GCMs-based spatiotemporal evolution of climate extremes during the 21st century in China

Authors

  • Jianfeng Li,

    1. Department of Geography and Resource Management, The Chinese University of Hong Kong, Hong Kong, China
    Search for more papers by this author
  • Qiang Zhang,

    Corresponding author
    1. Department of Water Resources and Environment, Sun Yat-sen University, Guangzhou, China
    2. Key Laboratory of Water Cycle and Water Security in Southern China of Guangdong High Education Institute, Sun Yat-sen University, Guangzhou, China
    3. School of Geography and Planning, and Guangdong Key Laboratory for Urbanization and Geo-simulation, Sun Yat-sen University, Guangzhou, China
    • Corresponding author: Q. Zhang, Ph.D. Professor, Department of Water Resources and Environment, Sun Yat-sen University, Guangzhou 510275, China. (zhangq68@mail.sysu.edu.cn)

    Search for more papers by this author
  • Yongqin David Chen,

    1. Department of Geography and Resource Management, The Chinese University of Hong Kong, Hong Kong, China
    Search for more papers by this author
  • Vijay P. Singh

    1. Department of Biological and Agricultural Engineering and Department of Civil and Environmental Engineering, Texas A & M University, College Station, Texas, USA
    Search for more papers by this author

Abstract

[1] Changes in the hydrological cycle being caused by human-induced global warming are triggering variations in observed spatiotemporal distributions of precipitation and temperature extremes, and hence in droughts and floods across China. Evaluation of future climate extremes based on General Circulation Models (GCMs) outputs will be of great importance in scientific management of water resources and agricultural activities. In this study, five precipitation extreme and five temperature extreme indices are defined. This study analyzes daily precipitation and temperature data for 1960–2005 from 529 stations in China and outputs of GCMs from the Coupled Model Intercomparison Project Phase 3 (CMIP3) and Phase 5 (CMIP5). Downscaling methods, based on QQ-plot and transfer functions, are used to downscale GCMs outputs to the site scale. Performances of GCMs in simulating climate extremes were evaluated using the Taylor diagram. Results showed that: (1) the multimodel CMIP5 ensemble performs the best in simulating observed extreme conditions; (2) precipitation processes are intensifying with increased frequency and intensity across entire China. The southwest China, however, is dominated by lengthening maximum consecutive dry days and also more heavy precipitation extremes; (3) warming processes continue with increasing warm nights, decreasing frost days, and lengthening heat waves during the 21st century; (4) changes in precipitation and temperature extremes exhibit larger changing magnitudes under RCP85 scenario; (5) for the evolution of changes in extremes, in most cases, the spatial pattern keeps the same, even though changing rates vary. In some cases, area with specific changing properties extends or shrinks gradually. The directions of trends may alter during the evolution; and (6) changes under RCP85 become more and more pronounced as time elapses. Under the peak-and-decline RCP26, changes in some cases do not decrease correspondingly during 2070–2099 even though the radiative forcing during 2070–2099 is less than during 2040–2069. The increase of radiative forcing triggers considerable regional variations in consecutive dry days, but causes only slight changes in the areal average in China. The results of this study imply higher flood risk across entire China but intensifying droughts in south China in the 21st century, and also more heat-related losses in east coasts of China.

1 Introduction

[2] The IPCC 4th assessment report predicts the temperature rises of 1.1–6.4°C by 2100 [IPCC, 2007]. Human-induced global warming is accelerating the global hydrological cycle [Allen and Ingram, 2002; Alan et al., 2003; Ziegler et al., 2003]. The acceleration of the hydrological cycle is altering spatiotemporal patterns of precipitation and which has the potential to trigger increased occurrences of climate extremes and subsequently increased frequency and intensity of floods and droughts in many regions of the world [e.g., Easterling et al., 2000; Mirza, 2002; Dore, 2005]. Climatic and hydrological changes, as a result of increased emission of greenhouse gases and related impacts on human society and ecological environment, have been widely discussed in recent years [Houghton et al., 2001; Oreskes, 2004].

[3] Public awareness of extreme events has risen sharply in recent years partly because of increasing catastrophic influences of natural hazards, such as floods, typhoon, and the sea level rise [Beniston and Stephenson, 2004; Zolina et al., 2004; Xu et al., 2009; Zhang et al., 2011, 2012c]. There are a multitude of publications addressing observed changes and behaviors of extreme precipitation and temperature around the world [e.g., Kunkel, 2003; Wang and Zhou, 2005; Alexander et al., 2006; Zhang et al., 2012b, 2012c]. These studies are greatly helpful in understanding changing properties, physical mechanisms, and possible implications of climate extremes at different spatial and temporal scales. The IPCC 4th assessment report indicated that the type, frequency, and intensity of extreme events are expected to change in future, even with relatively small mean climate changes [IPCC, 2007; Dankers and Hiederer, 2008]. In recent years, additional investigations about projections of future climate changes based on General Circulation Models (GCMs) were carried out. Yin [2005] analyzed an ensemble of 15 GCMs simulations, and reported that a consistent poleward and upward shift and intensification of storm tracks were found by the end of the 21st century. Evans [2009] used 18 GCMs to analyze the climate change in the Middle East during 21st century. According to Evans [2009], the overall temperature increases by 1.4°C in the midcentury, and by 4°C by late century under A2 scenario. A decrease in precipitation was found out in Eastern Mediterranean, Turkey, Syria, Northern Iraq, Northeastern Iran, and the Caucasus. Li et al. [2006] found out different changes of rainfall among 11 GCMs over the Amazon under the A1B scenario. They found 5 of 11 models predicted an increase of annual rainfall, while another three models predicted a decrease. Evaluation and investigation of future changing properties of climate extremes are of great significance in the planning and scientific management of water resources and agricultural activities.

[4] China is a country with the largest population in the world and also is a country heavily dependent on agricultural production [e.g., Zhang et al., 2012d]. Changes in precipitation and temperature extremes bring about uncertainties to food system by influencing food availability, food access, food utilization, and stability of food system [Wheeler and von Braun, 2013]. From this perspective, China is vulnerable to precipitation and temperature extremes. In China, there have been many investigations pertaining to spatiotemporal patterns of precipitation and temperature extremes at different spatiotemporal scales [e.g., Yang and Lau, 2004; Zhai et al., 2005; Zhang et al., 2012a]. In recent years, prediction and evaluation of possible changes of extreme precipitation and temperature in China by diverse techniques and models have drawn considerable concerns from hydrologists and meteorologists. Zhang et al. [2006] studied the distribution of extreme climate events over China under the IPCC B2 scenario for time intervals of 1961–1990 and 2071–2100 by a regional climate model system — PRECIS, and concluded that extreme precipitation events would occur more frequently over China. Xu et al. [2006] also used PRECIS to study changes in temperature and precipitation, including extreme maximum/minimum temperature and precipitation, during 2071–2100 under A2 and B2 scenarios, and their results showed that extreme maximum temperature and precipitation would increase, while the extreme minimum temperature would decrease under B2 scenario. Li et al. [2011a] analyzed possible changes of extreme precipitation in July–August in China under the CO2 doubling condition by using outputs of GCMs, and found out that the increase of temperature and water vapor content contributes to the increase of precipitation extreme over contiguous China. Li et al. [2012] applied outputs of GCMs and downscaling technique to assess variations in extreme precipitation and temperature on the Loess Plateau of China, and concluded that more frequent and intense precipitation, longer drought duration, less cold extremes, longer heat wave duration, and growing season length would happen during the 21st century. Jiang et al. [2012] investigated extreme climate events in China over 2071–2100 under A1B scenario based on the IPCC-AR4 multimodel ensemble, and indicated that more frequent and intense extreme precipitation would be expected by the end of the 21st century, and less maximum number of consecutive dry days would be found in northeast and northwest China.

[5] Recently, the availability of outputs of the Coupled Model Intercomparison Project Phase 5 (CMIP5) provides simulations from more sophisticated and higher spatial resolution models [Taylor et al., 2012]. The Representative Concentration Pathways (RCPs) are a new generation of scenarios in CMIP5 [Moss et al., 2010]. Due to improvements in models and scenarios, the new CMIP5 simulations may shed more insights to climate changes. Sillmann et al. [2013b] made an overview of projections of changes in climate extremes over the 21st century at the global scale based on CMIP5, and made an indirect comparison between the simulations from CMIP3 (the Coupled Model Intercomparison Project Phase 5) and CMIP5.

[6] However, reported studies were mainly concerned with projection and evaluation of precipitation or temperature extremes for a specific region based on one or two specific periods during the 21st century (usually the end of the 21st century). In fact, climate change is a gradual process. The spatiotemporal evolutions over the whole 21st century are very important and help scientists and governments to make different measures to face possible risks during different periods. Considering the improvements in CMIP5 and the importance of the spatiotemporal evolution of extremes in China, the current study analyzes the spatiotemporal evolution of precipitation and temperature extremes over China in the whole 21st century under RCP26 and RCP85 scenarios by using outputs of GCMs from CMIP5. Then, implications of the variations of extreme precipitation and temperature for China are discussed. This study can help to develop a comprehensive understanding of how extreme events may change and what challenges China would face in terms of water resources management and planning of agricultural development under influences of climate extremes during the 21st century. In this case, the major objective is to present a relatively clear picture concerning variations of climate extremes, e.g., precipitation and temperature extremes, in both space and time across entire China and also regional responses of China to global climate changes in the 21st century with respect to climate extremes.

2 Data

[7] Daily precipitation and temperature data covering the period of 1960–2005 from 529 stations in China provided by the National Climate Center of China Meteorological Administration were analyzed. Locations of these stations are shown in Figure 1. Outputs of GCMs provided by the World Climate Research Program's (WCRP's) CMIP5 and CMIP3 multimodel data sets are used (Tables 1 and 2) [Meehl et al., 2007; Taylor et al., 2012]. Only GCMs from CMIP5 with outputs covering the whole 2006–2099 under RCP26 and RCP85 are chosen in this study. Some institutes release more than one model. In this case, only one model of that institute is selected. Following this rule, 13 models from CMIP5 are used in this study. The CMIP3 models whose outputs provide extreme indices series directly were used [Xu et al., 2009; Li et al., 2011b]. Because CMIP3 models are used in order to compare their differences from CMIP5, only the CMIP3 models that have corresponding CMIP5 models (i.e., released by the same institute) under RCP45 are chosen. For outputs from CMIP3, 20C3M is a historical scenario assuming that greenhouse gases increase at the same level as observed through the 20th century [Zhang, 2005; Li et al., 2012]. B1 scenario is a future scenario in which the CO2 concentrations reach about 550 ppm by 2100. CMIP5 employs a new set of emission scenarios called RCPs, which are not correlated with predefined storylines like scenarios in CMIP3 [Moss et al., 2010]. The outputs from CMIP5 under RCP26, RCP45, and RCP85 are applied (Figure 2). RCP26 is a peak-and-decline scenario in which the 2°C global average warming target is satisfied, and the radiative forcing increases to about 3 W/m2 before 2100 and then decreases to 2.6 W/m2 by 2100 [van Vuuren et al., 2011]. In RCP45, the radiative forcing reaches about 4.5 W/m2 by the end of the 21st century, and it has comparable CO2 concentrations and median temperature increase as B1 scenario [Rogelj et al., 2012]. The radiative forcing in RCP85 increases the fastest, and reaches a level of 8.5 W/m2 by 2100 [Riahi et al., 2011]. The multimodel ensemble simulations are expected to provide more robust results and less uncertainties [Sillmann et al., 2013b]. In this study, the CMIP3 and CMIP5 ensembles are calculated after downscaling. Additionally, the results of Sillmann et al. [2013a, 2013b] from ETCCDI indices archive (EIA) hosted by the Canadian Centre for Climate Modelling and Analysis (www.cccma.ec.gc.ca/data/climdex/ climdex.shtml) are also used to investigate differences in climate extremes under RCP45 and B1 scenarios.

Figure 1.

The meteorological stations used in this study. The number denotes river basin: 1, Songhuajiang River; 2, Liaohe River; 3, Haihe River; 4, Yellow River; 5, Huaihe River; 6, Yangtze River; 7, southeast rivers; 8, Pearl River; 9, southwest rivers; and 10, northwest rivers. The black curve inside the domain denotes the boundary of the water basin.

Table 1. Details of Outputs of GCMs From CMIP5
GCMModeling CenterResolution (Lon × Lat)Data Duration
HistoricalRCP26RCP45RCP85
BCC-CSM1.1(m)BCC320 × 1601850–20122006–21002006–21002006–2100
BNU-ESMGCESS128 × 641950–20052006–21002006–21002006–2100
CanESM2CCCma128 × 641850–20052006–23002006–23002006–2100
CCSM4NCAR288 × 1921850–20052006–23002006–22992006–2300
CESM1(CAM5)NSF-DOE-NCAR(1)288 × 1921850–20052006–21002006–21002006–2100
CNRM-CM5CNRM-CERFACS256 × 1281850–20052006–21002006–21002006–2100
CSIRO-Mk3.6.0CSIRO-QCCCE192 × 961850–20052006–21002006–23002006–2300
GFDL-CM3NOAA GFDL144 × 901860–20052006–2100/2006–2100
IPSL-CM5A-MRIPSL144 × 1431850–20052006–21002006–23002006–2100
MIROC5TUT NIES JAMEST256 × 1281850–20122006–21002006–21002006–2100
MPI-ESM-MRMPI-M192 × 961850–20052006–21002006–21002006–2100
MRI-CGCM3MRI320 × 1601850–20052006–21002006–21002006–2100
NorESM1-MNCC144 × 961850–20052006–21002006–23002006–2100
Table 2. Details of Outputs of GCMs From CMIP3
GCMModeling CenterResolution (Lon × Lat)Data Duration
20C3MB1
CNRM-CM3CNRM128 × 641860–19992000–2299
IPSL-CM4IPSL96 × 721860–20002000–2300
MIROC3.2_MTUT NIES JAMEST128 × 641850–20002001–2100
PCMNCAR128 × 641890–19992000–2299
Figure 2.

Total radiative forcing (W/m2) of historical, RCP26, RCP45, and RCP85 scenarios. The bars denote the changes in the averages of radiative forcing (%) during 2010–2039, 2040–2069, and 2070–2100 relative to 1960–2005.

[8] Ten extreme indices, widely used in the literature, were applied to describe precipitation and temperature extremes in China [IPCC, 2007; Sillmann and Roeckner, 2008; Zhang et al., 2011; Li et al., 2012]. The details of these extreme indices are displayed in Table 3. These indices were first proposed by Frich et al. [2002]. The consecutive dry days (CDD), number of heavy precipitation days (R10), max 5 day precipitation amount (R5d), annual precipitation fraction due to very wet days (R95T), and simple daily intensity index (SDII) are indices reflecting extreme conditions of precipitation. The warm nights (TN90), frost day (FD), extreme temperature range (ETR), growing season length (GSL), and heat wave duration index (HWDI) indices represent different temperature extremes. In the outputs of CMIP3, extreme indices are provided directly. For CMIP5, extreme indices are not provided in the outputs. In this study, they are calculated based on the gridded daily precipitation and temperature.

Table 3. Definitions of Ten Extreme Indices
IndicesNameDefinitionUnit
CDDConsecutive dry daysMaximum number of consecutive days with daily precipitation < 1 mmday
R10Number of heavy precipitation daysNumber of days with daily precipitation ≥ 10 mmday
R5dMax 5 day precipitation amountMaximum total 5 day precipitationmm
R95TAnnual precipitation fraction due to very wet daysFraction of annual total precipitation due to events exceeding the 95th percentile of wet day (daily precipitation > 1 mm) during 1961–1990%
SDIISimple daily intensity indexAnnual total precipitation divided by the number of wet days (daily precipitation > 1 mm) in the yearmm/day
TN90Warm nightsFraction of time Tmin > the long-term 90th percentile value of daily Tmin during 1961–1990%
FDFrost dayThe number of days with Tmin < 0°Cday
ETRExtreme temperature rangeThe difference between the highest and the lowest temperature of the same calendar year°C
GSLGrowing season lengthPeriod between when Tday >5°C for >5 days and Tday < 5°C for >5 daysday
HWDIHeat wave duration indexMaximum period >5 consecutive days with Tmax > 5°C above the calendar day mean calculated for a 5 day window centered on each calendar day during 1961–1990day

3 Methodologies

3.1 Downscaling Techniques

[9] The outputs of GCMs are spatially downscaled from gridded to site scales. The downscaling method, proposed by Li et al. [2011b] and Zhang [2005], was applied in this study. The first step is to smooth the GCM grid data by calculating the inverse distance weighted average of four neighboring grid boxes. The second step is to develop transfer functions to project the GCMs grid data to the target station.

[10] The inverse distance weights wi can be computed as:

display math(1)

where ε is a very small value to prevent division by zero, and di represents the angular distance between grid box center and target station. The quantity di can be estimated by the following equation:

display math(2)

where dlat and dlong denote the latitude and longitude of a grid box center, respectively; slat and slong denote the latitude and longitude of the target station, respectively.

[11] Then the smoothed GCM data of the target station datasmoothed can be computed as:

display math(3)

where datai denotes the data in a GCM grid box. In order to calculate the smoothed GCM data, the data of four neighboring grids must be available. However, in some cases, especially in the area along the coast, the data of some of neighboring grids may be unavailable. Therefore, in this study, to downscale the GCM outputs when some of neighboring grids are unavailable, only the available grids (the number of available grids must be larger than 1) will be inputted to equation (3) to calculate the smoothed GCM data. For instance, if only two of four neighboring grids are available, the inverse distance weights and the smoothed GCM data will be calculated only by these two grids.

[12] After obtaining the smoothed GCM data, the method proposed by Zhang [2005] was applied to estimate transfer functions. This method generates probability distributions of local observations according to outputs of GCMs, instead of trying to detect strong one-by-one correlations between station values and outputs of GCMs. Thus, this downscaling method cannot reproduce actual values of each year at the site scale.

[13] The smoothed GCMs data and historical observations at the target station in the base climate period were paired according to their ranks or corresponding quartiles. Pairs were plotted to estimate a best fitting nonlinear function and a best fitting linear function. These two best fitting functions are called transfer functions. The nonlinear function is used to transform the GCMs outputs within the range of the nonlinear function. The linear function is applicable for the GCMs outputs out of the range. Finally, means of downscaled values under different periods under various scenarios are calculated.

3.2 Taylor Diagram

[14] Taylor diagrams are used to assess performances of models in simulating historical extremes [Taylor, 2001; Kwok, 2011]. A Taylor diagram provides a concise summary of comparisons between simulations and observation in terms of correlation (R), the root-mean-square difference (E2), and the ratio of variances of the model (σm) and observation (σo). These statistics are associated through the cosines relationship. The root-mean-square difference and the standard deviations of each variable are normalized by the standard deviation of the corresponding observation:

display math(4)

and

display math(5)

Then the polar coordinates inline image and cos−1R of a model can describe a point on the plot. According to the cosines relationship, the corresponding inline image can be obtained. After normalization, the observation will be always plotted at the unit distance from the origin along the abscissa. The inline image is represented by the dashed semicircles centered at the unity on the abscissa. Based on this, relationship between model and observation can be described by the degree to which how close the point representing the model and the point representing the observation are.

4 Results

4.1 Performances of Models

[15] Although previous studies did performance comparisons of different GCMs in extreme events for China [i.e., Jiang et al., 2012], no comparison is yet done for the CMIP5 models, and there is no comparison between the CMIP3 and CMIP5 models in China either. Therefore, before making projections, an assessment of the performance between CMIP3 and CMIP5 in downscaling extreme indices is made.

[16] The comparison aims to evaluate how closely downscaled simulations of considered models, including the CMIP3 ensemble and CMIP5 ensemble, resemble the observation based on Taylor diagrams. Considering the lengths of the 20C3M in CMIP3 and historical scenario in CMIP5, the observation period is divided into two subperiods: the base climate period (1960–1990) and the pseudo future climate period (1991–1999). The base climate period is the one during which downscaling relationships between individual station and corresponding grids of all indices are developed. Based on the downscaling relationship developed from the base climate period, the gridded extreme indices during pseudo future climate period are downscaled to the corresponding site. Then the correlation, root-mean-square difference, and the ratio of variances of the downscaled and observed extreme indices are calculated and then normalized (Figure 3). The case that the downscaling process fails or returns an invalidate value is also considered.

Figure 3.

Comparisons of model simulations and observation during 1991–1999 of (a) CDD, (b) R10, (c) R5d, (d) R95T, (e) SDII, (f) TN90, (g) FD, (h) ETR, (i) GSL, and (j) HWDI in Taylor diagrams.

[17] For CDD, the normalized standard deviations of all models approximate to 1, and the correlations between models and observation are around 0.95, indicating good performances of downscaling CDD of all models (Figure 3a). For both CMIP3 and CMIP5 ensembles, they have the best performances in downscaling CDD compared to other individual models, as their normalized standard deviations are the closest to 1 and the correlations are the highest. Similarly, models have very good performances in downscaling R10 (Figure 3b), R5d (Figure 3c), SDII (Figure 3e), FD (Figure 3g), and ETR (Figure 3h). The CMIP5 models perform generally better than CMIP3 ones. Among considered models, the CMIP5 ensemble has relatively better performances. For GSL, the MIROC5, MPI-ESM-MR, and the CMIP5 ensemble have the best performances, as their normalized standard deviations are close to 1 and their correlations are close to 0.85 – 0.90 (Figure 3i). But for other models, their correlations are relatively low and ranging between 0.70 and 0.80. For HWDI, the models' correlations all range from 0.60 to 0.80, and the normalized standard deviation ranges from 0.80 to 1.30, showing fairly good performances (Figure 3j). Simulations of CMIP5 models are better than the CMIP3, and the CMIP3 and CMIP5 ensembles have the best performance in downscaling HWDI. For R95T and TN90, the performances of all models are not so good, as their correlations are low (Figures 2d and 2f).

[18] According to the above assessment, the CMIP3 and CMIP5 ensembles generally perform better than individual models. Compared to the CMIP3, the CMIP5 simulations can represent better the site-scale extreme indices in China, implying an improvement of the CMIP5 simulations. In most cases, the CMIP5 ensemble has the best performance. CMIP5 ensemble can downscale most extreme indices with acceptable performances, except for R95T and TN90. The R95T and TN90 are indices correlated with the most extreme cases and percentiles (95th percentile precipitation for R95T, and 90th percentile minimum temperature for TN90). Based on this, following analyses are based on the CMIP5 ensemble.

4.2 Variations of Extreme Precipitation Under RCP26 Scenario

[19] The spatiotemporal variations of different extreme precipitation events under RCP26 scenario during 2010–2039, 2040–2069, and 2070–2099 are analyzed (Figures 4, 5 and Figures S1S3). The colors in these figures denote changes in averages of extreme indices during considered periods relative to 1960–2005. Downscaling relationships are trained based on the whole observation period: 1960–2005, and then extreme indices from the CMIP5 ensemble under RCPs are downscaled to the site scale. The relative change in each station is calculated at first. Then the Inverse Distance Weighted technique (IDW) is used to interpolate changing rates across China [De By, 2001; Gemmer et al., 2004].

Figure 4.

The changes (%) in R10 of (a) 2010–2039, (c) 2040–2069, and (e) 2070–2099 under RCP26, as well as of (b) 2010–2039, (d) 2040–2069, and (f) 2070–2099 under RCP85, relative to 1960–2005. The upward triangle denotes the index of that station increases, and the downward triangle denotes the index of that station decreases.

Figure 5.

The same as Figure 3, but for R95T.

[20] During 2010–2039 under RCP26, CDD in the south China increases (Figure S1a). The CDD increases slightly by 0 – 10% compared to 1960–2005, indicating a slight lengthening of maximum consecutive dry days. As for other parts in China, the CDD decreases by 0% – 10%. During 2040–2069, changes in CDD are similar to during 2010–2039, but there are more areas with decreasing CDD (Figures S1a and S1c). During 2070–2099, the increasing CDD occurs mainly in the southwest rivers basin, the southeast China, and the Huaihe River basin, and the increasing rates range during 0% – 10% (Figure S1e). The CDD variations indicate that the south China may become drier during the 21st century in terms of the maximum consecutive dry days, while the occurrence risk of droughts in north China may become smaller.

[21] During 2010–2039 under RCP26, the R10 increases in the west and the north China (Figure 4a). The maximum of the changing percentage is detected mainly in Qaidam Basin in west China, being around 70%, implying more heavy precipitation days in these regions. Decreasing R10 mainly occurs in the Yangtze River with decreasing magnitudes of −10% – 0%. During 2040–2069, R10 in the whole China increases (Figure 4c). In the middle of west China, R10 is 40% – 80% larger than the values during 1960–2005. Increasing rates of R10 in the northeast China raise to 10% – 20% during 2040–2069 from 0% to 10% during 2010–2039 (Figures 4a and 4c). At the same time, the R10 in south China begins to increase. The spatial distribution of R10 variations during 2070–2099 is very similar to that during 2040–2069 (Figure 4e).

[22] During 2010–2039, the increasing R5d is identified in most parts of China except for a small part in southwest river basin (Figure S2a). The relatively large increasing magnitude of R5d, which is about 20%, occurs in northwest rivers basin, the Liaohe River basin and the Haihe River basin, implying the maximum 5 day precipitation amount becomes larger. During 2040–2069, the R5d increases in entire China, and changing rates are larger than those during 2010–2039 (Figures S2a and S2c). The R5d in the majority of China increases by 10% – 20%. The maximum is detected in most western parts of west China, which reaches nearly 40%, showing that the maximum precipitation within 5 days increases by almost 40%, which may cause tremendous impacts on the local hydrological cycle and the biosphere. On the other hand, the original values of R5d are very low there due to the arid and semi-arid climate in most parts of west China, so the significant increase in the changing rate does not equal the remarkable change in the absolute value of the 5 day total precipitation. Therefore, how this change influences west China is unclear. The variations of R5d during 2070–2099 are almost the same as during 2040–2069 (Figures S2c and S2e).

[23] The entire territory of China is dominated by increasing R95T during 2010–2039, indicating that the precipitation from very wet days may contribute more to the annual total precipitation (Figure 5a). The west China, especially the Qaidam Basin, is featured by the largest increasing magnitudes of R95T, being around 80%. The increasing rates of R95T in the middle and southeast China are relatively small, ranging from 0% to 10%. During 2040–2069, the changing magnitudes of R95T are larger compared to during 2010–2039 (Figures 5a and 5c). In the western part of China, the increasing rates reach the maximum of about 100%. Therefore, compared to 1960–2005, much more proportion of precipitation occurs as extreme heavy precipitation with precipitation volume exceeding the 95th percentile in northwest China. In south China, increasing rates of R95T are primarily of 10% – 20%. At the same time, changing rates of the R95T in the northeast China are mainly around 30% – 40%. The variations of R95T during 2070–2099 are almost the same as during 2040–2069 (Figures 5c and 5e).

[24] The increasing SDII is detected in major parts of China with an increasing magnitude of 0% – 10%, showing that the overall precipitation may be more concentrated during 2010–2039 (Figure S3a). On the other hand, the decrease of SDII ranges between −10% and 0% in parts of southwest China and the southeast river basins. During 2040–2069, increasing rates of SDII in most China are still 0% – 10%, but in some places, such as the Liaohe River, the SDII increases by 10% – 20% (Figure S3c). The variations of SDII during 2070–2099 are almost the same as during 2040–2069, but there are less areas with SDII increasing by 10% – 20% (Figures S3c and S3e). The variations of SDII indicate generally intensifying precipitation regimes over the entire territory of China.

[25] As a peak-and-decline scenario, the radiative forcing in RCP26 reaches a peak during 2040–2069 and then declines (Figure 2). The spatial patterns and magnitudes of changes in extreme indices during 2070–2099 and 2040–2069 are almost the same. In some cases, the changes during 2070–2099 are even less dramatic. RCP26 is a scenario in which the 2°C global average warming target is satisfied. Even in this case, changes in extreme precipitation are still considerable, especially in the west China. Therefore, in general, extreme precipitation in the whole China under RCP26 becomes heavier and more intensive. In south China, at the same time, the potential of occurrence of droughts may increase.

4.3 Variations of Extreme Precipitation Under RCP85 Scenario

[26] The changing patterns of extreme precipitation under RCP85 scenario, of which the radiative forcing reaches the highest level by 2100 compared to other RCPs, are also investigated.

[27] During the period of 2010–2039 under RCP85, CDD in south China, the Huaihe River basin, and the south Haihe River basin increases by 0% – 10% (Figure S1b). Compared with the CDD variations during 2010–2039 under the RCP26, the area with increasing CDD is larger under RCP85: CDD in the Huaihe River basin and Haihe River basin also increases (Figures S1a and S1b). During 2040–2069, the CDD in south China and the Huaihe River basin increases (Figure S1d). The largest increasing magnitude of CDD is found in southeast China and the upper and lower Pearl River basin, ranging from 10% to 20%. North China is dominated mainly by decreasing CDD with magnitude of −10% – 0%. In the Songhuajiang River basin and the middle of the northwest rivers, the CDD decreases relatively dramatically, with decreasing rates of −20% – −10%. During 2070–2099, the CDD changes more dramatically (Figure S1f). Increasing CDD is detected in more regions in the south China with changing magnitude between 10% and 20%, implying higher occurrence probabilities of droughts in these regions. Meanwhile, the CDD decreases by 10% – 20% in more areas of the north China, implying lower probabilities of droughts.

[28] During the period of 2010–2039, the spatial distribution of R10 variations under RCP85 is similar to that under RCP26, except that the area with decreasing R10 is larger (Figures 4a and 4b). The R10 in the southern parts of the Yellow River basin and northern parts of the Pearl River basin also decreases under RCP85. During the period of 2040–2069, the R10 in the Qaidam Basin is double relative to during 1960–2005 (Figure 4d). The increasing rates of R10 in the south China range from 0% to 10%. Also, larger areas with increasing R10 are detected during 2040–2069 than during 2010–2039 (Figures 4b and 4d). Relative to 1960–2005, the R10 in west China during 2070–2099 is more than double, showing that there are much more heavy precipitation days in the west China by the end of the 21st century, which may help mitigate droughts (Figure 4f). Compared to 2040–2069, R10 in the north China increases with higher magnitudes (10% – 70%) during 2070–2099 (Figures 4d and 4f).

[29] During 2010–2039 under RCP85, change in R5d is quite similar to that under RCP26 (Figures S2a and S2b). Compared with the changes in R5d during 2010–2039, R5d increases more dramatically during 2040–2069 (Figures S2b and S2d). In the whole China, the range of increasing rates of R5d is mainly from 10% to 30% during 2040–2069. The R5d in the west and northeast China increases relatively more dramatically than in middle and south China. The R5d variation during 2070–2099 exhibits a similar spatial pattern when compared to those during 2040–2069, but the changing magnitudes are larger (Figures S2d and S2f). The R5d in the northeast China increases by 30% – 50%, implying increasing flooding risk. Besides, R5d in west China increases about 60% – 80%, showing that extreme precipitation events denoted by R5d may be much more intensive in west China during 2070–2099.

[30] The spatial patterns of changes in the R95T during 2010–2039 under RCP85 and RCP26 scenarios are quite similar (Figures 5a and 5b). Spatial pattern of the change in R95T during 2040–2069 is almost the same as that during 2010–2039, while changing magnitudes are much larger, mainly from 20% to 50% (Figures 5b and 5d). The R95T in the middle of west China increases the most (> 80%), showing that much more proportion of precipitation in this region occurs in the form of extreme heavy precipitation. During 2070–2099, the R95T in the lower Yangtze River basin increases by mainly 10% – 40%, and R95T in the upper Yangtze River increases by >80% (Figure 5f). This kind of imbalance in increase of R95T may alter the hydrological conditions in the Yangtze River basin and introduce uncertainties into water resources management. The R95T in the northeast China mainly increases by 50% – 70%, and that in the middle of west China is more than double of the observed values during 1960–2005.

[31] The SDII in many parts of China increases during 2010–2039 (Figure S3b). During 2040–2069, the SDII in the northeast China, the Huaihe River basin, and the southwest of west China increases more than during 2010–2039 (Figures S3b and S3d). The spatial distribution of SDII changes during 2070–2099 is the same as that during 2040–2069, but SDII increases more remarkably (Figures S3d and S3f). The SDII in the majority of China increases by 10% – 20%. The largest increasing rates of SDII are found in west China, which reaches about 50%. Changes in SDII indicate that precipitation in China may be more intensive, and in west and northeast China in particular.

[32] In fact, the variations of extreme indices under RCP85 and RCP26 scenarios during 2010–2039 are quite similar. To be more specific, changing magnitudes do not have much discrepancies, and spatial patterns are basically the same, although sometimes areas with specific changing properties extend or shrink under RCP85, i.e., the area with decreasing R10 under RCP85 extends to the southern part of the Yellow River basin (Figures 4a and 4b). Overall, as the radiative forcing under RCP85 scenario is higher, the extreme precipitation in the 21st century under RCP85 scenario is becoming more extreme, heavier, and more intensive. Like the radiative forcing under RCP85 increases monotonously, changing magnitudes become more and more dramatic with elapsing time.

4.4 Variations of Extreme Temperature Under RCP26 Scenario

[33] In the backdrop of global warming, extreme temperature variations represented by five extreme temperature indices during the periods of 2010–2039, 2040–2069, and 2070–2099 under RCP26 scenario are analyzed. It should be noted that some stations fail to be downscaled, or some temperature extremes never occur there (i.e., FD in south China). Such stations are excluded in these analyses.

[34] During 2010–2039, TN90 in the whole China increases (Figure 6a). Increasing rates of the whole China are mainly in the range of 0% – 100%, showing that warm nights occur much more often in the years to come. During 2040–2069, the increasing magnitudes of TN90 are larger than during 2010–2039 (Figures 6a and 6c). The TN90 in most parts of China increases by 50% – 100%, and those in the coasts of southeast China, the southwest China, and parts of the northwest China increase by 100% – 150%. During 2070–2099, the change in TN90 is quite similar to that during 2040–2069 (Figures 6c and 6e).

Figure 6.

The same as Figure 3, but for TN90.

[35] Figure S4a shows the changing pattern of FD during 2010–2039. Attention should be paid that the definition of FD is the number of days with minimum temperature < 0°C. But in some places of south China, the minimum temperature is impossible to be less than 0°C. So in those places, FD always equals to 0. Therefore, it is impossible to calculate the changing rates in those places. These stations have been excluded when analyzing the FD variations. The FD in the whole China decreases. The FD in north China decreases by 0% – 10%, indicating that there may have 0% – 10% less frost days in the coming future relative to the observation. The FD in south China decreases greatly, ranging from −80% to −30%. In the middle Yangtze River basin, where Sichuan Basin locates, the decreasing rate is around −80%, which means that FD is only 20% of that during 1960–2005. In fact, in south area, the occurrence of minimum temperature < 0°C is originally very low. In other words, the number of frost days in this area during 1960–2005 is very low. Therefore, FD in the future is very small correspondingly. Moreover, it should be noted that some stations, especially in south China, are excluded when interpolating FD, because it is impossible to have frost days there, so the bias of interpolation in south China may be relatively large. Figure S4c shows that the spatial variation of FD during 2040–2069 is similar to that during 2010–2039, but changing magnitudes are larger. The FD in west China, Yellow River basin and Haihe River basin decreases by 10% – 20%. In addition, more areas in south China are found with decreasing FD with decreasing range of −80% – −30%, indicating that more areas in south China may have very small FD. Some areas in south China have decreasing rates close to −100%, meaning that FD almost disappears there due to the increasing minimum temperature. The variation in FD by the end of the 21st century is similar to that during 2040–2069 (Figure S4e).

[36] The changes in ETR are smaller compared to other temperature extremes (Figure S5). At the beginning of the 21st century under RCP26, the ETR in most parts of China decreases by 0% – 5%, and that in the southwest China increases by 0% – 5% (Figure S5a). There is a small difference between spatial patterns of ETR changes during 2040–2069 and during 2010–2039: more area in the south China has increasing ETR and ETR in the southeast rivers basin begins to increase slightly (Figures S5a and S5c). During 2070–2099, about half of the China, including the south China and most parts of the northwest rivers basin, has increasing ETR (Figure S5e).

[37] Figure S6 illustrates the variations of GSL. According to the definition of GSL, Tday < 5°C for >5 days must be identified. Nevertheless, in some places of south China, it is almost impossible that Tday can keep below 5°C for more than 5 days. Consequently, lots of stations in south China are excluded from analysis in GSL. Again, the bias caused by the gap of GSL shall be paid good attention. Figure S6a shows that almost entire north China has increasing GSL with increasing rate of 0% – 10%, indicating that the growing season length in north China is a little bit longer during 2010–2039 under RCP26. A station in the southeast rivers basin has decreasing GSL, which causes decreasing GSL in the whole southeast China. In fact, the increase in GSL can be caused by the phenomenon that Tday > 5°C for > 5 days occurs earlier or the Tday < 5°C for > 5 days starts later. In the backdrop of global warming, the above two phenomena may happen more frequently. The change in GSL during 2040–2069 and 2070–2099 is exhibiting the same spatial pattern and nearly the same changing magnitudes as during 2010–2039, except that GSL in the upper southwest rivers basin, the upper Yangtze River, and the upper Yellow River is in a larger increasing magnitude (10% – 20%) (Figures S6a, S6c, and S6e).

[38] Under RCP26, the HWDI in entire China increases during 2010–2039 (Figure 7a). The increasing rates in southeast China are relatively small, ranging mainly between 0% and 50%. The increasing rates in northeast China are relatively larger, ranging primarily from 50% to 100%. As for HWDI in west China, it increases by 100% – 200%, the range of which is much higher than those in east China. HWDI is an extreme temperature index considering the duration of heat waves. The remarkable increase of HWDI indicates that the duration of the heat wave may be much longer in the whole China during 2010–2039 under RCP26. The HWDI increases more tremendously during 2040–2069 than during 2010–2039 (Figures 7a and 7c). The HWDI in southeast China increases by 50% – 100%. The HWDI along the coastal areas has relatively high increasing rates, being in the range from 100% to more than 200%. This result implies that the durations of heat wave in coastal areas may be two to three times of those during 1960–2005. The HWDI in west China also increases dramatically, ranging from 150% to 250%. The largest increasing rate of HWDI is found in the most southwest China, which is close to 500%.

Figure 7.

The same as Figure 3, but for HWDI.

[39] During 2070–2099, the radiative forcing declines after reaching the peak during 2040–2069. Actually, all temperature extremes during 2070–2099 keep the same spatial patterns and similar changing magnitudes as during 2040–2069. In some cases, changing magnitudes even decrease, for example, TN90 (Figure 6). Furthermore, the east coasts suffer more TN90 and longer HWDI continually, meaning that coastal areas in China during 2070–2099 may suffer more warm nights and lengthening heat waves (Figure 7). As lots of metropolises are located in coastal areas, such dramatic change may cause tremendous impacts on human society, even under a scenario where the 2°C global average warming target could be met.

[40] In the 21st century under RCP26, more and more warm nights occur in China. In addition, the global warming also leads to the decrease of the number of frost days, and lengthening of the growing season. The duration of heat wave is much longer than during 1960–2005. In general, extreme temperature events are intensifying and higher.

4.5 Variations of Extreme Temperature Under RCP85 Scenario

[41] During 2010–2039 under RCP85, the whole China is featured by increasing TN90 with the changing range of 50% – 100% (Figure 6b). Compared to RCP26, the area with relatively high increasing rates of TN90 is larger under RCP85 (Figures 6a and 6b). During 2040–2069, the TN90 increases much more compared to during 2010–2039 (Figures 6b and 6d). The majority of China has increasing TN90 ranging from 150% to 200%. The TN90 in the middle Yangtze River and the northeast China increases relatively less, while the TN90 in the most southwest China increases the most. During 2070–2099, TN90 across China increases very dramatically while the spatial pattern does not change too much (Figures 6d and 6f). The TN90 in west China increases by 250% – 350%, indicating that the frequency of warm nights increases to roughly two to three times more than that during 1960–2005. The TN90 in southeast and northeast China increases by 150% – 250%.

[42] The spatial pattern of FD changes during 2010–2039 under RCP85 scenario is similar to that during RCP26 scenario (Figures S4a and S4b). The FD changes more dramatically during 2040–2069 than during 2010–2039 (Figures S4b and S4d). The FD in the north China decreases more during 2040–2069, and ranges from −20% to −10%. The decreasing rate in the south China is very close to 100%, implying that FD may disappear in south China. During 2070–2099, the decreasing rate of FD in the entire south China is very close to 100% (Figure S4f). The FD in the whole China decreases much more dramatically during 2070–2099 compared to during 2040–2069. In general, FD in north China also decreases by 20% – 30%.

[43] The ETR in the south of the southwest rivers, the Pearl River basin, and the southeast rivers basin increases by 0% – 5%, while the ETR in the other parts decreases by 0% – 5% during 2010–2039 (Figure S5b). The biggest difference between ETR variations under RCP85 and RCP26 scenarios is that ETR in the Pearl River basin and the southeast rivers basin increases under RCP85 (Figures S5a and S5b). During 2040–2069, the ETR in more areas in the middle and lower Yangtze River basin begins to increase (Figure S5d). The change in ETR during 2070–2099 is quite similar to that during 2040–2069, except that the ETR decreases by 5% – 10% in the Songhuajiang River basin (Figures S5d and S5f). This rate is relatively dramatic, considering the changes in ETR are relatively stable.

[44] The China is dominated by increasing GSL ranging between 0% and 10% during 2010–2039 (Figure S6b). Compared to 2010–2039, the north and west China is featured by increasing GSL with larger magnitudes during 2040–2069 (Figures S6b and S6d). Although the spatial pattern of the increase of GSL during 2070–2099 is similar to that during 2040–2069, the increasing magnitudes increase to a higher level (Figures S6d and S6f). The GSL in most parts of north China increases by 10% – 20%. In the upper Yangtze River and Yellow River basins, GSL increases by 40% – 100%.

[45] The HWDI in the whole China increases under RCP85 (Figure 7). During 2010–2039, the HWDI in south and southeast China mainly increases by 0% – 100%, and the HWDI in southeast coasts increases by relatively larger magnitudes of 50% – 100%. The increasing rates of HWDI in west China are relatively high, and the HWDI in southwest China increases with the increasing rates of about 300%, showing that the duration of heat waves in southwest China during 2010–2039 under RCP85 scenario is already nearly four times larger than that during 1960–2005. The HWDI in the whole China increases much more dramatically during 2040–2069 compared with that during 2010–2039 (Figures 7b and 7d). The HWDI in southwest China and east coasts increases more significantly than other areas, with increasing rates larger than 300%. For south China, where the change in HWDI is relatively small compared to the whole China, the increasing rates also reach 100% – 200%. The HWDI in the majority of China, mainly in the west China and the east coasts, increases by more than 500%, implying that the duration of the heat wave in most parts of China during 2070–2099 is approximately six times more than the duration during 1960–2005 (Figure S6f). Such tremendous lengthening of the heat wave durations may have harmful impacts on human society and ecological environment.

[46] Basically, the patterns and tendencies of extreme temperature changes under RCP85 scenario are similar to those under RCP26 scenario. Warm night occurs more frequently and the duration of heat wave increases a lot across the entire territory of China. Frost days occur much less and even may disappear in south China. The growing season length becomes longer. However, the changing magnitudes of extreme temperature under RCP85 and RCP26 scenarios are quite different. During 2010–2039, changes magnitudes of extreme temperature events under RCP85 are almost the same as RCP26. During 2040–2069, extreme temperature events under RCP85 scenario begin to change relatively more dramatically. During 2070–2099, extreme temperature events change much more tremendously under RCP85. At the end of the 21st century, compared with RCP26 scenario, under RCP85 scenario, more warm nights, longer heat wave, less frost days, and longer growing season length may be detected.

4.6 Temporal Evolution of Precipitation and Temperature Extremes in China

[47] The areal averages and standard deviations of changing rates of precipitation and temperature extremes in China during periods of 2010–2039, 2040–2069, and 2070–2099 are calculated based on the interpolated maps described above. The areal averages and standard deviations illustrate temporal changes of extremes in the whole China (Figure 8). The averages of CDD in overall China decrease in the future (Figure 8a). The radiative forcing under RCP26 increases first and then declines. However, under RCP26, the overall averages of CDD decrease slightly at first, and then stay at −2.82% during 2040–2069 and 2070–2099. The standard deviations during 2040–2069 and 2070–2099 are larger than that during 2010–2039. Under RCP85, the averages of CDD do not change a lot yet, but as the time goes by, the standard deviations increase. This result indicates that global warming does not induce huge changes of the CDD areal averages in China, but causes relatively considerable regional variations. Figures. S1b, S1d, and S1f also imply that as the radiative forcing increases, the changing magnitudes of CDD increase correspondingly when the spatial patterns of changes in CDD are quite stable. Under RCP26, the average of changing rates of R10 is 10.10% during 2010–2039, and reaches 18.28% during 2040–2069 (Figure 8b). Then during 2070–2099, the average and standard deviation do not change a lot from those during 2040–2069. Under RCP85, the averages and standard deviation of R10 increase continuously. Similarly, for R5d, R95T, SDII, TN90, GSL, and HWDI, the averages and standard deviations increase correspondingly when the radiative forcing increases (Figures 2, 8c, 8d, 8e, 8f, 8i, and 8j). Nevertheless, in some cases, under RCP26, the averages and standard deviations do not decrease correspondingly when the radiative forcing decrease during 2070–2099 (Figures 2 and 8). In fact, the averages and standard deviations during 2070–2099 are equal to or slightly larger than those during 2040–2069 (e.g., CDD, R10, R5d, and R95T). Among all indices, the TN90 and HWDI increase most dramatically, indicating that China would suffer more warm nights and longer heat wave. The FD decreases more if the radiative forcing increases more (Figure 8g). The changing magnitudes of ETR are the smallest compared to other extreme indices, indicating the global warming has limited influences on the ETR (Figure 8h).

Figure 8.

The areal changing rates of (a) CDD, (b) R10, (c) R5d, (d) R95T, (e) SDII, (f) TN90, (g) FD, (h) ETR, (i) GSL, and (j) HWDI during 2010–2039, 2040–2069, and 2070–2099 in the whole China under RCP26 (blue bar) and RCP85 (orange bar) relative to 1960–2005. The bold horizontal line inside the bar denotes the average; the upper/lower edge denotes the average plus/minus the standard deviation. The horizontal dash line denotes the 0% changing rate. The gray background denotes the period of 2040–2069, and the white background on the left/right denotes period of 2010–2039/2070–2099. For FD and GSL, the stations of which FD or GSL never occurs in the observation are excluded in the areal changing rates.

5 Discussion

[48] Jiang et al. [2012] compared the IPCC-AR4 models, and concluded that the multimodel ensemble can represent the extremes very well in China. The assessment of this paper also indicates that the CMIP3 and CMIP5 ensembles perform better than individual models. And in most case, the CMIP5 ensemble has the best performance. In the above assessment, the performances of CMIP5 ensemble in downscaling the R95T and TN90 are not so good. However, Jiang et al. [2012] compared the observed and simulated extreme indices in gridded scale, and concluded that the GCMs model the TN90 and R95T relatively good, which is different from this study. In fact, the assessment in this study is based on the site scale after downscaling, while the comparison of Jiang et al. [2012] was based on the gridded scale. Also, the assessment methods are different: this study applies the Taylor diagram, while Jiang et al. [2012] used the mean absolute error.

[49] Most of previous studies about possible changes in extreme events were based on the CMIP3 data sets. When comparing this study to the previous studies, a simple understanding about differences of changes in extremes calculated from the CMIP3 and CMIP5 would be helpful. The RCP45 in CMIP5 and B1 scenario in CMIP3 have similar CO2 concentrations by the end of 21st century [Sillmann et al., 2013b]. Due to the differences in prescribed forcing agents between B1 and RCP45 scenarios, their simulations of climate changes are not directly comparable [Rogelj et al., 2012]. However, projected changes can be compared to provide an estimate of uncertainty related to different scenarios [Sillmann et al., 2013b]. They found out that changes in extreme temperature and precipitation are generally more pronounced in RCP45. In this study, a preliminary and straightforward comparison between simulations of extreme in China under RCP45 and B1 scenarios is made (Table 4). Because the difference in models of ensembles may cause discrepancy, the CMIP5 ensemble only includes CNRM-CM5, IPSL-CM5A-MR, MIROC5, and CCSM4, all of which are released by the institutes of CMIP3 models used in this study. The areal averages of changes in precipitation extremes under B1 and RCP45 are very close, except TN90 and HWDI. Basically, changes in CDD, TN90, and FD are more remarkable under RCP45, while changes in other indices are more pronounced under B1. This result is not completely in good agreement with the results of Sillmann et al. [2013b]. There are two reasons: first, they estimated the difference in global scale, while our result focuses on China, and; second, Sillmann et al. [2013b] calculated extremes indices on gridded scale, while this study calculates on site scale after downscaling. Some outputs from results of Sillmann et al. [2013a, 2013b] are used to better discuss the differences in changes between B1 and RCP45. It should be noted that their results are regridded to 2.5° × 2.5° grid, and only outputs of models used in this discussion are incorporated to the ensembles. Then the changes in some extremes estimated by the procedures of this study and Sillmann et al. [2013b] during 2070–2099 under RCP45 and B1 are compared (Figures 9, 10 and S7–S10). The increase of R10 in the northwest China during 2070–2099 is less remarkable under RCP45, but there are more places with increasing R10 in the south China under RCP45 (Figures 9a and 9b). Figures 9c and 9d indicate the temporal averages of R10 during 2070–2099 on the gridded scale, which is extracted from the result of Sillmann et al. [2013a, 2013b]. In both site and gridded scales, the northwest China has less R10 under RCP45, and the south China has larger R10 under RCP45, which are spatial differences under B1 and RCP45 scenarios. Figure 10 shows that the increase of TN90 in the whole China is more pronounced under RCP45. Relative to B1, the temporal averages of CDD are larger in the northwest China, and smaller in the south China under RCP45 (Figures S7c and S7d). Correspondingly, compared to the changes in B1, the decrease of CDD in the northwest China, and the increase of south China are less remarkable under RCP45. The differences in temporal averages of FD between under B1 and RCP45 also support the differences in changes in FD (Figure S10). The increase of R5d in the northwest China is less remarkable under RCP45 (Figures S8a and S8b), which can also be represented by the decrease of temporal averages of R5d in the northwest China (Figures S8c and S8d). Nevertheless, the increase of R5d in the south China is less prominent under RCP45, but the temporal average is larger there under RCP45 (Figure S8). Such discrepancy in south China is also found for SDII (Figure S9). Therefore, various spatial differences between extremes under B1 and RCP45 are identified in China. Basically, such spatial differences identified in this study are in good agreement with the outputs provided by Sillmann et al. [2013a, 2013b]. Nevertheless, it is hard to decide changes in extremes are more remarkable under B1 or RCP45 in terms of only areal average. This is a very preliminary comparison and provides some basic information about the differences in CMIP3 and CMIP5.

Table 4. The Areal Averages of the Changes (%) in Extreme Indices Under B1 and RCP45
IndexScenario2010–20392040–20692070–2099
CDDB1−0.94−1.51−1.82
RCP45−0.95−3.10−2.55
R10B119.0223.6334.77
RCP4510.6120.1722.86
R5dB18.5114.8116.47
RCP457.517.8610.40
R95TB126.2137.2546.70
RCP4525.6936.5541.75
SDIIB18.4012.3418.21
RCP453.397.168.51
TN90B12.5628.6056.35
RCP4554.95115.24149.10
FDB1−20.09−25.91−31.49
RCP45−17.19−28.26−32.80
ETRB1−0.56−0.42−1.11
RCP45−0.31−0.53−0.65
GSLB15.188.4611.46
RCP454.678.9410.52
HWDIB1171.94331.65534.79
RCP4587.93182.66247.47
Figure 9.

The comparison between R10 during 2070–2099 in terms of the changes (%) relative to 1960–2005 under (a) B1 and (b) RCP45, as well as in terms of the temporal averages of R10 (day) under (c) B1 and (d) RCP45.

Figure 10.

The same as Figure 9, but for TN90 (%).

[50] The abnormally long period of precipitation deficiency, reflected by the maximum consecutive dry days (CDD) in this study, is an important indicator of droughts [Heim, 2002]. In north China, the CDD decreases, implying that the risk of droughts may decrease. In south China, the CDD increases, indicating that the risk of droughts increases. There are also two contrary explanations for the increase of CDD: the south China may be dominated by decreasing precipitation, so the consecutive dry days would be lengthening; or precipitation in south China may be intensifying and extreme that less rainy days could precipitate most precipitation of the whole year while the total precipitation would not change much, so the consecutive dry days would tend to be lengthening. The second explanation implies that precipitation would be more likely to occur in the form of extreme heavy precipitation instead of normal precipitation. Relative to 1960–2005, the averages of the annual precipitation from wet days and the number of wet days in the south China during 2010–2039 decreases slightly (not shown here). Therefore, the lengthening of CDD during 2010–2039 is caused by the first explanation. Nevertheless, relative to 1960–2005, the annual precipitation in south China increases slightly while the number of wet days decreases slightly during 2040–2069 and 2070–2099 (not shown here). At the same time, changes in R5d, R95T, and SDII indicate that precipitation becomes more intensifying, and precipitation is more likely to occur as extreme heavy precipitation. Therefore, the second explanation is more reasonable for the lengthening of CDD.

[51] In general, the whole China may experience more extreme heavy precipitation and more extreme heat waves during the 21st century. For southwest China, it may experience more severe droughts, more extreme precipitation, and higher extreme temperature. Under RCP26 and RCP85 scenarios, spatial patterns of variations of climate extremes are roughly similar. According to the setting of scenarios, the radiative forcing by the end of the 21st century of RCP85 scenario is larger than those of RCP26 scenario. Therefore, generally, the changing magnitudes under RCP85 scenario are larger than under RCP26 scenario.

[52] This study not only investigates extremes by the end of the 21st century (usually 2070–2099), a period which previous studies usually focused on [Jiang et al., 2012], but also analyzes the spatiotemporal evolution of precipitation and temperature extremes in China during the whole 21st century. In fact, the change in extreme is a gradual process. Therefore, studying variations of extremes in different episodes can provide more details about the future climate change, and help to understand how extremes would change to the tremendous conditions by the end of 21st century. In most cases, during the intensifying changes in extremes, the spatial patterns are stable, but the changing magnitudes increase as radiative forcing raises. For example, changing rates of R95T in west China are always the largest compared to other regions, and as time goes by, the radiative forcing under RCP85 increases continuously, then the increasing rates increase correspondingly (Figure 5). In a few cases, the spatial pattern may alter. For instance, under RCP85, the R10 in south China does not increase during 2010–2039 (Figure 4b). Nevertheless, as the radiative forcing continues increasing during 2040–2069, the increasing R10 occurs in the south China (Figure 4d). Finally, during 2070–2099, the area with increasing R10 extends eastward (Figure 8B). This evolution process can also be found in CDD, SDII, R10, and ETR. Moreover, for CDD, the increasing of radiative forcing only changes the overall average of CDD in China slightly, but changes regional variations relatively more (Figures S1 and 8). The increasing of CDD in some parts of China offsets the decreasing of CDD in the other parts of China, which makes the over average changes slightly.

[53] Under RCP26, a scenario in which 2°C global average warming target by the end of 21st century is achieved, the increases in R10, R5d, and R95T in the north China, especially the northwest China, are still considerable. At the same time, the TN90 and HWDI in the whole China also increase considerably. Also, the FD in the south China decreases a lot. Therefore, even the global average warming target would be controlled, the changes in extremes in regional areas may be still remarkable.

[54] The CDD is an indicator of the vulnerability of a region to droughts [Sillmann and Roeckner, 2008]. During the 21st century, CDD increases in south China. From this perspective, south China may be much more vulnerable to droughts in the 21st century. In 2010 spring, a severe drought occurred in southwest China and affected 51 million Chinese [People's Daily Online's, 2010]. In 2012, this drought has lasted for three years [International Business Times, 2012]. This historical drought event may be an evidence showing the increasing CDD in southwest China. During 2040–2069 and 2070–2099, especially under RCP85 scenario, the CDD increases even more, and the largest increasing magnitudes of CDD are identified in the southwest China, implying that southwest China may be exposed to more severe droughts. According to the variations of the R10, R5d, R95T, and SDII, extreme precipitation in China may be heavier and occur with higher frequency. In other words, extreme heavy precipitation in China during the 21st century tends to be increasingly intensifying, which may cause other related hazards. For example, the Loess Plateau is located in middle China, dramatic extreme precipitation may cause not only floods, but also landslides and accelerated erosion of the loess. Consequently, sediment load in the lower Yellow River basin may correspondingly increase, which then may raise the riverbed in the lower Yellow River. Finally, the risk of and the vulnerability to floods may increase [Derbyshire, 2001]. This extremization of precipitation may be quite distinguishable in west China, an arid area. The potential of floods and related geological hazards, such as landslides, may also be raised a lot. In southeast China, where many cities are located, the extremization may be also considerable. This may add the possibility of waterlogging inside the cities, and may lead to social and economic losses.

[55] More extremely hot events happen in the 21st century. The TN90 increases continually through the whole 21st century. Besides, due to the increasing temperature, the possibility of frost days in the whole China decreases. The frost days in south China are disappearing, and the area where frost days may disappear extends northward, which may change or even damage the local agricultural structure completely in related regions. Additionally, the duration of heat wave in the whole China increases greatly, which is especially true in west China and east coasts. In northern parts of west China, a semi-arid or desert area, it is usually extremely hot in summer. The dramatic extension of the length of heat wave makes this area much hotter and increases the potential evaporation [Van Bavel, 1966]. There are many cities and huge population along the east coasts. The consistency of heat wave may directly induce lots of heat-related diseases and deaths [Semenza et al., 1996].

6 Conclusions

[56] In this study, the performances of GCMs of CMIP3 and CMIP5 in description of spatiotemporal patterns of extreme precipitation and temperature events are assessed based on Taylor diagram. Then spatiotemporal evolutions of precipitation and temperature extremes across China under RCP26 and RCP85 scenarios during the 21st century are investigated by using outputs of CMIP5 ensemble. The differences of extremes from CMIP3 and CMIP5, as well as the implications of changes in extremes are discussed. The following are drawn from this study:

  1. [57] The CMIP5 ensemble has the best ability in describing spatiotemporal variations of extreme events. Its performance is better than the CMIP3 ensemble. The performances in representing extreme indices defined by high percentiles are not as good as other indices.

  2. [58] During the 21st century, extreme precipitation in China is increasingly intensifying. The southwest China may suffer more droughts and at the same time more extreme heavy precipitation. Changes in extreme precipitation in west China are more pronounced. Compared to the RCP26 scenario, extreme precipitation events change more dramatically under the RCP85 scenario.

  3. [59] Extreme temperature events in China during the 21st century occur with increasingly large magnitude and with higher frequency. The number of warm nights and the duration of heat waves increase dramatically, and the number of frost days is decreasing. Extreme temperature events change more tremendously under the RCP85 scenario.

  4. [60] Changes in extreme precipitation and temperature are gradual processes. In most cases, the evolutions of changes are reflected by the variation of changing rates with stable spatial patterns. In some cases, spatial patterns change step by step, which means the area with specific properties may extend or shrink gradually. In addition, the changing sign may change. The R10 in south China first decreases, and only a small proportion area has increasing R10. As time goes by, the area with increasing R10 extends to the majority of south China.

  5. [61] Changing magnitudes highly depend on the radiative forcing. Under RCP85, changes in extremes become more and more pronounced as the radiative forcing increases. Under the peak-and-decline RCP26, the changes in China are more considerable during 2040–2069 than 2010–2039. However, in some cases, the changes during 2070–2099 do not decrease correspondingly although the radiative forcing during 2070–2099 is less than during 2040–2069. Moreover, the increase of radiative forcing makes considerable regional changes in CDD, but makes only slight change in the areal average of CDD in China.

  6. [62] The remarkable variations of extreme precipitation and temperature during the 21st century may have different implications for different regions. South China may suffer higher risks of droughts and floods at the same time in the future. There may be more floods, landslides, and erosion of loess in the Loess Plateau. More and more heat-related diseases and deaths may be induced in the cities along the east coast. Overall, the risk of floods may increase in the whole China. Results of this study will be relevant for mitigation of intensifying climate extremes, floods and droughts, and planning of agriculture development of China.

Acknowledgments

[63] This work was supported by The National Natural Science Foundation of China (grant no. 41071020), Program for New Century Excellent Talents in University (NCET), and a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project no. CUHK441313). We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 and CMIP5 multimodel data sets. We thank the climate modeling groups (listed in the Tables 1 and 2 of this paper) for developing and making their model output available. Support of data sets is provided by the Office of Science, U.S. Department of Energy. Our cordial gratitudes should also be extended to the editor-in-chief, Prof. Dr. Steve Ghan, and two reviewers for their pertinent and professional comments and suggestions which are greatly helpful for further improvement of the quality of this manuscript.

Ancillary