Ethics statement

No specific permissions were required for the described studies, because we did not carry out any experiments. All data used in this study are issued by China’s National Health Commission, China Meteorological Administration, Global Climate Dataset and Local City Weather Bureaus (see Table 1). All data is public and free to use.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. A summary of data resource.

https://doi.org/10.1371/journal.pone.0255229.t001

Data collection

In this study, we chose 29^th January to 29^th February 2020 as our studied period because China National Health Commission (NHC) began to collect and release epidemic data across the whole country after 25th January [38]. It was also because that the period prior to March, 2020 is considered as the rise phase of COVID-19 epidemic in China [4, 5]. Chinese government’s control measures were most strictly enforced during this period. The daily officially reported positively confirmed case and deaths of COVID-19 were obtained from the official website of China’s National Health Commission, and the provincial health commission (including municipalities, and special administrative regions) (Table 1). In the meantime, climatic factors of all China’s cities were recorded from China Meteorological Administration and Global Climate Dataset (Table 1). In this study, we selected the relative air humidity, temperature and solar radiation intensity as representing climate factors. This is because these parameters has been reported as the most influencing meteorological factors on COVID-19 transmission in previous studies [3, 24, 26, 29].

Socio-economic determinants including GDP and population, for which the data at the end of 2018 were collected from the China’s National Statistical Yearbook because data for year 2019 and 2020 have not been released. GDP and population were chosen because they have decisive effects on factors that directly affected transmission and prevention of COVID-19, such as population mobility, education level and medical conditions [2, 13, 39]. Assuming that GDP and population would not change dramatically between two consecutive years, the impact of using the 2018 Statistical Yearbook data on our analysis would be negligible. Spatial distance included the distance of the target city to Wuhan city (DW for short), and distance between the target city and their local provincial capital city (DLPC, for short). DW and DLPC were calculated through Baidu maps, referring to the shortest traffic distances between cities (Table 1).

Data analysis

The number of cities selected in this study is 334, which is far less than the total number of cities in China (approximately 661 in 2020). This is because the cities selected in this study are all cities where the conformed case of COVID-19 was reported from NHC, while the cities without the conformed case were not selected. All cities (324) were categorized into five types: small-, medium-, large-, very large- and super large-sized cities, as suggested from land transaction website of China (Table 1). The cities were mainly classified according to their resident population, GDP, cultural radiation area and urban built-up area (especially emphasizing on population). The Chinese State Council stated in 2014 that, cities with a population of more than 10 million, 5~10 million, 1~5 million, 0.5~1 million, and less than 0.5 million were defined as super large-sized, very large-, large-, medium- and small-sized cities, respectively. Previous studies have indicated that the epidemic intensity had trivial dependencies on population scale [2, 23]. In order to reduce this effect, the daily confirmed cases and deaths in population size of 100,000 were set as the initial data for data processing. Therefore, we used the positive confirmed cases and deaths over the population size of 100,000 at every 5 consecutive days (the CCOPSPTT and the DOPSPTT for short, respectively) as for the statistical analysis, with totally 8 time periods. In each time period, climatic factors are represented by their average of 5 consecutive days. Then, one-way ANOVA was used to compare the differences of CCOPSPTT and DOPSPTT among five city types at eight periods. Here we use eight periods to reduce the number of repetitive samples. Time was included as one analysis factor to investigate whether CCOPSPTT and DOPSPTT in different types of cities change with time. In one-way ANOVA, if the variance of the above indicators was homogeneous, the Tukey HSD was used to test the differences of CCOPSPTT and DOPSPTT among different cities as impacted by above mentioned influencing factors. Alternatively, if the variance was heterogeneous, Kruskal-Wallis test was used. The use and explanation of one-way ANOVA see Wilcox [40].

Based on the investigative data of 324 cities, GLMMs were used to test the multiple-influences of city sizes, socio-economic determinants, climate factors, spatial distance and time on CCOPSPTT and DOPSPTT of COVID-19. In GLMMs, city sizes were set as a random variable in order to reduce the dependencies of epidemic intensity on population scale. In addition, previous studies have suggested that variances of CCOPSPTT and DOPSPTT can be affected by time [1, 14]. After the outbreak of COVID-19, the spread of this virus from the outbreak area to other cities and the increase in infected persons within a city accumulated and shifted over time. Thus, in the course of data analysis, time was included. We set the starting day of data collection (25th January, 2019) as 0, and values of time were calculated as the interval of the other dates to this day. Hence, times of the eight studied periods were set as 0, 5, 10, 15, 20, 25, 30 and 35, respectively. Though the date of first officially confirmed case (29th January 2020) in Wuhan should be set to 0, our times were safe to use as they are mathematically logically equal with initially times of the first confirmed case. The only difference between the two referential times is a constant, which did not affect the statistical results. Moreover, impact of the confirmed cases on deaths cannot be ignored, because it is the most important direct factor causing death. Thus, we also set CCOPSPTT as an independent variable of DOPSPTT in GLMMs. The econometric models of CCOPSPTT and DOPSPTT are shown in (1)-(2).

(1)

(2)

In the models, C, SD and SED represent climate factors, spatial distance and socio-economic determinants, respectively. Their sets are respectively expressed in Formula 3–5. c and t represent city size and time respectively. SRI, AT and RAH are solar radiation intensity, average temperature and relative air humidity, respectively. ε is the random error; which is formed by the random action of many factors in the process of measurement and statistics. If the significance of regression coefficients, such as α and β, is less than 0.05, suggesting that they have the significantly impacts on CCOPSPTT or DOPSPTT.

(3)

(4)

(5)

Finally, we used HPA to compare the relative contributions of these factors on CCOPSPTT and DOPSPTT. Here, the relative contribution was the proportion of each independent variable from the goodness-of-fit measures across all variable combinations in a hierarchy, which indicated weight of one variable versus other independent variables that obviously contributed to COVID-19 transmission. When performing GLMMs and HPA, all data were standardized by logarithmic transformation in order to avoid the trivial influence of the non-normal data on statistical results (e.g., much data of CCOPSPTT and DOPSPTT in different city types did not conformed to the normal distribution, see Tables 2 and 3; SD > Mean indicates non-normal distribution). As COVID-19 control conditions were much more complicated in Hubei compared with other provinces, like lockdown and mass screening, data analysis of Hubei and other province were performed separately. The use and explanation of GLMMs and HPA see Dichmont [41], Lai and Peres-neto [42]. In the data analysis processes of GLMMs and HPA, socio-economic determinants, climate factors, spatial distance and time were independent variables, while CCOPSPTT and DOPSPTT were dependent variables. Remarkably, we had not defined one or several specific variables as control variables, and then used the econometric model to analyze relationships of explanatory variables with CCOPSPTT and DOPSPTT. Instead, we defined all independent variables as explanatory variables. In this case, seven independent variables, including solar radiation intensity, average temperature, relative air humidity, time, GDP, population, DW and DLDP, have the an equivalent logical impacts on CCOPSPTT and DOPSPTT. They were all explanatory variables, not control variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Differences in the confirmed cases over population size per 100000 at every 5 consecutive days (CCOPSPTT) among five types of cities outside and within Hubei province from 25th January to 29th February, 2020.

https://doi.org/10.1371/journal.pone.0255229.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Differences in deaths over population size per 100000 at every 5 consecutive days (DOPSPTT) among five types of cities outside and in Hubei province from 25th January to 29th February, 2020.

https://doi.org/10.1371/journal.pone.0255229.t003

The creditability of GLMMs and HPA results was determined by determinate coefficient (R²) and p-values. High R² and p-values <0.05 indicated that statistical results were credible. Besides, we also tested the robustness of the model to ensure the credibility of the research results. In this study, since independent variables, especially climate factors, are highly correlated between adjacent time observation points, their values at time t not only have an impact on CCOPSPTT and DOPSPTT, but also have a significant impact at time t+1. Therefore, our model is time-delayed. According to the results of Jiang et al [43], Xun and Halbert [44], the robustness of the time-delay model can be tested with the data of one lag period. More specifically, the dependent variables can be estimated by the independent variables with a lag of one period (i.e., lag regression or time-delay model). When the regression parameters, like p-value, R² and estimate coefficients, of time-delay and baseline (original) models are close to each other, it indicates that our model has high robustness. All data processing was conducted in R. 3.4.3 software.

Differences in the CCOPSPTT and DOPSPTT of COVID-19 among five city types

Outside Hubei province, there was a significant difference on CCOPSPTT among the five city types from 25^th January to 29^th February, 2020 (p < 0.05) (Table 2). In most cases, CCOPSPTT decreased in the order of super large- ≥ very large- > large- ≥ medium- > small-sized cities (Table 2). Within Hubei province, CCOPSPTT in the only super large-sized city (Wuhan) were obviously higher than other cities. Additionally, the gap of CCOPSPTT between Wuhan and other cities in Hubei province continuously increased with time (Table 2). For example, on 25^th January, 2020, CCOPSPTT in Wuhan was 6×E-3, which was significantly higher than that of large-sized cities (3×E-4), medium-sized cities (5×E-4) and small-sized cities (1×E-4). Wuhan was more than 12 times higher than other cities on CCOPSPTT (Table 2). On 29^th February, 2020, the CCOPSPTT in Wuhan, large-, medium- and small-sized cities of Hubei province were 0.02, 1×E-4 (average), 2×E-5 (average) and 2×E-5 (average), respectively. There was over 200 times higher CCOPSPTT in Wuhan and in other cities (Table 2). In all sampling dates, difference of CCOPSPTT among large-, medium- and small-sized types of cities in Hubei was not significant (p > 0.05) (Table 2).

DOPSPTT showed no significant difference among five city types outside Hubei province on 25^th January and 29^th February, 2020 (p > 0.05) (Table 3). But during the six sampling periods in between these two dates outside Hubei province, DOPSPTT in super large-sized cities were significantly higher than that in other cities (p < 0.05). DOPSPTT showed no significant difference among other four (very large-, large-, medium- and small-sized cities) types of cities outside Hubei province (p > 0.05) (Table 3). In Hubei province, DOPSPTT of the only super large-sized city (Wuhan) were obviously higher than that of other cities at all sampling dates. The gap of DOPSPTT between Wuhan and other Hubei cities also continuously increased with time. The DOPSPTT also showed no significant difference among large-, medium- and small-sized cities (p > 0.05) (Table 3).

Multiple relationships of socio-economic determinants, climate, spatial distance and time to the CCOPSPTT and DOPSPTT of COVID-19

The GLMMs results showed high R² values (> 0.10) and significant p-values (< 0.001) in all analysis (Tables 4 and 5). Our results also showed that the time-delay and the baseline models have similar regression parameters. For example, in the GLMMs of COPSPTT, the p-values of population, GDP, DW, relative air humidity, solar radiation intensity, average temperature and time of the two models, were all less than 0.05 outside Hubei provinces. The positive or negative forms of regression coefficients of all independent variables in the two models were consistent (Table 4). R² difference ranged from 0.01 to 0.04, and there was no obviously between the two models. These suggested that our model has high robustness (Tables 4 and 5). The variability of CCOPSPTT and DOPSPTT among five city types could be affected by socio-economic determinants, climate, spatial distance and time.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Results of GLMs in relationship of the confirmed cases over population size per 100000 at every 5 consecutive days (CCOPSPTT) with socio-economic determinants (GDP and population), spatial distance (DW and DLPC), climate factors (relative air humidity, solar radiation intensity and average temperature) and time in cities outside and within Hubei province.

https://doi.org/10.1371/journal.pone.0255229.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Results of GLMs in relationship of deaths over population size per 100000 at every 5 consecutive days (DOPSPTT) with the CCOPSPTT, socio-economic determinants, spatial distance, climate factors and time in cities outside and within Hubei province.

https://doi.org/10.1371/journal.pone.0255229.t005

Outside Hubei province, GDP (positive), population (positive), DW (negative), relative air humidity (negative), solar radiation intensity (positive), average temperature (positive) and time (positive) had a significant regression relationship with CCOPSPTT (p < 0.05). In contrast, DLPC had no significant correlation (p> 0.05) with CCOPSPTT (Table 4). In Hubei province, only GDP (positive) had a significant regression relationship with CCOPSPTT (p < 0.05), whereas other factors did not (p > 0.05) (Table 4).

Regarding DOPSPTT, GDP, population, DW, time and CCOPSPTT showed the significant positive influences on it outside Hubei province (p < 0.01) (Table 5). However, in Hubei province, the variability of DOPSPTT was significantly affected (positive) by CCOPSPTT and GDP (p < 0.05), but not related to other factors (p > 0.05) (Table 5).

The HPA results showed high R² values (R² of CCOPSPTT and DOPSPTT in Hubei and other provinces were 0.45, 0.30, 0.82 and 0.49, respectively) and significant p-values (< 0.001) in all analysis, indicating that HPA can be good used to decompose the relative contribution of each influencing factor to CCOPSPTT and DOPSPTT. This also suggested our results have high credibility (Fig 1). Among cities outside Hubei province, the results of HPA showed that the relative contributions of socio-economic determinants, climate, spatial distance and time to the change of CCOPSPTT were 78.90%, 3.20%, 6.40% and 11.50%, respectively. Specifically, the relative contribution of GDP (45.72%) was obviously higher than that of population (33.23%). DW (4.09%) contributed significantly higher than DLPC (2.30%) on CCOPSPTT (Fig 1A). All climatic factors had relatively lower relative contributions to CCOPSPTT (relative air humidity 1.45%; average temperature 0.90%; and solar radiation intensity 0.84%) (Fig 1A). In Hubei province, the relative contributions of different factors to variability of CCOPSPTT among cities were ordered as socio-economic determinants (91.31%) > spatial distance (6.52%) > climate (1.95%) > time (0.22%) (Fig 1B). The relative contribution of GDP (73.28%) was significantly higher than that of population (18.03%), while average temperature (1.06%) contributed slightly higher to that of relative air humidity (0.56%) and solar radiation intensity (0.33%) (Fig 1B). The relative contributions of GDP, spatial distance and average temperature to CCOPSPTT were obviously higher in Hubei province than those of other provinces (Fig 1A and 1C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The relative contributions of different factors to the confirmed cases (a and b) and deaths (c and d) over population size per 100000 at every 5 consecutive days (CCOPSPTT and DOPSPTT).

Instruction of DW and DLPC as suggested in Table 3. SED is socio-economic determinants. SRI, AT and PAH is solar radiation intensity, average temperature and relative air humidity, respectively.

https://doi.org/10.1371/journal.pone.0255229.g001

The relative contributions to DOPSPTT were higher from CCOPSPTT (44.06%) and socio-economic determinants (44.06%) than from spatial distance (2.10%), climate (1.83%) and time (2.84%) in cities outside Hubei province (Fig 1C). Particularly, the contribution of GDP (26.85%) was higher than that of population (22.32%). Contribution of DW (1.55%) was more than that of DLPC (0.55%), while the relative contributions of all climate factors were less than 0.76% (Fig 1C). In Hubei province, the relative contributions of CCOPSPTT, socio-economic determinants, climate, DW and time to DOPSPTT were 40.61%, 52.80%, 2.06%, 4.03% and 0.50%, respectively (Fig 1D). Among the sub-factors, the relative contribution from GDP (41.73%) was obviously higher than that of population (11.03%), and the relative contributions of all climatic factors were less than 1.20%. Meanwhile, GDP, spatial distance and average temperature showed higher contributions to DOPSPTT in Hubei than that in other provinces, while CCOPSPTT exhibited the opposite trend (Fig 1B and 1D).

Relationship between city types and the confirmed cases and deaths of COVID-19

The difference of CCOPSPTT among five city types outside Hubei province (Table 2) suggested that city size and developmental level linked with the infection and transmission of COVID-19. The population and economic activities of super large- and very large-sized cities are significantly higher than those of other types of cities [9, 40]. In addition, due to the developed traffic system, population mobility in super large- and very large-sized cities are higher than those in large-, medium- and small-sized cities [32, 41]. These make population density and the amount of interactions with other people in super large- and very large-sized cities are higher than those of other cities [45]. According to the theory of infectious disease, the increases of population density and the amount of interactions with other people will improve the transmission risk of the epidemic [2, 23, 39, 46]. As a result, novel coronavirus would spread faster in super large- and very large-sized cities, resulting in the large number of CCOPSPTT. In addition, as suggested from Zhu et al. [17], Wu et al. [18] and Heibati et al. [28], urban climate factors and pollution status also affected the spread of novel coronavirus. The increases of contaminant concentration and air humidity in cold season both are beneficial to the spread of novel coronavirus [29]. Under the same geographical conditions, such as the location from the ocean and altitude, the pollutant concentration and air humidity in larger-sized cities are significantly higher than those in small- and medium-sized cities due to the influences of urban “Heat Island Effect” and “Turbidity Island Effect” [17, 18, 47]. Therefore, differences in climate factors and pollution status can also cause differences in the spread of novel coronavirus among cities with different sizes.

Among five city types, CCOPSPTT showed no significant difference between super large- and very large-sized cities, as well as between large- and medium -sized cities (Table 2). This may be determined by the complex relationships among population, population mobility, and the degree of the implementation of control measures [15, 16, 24]. For example, although super large-sized cities have larger population and higher population mobility, where the implementation of epidemic control measures would be more difficult compared with very large-sized cities. Beside, at present, people’s cognition on the role of income (GDP or GDPP) on COVID-19 transmission would be controversial because at an aggregate level higher degrees of economic development (i.e., a higher GDP) might promote higher population mobility, e.g., potentially favoring the spread of epidemics. But at the same time, a richer region should have larger amount of resources to put into place more effective control interventions. On the other hand, during a sudden and rapid period of COVID-19 outbreak, a larger GDP implying a potentially more educated population; which might allow the adoption of individual protective measures at a higher rate. This suggests that the potential role of GDP on COVID-19 can be complicated between different city types. This would also explain the insignificant difference of CCOPSPTT between super large- and very large-sized cities, as well as between large- and medium-sized cities.

Except for Wuhan, CCOPSPTT among large-, medium- and small-sized cities in Hubei province was not significantly different (Table 2). This indicated the change of confirmed cases from large- to small-sized cities was affected little by the city size and development level; which showed the opposite trend with cities outside Hubei province that significantly differed (Table 2). The difference in change of CCOPSPTT between Hubei and other province with city sizes may be related to China’s extreme control measures, such as strict traffic control and home quarantine [33, 36]. Wuhan was shut down on 23^th January, while traffic controls in other cities of Hubei province started after 28^th January, 2020. Home quarantine and traffic control could block the transmission of coronavirus [36]. Thus, change in the confirmed cases would be the effects of population mobility before the closure of Wuhan. As Wuhan is the capital city serving as the economic and cultural center, many virus carriers would have moved from Wuhan to smaller cities surrounded before the traffic control started. Within Hubei province, there might be no significant difference in the number of population mobility from Wuhan to surrounding small cities due to the developed transportation and the large population base. After traffic control and home quarantine, the increase in confirmed cases might be due to intra-family infection [37, 48]. Therefore, similar population mobility and similar extent of intra-family infection caused an identical CCOPSPTT value between large-, medium- and small-sized cities in Hubei province (Table 2). However, outside Hubei province, the number of people from Hubei province was higher in larger cities than that in small and medium-sized cities, because the former have closer contact with Hubei province’s cities, especially Wuhan, in terms of personnel flow and business exchanges [5, 32]. Such difference made CCOPSPTT of larger cities higher than that of small and medium-sized cities.

The DOPSPTT in super large-sized cities were higher than that in other cities, whereas there was no significant difference between very large-, large-, medium- and small-sized cities both in Hubei and other provinces (Table 3). This may be explained by the complicated influence from many factors, such as the confirmed cases, medical conditions, Intensive care unit (ICU) per capita and China’s control measures [5, 33, 36]. Although super large-sized cities have better medical conditions, but a higher confirmed cases and lower ICU per capita may lead to higher rate of deaths than that of other small cities. China’s strong control measures, such as traffic control, home quarantine, free medical treatment, testing and isolation of asymptomatic cases, has made the vast majority of COVID-19 patients and carriers treated and controlled [33, 37]. Thus, DOPSPTT had not differed between cities with different sizes. In Hubei province, in addition to the above several control measures, a large number of doctors and medical supplies were provided to these cities, and a large number of mobile cabin hospital were constructed [19, 32, 49]. Local hospitals have sufficient medical resources and equipment to treat patients. Therefore, DOPSPTT were not significantly different between very large-, large-, medium- and small-sized cities (Table 3). As an exception, CCOPSPTT and DOPSPTT of COVID-19 in Wuhan were obviously higher than in other cities of Hubei province (Tables 2 and 3). This is because Wuhan was the only large-scale epidemic area in China within our data survey period (from 25^th January to 29^th February). Government disease controls and the spread patterns of COVID-19 were different between Wuhan and other cities [4, 20]. Specific explanations can be summarized in the following aspects: (1) the difference between the proportion of the elderly with novel coronavirus in the total population in Wuhan and other cities. The population with high susceptibility and mortality of novel coronavirus was mainly elderly people [4, 19]. Wuhan was the first city to report COVID-19. The virus has spread widely in Wuhan as local authorities try to contain all aspects of the epidemic, leading to a higher proportion of the elderly infected with the virus. But outside Wuhan, the early detection and isolation of the virus effectively controlled the spread of the epidemic among the elderly; (2) the influences of timely medical treatment and epidemic prevention on the spread of novel coronavirus [19, 49]. Shutdown of Wuhan helped other cities to win a valuable time for epidemic prevention and control, thereby reducing the number of infections and deaths. At the same time, the implementation of effective measures avoided a run on medical resources, leading to a relatively low fatality rate [5, 36, 37]. However, in the early stage of the outbreak of the epidemic in Wuhan, many patients may not receive timely treatment due to a sharp increase in the number of infected person and the shortage of medical resources, leading to a high mortality rate [19]; (3) the impact of the improvement of control measure, diagnosis and treatment techniques on preventing the spread of novel coronavirus. Since the COVID-19 first large-scale outbreak in Wuhan, scientists and medical staff had little knowledge about this newly emerging disease, and there were no effective diagnosis, treatment and control schemes, resulting in high CCOPSPTT and DOPSPTT in Wuhan. However, as the anti-epidemic work have developed, other cities have continuously improved control measures, diagnostic and treatment techniques, to ensure that patients and carriers have been diagnosed and treated timely, thus prevented effectively the spread of the epidemic [32, 33, 37]. As a result, these cities have fewer deaths and confirmed cases of COVID-19 than Wuhan.

It is the most noteworthy that we studied the rise period of COVID-19 when China implementing the strictest control. The impact of city types on confirmed cases and deaths through GDP, population and spatial distance may be the extension of their original impact before the implementation of government’s control measures. More specifically, nearly 5 million people left Wuhan for other cities before its shutdown [12]. After these people arrived at new cities, differences in GDP, population and other aspects among these cities would have led to the variance of virus transmission before the implementation of Chinese government’s epidemic control in a national scale [1]. In addition, there is an incubation period of COVID-19 and some infected persons are asymptomatic. Hence, carriers of the novel coronavirus continued to spread it during the period of government epidemic control. Thus, influence of city types on CCOPSPTT and DOPSPTT via GDP, population and spatial distance would not be fully eliminated by the implementation of control measures. In this case, city sizes had a significant positive correlation with CCOPSPTT and DOPSPTT in our studied periods.

Influences of socio-economic determinants, climate, spatial distance and time on the confirmed cases and deaths of COVID-19

Our results showed that socio-economic determinants have the positive and greatest contribution to variability of CCOPSPTT among cities outside Hubei province (Fig 1, Table 4). The higher contribution to CCOPSPTT from than that from population (Fig 1) may be due to that GDP reflects people’s income and consumption ability, better affecting population mobility and activity which was more closely related to the spread of the virus [46, 50]. Time and DW were found the second and third contributing factors affecting CCOPSPTT (Fig 1). This was consistent with the basic theory of epidemiology that the confirmed cases would increase continually over time [5, 51] and with the finding that regions closer to the outbreak of COVID-19 are more easily infected [33, 48]. GLMMs results also proved that DW has the negative relationship with the CCOPSPTT (Table 4). Wuhan was the outbreak source of COVID-19 and the economic and cultural center of central China. During the Chinese Spring Festival, a greater number of people working, studying, visiting and traveling in Wuhan would return to their hometown. According to the theories of economic geography, the gather of the surrounding residents to the economic and cultural centers to would decrease with the increase of spatial distance [32, 50]. In other words, the number of people who left Wuhan to other surrounding cities would decrease with spatial distance [30]. This may have explained the significant negative correlation between DW and CCOPSPTT (Table 2). The weak correlation between DLPC and the confirmed cases indicated that there was not a second city showed a large-scale COVID-19 outbreak in China from January to February. These could be due to China’s strong control. After January 24, all coronavirus patients and carriers either from Wuhan and Hubei province or contacted with people from this region were quarantined at home, potentially having prevented epidemics from a second outbreak [32, 36, 37]. In addition, our results showed that relative air humidity had the negative influence on CCOPSPTT, while average temperature and solar radiation intensity displayed positive influences (Table 4). This echoed the findings reported by a few previous studies [8, 10, 27]. Dalziel et al. [50] and Liu et al. [1] proved that slightly higher temperature and solar radiation intensity and lower relative air humidity in low temperature environment were conducive to the survival and transmission of virus. Qing et al. [8] analyzed the data of infectious diseases in the Ming and Qing dynasties (AD 1368–1901) across several hundred years, and concluded that the viral transmission and deaths were related to climate. However, in this study, we used the positive confirmed cases and deaths of COVID-19 in five consecutive days and the average of climate factors in analyzing their relationship. Compared with the data of daily-variety (or daily analysis unit), our treatment reduced the number of duplicate samples, but the use of average of climatic factors may ignore the effect of abrupt changes of climate factors on viral transmission in the short term (5 days); which may have slightly affected the accuracy of our research.

There was a subtle difference in the contributions of socio-economic determinants, spatial distance, climate and time to CCOPSPTT between cities in Hubei and other provinces (Fig 1 and Table 3). This might be determined by the scale effect. As previous studies suggested, climate changed little in a small region, but population mobility and frequent interpersonal communication caused by population and spatial distance would change more so that could affect viral transmission more [10, 51]. Thus, the contributions of socio-economic determinants and DW to CCOPSPTT in Hubei province were higher than that of other provinces (Fig 1). In comparison, all climate factors showed no significant correlation with CCOPSPTT in Hubei province, while an opposite trend was observed in other provinces (Fig 1 and Table 4).

The confirmed cases was the main contributor to the deaths, which accounted 40.61% and 44.06% for the variance of DOPSPTT in Hubei and other provinces, respectively (Fig 1, Table 4). This is consistent with people’s common agreement that the higher the number of confirmed cases, the higher the number of deaths when an epidemic goes on under the absence of an effective vaccine. Conversely, impacts of GDP, population and spatial distance on DOPSPTT varied between Hubei and other provinces (Fig 1, Table 5). This difference may be caused by China’s strong control [36, 37]. According to the epidemic model without government’s control, the death toll would significantly correlated with the confirmed cases, spatial distance, population mobility and time [33, 48]. After government’s control, the impact of these factors on deaths would be reduced. However, the extent of reduced impact of these factors on death varied with different control measures. For example, CCOPSPTT and GDP that were strongly associated with death (accounted for the largest variance of DOPSPTT among different cities), still affected deaths after implementing control measures (Fig 1). Conversely, the effect of other factors that low associated with deaths may have not been changed by the control measures [32, 37]. In the cities of Hubei province, due to the implementation of many strong control measures, DOPSPTT in Hubei province only showed association with CCOPSPTT and GDP.

Outside Hubei province, population, DW and time have significant correlation with DOPSPTT (Table 5). This may be because of the not as strict control measures in these places as that in Hubei province. Although China imposed nationwide limitation on population mobility at the end of January, the extent to which such controls limited the spread of the virus depended on many factors. For example, (i) the level of the local epidemics when control interventions (e.g., social distancing, lockdown, etc) were implemented; (ii) the promptness (e.g., whether the lockdown implementation was abrupt or slowly started) and intensity of such interventions measures; and (iii) the extent of further measures such as testing and isolation of asymptomatic cases [14, 22, 31]. On the national scale, these control measures might be different between provinces. The influence of the control measures on the reduced impact of our studied influencing factors on deaths was not as significant as that in Hubei province. In this case, population, DW and time had significant correlation with the deaths (Table 5). In terms of climate-related factors, due to their loose correlation with deaths, government’s control measures may have prevented them from having a significant impact on DOPSPTT in both Hubei and other provinces.