Why models run hot: results from an irreducibly simple climate model
Authors
Abstract
An irreducibly simple climate-sensitivity model is designed to empower even non-specialists to research the question how much global warming we may cause. In 1990, the First Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) expressed “substantial confidence” that near-term global warming would occur twice as fast as subsequent observation. Given rising CO_{2} concentration, few models predicted no warming since 2001. Between the pre-final and published drafts of the Fifth Assessment Report, IPCC cut its near-term warming projection substantially, substituting “expert assessment” for models’ near-term predictions. Yet its long-range predictions remain unaltered. The model indicates that IPCC’s reduction of the feedback sum from 1.9 to 1.5 W m^{−2} K^{−1} mandates a reduction from 3.2 to 2.2 K in its central climate-sensitivity estimate; that, since feedbacks are likely to be net-negative, a better estimate is 1.0 K; that there is no unrealized global warming in the pipeline; that global warming this century will be <1 K; and that combustion of all recoverable fossil fuels will cause <2.2 K global warming to equilibrium. Resolving the discrepancies between the methodology adopted by IPCC in its Fourth and Fifth Assessment Reports that are highlighted in the present paper is vital. Once those discrepancies are taken into account, the impact of anthropogenic global warming over the next century, and even as far as equilibrium many millennia hence, may be no more than one-third to one-half of IPCC’s current projections.
Keywords
Climate changeClimate sensitivityClimate modelsGlobal warmingTemperature feedbacksDynamical systems1 Introduction
The present paper describes an irreducibly simple but robustly calibrated climate-sensitivity model that fairly represents the key determinants of climate sensitivity, flexibly encompasses all reasonably foreseeable outcomes, and reliably determines how much global warming we may cause both in the short term and in the long term. The model investigates and identifies possible reasons for the widening discrepancy between prediction and observation.
Simplification need not lead to error. It can expose anomalies in more complex models that have caused them to run hot. The simple climate model outlined here is not intended as a substitute for the general-circulation models. Its purpose is to investigate discrepancies between IPCC’s Fourth (AR4) and Fifth (AR5) Assessment Reports and to reach a clearer understanding of how the general-circulation models arrive at their predictions, and, in particular, of how the balance between forcings and feedbacks affects climate-sensitivity estimates. Is the mainstream science settled? Or is there more debate [8] than Professor Garnaut suggests? The simple model provides a benchmark against which to measure the soundness of the more complex models’ predictions.
2 Empirical evidence of models running hot
How reliable are the general-circulation models the authority of whose output Professor Garnaut invites us to accept without question? In 1990, FAR predicted with “substantial confidence” that, in the 35 years 1991–2025, global temperature would rise by 1.0 [0.7, 1.5] K, equivalent to 2.8 [1.9, 4.2] K century^{−1}. Yet 25 years after that prediction the outturn, expressed as the trend on the mean of the two satellite monthly global mean surface temperature anomaly datasets [9, 10], is 0.34 °C, equivalent to 1.4 °C century^{−1}—half the central estimate in FAR and beneath the lower bound of the then-projected warming interval (Fig. 1). Global temperature would have to rise over the coming decade at a rate almost twice as high as the greatest supra-decadal rate observed since the global instrumental record began in 1850 to attain even the lower bound of the predictions in FAR, and would have to rise at more than thrice the previous record rate—i.e., at 0.67 K over the decade—to correspond with the central prediction.
Empirically based reports of validation failure in complex general-circulation models abound in the journals [14–29]. Most recently, Zhang et al. [30] reported that some 93.4 % of altocumulus clouds observed by collocated CALIPSO and CloudSat satellites cannot be resolved by climate models with a grid resolution >1° (110 km). Studies of paleo-vegetation and pollens in China during the mid-Holocene climate optimum 6,000 years ago find January (i.e., winter minimum) temperatures to have been 6–8 K warmer than present. Yet, Jiang et al. [31] showed that all 36 models in the Paleoclimate Modeling Intercomparison Project backcast winter temperatures for the mid-Holocene cooler than the present. Also, all but one model incorrectly simulated annual-mean mid-Holocene temperatures in China as cooler than the present [31]. Suggestions that current models accurately simulate the mid-Holocene climate optimum rely on comparisons between projected and observed summer warming only, overlooking models’ failure to represent winter temperatures correctly, perhaps through undue sensitivity to CO_{2}-driven warming.
3 An irreducibly simple climate-sensitivity model
An irreducibly simple climate-sensitivity model is now described. It is intended to enable even non-specialists to study why the models are running hot and to obtain reasonable estimates of future anthropogenic temperature change. The model is calibrated against the climate-sensitivity interval projected by the CMIP3 suite of models and against global warming since 1850. Its utility is demonstrated by its application to the principal outputs of the CMIP5 models and to other questions related to climate sensitivity.
This simple equation represents, in an elementary but revealing fashion, the essential determinants of the temperature response to any anthropogenic radiative perturbation of the climate and permits even the non-specialist to generate respectable approximate estimates of temperature response over time. It is not, of course, intended to replace the far more complex general-circulation models; rather, it is intended to illuminate them.
4 Parameters of the simple model
The parameters of the simple model are now described.
4.1 The CO_{2} fraction q_{t}
The principal direct anthropogenic radiative forcing is CO_{2}. Other influential greenhouse gases are CH_{4}, N_{2}O, and tropospheric O_{3}. In AR4, it was estimated that CO_{2} would contribute some 70 % of total net anthropogenic forcing from 2001 to 2100, so that q_{100} = 0.7. Likewise, AR5, on the RCP 8.5 business-as-usual radiative-forcing scenario, projects that CO_{2} concentration by 2100 will be 936 ppmv, but that the influence of other greenhouse gases will raise that value to 1,313 ppmv CO_{2} equivalent (CO_{2}e), again implying a CO_{2} fraction q_{100} = 0.7. Note that the discrepancy between ratios of forcings and of CO_{2} concentrations is small over the relevant intervals.
However, AR5 concludes at p. 165 that CO_{2} contributed 80 % of greenhouse-gas forcing from 2005 to 2011: “Based on updated in situ observations, this assessment concludes that these trends resulted in a 7.5 % increase in RF from GHGs from 2005 to 2011, with carbon dioxide (CO_{2}) contributing 80 %.” Furthermore, models have greatly exaggerated the growth of atmospheric CH_{4} concentration. It is reasonable to suppose that CO_{2} will represent not <83 % of total anthropogenic forcings over the twenty-first century: i.e., q_{t} ≥ 0.83. To retain compatibility with IPCC’s practice of expressing the CO_{2} fraction q_{t} as a percentage of total anthropogenic forcing, the convention has been retained here. Accordingly, the total anthropogenic forcing may be derived by taking the reciprocal of the CO_{2} fraction; thus, q_{t} ≥ 0.83 ⇒ q_{t}^{−1} ≤ 1.2. The CO_{2} radiative forcing (ΔF_{t}) is essentially being scaled by this factor, as a measure of weighting the CO_{2}.
4.2 The CO_{2} radiative forcing ΔF_{t}
4.3 The Planck climate-sensitivity parameter λ_{0}
However, [33], cited in AR4, pointed out that “[i]ntermodel differences in λ_{0} arise from different spatial patterns of warming; models with greater high-latitude warming, where the temperature is colder, have smaller values of λ_{0.}”
Accordingly [33], followed by AR4, gave λ_{0} = 3.2^{−1} = 0.3125 K W^{−1} m^{2} to allow for variation with latitude (note, however, that AR4 expresses λ_{0} in W m^{−2} K^{−1}). Other values of λ_{0} in the literature are 0.29–0.30 [34–37]. Though the value of λ_{0} may vary somewhat over time, IPCC’s value 0.3125 K W^{−1} m^{2} may safely be taken as constant at sub-millennial timescales.
4.4 The temperature-feedback sum f_{t}
4.5 The closed-loop gain g_{t} and the open-loop or system gain G_{t}
The effect of temperature feedbacks is to augment or diminish the instantaneous temperature response ΔT_{0} to a direct forcing. The closed-loop gain g_{t} is the product of the instantaneous or Planck climate-sensitivity parameter λ_{0} and the feedback sum f_{t}. The open-loop or system gain factor G_{t} is equal to (1 − g_{t}) ^{−1}. Both g_{t} and G_{t} are unitless. The equilibrium temperature response ΔT_{∞} is the product of the instantaneous temperature response ΔT_{0} and the system gain factor G_{t}.
4.6 The equilibrium climate-sensitivity parameter λ_{∞}
The equilibrium-sensitivity parameter λ_{∞}, in K W^{−1} m^{2}, is the product of the Planck parameter λ_{0} = 3.2^{−1} K W^{−1} m^{2} and the system gain factor G_{t}. Climate sensitivity ΔT_{∞} is the product of λ_{∞} and a given forcing ΔF_{∞}.
4.7 Derivation of G_{∞}, g_{∞}, and f_{∞} from ΔT_{∞}/ΔF_{∞}
Derivation of the equilibrium-sensitivity parameter λ_{∞} from the Planck parameter λ_{0} and the feedback sum f_{∞}, based on the lower, central and upper estimates of f_{∞} in AR4 (left) and AR5 (right)
AR4 | Derivation of λ_{∞} | AR5 | ||||||
---|---|---|---|---|---|---|---|---|
f_{∞} | g_{∞} | G_{∞} | λ_{∞} | f_{∞} | g_{∞} | G_{∞} | λ_{∞} | |
Unamplified feedback sum | Closed-loop gain | System gain factor | Equilibrium-sensitivity parameter | λ_{0} = 3.2^{−1} | Unamplified feedback sum | Closed-loop gain | System gain factor | Equilibrium-sensitivity parameter |
f_{1} + f_{2} + ··· + f_{n} | λ_{0}f_{∞} | (1–g_{∞})^{−1} | λ_{0}G_{∞} | Derivation | f_{1} + f_{2} + ··· + f_{n} | λ_{0}f_{∞} | (1–g_{∞})^{−1} | λ_{0}G_{∞} |
(W m^{−2} K^{−1}) | Unitless | Unitless | (K W^{−1} m^{2}) | Units | (W m^{−2} K^{−1}) | Unitless | Unitless | (K W^{−1} m^{2}) |
1.5 | 0.469 | 1.882 | 0.588 | Low est. | 1.0 | 0.313 | 1.455 | 0.455 |
1.9 | 0.594 | 2.462 | 0.769 | Best est. | 1.5 | 0.469 | 1.882 | 0.588 |
2.4 | 0.750 | 4.000 | 1.250 | High est. | 2.2 | 0.688 | 3.200 | 1.000 |
4.8 The transience fraction r_{t}
In [38], a simple climate model was used, comprising an advective–diffusive ocean and an atmosphere with a Planck sensitivity ΔT_{0} = 1.2 K, the product of the direct radiative forcing 5.35 ln 2 = 3.708 W m^{−2} in response to a CO_{2} doubling and the zero-feedback climate-sensitivity parameter λ_{0} = 3.2^{−1} K W^{−1} m^{2}. The climate object thus defined was forced with a 4 W m^{−2} pulse at t = 0, and the evolutionary curve of climate sensitivity (Fig. 4) was determined. Equilibrium sensitivity was found to be 3.5 K, of which 1.95 K is shown as occurring after 50 years, implying r_{50} = 0.56. For comparison, AR4 gave 3.26 K as its central estimate of equilibrium climate sensitivity to a doubling of CO_{2} concentration, implying λ_{∞} = 3.26/(5.35 ln 2) = 0.88 K W^{−1} m^{2}. The mean of projected concentrations on the six SRES emissions scenarios in AR4, obtained by enlarging the graphs and overlaying a precise grid on them and reading off and averaging the annual values, is 713 ppmv in 2100 compared with 368 ppmv in 2000.
Approximate values of r_{t} at values f_{∞} ≤ 0 and f_{∞} = 0.5, 1.3, 2.1, and 2.9 over periods t = 25–300 years, derived from [38]
Approximate values of r_{t} | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Years t | 25 | 50 | 75 | 100 | 125 | 150 | 175 | 200 | 225 | 250 | 275 | 300 |
f_{∞} ≤ 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
f_{∞} = 0.5 | 0.65 | 0.70 | 0.74 | 0.77 | 0.79 | 0.80 | 0.81 | 0.82 | 0.83 | 0.84 | 0.85 | 0.85 |
f_{∞} = 1.3 | 0.55 | 0.63 | 0.65 | 0.68 | 0.70 | 0.71 | 0.72 | 0.73 | 0.74 | 0.75 | 0.75 | 0.76 |
f_{∞} = 2.1 | 0.40 | 0.49 | 0.53 | 0.56 | 0.57 | 0.59 | 0.60 | 0.61 | 0.62 | 0.63 | 0.64 | 0.64 |
f_{∞} = 2.9 | 0.15 | 0.19 | 0.22 | 0.23 | 0.25 | 0.26 | 0.27 | 0.28 | 0.29 | 0.30 | 0.30 | 0.30 |
It is not possible to provide a similar table for values of f_{∞} given in AR4 or AR5, since IPCC provides no evolutionary curve similar to that in Fig. 4. Nevertheless, Table 2, derived from [38], allows approximate values of r_{t} to be estimated.
5 How does the model represent different conditions?
The simple model has only five tunable parameters: the CO_{2} fraction q_{t}, dependent on projected CO_{2} concentration change; the CO_{2} radiative forcing ΔF_{t}; the transience fraction r_{t}; the Planck sensitivity parameter λ_{0}, on which the instantaneous temperature response ΔT_{0} and the system gain G_{t} are separately dependent; and the feedback sum f_{t}, of which the equilibrium-sensitivity parameter λ_{∞} is a function.
These five parameters permit representation of any combination of anthropogenic forcings; of expected warming at any stage from inception to equilibrium after perturbation by forcings of any magnitude or sign; and of any combination of feedbacks, positive or negative, linear or nonlinear. The model makes explicit the relative contributions of forcings and feedbacks to projected anthropogenic global warming. Feedbacks, mentioned >1,000 times in AR5, are the greatest source of uncertainty in predicting anthropogenic temperature change.
6 Calibration against climate-sensitivity projections in AR4
Comparison of the Charney-sensitivity interval 2.9 [2.2, 4.6] K generated by the model on the basis of the feedback-sum interval f on 1.9 [1.5, 2.4] (AR5) with the CMIP3 sensitivity interval 3.26 [2.0, 4.5] K (AR4)
AR4 | f | λ_{∞} | ΔF_{2x} | ΔT_{2x} | ΔT_{2x} | Variance |
---|---|---|---|---|---|---|
2x CO_{2} | AR4 | Table 1 | 5.35 ln 2 | Model (λ_{∞}ΔF_{2x}) | AR4 Box 10.2 | AR4-model model |
(W m^{−2} K^{−1}) | (K W^{−1} m^{2}) | (W m^{−2}) | (K) | (K) | (%) | |
Lowest | 1.5 | 0.588 | 2.20 | 2.00 | −9 | |
Best | 1.9 | 0.769 | 3.708 | 2.85 | 3.26 | 14 |
Highest | 2.4 | 1.250 | 4.60 | 4.50 | −2 |
The chief reason why the central estimate in AR4 is 14 % greater than the model’s central estimate is that IPCC’s central estimate is close to the mean of the upper and lower bounds, while the model’s central estimate is closer to the lower than to the upper bound because it is derived from AR4’s central estimate of the feedback sum. This asymmetry is inherent in Eq. (1), but is not reflected in AR4’s central estimate. The sensitivity interval 2.9 [2.2, 4.6] K found by the simple model is accordingly close enough to the interval 3.26 [2.0, 4.5] K in AR4, Box 10.2, to calibrate the model.
7 Calibration against observed temperature change since 1850
Modeled and observed global warming, January 1850 to April 2014
1850–2014 | CO_{2} (1850) | CO_{2} (2014) | f | q_{t}^{−1} | r_{t} | λ_{∞} | ΔF_{t} | ΔT_{2x} (Model) | ΔT_{2x} (Obs.) | Variance |
---|---|---|---|---|---|---|---|---|---|---|
Basis | cf. 278 (1850) | NOAA (2014) | AR5 fig. 9.43 | 2.19 1.72 | Table 2 | Table 1 | 5.35 ln (393/285) | q_{t}^{−1}r_{t}λ_{∞} ΔF_{t} | HadCRUT4 | Obs-model model |
Units | (ppmv) | (ppmv) | (W m^{−2} K^{−1}) | (K W^{−1} m^{2}) | (W m^{−2}) | (K) | (K) | (%) | ||
1.0 | 0.7 | 0.455 | 0.7 | |||||||
285 | 400 | 1.5 | 1.27 | 0.6 | 0.588 | 1.72 | 0.8 | 0.8 | 0 | |
2.2 | 0.5 | 1.000 | 1.1 |
Assuming that all global warming since 1850 was anthropogenic, the model fairly reproduces the change in global temperature since then, suggesting that the 0.6 K committed but unrealized warming mentioned in AR4, AR5 is non-existent. If some global warming was natural, then a fortiori the likelihood of committed but unrealized warming is small.
8 Application of the model to global-warming projections in AR5
8.1 The climate-sensitivity interval
In FAR, the implicit central estimate of λ_{∞} was 0.769 K W^{−1} m^{2}, giving an equilibrium climate sensitivity 2.9 K in response to a CO_{2} doubling. The CMIP3 model ensemble in AR4, p. 798, box 10.2 gave as its central estimate an equilibrium sensitivity of 3.26 K, implying that λ_{∞} = 0.879 K W^{−1} m^{2} and consequently that f = 2.063 W m^{−2} K^{−1}, somewhat above the 1.9 W m^{−2} K^{−1} given in [32].
No best estimate for equilibrium climate sensitivity can now be given because of a lack of agreement on values across assessed lines of evidence and studies.
The simple model indicates that, as a result of the fall in the interval of estimates of f from 1.9 [1.5, 2.4] W m^{−2} K^{−1} in AR4 to 1.5 [1.0, 2.2] W m^{−2} K^{−1} in AR5, the Charney-sensitivity interval in response to a CO_{2} doubling should have been reduced from 3.26 [2.0. 4.5] K to 2.2 [1.7, 2.7] K. Yet, the CMIP5 climate-sensitivity interval given in AR5 is 3.2 [2.1, 4.7] K (AR5).
Comparison of the Charney-sensitivity interval 2.2 [1.7, 3.7] K generated by the model on the basis of the feedback-sum interval f on 1.0 [1.5, 2.2] W m^{−2} K^{−1} (AR5) with IPCC’s published climate-sensitivity interval 3.2 [2.1, 4.7] K (AR5)
AR5 | f_{∞} | λ_{∞} | ΔF_{2x} | ΔT_{2x} | ΔT_{2x} | Variance |
---|---|---|---|---|---|---|
2x CO_{2} | AR5 fig. 9.43 | Table 1 | 5.35 ln 2 | Model (λ_{∞} ΔF_{2x}) | AR5 (SPM) | AR5-model model |
(W m^{−2} K^{−1}) | (K W^{−1} m^{2}) | (W m^{−2}) | (K) | (K) | (%) | |
Lowest | 1.0 | 0.455 | 1.7 | 2.1 | 24 | |
Best | 1.5 | 0.588 | 3.708 | 2.2 | 3.2 | 46 |
Highest | 2.2 | 1.000 | 3.9 | 4.7 | 21 |
8.2 Projected warming in the RCP forcing scenarios
In AR5, IPCC introduces four new forcing scenarios, based on net anthropogenic forcings of 2.6, 4.5, 6.0, and 8.5 W m^{−2} over 1750–2100, of which approximately 2.3 W m^{−2} is shown as having occurred by 2011. There has also been global warming of approximately 0.9 K since 1750.
Comparison of projected warming under the RCP 2.6, 4.5, 6.0, and 8.5 radiative forcing scenarios, 2014–2100, as generated by the simple model and as given in AR5
RCP | CO_{2} (2100) | CO_{2}e (2100) | r_{t} | q_{t}^{−1} | λ_{∞} | ΔF_{t} | ΔT_{2x} (Model) | ΔT_{2x} (AR5) | Variance |
---|---|---|---|---|---|---|---|---|---|
Basis | Box SPM.1 | Box SPM.1 | Table 1 | CO_{2}e CO_{2} | Table 2 | 5.35 ln (CO_{2}/400) | q_{t}^{−1}r_{t}λ_{∞} ΔF_{t} | AR5 SPM | RCP-model model |
(ppmv) | (ppmv) | (K W^{−1} m^{2}) | (W m^{−2}) | (K) | (K) | (%) | |||
RCP 2.6 | 421 | 475 | 0.7 | 0.455 | 0.10 | 0.3 | +200 | ||
0.6 | 1.128 | 0.588 | 0.274 | 0.11 | 1.0 | +809 | |||
0.5 | 1.000 | 0.15 | 1.7 | +1,033 | |||||
RCP 4.5 | 538 | 630 | 0.7 | 0.455 | 0.59 | 1.1 | +86 | ||
0.6 | 1.171 | 0.588 | 1.586 | 0.66 | 1.8 | 73 | |||
0.5 | 1.000 | 0.93 | 2.6 | +180 | |||||
RCP 6.0 | 670 | 800 | 0.7 | 0.455 | 1.05 | 1.4 | +33 | ||
0.6 | 1.194 | 0.588 | 2.760 | 1.16 | 2.2 | +90 | |||
0.5 | 1.000 | 1.65 | 3.1 | +88 | |||||
RCP 8.5 | 936 | 1,313 | 0.7 | 0.455 | 2.00 | 2.6 | +30 | ||
0.6 | 1.402 | 0.588 | 4.548 | 2.25 | 3.7 | +39 | |||
0.5 | 1.000 | 3.19 | 4.8 | +50 |
8.3 An observationally based estimate of global warming to 2100
The simple model may be deployed to obtain observationally based best estimates of global warming to 2100, for instance, by adopting realistic values of the CO_{2} forcing ΔF_{t}, the feedback sum f, the CO_{2} fraction q_{t}, and the transience fraction r_{t}.
8.3.1 The CO_{2} forcing ΔF_{t}
RCP 8.5 is the “business-as-usual” scenario in AR5. However, the assumptions underlying it are unrealistic (see Discussion). In the more realistic RCP 6.0 scenario, atmospheric CO_{2} concentration, currently 400 ppmv, is projected to reach 670 ppmv by 2100, so that ΔF_{t} from 2015 to 2100 will be 5.35 ln(670/400), or 2.760 W m^{−2}.
8.3.2 The feedback sum f
A plausible upper bound to f may be found by recalling that absolute surface temperature has varied by only 1 % or 3 K either side of the 810,000-year mean [40, 41]. This robust thermostasis [42, 43], notwithstanding Milankovich and other forcings, suggests the absence of strongly net-positive temperature feedbacks acting on the climate.
8.3.3 The CO_{2} fraction q_{t}
IPCC’s implicit value for q_{t} falls on [0.71, 0.89], the higher values corresponding to the lower projected total anthropogenic forcings. A reasonable interval for q_{t} corresponding to low values of f_{t} is thus [0.8, 0.9], so that q_{t}^{−1} falls on [1.10, 1.25]. For comparison, on RCP 6.0 in AR5, the implicit value for q_{t}^{−1} is 1.194 (Table 6).
8.3.4 The transience fraction r_{t}
Where f_{t} ≤ 0.3, little error will arise if, for all t, r_{t} is taken as unity: For at sufficiently small f_{t}, there is little difference between instantaneous and equilibrium response.
8.3.5 Projected global warming from 2014 to 2100
Comparison of the climate-sensitivity interval 1.0 [0.8, 1.3] K generated by the model with IPCC’s climate-sensitivity interval 3.2 [2.1, 4.7] K [AR5, SPM]
8.4 How much post-1850 global warming was anthropogenic?
Assuming 285 ppmv CO_{2} in 1850 and 400 ppmv in 2014, and applying the observationally derived values of f_{t}, holding r_{t} at unity, and taking q_{t}^{−1} = 2.29/1.813 = 1.263 to allow for the greater fraction of past warming attributable to CH_{4}, the simple model determines the approximate fraction of the 0.8 K observed global warming since 1850 that was anthropogenic as 78 % [62 %, 104 %].
If it is assumed that g_{t} < +0.1, warming is already at equilibrium, since r_{t} → 1 for the implicit values f_{t} ≤ +0.3 W m^{−2} K^{−1}, on this scenario there is probably no committed but unrealized global warming. If AR4 is correct in its estimate that 0.6 K warming is in the pipeline, then <0.2 K anthropogenic warming has occurred since 1850, indicating that warming realized since then is substantially natural.
8.5 An observationally based estimate of Charney sensitivity
With the observationally derived values of f, the model provides a new estimate of the Charney sensitivity (Table 7). If temperature feedbacks are at most weakly net-positive, with loop gain g on [−0.5, +0.1] as Fig. 5 and 810,000 years of thermostasis suggest, Charney sensitivity may fall on 1.0 [0.8, 1.3] K. The model’s central estimate is one-third of the 3.2 K central estimate from the CMIP5 model ensemble in AR5, or of the 3.26 K central estimate from the CMIP3 model ensemble in AR4.
For comparison, in [44], g is found to fall on [−1.5, +0.7], so that, assuming the forcing at CO_{2} doubling is 4 W m^{−2}, a little above the 3.71 W m^{−2} that IPCC currently regards as canonical, the equilibrium Charney sensitivity ΔT_{2×} falls on [0.5, 4.2] K. The model’s climate-sensitivity interval is better constrained than the CMIP models’ intervals because across a broad interval of weakly positive to net-negative feedbacks there is little change in the temperature response.
8.5.1 Charney sensitivity: summary of results
Charney-sensitivity estimates from all five IPCC Assessment Reports and, in bold face, from the simple model
Climate-sensitivity estimates | Central (K) | Lower (K) | Upper (K) |
---|---|---|---|
FAR (Models) | 4.0 | 1.9 | 5.2 |
FAR (SPM) | 2.5 | 1.5 | 4.5 |
SAR (SPM) | 2.5 | 1.5 | 4.5 |
TAR (Models) | 3.0 | 1.7 | 4.2 |
TAR (SPM) | None | 1.5 | 4.5 |
AR4 (CMIP3 models) | 3.26 | 2.0 | 4.5 |
AR4 (SPM) | 3.0 | 2.0 | 4.5 |
AR5 (SPM) | None | 1.5 | 4.5 |
AR5 (CMIP5 models) | 3.2 | 2.1 | 4.7 |
AR5 (CMIP5: central estimate rebased to mean feedback sumf) | 2.9 | ||
AR5: adjusted for AR5 feedback sum f on 1.5 [1.0, 2.2] | 2.2 K | 1.7 | 3.9 |
Simple model: f on −0.64 [−1.6, +0.32] | 1.0 K | 0.8 | 1.3 |
As Table 8 shows, correcting the output of the CMIP5 models to determine the central estimate of temperature response from the central estimate of the feedback sum and to determine the entire sensitivity interval from the revised feedback-sum interval given in AR5 reduces the sensitivity interval from 3.2 [2.1, 4.7] K to 2.2 [1.7, 3.9] K, bringing the CMIP5 feedback-sum interval into line with IPCC’s interval.
If, however, the loop gain g is indeed below the process engineers’ limit for stability, namely +0.1, compatible with the results in [21, 23], then the simple model’s output giving a climate-sensitivity interval 1.0 [0.8, 1.3] K may be preferable.
9 How skillful is the model?
10 Discussion
The irreducibly simple model presented here aims specifically to study climate sensitivity. Though it is capable of representing in a rough and ready fashion all the forcings and feedbacks discussed in AR5, the question arises whether extreme simplicity renders such models altogether valueless in contrast to the more complex general-circulation models.
Recently, it was explained in [45] that although the complex models cover many physical, chemical and biological processes in their representation of the Earth’s climate system, the added complexity has naturally led to great difficulty in identifying the chains of causality in the climate object—what the authors call “the processes most responsible for a certain effect.”
Two recent examples of the substantial uncertainty in representing climate by complex models indicate that greater complexity does not necessarily entail improved performance, despite myriad improvements and intense scrutiny.
The first example: It was recently reported [46] that increased spatial resolution had led to improvements in simulations of sea-level pressure, surface temperatures, etc., in GISS’ latest model, E2, but that simultaneously, “some degradations are seen in precipitation and cloud metrics.” Increased spatial resolution in a model, therefore, does not automatically lead to improvement.
The second example: The IPSL-CM5A modeling group’s recent study [47] of the skill of horizontal and vertical atmospheric grid configuration in representing the observed climate reported that, when the number of atmospheric layers was increased from 19 to 39 to improve stratospheric resolution, a substantial global energy imbalance requiring retuning of model parameters resulted, but that, paradoxically, these significant impacts of the model’s grid resolution had not led to any significant changes in projected climate sensitivity.
It is not necessarily true, therefore, that improvements in the resolution of a model will refine the determination of climate sensitivity. By the same token, a reduction in complexity—even an irreducible reduction—does not necessarily entail a reduction in the reliability with which climate sensitivity is determined.
On the other hand, it would be inappropriate to claim that the simple model is preferable to the complex general-circulation models. Its purpose is more limited than theirs, being narrowly focused on determining the transient and equilibrium responses of global temperature to specified radiative forcings and feedbacks in a simplified fashion. The simple model is not a replacement for the general-circulation models, but it is capable of illuminating their performance. It also puts climate-sensitivity modeling within the reach of those who have no access to or familiarity with the general-circulation models. In effect, this paper is the user manual for the simple model, bringing it within the reach of all who have a working knowledge of elementary mathematics and physics.
Irreducible simplicity is the chief innovation embodied in the simple model. While it is rooted in the mainstream mathematics and physics of climate sensitivity and is capable of reflecting no less wide a range of scenarios than the general-circulation models, it allows a rapid but not unreliable determination of climate sensitivity by anyone even at undergraduate level, providing insights not only into the relevant physics but also into the extent to which the more complex models are adequately reflecting the physics.
The complex general-circulation models have been running hot for a quarter of a century. The simple model confirms the hot running and exposes several of the reasons for it.
Firstly, application of the simple model reveals that the central climate-sensitivity estimate in the CMIP5 ensemble is somewhat too high because IPCC has taken its mid-range climate-sensitivity estimate as the mean of its upper- and lower-bound estimates rather than determining it from the mean feedback sum f_{∞}. By contrast, in [38] the central climate-sensitivity estimate was perhaps more correctly derived from the central feedback-sum estimate f_{∞} = 2.1 W m^{−2} K^{−1}, the exact mean of the lower and upper bounds f_{∞} on [1.3, 2.9] W m^{−2} K^{−1}. Accordingly, in [38] the central climate-sensitivity estimate 3.5 K is significantly closer to the lower-bound estimate 2.0 K than to the upper-bound estimate 12.7 K. The rapidly increasing slope of climate sensitivity against loop gain g_{∞} as the value of g_{∞} approaches unity (the singularity in the Bode feedback-amplification equation [48]), is the reason for this asymmetry (Fig. 5), and is also the reason for the extremely high-sensitivity estimates sometimes presented in the journals. Implicitly, f_{∞} in the CMIP5 ensemble falls on 1.923 [1.434, 2.411] W m^{−2} K^{−1}. The mean of these two values is 1.923 W m^{−2} K^{−1}. Based on the mean feedback sum f_{∞} = 1.923 W m^{−2} K^{−1}, the CMIP5 central estimate of climate sensitivity should have been 2.9 K, not 3.2 K.
Secondly, the simple model reveals that the climate sensitivity 3.3 [2.0, 4.5] K in AR4 should have fallen sharply to 2.2 [1.7, 3.7] K in AR5 commensurately with the reduction of the feedback-sum interval between the two reports (Fig. 3). For the variance between the CMIP3 and CMIP5 projections of climate sensitivity is inferentially confined to the feedback-sum interval. If the CMIP5 models took account of significant net-positive feedbacks not included in AR5, Fig. 9.43, in the chart of climate-relevant feedbacks (Fig. 3), it is not clear why that chart was not updated to include them. The sharp reduction of the feedback-sum interval in CMIP5 and hence in AR5 compared with the interval in CMIP3 and hence in AR4 mandates a sharp reduction in the climate-sensitivity interval, which, however, was instead increased somewhat.
Thirdly, the simple model shows that even the reduced feedback-sum interval in CMIP5 and hence in AR5 seems implausibly high when set against the thermostasis over geological timescales shown in [40]. In Fig. 5, g ≤ +0.1 is consistent with the inferred thermostasis. Charney sensitivity would then be 1.3 K or less—below even the lower bound of the climate-sensitivity interval [1.5, 3] K in AR5.
Fourthly, the simple model demonstrates that, in AR5, the estimates of global warming to 2100 under the four RCP scenarios (Table 5) project much more warming over the twenty-first century than they should. For instance, under the RCP 2.6 scenario, it is expected that there will be no more than 2.6 W m^{−2} radiative forcing to 2100, of which some 2.3 W m^{−2} had already occurred by 2011. Even adding IPCC’s estimate of 0.6 K committed but unrealized warming to the small warming yet to be generated by the 0.3 W m^{−2} forcing still to come by 2100 under this scenario, it is not easy to understand why IPCC’s upper-bound warming estimate on RCP 2.6 is as high as 1.7 K.
Fifthly, application of the simple model raises the question why AR5 adopted the extreme RCP 8.5 scenario at all. On that scenario, atmospheric CO_{2} concentration is projected to reach 936 ppmv by 2100 on the basis of two implausible assumptions: first, that global population will be 12 billion by 2100, though the UN predicts that population will peak at little more than 10 billion by not later than 2070 and will fall steeply thereafter; and secondly, that coal will contribute as much as 50 % of total energy supply, though gas is rapidly replacing coal in many countries, a process that will accelerate as shale gas comes on stream. Furthermore, the observed increase in CH_{4} concentration at a mean rate of 3 ppbv year^{−1} from 1990 to 2011, taken with the history of very substantial over-prediction of the CH_{4} growth rate, does not seem to justify IPCC in projecting that, on the RCP 8.5 scenario, the mean rate of increase in CH_{4} concentration from 2015 to 2100 will be 21 ppbv year^{−1}, seven times the observed rate of increase over recent decades.
The utility of the simple model lies in identifying discrepancies such as those enumerated above. It should not be seen as a substitute for the more complex models, but as a simple benchmark against which the plausibility of their outputs may be examined.
11 Conclusion
Resolving the discrepancies between the methodology adopted by IPCC in AR4 and AR5 is vital. Once those discrepancies are corrected for, it appears that the impact of anthropogenic global warming over the next century, and even as far as equilibrium many millennia hence, may be no more than one-third to one-half of IPCC’s current projections.
Suppose, for instance, that the equilibrium response to a CO_{2} doubling is, as the simple model credibly suggests it is, <1 K. Suppose also that the long-run CO_{2} fraction proves to be as high as 0.9. Again, this possibility is credible. Finally, suppose that remaining affordably recoverable reserves of fossil fuels are as much as thrice those that have been recovered and consumed so far. Then, the total warming we shall cause by consuming all remaining recoverable reserves will be little more than 2.2 K, and not the 12 K imagined by IPCC on the RCP 8.5 scenario. If so, the case for any intervention to mitigate CO_{2} emissions has not necessarily been made: for the 2.2 K equilibrium warming we project would take place only over many hundreds of years. Also, the disbenefits of more extreme heat may well be at least matched by the benefits of less extreme cold. It is no accident that 90 % of the world’s living species thrive in the warm, wet tropics, while only 1 % live at the cold, dry poles. As a benchmark, AR5 estimates that adaptation to the 2–3 K global warming it expects by 2100 will cost 0.2 %–2.0 % of global GDP, broadly in line with the cost estimate of 0–3 % of GDP in Lord Stern’s report for the UK Government on the economics of climate change in 2006. However, the reviewed journals of economics generally report that the cost of mitigation today would be likely to exceed these low costs of adaptation to projected global warming, perhaps by as much as one or two orders of magnitude.
Under different assumptions, the simple model is of course capable of reaching conclusions more alarming (but arguably less reasonable) than those that have been sketched here. Be that as it may, the utility of the model lies in making accessible for the first time the distinction between the relative contributions of forcings and feedbacks; in exposing anomalies requiring clarification in the outputs of the general-circulation models, which seem to agree ever more closely with each other while departing ever farther from observation (Fig. 1); and, above all, in facilitating the rapid and simple estimation of both transient and equilibrium climate sensitivity under a wide range of assumptions and without the need either for climatological expertise or for access to the world’s most powerful computers and complex models. The simple model has its limitations, but it has its uses too.
Conflict of interest
The authors declare that they have no conflict of interest.