Contents lists available atScienceDirect
Weather and Climate Extremes
journal homepage:www.elsevier.com/locate/waceAttribution analyses of temperature extremes using a set of 16 indices
Nikolaos Christidis
⁎, Peter A. Stott
Met Office Hadley Centre, FitzRoy Road, Exeter EX1 3PB, UK
A R T I C L E I N F O
Keywords:Detection and attribution Temperature extremes Anthropogenic forcings
A B S T R A C T
Detection and attribution studies have demonstrated that anthropogenic forcings have been driving significant changes in temperature extremes since the middle of the 20th century. Moreover, new methodologies have been developed for the attribution of extreme events that assess how human influence may have changed their characteristics. Here we combine formal statistical analyses based on optimalfingerprinting to attribute observed long term changes in temperature extremes with an ensemble-based approach for event attribution. Our analyses are applied to 16 indices constructed with daily temperature data that focus on different characteristics of extremes and together build up a more complete representation of historical changes in warm and cold extremes than previous studies. For each index we compute an annual value for all years of the post-1960 period using data from observations and experiments with a coupled Earth System model for the analysis of multi-decadal changes and a high-resolution atmospheric model for event attribution. The models indicate that anthropogenic forcings have influenced almost all indices in recent decades and led to more prominent changes in the frequency of extremes. The optimalfingerprinting analyses show that for most indices the anthropogenic signal is detectable in changes during 1961–2010 both in Europe and on a quasi-global scale. The weaker natural effect, resulting mainly from volcanic eruptions, is in most cases not detectable, with the exception of large scale changes in indices linked to the frequency of cold night-time extremes. Our event analyses estimate how anthropogenic forcings alter the chances of getting new record index values in Europe andfind that such extremes would be markedly rare if human influence were not accounted for, whereas in the current climate their return times range from a few years to a few decades.
1. Introduction
Accumulating evidence from detection and attribution studies helped establish that anthropogenic forcings have significantly changed characteristics of daily temperature extremes in recent decades (Bindoffet al., 2013). Detailed assessments investigating temperature extremes in the context of anthropogenic climate change (Seneviratne et al., 2012), including studies of individual events (NAS, 2016), elicit great public and media interest, primarily because of the socio-economic impacts associated with extremes. For example, recent catastrophic heatwaves in Europe (Christidis et al., 2015a; Dole et al., 2011) exposed the vulnerability of communities, while an increased incidence of heat extremes worldwide would take its toll on human health (McMichael, 2013; Wolf and McGregor, 2013) increase fire risk (Yoon et al., 2015), exert stress on crops (Teixeira et al., 2013), exacerbate air pollution (Lelieveld et al., 2014) etc. Assessing the contribution of causal factors like anthropogenic forcings to observed changes in extremes or extreme events can be a valuable tool for decision making, e.g. by aiding effective adaptation planning, and an
integral part of the developing climate services (Hewitt et al., 2012). Attribution of extremes and extreme events is a growing research area marked by major advances over the last 10–15 years. The earlier work focussed on long term trends in simple indices of daily tempera-ture extremes like the coldest and warmest day and night of the year (Hegerl et al., 2004). Christidis et al. (2005) applied an optimal fingerprinting methodology (Allen and Stott, 2003) to formally estab-lish for the first time significant anthropogenic warming in daily temperature extremes since the 1950s. Subsequent work also consid-ered more sophisticated indices from the extreme value theory that better represent the tails of the distribution and confirmed that the anthropogenicfingerprint is detectable in observed changes of both hot and cold extremes (Christidis et al., 2011; Zwiers et al., 2011). Apart from global changes,fingerprinting analyses demonstrated detectable anthropogenic warming in extremely warm and cold nights in several continental and sub-continental regions (Min et al., 2013; Wen et al., 2013). Detectable changes in the frequency of daily temperature extremes due to human climatic influences have also been found on global and regional scales (Morak et al., 2013).
http://dx.doi.org/10.1016/j.wace.2016.10.003
Received 18 August 2016; Received in revised form 21 October 2016; Accepted 25 October 2016
⁎Corresponding author.
E-mail address:nikos.christidis@metoffice.gov.uk(N. Christidis).
Weather and Climate Extremes xx (xxxx) xxxx–xxxx
2212-0947/ © 2016 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by/4.0/). Available online xxxx
While long term changes in extremes have been mainly investigated by fingerprinting analyses, the new science of event attribution has developed novel methodologies to quantify how anthropogenic forcings may change characteristics of specific extreme events (Stott et al., 2016). A series of annual reports published in the Bulletin of the American Meteorological Society (BAMS) since 2012 explaining ex-treme events of the previous year from a climate perspective showed a substantial influence on the frequency and intensity of heat events by human-caused climate change (Herring et al., 2015). The large volume of attribution studies of hot and cold high-impact events around the world that have been published both in BAMS and elsewhere in the literature underpin the conclusion of the Intergovernmental Panel on Climate Change (IPCC) that“it is very likely that human influence has contributed to the observed changes in the frequency and intensity of daily temperature extremes on the global scale since the mid-20th century”(Bindoffet al., 2013).
The new attribution study presented here has a threefold aim: a) Provide a more comprehensive description of temperature extremes
and examine how anthropogenic climate change might have influenced their different characteristics. This is achieved by employing the complete suite of the original 16 temperature extreme indices introduced by the Expert Team on Climate Change Detection, Monitoring and Indices (ETCCDI; http:// etccdi.pacificclimate.org/). While some of these indices have been popular in attribution research, there is no study yet, to our knowledge, which has considered the entire range and about half of the indices are investigated here for thefirst time.
b) Produce a synthesis attribution assessment that examines the effect of human influence on both trends in extremes and extreme events. The popularfingerprinting methodology is employed to assess long term changes in characteristics of temperature extremes. The suite of indices used in the study leads to new definitions of extreme events. Most event analyses consider the mean temperature over a period and define a threshold above (or below) which an extremely hot (or cold) event occurs. For example, studies of the European heatwave of 2003 defined heatwaves using the summer mean temperature over a large European area (Stott et al., 2004; Christidis et al., 2015a). Here we look at the annual mean index value instead, which also provides a useful event definition. For example, a large number of tropical nights or frost days (two of the ETCCDI indices) in a region would define events that are more directly linked to certain health or agricultural impacts. The most suitable index for an attribution analysis would depend on the aspect of the event that the study concentrates on.
c) Utilise two of Hadley Centre's state-of-the art climate models in new analyses of temperature extremes. A major upgrade of the atmospheric model that forms the basis of Hadley Centre's event attribution system (Christidis et al., 2013) was recently undertaken by the EUCLEIA project (http://eucleia.eu). This resulted in a system that features the highest resolution global model used in event attribution studies.Shiogama et al. (2016)also employed a high-resolution model, which, however, has fewer vertical levels, while other high resolution analyses rely on regional models (e.g. Massey et al., 2015;Takayabu et al., 2015). A more complex Earth System model was utilised in the analyses of long term changes. The remainder of the paper comprises a description of the data and methods used in the study (Section 2), a presentation of results (Section 3) and a discussion of the mainfindings and future develop-ments (Section 4).
2. Data and methodology 2.1. The HadEX2 dataset
The observations used in our study come from HadEX2, a global gridded dataset of 27 ETCCDI temperature and precipitation climate extreme indices (Donat et al., 2013). This is an extension of the original HadEX dataset (Alexander, 2006), which included considerably fewer stations and a smaller spatial coverage. Here we examine only temperature extremes and use the original set of 16 indices shown in Table 1. The indices describe different aspects of temperature ex-tremes. While some of the indices that measure the intensity (TXx, TNx, TXn, TNn) and frequency (percentile based indices) of extremes have been popular in attribution studies, here we also consider those indices that employ critical temperature thresholds useful for impact studies. We also provide new analyses of changes in the growing season length and diurnal temperature range. HadEX2 and its predecessor have demonstrated significant changes since last century with extremes of the minimum daily temperature (TN) shown to have warmed more than those of the maximum temperature (TX). This asymmetry, also noted elsewhere in the literature (Morak et al., 2013), means that in a warming climate the temperature distribution does not simply shift as a whole to a warmer regime, but also changes in shape.
Although the HadEX2 data extend over the period 1901–2010, we only consider post-1960 years in this study. This is because some of the model experiments used in the analysis do not include earlier years, but also because the observational coverage was poorer in the early part of the dataset. We examine both global changes in extreme characteristics spanning the length of half a century (1960–2010), as well as regional changes in trends and in the present-day likelihood of extremes due to anthropogenic drivers. For illustration purposes we concentrate on a region that covers the European continent (30W-50E, 30-80N), which is of particular interest to the EUCLEIA project, though of course the same methodologies can also be applied to other regions.
European index timeseries are shown inFig. 1and the linear trends over the analysis period (1960–2010) averaged over the entire observational area and over Europe are listed inTable 2. The global trend patterns are illustrated inFig. 2. Testing the hypothesis that the
Table 1
The 16 ETCCDI indices of temperature extremes used in this study.
Abbreviation Index description
FD Number of frost days (Tmin < 0 °C) SD Number of summer days (Tmax > 25 °C) ID Number of icing days (Tmax < 0 °C) TN Number of tropical nights (Tmin > 20 °C)
GSL Growing season length (Number of days between first span from the beginning of winter of at least 6 days with Tmean > 5 °C andfirst span of at least 6 days after thefirst month of summer with Tmean < 5 °C)
TXx Maximum Tmax (warmest day) TNx Maximum Tmin (warmest night) TXn Minimum Tmax (coldest day) TNn Minimum Tmin (coldest night)
TN10p Percentage of days when Tmin < 10th climatological percentile (base period 1961–99)
TX10p Percentage of days when Tmax < 10th climatological percentile (base period 1961–99)
TN90p Percentage of days when Tmin > 90th climatological percentile (base period 1961–99)
TX90p Percentage of days when Tmax > 90th climatological percentile (base period 1961–99)
WSDI Warm spell duration (Annual count of days with at least 6 consecutive days when Tmax > 90th percentile) CSDI Cold spell duration (Annual count of days with at least 6
consecutive days when Tmin < 90th percentile)
DTR Daily temperature range (Mean difference between Tmax and Tmin)
Fig. 1.Timeseries of annual anomaly values (relative to 1961–1990) averaged over Europe for the 16 ETCCDI indices. Timeseries constructed with HadEX2 are shown in black. The red and blue lines represent the ensemble mean of the ALL and NAT simulations with the HadGEM3-A model and the coloured areas mark the ensemble spread. The modelled data were masked to include the same coverage as the observations. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)
Table 2
Index trends (index/year) during 1960–2010 computed with HadEX2 and the HadGEM3-A (atmospheric model) and HadGEM2-ES (coupled model) experiments. Numbers in bold indicate trends significantly different than zero (tested at the 5% significance level using a least square fit). Modelled trends correspond to the ensemble mean. The modelled fields were re-gridded onto the HadEX2 grid and masked to include the same coverage as the observations.
Globally averaged trends European trends
Index OBS ALL NAT OBS ALL NAT
Atmos Coupled Atmos Coupled Atmos Coupled Atmos Coupled
FD SD ID TN GSL TXx TNx TXn TNn TN10p TX10p TN90p TX90p WSDI CSDI DTR −0.162 0.046 −0.017 0.181 0.179 0.025 0.028 0.026 0.037 −0.132 −0.081 0.224 0.137 0.253 −0.056 −0.007 −0.150 0.215 −0.106 0.124 0.156 0.036 0.026 0.023 0.030 −0.100 −0.086 0.138 0.132 0.166 −0.047 0.001 −0.258 0.274 0.177 0.181 0.243 0.041 0.031 0.042 0.043 −0.124 −0.111 0.186 0.184 0.223 −0.057 0.002 0.073 −0.033 0.047 −0.020 −0.073 −0.006 −0.006 −0.006 −0.005 0.029 0.027 −0.036 −0.028 −0.024 0.024 −5×10−4 −0.071 0.048 −0.050 0.013 0.061 0.005 0.007 0.013 0.015 −0.036 −0.034 0.032 0.026 0.023 −0.022 2×10−4 −0.216 0.293 −0.145 0.099 0.223 0.039 0.032 0.031 0.042 −0.102 −0.099 0.204 0.168 0.282 −0.079 −10−4 −0.157 0.263 −0.121 0.062 0.188 0.046 0.029 0.027 0.030 −0.105 −0.106 0.162 0.162 0.231 −0.064 0.005 −0.319 0.333 −0.216 0.120 0.339 0.050 0.036 0.049 0.053 −0.139 −0.126 0.220 0.211 0.299 −0.091 0.003 0.100 −0.005 0.060 −0.006 −0.090 −5×10−4 −0.004 −0.008 −0.008 0.033 0.024 −0.041 −0.023 −0.019 0.027 0.001 −0.104 0.066 −0.064 0.028 0.111 0.007 0.008 0.027 0.029 −0.051 −0.045 0.044 0.037 0.049 −0.045 −6×10−4
least squares fit has a zero trend (2-sidedt-test), we found that the observed trends are significant at the 5% level for the majority of the indices. HadEX2 shows a warming of both warm and cold extremes accompanied by a corresponding increase and decrease in their frequencies. Indices of cold extremes display higher variability, but their trend may be larger (e.g. TXn and TNn vs. TXx and TNx inFig. 2). Note that despite the larger warming in TXn than TNx shown inFig. 2, the estimated TXn global trend inTable 1is somewhat smaller, because of the variable observational coverage allowed in this computation. Trends in frequency (as indicated by TN10p, TX10p, TN90p, TX90p) are found to be larger than trends in intensity (TXx, TNx, TXn, TNn). This can be confirmed using a simple signal-to-noise estimate com-puted as the ratio of the absolute 2001–2010 mean index anomaly and the standard deviation of the de-trended index during 1900–1950. The resulting estimates are found to be 1.6–3.7 times greater for the frequency indices than their intensity counterparts. As also shown in previous work (Christidis et al., 2007), the growing season has become longer in recent decades with the change over Europe being larger than the global change. The differential change in TN and TX has also led to a decrease in the DTR globally, though there is hardly any change in Europe. Similar global decreases in the DTR have been shown in other studies too (Zhou et al., 2010; Lewis and Karoly, 2013).
Attribution studies that assess the effect of human influence on various aspects of the climate essentially compare the real world, influenced by all (both natural and anthropogenic) external forcings
(ALL), and a hypothetical natural world influenced by natural forcings only (NAT), i.e. changes in the solar output and volcanic eruptions. Ensembles of model simulations representing the ALL and NAT climates are typically employed and, in studies of extremes, are used to either estimate the change in the likelihood of extreme events, or construct thefingerprint of external forcings associated with long term changes, as discussed in the following two sections.
2.2. Changes in the odds of extreme events
Attribution systems based on large ensembles of ALL and NAT simulations generated by atmospheric models are a very popular research tool in studies of extreme events (Stott et al., 2016) and were introduced byPall et al. (2011)in their analysis of the UKfloods in autumn 2000. Atmospheric model simulations condition the attribu-tionfindings on the state of the ocean at the time of the event under consideration. Here, however, we utilise long, multi-decadal simula-tions, and use the last 10 years to represent the present-day climate. Consequently, our results are not dependent on a specific pattern of sea surface temperature (SST) anomalies, or mode of oceanic variability present at the time an extreme event occurs. Instead, attribution assessments are made with reference to a wider (though not exhaus-tive) range of possible conditions. The two ensembles enable the construction of the distribution of the relevant climatic variable (e.g. temperature for a heatwave event) for the actual climate (ALL) and the
Fig. 2.Trends (index/year) during 1960–2010 computed with HadEX2 for the 16 ETCCDI indices. The red box on the upper left panel (FD) marks the region used in the European analyses. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)
climate without the effect of human influence (NAT). We then use these distributions to obtain estimates of the event's likelihood, or its inverse (return period). The change in the likelihood provides a measure of the effect of anthropogenic forcings. In this study, instead of temperature distributions, we construct distributions for each of the 16 extreme indices. We consider a range of extreme index values and estimate the return time of exceeding them (or going below them for indices with a negative trend) and thus construct return time diagrams that enable event attribution in real time. This means that if a new event occurs (e.g. record number of frost days in a year, or warm spell duration), then we can refer back to the diagram to measure the change in the likelihood that corresponds to the observed threshold. Similar fast-track attribution approaches have been developed byChristidis et al. (2015b)andLewis et al. (2014).
The Hadley Centre event attribution system (Christidis et al., 2013) was built on the HadGEM3-A model (Hewitt et al., 2011) and has been used to study a range of different types of extremes. After the major upgrade undertaken as part of the EUCLEIA project, the model now features the highest resolution used in attribution systems (N216, or about 60 km horizontal resolution and 85 vertical levels). In this study we use two ensembles (ALL and NAT) that comprise 15 simulations each and cover the historical period 1960–2013. The Hadley Centre system is envisaged to be eventually integrated into an operational framework that will provide attribution assessments on a seasonal basis, with simulations spun offfrom the long runs used in this work and extended at the end of every season thereafter. The ALL
experi-ment employs observed SSTs and sea-ice data (HadISST;Rayner et al., 2003) as boundary conditions. The boundary conditions for the NAT experiment are constructed by subtracting an estimate of the anthro-pogenic SST change from the observations, which was obtained using multi-model ensembles of ALL and NAT simulations conducted with several coupled models (Stone and Pall, 2016). The sea-ice amount prescribed in NAT simulations is also adjusted accordingly using simple empirical relationships (Christidis et al., 2013; Pall et al., 2011). The ensembles are generated using random parameter pertur-bations, as well as a stochastic kinetic energy scheme that accounts for sub-grid scale energy sources. Details about technical features of the system and forcings used in the simulations can be found inChristidis et al. (2013).
Global and European mean index trends during 1960–2010 estimated with the ensemble mean of the atmospheric model are listed inTable 2and the modelled index timeseries for Europe are shown together with the HadEX2 timeseries inFig. 1. The ALL experiment yields statistically significant trends that are consistent in size and magnitude with the HadEX2 trends, with the exception of the global DTR index, for which the model gives a positive non-significant trend (Table 2). The observational timeseries marked by the black lines in Fig. 1lie within the range corresponding to the ALL experiment, while the ALL ensemble mean agrees well with HadEX2, but of course displays less variability because of the averaging. The NAT experiment gives smaller, near-zero trends (Table 2). In certain cases (e.g. TN, TN90p) the observed values of the European indices in recent years
Fig. 3.Trends (index/year) over land estimated with the ensemble mean of the 15 ALL simulations with HadGEM3-A for the 16 ETCCDI indices.
(Fig. 1) have risen above the range of NAT values (but still remain within the ALL range), suggesting observed changes cannot be explained apart from anthropogenic forcings. Trend patterns over land corresponding to the ALL and NAT experiments are shown inFigs. 3 and 4respectively. The ALL patterns agree well with the HadEX2 ones (Fig. 2) and indicate a warming of the climate system. The only exception is again DTR, for which the observational patterns show a stronger and more widespread decrease, whereas the ALL show a weaker and mixed signal with extended areas, including areas with little observational coverage (parts of Africa and South America). The NAT patterns (Fig. 4) show smaller trends, in most cases consistent with a weak cooling, presumably due to the overall effect of the major volcanic eruptions of Agung (1963), El Chichón (1982) and Pinatubo (1991).
2.3. Optimalfingerprinting
Optimalfingerprinting, originally introduced in climate science by Hasselmann (1993)and further developed byAllen and Stott (2003), is a popular methodology in detection and attribution studies that helped demonstrate the prominent role of human influence in global and regional temperature changes (Bindoffet al., 2013). As applied to our study, the method uses a total least squares regression algorithm to decompose an observed change into components of the modelled forced climate response and unforced variability. The model used for ourfingerprinting analyses is the Hadley Centre's Earth System model
HadGEM2-ES (Jones et al., 2011). Similarly to the analyses of events discussed inSection 2.1, model experiments with and without the effect of human influence (ALL and NAT) were carried out. Fingerprinting analyses of the recent warming of the surface (Jones et al., 2013) and free atmosphere (Lott et al., 2013) with HadGEM2-ES gave evidence of a major anthropogenic contribution, highlighted in the last IPCC report. For each ETCCDI index, we carry out two separate analyses to examine changes on global scale and in Europe. In our global analyses, the change is represented by 5-year mean index anomalies (relative to 1961–1990) in ten successive time segments: 1961–1965, 1966–1970, …, 2006–2010. The index pattern for each segment is spatially smoothed using spherical harmonics at T8 truncation and, for the modelled change, is masked to include only grid-points where observations are available. Analyses of changes in Europe include no spatial information and simply employ the 5-year mean anomalies averaged over the region, i.e. vectors of 10 index mean values over the analysis period.
We perform a two-way regression that partitions the observed change between the modelled response to anthropogenic forcings (ANT), natural forcings (NAT) and internal climate variability. We organise the observations into a vectoryand similarly the ALL and NAT responses, orfingerprints, into vectorsxALLandxNAT. Assuming
that the ANTfingerprint,xANT, can be approximated by the difference
xALL-xNATwe then have the expression:
Fig. 4.Trends (index/year) over land estimated with the ensemble mean of the 15 NAT simulations with HadGEM3-A for the 16 ETCCDI indices.
β β β β y x u x u u x x u x u u = ( − ) + ( − ) + = ( + − ) + ( − ) +
ALL NAT ANT NAT
NAT
1 1 2 2 0 1 1
2 2 0 (1)
where the noise term u0 accounts for the effect of internal climate variability. The fingerprints are typically derived from the ensemble mean of model simulations and are also affected by unforced variability (noise termsu1andu2) to an extent that depends on the ensemble size. Here we use the ensemble mean of four simulations for each experi-ment (ALL and NAT) to construct the fingerprints, as well as 1013 years from a control simulation of the pre-industrial climate to estimate the effect of internal variability. We extract 39 overlapping 50-year long segments from the control simulation, process and organise them into vectors similar to the observations and model fingerprints, and use them to construct the variance-covariance matrix of the noise terms in Eq. (1). The analysis aims to estimate the scaling factors associated with the ANT and NATfingerprints:βANT=β1and
βNAT=β1+β2, from which signal detectability is inferred. Scaling factors that do not encompass zero indicate a detectable signal. Moreover, scaling factors consistent with unity that have a small uncertainty range suggest good agreement between the model and the observations. The uncertainty in the scaling factors is estimated using the segments extracted from the control simulation. Fingerprinting analyses are restricted to a transformed space defined by the leading Empirical Orthogonal Functions (EOFs) of the internal variability. Here we keep
thefirst 15 EOFs for the global analyses and 7 EOFs for the European analyses, explaining about 85% and 95% of the observed variance respectively. Details on the implementation of the methodology can be found inAllen and Stott (2003).
The 1960–2010 trends averaged globally and over Europe com-puted with the ensemble mean of the ALL and NAT HadGEM2-ES experiments are shown in Table 1. The ALL trends are generally consistent in sign and magnitude with the observations. The coupled model is also found to produce greater ALL trends (both positive and negative) than HadGEM3-A. The NAT trends are again found to be smaller (near-zero), but of a more mixed sign than the NAT trends from the atmospheric model. European timeseries similar to the ones inFig. 1constructed with HadGEM2-ES data (not shown here) are found to be similar to HadGEM3-A for the ALL experiment (ensemble mean correlation coefficient of 0.7 averaged across all the indices) and also display no long term trend for the NAT experiment (correlation coefficient of 0.01). Global trend patterns corresponding to the HadGEM2-ES ensemble mean have an average correlation coefficient of 0.6 with the HadGEM3-A patterns for the ALL experiment and −0.07 for the NAT. The low correlations between the NAT patterns indicate that they are largely manifestations of unforced variability rather than forced changes. The ALL experiments from both models are found to overall reproduce well the observed forced response and trends in recent decades. A more detailed model evaluation is discussed
Fig. 5.Normalised distributions of the annual mean anomaly in 1960–2010 averaged over Europe for the 16 ETCCDI indices. The distributions are estimated with data from HadEX2 (histograms) and from ensembles of historical forcing simulations (ALL) with HadGEM3-A (red lines) and HadGEM2-ES (green lines). The modelled data were masked to include the same coverage as the observations. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)
next.
2.4. Model evaluation
Attribution assessments largely rely on model simulations, espe-cially in event analyses where no observational constraints are applied. It is therefore essential to evaluate the models against observations to increase confidence in the robustness of the results (Stott et al., 2016). Observations are used in event studies to set critical thresholds and in fingerprinting analyses to provide constraints applied to the modelled response to external forcings. In addition, we have already shown that index trends estimated with the two models used in our study are consistent with the observations. Two common evaluation assessments are also carried out here to examine how well the models represent the climatological distribution of the 16 indices and the observed varia-bility over different timescales.Fig. 5illustrates the probability density functions (PDFs) of the index anomalies in Europe over the period 1960–2010. Although 51 years of index values cannot adequately describe details of the distributions such as their tails, the histograms constructed with the observations (Fig. 5) give a general representation of their shape and spread. The models provide more data because of the multiple simulations in each ensemble (e.g. from the 15-member ensembles of HadGEM3-A, we get 15×51 years=765 index values per experiment). The coupled and the atmospheric models yield similar PDFs for most indices (red and green lines inFig. 5), though there are
also some discrepancies, most notably the higher variability indicated by HadGEM2-ES for the TN index, which is also more consistent with the observations. We performed two-sided Kolmogorov-Smirnov tests to assess whether the observations are significantly different from the data of each model simulation and, for both models and all indices, found no significant differences (5% significance level). Finally, we performed power spectra analyses (Gillett et al., 2000; Christidis et al., 2015b) to examine whether the models provide reasonable variability estimates. We compute power spectra from the European index time-series during 1960–2010 using both the observations and each simulation of the ALL experiment. The results are shown inFig. 6. The observations are found to generally lie within the range of the simulated spectra, though there are also a few exceptions. For example, the models appear to produce more variability for the TN10p index at interdecadal timescales, which would result in a more stringent testing of signal detection. Another interesting index is TN, for which HadGEM2-ES seems to overestimate variability and HadGEM3-A to underestimate it. Despite these discrepancies, the level of agreement between observed and modelled variability is certainly sufficient for our attribution analyses.
Fig. 6.Power spectra for the annual mean anomaly in Europe during 1960–2010 for the 16 ETCCDI indices. Spectra based on observed index timeseries are plotted in black, whereas spectra based on simulated timeseries from the ALL ensembles generated with HadGEM3-A and HadGEM2-ES are plotted in orange and green respectively. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)
3. Results
3.1. Long-term changes in global and european extremes
We investigate the effect of external climatic forcings on spatio-temporal changes in extreme indices on a global scale and spatio-temporal changes in Europe during 1961–2010 using optimalfingerprinting, as discussed inSection 2.3. Grid point index anomalies are computed with data from the HadGEM2-ES experiments and are used to construct the modelfingerprints. Two-way regression analyses carried out for each index provide the scaling factors for the anthropogenic and natural fingerprints shown inFig. 7. The anthropogenic signal is detected in changes of 11 indices on a global scale and 10 indices in Europe. Changes in the warmest and coldest days and nights of the year due to human influence are detected on a global scale, but changes in the cold extremes are not detected in Europe, where variability is larger (see TXn and TNn timeseries in Fig. 1). The use of more sophisticated indices derived from the extreme value theory may be more suitable for the detection of such regional changes (Zwiers et al., 2011). The anthropogenic signal is more prominent in changes of the frequency of daily extremes and is detected in all four percentile indices, both globally and in Europe.Morak et al. (2013)found that changes in the frequency of extremes may also be detectable when examined in different seasons. The effect of human influence cannot be detected for the cold indices FD and ID. This could be because of high variability, but also due to the fact that the indices are defined relative to a temperature threshold (0 °C) that would not be relevant in warmer regions. An earlier optimal fingerprinting analysis (Christidis et al., 2007) demonstrated a detectable lengthening in the growing season globally due to anthropogenic forcings, a result we confirm here with a different model. The warm spell duration (WSDI) has markedly increased in Europe over the last 20 years (Fig. 1) and the contribution of external forcings to the observed change can be detected, though this is not found to be the case in the global analysis. Similarly, the decrease in cold spell duration is only detectable in the regional analysis. The weaker NAT signal cannot be distinguished from internal climate variability for the majority of the indices. On a global scale, the NAT fingerprint is only detectable for TNn, TN10p, and CSDI, though the uncertainty in the TNn result is high. Interestingly, both TN10p and
CSDI relate to the frequency of night-time cold extremes which may spike after major volcanic eruptions, as found in index timeseries constructed with HadEX2 and NAT over the total observational area (not shown here). This increase in the number cold nights and cold night spells after an eruption could be associated with decreases in cloudiness, at least over land regions where observations are available, though more research is needed to establish any such effect. Over the smaller European region the NAT scaling factors have generally large uncertainties and robust detection of the NAT signal is not possible, with the exception of the WSDI index, for which both the observations and the model show a decrease following the Pinatubo eruption (Fig. 1). Even in cases where the NAT effect can be detected, it is still much weaker than the response to anthropogenic forcings which therefore constitute the main driver of observed changes. Overall, our analyses provide a strong indication that human influence has sig-nificantly influenced characteristics of temperature extremes in recent decades and the anthropogenic signal may rise above internal varia-bility not only quasi-globally, but also on continental scales.
3.2. The changing likelihood of european extreme events
We next compute extreme index anomalies from the HadGEM3-A temperature data for each simulated year and use the most recent decade (2004–2013) to construct index distributions for the near present-day climate with and without the anthropogenic effect. We then define events on the basis of preselected index values and compute the probability of either exceeding these thresholds (for indices with a positive trend), or going below them (for indices with a negative trend). As in previous work, we employ the Generalised Pareto Distribution to estimate probabilities of rare events. The analysis presented here, as already mentioned, focuses on the European region, but the same approach can also be applied to other areas. Return times of events are estimated from the inverse of the probability estimates. As in other event attribution studies, the uncertainty in the computed probabilities is derived using a Monte Carlo bootstrap procedure (Christidis et al., 2013), whereby the index values are randomly resampled (in this work 1000 times), new probability estimates are obtained and the 5–95% range is quantified using simple order statistics.Fig. 8illustrates the resulting return time plots for each index, together with the record
Fig. 7.Best estimates of the scaling factors and their 5–95% uncertainty range from global (top panel) and European (bottom panel) optimalfingerprinting analyses for the 16 ETCCDI indices. Two-way regressions are carried out that separate the response to external forcings between its anthropogenic (orange scaling factors) and natural (green scaling factors) components. The horizontal lines mark the zero and unity values, which help determine whether a signal is detectable and in good agreement with the observed change. (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)
values in the HadEX2 dataset and the year they correspond to. We expect that some of the records used here may have been broken in post-2010 years, considering, for example, that 2014 was a record hot year in Europe (Uhe et al., 2016). We note that for 12 out of the 16 indices record years fall after 1990, when accelerating changes in the anthropogenically forced climate become more evident for the majority of indices (Fig. 1). Exceptions include DTR, for which there is no significant trend in recent decades, and ID, which has a relatively high
variability as an index of cold extremes. The coldest day and night of the year (TXn and TNn) also had their minimum in 1974, apparently a manifestation of a cold winter in Europe, but it should also be noted that for these two indices the anthropogenic effect appears to be weaker than their warm counterparts (smaller separation between the ALL and NAT timeseries shown inFig. 1).
The most striking result of our event analysis is the wide separation between the ALL and NAT return times illustrated inFig. 8. With the exception of DTR, there is no overlap between the expected range of indices for events with multi-decadal return times. Events expected to occur every few decades under the influence of anthropogenic forcings would have at least an order of magnitude longer return times in the natural climate. The NAT probabilities are indeed so small that cannot be reliably estimated with the limited sample of indices we get from our 15 simulations and much larger ensembles would be required for that purpose.Table 3shows the best estimates of the return times from the ALL experiment associated with HadEX2 record events together with their uncertainties. Apart from the TNx index which has return time of about 2000 years, record events are suggested to occur every few decades or years for the rest of the indices. Return times of percentile based indices (TN10p, TX10p, TN90p, TX90p) are at least an order of magnitude smaller than those of annual day- and night-time extremes
Fig. 8.Present-day return times of events in Europe estimated over a range of anomalies for the 16 ETCCDI indices. The events are defined as the likelihood of going above (for indices with positive trends in recent decades), or below (for indices with negative trends) an anomaly threshold. Results from the ALL experiment (actual climate) are shown in red and from the NAT experiment (climate without the effect of human influence) in blue. Three lines are plotted for each index and each experiment representing the best estimate (middle line) and the 5–95% uncertainty range. The grey horizontal lines correspond to the record index anomalies estimated with HaDEX2 and the years they occurred are also marked the panels.
Table 3
Return times of record index anomalies in the near present climate (2004–2013).
Index Return time in years best estimate (5–95% uncertainty range)
Index Return time in years best estimate (5–95% uncertainty range) FD 22.0 (9.8–47.1) TNn 51.2 (25.9–> 103) SD 52.9 (17.6–127.6) TN10p 2.7 (2.3–3.2) ID 41.9 (19.0–266.9) TX10p 3.3 (2.8–3.7) TN 21.4 (12.5–37.4) TN90p 8.3 (5.1–13.8) GSL 18.4 (9.6–40.0) TX90p 6.5 (5.1–8.8) TXx 33.9 (20.2–133.0) WSDI 63.0 (33.4–197.0) TNx ~2000 (302.7–> 103) CSDI 14.9 (9.9–25.3) TXn 50.8 (19.1–> 103) DTR 29.9 (11.9–243.2)
(TXx, TNx, TXn, TNn), suggesting that the frequency of very hot and cold days and nights (percentile based indices) has been affected more by anthropogenic climate change than their intensity. With the exception of DTR, record index values appear to be extremely unlikely without the effect of human influence. For indices with the widest the gap between the ALL and NAT lines in Fig. 8(e.g. SD and TX90p) breaking the record appears to be almost impossible without the anthropogenic effect, even when other key factors like unforced variability and the state of the ocean are taken into consideration. Our analysis provides clear evidence of a major shift in the European climate as far as hot and cold extremes are concerned and with continental temperatures projected to continue to rise during the course of the century (IPCC, 2013: Annex I), new extremes of unprecedented frequency and intensity are to be expected.
4. Discussion
Understanding the causes behind changes in temperature extremes is valuable to both decision makers aiming to alleviate adverse impacts, as well as the public learning to adjust to a changing climate. Our study is the first synthesis of two popular attribution methodologies that investigates the effect of human influences on both observed trends since the 1960s and present-day extreme events. Moreover, we focus not only on a single aspect of extremes (e.g. intensity, frequency), but utilise a raft of indices that offers a more complete description of hot and cold extremes, as well as information relevant to impacts. Observations indicate that the warming in extreme temperatures has been continuing unabated (Seneviratne et al., 2014) and, unsurpris-ingly, ourfindings underline the pivotal role of human activity behind this change. Our work also establishes in a formal statistical way that observed multi-decadal trends in extremes over the entire observa-tional area and the European region are largely driven by anthropo-genic forcings. The weaker natural signal cannot be detected for most indices, though it is detected in global changes in the frequency of cold night-time extremes. With respect to annual extremes in Europe, we attempted to construct pre-computed tables, or diagrams, that help estimate the change in the likelihood of rare events as soon as they happen. This, however, turns out not to be possible, as for several indices the change has been so great that current extremes would almost be impossible to occur in the“natural”world. Record breaking events are now estimated to take place every few decades or years and their return times are expected to further decrease as the climate continues to get warmer. Attribution assessments still remain challen-ging for events with a smaller spatial extent for which variability could largely mask the anthropogenic signal and future research is undoubt-edly going to concentrate more on local scales.
Moving forward, we identify several areas where future work is required. Firstly, as attribution assessments rely heavily on models, verifying our results with other models would be an advantage. Here we use two state-of-the-art models that have been evaluated in terms of their representation of extremes. HadGEM3-A is now run at the highest resolution used in event attribution studies and a multi-model analysis would ideally require an ensemble of similarly high-resolution models that is not currently available. While the use of high resolution is expected to provide a more accurate representation of the climate system, multi-model approaches have the advantage of being less affected by model errors, assuming that such errors are not common to most of the models employed. Another practical challenge is the amount of offline data processing required to construct extreme indices from model output, which significantly increases with spatial resolu-tion. Apart from experiments with different external forcings,fi nger-printing analyses also require hundreds, if not thousands of years of control simulations of the pre-industrial climate, while a comprehen-sive study of several indices, as presented here, would be extremely resource-intensive, if multiple models were to be used. An obvious way round would be that modelling centres provide extreme indices as part
of their disseminated output, though this would require coordinated effort. The role of ETCCDI would be pivotal to that end. An extension of this study to rainfall extremes also needs to be undertaken. HadEX2 includes 12 indices of extreme precipitation that can be utilised for this purpose. Although signal detection is more challenging for precipita-tion than temperature, significant progress has been made recently in that area (Min et al., 2011; Wu et al., 2013; Salzmann, 2016). While here we examine changes in European extremes, similar assessments need to be produced for other areas, while regional stakeholders would alsofind great value in analyses of extremes on smaller spatial scales (enabled, for example, by utilising downscaling techniques) which take their toll on local communities. Moreover, further partitioning of the anthropogenic response between individual forcing components might be more meaningful, for example in areas where extremes are strongly linked to changes in land use, or aerosol emissions. Finally, as the warming signal continues to intensify, it is important to use more updated observations in attribution analyses and there are indeed plans to include post-2010 years in the next version of HadEX2 (Alexander et al., 2016).
Acknowledgements
This work was supported by the Joint BEIS/Defra Met Office Hadley Centre Climate Programme (GA01101) and the EUCLEIA project funded by the European Union's Seventh Framework Programme [FP7/2007–2013] under Grant agreement no. 607085. References
Alexander, L.V., Zhang, X., Peterson, T.C., Caesar, J., Gleason, B., Klein Tank, A.M.G., Haylock, M., Collins, D., Trewin, B., Rahimzadeh, F., Tagipour, A., Rupa Kumar, K., Revadekar, J., Griffiths, G., Vincent, L., Stephenson, D.B., Burn, J., Aguilar, E., Brunet, M., Taylor, M., New, M., Zhai, P., Rusticucci, M., Vazquez-Aguirre, J.L., 2006. Global observed changes in daily climate extremes of temperature and precipitation. J. Geophys. Res. 111.http://dx.doi.org/10.1029/2005JD006290. Alexander, L.V., Zhang, X., Hegerl, G., Seneviratne, S., 2016. Implementation Plan for
WCRP Grand Challenge on Understanding and Predicting Weather and Climate Extremes.〈http://www.wcrp-climate.org/images/documents/grand_challenges/ WCRP_Grand_Challenge_Extremes_Implementation_Plan_v20150203.pdf〉.
Allen, M.R., Stott, P.A., 2003. Estimating signal amplitudes in optimalfingerprinting, part I: theory. Clim. Dyn. 21, 477–491.
Bindoff, N.L., Stott, P.A., AchutaRao, K.M., Allen, M.R., Gillett, N., Gutzler, D., Hansingo, K., Hegerl, G., Hu, Y., Jain, S., Mokhov, I.I., Overland, J., Perlwitz, J., Sebbari, R., Zhang, X., 2013. Detection and attribution of climate change: from global to regional. In: Stocker, T.F. (Ed.), Climate Change 2013: The Physical Science Basis. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 867–952.
Christidis, N., Stott, P.A., Brown, S., Hegerl, G.C., Caesar, J., 2005. Detection of changes in temperature extremes during the second half of the 20th century. Geophys. Res. Lett. 32.http://dx.doi.org/10.1029/2005GL023885.
Christidis, N., Stott, P.A., Brown, S., Karoly, D.J., Caesar, J., 2007. Human contribution to the lengthening of the growing season during 1950–1999. J. Clim. 20, 5441–5454.
Christidis, N., Stott, P.A., Brown, S.J., 2011. The role of human activity in the recent warming of extremely warm daytime temperatures. J. Clim. 24, 1922–1930.
Christidis, N., Stott, P.A., Scaife, A., Arribas, A., Jones, G.S., Copsey, D., Knight, J.R., Tennant, W.J., 2013. A new HadGEM3-A based system for attribution of weather and climate-related extreme events. J. Clim. 26, 2756–2783.
Christidis, N., Jones, G.S., Stott, P.A., 2015a. Dramatically increasing chance of extremely hot summers since the 2013 European heatwave. Nat. Clim. Change 5, 46–50.
Christidis, N., Stott, P.A., Zwiers, F.W., 2015b. Fast-track attribution assessments based on pre-computed estimates of changes in the odds of warm extremes. Clim. Dyn. 45, 1547–1564.
Dole, R., Hoerling, M., Perlwitz, J., Eischeid, J., Pegion, P., Zhang, T., Quan, X.W., Xu, T., Murray, D., 2011. Was there a basis for anticipating the 2010 Russian heat wave? Geophys. Res. Lett. 38.http://dx.doi.org/10.1029/2010GL046582.
Donat, M.G., Alexander, L.V., Yang, H., Durre, I., Vose, R., Dunn, R.J.H., Willett, K.M., Aguilar, E., Brunet, M., Caesar, J., Hewitson, B., Jack, C., Klein Tank, A.M.G., Kruger, A.C., Marengo, J., Peterson, T.C., Renom, M., Oria Rojas, C., Rusticucci, M., Salinger, J., Elrayah, A.S., Sekele, S.S., Srivastava, A.K., Trewin, B., Villarroel, C., Vincent, L.A., Zhai, P., Zhang, X., Kitching, S., 2013. Updated analyses of temperature and precipitation extreme indices since the beginning of the twentieth century: the HadEX2 dataset. J. Geophys. Res. Atmos. 118, 2098–2118.
Gillett, N.P., Allen, M.R., Tett, S.F.B., 2000. Modelled and observed variability in atmospheric vertical temperature structure. Clim. Dyn. 16, 49–61.
Hasselmann, K., 1993. Optimalfingerprints for the detection of the time dependent
climate change. J. Clim. 6, 1957–1971.
Hegerl, G.C., Zwiers, F.W., Stott, P.A., Kharin, V.V., 2004. Detectability of anthropogenic changes in annual temperature and precipitation extremes. J. Clim. 17, 3683–3700.
Herring, S.C., Hoerling, M.P., Kossin, J.P., Peterson, T.C., Stott, P.A., 2015. Explaining extreme events of 2014 from a climate perspective. Bull. Am. Meteor. Soc. 96 (12), S1–S172.
Hewitt, C., Mason, S., Walland, D., 2012. The global framework for climate services. Nat. Clim. Change 2, 831–832.
Hewitt, H.T., Copsey, D., Culverwell, I.D., Harris, C.M., Hill, R.S.R., Keen, A.B., McLaren, A.J., Hunke, E.C., 2011. Design and implementation of the infrastructure of HadGEM3: the next-generation Met Office climate modelling system. Geosci. Model Dev. 4, 223–253.
IPCC, 2013. Annex I. atlas of global and regional climate projections. In: van Oldenborgh, G.J., Collins, M., Arblaster, J., Christensen, J.H., Marotzke, J., Power, S.B., Rummukainen, M., Zhou, T. (Eds.), Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA.
Jones, C.D., Hughes, J.K., Bellouin, N., Hardiman, S.C., Jones, G.S., Knight, J., Liddicoat, S., O’Connor, F.M., Andres, R.J., Bell, C., Boo, K.-O., Bozzo, A., Butchart, N., Cadule, P., Corbin, K.D., Doutriaux-Boucher, M., Friedlingstein, P., Gornall, J., Gray, L., Halloran, P.R., Hurtt, G., Ingram, W.J., Lamarque, J.-F., Law, R.M., Meinshausen, M., Osprey, S., Palin, E.J., Parsons Chin, L., Raddatz, T., Sanderson, M.G., Sellar, A.A., Schurer, A., Valdes, P., Wood, N., Woodward, S., Yoshioka, M., Zerroukat, M., 2011. The HadGEM2-ES implementation of CMIP5 centennial simulations. Geosci. Model Dev. 4, 543–570.
Jones, J.S., Stott, P.A., Christidis, N., 2013. Attribution of observed historical near surface temperature variations to anthropogenic and natural causes using CMIP5 simulations. J. Geophys. Res. 118, 4001–4024.
Lelieveld, J., Hadjinicolaou, P., Kostopoulou, E., Giannakopoulos, C., Pozzer, A., Tanarhte, M., Tyrlis, E., 2014. Model projected heat extremes and air pollution in the eastern Mediterranean and Middle East in the twenty-first century. Reg. Eniron. Change 14, 1937–1949.
Lewis, S.C., Karoly, D.J., 2013. Evaluation of historical diurnal temperature range trends in CMIP5 models. J. Clim. 26, 9077–9089.
Lewis, S.C., Karoly, D.J., Yu, M., 2014. Quantitative estimates of anthropogenic contributions to extreme national and state monthly, seasonal and annual average temperature for Australia. Aust. Met. Ocean. J. 64, 215–230.
Lott, F., Stott, P.A., Mitchell, D.M., Christidis, N., Gillett, N., Haimberger, L., Perlwitz, J., Thorne, P.W., 2013. Models versus radiosondes in the free atmosphere: a new detection and attribution analysis of temperature. J. Geophys. Res. 118, 2609–2619.
Massey, N., Jones, R., Otto, F.E.L., Aina, T., Wilson, S., Murphy, J.M., Hassell, D., Yamazaki, Y.H., Allen, M.R., 2015. Weather@home development and validation of a very large ensemble modelling system for probabilistic event attribution. Q. J. Meteor. Soc. 141, 1528–1545.
McMichael, A.J., 2013. Globalization, climate change, and human health. N. Engl. J. Med. 368, 1335–1343.
Min, S.-K., Zhang, X., Zwiers, F.W., Hegerl, G.C., 2011. Human contribution to more-intense precipitation extremes. Nature 470, 378–381.
Min, S.-K., Zhang, X., Zwiers, F., Wehner, M., 2013. Multimodel detection and attribution of extreme temperature changes. J. Clim. 26, 7430–7451.
Morak, S., Hegerl, G.C., Christidis, N., 2013. Detectable changes in the frequency of temperature extremes. J. Clim. 26, 1561–1574.
National Academies of Sciences (NAS), Engineering and medicine, Attribution of Extreme Weather Events in the Context of Climate Change. The National Acedemies Press, Washington DC, 2016.http://dx.doi.org/10.17226/21852
Pall, P., Aina, T., Stone, D.A., Stott, P.A., Nozawa, T., Hilberts, A.G.J., Lohmann, D., Allen, M.R., 2011. Anthropogenic greenhouse gas contribution toflood risk in England and Wales in autumn 2000. Nature 470, 382–385.
Rayner, N.A., Parker, D.E., Horton, E.B., Folland, C.K., Alexander, L.V., Rowell, D.P., Kent, E.C., Kaplan, A., 2003. Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res. 108.
http://dx.doi.org/10.1029/2002JD002670.
Salzmann, M., 2016. Global warming without Global mean precipitation increase? Sci. Adv., 24.http://dx.doi.org/10.1126/sciadv.1501572.
Seneviratne, S.I., Nicholls, N., Easterling, D., Goodess, C.M., Kanae, S., Kossin, J., Luo, Y., Marengo, J., McInnes, K., Rahimi, M., Reichstein, M., Sorteberg, A., Vera, C., Zhang X., 2012. Changes in climate extremes and their impacts on the natural physical environment. In: C.B., Barros, V., Stocker, T.F., Qin, D., Dokken, D.J., Ebi, K.L., Mastrandrea, M.D., Mach, K.J., Plattner, G.-K., Allen, S.K., Tignor, M., Midgley, P.M. (Eds.), Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation Field, A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change, pp. 109–230.
Seneviratne, S.I., Donat, M.G., Mueller, B., Alexander, L.V., 2014. No pause in the increase of hot temperature extremes. Nat. Clim. Change 4, 161–163.
Shiogama, H., Imada, Y., Mori, M., Mizuta, R., Stone, D., Yoshida, K., Arakawa, O., Ikeda, M., Takahashi, C., Arai, M., Ishii, M., Watanabe, M., Kimoto, M., 2016. Attributing historical changes in probabilities of record-breaking daily temperature and precipitation extreme events. Sci. Online Lett. Atmos. 12, 225–231.
Stone, D.A., Pall, P., 2016. A benchmark estimate of the effect of anthropogenic emissions on the ocean surface. J. Adv. Model. Earth Sys, (submitted for publication).
Stott, P.A., Stone, D.A., Allen, M.R., 2004. Human contribution to the European heatwave of 2003. Nature 432, 610–613.
Stott, P.A., Christidis, N., Otto, F., Sun, Y., Vanderlinden, J.-P., van Oldenborgh, G.-J., Vautard, R., von Storch, H., Walton, P., Yiou, P., Zwiers, F.W., 2016. Attribution of extreme weather and climate-related events. WIREs Clim. Change 7, 23–41. Takayabu, I., Hibino, K., Sasaki, H., Shiogama, H., Mori, N., Shibutani, Y., Takemi, T.,
2015. Climate change effects on the worst-case storm surge: a case study of Typhoon Haiyan. Environ. Res. Lett., 10.http://dx.doi.org/10.1088/1748-9326/10/6/ 064011.
Teixeira, E.I., Fischer, G., van Velthuizen, H., Walter, C., Ewert, F., 2013. Global hot-spots of heat stress on agricultural crops due to climate change. Agric. For. Meteorol. 170, 206–215.
Uhe, P., Otto, F.E.L., Haustein, K., van Oldenborgh, G.J., King, A.D., Wallom, D.C.H., Allen, M.R., Cullen, H., 2016. Comparison of methods: attributing the 2014 record European temperatures to human influences. Geophys. Res. Lett. 43, 8685–8693.
Wen, Q.H., Zhang, X., Xu, Y., Wang, B., 2013. Detecting human influence on extreme temperatures in China. Geophys. Res. Lett. 40, 1171–1176.
Wolf, T., McGregor, G., 2013. The development of a heat wave vulnerability index for London, United Kingdom. Weather Clim. Extrem. 1, 59–68.
Wu, P., Christidis, N., Stott, P.A., 2013. Anthropogenic impact on the Earth’s hydrological cycle. Nature. Clim. Change 3, 807–810.
Yoon, J.-H., Wang, S., Gillies, R.R., Hipps, L., Kravitz, B., Rasch, P.J., 2015. Extremefire season in California: A glimpse into the future? in: Explaining Extremes of 2014 from a Climate Perspective. Bull. Am. Meteor. Soc., vol. 96, 12, 2015, pp. S5–S9
Zhou, L., Dickinson, R.E., Dai, A., Dirmeyer, P., 2010. Detection and attribution of anthropogenic forcing to diurnal temperature range changes from 1950 to 1999: comparing multi-model simulations with observations. Clim. Dyn. 35, 1289–1307.
Zwiers, F.W., Zhang, X., Feng, Y., 2011. Anthropogenic influence on long return period daily temperature extremes at regional scales. J. Clim. 24, 296–307.