An interdecadal shift in the number of hot nights around 1997 over Eastern China

In this study, we investigate the interdecadal variation in summer nighttime hot extremes over eastern China using observational daily minimum temperature during 1979–2013. Results show a statistically significant shift in the number of hot nights (NHN) around 1997 with averaged NHN over eastern China of 6 days more during 1997–2013 than 1979–1996. The time series of the first leading Empirical Orthogonal Function mode of the NHN is closely related with sea surface temperature anomalies over the tropical western pacific warm pool, Pacific Decadal Oscillation and Atlantic Multidecadal Oscillation, which all experienced substantial interdecadal changes in the late 1990s. Other factors such as the Urban Heat Island (UHI) effects may also contribute to the interdecadal change of the NHN around 1997.


Introduction
High-temperature extremes have much more substantial effects on human lives and property than mean temperature (e.g. Meehl and Tebaldi, 2004;Fischer et al., 2007;Seneviratne et al., 2014). This was evident in several recent heat waves, which generally refer to three or more consecutive hot days, including the devastating summer of 2003, with 66 000 deaths in Europe (e.g. Luterbacher et al., 2004), the disastrous high-temperature extreme that struck the Russian Federation in 2010, causing over 55 000 deaths (e.g. Matsueda, 2011), and the record-breaking heat wave that swept the China entirely in 2013, resulting in 59 billion Yuan losses (e.g. Xia et al., 2016). Observational data shows that heat extremes continue to increase over land, even during the so-called global warming hiatus period of 1998-2012 (e.g. Seneviratne et al., 2014).
In China, the mean temperature has been increasing, especially after the 1980s (e.g. Li et al., 2012). Temperature extremes become more likely to occur frequently with the increase in mean temperature (e.g. Gong et al., 2004;Ren et al., 2010a;You et al., 2011;Xia et al., 2016). China has not only experienced significant increase in hot days, but also witnessed an increase in the frequency of hot nights in recent decades (e.g. Cao et al., 2013;Ren and Zhou, 2014;Sun et al., 2014). Ding et al. (2010) demonstrated that extreme hot days have increased significantly in most of China, with a rising rate of 4 days per decade over the northern China between 1961 and 2007. Hot days exhibited a notable increase after the late 1990s in China (e.g. Wei and Chen, 2011;Li et al., 2012). Zhou and Ren (2011) found that warm nights increased at a rate of 8.16 days/10 years, which is faster than that of warm days (5.22 days/10 years) in China between 1961 and2008. Previous studies focus mainly on the climatology, interannual variability and trends of high-temperature extremes, particularly on daytime hot extremes in China (e.g. Qian et al., 2007;Ren et al., 2010a;Qian et al., 2011;Yan et al., 2011;Zhang and Wu, 2011;Cao et al., 2013;Huang et al., 2014a;Ren and Zhou, 2014). A few studies have analyzed the interdecadal variations of daytime hot extremes (e.g. Wei and Chen, 2011;Li et al., 2012). Less attention, however, has been devoted to the interdecadal variations of hot extremes at night. In this study, we investigate the interdecadal variation of summer nighttime hot extremes over eastern China, which is the most populated and developed area of China, and further discuss possible physical mechanisms involved.

Data and methods
In this study, we use updated homogenized daily minimum temperature data from the China Homogenized Temperature dataset based on Multiple Analysis of Series for Homogenization (MASH), namely CHTM 3.0 between 1960 to 2013, at 753 national reference meteorological stations in China (e.g. Li et al., 2016). The CHTM 3.0 data was produced using the MASH method to detect and adjust inhomogeneities and using the technique proposed by Li et al. (2015) to maintain physical consistency in temperature series. It has been suggested that CHTM 3.0 is more reasonable and suitable for large-scale climate change analyses in China compared to other datasets (e.g. Li et al., 2016). Previous studies have indicated an abrupt climatic shift around 1980 (e.g. Ding et al., 2008, Tu et al., 2010. In this study, we focus on the period of 1979-2013, after this climatic shift. Nighttime hot extremes were estimated using the number of hot nights (NHN) in summer (June-July-August, JJA) at 431 stations in the east of 110 ∘ E in China (Figure 1). Following previous studies, the NHN index was defined as the number of days at each station where the daily minimum temperature meets or exceeds the long-term mean 90th percentile of daily minimum temperature (e.g. Fischer et al., 2007;Zhang and Wu, 2011). For each of the 35 years analyzed between 1979 and 2013, the 90th percentile was based on the 92-day JJA period. The long-term mean 90th percentile was calculated as the mean of the 90th percentile daily minimum temperatures over these 35 years. Similar conclusions were obtained using the thresholds of the 95th and 99th percentiles for the NHN, therefore not presented. The sea surface temperature (SST) data was acquired from the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST (ERSST v4; e.g. Huang et al., 2014b) from 1979 to 2013. The Atlantic Multidecadal Oscillation (AMO) index is derived from http://www.esrl.noaa.gov/psd/data/timeseries/AMO/ for the period of 1979-2013, while the Pacific Decadal Oscillation (PDO) index for the same period is obtained from http://research.jisao.washington.edu/pdo/. The geopotential height and total column water vapor data are acquired from the European Center Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis datasets (e.g. Berrisford et al., 2011) from http://apps.ecmwf.int/datasets/data/interim-full-daily/ levtype=sfc/ for the period covering 1979-2013.
Empirical Orthogonal Function (EOF) analysis (e.g. Lorenz, 1956) was used to obtain the first dominant mode of the NHN and associated time series over eastern China during 1979-2013. The EOF analysis was performed based on the covariance matrix of normalized departures of NHN. The fast Fourier transform (FFT) method was used to separate the relevant time series of the NHN and SST on different scales. The Mann-Kendall (MK) technique (e.g. Mann, 1945;Kendall, 1975) was used to test the abrupt change point of the trend (see Appendix). According to the theory of the MK's test, when the value of statistic UF is positive, it indicates that the factor being tested has an increasing trend. In contrast, a negative value indicates a decreasing trend. Furthermore, the trend passes the 95% confidence level when the value of UF is greater (lower) than 1.96 (−1.96). Figure 2 shows the spatial variation of the first leading EOF mode (EOF1, Figure 2(a)) and the climate mean This indicates that nighttime hot extremes occurred more in the middle than other parts of eastern China. The spatial distribution of the climate mean state of the NHN shows that NHN also occurred more around the middle part of eastern China between 1979 and 2013 ( Figure 2(b)), which overall coincides with the spatial pattern of EOF1. Besides, high occurrence areas where the occurrence of the NHN exceeds 10.5 days are also scattered across the North China.

Discussion of possible physical mechanisms
SST anomalies may be one possible cause for the interdecadal increase of the NHN after 1997 by influencing atmospheric circulation and climate anomalies (e.g. Xie et al., 2009;Li et al., 2010;Wu et al., 2010;Huang et al., 2011). SST anomalies can affect atmospheric circulation via the air-sea interactions including convective activities and surface energy fluxes (e.g. Xie et al., 2009;Hu et al., 2013). Air-sea interactions force the atmosphere to adjust itself through developing anomalous anticyclone or cyclone. Anomalous anticyclones or cyclones subsequently influence surface air temperature by changing the temperature advection, vertical motion and adiabatic heating (e.g. . Figure 5(a) shows the spatial pattern of correlation coefficients of the EOF1 time series of summer NHN with simultaneous SST. The strongest relationship can be seen observed in the Tropical Western Pacific Warm Pool (TWPWP) region (0-20 ∘ N, 130-160 ∘ E). The correlation coefficient between summer mean SST averaged over the TWPWP region and the EOF1's time series is 0.74 which is statistically significant at the 99% confidence level. There is an increasing trend in the time series of SST ( Figure 5(b)). The interannual and interdecadal variations, which are separated by using FFT, of the summer mean SST averaged over the TWPWP region are significant in Figure 5 1979-1996 and 1997-2013 (1997-2013 minus 1979-1996) over eastern China.
the SST is observed around the mid-1990s, both in the time series (Figure 5(b)) and interdecadal variations ( Figure 5(c)).The MK test of the SST (not shown) indicates that an abrupt increase of SST occurs in 1994 and significant increases start from 1997. Summer mean SST averaged over the TWPWP between 1979 and 1996 is about 29.3 ∘ C, while it rises up to 29.6 ∘ C between 1997 and 2013 ( Figure 5(b)). Considering the high correlation between EOF1's time series of the NHN and summer mean SST averaged over the TWPWP, the interdecadal increase of the SST around the mid-1990s may represent an important contribution to the interdecadal increase of the NHN over eastern China.
Previous studies have demonstrated that the SST anomaly in the TWPWP has significant impacts on summer climate over East Asia (e.g. Huang and Sun, 1994b;Huang and Sun, 1994a;Zhao et al., 2002;Zhao et al., 2002;Huang et al., 2006;Zhang et al., 2006). When the SST gets warmer around the TWPWP, the western Pacific subtropical high (WPSH) shifts northward, extends westward and intensifies by the propagation of quasi-stationary planetary waves caused by accelerated convection and the enhancement of a Hadley cell (e.g. Huang and Li, 1988;Huang and Lu, 1989;Huang and Sun, 1994b;Huang and Sun, 1994a;Zhao et al., 2002;Zhang et al., 2006). The interdecadal increase of SST in the TWPWP tends to intensify the WPSH and increase the geopotential height at 500 hPa (Figure 6(a)), and subsequently rise up the daytime surface air temperature through increasing surface solar radiation absorption and enhancing subsidence warming (e.g. Zhang et al., 2005;Baldi et al., 2006;Lei et al., 2009;Wei and Chen, 2011). With daytime temperature rising, the temperature persistence between day and night will contribute to the nighttime high temperature which favors nighttime hot extremes (e.g. Chen and Lu, 2014). The SST anomaly in the TWPWP is closely associated with the atmospheric water vapor in East Asia by influencing the Asian summer monsoon and atmospheric circulation (e.g. Huang and Sun, 1994b;Huang et al., 2006;Zhang et al., 2006;Xiang and Wang, 2013). Previous studies show that positive SST anomalies in the TWPWP intensify local convective activities and create a very strong anticyclone circulation in the lower troposphere over the western Pacific Ocean (e.g. Huang and Sun, 1994b;Huang and Sun, 1994a;Jin and Chen, 2002), which strengthen the Asian summer monsoon (e.g. Jin and . Thus, more water vapor can be transported by the strong Asian summer monsoon flow from the TWPWP into eastern China (e.g. Huang and Chen, 2010). Figure 6(b) shows that the interdecadal increase of SST in the TWPWP tends to increase the nighttime total column water vapor over North China and some other areas of eastern China. The increased atmospheric water vapor condition can result in more downward longwave radiation and warm up the nighttime surface air temperature, favoring the occurrence of nighttime hot extremes (e.g. Wei and Sun, 2007;Gershunov et al., 2009;Ha and Yun, 2012;Chen and Lu, 2014;Lu and Chen, 2016).
Other factors such as the large-scale patterns and anthropogenic forcing may be contributing to the abrupt increase of summer NHN over eastern China. The Atlantic Multidecadal Oscillation (AMO) and the Pacific Decadal Oscillation (PDO) are large-scale patterns in regulating decadal changes of surface air temperature (e.g. Knight et al., 2006;Deser et al., 2010) for many regions such as East Asia (e.g. Zhu and Yang, 2003;Gao et al., 2015;Xia et al., 2016). Correlation coefficients between EOF1's time series of NHN and the AMO and PDO indices are 0.58 and −0.46 (both statistically significant at the 99% confidence level), respectively, for the 1979-2013 period, indicating that AMO and PDO are closely related to NHN over eastern China. In Figure 7, interdecadal variations around the late 1990s are seen in both the AMO and the PDO time series. The long-term mean AMO (PDO) index is about −0.06 (0.8) over 1979-1996 and it changes to be about 0.2 (−0.2) over 1997-2013 (Figure 7). There are substantial interdecadal differences in AMO and PDO between 1997 and 2013 compared to these observed between 1979 and 1996. The changes in AMO and PDO are conducive of the interdecadal increase of NHN after 1997.
The impacts of anthropogenic forcing such as the Urban Heat Island (UHI) effects on nighttime hot extremes in China also cannot be ignored (e.g. Zhou and Ren, 2009;Ren et al., 2010b;Wu and Yang, 2013;Ren and Zhou, 2014;Zhou et al., 2014). Ren and Zhou (2014) indicated that the increasing trend of nighttime hot extremes in summer contributed to by UHI reaching up to 3.435 days (10 years) −1 , which accounts for 33.7% of the overall increase in mainland China between 1961 and 2008. The UHI-induced increase of nighttime hot extremes dominantly appeared after the beginning of the 1990s. Considering that eastern China is densely populated and hosts many megacities, the UHI effects are favorable to the interdecadal variation of NHN. It should be kept in mind that physical mechanisms cannot be fully extracted by using statistical analyses. In the future, further studies on the model simulations to test the proposed mechanisms are needed to explain the abrupt interdecadal variation.

Conclusions
In this study, we investigate the interdecadal variation of summer nighttime hot extremes for the period of 1979-2013 over eastern China. Previous analyses and this study show consistently increasing nighttime hot extremes in recent decades (e.g. Qian et al., 2011;You et al., 2011;Zhou and Ren, 2011). We further identify a statistically significant shift of the number of hot nights (NHN) that occurred around 1997 over eastern China. The averaged NHN over eastern China between 1997 and 2013 was 6 days more than that measured between 1979 and 1996, with most regions having experienced an increase of more than 5 days after 1997. The possible physical mechanisms that may be invoked to explain the interdecadal variation of NHN around 1997 over eastern China are discussed. The time series of the first leading Empirical Orthogonal Function mode of summer NHN is closely related to the simultaneous sea surface temperatures (SST) in the tropical western pacific warm pool (TWPWP) which also experience substantial interdecadal increase around the late-1990s (r = 0.74). The interdecadal increase of SST in the TWPWP tend to increase the geopotential height at 500 hPa and the total column water vapor by intensifying the western Pacific subtropical high (WPSH) and influencing the Asian summer monsoons and atmospheric circulation. Increased geopotential heights at daytime and water vapor at nighttime favored the occurrence of nighttime hot extremes after 1997. The interdecadal variation of NHN is also closely associated with the summer Atlantic Multidecadal Oscillation (AMO) and Pacific Decadal Oscillation (PDO) with correlation coefficients of 0.58 and −0.46, respectively. The same shifts are revealed in both the AMO and PDO indices around the late-1990s. The effects of the Urban Heat Island (UHI) on nighttime hot extremes may also contribute to the interdecadal variation of the NHN. Causes for the abrupt increase of summer NHN, over eastern China, can be mixed and complicated. The proposed physical mechanisms cannot account for the whole of the interdecadal variation around 1997, and other factors such as The Tibetan Plateau (TP) snow and soil moisture may also play roles (e.g. Ding et al., 2009;Zhang and Dong, 2010). These proposed mechanisms explaining the interdecadal variation of the NHN over eastern China need to be further tested and clarified using model simulations in future.

Appendix: Mann-Kendall (MK) test
The MK test is based on the statistic S k defined as follows: and x i are the sequential values and n is the length of the data set. The statistic UF is computed by √ var ( S k ) , k = 1, 2, … , n where UF 1 = 0, the E(S k ) and var(S k ) are mean and variance of S k , respectively, which can be computed as follow: The statistic UF k follows the standard normal distribution and is calculated from x in the time series of x 1 , x 2 , … , x n .
The other statistic UB used in this paper is computed in a similar manner as UF, from x in reverse order (x n , x n − 1 , … , x 1 ), and UB k = −UF k (k = n, n − 1, … , 1) , UB 1 = 0