4418 JOURNAL OF CLIMATE VOLUME 19

Modeling Northern Hemisphere Summer Heat Extreme Changes and Their Uncertainties Using a Physics Ensemble of Climate Sensitivity Experiments

ROBIN T. CLARK,SIMON J. BROWN, AND JAMES M. MURPHY Hadley Centre, Met Office, Exeter, United Kingdom

(Manuscript received 3 June 2005, in final form 1 December 2005)

ABSTRACT

Changes in extreme daily temperature events are examined using a perturbed physics ensemble of global

model simulations under present-day and doubled CO2 climates where ensemble members differ in their representation of various physical processes. Modeling uncertainties are quantified by varying poorly con- strained model parameters that control atmospheric processes and feedbacks and analyzing the ensemble spread of simulated changes. In general, uncertainty is up to 50% of projected changes in extreme heat events of the type that occur only once per year. Large changes are seen in distributions of daily maximum temperatures for June, July, and August with significant shifts to warmer conditions. Changes in extremely hot days are shown to be significantly larger than changes in mean values in some regions. The intensity, duration, and frequency of summer heat waves are expected to be substantially greater over all continents. The largest changes are found over Europe, North and South America, and East Asia. Reductions in soil moisture, number of wet days, and nocturnal cooling are identified as significant factors responsible for the changes. Although uncertainty associated with the magnitude of expected changes is large in places, it does not bring into question the sign or nature of the projected changes. Even with the most conservative simula- tions, hot extreme events are still expected to substantially increase in intensity, duration, and frequency. This ensemble, however, does not represent the full range of uncertainty associated with future projections; for example, the effects of multiple parameter perturbations are neglected, as are the effects of structural changes to the basic nature of the parameterization schemes in the model.

1. Introduction eral papers have also reviewed the extremely unusual 2003 European summer. Trigo (2005), Luterbacher et Several authors have examined changes in extreme al. (2004), Schar et al. (2004), and Fink et al. (2004) temperature events observed during the twentieth cen- concluded that the anomalously hot event was excep- tury. Horton et al. (2001) found an increase in warm tional both in terms of its duration and intensity. Pos- extremes in England although a reduction in total inci- sible mechanisms behind the summer, examined by dence of extremes. Frich et al. (2002) reported signifi- Black et al. (2004), Ogi et al. (2005), Beniston and Diaz cantly longer durations observed in several (2004), and Trigo et al. (2005) point to soil moisture, areas of the world and a global upward trend in heat long periods of clear-sky conditions, large-scale atmo- wave duration although this was not statistically signifi- spheric blocking, subsidence, the Northern Hemisphere cant. Easterling et al. (2000a,b) found a cluster of heat annular mode, northward extension of the Hadley cell, waves during the 1930s but no trend in heat extremes in and an upper-troposphere double jet as factors. the United States, although there appears to have been Future changes in extreme events, resulting from in- general warming trends in minimum and average tem- creased atmospheric concentrations of CO , other trace peratures. Manton et al. (2001) found significant in- 2 gases, and aerosols, have been examined far less than creases in the annual number of hot days in Southeast changes in the mean. This is due to the very nature of Asia and the South Pacific between 1961 and 1998. Sev- extreme events requiring very long simulations in order to produce robust results. It is only since the mid-1990s Corresponding author address: Robin Clark, Hadley Centre, that advances in supercomputer speed have allowed Met Office, FitzRoy Road, Exeter, EX1 3PB, United Kingdom. such exercises to take place. Zwiers and Kharin (1998), E-mail: [email protected] examining two global 20-yr simulations of doubled

JCLI3877 1SEPTEMBER 2006 C L A R K E T A L . 4419

CO2, reported a 1-in-20-yr event intensity increase of fixed threshold defined by the 99th percentile of the between 5° and 10°C for maximum temperature over present-day simulation distribution. Here, we use the North and South America and Eurasia. Meehl and same ensemble to examine changes in extremes from a Tebaldi (2004) examined changes to 3-day heat events different standpoint. We focus on the change in the using the mean of a five-member global coupled model temperature associated with an event of given fre- ensemble driven by changes in CO2,SO2,CH4, and quency in the doubled CO2 simulation, compared to its N2O emissions following a “business-as-usual” path value in the present-day simulation. This is referred to through to 2100. They found significant increases in the throughout as the intensity of the event. We further intensity, duration, and frequency of these events over extend the work of Barnett et al. by considering parts of North America and Europe. Kharin and changes for all percentiles of the distribution of daily Zwiers (2000) found that substantially greater changes events, and by considering heat waves lasting several occur for daily minimum temperature extremes than days as well as individual daily events. These extensions daily maximum temperature extremes. The greatest make the analysis potentially useful for a wide range of change in daily maximum temperatures was found in users in determining possible impacts. central and southeast North America, central and Southeast Asia, and tropical Africa where there were 2. Model projections also substantial decreases in summertime soil moisture content. Hayhoe et al. (2004) expect extreme heat The 53-member PPE used the Third Hadley Centre events over California to become 4 to 8 times more Atmospheric Model (HadAM3) global circulation common by 2100, with the range predominantly due to model (Pope et al. 2000) coupled to a 50-m nondynamic uncertainty in future greenhouse gas emissions. Schar mixed layer (“slab”) ocean. Model resolution was 2.5° et al. (2004) reported that a regional climate model latitude by 3.75° longitude with a vertical resolution of simulation for Europe, driven by a single greenhouse 19 levels. In each ensemble member, a single uncertain gas scenario, shows that toward the end of the twenty- parameter in the atmospheric model was perturbed first century every second summer could be as warm or away from a standard setting to a high, medium, or low warmer than the extremely unusual 2003 European value. These perturbations were chosen by experts with summer [for an assessment of this event see Fink et al. the aim of giving a range of simulations, which, al- (2004)]. Stott et al. (2004) report a hundredfold in- though different from the standard setting, would still crease in the expected frequency of 2003-type summers be just as plausible. The results from this experiment by the mid-twenty-first century under the Special Re- should be regarded as a lower limit on the range of port on Emissions Scenarios (SRES) A2 scenario (Na- possible changes for several reasons. First, only one kicenovic and Swart 2000). parameter was perturbed at a time, so the ensemble However, the important issue of the uncertainty in does not sample the entire model parameter space. Sec- the predicted changes has generally been neglected, es- ond, structural perturbations, for example, changes in pecially in earlier papers, because of limited computing grid resolution or changes in the fundamental physical resources. Uncertainty in producing projections of fu- assumptions employed in the model parameterization ture climate can be characterized into three main schemes, were not considered. Finally, the use of a slab sources—natural variability, uncertainties in green- ocean precluded quantification of uncertainties arising house gas and aerosol emissions, and their resulting from changes in ocean circulation and reduced the in- atmospheric concentrations and uncertainties in arising fluence of sea surface temperature variability. from imperfect representation of the real climate sys- Each of the 53 model formulations were run for 20 yr tem in models. Ideally, probabilistic projections of fu- beyond equilibrium for the present-day and doubled ture extreme event frequency and intensity are re- CO2 simulation. Murphy et al. (2004) and Barnett et al. quired; however, this would require a very large en- (2006) give more details in the supporting information semble to sample sufficiently all sources of uncertainty for these papers. Twenty years of daily maximum 1.5-m and to characterize their likelihood. As a step toward air temperature data, covering June, July, and August such projections, Barnett et al. (2006) used model data (JJA) from each ensemble member, were extracted from the 53-member Hadley Centre perturbed physics from both the present-day and doubled CO2 integra- ensemble (PPE) of Murphy et al. (2004) to examine tions. This gave two samples of 1800 values for each regional changes in frequency of extreme heat events model grid point and member, which were analyzed for associated with the equilibrium response to doubled changes in extreme events. Although periods of data

CO2. They found globally averaged increases of at least longer than 20 yr would have been more suited to ex- 20 in the number of occurrences of days warmer than a treme event analysis, they were not available as a result 4420 JOURNAL OF CLIMATE VOLUME 19 of restricted computer resources. Nevertheless, the da- simulations do show more areas of positive asymmetry tabase of model output corresponds to nearly 200 000 than are observed, notably over central and eastern simulated days. Russia, the eastern United States, and the Indonesian In general, differences between the responses of dif- region. ferent ensemble members are driven both by the effects In general, the biases shown in Fig. 1 arise from com- of the parameter perturbations and by the effects of plex interactions between different processes operating natural variability (Murphy et al. 2004). Here we do not in the model and are difficult to ascribe unambiguously attempt to quantify their relative contributions, though to specific causes. We note, however, that the magni- this is recognized as an important issue for future work. tudes of the regional errors in long-term averaged tem- perature are typical of those found in other climate Validation of simulated daily distributions models (e.g., Giorgi and Francisco 2000). The warm For the purposes of validation we follow the ap- biases over some southern and central regions of the proach similar to Ferro et al. (2005) by comparing the North American and Eurasian continents coincide with median, interquartile range and Yule–Kendall skew- positive biases in the shortwave radiative forcing due to Ϫ2 ness measure (termed location, scale, and asymmetry cloud, exceeding 15 Wm in places. These regions also hereafter) of the simulated daily distributions of diurnal show some of the largest biases in simulated scale, par- maximum surface air temperature with daily observa- ticularly where warm biases in temperature coincide tions from 1950 to 2000 (Caesar et al. 2006). Here, how- with negative biases in simulated precipitation (not ever, the scale is taken as the difference between the shown). This is likely to reflect deficits in evaporation 95th and 5th percentile and the Yule–Kendall skewness arising from insufficient soil moisture in the model is defined as (Brabson et al. 2005). In summary, although the simulated large-scale pat- ϭ ͓͑ Ϫ ͒ Ϫ ͑ Ϫ ͔͒ր͑ Ϫ ͒ a pc95 pc50 pc50 pc5 pc95 pc5 , terns agree with observations quite well, there are sig- nificant regional differences that need to be taken into where pc represents the ith percentile. i account when assessing the modeled changes due to We calculate the location, scale, and asymmetry di- increased CO . The discrepancies between the simu- agnostics simulated for present-day climate by each en- 2 lated and observed present-day distributions illustrate semble member, form the ensemble mean values, and the point made earlier that not all sources of modeling plot these alongside corresponding observations in Fig. uncertainty are captured by this ensemble. For ex- 1. The simulated and observed geographical distribu- ample, at 50% of the grid points, the observed location tions agree generally quite well, but with some regional differences. For most areas for which there are obser- lies outside the range of the 53 location values spanned vations the simulated mean daily temperatures are too by the ensemble. Hopefully future ensembles will cool (location too small; see Fig. 1a) particularly at high sample modeling uncertainty more comprehensively, northern latitudes where the model is too cool by typi- resulting in ensemble distributions of location, scale, cally 4°C, although the bias reaches up to 8°C in places and asymmetry that encompass the observed values to (e.g., Alaska). In many other regions (southern North a greater extent than found here. Nevertheless, we find America, eastern Europe, northern Australia, northern that the model used in our study is able to reproduce South Africa, and southern Brazil) the simulated loca- observed changes in extreme daily temperatures during tion is larger than observations (typically by 2° to 4°C). the twentieth century (Christidis et al. 2005), suggesting For scale, the geographic variation is similar between that the model physics is capable of capturing climate observations and model, but the model values are gen- feedback processes relevant to the prediction of future erally larger. This is particularly so for the central changes in extremes. If, however, the model bias results United States, central Eurasia, South Africa, the Ibe- in a region making a transition to or from a climatic rian Peninsula, and the interior of Australia where the regime with increased CO2 that would not have oc- simulated scale is too large by 40% or more. On the curred had the model not had such biases, then the other hand, the simulated scale is too small for central projected results will be in error. In section 5, for ex- Canada and small subregions of Europe, Brazil, and ample, we show that changes in the ability of evapora- Africa, by typically 10%–20%. The observations show tion to restrict increases in temperature play an impor- negative asymmetry in all regions apart from the north- tant role in determining the magnitude of simulated west coasts of northern continents, southern Brazil, and changes in extreme temperatures. The changes in southern Australia. The model captures the general evaporation would in turn depend on the levels of soil predominance of negative asymmetry, although the moisture present in the present-day integrations, so bi- 1SEPTEMBER 2006 C L A R K E T A L . 4421

FIG. 1. Ensemble average values of (a), (b) location, (c), (d) scale (°C), and (e), (f) asymmetry (as fraction of scale), for (left) present-day model simulations and (right) values from 1950–2000 observations (Caesar et al. 2006).

For a given model integration or observed time series, location is defined as pc50, scale as pc95 minus pc5, and Ϫ Ϫ Ϫ asymmetry as [(pc95 pc50) (pc50 pc5)]/scale, where pci represents the ith percentile using all JJA daily data for a single member and CO2 forcing. ases in soil moisture would reduce the plausibility of the The grid points selected for the main analysis are taken simulated changes in some regions. from southwestern France, the Czech Republic, the Great Lakes region of North America, and eastern 3. Methodology China. The southwestern France and Great Lakes grid points were chosen since these two areas experienced To save space we use results from a small number of very large societal impacts resulting from heat extremes grid boxes to illustrate our main points. Where appro- in 2003 (INVS 2003) and 1995 (Kunkel et al. 1996), priate, however, we support these by maps of global respectively. The Czech Republic grid point was chosen changes to emphasize the generality of key conclusions. since this region is representative of large areas of cen-

Fig 1 live 4/C 4422 JOURNAL OF CLIMATE VOLUME 19 tral Europe that could suffer substantial damaging im- the distributions gives information about general pacts should the incidence and/or intensity of heat changes in the distribution shape and, in particular, ex- waves increase. Finally, the inclusion of a grid point treme behavior. These are shown for the four example from eastern China was warranted by its very large grid points in Figs. 2a–d for both the present-day and populations, particularly in large cities with the poten- doubled CO2 ensembles of simulations. In the Czech tial for substantial impacts on health from heat ex- Republic (Fig. 2a), most ensemble members show a tremes. While it is somewhat questionable to take general shift to warmer conditions under CO2 doubling single gridpoint values from a GCM as representative with little change in distribution shape apart from a of local climates, we do it here merely to illustrate the slight extension of the warm tail. In the Great Lakes main characteristics of changes and how these may vary region of the Unites States (Fig. 2b), a general shift to from one region to another. warmer conditions is accompanied by a subtle change The following measures were used to characterize in shape with a wider distribution for temperatures the changes in extreme heat events in June, July, and greater than the mode and a broader maximum fre- August: frequency distributions of all daily maximum quency peak. More complex changes in distribution temperatures, changes in the intensity of daily extreme peak, width, and shape are found in the distributions events measured by temperature values associated with for eastern China (Fig. 2c) and especially southwestern distribution percentiles, and changes in the intensity France (Fig. 2d) with the distribution exhibiting a bi- and frequency of heat wave events. modal structure under doubled CO2 with the two The advent of ensembles of simulations of climate modes, peaking at 32° and 40°C in eastern China and change, either using models developed at different cen- 25° and 40°C in southwestern France. We searched the ters (e.g., Räisänen 2001), or perturbed physics en- present-day simulations for evidence of bimodality un- sembles (Murphy et al. 2004; Stainforth et al. 2005), der present-day conditions. Only three grid points (out raises the issue of how different model versions should of 2400) were found with distinct bimodality, although be weighted when using the ensemble results to con- several had broad unimodal distributions. Furthermore, struct probabilistic predictions. Murphy et al. (2004) observations of terrestrial daily maximum temperature estimated a probability distribution for global climate distributions covering much of the globe for 1950–2000 sensitivity, weighting ensemble members according to a (Caesar et al. 2006) do not show any evidence of bimo- metric based on global spatial distributions of present- dality. This demonstrates that the evidence of bimodal- day climate means of 32 different variables. On the ity in the simulated response to doubled CO2 repre- other hand, Tebaldi et al. (2004, 2005) used regional sents a clear regime change in the nature of regional time and spatial means of present-day precipitation and climate variability. Detailed analysis of the mechanisms surface temperature, respectively, to weight predictions driving these changes is beyond the scope of this paper of regional changes in the corresponding variable. but will be undertaken in future. However, methods of producing likelihood-weighted b. Percentile analysis probabilistic predictions are still in their infancy and are subject to significant assumptions surrounding issues Daily maximum temperatures were ranked on a grid- such as the choice of observational constraints: the de- point by gridpoint basis, from which percentile values sign of ensemble and choices made in the Bayesian were calculated for each individual ensemble member methodology to convert the ensemble results and ob- and CO2 concentration. Changes in the temperature servational constraints into posterior probabilities. values associated with each percentile (referred to as Pending further investigation of these issues we take a changes in intensity) were then calculated for each simple approach of regarding all ensemble members as member. Uncertainty in changes for each percentile is equally plausible, reserving the construction of proba- expressed as a 10%–90% range by ranking the changes bilistic estimates of changes in extremes for future and taking the 5th and 47th of the 53 members. work. Ensemble mean changes for each percentile are shown in Figs. 3a–d together with the derived 10%– 4. Results 90% ensemble range, for our selected grid points. All the changes are positive. A general trend is seen in a. Frequency distributions of daily data which increases become larger toward the hotter per- Frequency distributions of maximum daily tempera- centiles. In the Czech Republic (Fig. 3a), this trend ture were produced using a set of 100 equally sized bins continues all the way to the most extreme percentiles. defined over the range from the ensemble at each grid However, in the other three regions (Figs. 3b,c,d), the point. Examining the changes in shape, size, and tails of largest increases occur in more moderate percentiles. In 1SEPTEMBER 2006 C L A R K E T A L . 4423

FIG. 2. Daily max JJA temperature distributions of all individual ensemble members for grid points representing (a) the Czech Republic, (b) the Great Lakes region of the United States, (c) eastern China, and (d) southwestern

France under present-day (blue lines) and doubled (red lines) CO2 climates. Black lines show equivalent tem- perature distribution of gridded daily station data from 1950 to 2000 from Caesar et al. (2006). section 5a we demonstrate that the largest changes oc- Fig. 3 by the vertical distances between the lower and cur for percentile ranges that have soil moisture avail- upper curves, which represent 80% of the ranked en- able for cooling through evaporation and evapotrans- semble spread. This uncertainty is up to 50% of the piration during present-day climate, but not under CO2 magnitude of the increases in temperature but does not doubling. On the other hand, the most extreme percen- question the sign of changes and the general tendency tiles are already associated with dry conditions during of greater changes at higher percentiles. present-day climate and so experience more moderate Figure 4 places the results of Fig. 3 in a global context increases as there is no change from a regime with by showing maps of changes in the intensity of extreme moisture-derived cooling to one without. hot days, expressed by the 99th percentile (the thresh- Uncertainties in percentile changes are quantified in old reached approximately once a year or less). Simu-

Fig 2 live 4/C 4424 JOURNAL OF CLIMATE VOLUME 19

FIG. 3. Changes in percentile values of JJA daily max temperature between simulated present-day and

doubled CO2 climates for (a) the Czech Republic, (b) the Great Lakes region of the United States, (c) eastern China, and (d) southwesterns France. The middle line shows ensemble mean changes. The upper and lower lines represent the 10%–90% uncertainty estimate of ensemble spread. The 99th percentile can be interpreted as an event that occurs no more than once per year. lated intensities show substantial increases in most ter- tile are substantially different from changes in the me- restrial regions. Even the lower end of the ensemble dian. Correlations between 99th percentile and median distribution of changes (Fig. 4b) shows significant in- changes (Fig. 5b) are found to be lower than those creases in many areas, exceeding 6°C for much of Eu- found by Barnett et al. (2006). rope, North and South America, and eastern Asia. The To investigate these changes in distributional charac- upper end of the ensemble distribution (Fig. 4a) shows teristics in more detail, ensemble mean changes in the that increases greater than 10°C could occur in these location, scale, and asymmetry (defined in section 2) areas. Although the uncertainty across the ensemble, are plotted in Fig. 6. Areas where a uniform shift in the expressed as the difference between Figs. 4a and 4b, distribution of daily temperatures is insufficient to ex- can be large in nearly all regions, the sign of the changes plain the changes in extreme values include the central is not brought into question. The change in extreme United States, Brazil, Europe, and China. These are temperatures is principally due to the general shift of areas where changes in scale and asymmetry are rela- the daily distribution, with changes in the median cap- tively larger. The changes in scale and asymmetry are turing most of the change in the extremes (Fig. 5a) as found to be positively correlated, with increases in scale pointed out by Barnett et al. (2006). However, in some generally accompanied with increases in asymmetry areas, including parts of Europe, northeastern North and vice versa. Exceptions to this are northeastern Bra- America, and eastern Asia, changes in the 99th percen- zil and the Balkans, where increases in scale are accom- 1SEPTEMBER 2006 C L A R K E T A L . 4425

FIG. 4. (a) Upper 90% and (b) lower 10% values of the ensemble range of changes in the 99th percentile value

of JJA daily max surface temperatures due to a doubling of atmospheric CO2. Units: °C. Areas of red indicate

warmer conditions under double CO2. The 99th percentile can be interpreted as an event that occurs no more than once per year. panied by reductions in asymmetry. In these areas annual variability is defined as the average of the 95th (which include the four grid points presented in Figs. 2 minus 5th percentile differences calculated from indi- and 3) it is clear that changes in the shape of the daily vidual years. The results shown in Fig. 6b are the dif- temperature distributions are having a greater contri- ferences between the doubled CO2 and present-day bution to the changes in the intensity of extremes. values averaged over all ensemble members. We calcu- Changes in scale and asymmetry exert a significant in- late the change in interannual variability as the differ- fluence on the changes in extremes found here, since we ence between the change in scale (Fig. 6b) and the consider changes in the intensity of events of a given change in intra-annual variability (Fig. 6d). The frequency. In contrast, Barnett et al. (2006) considered changes in intra-annual variability are typically consid- changes in the frequency of exceedance of a fixed erably larger than changes in interannual variability, threshold, and obtained results showing less sensitivity with many regions increasing by Ͼ4°C (Europe, China, to changes in the shape of the daily distributions. Brazil, and the U.S./Canadian border). Schar et al. Measures of changes in intra-annual variability and (2004) find substantial relative increases in interannual interannual variability are also plotted, (Figs. 6d and 6e, variability for some regions, particularly Europe, and respectively). For a given ensemble member, intra- similar increases are found here when comparable vari-

Fig 4 live 4/C 4426 JOURNAL OF CLIMATE VOLUME 19

FIG. 5. (a) Ensemble mean difference between changes in 99th and 50th percentiles as percentage proportion of 50th percentile change for JJA daily max surface temperature distributions in response to a doubling of atmo-

spheric CO2. Units: %. Areas of red indicate that increase in 99th percentile is greater than of 50th percentile. (b) Correlation between changes in simulated 99th percentile and changes in pseudo-99th percentile, calculated by applying 50th percentile changes to the 99th percentile of the present-day conditions. Areas of red show stronger correlation between simulated and pseudo-99th percentile values. ability measures are used. However, this analysis sug- single long-duration, moderately extreme event can gests that it is changes in intra-annual variability that also have a greater impact, particularly on health, than dominate the model response. multiple short-lived, more extreme events (WHO 2004). Here we define a heat wave event as a consecu- c. Heat wave analysis tive sequence of days in which each day exceeds a given Here we analyze changes in the nature of extreme temperature threshold. To quantify changes in heat heat wave events, reflecting their potentially damaging wave events, time series were analyzed at each grid impacts on society (García-Herrera et al. 2005; Trigo et point to determine the threshold associated with a 1-in- al. 2005; Changnon et al. 1996). Event duration and 20-yr event for heat waves of a given duration. Figures timing is often as or more important than intensity and/ 7a–d show plots of the change in intensity of events in or event frequency. Heat waves occurring early in the response to doubled CO2 as a function of heat wave year before people have acclimatized to summer can duration. As mentioned previously, the 10%–90% un- have greater impacts than those occurring later. A certainty range is shown by ranking across the en-

Fig 5 live 4/C 1SEPTEMBER 2006 C L A R K E T A L . 4427

semble. The simulations show increases in the intensity of heat waves of all durations. For example, in the Great Lakes region (Fig. 7b), a 1-in-20-yr event lasting 5 days has an intensity range of between 28° and 34°C (10%–90% ensemble range) under present-day condi- tions. This range becomes 38° to 44°C in response to

CO2 doubling, that is, a general increase of 10°C. The mean and spread of the ensemble of simulated changes for more moderate 1-in-2-yr events (not shown) is typi- cally about 50% smaller. A notable result in all four regions is the constant nature of the changes with du- ration—that is, intensity changes of longer-duration events are very similar to those of shorter durations. Furthermore, for a given threshold, heat wave duration

increases with CO2 doubling. For example, an event of between 30° and 36°C (10%–90% ensemble range) in the Great Lakes region has a duration of 2 days under

present-day conditions. This becomes 17 days with CO2 doubling. This is also seen in many other regions (not shown). Global maps of upper (90%) and lower (10%) values of the ensemble range of intensity changes for hottest 1-in-20-yr 10-day events are presented in Figs. 8a,b. The lower estimate of intensity change (Fig. 8b) shows increases greater than 6°C across large areas of Europe, North and South America, and eastern Asia. Upper-limit increases exceed 10°C in these areas (Fig. 8a) with remaining areas of the globe predominantly greater than 4°C. Figures 8c,d present changes in heat waves measured in a different way. Here we define the threshold associated with a 1-in-20-yr 10-day event from the present-day simulation and calculate the change in the frequency of exceedance of that threshold

when CO2 is doubled. From Fig. 8d, the frequency in- creases by a factor of 50 according to the lower end of the ensemble range over western North America as well as large parts of Africa and South America. Inter- estingly, regions of greatest frequency changes are not necessarily the same as those of largest intensity change. The upper estimate of the ensemble range (Fig. 8c) shows such events becoming at least 50 times more frequent over 85% of non-Antarctic land areas.

5. Exploration of mechanisms

Increases in extreme high temperatures are found to

FIG. 6. Ensemble mean change in (a) location, (b) scale, (c) be larger than increases in average temperatures for asymmetry, (d) intra-annual variability, and (e) interannual vari- certain regions in this ensemble. Two key factors con- ability of daily JJA max temperature due to CO2 doubling. Lo- tributing to this result are described here. Ϫ cation is given by pc50, scale by pc95 pc5, and asymmetry by Ϫ Ϫ Ϫ [(pc95 pc50) (pc50 pc5)]/scale, where pci represents the ith percentile using all JJA daily data for a given model integration. a. Effects due to soil moisture Intra-annual variability is calculated as the mean of (pc Ϫ pc ) 95 5 Cooling through direct evaporation of soil moisture, differences between doubled CO2 and present-day integrations averaged over all individual years and all ensemble members. or through the evapotranspiration of plants, has a con- Interannual variability is defined as (b) minus (d).

Fig 6 live 4/C 4428 JOURNAL OF CLIMATE VOLUME 19

FIG. 7. Simulated intensities as a function of duration of 1-in-20-yr return period heat waves in JJA, defined as sequences of days on which the max daily temperature exceeds a given temperature thresh- old. Results are shown for (a) the Czech Republic, (b) the Great Lakes region of the United States, (c)

eastern China, and (d) southwestern France under present-day (light lines) and double CO2 (bold lines) concentrations. The vertical bars represent the 10%–90% uncertainty range across the ensemble. trolling effect on maximum air temperature close to the against daily maximum temperature percentile (not surface, limiting high values. However, if this process shown) suggested that the highest temperatures tend to reduces or even ceases due to absence of moisture, per- coincide with the smallest soil moisture values, albeit haps resulting from a prolonged drought, then air tem- with significant scatter in the daily data. To reduce peratures can rise disproportionately. Indeed, this is noise, ranked temperature values were subdivided into partly thought to be a cause of the anomalously dry and equal portions each containing 2% of the data. Corre- hot European summer of 2003 according to Black et al. sponding average daily soil moisture was then calcu- (2004). Furthermore, Brabson et al. (2005) and Durre lated for each subdivision of temperature. This was et al. (2000) also relate temperature extremes to low done separately for each ensemble member. Results for soil moisture. To assess whether such mechanisms are southwestern France are given in Fig. 9a and show a at work here, maximum daily temperatures have been clear relationship between soil moisture and tempera- compared with soil moisture and the frequency of dry ture percentile. This relationship was present in all en- days. semble members with other regions showing similar re- An initial analysis of gridbox JJA soil moisture sults (not shown). Beyond the 80th temperature per- 1SEPTEMBER 2006 C L A R K E T A L . 4429

FIG. 8. (a) Upper 90% and (b) lower 10% values of the ensemble range of

changes in intensity (°C), due to a doubling of atmospheric CO2, associated with 1-in-20-yr JJA temperature heat wave events lasting 10 consecutive days. (c), (d) The 90% and 10% values of the ensemble range in the relative fre-

quency of 1-in-20-yr events in the doubled CO2 simulations relative to the present-day simulation. Areas of red indicate warmer events in (a) and (b) and

larger increases in frequency in (c) and (d) under doubled CO2.

Fig 8 live 4/C 4430 JOURNAL OF CLIMATE VOLUME 19

FIG. 9. Soil moisture vs daily max temperature relationships for the southwestern France grid point in

JJA under present-day (thin lines) and doubled (thick lines) CO2 conditions. (a) Max temperature percentile vs soil moisture. (b) Soil moisture vs max temperature. Plots were produced by ranking the x-axis parameter separately for each ensemble member and calculating the average corresponding y-axis parameter for each 2% block of x-axis parameter. The middle line of each set of three shows the ensemble median relationship with upper and lower lines showing the 10%–90% ranked ensemble range of the y-axis parameter. centile, the soil moisture–temperature relationship Jones (2006) identify an earlier decline in soil moisture changes between the present-day and doubled CO2 cli- during spring, enhanced land–sea contrast in lower- mates. Whilst soil moisture continues to fall as tempera- tropospheric temperatures, remotely forced circulation ture increases under present-day conditions, the depen- changes, and local rainfall changes as contributors to dence of temperature on soil moisture under doubled summer drying over Europe. Here we do not attempt a

CO2 is increasingly reduced as soil moisture approaches detailed analysis, but we do note that an analysis (not zero. shown) of the number of dry JJA days versus extreme

The change in soil moisture with CO2 doubling is temperatures (similar to that performed above for soil largest for the 60th to 80th percentiles in Fig. 9a, which moisture) indicates a clear tendency for higher maxi- corresponds to the largest changes in the maximum mum daily temperatures in years with greater numbers temperature percentile values shown in Fig. 3d. This of dry days. Doubling CO2 concentrations results in suggests that for this range of percentile values there is large increases in the number of dry JJA days in Eu- a transition from a regime where moisture can limit rope and many other large areas (not shown), which temperature to one where there is insufficient moisture contributes to increases in extreme maximum tempera- to do so. This close correspondence between the per- tures through reduced recharge of soil moisture from centiles showing the largest changes in temperature and precipitation. soil moisture is found to apply to many locations (Fig. A similar analysis to that in Fig. 9a, showing soil 10), particularly in the summer hemisphere. An addi- moisture against actual daily maximum temperature, is tional indicator of such a pattern is shown by the geo- presented in Fig. 9b. From this one can see that for a graphical distribution of the 99th–50th percentile signal given soil moisture value, maximum temperatures are shown in Fig. 5a. In the northern extratropics, this sig- greater under increased CO2, with the largest changes nal is best described as a transition zone (located be- occurring for the lowest soil moisture values. The mag- tween 40°and 50°N) between the semiarid climate to nitudes of these temperature changes are approxi- the south and more temperate climate farther north. mately half the total changes seen in the model. Similar transition zones have also been found to be b. Consecutive daytime temperatures and nocturnal preferred locations for strong land surface–atmosphere cooling interactions in Koster et al. (2004). Regional changes in soil moisture are potentially af- In this section the change in temperature between fected by several factors. For example, Rowell and consecutive days and the intervening nocturnal cooling 1SEPTEMBER 2006 C L A R K E T A L . 4431

FIG. 10. Daily JJA max temperature percentile (a) that experiences the greatest changes with CO2 doubling and (b) at which the soil moisture changes the most.

(defined as the difference between the maximum tem- varies with minimum temperature for southwestern perature and the minimum value on the following day) France. With CO2 doubling, slight increases in daytime is analyzed to see how the modeled increase in heat heating occur for most minimum temperatures but with wave events might be occurring. The magnitude of av- little change for higher minimum temperatures. On a erage nocturnal cooling increases with the preceding global scale (not shown) there was a general increase in day’s temperature for both levels of CO2. This is shown daytime heating, but (with the exception of eastern in Fig. 11a for southwestern France but is typical of North America where increases of 8°C were seen) in- other land points. However, under increased CO2, the creases were generally smaller than 4°C. As a result, on following night’s cooling is found to be considerably a global scale (with the exceptions of eastern North reduced for all daytime temperatures (the vertical shift America and Greenland), the magnitude of nocturnal in Fig. 11a between the two sets of lines). On a global cooling reduction was generally much greater than in- scale, reductions were found over all land areas except creases in daytime heating. To see the effect of reduced Greenland. Reductions over Europe and North nocturnal cooling on heat waves, increases in daily America ranged from 4° to 16°C according to location. maximum temperature for consecutive days (“upsteps” Figure 11b shows how daytime heating (defined as hereafter) are compared with the intervening nocturnal maximum temperature minus minimum temperature) cooling for southwestern France (Fig. 11c). This shows

Fig 10 live 4/C 4432 JOURNAL OF CLIMATE VOLUME 19

FIG. 11. Relationships between daily max temperature, nocturnal cooling on the following night, and max temperature on the following day for the southwestern France grid point during JJA under present-

day (thin lines) and doubled (thick lines) CO2 conditions. (a) Max temperature vs nocturnal cooling. (b) Min temperature vs daytime heating. (c) Nocturnal cooling vs positive changes in daily max temperature between the days on either side of nocturnal cooling. Plots produced by ranking x-axis parameter separately for each ensemble member and calculating average corresponding y-axis parameter for each 2% block of x-axis parameter. The middle line of each set of three lines shows the ensemble median relationship with upper and lower lines showing the 10%–90% ranked ensemble range in the y-axis parameter. (d) A schematic of the effects of changes in nocturnal cooling and daytime heating. that on average upsteps generally increase with reduced were examined during periods whenever daily maxi- nocturnal cooling, and also that doubling CO2 results in mum temperatures exceeded 30°C. A frequency distri- larger upsteps for nocturnal cooling values below bution of both upsteps and downsteps (negative daily ϳ14°C. For the high daily temperatures of interest here maximum temperature changes for consecutive days) is (typically ϳ25°C or more), simulated nocturnal cooling shown in Fig. 12 during these conditions. The effect of reduces by several degrees from present-day simulation doubling CO2 increases the probability of upsteps ex- values of ϳ15°C or more, implying (from Fig. 11c) ceeding 2°C. This suggests that during a sequence of larger upsteps and thus facilitating the increased occur- days above 30°C, extreme temperatures can be reached rence and severity of heat waves. A schematic of the quicker with increased CO2. Table 1 shows the effect of effects of changes in nocturnal cooling and daytime CO2 doubling on selected statistics associated with tem- heating is shown in Fig. 11d. perature changes between consecutive days during epi- To investigate these effects in more detail, upsteps sodes when the temperature consistently exceeds 30°C. 1SEPTEMBER 2006 C L A R K E T A L . 4433

heat flux from the surface in the 2003 European sum- mer heat wave by reducing the volume of air being heated by the surface. Such a mechanism may be at work here. However, we cannot yet investigate such possible mechanisms as suitable diagnostics (e.g., sub- daily radiation fluxes, surface fluxes, and cloud cover) are unavailable for this ensemble. Further study will be required. Attempts were also made to explain the role of simu- lated circulation changes on the intensity of heat waves. However, it appeared that many different circulation types can give rise to days greater than the 80th maxi- mum temperature percentile at any particular grid point. Also, the range of relevant circulation types var- ied significantly from region to region. As a result, at- tributing changes in heat events to circulation changes proved inconclusive, unlike the findings in Meehl and Tebaldi (2004) and Pal et al. (2004).

FIG. 12. Probability distribution of simulated changes in daily 6. Conclusions max temperature between consecutive days for all pairs of days when max temperature exceeds 30°C on both days, under present- Changes in extreme maximum temperature for bo- day (thin line) and doubled (thick line) CO2 conditions for the southwestern France grid point. Positive changes indicate pairs of real summer (JJA) due to a doubling of CO2 concen- days when the second day was warmer than the first. tration have been examined using a 53-member en- semble of global simulations from the HadAM3 climate model coupled to a slab ocean where each member The three values for each statistic represent the range differs in one uncertain parameter controlling a specific across the ensemble members. The number of upsteps aspect of the physical representation of the climate sys- as a percentage proportion of all steps increases by tem. Although not sampling all possible sources of un- 3.8% with upstep magnitude increasing by 0.3°C for the certainty, this ensemble and analysis is a significant step ensemble mean. While these figures may seem small, toward probabilistic projections of the future risk of hot the large 30% increase in the number of consecutive extreme temperature events. The significant ensemble upsteps could enhance the impact of upstep changes by spread that is seen does not, however, undermine the their increased collation, enhancing the likelihood of robustness of the result that hot temperature extremes extreme temperatures. in JJA will be more frequent, more severe, and of

There are potentially many mechanisms that could longer duration under increased CO2. be causing these changes in consecutive daily tempera- We analyzed changes in daily extremes in terms of tures and nocturnal cooling, for example, increased the change in the temperature associated with an event downwelling radiation either through greater atmo- corresponding to a given percentile of the simulated spheric concentrations of greenhouse gases (including doubled CO2 distribution, compared to the value of the water vapor) or increased cloudiness. Additionally, same percentile in the present-day simulation distribu- Black et al. (2004) suggested that a thinning of the noc- tion. The simulations show large spatial variability in turnal boundary layer may have increased the impact of the percentile that exhibits the largest change. Extreme

TABLE 1. Summary statistics of temperature changes with CO2 doubling for the grid box representing southwestern France during periods of at least 2 days when daily max temperature reaches at least 30°C every day. Upsteps and downsteps are defined as being pairs of days when the daily max temperature of the second day of each pair is greater or less than the first, respectively.

Percentile of ensemble distribution 10% Avg 90% Average increase in upstep magnitude (°C change) Ϫ0.2 0.3 0.7 Average number of upsteps as proportion of total number of steps (% change) Ϫ3.0 3.8 9.9 Average number of consecutive upsteps (% change) 1.7 30.2 63.2 4434 JOURNAL OF CLIMATE VOLUME 19 hot temperatures show the greatest increase in Europe, sured as the change in the temperature threshold asso- parts of North and South America, and eastern Asia. ciated with a 1-in-20-yr heat wave for events of a given In many parts of the world the results give changes in duration, the ensemble simulates substantial increases the 50th percentile similar to changes in the 99th per- with increases in many areas found to be relatively in- centile. This suggests that the response of extremes can dependent of heat wave duration. The duration of heat be understood, to first order, as a simple shift in the waves above a given threshold is found to increase sub- distribution of daily values as found by Barnett et al. stantially in many areas, and large increases in their (2006) although to a lesser extent here. However, in frequency, measured as the change in occurrence of some regions differences of several degrees were found events over a specific temperature threshold, are also between the 50th and 99th percentile changes, due to found. We analyze sequences of daily temperatures more complex responses involving significant changes during simulated heat waves, finding that increases in in the scale and asymmetry of the daily distributions. intensity and frequency in response to doubled CO2 are Substantial increases in intra-annual and interannual explained mainly by reductions in nocturnal cooling variability in daily maximum temperature have also during hot spells, rather than by increases in daytime been shown. Results for four selected grid points illus- heating. trate such changes, which include substantial changes in In summary, the ensemble results suggest that im- distribution shape with examples of the development of pacts from hotter and more frequent extreme tempera- bimodal characteristics not found in the present-day ture events could be severe in many parts of the world, simulations. implying a need for substantial adaptation measures Uncertainty in percentile changes was estimated should efforts to mitigate future greenhouse gas emis- from the ensemble spread of changes, which arises both sions fail. Further research is required to explore addi- from the perturbed model parameters and the effects of tional uncertainties arising from the effects of multiple natural variability. The range has generally been found parameter perturbations, structural model uncertain- to be up to approximately 50% of the magnitude of the ties, the transient response of extremes to anthropo- changes, however, this does not question the sign or genic forcing, and ocean circulation feedbacks. The ul- nature of the changes. This uncertainty does not take timate aim should be to provide plausible probabilistic into account uncertainty in structural elements of the predictions consistent with the best possible sampling model or uncertainty that might arise if multiple pa- of modeling uncertainties and also with suitable obser- rameters were perturbed. For some areas the greatest vational constraints. changes were found for more moderate percentiles. This was attributed to reductions in soil moisture lead- Acknowledgments. We thank Mat Collins and David ing to a shift from a wet to a dry regime on doubling Sexton for suggestions and discussions. This work was

CO2 at these percentiles. The more extreme percen- supported by the Department for Environment, Food tiles, by contrast, were already in a dry regime in the and Rural Affairs, under the Climate Prediction Pro- present-day simulation, so the change on doubling CO2 gramme Contract PECD 7/12/37. was smaller. Regional climate modeling studies have also found REFERENCES substantial changes in extreme temperatures with simi- lar sensitivity to soil moisture (Moberg and Jones 2004; Barnett, D. N., S. J. Brown, J. M. Murphy, D. M. H. Sexton, and M. J. Webb, 2006: Quantifying uncertainty in changes in ex- Räisänen et al. 2004) The Prediction of Regional Sce- treme event frequency in response to doubled CO2 using a narios and Uncertainties for Defining European Cli- large ensemble of GCM simulations. Climate Dyn., 26, mate Change Risks and Effects (PRUDENCE) project doi:10.1007/s00382-005-0097-1. (Räisänen et al. 2004) is an enhancement to this study Beniston, M., and H. F. Diaz, 2004: The 2003 heat wave as an that will help in the understanding of these local feed- example of summers in a greenhouse climate? Observations and climate model simulations for Basel, Switzerland. Global backs and their dependence on model formulation. Planet. Change, 44, 73–81. This effort will be extended as part of the ENSEMBLES Black, E., M. Blackburn, G. Harrison, B. Hoskins, and J. Meth- project, in which the perturbed parameter approach de- ven, 2004: Factors contributing to the summer 2003 European scribed in this paper will be applied to generate an heatwave. Weather, 59, 217–223. ensemble of regional climate model simulations suit- Brabson, B. B., D. H. Lister, P. D. Jones, and J. P. Palutikof, 2005: able for the study of changes in extremes at spatial Soil moisture and predicted spells of extreme temperatures in Britain. J. Geophys. Res., 110, D05104, doi:10.1029/ scales relevant to studies of climate impacts. 2004JD005156. We also investigate changes in the intensity, dura- Caesar, J., L. Alexander, and R. Vose, 2006: Large-scale changes tion, and frequency of heat waves. For intensity, mea- in observed daily maximum and minimum temperatures: 1SEPTEMBER 2006 C L A R K E T A L . 4435

Creation and analysis of a new gridded data set. J. Geophys. fall and temperature in Southeast Asia and the South Pacific: Res., 111, D05101, doi:10.1029/2005JD006280. 1961–1998. Int. J. Climatol., 21, 269–284. Changnon, S. A., K. E. Kunkel, and B. C. Reinke, 1996: Impacts Meehl, G. A., and C. Tebaldi, 2004: More intense, more frequent and responses to the 1995 heat wave: A call to action. Bull. and longer lasting heat waves in the 21st century. Science, 305, Amer. Meteor. Soc., 77, 1497–1506. 994–997. Christidis, N., P. A. Stott, S. Brown, G. C. Hegerl, and J. Caesar, Moberg, A., and P. D. Jones, 2004: Regional climate model simu- 2005: Detection of changes in temperature extremes during lations of daily maximum and minimum near-surface tem- the second half of the 20th century. Geophys. Res. Lett., 32, peratures across Europe compared with observed station L20716, doi:10.1029/2005GL023885. data 1961–1990. Climate Dyn., 23, 695–715. Durre, I., J. M. Wallace, and D. P. Lettenmaier, 2000: Depen- Murphy, J. M., D. M. H. Sexton, D. N. Barnett, G. S. Jones, M. J. dence of extreme daily maximum temperatures on anteced- Webb, M. Collins, and D. A. Stainforth, 2004: Quantification ent soil moisture in the contiguous United States during sum- of modelling uncertainties in a large ensemble of climate mer. J. Climate, 13, 2641–2651. change simulations. Nature, 430, 768–772. Easterling, D. R., J. L. Evans, P. Y. Groisman, T. R. Karl, K. E. Nakicenovic, N., and R. Swart, Eds., 2000: Special Report on Kunkel, and P. Ambenje, 2000a: Observed variability and Emission Scenarios. Cambridge University Press, 570 pp. trends in extreme climate events: A brief review. Bull. Amer. Ogi, M., K. Yamazaki, and Y. Tachibana, 2005: The summer Meteor. Soc., 81, 417–425. northern annular mode and abnormal summer weather in ——, G. A. Meehl, C. Parmesan, S. A. Changnon, T. R. Karl, and 2003. Geoophys. Res. Lett., 32, L04706, doi:10.1029/ L. O. Mearns, 2000b: Climate extremes: Observations, mod- 2004GL021528. eling and impacts. Science, 289, 2068–2074. Pal, J. S., F. Giorgi, and B. Xunqiang, 2004: Consistency of recent Ferro, C. A. T., A. Hannachi, and D. B. Stephenson, 2005: Simple European summer precipitation trends and extremes with nonparametric techniques for exploring changing probability future regional climate projections. Geophys. Res. Lett., 31, distributions of weather. J. Climate, 18, 4344–4354. L13202, doi:10.1029/2004GL019836. Fink, A. H., T. Brucher, A. Kruger, G. C. Leckebusch, J. G. Pinto, Pope, V. D., M. Gallani, P. R. Rowntree, and R. A. Stratton, 2000: and U. Ulbrich, 2004: The 2003 European summer heatwaves The impact of new physical parameterisations in the Hadley and drought—Synoptic diagnosis and impacts. Weather, 59, Centre climate model—HadAM3. Climate Dyn., 16, 123–146.

209–215. Räisänen, J., 2001: CO2-induced climate change in CMIP2 experi- Frich, P., L. V. Alexander, P. Della-Marta, B. Gleason, M. Hay- ments: Quantification of agreement and role of internal vari- lock, A. M. G. Klein Tank, and T. Peterson, 2002: Observed ability. J. Climate, 14, 2088–2104. coherent changes in climatic extremes during the second half ——, and Coauthors, 2004: European climate in the late twenty- of the twentieth century. Climate Res., 19, 193–212. first century: Regional simulations with two driving global García-Herrera, R., J. Diaz, R. M. Trigo, and E. Hernandez, 2005: models and two forcing scenarios. Climate Dyn., 22, 13–31. Extreme summer temperatures in Iberia: Health impacts and Rowell, D. P., and R. G. Jones, 2006: The causes and uncertainty associated synoptic conditions. Ann. Geophys., 23, 239–251. of future summer drying over Europe. Climate Dyn., 27, Giorgi, F., and R. Francisco, 2000: Evaluating uncertainties in the doi:10.1007/s00382-006-0125-9. prediction of regional climate change. Geophys. Res. Lett., 27, Schar, C., P. Luigi Vidale, D. Luthi, C. Frei, C. Haberli, M. A. 1295–1298. Liniger, and C. Appenzeller, 2004: The role of increasing Hayhoe, K., and Coauthors, 2004: Emissions pathways, climate temperature variability in European summer heatwaves. Na- change, and impacts on California. Proc. Natl. Acad. Sci. ture, 427, 332–336. USA, 101, 12 422–12 427. Stainforth, D. A., and Coauthors, 2005: Uncertainty in predictions Horton, E. B., C. K. Folland, and D. E. Parker, 2001: The chang- of the climate response to rising levels of greenhouse gases. ing incidence of extremes in worldwide and central England Nature, 433, 403–406. temperatures to the end of the 20th century. Climate Change, Stott, P. A., D. A. Stone, and M. R. Allen, 2004: Human contri- 50, 267–295. bution to the European heatwave of 2003. Nature, 432, 610– INVS, 2003: Impact sanitaire de la vague de chaleur en France 614. survenue en août 2003. Rapport de l’Institut de veille sani- Tebaldi, C., R. W. Smith, D. Nychka, and L. O. Mearns, 2004: taire (National Institute of Public Health Surveillance), 75 Regional probabilities of precipitation change: A Bayesian pp. analysis of multimodel simulations. Geophys. Res. Lett., 31, Kharin, V. V., and F. W. Zwiers, 2000: Changes in the extremes in L24213, doi:10.1029/2004GL021276. an ensemble of transient climate simulations with a coupled ——, ——, ——, and ——, 2005: Quantifying uncertainty in pro- atmosphere–ocean GCM. J. Climate, 13, 3760–3788. jections of regional climate change: A Bayesian approach to Koster, R. D., and Coauthors, 2004: Regions of strong coupling the analysis of multimodel ensembles. J. Climate, 18, 1524– between soil moisture and precipitation. Science, 305, 1138– 1540. 1140. Trigo, R. M., R. García-Herrera, J. Diaz, I. F. Trigo, and M. A. Kunkel, K. E., S. A. Changnon, B. C. Reinke, and R. W. Arritt, Valente, 2005: How exceptional was the early August 2003 1996: The July 1995 heat wave in the Midwest: A climatic heatwave in France? Geophys. Res. Lett., 32, L10701, perspective and critical weather factors. Bull. Amer. Meteor. doi:10.1029/2005GL022410. Soc., 77, 1507–1518. WHO, 2004: Heat waves: Risks and responses. Health and Global Luterbacher, J., D. Dietrich, E. Xoplaki, M. Grosjean, and H. Environmental Change, World Health Organization, Series Wanner, 2004: European seasonal and annual temperature No. 2, 125 pp. variability, trends and extremes since 1500. Science, 303, Zwiers, F. W., and V. V. Kharin, 1998: Changes in the extremes of

1499–1503. the climate simulated by CCC GCM2 under CO2 doubling. J. Manton, M. J., and Coauthors, 2001: Trends in extreme daily rain- Climate, 11, 2200–2222.