Spatiotemporal variation characteristics of precipitation and urbanization effects in Shenzhen from 1960 to 2019

Under the dual influence of climate change and human activities, the regional hydrological cycle process has undergone significant changes (He et al. 2019; Omer et al. 2020). Abnormal changes in precipitation are significant, resulting in frequent regional/urban droughts and floods, which have a profound impact on regional/urban water resources development and utilization, urban water supply, production, living, and ecological environment (Pathak et al. 2016). With the development of society and the economy, the scale of the city continues to expand, and urban water safety has become a hot topic of public concern. Urban water security mainly includes water supply, flooding, and other aspects. Under the background of climate change and high urbanization, extreme events occur frequently, resulting in a reduction in the amount of water resources available for human control and an increase in the spatial and temporal differences in water resources (Coffel et al. 2019). At the same time, the seasonal distribution and amount of precipitation are key factors affecting urban water supply. Extreme precipitation, such as the maximum daily precipitation, is a key factor affecting urban floods. Therefore, analyzing the spatiotemporal variation characteristics of characteristic precipitation in a certain region has become a hot topic for scholars.

Domestic and foreign research on the spatiotemporal evolution of precipitation mainly focuses on large-scale river basins and economically developed large cities or urban agglomerations. For example, Silva Dias et al. (2012) used daily rainfall data in Sao Paulo, Brazil to analyze its precipitation trends. The results showed that precipitation in this region showed significant increasing trends. Caloiero (2013) found different results between the North Island and the South Island of New Zealand, where rainfall is most concentrated, with precipitation concentration on the eastern side of the South Island being comparable to that of the North Island, while precipitation concentration on the west side is the lowest. Royé & Martin-Vide (2017) studied the concentration of precipitation days in the United States and found that the less annual precipitation, the more concentrated the precipitation days are. Zhang et al. (2023) found that the overall winter precipitation on the Qinghai-Tibet Plateau mainly increased at a large scale, with the largest increase rate in the central and southern regions. Wang et al. (2023) analyzed the spatiotemporal variation characteristics of precipitation in the Jianghuai region and found that annual and summer precipitation showed an insignificant increasing trend, and spring and autumn precipitation showed an insignificant decreasing trend. Shifteh Some'e et al. (2012) examined the trend of precipitation in Iran using the MK method and estimated the magnitude of the precipitation trend using the Theil-Sen slope and found that there is a significant increase in summer precipitation in Iran and an increase in the concentration of precipitation. In terms of spatial changes in precipitation, Chen & Zhang (2012) analyzed the precipitation changes at stations in the Hanjiang Basin and found that the changes in precipitation were spatially non-consistent with a decrease in precipitation in the upper reaches and an increase in precipitation in the lower reaches. Deng et al. (2018) used empirical orthogonal function (EOF) decomposition to study the precipitation changes in the Hanjiang Basin and found that precipitation increased in the southwestern and downstream areas of the basin, which further led to the conclusion that the spatial changes in precipitation in the Hanjiang basin are non-consistent from mathematical methods. From the above analysis, it is found that in the context of climate change, precipitation in different regions has changed significantly, mainly in the form of an increase in the concentration of precipitation on the time scale and an increase in the heterogeneity of the temporal distribution of precipitation within the year. On the spatial scale, the spatial variation of precipitation is non-consistent.

As the urbanization process continues to advance, corresponding research on precipitation in urbanized areas has gradually increased. Changnon et al. (1976) initiated and implemented the Metropolitan Meteorological Observation Experimental Program. The observation results showed that urbanization increased urban downwind precipitation relative to the background area. At the same time, in a study on summer precipitation changes in St. Louis, it was found that urbanization caused. There was a significant increase in rainfall downwind of the city (Changnon 1979). Seino et al. (2018) studied the precipitation characteristics of Tokyo and found that the heat island effect caused by urbanization will cause precipitation in urban Tokyo to increase by about 10%. In a case study of Houston, Shepherd et al. (2010) found that urban expansion induced a rainfall enhancement effect. Zhang (2020) found that the frequency and intensity of extreme precipitation events in urbanized areas in southeastern China showed a significant increasing trend. Han et al. (2015) discussed the changing characteristics of extreme precipitation in the Yangtze River Delta urban agglomeration and believed that the frequency of floods in key cities in the region was increasing. Li et al. (2021) analyzed the characteristics of extreme precipitation in the Guangdong-Hong Kong-Macao Bay Area and concluded that extreme precipitation in this area has increased significantly. Ding et al. (2019) compared the difference in precipitation during the rainy season between urban and suburban areas of Beijing and found that although precipitation during the rainy season showed a decreasing trend, urbanization still has the effect of increasing rainfall. Kang et al. (2021) found that urbanization in the Taihu Basin leads to an increase in extreme precipitation. Zhao et al. (2021) analyzed the impact of urbanization on temperature and precipitation in Shenzhen from 1979 to 2017, showing that there is a significant rain island effect in Shenzhen.

In summary, it is found that previous studies on the effect of urbanization on increasing rainfall have focused on qualitative analysis and lacked quantitative research, while the spatial scale of the relevant studies has focused on urban agglomerations and provincial-level administrative districts, and lacked quantitative research on the individual city. Shenzhen is located on the southern of Guangdong Province, the east coast of the Pearl River Delta. It is a window for China's reform and opening up, a national economic center city of China, and an international city. After more than 40 years of rapid urbanization, Shenzhen has become one of the most urbanized areas in China. Under the background of high urbanization, precipitation characteristics in Shenzhen have also undergone certain changes. Therefore, this article analyzes the spatiotemporal changes in precipitation in Shenzhen and the impact of urbanization based on Shenzhen rainfall station data. In this study, the trend-free pre-whitening MK (TFPW-MK), empirical orthogonal function, and suburban site classification are used to analyze the spatial and temporal patterns of precipitation changes in Shenzhen and the impact of urbanization on precipitation. An in-depth study of precipitation changes at different scales in Shenzhen has important practical significance for Shenzhen's water resources management, water security strategy, and flood control. Meanwhile, this study may provide a research example to study the urbanization effect of precipitation in related highly urbanized cities.

Shenzhen is located on the southern of Guangdong Province, the east coast of the Pearl River Delta. (113°43′ ∼ 114°38′E, 22°24′ ∼ 22°52′N). It is one of the four central cities in the Guangdong-Hong Kong-Macao Greater Bay Area. Shenzhen has a subtropical monsoon climate, with long summers and short winters, a mild climate, sufficient sunshine, and abundant rainfall. The average annual precipitation is 1,830 mm. The distribution of precipitation time is uneven. During the rainy season (the rainy season in Shenzhen is from April to October), the multi-year average precipitation is 1,633.62 mm, which is about 85% of the annual precipitation. The city covers an area of 1,997.47 km², with complex terrain and diverse landform types. The terrain gradually decreases from southeast to northwest. As of 2020, the city's built-up area is 927.96 km². As a typical representative of rapid urbanization, Shenzhen has experienced rapid economic development, sudden population increase, and large-scale urban expansion. The city's surface coverage has changed tremendously, and the urban climate has gradually changed.

The TFPW-MK test method based on preset whitening processing is used to conduct trend tests on the precipitation indicators series in Shenzhen. This method can effectively overcome the shortcomings of the Mann–Kendall (M–K) test method that leads to inaccurate analysis results due to autocorrelation (Gebru et al. 2022). The specific calculation steps of TFPW-MK are as follows:

Finally, the trend term is merged with the independent sequence to obtain a new sequence ⁠, and the obtained new sequence is used to perform the M–K test. For details on the M–K test method, please refer to the references (Güçlü 2020).

EOF is a commonly used method for spatial distribution mode analysis. It is a method for analyzing the structural characteristics of matrix data and extracting main data features. It is widely used in meteorological and climate research. Concentrate the information of the original multiple variables to the maximum extent on the principal components of a few independent variables. The greater the cumulative variance contribution rate of the principal components, the greater the information the principal component accounts for in the original variables. EOF can reflect the spatial distribution characteristics of the element field to a certain extent (Eom et al. 2017). The specific algorithm is as follows:

• (1) Preprocess the original data of m spatial fields with a time scale of n into an anomaly form to obtain a data matrix ⁠.

• (2) Calculate the cross product of matrix X and its transposed matrix X^T to obtain the square matrix ⁠:

  

  (4)
• (3) Calculate the characteristic roots (⁠⁠) and eigenvector of the square matrix C, both of which satisfy:

  

  (5)

  where E is an m-dimensional diagonal matrix. The above characteristic roots are all greater than 0, and⁠. The eigenvector corresponding to each characteristic root is an EOF mode. For example, corresponds to the first main mode of the original data.
• (4) Calculate time coefficients: Project EOF onto the original data matrix to obtain the time coefficients corresponding to all spatial vectors. The calculation formula is as follows:

  

  (6)

Each row of data in the matrix is the time coefficient of the corresponding feature vector.

• (5) Calculation contribution rate: the variance size of matrix X can be simply expressed by the numerical value of the characteristic root. The higher the⁠, the more important the corresponding mode is and the greater the contribution to the total variance. The contribution rate of the total variance corresponding to the ith mode is as follows:

  

  (7)
• (6) Significance test: The decomposed EOF needs to be tested for significance. Generally, it needs to pass the North criterion. The characteristic root error at the 95% level:

  

  (8)

In the formula: is the eigenvector value. N is the effective degree of freedom of the sample. When adjacent eigenvalues are satisfied ⁠, the EOF values corresponding to the two eigenvalues are considered to be valuable signals.

From Figure 3, the change patterns of the precipitation indicators in Shenzhen are consistent with the stations, and the precipitation indicators in Shenzhen have obvious oscillation periods, and the oscillation period is about 10–20 years (Fu et al. 2022). Although there are fluctuations, except for SDII, the rest of the Shenzhen precipitation indicators show an overall upward trend. From the results of the M–K mutation test of the precipitation indicators (Figure 4), it can be seen that for SDII, the intersection of the UF statistic sequence and the UB statistic sequence appeared in 1972, indicating that the precipitation intensity mutated during this period. For the four indicators PRCPTOT, RRS, RX1day, and RX7day, the UF and UB curves had multiple intersection points between 1960 and 2019. However, after 1991, the UF and UB curves had intersection points and their relative positions changed. It shows that significant mutations occurred in PRCPTOT, RRS, RX1day, and RX7day in Shenzhen City in 1991. Combined with Figure 3, it can be found that precipitation changed from a year-by-year decrease before 1991 to a year-by-year increase trend after 1991.

Taking into account the physical meaning of different precipitation indicators, the EOF method is used to decompose the RRS and RX7day indicators in Shenzhen. Among them, RRS reflects the changing characteristics of long-duration precipitation in Shenzhen, and RX7day can reflect the distribution characteristics of short-duration heavy rains in Shenzhen. The cumulative variance contribution of RRS and RX7day eigenvectors is shown in Table 3. It can be seen from Table 3 that the precipitation fields corresponding to RRS and RX7day converge quickly. The cumulative variance contribution rates of the first two principal components account for 77.6 and 64.3%, respectively. Both of them passed the North criterion significance test. Therefore, the distribution modes of the first two principal components can characterize the variability distribution structure of the precipitation variable field in Shenzhen.

Table 3

Cumulative variance contribution rate of the top five eigenvectors of EOF decomposition of two precipitation indicators

Extreme precipitation indicator .	Mode .	Eigenvector value .	Variance contribution rate (%) .	Cumulative variance contribution rate (%) .	.	.
RRS	1	571741.3	66.4	66.4	475071.7	255690.5
	2	96669.6	11.2	77.6	44401.6	43232.0
	3	52268.0	6.1	83.6	7557.9	23375.0
	4	44710.2	5.2	88.8	18550.5	19995.0
	5	26159.7	3.0	91.9	3596.1	11699.0
RX7day	1	38440.0	48.4	48.4	25839.7	16390.9
	2	12600.3	15.9	64.3	5482.5	5372.8
	3	7117.8	9.0	73.3	594.1	3035.4
	4	6523.7	8.2	81.5	2318.4	2781.7
	5	4205.3	8.3	86.8	826.2	1793.1

Extreme precipitation indicator .	Mode .	Eigenvector value .	Variance contribution rate (%) .	Cumulative variance contribution rate (%) .	.	.
RRS	1	571741.3	66.4	66.4	475071.7	255690.5
	2	96669.6	11.2	77.6	44401.6	43232.0
	3	52268.0	6.1	83.6	7557.9	23375.0
	4	44710.2	5.2	88.8	18550.5	19995.0
	5	26159.7	3.0	91.9	3596.1	11699.0
RX7day	1	38440.0	48.4	48.4	25839.7	16390.9
	2	12600.3	15.9	64.3	5482.5	5372.8
	3	7117.8	9.0	73.3	594.1	3035.4
	4	6523.7	8.2	81.5	2318.4	2781.7
	5	4205.3	8.3	86.8	826.2	1793.1

View Large

The variance contribution rates of the second main mode eigenvectors of RRS and RX7day are 11.2 and 15.9%, respectively. The distribution is roughly divided between Shenzhen Reservoir Station and Qinglinjing Station. The eigenvector has negative values to the east and positive values to the west. The positive value center appears near Gaofeng and Luotian in the northwest, and the negative value center appears near Nan'aowei in the Dapeng Bay District in the southeast, showing an ‘East-West’ reverse distribution characteristic. This shows that Shenzhen has a spatial distribution of floods in the west and drought in the east or drought in the west and floods in the east. The time coefficients of the second main mode of RRS and RX7day over the years show that under this mode, rainfall in the east and west changes alternately with high and low rainfall, and has certain periodic characteristics.

The rainfall stations were divided into urban and suburban stations based on Shenzhen socioeconomic data and land use data (Wang et al. 2021). Shenzhen Reservoir Station was selected as an urban station and Sanzhoutian Station was selected as a suburban station (Table 4). The differences in the evolution of precipitation indicators in urban and suburban areas were analyzed from two perspectives: a long series comparison of extreme precipitation indicators between urban and suburban rainfall stations and a comparison at different development stages.

Shenzhen's urbanization is mainly divided into two obvious stages. The main turning point is 1991. Based on previous research, the first main cycle of annual precipitation in the study area is about 10, the main mutation year is 1991, and there is a 60-year annual precipitation sequence from 1960 to 2019. Therefore, the 60-year precipitation series in Shenzhen can be divided into two periods with 1991 as the boundary. The lengths of the two periods before and after are 31a and 29a, respectively. The two periods happen to be roughly 3 periods, which can avoid the influence of the periodic sequence on the precipitation change results due to the random selection of periods.

Taking 1991 as the dividing point, the precipitation changes in Shenzhen at different urbanization stages were studied. Table 5 shows the comparison of precipitation indicators between urban and suburban stations in Shenzhen at different stages. It can be seen from Table 5 that for PRCPTOT, the precipitation difference between urban and suburban areas in Shenzhen develops in a decreasing trend. A longitudinal comparison of urban and suburban areas in different periods shows that the average precipitation in the urban area from 1991 to 2019 increased by 11.96% compared with the average precipitation from 1960 to 1990, while the annual precipitation in the suburbs increased by 8.66% in the two periods. In terms of average annual precipitation, the difference between the annual precipitation in the suburbs and the urban area from 1960 to 1990 was 157.67 mm. From 1991 to 2019, the difference narrowed to 108.69 mm, a decrease of 31.06%. In terms of SDII, precipitation intensity in both urban and suburban areas decreased after 1991, but the decrease in suburban areas was greater, at −2.02%. The RRS in urban areas and suburbs from 1991 to 2019 increased by 11.58 and 10.99%, respectively, compared with before 1991. The RRS of urban stations is larger than that of suburban stations. With the development of urbanization, the urban–suburban difference in rainfall during flood season in Shenzhen has also expanded from 242.0 mm before 1991 to 283.14 mm. In terms of extreme precipitation, in the past 60 years, the extreme precipitation in Shenzhen's suburbs was much greater than that in urban stations. Before 1991, the differences in RX1day, RX3day, and RX7day between suburban stations and urban stations were 12.6, 30.1, and 40.7 mm, respectively. After 1991, the value shrank to 7.4, 25.3, and 39.2 mm.

The above analysis shows that except for RRS, the PRCPTOT, SDII, RX1day, RX3day, and RX7day of urban stations in Shenzhen are smaller than those of suburban stations. However, with the development of urbanization, the differences between urban and suburban stations are decreasing. The RRS of urban areas is larger than that of suburban stations, and the difference is constantly expanding, indicating that urbanization in Shenzhen has a certain rainfall-enhancing effect on various precipitation indicators. In terms of extreme precipitation, the smaller the precipitation scale, the stronger the rainfall-enhancing effect of urbanization.

• (a) Shenzhen's PRCPTOT, RRS, RX1day, RX3day, and RX7day all showed an insignificant upward trend. SDII showed an insignificant decreasing trend. Except for PRCPTOT and RRS, the changing trends of other indicators have certain spatial heterogeneity. Among them, some stations showed a downward trend in terms of extreme precipitation, and they were mainly located in the west of Shenzhen City.

• (b) Shenzhen's precipitation indicators all have obvious oscillation periods, and the oscillation period is about 10–20 years. Shenzhen's SDII mutated in 1972; the four precipitation indicators PRCPTOT, RRS, RX1day, and RX7day mutated in 1991. The corresponding precipitation changed from a yearly decrease before 1991 to a yearly increase after 1991.

• (c) Different precipitation indicators in Shenzhen have similar distribution modes. The first main mode shows global consistency, showing that the precipitation in Shenzhen is either rainy or less rainy, and reflects that the precipitation center of Shenzhen has changed from the northwest before 1990 to the southeast after 1990. The second main mode shows an ‘East-West’ reverse distribution, that is, precipitation increases in the east and decreases in the west, or precipitation increases in the west and decreases in the east.

• (d) Urbanization has a significant impact on the spatial distribution of precipitation in Shenzhen and has an increasing effect on the city's PRCPTOT, RRS, and extreme precipitation. The impact of urbanization on extreme precipitation on smaller time scales is more severe, making cities more vulnerable to the risk of extreme floods.

All relevant data are available from https://github.com/ssw16/-.git.

The authors declare there is no conflict.