The sensitivity of the meridional overturning circulation to modelling uncertainty in a perturbed physics ensemble without flux adjustment

Ocean dynamics play a key role in the climate system, by redistributing heat and freshwater. The uncertainty of how these processes are represented in climate models, and how this uncertainty affects future climate projections can be investigated using perturbed physics ensembles of global circulation models (GCMs). Techniques such as flux adjustments should be avoided since they can impact the sensitivity of the ensemble to the imposed forcing. In this study a method for developing an coupled ensemble with a GCM that does not use flux adjustment is presented. The ensemble is constrained by using information from a prior ensemble with a mixed layer ocean coupled to an atmosphere GCM, to reduce drifts in the coupled ensemble. Constraints on parameter perturbations are derived by using observational constraints on surface temperature, and top of the atmosphere radiative fluxes. As an example of such an ensemble developed with this methodology, uncertainty in response of the meridional overturning circulation (MOC) to increased CO₂ concentrations is investigated. The ensemble mean MOC strength is 17.1?Sv and decreases by 2.1?Sv when greenhouse gas concentrations are doubled. No rapid changes or shutdown of the MOC are seen in any of the ensemble members. There is a strong negative relationship between global mean temperature and MOC strength across the ensemble which is not seen in a multimodel ensemble. A positive relationship between climate sensitivity and the decrease of MOC strength is also seen.     

Potential impacts of climate change on storage conditions for commercial agriculture: an example for potato production in Michigan

Climate change is anticipated to influence all parts of agricultural production systems. However, the potential impacts on crop storage have rarely been assessed, even though storage is an important component of a grower’s marketing strategy and is essential for the continuous supply of a commodity for processors, exporters, and consumers. The Michigan chip-processing potato industry provides an example of the importance of crop storage. Michigan is the largest producer of chip-processing potatoes in the USA, and potatoes are stored on farms from September to June. We use an ensemble of climate projections developed for three future time slices (early, mid, and late century) from 16 climate models forced by three greenhouse gas concentration pathways to assess future changes in potato storage conditions. Our findings indicate an increased future demand for ventilation and/or refrigeration immediately after harvest and again in spring and early summer, even for the early-century time slice. Furthermore, the period of reliably cold storage temperatures during winter is anticipated, when averaged across all models, to shorten by 11–17 days in Michigan’s primary production area and 14–20 days in the more southern secondary area by mid-century, and by 15–29 days and 31–35 days, respectively, for the northern and southern production areas by late century. The level of uncertainty, as indicated by the ensemble range, is large, although the sign of the projected changes in storage parameters is consistent. This case study provides an example of the potentially large effects of climate change on the storage conditions for agricultural commodities.     

中国CMIP5模式对未来北极海冰的模拟偏差

本文对中国参加CMIP5的6个气候模式对未来北极海冰的模拟情况进行了评估。通过与1979-2005年海冰的观测值以及2050年代的多模式集合平均值对比发现,中国的气候模式对海冰范围的模拟结果与CMIP5模式的平均水平存在一定差距,具体表现为：BNU-ESM和FGOALS-s2对当前海冰范围估计很好,但对温度敏感性略偏高;FIO-ESM对当前海冰范围估计很好,但由于海冰对温度的敏感性偏低,导致其模拟的未来海冰在各种RCP情景中都融化缓慢;FGOALS-g2（BCC-CSM1-1和BCC-CSM1-1-m）对当前海冰范围的模拟存在显著偏多（显著偏少）的问题,这导致其对未来海冰融化的估计也持续偏多（偏少）。中国模式对北极海冰的模拟偏差导致它们对极区地表大气温度和湿度的模拟出现偏差,并且这些极区气象要素的偏差会进一步通过动力过程传导到对秋、冬季西风带、极涡的模拟中去。研究表明：从对海冰本身的模拟以及海冰偏差带来的气候影响这两个角度看,BNU-ESM在中国模式中水平较高,但总体上中国6个气候模式在海冰分量的模拟上仍与世界平均水平存在差距,这需要中国各模式中心的持续改进。     

On the generation of climate model ensembles

Climate model ensembles are used to estimate uncertainty in future projections, typically by interpreting the ensemble distribution for a particular variable probabilistically. There are, however, different ways to produce climate model ensembles that yield different results, and therefore different probabilities for a future change in a variable. Perhaps equally importantly, there are different approaches to interpreting the ensemble distribution that lead to different conclusions. Here we use a reduced-resolution climate system model to compare three common ways to generate ensembles: initial conditions perturbation, physical parameter perturbation, and structural changes. Despite these three approaches conceptually representing very different categories of uncertainty within a modelling system, when comparing simulations to observations of surface air temperature they can be very difficult to separate. Using the twentieth century CMIP5 ensemble for comparison, we show that initial conditions ensembles, in theory representing internal variability, significantly underestimate observed variance. Structural ensembles, perhaps less surprisingly, exhibit over-dispersion in simulated variance. We argue that future climate model ensembles may need to include parameter or structural perturbation members in addition to perturbed initial conditions members to ensure that they sample uncertainty due to internal variability more completely. We note that where ensembles are over- or under-dispersive, such as for the CMIP5 ensemble, estimates of uncertainty need to be treated with care.     

Extended regional climate model projections for Europe until the mid-twentyfirst century: combining ENSEMBLES and CMIP3

This study aims at sharpening the existing knowledge of expected seasonal mean climate change and its uncertainty over Europe for the two key climate variables air temperature and precipitation amount until the mid-twentyfirst century. For this purpose, we assess and compensate the global climate model (GCM) sampling bias of the ENSEMBLES regional climate model (RCM) projections by combining them with the full set of the CMIP3 GCM ensemble. We first apply a cross-validation in order to assess the skill of different statistical data reconstruction methods in reproducing ensemble mean and standard deviation. We then select the most appropriate reconstruction method in order to fill the missing values of the ENSEMBLES simulation matrix and further extend the matrix by all available CMIP3 GCM simulations forced by the A1B emission scenario. Cross-validation identifies a randomized scaling approach as superior in reconstructing the ensemble spread. Errors in ensemble mean and standard deviation are mostly less than 0.1 K and 1.0 % for air temperature and precipitation amount, respectively. Reconstruction of the missing values reveals that expected seasonal mean climate change of the ENSEMBLES RCM projections is not significantly biased and that the associated uncertainty is not underestimated due to sampling of only a few driving GCMs. In contrast, the spread of the extended simulation matrix is partly significantly lower, sharpening our knowledge about future climate change over Europe by reducing uncertainty in some regions. Furthermore, this study gives substantial weight to recent climate change impact studies based on the ENSEMBLES projections, since it confirms the robustness of the climate forcing of these studies concerning GCM sampling.     

Terrestrial biosphere carbon storage under alternative climate projections

This study investigates commonalities and differences in projected land biosphere carbon storage among climate change projections derived from one emission scenario by five different general circulation models (GCMs). Carbon storage is studied using a global biogeochemical process model of vegetation and soil that includes dynamic treatment of changes in vegetation composition, a recently enhanced version of the Lund-Potsdam-Jena Dynamic Global Vegetation Model (LPJ-DGVM). Uncertainty in future terrestrial carbon storage due to differences in the climate projections is large. Changes by the end of the century range from −106 to +201 PgC, thus, even the sign of the response whether source or sink, is uncertain. Three out of five climate projections produce a land carbon source by the year 2100, one is approximately neutral and one a sink. A regional breakdown shows some robust qualitative features. Large areas of the boreal forest are shown as a future CO₂ source, while a sink appears in the arctic. The sign of the response in tropical and sub-tropical ecosystems differs among models, due to the large variations in simulated precipitation patterns. The largest uncertainty is in the response of tropical rainforests of South America and Central Africa.     

A Demonstration of the Uncertainty in Projections of UK Climate Change Resulting from Regional Model Formulation

Regional climate models (RCMs) are now commonly used to downscale climate change projections provided by global coupled models to resolutions that can be utilised at national and finer scales. Although this extra tier of complexity adds significant value, it inevitably contributes a further source of uncertainty, due to the regional modelling uncertainties involved. Here, an initial attempt is made to estimate the uncertainty that arises from typical variations in RCM formulation, focussing on changes in UK surface air temperature (SAT) and precipitation projected for the late twenty-first century. Data are provided by a relatively large suite of RCM and global model integrations with widely varying formulations. It is found that uncertainty in the formulation of the RCM has a relatively small, but non-negligible, impact on the range of possible outcomes of future UK seasonal mean climate. This uncertainty is largest in the summer season. It is also similar in magnitude to that of large-scale internal variations of the coupled climate system, and for SAT, it is less than the uncertainty due to the emissions scenario, whereas for precipitation it is probably larger. The largest source of uncertainty, for both variables and in all seasons, is the formulation of the global coupled model. The scale-dependency of uncertainty due to RCM formulation is also explored by considering its impact on projections of the difference in climate change between the north and south of the UK. Finally, the implications for the reliability of UK seasonal mean climate change projections are discussed.     

Climate sensitivity estimated from ensemble simulations of glacial climate

The concentration of greenhouse gases (GHGs) in the atmosphere continues to rise, hence estimating the climate system’s sensitivity to changes in GHG concentration is of vital importance. Uncertainty in climate sensitivity is a main source of uncertainty in projections of future climate change. Here we present a new approach for constraining this key uncertainty by combining ensemble simulations of the last glacial maximum (LGM) with paleo-data. For this purpose we used a climate model of intermediate complexity to perform a large set of equilibrium runs for (1) pre-industrial boundary conditions, (2) doubled CO₂ concentrations, and (3) a complete set of glacial forcings (including dust and vegetation changes). Using proxy-data from the LGM at low and high latitudes we constrain the set of realistic model versions and thus climate sensitivity. We show that irrespective of uncertainties in model parameters and feedback strengths, in our model a close link exists between the simulated warming due to a doubling of CO₂, and the cooling obtained for the LGM. Our results agree with recent studies that annual mean data-constraints from present day climate prove to not rule out climate sensitivities above the widely assumed sensitivity range of 1.5–4.5°C (Houghton et al. 2001). Based on our inferred close relationship between past and future temperature evolution, our study suggests that paleo-climatic data can help to reduce uncertainty in future climate projections. Our inferred uncertainty range for climate sensitivity, constrained by paleo-data, is 1.2–4.3°C and thus almost identical to the IPCC estimate. When additionally accounting for potential structural uncertainties inferred from other models the upper limit increases by about 1°C.     

Going to the Extremes

Projections of changes in climate extremes are critical to assessing the potential impacts of climate change on human and natural systems. Modeling advances now provide the opportunity of utilizing global general circulation models (GCMs) for projections of extreme temperature and precipitation indicators. We analyze historical and future simulations of ten such indicators as derived from an ensemble of 9 GCMs contributing to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC-AR4), under a range of emissions scenarios. Our focus is on the consensus from the GCM ensemble, in terms of direction and significance of the changes, at the global average and geographical scale. The climate extremes described by the ten indices range from heat-wave frequency to frost-day occurrence, from dry-spell length to heavy rainfall amounts. Historical trends generally agree with previous observational studies, providing a basic sense of reliability for the GCM simulations. Individual model projections for the 21st century across the three scenarios examined are in agreement in showing greater temperature extremes consistent with a warmer climate. For any specific temperature index, minor differences appear in the spatial distribution of the changes across models and across scenarios, while substantial differences appear in the relative magnitude of the trends under different emissions rates. Depictions of a wetter world and greater precipitation intensity emerge unequivocally in the global averages of most of the precipitation indices. However, consensus and significance are less strong when regional patterns are considered. This analysis provides a first overview of projected changes in climate extremes from the IPCC-AR4 model ensemble, and has significant implications with regard to climate projections for impact assessments. An erratum to this article is available at . An erratum to this article can be found at     

An efficient climate forecasting method using an intermediate complexity Earth System Model and the ensemble Kalman filter

We present the implementation and results of a model tuning and ensemble forecasting experiment using an ensemble Kalman filter for the simultaneous estimation of 12 parameters in a low resolution coupled atmosphere-ocean Earth System Model by tuning it to realistic data sets consisting of Levitus ocean temperature/salinity climatology, and NCEP/NCAR atmospheric temperature/humidity reanalysis data. The resulting ensemble of tuned model states is validated by comparing various diagnostics, such as mass and heat transports, to observational estimates and other model results. We show that this ensemble has a very reasonable climatology, with the 3-D ocean in particular having comparable realism to much more expensive coupled numerical models, at least in respect of these averaged indicators. A simple global warming experiment is performed to investigate the response and predictability of the climate to a change in radiative forcing, due to 100 years of 1% per annum atmospheric CO₂ increase. The equilibrium surface air temperature rise for this CO₂ increase is 4.2±0.1°C, which is approached on a time scale of 1,000 years. The simple atmosphere in this version of the model is missing several factors which, if included, would substantially increase the uncertainty of this estimate. However, even within this ensemble, there is substantial regional variability due to the possibility of collapse of the North Atlantic thermohaline circulation (THC), which switches off in more than one third of the ensemble members. For these cases, the regional temperature is not only 3–5°C colder than in the warmed worlds where the THC remains switched on, but is also 1–2°C colder than the current climate. Our results, which illustrate how objective probabilistic projections of future climate change can be efficiently generated, indicate a substantial uncertainty in the long-term future of the THC, and therefore the regional climate of western Europe. However, this uncertainty is only apparent in long-term integrations, with the initial transient response being similar across the entire ensemble. Application of this ensemble Kalman filtering technique to more complete climate models would improve the objectivity of probabilistic forecasts and hence should lead to significantly increased understanding of the uncertainty of our future climate.     

Regional and seasonal intercomparison of CMIP3 and CMIP5 climate model ensembles for temperature and precipitation

Regional and seasonal temperature and precipitation over land are compared across two generations of global climate model ensembles, specifically, CMIP5 and CMIP3, through historical twentieth century skills and multi-model agreement, and twenty first century projections. A suite of diagnostic and performance metrics, ranging from spatial bias or model-consensus maps and aggregate time series plots, to measures of equivalence between probability density functions and Taylor diagrams, are used for the intercomparisons. Pairwise and multi-model ensemble comparisons were performed for 11 models, which were selected based on data availability and resolutions. Results suggest little change in the central tendency or variability or uncertainty of historical skills or consensus across the two generations of models. However, there are regions and seasons, at different levels of aggregation, where significant changes, performance improvements, and even degradation in skills, are suggested. The insights may provide directions for further improvements in next generations of climate models, and in the meantime, help inform adaptation and policy.     

Sources of uncertainty in projections of twenty-first century westerly wind changes over the Amundsen Sea,West Antarctica,in CMIP5 climate models

The influence of changes in winds over the Amundsen Sea has been shown to be a potentially key mechanism in explaining rapid loss of ice from major glaciers in West Antarctica, which is having a significant impact on global sea level. Here, Coupled Model Intercomparison Project Phase 5 (CMIP5) climate model data are used to assess twenty-first century projections in westerly winds over the Amundsen Sea (U _AS ). The importance of model uncertainty and internal climate variability in RCP4.5 and RCP8.5 scenario projections are quantified and potential sources of model uncertainty are considered. For the decade 2090–2099 the CMIP5 models show an ensemble mean twenty-first century response in annual mean U _AS of 0.3 and 0.7 m s^?1 following the RCP4.5 and RCP8.5 scenarios respectively. However, as a consequence of large internal climate variability over the Amundsen Sea, it takes until around 2030 (2065) for the RCP8.5 response to exceed one (two) standard deviation(s) of decadal internal variability. In all scenarios and seasons the model uncertainty is large. However the present-day climatological zonal wind bias over the whole South Pacific, which is important for tropical teleconnections, is strongly related to inter-model differences in projected change in U _AS (more skilful models show larger U _AS increases). This relationship is significant in winter (r = ?0.56) and spring (r = ?0.65), when the influence of the tropics on the Amundsen Sea region is known to be important. Horizontal grid spacing and present day sea ice extent are not significant sources of inter-model spread.     

Applying probabilistic projections of climate change with impact models: a case study for sub-arctic palsa mires in Fennoscandia

A comparison of two approaches for determining probabilistic climate change impacts is presented. In the first approach, ensemble climate projections are applied directly as inputs to an impact model and the risk of impact is computed from the resulting ensemble of outcomes. As this can involve large numbers of projections, the approach may prove to be impractical when applied to complex impact models with demanding input requirements. The second approach is to construct an impact response surface based on a sensitivity analysis of the impact model with respect to changes in key climatic variables, and then to superimpose probabilistic projections of future climate onto the response surface to assess the risk of impact. To illustrate this comparison, an impact model describing the spatial distribution of palsas in Fennoscandia was applied to estimate the risk of palsa disappearance. Palsas are northern mire complexes with permanently frozen peat hummocks, located at the outer limit of the permafrost zone and susceptible to rapid decline due to regional warming. Probabilities of climate changes were derived from an ensemble of coupled atmosphere–ocean general circulation model (AOGCM) projections using a re-sampling method. Results indicated that the response surface approach, though introducing additional uncertainty, gave risk estimates of area decline for palsa suitability that were comparable to those obtained using multiple simulations with the original palsa model. It was estimated as very likely (>90% probability) that a decline of area suitable for palsas to less than half of the baseline distribution will occur by the 2030s and likely (>66%) that all suitable areas will disappear by the end of the twenty-first century under scenarios of medium (A1B) and moderately high (A2) emissions. For a low emissions (B1) scenario, it was more likely than not (>50%) that conditions over a small fraction of the current palsa distribution would remain suitable until the end of the twenty-first century.     

The impact of perturbations to ocean-model parameters on climate and climate change in a coupled model

Assessments of the impacts of uncertainties in parameters on mean climate and climate change in complex climate models have, to date, largely focussed on perturbations to parameters in the atmosphere component of the model. Here we expand on a previously published study which found the global impacts of perturbed ocean parameters on the rate of transient climate change to be small compared to perturbed atmosphere parameters. By separating the climate-change-induced ocean vertical heat transport in each perturbed member into components associated with the resolved flow and each parameterisation scheme, we show that variations in global mean heat uptake in different perturbed versions are an order of magnitude smaller than the average heat uptake. The lack of impact of the perturbations is attributed to (1) the relatively small impact of the perturbation on the direct vertical heat transport associated with the perturbed process and (2) a compensation between those direct changes and indirect changes in heat transport from other processes. Interactions between processes and changes appear to combine in complex ways to limit ensemble spread and uncertainty in the rate of warming. We also investigate regional impacts of the perturbations that may be important for climate change predictions. We find variations across the ensemble that are significant when measured against natural variability. In terms of the experimental set-up used here (models without flux adjustments) we conclude that perturbed physics ensembles with ocean parameter perturbations are an important component of any probabilistic estimate of future climate change, despite the low spread in global mean quantities. Hence, careful consideration should be given to assessing uncertainty in ocean processes in future probabilistic assessments of regional climate change.     

Assessing uncertainties in climate change impact analyses on the river flow regimes in the UK. Part 2: future climate

The first part of this paper demonstrated the existence of bias in GCM-derived precipitation series, downscaled using either a statistical technique (here the Statistical Downscaling Model) or dynamical method (here high resolution Regional Climate Model HadRM3) propagating to river flow estimated by a lumped hydrological model. This paper uses the same models and methods for a future time horizon (2080s) and analyses how significant these projected changes are compared to baseline natural variability in four British catchments. The UKCIP02 scenarios, which are widely used in the UK for climate change impact, are also considered. Results show that GCMs are the largest source of uncertainty in future flows. Uncertainties from downscaling techniques and emission scenarios are of similar magnitude, and generally smaller than GCM uncertainty. For catchments where hydrological modelling uncertainty is smaller than GCM variability for baseline flow, this uncertainty can be ignored for future projections, but might be significant otherwise. Predicted changes are not always significant compared to baseline variability, less than 50% of projections suggesting a significant change in monthly flow. Insignificant changes could occur due to climate variability alone and thus cannot be attributed to climate change, but are often ignored in climate change studies and could lead to misleading conclusions. Existing systematic bias in reproducing current climate does impact future projections and must, therefore, be considered when interpreting results. Changes in river flow variability, important for water management planning, can be easily assessed from simple resampling techniques applied to both baseline and future time horizons. Assessing future climate and its potential implication for river flows is a key challenge facing water resource planners. This two-part paper demonstrates that uncertainty due to hydrological and climate modelling must and can be accounted for to provide sound, scientifically-based advice to decision makers.     

Addressing sources of uncertainty in runoff projections for a data scarce catchment in the Ecuadorian Andes

Future climate projections from general circulation models (GCMs) predict an acceleration of the global hydrological cycle throughout the 21st century in response to human-induced rise in temperatures. However, projections of GCMs are too coarse in resolution to be used in local studies of climate change impacts. To cope with this problem, downscaling methods have been developed that transform climate projections into high resolution datasets to drive impact models such as rainfall-runoff models. Generally, the range of changes simulated by different GCMs is considered to be the major source of variability in the results of such studies. However, the cascade of uncertainty in runoff projections is further elongated by differences between impact models, especially where robust calibration is hampered by the scarcity of data. Here, we address the relative importance of these different sources of uncertainty in a poorly monitored headwater catchment of the Ecuadorian Andes. Therefore, we force 7 hydrological models with downscaled outputs of 8 GCMs driven by the A1B and A2 emission scenarios over the 21st century. Results indicate a likely increase in annual runoff by 2100 with a large variability between the different combinations of a climate model with a hydrological model. Differences between GCM projections introduce a gradually increasing relative uncertainty throughout the 21st century. Meanwhile, structural differences between applied hydrological models still contribute to a third of the total uncertainty in late 21st century runoff projections and differences between the two emission scenarios are marginal.     

Future changes of the terrestrial ecosystem based on a dynamic vegetation model driven with RCP8.5 climate projections from 19 GCMs

Future changes of terrestrial ecosystems due to changes in atmospheric CO₂ concentration and climate are subject to a large degree of uncertainty, especially for vegetation in the Tropics. Here, we evaluate the natural vegetation response to projected future changes using an improved version of a dynamic vegetation model (CLM-CN-DV) driven with climate change projections from 19 global climate models participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5). The simulated equilibrium vegetation distribution under historical climate (1981–2000) has been compared with that under the projected future climate (2081–2100) scenario for Representative Concentration Pathway 8.5 (RCP8.5) to qualitatively assess how natural potential vegetation might change in the future. With one outlier excluded, the ensemble average of vegetation changes corresponding to climates of 18 GCMs shows a poleward shift of forests in northern Eurasia and North America, which is consistent with findings from previous studies. It also shows a general “upgrade” of vegetation type in the Tropics and most of the temperate zones, in the form of deciduous trees and shrubs taking over C3 grass in Europe and broadleaf deciduous trees taking over C4 grasses in Central Africa and the Amazon. LAI and NPP are projected to increase in the high latitudes, southeastern Asia, southeastern North America, and Central Africa. This results from CO₂ fertilization, enhanced water use efficiency, and in the extra-tropics warming. However, both LAI and NPP are projected to decrease in the Amazon due to drought. The competing impacts of climate change and CO₂ fertilization lead to large uncertainties in the projection of future vegetation changes in the Tropics.     

Future heat vulnerability in California, Part II: projecting future heat-related mortality

Through the 21^st century, a significant increase in heat events is likely across California (USA). Beyond any climate change, the state will become more vulnerable through demographic changes resulting in a rapidly aging population. To assess these impacts, future heat-related mortality estimates are derived for nine metropolitan areas in the state for the remainder of the century. Heat-related mortality is first assessed by initially determining historical weather-type mortality relationships for each metropolitan area. These are then projected into the future based on predicted weather types created in Part I. Estimates account for several levels of uncertainty: for each metropolitan area, mortality values are produced for five different climate model-scenarios, three different population projections (along with a constant-population model), and with and without partial acclimatization. Major urban centers could have a greater than tenfold increase in short-term increases in heat-related mortality in the over 65 age group by the 2090s.     

Higher precision estimates of regional polar warming by ensemble regression of climate model projections

This study presents projections of twenty-first century wintertime surface temperature changes over the high-latitude regions based on the third Coupled Model Inter-comparison Project (CMIP3) multi-model ensemble. The state-dependence of the climate change response on the present day mean state is captured using a simple yet robust ensemble linear regression model. The ensemble regression approach gives different and more precise estimated mean responses compared to the ensemble mean approach. Over the Arctic in January, ensemble regression gives less warming than the ensemble mean along the boundary between sea ice and open ocean (sea ice edge). Most notably, the results show 3?°C less warming over the Barents Sea (~7?°C compared to ~10?°C). In addition, the ensemble regression method gives projections that are 30?% more precise over the Sea of Okhostk, Bering Sea and Labrador Sea. For the Antarctic in winter (July) the ensemble regression method gives 2?°C more warming over the Southern Ocean close to the Greenwich Meridian (~7?°C compared to ~5?°C). Projection uncertainty was almost half that of the ensemble mean uncertainty over the Southern Ocean between 30° W to 90° E and 30?% less over the northern Antarctic Peninsula. The ensemble regression model avoids the need for explicit ad hoc weighting of models and exploits the whole ensemble to objectively identify overly influential outlier models. Bootstrap resampling shows that maximum precision over the Southern Ocean can be obtained with ensembles having as few as only six climate models.