Feeds:
Posts
Comments

Archive for the ‘Climate Models’ Category

In Impacts – II – GHG Emissions Projections: SRES and RCP we looked at projections of emissions under various scenarios with the resulting CO2 (and other GHG) concentrations and resulting radiative forcing.

Why do we need these scenarios? Because even if climate models were perfect and could accurately calculate the temperature 100 years from now, we wouldn’t know how much “anthropogenic CO2” (and other GHGs) would have been emitted by that time. The scenarios allow climate modelers to produce temperature (and other climate variable) projections on the basis of each of these scenarios.

The IPCC AR5 (fifth assessment report) from 2013 says (chapter 12, p. 1031):

Global mean temperatures will continue to rise over the 21st century if greenhouse gas (GHG) emissions continue unabated.

Under the assumptions of the concentration-driven RCPs, global mean surface temperatures for 2081–2100, relative to 1986–2005 will likely be in the 5 to 95% range of the CMIP5 models:

  • 0.3°C to 1.7°C (RCP2.6)
  • 1.1°C to 2.6°C (RCP4.5)
  • 1.4°C to 3.1°C (RCP6.0)
  • 2.6°C to 4.8°C (RCP8.5)

Global temperatures averaged over the period 2081– 2100 are projected to likely exceed 1.5°C above 1850-1900 for RCP4.5, RCP6.0 and RCP8.5 (high confidence), are likely to exceed 2°C above 1850-1900 for RCP6.0 and RCP8.5 (high confidence) and are more likely than not to exceed 2°C for RCP4.5 (medium confidence). Temperature change above 2°C under RCP2.6 is unlikely (medium confidence). Warming above 4°C by 2081–2100 is unlikely in all RCPs (high confidence) except for RCP8.5, where it is about as likely as not (medium confidence).

I commented in Part II that RCP8.5 seemed to be a scenario that didn’t match up with the last 40-50 years of development. Of course, the various scenario developers give their caveats, for example, Riahi et al 2007:

Given the large number of variables and their interdependencies, we are of the opinion that it is impossible to assign objective likelihoods or probabilities to emissions scenarios. We have also not attempted to assign any subjective likelihoods to the scenarios either. The purpose of the scenarios presented in this Special Issue is, rather, to span the range of uncertainty without an assessment of likely, preferable, or desirable future developments..

Readers should exercise their own judgment on the plausibility of above scenario ‘storylines’..

To me RCP6.0 seems a more likely future (compared with RCP8.5) in a world that doesn’t have any significant attempt to tackle CO2 emissions. That is, no major change in climate policy to today’s world, but similar economic and population development (note 1).

Here is the graph of projected temperature anomalies for the different scenarios. :

From AR5, chapter 12

From AR5, chapter 12

Figure 1

That graph is hard to make out for 2100, here is the table of corresponding data. I highlighted RCP6.0 in 2100 – you can click to enlarge the table:

ar5-ch12-table12-2-temperature-anomaly-2100-499px

Figure 2 – Click to expand

Probabilities and Lists

The table above has a “1 std deviation” and a 5%-95% distribution. The graph (which has the same source data) has shading to indicate 5%-95% of models for each RCP scenario.

These have no relation to real probability distributions. That is, the range of 5-95% for RCP6.0 doesn’t equate to: “the probability is 90% likely that the average temperature 2080-2100 will be 1.4-3.1ºC higher than the 1986-2005 average”.

A number of climate models are used to produce simulations and the results from these “ensembles” are sometimes pressed into “probability service”. For some concept background on ensembles read Ensemble Forecasting.

Here is IPCC AR5 chapter 12:

Ensembles like CMIP5 do not represent a systematically sampled family of models but rely on self-selection by the modelling groups.

This opportunistic nature of MMEs [multi-model ensembles] has been discussed, for example, in Tebaldi and Knutti (2007) and Knutti et al. (2010a). These ensembles are therefore not designed to explore uncertainty in a coordinated manner, and the range of their results cannot be straightforwardly interpreted as an exhaustive range of plausible outcomes, even if some studies have shown how they appear to behave as well calibrated probabilistic forecasts for some large-scale quantities. Other studies have argued instead that the tail of distributions is by construction undersampled.

In general, the difficulty in producing quantitative estimates of uncertainty based on multiple model output originates in their peculiarities as a statistical sample, neither random nor systematic, with possible dependencies among the members and of spurious nature, that is, often counting among their members models with different degrees of complexities (different number of processes explicitly represented or parameterized) even within the category of general circulation models..

..In summary, there does not exist at present a single agreed on and robust formal methodology to deliver uncertainty quantification estimates of future changes in all climate variables. As a consequence, in this chapter, statements using the calibrated uncertainty language are a result of the expert judgement of the authors, combining assessed literature results with an evaluation of models demonstrated ability (or lack thereof) in simulating the relevant processes (see Chapter 9) and model consensus (or lack thereof) over future projections. In some cases when a significant relation is detected between model performance and reliability of its future projections, some models (or a particular parametric configuration) may be excluded but in general it remains an open research question to find significant connections of this kind that justify some form of weighting across the ensemble of models and produce aggregated future projections that are significantly different from straightforward one model–one vote ensemble results. Therefore, most of the analyses performed for this chapter make use of all available models in the ensembles, with equal weight given to each of them unless otherwise stated.

And from one of the papers cited in that section of chapter 12, Jackson et al 2008:

In global climate models (GCMs), unresolved physical processes are included through simplified representations referred to as parameterizations.

Parameterizations typically contain one or more adjustable phenomenological parameters. Parameter values can be estimated directly from theory or observations or by “tuning” the models by comparing model simulations to the climate record. Because of the large number of parameters in comprehensive GCMs, a thorough tuning effort that includes interactions between multiple parameters can be very computationally expensive. Models may have compensating errors, where errors in one parameterization compensate for errors in other parameterizations to produce a realistic climate simulation (Wang 2007; Golaz et al. 2007; Min et al. 2007; Murphy et al. 2007).

The risk is that, when moving to a new climate regime (e.g., increased greenhouse gases), the errors may no longer compensate. This leads to uncertainty in climate change predictions. The known range of uncertainty of many parameters allows a wide variance of the resulting simulated climate (Murphy et al. 2004; Stainforth et al. 2005; M. Collins et al. 2006). The persistent scatter in the sensitivities of models from different modeling groups, despite the effort represented by the approximately four generations of modeling improvements, suggests that uncertainty in climate prediction may depend on underconstrained details and that we should not expect convergence anytime soon.

Stainforth et al 2005 (referenced in the quote above) tried much larger ensembles of coarser resolution climate models, and was discussed in the comments of Models, On – and Off – the Catwalk – Part Four – Tuning & the Magic Behind the Scenes. Rowlands et al 2012 is similar in approach and was discussed in Natural Variability and Chaos – Five – Why Should Observations match Models?

The way I read the IPCC reports and various papers is that clearly the projections are not a probability distribution. Then the data gets inevitably gets used as a de facto probability distribution.

Conclusion

“All models are wrong but some are useful” as George Box said, actually in a quite unrelated field (i.e., not climate). But it’s a good saying.

Many people who describe themselves as “lukewarmers” believe that climate sensitivity as characterized by the IPCC is too high and the real climate has a lower sensitivity. I have no idea.

Models may be wrong, but I don’t have an alternative model to provide. And therefore, given that they represent climate better than any current alternative, their results are useful.

We can’t currently create a real probability distribution from a set of temperature prediction results (assuming a given emissions scenario).

How useful is it to know that under a scenario like RCP6.0 the average global temperature increase in 2100 has been simulated as variously 1ºC, 2ºC, 3ºC, 4ºC? (note, I haven’t checked the CMIP5 simulations to get each value). And the tropics will vary less, land more? As we dig into more details we will attempt to look at how reliable regional and seasonal temperature anomalies might be compared with the overall number. Likewise rainfall and other important climate values.

I do find it useful to keep the idea of a set of possible numbers with no probability assigned. Then at some stage we can say something like, “if this RCP scenario turns out to be correct and the global average surface temperature actually increases by 3ºC by 2100, we know the following are reasonable assumptions … but we currently can’t make any predictions about these other values..

References

Long-term Climate Change: Projections, Commitments and Irreversibility, M Collins et al (2013) – In: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change

Scenarios of long-term socio-economic and environmental development under climate stabilization, Keywan Riahi et al, Technological Forecasting & Social Change (2007) – free paper

Error Reduction and Convergence in Climate Prediction, Charles S Jackson et al, Journal of Climate (2008) – free paper

Notes

Note 1: As explored a little in the last article, RCP6.0 does include some changes to climate policy but it seems they are not major. I believe a very useful scenario for exploring impact assessments would be the population and development path of RCP6.0 (let’s call it RCP6.0A) without any climate policies.

For reasons of”scenario parsimony” this interesting pathway avoids attention.

Read Full Post »

In one of the iconic climate model tests, CO2 is doubled from a pre-industrial level of 280ppm to 560ppm “overnight” and we find the new steady state surface temperature. The change in CO2 is an input to the climate model, also known as a “forcing” because it is from outside. That is, humans create more CO2 from generating electricity, driving automobiles and other activities – this affects the climate and the climate responds.

These experiments with simple climate models were first done with 1d radiative-convective models in the 1960s. For example, Manabe & Wetherald 1967 who found a 2.3ºC surface temperature increase with constant relative humidity and 1.3ºC with constant absolute humidity (and for many reasons constant relative humidity seems more likely to be closer to reality than constant absolute humidity).

In other experiments, especially more recently, more more complex GCMs simulate 100 years with the CO2 concentration being gradually increased, in line with projections about future emissions – and we see what happens to temperature with time.

There are also other GHGs (“greenhouse” gases / radiatively-active gases) in the atmosphere that are changing due to human activity – especially methane (CH4) and nitrous oxide (N2O). And of course, the most important GHG is water vapor, but changes in water vapor concentration are a climate feedback – that is, changes in water vapor result from temperature (and circulation) changes.

And there are aerosols, some internally generated within the climate and others emitted by human activity. These also affect the climate in a number of ways.

We don’t know what future anthropogenic emissions will be. What will humans do? Build lots more coal-fire power stations to meet energy demand of the future? Run the entire world’s power grid from wind and solar by 2040? Finally invent practical nuclear fusion? How many people will there be?

So for this we need some scenarios of future human activity (note 1).

Scenarios – SRES and RCP

SRES was published in 2000:

In response to a 1994 evaluation of the earlier IPCC IS92 emissions scenarios, the 1996 Plenary of the IPCC requested this Special Report on Emissions Scenarios (SRES) (see Appendix I for the Terms of Reference). This report was accepted by the Working Group III (WGIII) plenary session in March 2000. The long-term nature and uncertainty of climate change and its driving forces require scenarios that extend to the end of the 21st century. This Report describes the new scenarios and how they were developed.

The SRES scenarios cover a wide range of the main driving forces of future emissions, from demographic to technological and economic developments. As required by the Terms of Reference, none of the scenarios in the set includes any future policies that explicitly address climate change, although all scenarios necessarily encompass various policies of other types.

The set of SRES emissions scenarios is based on an extensive assessment of the literature, six alternative modeling approaches, and an “open process” that solicited wide participation and feedback from many groups and individuals. The SRES scenarios include the range of emissions of all relevant species of greenhouse gases (GHGs) and sulfur and their driving forces..

..A set of scenarios was developed to represent the range of driving forces and emissions in the scenario literature so as to reflect current understanding and knowledge about underlying uncertainties. They exclude only outlying “surprise” or “disaster” scenarios in the literature. Any scenario necessarily includes subjective elements and is open to various interpretations. Preferences for the scenarios presented here vary among users. No judgment is offered in this Report as to the preference for any of the scenarios and they are not assigned probabilities of occurrence, neither must they be interpreted as policy recommendations..

..By 2100 the world will have changed in ways that are difficult to imagine – as difficult as it would have been at the end of the 19th century to imagine the changes of the 100 years since. Each storyline assumes a distinctly different direction for future developments, such that the four storylines differ in increasingly irreversible ways. Together they describe divergent futures that encompass a significant portion of the underlying uncertainties in the main driving forces. They cover a wide range of key “future” characteristics such as demographic change, economic development, and technological change. For this reason, their plausibility or feasibility should not be considered solely on the basis of an extrapolation of current economic, technological, and social trends.

The RCPs were in part a new version of the same idea as SRES and published in 2011. My understanding is that the Representative Concentration Pathways worked more towards final values of radiative forcing in 2100 that were considered in the modeling literature, and you can see this in the names of each RCP.

from A special issue on the RCPs, van Vuuren et al (2011)

By design, the RCPs, as a set, cover the range of radiative forcing levels examined in the open literature and contain relevant information for climate model runs.

[Emphasis added]

From The representative concentration pathways: an overview, van Vuuren et al (2011)

This paper summarizes the development process and main characteristics of the Representative Concentration Pathways (RCPs), a set of four new pathways developed for the climate modeling community as a basis for long-term and near-term modeling experiments.

The four RCPs together span the range of year 2100 radiative forcing values found in the open literature, i.e. from 2.6 to 8.5 W/m². The RCPs are the product of an innovative collaboration between integrated assessment modelers, climate modelers, terrestrial ecosystem modelers and emission inventory experts. The resulting product forms a comprehensive data set with high spatial and sectoral resolutions for the period extending to 2100..

..The RCPs are named according to radiative forcing target level for 2100. The radiative forcing estimates are based on the forcing of greenhouse gases and other forcing agents. The four selected RCPs were considered to be representative of the literature, and included one mitigation scenario leading to a very low forcing level (RCP2.6), two medium stabilization scenarios (RCP4.5/RCP6) and one very high baseline emission scenarios (RCP8.5).

Here are some graphs from the RCP introduction paper:

Population and GDP scenarios:

rcp-population-and-gdp-fig2-499px

Figure 1 – Click to expand

I was surprised by the population graph for RCP 8.5 and 6 (similar scenarios are generated in SRES). From reading various sources (but not diving into any detailed literature) I understood that the consensus was for population to peak mid-century at around 9bn people and then reduce back to something like 7-8bn people by the end of the century. This is because all countries that have experienced rising incomes have significantly reduced average fertility rates.

Here is Angus Deaton, in his fascinating and accessible book for people interested in The Great Escape as he calls it (that is, our escape from poverty and poor health):

In Africa in 1950, each woman could expect to give birth to 6.6 children; by 2000, that number had fallen to 5.1, and the UN estimates that it is 4.4 today. In Asia as well as in Latin America and the Caribbean, the decline has been even larger, from 6 children to just over 2..

The annual rate of growth of the world’s population, which reached 2.2% in 1960, was only half of that in 2011.

The GDP graph on the right (above) is lacking a definition. From the other papers covering the scenarios I understand it to be total world GDP in US$ trillions (at 2000 values, i.e. adjusted for inflation), although the numbers don’t seem to align exactly.

Energy consumption for the different scenarios:

Figure 2 – Click to expand

Annual emissions:

Figure 3 – Click to expand

Resulting concentrations in the atmosphere for CO2, CH4 (methane) and N2O (nitrous oxide):

rcp-fig3-ghg-concentrations-499px

Figure 4 – Click to expand

Radiative forcing (for explanation of this term, see for example Wonderland, Radiative Forcing and the Rate of Inflation):

rcp-fig10-rf-499px

Figure 5  – Click to expand

We can see from this figure (fig 5, their fig 10) that the RCP numbers refer to the expected radiative forcing in 2100 – so RCP8.5, often known as the “business as usual” scenario has a radiative forcing, compared to pre-industrial values, of 8.5 W/m². And RCP6 has a radiative forcing in 2100, of 6 W/m².

We can also see from the figure on the right that increases in CO2 are the cause of almost all of most of the increase from current values. For example, only RCP8.5 has a higher methane (CH4) forcing than today.

Business as usual – RCP 8.5 or RCP 6?

I’ve seen RCP8.5 described as “business as usual” but it seems quite an unlikely scenario. Perhaps we need to dive into this scenario more in another article. In the meantime, part of the description from Riahi et al (2011):

The scenario’s storyline describes a heterogeneous world with continuously increasing global population, resulting in a global population of 12 billion by 2100. Per capita income growth is slow and both internationally as well as regionally there is only little convergence between high and low income countries. Global GDP reaches around 250 trillion US2005$ in 2100.

The slow economic development also implies little progress in terms of efficiency. Combined with the high population growth, this leads to high energy demands. Still, international trade in energy and technology is limited and overall rates of technological progress is modest. The inherent emphasis on greater self-sufficiency of individual countries and regions assumed in the scenario implies a reliance on domestically available resources. Resource availability is not necessarily a constraint but easily accessible conventional oil and gas become relatively scarce in comparison to more difficult to harvest unconventional fuels like tar sands or oil shale.

Given the overall slow rate of technological improvements in low-carbon technologies, the future energy system moves toward coal-intensive technology choices with high GHG emissions. Environmental concerns in the A2 world are locally strong, especially in high and medium income regions. Food security is also a major concern, especially in low-income regions and agricultural productivity increases to feed a steadily increasing population.

Compared to the broader integrated assessment literature, the RCP8.5 represents thus a scenario with high global population and intermediate development in terms of total GDP (Fig. 4).

Per capita income, however, stays at comparatively low levels of about 20,000 US $2005 in the long term (2100), which is considerably below the median of the scenario literature. Another important characteristic of the RCP8.5 scenario is its relatively slow improvement in primary energy intensity of 0.5% per year over the course of the century. This trend reflects the storyline assumption of slow technological change. Energy intensity improvement rates are thus well below historical average (about 1% per year between 1940 and 2000). Compared to the scenario literature RCP8.5 depicts thus a relatively conservative business as usual case with low income, high population and high energy demand due to only modest improvements in energy intensity.

When I heard the term “business as usual” I’m sure I wasn’t alone in understanding it like this: the world carries on without adopting serious CO2 limiting policies. That is, no international agreements on CO2 reductions, no carbon pricing, etc. And the world continues on its current trajectory of growth and development. When you look at the last 40 years, it has been quite amazing. Why would growth slow, population not follow the pathway it has followed in all countries that have seen rising prosperity, and why would technological innovation and adoption slow? It would be interesting to see a “business as usual” scenario for emissions, CO2 concentrations and radiative forcing that had a better fit to the name.

RCP 6 seems to be a closer fit than RCP 8.5 to the name “business as usual”.

RCP6 is a climate-policy intervention scenario. That is, without explicit policies designed to reduce emissions, radiative forcing would exceed 6.0 W/m² in the year 2100.

However, the degree of GHG emissions mitigation required over the period 2010 to 2060 is small, particularly compared to RCP4.5 and RCP2.6, but also compared to emissions mitigation requirement subsequent to 2060 in RCP6 (Van Vuuren et al., 2011). The IPCC Fourth Assessment Report classified stabilization scenarios into six categories as shown in Table 1. RCP6 scenario falls into the border between the fifth category and the sixth category.

Its global mean long-term, steady-state equilibrium temperature could be expected to rise 4.9° centigrade, assuming a climate sensitivity of 3.0 and its CO2 equivalent concentration could be 855 ppm (Metz et al. 2007).

Some of the background to RCP 8.5 assumptions is in an earlier paper also by the same lead author – Riahi et al 2007, another freely accessible paper (reference below) which is worth a read, for example:

The task ahead of anticipating the possible developments over a time frame as ‘ridiculously’ long as a century is wrought with difficulties. Particularly, readers of this Journal will have sympathy for the difficulties in trying to capture social and technological changes over such a long time frame. One wonders how Arrhenius’ scenario of the world in 1996 would have looked, perhaps filled with just more of the same of his time—geopolitically, socially, and technologically. Would he have considered that 100 years later:

  • backward and colonially exploited China would be in the process of surpassing the UK’s economic output, eventually even that of all of Europe or the USA?
  • the existence of a highly productive economy within a social welfare state in his home country Sweden would elevate the rural and urban poor to unimaginable levels of personal affluence, consumption, and free time?
  • the complete obsolescence of the dominant technology cluster of the day-coal-fired steam engines?

How he would have factored in the possibility of the emergence of new technologies, especially in view of Lord Kelvin’s sobering ‘conclusion’ of 1895 that “heavier-than-air flying machines are impossible”?

Note on Comments

The Etiquette and About this Blog both explain the commenting policy in this blog. I noted briefly in the Introduction that of course questions about 100 years from now mean some small relaxation of the policy. But, in a large number of previous articles, we have discussed the “greenhouse” effect (just about to death) and so people who question it are welcome to find a relevant article and comment there – for example, The “Greenhouse” Effect Explained in Simple Terms which has many links to related articles. Questions on climate sensitivity, natural variation, and likelihood of projected future temperatures due to emissions are, of course, all still fair game in this series.

But I’ll just delete comments that question the existence of the greenhouse effect. Draconian, no doubt.

References

Emissions Scenarios, IPCC (2000) – free report

A special issue on the RCPs, Detlef P van Vuuren et al, Climatic Change (2011) – free paper

The representative concentration pathways: an overview, Detlef P van Vuuren et al, Climatic Change (2011) – free paper

RCP4.5: a pathway for stabilization of radiative forcing by 2100, Allison M. Thomson et al, Climatic Change (2011) – free paper

An emission pathway for stabilization at 6 Wm−2 radiative forcing,  Toshihiko Masui et al, Climatic Change (2011) – free paper

RCP 8.5—A scenario of comparatively high greenhouse gas emissions, Keywan Riahi et al, Climatic Change (2011) – free paper

Scenarios of long-term socio-economic and environmental development under climate stabilization, Keywan Riahi et al, Technological Forecasting & Social Change (2007) – free paper

Thermal equilibrium of the atmosphere with a given distribution of relative humidity, S Manabe, RT Wetherald, Journal of the Atmospheric Sciences (1967) – free paper

The Great Escape, Health, Wealth and the Origins of Inequality, Angus Deaton, Princeton University Press (2013) – book

Notes

Note 1: Even if we knew future anthropogenic emissions accurately it wouldn’t give us the whole picture. The climate has sources and sinks for CO2 and methane and there is some uncertainty about them, especially how well they will operate in the future. That is, anthropogenic emissions are modified by the feedback of sources and sinks for these emissions.

Read Full Post »

The subject of EMICs – Earth Models of Intermediate Complexity – came up in recent comments on Ghosts of Climates Past – Eleven – End of the Last Ice age. I promised to write something about EMICs, in part because of my memory of a more recent paper on EMICs. This article will just be short as I found that I have already covered some of the EMIC ground.

In the previous 19 articles of this series we’ve seen a concise summary (just kidding) of the problems of modeling ice ages. That is, it is hard to model ice ages for at least three reasons:

  • knowledge of the past is hard to come by, relying on proxies which have dating uncertainties and multiple variables being expressed in one proxy (so are we measuring temperature, or a combination of temperature and other variables?)
  • computing resources make it impossible to run a GCM at current high resolution for the 100,000 years necessary, let alone to run ensembles with varying external forcings and varying parameters (internal physics)
  • lack of knowledge of key physics, specifically: ice sheet dynamics with very non-linear behavior; and the relationship between CO2, methane and the ice age cycles

The usual approach using GCMs is to have some combination of lower resolution grids, “faster” time and prescribed ice sheets and greenhouse gases.

These articles cover the subject:

Part Seven – GCM I – early work with climate models to try and get “perennial snow cover” at high latitudes to start an ice age around 116,000 years ago

Part Eight – GCM II – more recent work from the “noughties” – GCM results plus EMIC (earth models of intermediate complexity) again trying to produce perennial snow cover

Part Nine – GCM III – very recent work from 2012, a full GCM, with reduced spatial resolution and speeding up external forcings by a factors of 10, modeling the last 120 kyrs

Part Ten – GCM IV – very recent work from 2012, a high resolution GCM called CCSM4, producing glacial inception at 115 kyrs

One of the the papers I thought about covering in this article (Calov et al 2005) is already briefly covered in Part Eight. I would like to highlight one comment I made in the conclusion of Part Ten:

What the paper [Jochum et al, 2012] also reveals – in conjunction with what we have seen from earlier articles – is that as we move through generations and complexities of models we can get success, then a better model produces failure, then a better model again produces success. Also we noted that whereas the 2003 model (also cold-biased) of Vettoretti & Peltier found perennial snow cover through increased moisture transport into the critical region (which they describe as an “atmospheric–cryospheric feedback mechanism”), this more recent study with a better model found no increase in moisture transport.

So, onto a little more about EMICs.

There are two papers from 2000/2001 describing the CLIMBER-2 model and the results from sensitivity experiments. These are by the same set of authors – Petoukhov et al 2000 & Ganopolski et al 2001 (see references).

Here is the grid:

From Petoukhov et al (2000)

From Petoukhov et al (2000)

The CLIMBER-2 model has a low spatial resolution which only resolves individual continents (subcontinents) and ocean basins (fig 1). Latitudinal resolutions is the same for all modules (10º). In the longitudinal direction the Earth is represented by seven equal sectors (roughly 51º􏰖 longitude) in the atmosphere and land modules.

The ocean model is a zonally averaged multibasin model, which in longitudinal direction resolves only three ocean basins Atlantic, Indian, Pacific). Each ocean grid cell communicates with either one, two or three atmosphere grid cells, depending on the width of the ocean basin. Very schematic orography and bathymetry are prescribed in the model, to represent the Tibetan plateau, the high Antarctic elevation and the presence of the Greenland-Scotland sill in the Atlantic ocean.

The atmospheric model has a simplified approach, leading to the description 2.5D model. The time step can be relaxed to about 1 day per step. The ocean grid is a little finer in latitude.

On selecting parameters and model “tuning”:

Careful tuning is essential for a new model, as some parameter values are not known a priori and incorrect choices of parameter values compromise the quality and reliability of simulations. At the same time tuning can be abused (getting the right results for the wrong reasons) if there are too many free parameters. To avoid this we adhered to a set of common-sense rules for good tuning practice:

1. Parameters which are known empirically or from theory must not be used for tuning.

2. Where ever possible parametrizations should be tuned separately against observed data, not in the context of the whole model. (Most of the parameters values in Table 1 were obtained in this way and only few of them were determined by tuning the model to the observed climate).

3. Parameters must relate to physical processes, not to specific geographic regions (hidden flux adjustments).

4. The number of tuning parameters must be much smaller than the degrees of freedom predicted by the model. (In our case the predicted degrees of freedom exceed the number of tuning parameters by several orders of magnitude).

To apply the coupled climate model for simulations of climates substantially different from the present, it is crucial to avoid any type of ̄flux adjustment. One of the reasons for the need of ̄flux adjustments in many general circulation models is their high computational cost, which makes optimal tuning􏱃 difficult. The high speed of CLIMBER-2 allows us to perform many sensitivity experiments required to identify the physical reasons for model problems and the best parameter choices. A physically correct choice of model parameters is fundamentally different from a flux adjustment; only in the former case the surface fluxes are part of the proper feedbacks when the climate changes.

Note that many GCMs back in 2000 did need to use flux adjustment (in Natural Variability and Chaos – Three – Attribution & Fingerprints I commented “..The climate models “drifted”, unless, in deity-like form, you topped up (or took out) heat and momentum from various grid boxes..)

So this all sounds reasonable. Obviously it is a model with less resolution than a GCM, and even the high resolution (by current standards) GCMs need some kind of approach to parameter selection (see Models, On – and Off – the Catwalk – Part Four – Tuning & the Magic Behind the Scenes).

What I remembered about EMICs and suggested in my comment was based on this 2010 paper by Ganopolski, Calov & Claussen:

We will start the discussion of modelling results with a so-called Baseline Experiment (BE). This experiment represents a “suboptimal” subjective tuning of the model parameters to achieve the best agreement between modelling results and palaeoclimate data. Obviously, even with a model of intermediate complexity it is not possible to test all possible combinations of important model parameters which can be considered as free (tunable) parameters.

In fact, the BE was selected from hundred model simulations of the last glacial cycle with different combinations of key model parameters.

Note, that we consider “tunable” parameters only for the ice-sheet model and the SEMI interface, while the utilized climate component of CLIMBER-2 is the same in previous studies, such as those used by C05 [this is Calov et al. (2005)]. In the next section, we will discuss the results of a set of sensitivity experiments, which show that our modelling results are rather sensitive to the choice of the model parameters..

..The ice sheet model and the ice sheet-climate interface contain a number of parameters which are not derived from first principles. They can be considered as “tunable” parameters. As stated above, the BE was subjectively selected from a large suite of experiments as the best fit to empirical data. Below we will discuss results of a number of additional experiments illustrating the sensitivity of simulated glacial cycle to several model parameters. These results show that the model is rather sensitive to a number of poorly constrained parameters and parameterisations, demonstrating the challenges to realistic simulations of glacial cycles with a comprehensive Earth system model.

And in their conclusion:

Our experiments demonstrate that the CLIMBER-2 model with an appropriate choice of model parameters simulates the major aspects of the last glacial cycle under orbital and greenhouse gases forcing rather realistically. In the simulations, the glacial cycle begins with a relatively abrupt lateral expansion of the North American ice sheets and parallel growth of the smaller northern European ice sheets. During the initial phase of the glacial cycle (MIS 5), the ice sheets experience large variations on precessional time scales. Later on, due to a decrease in the magnitude of the precessional cycle and a stabilising effect of low CO2 concentration, the ice sheets remain large and grow consistently before reaching their maximum at around 20 kyr BP..

..From about 19 kyr BP, the ice sheets start to retreat with a maximum rate of sea level rise reaching some 15 m per 1000 years around 15kyrBP. The northern European ice sheets disappeared first, and the North American ice sheets completely disappeared at around 7 kyr BP. Fast sliding processes and the reduction of surface albedo due to deposition of dust play an important role in rapid deglaciation of the NH. Thus our results strongly support the idea about important role of aeolian dust in the termination of glacial cycles proposed earlier by Peltier and Marshall (1995)..

..Results from a set of sensitivity experiments demonstrate high sensitivity of simulated glacial cycle to the choice of some modelling parameters, and thus indicate the challenge to perform realistic simulations of glacial cycles with the computationally expensive models.

My summary – the simplifications of the EMIC combined with the “trying lots of parameters” approach means I have trouble putting much significance on the results.

While the basic setup, as described in the 2000 & 2001 papers seems reasonable, EMICs miss a lot of physics. This is important with something like starting and ending an ice age, where the feedbacks in higher resolution models can significantly reduce the effect seen by lower resolution models. When we run 100’s of simulations with different parameters (relating to the ice sheet) and find the best result I wonder what we’ve actually found.

That doesn’t mean they are of no value. Models help us to understand how the physics of climate actually works, because we can’t do these calculations in our heads. GCMs require too much computing resources to properly study ice ages.

So I look at EMICs as giving some useful insights that need to be validated with more complex models. Or with further study against other observations (what predictions do these parameter selections give us that can be verified?)

I don’t see them as demonstrating that the results “show” we’ve now modeled ice ages. The exact same comment also goes for another 2007 paper which used a GCM coupled to an ice sheet model that we covered in Part Nineteen – Ice Sheet Models I. An update of that paper in 2013 came with a excited Nature press release but to me simply demonstrates that with a few unknown parameters you can get a good result with some specific values of those parameters. This is not at all surprising. Let’s call it a good start.

Perhaps Abe Ouchi et al 2013 was the paper that will be verified as the answer to the question of ice age terminations – the delayed isostatic rebound.

Perhaps Ganopolski, Calov & Claussen 2010 with the interaction of dust on ice sheets will be verified as the answer to that question.

Perhaps neither will be.

Articles in this Series

Part One – An introduction

Part Two – Lorenz – one point of view from the exceptional E.N. Lorenz

Part Three – Hays, Imbrie & Shackleton – how everyone got onto the Milankovitch theory

Part Four – Understanding Orbits, Seasons and Stuff – how the wobbles and movements of the earth’s orbit affect incoming solar radiation

Part Five – Obliquity & Precession Changes – and in a bit more detail

Part Six – “Hypotheses Abound” – lots of different theories that confusingly go by the same name

Part Seven – GCM I – early work with climate models to try and get “perennial snow cover” at high latitudes to start an ice age around 116,000 years ago

Part Seven and a Half – Mindmap – my mind map at that time, with many of the papers I have been reviewing and categorizing plus key extracts from those papers

Part Eight – GCM II – more recent work from the “noughties” – GCM results plus EMIC (earth models of intermediate complexity) again trying to produce perennial snow cover

Part Nine – GCM III – very recent work from 2012, a full GCM, with reduced spatial resolution and speeding up external forcings by a factors of 10, modeling the last 120 kyrs

Part Ten – GCM IV – very recent work from 2012, a high resolution GCM called CCSM4, producing glacial inception at 115 kyrs

Pop Quiz: End of An Ice Age – a chance for people to test their ideas about whether solar insolation is the factor that ended the last ice age

Eleven – End of the Last Ice age – latest data showing relationship between Southern Hemisphere temperatures, global temperatures and CO2

Twelve – GCM V – Ice Age Termination – very recent work from He et al 2013, using a high resolution GCM (CCSM3) to analyze the end of the last ice age and the complex link between Antarctic and Greenland

Thirteen – Terminator II – looking at the date of Termination II, the end of the penultimate ice age – and implications for the cause of Termination II

Fourteen – Concepts & HD Data – getting a conceptual feel for the impacts of obliquity and precession, and some ice age datasets in high resolution

Fifteen – Roe vs Huybers – reviewing In Defence of Milankovitch, by Gerard Roe

Sixteen – Roe vs Huybers II – remapping a deep ocean core dataset and updating the previous article

Seventeen – Proxies under Water I – explaining the isotopic proxies and what they actually measure

Eighteen – “Probably Nonlinearity” of Unknown Origin – what is believed and what is put forward as evidence for the theory that ice age terminations were caused by orbital changes

Nineteen – Ice Sheet Models I – looking at the state of ice sheet models

References

CLIMBER-2: a climate system model of intermediate complexity. Part I: model description and performance for present climate, V Petoukhov, A Ganopolski, V Brovkin, M Claussen, A Eliseev, C Kubatzki & S Rahmstorf, Climate Dynamics (2000)

CLIMBER-2: a climate system model of intermediate complexity. Part II: model sensitivity, A Ganopolski, V Petoukhov, S Rahmstorf, V Brovkin, M Claussen, A Eliseev & C Kubatzki, Climate Dynamics 􏱄(2001)

Transient simulation of the last glacial inception. Part I: glacial inception as a bifurcation in the climate system, Reinhard Calov, Andrey Ganopolski, Martin Claussen, Vladimir Petoukhov & Ralf Greve, Climate Dynamics (2005)

Simulation of the last glacial cycle with a coupled climate ice-sheet model of intermediate complexity, A. Ganopolski, R. Calov, and M. Claussen, Climate of the Past (2010)

Read Full Post »

In Part Seven we had a look at a 2008 paper by Gettelman & Fu which assessed models vs measurements for water vapor in the upper troposphere.

In this article we will look at a 2010 paper by Chung, Yeomans & Soden. This paper studies outgoing longwave radiation (OLR) vs temperature change, for clear skies only, in three ways (and comparing models and measurements):

  • by region
  • by season
  • year to year

Why is this important and what is the approach all about?

Let’s suppose that the surface temperature increases for some reason. What happens to the total annual radiation emitted by the climate system? We expect it to increase. The hotter objects are the more they radiate.

If there is no positive feedback in the climate system then for a uniform global 1K (=1ºC) increase in surface & atmospheric temperature we expect the OLR to increase by 3.6 W/m². This is often called, by convention only, the “Planck feedback”. It refers to the fact that an increased surface temperature, and increased atmospheric temperature, will radiate more – and the “no feedback value” is 3.6 W/m² per 1K rise in temperature.

To explain a little further for newcomers.. with the concept of “no positive feedback” an initial 1K surface temperature rise – from any given cause – will stay at 1K. But if there is positive feedback in the climate system, an initial 1K surface temperature rise will result in a final temperature higher than 1K.

If the OLR increases by less than 3.6 W/m² the final temperature will end up higher than 1K – positive feedback. If the OLR increases by more than 3.6 W/m² the final temperature will end up lower than 1K – negative feedback.

Base Case

At the start of their paper they show the calculated clear-sky OLR change as the result of an ideal case. This is the change in OLR as a result of the surface and atmosphere increasing uniformly by 1K:

  • first, from the temperature change alone
  • second, from the change in water vapor as a result of this temperature change, assuming relative humidity stays constant
  • finally, from the first and second combined
From Chung et al (2010)

From Chung et al (2010)

Figure 1 – Click to expand

The graphs show the breakdown by pressure (=height) and latitude. 1000mbar is the surface and 200mbar is approximately the tropopause, the place where convection stops.

The sum of the first graph (note 1) is the “no feedback” response and equals 3.6 W/m². The sum of the second graph is the “feedback from water vapor” and equals -1.6 W/m². The combined result in the third graph equals 2.0 W/m². The second and third graphs are the result if relative humidity is constant.

We can also see that the tropics is where most of the changes take place.

They say:

One striking feature of the fixed-RH kernel is the small values in the tropical upper troposphere, where the positive OLR response to a temperature increase is offset by negative responses to the corresponding vapor increase. Thus under a constant RH- warming scenario, the tropical upper troposphere is in a runaway greenhouse state – the stabilizing effect of atmospheric warming is neutralized by the increased absorption from water vapor. Of course, the tropical upper troposphere is not isolated but is closely tied to the lower tropical troposphere where the combined temperature-water vapor responses are safely stabilizing.

To understand the first part of their statement, if temperatures increase and overall OLR does not increase at all then there is nothing to stop temperatures increasing. Of course, in practice, the “close to zero” increase in OLR for the tropical upper troposphere under a temperature rise can’t lead to any kind of runaway temperature increase. This is because there is a relationship between the temperatures in the upper troposphere and the lower- & mid- troposphere.

Relative Humidity Stays Constant?

Back in 1967, Manabe & Wetherald published their seminal paper which showed the result of increases in CO2 under two cases – with absolute humidity constant and with relative humidity constant:

Generally speaking, the sensitivity of the surface equilibrium temperature upon the change of various factors such as solar constant, cloudiness, surface albedo, and CO2 content are almost twice as much for the atmosphere with a given distribution of relative humidity as for that with a given distribution of absolute humidity..

..Doubling the existing CO2 content of the atmosphere has the effect of increasing the surface temperature by about 2.3ºC for the atmosphere with the realistic distribution of relative humidity and by about 1.3ºC for that with the realistic distribution of absolute humidity.

They explain important thinking about this topic:

Figure 1 shows the distribution of relative humidity as a function of latitude and height for summer and winter. According to this figure, the zonal mean distributions of relative humidity closely resemble one another, whereas those of absolute humidity do not. These data suggest that, given sufficient time, the atmosphere tends to restore a certain climatological distribution of relative humidity responding to the change of temperature.

It doesn’t mean that anyone should assume that relative humidity stays constant under a warmer world. It’s just likely to be a more realistic starting point than assuming that absolute humidity stays constant.

I only point this out for readers to understand that this idea is something that has seemed reasonable for almost 50 years. Of course, we have to question this “reasonable” assumption. How relative humidity changes as the climate warms or cools is a key factor in determining the water feedback and, therefore, it has had a lot of attention.

Results From the Paper

The observed rates of radiative damping from regional, seasonal, and interannual variations are substantially smaller than the rate of Planck radiative damping (3.6W/m²), yet slightly larger than that anticipated from a uniform warming, constant-RH response (2.0 W/m²).

The three comparison regressions can be seen, with ERBE data on the left and model results on the right:

From Chung et al (2010)

From Chung et al (2010)

Figure 2 – Click to expand

In the next figure, the differences between the models can be seen, and compared with ERBE and CERES results. The red “Planck” line is the no-feedback line, showing that (for these sets of results) models and experimental data show a positive feedback (when looking at clear sky OLR).

From Chung et al (2010)

From Chung et al (2010)

Figure 3 – Click to expand

Conclusion

At the least, we can see that climate models and measured values are quite close, when the results are aggregated. Both the model and the measured results are a long way from neutral feedback (the dashed slope in figure 2 and the red line in figure 3), instead they show positive feedback, quite close to what we would expect from constant relative humidity. The results indicate that relative humidity declines a little in the warmer case. The results also indicate that the models calculate a little more positive feedback than the real world measurements under these cases.

What does this mean for feedback from warming from increased GHGs? It’s the important question. We could say that the results tell us nothing, because how the world warms from increasing CO2 (and other GHGs) will change climate patterns and so seasonal, regional and year to year changes in periods from 1985-1988 and 2005-2008 are not particularly useful.

We could say that the results tell us that water vapor feedback is demonstrated to be a positive feedback, and matches quite closely the results of models. Or we could say that without cloudy sky data the results aren’t very interesting.

At the very least we can see that for current climate conditions under clear skies the change in OLR as temperature changes indicates an overall positive feedback, quite close to constant relative humidity results and quite close to what models calculate.

The ERBE results include the effect of a large El Nino and I do question whether year to year changes (graph c in figs 2 & 3) under El Nino to La Nino changes can be considered to represent how the climate might warm with more CO2. If we consider how the weather patterns shift during El-Nino to La Nina it has long been clear that there are positive feedbacks, but also the weather patterns end up back to normal (the cycle ends). I welcome knowledgeable readers explaining why El Nino feedback patters are relevant to future climate shifts, perhaps this will help me to clarify my thinking, or correct my misconceptions.

However, the CERES results from 2005-2008 don’t include the effect of a large El Nino and they show an overall slightly more positive feedback.

I asked Brian Soden a few question about this paper and he was kind enough to respond:

Q. Given the much better quality data since CERES and AIRS, why is ERBE data the focus?
A. At the time, the ERBE data was the only measurement that covered a large ENSO cycle (87/88 El Nino event followed by 88/89 La Nina)

Q. Why not include cloudy skies as well in this review? Collecting surface temperature data is more challenging of course because it needs a different data source. Is there a comparable study that you know of for cloudy skies?
A. The response of clouds to surface temperature changes is more complicated. We wanted to start with something relatively simple; i.e., water vapor. Andrew Dessler at Texas AM has a paper that came out a few years back that looks at total-sky fluxes and thus includes the effects on clouds.

Q. Do you know of any studies which have done similar work with what must now be over 10 years of CERES/AIRS.
A. Not off-hand. But it would be useful to do.

Articles in the Series

Part One – introducing some ideas from Ramanathan from ERBE 1985 – 1989 results

Part One – Responses – answering some questions about Part One

Part Two – some introductory ideas about water vapor including measurements

Part Three – effects of water vapor at different heights (non-linearity issues), problems of the 3d motion of air in the water vapor problem and some calculations over a few decades

Part Four – discussion and results of a paper by Dessler et al using the latest AIRS and CERES data to calculate current atmospheric and water vapor feedback vs height and surface temperature

Part Five – Back of the envelope calcs from Pierrehumbert – focusing on a 1995 paper by Pierrehumbert to show some basics about circulation within the tropics and how the drier subsiding regions of the circulation contribute to cooling the tropics

Part Six – Nonlinearity and Dry Atmospheres – demonstrating that different distributions of water vapor yet with the same mean can result in different radiation to space, and how this is important for drier regions like the sub-tropics

Part Seven – Upper Tropospheric Models & Measurement – recent measurements from AIRS showing upper tropospheric water vapor increases with surface temperature

Part Eight – Clear Sky Comparison of Models with ERBE and CERES – a paper from Chung et al (2010) showing clear sky OLR vs temperature vs models for a number of cases

Part Nine – Data I – Ts vs OLR – data from CERES on OLR compared with surface temperature from NCAR – and what we determine

Part Ten – Data II – Ts vs OLR – more on the data

References

An assessment of climate feedback processes using satellite observations of clear-sky OLR, Eui-Seok Chung, David Yeomans, & Brian J. Soden, GRL (2010) – free paper

Thermal equilibrium of the atmosphere with a given distribution of relative humidity, Manabe & Wetherald, Journal of the Atmospheric Sciences (1967) – free paper

Notes

Note 1: The values are per 100 mbar “slice” of the atmosphere. So if we want to calculate the total change we need to sum the values in each vertical slice, and of course, because they vary through latitude we need to average the values (area-weighted) across all latitudes.

Read Full Post »

In one stereotypical view of climate, the climate state has some variability over a 30 year period – we could call this multi-decadal variability “noise” – but it is otherwise fixed by the external conditions, the “external forcings”.

This doesn’t really match up with climate history, but climate models have mostly struggled to do much more than reproduce the stereotyped view. See Natural Variability and Chaos – Four – The Thirty Year Myth for a different perspective on (only) the timescale.

In this stereotypical view, the only reason why “long term” (=30 year statistics) can change is because of “external forcing”. Otherwise, where does the “extra energy” come from (we will examine this particular idea in a future article).

One of our commenters recently highlighted a paper from Drijfhout et al (2013) –Spontaneous abrupt climate change due to an atmospheric blocking–sea-ice–ocean feedback in an unforced climate model simulation.

Here is how the paper introduces the subject:

Abrupt climate change is abundant in geological records, but climate models rarely have been able to simulate such events in response to realistic forcing.

Here we report on a spontaneous abrupt cooling event, lasting for more than a century, with a temperature anomaly similar to that of the Little Ice Age. The event was simulated in the preindustrial control run of a high- resolution climate model, without imposing external perturbations.

This is interesting and instructive on many levels so let’s take a look. In later articles we will look at the evidence in climate history for “abrupt” events, for now note that Dansgaard–Oeschger (DO) events are the description of the originally identified form of abrupt change.

The distinction between “abrupt” changes and change that is not “abrupt” is an artificial one, it is more a reflection of the historical order in which we discovered “slow” and “abrupt” change. 

Under a Significance inset box in the paper:

There is a long-standing debate about whether climate models are able to simulate large, abrupt events that characterized past climates. Here, we document a large, spontaneously occurring cold event in a preindustrial control run of a new climate model.

The event is comparable to the Little Ice Age both in amplitude and duration; it is abrupt in its onset and termination, and it is characterized by a long period in which the atmospheric circulation over the North Atlantic is locked into a state with enhanced blocking.

To simulate this type of abrupt climate change, climate models should possess sufficient resolution to correctly represent atmospheric blocking and a sufficiently sensitive sea-ice model.

Here is their graph of the time-series of temperature (left) , and the geographical anomaly (right) expressed as the change during the 100 year event against the background of years 200-400:

From Drijfhout et al 2013

From Drijfhout et al 2013

Figure 1 – Click to expand

In their summary they state:

The lesson learned from this study is that the climate system is capable of generating large, abrupt climate excursions without externally imposed perturbations. Also, because such episodic events occur spontaneously, they may have limited predictability.

Before we look at the “causes” – the climate mechanisms – of this event, let’s briefly look at the climate model.

Their coupled GCM has an atmospheric resolution of just over 1º x 1º with 62 vertical levels, and the ocean has a resolution of 1º in the extra-tropics, increasing to 0.3º near the equator. The ocean has 42 vertical levels, with the top 200m of the ocean represented by 20 equally spaced 10m levels.

The GHGs and aerosols are set at pre-industrial 1860 values and don’t change over the 1,125 year simulation. There are no “flux adjustments” (no need for artificial momentum and energy additions to keep the model stable as with many older models).

See note 1 for a fuller description and the paper in the references for a full description.

The simulated event itself:

After 450 y, an abrupt cooling event occurred, with a clear signal in the Atlantic multidecadal oscillation (AMO). In the instrumental record, the amplitude of the AMO since the 1850s is about 0.4 °C, its SD 0.2 °C. During the event simulated here, the AMO index dropped by 0.8 °C for about a century..

How did this abrupt change take place?

The main mechanism was a change in the Atlantic Meridional Overturning Current (AMOC), also known as the Thermohaline circulation. The AMOC raises a nice example of the sensitivity of climate. The AMOC brings warmer water from the tropics into higher latitudes. A necessary driver of this process is the intensity of deep convection in high latitudes (sinking dense water) which in turn depends on two factors – temperature and salinity. More importantly, more accurately, it depends on the competing differences in anomalies of temperature and salinity

To shut down deep convection, the density of the surface water must decrease. In the temperature range of 7–12 °C, typical for the Labrador Sea, the SST anomaly in degrees Celsius has to be roughly 5 times the sea surface salinity (SSS) anomaly in practical salinity units for density compensation to occur. The SST anomaly was only about twice that of the SSS anomaly; the density anomaly was therefore mostly determined by the salinity anomaly.

In the figure below we see (left) the AMOC time series at two locations with the reduction during the cold century, and (right) the anomaly by depth and latitude for the “cold century” vs the climatology for years 200-400:

From Drijfhout et al 2013

From Drijfhout et al 2013

Figure 2 – Click to expand

What caused the lower salinities? It was more sea ice, melting in the right location. The excess sea ice was caused by positive feedback between atmospheric and ocean conditions “locking in” a particular pattern. The paper has a detailed explanation with graphics of the pressure anomalies which is hard to reduce to anything more succinct, apart from their abstract:

Initial cooling started with a period of enhanced atmospheric blocking over the eastern subpolar gyre.

In response, a southward progression of the sea-ice margin occurred, and the sea-level pressure anomaly was locked to the sea-ice margin through thermal forcing. The cold-core high steered more cold air to the area, reinforcing the sea-ice concentration anomaly east of Greenland.

The sea-ice surplus was carried southward by ocean currents around the tip of Greenland. South of 70°N, sea ice already started melting and the associated freshwater anomaly was carried to the Labrador Sea, shutting off deep convection. There, surface waters were exposed longer to atmospheric cooling and sea surface temperature dropped, causing an even larger thermally forced high above the Labrador Sea.

Conclusion

It is fascinating to see a climate model reproducing an example of abrupt climate change. There are a few contexts to suggest for this result.

1. From the context of timescale we could ask how often these events take place, or what pre-conditions are necessary. The only way to gather meaningful statistics is for large ensemble runs of considerable length – perhaps thousands of “perturbed physics” runs each of 100,000 years length. This is far out of reach for processing power at the moment. I picked some arbitrary numbers – until the statistics start to converge and match what we see from paleoclimatology studies we don’t know if we have covered the “terrain”.

Or perhaps only five runs of 1,000 years are needed to completely solve the problem (I’m kidding).

2. From the context of resolution – as we achieve higher resolution in models we may find new phenomena emerging in climate models that did not appear before. For example, in ice age studies, coarser climate models could not achieve “perennial snow cover” at high latitudes (as a pre-condition for ice age inception), but higher resolution climate models have achieved this first step. (See Ghosts of Climates Past – Part Seven – GCM I & Part Eight – GCM II).

As a comparison on resolution, the 2,000 year El Nino study we saw in Part Six of this series had an atmospheric resolution of 2.5º x 2.0º with 24 levels.

However, we might also find that as the resolution progressively increases (with the inevitable march of processing power) phenomena that appear at one resolution disappear at yet higher resolutions. This is an opinion, but if you ask people who have experience with computational fluid dynamics I expect they will say this would not be surprising.

3. Other models might reach similar or higher resolution and never get this kind of result and demonstrate the flaw in the EC-Earth model that allowed this “Little Ice Age” result to occur. Or the reverse.

As the authors say:

As a result, only coupled climate models that are capable of realistically simulating atmospheric blocking in relation to sea-ice variations feature the enhanced sensitivity to internal fluctuations that may temporarily drive the climate system to a state that is far beyond its standard range of natural variability.

Articles in the Series

Natural Variability and Chaos – One – Introduction

Natural Variability and Chaos – Two – Lorenz 1963

Natural Variability and Chaos – Three – Attribution & Fingerprints

Natural Variability and Chaos – Four – The Thirty Year Myth

Natural Variability and Chaos – Five – Why Should Observations match Models?

Natural Variability and Chaos – Six – El Nino

Natural Variability and Chaos – Seven – Attribution & Fingerprints Or Shadows?

Natural Variability and Chaos – Eight – Abrupt Change

References

Spontaneous abrupt climate change due to an atmospheric blocking–sea-ice–ocean feedback in an unforced climate model simulation, Sybren Drijfhout, Emily Gleeson, Henk A. Dijkstra & Valerie Livina, PNAS (2013) – free paper

EC-Earth V2.2: description and validation of a new seamless earth system prediction model, W. Hazeleger et al, Climate dynamics (2012) – free paper

Notes

Note 1: From the Supporting Information from their paper:

Climate Model and Numerical Simulation. The climate model used in this study is version 2.2 of the EC-Earth earth system model [see references] whose atmospheric component is based on cycle 31r1 of the European Centre for Medium-range Weather Forecasts (ECMWF) Integrated Forecasting System.

The atmospheric component runs at T159 horizontal spectral resolution (roughly 1.125°) and has 62 vertical levels. In the vertical a terrain-following mixed σ/pressure coordinate is used.

The Nucleus for European Modeling of the Ocean (NEMO), version V2, running in a tripolar configuration with a horizontal resolution of nominally 1° and equatorial refinement to 0.3° (2) is used for the ocean component of EC-Earth.

Vertical mixing is achieved by a turbulent kinetic energy scheme. The vertical z coordinate features a partial step implementation, and a bottom boundary scheme mixes dense water down bottom slopes. Tracer advection is accomplished by a positive definite scheme, which does not produce spurious negative values.

The model does not resolve eddies, but eddy-induced tracer advection is parameterized (3). The ocean is divided into 42 vertical levels, spaced by ∼10 m in the upper 200 m, and thereafter increasing with depth. NEMO incorporates the Louvain-la-Neuve sea-ice model LIM2 (4), which uses the same grid as the ocean model. LIM2 treats sea ice as a 2D viscous-plastic continuum that transmits stresses between the ocean and atmosphere. Thermodynamically it consists of a snow and an ice layer.

Heat storage, heat conduction, snow–ice transformation, nonuniform snow and ice distributions, and albedo are accounted for by subgrid-scale parameterizations.

The ocean, ice, land, and atmosphere are coupled through the Ocean, Atmosphere, Sea Ice, Soil 3 coupler (5). No flux adjustments are applied to the model, resulting in a physical consistency between surface fluxes and meridional transports.

The present preindustrial (PI) run was conducted by Met Éireann and comprised 1,125 y. The ocean was initialized from the World Ocean Atlas 2001 climatology (6). The atmosphere used the 40-year ECMWF Re-Analysis of January 1, 1979, as the initial state with permanent PI (1850) greenhouse gas (280 ppm) and aerosol concentrations.

Read Full Post »

In Part Three – Attribution & Fingerprints we looked at an early paper in this field, from 1996. I was led there by following back through many papers referenced from AR5 Chapter 10. The lead author of that paper, Gabriele Hegerl, has made a significant contribution to the 3rd report, 4th and 5th IPCC reports on attribution.

We saw in Part Three that this particular paper ascribed a probability:

We find that the latest observed 30-year trend pattern of near-surface temperature change can be distinguished from all estimates of natural climate variability with an estimated risk of less than 2.5% if the optimal fingerprint is applied.

That paper did note that greatest uncertainty was in understanding the magnitude of natural variability. This is an essential element of attribution.

It wasn’t explicitly stated whether the 97.5% confidence was with the premise that natural variability was accurately understood in 1996. I believe that this was the premise. I don’t know what confidence would have been ascribed to the attribution study if uncertainty over natural variability was included.

IPCC AR5

In this article we will look at the IPCC 5th report, AR5, and see how this field has progressed, specifically in regard to the understanding of natural variability. Chapter 10 covers Detection and Attribution of Climate Change.

From p.881 (the page numbers are from the start of the whole report, chapter 10 has just over 60 pages plus references):

Since the AR4, detection and attribution studies have been carried out using new model simulations with more realistic forcings, and new observational data sets with improved representation of uncertainty (Christidis et al., 2010; Jones et al., 2011, 2013; Gillett et al., 2012, 2013; Stott and Jones, 2012; Knutson et al., 2013; Ribes and Terray, 2013).

Let’s have a look at these papers (see note 1 on CMIP3 & CMIP5).

I had trouble understanding AR5 Chapter 10 because there was no explicit discussion of natural variability. The papers referenced (usually) have their own section on natural variability, but chapter 10 doesn’t actually cover it.

I emailed Geert Jan van Oldenborgh to ask for help. He is the author of one paper we will briefly look at here – his paper was very interesting and he had a video segment explaining his paper. He suggested the problem was more about communication because natural variability was covered in chapter 9 on models. He had written a section in chapter 11 that he pointed me towards, so this article became something that tried to grasp the essence of three chapters (9 – 11), over 200 pages of reports and several pallet loads of papers.

So I’m not sure I can do the synthesis justice, but what I will endeavor to do in this article is demonstrate the minimal focus (in IPCC AR5) on how well models represent natural variability.

That subject deserves a lot more attention, so this article will be less about what natural variability is, and more about how little focus it gets in AR5. I only arrived here because I was determined to understand “fingerprints” and especially the rationale behind the certainties ascribed.

Subsequent articles will continue the discussion on natural variability.

Knutson et al 2013

The models [CMIP5] are found to provide plausible representations of internal climate variability, although there is room for improvement..

..The modeled internal climate variability from long control runs is used to determine whether observed and simulated trends are consistent or inconsistent. In other words, we assess whether observed and simulated forced trends are more extreme than those that might be expected from random sampling of internal climate variability.

Later

The model control runs exhibit long-term drifts. The magnitudes of these drifts tend to be larger in the CMIP3 control runs than in the CMIP5 control runs, although there are exceptions. We assume that these drifts are due to the models not being in equilibrium with the control run forcing, and we remove the drifts by a linear trend analysis (depicted by the orange straight lines in Fig. 1). In some CMIP3 cases, the drift initially proceeds at one rate, but then the trend becomes smaller for the remainder of the run. We approximate the drift in these cases by two separate linear trend segments, which are identified in the figure by the short vertical orange line segments. These long-term drift trends are removed to produce the drift corrected series.

[Emphasis added].

Another paper suggests this assumption might not be correct. Here is Jones, Stott and Christidis (2013) – “piControl” are the natural variability model simulations:

Often a model simulation with no changes in external forcing (piControl) will have a drift in the climate diagnostics due to various flux imbalances in the model [Gupta et al., 2012]. Some studies attempt to account for possible model climate drifts, for instance Figure 9.5 in Hegerl et al. [2007] did not include transient simulations of the 20th century if the long-term trend of the piControl was greater in magnitude than 0.2 K/century (Appendix 9.C in Hegerl et al. [2007]).

Another technique is to remove the trend, from the transient simulations, deduced from a parallel section of piControl [e.g., Knutson et al., 2006]. However whether one should always remove the piControl trend, and how to do it in practice, is not a trivial issue [Taylor et al., 2012; Gupta et al., 2012]..

..We choose not to remove the trend from the piControl from parallel simulations of the same model in this study due to the impact it would have on long-term variability, i.e., the possibility that part of the trend in the piControl may be long-term internal variability that may or may not happen in a parallel experiment when additional forcing has been applied.

Here are further comments from Knutson et al 2013:

Five of the 24 CMIP3 models, identified by “(-)” in Fig. 1, were not used, or practically not used, beyond Fig. 1 in our analysis. For instance, the IAP_fgoals1.0.g model has a strong discontinuity near year 200 of the control run. We judge this as likely an artifact due to some problem with the model simulation, and we therefore chose to exclude this model from further analysis

From Knutson et al 2013

From Knutson et al 2013

Figure 1

Perhaps this is correct. Or perhaps the jump in simulated temperature is the climate model capturing natural climate variability.

The authors do comment:

As noted by Wittenberg (2009) and Vecchi and Wittenberg (2010), long-running control runs suggest that internally generated SST variability, at least in the ENSO region, can vary substantially between different 100-yr periods (approximately the length of record used here for observations), which again emphasizes the caution that must be placed on comparisons of modeled vs. observed internal variability based on records of relatively limited duration.

The first paper referenced, Wittenberg 2009, was the paper we looked at in Part Six – El Nino.

So is the “caution” that comes from that study included in the probability of our models ability to simulate natural variability?

In reality, questions about internal variability are not really discussed. Trends are removed, models with discontinuities are artifacts. What is left? This paper essentially takes the modeling output from the CMIP3 and CMIP5 archives (with and without GHG forcing) as a given and applies some tests.

Ribes & Terray 2013

This was a “Part II” paper and they said:

We use the same estimates of internal variability as in Ribes et al. 2013 [the “Part I”].

These are based on intra-ensemble variability from the above CMIP5 experiments as well as pre-industrial simulations from both the CMIP3 and CMIP5 archives, leading to a much larger sample than previously used (see Ribes et al. 2013 for details about ensembles). We then implicitly assume that the multi-model internal variability estimate is reliable.

[Emphasis added]. The Part I paper said:

An estimate of internal climate variability is required in detection and attribution analysis, for both optimal estimation of the scaling factors and uncertainty analysis.

Estimates of internal variability are usually based on climate simulations, which may be control simulations (i.e. in the present case, simulations with no variations in external forcings), or ensembles of simulations with the same prescribed external forcings.

In the latter case, m – 1 independent realisations of pure internal variability may be obtained by subtracting the ensemble mean from each member (assuming again additivity of the responses) and rescaling the result by a factor √(m/(m-1)) , where m denotes the number of members in the ensemble.

Note that estimation of internal variability usually means estimation of the covariance matrix of a spatio-temporal climate-vector, the dimension of this matrix potentially being high. We choose to use a multi-model estimate of internal climate variability, derived from a large ensemble of climate models and simulations. This multi-model estimate is subject to lower sampling variability and better represents the effects of model uncertainty on the estimate of internal variability than individual model estimates. We then simultaneously consider control simulations from the CMIP3 and CMIP5 archives, and ensembles of historical simulations (including simulations with individual sets of forcings) from the CMIP5 archive.

All control simulations longer than 220 years (i.e. twice the length of our study period) and all ensembles (at least 2 members) are used. The overall drift of control simulations is removed by subtracting a linear trend over the full period.. We then implicitly assume that this multi- model internal variability estimate is reliable.

[Emphasis added]. So two approaches to evaluate internal variability – one approach uses GCM runs with no GHG forcing; and the other approach uses the variation between different runs of the same model (with GHG forcing) to estimate natural variability. Drift is removed as “an error”.

Chapter 10 on Spatial Trends

The IPCC report also reviews the spatial simulations compared with spatial observations, p. 880:

Figure 10.2a shows the pattern of annual mean surface temperature trends observed over the period 1901–2010, based on Hadley Centre/ Climatic Research Unit gridded surface temperature data set 4 (Had- CRUT4). Warming has been observed at almost all locations with sufficient observations available since 1901.

Rates of warming are generally higher over land areas compared to oceans, as is also apparent over the 1951–2010 period (Figure 10.2c), which simulations indicate is due mainly to differences in local feedbacks and a net anomalous heat transport from oceans to land under GHG forcing, rather than differences in thermal inertia (e.g., Boer, 2011). Figure 10.2e demonstrates that a similar pattern of warming is simulated in the CMIP5 simulations with natural and anthropogenic forcing over the 1901–2010 period. Over most regions, observed trends fall between the 5th and 95th percentiles of simulated trends, and van Oldenborgh et al. (2013) find that over the 1950–2011 period the pattern of observed grid cell trends agrees with CMIP5 simulated trends to within a combination of model spread and internal variability..

van Oldenborgh et al (2013)

Let’s take a look at van Oldenborgh et al (2013).

There’s a nice video of (I assume) the lead author talking about the paper and comparing the probabilistic approach used in weather forecasts with that of climate models (see Ensemble Forecasting). I recommend the video for a good introduction to the topic of ensemble forecasting.

With weather forecasting the probability comes from running ensembles of weather models and seeing, for example, how many simulations predict rain vs how many do not. The proportion is the probability of rain. With weather forecasting we can continually review how well the probabilities given by ensembles match the reality. Over time we will build up a set of statistics of “probability of rain” and compare with the frequency of actual rainfall. It’s pretty easy to see if the models are over-confident or under-confident.

Here is what the authors say about the problem and how they approached it:

The ensemble is considered to be an estimate of the probability density function (PDF) of a climate forecast. This is the method used in weather and seasonal forecasting (Palmer et al 2008). Just like in these fields it is vital to verify that the resulting forecasts are reliable in the definition that the forecast probability should be equal to the observed probability (Joliffe and Stephenson 2011).

If outcomes in the tail of the PDF occur more (less) frequently than forecast the system is overconfident (underconfident): the ensemble spread is not large enough (too large).

In contrast to weather and seasonal forecasts, there is no set of hindcasts to ascertain the reliability of past climate trends per region. We therefore perform the verification study spatially, comparing the forecast and observed trends over the Earth. Climate change is now so strong that the effects can be observed locally in many regions of the world, making a verification study on the trends feasible. Spatial reliability does not imply temporal reliability, but unreliability does imply that at least in some areas the forecasts are unreliable in time as well. In the remainder of this letter we use the word ‘reliability’ to indicate spatial reliability.

[Emphasis added]. The paper first shows the result for one location, the Netherlands, with the spread of model results vs the actual result from 1950-2011:

from van Oldenborgh et al 2013

from van Oldenborgh et al 2013

Figure 2

We can see that the models are overall mostly below the observation. But this is one data point. So if we compared all of the datapoints – and this is on a grid of 2.5º – how do the model spreads compare with the results? Are observations above 95% of the model results only 5% of the time? Or more than 5% of the time? And are observations below 5% of the model results only 5% of the time?

We can see that the frequency of observations in the bottom 5% of model results is about 13% and the frequency of observations in the top 5% of model results is about 20%. Therefore the models are “overconfident” in spatial representation of the last 60 year trends:

van Oldenborgh-2013-fig3

From van Oldenborgh et al 2013

Figure 3

We investigated the reliability of trends in the CMIP5 multi-model ensemble prepared for the IPCC AR5. In agreement with earlier studies using the older CMIP3 ensemble, the temperature trends are found to be locally reliable. However, this is due to the differing global mean climate response rather than a correct representation of the spatial variability of the climate change signal up to now: when normalized by the global mean temperature the ensemble is overconfident. This agrees with results of Sakaguchi et al (2012) that the spatial variability in the pattern of warming is too small. The precipitation trends are also overconfident. There are large areas where trends in both observational dataset are (almost) outside the CMIP5 ensemble, leading us to conclude that this is unlikely due to faulty observations.

It’s probably important to note that the author comments in the video “on the larger scale the models are not doing so badly”.

It’s an interesting paper. I’m not clear whether the brief note in AR5 reflects the paper’s conclusions.

Jones et al 2013

It was reassuring to finally find a statement that confirmed what seemed obvious from the “omissions”:

A basic assumption of the optimal detection analysis is that the estimate of internal variability used is comparable with the real world’s internal variability.

Surely I can’t be the only one reading Chapter 10 and trying to understand the assumptions built into the “with 95% confidence” result. If Chapter 10 is only aimed at climate scientists who work in the field of attribution and detection it is probably fine not to actually mention this minor detail in the tight constraints of only 60 pages.

But if Chapter 10 is aimed at a wider audience it seems a little remiss not to bring it up in the chapter itself.

I probably missed the stated caveat in chapter 10’s executive summary or introduction.

The authors continue:

As the observations are influenced by external forcing, and we do not have a non-externally forced alternative reality to use to test this assumption, an alternative common method is to compare the power spectral density (PSD) of the observations with the model simulations that include external forcings.

We have already seen that overall the CMIP5 and CMIP3 model variability compares favorably across different periodicities with HadCRUT4-observed variability (Figure 5). Figure S11 (in the supporting information) includes the PSDs for each of the eight models (BCC-CSM1-1, CNRM-CM5, CSIRO- Mk3-6-0, CanESM2, GISS-E2-H, GISS-E2-R, HadGEM2- ES and NorESM1-M) that can be examined in the detection analysis.

Variability for the historical experiment in most of the models compares favorably with HadCRUT4 over the range of periodicities, except for HadGEM2-ES whose very long period variability is lower due to the lower overall trend than observed and for CanESM2 and bcc-cm1-1 whose decadal and higher period variability are larger than observed.

While not a strict test, Figure S11 suggests that the models have an adequate representation of internal variability—at least on the global mean level. In addition, we use the residual test from the regression to test whether there are any gross failings in the models representation of internal variability.

Figure S11 is in the supplementary section of the paper:

From Jones et al 2013, figure S11

From Jones et al 2013, figure S11

Figure 4

From what I can see, this demonstrates that the spectrum of the models’ internal variability (“historicalNat”) is different from the spectrum of the models’ forced response with GHG changes (“historical”).

It feels like my quantum mechanics classes all over again. I’m probably missing something obvious, and hopefully knowledgeable readers can explain.

Chapter 9 of AR5 – Climate Models’ Representation of Internal Variability

Chapter 9, reviewing models, stretches to over 80 pages. The section on internal variability is section 9.5.1:

However, the ability to simulate climate variability, both unforced internal variability and forced variability (e.g., diurnal and seasonal cycles) is also important. This has implications for the signal-to-noise estimates inherent in climate change detection and attribution studies where low-frequency climate variability must be estimated, at least in part, from long control integrations of climate models (Section 10.2).

Section 9.5.3:

In addition to the annual, intra-seasonal and diurnal cycles described above, a number of other modes of variability arise on multi-annual to multi-decadal time scales (see also Box 2.5). Most of these modes have a particular regional manifestation whose amplitude can be larger than that of human-induced climate change. The observational record is usually too short to fully evaluate the representation of variability in models and this motivates the use of reanalysis or proxies, even though these have their own limitations.

Figure 9.33a shows simulated internal variability of mean surface temperature from CMIP5 pre-industrial control simulations. Model spread is largest in the tropics and mid to high latitudes (Jones et al., 2012), where variability is also large; however, compared to CMIP3, the spread is smaller in the tropics owing to improved representation of ENSO variability (Jones et al., 2012). The power spectral density of global mean temperature variance in the historical simulations is shown in Figure 9.33b and is generally consistent with the observational estimates. At longer time scale of the spectra estimated from last millennium simulations, performed with a subset of the CMIP5 models, can be assessed by comparison with different NH temperature proxy records (Figure 9.33c; see Chapter 5 for details). The CMIP5 millennium simulations include natural and anthropogenic forcings (solar, volcanic, GHGs, land use) (Schmidt et al., 2012).

Significant differences between unforced and forced simulations are seen for time scale larger than 50 years, indicating the importance of forced variability at these time scales (Fernandez-Donado et al., 2013). It should be noted that a few models exhibit slow background climate drift which increases the spread in variance estimates at multi-century time scales.

Nevertheless, the lines of evidence above suggest with high confidence that models reproduce global and NH temperature variability on a wide range of time scales.

[Emphasis added]. Here is fig 9.33:

From IPCC AR5 Chapter 10

From IPCC AR5 Chapter 10

Figure 5 – Click to Expand

The bottom graph shows the spectra of the last 1,000 years – black line is observations (reconstructed from proxies), dashed lines are without GHG forcings, and solid lines are with GHG forcings.

In later articles we will review this in more detail.

Conclusion

The IPCC report on attribution is very interesting. Most attribution studies compare observations of the last 100 – 150 years with model simulations using anthropogenic GHG changes and model simulations without (note 3).

The results show a much better match for the case of the anthropogenic forcing.

The primary method is with global mean surface temperature, with more recent studies also comparing the spatial breakdown. We saw one such comparison with van Oldenborgh et al (2013). Jones et al (2013) also reviews spatial matching, finding a better fit (of models & observations) for the last half of the 20th century than the first half. (As with van Oldenborgh’s paper, the % match outside 90% of model results was greater than 10%).

My question as I first read Chapter 10 was how was the high confidence attained and what is a fingerprint?

I was led back, by following the chain of references, to one of the early papers on the topic (1996) that also had similar high confidence. (We saw this in Part Three). It was intriguing that such confidence could be attained with just a few “no forcing” model runs as comparison, all of which needed “flux adjustment”. Current models need much less, or often zero, flux adjustment.

In later papers reviewed in AR5, “no forcing” model simulations that show temperature trends or jumps are often removed or adjusted.

I’m not trying to suggest that “no forcing” GCM simulations of the last 150 years have anything like the temperature changes we have observed. They don’t.

But I was trying to understand what assumptions and premises were involved in attribution. Chapter 10 of AR5 has been valuable in suggesting references to read, but poor at laying out the assumptions and premises of attribution studies.

For clarity, as I stated in Part Three:

..as regular readers know I am fully convinced that the increases in CO2, CH4 and other GHGs over the past 100 years or more can be very well quantified into “radiative forcing” and am 100% in agreement with the IPCCs summary of the work of atmospheric physics over the last 50 years on this topic. That is, the increases in GHGs have led to something like a “radiative forcing” of 2.8 W/m²..

..Therefore, it’s “very likely” that the increases in GHGs over the last 100 years have contributed significantly to the temperature changes that we have seen.

So what’s my point?

Chapter 10 of the IPCC report fails to highlight the important assumptions in the attribution studies. Chapter 9 of the IPCC report has a section on centennial/millennial natural variability with a “high confidence” conclusion that comes with little evidence and appears to be based on a cursory comparison of the spectral results of the last 1,000 years proxy results with the CMIP5 modeling studies.

In chapter 10, the executive summary states:

..given that observed warming since 1951 is very large compared to climate model estimates of internal variability (Section 10.3.1.1.2), which are assessed to be adequate at global scale (Section 9.5.3.1), we conclude that it is virtually certain [99-100%] that internal variability alone cannot account for the observed global warming since 1951.

[Emphasis added]. I agree, and I don’t think anyone who understands radiative forcing and climate basics would disagree. To claim otherwise would be as ridiculous as, for example, claiming that tiny changes in solar insolation from eccentricity modifications over 100 kyrs cause the end of ice ages, whereas large temperature changes during these ice ages have no effect (see note 2).

The executive summary also says:

It is extremely likely [95–100%] that human activities caused more than half of the observed increase in GMST from 1951 to 2010.

The idea is plausible, but the confidence level is dependent on a premise that is claimed via one graph (fig 9.33) of the spectrum of the last 1,000 years. High confidence (“that models reproduce global and NH temperature variability on a wide range of time scales”) is just an opinion.

It’s crystal clear, by inspection of CMIP3 and CMIP5 model results, that models with anthropogenic forcing match the last 150 years of temperature changes much better than models held at constant pre-industrial forcing.

I believe natural variability is a difficult subject which needs a lot more than a cursory graph of the spectrum of the last 1,000 years to even achieve low confidence in our understanding.

Chapters 9 & 10 of AR5 haven’t investigated “natural variability” at all. For interest, some skeptic opinions are given in note 4.

I propose an alternative summary for Chapter 10 of AR5:

It is extremely likely [95–100%] that human activities caused more than half of the observed increase in GMST from 1951 to 2010, but this assessment is subject to considerable uncertainties.

Articles in the Series

Natural Variability and Chaos – One – Introduction

Natural Variability and Chaos – Two – Lorenz 1963

Natural Variability and Chaos – Three – Attribution & Fingerprints

Natural Variability and Chaos – Four – The Thirty Year Myth

Natural Variability and Chaos – Five – Why Should Observations match Models?

Natural Variability and Chaos – Six – El Nino

Natural Variability and Chaos – Seven – Attribution & Fingerprints Or Shadows?

Natural Variability and Chaos – Eight – Abrupt Change

References

Multi-model assessment of regional surface temperature trends, TR Knutson, F Zeng & AT Wittenberg, Journal of Climate (2013) – free paper

Attribution of observed historical near surface temperature variations to anthropogenic and natural causes using CMIP5 simulations, Gareth S Jones, Peter A Stott & Nikolaos Christidis, Journal of Geophysical Research Atmospheres (2013) – paywall paper

Application of regularised optimal fingerprinting to attribution. Part II: application to global near-surface temperature, Aurélien Ribes & Laurent Terray, Climate Dynamics (2013) – free paper

Application of regularised optimal fingerprinting to attribution. Part I: method, properties and idealised analysis, Aurélien Ribes, Serge Planton & Laurent Terray, Climate Dynamics (2013) – free paper

Reliability of regional climate model trends, GJ van Oldenborgh, FJ Doblas Reyes, SS Drijfhout & E Hawkins, Environmental Research Letters (2013) – free paper

Notes

Note 1: CMIP = Coupled Model Intercomparison Project. CMIP3 was for AR4 and CMIP5 was for AR5.

Read about CMIP5:

At a September 2008 meeting involving 20 climate modeling groups from around the world, the WCRP’s Working Group on Coupled Modelling (WGCM), with input from the IGBP AIMES project, agreed to promote a new set of coordinated climate model experiments. These experiments comprise the fifth phase of the Coupled Model Intercomparison Project (CMIP5). CMIP5 will notably provide a multi-model context for

1) assessing the mechanisms responsible for model differences in poorly understood feedbacks associated with the carbon cycle and with clouds

2) examining climate “predictability” and exploring the ability of models to predict climate on decadal time scales, and, more generally

3) determining why similarly forced models produce a range of responses…

From the website link above you can read more. CMIP5 is a substantial undertaking, with massive output of data from the latest climate models. Anyone can access this data, similar to CMIP3. Here is the Getting Started page.

And CMIP3:

In response to a proposed activity of the World Climate Research Programme (WCRP) Working Group on Coupled Modelling (WGCM), PCMDI volunteered to collect model output contributed by leading modeling centers around the world. Climate model output from simulations of the past, present and future climate was collected by PCMDI mostly during the years 2005 and 2006, and this archived data constitutes phase 3 of the Coupled Model Intercomparison Project (CMIP3). In part, the WGCM organized this activity to enable those outside the major modeling centers to perform research of relevance to climate scientists preparing the Fourth Asssessment Report (AR4) of the Intergovernmental Panel on Climate Change (IPCC). The IPCC was established by the World Meteorological Organization and the United Nations Environmental Program to assess scientific information on climate change. The IPCC publishes reports that summarize the state of the science.

This unprecedented collection of recent model output is officially known as the “WCRP CMIP3 multi-model dataset.” It is meant to serve IPCC’s Working Group 1, which focuses on the physical climate system — atmosphere, land surface, ocean and sea ice — and the choice of variables archived at the PCMDI reflects this focus. A more comprehensive set of output for a given model may be available from the modeling center that produced it.

With the consent of participating climate modelling groups, the WGCM has declared the CMIP3 multi-model dataset open and free for non-commercial purposes. After registering and agreeing to the “terms of use,” anyone can now obtain model output via the ESG data portal, ftp, or the OPeNDAP server.

As of July 2009, over 36 terabytes of data were in the archive and over 536 terabytes of data had been downloaded among the more than 2500 registered users

Note 2: This idea is explained in Ghosts of Climates Past -Eighteen – “Probably Nonlinearity” of Unknown Origin – what is believed and what is put forward as evidence for the theory that ice age terminations were caused by orbital changes, see especially the section under the heading: Why Theory B is Unsupportable.

Note 3: Some studies use just fixed pre-industrial values, and others compare “natural forcings” with “no forcings”.

“Natural forcings” = radiative changes due to solar insolation variations (which are not known with much confidence) and aerosols from volcanos. “No forcings” is simply fixed pre-industrial values.

Note 4: Chapter 11 (of AR5), p.982:

For the remaining projections in this chapter the spread among the CMIP5 models is used as a simple, but crude, measure of uncertainty. The extent of agreement between the CMIP5 projections provides rough guidance about the likelihood of a particular outcome. But—as partly illustrated by the discussion above—it must be kept firmly in mind that the real world could fall outside of the range spanned by these particular models. See Section 11.3.6 for further discussion.

And p. 1004:

It is possible that the real world might follow a path outside (above or below) the range projected by the CMIP5 models. Such an eventuality could arise if there are processes operating in the real world that are missing from, or inadequately represented in, the models. Two main possibilities must be considered: (1) Future radiative and other forcings may diverge from the RCP4.5 scenario and, more generally, could fall outside the range of all the RCP scenarios; (2) The response of the real climate system to radiative and other forcing may differ from that projected by the CMIP5 models. A third possibility is that internal fluctuations in the real climate system are inadequately simulated in the models. The fidelity of the CMIP5 models in simulating internal climate variability is discussed in Chapter 9..

..The response of the climate system to radiative and other forcing is influenced by a very wide range of processes, not all of which are adequately simulated in the CMIP5 models (Chapter 9). Of particular concern for projections are mechanisms that could lead to major ‘surprises’ such as an abrupt or rapid change that affects global-to-continental scale climate.

Several such mechanisms are discussed in this assessment report; these include: rapid changes in the Arctic (Section 11.3.4 and Chapter 12), rapid changes in the ocean’s overturning circulation (Chapter 12), rapid change of ice sheets (Chapter 13) and rapid changes in regional monsoon systems and hydrological climate (Chapter 14). Additional mechanisms may also exist as synthesized in Chapter 12. These mechanisms have the potential to influence climate in the near term as well as in the long term, albeit the likelihood of substantial impacts increases with global warming and is generally lower for the near term.

And p. 1009 (note that we looked at Rowlands et al 2012 in Part Five – Why Should Observations match Models?):

The CMIP3 and CMIP5 projections are ensembles of opportunity, and it is explicitly recognized that there are sources of uncertainty not simulated by the models. Evidence of this can be seen by comparing the Rowlands et al. (2012) projections for the A1B scenario, which were obtained using a very large ensemble in which the physics parameterizations were perturbed in a single climate model, with the corresponding raw multi-model CMIP3 projections. The former exhibit a substantially larger likely range than the latter. A pragmatic approach to addressing this issue, which was used in the AR4 and is also used in Chapter 12, is to consider the 5 to 95% CMIP3/5 range as a ‘likely’ rather than ‘very likely’ range.

Replacing ‘very likely’ = 90–100% with ‘likely 66–100%’ is a good start. How does this recast chapter 10?

And Chapter 1 of AR5, p. 138:

Model spread is often used as a measure of climate response uncertainty, but such a measure is crude as it takes no account of factors such as model quality (Chapter 9) or model independence (e.g., Masson and Knutti, 2011; Pennell and Reichler, 2011), and not all variables of interest are adequately simulated by global climate models..

..Climate varies naturally on nearly all time and space scales, and quantifying precisely the nature of this variability is challenging, and is characterized by considerable uncertainty.

[Emphasis added in all bold sections above]

Read Full Post »

In (still) writing what was to be Part Six (Attribution in AR5 from the IPCC), I was working through Knutson et al 2013, one of the papers referenced by AR5. That paper in turn referenced Are historical records sufficient to constrain ENSO simulations? [link corrected] by Andrew Wittenberg (2009). This is a very interesting paper and I was glad to find it because it illustrates some of the points we have been looking at.

It’s an easy paper to read (and free) and so I recommend reading the whole paper.

The paper uses NOAA’s Geophysical Fluid Dynamics Laboratory (GFDL) CM2.1 global coupled atmosphere/ocean/land/ice GCM (see note 1 for reference and description):

CM2.1 played a prominent role in the third Coupled Model Intercomparison Project (CMIP3) and the Fourth Assessment of the Intergovernmental Panel on Climate Change (IPCC), and its tropical and ENSO simulations have consistently ranked among the world’s top GCMs [van Oldenborgh et al., 2005; Wittenberg et al., 2006; Guilyardi, 2006; Reichler and Kim, 2008].

The coupled pre-industrial control run is initialized as by Delworth et al. [2006], and then integrated for 2220 yr with fixed 1860 estimates of solar irradiance, land cover, and atmospheric composition; we focus here on just the last 2000 yr. This simulation required one full year to run on 60 processors at GFDL.

First of all we see the challenge for climate models – a reasonable resolution coupled GCM running just one 2000-year simulation consumed one year of multiple processor time.

Wittenberg shows the results in the graph below. At the top is our observational record going back 140 years, then below are the simulation results of the SST variation in the El Nino region broken into 20 century-long segments.

From Wittenberg 2009

From Wittenberg 2009

 Figure 1 – Click to Expand

What we see is that different centuries have very different results:

There are multidecadal epochs with hardly any variability (M5); epochs with intense, warm-skewed ENSO events spaced five or more years apart (M7); epochs with moderate, nearly sinusoidal ENSO events spaced three years apart (M2); and epochs that are highly irregular in amplitude and period (M6). Occasional epochs even mimic detailed temporal sequences of observed ENSO events; e.g., in both R2 and M6, there are decades of weak, biennial oscillations, followed by a large warm event, then several smaller events, another large warm event, and then a long quiet period. Although the model’s NINO3 SST variations are generally stronger than observed, there are long epochs (like M1) where the ENSO amplitude agrees well with observations (R1).

Wittenberg comments on the problem for climate modelers:

An unlucky modeler – who by chance had witnessed only M1-like variability throughout the first century of simulation – might have erroneously inferred that the model’s ENSO amplitude matched observations, when a longer simulation would have revealed a much stronger ENSO.

If the real-world ENSO is similarly modulated, then there is a more disturbing possibility. Had the research community been unlucky enough to observe an unrepresentative ENSO over the past 150 yr of measurements, then it might collectively have misjudged ENSO’s longer-term natural behavior. In that case, historically-observed statistics could be a poor guide for modelers, and observed trends in ENSO statistics might simply reflect natural variations..

..A 200 yr epoch of consistently strong variability (M3) can be followed, just one century later, by a 200 yr epoch of weak variability (M4). Documenting such extremes might thus require a 500+ yr record. Yet few modeling centers currently attempt simulations of that length when evaluating CGCMs under development – due to competing demands for high resolution, process completeness, and quick turnaround to permit exploration of model sensitivities.

Model developers thus might not even realize that a simulation manifested long-term ENSO modulation, until long after freezing the model development. Clearly this could hinder progress. An unlucky modeler – unaware of centennial ENSO modulation and misled by comparisons between short, unrepresentative model runs – might erroneously accept a degraded model or reject an improved model.

[Emphasis added].

Wittenberg shows the same data in the frequency domain and has presented the data in a way that illustrates the different perspective you might have depending upon your period of observation or period of model run. It’s worth taking the time to understand what is in these graphs:

From Wittenberg 2009

From Wittenberg 2009

Figure 2 – Click to Expand

The first graph, 2a:

..time-mean spectra of the observations for epochs of length 20 yr – roughly the duration of observations from satellites and the Tropical Atmosphere Ocean (TAO) buoy array. The spectral power is fairly evenly divided between the seasonal cycle and the interannual ENSO band, the latter spanning a broad range of time scales between 1.3 to 8 yr.

So the different colored lines indicate the spectral power for each period. The black dashed line is the observed spectral power over the 140 year (observational) period. This dashed line is repeated in figure 2c.

The second graph, 2b shows the modeled results if we break up the 2000 years into 100 x 20-year periods.

The third graph, 2c, shows the modeled results broken up into 100 year periods. The probability number in the bottom right, 90%, is the likelihood of observations falling outside the range of the model results – if “the simulated subspectra independent and identically distributed.. at bottom right is the probability that an interval so constructed would bracket the next subspectrum to emerge from the model.

So what this says, paraphrasing and over-simplifying: “we are 90% sure that the observations can’t be explained by the models”.

Of course, this independent and identically distributed assumption is not valid, but as we will hopefully get onto many articles further in this series, most of these statistical assumptions – stationary, gaussian, AR1 – are problematic for real world non-linear systems.

To be clear, the paper’s author is demonstrating a problem in such a statistical approach.

Conclusion

Models are not reality. This is a simulation with the GFDL model. It doesn’t mean ENSO is like this. But it might be.

The paper illustrates a problem I highlighted in Part Five – observations are only one “realization” of possible outcomes. The last century or century and a half of surface observations could be an outlier. The last 30 years of satellite data could equally be an outlier. Even if our observational periods are not an outlier and are right there on the mean or median, matching climate models to observations may still greatly under-represent natural climate variability.

Non-linear systems can demonstrate variability over much longer time-scales than the the typical period between characteristic events. We will return to this in future articles in more detail. Such systems do not have to be “chaotic” (where chaotic means that tiny changes in initial conditions cause rapidly diverging results).

What period of time is necessary to capture natural climate variability?

I will give the last word to the paper’s author:

More worryingly, if nature’s ENSO is similarly modulated, there is no guarantee that the 150 yr historical SST record is a fully representative target for model development..

..In any case, it is sobering to think that even absent any anthropogenic changes, the future of ENSO could look very different from what we have seen so far.

Articles in the Series

Natural Variability and Chaos – One – Introduction

Natural Variability and Chaos – Two – Lorenz 1963

Natural Variability and Chaos – Three – Attribution & Fingerprints

Natural Variability and Chaos – Four – The Thirty Year Myth

Natural Variability and Chaos – Five – Why Should Observations match Models?

Natural Variability and Chaos – Six – El Nino

Natural Variability and Chaos – Seven – Attribution & Fingerprints Or Shadows?

Natural Variability and Chaos – Eight – Abrupt Change

References

Are historical records sufficient to constrain ENSO simulations? Andrew T. Wittenberg, GRL (2009) – free paper

GFDL’s CM2 Global Coupled Climate Models. Part I: Formulation and Simulation Characteristics, Delworth et al, Journal of Climate, 2006 – free paper

Notes

Note 1: The paper referenced for the GFDL model is GFDL’s CM2 Global Coupled Climate Models. Part I: Formulation and Simulation Characteristics, Delworth et al, 2006:

The formulation and simulation characteristics of two new global coupled climate models developed at NOAA’s Geophysical Fluid Dynamics Laboratory (GFDL) are described.

The models were designed to simulate atmospheric and oceanic climate and variability from the diurnal time scale through multicentury climate change, given our computational constraints. In particular, an important goal was to use the same model for both experimental seasonal to interannual forecasting and the study of multicentury global climate change, and this goal has been achieved.

Two versions of the coupled model are described, called CM2.0 and CM2.1. The versions differ primarily in the dynamical core used in the atmospheric component, along with the cloud tuning and some details of the land and ocean components. For both coupled models, the resolution of the land and atmospheric components is 2° latitude x 2.5° longitude; the atmospheric model has 24 vertical levels.

The ocean resolution is 1° in latitude and longitude, with meridional resolution equatorward of 30° becoming progressively finer, such that the meridional resolution is 1/3° at the equator. There are 50 vertical levels in the ocean, with 22 evenly spaced levels within the top 220 m. The ocean component has poles over North America and Eurasia to avoid polar filtering. Neither coupled model employs flux adjustments.

The control simulations have stable, realistic climates when integrated over multiple centuries. Both models have simulations of ENSO that are substantially improved relative to previous GFDL coupled models. The CM2.0 model has been further evaluated as an ENSO forecast model and has good skill (CM2.1 has not been evaluated as an ENSO forecast model). Generally reduced temperature and salinity biases exist in CM2.1 relative to CM2.0. These reductions are associated with 1) improved simulations of surface wind stress in CM2.1 and associated changes in oceanic gyre circulations; 2) changes in cloud tuning and the land model, both of which act to increase the net surface shortwave radiation in CM2.1, thereby reducing an overall cold bias present in CM2.0; and 3) a reduction of ocean lateral viscosity in the extra- tropics in CM2.1, which reduces sea ice biases in the North Atlantic.

Both models have been used to conduct a suite of climate change simulations for the 2007 Intergovern- mental Panel on Climate Change (IPCC) assessment report and are able to simulate the main features of the observed warming of the twentieth century. The climate sensitivities of the CM2.0 and CM2.1 models are 2.9 and 3.4 K, respectively. These sensitivities are defined by coupling the atmospheric components of CM2.0 and CM2.1 to a slab ocean model and allowing the model to come into equilibrium with a doubling of atmospheric CO2. The output from a suite of integrations conducted with these models is freely available online (see http://nomads.gfdl.noaa.gov/).

There’s a brief description of the newer model version CM3.0 on the GFDL page.

Read Full Post »

Older Posts »