Anthropogenic climate change has gained much visibility in both scientific circles and public discourse over the last two decades. Climate scenarios can claim some credit in raising this awareness by helping scientists, policy-makers and lay people imagine what future climate may look like. Increasingly, climate scenarios are being used in long-term planning to help society adapt to a changing climate. In a separate paper (Hulme and Dessai, 2008b), we have postulated that the success of scenarios can be evaluated on the basis of: predictive success (has the future turned out as envisaged), decision success (have decisions subsequently made turned out to be robust) and learning success (have scenarios engendered participation and learning). In this paper, we evaluate predictive success by comparing recent observations with projections from four generations of national UK climate scenarios published between 1991 and 2002.
Climate scenarios portray how the future climate may evolve under a given set of assumptions about world development, parameterisations of climate models and various other factors. The Intergovernmental Panel on Climate Change (IPCC) Third Assessment Report (TAR) devoted an entire chapter to the development of climate scenarios (Mearns et al., 2001). They defined a climate scenario as a plausible future climate that has been constructed for explicit use in investigating the potential consequences of anthropogenic climate change. Climate scenarios often make use of climate projections (descriptions of the modelled response of the climate system to scenarios ofgreenhouse gas (GHG) and aerosol concentrations), by manipulating model outputs and combining them with observed climate data. The more recent Fourth Assessment Report (AR4) of the IPCC defines a climate scenario as a plausible and often simplified representation of the future climate, based on an internally consistent set of climatological relationships and assumptions of radiative forcing, typically constructed for explicit use as input to climate change impact models(IPCC, 2007a,b).
Climate models are often evaluated (sometimes called validation) by comparing a model simulation of a control period (e.g. 1961–1990) with observations using various possible metrics (e.g. seasonal temperature and precipitation, Northern Atlantic Oscillation Index, weather types). There is no agreed universal standard by which to evaluate climate models (cf. Phillips et al., 2006; Randall et al., 2007), therefore expert judgment, pragmatic reasoning and political considerations are often used to select which climate model projections are used in the construction of national climate scenarios (Hulme and Dessai, 2008a). Because of the time horizons of climate projections—in the order of decades up to a century and sometimes beyond—climate scenarios have rarely been compared with observations. Weather forecasts of daily to seasonal time scales are amenable to verification; once the forecast has been produced it is possible to assess how well the model has performed the following day or season (for seasonal climate see, for example, Troccoli et al., 2008). For climate scenarios this form of verification is not feasible because of the long time scales involved.
More recently, early climate change projections of global temperature change have been compared with observations (Hansen et al., 2006). Similarly, IPCC TAR projections of carbon dioxide concentrations, global mean air temperature and global sea level have been compared with recent observations (Rahmstorf et al., 2007). The IPCC AR4 compared the model projections of global warming (from previous assessment reports) with observed warming, providing broad confidence in such short-term global projections (IPCC, 2007b). Pielke (2008) compared global mean temperature projections from all IPCC assessment reports with observational surface and satellite data. Here, we compare a sequence of regional climate scenarios for the UK with recent observations.
The UK has been at the forefront of regional climate scenario development, with the publication of four generations of national climate change scenarios between 1991 and 2002. The scenarios were developed in parallel with the various IPCC assessment reports. Given the scientific advancements reported in IPCC AR4, a new set of scenarios is scheduled to be published in spring 2009. It is important to note that climate scenarios are not the direct result of a climate modelling experiment, but the product of a negotiation between scientists, policy-makers and diverse stakeholders in society (Hulme and Dessai, 2008a explain the process of climate scenario construction in the UK).
The comparison of climate scenarios with recent observations is a very different exercise to evaluating the performance of a climate model simulation to past observations (e.g. Airey and Hulme, 1995) or to a detection and attribution exercise (e.g. Karoly and Stott, 2006). It is different to climate model evaluation or validation because the comparison with observations is made with the climate scenarios and not with the underlying climate model simulations. It is a worthy endeavour because a ‘good’ climate model does not necessarily translate into climate scenarios of equal skill or performance because of the assumptions and manipulations that take place in the climate scenario construction process (Hulme and Dessai, 2008a). This paper is also distinct from detection and attribution exercises since we are not trying to find an anthropogenic signal in recent observations. The purpose of this paper is to assess whether successive national UK climate scenarios have been reflective of the actual evolving climate. This is important because these scenarios are increasingly used to climate-proof multi-million pound investments and to develop new risk and resource management strategies.
2. Data and methods
The monthly Central England Temperature (CET) record (Parker and Horton, 2005, and references therein) and the England and Wales Precipitation (EWP) record (Alexander and Jones, 2001, and references therein) are used to describe recent changes in the British climate. Monthly mean CET begins in 1659 and is representative of a wide area of the country centred on the English Midlands. Monthly mean EWP starts in 1766 and is representative of the land area of England and Wales. Both CET and EWP have been shown to be representative of the UK climate with CET having larger spatial correlations than EWP, but even the latter was highly representative for the most part (Jones and Hulme, 1997; Croxton et al., 2006). Given the length of these two records, it is possible to estimate an approximation of ‘natural’ climate variability. For multi-decadal climate variability this was done by calculating two standard deviations from a series of non-overlapping 30-year means from 1660–1689 to 1900–1929 for CET and from 1767–1796 to 1917–1946 for EWP (cf. Hulme and Brown, 1998). For inter-annual climate variability we used yearly data from 1660 to 1950 for CET and from 1767 to 1950 for EWP. The year 1950 was selected as the cut-off in order to avoid significant anthropogenic influence in the estimation of natural variability. This year was chosen because of the increasing predominance of GHG forcing over the last 60 years in both global (IPCC, 2007b) and CET (Karoly and Stott, 2006) warming.
The climate change scenario information was extracted from the scientific reports (CCIRG, 1991, 1996; Hulme and Jenkins, 1998; Hulme et al., 2002) and the UK Climate Impacts Programmes (UKCIP) website (http://data.ukcip.org.uk/UKCIP02_dataarchive/). Table 1 summarises the major characteristics of the four sets of climate scenarios. CCIRG91 and CCIRG96 each presented one climate scenario, whereas UKCIP98 and UKCIP02 each presented four. UKCIP98 included uncertainties in GHG emissions and climate sensitivity, whereas UKCIP02 only sampled uncertainties in GHG emissions. Each set of climate scenarios was developed after significant advances in climate modelling (from the first atmospheric general circulation models (GCMs) in the 1980s to the first regional climate model results in the late 1990s), paralleled by increases in spatial and temporal resolution (from 500 to 50 km and from seasonal averages to daily changes).
Table 1. Summary characteristics of the four generations of UK climate scenarios
Number of scenarios
1.5% p.a. growth in GHG concentration
IPCC IS92a emissions scenario
0.5 and 1.0% p.a. growth in GHG concentration
IPCC SRES scenarios: B1, B2, A2, A1FI
2.5 °C, plus IPCC range of 1.5 °C and 4.5 °C
Five atmospheric GCMs (including UKLO)
One atmosphere-ocean GCM, HadCM1 (plus a table with 11 other GCMs of various designations)
One atmosphere-ocean GCM, HadCM2 (plus some results from three coupled GCMs)
One regional climate model (HadRM3), conditioned by HadAM3H, conditioned by HadCM3 (simple maps from eight other coupled GCMs)
5° by 5° grid (ca 500 km)
2.5° by 3.75° grid (ca 300 km)
2.5° by 3.75° grid (ca 300 km)
0.44° by 0.44° grid (ca 50 km)
Monthly/seasonal averages, plus some daily weather variables
Monthly/seasonal averages and inter-annual variability, plus some daily weather variables
Simulated winter and summer mean temperature and precipitation changes that most closely matched the CET and EWP geographies were compiled for all the scenarios. For CCIRG91 and CCIRG96 this was a simple case of reading the changes from the maps published in the reports. For UKCIP98 CET, the grid box centred in 0° longitude and between 51° and 54° latitude was used. For EWP an average of three grid boxes that cover Northern England, Wales and Central England was used. For UKCIP02 CET, the following grids were used: 332–334, 352–354 and 372–374 (http://data.ukcip.org.uk/UKCIP02_dataarchive/50km_resolution/UKCIP02_50km_grid.pdf), whereas for EWP, all grid boxes in England and Wales were used. The earliest period portrayed in each climate scenario was used, which was 2010 for CCIRG91 and was the 2020s (i.e. period mean 2011–2040) for the other three scenario generations. The climate scenario projections data were linearly interpolated from 1990 to 2010 (for CCIRG91) and from 1961–1990 to the 2020s (for the other scenarios). This is consistent with the construction of the climate scenarios, which for various time slices (2020s and 2050s) and emissions scenarios have been linearly pattern-scaled from model experiments of the 2080s. Pattern-scaling simultaneously draws out the anthropogenic climate change signal (by maximising signal-to-noise) and reduces the influence of initial model conditions.
For UKCIP98 and UKCIP02, projected changes in seasonal CET and EWP inter-annual variability are also compared with recent changes. These data were read from the maps in the respective reports.
3. Results and discussion
Figure 1 shows inter-annual winter and summer CET changes compared with the 1961–1990 period for the observed record (grey bars). There have been no negative observed annual anomalies since 1998 in winter. In summer, five out of 48 years (1976, 1983, 1995, 2003 and 2006) are above the two standard deviations of inter-annual natural variability (as represented by data from 1660–1950), which is four times what one would expect by chance. In the period 1976–2007, inter-annual variability has decreased in winter (by 15%) and increased in summer (by 4%) compared with the 1961–1990 period. The sign of changes in the observed seasonal inter-annual variability are consistent with UKCIP98 Medium-high scenario for the 2020s (3% decrease for winter and 11% increase for summer). UKCIP02 only presented inter-annual variability results for the 2080s, but in terms of sign the projections are also consistent with observations.
The thick black lines in Figure 1 represent the observed 30-year overlapping means from 1946–1975 to 1978–2007. This is the only data that are directly comparable with the climate scenario projections (lines in colour). During this short period there is a warming trend for CET, which is particularly pronounced in summer. Observed CET has been warming faster than most scenarios, except for a few periods when compared with UKCIP98 High and Medium-high. For summer, five of the thirty-three 30-year means have been above the two standard deviations of natural multi-decadal variability, which is over six times what one would expect by chance. This is consistent with climate-model-based studies that have identified an anthropogenic climate change signal in annual-mean CET (Sexton et al., 2004; Karoly and Stott, 2006). The trends projected by the various climate scenarios are generally consistent with the 30-year observed mean of CET. The slower warming in the various scenarios relative to the observed CET could result from missing processes in climate models, uncertain parametrisations and/or uncertainties in initial conditions. For summer, this could be related to atmosphere–surface interactions such as soil moisture deficit, which can amplify summer temperature extremes (Fischer et al., 2007), and which Ferranti and Viterbo (2006) have shown to have a larger impact than has the ocean boundary forcing.
Figure 1 also shows in dotted black lines the observed means of declining length from the period centred on 1993 (1979–2007; 29 year mean) until 2007 (1993–2007; 15 year mean). Although these dotted lines are not directly comparable with the scenario projections (which are based on 30-year means), they provide a more recent picture of changes in CET. For winter, they show recent warming slightly above the scenario projections, whereas for summer the recent observations lie within the range of scenario projections.
Figure 2 shows observed winter and summer EWP, displaying high inter-annual variability in EWP for both seasons. During the period shown, three winters are above the two standard deviations of inter-annual natural variability, which is twice what one would expect by chance. Compared with the 1961–1990 average, inter-annual variability for the period 1976–2007 has increased for winter (by 5%) and even more so for summer (by 17%). Projections of seasonal inter-annual variability for the UKCIP98 Medium-high for the 2020s are consistent for summer (an increase of 10%), but not for winter (a decrease by 2%). For winter, both UKCIP98 Medium-High and UKCIP02 projections for the 2080s show an increase (around 20%) in inter-annual variability. For the summer of the 2080s, UKCIP98 Medium-high projects an overall increase in inter-annual variability (with a slight decrease for eastern and Central England), whereas UKCIP02 projects a decrease.
Unlike CET, there are no major observed precipitation trends at the 30-year time-scale. The observed data show a slight tendency towards wetter winters, but almost no tendency for summers (cf. Osborn and Hulme, 2002; Jenkins et al., 2007). For most of the period examined until 1993, observed winter EWP has been slightly wetter than projected by the climate scenarios; thereafter observations have remained roughly within the range of scenario projections. Summer EWP until 1989 was slightly drier than projected by the climate scenarios; from 1992, however, almost all observed periods have been slightly wetter than the scenario projections, in particular compared with UKCIP02. Although scenario changes are partially consistent with precipitation observations, both winter and summer trends still lie well within natural multi-decadal climate variability. This makes the comparison of precipitation scenarios with observations problematic because of high observed inter-annual variability and low signal-to-noise ratios in model output. Only recently have anthropogenic influences been detected in precipitation trends, but only at the global scale (Lambert et al., 2004, 2005) or within broad latitudinal bands over land (Zhang et al., 2007).
Climate scenarios serve a number of functions in society including raising awareness about climate change, guiding public policy and supporting investment decisions. Comparing earlier (from the 1990s) published climate scenarios with recent observations is an important exercise that provides additional information over and above conventional climate model evaluation. If regional climate scenarios are presumed to have any skill (cf. Osborn et al., 1999) or legitimacy (cf. Hulme and Dessai, 2008a) in capturing the range of possible future climates then it is important to subject them to this form of comparison.
This analysis has shown that recent trends in the observed UK climate fall broadly within the range of published climate scenario projections, the greatest ambiguity occurring with summer precipitation. This latter result suggests that using the UK national scenarios on their own to guide adaptation decisions might not always lead to robust adaptation. Using results from different climate models and applying different downscaling techniques has been shown to produce a wider range of climate impacts (Wilby and Harris, 2006) and to lead to different adaptation decisions (Dessai and Hulme, 2007). Therefore, it is vital that climate uncertainties not necessarily represented in published national scenarios are explored when making adaptation decisions.
This analysis has also shown that natural multi-decadal climate variability should be included in adaptation assessments with a planning horizon of 30 years (or more) since its influence in precipitation change is larger than the anthropogenic signal estimated by climate scenarios in the near future (cf. Hulme et al., 1999). This poses major challenges for adaptation planning as precipitation observations may diverge from climate scenarios for years to come because of natural multi-decadal climate variability. Even if decadal forecasting skill is substantially improved (e.g. Smith et al., 2007), planners will still have to take adaptation decisions under considerable uncertainty, particularly if precipitation is a significant driver of the system being considered (e.g. water resources management). Fortunately, there are tools and methods that allow decision-makers to identify decisions that work reasonably well under conditions of deep uncertainty (Dessai et al., 2009).
The reliance of (three of) the national UK scenarios on one climate modelling framework, successive generations of the Hadley Centre model, suggests a possible weakness in the way uncertainties have previously been handled. Although future scenarios (e.g. UKCIP09) will present some results using probability density functions as a way of better capturing some of these uncertainties (which will include natural climate variability and projections from other climate models), verifying such probabilistic scenarios will be more problematic.
Dessai was supported by funding from the Tyndall Centre core contract with the NERC, EPSRC and ESRC, and the EPSRC funded project ‘Simplicity, Complexity and Modelling’ (EP/E018173/1). Roger Jones and five anonymous reviewers are thanked for their comments.