Abstract
As European heatwaves become more severe, summers in the United Kingdom (UK) are also getting warmer. The UK record temperature of 38.7 °C set in Cambridge in July 2019 prompts the question of whether exceeding 40 °C is now within reach. Here, we show how human influence is increasing the likelihood of exceeding 30, 35 and 40 °C locally. We utilise observations to relate local to UK mean extremes and apply the resulting relationships to climate model data in a riskbased attribution methodology. We find that temperatures above 35 °C are becoming increasingly common in the southeast, while by 2100 many areas in the north are likely to exceed 30 °C at least once per decade. Summers which see days above 40 °C somewhere in the UK have a return time of 100300 years at present, but, without mitigating greenhouse gas emissions, this can decrease to 3.5 years by 2100.
Introduction
Intensification of hot extremes has continued unabated in recent decades^{1}, posing a threat to human health^{2,3} and bringing forth a raft of further socioeconomic impacts^{4,5}. Europe is gearing up for more frequent and intense heatwaves^{6} and while the UK has not yet borne the brunt of extreme continental heat, its summer temperatures are decidedly on the rise^{7,8}. Attribution research provides strong evidence that hot extremes are becoming more frequent and intense^{9} under the influence of humancaused climate change^{10,11}. The UK summer temperature of 2018 was a joint record, estimated to have become 30 times more likely due to anthropogenic causes^{12}. A year later, during a severe heatwave in western Europe^{13}, the warmest daily temperature averaged over the UK reached a new peak (Fig. 1a) and the highest temperature in the country ever recorded was registered in Cambridge. These consecutive summer extremes are exposing the UK’s vulnerability to such weather with ensuing impacts highlighted in the media, including a mortality spike in tandem with the 2019 event^{14,15}, and a sharp heatwavedriven fall in overseas holiday demand that might have contributed to the collapse of the Thomas Cook travel group^{16}. Therefore, the need to understand how the likelihood of extremely hot temperatures is changing under the anthropogenic effect on the climate is pressing and essential to decisionmakers planning the UK’s adaptation strategy.
To respond to this need, we compute observed and modelled values of the warmest daily maximum temperature in individual years (tx01) and estimate how the likelihood of exceeding extreme thresholds has been changing since 1900 and how it may further change in the remaining of this century under different emission scenarios^{17}. We find that the likelihood of extremely warm days in the UK has been increasing and will continue to do so during the course of the century with the most extreme temperatures expected to be observed in the southeast England. The likelihood of exceeding 40 °C anywhere in the UK in a given year has also been rapidly increasing, and, without curbing of greenhouse gas emissions, such extremes could be taking place every few years in the climate of 2100.
Results
Observed changes in UK tx01 extremes
Limitations arising from the spatial resolution of climate models and the coverage of observation stations often prevent attribution studies on local scales and have kept the focus on extremes over larger, subcontinental areas^{18}. Although downscaling of model output has occasionally been employed to investigate local events^{19}, the lack of reliable observations makes it difficult to evaluate the models. Here, we take advantage of the recent upgrade of the HadUKGrid dataset^{20} for daily maximum temperature, which now provides observational data on a highresolution grid of 1 × 1 km. The resolution of the dataset enables us to model the relationship between local and UK mean tx01. This simple downscaling technique allows us to estimate changes in the likelihood of UK extreme temperatures locally from model experiments with and without anthropogenic forcings that provide data on a relatively coarse resolution.
The HadUKGrid data cover the entire UK and are available for the period 1960–present. Annual values of tx01 are calculated for all the grid boxes, and a map of the trends over the observational period is illustrated in Fig. 1b. Warming trends dominate and are most prominent in the southeast, where they may locally reach 1 °C decade^{−1}. Testing the significance of the trends with the Mann–Kendall test, indicates they are significantly different than zero in most regions, but not in parts of Scotland where there is a weaker warming and also areas of cooling. It should be noted that although this is a useful qualitative assessment, the trend estimates are sensitive to the start and end dates. The UK’s warmest day in a year is also estimated (Fig. 1a) after averaging daily maximum temperatures of each day over the observational area. Year 2019 has the highest UK mean tx01 value on record, though climate models indicate that due to internal variability there is a current risk of even higher temperatures.
Transfer functions for the estimation of local tx01
Even though climate models can provide reliable estimates of the UK mean tx01, as discussed later, their spatial resolution is still too coarse to yield local estimates on a 1km grid. We therefore derive observationally based transfer functions to obtain local tx01 values from the UK mean that we can later apply to model data. A simple linear fit is applied to all the grid boxes of HadUKGrid to model the relationship between the gridbox tx01 and its UKmean counterpart. An example for a gridbox in London is shown in Fig. 2a. We account for the range of values of the response variable (local tx01) by estimating its confidence bounds for each percentile^{21} (orange lines in Fig. 2b) and so end up with a set of 100 possible transfer functions at each gridbox ('Methods'). Moreover, given the limited sample of 60 years, our analysis also investigates the uncertainty in the transfer functions, by applying a Monte Carlo bootstrap procedure that resamples the observational data. This procedure offers alternative transfer functions (grey lines in Fig. 2c), each with an associated set of 100 variants, as previously explained. Finally, it is important to establish whether gridbox temperatures on spatial scales of 1 km can adequately represent local temperatures. To this end, we compare station observations across the UK with the HadUKGrid temperature of the gridbox where each station is located and confirm a good agreement in all cases (Fig. 2d; Supplementary Note 1 and Supplementary Fig. 1).
The CMIP5 ensemble
Estimates of the UK mean tx01 are next obtained from simulations with 16 climate models that participated in the World Climate Research Programme’s Coupled Model Intercomparison Project phase 5 (CMIP5)^{22}. The models provide simulations of the actual climate (all forcings) under the influence of both natural and anthropogenic forcings, as well as of a hypothetical natural world without the effect of human influence. Anthropogenic forcings include historical changes in wellmixed greenhouse gases, aerosols, ozone and landuse. Natural forcings include only volcanic aerosol emissions and changes in the solar irradiance. Data from the following atmosphere–ocean coupled CMIP5 models are used in the analysis:
ACCESS13, bcccsm11, CCSM4, CESM1CAM5, CNRMCM5, CSIROMk360, CanESM2, GFDLCM3, GFDLESM2M, HadGEM3ES, IPSLCM5ALR, IPSLCM5AMR, MIROCESM, MIROCESMCHEM, MRICGCM3, NorESM1M.
The allforcings experiment was also extended to the end of the twentyfirst century with projections that follow the representative concentration pathway (RCP) scenarios^{17} RCP 4.5 and 8.5 scenarios. Given the substantial volume of the simulated daily data, we employ only one simulation per model and per experiment (r1i1p1), thus placing equal weight on all models. We apply a simple biascorrection to all the model data to make sure that the mean tx01 value in the allforcing simulations during the base period 1961–1990 agrees with the observed mean value. For consistency, we regrid all the models on a common 60km resolution grid, on which the observations are also available, and then mask the model data with the observations on the same grid and compute the UKmean value. For each model, we subtract the mean observed from the mean modelled tx01 over the base period and remove the resulting bias from the tx01 estimates with and without anthropogenic forcings derived from the same model.
Model evaluation
Evaluation of the models against observations in attribution analyses is essential, in order to determine whether they are fitforpurpose. Here we compare the data of the UK mean tx01 during 1960–2019 simulated by the allforcings experiment with observationally based data from HadUKGrid. We apply a set of standard evaluation tests to assess how well the models represent the trends, variability and distribution of tx01. Results are shown in Fig. 3. First, we estimate the trends in tx01 over the observational period and its associated ± 2 standard deviation range computed with leastsquare fits (Fig. 3a). The observations indicate a small positive trend, but its precise value is uncertain because of the effect of variability. Although some models produce weaker trends than the observations, the relatively short length of the observational period prevents a more detailed assessment, and since all models have a range that overlaps with the one from the observations, they are all included in the analysis. We next assess the simulated variability over different timescales with power spectra from detrended tx01 timeseries (Fig. 3b). The observed spectrum is found to be within the range of the modelled spectra, albeit towards the higher end, though again sampling limitations need to be taken into consideration. The observed and modelled distribution of the UK mean tx01 in the period 1960–2019 is illustrated in Fig. 3c. The modelled distribution is constructed with data from all 16 models and is found to be indistinguishable from the observed distribution, when a Kolmogorov–Smirnov (KS) test is applied (Pvalue greater than 0.1). The shape of the observed distribution (histogram) indicates that temperatures in upper tail are not well sampled because of the limited length of the record. This is also reflected in the quantile–quantile plot of Fig. 3d, which nonetheless still indicates that the models can realistically represent the distribution of tx01. In conclusion, the simple evaluation tests described here do not raise concerns about the ability of the multimodel ensemble to represent the UK tx01, but, on the contrary, indicate it is a sufficiently good dataset for the attribution analysis.
Testing the transfer functions with the CMIP5 models
A main assumption in our methodology is that the transfer functions we derive from the observations are not sensitive to the nonstationarity of the climate to an extent that would weaken our results. Local and regional temperatures are influenced by both internal variability and external forcings, and the interplay between these two factors would be different in different periods^{23}. Using the CMIP5 models, we test the effect of nonstationarities and the interplay of variability and external forcings. First, we establish that the choice of the training period for which the transfer functions are derived does not compromise the analysis. We derive a set of three transfer functions from the models, corresponding to three different scenarios:
 1.
Strong forcing influence: The transfer functions are derived from simulations with all forcings for future years 2020–2100, characterised by a strong anthropogenic influence.
 2.
Variability influence only: The transfer functions are derived from simulations with natural forcings only, i.e., in the absence of any anthropogenic influence.
 3.
Mixed response: The transfer functions are derived from simulations with all forcings for the period 1960–2019, i.e., the same as the observational period used in our main analysis, for which anthropogenic influence increases with time.
The transfer functions obtained for four different grid boxes (in the Southeast London area, Scotland, Central England and Northern Ireland) are shown in Fig. 4. It is evident that different training periods yield similar transfer functions. The largest discrepancies are generally within a degree (in most cases much smaller) and are eclipsed by uncertainties due to sampling (Fig. 2c) and internal variability (Fig. 2b), which have been included in our approach. For example, the uncertainty range in Fig. 2c spans ~10 °C, and is accounted for by using 100 variants of the transfer function. We repeated the sensitivity tests over different grid boxes, training periods and with individual models, and found no indication of a large uncertainty due to the nonstationary climate that would adversely affect our results.
We also test whether internal variability may significantly change in a nonstationary climate over the course of the century. We use the multimodel ensemble mean of the UK tx01 timeseries as an estimate of the forced response and subtract it from all the timeseries by individual models. We then compute the standard deviation in 5year rolling windows, which provides the timeseries of the standard deviation shown in Fig. 5. The models indicate no major change in variability over the period 1900–2100. Finally, as a way of assessing the quality of the of the simulated UK mean tx01 samples used in the analysis, we examine how different model combinations might affect the local tx01 distributions. We find that different samples yield similar distributions (Supplementary Note 2 and Supplementary Fig. 2) and conclude that the sample choice does not introduce a considerable uncertainty.
Attribution on local scales
We adopt a popular riskbased event attribution framework^{24}, whereby tx01 estimates from the two multimodel ensembles (with and without human influence) are used to generate probability distributions for the actual and natural climate. Local distributions are constructed on the observational (1 × 1 km) grid by applying the previously derived transfer functions to the simulated UK mean tx01 and the uncertainty in the transfer functions is accounted for by the bootstrapping procedure. Details on the construction of the local distributions are given in the ‘Methods’.
Results from our analysis for a gridbox in London are illustrated in Fig. 6. For this example, the allforcing simulations were extended to 2100 with the RCP 4.5 scenario. The cumulative distribution function (CDF) of tx01 shifts to higher values with time, increasing the likelihood of exceeding 40 °C, which is nearzero in the natural climate (Fig. 6a). Exceeding the lower threshold of 30 °C in London is common and occurs almost every year, even without the anthropogenic effect (Fig. 6b). However, temperatures above 35 °C are now 2–3 times more likely than in the natural climate (Fig. 6f), and model projections suggest they will occur at least twice a decade at the end of the century (Fig. 6c). The likelihood of exceeding 40 °C in the reference location is still extremely low, but is rapidly increasing, with the return time falling from thousands of years in the natural world to hundreds, or even tens of years by 2100 (Fig. 6d). For London, these likelihoods could increase even further as a consequence of increased urbanisation in future or from higher rates of local anthropogenic heat release^{25}, for example from wider adoption of air conditioning during heatwaves.
Repeating the analysis on all grid boxes, we also produce maps showing the return time for local exceedances of the three temperature thresholds (Fig. 7). Given the high spatial resolution of the plotted fields, certain topographic or coastal effects become evident on close inspection. Besides the noticeable contrast between warmer summers in the south and cooler in the north, the southeast England clearly stands out as the region where high temperature extremes are most likely to occur. Compared with the natural world, there are now more areas likely to see temperatures exceeding 30 or 35 °C, while the 40 °C threshold is still very rare, even in the southeast. By the end of the century, most areas in the north of the UK will also be regularly experiencing days with temperatures at least as warm as 30 °C, while crossing the 35 °C becomes common in the southeast under RCP 4.5 and over most of England under RCP 8.5. The highest threshold of 40 °C is to be exceeded at least once a century in the London area under RCP 4.5, and several times a century over most of southeast England under RCP 8.5. The effect of the uncertainty in the transfer functions has also been assessed (Supplementary Note 3 and Supplementary Figs. 3 and 4), and although it may change to some extent the intensity and spread of the map features in Fig. 7, the main conclusions still hold.
Chances of exceeding extreme thresholds anywhere in the UK
We finally compute the likelihood of exceeding an extreme threshold in a given year, not at a specific location, but anywhere in the UK. The chance of rising above, for example, 40 °C, the most extreme threshold examined here, might still be very low for a given location, but has been increasing in most areas under the influence of warming trends (Fig. 1b). When all grid boxes are examined together, the likelihood of getting at least one gridbox that exceeds 40 °C in a specific year is expected to be higher than a local likelihood. The CMIP5 models provide a large number of alternative representations for every year. Each representation may yield a hit, i.e., at least one location where the reference tx01 threshold is exceeded, or not, and the likelihood of exceeding the threshold anywhere in the UK may thus be determined by the count of hits ('Methods'). Figure 8 depicts timeseries of the return time for different threshold exceedances and its expected range in the natural climate. Rising above 35 °C is estimated to occur once every 5 years at present and almost every year by the end of the century (Fig. 8b). Also, the probability of recording 40 °C, or above, in the UK is now rapidly accelerating and begins to rise clearly above the range of the natural climate (Fig. 8c). The return time for the 40 °C threshold is reduced from 100–1000s of years in the natural climate to 100–300 years in the present climate and to only about 15 years by 2100 under the mediumemissions scenario (RCP 4.5) and 3.5 years under the highemissions scenario (RCP 8.5).
Discussion
Our study demonstrates that humancaused climate change has set hotday extremes in the UK on a course towards temperatures that would be too high to be observed in the natural climate. As the warming continues, new records are expected in coming decades, with the most severe extremes likely to occur in the southeast of the UK. Our attribution analysis derives local information from observations rather than regional models and investigates highimpact extremes that could break out anywhere in the UK rather than in prescribed locations. There are, of course, uncertainties in our analysis, some of which we have explored and tried to address, including uncertainties in the transfer functions and the limited number of years they are based on, or the limited number of models employed and their ability to represent the UK climate. Although the transfer functions make a distinction between urban and rural locations, large future changes in the UK’s urban landscape could present a caveat in the analysis, though this is likely to affect only a small fraction of grid boxes. Future probability estimates are found to be sensitive to the choice of the RCP in the model simulations. Here, we estimate the future likelihood of extremes under both a mid and high range RCP scenario. However, if emissions are reduced in line with the Paris climate agreement, the future probabilities are expected to be lower. Despite these uncertainties, our analysis still clearly establishes the nature of already realised and future changes in extreme temperatures including their spatial characteristics, information that can help the UK plan its resilience to heat extremes.
Methods
Transfer functions
The local and UK mean tx01 is computed from the observations for every year in the period 1960–2019. For a given gridbox, we represent the dependence of the local tx01 on the UK mean with a simple linear model:
The linear regression is fitted to the n = 60 observed annual values of tx01(local) and tx01(UK) and ordinary least squares are used to estimate the coefficients α_{0} and α_{1}.
If y_{i}^{obs} and y_{i}^{fit} denote the observed and fitted values of tx01(local) in year i, and SSE the sum of squared errors:
then the confidence interval for the response variable and the (1 + p)/2 quantile of the t(n − 2) distribution is estimated^{19} as:
where x_{i}^{obs} denotes the observed value of tx01(UK) in year i, \(\overline {\mathrm{X}}\) the mean of the observed tx01(UK) values and SXX is calculated as:
The confidence bounds are represented by very shallow hyperbolas that can be almost perfectly approximated by straight lines, as done in this study. Using the best fit to the observed data and the 1st to 99th percentiles for the estimation of the uncertainty range for the response variable, we end up with a set of 100 transfer functions per gridbox. Therefore, when we apply the transfer functions to a model simulated value of tx01(UK), we obtain 100 values that represent the possible range of the tx01 at the reference location.
Uncertainty in the transfer functions
Although the 60year long observational dataset used to derive the transfer functions is deemed large enough to provide reliable estimates of the linear fits at every gridbox, sampling uncertainty will still have some effect on the analysis results. This kind of uncertainty is commonly accounted for by a simple a Monte Carlo bootstrap procedure^{26} that we also employ here. The procedure involves random resampling of the 60 annual pairs of tx01(UK) and tx01(local) and deriving a new set of transfer functions from the resampled data. Multiple resampling provides multiple sets of transfer functions. Each set, as explained next, can provide an estimate of the likelihood of the local tx01 exceeding a certain temperature threshold and by repeating the calculations with all the bootstrapped sets of transfer functions, we obtain multiple estimates of the likelihood, which enables us to estimate its uncertainty range.
Estimation of the local tx01 probabilities
For every year of each experiment we obtain samples of 1600 tx01 values for every gridbox (16 CMIP5 models that provide estimates of the UK mean tx01 × 100 transfer functions). We further increase the sample size in the allforcings experiment to 32,000 by calculating the probabilities in 20year rolling windows during the period 1900–2100 (i.e., in time segments 1900–1919, 1901–1920, …, 2081–2100). We select years 2011–2030 to represent the presentday climate and 2081–2100 to represent the late twentyfirst century climate. Future probabilities are estimated with both RCP 4.5 and 8.5, whereas only simulations with RCP 4.5 were used to estimate presentday probabilities. For the natural world, we aggregate all simulated years (1900–2005), assuming that the natural climate is stationary in the long run, which yields samples of 169,600 tx01 values (16 models × 100 transfer functions × 106 years) for every gridbox. The resulting samples provide estimates of the likelihood of exceeding the preselected thresholds of 30, 35 and 40 °C. Given the large sample sizes, probabilities are computed by a simple count of threshold exceedances. Using alternative sets of the transfer functions from the bootstrapping procedure described earlier, we recalculate the probabilities multiple times and estimate their 5–95% range to account for the uncertainty in the empirical relationships.
Probability of exceeding a threshold anywhere in the UK
The 16 CMIP5 models provide 320 simulated years in consecutive 20year rolling windows. The sample increases to 32,000 when we apply the set of 100 transfer functions for each gridbox to obtain highresolution annual maps of the tx01. Counting how many of these 32,000 maps include at least one location where the reference threshold is exceeded, allows us to the calculate the probability estimate. As before, the probability is recomputed with alternative sets of the transfer functions to assess their uncertainty. The natural probabilities are also aggregated to provide the 5–95% range for the natural climate.
Data availability
The HadUKGrid temperature data and station temperature data from the Met Office Integrated Data Archive System (MIDAS) that support the findings of this study are available from the CEDA Archive, http://archive.ceda.ac.uk. The CMIP5 simulated temperature data that support the findings of this study are available from the Earth System Grid Federation (ESGF) Archive, https://esgf.llnl.gov/.
Code availability
IDL code used for the analysis is available upon request.
References
Seneviratne, S. I., Donat, M., Mueller, B. & Alexander, L. V. No pause in the increase of hot temperature extremes. Nat. Clim. Change 4, 161–163 (2014).
Christidis, N., Mitchell, D. & Stott, P. A. Anthropogenic climate change and heat effects on health. Int. J. Climatol. 39, 4751–4768 (2019).
Christidis, N., Donaldson, G. C. & Stott, P. A. Causes for the recent changes in coldrelated and heatrelated mortality in England and Wales. Clim. Change 102, 539–553 (2010).
Easterling, D. R. et al. Climate extremes: observations, modelling and impacts. Science 289, 2068–2074 (2000).
Carleton, T. A. & Hsiang, S. M. Social and economic impacts of climate. Science 353, 6304 (2016).
Christidis, N., Jones, G. S. & Stott, P. A. Dramatically increasing chance of extremely hot summers since the 2003 European heatwave. Nat. Clim. Change 5, 46–50 (2015).
Lowe, J. A. et al. UKCP18 science overview report. Available from: https://www.metoffice.gov.uk/pub/data/weather/uk/ukcp18/sciencereports/UKCP18Overviewreport.pdf (2018).
Kendon, M., McCarthy, M., Jevrejeva, S., Matthews, A. & Legg, T. State of the UK climate 2018. Int. J. Climatol. 39(Suppl.1), 1–55 (2019).
Herring, S. C., Christidis, N., Hoell, A., Hoerling, M. & Stott, P. A. Introduction to explaining extreme events of 2017 from a climate perspective. In “Explaining extreme events of 2017 from a climate perspective”. Bull. Am. Meteorol. Soc. 100, S1–S4 (2019).
Stott, P. A. Weather risks in a warming world. Nat. Clim. Change 5, 516–517 (2015).
Fischer., E. M. & Knutti, R. Anthropogenic contribution to global occurrence of heavyprecipitation and hightemperature extremes. Nat. Clim. Change 5, 560–564 (2015).
McCarthy, M. et al. Drivers of the UK summer heatwave of 2018. Weather 74, 390–396 (2019).
Vautard, R. et al. Human contribution to the recordbreaking July 2019 heat wave in Western Europe. World Weather Attribution report available from: https://www.worldweatherattribution.org/humancontributiontotherecordbreakingjuly2019heatwaveinwesterneurope (2019).
Office for National Statistics. Do summer heatwaves lead to an increase in deaths? https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/causesofdeath/articles/dosummerheatwavesleadtoanincreaseindeaths/20191007 (2019).
Roberts, M. Summer heatwave ‘deathspike’ seen in UK. BBC News online. https://www.bbc.co.uk/news/health49958201 (2019).
Collinson, P. Why did Thomas Cook collapse after 178 years in business? The Guardian. https://www.theguardian.com/business/2019/sep/23/thomascookastheworldturnedthesunceasedtoshineonvenerabletouroperator (2019).
Moss, R. H. et al. The next generation of scenarios for climate change research and assessment. Nature 463, 747–756 (2010).
Angélil, O. et al. On the nonlinearity of spatial scales in extreme weather attribution statements. Clim. Dyn. 50, 2739–2752 (2018).
Pall, P. et al. Anthropogenic greenhouse gas contributution to flood risk in England and Wales in autumn 2000. Nature 470, 382–385 (2011).
Hollis, D., McCarthy, M. P., Kendon, M., Legg, T. & Simpson, I. HadUK‐Grid—A new UK dataset of gridded climate observations. Geosci Data J. https://doi.org/10.1002/gdj3.78 (2019).
von Storch, H. & Zwiers, F. Statistical Analysis in Climate Research Ch.8 (Cambridge Univ. Press, Cambridge, 2003).
Taylor, K., Stouffer, R. J. & Meehl, G. A. An overview of CMIP5 and the experimental design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012).
Deser, C., Knutti, R., Solomon, S. & Phillips, A. S. Communication of the role of natural variability in future North American climate. Nat. Clim. Change 2, 775–779 (2012).
Christidis, N., Ciavarella, A. & Stott, P. A. Different ways of framing event attribution questions: the example of warm and wet winters in the United Kingdom similar to 2015/16. J. Clim. 31, 4827–4845 (2018).
McCarthy, M. P., Harpham, C., Goodess, C. M. & Jones, P. D. Simulating climate change in UK cities using a regional climate model, HadRM3. Int. J. Climatol. 32, 1875–1888 (2012).
Christidis, N. et al. A new HadGEM3A based system for attribution of weather and climaterelated extreme events. J. Clim. 26, 2756–2783 (2013).
Acknowledgements
This work was supported by the Met Office Hadley Centre Climate Programme funded by BEIS and Defra and the EUPHEME project, which is part of ERA4CS, an ERANET initiated JPI Climate and cofunded by the European Union (Grant 690462).
Author information
Authors and Affiliations
Contributions
N.C. organised the research and carried out the analysis. M.M. provided data and together with PAS helped design the study. N.C. wrote the paper with help from all coauthors.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Peer review information Nature Communications thanks Jakob Zscheischler and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Christidis, N., McCarthy, M. & Stott, P.A. The increasing likelihood of temperatures above 30 to 40 °C in the United Kingdom. Nat Commun 11, 3093 (2020). https://doi.org/10.1038/s41467020168340
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467020168340
This article is cited by

Extreme heatwaves: surprising lessons from the record warmth
Nature (2022)

Future extension of the UK summer and its impact on autumn precipitation
Climate Dynamics (2022)

An earlypostmortem metabolic comparison among three extreme acute heat stress temperature settings in chicken breast muscle
Journal of Food Science and Technology (2021)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.