Projected Changes in Temperature and Precipitation Extremes over China as Measured by 50-yr Return Values and Periods Based on a CMIP5 Ensemble

Increasing attention is being paid to projections of future changes in weather and climate extremes because of their profound impacts on human society and natural ecosystems. As concluded by IPCC in its Special Report on Managing the Risks of Extreme Events to Advance Climate Change Adaptation (Seneviratne et al., 2012), the impact of climate change is greater on extreme climates than the mean climate. The probability of occurrence of extreme climate events will continue to increase in the future (IPCC, 2013), and climate models are the main tools to address this issue. Climate change simulations with the latest generation of AOGCMs and ESMs are available for analysis within CMIP5 (Taylor et al., 2012). In general, compared to the previous generation of AOGCMs (CMIP3) (Meehl et al., 2007), the CMIP5 models are of higher spatial resolution (e.g., a highest horizontal resolution of up to 0.56^° and 95 vertical levels) and more comprehensive. Many studies have been conducted to analyze future changes in extremes based on ensembles of CMIP5 models (Chen, 2013; Chen et al., 2013, 2014; Kharin et al., 2013; Kunkel et al., 2013; Sillmann et al., 2013a, 2013b; Toreti et al., 2013; Fischer and Knutti, 2014; Wuebbles et al., 2014). In general, significant increases in hot spells and intensification of precipitation extremes in the middle and high latitudes throughout the globe are reported in these studies.

There are several ways to quantify temperature and precipitation extremes, e.g., the indices from the Expert Team on Climate Change Detection and Indices (ETCCDI) (Frich et al., 2002; Zhang et al., 2011; Giorgi et al., 2014a; Zhou et al., 2014; Xu et al., 2015a, 2015b). Another approach is, for example, that of Kharin et al. (2005, 2007, 2013), who analyzed changes in return values and periods using Generalized Extreme Value (GEV) distributions.

Located in the East Asian monsoon region, China is characterized by a climate with large natural variability and the region is vulnerable to climate change due to its relatively low adaptive capacity (Qin, 2012). More and more research is being devoted to extremes over China, from observational changes in past decades (Zhai et al., 1999; Zhai and Pan, 2003; Su et al., 2005; Qian et al., 2007; Zhang and Zhai, 2011; Chen and Zhai, 2013) to future projections based on either global (CMIP3 and CMIP5) or regional climate models (Gao et al., 2002; Jiang et al., 2004; Zhang et al., 2006; Xu et al., 2009a; Chen, 2013; Xu et al., 2013; Zhou et al., 2014; Sun et al., 2015). In particular, projected changes in extremes of 20-yr return levels have been investigated by (Xu, 2010) and (Wu et al., 2012) using CMIP3 ensembles and high-resolution regional climate simulations, respectively. Changes in return values indicate a general increase in heat waves and decrease in cold spells in the future in this region, both in terms of temporal frequency and spatial spread, as well as an intensification of extreme precipitation. However, few studies have thus far been conducted using the current state-of-the-art CMIP5 model simulations. In addition, a gridded dataset at the daily scale based on denser station observations over China recently became available (Wu and Gao, 2013), which helps in better evaluating models' performances in simulating extremes.

The main objective of this paper is therefore to investigate future changes in temperature and precipitation extremes over China based on the most recent CMIP5 multi-model ensemble. Building on previous work, the analysis focuses on the very-high-risk events defined by the 50-yr return levels of the extreme indices from ETCCDI. We compare the simulated return values in the present against observations to validate the model performance, and assess projected changes in return values and return periods for these very extreme events under forcings from three greenhouse gas future scenarios: RCP2.6, RCP4.5 and RCP8.5 (Moss et al., 2010).

Following this introduction, the datasets and methodology are described in section 2, followed by the validation of present-day simulations in section 3. Projected future changes in 50-yr return values and periods are presented in section 4, with a summary concluding the paper in section 5.

The CMIP5 ensemble includes 32 models for the historical period 1950-2005, 22 for RCP2.6, 31 for RCP4.5, and 29 for RCP8.5 (over the period 2006-2100). We select the 22 models that provide daily output and only consider one run per scenario. Information on the 22 models is given in Table 1, and more detail can be found at http://cmip-pcmdi.llnl.gov/cmip5/.

The gridded observation dataset CN05.1 developed by (Wu and Gao, 2013) is employed to validate the model performance in simulating the extremes. The dataset is composed of daily mean, minimum and maximum temperature, and daily precipitation, over China. It is a further development of CN05 (Xu et al., 2009b), being based on an interpolation from more station observations (760 stations for CN05 and 2416 for CN05.1) at different spatial resolutions of 0.25^°, 0.5^°, 1^° and 2^°. Here, we use the 0.5^° resolution to match the highest resolution of the models.

We use the following indices defined by ECTTDI (http://etccdi.pacificclimate.org/indices.shtml): annual TXx (the maximum of daily maximum 2-m surface air temperature); annual TNn (the minimum of daily minimum temperature); annual RX5day (the maximum consecutive 5-d precipitation); and annual CDD (maximum number of consecutive dry days, or days with <1 mm of precipitation). The indices are among the most commonly used out of the tens of indices recommended by ECTTDI (e.g., Tebaldi et al., 2006; Sillmann et al., 2013a, 2013b).

The analysis of return periods is employed in this study, following the approach of Kharin et al. (2005, 2007, 2013). A GEV model is used to fit the probability of occurrence of the extreme events, and the parameters of the distribution (e.g., return period) are estimated based on the L-moment method, which is computationally fast and suitable for small-sample data (Hosking, 1990). We focus on events with a 50-yr return period in the present-day climate as a metric for the rare extremes of temperature and precipitation. The 50-yr return values of the abovementioned annual extremes are referred to as TXx-50, TNn-50, RX5day-50 and CDD-50, respectively.

It is noted that the return periods and return values are estimated using statistical models that describe the expected probability of an event. An event with a 50-yr return period is not necessarily expected to happen at intervals of 50 yr, but may also happen during a 20-yr period or several times within a 50-yr period. The 50-yr return period indicates that the expected probability of the event happening in a given year is 1/50 (=2%).

Return values are first estimated for the CN05.1 data on its 0.5^°× 0.5^° latitude——longitude grids, and each model on its native grid, and are then bilinearly interpolated onto the 0.5^°× 0.5^° grids for computing the multi-model ensemble statistics. In addition, regionally aggregated extreme value statistics are assessed over the eight sub-regions of China shown in Table 2. For the multi-model ensemble, the multi-model median is used in the analysis instead of the mean (average), where the median is the value separating the top and bottom halves of the model values in the ensemble, thereby defining the "middle" value.

Changes in temperature and precipitation extremes in the future are compared to the reference (present-day) period of 1986-2005, while 2016-35, 2046-65 and 2080-99 are considered as the early, mid and end of the 21st century, respectively. The term "change" in the paper indicates the difference (anomaly) between values for a future period and the present-day one.

We use box-and-whisker plots to illustrate the inter-model agreement or disagreement in the projected changes, which consist of the multi-model median, the interquartile model range (the range between the 25th and 75th quantiles, i.e., the box) and the full inter-model range (i.e., the whiskers). The interquartile model spread corresponds to an agreement across at least 75% of the models, which is referred to as "the majority of models" in this study (Sillmann et al., 2013a, 2013b).

The multi-model medians of TXx-50, TNn-50, RX5day-50 and CDD-50 from the reference period (1986-2005) simulations, as well as the corresponding bias compared to the observations, are presented in Fig. 1. The largest values of TXx-50 are found over the basins in Northwest and North China, with minima over the Tibetan Plateau. For TNn-50, the largest negative values are found in a band extending from Northeast China to the Tibetan Plateau, as well as the northern part of Northwest China, with the smallest values over southern China. Compared to observations, the CMIP5 model median shows a prevailing tendency to overestimate TXx-50 and to underestimate TNn-50, with biases typically within a few degrees over most of China, except in the west with its complex topography. Averaged over the entire China territory, there is a warm bias of 1.2^°C for TXx-50 and a cold bias of -4.1^°C for TNn-50. The largest warm bias for TXx-50 and the largest cold bias for TNn-50, both greater than 10^°C, are found over the mountains in Northwest China and the Tibetan Plateau, and in the Sichuan Basin. These are most likely due to the coarse model resolution and thus an excessively smoothed representation of the topography, which, along with a tendency of observing stations to lie in valleys rather at high elevations, leads to a generally warm bias in high mountain regions and cold bias in valley basins (Gao et al., 2013). We also note that the biases of TXx-50 and TNn-50 show consistent patterns with the biases in TXx and TNn, although the magnitudes are larger in some areas (Dong et al., 2015).

Figure 1.  Observed extreme indices (left-hand column) and differences between the CMIP5 multi-model median and observation (right-hand column) over China in the present day (1986-2005): (a, b) TXx-50 (units: ^°C); (c, d) TNn-50 (units: ^°C); (e, f) RX5day-50 (units: mm); (g, h) CDD-50 (units: d).

Moving to the precipitation based extremes, the largest observed values of RX5day-50 are found over the monsoon areas of eastern China, while the largest values of CDD-50 are found over the desert areas of Northwest China (and smallest over eastern China). In general, the bias in the multi-model median appears to be negatively correlated with the patterns of both RX5day-50 and CDD-50, indicating that the models tend to underestimate the magnitude of precipitation wet and dry extremes. In particular, large underestimations of RX5day-50 are found in eastern China centered over the middle and lower reaches of the Yangtze River, where monsoon precipitation dominates (Fig. 1f), which has also been found in previous simulations (Xu et al., 2010). The overestimation of RX5day-50 over central and western China is also a typical deficiency of climate model simulations over these regions (Zhang et al., 2008; Xu et al., 2013). As reported by (Gao et al., 2008), the overestimation of RX5day-50 at the southern edge of the Tibetan Plateau is due to the low resolution of the GCMs, which allows the penetration of precipitation fronts from the southern slope of the Himalaya into the region. The China-wide average bias of the RX5day-50 simulation over China is about 21% of the observed value.

Figure 2.  Box-and-whisker plots for the observation and multi-model CMIP5 ensemble of (a) TXx-50 (units: ^°C), (b) TNn-50 (units: ^°C), (c) RX5day-50 (units: mm) and (d) CDD-50 (units: d) over China and the eight sub-regions in the present day (1986-2005). The observation and models' median are indicated by the red stars and black bars, respectively.

For CDD-50, greater values in the range of 0-25 d compared to observations are found in eastern and southern China, and large negative biases are found at the northern edges of the Tibetan Plateau. It should be noted that the later biases occur in the region with the poorest observational network (Wu et al., 2011; Wu and Gao, 2013), which can also cause some uncertainty. Averaged over China, the bias of CCD-50 is about -36 d compared to observations. Similar to the temperature indices, the biases of Rx5day-50 and CDD-50 show consistencies with those for Rx5day, and CDD, but with larger values (Chen et al., 2014).

A summary of the observed and CMIP5 multi-model simulations for the extreme indices of TXx-50, TNn-50, RX5day-50 and CDD-50, over the whole of China and the eight sub-regions, is provided in Fig. 2. Information on the model simulations in the figure include the multi-model medians, the interquartile model ranges spanned by the 25th and 75th quantiles, and the total inter-model range.

The median values of TXx-50 are typically in the range from 34^°C to 40^°C across the sub-regions, indicating the occurrence of warm summers throughout the country, except in SWC1 (23^°C) where the Tibetan Plateau is located (Fig. 2a). More diverse values of TNn-50 are found across the sub-regions, with median values lower than -35^°C in NEC and SWC1, and around -8^°C in SC (Figs. 2b). For TXx-50, in five (NEC, NC, EC, SWC1, NWC) out of the eight sub-regions, as well as the whole country (CN), the observations are within the interquartile model spread. TXx-50 is slightly overestimated in these sub-regions, and largely overestimated in the other three regions (CC, SC and SWC2), where the observations fall outside the interquartile model spread. In general, more extreme warm events are simulated by the models. Conversely, an underestimation of TNn-50 corresponding to more cold events in the models can be found over most sub-regions, with the observations falling within the interquartile model spread in four sub-regions (CC, SC, SWC1 and SWC2) and the whole of China (Fig. 2b).

The models show lower performance in reproducing the observed precipitation extremes. They tend to overestimate RX5day-50 (Fig. 2c) in most sub-regions, with the observations falling outside the interquartile model spread. The largest overestimation is found in SWC1, SWC2 and NC. Conversely, a greatly underestimated RX5day-50 is found in EC, where monsoon climate dominates. In fact, only the value from one model simulation is found to be close to the observations. Note that EC is also a sub-region with large inter-model spread, indicating a large difference in model performance over the area. Overall, the models simulate shorter CDD-50 for the whole of China compared to observations (Fig. 2d), which is in line with the general finding that models tend to produce too many low precipitation events (e.g., Sillmann et al., 2013a). In CN, SWC1 and NWC, the observations in fact have values that are about twice the multi-model median, and are far from the interquartile model spread. It is interesting to note that, in NEC and SC, where a large bias of CDD-50 is found, the observed RX5day-50 is within the interquartile model spread, showing the different behaviors of the models in simulating the two ends of the precipitation extremes.

In summary, the CMIP5 ensemble shows reasonably good performance in simulating temperature extremes over China, with a prevailing tendency for overestimating both warm and cold extremes. Lower performance is found for the precipitation extremes, with observations lying outside the interquartile model range in the majority of regions, and the models showing a prevailing tendency to overestimate (underestimate) the wet (dry) extreme indices analyzed.

In this section we begin by calculating the 50-yr return values of temperature (TXx-50 and TNn-50) and precipitation (RX5day-50 and CDD-50) extremes from the CMIP5 multi-model ensemble for different periods of the 21st century, and then we compare them to the values calculated for the corresponding present-day period.

Figure 3 presents the spatial distribution of the multi-model ensemble median changes in TXx-50 and TNn-50 for the end of the century (2080-99) relative to the present-day period (1986-2005) under the RCP2.6, RCP4.5 and RCP8.5 scenarios. As shown in the figure, the warming causes an increase in the magnitude of both TXx-50 and TNn-50 following the increase in greenhouse gas forcings, which is minimum under RCP2.6, maximum under RCP8.5, and intermediate under RCP4.5.

The change in warm and cold extremes shows different spatial patterns. Cold extremes index (TNn-50) increases considerably faster over the high-latitude (Northeast and Northwest China) and high-altitude (Tibetan Plateau) areas, as related to the snow and ice albedo feedbacks (Giorgi et al., 1997). Most notably, a pronounced increase in TNn-50 of up to 8^°C is found in Northeast China under RCP8.5 (Fig. 3f). Conversely, the projected increase in warm extremes is more evenly distributed. The increase in TXx-50 is mostly in the range of 5^°C to 7^°C under RCP8.5 over the whole of the country, except in parts of Inner Mongolia, the Tibetan Plateau, and the southern coast.

Figure 3.  CMIP5 multi-model median changes in TXx-50 (left) and TNn-50 (right) by the end of the 21st century (2080-99) relative 1986-2005 under RCP2.6 (top), RCP4.5 (middle) and RCP8.5 (bottom) (units: ^°C).

Figure 4.  Box-and-whisker plots for multi-CMIP5 ensemble of (a) TXx-50 and (b) TNn-50 changes over China and the eight sub-regions by the end of the 21st century under different scenarios (units: ^°C). The colors indicate the different scenarios, with blue for RCP2.6, green for RCP4.5 and red for RCP8.5. The model median is indicated by the bars.

Figure 5.  Box-and-whisker plots for the multi-model CMIP5 ensemble of return periods of the present-day TXx-50 values in different periods (2016-35, 2046-65, 2080-99) of the 21st century over China and the eight sub-regions under different scenarios (units: yr). The colors indicate the different scenarios, with blue for RCP2.6, green for RCP4.5 and red for RCP8.5. The model median is indicated by the bars.

Figure 4 shows box-and-whisker plots of the projected sub-regional changes in TXx-50 and TNn-50 by the end of the 21st century under different scenarios. This figure confirms the greater spatial variability in the change of TNn-50 compared to TXx-50, and shows how the upper tail (larger warming) of the distribution of changes is more pronounced than the lower tail, i.e., it shows the presence of individual models with very high temperature sensitivity. The most significant increases in TXx-50 under RCP8.5 are found in EC and CC, with a multi-model median value of 5.9^°C and 6.6^°C, respectively. However, relatively large model spreads are also found in these two sub-regions. The increase in TXx-50 is in the range of 5.1^°C to 5.8^°C in other sub-regions (Fig. 4a). China-wide median changes in TXx-50 are 1.5^°C, 3.0^°C and 5.4^°C under the RCP2.6, RCP4.5 and RCP8.5 scenarios, respectively.

For TNn-50, the increases are more pronounced in NEC and NC, with values of 8.5^°C and 7.1^°C (RCP8.5), respectively, while a minimum warming of 4.1^°C is found in SC. The maximum inter-model spread in TNn-50 occurs over different regions compared to TXx-50, and specifically over the northern regions of NWC and NEC. This is likely related to the different strength of the snow-feedback effect in the different models. Median changes over China are 1.6^°C, 2.7^°C and 5.6^°C under the RCP2.6, RCP4.5 and RCP8.5 scenarios, respectively.

Return periods of the present-day TXx-50 in different periods of the 21st century under the three scenarios over China and different sub-regions are summarized in the box-and-whisker plots of Fig. 5. As shown in the figure, general decreases in the TXx-50 return periods are found in all the scenarios and across all the sub-regions throughout the 21st century. Changes in the return periods show less dependency on the emission scenarios in the early period of the 21st century, and then become more pronounced in the high-forcing scenarios. By the end of the century, the multi-model median return periods of the present day (50 yr) are projected to reduce dramatically to 5.5, 2.8 and 1.2 yr under RCP2.6, RCP4.5 and RCP8.5, respectively, which will lead to a severe increase in extreme warm events. The largest reductions in return periods are found in the western China regions (SWC1, SWC2 and NWC), followed by Central China (CC). The model spreads tend to be lowest by the end of the century and under RCP8.5.

Cold extremes are projected to be much less frequent in the future. By the end of the 21st century, the multi-model median return period of TNn increases to over 100 and 500 years under RCP2.6 and RCP4.5, respectively, and to infinite numbers in RCP8.5, essentially indicating the disappearance of such events. Inter-model uncertainties in return periods of TNn are in general greater than those of TXx (figures not shown for brevity).

Moving our attention to the precipitation extremes, the CMIP5 multi-model median changes of RX5day-50 and CDD-50 at the end of the 21st century under the different scenarios are presented in Fig. 6. With a mixture of positive and negative changes under RCP2.6, a dominant increase in RX5day-50 can be found under the higher forcing scenarios of RCP4.5 and RCP8.5. The increase under RCP8.5 is mostly in the range of 10%-25% throughout the country, with the largest increase of up to 50% found in Yunnan Province of Southwest China.

The pattern of CDD-50 changes shows, to some extent, consistencies across the scenarios, characterized by a dipole structure of increases over most of the north and decreases in the south. The dipole pattern is most evident in the RCP8.5 scenario, where, by the end of the century, the CDD-50 is 10-25 d longer than in the present day over most of the southern areas. Conversely, the CDD-50 is 10-25 d shorter than present day in the north. We stress that the increase in CDD-50 over southern China is combined with an increase of RX5day-50, indicating a shift toward a regime of greater occurrence of both flood-producing and drought-producing events over the region (Giorgi et al., 2011, 2014b). This response is most pronounced in Southwest China, where the largest changes in both RX5day-50 and CDD-50 are found.

Figure 6.  The CMIP5 multi-model median changes in RX5day-50 (units: %) (left) and CDD-50 (units: d) (right) by the end of the 21st century (2080-99) relative to 1986-2005 under RCP2.6 (top), RCP4.5 (middle) and RCP8.5 (bottom).

Figure 7.  Box-and-whisker plots for the multi-model CMIP5 ensemble changes in (a) RX5day-50 and (b) CDD-50 over China and the eight sub-regions by the end of the 21st century (2080-99) relative to 1986-2005 under different scenarios (units: % and d). The colors indicate the different scenarios, with blue for RCP2.6, green for RCP4.5 and red for RCP8.5. The model median is indicated by the bars.

Late 21st century projected changes in precipitation extremes (RX5day-50 and CDD-50) under the three scenarios over China and its sub-regions are summarized in Fig. 7. For RX5day-50, the median change is positive in all regions and scenarios, as is the interquartile range (except for NC and SC under RCP2.6). The full inter-model spread is relatively large, but still above the zero line in the majority of regional cases. The largest multi-model median increase for RCP8.5 is found in SWC2 (42%), and the minimum in NC (24%). The inter-model spreads are in general wider under RCP8.5 compared to RCP4.5 and RCP2.6, with the largest inter-model spread occurring over SC and SWC2. The median changes of RX5day-50 averaged over China are 7%, 15% and 29% under RCP2.6, RCP4.5 and RCP8.5, respectively. Thus, the signal of increase in wet extremes over the country is robust.

For CDD-50, the dipole structure of the change identified above produces predominant declines in the northern regions (NEC, NC, NWC) and increases in the central and southern ones, albeit with more inter-regional variability than for the wet extremes. The interquartile and full inter-model spreads are large and in most cases cross the zero line, indicating a lack of agreement in the sign of the response across the ensemble. When averaging over the entire China territory, we find a slight reduction in the median of 3.5, 4 and 5 d under RCP2.6, RCP4.5 and RCP8.5, respectively. Therefore, a larger uncertainty in the response of dry extremes than wet extremes is found in the CMIP5 ensemble.

Figure 8.  Box-and-whisker plots for the multi-model CMIP5 ensemble return periods of the present-day (1986-2005) (a) RX5day-50 and (b) CDD-50 values in different periods (2016-35, 2046-65, 2080-99) of the 21st century over China and the eight sub-regions under different scenarios (units: yr). The colors indicate the different scenarios, with blue for RCP2.6, green for RCP4.5 and red for RCP8.5. The model median is indicated by the bars.

Sub-regional and China-wide changes in return periods of present-day RX5day-50 and CDD-50 are summarized in Figs. 8a and b, respectively. A reduction in RX5day return periods is projected by all models under the different future scenarios (Fig. 8a). Averaged over China, the median return period of RX5day is reduced from 50 yr in the present day to ∼20 years during 2016-35 under the three scenarios, with a further drop to 17, 13 and 7 yr by the end of the century under RCP2.6, RCP4.5 and RCP8.5, respectively. The largest decrease in the return period is found over the Tibetan Plateau (SWC1), with a value of 4.8 yr in 2080-99 under RCP8.5. The model spreads are in general small by the end of the 21st century under RCP8.5, except for a few models projecting very large values over SWC2 and SC.

The changes in the CDD return periods are smaller compared to RX5day, and show less evident scenario dependence (Fig. 8b). When averaged over China, the median of the CDD return period during 2016-35 is reduced from the present-day value of 50 yr to ∼ 32 yr under the three scenarios. By the end of the 21st century, the projected return periods are 38, 36 and 29 yr under RCP2.6, RCP4.5 and RCP8.5, respectively. For the different sub-regions, decreases in the return periods in the south and increases in the north are found, corresponding to the dipole pattern shown in Fig. 6. The largest increase, up to ∼85 years, is found in NEC, and the largest decrease, ∼13 years, in SWC2, under RCP8.5. The inter-model spreads are in general larger than RX5day.

This paper investigates the performance of a CMIP5 global model ensemble in simulating present-day 50-yr return values of temperature and precipitation extreme indices (TXx, TNn, RX5day and CDD), along with the corresponding projected changes in return values and return periods (reoccurrence times) under different greenhouse gas concentration scenarios over China in the 21st century. The main conclusions can be summarized as follows:

(1) The CMIP5 multi-model median estimates of the 50-yr return values of TXx and TNn (TXx-50 and TNn-50) in general agree with observations over China, albeit with a tendency for overestimation of TXx-50 and underestimation of TNn-50. The differences between model simulations and observations are usually within a few degrees, except over areas with complex topography. Better performance and smaller inter-model spread are found for TXx-50 compared to TNn-50. The simulation of 50-yr return values of precipitation extremes (RX5day-50 and CDD-50) is not as good as that for temperature. A predominant underestimation of RX5day-50 in South China and overestimation in North China can be found, indicating a relatively low performance of the models in reproducing the monsoonal precipitation over the region. Generally, an underestimation of CDD-50 is found over most of China.

(2) Significant increases in TXx-50 and TNn-50 are found in the 21st century. The increase in TXx-50 is spatially more evenly distributed throughout China than TNn-50, while larger increases in TNn-50 are found in Northeast China and the Tibetan Plateau. Substantial decreases in the return periods of the present-day TXx-50 are projected in the future, with values reduced from 50 to 1.2 yr under RCP8.5 by the end of 21st century. This is indicative of much more frequent extreme warm events in the future. Conversely, extreme cold events become rarer, and in fact the present-day 50-yr return period events may no longer exist by the end of 21st century under RCP8.5.

(3) General increases in RX5day-50 values are projected in the future, being more pronounced under the higher emission scenarios and in the late century, evenly distributed across the country, and with good inter-model agreement. The change in CDD-50 shows a dipole structure, with a reduction in the north and increment in the south. The median of the return periods for RX5day decreases from 50 yr in the present day to only a few years by the end of the century under RCP8.5, while for CDD the return period increases in the north and decreases in the south. Therefore, an increase in both wet and dry extremes can be found over southern China.

Finally, although the resolutions of the CMIP5 models are in general higher than those of the previous-generation CMIP3 ensemble, they are still insufficient to reproduce the monsoonal climate over East Asia well (Gao et al., 2006, 2008); and in fact, our analysis shows that GCMs need to be improved to better simulate precipitation extremes especially. In this respect, we plan to extend our study to the multi-RCM ensemble simulations performed under the CORDEX framework (Giorgi et al., 2009; Gao et al., 2016). Future studies should also include more targeted analyses to better understand the physical processes underlying both the model biases and the changes in the regional distribution of extremes over the Chinese territory.