In this research, we examine the changes in compound extreme heat and precipitation events in South China, by taking Guangdong Province as an example since it possesses the densest population and is the most urbanized province in South China. It is characterized by a subtropical monsoon climate with hot-humid summer and cool-dry winter. Guangdong has experienced rapid urbanization and industrialization since the commence of China’s economic reform and opening-up policy (Xiong et al., 2012). Among all provincial units of China, Guangdong has been holding the largest Gross Domestic Product (GDP) since 1989. Its urbanization level reached 69.85% in 2017. Of Guangdong, the PRD region (as denoted by the red boundary in Figure 1) exhibits the highest urbanization level of 85.29% and the largest population density (Statistics Bureau of Guangdong Province, 2017). Moreover, this area has been severely suffering from dramatic increases in extreme weather and climate events over the past decades (Chen et al., 2015; Lin et al., 2019).

In this study, the compound extreme heat and precipitation events are derived from daily maximum temperature (T_max) and daily precipitation. Observations recorded at 86 meteorological stations in Guangdong from 1961 to 2017 are obtained from the China Meteorological Data Service Center (http://data.cma.cn). The raw data have been homogenized using a statistical approach proposed by Xu et al. (2013). Their temporal inhomogeneity has been evaluated by the Easterling-Peterson method (Li et al., 2004; You et al., 2010). In this study, stations with ≥3 missing days in any month from June to August are excluded.

Compound events are defined as comprising a combination of two or more different extremes occurring coincidentally or sequentially within a certain period of time (Mueller and Seneviratne, 2012; AghaKouchak et al., 2014; Leonard et al., 2014; Wahl et al., 2015). These extremes are considered contributing to complex interactions of multiple hazards such as widespread wildfires (Witte et al., 2011), large-scale air pollution (Konovalov et al., 2011) to human society/ecosystems (Weber et al., 2020). In this study, compound heat and precipitation extremes are defined for each station individually. An extreme precipitation event is first detected when daily precipitation is larger than the 90^th percentile value for all rainy days (≥0.1 mm) in the summers of the reference period of 1961–1990. Then the compound event is counted if the extreme precipitation event is preceded by an extreme heat event within three days. Here, a heat event is defined when daily T_max is larger than the 90^th percentile of the reference period. To quantify the compound events, we adopt a probabilistic metric by using the fraction of the compound events accounting for all extreme precipitation events in a calendar year.

In order to evaluate the possible influences of urbanization on compound events, we classify all meteorological stations into urban and non-urban types, as suggested by previous studies (Mishra et al., 2015; Luo and Lau, 2018; Wang et al., 2019). Stations are tagged as urban type if they are located in urban areas or urban buffers of 25 km that have a population more than 250,000; otherwise, they are classified as non-urban type. The urban area extents are derived from the DeLorme World Base Map dataset (https://www.baruch.cuny.edu/confluence/display/geoportal/ESRI+International+Data), which has been validated by the urban extents extracted from Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data (Mishra et al., 2015).

The urbanization effects are quantified by calculating the differences in the trends between the urban and non-urban series (Ren and Zhou, 2014; Luo and Lau, 2019b). The urban (non-urban) series of the frequency and fraction of compound events are obtained by averaging all urban (non-urban) stations. The secular trend of the series of compound events is estimated by the conventional linear regression, and its significance is evaluated by the modified nonparametric Mann-Kendall (mMK) test. The mMK method considers the autocorrelation in the time series to provide an unbiased evaluation of the trend (Hamed and Rao, 1998). It has been widely used in hydrological and climatological studies (e.g., Luo and Lau (2019a); Sa'adi et al. (2019))

Based on the above definition, we search for the compound events at all stations from 1961 to 2017, and obtain the multi-year mean frequency and fraction of these events. As shown in Figure 2, compound events with extreme precipitation and hot days have occurred in all parts of Guangdong. On average, 28.25% of extreme precipitation events are preceded by a heat extreme within three days, and the study area experiences 1.26 compound events per year. The frequency and fraction of compound events demonstrate obvious spatial variations across the study area. Specifically, compound events are more prominent in densely populated and highly urbanized areas such as PRD, in which the highest frequency and fraction of compound events are observed. The PRD region has 1.49 compound events per year, and 31.81% of its precipitation extremes occur following a previous extreme heat day within a short period. The larger (smaller) frequency and fraction in more (less) urbanized areas indicate that local urbanization may increase the occurrence of compound events.

Nearly all stations have experienced the compound events (with the frequency of compound events > 0). This result is in accordance with the findings of Hao et al. (2013) that heat and precipitation extremes have co-occurred in the high latitudes and tropical regions. The spatial variations of the compound heat and extreme precipitation events are highly consistent with the distribution of the urban and non-urban stations, i.e., compound events tend to occur more frequently at the urban than non-urban stations.

To understand the temporal evolution of compound events, we calculate the regional mean frequency and fraction of compound events by averaging all stations (Figure 3). Both the frequency and fraction show dramatic increasing trends, i.e., statistically significant at the 0.05 level. The regional mean frequency of compound events has increased by 0.18 events per decade, and the fraction has risen by 3.17% per decade over the study period. It indicates that the probability of extreme precipitation events following a heat event tends to increase.

It is also shown in Figure 3 that the increasing trends of the compound frequency and fraction tend to be accelerated since the 1960s. The magnitudes of the trends are 0.25 and 4.49% per decade during 1971–2017 for the frequency and fraction of compound events, respectively. These trend magnitudes become larger during 1981–2017, i.e., 0.30 and 5.23%, respectively. The trends remain significant and continue to increase till at least the 1990s, since which the frequency and fraction increased by 0.29 and 7.60%, respectively. These results suggest that the increasing speed of the proportion of extreme precipitation that follows a heat event has elevated.

Figure 4 depicts the spatial distribution of the secular trends of the frequency and fraction of compound events at the individual station during 1961–2017. Compound events exhibit increasing tendencies in measure of frequency (fraction) since the 1960s at nearly all stations, of which 55.81% (56.98%) are significant at the 0.05 level. The upward trends show regional disparities with stronger magnitude in more populated and urbanized areas and weaker in less developed regions. In particular, the PRD region with the densest population and highest urbanization level has the most substantial increasing tendency, whereas other less urbanized areas such as the northern parts of Guangdong possess relatively weaker trends. These features indicate that the residents living in PRD are facing intensifying threats induced by compound heat-precipitation events. More substantial intensification of compound events in faster-urbanized areas implies that the urbanization process possibly plays an important role in accelerating this phenomenon.

The trends of compound heat-precipitation events revealed above are consistent with the findings of Scherrer et al. (2016) in such a way that the increasing trends of hot days and heavy precipitation were found in warmer places. Hao et al. (2013) used the Coupled Model Intercomparison Project phase 5 (CMIP5) climate model to simulate concurrent wet and warm events, and demonstrated that those events increased significantly in high-latitude and tropical regions (Hao et al., 2013). The reason for the increasing trend of compound events in South China is likely that the preconditioned extreme hot weather with higher temperature leads to increased higher evaporation rates and vapor content, thus accelerating the hydrological cycle under the context of global warming (Menzel and Bürger, 2002). Moreover, the preceded heat can enhance the moisture flux and the convective available potential energy (CAPE) and thereby provide a suitable environment for extreme precipitation and flooding in several subsequent days (Zhang and Villarini, 2020). In our study, we also find that the upward trends of frequency and fraction of the compound climate events in South China became even steeper from 1990 to 2017. A possible reason for the acceleration since the 1990s is that the increased Tibetan Plateau snow cover and sea surface temperature in the equatorial Indian Ocean boosted the precipitation in South China (Wu et al., 2010).

To quantify the impacts of urbanization on the increases in the frequency and fraction of compound events, all stations are categorized into urban and non-urban types (Figure 1), and we calculate the annual mean values for the two types of stations from 1961 to 2017 (see Figure 5). Both the urban and non-urban areas exhibit rising trends in terms of the compound frequency and fraction. It is noteworthy that the urban stations (as shown in pink shading) exhibit even steeper trends than those in the non-urban areas (as shown in cyan shading), demonstrating a remarkable contribution of urbanization. The frequency (fraction) of compound events in the urban and non-urban areas increased by 0.22 events (4.05%) and 0.13 events (2.05%) per decade, respectively. The differences in the trend between the urban and non-urban regions are 0.09 events for the compound frequency and 2.00% for the compound fraction. Since possible influences by other impact factors such as global warming and large-scale circulations are comparable at the local scale, the differences between trends for the urban and non-urban regions mainly result from local urbanization. Accordingly, we estimate that urbanization accounts for 40.91% (49.38%) of the total increasing trend in the frequency (fraction) of the compound events in the urban region.

These results indicate that the urbanization process tends to exert intensifying impacts on compound heat-precipitation events. Previous studies such as Chen et al. (2015); Wang et al. (2018) found that the PRD region experienced much stronger precipitation compared to its surrounding rural areas and attributed this difference to urbanization. Other studies have also confirmed that urbanization can affect extreme precipitation by influencing UHI (Oke, 1982; Dixon and Mote, 2003), urban canopy (Miao et al., 2009), and urban aerosols (Han and Baik, 2008). Specifically, UHI effects can be enhanced by the increases in anthropogenic heat release, as suggested in many modeling studies by the Weather Research and Forecasting (WRF) model coupled with the Urban Canopy Model (UCM) (Feng et al., 2012; Chen et al., 2014; Yang et al., 2019). The urban development enhances the total thermal discomfort hours by 27% in the urban core areas of YRD, and anthropogenic heat release and urban land use change contribute nearly equally to this change (Yang et al., 2019). Warmer temperature in urban areas enhances the disturbance above the ground and strengthens the upward motion and convective activities (Collier, 2006). Moreover, urban canopy disturbs the water vapor and energy balance in urban boundary layers and impacts heavy convective precipitation by increasing the surface roughness, which reduces surface wind, bifurcates the approaching moist air mass upward, and then aggregates them in the downwind of urban areas (Cotton and Pielke, 2007; Zhang et al., 2014). Additionally, extreme precipitation is contributed by the interactions of urban aerosols with radiation and clouds (Liu et al., 2020b). Urban aerosols absorb and scatter solar radiation to generate condensation nuclei, which can influence deep convection and hence precipitation (Li et al., 2011; Liu et al., 2020b). Consequently, via these processes, the local urbanization provides favorable conditions for extreme compound heat and precipitation events.