In the context of our discussion of multi-decadal climate, it is the forecast of the evolution of climate variables over this time period of such climate variables as temperature, precipitation, soil moisture, sea ice, vegetation etc over local, regional and global average scales. Such a prediction is unconstrained in its evolution as observed climate variables are not used except as initial conditions.

The prediction can be expressed as an ensemble of model runs (i.e. a set of realizations with perturbed initial conditions, forcings etc) or as one realization. The climate variables can be expressed as values with different time and space averages (e.g. such as a decadal average regional maximum 2m temperature for Amsterdam).

You write that

“1. One should realize that there is ALWAYS a chance that predictions do not come true, even if the model has shown skill in hindcast studies; 2.There are a number of tests (but more than just the skill in a hindcast) that feed our belief in the usefulness of a model, indeed in a sort of Bayesian manner.”

If the predictions are inaccurate in hindcast studies, they certainly should not be claimed to be able to have skill in any projections for the coming decades. Of course, a prediction, even if it shows skill in a hindcast set of runs, can still be inaccurate in the future. But a NECESSARY condition to accept a model prediction as accurate is that they show skill.

On your second comment, what tests are you referring to? Also, the use of the word ‘belief’ is not a robust scientific demonstration of added value. Please be more specific.

Question 1b: What exactly is meant by the word skill?

In the context of multi-decadal climate predictions of climate change, skill is defined as an ability to produce model results for climate variables that are at least as accurate as achieved from reanalyses (averaged to correspond to the time period of the model prediction). The skill needs to be tested using hindcast runs against i) the average climate over a multi-decadal time period and ii) CHANGES in the average climate over this time period.

Moreover, a model cannot be said to be accurate if it deviates less from reality for one climate variable (e.g. temperature) than does another variable (e.g. precipitation). They are all interrelated. If precipitation is incorrect, clouds, near surface soil moisture, vegetation condition etc will be affected. This will affect the surface sensible and latent heat fluxes, and then must alter temperature. If the temperature were closer to the observed but clouds and precipitation were not, this clearly should be a red flag that the model is not obtaining its temperature results for the right reason.

Question 2: How robust is the distinction Roger makes between 4 different downscaling types?

The distinction is based on the degree of constraint provided by real-world observational data. With Type 2, real world data (either that retained from a global model in terms of long wave circulation patterns, sea surface temperature, etc, or from reanalyses) provides a control on how much the model results can deviate from reality. With Type 4, except for the prescribed external forcing, terrain, deep soil type, deep ocean temperature and salinity etc which vary over much longer time periods, all other climate variables have no real world constraint.

The four types of downscaling provide a robust delineation of downscaling in that the time period of the integration and the degree of independence from real world data are clearly distinct.

Question 3: (for Bart and Jason): Would it be possible to be more specific about the added value of RCMs by giving a list of references and quotes?

I am also still waiting for examples where Type 4 runs in hindcast provide skill in predicting climate change. Even predicting current climate statistics is a daunting challenge as I illustrated with the examples from the literature.

Question 4: Would it be feasible to focus on one specific vulnerability case, and to explore what would be the difference between the approaches of Roger on the one hand, and of Bart and Jason on the other hand?

I agree.

Let’s select the vulnerability of Amsterdam to future (multi-decadal) environmental conditions, including climate.

On the example with respect to traffic accidents, this is an excellent example of the bottom-up approach to vulnerability. If an insurance company runs a sophisticated model to investigate whether the accident rate goes up, if, for example, the traffic density increases, they, of course, would use a model that was validated as having skill for this type of prediction.

We can use the same bottom-up approach with climate. For example, if sea level rose by 50cm, 1m etc over the next 50 year what would be the consequences for Amsterdam?

Sea level rise is, of course, the climate variable that the global models should be able to project on multi-decadal time scales (if they can for any climate variable) since the sea level rise is based to a significant amount on large scale average heating of the oceans. Unfortunately, however, the multi-decadal global climate model have shown no skill at predicting changes in sea level beyond what is obtained by just extrapolating the observed sea level rise into the future.

Even with the absence of skillful projections, however, we can address the question ‘what level of sea level rise is plausible’? We can do this by

i) assuming a continuation of the observed sea level rise for the Netherlands over the last century;

ii) assume a release of fresh water from glaciers in Antarctica and Greenland that is double, triple etc what it was estimated to be over the last several decades,

iii) assume part of the ice in Antarctica falls into the sea, etc.

Each of these would increase the threat to Amsterdam but with (probably) less likelihood. We would like to have quantitative estimates of the probability of these events occurring, but, unfortunately, our level of skill is not there yet.

The global and regional climate models, without showing skill at prediction on multi-decadal time scales, is only providing an illusion of skill to the policymakers with respect to this climate variable (i.e. sea level change) as well as changes in other climate variables. They cannot provide us with verifiably robust estimates of future sea level rise beyond what we have observed in recent decades.

By presenting the global, regional, and local climate projections as robust (skillful) to the impacts and policy communities we are misleading them on the actual level of our scientific capability. The large amount of funds that are being used to make these multi-decadal projections could be better spent on adaptation to climate and other environmental risks, regardless of our degree of confidence that we can predict changes in risk in the coming decades. Indeed, this is a much better insurance policy than relying on global climate models for this information.

The recent tragic tornado in Oklahoma USA is a good example of how money that is being used for multi-decadal impact studies based on global climate models could have been reallocated to provide tornado shelters for schools in Oklahoma, with a resultant much more beneficial result (i.e. it could have saved lives). In my view, we need a reassessment of funding priorities with respect to the risks that society faces.

Gerbrand – thank you again for your excellent questions and comments! Roger

Guest comment by Gerbrand Komen:

Hi Marcel,

I am following the discussion between Bart, Jason and Roger with great interest. This is a very important topic indeed, and I hope that this discussion will not only lead to better mutual understanding but also indeed to a better insight in the way in which model results should be interpreted.

Perhaps you could benefit from some of my observations.

First of all, Bart, Jason and Roger seem to agree that combining observations and models is the best way forward if you try to understand the climate system.

But there is also some real disagreement.

Roger repeatedly makes the following points:

· There is no difference between predictions and projections.

· Climate models have no skill in a hindcast mode and are therefore not useful for ‘type 4’ predictions.

· Adaptation (always important, with and without climate change) should be done on the basis of a vulnerability analysis.

Roger also feels that the use of RCMs for the development of adaptation strategies is cost-inefficient.

Jason and Bart both maintain that RCMs can (or may?) provide useful additional information. As an example Bart mentions the case where excessive precipitation and an extreme storm surge coincide.

I hope the discussion will clarify a few questions that came to my mind.

Question 1a: What exactly is a prediction?

I tend to agree with Roger that there is no sharp distinction between predictions and projections. They are both modelled constructions of the state of the climate system at a future time. The real question is about the belief in the possibility that these predictions may verify. This is not black or white, because the prediction may be approximately correct for some variables and wrong for some other variables, and because one can distinguish different degrees of beliefs about the likelihood that a prediction will (or might) come true [very uncertain, possible, very unlikely, . . very likely . .]. Roger seems to assume that skill in hindcasts is necessary for usefulness. I can see that skill strengthens our belief in the likelihood, but in my essay [http://bit.ly/cCb1n2 or http://home.kpn.nl/g.j.komen/Uncertainties.pdf] on Rogers website I have made two points that may be relevant for the present discussion: 1. One should realize that there is ALWAYS a chance that predictions do not come true, even if the model has shown skill in hindcast studies; 2.There are a number of tests (but more than just the skill in a hindcast) that feed our belief in the usefulness of a model, indeed in a sort of Bayesian manner.

Question 1b: What exactly is meant by the word skill?

There are technical definitions, such as

http://www.ecmwf.int/products/forecasts/guide/Measure_of_skill_the_anomaly_correlation_coefficient.html, but, more loosely speaking, the word skill could also be used to indicate agreement between modelled and observed quantities. It seems important that Bart, Jason and Roger use the same definition.

Question 2: How robust is the distinction Roger makes between 4 different downscaling types?

I have my doubts. Bart also, apparently. He somewhere called his scenarios as ‘between type 2 and 4’. To me it seems that there are different ‘dimensions’ to the problem:

1. how are we nesting, initializing and forcing?

2. how are we validating?

3. what prediction horizon is considered?

4. how are we using the results?

Roger seems to project this on a one-dimensional space (type 1 – 4), but perhaps a more subtle division could benefit the present discussion.

Question 3: (for Bart and Jason): Would it be possible to be more specific about the added value of RCMs by giving a list of references and quotes?

This is exactly what Roger has been asking for several times. Roger gave specific lists of model failures, but I suspect that he has been cherry picking. It would be very useful therefore to also have a list of model successes. From my own extensive modelling experience I know that there always are aspects that are not well-described by a model. By developing one’s belief in a model one has to weigh failures and successes. Roger – and many others – seem to have the outdatedPopperian idea that you can refute a model of a complex system. This is simply not true. For example, in our ocean wave prediction model the wave height was generally better computed than the wave period. So we knew that reality was more complex than our model, but this did not make our model useless. In fact, the results have been widely and successfully used. So for the present discussion, it seems essential to have lists of both successes and failures of RCMs

Question 4: Would it be feasible to focus on one specific vulnerability case, and to explore what would be the difference between the approaches of Roger on the one hand, and of Bart and Jason on the other hand?

An interesting case might be the precipitation/storm surge case mentioned by Bart, but any other specific case would also be helpful.

So, these are my questions. It would be great if Bart, Jason and Roger could come up with answers.

To end, it may be helpful when I indicate how I try to understand the different viewpoints in a simplified, daily life example. In daily life we make scenarios all the time. For example, I can imagine getting involved in a traffic accident at some future time. So this is a realistic scenario, and a statement about what might happen. As a result I take insurance. Fortunately, I do not need to run a sophisticated agent-based model of traffic in the Netherlands to generate this scenario. The observation that traffic accidents occur and the fact that I am not very different from other people suffices. This is more or less what Roger says about RCMs. However, I can imagine that an insurance company would want to run sophisticated models to investigate whether the accident rate goes up, if, for example, the traffic density increases. And this is the position of Bart and Jason. This suggests that the value of using models not only depends on the trust one has in the model but also on the particular question that is being asked. Of course, I am not sure whether this analogy stands, but perhaps reactions to question 4 can shed light on this.

So these are my observations. I hope they are helpful, and, above all, I really hope that the discussion between Bart, Jason and Roger on climatedialogue will bring us forward.

All the best,
Gerbrand

To me it lacks a clear definition of Long-Term Persistence. Two of the invited debaters use it but both do not seem to be able to explain it in simple terms what is and how it comes about. Why is that? The rest of the discussion seems to be a lot of talking to one another on different levels. Telling others to read the papers doesn’t help if you can’t explain succinctly what you are on about. I do notice the blog operators trying to get something more out of it, but i feel their efforts failed. Pity…

There appear to be different interpretations of natural variability and of detection which may be a frequent cause of misunderstanding in this dialogue and beyond. Below I’ll try to describe these different interpretations in an effort to elucidate where the different opinions may (partly) be coming from.

In general, the following processes involved in climate change can be distinguished:
- natural unforced variability (e.g. internal variability involving a redistribution of energy)
- natural forced variability (e.g. changes in the output of the sun or in volcanism)
- anthropogenic forced variability (e.g. changes in greenhouse gas or aerosol concentrations)

where a forcing refers to a process causing an energy imbalance, which in turn causes a temperature change. Internal variability on the other hand causes a temperature change arising from semi-random internal processes. This temperature change can then cause an energy imbalance (since outgoing energy scales as T^4), but the cause-effect chain (linking temperature change and energy imbalance) is opposite to a radiative forcing)

As we wrote in the introductory text, according to AR4 “an identified change is ‘detected’ in observations if its likelihood of occurrence by chance due to internal variability alone is determined to be small.”

In other words, detection is based on distinguishing the forced (natural and anthropogenic) from the unforced (natural) component.

Demetris seems to argue that these different processes can not be distinguished, or at least that internal (unforced) variability and natural forcings can not be distinguished. Anthropogenic forcings can only be distinguished by virtue of them not having been acting on the system prior to ~1850. Armin seems to take a somewhat similar view, in combining natural unforced and forced changes in what he terms natural fluctuations. Rasmus seems to take the view as I outlined above (the distinction in three main types of processes).

Demetris argues that the current temperature signal is not outside of the bounds of what could be expected from natural forced and unforced changes, thereby using a higher bar than the standard definition of “detection”. He bases his statement on a higher Hurst coefficient than Armin does, which increases the bar further.

This may clarify how the statement that climate forcings introducing LTP and climate forcings being omnipresent (which all three agreed on) can still lead to different conclusions regarding the presence of LTP saying anything about internal variability, because different operational definitions of detection and internal variability (and perhaps also of LTP, as has been put forward by Armin) are used (where in one view internal variability is only the unforced component of change, where in another view internal variability also includes natural forcings).

This brings up the question, if (according to Demetris) the recent warming is not outside of the bounds of natural forced and unforced variability, where does all the excess energy come from that is observed to accumulate in the climate system? It doesn’t seem to be due to natural forcings (which show no warming trend over the past 50 years), nor is there any sign of a redistribution of energy within the climate system (everywhere we look it’s warming). Where is the energy hiding, or where is it coming from (if not from excess greenhouse gases inhibiting planetary heat loss)?

“…they assume that the derived statistical relationship will not change due to climate change.

Dynamical downscaling, or the use of Regional Climate Models (RCMs), does not share the limitations of statistical downscaling.”

I think that RCMs very much share the potential caveat of non-stationarity, as they too involve statistically fitted models, known as parameterisation schemes. The bulk effects of unresolved small-scale processes are essentially represented by statistical models, e.g. clouds and surface fluxes. What’s more, imperfections in these feed back into the calculations in RCMs, whereas such discrepancies merely cause a growing bias in the statistical schemes.

It is also possible to evaluate the downscaling for these effects, e.g. by applying them to the past. Furthermore, since the RCMs and statistical downscaling draw information from independent sources, and one need to look at both (do they converge/diverge) before using the results for decision making.

My best wishes for the continuation and further development of the Climate Dialogue forum. Even with the difficulties encountered, dialogue is the only way forward. Besides, as Heraclitus said, “Tο αντíξουν συμφέρoν και εκ των διαφερόντων καλλíστην αρμoνíαν και πάντα κατ’ έριν γíνεσθαι” (Opposition unites, the finest harmony springs from difference, and all comes about by strife).

If I may offer a simple suggestion for the future dialogues, I would propose to merge the two sections “Expert comments” and “Public comments”. First, these section titles are not very accurate; it would be more accurate to say “Editors and guests” rather than “experts”. My feeling is that everybody who contributes in this dialogue is an expert—both the eponymous and the pseudonymous discussers. Second, the reading of the comments would be more convenient and sensible if the comments were in chronological order rather than separated into two sections.

D.

What is the chance that the observed changes are due to internal variability? (meaning a redistribution of energy within the climate system – there seems to be quite some confusion about what the different terms forced vs unforced/internal var. mean, which I will come back to in a later comment)

In more complex systems (yet the most common ones) whose study needs to abandon linear models, the contribution of each cause or forcing is not straightforward.

So I may not be able to calculate the particular contribution of external and internal forcings. For me it suffices to say that the climate was never static, which implies variability — particularly due to internal dynamics. If you can do such separation, please do and let me know if you find that in the entire Earth’s history the ever changing climate has been driven by external forcing only.

D.