Changing Trends and Abrupt Features of Extreme Temperature in Mainland China from 1960 to 2010

Studies based on the 10th (90th) percentiles as thresholds have been presented to assess moderate extremes in China and globally. However, there has been notably little research on the occurrences of high extremes of warm days and hot days (TX95p and TX99p) and cold nights and very cold nights (TN05p and TN01p), based on the 95th and 99th (5th and 1st) percentiles of the daily maximum (minimum) temperature data at a certain station in the period 1971–2000, which have more direct impacts on society and the ecosystem. The trends analyses of cool nights or warm days are based upon the hypothesis that expects a linear trend and no abrupt change. However, abrupt changes in the climate, especially in extreme temperatures, have been pointed to as a major threat to ecosystem services. This study demonstrates that (1) the mean frequencies of TX95p and TX99p increased by 1.80 day/10 year and 0.62 day/10 year, respectively, and that those of TN05p and TN01p decreased by 3.18 day/10 year and 1.01 day/10 year, respectively, in mainland China. Additionally, the TX95p and TX99p increased significantly by 50.42% and 58.21%, respectively, while the TN05p and TN01p of all of the stations decreased significantly by 83.76% and 76.48%, respectively. Finally, (2) the TX95p and TX99p trends underwent abrupt changes in the 1990s or 2000s, but the trends of TN05p and TN01p experienced abrupt changes in the late 1970s and early 1980s. After the abrupt change points, the trend of warm and hot days increased more rapidly than before in most regions, but the trend of cold days and very cold days decreased more slowly than before in most regions, which indicates a greater risk of heat waves in the future.

has more directly impacts on society and ecosystem systems. The study showed: (1) the frequencies of TX95p and TX99p averagely increased by 1.80 days/10 a and 0.62 days/10 a respectively in all stations of mainland China, and TX95p in 50.42 % and TX99p in 58.21 % of the stations increased significantly, but TN05p in 83.76 % and TN01p in 76.48 % of stations decreased significantly, and the frequencies of TN05p and summers can drastically reduce agricultural production (Asseng et al., 2011;Farooq et al., 2011), increase energy consumption (Hadley et al., 2006), and lead to hazardous health conditions (Dematte et al., 1998;Pantavou et al., 2011). Thus, understanding and predicting the spatial and temporal variability and trends of extreme weather events is crucial for the protection of socio-economic well-being, and is also crucial for under-5 standing extreme weather events and mitigating its regional impact.
To analyze the variations in extreme climate, Expert Team of Climate Change Detection, Monitoring and Index (ETCCDMI) defined 27 extreme temperature and precipitation indices (Klein Tank et al., 2009). Two main types of extremes indices were developed by the calculation of the number of days in a year exceeding specific thresholds that have fixed values (absolute thresholds) and thresholds that are relative value (percentile thresholds) to a base period climate (Zhang et al., 2011). These indices of the number of days above or below percentile thresholds are more suitable for spatial comparisons of extremes than those based on certain absolute thresholds (Klein Tank et al., 2009). Thus, extreme temperature indexes which based on minimum temperature below the longterm 10th percentile and/or maximum temperature above the longterm 90th percentile were widely used and published in global scale, North America, South America, Europe, Asia and Australia (Alexander et al., 2006;Bonsal et al., 2001;DeGaetano and Allen, 2002;Klein Tank and Können, 2003;Zhou and Ren, 2011;Kothawale et al., 2010;Aguilar et al., 2005;Rusticucci, 2012;Nemec et al., 2013). 20 Most of the researches based on the 10th (90th) percentiles as thresholds set to assess moderate extremes that averagely occur 36.5 times every year (10 percentage of 365 days) rather than high impact, once or twice-in-a-year weather events. Compared to moderate extremes, the high extremes temperature that based on 5th or 1st (95th or 99th) percentiles have higher potential risks on people's health and lifestyles, the 25 economy, society, and the environment. However, there has been very little research reported on the occurrences of high extremes warm days and cold nights according to 5th or 1st (95th or 99th) percentiles. More than 70 % of the Earth's land area underwent a significant reduction in the number of cool nights but insignificant increase in warm days (Alexander et al., 2006). But in China, regional analyses reported that a significant reduction occurred for cool nights and a significant increase occurred for warm days (You et al., 2013;Liang et al., 2014;Wang et al., 2013;Yu and Li, 2015). Along with these regional analyses in China, 5 the changes are much less spatially coherent even though significant trends are found in more than half of stations in the entire China mainland. Further updated studies need to amplify the spatially heterogeneous of temporal-trends on temperature extremes among different climate regions in China.
In this paper, two questions are studied in terms of temperature extreme: how the 10 temperature extreme trends spatially distributed in different regions in mainland China; when the change point of temperature extreme trends were happened in the annual t series during 1960-2010. In this study, the 5th (95th) and 1st (99th) percentile were individually chosen to get the thresholds of 4 indices (TX95p, TX99p, TN05p, TN01p) based on the 95th and 99th percentiles of daily Tmax and at 5th and 1st of daily Tmin, 15 respectively. The spatial heterogeneity and abrupt features of temporal-trends of the 4 indices was embodied among nine climate regions in mainland China during 1960-2010. 20 Daily temperature records were provided by the National Meteorological Information Center of China Meteorological Administration (CMA), including maximum and minimum surface temperature records of China from 1 January 1960 to 31 December 2010. A series of control methods were employed and the errors were corrected by the National Meteorological Information Center, which includes extreme value control and 25 consistency check (Liu and Li, 2003;li and Xiong, 2004 good quality data were chosen to use to analyze (Fig. 1). Data homogeneity was tested by the software RHtest V3 (http://etccdi.pacificclimate.org/software.shtml). In this study, no direct relationship between the year of data inhomogeneity and metadata was found and no adjustment was attempted for any stations.

5
The spatial heterogeneity of temporal-trends need define the climate zones in China. Based on climate zones of China that calculated by using monthly temperature and precipitation data (Zhang and Yan, 2014), considering the coincidence with administrative division of province, mainland of china was regionalized into 9 climate zones as follows (

Extreme temperature indices
We used 5th (95th) and 1st (99th) percentile were individually chosen to get the thresholds of the 4 indices (Table 1), which is different with the indices recommended by the Expert Team on Climate Change Detection and Indices (ETCCDI) that used 10th percentile to define the occurrences of cold nights (days) and warm days (nights

Trend analysis, significance test and abrupt changes detection
Annual time-series of the indices were calculated for each station. Trends in the annual indices were calculated using linear trend estimated for each station, using all available years from 1960 to 2010. Statistical significance of the trends is evaluated at the 5 % level of significance against the null hypothesis.

5
The possible abrupt changes in trends of the indices have been examined by using Mann-Kendall method and 5 yr moving T test (MTT) method (Alexander et al., 2006). For the emergence of multiple point mutations after using Mann-Kendall method, the 5 yr moving T test method were used to verify the authenticity of the mutation point to enhance the credibility of the results (Klein Tank and Können, 2003;Wei, 2007).

Trends of temperature extreme
This section gives an overview of time series for 4 temperature extreme indices averaged by all the stations in mainland of China (Fig. 2). Figure 3 summarizes the results of anomalies in annual temperature extreme indices averaged by all the sta- 15 tions in mainland of China. The warm days (TX95p) and hot days (TX99p) days underwent increase trends in recent 51 years, with a linear trend +1.8 day/10 a and linear trend +0.62 day/10, respectively; and the cold nights (TN05p) and frozen nights (TN01p) showed downward trends with a linear trend −3.18 day/10 a and linear trend −1.01 day/10 a, respectively (Fig. 2). Both warm days and hot days days showed rapid 20 increases trends after mid-1980s although slight decrease tendencies appeared before mid-1980s, and cold nights and frozen nights almost fallowed downward trends despite a short-term upward trend before 1970s (Fig. 3)

Spatial variations of trends in extreme temperature
This section showed the trends' spatial variations of 4 indices in mainland China (Fig. 4) AND the stations' percentage of trends and passed the significant test (P < 0.05) among 9 climate zones (Table 2).

5
TX99p and trends TX95p have similar spatial patterns, although the trend values may be different, and the TX95P and TX99P showed an increasing trend in almost all the regions of mainland but Central China (CC) and its surrounding areas where presented a decreasing trend or insignificant trends ( Fig. 4a and b). Except the junction area of North and Northwest China (NC and NWC), the northwest of Southwest China (SWC), 10 and a few sites in northeast of East China (EC), most of the site was to reduce TN05p and TN01p trend ( Fig. 4c and d). Both the largest increases in TX95P and TX99P frequency, and biggest decreases in TN05p and TN01p, are mostly located in similar regions in NC, NWC, WC, TP, SWC, SC north of CC, WC, especially in up reaches of Yellow River and Yangtze River, and estuary of Yangtze River.

Spatial variations of significantly trends for 4 indices in China
TX95p and TX99p in more than half of the stations increased significantly, and TN05p and TN01p in more than three-quarters of stations decreased significantly (

Characteristics of mutation in TX95p and TX99p trends
The mutation in TX95p and TX99p trends were mainly in 1990s and 2000s (Table 3). 10 The abrupt changes in trends of TX95p early occurred in the 1980s in NEC and CC, and latest occurred in the 2000s in the SWC, SC and EC, and mutations in the 1990s was NC, TP, WC and NWC. The TX99p mutation early occurred in SC and CC in the mid-1980s and latest occurred in the 2000s in the NEC and SWC and EC, and the observed abrupt changes in the 1990s was NC, TP, WC and NWC (Table 3). 15 All the trends values of TX99p among the climate zones after mutation were bigger than before mutation, but the trends values of TX9p in more than half the number of climate zones (NEC, NC, WC, NWC and EC) after mutation were bigger than before mutation (Table 3). 20 Except for TP that mutation of TN05p occurred in 2001, other climate zones' mutation of TN05p and TN01p occurred in the 1980s and the end of 1970s (Table 4). The decreasing trends of TN01p among the climate zones after mutation were shrunk than before mutation except for WC, but the decreasing trends of TN05p in more than half the number of climate zones (NC, TP, SC, NWC and EC) after mutation were shrunk

Discussion
The last IPCC report points out that there is an increasing concern about temperature extremes, which are expected to be more frequent (IPCC, 2012). In mainland China, this study showed that the frequencies of TX95p and TX99p averagely increased by 1.80 days/10a and 0.62 days/10 a respectively, and the frequencies of TN05p and 5 TN01p averagely decreased by 3.18 days/10 a and 1.01 days/10 a respectively, and Zhou and Ren (2011) show a increase at a rate of 5.22 days/10 a occurred for warm days (TX90p) and a reduction at a rate of −8.23 days/10 a occurred for cool nights (TN90p) during 1961-2008. The increase rates of warm days in 90th, 95th and 99th percentiles are much less than the decrease rates of cold nights in 10th, 5th and 1st 10 percentiles, respectively, which seems to be associated with asymmetric warming characteristic that the rate of increase in daily minimum temperature is significantly higher than that of daily maximum temperature (Easterling et al., 1997;Karl et al., 1993;Vose et al., 2005). More warm days increase and less cold nights were also detected in different seasons or subareas of China (You et al., 2013;Li et al., 2010;Zhai and Pan, 15 2003;Shi and Cui, 2012). The increase (decrease) of warm days (cold nights) whether 90th (10th), 95th (5th) or 99th (1st) percentiles in China are consistent with all other global or regional studies that shown that the occurrence of warm days increased, but cold days decreased (Bonsal et al., 2001;DeGaetano and Allen, 2002;Klein Tank and Können, 2003;Alexander et al., 2006). 20 This study showed the frequency of warm and hot days was an increasing trend but the cold and frozen days was a decreasing tendency in almost everywhere except for Central China and its surrounding areas where the warm and hot days tended to decrease. In Central China, T max or warm days in summer shows a cooling trend also were found by Qi and Wang (2012) and (You et al., 2013). The trends of TX95p and TX99p  . The research showed that after the mutation, the increasing rate of warm days and hot days is much greater than before in most areas of China. And extreme warm days (with daily maximum temperature > 35 • C) increased significantly in most of China during 1961during -2007during (Ding et al., 2010. Ensemble multi-model projected more extreme warm events and less cold events are expected over China in future under the RCP4.5 scenario (Yao et al., 2012). All these indicated that more potential risk of heatwaves in future, which not only affects human health and disease but also can change the probability of agrometeorological disasters. However, crops' growth duration was shortened, as well as the sowing date or the phenology was shifted because of climate warming (Fang et al., 2015(Fang et al., , 2013Richardson et al., 2013), limited knowledge in crops response 10 to climate warming and deserve more attention in the future.

Conclusions
1. It is showed that warm days and hot days underwent an increase trend in recent 51 years, and a rapid increase after mid-1980s. The cold days and frozen days underwent a decrease trend in recent 51 years, and a rapid decrease from 1960s 15 to 1990s.
2. The warm days (cold days) and hot days (frozen days) showed an upward (downward) tendency in most area of China, but Central China and its surrounding areas showed an decline tendency in warm days and hot days.