* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Predicting regional climate change: living with uncertainty
Myron Ebell wikipedia , lookup
Economics of climate change mitigation wikipedia , lookup
Mitigation of global warming in Australia wikipedia , lookup
Numerical weather prediction wikipedia , lookup
Global warming hiatus wikipedia , lookup
Climatic Research Unit email controversy wikipedia , lookup
Instrumental temperature record wikipedia , lookup
German Climate Action Plan 2050 wikipedia , lookup
2009 United Nations Climate Change Conference wikipedia , lookup
Michael E. Mann wikipedia , lookup
Heaven and Earth (book) wikipedia , lookup
Atmospheric model wikipedia , lookup
Global warming controversy wikipedia , lookup
Fred Singer wikipedia , lookup
Soon and Baliunas controversy wikipedia , lookup
ExxonMobil climate change controversy wikipedia , lookup
Effects of global warming on human health wikipedia , lookup
Climate resilience wikipedia , lookup
Climate change denial wikipedia , lookup
Climatic Research Unit documents wikipedia , lookup
Global warming wikipedia , lookup
Climate change adaptation wikipedia , lookup
Politics of global warming wikipedia , lookup
Economics of global warming wikipedia , lookup
United Nations Framework Convention on Climate Change wikipedia , lookup
Climate change feedback wikipedia , lookup
Effects of global warming wikipedia , lookup
Climate change in Tuvalu wikipedia , lookup
Climate change and agriculture wikipedia , lookup
Climate engineering wikipedia , lookup
Citizens' Climate Lobby wikipedia , lookup
Climate governance wikipedia , lookup
Carbon Pollution Reduction Scheme wikipedia , lookup
Media coverage of global warming wikipedia , lookup
Climate sensitivity wikipedia , lookup
Climate change in the United States wikipedia , lookup
Global Energy and Water Cycle Experiment wikipedia , lookup
Public opinion on global warming wikipedia , lookup
Attribution of recent climate change wikipedia , lookup
Scientific opinion on climate change wikipedia , lookup
Effects of global warming on humans wikipedia , lookup
Solar radiation management wikipedia , lookup
General circulation model wikipedia , lookup
Climate change and poverty wikipedia , lookup
Climate change, industry and society wikipedia , lookup
Surveys of scientists' views on climate change wikipedia , lookup
Progress in Physical Geography 23,1 (1999) pp. 57–78 Predicting regional climate change: living with uncertainty Timothy D. Mitchell and Mike Hulme Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK Abstract: Regional climate prediction is not an insoluble problem, but it is a problem characterized by inherent uncertainty. There are two sources of this uncertainty: the unpredictability of the climatic and global systems. The climate system is rendered unpredictable by deterministic chaos; the global system renders climate prediction uncertain through the unpredictability of the external forcings imposed on the climate system. It is commonly inferred from the differences between climate models on regional scales that the models are deficient, but climate system unpredictability is such that this inference is premature; the differences are due to an unresolved combination of climate system unpredictability and model deficiencies. Since model deficiencies are discussed frequently and the two sources of inherent uncertainty are discussed only rarely, this review considers the implications of climatic and global system unpredictability for regional climate prediction. Consequently we regard regional climate prediction as a cascade of uncertainty, rather than as a single result process sullied by model deficiencies. We suggest three complementary methodological approaches: (1) the use of multiple forcing scenarios to cope with global system unpredictability; (2) the use of ensembles to cope with climate system unpredictability; and (3) the consideration of the entire response of the climate system to cope with the nature of climate change. We understand regional climate change in terms of changes in the general circulations of the atmosphere and oceans; so we illustrate the role of uncertainty in the task of regional climate prediction with the behaviour of the North Atlantic thermohaline circulation. In conclusion we discuss the implications of the uncertainties in regional climate prediction for research into the impacts of climate change, and we recognize the role of feedbacks in complicating the relatively simple cascade of uncertainties presented here. Key words: climate change, climate modelling, downscaling, external forcing, regional climate prediction, uncertainty, unpredictability. I Introduction When a person, city or nation considers climate change, their primary interest is in changes in the warmth, rain and wind that affect them and their local environments. © Arnold 1999 0309–1333(99)PP214RA 58 Predicting regional climate change: living with uncertainty Nevertheless, it was the argument that greenhouse gases emitted by humans (‘anthropogenic emissions’) might substantially alter global climate that persuaded the global community to establish the UN Framework Convention on Climate Change (UN/FCCC) (Hecht and Tirpak, 1995) with the following objective: ‘Stabilisation of greenhouse gas concentrations at a level that would prevent dangerous anthropogenic interference with the climate system’ (UN/FCCC Article 2). The scientific evidence that prompted countries to sign and ratify the UN/FCCC and then to negotiate the Kyoto Protocol in 1997 (which sets specific targets for reducing greenhouse gas emissions – Masood, 1997) has been summarized in a series of influential reports prepared by the Intergovernmental Panel on Climate Change (IPCC) (Houghton et al., 1990; 1996). The majority of the scientific evidence contained in these reports is concerned with the global response of the climate system to anthropogenic forcing, but people, cities and nations will be affected by the regional1 responses. The IPCC (Watson et al., 1998) was therefore asked to prepare a special report on the vulnerability of natural environments and human societies to climate change in different regions of the world, which was to provide: ‘A common base of information regarding the potential costs and benefits of climatic change, including the evaluation of uncertainties, to help the COP [Conference of the Parties to the UN/FCCC] determine what adaptation and mitigation2 measures might be justified’ (Obasi and Dowdeswell, 1998: vii). Such accurate and precise information is highly desirable, but the chair of the IPCC was forced to acknowledge that it could not be provided: ‘Because of the uncertainties associated with regional projections of climate change, the report necessarily takes the approach of assessing sensitivities and vulnerabilities of each region, rather than attempting to provide quantitative predictions of the impacts of climate change’ (Bolin et al., 1998: x). Underlying this acknowledgement is the belief that ‘models cannot predict on regional scales’. Examples of this frequently heard refrain may be taken from a single volume of Progress in Physical Geography (Joubert and Hewitson, 1997: 52; Schulze, 1997: 118; Wilby and Wigley, 1997: 531) or from Henderson-Sellers’ (1996: 59) review of the climate research process: ‘Unfortunately, numerical climate models and the associated tools currently used to predict climate change, have no regional prediction skill.’ If we regard GCMs (‘general circulation models’ or ‘global climate models’) as unreliable at regional scales, and if we regard the palaeo-analogue approach to regional climate prediction as inappropriate (Covey, 1995; Crowley, 1997), then it appears that regional climate prediction is not yet possible. In this review we demonstrate that if we equip ourselves with an understanding of the fundamental role of uncertainty, then the prediction of regional climate change is reduced from an insoluble to a difficult problem. Since it is commonly assumed that model deficiencies are responsible for any differences between models at regional scales, such differences are commonly used as ‘proof’ that ‘models cannot predict on regional scales’. Having outlined in section II the differences between models at regional scales, we discuss in section III the potential for downscaling techniques to overcome the difficulty posed by the differences; we conclude that downscaling techniques cannot correct for model inaccuracies and so they fail to overcome the difficulty. In section IV we consider the possibility that model deficiencies are responsible for the differences between models at regional scales. However, there is an alternative explanation provided by climate system unpredictability, and in section V we show that the dependence on initial conditions of a climate simulation means that Timothy D. Mitchell and Mike Hulme 59 it is premature to infer from the differences between models at regional scales that the models are deficient. Differences between simulations may (at least partly) be due to the inherent unpredictability of the climate system. Therefore, if we consider the problem of predicting regional climate change in more general terms we find that in addition to the possibility that model deficiencies may render our predictions uncertain (section IV), there are two sources of inherent uncertainty in regional climate prediction: climate system unpredictability (section V) and global system unpredictability (section VI). By ‘global system unpredictability’ is meant the inherent unpredictability of the external3 forcings imposed on the climate system, whether anthropogenic (e.g., greenhouse gas emissions) or natural (e.g., solar variability or volcanic eruptions). The unpredictability of the climatic and global systems is such that even with a perfect model, uncertainty would remain inherent to regional climate prediction. Yet the climatic and impacts research conducted thus far has largely considered uncertainty in terms of model deficiencies. Therefore we suggest (in section VII) that the problem of regional climate prediction should be viewed in terms of a cascade of uncertainty that stems from the unpredictability of the climatic and global systems, instead of a single result process sullied by model deficiencies. Given this understanding of the uncertainties involved in regional climate prediction, in section VIII we suggest three methodological approaches to deal with them: (1) the use of multiple forcing scenarios to cope with global system unpredictability; (2) the use of ensembles to cope with climate system unpredictability; and (3) the consideration of the entire response of the climate system to cope with the nature of climate change. All three approaches apply equally to global and regional climate prediction, but the entire response of the system to forcing (the third approach) is treated differently at global and regional scales. So in section IX we show how regional climate prediction is concerned in particular with the response of the general circulations of the atmosphere and oceans. In section X we illustrate this discussion by considering the North Atlantic thermohaline circulation, and by providing some examples of the use of ensembles. In conclusion (section XI), we consider the implications of the uncertainty of regional climate prediction for research into the impacts of climate change, and we point out the role of feedbacks in complicating the relatively simple cascade of uncertainties presented here. II Intermodel differences on regional scales In the early 1990s there was very little confidence in the regional climate information extracted from GCM integrations, and the models were held responsible because of • • • • • • the equilibrium method of forcing models with CO2-equivalent gases; the lack of fully coupled atmosphere–ocean GCMs (AOGCMs); coarse model resolution; deficiencies in the model physics; an inability to simulate present-day regional climate features; and intermodel differences in simulations of regional climate change. 60 Predicting regional climate change: living with uncertainty Since then there have been a number of improvements in the climate models, notably in the adoption of transient methods of forcing with CO2-equivalent gases, and in the development of fully coupled AOGCMs. Despite the improvements in the climate models, intermodel comparisons still reveal substantial intermodel differences on regional scales. The IPCC Second assessment report included a comparison of simulations from a number of AOGCMs for five particular regions (Kattenberg et al., 1996), and more recent work (Kittel et al., 1998) has confirmed the results reported by the IPCC. The models’ representation of the real world may be judged by comparing the control simulations4 with observed climatologies; such comparisons reveal that the biases in the seasonal cycle range between –7 °K and 10 °K. These recent model experiments exhibit biases in the same range as those in the IPCC First assessment report for control simulations with older models (Gates et al., 1990). If we compare the models we find that over most regions the intermodel range of temperature bias is of the order of 10 °K, and when AOGCMs are forced with greenhouse gases, the climate ‘scenarios’5 of regional temperature and precipitation change vary widely from model to model (Kattenberg et al., 1996; Kittel et al., 1998). Figure 1 illustrates this variety of regional responses to forcing that a collection of models typically exhibits. Even when one considers a global-scale phenomenon such as ENSO (El Niño Southern Oscillation), both the model differences from observations, and the wide range of results among models and between models and observations, Figure 1 The differences between models on regional scales. Nine model experiments (represented by different symbols) were reviewed by Kattenberg et al. (1996) who presented the regional biases in summer (JJA) temperature and precipitation for seven regions (the eighth (GL) is a global average). Temperature biases are differences (control minus observed); precipitation biases are differences as percentages of the observed. Square brackets about the zero-bias line span +σ to –σ for temperature, or the coefficient of variation for precipitation, of observed regional averages Source: Modified after Kittel et al. (1998) Timothy D. Mitchell and Mike Hulme 61 remain; a model’s ENSO may be too strong (Tett et al., 1997), too weak (Knutson and Manabe, 1994; Schneider and Kinter, 1994), or absent (von Storch, 1994). Intermodel comparisons such as these have reduced our confidence in AOGCM scenarios of regional climate change to a ‘low’ level (Kattenberg et al., 1996: 339), and the culprit is generally held to be the AOGCMs’ low horizontal resolution and physical parameterizations (Gates et al., 1996). III Can downscaling help? Intermodel differences have discouraged many researchers from using AOGCM output on regional scales. Being aware of the acute need for reliable predictions of regional climate change, some have therefore turned to downscaling techniques in their attempts to overcome the differences between models on regional scales. Recent reviews of downscaling techniques may be found in Hewitson and Crane (1996), Kattenberg et al. (1996) and Wilby and Wigley (1997). Statistical downscaling techniques commonly develop statistical relationships between local climate variables and model predictors, and then apply those relationships to model climate scenarios to derive estimations of localized climate change. Such statistical links may be made directly by regression (Kim et al., 1984), via circulation patterns (Wilby, 1994) or via stochastic weather generators (Wilks, 1992). Dynamical downscaling techniques commonly use output from GCM simulations to provide the boundary and initial conditions for a ‘one-way nested’ regional model with a higher spatial resolution (Giorgi, 1990); a variant approach is to develop a variable-resolution GCM (Deque and Piedelievre, 1995; Fox-Rabinovitz et al., 1997). When applied to regional climate change prediction, the one element these downscaling techniques have in common is their dependence upon the accuracy of the large-scale information with which they are ‘driven’. This is true whether the relationship between the predictor and predictand is statistical or dynamical; both assume that the large-scale information from the GCM is ‘correct’ and on that basis they attempt to estimate the corresponding local change in climate. If the large-scale changes in the GCMs are inaccurate, then the derived local changes merely add misleading precision to the GCM results, and the downscaling techniques fail to add any predictive skill. Figure 2 illustrates how closely a downscaled variable depends on the GCM variable from which it is derived. We find that if the GCM variable is an accurate representation of the corresponding observed variable then the downscaled variable is also an accurate representation of the observed variable; conversely, where the GCM variable is inaccurate, so is the downscaled variable. We conclude that a prerequisite of downscaling is an accurate GCM to ‘drive’ the downscaling technique. Leonard Bengtsson (1995), former Director of the Max Planck Institute and ECMWF, concurs: ‘Regional climate prediction requires first and for all [sic] good global prediction by coupled climate models – without it regional simulations are useless.’ The development of downscaling techniques does not overcome the lack of confidence in AOGCMs at regional scales; it merely accentuates the need for the inherent regional uncertainties to be understood. 62 Predicting regional climate change: living with uncertainty Figure 2 The dependence of downscaled results on the large-scale information supplied by the GCM for three variables: maximum daily temperature in degrees Celsius (TMX), fog in units 0–1 (FOG) and snow height in cm (SNW). The observed data (OBS), the GCM data (CTL) and the downscaled data (SCA) are presented for the weather station at Potsdam, Germany Source: Modified after Bürger (1996) IV AOGCM deficiencies Any attempt to explain the wide range of AOGCM outcomes at regional scales described in section II must consider a number of competing potential reasons, the first of which is provided by model deficiencies. Most scientists accept that it is unreasonable to suggest that AOGCMs simply have poor physics and are unrepresentative of reality. If one allows for the coarse resolution of AOGCMs then the models are remarkably good, simulating the seasonal cycles in temperature, precipitation rates and mean sea-level pressure (Gates et al., 1996). However, it is possible that the parameterizations, coarse resolution and unrepresented feedbacks that limit AOGCMs may generate the intermodel variety exhibited in AOGCM simulations at regional scales. This may particularly apply to hydrological variables, whose key processes (evaporation, convection, condensation and precipitation) are all highly discontinuous in space and time and are heavily parameterized in AOGCMs. Such model inaccuracies are a commonly recognized source of uncertainty in climate prediction. However, in practice it is difficult to judge the effect of such model weaknesses in isolation from another source of uncertainty that provides a further potential reason for the wide range of model outcomes at regional scales: climate system unpredictability. Timothy D. Mitchell and Mike Hulme V 63 Climate system unpredictability We may generalize the problem of regional climate prediction to all natural systems and state that if the natural system bears the characteristics of a chaotic system6 then if (in a hypothetical situation) it evolved a number of times from slightly different initial conditions, it would exhibit different characteristics in each evolution. The different outcomes might be described in terms of a frequency distribution of possible outcomes. If we model such a natural system then we expect (and hope) that successive evolutions, with slightly different initial conditions, will also exhibit different characteristics. Our evaluation of the worth of the model will not be on the grounds of whether or not the model possesses a frequency distribution of possible outcomes, but of whether or not the model’s distribution corresponds to the distribution of the natural system. Since the climate system is commonly regarded as a chaotic system (Lorenz, 1963; 1993) we expect that multiple integrations of an AOGCM, with slightly different initial conditions, will yield a frequency distribution of possible outcomes. Figure 3 displays four separate evolutions of a single model where only the initial conditions varied between evolutions; the wide divergence between the outcomes illustrates the unpredictability inherent to the climate system. So we are not surprised at the occurrence of intermodel differences at regional scales; we would expect differences to occur even if the AOGCMs were identical. It follows that in the presence of such unpredictability, the argument that ‘there is a wide range of model outcomes, therefore the simulations are inaccurate’ (often with the thought that this is due to parameterizations or coarse resolutions) is a non sequitur. An unexplored possibility in such an argument is that the unpredictability of the climate system may support a frequency distribution of possible outcomes, and that the wide range of model outcomes may be due to accurate modelling of this distribution, rather than to inaccurate AOGCMs diverging from a single solution. Therefore it is premature to state that ‘models cannot predict on regional scales’ merely from the evidence reviewed in section II. Rather, we must ask whether the model distribution of outcomes is a good representation of the distribution of outcomes possessed (theoretically) by the natural system or whether it has been corrupted by poor parameterization schemes, by coarse AOGCM spatial resolution or by unrepresented external phenomena (such as vegetation feedbacks or solar variability). The extent to which the wide range of model outcomes is due to the unpredictability inherent to the climate system over against AOGCM deficiencies is still an open question. VI Global system unpredictability A second source of inherent uncertainty in regional climate prediction is ‘global system unpredictability’, which encompasses all the unpredictabilities bound up in the future external forcings imposed on the climate system. These external forcings may be anthropogenic or natural; some of the most important include anthropogenic greenhouse gas emissions, solar variability and volcanic eruptions. Any accurate climate prediction must take into account the effect of such external forcings on the climate system, and our second source of inherent uncertainty in regional climate 64 Predicting regional climate change: living with uncertainty Figure 3 The unpredictability inherent in the climate system. The time evolution (left to right) of overlapping 30-year mean climate anomalies for four ensemble members, relative to observations. The final years of the model periods are 2070–99; the last observed period is 1967–96. The model used is HadCM2, forced with historical changes in compounded greenhouse gases until 1990, and with increases of 1% per annum thereafter. The anomalies are of summer mean temperature and precipitation for the UK, using an average from three grid boxes centred over the UK Source: Modified after Hulme and Brown (1998) prediction is introduced by the fact that such forcings may themselves be unpredictable. In order to show the magnitude of this source of uncertainty, each of the three external forcings mentioned above is considered in turn. We cannot predict with certainty the future course of atmospheric greenhouse gas concentrations. The growth rate of concentrations (Keeling et al., 1995; Schimel et al., 1995) is unpredictable, despite the seasonal regularity of the trend in the concentration of atmospheric CO2. It is particularly the unpredictabilities of the global carbon cycle and the difficulties in constraining the carbon budget (reviewed by Joos, 1996) that make concentrations unpredictable. Yet even with a perfect understanding of the carbon cycle, we could not predict concentrations with certainty because of the unpredictability of anthropogenic greenhouse gas emissions; the IPCC describe the magnitude of this uncertainty in very strong terms: ‘It is our view that there is no objective method of assessing the likelihood of all the key assumptions that will influence future emission estimates . . . Thus we do not attempt to identify the most likely emission scenario.’ (Alcamo et al., 1995: 258). We have reasonable reconstructions of historical variations in anthropogenic greenhouse gas emissions, but not for solar variability; we are uncertain how the Timothy D. Mitchell and Mike Hulme 65 magnitude of solar forcing has varied in the past. The minimum magnitude of change we can conceive is the irradiance change (0.15%) in the 11-year Schwabe cycle that has been directly observed from satellites (Willson and Hudson, 1991) over the past two decades. The maximum conceivable magnitude was estimated by Reid (1997) to be 0.65%, by assuming that all the estimated global temperature variability since the Maunder Minimum of the seventeenth century may be attributed solely to solar variability and to anthropogenic greenhouse gas emissions. More conservative reconstructions of solar variability have been published by Hoyt and Schatten (1993) and by Lean et al. (1995), who calculate the difference in irradiance between the Maunder Minimum and the present to be 0.30% and 0.24%, respectively. If the historical variations in solar forcing are so uncertain – Reid describes the reconstructions as instances of ‘a highly speculative activity’ (1997: 392) – then it is difficult to evaluate the effects of solar variability on the climate system; indeed, Hoyt and Schatten admit on the cover of their work entitled The role of the sun in climate change (1997) that they cannot answer the question posed by the title. However, a succession of model experiments (Rind and Overpeck, 1993; Crowley and Kim, 1996) has suggested that solar variability is sufficient to force global climate changes of several tenths of a degree over centennial timescales. This succession has culminated in AOGCM experiments (Cubasch et al., 1997; Tett et al., 1998) that confirm the potential importance of solar variability, but suggest that solar forcing changes are insufficient by themselves to explain the temperature record of the twenthieth century. Since the historical record of solar variability is so uncertain, and since the historical effects of solar variability on the earth’s radiation budget are unknown, the solar forcing of climate in the future must be regarded as unpredictable. Although the climatic effects of certain volcanic eruptions are well known, the climatic forcing from volcanoes is unpredictable. Observations of the eruption plume of Mount Pinatubo in the Philippines (15 June 1991) demonstrated the atmospheric effects of a major eruption in a remarkable way (McCormick et al., 1995); the albedo changes to the atmosphere resulted in a peak decrease in August 1991 of 8 Wm–2 in radiative forcing between 5 °S and 5°N, and of 3 Wm–2 globally (Minnis et al., 1993). Despite the global cooling, this eruption – like others – produced a pronounced warming in the Northern Hemisphere (Robock and Mao, 1992; Parker et al., 1996) that illustrates the contrast in climatic effects that may appear between global and subglobal scales. Although Mount Pinatubo was well observed, it is difficult to reconstruct the history of volcanic eruptions; even recent volcanoes may escape observation (e.g., Sedlacek et al., 1981; Mroz et al., 1983) and we are reliant upon proxy measurements for identifying the dates of major eruptions prior to the commencement of satellite observations (e.g., Briffa et al., 1998). It is even more difficult to reconstruct the radiative forcing exerted upon the atmosphere by an eruption. The radiative forcing depends upon the nature of the injection of aerosols and aerosol precursors into the stratosphere, and upon the spatial and temporal patterns of their residence in the atmosphere. The location, altitude and period of injection all contribute to the former, and the particulate microphysics, particulate chemistry and atmospheric circulation all contribute to the latter. Whereas there is a comparatively wide range of sources of information for identifying the dates of eruptions, the scope for identifying the characteristics of the plumes from individual eruptions is severely limited by the requirement for a wide range of evidence (e.g., de Silva and Zielinski, 1998). Bradley and Jones (1992: 618) conclude their 66 Predicting regional climate change: living with uncertainty evaluation of volcanic forcing reconstructions as follows: ‘The record of large explosive eruptions since AD 1500 is probably quite incomplete, making it difficult to assess their overall impact on climate.’ Predicting the characteristics of future volcanic eruptions is even more difficult than reconstructing those of the past, and therefore the volcanic forcing of climate in the future is inherently unpredictable. VII The cascade of uncertainty The uncertainty concerning future regional climate change that results from climate model inaccuracies is commonly recognized. Yet the two sources of uncertainty that are inherent to regional climate prediction – the unpredictability of the climatic and global systems – have not been adequately dealt with in regional climate scenarios thus far, either in scenario construction, or in considering the environmental effects of the scenarios. Much of the research concerning the environmental effects of future climatic change may be categorized into the ‘timeless’ and the ‘single scenario’ approaches. The research summarized by the IPCC First assessment report (Houghton et al., 1990), and much of that summarized by the IPCC Second assessment report (Watson et al., 1996), considered the environmental impacts of a single physical event – such as the doubling of atmospheric concentrations of CO2 – in a ‘timeless’ approach that made no attempt to attach any time to the physical event. The introduction of transient perturbed climate change experiments has permitted a time dimension to be introduced to impact studies, which is incorporated into some of the Second assessment report and the majority of the IPCC Regional report (Watson et al., 1998). However, such impact studies are commonly limited to a ‘single scenario’, considering the environmental impacts at a number of points in time of only a single climate realization produced by forcing only a single AOGCM with only a single emissions scenario. Thus our understanding of the regional consequences of climate change is marked by the omission of any consideration of the uncertainties introduced by the unpredictability of the climatic and global systems. This neglect has permitted a ‘single result’ approach to prediction to permeate the climate change impacts literature, an approach that is condemned by Henderson-Sellers (1996: 60): ‘Since prediction is not a single result process, all predictions are (un)certain, contain and/or are based on, incomplete information and are debatable.’ The unpredictability of the climatic and global systems introduces a cascade of uncertainty to regional climate prediction (Figure 4) that cannot be squeezed to the single point of the ‘single result’ approach to prediction. VIII Three methodological responses Since unpredictability is a system characteristic, even with a perfect model the problem of regional climate prediction would still be characterized by uncertainty. To live with this uncertainty demands that we make three specific methodological responses: (1) the use of multiple forcing scenarios to cope with global system unpredictability; (2) the use of ensembles to cope with climate system unpredictability; and (3) the consideration of the entire response of the system to cope with the nature of climate change. We will comment briefly on each of these in turn. Timothy D. Mitchell and Mike Hulme 67 Figure 4 The magnitude of external radiative forcing is uncertain and so is presented as a range of possibilities in the top triangle. Two of these possibilities are selected as the starting points from which is presented the uncertainty concerning the magnitude of the climatic response. Thus we find that the cascade of uncertainties ultimately presents us with a wide variety of possible environmental futures at the base of the lowest triangles 1 Multiple scenarios Since the future course of anthropogenic emissions is inherently unpredictable, those responsible for the IS92 emissions scenarios made no attempt to identify one scenario as more likely than another, and made a specific recommendation to the contrary: ‘Given the degree of uncertainty about future emissions, we recommend that analysts use the full range of IS92 Scenarios rather than a single scenario as input to atmosphere/climate models’ (Alcamo et al., 1995: 255). (Since the anthropogenic emission of greenhouse gases is only one of a number of unpredictable external forcings, the full range of possible climatic forcings is even wider.) Yet – with a few exceptions – this clear recommendation has not been followed in climate prediction studies, and modellers have concentrated on a single emissions scenario, typically a 1% per annum increase in compounded greenhouse gases that corresponds approximately to the IS92a emissions scenario (Kattenberg et al., 1996: Table 6.3). Partly because of their dependence upon the limited range of model experiments, impacts researchers have taken their lead from the climate modellers and have also concentrated on the environmental consequences of a single emissions scenario: ‘Most commonly, the estimates are based upon changes in equilibrium climate that have been simulated to result from an equivalent doubling of carbon dioxide (CO2) in the atmosphere.’ (Watson et al., 1998: 4). Our uncertainty about climatic forcings is such that AOGCMs must be perturbed with multiple forcing scenarios to represent a wide range of possible forcings. Only then may we proceed to evaluate with any confidence the regional climatic effects of any future external forcing of the climate system. 68 2 Predicting regional climate change: living with uncertainty Ensembles The inherent unpredictability of the climate system is such that our understanding of what it means to ‘predict’ climate is somewhat different from the popular perception of prediction.7 Even if we were able to force a perfect model with a perfectly specified external forcing scenario, the model would not follow a single trajectory through time that might be calculated beforehand, in the manner of a frictionless pendulum. Rather, the dependence of the climate system on initial conditions means that inaccuracies in the initial state grow through time, so that there is a multiplicity of future states that are possible, even under a perfect forcing specification (Lorenz, 1963). This characteristic imposes limits of a few days on numerical weather prediction (James, 1994: 292–93), as our ability to forecast weather is dependent, not merely upon observational accuracy, computational power and the accuracy of the model, but more fundamentally upon the ‘chaotic’ nature of the system. For this reason an operational meteorologist will construct an ‘ensemble’ of simulations when making a forecast. The single GCM is integrated a number of times, the only difference between each ensemble ‘member’ being the initialization conditions. The aim of constructing the ensemble is to sample the full probability distribution of the future course of the weather. Likewise, if we are making predictions about future climate then we must accept that the climate system is unpredictable and that ensemble techniques should be used to sample the ‘full’ probability distribution of the future course of climate. An examination of Figure 3 illustrates the importance of the ensemble approach. If only one simulation had been performed with HadCM2 rather than four, then only a single climate scenario would have been available and a misleading impression would have been gained of the regional climatic response to the perturbation. 3 Entire response of the system We have addressed the two sources of uncertainty that are inherent to regional climate prediction, but we have not yet considered the nature of the regional climate changes that we expect to find. Such climate changes are conceptually different from daily changes in weather and so require a different kind of prediction. A change in the weather from Wednesday to Thursday may be understood in terms of a trajectory (the weather) in a fixed attractor8 (the coupled atmosphere–ocean system). A weather forecast is an instance of what Lorenz (1975) described as predictions ‘of the first kind’. It is an initial value problem, where we are given a set of initial conditions from Wednesday and we must calculate from them the future behaviour of the atmosphere in order to derive Thursday’s conditions. In contrast, we may understand a change in radiative forcing (perhaps from increased greenhouse gas concentrations or from changes in solar irradiance) as effecting a change in the attractor. We do not expect the attractor of the climate system to change in response to forcing in a linear manner, such as in a linear shift of the time-averaged state against an unchanged variability background (Figure 5); rather, we expect the system response to perturbation to depend upon the convolution of the forcing with the ‘natural’ variability of the system (Figure 6). A climate prediction is an attempt to predict how the attractor will change (a prediction ‘of the second kind’9 – Lorenz, 1975); in this sense any climate prediction must consider the entire response of the system. Timothy D. Mitchell and Mike Hulme 69 Figure 5 An erroneous linear-thinking perspective might suppose that a forcing (F) imposed on the Lorenz model (a) would produce an impact similar to (b) Source: Palmer (1993) IX Characterizing the entire response of the system Climate system predictions of the second kind may be examined from global or regional perspectives. These perspectives characterize the entire response of the system in different ways. It is worth considering this contrast between the global and the regional, not merely to avoid confusion, but to illustrate what is meant by considering the ‘entire response of the system’. In global climate analyses we often consider a single change characterized – in a highly simplified manner – as a linear shift. We treat the variability as ‘noise’, we treat the shift as ‘signal’ and we use noise reduction techniques to increase the signal-tonoise ratio so that we may claim ‘detection’ of a statistically significant shift (Santer et 70 Predicting regional climate change: living with uncertainty Figure 6 If a forcing (F) is imposed on the Lorenz model along the positive T-axis, then the (counterintuitive) impact is a decrease in T. This can be understood by noting that near the region of maximum instability, positive T forcing increases the probability that the system will evolve towards the regime whose centroid has the smaller value of T Source: Palmer (1993) al., 1996). One particular group of ‘detection studies’ adopting this approach (Santer et al., 1994; Hegerl et al., 1996; 1997) perturbed an AOGCM with increased atmospheric greenhouse gas concentrations, employed noise reduction methods to extract a pattern of response from the model output, and compared the pattern with patterns derived from observations. Such ‘fingerprint studies’ culminated in the tentative claim (Houghton et al., 1996) that anthropogenically induced climate change had been detected. The motive behind these detection studies was to encourage the adoption of mitigation policies by quantifying and confirming the conclusion that may be drawn from a simple consideration of the radiative properties of the global climate system (Arrhenius, 1896a; 1896b). Since the detection studies described above consider a single global response to external forcing, it is only at the continental, hemispheric and global scales that they may be expected to achieve the identification of a response (Stott and Tett, 1998). However, the information required for adaptive responses to climate change must be on regional scales. If we consider the response to radiative forcing of a particular region then we can no longer treat changes in the general circulations of the atmosphere and oceans as ‘noise’ to be disregarded in favour of a single linear shift, because that region’s climate is a result of those general circulations. It is through the general circulations that energy is redistributed in such a way as to maintain in equilibrium the zones of permanent radiation surplus and deficit, and any radiative forcing of the climate system ‘reaches’ a particular region through these dissipative mechanisms. So Timothy D. Mitchell and Mike Hulme 71 we are required to understand the behaviour of, and the effects of radiative forcing upon, the general circulations if we are to understand the consequent climatic changes in a particular region. In short, we need to understand the ‘entire response of the system’ in more detail than at the global scale. This explanation of regional climate change in terms of changes in the general circulations of the atmosphere and oceans is a ‘geographical’ equivalent to the ‘mathematical’ explanation, given above, in terms of a change in the climate system attractor. The general circulation changes correspond to the changes in the attractor. The general circulations and the attractor incorporate a natural variability that extends from seconds to millennia, and the trajectory through the attractor is inherently unpredictable. The regional climate change problem is – assuming that the radiative forcing is perfectly specified – to elucidate how the general circulations will change or, expressed alternatively, to elucidate how the shape of the climate system attractor will change. It is in this sense that we need to understand the ‘entire response of the system’. However, on the interdecadal timescales of interest for climate prediction we hardly understand the current behaviour of the general circulations – Latif (1998) provides a good review – let alone the effects of external forcing upon them. The position is summed up well by Anderson and Willebrand (1996: v) in their work on decadal climate variability: ‘The causes of natural variability of the climate system on decadal time scales are presently not well known; understanding the mechanisms is however a prerequisite for any climate prediction of changes on these time scales.’ X Some examples The discussion in this review has necessarily been of a rather theoretical nature, and it is appropriate to introduce some examples. A brief consideration of the thermohaline circulation (THC) of the North Atlantic serves to illustrate a number of points. In 1961 Stommel reported a very simple experiment with profound results that was forgotten until the 1980s. He considered the convective behaviour between two interconnected reservoirs forced by density differences maintained by salt and heat transfers and found that two distinct stable regimes were possible. Building on this work, Welander (1986) later presented an assortment of thermohaline oscillators and multiple steady states applicable to the world’s oceans. So multiple steady states are conceivable under natural variability. More recently, researchers have shown that anthropogenic external forcing may affect the behaviour of the THC to the extent of inducing a switch between one climatic state and another. Manabe and Stouffer (1993) chose two different emissions scenarios with which to force a coupled model, and examined the response in the North Atlantic THC; they found two distinct stable regimes, the occupancy of which depended on the magnitude of the equilibrium change in atmospheric CO2 concentration. In a similar experiment, Stocker and Schmittner (1997) showed that the stable regime obtained is also dependent upon the rate of change of atmospheric CO2 concentration. Rahmstorf (1995) conducted a more general examination of the hysteresis behaviour of the North Atlantic THC and found a number of forms of nonlinear behaviour. Schiller et al. (1997) considered the stability of the North Atlantic THC against meltwater input and found that four elements of the climate system exhibited feedbacks with the THC: oceanic heat 72 Predicting regional climate change: living with uncertainty transport, precipitation and runoff, the atmospheric circulation and the wind-driven circulation. This nonlinear behaviour of the North Atlantic THC on decadal and centennial timescales illustrates the unpredictability of the coupled ocean–atmosphere system that was discussed in section V. It suggests that climate predictions for regions dependent upon the North Atlantic THC (principally Europe) must be made on the understanding that an external forcing may result in a number of possible outcomes (section VII). Since the behaviour of the North Atlantic THC depends on the magnitude and rate of external forcing, any predictions should be made on the basis of AOGCM simulations in which the model is forced with multiple forcing scenarios (section VIII 1). Since the behaviour of the North Atlantic THC depends on the initial conditions, an ensemble experiment should be performed for each forcing scenario (section VIII 2). Since any external forcing affects the North Atlantic THC through the atmosphere and continental ice sheets, and since the THC itself influences the atmosphere and continental ice sheets, the North Atlantic THC response to forcing can only be understood in terms of the entire response of the system to forcing (sections VIII 3 and IX). Our second example concerns the application of ensemble methods that (as we suggested in section VIII 2) allow us to deal with climate system unpredictability in regional climate prediction. There are not yet many instances of ensemble techniques being used in the climate change problem, chiefly because of the expense of performing multicentury AOGCM experiments. (Therefore it is also worth considering the use of ensemble techniques in the seasonal prediction problem – Dix and Hunt, 1995; Rowell, 1998.) The Max-Planck Institute (Cubasch et al., 1997) have performed a two-member ensemble experiment that is relevant to the climate change problem, in which ECHAM3 was forced with an estimated historical record of solar variability. The Hadley Centre have performed a wider range of four-member ensemble experiments in which HadCM2 was forced with estimated historical and future atmospheric CO2 and sulphate concentrations (Mitchell et al., 1999), and with estimated historical records of solar and volcanic variability (Tett et al., 1998). An example of the uses to which the Hadley Centre ensembles have been put is provided by the analysis on the Northern Hemisphere storm tracks in general (Carnell and Senior, 1998), on the North Atlantic Oscillation in particular (Osborn et al., 1999), and on water and crop responses to climate change in Europe (Hulme et al., 1999). XI Conclusion This discussion of the uncertainty bound up with regional climate prediction has many implications for research, both of the climate system itself and of the impacts of climate change. It has been shown that even with a perfect model, uncertainty would remain inherent to regional climate prediction, because of the inherent unpredictabilities of external forcing – both anthropogenic and natural – and of the climate system itself. Only if we were able to alter the fundamental natures of the climate and global systems to make them predictable might we hope to predict with certainty. It seems that if we are to be successful in our aim of predicting regional climate change then we must understand that uncertainty must arise from systemic unpredictability. The first implication of this insight is that to avoid misunderstanding we must adopt Timothy D. Mitchell and Mike Hulme 73 the precise terminological language that contrasts unpredictability (a system characteristic) with uncertainty (what we have to live with as a consequence of unpredictability), and that treats ‘prediction’ in the context of inherent uncertainty. A further implication is that we must not treat any model output as the one true climate prediction, but must take the different – and equally probable – climate scenarios illustrated in Figure 3 and use them as equally probable inputs when researching the impacts of climate change. The uncertainties involved complicate the communication of vital information, making scientists instinctively wary of making any assessment of the changes that may follow anthropogenic forcing. So when pressed for information, scientists are faced with a classic dilemma: predict and risk being wrong, or remain silent and risk failing to release useful information. Fowler and Hennessy (1995: 284) discuss the problem for the specific issue of extreme precipitation events and are convinced of the need for scientists to be vocal: Although there are good reasons for scientific reticence concerning possible changes . . . there is also a pressing need for information . . . In the absence of contrary advice from atmospheric scientists, the norm will prevail of assuming that past experience is a reliable guide. Given that the implicit assumption of a stationary climate underlying such an approach contradicts expectations of a rapidly changing climate over the next several decades, clearly there is an onus on atmospheric scientists to provide as much information as possible. The inherent uncertainty in regional climate prediction is such that scientists must counter the damaging norm of assuming a stationary climate. In the past, conventional understanding has adopted a 30-year mean as the one true climate and this human artifact has been used as the basis upon which return periods are calculated, infrastructure is designed and policy decisions are reached. Thus humans are only adapted to the latest 30-year mean, rather than to true levels of natural variability. To be in a position to adapt to the climatic changes that will follow any future anthropogenic forcing of the climate system, we must first be better adapted to the levels of variability present in an unperturbed climate system. Achieving this requires a better understanding both of climatic variability and of the sensitivity of natural environments and human societies to that climatic variability. The problem for researchers is to close the gap between the demand for, and supply of, reliable information on regional climate change and its impacts. At this point it is helpful to make a distinction between the information necessary for mitigation rather than for adaptation by policy decisions. Since CO2 emissions are rapidly mixed in the atmosphere, the climate system response to CO2 emissions is spatially global in scale. So mitigation policy, which seeks to reduce CO2 emissions, requires information about the global-scale response to CO2 emissions (the existence of a ‘shift’ in the climate system attractor); the ‘detection’ work described in section IX has sufficiently supplied that requirement for the global community to sign the UN/FCCC and the Kyoto Protocol. In contrast, adaptive responses to climate change require information on regional scales. Yet regional climate prediction is still in its infancy, so a large gap remains between the demand for, and supply of, reliable information. There is also a contrast between the objects of the mitigative and adaptive responses. The adaptive strategies require that a response be made to the climate system’s behaviour, whether the system is being forced by humans or by natural phenomena. Mitigative strategies, however, can only deal with the anthropogenic forcing of the climate system – unless we are so bold as to contemplate altering solar variability or volcanic eruptions. 74 Predicting regional climate change: living with uncertainty This review has presented the cascade of uncertainties (Figure 4) as the reality with which our methodologies must live. None the less we recognize that there are two problems with limiting the regional climate prediction problem to the resolution of this relatively simple cascade of uncertainties. The first problem is the presence of feedbacks between most of the nonhuman elements of the cascade. This is well exemplified by the exchanges between the coupled atmosphere–ocean–cryosphere system (simulated by AOGCMs) and the land surface, such as the posited vegetation feedbacks in the Sahel (reviewed by Nicholson, 1988) that may be partly responsible for the prolonged droughts of recent decades (Eltahir and Gong, 1996). Another example refers to the potential, described in section X, for multiple steady states in the thermohaline circulation. Rahmstorf (1995) has shown with a coupled model that it is conceivable that precipitation changes may change freshwater inputs into the oceans, triggering convective instability, and thus inducing transitions between different equilibrium states of the thermohaline circulation, with substantial climatic effects. The second problem is the lack of attention to humans. The human world is unpredictable; only the most extreme reductionist can envisage the predictability of human thought and behaviour. Since humans affect the climate system and the consequences in turn affect humans, we cannot completely account for all the interactions that are of relevance to regional climate prediction. Of course, we try to include all the possibilities of human influence on the climate system through the emissions scenarios, but this neglects the feedbacks between humans and their environment. Integrated assessment (Peck and Teisberg, 1992; Nordhaus, 1994; Hasselmann et al., 1997) is another venture still in its infancy, but in the long run it may make an invaluable contribution to unravelling the web of feedbacks and to managing global change. Whatever the contribution made by integrated assessment, we must still however carefully examine each environmental system for its response to regional climate change, with a keen awareness of the inherent uncertainties involved. If we are to adapt sensitively to climate change then we require information now; the many uncertainties involved should not stop us from providing the best information possible. The heart of the task of regional climate prediction lies in bringing to light the different possibilities the future may hold for us; we must elucidate the climatic and environmental changes that are possible and enlighten nonscientists in ways that communicate the inherent uncertainties of regional climate change. Acknowledgements Tim Mitchell is supported by a grant from the Natural Environment Research Council (GT 04/97/81/MAS) in conjunction with the UK Meteorological Office (Met 1b/2437), but the views expressed in this article are solely those of the authors, not of NERC or UKMO. We are grateful to Dr Declan Conway for his comments on a draft of this article, and to Dr John Mitchell for his support and constructive discussions. Notes 1. In this review ‘regional’ refers to subcontinental and local scales. 2. In current terminology ‘adaptation’ measures are adaptive responses to an actual climate Timothy D. Mitchell and Mike Hulme 75 change, past, present or future; ‘mitigation’ measures are responses to a perceived potential future climate change that attempt to reduce the likelihood of that change occurring. An example of the former is building more coastal defences; and an example of the latter is reducing greenhouse gas emissions. 3. In this review any agent or system beyond the scope of the atmosphere, oceans and cryosphere is considered to be ‘external’ to the climate system. Thus external forcings on the climate system include both anthropogenic and natural forcings, exemplified by CO2 emissions and solar variability, respectively. 4. A control simulation is conducted without any ‘perturbation’ from humans, the sun or any other external agent. 5. The term ‘scenario’ has been much abused (see Henderson-Sellers, 1996). In this review an ‘emissions scenario’ refers to a description of a possible future temporal pattern of anthropogenic emissions of greenhouse gases; a ‘climate scenario’ refers to a description of a possible future climate. 6. An introduction to deterministic chaos is beyond the scope of this review, and the reader is referred to Hall’s (1991) introduction to chaos and Lorenz’s (1993) introduction to chaos with its particular reference to the atmosphere. 7. Palmer (1993) provides a good introduction to these predictability concepts. 8. The climate system ‘attractor’ may be loosely defined as the complete set of possible instantaneous states of the climate system, or more precisely as ‘a limit set that is not contained in any larger limit set, and from which no orbits emanate’ (Lorenz, 1993: 206). 9. These two ‘kinds’ of prediction – discussed at greater length by Palmer (1996) – are fundamentally different, and it is crucial to understand this difference when interpreting GCM output, especially since the same GCM may be used for both kinds of prediction (Cullen, 1993). References Alcamo, J., Bouwman, A., Edmonds, J., Grubler , A., Morita, T. and Sugandhy, A. 1995: An evaluation of the IPCC IS92 emission scenarios. In Houghton, J.T., Meira Filho, L.G., Bruce, J., Hoesung Lee, B.A., Callander, B.A., Haites, E., Harris, N. and Maskell, K., editors, Climate change 1994: radiative forcing of climate change and an evaluation of the IPCC IS92 emission scenarios, Cambridge: Cambridge University Press. Anderson, D.L.T. and Willebrand, J. 1996: Preface. In Anderson, D.L.T. and Willebrand, J., editors, Decadal climate variability: dynamics and predictability. NATO ASI Series. Vol. I 44, Berlin: Springer-Verlag. Arrhenius, S. 1896a: Ueber den Einfluss des Atmosphärischen Kohlensäuregehalts auf die Temperatur der Erdoberfläche. Proceedings of the Royal Swedish Academy of Sciences 22. ––––– 1896b: On the influence of carbonic acid in the air upon the temperature of the ground. The London, Edinburgh and Dublin Philosophical Magazine and Journal of Science 41, 237–76. Bengtsson, L. 1995: On the prediction of regional climate change. In 3rd International Conference on Modelling of Climate Change and Variability, Max Planck Institut fur Meteorologie, 4–8 September. Bolin, B., Watson, R.T., Zinyowera, M.C., Sundararaman, N. and Moss, R.H. 1998: Preface to Watson, R.T., Zinyowera, M.C., Moss, R.H. and Dokken, D.J., editors, The regional impacts of climate change: an assessment of vulnerability, Cambridge: Cambridge University Press. Bradley, R.S. and Jones, P.D. 1992: Records of explosive volcanic eruptions over the last 500 years. In Bradley, R.S. and Jones, P.D., editors, Climate since AD 1500, London: Routledge, London. Briffa, K.R., Jones, P.D., Schweingruber, F.H. and Osborn, T.J. 1998: Influence of volcanic eruptions on Northern Hemisphere summer temperature over the past 600 years. Nature 393, 450–55. Bürger, G. 1996: Expanded downscaling for generating local weather scenarios. Climate Research 7, 111–28. Carnell, R.E. and Senior, C.A. 1998: Changes in mid-latitude variability due to increasing greenhouse gases and sulphate aerosols. Climate Dynamics 14, 369–83. Covey, C. 1995: Using paleoclimates to predict future climate: how far can analogy go? An editorial comment. Climatic Change 29, 403–407. 76 Predicting regional climate change: living with uncertainty Crowley, T.J. 1997: Commentary: the problem of paleo-analogs. Climatic Change 35, 119–21. Crowley, T.J. and Kim, K.-Y. 1996: Comparison of proxy records of climate change and solar forcing. Geophysical Research Letters 23, 359–62. Cubasch, U., Voss, R., Hegerl, G.C., Waszkewitz, J. and Crowley, T.J. 1997: Simulation of the influence of solar radiation variations on the global climate with an ocean–atmosphere general circulation model. Climate Dynamics 13, 757–67. Cullen, M.J.P . 1993: The unified forecast/climate model. Meteorological Magazine 122, 81–94. Deque, M. and Piedelievre, J.Ph. 1995: High resolution climate simulation over Europe. Climate Dynamics 11, 321–29. de Silva, S.L. and Zielinski, G.A. 1998: Global influence of the AD 1600 eruption of Huaynaputina, Peru. Nature 393, 455–58. Dix, M.R. and Hunt, B.G. 1995: Chaotic influences and the problem of deterministic seasonal prediction. International Journal of Climatology 15, 729–52. Eltahir, E.A.B. and Gong, C. 1996: Dynamics of wet and dry years in west Africa. Journal of Climate 9, 1030–42. Fowler, A.M. and Hennessy, K.J. 1995: Potential impacts of global warming on the frequency and magnitude of heavy precipitation. Natural Hazards 11, 283–303. Fox-Rabinovitz, M.S., Stenchikov, G.L., Suarez, M.J. and Takacs, L.L. 1997: A finite-difference GCM dynamical core with a variable-resolution stretched grid. Monthly Weather Review 125, 2943–68. Gates, W.L., Henderson-Sellers, A., Boer, G.J., Folland, C.K., Kitoh, A., McA vaney, B.J., Semazzi, F., Smith, N., Weaver, A.J. and Zeng, Q.-C. 1996: Climate models – evaluation. In Houghton, J.T., Meira Filho, L.G., Callander, B.A., Harris, N., Kattenberg, A. and Maskell, K., editors, Climate change 1995: the science of climate change, Cambridge: Cambridge University Press. Gates, W.L., Rowntree, P.R. and Zeng, Q.-C. 1990: Validation of climate models. In Houghton, J.T., Jenkins, G.J. and Ephraums, J.J., editors, Climate change: the IPCC scientific assessment, Cambridge: Cambridge University Press. Giorgi, F. 1990: Simulation of regional climate using a limited area model nested in a general circulation model. Journal of Climate 3, 941–63. Hall, N., editor, 1991: The New Scientist guide to chaos, Harmondsworth: Penguin Books. Hasselmann, K., Hasselmann, S., Giering, R., Ocana, V. and von Storch, H. 1997: Sensitivity study of optimal CO2 emission paths using a simplified structural integrated assessment model. Climatic Change 37, 345–86. Hecht, A.D. and Tirpak, D. 1995: Framework agreement on climate change: a scientific and policy history. Climatic Change 29, 371–402. Hegerl, G.C., Hasselmann, K., Cubasch, U., Mitchell, J.F.B., Roeckner, E., Voss, E. and Waszkewitz, J. 1997: Multi-fingerprint detection and attribution analysis of greenhouse gas-plus-aerosol and solar forced climate change. Climate Dynamics 13, 613–34. Hegerl, G.C., von Storch, H., Hasselmann, K., Santer, B.D., Cubasch, U. and Jones, P.D. 1996: Detecting greenhouse-gas-induced climate change with an optimal fingerprint method. Journal of Climate 9, 2281–306. Henderson-Sellers, A. 1996: Can we integrate climatic modelling and assessment? Environmental Modeling and Assessment 1, 59–70. Hewitson, B.C. and Crane, R.G. 1996: Climate downscaling: techniques and application. Climate Research 7, 85–95. Houghton, J.T., Jenkins, G.J. and Ephraums, J.J., editors, 1990: Climate change: the IPCC scientific assessment, Cambridge: Cambridge University Press. Houghton, J.T., Meira Filho, L.G., Callander , B.A., Harris, N., Kattenberg, A. and Maskell, K., editors, Climate change 1995: the science of climate change, Cambridge: Cambridge University Press. Hoyt, D.V. and Schatten, K.H. 1993: A discussion of plausible solar irradiance variations, 1700–1992. Journal of Geophysical Research 98, 18895–906. ––––– 1997: The role of the sun in climate change. New York: Oxford University Press. Hulme, M., Barrow, E.M., Arnell, N.W ., Downing, T.E., Harrison, P.A. and Johns, T.C. 1999: Assessing environmental impacts of anthropogenic climate change: distinguishing signal from noise. Nature, in press. Hulme, M. and Brown, O. 1998: Portraying climate scenario uncertainties in relation to tolerable regional climate change. Climate Research 10, 1–14. James, I.N. 1994: Introduction to circulating atmospheres, Cambridge: Cambridge University Press. Joos, F. 1996: The atmospheric carbon dioxide perturbation. Europhysics News 27, 213–18. Timothy D. Mitchell and Mike Hulme Joubert, A.M. and Hewitson, B.C. 1997: Simulating present and future climates of southern Africa using general circulation models. Progress in Physical Geography 21, 51–78. Kattenberg, A., Giorgi, F., Grassl, H., Meehl, G.A., Mitchell, J.F.B., Stouffer, R.J., Tokioka, T., Weaver, A.J. and Wigley, T.M.L. 1996: Climate models – projections of future climate. In Houghton, J.T., Meira Filho, L.G., Callander, B.A., Harris, N., Kattenberg, A. and Maskell, K., editors, Climate change 1995: the science of climate change, Cambridge: Cambridge University Press. Keeling, C.D., Whorf, T.P., Wahlen, M. and van der Plicht, J. 1995: Interannual extremes in the rate of rise of atmospheric carbon dioxide since 1980. Nature 375, 666–70. Kim, J.-W., Chang, J.-T., Baker, N.L., Wilks, D.S. and Gates, W.L. 1984: The statistical problem of climate inversion: determination of the relationship between local and large-scale climate. Monthly Weather Review 112, 2069–77. Kittel, T.G.F., Giorgi, F . and Meehl, G.A. 1998: Intercomparison of regional biases and doubled CO2-sensitivity of coupled atmosphere–ocean general circulation model experiments. Climate Dynamics 14, 1–15. Knutson, T.A. and Manabe, S. 1994: Impact of increased CO2 on simulated ENSO-like phenomena. Geophysical Research Letters 21, 2295–98. Latif, M. 1998: Dynamics of interdecadal variability in coupled ocean–atmosphere models. Journal of Climate 11, 602–24. Lean, J., Beer, J. and Bradley, R. 1995: Reconstruction of solar irradiance since 1610: implications for climate change. Geophysical Research Letters 22, 3195–98. Lorenz, E.N. 1963: Deterministic nonperiodic flow. Journal of the Atmospheric Sciences 20, 130–41. ––––– 1975: Climate predictability. The physical basis of climate modelling. WMO, GARP Publication Series 16, 132–36. ––––– 1993: The essence of chaos. London: UCL Press. Manabe, S. and Stouffer, R.J. 1993: Century-scale effects of increased atmospheric CO2 on the ocean–atmosphere system. Nature 364, 215–18. Masood, E. 1997: Kyoto agreement creates new agenda for climate research. Nature 390, 649–50. McCormick, M.P., Thomason, L.W . and Trepte, C.R. 1995: Atmospheric effects of the Mt Pinatubo eruption. Nature 373, 399–404. 77 Minnis, P., Harrison, E.F., Stowe, L.L., Gibson, G.G., Denn, F.M., Doelling, D.R. and Smith, W.L. Jr 1993: Radiative climate forcing by the Mount Pinatubo eruption. Science 259, 1411–15. Mitchell, J.F.B., Johns, T.C., Eagles, M., Ingram, W.J. and Davis, R.A. 1999: Towards the construction of climate change scenarios. Climatic Change, in press. Mroz, E.J., Mason, A.S. and Sedlacek, W.A. 1983: Stratospheric sulphate from El Chichon and the mystery volcano. Geophysical Research Letters 10, 873–76. Nicholson, S.E. 1988: Land surface atmosphere interaction: physical processes and surface changes and their impact. Progress in Physical Geography 12, 36–65. Nordhaus, W .D. 1994: Managing the global commons: the economics of climate change, Cambridge, MA: MIT Press. Obasi, G.O.P. and Dowdeswell, E. 1998: Foreword to Watson, R.T., Zinyowera, M.C., Moss, R.H. and Dokken, D.J., editors, The regional impacts of climate change: an assessment of vulnerability, Cambridge: Cambridge University Press. Osborn, T.J., Briffa, K.R., Tett, S.F.B., Jones, P.D. and Trigo, R.M. 1998: Evaluation of the North Atlantic Oscillation as simulated by a climate model. Climate Dynamics, in press. Palmer, T.N. 1993: A nonlinear dynamical perspective on climate change. Weather 48, 314–26. ––––– 1996: Predictability of the atmosphere and oceans: from days to decades. In Anderson, D.L.T. and Willebrand, J., editors, Decadal climate variability, Berlin: Springer, Berlin. Parker, D.E., Wilson, H., Jones, P.D., Christy, J.R. and Folland, C.K. 1996: The impact of Mt Pinatubo on world-wide temperatures. International Journal of Climatology 16, 487–97. Peck, S.C. and Teisberg, T .J. 1992: CETA: A model for carbon emissions trajectory assessment. Energy Journal 13, 55–77. Rahmstorf, S. 1995: Bifurcations of the Atlantic thermohaline circulation in response to changes in the hydrological cycle. Nature 378, 145–49. Reid, G.C. 1997: Solar forcing of global climate change since the mid-17th century. Climatic Change 37, 391–405. Rind, D. and Overpeck, J. 1993: Hypothesised causes of decade-to-century climate variability: climate model results. Quaternary Science 78 Predicting regional climate change: living with uncertainty Reviews 12, 357–74. Robock, A. and Mao, J. 1992: Winter warming from large volcanic eruptions. Geophysical Research Letters 12, 2405–408. Rowell, D.P . 1998: Assessing potential seasonal predictability with an ensemble of multidecadal GCM simulations. Journal of Climate 11, 109–20. Santer, B.D., Bruggemann, W., Cubasch, U., Hasselmann, K., Hock, H., Maier-Reimer, E. and Mikolajewicz, U. 1994: Signal-to-noise analysis of time-dependent greenhouse warming experiments. Climate Dynamics 9, 267–85. Santer, B.D., Taylor, K.E., Wigley, T.M.L., Johns, T.C., Jones, P.D., Karoly, D.J., Mitchell, J.F.B., Oort, A.H., Penner, J.E., Ramaswamy, V ., Schwarzkopf, M.D., Stouffer, R.J. and Tett, S.F.B. 1996: A search for human influences on the thermal structure of the atmosphere. Nature 384, 39–46. Schiller, A., Mikolajewicz, U. and Voss, R. 1997: The stability of the North Atlantic thermohaline circulation in a coupled ocean–atmosphere general circulation model. Climate Dynamics 13, 325–47. Schimel, D., Enting, I., Heimann, M., Wigley , T.M.L., Raynaud, D., Alves, D. and Siegenthaler, U. 1995: CO2 and the carbon cycle. In Houghton, J.T., Meira Filho, L.G., Bruce, J., Hoesung Lee, B.A., Callander, B.A., Haites, E., Harris, N. and Maskell, K., editors, Climate change 1994: radiative forcing of climate change and an evaluation of the IPCC IS92 emission scenarios, Cambridge: Cambridge University Press. Schneider , E.K. and Kinter , J. 1994: An examination of internally generated variability in long climate simulations. Climate Dynamics 10, 181–204. Schulze, R.E. 1997: Impacts of global climate change in a hydrologically vulnerable region: challenges to South African hydrologists. Progress in Physical Geography 21, 113–36. Sedlacek, W.A., Mroz, E.J. and Heiken, G. 1981: Stratospheric sulfate from the Gareloi eruption, 1980: contribution to the ‘ambient’ aerosol by a poorly documented volcanic eruption. Geophysical Research Letters 8, 761–64. Stocker, T.F . and Schmittner, A. 1997: Influence of CO2 emission rates on the stability of the thermohaline circulation. Nature 388, 862–65. Stommel, H. 1961: Thermohaline circulation with two stable regimes of flow. Tellus 13, 224–41. Stott, P.A. and Tett, S.F.B. 1998: Scale-dependent detection of climate change. Journal of Climate 11, 3282–94. Tett, S.F.B., Johns, T.C. and Mitchell, J.F.B. 1997: Global and regional variability in a coupled AOGCM. Climate Dynamics 13, 303–23. Tett, S.F.B., Stott, P.A., Allen, M.R., Ingram, W.J. and Mitchell, J.F.B. 1998: Causes of twentieth century temperature change. Submitted. von Storch, J.S. 1994: Interdecadal variability in a global coupled model. Tellus 46A, 419–32. Watson, R.T., Zinyowera, M.C., Moss, R.H. and Dokken, D.J., editors, 1996: Climate change 1995: impacts, adaptations and mitigation of climate change: scientific-technical analyses. Cambridge: Cambridge University Press. ––––– editors, 1998: The regional impacts of climate change: an assessment of vulnerability. Cambridge: Cambridge University Press. Welander, P. 1986: Thermohaline effects in the ocean circulation and related simple models. In Willebrand, J.T. and Anderson, D.L.T., editors, Large-scale transport processes in oceans and atmosphere, Dordrecht: Reidel. Wilby, R.L. 1994: Stochastic weather type simulation for regional climate change impact assessment. Water Resources Research 30, 3395–403. Wilby, R.L. and Wigley, T.M.L. 1997: Downscaling general circulation model output: a review of methods and limitations. Progress in Physical Geography 21, 530–48. Wilks, D.S. 1992: Adapting stochastic weather generation algorithms for climate change studies. Climate Change 22, 67–84. Willson, R.C. and Hudson, H.S. 1991: The sun’s luminosity over a complete solar cycle. Nature 351, 42–44.