Journal of Geophysical Research: Atmospheres

Comparing the impacts of mitigation versus non-intervention scenarios on future temperature and precipitation extremes in the HadGEM2 climate model


Corresponding author: J. Caesar, Met Office Hadley Centre, FitzRoy Road, Exeter EX1 3PB, UK. (


[1] Although international climate change negotiations focus on global mean temperature targets, it is also important to assess the impact of emission scenarios on climate extremes at the global and regional scale. This paper examines how temperature and precipitation extremes indices are projected to change around the globe during the 21st Century under an aggressive climate change mitigation scenario in the United Kingdom Met Office Hadley Centre HadGEM2-AO coupled climate model, and how these changes might differ from a mid-range non-intervention scenario. Even under an aggressive mitigation strategy there are projected increases in warm temperature extremes on a global and regional scale up to the mid-21st Century, and temperature extremes tend to follow a similar trajectory to the projected global mean temperature change associated with each scenario. Changes in precipitation-related extremes are projected to be more variable. There are regional differences in the direction of changes and it appears that the aerosol forcing associated with the scenarios could have an important influence. The regions which are projected to benefit most from mitigation vary depending on the index being considered, but in general absolute increases in temperature extremes are reduced in the northern midlatitudes, whereas for frequency based indices it is northern South America, parts of the USA, Africa and Asia which see the largest avoided increases. For precipitation indices, northern South America sees the most consistent signal toward avoided drying conditions, along with the Mediterranean and parts of Russia and central Asia.

1. Introduction

[2] Toward the end of 2010, Parties to the United Nations Framework Convention on Climate Change (UNFCCC) agreed to work toward limiting global mean temperature rise to a maximum of 2°C above pre-industrial levels (Durban Platform for Enhanced Action,) United Nations Framework Convention on Climate Change, 2011, Research continues into assessing the impacts of different future greenhouse gas emissions strategies upon the trajectory of global mean temperature, but there is also considerable interest in changes in the frequency and severity of climate extremes which are often associated with high impacts upon economic, social and ecological systems [Easterling et al., 2000].

[3] Until recently, most advice on the outcomes of emissions pathways that could limit warming to 2°C has been based on simple climate models [e.g., Wigley and Raper, 2001; Allen et al., 2009; Meinshausen et al., 2009], but these are unable to explicitly take account of many key feedbacks and interactions within the large-scale climate system. One of the first groups to consider a global climate model (GCM) assessment of climate change mitigation was Washington et al. [2009]. Their low emissions scenario represented a stabilized CO2 concentration of about 450 ppm in 2100, with emissions peaking in the early 21st Century and then gradually declining. Under the mitigation scenario they found an increase in global mean temperature of about half that which would be seen under their non-mitigation scenario, and which remained within the 2°C above pre-industrial target. A study by Arora et al. [2011], using the second-generation Canadian earth system model (CanESM2), found a simulated warming of 2.3°C over the 1850–2100 period using a low-emissions scenario (RCP2.6, see Section 2.1), slightly above the 2°C target.

[4] Washington et al. [2009] also explicitly considered changing climate extremes in the form of heat waves and found that the increase in the intensity of heat waves would be 55% less in the mitigation scenario compared with the non-mitigation scenario. The greatest reduction in regional heat wave intensity with mitigation could be seen over the western USA, Canada, and most of Europe, Russia and northern Africa. Clark et al. [2010] also examined future changes in heat waves, but used a large ensemble of GCM simulations with perturbations to parameters governing key processes such as large-scale cloud, convection, sea ice and boundary layer processes. A subset of ensemble members were chosen which gave globally averaged temperature increases of around 2°C, and these were used to represent changes that might be consistent with a climate change mitigation scenario. From these they found that increases in the temperature on extreme hot days ranged from between 2°C to 6°C for parts of Europe, North America and Asia. Soil moisture changes have been linked with heatwaves and the surface energy balance [e.g., Brabson et al., 2005] and Clark et al. [2010] found that important contributing factors to these changes were found to be related to the direct or indirect influences on the simulation of soil moisture, with the additional warming associated with dryer conditions.

[5] There has been limited research to date on the effects of climate change mitigation on precipitation and associated extremes. Washington et al. [2009] found that the projected mean precipitation changes under mitigation were about half that of the changes under their non-mitigation scenario. Using both a multimodel ensemble of GCMs and an Earth System Model of Intermediate Complexity, Solomon et al. [2009] investigated the impacts on precipitation if a cessation of CO2 emissions were to occur immediately following a set rise of 2% per year up to various concentration targets (ranging from 450 ppmv to 1200 ppmv), with a subsequent decline in CO2 concentrations. They found that irreversible (on the timescale of the next millennium) dry season reductions in rainfall (∼20% for a 2°C global temperature increase) could be experienced in several regions, including northern Africa, southern Europe and western Africa.

[6] A key question regarding changing climate extremes is whether the changes in extremes scale with changes in the mean climate. This is partly because many current studies of mitigation options tend to use a pattern-scaling approach where a pattern of climate change, such as mean precipitation, is scaled linearly with global temperature change [Mitchell, 2003]. A clear demonstration of a potential nonlinearity in the hydrological cycle is provided by Wu et al. [2010], who used GCM simulations to show that stabilizing or reducing atmospheric CO2 concentrations could lead to a temporary strengthening of the global hydrological cycle, while reducing rainfall over parts of the tropics and sub-tropics. In addition to establishing if global changes in extremes scale with global mean temperature change, it is also useful to establish how regional changes scale with temperature change. Temperature changes are not spatially uniform [e.g., Joshi et al., 2008], and the sign of precipitation changes may differ between regions. Lenton [2011] notes that changes to phenomena such as the south Asian monsoon may not be directly dependent on global mean warming, but are likely to be affected by localized warming which alters temperature gradients in the vicinity. One cause of these gradients is an uneven distribution of aerosols from anthropogenic sources in the atmosphere, and these regional differences in warming may also impact upon temperature extremes in those areas. When using climate change projections to assess potential future changes in climate impacts, it becomes particularly important to consider changes on a regional or country level due to the scale of the impacts being considered (socio-economic, ecological etc.). While much of the debate at the policy level is related to global mean temperature thresholds, adaptation to climate change will have to address changes in climate and extreme events at the regional and local scales.

[7] This paper aims to investigate the projected global changes in climate extremes which could be avoided by adopting an aggressive mitigation scenario using a selection of widely used climate extremes indices derived from daily data output from HadGEM2-AO simulations. We also consider regional differences in the patterns of projected extremes.

2. Models and Methods

2.1. Climate Model and Emissions Scenarios

[8] The model used for this study is the Met Office Hadley Centre Global Environment Model version 2 (HadGEM2) [Collins et al., 2008] which is a development of the HadGEM1 model [Johns et al., 2006] used in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) [Intergovernmental Panel on Climate Change (IPCC), 2007]. We use the HadGEM2-AO configuration with the atmosphere coupled to a fully dynamical ocean. The atmospheric component has 38 levels extending to ∼40 km height, with a horizontal resolution of 1.25 degrees of latitude by 1.875 degrees of longitude.

[9] To compare the effects of reducing greenhouse gas emissions in the future we use GCM experiments with three different emissions pathways; two based upon a non-mitigation scenario, and the third using an aggressive mitigation scenario, with all simulations covering the period 1861–2100. The primary non-mitigation reference scenario is the A1B-SRES scenario [Nakicenovic and Swart, 2000], which represents a mid-range non-intervention (no explicit climate policy) scenario. It is a medium-high emissions scenario which assumes a future of strong economic growth leading to an increase in the rate of greenhouse gas emissions, and the atmospheric carbon dioxide equivalent (CO2eq) concentration rises throughout the 21st Century to around 900 ppm by 2100 (Figure 1). We use the A1B-SRES scenario as it provides overlap and consistency with much existing climate modeling work, and it is fairly consistent with observed carbon emissions over the past two decades [van Vuuren and Riahi, 2008; Le Quéré et al., 2009]. Two simulations using the A1B-SRES simulation were available for this study in addition to which we have available a single simulation of the A1B-IMAGE scenario [van Vuuren et al., 2007]. The A1B-IMAGE and A1B-SRES scenarios were developed using different Integrated Assessment Model (IAMs) which make different assumptions regarding a variety of inputs including globalization, population changes, economic growth and technological development; additionally A1B-IMAGE has been updated against new information compared to the older SRES scenario [van Vuuren et al., 2007]. This accounts for differences in the projected emissions between the two scenarios and also the sulphate aerosol burden, which is markedly different during the early 21st Century with the A1B-IMAGE scenario containing lower sulphur emissions [Johns et al., 2011].

Figure 1.

Global mean CO2-equivalent concentration used to drive the (top) A1B and E1 simulations, and (bottom) corresponding radiative forcing. Profiles for RCP8.5 and RCP2.6 are also shown. Adapted from Johns et al. [2011].

[10] The range of SRES scenarios include examples of lower emissions scenarios (e.g., the B1 and B2 families) but these do not explicitly take account of mitigation as a result of climate change policy. However, the European Union ENSEMBLES project has developed an aggressive mitigation scenario known as E1 [Lowe et al., 2009], and was the first international multimodel inter-comparison project to make use of such a scenario [Johns et al., 2011]. The E1 scenario has a peak in the CO2eq concentration at around 535 ppm (ppm) in 2045, before stabilizing at around 450 ppm during the 22nd Century (Figure 1). CO2eq emissions start to reduce early in the 21st Century, and decline to almost zero by 2100. The E1 scenario also has a lower sulphur burden compared to the SRES scenarios as it is derived from the A1B-IMAGE scenario, and because of the mitigation policies used to construct the scenario [Johns et al., 2011].

[11] We also acknowledge that a range of scenarios have been defined for use in the IPCC Fifth Assessment Report (AR5) and are being implemented in GCM experiments at climate modeling groups around the world [Moss et al., 2010; Arora et al., 2011]. These Representative Concentration Pathways (RCPs) use defined radiative forcing levels as a starting point whereas the SRES scenarios were developed by working “forwards” from their socio-economic assumptions to determine emissions and then radiative forcings [Moss et al., 2010]. The E1 mitigation scenario results will provide a useful comparison with the results from the RCP 2.6 scenario (sometimes also referred to as RCP 3-PD) since both scenarios follow a similar trajectory in total radiative forcing [van Vuuren et al., 2007]. RCP 2.6 is the low-end forcing scenario of 2.6 Wm−2, the others being RCP 4.5, RCP 6.0, and RCP 8.5.

2.2. Climate Extremes Indices

[12] There are a variety of approaches to defining climate extremes. In this paper we make use of internationally defined climate extremes indices [Zhang et al., 2011] which have been used extensively in the assessment of observed changes in global climate extremes [Frich et al., 2002; Alexander et al., 2006] as well as being used for the assessment of extremes in GCMs [e.g., Tebaldi et al., 2006; Sillmann and Roeckner, 2008; Russo and Sterl, 2011]. Most of these indices are based upon relatively frequently occurring extremes (e.g., once a year or more frequently) and therefore give a relatively robust baseline from which to determine changes due to the larger sample size of events. A range of indices are used to capture differing characteristics of climate extremes including frequency, intensity and duration. Percentile based indices allow a comparison of changes in the frequency of extreme events across the globe, as they are defined relative to a local baseline, and are robust to outliers which is a key benefit when analyzing observations. Absolute indices, such as maximum annual temperature, are more useful for determining local impacts.

[13] The selected indices (Table 1) are calculated using a modified version of software originally developed for use at international climate extremes workshops [Peterson and Manton, 2008; Zhang et al., 2011]. The software uses as input the time series from each individual model grid box of maximum and minimum daily 1.5 m temperatures, and also the daily total precipitation. Precipitation is only totaled for days where the amount is 1 mm or above as this provides consistency with the observational indices where the quality control procedure filters out “trace” values. In the case of the percentile indices, where it is required to define a baseline period during which to calculate the relevant percentiles, we use 1961–1990 to represent the ‘present-day’ climatology. Frequency of exceedence of these baseline percentile thresholds are then calculated for all years from 1861 to 2100 and the bootstrap methods of Zhang et al. [2005] are applied to prevent inhomogeneities in the indices occurring between the base period and the remainder of the series.

Table 1. Definitions of the Climate Indices Used in This Papera
Index IDIndicator Name and Brief DefinitionUnits
TXxMaximum annual daily maximum temperature°C
TNnMinimum annual daily minimum temperature°C
TX90pPercentage of warm days (daily maximum temperature above the 90th percentile of 1861–90 daily maximum temperature)%
TN90pPercentage of warm nights (daily minimum temperature above the 90th percentile of 1861–90 daily minimum temperatures)%
PRCPTOTAnnual total precipitation on days with ≥1 mm
SDIISimple daily intensity index (mean rainfall of days with ≥1 mm rain)mm
RX5dayAnnual maximum consecutive 5-day precipitationmm
R95pAnnual total precipitation on days when daily precipitation >95th
R20mmVery heavy precipitation days (days with ≥20 mm)Days

3. Results

3.1. Observational Comparison

[14] To provide an indication of how well the HadGEM2-AO simulations represent present-day climate extremes, we carried out a comparison with the HadEX observational climate extremes data set [Alexander et al., 2006] for the period of 1961–1990. Figures 2a and 2b show the results for the annual maximum daily temperature (TXx). Due to the variation in the availability of data from weather stations, some regions, including much of Africa, the Middle East, Greenland and northern South America, are not represented in HadEX. The simulated spatial pattern of TXx is fairly consistent with the observations. Notable differences are that areas of eastern Asia and the central and eastern United States are warmer in the model than the observations. Other studies have found inadequacies in the ability of GCMs to simulate the “warming hole” over the USA [Portmann et al., 2009] during the recent observational period and also biases in changes in TXx in several regions. For example, the comparisons of Christidis et al. [2005, 2011] indicate that HadCM3 simulates greater warming over the eastern USA and China than indicated by observations.

Figure 2.

Observations and model results for the 1961–1990 period: (a and b) annual maximum daily temperature, TXx, (c and d) total annual precipitation (PRCPTOT). Observations are from the HadEX data set [Alexander et al., 2006]. The HadGEM2 data have been re-gridded to the resolution of HadEX (3.75° × 2.5°). Non-gridded land areas represent non-availability of data in HadEX.

[15] Previous work has highlighted that there is generally more uncertainty associated with precipitation projections from GCMs than for temperature, both in terms of the mean fields and annual cycles, as well as intensity, frequency and duration statistics [Trenberth, 2011]. Even in terms of the simulated present-day climate, there is a large spread among the precipitation extremes as represented by GCMs, particularly in the tropics and sub-tropics [Kharin et al., 2007]. GCMs may simulate precipitation which is too frequent and with underestimated intensity as a result of parameterizations of sub-grid scale convection [e.g., Sun et al., 2006; Dai, 2006].

[16] Figures 2c and 2d show the results for PRCPTOT (total annual precipitation on days with over 1 mm) which again shows good spatial agreement. One key difference is that simulated rainfall is lower over India in the model than in the observations. This same difference can be observed in other precipitation indices, such as RX5day (not shown) which shows much lower totals over India in the model. Studies have shown that all configurations of HadGEM2, in common with the development version of HadGEM3 and other CMIP3/CMIP5 models, underestimate monsoon rainfall over central India and display consistent errors in simulating the monsoon circulation [Levine and Turner, 2012; R. C. Levine et al., The role of northern Arabian surface temperature biases in CMIP5 model simulations and future predictions of Indian summer monsoon rainfall, submitted to Climate Dynamics, 2012].

[17] A number of detection and attribution studies have considered extremes of temperature [Christidis et al., 2005, 2011; Morak et al., 2011; Zwiers et al., 2011] and precipitation [Min et al., 2011]. Christidis et al. [2005], for example, find indications of significant observed warming trends in warm-day extremes, and that they are linked to human activity rather than internal climate variability, and Zwiers et al. [2011] detect a significant anthropogenic influence upon temperature extremes, including warm days. However, these two studies find that the models used in the analyses overestimate the observed signal in max Tmax extremes. On the other hand, Morak et al. [2011], who investigate attribution of warm nights (TN90p), found that the ensemble of CMIP3 models used tended to show weaker changes than the observations, though the difference was not significant. A recent study by Min et al. [2011] compared multimodel simulations (though not including HadGEM2) and found that the models tended to underestimate the observed increasing trends in the RX1day and RX5day indices during the latter half of the 20th Century over the northern hemisphere, which suggests that increases in heavy precipitation could be underestimated in GCM projections.

3.2. Global Changes in Extremes Indices

3.2.1. Temperature

[18] To put the results into context, Figure 3 shows the changes in global mean annual temperature as simulated by HadGEM2-AO from 1861 to 2000 for the historic part of the experiment, and for 2000 to 2100 for the mitigation (E1) and non-mitigation (A1B-SRES and A1B-IMAGE) scenarios. Around the year 2050 there is a discernible divergence of the scenarios. By 2100 the A1B-SRES simulations have projected temperature anomalies approaching 4°C above the pre-industrial (defined here as 1861–1890) mean and the A1B-IMAGE projection has a higher temperature anomaly of nearly 4.5°C. The temperature anomalies in the E1 simulations increase at a slower rate to reach just over 2°C by 2100. The A1B-SRES projections have a similar trajectory to the A1B-IMAGE projection, but have a period of less rapid temperature increase during the first few decades of the 21st Century.

Figure 3.

Globally averaged 1.5 m temperature anomalies calculated relative to the 1861–1890 reference period.

[19] The global mean (land-only) time series of TXx (annual maximum daily temperature; Figure 4a) shows similarities compared with the global mean temperature projections in Figure 3. TXx increases from a simulated 1961–1990 mean of 31.1°C until about 2050. At this point, TXx in the non-mitigation experiments continues to rise to around 4°C above the 1961–90 mean by 2100 (nearly 5°C under A1B-IMAGE), whereas with mitigation it stabilizes at just below 2°C above the 1961–90 mean. A similar change can be found in annual minimum daily temperatures (TNn, not shown).

Figure 4.

Simulated and projected extremes indices from 1961 to 2100 for land area only: (a) annual maximum daily temperature, TXx, (b) number of warm days (TX90p), (c) number of warm nights (TN90p), (d) total annual precipitation (PRCPTOT), (e) simple daily intensity index (SDII), and (f) maximum consecutive 5-day precipitation (RX5day). The precipitation indices are fitted with a cubic function.

[20] An index capturing the frequency of warm days is TX90p which represents the number of days per year that the daily temperature exceeds the local 90th percentile (in this case relative to the 1961–1990 baseline) shown in Figure 4b. The definition of the index means that for the 1961–90 base period, 10% of days per year would exceed the threshold, which is approximately 36 days per year. By the year 2000, the simulations indicate a global mean value of around 60 days which already represents a near doubling of the frequency. The observational results of Alexander et al. [2006] for HadEX for the year 2000 show that the number of warm nights per year had increased by around 15 days compared with the 1961–90 mean (i.e., an absolute value of approximately 51 days per year) which is lower than the results indicate in the model projections, though it should be noted that large areas of Africa and South America were excluded from their analysis.

[21] As with the other temperature indices, the mitigation and non-mitigation scenarios clearly begin to diverge around the year 2050, with the projected frequency of warm days in the year 2100 reaching 200 days per year under the A1B-SRES simulations, 220 days under A1B-IMAGE, and stabilizing at around 130 days per year under the E1 scenario. Changes in the frequency of warm nights (TN90p) are shown in Figure 4c. This is based upon changes in daily minimum temperatures, but displays similar projected changes to the TX90p index.

[22] In terms of the scaling of the changing extremes with global mean temperature, Figure 5a indicates a strong linear relationship between changes in global mean temperature and changes in the TXx index. Increases in TXx are slightly higher than increases in global mean temperature e.g., an increase of around 4°C in TXx for a global mean increase of around 3°C. Figure 5b shows that there is also a strong linear relationship between global mean temperature and the frequency of warm days index (TX90p). This relationship is replicated across the other temperature extremes indices examined (Table 1) including TNn and TN90p (not shown). This supports other findings including Morak et al. [2011] who find that on a global scale, and for many regions, more than half of the observed trend in TN90p can be explained by the observed trend in mean temperatures.

Figure 5.

Relationship between annual global mean temperature and (a) annual maximum daily temperature, TXx, (b) annual number of warm days, TX90p, and (c) maximum consecutive 5-day precipitation, RX5day. (d) Relationship between global mean precipitation anomaly and maximum consecutive 5-day precipitation, RX5day. Asterisk symbols indicate second ensemble members of A1B-SRES and E1.

3.2.2. Precipitation

[23] Annual mean (land-only) precipitation is shown in Figure 4d. To help identify the trajectories of global precipitation indices through the 21st Century, we have fitted a cubic function to the time series. It is difficult to determine differences between many of the precipitation indices due to the large inter-annual variability, but the fitting of the functions to the data does indicate some consistent patterns of change between some of the indices. From a global mean annual precipitation value of around 700 mm in 2000, the A1B-IMAGE and E1 simulations show a gradual increase to around 740 mm by 2100. However, precipitation under A1B-SRES decreases to around 680 mm by 2050, then begins to increase at a comparable rate to the IMAGE derived scenarios up to around 700 mm by 2100. This suggests a similarity between the two IMAGE derived scenarios (i.e., A1B-IMAGE and E1) which is not shared with A1B-SRES.

[24] For the simple-daily intensity index (SDII) (Figure 4e) there is a clear lag in the increase in intensity in the A1B-SRES simulations, compared with A1B-IMAGE. After about 2050 the rates of increase are similar for both of the A1B scenarios, but under E1 the SDII stabilizes by the year 2100.

[25] The more severe multiday precipitation events may be represented by the annual consecutive 5-day (RX5day) maximum precipitation (Figure 4f). The time series of RX5day indicates a small increase in rainfall totals under the scenarios, from around 70 mm in 2000 to between 75 mm and 80 mm by 2100. Again, a similar pattern to SDII emerges with a delay in the increase in RX5day under the A1B-SRES scenario. Under both A1B scenarios RX5day increases throughout the 21st Century, whereas under E1 the values stabilize by the latter part of the century. A similar pattern can be seen in the number of days with over 20 mm of rainfall (R20mm; not shown).

[26] The relationship between precipitation extremes and global mean temperature is weaker than for the temperature extremes. There is a small increase in the RX5day values at higher global mean temperatures (Figure 5c). The scaling of RX5day with global mean precipitation (Figure 5d), expressed as a percentage change from pre-industrial (1861–90), shows high variability, but there is a tendency for higher extremes to be associated with years of higher global mean precipitation. This pattern is repeated across other precipitation indices (not shown). It is worth considering previous work which has looked at the mechanisms underlying the relationship between changes in precipitation (mean and extremes) and the relationship with temperature change. Allen and Ingram [2002] found that global-mean precipitation is primarily constrained by the energy budget, but that the intensity of heavy rainfall events is likely to increase at a faster rate with the availability of moisture. Pall et al. [2007] investigated whether this mechanism operates at regional and seasonal scales, and found that changes in extremes, particularly between 60°N and 60°S, were better predicted by Clausius-Clapeyron than by changes in mean precipitation.

3.3. Regional Changes in Extremes

3.3.1. Temperature

[27] Considering the projected spatial pattern of changes in TXx between 1961 and 1990 and the end of the 21st Century (2070–2099), as shown in Figure 6, the largest changes under the non-mitigation scenarios are over the northern midlatitudes where the annual daily maximum temperature is projected to increase by over 6°C. Under mitigation these changes are lower, typically around 5°C or less. If a mitigation scenario is followed, it is Canada, the northeastern United States, Europe and Russia, which are projected to have the largest avoided changes. Changes are statistically significant at most grid boxes for TXx and the other temperature based indices shown here.

Figure 6.

Projected global changes in TXx (annual maximum temperature). Changes are shown for the period 2070–2099 relative to 1961–1990. Avoided changes show the difference in the change for A1B-SRES minus E1 scenarios. A1B-SRES and E1 maps represent mean of the two ensemble members. Dots indicate regions where changes between the two periods are significant at the 5% level (using a Student's t-test).

[28] Figure 7 shows the spatial pattern of changes in TX90p, and the changes avoided by mitigation by the year 2100. There is a consistent pattern of warming globally, though the greatest increases are seen over northern South America, Africa, the Middle East and parts of southern Asia. Mitigation limits the increase in frequency of warm nights over North America, Europe and northern Asia, as well as Australia. Interestingly, increases over northern South America and central Africa remain large under mitigation though lower than the A1B scenarios. A similar change can be seen for warm nights (TN90p, Figure 8). In studies of changing extremes using observational data, TN90p showed the most globally coherent signal of increasing temperatures of all of the extremes indices considered in this work [Alexander et al., 2006]. Compared to the TXx index, where the larger changes are generally seen over the northern midlatitude regions, the largest changes in the TX90p and TN90p indices are seen closer to the Equator over Africa, the Middle East and southern Asia. Absolute increases in mean temperature and extremes are larger at higher latitudes, but as a result of the lower variability in daily temperatures in the tropics there tends to be a larger increase in the frequency of days exceeding the 90th percentile baseline by a small margin, compared with relatively fewer days exceeding the threshold by a larger margin at higher latitudes [Russo and Sterl, 2011].

Figure 7.

As in Figure 6 but showing projected global changes in the TX90p (warm days) index.

Figure 8.

As in Figure 6 but showing projected global changes in the TN90p (warm nights) index.

[29] Figure 9 shows projections for the 21st Century for the annual warmest day (TXx), this time broken down by the selected countries and regions. Here we consider the ‘BRIC’ countries, of Brazil, Russia, India and China, with the addition of the United States of America, Europe, South Africa and Australia. It is clear that most regions show a similar pattern to the global picture with divergence between the mitigation and non-mitigation scenarios around mid-century, then stabilization under the mitigation scenario, and further increases under the non-mitigation scenarios. By the year 2100 the largest difference between the scenarios is around 4°C, seen over the USA, Europe, and Russia. Other regions show slightly smaller differences, with less of a clear difference showing over India.

Figure 9.

Regional mean time series of annual TXx.

[30] Hegerl et al. [2004] found evidence that changes in seasonal mean temperatures are significantly different from changes in annual temperature extremes over many parts of the globe. We did not investigate the relationship between seasonal mean temperatures and extremes, but do find linearity between temperature extremes (TXx and TN90p) and regional annual mean temperatures over the regions assessed, though the relationships are generally weaker than for the global means (not shown). Of the regions considered, Europe indicates the largest changes in TXx of around 1.6°C per degree of mean warming under the A1B-SRES scenario, assessed using a linear least squares fit over the 21st Century. Russia shows the lowest increase in TXx of around 0.7°C per degree of mean warming. For both of these regions, increases in TXx are lower under mitigation (1.1°C and 0.4°C respectively), and lower rates of increase are also projected over the other regions with the exception of Brazil and India.

3.3.2. Precipitation

[31] The spatial patterns for changes in annual mean precipitation are shown in Figure 10. Under the non-mitigation scenario there are large reductions in precipitation over the Amazon, the Mediterranean region and central Africa. There are also reductions over North America, Europe, Southeast Asia and Australia. As already noted, the E1 and A1B-IMAGE scenarios project a similar precipitation trajectory when considered on the global scale. Compared to A1B-SRES, A1B-IMAGE has a smaller areal extent where large decreases in precipitation occur, such as the Pacific Northwest of the USA and Canada, western Russia, central Africa and parts of South America. Due to the higher spatial variability of precipitation, changes are not statistically significant at all grid boxes. Mitigation acts to reduce the non-intervention reductions in precipitation over northern South America, the eastern USA, Europe, Southeast Asia and much of Africa and the Middle East, and also limits the increases projected over the Saharan region.

Figure 10.

As in Figure 6 but showing projected global changes in PRCPTOT (total precipitation).

[32] Figure 11 shows the projected RX5day spatial pattern of 21st Century change, which is very variable but dominated by increasing rainfall amounts. Under the A1B-SRES scenario, regions with significant projected RX5day decreases are over the Amazon region, North Africa and over some grid boxes in southwestern Africa, and Western Australia. These regions experience smaller rainfall decreases (in some cases actual increases) under the mitigation scenario. However, under mitigation there are smaller increases in rainfall over central Africa than there are under the non-mitigation scenarios. The projected impacts of mitigation suggest avoided increases in RX5day over northern South America, parts of Africa, especially in the north, and central Asia. We also see similar patterns of change between the non-mitigation and mitigation scenarios, in the amount of precipitation falling on days above the 95th percentile, R95p (Figure 12), though the number of grid boxes indicating significant change is low. The main region with avoided decreases in R95p is central Asia, with avoided increases in R95p indicated for the southern USA, eastern South America, central Africa and some Arctic regions.

Figure 11.

As in Figure 6 but showing projected global changes in the RX5day (annual maximum consecutive 5-day rainfall) index.

Figure 12.

As in Figure 6 but showing projected global changes in the R95p (precipitation above the 95th percentile) index.

[33] Due to the smaller spatial scales of precipitation compared with temperature there is much greater variability within the patterns of change of mean precipitation and precipitation extremes indices. However, in terms of the large-scale geographical patterns there is broad agreement between the changes in mean precipitation and those in the precipitation extremes, though some regions indicate differences. For example, much of the USA and Europe demonstrate decreases in mean precipitation, but increases in the R95p index under the A1B scenarios.

4. Discussion and Conclusions

[34] We have compared projected changes in climate extremes under an aggressive mitigation scenario with changes under versions of a non-mitigation scenario using the HadGEM2-AO climate model. Changes in temperature extremes, both globally and regionally, appear to be strongly linked to changes in global mean temperature. There is a clear division between the scenarios in global mean temperature around the mid-21st Century, with temperatures projected to stabilize under the mitigation scenario, but increase by a further 1.5°C by the year 2100 under the non-mitigation scenarios. Temperature related extremes are markedly reduced under a mitigation scenario compared with the non-mitigation scenarios. Under the mitigation and non-mitigation scenarios the Amazon region displays notable increases in absolute daily temperature extremes and in the frequency of warm days and nights. The USA, Europe and Russia are projected to be particularly affected by increasing absolute temperature extremes, and could potentially see the largest limitations of these increases through mitigation compared with other regions. In common with previous studies [e.g., Tebaldi et al., 2006] we find that geographical patterns of change for each extremes index are relatively stable across the different emissions scenario projections, and that the magnitude of change of temperature extremes exhibits a strong relationship with global mean temperature change. As noted by Tebaldi et al. [2006], this supports the use of pattern scaling for future climate projections, particularly for temperature related indices.

[35] The avoided change maps indicate the benefits of the mitigation strategy. For TXx the largest avoided changes occur in the northern midlatitudes, particularly over Russia, but also Canada, the eastern USA, and Europe where avoided changes are of the order of 3°C or above. For TX90p, the key region of avoided change is northern South America, but other regions include southern Africa, the eastern USA, and parts of central and East Asia. Avoided changes in TN90p show some similarity, but affect larger areas of the USA, north and east Africa and central and East Asia. Over the USA projected annual increases in warm nights of at least 50 days are avoided through mitigation by 2100 based upon the scenarios considered here.

[36] Changes in precipitation extremes appear to be less strongly linked to global temperature change, with some regions projected to experience increasing precipitation extremes and some decreasing extremes though it is noted again that the indices used here consider only land-based precipitation, whereas numerous studies have shown that the clearest climate change related precipitation increases tend to occur over ocean regions [e.g., Allen and Ingram, 2002]. The reduction in global mean precipitation during the early 21st Century in HadGEM2-AO under the A1B-SRES scenario, compared to the IMAGE-derived scenarios, has also been noted by Johns et al. [2011]. They explain this as being due to A1B-SRES having a peak in aerosol emissions in 2020, which does not appear in the IMAGE derived A1B and E1 scenarios. HadGEM2-AO appears to have a greater response to this aerosol forcing than other GCMs which they suggest may be due to the inclusion of black carbon and biomass aerosol or the indirect sulphate aerosol effect. The differences in behavior between the precipitation changes under A1B-SRES and A1B-IMAGE point toward a requirement for further investigation of the effect of aerosols on precipitation changes. Min et al. [2011], in their supplementary material, note that the aerosol influence on extreme precipitation is still poorly understood and requires further investigation [Khain, 2009]. They discuss relevant mechanisms related to aerosols which may affect precipitation, which include changes in shortwave forcing. Shiogama et al. [2010] note that the uncertainty in the hydrological sensitivity (i.e., the relationship between precipitation change per unit global mean temperature change) has been considered, but little work has been done in comparisons under different emissions scenarios and aerosols. They show a robust emissions scenario dependency, with lower sensitivities occurring under higher greenhouse gas and aerosol emissions scenarios, and that it is the differing aerosol emissions that lead to this difference. They also highlight the potential implications for the use of pattern scaling approaches and the assumption that the scaling pattern of a model is consistent under different emissions scenarios. From the experiments and analyses presented here, it appears that the emissions scenario signal in future precipitation extremes, as based upon prescribed CO2 concentrations, may be complicated by differences in the aerosol loading between the IMAGE and SRES derived scenarios Understanding the influence of aerosols on precipitation could also be of considerable value in determining the impacts of using certain geo-engineering approaches to mitigating climate change, which we did not consider in this study. Recent studies of schemes which propose injecting aerosols into the atmosphere have highlighted the issue of potential regional impacts resulting from this form of climate modification, particularly in relation to precipitation patterns [e.g., Jones et al., 2009].

[37] For total precipitation, the projections indicate that large absolute decreases in precipitation could be avoided over northern South America through mitigation. Other regions with avoided drying include west central Africa, China and Indochina, the eastern USA, and western Europe. There are avoided increases in precipitation over parts of northern Africa, South America, the Himalayas, and Arctic regions. RX5day shows large-scale similarities to total precipitation, but with more spatial variability. Projected increases in RX5day extremes are avoided over parts of the USA, China and eastern Australia. R95p again shows broadly similar large-scale patterns, but indicates avoided increases over Saharan Africa and the Middle East. Increases in R95p are avoided in the USA east of the Rocky Mountains.

[38] The mechanisms contributing to the occurrence of extreme events are varied, and may depend on the time of year or the climatic mechanisms and interactions which influence a particular region. Kenyon and Hegerl [2008, 2010] examined the influence of large-scale modes of climate variability, such as the El Niño-Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO), on global temperature and precipitation extremes. They found that temperature extremes are strongly influenced by large-scale circulation patterns. There are particular regional patterns, with ENSO showing a clear influence around the Pacific Rim and in North America and the NAO has a strong impact upon Eurasia. For precipitation extremes similar relationships were found. Alexander et al. [2009] also found that global sea surface temperature anomaly patterns (linked to ENSO) are important for the modulation of extreme temperature and precipitation around the globe. Therefore to better understand future changes in climate extremes it would also be necessary to investigate the behavior of these large-scale modes of variability in the future under different emissions scenarios, although recent studies suggest that the effect of climate change on the tropical Pacific climate system remains uncertain [Latif and Keenlyside, 2009].

[39] This study has only utilized a single configuration of the HadGEM2 model. Previous studies have found that simulated extremes in models can share large-scale features, but can still be very model dependent [e.g., Allen and Ingram, 2002; Hegerl et al., 2004]. As we have established, HadGEM2-AO does a good job at simulating the large-scale features seen in observations, but we have identified areas where it may not perform so well such as the Indian monsoon region, though this is seen across other configurations of the HadGEM model as well as other CMIP5 models (e.g., R. C. Levine et al., submitted manuscript, 2012).

[40] The results presented here from the HadGEM2-AO model indicate that, even under an aggressive mitigation emissions scenario, there are projected increases in warm extremes on a global and regional scale up to the mid-21st Century which indicates that adaptation measures would still be required. However, the projections indicate that beyond the mid-21st Century these increases tend to stabilize along with global mean temperatures. For precipitation extremes the results are less clear. There are regional differences in the direction of change, some of the indices display a high level of inter-annual variability, and it appears that the aerosol forcing associated with the scenarios may have an important influence.


[41] This work was supported by the AVOID program (DECC and Defra) under contract GA0215. The model experiments used in this study were run as part of the European Commission funded ENSEMBLES project. We acknowledge the contributions of Yang Feng (Environment Canada) for support with the FClimDex software and the two anonymous reviewers for their insightful comments and suggestions.