Asymmetrical Impact of Daytime and Nighttime Warming on the Interannual Variation of Urban Spring Vegetation Phenology

Urban warming significantly advances spring vegetation phenology. However, the potential effect of daytime and nighttime warming on the start of the urban vegetation growing season (SOS) remains to be determined. Here, we characterized the interannual response of SOS to daytime and nighttime warming from 2003 to 2020 using remotely sensed phenological observations across cities in the Northern Hemisphere (>30°N). We implemented the partial correlation analysis and the process‐based phenology model to quantify the effects of daytime and nighttime warming on vegetation. We found that either daytime or nighttime warming can promote an earlier urban spring SOS for Northern Hemisphere cities, while the phenological response varies across cities. Additionally, the response of SOS to daytime and nighttime urban warming (ST, expressed in advance days of SOS per °C warming) offset each other at an average rate of 0.08 days/°C per decade. Our results suggest that daytime warming predominates the temporal variation of SOS for high‐latitude cities, whereas nighttime warming is the primary driver of SOS change in low‐latitude cities. By revealing the effect and contributions of daytime and nighttime warming on the urban SOS, our results highlight the importance of considering daytime and nighttime temperatures separately in the response of vegetation phenology to urban thermal warming and climate variability.

Multi-source observations are employed to monitor vegetation phenology changes.Data sets that can be used for phenology research and change detection include site-specific observations of individual plant species (Chmielewski, 1996;Fitchett et al., 2015;Pellikka, 2001;Richardson et al., 2013) and satellite-derived phenological metrics (e.g., estimated based on vegetation indices) (Ganguly et al., 2010;Hmimina et al., 2013;X. Li, Zhou, Asrar, & Meng, 2017;Zhang et al., 2003Zhang et al., , 2018)).Observing individual plants and species directly for an extended period is often more prolonged than satellite images or repeat photographs currently accessible (X.Li, Zhou, Asrar, Mao, et al., 2017;Piao et al., 2019).However, the limitations of field phenology observations in their geographic scope, as well as the intense labor, make it challenging to draw generalized conclusions about the seasonal patterns of species and plant communities at broader geographical and climatic scales (Berra & Gaulton, 2021;Dronova & Taddeo, 2022;Pereira et al., 2013;Thackeray et al., 2016).Presently, time series data of vegetation indices or biophysical variables obtained from remotely sensed observations have been commonly used to derive specific phenological metrics (de Beurs & Henebry, 2008;Jönsson & Eklundh, 2004;X. Li et al., 2019b;Reed et al., 1994;Roerink et al., 2000;White et al., 2009).Advances in remote sensing-based land surface phenology can consistently capture phenological monitoring at broader ecosystem scales (Brooks et al., 2020;Dash & Ogutu, 2016;Oehri et al., 2017).Accordingly, satellite-based phenological observations play a pivotal role in exploring the response of terrestrial ecosystems to environmental changes across various levels, ranging from local to global scales (Caparros-Santiago et al., 2021;Radeloff et al., 2019;Zeng et al., 2020).
Understanding the shifting vegetation phenology caused by the urban thermal environment is important in the ecological, climatic, and health aspects (Calfapietra et al., 2015;Jochner & Menzel, 2015;Zhou, 2022).Urbanization modifies the surface radiation and energy balance, resulting in elevated temperatures in urban domains compared to natural ecosystems (Chrysoulakis et al., 2018;Oke, 1988;Oliveira et al., 2022).The urban thermal environment can impact vegetation phenology both within and around cities. Studies have shown a significant correlation between phenology and surface temperature in a wide range of scenarios (Jensen et al., 2022;Jia et al., 2021;D. Li et al., 2019;L. Li et al., 2022;Meng et al., 2020a;Yin et al., 2023Yin et al., , 2024;;Zhang et al., 2004).Given that vegetation phenology serves as the primary biotic factor regulating the urban microclimate (Schwaab et al., 2021;Su et al., 2020), the urban environment has been regarded as an "open laboratory" that offers the possibility for predicting how vegetation will perform in future warming (Calfapietra et al., 2015;Yang et al., 2023;Yin et al., 2024).The compound effect in the urban environment caused by the urban heat island effect and global warming can significantly impact the phenological growth stages and thus further alter the pollen release dynamics and increase the risk of respiratory allergies among sensitive populations (D'Amato et al., 2020;Pacheco et al., 2021;Zhang & Steiner, 2022).Moreover, vegetation phenology is also a crucial factor that affects the seasonal dynamics of terrestrial carbon balance in urban domains (Hardiman et al., 2017;Jochner & Menzel, 2015;Schwartz et al., 2020;Zhou, 2022).
Spring phenology simulations can enhance the comprehension of the global carbon balance and ecosystems' response to climate change (Calfapietra et al., 2015;Descals et al., 2023;Hufkens et al., 2018).Traditional phenological models driven by the daily mean air temperature might not be universally amenable for simulating and predicting the temporal dynamics of spring vegetation phenology (Fu et al., 2016;Piao et al., 2015).Several studies have documented that the response of spring phenology to daytime and nighttime warming is asymmetrical, showing distinct geographical variation across different natural ecosystems (Meng et al., 2020b;Shen et al., 2018Shen et al., , 2023;;Wang et al., 2021).Many studies have also explored the relationships between spring phenology and the urban thermal environment using a single temperature variable, for example, daily mean, daytime, nighttime temperature, and even the land surface temperature.For example, the correlation between urban phenology and land surface temperature using the impervious surface area as urban-rural urbanization gradients (Jia et al., 2021;Melaas et al., 2016;Walker et al., 2015;Y. Zhang et al., 2022) and the study of phenological differences between urban and rural areas using buffers generated outward from urban boundaries (Yin et al., 2023;Zhang et al., 2004;Zhou et al., 2016).Nevertheless, it remains unclear how daytime and nighttime temperatures influence the interannual variation in spring vegetation phenology in urban domains, which is important in predicting the spring phenological response in the projected temperature changes caused by the urbanization effect and climate warming.
In this study, we investigated the effect of daytime and nighttime temperatures on interannual variations of the start of the season (SOS) across cities in the Northern Hemisphere (>30°N) from 2003 to 2020 using remotely sensed phenology observations.We conducted a four-part analysis: (a) intercomparing the differences in optimal preseason length for urban SOS computed by daytime and nighttime temperatures; (b) interpreting the response of urban SOS to daytime and nighttime temperatures by performing partial correlation analysis; (c) assessing the interannual variation in temperature sensitivity of urban SOS; and (d) quantifying the contributions of daytime and nighttime temperature to urban SOS using temperature forced models.

Data Sets
The urban extent was obtained from the Global Urban Boundary data set (X. Li et al., 2020).Urban clusters with areas greater than 100 km 2 and located in the Northern Hemisphere (>30°N) were selected in this study.All of them are mid-latitude (i.e., between 30˚N and 65˚N) cities.We derived the SOS (2003-2020) from MODIS Land Cover Dynamics (MCD12Q2) with a spatial resolution of 500 m × 500 m (Friedl et al., 2022).The MCD12Q2 phenology data set has been validated with multiple independent in situ measurements (i.e., PhenoCam) and various satellite-based extraction algorithms to ensure realistic estimates of phenological metrics (Moon et al., 2019).The SOS was when the vegetation index (i.e., EVI2) curve first crossed 15% of the segment amplitude, which closely aligns with the initiation of budburst in field phenological observations (Meng et al., 2020b).We used the daily maximum and minimum near-surface air temperatures as T day and T night with a spatial resolution of 1 × 1 km from 2003 to 2020 (Zhang & Zhou, 2022).The daily precipitation and shortwave radiation were obtained from ERA5 provided by the Copernicus Climate Change Service.The MODIS Land Cover Type Yearly Global (MCD12Q1) was used to classify pixels of crop and urban (Friedl & Sulla-Menashe, 2022).We resampled all data sets to 1 km to match the resolution of air temperature data for analyses.We excluded crop-type vegetation because its spring phenology was subject to the timing of sowing rather than the climatic temperature (Yin et al., 2023).We also removed urban pixels that changed due to urbanization from 2003 to 2020 to abate the potential influence on SOS caused by the land cover change (Pan et al., 2023).To exclude potential biases caused by outliers and unreliable observations, we removed pixels with dates of SOS later than 150 days of the year (DOY) and earlier than 30 DOY.We also performed quality control on the SOS data using the quality assurance flags provided along with the MCD12Q2 data so that only pixels labeled with the best and good quality were used in our analysis.

Empirical Analyses
We used partial correlation analyses to investigate the effects of daytime (T day ) and nighttime temperature (T night ) on SOS.The correlation between SOS and T day (or T night ) was quantified after controlling for other confounding variables (i.e., T night (or T day ), precipitation, and shortwave radiation) (Piao et al., 2015;Wu et al., 2018).As spring phenology is relevantly related to the temperature of the preceding period (Menzel et al., 2006;H. Zhang et al., 2022), we implemented a 5-day interval approach from January 1st to the average SOS date to determine the preseason length.For each city, the period for which the absolute value of the partial correlation coefficient between SOS and T day (or T night ) was highest was considered the optimal length of the preseason most relevant to SOS.We performed a multiple linear regression model of SOS against mean T day , T night , precipitation, and shortwave radiation over the preseason to compute the temperature sensitivity of SOS (S T ).The multiple linear regression slope coefficients for T day and T night were defined as SOS's daytime temperature sensitivity (S Tday ) and nighttime temperature sensitivity (S Tnight ).Similarly, we also computed the optimal preseason length, partial correlation coefficients, and temperature sensitivity of SOS (S Tmean ) to daily mean temperature (the average of daytime and nighttime temperature) to explore the contrasting variations in responses of spring phenology to daytime and nighttime warming.

Process-Based Modeling
Because vegetation needs to accumulate a specific amount of heat during the end of winter or the beginning of spring before green-up (Chuine, 2000), we used the process-based phenology model (i.e., forcing) to simulate the physiological process of vegetation SOS in each city by the daily daytime, nighttime, and mean temperature.Here, we only applied the forcing requirement for spring phenology modeling because most chilling models overlook the effect of freezing temperature on the dormancy release (Wang et al., 2020).Models that assume a fixed start date for forcing accumulation in the spring have shown better performance than those based on chilling and forcing accumulation (Linkosalo et al., 2006).All models used in this study precisely captured the satellitederived SOS variations over time and across cities (Figure 1).
Earth's Future where S f is the states of forcing, t 1f is the initial days of the forcing period, R f is the rate of change of S f , T is the daily temperature (i.e., T day , T mean , and T night ), T opt is the optimal temperature for vegetation to start forcing day accumulation, and F* is the forcing threshold when the vegetation SOS is triggered.
To quantify the contributions of T day and T night to SOS, we developed an improved model based on the forcing model.In the day-night forcing model, Equation 2 is replaced as follows: if where k is the weighting factor of T day , and (1 k) is the weighting factor of T night .
Each parameter for the forcing model was optimized for each city using a simulated annealing algorithm from 2003 to 2020.The ecological meaning of each parameter was used to establish the range for each parameter.The optimal temperature for forcing days accumulation was set no smaller than 0 (°C) (Wang et al., 2021).k ranges from 0 to 1 to weigh the impact of T day and T night on the day-night forcing model accumulation.RMSE between simulated SOS sim i and observed SOS obs i was used as the objective function.The optimal parameters were determined based on the ones that generated the lowest RMSE value.
where n is the number of years for each city, SOS sim i is the ith SOS simulated by the model, SOS obs i is the ith observed SOS.
The accumulated growing days were calculated for each city in each year from the T opt values: where AGD D is the number of growing days accumulated from the beginning of the year (January 1st) until time of SOS, T opt is the optimal temperature for vegetation to start forcing day accumulation.

Differences in Optimal Preseason Length
The optimal length of preseason for T day , T mean , and T night exhibited significant differences from each other (Figure 2).The optimal preseason length of urban spring phenology calculated by T mean was the most extended across all cities in the Northern Hemisphere, averaging 45.78 ± 12.89 days, followed by T day (around 37.86 ± 15.01 days).By contrast, T night -based preseason length is about 32.00 ± 13.53 days.The Statistical distributions of optimal preseason length for T day , T mean , and T night were positively skewed (with skewness equaled to 1.05, 0.90, and 1.22, respectively).Even though the preseason lengths for T day , T mean , and T night were distributed diversely across geographical space, they were spatially correlated with latitudes (Figure 3).Cities at higher latitudes required longer preseason lengths than those at lower latitudes.The 1°increases in latitudes approximately extended the preseason length calculated by T day and T night of 0.65 days, and 0.55 days for T meanbased optimal preseason length.

Response of SOS to Temperature
Although warming in either T day or T night promotes an earlier spring SOS for most cities in the Northern Hemisphere, the phenological response of SOS varies across cities (Figure 4).In approximately 69.30% of cities, SOS was negatively correlated with T day .Similarly, negative correlations were found between SOS and T night in 63.92% of cities, and SOS negatively responded to T mean in up to 96.52% of cities.The results suggested that the higher temperature during the preceding preseason length caused the earlier SOS in most cities.We also found that at least one negative correlation between SOS and T day or T night was observed in almost all cities (>99%).
High T day and T night during the preseason advanced the SOS in 33.39% of cities.For 35.92% of cities, the increase in T day caused the advancement of SOS, while the increase in T night played a delayed role in controlling SOS.By contrast, increased T night and decreased T day resulted in advanced SOS in 30.54% of cities.However, we found no geospatial relation between the effects of T day and T night on the urban vegetation SOS.

Temporal Changes in S T
The opposite interannual S Tday and S Tnight offset each other, resulting in no significant temporal changes in S Tmean (Figure 5).Warming T day , T mean , and T night substantially caused SOS to advance in the Northern Hemisphere cities (S T < 0).We found that the declining S T of SOS based on T day with an average decreasing rate of 0.08 days/°C per decade and the rising S T of SOS based on T night with an average increasing rate of 0.08 days/°C per decade completely offset each other.Therefore, no temporal relationships were found for the S T of SOS based on T mean when the significance was set at p < 0.05.

Contributions of T day and T night to SOS
T day predominated in controlling the temporal variation of SOS for cities located at high latitudes, while T night is the primary driver of SOS in low-latitude cities (Figure 6).With a rate of 0.004, K a progressively increased along the latitudinal gradients while K b decreased.The spatial distribution generally showed greater K a of T day in higher latitudes cities (>0.8) compared to K b of T night (<0.2), indicating T day had a more significant impact on SOS compared to T night .As K b and K a were opposite, T night was dominant in regulating SOS for cities at low latitudes.

Discussion and Conclusion
This study uses direct comparisons of urban spring vegetation phenology in response to daytime, nighttime, and daily mean temperature from satellite observations by using partial correlation analysis.Preseason temperatures play a crucial role in the triggering of SOS.The phenological responses of preseason length to daytime, nighttime, and daily mean temperature are significantly different, indicating that daytime and nighttime warming affect vegetation phenology to different degrees.We provide evidence that interannual warming at night or during the day can cause SOS to start earlier in most cities.However, daytime and nighttime warming may have the opposite effect on SOS.It is also found that the decreasing daytime temperature sensitivity of SOS has been offset by a rise in nighttime temperature sensitivity, resulting in the balanced status of SOS to daily mean temperature in the last two decades.Our mixed day-night forcing model reveals the asymmetrical contribution of daytime and nighttime temperatures to the optimal forcing requirements of urban spring vegetation phenology in terms of latitudes.Daytime warming dominates the SOS in high-latitudinal cities.In contrast, SOS of low-latitudinal cities is much more influenced by nighttime warming.
While the declining trends of temperature sensitivity of leaves unfolding in response to global warming have been documented in temperate regions (Chen et al., 2019;Fu et al., 2015), the phenological response of vegetation to temperature is asymmetric in terms of daytime and nighttime temperature (Meng et al., 2020b;Piao et al., 2015).However, there has been no significant decrease in the sensitivity of the daily mean temperature of SOS in global cities over the past 20 years.Our findings are consistent with the results of recent studies that urban warming advances spring phenology (Jia et al., 2021;Meng et al., 2020a;Zhou et al., 2016).We also concluded that the rising temperature (daytime or nighttime or daily mean temperature) in urban environments enables spring SOS to begin earlier over time but toward different directions of changing rate.In keeping with the contrasting asymmetry of daytime and nighttime temperature effects on natural vegetation (Rossi & Isabel, 2017;Shen et al., 2018), the phenological response of urban vegetation to daytime and nighttime warming is also different.Despite evidence showing that, for specific tree species, the slowdown in the progression of spring phenology is ) smaller and greater (or equal) than 0, respectively.The negative and positive coefficients of partial correlation represent the rising temperature advances and delays SOS, respectively.The pink gradient colors represent the categories of the partial correlation coefficient.The--represents that T day and T night negatively correlated with SOS after controlling for precipitation and radiation.The -+ represents that T day negatively correlated with SOS, and T night positively correlated with SOS after controlling for precipitation and radiation.The +represents that T day positively correlated with SOS, while T night negatively correlated with SOS after controlling for precipitation and radiation.The ++ represents that T day and T night positively correlated with SOS after controlling for the precipitation and radiation.attributed to an increase in daytime temperatures rather than nighttime temperatures (Fu et al., 2016;Piao et al., 2015), our results demonstrate that either daytime or nighttime temperature can lead to the advancement of urban vegetation SOS, which is different from natural tree species.Moreover, the earlier occurrence of urban spring phenology due to warming nighttime temperatures offset the deceleration of spring phenology advancement caused by warming daytime temperatures.According to our models, cities located at high latitudes are mainly affected by T day in regulating the temporal variation of SOS, whereas T night is the primary factor controlling SOS in cities situated at low latitudes.
Vegetation needs a critical level of forcing temperature to trigger spring phenology (Chuine, 2000;Miller et al., 2001).Only temperatures above a specific threshold count in accumulating growing degree days (Fu et al., 2016).Due to the heterogeneity and fluctuation of temperatures among cities, daytime or nighttime temperatures may exceed the optimal temperature range for vegetation to sprout in some specific cities.As a result, a few cities could exhibit a weak positive interannual correlation between daytime and nighttime temperatures.Furthermore, as higher-latitude cities are colder than lower-latitude cities and daytime temperature is higher than nighttime temperature, daytime rather than nighttime warming during the preseason length fulfills the forcing requirement more efficiently to trigger leaf onset for vegetation in higher-latitude cities.In contrast, the daytime temperature in lower-latitude cities may exceed the optimal temperature threshold for vegetation to unfold leaves, while the relatively cooler nighttime temperature can easily meet the temperature requirement and thus starts to contribute to the spring vegetation onset predominantly.
The unique urban environment could be a potential mechanism for the opposite SOS sensitivity to T day and T night .On the one hand, the urban heat island effect arises from human activities and urban infrastructure, including concrete structures and railways, which absorb and retain heat, releasing it during nighttime hours and thus causing urban areas to maintain higher temperatures at night (Deilami et al., 2018;Gartland, 2012;Qin, 2015).As global climate warming continues, the rising temperatures at night fulfill the optimal temperature threshold for vegetation to trigger their sprout, whereas the rising daytime temperature exceeds this threshold.These could contribute to the declining daytime warming and increasing nighttime warming effects on the spring vegetation phenology in urban areas.On the other hand, although artificial light at night may disrupt the natural cycles of light and thus potentially prompt the advance in phenological timing, its impact on SOS is weaker than temperatures (Zheng et al., 2021).
The variable presence of multiple factors in the complex urban environment might affect the spring vegetation phenology to climate warming.In addition to temperatures, atmospheric emissions from cities are another influencing factor that affects the photosynthetic phenology of vegetation (Calfapietra et al., 2015).For example, carbon dioxide can not only alter the phenological timing directly but also drive changes in climatic factors such as temperature and precipitation that indirectly lead to phenological shifts (Cleland et al., 2007).Although vapor pressure deficit and soil moisture might affect vegetation growth for particular climate conditions (Cleverly et al., 2016;Sanginés de Cárcer et al., 2018), we did not include the two variables because they co-vary with temperature and precipitation (Piao et al., 2015;Prince et al., 1998;Sehler et al., 2019).Photoperiods may also jointly regulate the spring phenology (Körner & Basler, 2010).It shows steady interannual variations compared to temperature and, thus, is considered a hard limit preventing vegetation from frost damage due to earlier bud break (Laube et al., 2014;Richardson et al., 2018;Way & Montgomery, 2015).However, vegetation in cities is exposed to a higher thermal environment during the preseason than natural vegetation due to the urban heat island effect (Wohlfahrt et al., 2019;Zhang et al., 2004), so the temperature is supposed to predominate the spring phenology over photoperiod.The hard limit of photoperiod has not yet been reached in this study, which is partly supported by evidence that the forcing accumulated growing days required for the SOS have been continuously decelerating for the past decades (Figure 7).
Overall, including Northern Hemisphere cities in this study has substantially increased the diversity of city sample size, allowing the assessment of a broader range of outcomes and contributing to scientifically robust conclusions.Phenological models designed to simulate the spring transition date captured by parameters that have ecological meanings regarding the climate warming effect of the asymmetrical daytime and nighttime warming on the spring phenology are a priority for future research.As the PhenoCam data time series accumulates and the number of observation sites within urban areas is employed, integrating site and satellite observations will offer substantial data support to derive robust and insightful findings in the research on urban vegetation phenology.Future works also need to focus on disentangling the influence of the complicated and changeable environment caused by anthropogenic factors on vegetation phenology to comprehensively understand the response of urban vegetation to future climate change.

Figure 1 .
Figure 1.Comparison between satellite-derived and predicted the start of the season (SOS) (days of the year).(a-d), Compare satellite-derived SOS and SOS simulated by T day & T night , T day , T mean , and T night , respectively.The solid black lines represent linear regression.The dotted black line represents the 1:1 line.The color gradient represents the scatter density, with red indicating high density and dark blue indicating the low density of observations.

Figure 2 .
Figure 2. Distributions of optimal preseason length of urban vegetation SOS in the Northern Hemisphere.(a), the average optimal preseason length for T day , T mean , and T night .(b-d) The optimal preseason length frequency distribution for T day , T mean , and T night , respectively.The error bars represent one standard deviation.The * denotes a significant difference at p < 0.01 based on the paired-samples t-test.

Figure 3 .
Figure 3. Spatial distribution of optimal preseason length.(a-c), the optimal preseason length calculated by T day , T mean , and T night , respectively.The circles represent city locations.The gradient colors represent the optimal preseason length, ranging from 5 to 140 days.The dotted black lines represent linear regressions.The light shade areas represent one standard deviation.

Figure 4 .
Figure 4. Spatial distribution and Frequency of partial correlation coefficient between the start of the season (SOS) and temperatures.(a-c), the partial correlation coefficient between SOS and T day , T mean , and T night , respectively.(d) Categories of the partial correlation coefficient.The circles represent city locations.The blue and red circles represent the coefficient of partial correlation between SOS and temperature (i.e., T day , T mean , and T night .)smaller and greater (or equal) than 0, respectively.The negative and positive coefficients of partial correlation represent the rising temperature advances and delays SOS, respectively.The pink gradient colors represent the categories of the partial correlation coefficient.The--represents that T day and T night negatively correlated with SOS after controlling for precipitation and radiation.The -+ represents that T day negatively correlated with SOS, and T night positively correlated with SOS after controlling for precipitation and radiation.The +represents that T day positively correlated with SOS, while T night negatively correlated with SOS after controlling for precipitation and radiation.The ++ represents that T day and T night positively correlated with SOS after controlling for the precipitation and radiation.

Figure 5 .
Figure 5. Temperature sensitivity of the start of the season (SOS) over time.(a-c), temporal changes in temperature sensitivity of SOS calculated by T day , T mean , and T night with a 10-year moving window from 2003 to 2020, respectively.The dotted black lines represent linear regression.The light shade areas represent one standard deviation.Significance was set at p < 0.05.

Figure 6 .
Figure 6.Spatial distribution of the contribution of T day and T night to the start of the season (SOS).(a, b), the weighting factors of T day (K a ) and T night (K b ) that contributed to the SOS, respectively.The circles represent city locations.The gradient colors represent the size of the weighting factors.The dotted black lines represent linear regressions.The light shade areas represent one standard deviation.

Figure 7 .
Figure 7. Accumulated growing days of the start of the season (SOS) over time.(a-c), temporal changes in accumulated growing days of SOS calculated by T day , T mean , and T night with a 10-year moving window from 2003 to 2020, respectively.The dotted black lines represent linear regression.The light shade areas represent one standard deviation.Significance was set at p < 0.05.