1. Introduction
The Fifth Assessment Report (AR5) of Working Group I of the Intergovernmental Panel on Climate Change (IPCC) concludes that “warming of the climate system is unequivocal” (IPCC 2013, p. 4). Climate extremes display clear changes, and all temperature-related indices show significant and widespread warming trends. Since the mid-twentieth century, the numbers of cold days and nights have decreased, whereas the numbers of warm days and nights have increased globally (Alexander et al. 2006; Zhang et al. 2011; Donat et al. 2014). The extreme temperature indices calculated from daily minimum (nighttime) temperatures generally show stronger trends than those calculated from daily maximum (daytime) temperatures. These changes in extreme temperatures have had substantial impacts on society, human health, the economy, and ecosystems. Many authors have investigated different aspects of extreme temperature changes, including their intensity, frequency, and duration. Most studies focus on changes in the intensity and frequency of extreme temperatures because of the high representativeness of these indices. Although the fixed threshold indices are not necessarily meaningful for all climate zones because of the fixed thresholds used in the definitions of the indices, changes in these indices can have profound impacts on particular sectors of society or ecosystems (Alexander et al. 2006). For example, changes in frost days may affect agricultural practices and engineering applications (e.g., Terando et al. 2012). Increases in tropical nights may have serious impacts on human health and are thought to occur primarily in combination with extended heat wave events, particularly in subtropical regions (Sillmann and Roeckner 2008). However, these fixed threshold indices have received less attention than other indices in previous studies, although a few studies indicate that the fixed threshold indices display pronounced changes at the global and regional scales (Donat et al. 2013; Zhou et al. 2016). Sillmann et al. (2013a,b) show that the CMIP5 models are generally able to reproduce the increasing trends in the warm fixed threshold indices (those for summer days and tropical nights) and the decreasing trends in the cold indices (ice days and frost days; these and the preceding indices are defined in section 2) when their anomalies relative to a reference period are considered. These models also project future increases in warm days and decreases in cold days.
There is a clear societal interest in whether changes in extreme events can be linked to anthropogenic forcing or natural climate variability. Many studies have indicated the existence of a human influence on the climate system. Previous studies have shown that the anthropogenic forcing can be identified in long-term changes in extreme temperatures at the global and continental scales, as well as within some countries (Christidis et al. 2005, 2011; Zwiers et al. 2011; Stott et al. 2011; Wen et al. 2013; Kim et al. 2016; Sun et al. 2016; Lu et al. 2016; Yin et al. 2017; Dong et al. 2018). Christidis et al. (2005) applied the optimal fingerprinting method to detect the anthropogenic influence on long-term changes in extreme temperature at the global scale for the first time. Different researchers subsequently used various sets of climate models and methods to identify clear anthropogenic fingerprints in the intensity and frequency of extreme temperatures at the continental and national scales (Min et al. 2013; Wen et al. 2013; Morak et al. 2013; Lu et al. 2016; Yin et al. 2017; Dong et al. 2018). In addition, attribution studies that investigate individual extreme events show that anthropogenic forcing clearly increases the occurrence probability of extreme events associated with high temperatures in different countries (Lewis and Karoly 2013; Sun et al. 2014). Recently, Christidis and Stott (2016) focused on 16 extreme temperature indices and detected anthropogenic forcing signals in these indices in Europe and at the quasi-global scale using HadGEM2-ES Earth system model for the analysis of multidecadal changes and HadGEM3-A atmospheric model for event attribution. This study provides a strong indication that the human influence has significantly influenced the characteristics of temperature extremes in recent decades, and the anthropogenic signal may rise above the internal variability both quasi-globally and on continental scales. However, Christidis and Stott (2016) also note that the effects of the human influence cannot be detected for the frost days and ice days threshold indices. They suggest that the reasons include a high degree of natural variability and the fact that the indices are defined relative to a temperature threshold (0°C) that is not relevant in warmer regions.
These previous studies show clearly that the detection of intensity-based (Min et al. 2013; Wen et al. 2013, Kim et al. 2016; Yin et al.2017) and frequency-based (Morak et al. 2013; Lu et al. 2016) indices of extreme temperature have been widely investigated, whereas other indices, such as fixed threshold indices, have not. The study by Christidis and Stott (2016) is based on only one climate model and focuses on the globe and Europe. The optimal fingerprint method developed by Allen and Stott (2003) was used in this study, and the spatial–temporal evolution of the extreme temperatures were projected to the first few leading empirical orthogonal functions (EOFs) of the model-simulated variability. Whether the changes in these fixed threshold indices in other regions of the world can be detected using the multimodel results is unclear. Here, we focus on four fixed threshold indices, as defined by the Expert Team on Climate Change Detection and Indices (ETCCDI), at the global and continental scales, as well as China. We use the outputs from multiple models that participated in phase 5 of the Coupled Model Intercomparison Project (CMIP5) (Taylor et al. 2012), the HadEX2 observational dataset, and observational data from stations in China. Multimodel ensembles generally provide better estimates of the response of the climate system to external forcings than individual models because the multimodel ensemble has very little influence of internal climate variability. The use of different observational datasets provides an opportunity to test the influence of observational data on the detection results. We also employ an improved detection method that is based on regularized covariance estimation (Ribes and Terray 2013); moreover, we conduct dimension reduction prior to the detection analysis, rather than relying on EOF truncation.
The structure of this paper is as follows. Section 2 describes the observational datasets and the output from the model simulations. The detection methods and data processing procedures are described in section 3. The main results are presented in section 4. The discussion and conclusions are provided in section 5.
2. Data
a. Study region and extreme temperature indices
In this study, detection and attribution are conducted for the globe (GLB) and five continental regions defined by Giorgi and Francisco (2000), specifically Europe (EUR; 30°–75°N, 10°W–40°E), Asia (ASI; 20°–70°N, 40°–180°E), Australia (AUS; 45°–11°S, 110°–155°E), North America (NAM; 25°–72°N, 170°–50°W), and South America (SAM; 56°S–25°N, 116°–40°W) (boxes in Fig. 1). China (CHI) is also used as a study region, since we have a strong interest in changes in extreme temperatures in that country.
Geographical distributions of the (left) observed and also the multimodel mean simulated trends under the (middle) ALL and (right) NAT forcings for four extreme temperature indices during 1961–2010. The trends were computed for grid cells with at least 40 years of data available. The black dots in the figure indicate land grid boxes where observations are not available.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
The ETCCDI has defined a set of core extreme indices, including percentile-based, absolute, fixed threshold, and duration indices, which can be used to assess different aspects of changes in climate extremes (e.g., Alexander et al. 2006; Zhang et al. 2011). In this study, we focus on the fixed threshold indices of extreme temperature, specifically the annual number of days when the daily maximum temperature is less than 0°C [ice days (ID)], the annual number of days when the daily minimum temperature is less than 0°C [frost days (FD)], the annual number of days when the daily maximum temperature exceeds 25°C [summer days (SU)], and the annual number of days when the daily minimum temperature exceeds 20°C [tropical nights (TR)]. These four fixed threshold indices are calculated for observations and models using the standard code provided by ETCCDI for generating extreme indices (Zhang et al. 2011).
b. Observations and model simulations
Daily observational data from HadEX2 and observations made at stations within China are used to calculate the extreme temperature indices. HadEX2 is the second generation of the global land extreme indices dataset developed by Donat et al. (2013) (https://www.climdex.org/gewocs.html). It is based on high-quality observations from over 7000 stations distributed over the global land area with the angular distance weighting interpolation method (Shepard 1968) applied to grid the station data into 3.75° longitude × 2.5° latitude grid box values. The data cover the period of 1901–2010; however, we only use the information after 1961, given the good-quality, better, and more consistent data density/sampling of the data in that period. The extreme indices calculated from these data within grid boxes are used for the global and continental analyses. For the analyses in China, both the HadEX2 data and the observational data from 2419 stations within China are used for the detection analyses, and the results are compared. The data from stations within China have been homogenized (Cao et al. 2016) using the RHtest V3 software package (Wang and Feng 2010; available at http://etccdi.pacificclimate.org/software.shtml). It is a set of high-quality homogenized data that has undergone strict quality control and has been widely used operationally and in scientific research (e.g., Sun et al. 2016). The values of the extreme indices are first calculated at each station, and the 1961–90 mean is then removed from the individual station series. All of the index data are then averaged onto the 3.75° longitude × 2.5° latitude grid used in HadEX2.
The model data include daily outputs from the CMIP5 model experiments performed under combined historical anthropogenic and natural (ALL) forcings, natural-only (NAT) forcing, greenhouse gas (GHG) forcing, and control experiments. The extreme indices calculated by Sillmann et al. (2013a,b) for the twentieth and the twenty-first centuries—that is, the historical experiments and the representative concentration pathway (RCP) experiments, respectively—are used in this study. The index data that are not available from this earlier study are calculated using the same calculation code for generating extreme indices (Zhang et al. 2011). In total, the ALL forcing experiments include 82 historical simulations performed using 18 climate models for which at least 3 simulations are available. The NAT and GHG forcing experiments include 26 historical simulations from 6 models and 23 simulations from 5 models, respectively. Table 1 lists the number of model simulations that are available for the different forcing experiments used in this study. Because some of the ALL forcing simulations end in 2005, the RCP4.5 simulations for the period of 2006–10 are used to extend the ALL forcing simulations since the differences among the different RCP scenarios during the early twenty-first century are negligible. The response to the anthropogenic forcing (ANT) is estimated from the differences between the model responses to the ALL and NAT forcings. Some studies have shown that temperature responses to different components of external forcings are generally linearly additive at the global and continental scales (Gillett et al. 2004; Shiogama et al. 2013; Marvel et al. 2015). All of the simulations used in this study cover the period from 1961 to 2010.
List of multimodel simulations used in this study. Numbers represent the sizes (numbers of simulations) of the ALL, GHG, and NAT ensembles or the number of 50-yr chunks for the CTL simulations. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)
All of the available model data are regridded using the same 3.75° × 2.5° grid as the HadEX2 dataset. The data are also masked according to the availability of the gridded observed data. Global and regional averages are then calculated to obtain mean time series of the annual temperature indices. The multimodel ensemble mean is computed by first calculating ensemble means from the individual model runs and then averaging over all the available models. The linear trends of observation and multimodel ensemble means are calculated using Kendall’s tau method (Sen 1968) because some of the threshold index data in some of the regions are zero.
3. Methods and data processing
a. Detection methods
The observed extreme temperature indices are compared with the model simulations using a standard optimal fingerprinting method (Allen and Stott 2003). The optimal fingerprinting method expresses the observed changes (Y) as the sum of signals (X) and internal unforced variability (ε); that is, Y = (X − v)β + ε. Here, the regression coefficients (or scaling factors) β that provide the best match between the model responses and the observations are determined using the total least squares method (Ribes and Terray 2013). The parameter v indicates the noise in the signal. The regression residual ε represents the internal unforced variability. If the 90% confidence interval of the scaling factor is above zero, the signals of the external forcings are considered to be detected in the observations. If the 90% ranges of the scaling factors also include unity, we conclude that the observed changes are consistent with the response of the model simulations to external forcing.
We conduct both single-signal and two-signal analyses to detect the influence of the different forcings on the changes in extreme temperature. In the single-signal analysis, the observations are regressed onto the multimodel mean responses to the individual forcings (ALL, ANT, GHG, and NAT). In the two-signal analysis, the observations are regressed onto the ALL and NAT responses simultaneously, and linear transformations are applied to the ALL and NAT scaling factors to examine the relative contributions of the NAT and ANT forcings to the observations (Tett et al. 2002; Jones et al. 2013).
b. Data processing
We conduct detection analyses on the four extreme indices, FD, ID, SU, and TR, based on their spatial and temporal evolution. The anomalies of each index are first computed relative to the 1961–90 mean in each grid box. The regional means for the individual model runs are calculated, and the multimodel ensemble mean is subsequently obtained for each region. The detection procedure is applied to the nonoverlapping 3- and 5-yr mean series that cover 1961–2010 for all of the continents and China. The use of 5- and 3-yr mean series reduces the temporal dimension but retains the influence of important natural forcings, such as volcanic eruptions.
To estimate the internal variability, the preindustrial control (CTL) simulations that lack external forcing and performed using 28 models, are split into 50-yr segments. Together with the within-ensemble difference (i.e., the residuals of the ensemble simulations after the ensemble mean has been removed), there are 284 fifty-year chunks in total. These chunks are used to estimate the scaling factor and the 90% confidence interval, as well as to test the consistency of the residuals (Ribes and Terray 2013).
4. Results
a. Observed and modeled trends
Figure 1 displays the linear trends of the four fixed threshold indices determined from the observations and the model simulations under ALL and NAT forcing. The observational data (left column) show that the cold extreme indices (FD and ID) predominantly display decreasing trends, whereas the warm extreme indices (SU and TR) generally display increasing trends. Observed trends in these indices are significant at the 5% level for most of Northern Hemisphere land grids where the indices can be computed. For the cold extreme indices, the largest negative trends are observed over the Eurasian continent at middle to high latitudes. The decreases in FD and ID in North America are smaller than those in other areas at the same latitudes, especially for ID. A positive trend is seen in the Arctic that may be related to the sparsity of data in this region and may not be real. The very small trends in the cold extremes at the low latitudes are related to the small or zero values of these indices in these areas and its trend may not be real. For the warm extreme indices, increasing trends appear almost everywhere over the global land area, except that a cooling trend is seen over the North and South Americas for SU. This warming hole in the warm extremes of North America has also been found by other studies (e.g., Sillmann et al. 2013a; Vose et al. 2017). Large positive trends are generally observed in Europe and Asia, especially in East Asia and the region of Siberia. Tropical nights appear only at middle to low latitudes, and larger trends are seen at the low latitudes than at the middle latitudes.
The multimodel simulations performed using the ALL and NAT forcings (Fig. 1, middle and right columns) show spatially smoothed distributions compared with the observations because a substantial amount of the internal variability has been removed for the model ensemble mean. The ALL results generally exhibit quite good consistency with the observations, whereas the NAT results do not reproduce most of the observed changes. For the cold extreme indices, the model simulations under ALL forcing reproduce the large negative trends in western Europe but underestimate them in Asia. The trends in ID in North America show much larger decrease in the models than in the observations. The models also simulate the small changes seen at low latitudes reasonably well. The NAT results show only very small changes in the cold extremes compared with the observations. For the warm extreme indices, the models largely reproduce the observations in western Europe, South Asia, Africa, and Australia. However, the models underestimate the large positive trends in SU in East Asia and Siberia. The models also cannot simulate the observed negative trends in SU in North and South America and overestimate the positive trends in TR in the Americas. The NAT results show much smaller changes compared with the observations in most areas.
Figure 2 shows the global 5-yr mean anomalies of the four fixed threshold indices from observations (HadEX2) and the model simulations under the ALL, GHG, and NAT forcings. The long-term changes in multimodel responses to the ALL forcing generally exhibit good correspondence with the observations. The observations all fall well within the spread of the multimodel ensemble. For the cold extremes (FD and ID), the ALL results evolve similarly to the observations. The simulated increases in the cold extremes around the early 1990s may be related to volcanic eruptions, but this increase is not clearly seen in the 5-yr mean series for observations (although visible in the annual series). The GHG results generally display larger trends, and the NAT simulations show very small trends compared with the observed changes. For the warm extremes (TR and SU), the ALL runs slightly overestimate the observed changes after the 1990s. When compared with the observations, the GHG results show larger changes after the late 1980s but smaller changes before the early 1960s. Although the NAT results exhibit poor agreement with the observed warm extremes, they simulate distinct changes after volcanic eruptions, such as the cooling that followed the Mt. Pinatubo eruption in approximately 1991. Volcanic eruptions appear to have a smaller effect on the cold extremes than the warm extremes in the observations.
(left) Global mean 5-yr mean anomalies (relative to the 1961–90 average) and (right) trends of four fixed threshold indices from the HadEX2 observational dataset (black) and the multimodel response to the ALL (red), GHG (green), and NAT (blue) forcings, respectively. Shading indicates the 5%–95% ranges of all of the individual simulations. Gray error bars (right) show the 5%–95% confidence intervals of the linear trends.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
For the other continents (not shown), the ALL simulations largely display quite good consistency with the observations for the cold extremes. Very small values can be seen in both the observations and the models at the low latitudes. The GHG results display less severe cold extremes than the ALL results, whereas the NAT results are not consistent with the observations. For the warm extremes, the ALL results reproduce most of the observed changes, except in the Americas. Previous studies (e.g., Knutson et al. 2017) have shown that the observed trends in some extreme temperatures in parts of the Americas are inconsistent with CMIP5 historical runs driven by the ALL forcing; the reason for this discrepancy is unknown, but it may be due to the effects of anthropogenic aerosols, atmospheric circulation, internal climate variability, and changes in land use, among other factors. This study shows that the warming hole also cannot be simulated by the changes in the warm fixed threshold indices. In summary, all of these results confirm and support the findings from Fig. 1. In addition, we compare the 5-yr anomaly time series in China based on HadEX2 and the homogenized within-China station data (Fig. 3) and find that these two different datasets display very similar changes. This result indicates that these two observational datasets are consistent and are of quite good quality to be used for detection studies.
China mean 5-yr mean anomalies (relative to the 1961–90 average) from two observation datasets CHI (black) and CHI-H (purple) and the multimodel response to the ALL (red), GHG (green), and NAT (blue) forcings, respectively. Shading indicates the 5%–95% ranges of all of the individual simulations. CHI is calculated based on data from more than 2000 stations within China that were obtained from the National Meteorological Information Center (NMIC) of China, whereas CHI-H is based on the HadEX2 dataset.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
b. Detection results
The spatial–temporal evolution of the observations and model simulations under the ALL, ANT, GHG, and NAT forcings are compared using the optimal fingerprinting method. Figure 4 shows the scaling factors and their 90% confidence intervals obtained using single-signal detection. For the cold extremes, the ALL signal can be clearly detected on the global scale, over all of the continents analyzed, and within China. The residual consistency test is generally passed; however, the models show greater (black downward triangles in Fig. 4) and smaller variability (blue upward triangles) than the observations in NAM and ASI, implying that the uncertainty range of the scaling factors can be conservative or overestimated, respectively. The best estimates of the scaling factors are close to unity over the globe and EUR, suggesting a good consistency between observations and model simulations, while it is greater than unity in Asia and China and smaller than 1 in NAM, indicating an underestimate of model simulations in Asia and an overestimate for the models in NAM when compared with the observations, respectively. The ANT and GHG detection results are similar to those for the ALL signals, although their confidence intervals are slightly larger than ALL. NAT can be detected in most areas, but its scaling factors have large values and confidence intervals compared with other signals, indicating that the modeled NAT response is much smaller than the observed changes. For the warm extremes, the ALL signals cannot be detected in the changes in SU in NAM and SAM. For the other areas, the ALL signals can be detected, and the residual variability is consistent with or larger than the observations. The ANT and GHG results are generally similar to those obtained using ALL, although the residual consistency test shows slightly different results in a few regions. The NAT results show quite large values in most regions and cannot be detected in a few regions.
Best estimates of scaling factors and their 5%–95% confidence intervals from single-signal analyses from 1961 to 2010 for the globe (GLB), Asia (ASI), Europe (EUR), North America (NAM), South America (SAM), Australia (AUS), and China (CHI). See the caption of Fig. 3 for information on CHI and CHI-H data. The black downward triangles indicate that the variability in the model simulations is overestimated, according to the residual consistency test, whereas the blue upward triangles indicate small variability in the model simulations.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
To simultaneously assess the separate effects of ANT and NAT, we carry out a two-signal analysis; the results are shown in Fig. 5. For the cold extremes, the ANT forcing can be detected, whereas NAT cannot be detected, on the global scale and over all of the continents. This result indicates the dominant role of the ANT forcing in producing changes in the cold extremes. The residual variability of the models is consistent with or larger than that of the observations, reflecting good detectability for the cold extremes. For the warm extremes, the ANT can be detected on the global scale and over all of the continents in TR, but it is not detected in SU over NAM and SAM. NAT cannot be detected in most regions. All of these results show the strong effects of the human influence on the fixed threshold indices and confirm the robustness of the detection results obtained using single-signal detection. We also note that very similar detection results are obtained in China using the two different datasets. This result suggests a robust influence of anthropogenic forcing on the fixed threshold indices of extreme temperatures in China.
As in Fig. 4, but for two-signal analyses.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
We further estimate the attributable contributions from the ALL, ANT, and NAT signals to the long-term trends of the observed indices from 1961 to 2010 for the globe, the continents and China (Fig. 6). These are estimated as the linear trends of model-simulated signals multiplied by corresponding scaling factors along with their 90% confidence intervals, which are the part of the observed trend explained by model simulations from different forcings. The best estimates of the ANT contributions to the observed trends in the four temperature indices exceed 90% for most of the regions where the ALL signal can be detected. The ANT attributable contributions for FD, ID, SU, and TR at the global scale are 11.5 days, 5.5 days, 8.0 days, and 9.6 days, and the corresponding 90% ranges are 9.0–14.2 days, 3.5–7.7 days, 5.7–10.4 days, and 7.4–11.9 days, respectively. The NAT contribution is less than 10% in most of the regions.
The warming attributable to the ALL, ANT, and NAT forcings of the observed extreme temperature indices (OBS) and its 5%–95% confidence interval. The attributable contribution is estimated by multiplying the linear least squares trends in the relevant time series by the corresponding scaling factors.
Citation: Journal of Climate 31, 16; 10.1175/JCLI-D-17-0853.1
We conduct a series of sensitivity tests to examine the robustness of our detection results. We first evaluate the effects of different methods for reducing the temporal dimension on the detection results, including the use of 3-yr and 5-yr mean nonoverlapping series in the analyses. The scaling factors are quite similar for the two methods, reflecting the robustness of the detection results. We also found the order of operations for estimating trends from multimodel ensembles has little effect: the trend based on multimodel average of the indices is similar to the ensemble mean of trends computed from individual models. We then evaluate the influence of different global averages on the global detection results. Specifically, we consider the global average of the indices over all of the continents and the global average considering only the regions where the fixed threshold indices have observed values; for example, FD and ID are seen only at middle and high latitudes, whereas SU and TR occur only at low latitudes. We find the detection results are similar when different global averages are considered. We also test the influence of the different observational datasets on the detection results in China and obtain robust conclusions about the human influence, as indicated in the previous paragraphs.
5. Discussion and conclusions
Understanding the causes of observed climate changes and the underlying mechanisms provides an important basis for properly projecting future climate conditions and for making decisions to adapt to climate change. This study focuses on the detection of the fixed threshold indices of extreme temperature at the global scale, as well as over five of the continents and China, using the results of a multimodel ensemble for the first time. The results show that the CMIP5 multimodel simulations generally reproduce the observed changes in the fixed threshold indices of extreme temperature, both at the global scale and over most of the continents. The indices related to the daily maximum temperature in the Americas show some degree of inconsistency with the observations; the reasons for this inconsistency are unknown. The detection analyses show that an anthropogenic signal can be identified for most of the fixed threshold indices, whereas the natural signal cannot be detected. The models generally underestimate the changes in the cold fixed threshold indices but show different features in reproducing the warm fixed threshold indices for different continents. The ANT signal cannot be detected in the changes in summer days in North and South America, which lends additional support to the conclusion from the Fourth National Climate Assessment in the United States (Knutson et al. 2017) that the anthropogenic forcing in temperature extremes in the United States can be detected with only medium confidence. In addition, our study show that the human influence on extreme temperature changes in China is clear and robust based on two different observational datasets. Together with other previous studies (Lu et al. 2016; Yin et al. 2017), our results indicate that the anthropogenic forcing has affected various aspects of extreme temperatures in China.
Compared with previous findings by Christidis and Stott (2016), this study provides new insights into the anthropogenic influence on the fixed threshold indices at the global scale and continental scales using multimodel results. The human influence is clearly seen in the changes in FD and ID, which may be due to the use of a multimodel ensemble and/or the use of an improved detection method that does not employ EOF truncation and includes a different regional averaging method. The detection over five continents and China provides new evidence of human influence on the fixed threshold indices changes at the continental scale and over China. However, more studies will be conducted in the near future to identify the reasons why the anthropogenic signal is not detected in the changes of summer days over the Americas.
Acknowledgments
We thank three anonymous reviewers for their helpful comments. We thank Xuebin Zhang for his constructive suggestions. We acknowledge the Program for Climate Model Diagnosis and Intercomparison and the Working Group on Coupled Modeling of the World Climate Research Programme (WCRP) for their roles in making the WCRP CMIP multimodel datasets available. This study was supported by the National Key R&D Program of China (2018YFA0605604), the National Science Foundation of China (41675074), and Climate Project (CCSF201805).
REFERENCES
Alexander, L. V., and Coauthors, 2006: Global observed changes in daily climate extremes of temperature and precipitation. J. Geophys. Res., 111, D05109, https://doi.org/10.1029/2005JD006290.
Allen, M. R., and P. A. Stott, 2003: Estimating signal amplitudes in optimal fingerprinting. Part I: Theory. Climate Dyn., 21, 477–491, https://doi.org/10.1007/s00382-003-0313-9.
Cao, L., Y. Zhu, G. Tang, F. Yuan, and Z. Yan, 2016: Climatic warming in China according to a homogenized dataset from 2419 stations. Int. J. Climatol., 36, 4384–4392, https://doi.org/10.1002/joc.4639.
Christidis, N., and P. A. Stott, 2016: Attribution analyses of temperature extremes using a set of 16 indices. Wea. Climate Extremes, 14, 24–35, https://doi.org/10.1016/j.wace.2016.10.003.
Christidis, N., P. A. Stott, S. J. Brown, G. C. Hegerl, and J. Caesar, 2005: Detection of changes in temperature extremes during the second half of the 20th century. Geophys. Res. Lett., 32, L20716, https://doi.org/10.1029/2005GL023885.
Christidis, N., P. A. Stott, and S. J. Brown, 2011: The role of human activity in the recent warming of extremely warm daytime temperatures. J. Climate, 24, 1922–1930, https://doi.org/10.1175/2011JCLI4150.1.
Donat, M. G., L. V. Alexander, H. Yang, I. Durre, R. Vose, and J. Caesar, 2013: Global land-based datasets for monitoring climatic extremes. Bull. Amer. Meteor. Soc., 94, 997–1006, https://doi.org/10.1175/BAMS-D-12-00109.1.
Donat, M. G., J. Sillmann, S. Wild, L. V. Alexander, T. Lippmann, and F. W. Zwiers, 2014: Consistency of temperature and precipitation extremes across various global gridded in situ and reanalysis data sets. J. Climate, 27, 5019–5035, https://doi.org/10.1175/JCLI-D-13-00405.1.
Dong, S., Y. Sun, E. Aguilar, X. Zhang, T. C. Peterson, L. Song, and Y. Zhang, 2018: Observed changes in temperature extremes over Asia and their attribution. Climate Dyn., 51, 339–353, https://doi.org/10.1007/s00382-017-3927-z.
Gillett, N. P., M. F. Wehner, S. F. B. Tett, and A. J. Weaver, 2004: Testing the linearity of the response to combined greenhouse gas and sulfate aerosol forcing. Geophys. Res. Lett., 31, L14201, https://doi.org/10.1029/2004GL020111.
Giorgi, F., and R. Francisco, 2000: Evaluating uncertainties in the prediction of regional climate change. Geophys. Res. Lett., 27, 1295–1298, https://doi.org/10.1029/1999GL011016.
IPCC, 2013. Climate Change 2013: The Physical Science Basis. T. F. Stocker et al., Eds., Cambridge University Press, 1535 pp.
Jones, G. S., P. A. Stott, and N. Christidis, 2013: Attribution of observed historical near-surface temperature variations to anthropogenic and natural causes using CMIP5 simulations. J. Geophys. Res. Atmos., 118, 4001–4024, https://doi.org/10.1002/jgrd.50239.
Kim, Y.-H., S.-K. Min, X. B. Zhang, F. Zwiers, L. V. Alexander, M. K. Donat, and Y.-S. Tung, 2016: Attribution of extreme temperature changes during 1951–2010. Climate Dyn., 46, 1769–1782, https://doi.org/10.1007/s00382-015-2674-2.
Knutson, T., J. P. Kossin, C. Mears, J. Perlwitz, and M. F. Wehner, 2017: Detection and attribution of climate change. Climate Science Special Report: Fourth National Climate Assessment, Vol. I, D. J. Wuebbles et al., Eds., U.S. Global Change Research Program, 114–132, https://doi.org/10.7930/J01834ND.
Lewis, S. C., and D. J. Karoly, 2013: Anthropogenic contributions to Australia’s record summer temperatures of 2013. Geophys. Res. Lett., 40, 3705–3709, https://doi.org/10.1002/grl.50673.
Lu, C. H., Y. Sun, H. Wan, X. B. Zhang, and H. Yin, 2016: Anthropogenic influence on the frequency of extreme temperatures in China. Geophys. Res. Lett., 43, 6511–6518, https://doi.org/10.1002/2016GL069296.
Marvel, K., G. A. Schmidt, D. Shindell, C. Bonfils, A. N. LeGrande, L. Nazarenko, and K. Tsigaridis, 2015: Do responses to different anthropogenic forcings add linearly in climate models? Environ. Res. Lett., 10, 104010, https://doi.org/10.1088/1748-9326/10/10/104010.
Min, S.-K., X. Zhang, F. W. Zwiers, H. Shiogama, Y.-S. Tung, and M. Wehner, 2013: Multimodel detection and attribution of extreme temperature changes. J. Climate, 26, 7430–7451, https://doi.org/10.1175/JCLI-D-12-00551.1.
Morak, S., G. C. Hegerl, and N. Christidis, 2013: Detectable changes in the frequency of temperature extremes. J. Climate, 26, 1561–1574, https://doi.org/10.1175/JCLI-D-11-00678.1.
Ribes, A., and L. Terray, 2013: Application of regularised optimal fingerprint to attribution. Part II: Application to global near-surface temperature. Climate Dyn., 41, 2837–2853, https://doi.org/10.1007/s00382-013-1736-6.
Sen, P. K., 1968: Estimates of the regression coefficient based on Kendall’s tau. J. Amer. Stat. Assoc., 63, 1379–1389, https://doi.org/10.1080/01621459.1968.10480934.
Shepard, D., 1968: A two-dimensional interpolation function for irregularly spaced data. Proc. 23rd ACM Natl. Conf., New York, New York, Association for Computing Machines, 517–524.
Shiogama, H., D. Stone, N. Tatsuya, N. Toru, and E. Seita, 2013: On the linear additivity of climate forcing-response relationships at global and continental scales. Int. J. Climatol., 33, 2542–2550, https://doi.org/10.1002/joc.3607.
Sillmann, J., and E. Roeckner, 2008: Indices for extreme events in projections of anthropogenic climate change. Climatic Change, 86, 83–104, https://doi.org/10.1007/s10584-007-9308-6.
Sillmann, J., V. V. Kharin, X. Zhang, F. W. Zwiers, and D. Bronaugh, 2013a: Climate extremes indices in the CMIP5 multimodel ensemble: Part 1. Model evaluation in the present climate. J. Geophys. Res. Atmos., 118, 1716–1733, https://doi.org/10.1002/jgrd.50203.
Sillmann, J., V. V. Kharin, F. W. Zwiers, X. Zhang, and D. Bronaugh, 2013b: Climate extremes indices in the CMIP5 multimodel ensemble: Part 2. Future climate projections. J. Geophys. Res. Atmos., 118, 2473–2493, https://doi.org/10.1002/jgrd.50188.
Stott, P. A., G. S. Jones, N. Christidis, F. W. Zwiers, G. Hegerl, and H. Shiogama, 2011: Single-step attribution of increasing frequencies of very warm regional temperatures to human influence. Atmos. Sci. Lett., 12, 220–227, https://doi.org/10.1002/asl.315.
Sun, Y., X. B. Zhang, F. W. Zwiers, L. C. Song, H. Wan, T. Hu, H. Yin, and G. Ren, 2014: Rapid increase in the risk of extreme summer heat in eastern China. Nat. Climate Change, 4, 1082–1085, https://doi.org/10.1038/nclimate2410.
Sun, Y., L. C. Song, H. Yin, B. T. Zhou, T. Hu, X. B. Zhang, and P. Stott, 2016: Human Influence on the 2015 extreme high temperature events in western China. Bull. Amer. Meteor. Soc., 97, S102–S106, https://doi.org/10.1175/BAMS-D-16-0158.1.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, https://doi.org/10.1175/BAMS-D-11-00094.1.
Terando, A., K. Keller, and W. E. Easterling, 2012: Probabilistic projections of agro-climate indices in North America. J. Geophys. Res., 117, D08115, https://doi.org/10.1029/2012JD017436.
Tett, S. F. B., and Coauthors, 2002: Estimation of natural and anthropogenic contributions to twentieth century temperature change. J. Geophys. Res., 107, 4306, https://doi.org/10.1029/2000JD000028.
Vose, R. S., D. R. Easterling, K. E. Kunkel, A. N. LeGrande, and M. F. Wehner, 2017: Temperature changes in the United States. Climate Science Special Report: Fourth National Climate Assessment, Vol. I, D. J. Wuebbles et al., Eds., U.S. Global Change Research Program, 185–206, https://doi.org/10.7930/J0N29V45.
Wang, X. L., and Y. Feng, 2010: RHtestsV3 User Manual. Climate Research Division, Atmospheric Science and Technology Directorate, Science and Technology Branch, Environment Canada, 27 pp.
Wen, H. Q., X. Zhang, Y. Xu, and B. Wang, 2013: Detecting human influence on extreme temperatures in China. Geophys. Res. Lett., 40, 1171–1176, https://doi.org/10.1002/grl.50285.
Yin, H., Y. Sun, H. Wan, X. Zhang, and C. Lu, 2017: Detection of anthropogenic influences on the intensity of extreme temperatures in China. Int. J. Climatol., 37, 1229–1237, https://doi.org/10.1002/joc.4771.
Zhang, X. B., L. Alexander, G. Hegerl, P. Jones, A. Klein Tank, T. C. Peterson, B. Trewin, and F. W. Zwiers, 2011: Indices for monitoring changes in extremes based on daily temperature and precipitation data. Wiley Interdiscip. Rev.: Climate Change, 2, 851–870, https://doi.org/10.1002/wcc.147.
Zhou, B. T., Y. Xu, J. Wu, S. Y. Dong, and Y. Shi, 2016: Changes in temperature and precipitation extreme indices over China: Analysis of a high-resolution grid dataset. Int. J. Climatol., 36, 1051–1066, https://doi.org/10.1002/joc.4400.
Zwiers, F. W., X. B. Zhang, and Y. Feng, 2011: Anthropogenic influence on long return period daily temperature extremes at regional scales. J. Climate, 24, 881–892, https://doi.org/10.1175/2010JCLI3908.1.
