4.2. Analysis of Regression Results
With the adaptation scores of all cities under four types of climate events as dependent variables and four indicators as independent variables, the OLS stepwise regression analysis was conducted. The regression results showed that at the 1% significance level, only the proportion of urban residents was included in the high-temperature OLS model; the proportion of urban residents and per capita disposable income of urban residents were included in the low-temperature OLS model; the proportion of urban residents, per capita disposable income of urban residents, and the proportion of secondary and tertiary industries were included in the drought OLS model; and the proportion of urban residents, per capita disposable income of urban residents, and the ratio of urban built-up area to total city area were included in the flooding OLS model. All the above OLS models included independent variables that had positive coefficients, except for the ratio of urban built-up area to total city area, indicating that all three variables contributed positively to cities’ adaptation to climate events.
However, from the previous spatial autocorrelation analysis, it can be seen that the spatial distribution pattern of adaptation was not completely random but showed significant characteristics of spatial agglomeration and heterogeneity, which confirms Tobler’s first law of geography: everything is related to everything else, but near things are more related to each other [
59]. Moreover, it also shows that the results and inferences estimated by the classical linear regression model were likely to be less reliable because the OLS model can only estimate the parameters in the “global” or “average” level. Hence, this model failed to reflect the local changes in space, revealed the spatial dependence of cities’ adaptation, and showed that modification by introducing spatial factors (spatial differences and dependencies) is extremely necessary and meaningful. Therefore, on the basis of the OLS model, the component (coordinates) reflecting heterogeneity were included and the GWR models were used for regression analysis. The results of the OLS and GWR models were compared in
Table 4, and the regression parameters of the GWR model were statistically described in
Table 5.
It can be seen from
Table 4 that the adjusted goodness-of-fit indexes (R
adj2) of the GWR models for the four types of climate events were 0.744, 0.695, 0.680, and 0.733 for high temperature, low temperature, drought, and flooding, respectively, which were greater than the values of the corresponding OLS models. Meanwhile, the RSS values of the GWR models were far less than those of the OLS models, indicating that the fitting effects of the GWR models were better than the OLS models, with small errors and a better interpretation effect. In addition, the AICc values of the GWR models were smaller and the differences from the OLS models were much greater than 3, which further showed that the GWR models had better performance [
60]. As mentioned above, the GWR models consider the local aspect; therefore, the measurement results are better than the OLS models, which only consider the mean parameters [
46].
Table 5 shows the quintile statistics results. The mean values reflected the average level of contribution of the impact factors to cities’ adaptation to climate change, and it can be seen that all factors, except for the land urbanization factor, contributed positively to adaptation in general, which was consistent with the analysis results of the OLS models. There were significant directional heterogeneities in the maximum and minimum values of the regression coefficients of some impact factors, which also indicated that the regression coefficients were non-stationary according to the geographical locations of cities, i.e., there were significant spatial differences in the degree and direction of the role of different impact factors on cities’ adaptation to climate change.
The positive and negative distributions of the regression coefficients of the four types of urbanization are statistically presented in
Figure 7. It can be seen that PU, EU, and IU had positive impacts on the adaptation to climate events in most cities, where PU played a positive role in enhancing adaptation to high temperature in all cities, while land urbanization has a negative impact, i.e., a hindering effect on adaptation to flooding in most cities.
However, simple statistical analysis cannot reflect the specific differences between cities. In order to more intuitively analyze the spatial variations of the impact factors on cities’ adaptation to climate change in China, ArcGIS was used to draw spatial distribution maps of the regression coefficients of each variable in the GWR models.
Population urbanization: As shown in
Figure 8, the regression coefficients of PU decreased from southwest to northeast. The Central Plains region, connecting the two directions, was always at a high value, while the economically developed Pearl River Delta region was always at a low–medium value. It is not difficult to see that Northeastern China took the lead in industrialization and had an early and high starting point in urbanization development [
61], and as the vanguard of China’s reform and opening up, the Pearl River Delta region has a high level of economic integration and complete infrastructure [
62], both of which are highly attractive to the population. Therefore, their urbanization rates of the permanent resident population rank among the top in China. While Southwestern China and Central Plains are both underdeveloped areas, and the urbanization rates still lag behind the national average level, they are currently in the stage of rapid development. Population overurbanization is characterized by massive population inflow or natural population growth that outpaces the development of urban infrastructure, systems, and services, meaning city governance will become more difficult, thereby increasing vulnerability or decreasing readiness to climate change. The floating population in the Pearl River Delta is mainly educated at the junior high school level, and most of them are engaged in labor-intensive industries and business service industries rather than scientific and technological industries, which does not help to improve the adaptation level from the technical aspect. Meanwhile, a large number of urban villages pose challenges to land use, urban landscape, planning and management, and community security. Nowadays, the overall economy of Northeastern China is in recession, and the mismatch between the economic development level and urbanization level means climate change may bring greater risks and challenges [
63]. Due to the existence of the marginal diminishing effect, PU has weak and even negative effects on the adaptation of the two regions, while it has positive effects on Southwestern China and the Central Plains. In addition, it is worth noting that PU has a limited role in enhancing cities’ adaptation to high temperature in Northwestern China, while it has a significant positive contribution to their adaptation to low temperature and flooding, which also indicates that the PU sensitivity of cities’ adaptation varied significantly depending on climate events.
Economic urbanization: We can see from
Figure 9 that the regression coefficients of PU of adaptation to low temperature, drought, and flooding showed highly similar distribution trends: the middle–high-value areas were mainly located to the east of the Hu Line, while the low-value areas were distributed to the west of the line, with most cities having negative coefficients. West of the Hu Line is the less developed area in China, where the economic development is relatively backward. Therefore, the disposable income of urban residents there is not high, and the adaptation to low temperature and drought are also at a low level, which indicates that the development of EU in this area has limited pulling power and a weak impact on the level of adaptation to low temperature and drought. The cities with high values are located in the belt-shaped region consisting of Southern China, Central China, Northern China, and Northeastern China, and the development of EU had a significant positive impact on the improvement of their adaptation level. Since EU is expressed by per capita disposable income of urban residents, it is possible that with more income, residents can buy more facilities and services to protect against climate events, thereby reducing personal and property damage and improving their adaptation ability.
Land urbanization: As shown in
Figure 10, the coefficients of LU varied greatly between cities, with the range reaching 3.07, indicating that the process of LU varied greatly between these cities. LU played an extremely critical role in cities’ adaptation to flooding. High-value cities were clustered in Northwestern China, Southern China, Central China, and Northeastern China; most cities with low coefficients were located in Eastern China, especially in coastal provinces, and the coefficients were even negative. The eastern cities have superior geographical locations, rapid economic development, and early urban planning and infrastructure construction, which has lasted for a long time. However, in recent years, the continuous expansion of urban boundaries has accelerated the concentrated outbreak of urban problems, such as traffic congestion and housing shortages. The resources and environment of these cities are under great pressure, and the disaster resistance abilities and disaster management level are significantly inadequate. In this context, it is not surprising that LU had a negative impact on these cities’ adaptation to flooding.
Industrial urbanization:
Figure 11 shows that the high-value areas presented a belt-shaped distribution, mainly located in Jiangsu, Zhejiang, Shanghai, Anhui, and Jiangxi in Eastern China and Hunan and Hubei in Central China, while the low-value areas were mainly located in Fujian Province and the Shanxi–Hebei–Henan region. In addition, some cities in Xinjiang, Gansu, and Shandong were also at a high-value level. The high coefficient cities had a high level of IU relative to the rest of the country, with a diversified industrial pattern or a large proportion of the output value of the dominant industry and, thus, IU had a significant positive driving impact on their adaptation to drought. For example, the city of Karamay in Xinjiang is an industrial city with petroleum and petrochemical industries as the main industry. In 2019, the secondary industry accounted for 70%, while agriculture only accounted for 1.5%. In comparison, low-value areas are either weak in terms of their industrial economy or singular in their industrial structure. For example, Henan is a typical agricultural province, and the level of secondary and tertiary industries in its cities is relatively backward. From the assessment results, most of these cities have low values, and the IU process cannot play a significant role in promoting adaptation to drought. Moreover, Hebei is a large industrial province, not a powerful one, and mainly contains iron and steel industries with high pollution, high energy consumption, and low added value. Thus, the IU of Hebei did not play a significant positive role in economic development and environmental health. The low-quality IU did not improve the adaptation to drought but became an obstacle factor. To a certain extent, this also indicated that it is very urgent to enhance the industrial output values of cities with low IU coefficients and optimize and adjust their singular or unreasonable industrial structure.