Introduction

Global surface temperature in 2001–2020 was 0.99 °C higher than that in 1850–1900 (IPCC 2023), which had a profound effect on ecological processes. In the terrestrial ecosystem, the periodic and seasonal variations of meteorological factors have impacts on plant phenology (Schwartz et al. 2013; Yu et al. 2024). Understanding the correlation between phenology and climatic factors is essential for forecasting vegetation phenology and the response of terrestrial ecosystems to warming temperatures (Delpierre et al. 2016; Jiang et al. 2025). Besides, different species showed diverse responses to climate change, which may cause a reshaping of community structures and a change in the interactions among plants (Yang and Rudolf 2010; Renner and Zohner 2018). Leaf unfolding date (LUD) is one of the most conspicuous and accurate bio-indicators of ongoing climate change in terrestrial ecosystems, which has important effects on the development and reproductive success of trees (Zohner et al. 2016; Kharouba et al. 2018). Some previous studies have concluded that warming spring temperatures may lead to the advancement of leaf unfolding (Bigler and Vitasse 2019; Piao et al. 2019). However, warming winter results in insufficient chilling, and the sensitivity of LUD to mean temperature exhibits notable declining trends (Fu et al. 2015a; Chen et al. 2019; Dai et al. 2019).

The warming rate of nighttime temperature was more than that of daytime temperature during the recent decades, and daily minimum temperature (Tmin) has increased approximately 40% more quickly than daily maximum temperature (Tmax) (Davy et al. 2017). Temporal and spatial variations in LUD were triggered by asymmetric warming during the day and night (Rossi and Isabel 2016). Mean temperatures may not accurately reflect the warming trend in terms of the correlation between vegetation phenology and rising temperatures, and it is less useful in identifying diurnal asymmetric warming (Balducci et al. 2015; Rossi and Isabel 2016). Daily Tmin had a stronger and more positive effect on spring phenology and plant development than did daily Tmax in the northern hemisphere (Peng et al. 2013; Piao et al. 2015).

Plant buds of the woody species go through a dormant period before leafing out (Polgar and Primack 2011). Temperature has a pronounced impact on endodormancy and ecodormancy periods of overwintering buds for woody species (Delpierre et al. 2016). Endodormancy is a physiological adaptation mechanism for plants to avoid frost damage, and plants can only complete the subsequent phenological periods if they enter endodormancy normally. Low winter temperature is an important factor inducing dormancy initiation and termination of dormancy, and sufficient cold conditions are necessary for plants to end endodormancy (Hänninen 2016). Warming temperatures accelerate the growth of meristem cells and cell elongation during ecodormancy. When forcing requirements are met, ecodormancy is broken, and hence the plants begin to leaf out (Kramer et al. 2017). However, there are no obvious exact physical characteristics for the two stages of bud dormancy, making it difficult to determine when endodormancy ends and ecodormancy begins (Hänninen 2016). Warming temperatures result in advancing LUD by accelerating the forcing accumulation needed to break ecodormancy (Shen et al. 2016; Chen et al. 2019). Conversely, warming may reduce the amount of chilling requirement to induce endodormancy and ultimately delay spring phenology (Harrington and Gould 2015; Ford et al. 2016). Consequently, forcing requirement increases whereas chilling accumulation decreases, which makes LUD simulation complex under global warming (Fu et al. 2016). Asymmetric diurnal warming changed chilling and forcing accumulation rates (Peng et al. 2013), which caused the difference in LUD sensitivity to temperatures between day and night. For example, LUD was more strongly correlated with daytime temperatures than with nighttime temperatures in the majority of the northern hemisphere (Piao et al. 2015). The experimental warming studies showed that spring phenology was more sensitive to daytime temperatures than to nighttime temperatures (Fu et al. 2016; Rossi and Isabel 2016). At the middle and high latitudes of the Northern Hemisphere, (Deng et al. 2022) found that the sensitivity of the start of growing season to daytime temperatures was higher than to nighttime temperatures. Whether the asymmetric warming has opposing impacts on vegetative activity (e.g., phenology, productivity) has been a matter of debate (Zhu et al. 2022). The effects of warming during the daytime and nighttime on spring phenology are complex (Wan et al. 2009; Balducci et al. 2015). However, the chilling and forcing processes of Tmax and Tmin controlling the variations of LUD are not clear. Moreover, chilling and forcing requirements are different among tree species (Harrington and Gould 2015). Therefore, in order to fully understand the response of spring phenology to climate change, it is crucial to investigate and contrast the forcing and chilling requirements for various species.

Early leaf unfolding species (EFS) have lower heat requirement for leaf expansion than late leaf unfolding species (LFS) (Xu et al. 2021). For EFS, the LUD advances rapidly with warming temperatures, while LFS usually has a lower temperature sensitivity to avoid frost damage (Körner and Basler 2010; Stuble et al. 2021). The advance of LUD for EFS was significantly greater than that for LFS, which was due to higher dependence on photoperiod and chilling accumulation for LFS (Geng et al. 2020). Compared to EFS, LFS had higher photoperiod sensitivity of LUD and a more conservative response to temperature rise (Chen et al. 2019). The difference in phenological response between EFS and LFS leads to a significant extension of interspecific differences in spring phenology under warming temperatures (Geng et al. 2020). The dormancy state of EFS and LFS may be different, resulting in the divergent sensitivity of LUD to temperatures (Song et al. 2021). EFS is dependent on low temperatures to release dormancy and has a lower heat threshold, making it more sensitive to temperature rise, whereas LFS is sensitive to photoperiod and avoids premature leaf expansion to reduce frost risk. (Yu et al. 2023) found that several species in the subtropical regions have delayed leaf development dates due to a reduction in the amount of chilling, indicating that the impact of insufficient chilling caused by warming temperatures on leaf development in the subtropical regions was more pronounced than in the temperate regions. The impact of warming temperatures on phenology largely depends on local climate and plant types, and the response of phenology to warming may also vary among species (Lu et al. 2025). Therefore, it is necessary to explore how various factors regulate the effects of warming on the phenology of EFS and LFS over different warming magnitudes.

Different theories have been put out explaining how forcing and chilling requirements affect LUD. Some process-based phenological models, which incorporate endodormancy and ecodormancy, have been developed to estimate LUD (Chuine and Regniere 2017). These models include three types: only considering forcing temperatures (e.g., growing-degree-days (GDD) model), taking chilling and forcing temperatures into account (e.g., Unichill model), and adding photoperiod as an extra element interacting with the environment throughout both the endodormancy and ecodormancy phases (e.g., DORMPHOT model). However, the simple models (e.g., GDD model) usually performed better than the complicated ones (Olsson and Jönsson 2014; Basler 2016). Based on the premise that daily temperature is the most significant factor among all causative agents of phenological variations, the GDD model is applied to predict phenological phases. The GDD model is one of the earliest plant phenology models (Chen et al. 2015; Fu et al. 2015a), which assumes that phenological occurrence is driven by cumulative temperatures above a predefined threshold during a specific period (Sharratt et al. 1989; Rauschkolb et al. 2025). LUD estimated by the GDD model, however, still deviates from in situ measurements, leading to significant uncertainty in predicting how water and carbon balances of the terrestrial ecosystems respond to climate change (Levis and Bonan 2004; Richardson et al. 2012). At present, plant phenology is estimated according to functional type instead of species in most of the process-based models (Chamberlain and Wolkovich 2023). It is important to predict spring phenology across functional types by the GDD requirement. A previous study showed that spring phenology for early-active species (e.g., shrubs) was less accurately estimated by the GDD model than for the later-active species (Chamberlain and Wolkovich 2023). In the traditional GDD model, only mean temperatures are used to estimate LUD. Daytime and nighttime temperatures are equally weighted to determine the GDD threshold, which leads to an imprecise prediction of heat demands (Miller et al. 2001). The warming rate of daily Tmax was substantially increasing, while the warming rate of daily Tmin remained relatively stable (Zhong et al. 2023). Daytime and nighttime temperatures had different impacts on spring phenology, and the distinct phenological responses on Tmax and Tmin may vary at different rates. In agricultural ecosystems, high and low temperature thresholds have been incorporated into the phenological models to improve the prediction accuracy of plant growth (Zhou and Wang 2018). To deeply understand the impact of plant phenology on the terrestrial ecosystem function, phenological models that accurately predict future phenological changes and incorporate biosphere–atmosphere feedbacks are indispensable (Piao et al. 2019).

Most phenology models are developed for specific species and primarily use average temperature to predict spring phenology (Wang et al. 2020). Moreover, some phenology models that incorporate diurnal temperature variation effects are primarily used to simulate autumn leaf coloring date. However, the estimation of LUD is still relatively limited. Furthermore, compared with satellite remote sensing, which typically has a spatial resolution coarser than 250 m and captures canopy-scale phenological information (Peng et al. 2017), ground-based phenological observations enable more accurate differentiation of tree species and are unaffected by challenges such as mixed pixels and spatial heterogeneity. Consequently, they provide a more reliable basis for investigating the response of different tree species to climate change, thereby improving model accuracy. In this study, we employed 4,320 records of LUD in China to evaluate the effect of temperature on LUD. We hypothesized that there was a difference in the influence of daytime versus nighttime temperatures on LUD. We determined temperature-relevant period (TRP) to explore the relationship between LUD and preseason temperatures, and compared the sensitivity of LUD to daytime and nighttime temperatures between EFS and LFS. We further calculated forcing and chilling accumulations for two distinct periods (1983–1997 and 2000–2014) to explore whether chilling and forcing regulate leaf unfolding. In addition, we employed statistical model and process-based phenological models to analyze the divergence in the sensitivity of woody LUD to daytime versus nighttime temperatures.

Materials and methods

Phenological and meteorological data

The LUD data for 10 deciduous tree species and 2 shrubs across 12 sites in China were obtained from the National Earth System Science Data Center (http://www.geodata.cn/). The day of year was used to describe LUD. We excluded sites with records less than 22 years between 1983 and 2014, or fewer than 7 years during 1983–1997 and 2000–2014. The 12 phenology observation sites are shown in Fig. 1. The 12 species were composed of six EFS and six LFS (Table 1).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Locations of phenological observation sites in China

Table 1 The information on early leaf unfolding species (EFS) and late leaf unfolding species (LFS) across all sites
Full size table

Each phenological site was matched with a single meteorology observation station based on geographic coordinates, altitude, and topographic position. Daily mean air temperature, Tmax, Tmin, precipitation, and solar radiation at 12 sites from 1983 to 2014 were obtained from the China Meteorological Data Service Center (http://data.cma.cn/).

Temperature sensitivity of leaf unfolding date

To analyze the effects of different climatic variables on LUD, we first controlled two of three variables (temperature, precipitation and radiation), calculating the coefficient of partial correlation between the rest climatic variables and LUD, further proving that temperature was the major driving factor. Temperature sensitivity (ST) has been extensively used to evaluate phenological response to climate change (Fu et al. 2015a; Chen et al. 2020), which was the slope of linear regression between LUD and optimal preseason mean temperatures for many years (Dai et al. 2019). TRP was a period before leaf unfolding date, and it was determined by the highest absolute value of the coefficient of partial correlation between LUD and a 15-day window mean temperature before leaf unfolding date (Fu et al. 2015b). In order to exclude the impacts of precipitation and radiation on LUD, we used partial-correlation analyses to calculate TRP. Specifically, when the TRP of Tmax (or Tmin) was calculated, both precipitation and radiation were set as covariates.

Then, we calculated the temperature sensitivities (STmax and STmin) based on the TRP:

$$S_{{\rm T}}= - r_{{{\text{dt}}}} \, \times \,\frac{{S_{{\rm{d}}} }}{{S_{{{\rm{tp}}}} }}$$
(1)

where rdt is Pearson’s correlation between phenological onset dates and preseason temperatures, sd is the empirical standard deviation of phenological onset dates, and stp is the empirical standard deviation of preseason temperatures (Güsewell et al. 2017). In 1998 and 1999, there were data gaps because of a lack of phenology observations. We computed the sensitivities of LUD to Tmax and Tmin (STmax and STmin) during 1983–1997 and 2000–2014. Increased diurnal temperature range in spring suggested that the warming rates of daytime and nighttime were asymmetric. Therefore, we explored the difference in LUD in response to daytime and nighttime warming.

Calculations for chilling and forcing requirements

Woody plants require chilling and forcing accumulation to break dormancy. The previous study has concluded that the temperature slightly more than freezing can meet chilling requirement (Kramer et al. 2017), which ranges between 0 and 5 ℃ across all species (Wang et al. 2020). Chilling requirements were obtained by the total number of days when mean temperature higher than 0 °C and lower than 5 °C. At each site, the total number of chilling requirements was determined for each species ranging from 1st November to the mean LUD for the years. It was estimated as (Fu et al. 2015a; Wang et al. 2020):

$$T_{\max } \text{-based chilling requirement} = \,\sum\nolimits_{{t_{0} }}^{{t_{1} }} {1\quad \left( {0\, \le \,T_{\max } \, \le \,5} \right)}$$
(2)
$$T_{\min } \text{-based chilling requirement} = \,\sum\nolimits_{{t_{0} }}^{{t_{1} }} {1\quad \left( {0\, \le \,T_{\min } \, \le \,5} \right)}$$
(3)

where Tmax- and Tmin-based chilling requirements are the chilling requirements of Tmax and Tmin, respectively, t0 is November 1st of the previous year, t1 is mean LUD. We calculated chilling accumulation based on Tmax and Tmin during 1983–1997 and 2000–2014, respectively.

When daily Tmax (or Tmin) was more than the threshold temperature (0 ℃), the accumulated growing degree days were used to calculate Tmax (or Tmin)-based forcing requirements. Forcing accumulation was calculated from January 1st to average LUD (Fu et al. 2015a):

$$T_{\max } \text{-based forcing requirement} = \,\sum\nolimits_{{t_{0} }}^{{t_{1} }} {T_{\max } \quad \left( {0\, \le \,T_{\max } } \right)}$$
(4)
$$T_{\min } \text{-based forcing requirement} = \,\sum\nolimits_{{t_{0} }}^{{t_{1} }} {T_{\min } \quad \left( {0\, \le \,T_{\min } } \right)}$$
(5)

where Tmax- and Tmin-based forcing requirements are the forcing accumulations of Tmax and Tmin, respectively, t0 is January 1st of the current year, t1 is mean LUD. We calculated forcing accumulation based on Tmax and Tmin during 1983–1997 and 2000–2014. The differences in chilling and forcing accumulation between two periods were compared by paired t-tests.

Models for predicting leaf unfolding date

To examine the divergent impact of daytime and nighttime temperatures on LUD, we compared the threshold, GDD and day-night-temperature GDD (DNGDD) models. The threshold model was based on mean temperature (Krinner et al. 2005), and the threshold was determined by a simple statistical analysis. The average temperature for five days before LUD was determined, and the multiyear average temperature was used as the threshold. The GDD model was also used mean temperature to simulate LUD (Krinner et al. 2005):

$$GDD\left( t \right) = {\text{max}}\left( {T\left( t \right) - T_{{{\text{b}},0}} } \right)$$
(6)
$$GDD_{{{\text{threshold}}}} \, = \,\sum\nolimits_{{t\, = \,t_{0} }}^{y} {GDD\,\left( t \right)}$$
(7)

where Tb,0 is the base temperature (= 0 °C), T(t) is the mean daily temperature, t0 is the day on which the accumulation begins (set to January 1st), y is LUD and GDD(t) is the growing degree days at a date t, GDDthreshold is the requirement of cumulative growing degree from t0 to LUD. When GDD(t) was greater than the multiyear average GDD threshold, the date was LUD. The original GDD model was improved as the DNGDD model (Wu et al. 2018):

$$GDD\left( t \right)\, = \,k\, \times \,\max \left[{T_{{\text{day}}}\left( t \right) - T_{b,\,0}} \right] + \,\left( {1 - k} \right)\, \times \,\max\left[ {T_{{\text{night}}}\left( t \right) - T_{b,\,0}}\right]$$
(8)

where Tday(t) and Tnight(t) are daily maximum and minimum temperatures at a date t, respectively. k is the weighting factor and it is calculated by the ratio of RTmax to RTmin. RTmax and RTmin are the coefficients of partial correlations between LUD and Tmax, Tmin. If k > 1 or k < 0, Tday and Tnight have opposite impacts on LUD; if 0 < k < 1, they have the same effects on LUD (Wu et al. 2018). The above analyses were conducted using Matlab 2020b.

Model evaluation and data analysis

Phenological and meteorological data of the odd and even years were used for model fitting and validation, respectively. The coefficient of correlation (R2) and root mean square error (RMSE) between simulated and measured LUD were applied to evaluate the model fitting performance. The degree of agreement between the measured and simulated LUD was displayed using a Taylor diagram (Taylor 2001). The correlation coefficient, the root mean square difference between simulated and measured LUD, and the ratio of the standard deviations were all represented by a single point on a 2D plot.

Results

Responses of leaf unfolding date to T max and T min

Figure 2 shows the temporal trends of average monthly maximum and minimum temperatures from November of the prior year to May of the current year in 1983–2014. Tmax and Tmin from January to April increased significantly (p < 0.05). In March, Tmax and Tmin showed a remarkable rising trend of 1.1 and 1.6 °C per decade, respectively (p < 0.001). The diurnal temperature range in March increased significantly by 0.3 °C per decade during 1983–2014 (p < 0.05). It is indicated that the daytime and nighttime warming rates were asymmetric.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Long-term trends in monthly average maximum, minimum temperatures and diurnal temperature range at 12 sites from November to May in 1983–2014. Straight lines represent the changed trends of temperatures across all sites

For EFS, at most of the sites, TRP-Tmax ranged from 30 to 75 days and the average value was 45 days, and TRP-Tmin was similar to TRP-Tmax (Fig. 3a and b). For LFS, most of the TRP-Tmax values ranged from 30 to 105 days, with an average of 60 days, and the range of TRP-Tmin was the same as that of TRP-Tmax. The TRP of EFS was shorter than that of LFS. For most of EFS and LFS, LUD exhibited negative correlations with daytime and nighttime temperatures, suggesting that daytime and nighttime warming may advance LUD. Daytime and nighttime warming showed identical effects on LUD. The RTmax and RTmin were −0.63 and −0.67 for EFS, −0.49 and −0.51 for LFS, indicating Tmin had a slightly higher impact on LUD than Tmax (Fig. 3c and d).

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Differences in the temperature-relevant period (TRP) and the partial-correlation coefficients between early leaf unfolding species (EFS) and late leaf unfolding species (LFS) across all sites over the period 1983–2014. TRP-Tmax and TRP-Tmin represent the temperature-relevant periods for maximum and minimum temperatures, RTmax and RTmin indicate partial-correlation coefficients for maximum and minimum temperatures. The horizontal lines refer to the average value

Temporal variations of STmax and STmin in leaf unfolding date

The STmax values of LUD were negative during 1983–1997 and 2000–2014 (Fig. 4), suggesting that the increasing Tmax advanced LUD. Across all species and sites, average STmax decreased by 4.33% from −2.54 days ℃−1 during 1983–1997 to −2.43 days ℃−1 during 2000–2014. The STmax for EFS was higher than that for LFS. For EFS, LUD showed a more significant advancing trend during the warming period (2000–2014) than during the cooling period (1983–1997) when daytime temperatures increased by 1 ℃. Mean STmax for half of the species (primarily LFS) in 1983–1997 was significantly more negative than in 2000–2014 (p < 0.01), indicating a declining effect of Tmax on LUD in 2000–2014. For different species, Metasequoia glyptostroboides had the highest STmax (−3.43 days ℃−1) in the period of 1983–1997, followed by Fraxinus chinensis, and Firmiana simplex had the lowest STmax (−1.56 days ℃−1). During the two 15-year periods, the advance in LUD for Melia azedarach resulting from warming daytime temperatures decreased more than that for other species. However, the differences in STmax of LUD for Fraxinus chinensis and Robinia pseudoacacia during two 15-year periods were not significant.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Comparison of the sensitivity of leaf unfolding date to maximum (STmax) and minimum temperatures (STmin) across all species from 1983 to 2014. EFS early leaf unfolding species, LFS late leaf unfolding species. Please refer to Table 1 for the species codes shown in the figure

Similarly, the STmin values of LUD during 1983–1997 and 2000–2014 were negative (Fig. 4), showing that increased Tmin also resulted in the advance of LUD. Across all species and sites, average STmin declined by 40.3% from −4.32 days ℃−1 during 1983–1997 to −2.58 days ℃−1 during 2000–2014. The STmin for EFS was greater than that for LFS. Except for Salix babylonica, STmin of other species decreased significantly (p < 0.05). The advancing effect of nighttime temperatures on LUD in 1983–1997 was significantly stronger than that in 2000–2014. Fraxinus chinensis had the strongest STmin (−5.73 days ℃−1) in 1983–1997, followed by Metasequoia glyptostroboides, and Firmiana simplex had the lowest STmin (−2.72 days ℃−1). During 2000–2014, Fraxinus chinensis had the highest STmin (−5.39 days ℃−1), and Melia azedarach had the lowest STmin (−0.83 days ℃−1). For LFS, the difference in the sensitivity of LUD of Lagerstroemia indica to nighttime temperatures was the greatest among all species during the periods of 1983–1997 and 2000–2014. STmin of Lagerstroemia indica was positive in 2000–2014, showing that nighttime temperatures delayed LUD. The difference in STmax between 1983–1997 and 2000–2014 was lower than that in STmin. The advancement of LUD due to nighttime warming slowed down in 2000–2014.

Comparison of forcing and chilling requirements between two periods

The differences in Tmax- and Tmin- forcing requirements between 1983–1997 and 2000–2014 are shown in Fig. 5. For most of the EFS, mean Tmax-forcing requirements of other EFS in the cooling years (1983–1997) were higher than that in the warming years (2000–2014) (Fig. 5a, p < 0.05). During 2000–2014, the mean Tmax-forcing requirements of Metasequoia glyptostroboides and Cercis chinensis were higher than those of other species in EFS. For Prunus armeniaca, the difference in the mean Tmax-forcing requirements between 1983–1997 and 2000–2014 showed the most significant level among EFS. We found that the mean Tmax-forcing requirements increased remarkably for Melia azedarach and Firmiana simplex among EFS from 1983–1997 to 2000–2014. Across all LFS, the mean Tmax-forcing requirement of Robinia pseudoacacia was the lowest, and the difference in its mean Tmax-forcing requirement between the two periods was the most significant (Fig. 5a). Except for Robinia pseudoacacia, Morus alba and Salix babylonica, the mean Tmin-forcing requirements notably increased across most of the species from the cooling years (1983–1997) to the warming years (2000–2014) (Fig. 5b, p < 0.05). The mean Tmin-forcing requirement of Albizia julibrissin during the two periods was the highest. The differences in the mean Tmin-forcing requirements between the two periods were significantly greater for LFS than for EFS. We found that the mean forcing requirements of LFS were higher than those of EFS.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Differences in forcing and chilling requirements between 1983–1997 and 2000–2014 a daytime forcing requirements (Tmax-forcing), b nighttime forcing requirements (Tmin-forcing), c daytime chilling requirements (Tmax-chilling), d nighttime chilling requirements (Tmin-chilling). The triangle represents the mean value across all sites. Please refer to Table 1 for the species codes shown in the figure

Across all species, the mean Tmax-chilling requirements decreased significantly from the cooling years to the warming years (Fig. 5c, p < 0.05). The mean Tmax-chilling requirement of Melia azedarach was the lowest among all species during 1983–2014. The differences in average Tmax-chilling requirements for EFS were significantly lower than those for LFS during two periods. The mean Tmin-chilling requirements were significantly lower in 2000–2014 than in 1983–1997 (Fig. 5d, p < 0.05). The mean Tmin-chilling requirement of Melia azedarach during the two periods was the highest among all species. In particular, the differences in the mean Tmin-chilling requirements for LFS were larger than those for EFS during the two periods.

Performance of the threshold, GDD and DNGDD models

The threshold, GDD and DNGDD models underestimated LUD (Fig. 6a–c). The DNGDD model showed the best performance in simulating LUD across all species (R2 = 0.93, RMSE = 6.33 days, p < 0.01) compared to the threshold and GDD models (Fig. 6c). The simulated accuracy of three models was higher in EFS (threshold model R2 = 0.71, p < 0.01; GDD model R2 = 0.83, p < 0.01; DNGDD model R2 = 0.91, p < 0.01) than that in LFS. The R2 value revealed a more pronounced difference in LUD prediction between GDD and DNGDD models for LFS (p < 0.05) (Fig. 6b). LUD simulated by the DNGDD model was closer to the measured values than that estimated by the threshold and GDD models. The correlation coefficients between LUD estimated by the DNGDD and GDD models and the observed values were over 0.95 and 0.85, respectively, and the performances of the two models on LUD estimation in EFS were better than those in LFS (Fig. 6d–f).

Fig. 6
Fig. 6The alternative text for this image may have been generated using AI.
Full size image

Comparison of measured and simulated values of leaf unfolding date (LUD) using the threshold, GDD and DNGDD models

Discussion

Response of LUD to daytime and nighttime warming

Temperature is the dominant factor driving interannual variability of LUD. In the temperate climate regions, when temperature is above a certain threshold, plants can be triggered to initiate leaf unfolding (Chuine et al. 2000). A previous study showed that the variations in daytime and nighttime temperatures during the preseason may have opposite effects on spring phenophases (Meng et al. 2019). Piao et al. (2015) studied the impacts of daytime and nighttime temperatures on LUD, and they found that LUD was more strongly regulated by daytime temperatures than by nighttime temperatures. However, our results showed that preseason Tmin had a greater control on LUD than preseason Tmax. Zhu et al. (2022) solved the multicollinearity problem and concluded that warming during the day and night had similar impacts on vegetation phenology and productivity in the northern hemisphere. The ongoing trend of warming in the day and at night would lead to higher preseason accumulated temperatures, thus advancing LUD (Li et al. 2017; Meng et al. 2020b). LUD had a stronger partial correlation with Tmin than with Tmax. Nighttime warming advanced LUD, and the advance of leaf unfolding decreased from 1983–1997 to 2000–2014 (Fig. 4). The potential reason was that Tmin in the preseason (March) increased more rapidly than Tmax over 1983–2014. LUD had a negative relationship with Tmin as the warming of Tmin accelerated leaf unfolding by increasing the pace of phenological development. Similarly, Shen et al. (2016) reported that the vegetation green-up date had a larger negative partial correlation with Tmin than with Tmax in the Tibetan Plateau. The negative correlation between LUD and Tmax was because preseason daytime warming could advance spring phenology by increasing heat accumulation (Deng et al. 2022). Spring phenological information retrieved from remote sensing can provide various biophysical and chemical parameters at the regional and global scales (White et al. 2009). In contrast, spring phenology obtained by in situ observations reflects plant growth and development processes of individual and specific species (Menzel et al. 2006). Moreover, it can be used to distinguish the divergence in temperature sensitivity among various species, giving insights into the regulation of asymmetric climate warming on LUD of EFS and LFS.

We explored the divergence in TRP of LUD between EFS and LFS over the past 30 years under asymmetric daytime and nighttime warming, and calculated temperature sensitivity to quantify the trends in LUD. TRP of LUD for EFS was shorter than that for LFS, and temperature sensitivity was higher in the cooling years than in the warming years. The temperature sensitivity of LUD for temperate trees declined during 1981–2013 (Chen et al. 2019). It is suggested that the relative importance of rising average temperatures on spring phenology has decreased (Meng et al. 2020a). Additionally, Fu et al. (2015b) found a diminishing warming effect on leaf unfolding date of European tree species, which was a result of the reduction in chilling accumulation. In this study, it was found that STmin of LUD was higher than STmax during the period of 1983–1997. The STmax values of LUD in 1983–1997 and 2000–2014 were negative (Fig. 4). For the majority of LFS, the advancing effect of warming temperatures on LUD decreased, whereas the sensitivity of LUD for EFS to daytime temperatures slightly increased (Fig. 4). In Europe, LUD showed an advancing trend with rising daytime temperatures during 1951–2013, and the advancement of leaf unfolding significantly declined from 1951–1980 to 1981–2013 because of a lower chilling accumulation (Wang et al. 2021).

Effects of variations in chilling and forcing requirements

The advancement in LUD declined for most of the species with the warming minimum temperatures from 1983–1997 to 2000–2014. The LUD sensitivities to Tmax and Tmin decreased for most of LFS. We compared temperature sensitivity of LUD in 1983–1997 and 2000–2014, and found that decreased STmax and STmin were likely due to declining chilling requirements (Fig. 5c, d). Temperature sensitivity showed a synchronous change with chilling requirements from 1983 to 2014. Murray et al. (1989) demonstrated that heat requirements of 15 Britain woody plants decreased with the accumulation of chilling, and LFS was especially sensitive to chilling. Previous studies also proved that winter warming resulted in insufficient chilling (Laube et al. 2013; Fu et al. 2014, 2015b). Insufficient chilling not only affects the time of bud burst but also has an adverse impact on the health of the tree by reducing bud survival. Plants have conservative thermal traits to adapt to local temperatures, and forcing and chilling requirements reflect eco-physiological adaptation of leaf unfolding (Bennie et al. 2010; Zhang et al. 2017; Ren et al. 2020). The conservative thermal traits make plants withstand cold stress and ensure enough growing length for survival under climate change (Tylewicz et al. 2018). For some species, the advancing influence of daytime temperatures on LUD declined between 1983–1997 and 2000–2014, which was likely due to the difference in species-level thermal adaptation. Plants in the warm regions modify their enzyme activities to have higher forcing (or lower chilling) compared to those in the cool regions. Wang et al. (2023) used the linear mixed models to explore the interactions between chilling and forcing accumulation on LUD, and they found that the interactions were negative before leaf unfolding. It is indicated that greater forcing is required to trigger leaf expansion because of fewer chilling requirements (Harrington and Gould 2015).

Under future warming conditions, due to the decrease in chilling accumulations, plants require more forcing to compensate for the effects of declining chilling requirements on LUD during the same period. However, the increasing percentage of mean forcing requirements was not high (Fig. 5). In this case, a long time is needed for forcing accumulation to meet heat requirements. As a result of the decrease in chilling requirements, the advancing impact of rising Tmin on LUD slowed down (Du et al. 2019; Yin et al. 2021). When temperature is higher than the threshold value, plants start to accumulate forcing. The leaves begin to unfold only when the forcing requirements exceed the specific threshold (Chuine 2000; Harrington et al. 2010). The forcing requirements of daytime temperatures decreased from the cooling years to the warming years. Especially, the chilling requirements of EFS (e.g., Ulmus pumila, Fraxinus chinensis, and Prunus armeniaca) were higher than those of other species, and the decreased trends of forcing for Prunus armeniaca during 1983–1997 and 2000–2014 were more than those for other species. If chilling is sufficient, plants will need less forcing in the temperate and subtropical regions (Laube et al. 2013; Flynn and Wolkovich 2018; Du et al. 2019; Baumgarten et al. 2021), resulting in advancing LUD during daytime warming (Fig. 4). To avoid freezing damage, the temperature sensitivity of LFS was lower than that of EFS (Wang et al. 2014). LFS showed less sensitivity to chilling and forcing, and the end of chilling accumulation for LFS was earlier than that for EFS. LUD of EFS had higher temperature sensitivity than that of LFS, consistent with the findings of the previous study (Wolkovich et al. 2012). In 1983–1997, LUD was controlled by nighttime temperatures instead of daytime temperatures. Under warming conditions, nighttime temperatures had more significant impacts on LUD of EFS than daytime temperatures, while daytime temperatures had stronger effect on LUD of LFS (Fig. 4). Since LUD of EFS responded to climate warming differently from that of LFS, their ability to track rising temperatures may have a profound effect on species and forest ecosystems. The strong control of Tmin on spring phenology was also found in the Tibetan Plateau (Shen et al. 2016). The rising Tmin reduces the risk of frosts. Additionally, increasing Tmin promotes microbial activity, which helps plants emerge from dormancy (Heberling et al. 2019; Lee and Ibanez 2021). Therefore, nighttime temperatures had greater impacts on LUD than daytime temperatures.

Improvements of model performance

We can better understand how spring phenology responds to global warming by using the phenological models. Many physiological and ecological process-based models contain the phenology module, proving that changed phenology influences carbon cycling considerably. Most of the phenology models are developed based on daily mean temperature. The GDD model is developed based on the accumulation of forcing requirement over a certain threshold. Besides, forcing and chilling requirements are incorporated into the more complex models, including the sequential and parallel models (Basler 2016). Previous studies found that basic one-phase forcing models that only take into account the spring warming effect demonstrated comparable or even greater accuracy in predicting spring phenology than chilling forcing models, into which both the spring warming and autumn/winter chilling effects were incorporated (Linkosalo et al. 2006, 2008). Consequently, we predicted LUD based on one-phase forcing models.

Mean temperature was considered as the driver of leaf unfolding in the majority of models (Linkosalo et al. 2008). In the GDD model, only mean temperature was used to simulate LUD, but the response of phenology to other factors was ignored. The findings of Liu et al. (2016) showed that the temperature variability was more useful than mean temperature when the start of growing season of evergreen needleleaf forests was estimated. Additionally, Piao et al. (2015) concluded that Tmax instead of mean temperature resulted in interannual anomalies of LUD. Given the differential contributions of Tmax versus Tmin to LUD, the models based solely on mean temperature may perform poorly when we examine how spring phenology responds to temperatures. Considering the stronger control of nighttime temperatures on LUD in comparison with the daytime temperatures, we improved the traditional GDD model. In the DNGDD model, the weight of daytime and nighttime temperature forcing rate was adjusted according to the effects of RTmax and RTmin on LUD. The DNGDD model performed better in simulating LUD of EFS compared with that of LFS, which was because LUD for EFS was more sensitive to nighttime temperatures than to daytime temperatures during 1983–1997 and 2000–2014 (Figs. 4 and 6).

Conclusions

In this study, we explored the impacts of daytime and nighttime temperatures on woody LUD across 12 sites in China during 1983–1997 and 2000–2014. LUD was significantly regulated by temperatures, especially preseason temperatures. For most of the species, LUD was mainly controlled by Tmin rather than Tmax. Across all species, LUD was more sensitive to Tmin than to Tmax during the period of 1983–1997. Nighttime warming advanced LUD, while the advancing effect of nighttime temperatures on LUD decreased from 1983–1997 to 2000–2014. The DNGDD model performed better in estimating LUD across all species. These findings highlight the divergence in STmax and STmin of woody LUD. Therefore, the maximum and minimum temperatures should be incorporated into the phenology prediction of terrestrial models to accurately estimate ecosystem productivity in future studies.