But when was the climate stationary? (if I understand correctly?)... end of the last century we emerged from the little ice age, with changes in solar forcing and volcanic forcing and (of course) since the industrial revolution there was the slow buildup (now rapid) of CO2 and other anthropogenic greenhouse gases... to my mind global climate has not been stable since we entered the ice house world 30+ million years ago. If models are able to reproduce Earth climate response given those changing parameters, then I think we need to give the some credibility... but at the end of the day, they are imprefect models that can only project long-term future change. We are still some teraflops way from accurate prediction on a local or intra decadal scale.

actually it is not quite sure which model could gives better result in general. however, still by compering performance of the different climate models for specific areas/locations to predict the climate change scenarios we can suggest one climate model is better manifesting it self than the other for that specific areas. i think so!!  

But when was the climate stationary? (if I understand correctly?)... end of the last century we emerged from the little ice age, with changes in solar forcing and volcanic forcing and (of course) since the industrial revolution there was the slow buildup (now rapid) of CO2 and other anthropogenic greenhouse gases... to my mind global climate has not been stable since we entered the ice house world 30+ million years ago. If models are able to reproduce Earth climate response given those changing parameters, then I think we need to give the some credibility... but at the end of the day, they are imprefect models that can only project long-term future change. We are still some teraflops way from accurate prediction on a local or intra decadal scale.

you are corrrect. first, most GCMs I am aware of are poor performers of current climate conditions in many places, from surface T to precipitation. Secondly, to work well over say the last 30 years they also would have had to get the climate change trends right. And here the may get the T well, but T signals will be worn in many places.

Of course this entire discourse misses the point. GCMs are used only because they provide internally coherent representation of the climate systems at most places, and because of this what matters is the change signals they provide, which are then applied to historical time series with a variety of techniques.

A better criteria instead of the very hard (and not definitive) one of representing local conditions, would be to work with the standard 3-4 (and counting of course) that have been used in AR5, and perhaps indeed select 2 that get closer to capturing large scale features of your location of interest.

Short answer for now

there is a real issue of timescales here.

So while I agree that the climate is most certainly "non-stationary" but on what timescale?   There are many possible timescales.  The shortest ones are driven by the PDO or the AMO or the ESNO (these are all sea surface temperature oscilliations)

So a climate model, as applied to the past, can likely reproduce some time averaged, regionally averaged, climate but it falls well short of reproducing the variability withing that time window.

For me, climate change can be associated with increasing weather volatility which remains very difficult to predict for the future.   The real world is far more affected by weather volatility than it is by a 30 year averaged regional climate.

Hi Haykel,

if it is true that a hypothesis is disproved by just one counterexample, your question has to be answered by "No!". Apparently none of the present-day General Circulation Models used for climate projection predicted the actual thermal stagnation that already lasts for about 15-20 years - notwithstanding reasonable matching of the thermal past over the whole instrumental period. Though one might speculate that 2014 could mark the startup of another global warming shove, the bidecadal hiatus has not been captured by the models. Present-day GCMs apparently do not correctly simulate unforced multidecadal variability, nothing to say about intraseasonal dynamics, e.g., notably of the water cycle - a major drive of atmospheric circulation across the timescales.

Kind regards, Peter

I agree with Peter, with one consideration: climate is not the same as weather. Most recent generation GCMs may be good at reproducing climate features (although this usually remians limited to basic T and P dynamics), but yet critical features of, say, extreme events, may be, and in fact are less well captured. Also, the majority of GCM will perform better in one region rather than in all of the world. And where they perform exceptionally well they may be heavily parameterized there--so again, no guarantee their future predictions are any better.

Bottom line: I agree that some preliminary testing of GCM performance is useful and lends some credence to projections, but as you say this is no particular guarantee...again, projections muct be seen as plausible and possible scenarios: they will never be exact predictions of what will happen in 2030 or 2050.

Do the testing, but stick to the top 4-5 GCMs in the IPCC reports.

In particular the plausibles direction of local precipitation projecitons will be a large range form positives to negatives---and all need to be considered in order to assess the robustness of the systems being considered in impact studies.

Hi,

first I would like to emphasize Peter's hint at possibly changing feedbacks in future climates. Natural or man-made forcing drives the climate system through the (attractor) basins of climate regimes, which are made up of the full set of internal feedbacks at their actual strengths. Within the stability margins of a regime, gradual changes may be observed, but then there may occur regime changes via critical transition. This might  have been the case not only in the distant climate past, but even during the instrumental period. In the early 1970, for example, the climate system escaped a more than 30 years long stagnation period which was characterized by large, but recurrent, decadal-scale oscillations. The escape left a clear signature in the climate record, i.e. may have been a regime transition indeed. In the early 1990s, the system appears to have entered a new stagnation period. Again, a kind of regime transition might be inferred e.g. from fundamentally changing dynamics in the greenhouse gas records, notably methane. Whether this observed regime-like behaviour is lately borne in natural or forced multidecadal variability, deserves closer scrutiny. To come to the point: If the capability of a GCM to simulate the last century's climate is reached just by clever forcing scenarios (recall: 'post mortem'), there is no guarantee that it will be a good tool for climate projection. I would be inclined to better trust a model that shows realistic internal variability over multiple scales, from intraseasonal to multidecadal (and beyond).

At this point, I would like to add a remark to Francesco's answer: By intraseasonal climate dynamics planetary scale motions are meant, not just synoptic variability and weather - though weather regimes are often bound to planetary-scale processes. The intraseasonal (30-60 day) activity in boreal summer of the global monsoon, with its regional branches that interact to one another, is certainly the most prominent and highly influential dynamic regime which the climate system passes each year, lately driven by orbital forcing. These oscillatory intraseasonal motions reflect a dynamic organization via certain feedback chains at a certain degree of thermal (energetic) excitation of the climate system. Suffice it to say that 2003 was an exceptional year just in this "topological" sense: the major intraseasonal activity of the monsoon system was very weak or not present at all; that is, another interplay of those feedbacks occured. Does anyone know a GCM that predicted this dramatic change in the system's motion, based on a proper initialization? A fundamental deficit of present-day GCMs is their inability to correctly represent intraseasonal monsoon dynamics. I have made conceptual studies that may somehow explain why this turned out to be so notoriously difficult to achieve (cf. my RG account).

Best, Peter

There exists the possibility to diminish the range of climate forecasts based on ensemle of GGMs using the Bayesion approach and historical climate data. For example you can find details in Arzhanov M.M. et al Global planet change. 2012. v. 86-87. P. 57-65.

Climate models are intended to correctly capture the dynamics of the atmosphere-land-ocean system.   That system is chaotic, but also subject to energy conservation and the strong over-all negative feedback of black-body radiation.  As a result, on long time scales the system has definable equilibrium states that depend on things like the total solar irradiance (averaged over a suitably long time), concentrations of greenhouse gases (again, averaged over a suitably long time), and continental positions.   When we are trying to estimate the equilibrium climate sensitivity of the earth to, say, a doubling of carbon dioxide, it's very reasonable to compare two equilibrium periods in the earth's past (for example, the Last Glacial Maximum and the Holocene Optimum) and ask how well our climate models capture the observed differences between those periods.    The problem there, is that those equilibrium periods were a long time ago, and we only know the state of the climate during those times to limited precision.   You can read about efforts to test climate models against these observed climate equilibria at the website of the Paleoclimate Model Intercomparison Project (PMIP).   

When we compare climate models against changes in the recent past, we need to understand that due to the chaotic nature of the atmosphere-land-ocean system's governing physics, the actual climate history is only one of a cloud of climate histories that might have been observed, but weren't.  Just as the day-to-day weather for next month is fundamentally unknowable, within the distribution prescribed by energy conservation,  the year-to-year fluctuations due to El Nino-type variability are not knowable more than a few months in advance.   So we need to compare climate model performance over a period long enough that the unforced variability of the climate system is averaged out.   We're not completely sure how long that is, because we're not completely sure how large the unforced variability of the ocean circulations is over timescales of a decade or more.    

So we need to ask, for a given set of climate model design and parameter choices, how likely is it that the observed trajectory would arise?    We need to run our climate models with a large range of initial conditions and we need to run a lot of climate models with different ways of handling aspects of the climate system's physics (all within the uncertainty of our understanding of the system), and then ask, how likely would each of those choices be to produce the observed climate history?  That's what the IPCC report asks in its chapter on the evaluation of climate models, which you can read here.

http://www.ipcc.ch/pdf/assessment-report/ar5/wg1/WG1AR5_Chapter09_FINAL.pdf

https://pmip3.lsce.ipsl.fr/

Hi,

for an in-depth - though incomplete for good reasons - discussion of the intriguing and complicated issue to separate the signs of climate change from those of climate variability, you might wish to consult another discussion that ran a couple of weeks ago on RG:

https://www.researchgate.net/post/Can_anyone_help_with_the_difference_between_climate_change_and_climate_variability

To come back to the question posed, let me say: I'm not happy with the perspective that we inevitably might have to run thousands of GCM projections, i.e. to produce mountain-ranges of data that no-one might ever be able to seriously analyze (e.g. in order to understand mechanisms), when we wish to decide which of the models to prefer over others. The question was simply: Should we prefer models that have shown the ability to match the recent climate past, i.e. is this a reliable selection criterion? The answer is "no" to my mind, but this should not be confused with a statement of mistrust against GCMs in general (I'm a modeller myself for long enough and sufficiently know that business to respect it). It depends on ... the purpose of use - like it is the case for all models, not only GCMs (and also for all methods of data analysis).

Some understanding of the qualitative dynamics the climate system may develop can be gained from observation in conjunction with modelling exercises, but hardly from mass production. At the outset, a clear terminology should also be chosen. A basic question hereby is the conceptual understanding of climate change. Is the system viewed as one at some kind of (quasi-)equilibrium, with largely stochasic fluctuations around, or is it viewed as a system in a state of higher (thermal and dynamic) excitation, structurally (or, topologically) apart from equilibrium? Is it even seen to have already reached a chaotic state along some of these routes of excitation ("routes to chaos")? The physics of the system is not chaotic per se, and a system cannot be in equilibrium and in a chaotic state at the same time and over the same spatial scales (we do not speak of weather systems here). It depends on the degree of forcing (flow of energy through this open system's boundaries) which dynamical status the climate system attains (given a sufficient set of feedbacks in action). I have seen indeed chaotic GCM solutions, and had a sufficient degree of trouble with model stability in those regimes, but the basic message that I have received from both modelling and observational data analysis (of the instrumental period) is complexity, and remarkably: apparently at much lower dimensionality of the system than usually presumed of the climate system's motion. If low-dimensional climate dynamics plays a substantial role indeed in the system's variability and change, there is a better chance, I mean, to understand its potential of evolution, i.e. its future qualitative dynamics. This observation may also provide hints at model preferences, and I have learned that both the unforced variability a model generates and its performance with respect to the water cycle are among a handful of basic criteria which may reflect qualitatively correct dynamics. 

Regards, Peter

The dominant part of future climate change in the coming, say, 100 years is relateed to man-made drivers, first of all greenhouse gases; the component is somewhat present in the recent decades, but not in a significant way prior to that. Thus, a good performance of models in the historical past is not indicative that the Greenhouse part is simulated realistically - which is what you ask for. Otherwise, the "futures" presented by models are not predictions, i.e, most probable developments, but mere "scenarios", i.e, possibe, and consistent developments. (see our paper Bray, D., and H. von Storch, 2009: 'Prediction' or 'Projection'? The nomenclature of climate science. Sci. Comm. 30, 534-543, doi:10.1177/1075547009333698 somewhere her on ResearchGate)

Climate models show remarkable agreement in their projections, for example, they unanimously suggest future long term global warming. However, there is no way to absolutely guarantee that future projections of climate change on regional scales will give accurate answers.  The easiest way to illustrate this is by the fact that for some regions the future projections from different (state of the art) climate models give different answers.  However, by testing models against past climate we can gain confidence in models by verifying that they are able to simulate historical climate.  Given that the physics of the climate system is constant, we can then infer confidence in the future projections.  There are increasingly sophisticated ways of performing these tests:

1) Testing modeled mean climate over recent decades.  Does the climate model simulate the average values of surface temperature, winds, pressure etc? climate models do a remarkable job here, with many key features of the climate emerging spontaneously from the model physics.  Nevertheless, there are stubborn errors, for example in the structure of jet streams, the mean sea level pressure and seasonal mean precipitation amounts that are common to many leading models.

2) Testing modeled climate variability against recent observations.  Do climate models exhibit the characteristics of recent variability? again the models do remarkably well in reproducing the major patterns of variability such as El Nino, variations in the North Atlantic Oscillation, and other patterns of variability.  However, there are still inevitable errors and models do not do a good job of simulating the Madden Julian Oscillation for example.

3) Testing climate model predictions against recent observations.  In this case we can start the models from actual observed climate at various points in the historical record and test how well they predict the subsequent climate evolution.  If the models do well in predicting past climate variability, and if the processes are similar, they will do well in capturing future climate, at least in the near term out to years or perhaps decades ahead.  However, we have to be very careful with these tests because the future processes may be different and not everything is predictable anyway.  So testing whether such predictions span the range of observed variability, even if it is unpredictable, is therefore important.

In the end, if we can understand the mechanisms at work in the aspect of the prediction that you are particularly interested in, and if we can verify that these mechanisms agree with the observations we have so far, then it is reasonable to assume that those climate models will give better answers.

Yes "weather and climate" on some timescales is a statistical effect and so  mean values and changes probably can be will predicted by the models.

It this kind of stuff:

"the structure of jet streams". which represents dynamical change rather than statistical change and here, at the moment, most of the models fail.  This is understandable 

a) this is hard

b) we don't yet know all the relevant input physics.

Hi,

I think your question mean is: Does the historical data be reliable?

You know historical data are the outputs of models. So, there isn't any guarantee that their outputs have proper results. Because of climate change has many different responses & reasons. In a nutshell you need station's data. such as climatology stations, synoptic stations, etc for comparing results of historical data & observation data. I think it is the best way to find the errors of historical GCM output.

Best regards,

Nasrin