ensembles. Journal of Geophysical Analysis: Atmospheres, 120(17), pp.8575-8596.

Like this:

Like Loading…