**3. Results**

#### *3.1. Spatial Patterns of the LST and SOS in Hangzhou*

The average LST in spring of 2018 was utilized to explore the difference of LST between that in the urban and the rural Figure 4. The results showed that there was a significant spatial heterogeneity in the LST of Hangzhou, showing a gradient of high in the urban and low in the rural. In terms of spatial distribution, the area with a LST greater than 23.5 ◦C accounted for 10.9%, mainly distributed in the urban area; 16.6%, 32.9% and 26.6% of the area with a LST of 21.5–23.5, 19.5–21.5 and 17.5–19.5 ◦C distributed in the middle area of Hangzhou, respectively. The area with a LST less than 17.5 ◦C (accounting for 13%) was mainly distributed in the northern and southern edges of Hangzhou. Moreover, as shown in the inset chart in Figure 4, the LST followed a generally decreasing urban–rural gradient, showing a significant urban heat island effect that the LST was highest (24 ◦C) in the urban and lowest (18.6 ◦C) in the rural. In the range of 0–6 km from the urban, the LST decreased fast (0.72 ◦C/km), and the decrease tended to slow down (0.08 ◦C/km) after 6 km.

**Figure 4.** Spatial distribution of land surface temperature in spring of 2018 in Hangzhou. The histogram denotes the frequency distribution of difference temperature. The inset graph denotes the variation trend of temperature in the buffer zone at different distances from urban. The solid lines denote the urban boundaries, and the wide and short dotted lines denote the 10 km and 20 km buffers of the urban boundaries, respectively.

As shown in Figure 5, the annual average SOS of Hangzhou from 2006 to 2018 showed a significant spatial heterogeneity. The SOS was earlier in the urban and later in the rural. In terms of spatial distribution, the area with the SOS less than 76 day of year (termed DOY) accounted for 8.9%, mainly distributed in the urban and the area within 2 km from the urban, located in the east and south of Hangzhou. About 23.1%, 31.1% and 26.6% of the area with SOS of 76–83, 83–88 and 88–94 DOY were distributed in the middle area of Hangzhou, respectively. The area with SOS more than 94 DOY (accounting for 10.3%) mainly distributed in the northern edges of Hangzhou, in the mountainous areas with

higher elevations. Besides, as shown in the inset chart in Figure 5, the SOS followed a generally increasing urban–rural gradient, that is, from urban (79 DOY) to rural (87 DOY), the SOS was continuously delayed by 8 days. In the range of 0–6 km from the urban, the SOS increased fast (1.02 days/km), and the increase tended to slow down (0.14 days/km) after 6 km.

**Figure 5.** Spatial distribution of annual average of the start of the growing season (SOS) from 2006 to 2018 in Hangzhou. The histogram denotes the frequency distribution of difference SOS. The inset graph denotes the variation trend of SOS in the buffer zone at different distances from urban. The DOY denotes the day of year. The solid lines denote the urban boundaries, and the wide and short dotted lines denote the 10 km and 20 km buffers of the urban boundaries, respectively.

We further explored the spatial difference of SOS between that in the urban and the rural of Hangzhou from 2006 to 2018, and we found that the results from each year had little significant fluctuations. Therefore, in order to avoid information redundancy and excessively long images, we displayed the results every 4 years (2006, 2010, 2014 and 2018) Figure 6. The results showed that although the absolute value of SOS varied from year to year, the spatial differentiation of SOS yearly was consistent with the annual average SOS from 2006 to 2018 in Figure 5. They both showed a significant spatial heterogeneity that the SOS was smaller in the eastern and southern area of Hangzhou and larger in the northern marginal area. As shown in the inset chart in Figure 6, the SOS of the urban was 9, 9, 6 and 6 days earlier than the rural in 2006, 2010, 2014 and 2018, respectively. In addition, the SOS followed a generally increasing urban–rural gradient. In 2006, 2010, 2014 and 2018, the SOS increased fast (1.25, 1.05, 0.83 and 0.93 days/km, respectively) within the range of 0–6 km from the urban, while it tended to be stable (0.07, 0.21, 0.05 and 0.04 days/km, respectively) after 6 km.

Combined with the analysis of the average LST in spring of 2018 and SOS across Hangzhou above, the difference in LST and SOS presented an opposite state and change trend. That is, the LST tended to be higher in the urban and lower in the rural, while SOS tended to be earlier in the urban and later in the rural. The LST followed a generally decreasing urban–rural gradient, while the opposite occurred for SOS, but both the LST and SOS varied greatly within the range of 0–6 km and then tended to be stable. In summary, the LST and SOS showed a negative correlation, and the coupling relationship between them would be further explored below.

**Figure 6.** Spatial distribution of the start of the growing season (SOS) in Hangzhou (left charts), and variation trend of SOS in the buffer zone at different distances from urban (right charts) in 2006, 2010, 2014, and 2018. The black line denotes the urban boundary, dark gray denotes the buffer of 10 km, and light grey denotes the buffer of 20 km. The DOY denotes the day of year. The solid lines denote the urban boundaries, and the wide and short dotted lines denote the 10 km and 20 km buffers of the urban boundaries, respectively.

#### *3.2. Relationship between LST and SOS in Hangzhou*

The results above showed that the spatial distribution and urban–rural gradient of the average LST in spring and SOS showed an opposite trend. Therefore, it could be inferred that there was negative correlation between LST and SOS. Figure 7 shows the coupling relationship between LST and SOS, and the 94.6% of the area conformed to the inference above. The LST and SOS showed a significant negative correlation accounted for 53.9%, more than a half of the total area. Among them, the low LST-high SOS (accounting for 6.2%) mainly distributed in the northern edge of Hangzhou, the outer rural farthest from urban. The medium LST–medium SOS (45.3%) was mainly distributed in the middle part of Hangzhou, with a moderate distance from urban. The high LST–low SOS (2.4%) was mainly distributed in urban and within 2 km from urban. At the same time, there was 40.7% of the area that LST and SOS showed a weaker negative correlation, including low LST–medium SOS (10.9%), medium LST–low SOS (12.2%), medium LST–high SOS (10.1%), and high LST–medium SOS (7.5%), mainly distributed in the area between urban and outer rural. Besides, there was 5.4% of the area that exhibited a contrary relationship to the inference. That is, the LST and SOS showed a significantly positive correlations, which were low LST–low SOS (3.3%) and high LST–high SOS (2.1%). It might be related to the threshold of LST and SOS for the classification. In addition, in areas with low, moderate and high LST, 83%, 67% and 82% of the SOS has a medium-high, moderate and medium-low distribution, respectively. Overall, the results above confirmed the inference that spring LST was negatively correlated with SOS.

**Figure 7.** Coupling relationship between land surface temperature in spring and the start of the growing season (SOS) of 2018 in Hangzhou. The 3D pie chart denotes the percentage of each coupling relationship, and the flat pie charts denote the percentage of SOS by level at different temperatures. The solid lines denote the urban boundaries, and the wide and short dotted lines denote the 10 km and 20 km buffers of the urban boundaries, respectively.

In order to further verify the inference above, the annual average SOS of 2006–2018 and its change trend at different levels of LST was calculated to explore the relationship between LST and SOS. As shown in Figure 8a, the annual average SOS at low, medium, and high LST were 88.9, 85.8, and 85.0 DOY, respectively. The SOS continently decreased with the increase of the LST. That is, the spring phenology continued to advance. As shown in Figure 8b, the change trend of SOS at low, medium, and high LST were −5.1, −3.9, and

−2 days/10 years, respectively. The downward trend of SOS decreased with the increase of LST. That is, the rate of advancement of phenology continuously slowed down. In general, the inference established that there was a negative correlation between SOS and LST, and it developed in a consistent direction, showing a trend of convergence.

**Figure 8.** Distribution of (**a**) the annual average of the start of growing season (SOS) of 2006–2018 and (**b**) its interannual variation trend under different temperatures. Slope denotes the slope of the linear regression line between SOS and year. Low, Medium and High denote different temperatures according to the natural breakpoint classification method. The DOY denotes the day of year. In the box charts, the box denote the values of median, lower quartile (Q1) and upper quartile (Q3), respectively; the error bars denote the values of Q1 − 1.5(Q3 − Q1) and Q3 + 1.5(Q3 − Q1), respectively.

#### *3.3. Relationship between LST and SOS at Sample Points*

The relationship between LST and SOS at the sample point scale was further employed Figure 9. The LST in the urban of Hangzhou was significantly higher than that in the rural, exhibiting a significant urban heat island effect. In terms of the LST during the daytime in spring, the difference between that in the urban (23.0 ± 1.2 ◦C) and the rural (19.7 ± 1.0 ◦C) was significant, reaching 3.3 ± 1.0 ◦C. From 2006 to 2018, the difference continued to increase, with an average annual increase of 0.2 ◦C (*p* < 0.01). For the LST during the nighttime in spring, the difference between that in the urban (8.2 ± 1.0 ◦C) and the rural (7.0 ± 0.7 ◦C) reached 1.2 ± 0.3 ◦C. Compared with the daytime, the urban heat island effect was weaker at night, reduced by 2.1 ◦C. Integrating the temperature during the daytime and nighttime Figure 9c, the difference of the daily average temperature between that in the urban (15.6 ± 0.8 ◦C) and the rural (13.3 ± 0.8 ◦C) reached 2.3 ± 0.5 ◦C, and it continued to increase from 2006 to 2018 (Slope = 0.09 ◦C/years, *p* < 0.01). To further compare the difference of SOS between that in the urban and the rural Figure 9d, the SOS of urban (79.3 ± 5.6 DOY) was 7.4 ± 2.7 days earlier than that in the rural (86.7 ± 5.0 DOY). It indicated that the plant phenology changed significantly under different environmental backgrounds in the urban and the rural. Besides, it was worth noting that, except 2012 (1.5 days) and 2015 (12.9 days), the difference of SOS between that in the urban and the rural was relatively stable from 2006 to 2018, with a difference of 7.5 ± 1.6 days, and it was consistent with the difference of 6–9 days in Hangzhou Figure 6.

As shown in Figure 10, the LST during the daytime rather than nighttime showed a statistically significantly negative correlation with SOS both in the urban and the rural. In terms of the LST during the daytime Figure 10a, the sample points of LST–SOS in the urban and the rural distributed significantly separately, with LST of 19–28 ◦C and 16–23 ◦C, and SOS of 40–100 DOY and 50–110 DOY in the urban and the rural, respectively. Besides, the SOS in the rural was statistically significantly negatively correlated with the LST (Slope = −1.64 days/◦C; *p* < 0.01). The correlation in urban had a degree of significance (Slope = −1.01 days/◦C, *p* < 0.05), and the change trend of the SOS with LST increasing in the urban was smaller than that in the rural. For the LST during the nighttime Figure 10b, the sample points of LST-SOS in the rural and the urban had poor separation and high similarity. There was less statistically significant correlation between SOS and LST (*p* > 0.05), indicating that the LST during the nighttime had little effect on SOS. However, the fitted trends all showed that the SOS decreased with the increasing LST, which was consistent with the response of SOS to LST in Figure 8.

**Figure 9.** The interannual changes and differences of (**a**) the daytime land surface temperature in spring, (**b**) the nighttime land surface temperature in spring, (**c**) the daily average temperature, and (**d**) the start of growing season (SOS) in the urban and the rural from 2006 to 2018. The red/blue dotted lines denote the interannual changes, and the gray bars denote the differences. The DOY denotes the day of year.

**Figure 10.** The response of SOS in the urban and the rural to (**a**) daytime temperature and (**b**) nighttime temperature. The SOS means the start of the growing season. The red/blue solid points denote urban/rural sample data, the red/blue solid lines denote linear regression lines, and the regression equation and significance are shown in the illustration. The DOY denotes the day of year.

#### *3.4. Relative Contributions of* Δ*LST to* Δ*SOS at Sample Points*

The temperature contribution separation model was utilized to explore the contributions of the ΔLST to the ΔSOS under urbanization. The difference of SOS under urbanization between predicted by the model (7.3 ± 1.3 days) and observed through data (7.4 ± 2.7 days) was −0.16 ± 1.4 days Figure 11a. For the results predicted by the model Figure 11b, the ΔLST played a significant role in the advance of SOS under urbanization and the proportion of its contributions was relatively stable, despite the interannual fluctuations. Besides, we found that the ΔSOS dominated by the ΔLST contributed 72 ± 13.3% (5.3 ± 1.7 of 7.3 ± 1.3 days) to the difference of SOS between that in the urban and the rural Figure 11a. The advance of 28 ± 13.3% (1.9 ± 0.7 days) of SOS under urbanization was dominated by other factors such as photoperiod and air pollution. Overall, the local warming effect induced by the urban heat island effect produced substantial impacts on plant phenology under urbanization, but the impact of other factors also cannot be ignored.

**Figure 11.** (**a**) The contributions of the difference of land surface temperature between urban and rural (ΔLST) and other factors to the difference of plant phenology under urbanization and (**b**) their interannual variations from 2006 to 2018. The SOS means the start of the growing season. Quantile chart denotes the distribution of SOS differences from 2006 to 2018. Solid red points represent abnormal values. The DOY denotes the day of year.
