Yang Yang a,b,c,Zhao hui Lin a,b,*Lifeng Luosc,Yan Zhang c，Zhen Li'd

a Intermational Center for Climate and Enironment Sciences, Institute of Atmospheric Physics, Chinese Academy of Sciences, Bejing, China

b University of Chinese Academy of Sciences, Beijing, China

c Department of Geography, Environment, and Spatial Sciences, Michigan State University, East Lansing, MI, USA

d Key Laboratory of Regional Climate Environment for Temperate East Asia, Instute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China

cept the southwest part. In these studies, the relative thresholds based on temperature percentiles were adopted for identifying FirstEHD, which can ensure the FirstEHD is obtained for all stations, but the shortcoming of the method is that it is not convincing for comparing FirstEHD over different parts of China, as threshold temperatures vary with stations or regions. It is also well recognized that common heat alerts are usually based on the absolute temperature thresholds ( Amengual et al.,2014 ; Smith et al., 2013 ); for example, a

T

max of 35°C is the threshold for the issue of heat warnings by China Meteorological Administration(CMA). However, You et al. (2017) found that the choice of heat waves definition, based on relative and absolute threshold values respectively,can lead to contrasting trends for the frequency and duration of heat waves. So, it is worthwhile to investigate the spatiotemporal distribution of FirstEHD and its trend in China, based on a fixed absolute temperature threshold. It has also been noted that there exist decadal changes in extreme hot events across China (e.g., Sun et al., 2011 ), as well as the trend of temperature and heat wave frequency in the 1990s ( Jia and Hu, 2017 ; Qian et al., 2011 ; Wang and Gong, 2000 ). It is interesting to try to understand whether such decadal changes could also be found for the FirstEHD trend in China.Previous studies have attributed the long-term trend of temperature extremes to anthropogenic forcing, such as the greenhouse gas emissions, irrigation, and urbanization effects ( Lobell et al., 2008 ; Sun et al.,2016 ; Wang et al., 2020 ). It has been suggested that increasing temperature variability also plays an important role in summer heat waves over Europe ( Schär et al., 2004 ). Wang et al. (2020) further pointed out that the changing variabilities in temperature can affect the trends of compound hot extremes over China. In this paper, we would also like to evaluate whether the

T

variability can have an impact on the longterm changes of FirstEHD.

To sum up, the paper is organized as follows: Section 2 briefly describes the data and methods used. Main results are summarized in Section 3 , and then conclusions and discussions are presented in Section 4.

2. Data and methods

The

T

dataset used in this study is CHTM4.0, a homogenized station temperature dataset with records from 754 stations in mainland China (Fig. S1) spanning from 1960 to 2018. This dataset is an updated version from its predecessor, CHTM3.0, which has been described in Li et al. (2016) . The original data used to generate the CHTM4.0 homogenized dataset are based on those from 825 observational stations provided by the National Meteorological Information Center (NMIC)of CMA (available at http://data.cma.gov.cn ), which were obtained after initial quality control from 2419 weather stations across China by NMIC. Meanwhile, the homogenization method (including change-point detection, missing-value interpolation, etc.) is Multiple Analysis of Series for Homogenization (MASH) with the latest version program of MASHv3.03. Detailed information on the homogenization procedure can be found in Szentimrey (1999) and Li et al. (2016) . Following the temperature criteria for issuing the heat alert by CMA, the absolute threshold of 35°C was adopted for identifying EHDs in China, which has also been used in many other studies (e.g., Li and Huang, 2011 ;Sun et al., 2018 ; Yu et al., 2021 ). Hence, the first occurrence date of EHDs during a calendar year is identified as FirstEHD in this study.The non-parametric Theil—Sen (T—S) slope method ( Sen, 1968 ;Theil, 1950 ) was adopted to compute the trends of regionally averaged FirstEHD series, and then we performed the non-parametric Mann—Kendall (M—K) test for examining the significance level of trends( Mann, 1945 ; Kendall, 1975 ). To guarantee the robustness of analysis related to FirstEHD, we filtered out those stations with EHDs occurring for fewer than 40 years during 1960—2018, and this left us with 415 (out of 754) stations for analysis (Fig. S1). The moving

t

-test method was employed to examine the possible changes in trends ( Gentle, 1982 ).

Five general circulation indices from the National Climate Center of CMA were adopted to explore the possible reasons for the decadal differences in FirstEHD trends. Their definitions are listed in Table S1 (available at http://data.cma.gov.cnhttp://cmdp.ncccma.net/Monitoring/cn_index_130.php ), including the western Pacific subtropical high (WPSH) intensity, area and western ridge point indices in June, and the Tibetan Plateau index and East Asian trough index in May, during 1960—2016.

In order to estimate the respective roles of temperature trends and changing temperature variabilities in determining the trend of FirstEHD,we recalculated the trend of FirstEHD based on the

T

max residuals after removing the linear trend from

T

series at each meteorological station, as suggested by Wang et al. (2020) . The trends for the recomputed FirstEHD were assumed to be dictated by evolving variabilities of

T

,while the remaining proportion of trends in FirstEHD from the original series was assumed to be ascribed to the long-term trends of

T

.

3. Results

3.1. Climatology of FirstEHD over China

In order to identify the spatial pattern of EHDs, we firstly show the climatological distribution of the annual total EHDs at all meteorological stations in China averaged over 1960—2018 ( Fig. 1 (a)). We can see that the annual EHDs vary substantially across China, with some regions frequently affected by heat waves while others in cooler climate zones (such as northeastern China) hardly experiencing any. In eastern China, annual total numbers decrease northwards from around 40 EHDs in southern China to about 10 EHDs at stations over the North China Plain, and then diminish further north. This variation is largely driven by the latitudinal gradient related to solar radiation and general circulation. Regional topographic features and land cover also matter.For example, several stations inside the Sichuan Basin show more EHDs(around 40 days) than other stations at the same latitude. In Xinjiang,large annual total of EHDs can also be found, which is mainly ascribed to the arid climate in this region, with

T

easily exceeding 35°C.

Based on the selected 415 stations with EHDs occurring for more than 40 years, the climatological distribution of FirstEHD averaged over 1960—2018 is presented in Fig. 1 (b), and the FirstEHD has been aggregated into half-month intervals. It is found that FirstEHD dates vary remarkably across mainland China. In particular, FirstEHD occurs in March for four stations in South and Southwest China, including Yunnan, Guangxi, and Hainan provinces. However, for another eight stations in the same region, FirstEHD is generally found in April, and this can be ascribed to the complex terrain in these regions, which probably plays a crucial role in affecting the local distribution of temperature.

During the month of May, EHDs start to occur over stations scattered in east Xinjiang, the North China Plain, and Sichuan Basin, and most of them take place in the second half of May. Then in June, about two-thirds of the total stations concentrated in eastern China have their FirstEHDs. Since June marks the beginning of the summer season (i.e.,June—August) in the region, it is reasonable to expect a large number of EHDs begin to break out in this period. Lastly, the remaining stations along the coast and Yangtze River, and in Northwest China, see their FirstEHDs in July, mostly during the first half. Like

T

, FirstEHD can be easily affected by both large-scale and local factors. As a result, the overall spatial pattern of FirstEHD is quite complex. Stations in same region may experience their FirstEHDs many days apart. The stations along the coast, although in lower latitudes, generally have a lower Tmax due to mesoscale sea breezes during the day; hence their FirstEHDs are much later than those of inland stations at the same latitude.

3.2. Inhomogeneous trend of FirstEHD

Fig. 1. (a) Annual total EHDs (units: days) and (b) FirstEHD (units: Julian days) averaged over 1960—2018 in China. The dates in (b) are classified into the first and second half of a month. Triangles (crosses) represent the corresponding stations belonging to the first (second) half of a month. Key provinces mentioned in the study are displayed and listed (in the legend on the right) in (b).

Fig. 2. The long-term trend of FirstEHD during 1960—2018 (units: d/10 yr). Blue boxes denote the four key regions: Xinjiang (XJ; 36°—43°N, 75°—90°E); North-Central China (NC; 33°—42°N, 114°—120°E); Yangtze River Basin (YR; 26°—33°N; 108°—122°E); and South China (SC; 21°—25°N, 107°—120°E). Solid dots indicate the sites whose linear trends are significant at the 95% confidence level based on the two-tailed Student’s t -test.

Table 1 The regional mean FirstEHD and its trend over four key regions during 1960—2018.

In Fig. 2 we present the trends of FirstEHD during 1960—2018. Negative trends are observed in most parts of the country, indicating that FirstEHD happened gradually earlier in recent decades over these regions. Generally, stations with significant trend at the 95% confidence level are mostly located in southeastern China, and a limited number of stations with significant trends can be found in Xinjiang. This is not surprising because global warming is expected to raise surface air temperature in most regions; hence it is easier for

T

max to exceed the threshold at the beginning of warm season, thus pushing the FirstEHD earlier.However, one unique spatial feature is that a group of stations between the lower Yellow River Basin and lower Huaihe River Basin show weak positive trends, implying that FirstEHD there has been delayed, which is remarkably different from the other areas.

To understand the spatial difference in the long-term trend of FirstEHD across the country, we defined four key regions that are outlined in Fig. 2: Xinjiang (XJ), Yangtze River Basin (YR), South China(SC), and North-Central China (NC). These regions are more or less homogeneous in terms of their climate and other surface characteristics.

The regional mean results in Table 1 show that the mean date of FirstEHD is 17 June over SC, and FirstEHD advanced the most during 1960—2018, with a rate up to − 4.25 d/10 yr. The YR region, the largest areas among the four, is similar to SC in this regard since its mean FirstEHD is also in the second half of June (i.e., 21 June), and the mean trend during the period is − 2.49 d/10 yr, which is significant at the 95% confidence level. FirstEHD in XJ occurs on 8 June on average,the earliest of the four, but its trend is slightly weaker with a rate of− 2.21 d/10 yr.

NC shows an opposite trend to the other regions. Regional mean FirstEHD here happens on 10 June, but the FirstEHD in this region has been shifting later with a positive trend of 0.57 d/10 yr during 1960—2018. Among the total 46 stations in this region, five of them (Table S2) show statistically significant positive trends for the onset date of EHDs, and they are mainly located at the juncture of Shandong, Henan,Jiangsu, and Anhui provinces, as well as along the lower reaches of Yellow River. This could possibly be ascribed to the cooling trends of maximum temperature in this region, as can be found in Fig. S2, and this is also consistent with previous studies ( Li et al. 2017 ). The cooling trend of

T

in this region could possibly be explained by the enhanced cooling effect of aerosol pollution since the 1960s, as demonstrated by many previous studies (e.g., Qian et al., 2006 ; Wang et al., 2012 ).

3.3. Decadal differences of FirstEHD trend

It has been found that there was a significant decadal transition of temperature extremes over China in the 1990s (e.g., Qi and Wang, 2012 ). In order to understand whether there exists a decadal difference in FirstEHD trend throughout the study period, we firstly present the time series of FirstEHD in the abovementioned four regions.

Fig. 3. Regional mean time series of FirstEHD in the (a) XJ, (b) YR, (c) SC, and (d) NC regions. Blue and green curves show the time series before and after the transition year (i.e., 1997), respectively. Red lines depict the linear trends before and after the transition year. Purple line depicts the linear trend over the whole period of 1960—2018. The trend was calculated and tested by T—S slope and M—K test, respectively.

From Fig. 3 we can find that FirstEHD and its trend are generally not the same during the whole study period, either in terms of the magnitude or trend direction. We then conducted a nine-year moving

t

- test for the FirstEHD series (Fig. S3) by taking China as a whole, and it was found that there is a significant decadal transition around 1997 for the whole region. Then, the FirstEHD trends before and after the transition year over the XJ, YR, SC, and NC regions were calculated, and the results are shown in Table 1 .

Combining Fig. 3 and Table 1 , we find that three out of four key regions show negative trends before the mid-1990s, with − 1.57 d/10 yr in XJ, − 0.77 d/10 yr in YR, and − 1.04 d/10 yr in SC, suggesting that the onset dates for EHDs all advanced during the 1960s—1990s in XJ,YR, and SC regions. Conversely, a positive trend of 2.0 d/10 yr can be seen in NC before the mid-1990s, suggesting a delay of FirstEHD in this region.

After the mid-1990s, SC experiences a significant negative trend of FirstEHD —namely, a substantial advanced onset of EHDs. The rate of the decreasing trend can reach up to − 10.17 d/10 yr, indicating that the sharp decreasing trend after the mid-1990s contributes more to the overall trend pattern during 1960—2018 in the SC region. This is understandable considering the impact from enhanced warming induced by increased greenhouse gases and accelerated urbanization after the 1990s, especially over densely-populated regions.

In the YR region, the FirstEHD trend after the mid-1990s is weakly positive, with a magnitude of 0.75 d/10 yr, suggesting a slightly delayed occurrence of FirstEHD after the 1990s. In XJ, an increasing trend of FirstEHD can also be found after the mid-1990s, with a magnitude of 0.2 d/10 yr. Comparing with the trend before the mid-1990s, we can find that the advancing rates for FirstEHD over both the YR and XJ regions are reversed after the 1990s. However, the overall negative trends in YR and XJ during the whole period suggest that the FirstEHD trend is dominated by that before the mid-1990s.

As for NC, the trend of FirstEHD is − 1.67 d/10 yr after the mid-1990s, which is opposite in sign compared with that before the 1990s.This implies that the overall positive FirstEHD trend in NC during 1960—2018 is mainly dominated by that before the 1990s.

3.4. Possible factors for the decadal differences in FirstEHD trend

It is well understood that the occurrence of heat waves in China is closely associated with anomalous atmospheric circulations over East Asia (e.g., Sun et al., 2011 ). Amongst them, the WPSH is one of the dominant circulations that can have a considerable impact on extreme hot days in the eastern part of the country, especially in the Yangtze River Basin (e.g., Zou et al., 2015 ).

Based on the definitions of three WPSH indices (explained in Table S1), the time series of the WPSH intensity, area and western ridge point indices in June during 1960—2016 were calculated and presented in Fig. S4(a—c). It can be seen that the WPSH intensity and area in June were getting stronger/larger from 1960 to the mid-1990s, and the WPSH western ridge point index was getting smaller (i.e., westward shift in the trend of WPSH), which are all favorable for the advanced trend of FirstEHD in the YR and SC regions. After the mid-1990s, there is no remarkable trend for the above three WPSH indices, which is consistent with the negligible trend of FirstEHD from the mid-1990s to 2018 in YR. Another possible reason for the negligible trend of FirstEHD in YR could be related to the stagnation of global warming from the late 1990s ( Qu et al., 2017 ; Wang et al., 2010 ). It is further suggested that the sharp decreases in FirstEHD over SC could be ascribed to the enhanced urbanization after the mid-1990s over this region ( Luo and Lau,2017 ).

Chen et al. (2011) found that the Tibetan Plateau high pressure(TPHP) system has close connection with high temperatures in northwestern China, and large-scale extreme hot events frequently occur when TPHP is strengthened and moves northward. Fig. S4(d) shows the time series of the Tibetan Plateau index in May, from which we find that the Tibetan Plateau index was getting stronger from 1960 to the mid-1990s, which was favorable for the relatively larger advanced trend of FirstEHD before the mid-1990s in XJ. After the transition period, a weakened trend of the TPHP system can be found, which is also consistent with the delayed trend there of FirstEHD.

Fig. 4. Trends of FirstEHD (units: d/10 yr) contributed by the (a) T max variabilities and (b) long-term trend of T max during 1960—2018. The FirstEHD trend in (a) is based on T max residuals after removing the linear trend from T max at each station, and hence is assumed to be dictated by evolving variabilities of T max . The remaining proportion of FirstEHD trend for the original series is assumed to be ascribed to the long-term trends of T max as shown in (b).

Fig. 5. Possible influencing factors for the FirstEHD trend and its decadal changes over China during the 1960s—2010s.

It has been pointed out that the springtime East Asian trough can have an impact on the springtime climate over North China( Huang et al., 2019 ), so we present the time series of East Asian trough intensity (EATI) in May in Fig. S4(e). An intensified trend of the EATI from 1960 to the mid-1990s can be found, suggesting that there might have been more cold disturbances from higher latitudes affecting the NC region, and this would have been favorable for the delayed trend of FirstEHD over this region before the mid-1990s. After the mid-1990s,no remarkable trend in EATI can be found, and the advanced trend of FirstEHD could be associated with the enhanced urbanization effects, as well as the decreased aerosol emissions in this region.

3.5. Contribution of changing variability to the FirstEHD trend

Using the methodology proposed by Wang et al. (2020) , the trend of FirstEHD was recalculated with the detrended

T

time series( Fig. 4 (a)), and the residual FirstEHD trend associated with the longterm trend of

T

is also presented in Fig. 4 (b).From Fig. 4 we can see that the signal from changing variabilities of

T

(including diurnal variability, intraseasonal—seasonal cycles, interannual variability, etc.) induced a non-uniform FirstEHD trend, with scattered positive and negative trends around China. However, FirstEHD trends induced by the long-term trend of

T

are much more uniform,with negative trends over most parts of China and positive trends mostly located in NC, and the associated FirstEHD trends are generally consistent with the spatial patterns of the

T

trend (Fig. S2). Comparing Fig. 4 with Fig. 2 , we can further find that the spatial pattern of the overall trend in Fig. 2 is more similar to the trend pattern attributed to

T

max trend ( Fig. 4 (b)) than to the pattern attributed to

T

max variabilities ( Fig. 4 (a)). This suggests that the FirstEHD trend during 1960—2018 could be largely explained by the trends in maximum temperature over most parts of China.In order to quantify the relative contribution from

T

variabilities to the total trend of FirstEHD over China, we calculated the percentage contributions for XJ, YR, SC, and NC respectively. It was found that the relative contribution percentages from Tmax variability are 12.9%,17.6%, and 25.9% in the XJ, YR, and SC regions, respectively, while the relative contribution from

T

max variabilities can account for 75.5% of the overall trend in NC. This further implies that the FirstEHD trend is mainly contributed by the long-term changes in

T

, but

T

variabilities can also play an important role, especially in the NC region.

4. Discussion and conclusions

In this study, we investigated the onset dates of EHDs in mainland China during 1960—2018 based on a homogenized station dataset. Results showed that EHDs firstly initiate in Southwest China during March and April. In the following months of May to early August, northern China and Xinjiang, as well as most places in southeastern China successively reach their FirstEHDs. For the FirstEHD trend, an advancing trend is dominant over the majority of China, except for the NC region where there exists a delayed trend of 0.57 d/10 yr during last 59 years.It is further revealed that the

T

max trend contributes most to the advance/delay in FirstEHD over China, while the contribution from

T

variabilities is also non-negligible, and can even play a dominant role for the FirstEHD trend in NC.

Moreover, it was found that the mid-1990s marks a transition period for the FirstEHD trend in China. The different magnitudes and signs of FirstEHD trends before and after the mid-1990s can have different relative importance to the overall FirstEHD trend patterns in four subregions of China during 1960—2018. The possible factors responsible for the decadal changes in FirstEHD trend in the mid-1990s have been discussed and a summary is provided in Fig. 5 . It was found that the decadal changes in FirstEHD trend during the mid-1990s was closely associated with similar decadal changes in the circulation indices. In the YR and SC regions, FirstEHD trend is mainly influenced by the WPSH, while the movement of TPHP system is closely associated with the FirstEHD trend in XJ, and the FirstEHD trend over NC could be linked with changes in the intensity of the East Asian trough. It is further suggested that the long-term trend of FirstEHD could also be influenced by the urbanization effect and aerosol cooling effect, as well as the stagnation of global warming in the late-1990s. Generally, the aerosol cooling effect could be the main reason for the delayed trend of FirstEHD over NC, while the urbanization effect could be an important factor for the sharp advancing trend of FirstEHD in SC, as well as the negative trends of FirstEHD around the Beijing—Tianjin megacities. However, further studies with climate model simulations are needed in order to quantitatively identify the impact from aerosol emission changes and urbanization processes on the trends of Tmax and FirstEHD over different parts of China, and the possible underlying mechanisms.

Disclosure statement

The authors declare no conflict of interest.

Funding

This work was funded by the National Key Research and Development Program of China [Grant number 2017YFA0604304] and the National Natural Science Foundation of China [Grant number 41661144032 ].

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.aosl.2021.100080 .