Late to bed, late to rise—Warmer autumn temperatures delay spring phenology by delaying dormancy

Spring phenology of temperate forest trees has advanced substantially over the last decades due to climate warming, but this advancement is slowing down despite continuous temperature rise. The decline in spring advancement is often attributed to winter warming, which could reduce chilling and thus delay dormancy release. However, mechanistic evidence of a phenological response to warmer winter temperatures is missing. We aimed to understand the contrasting effects of warming on plants leaf phenology and to disentangle temperature effects during different seasons. With a series of monthly experimental warming by ca. 2.4°C from late summer until spring, we quantified phenological responses of forest tree to warming for each month separately, using seedlings of four common European tree species. To reveal the underlying mechanism, we tracked the development of dormancy depth under ambient conditions as well as directly after each experimental warming. In addition, we quantified the temperature response of leaf senescence. As expected, warmer spring temperatures led to earlier leaf‐out. The advancing effect of warming started already in January and increased towards the time of flushing, reaching 2.5 days/°C. Most interestingly, however, warming in October had the opposite effect and delayed spring phenology by 2.4 days/°C on average; despite six months between the warming and the flushing. The switch between the delaying and advancing effect occurred already in December. We conclude that not warmer winters but rather the shortening of winter, i.e., warming in autumn, is a major reason for the decline in spring phenology.

However, this advancing trend is slowing down. With increasing climate warming, spring phenology is not following at the same rate, as shown in observational Vitasse et al., 2018) and experimental studies (Chung et al., 2013;Morin et al., 2010). While spring temperature is known to be the main driver of spring phenology for a very long time (Polgar & Primack, 2011), there are further parameters influencing the timing of leaf-out. In addition to photoperiodic constraints , increasing winter temperatures and consequently insufficient chilling are discussed as reasons for the decline in spring advancement. (Asse et al., 2018;Laube et al., 2014;Polgar & Primack, 2011).
Temperate trees rely on safety mechanisms to prevent them from leafing out at the wrong time. When buds are formed during middle and late summer, they are initially only prevented from flushing by hormonal control of the active leaves (i.e., paradormancy) (Lang, 1994). At the time of leaf senescence and leaf fall, however, the buds need to be in a deep (endo-) dormant state, so that they will not open until spring, even if the weather should become temporarily warm again in autumn or winter. Thus, for dormancy release (also referred to as rest break), the buds need to be exposed to cool temperatures for a certain time (i.e., chilling) before becoming increasingly sensitive to warm (i.e., forcing) temperature and finally leaf-out (Polgar & Primack, 2011). The shift between the chilling and forcing period is seen as a gradual transition: With increasing accumulation of chilling, trees become more sensitive to forcing (Harrington et al., 2010). In other words, if they experienced less chilling due to winter warming, trees need more forcing temperatures to leaf-out Murray et al., 1989). Therefore, an increase in temperature may have contrasting effects on spring phenology, depending on the time at which it occurs.
Besides winter chilling, more recent studies have also reported on the impact of autumn temperature on spring phenology. A positive correlation between temperatures during autumn dormancy induction and spring leaf-out in the following year was found with warm autumn temperatures delaying spring phenology (Granhus et al., 2009;Heide, 2003;Søgaard et al., 2008;Westergaard & Eriksen, 1997). However, the causal link between autumn warming and subsequent spring phenology effects remains unclear. In addition, later autumn senescence was found to cause later spring leafout (Fu et al., 2014;Shen et al., 2020). While there is a growing body of research on spring phenology and the release of bud dormancy, the processes in autumn are comparatively less investigated. Not only are the environmental drivers of autumn leaf senescence not fully understood (Gallinat et al., 2015), also the timing of dormancy induction in tree buds and its environmental drivers are largely uncertain. The timing of dormancy induction, however, might be of particular interest, as it is very likely related to the time at which the accumulation of chilling temperatures starts. Therefore, we investigated the temperature sensitivity of the phenological cycle as a whole.
In order to project future growing season lengths and, consequently, terrestrial carbon sequestration, a causal understanding of seasonal differences in temperature sensitivity of leaf phenology is urgently needed. Most of our knowledge to date comes from extensive observations . However, experiments are needed to disentangle the environmental factors from each other and to provide a mechanistic understanding of plants' phenological response to increasing temperature , allowing for predictions of phenological shifts beyond the range of current conditions.
In this study, we investigated the phenological response of temperate trees to warming by disentangling temperature effects during different seasons from each other. We answer the question, during which period of the year elevated temperatures have an advancing or a delaying effect on leaf phenology. In a series of monthly warming manipulations at a realistic level of ca. 2.4°C from late summer until leaf-out in spring, we quantified the effects on leaf senescence, dormancy, and spring phenology in seedlings of two early and two late flushing common European forest tree species. To reveal the underlying mechanism, we tracked the progression of bud dormancy by repeatedly quantifying dormancy depth and tested how its induction and release are modulated by temperature. We hereby contribute to a deeper mechanistic understanding of the complex effects of climate warming on plant leaf phenology.

| Plant material
As study species we chose four common European tree species: Fagus sylvatica (European beech), Quercus robur (pedunculate oak), Betula pendula (silver birch), and Alnus glutinosa (black alder). Fagus and Quercus are late flushing species, Alnus and Betula flush early.
We obtained 2-year-old seedlings with seed sources from northern Germany from a local nursery in June 2018, planted them in 2 l pots each, and placed them in a common garden under ambient conditions. The species were evenly distributed among each other in blocks of four plants of each species. Each block was randomly assigned to a part of the experiment before the start. The pots were placed into sand to stabilize them and to prevent the roots from frost damage in winter. They were watered as needed. In total, we worked with 1324 seedlings, 294 each of the early flushing Betula and Alnus, and 356 each of the late flushing Fagus and Quercus.

| Dormancy development under ambient conditions
To track the pattern of dormancy induction and development, we assessed bud dormancy depth from August 2018 onward until spring flushing. Bud dormancy depth is defined as the time or growing degree days (GDD) required for a bud to flush under standardized optimal growing conditions (Halbritter et al., 2020). In late summer, the buds undergo a transition from shallow paradormancy to deep (endo-) dormancy, which is then gradually released over winter. High dormancy depth indicates that a plant is far away from leafing out, while low dormancy depth values indicate that the plant will soon leaf out with only a low input of additional warmth.
We started the dormancy depth assessment already in August, because we aimed to include the period of (endo-) dormancy induction, recognizable by increasing dormancy depth. Every week, we transferred three plants of each species from the common garden into a warm greenhouse with optimal growing conditions (21.5°C ± 2.3°C SD and 16 h light). If there were still leaves on the plants, we removed them to eliminate any inhibiting influence of leaves on the buds (Halbritter et al., 2020). Dormancy depth was quantified as the amount of GDD required by the seedling from its transfer into this warm greenhouse onward until unfolding of its first new leaf. GDDs were calculated as: where t 0 is the starting day at which the seedling was transferred to the warm, t 1 is the day at which leaf unfolding was observed, T mean is the daily mean temperature, and T base is a constant set to 5°C, representing a minimum temperature threshold required for stimulating budburst Polgar & Primack, 2011). To minimize potential smallscale environmental differences in the greenhouse, we reshuffled the position of the single pots three times a week. Note: Values for forcing units before January 1 are given in brackets because temperature before January 1 is not assumed to have a forcing effect. Note that for the early flushing Betula and Alnus, the experiment was only conducted until February.

| Temperature manipulation
We applied 4-week warming manipulations from September onward until February for the early flushing species and until March for the late flushing species. Two compartments of a greenhouse were used with natural light and fluctuating temperatures according to outside conditions. The control compartment with open windows and no heating had temperatures slightly above outside (0.6°C) and the warming compartment's temperature was maintained on average 2.4°C above the control (Table 1).
Each month, we transferred the preselected plants ( while the other half was used to destructively assess the dormancy depth directly after the treatment by the same method as described above for the dormancy development under ambient conditions ( Figure S1).
The resulting chilling and forcing units accumulated during each monthly warming period for each treatment are given in Table 2.
Chilling units were calculated as the number of days with daily mean temperatures between −2 and 10°C, according to recent findings of Baumgarten et al. (2021). Forcing units were calculated as GDD according to equation (1), where t 0 is the starting day and t 1 is the end of the respective manipulation period.

| Assessment of phenology
We observed autumn phenology at weekly intervals before, during, and after the September and October warming manipulation on those seedlings that were not used for dormancy depth assess- For spring phenology, we observed the plants three times per week and recorded the date at which the first leaf was unfolded, which was defined by the petiole being visible (Meier, 2001).

| Statistical analysis
The effect of warming on chlorophyll degradation, dormancy depth, and leaf-out dates was tested with linear mixed-effects model ANOVAs with the warming treatment, the month of the treatment, and the species as explanatory variables (R-package "lmerTest," (Kuznetsova et al., 2017)). Since data of dormancy depth and leaf-out were assessed only until February for early flushing Betula and Alnus and until March and April for late flushing Fagus and Quercus, two separate models were applied-one for all species until February and one for the late flushing species for the remaining months, respectively. Heteroscedasticity and normal distribution of the residuals were tested for each model by visual diagnostics, that is, inspecting the residuals versus fitted plots and the qq-plots of the residuals (Faraway, 2005). If necessary (i.e., for dormancy depth of all species until February), data were rank transformed. A Tukey's HSD post hoc test (R-package "emmeans," (Lenth, 2020)) was used to test for significance of differences between warming and control treatment for each month.
The effect size of warming on autumn and spring phenology was calculated as the difference in timing of phenological events divided by the difference in temperature between warming and control treatment. All statistical analyses were done in R version 4.0.2 (R Core Team, 2020). For graphical visualizations, the R packages gg-plot2 (Wickham, 2016) was used.

| Effect of warming at different periods on spring leaf phenology
We found contrasting effects of warming, depending on the time at which it was applied, while no difference between the species' responses to warming was observed (Table 3). From January onward, the seedlings responded with earlier spring leaf-out ( Figure 1 and Figure S2). This advancing temperature effect increased toward the time of flushing, reaching 2.5 days/°C in the month before leaf-out on average across all four species. However, while warming in winter did not postpone spring phenology at any time, we found a surprisingly strong delaying effect of warming in October on spring leafout ( Figure 1). Averaged over all species, leaf unfolding in spring was delayed by 2.4 days for every 1°C temperature increase in October. Figure S2 provides leaf unfolding dates of all trees and allows for comparison between warming and control of each manipulation period.

| Dormancy development under ambient conditions
Bud (

| Dormancy depth under warming
Warming had significant effects on dormancy development at the beginning and at the end of the dormant period ( Figure 3 and Table 4).

| Autumn senescence and leaf fall
While warming in September had no effect on autumn phenology, TA B L E 3 Effect of warming, the month in which the warming manipulation was applied and the species on the timing of spring leaf unfolding with all interactions (ANOVA statistics, two models were applied: one for all species until February and one for the late flushing species for March and April) F I G U R E 1 Effect of warming on spring leaf unfolding in days per °C warming at different periods during autumn, winter, and spring. Each warming manipulation lasted 4 weeks. Positive values indicate a delaying effect of warming on spring leaf unfolding. Asterisks denote significant differences in leaf unfolding days between warming and control with *p < .05, **p < .01, ***p < .001 (Tukey's HSD). The line is a LOESS curve integrating all species, for illustration of the overall trend. Note that no significant interaction between species and warming occurred in the first model until February (  Figure 4, Table 6). The day when SPAD values reached 50% was on average 11.7 days later after October warming compared to the control (Tukey's HSD p-value <.001, One potential explanation for the delay in spring phenology after autumn warming would be reduced chilling. However, warming in October did not affect the amount of chilling in comparison to the control, as neither under warming nor under control conditions any chilling units accumulated (Table 2). Also, when using different models (Hänninen, 1990;Wang et al., 2020), there was   (Table 2). Therefore, a lack of chilling cannot explain the later leafout after October warming.
Furthermore, the term "chilling" refers to the gradual decrease of dormancy depth under cool temperatures. At the beginning of the October warming period (27 September to 25 October), however, dormancy was still increasing and had not yet reached its peak ( Figure 2). Therefore, we propose that it was the establishment of dormancy, not its release, that was affected by warming in October. Dormancy establishment has rarely been studied to date, as neither the process itself nor its beginning or end is visible to our eyes.

| Delayed dormancy establishment due to warmer autumn temperatures
Warm temperatures during the establishment of bud dormancy in September and October caused a lower dormancy depth (Figure 3).
This could be interpreted in two different ways: (1) as an induction of weaker dormancy or (2) as a delay in dormancy induction.
If the first interpretation was correct, this weaker dormancy would result in faster dormancy release and in earlier leaf-out, which is not supported by our results. Therefore, we follow the second interpretation and conclude that autumn warming caused a delay in dormancy induction. This inference provides a mechanistic link between autumn warming and the delay of spring phenology: A later dormancy establishment delays the whole dormant phase, with a later peak and later release of dormancy. The buds become sensitive to warm temperatures later and, consequently, leaf-out later ( Figure 5). Mean 9.5 11.7 13.6 Note: The values are averages of eight tree seedlings, respectively. Bold numbers indicate significance (p < .05).

F I G U R E 5
Schematic illustration of the effect of warming on bud dormancy during its induction and its release. Autumn warming delays dormancy induction (resulting in lower dormancy depth values before (e.g., at time A), and higher dormancy depth values after (e.g., at time B) peak of dormancy and hereby reduced spring advancement of leaf phenology. The x-axis represents time and crosses the y-axes at 0 so that at the point where the dormancy depth curve crosses the x-axis, represents the time of a dormancy depth value of 0, that is, the time of leaf out [Colour figure can be viewed at wileyonlinelibrary.com] In contrast to our results, some earlier studies found deeper dormancy depth after warming during dormancy induction, but also associated with later leaf-out (Granhus et al., 2009;Heide, 2003;Malyshev, 2020;Søgaard et al., 2008;Westergaard & Eriksen, 1997).
This has been interpreted as a reinforcement of dormancy induction under warm temperatures (Hänninen & Tanino, 2011;Olsen et al., 2014). However, in all these studies, dormancy depth assessments were conducted either several weeks after the beginning of dormancy induction or the time of dormancy induction was unknown. In both cases, dormancy depth assessment very likely have taken place after the peak of dormancy depth, which is also supported by the continuously descending dormancy depth over time in the studies of Granhus et al. (2009) andHeide (2003). This suggests an alterna- In our experiment, October was the most sensitive month for dormancy induction, but in general it can be assumed that the crucial period is determined by the timing of autumnal temperature decline.
Which temperatures trigger dormancy induction and how they interplay with other environmental factors such as photoperiod clearly needs more research. In addition, although the dormancy delaying effect was apparent in all tested species, future studies on other species will examine the generality of our findings. In particular, the species and ecotypes are likely to be adapted in this regard to the climate and the typical duration of the non-growing at their place of origin.

| Delayed leaf senescence
Like bud dormancy establishment, autumn leaf senescence and leaf fall were delayed by warm temperatures in autumn. This is in line with previous studies which report a delay in leaf senescence due to warm temperatures in late summer and autumn (Delpierre et al., 2017;Fracheboud et al., 2009;Gunderson et al., 2012;Zani et al., 2020;Zohner & Renner, 2019). Fu et al. (2018) found in their experimental study a delaying temperature effect on leaf senescence by 6-8 days/°C in Fagus, which is well comparable to our results (Table S3).
Leaf senescence is not only important with respect to an elongation of the growing season, but it is functionally related to bud dormancy: As long as the leaves are active, their hormone release prevents the buds from opening (paradormancy, Howe et al., 2015).
However, from the time of leaf senescence onward, the buds depend on another reliable mechanism to avoid leaf-out at the wrong time, even if relatively warm temperatures still might occur. Therefore, bud dormancy induction and leaf senescence are very likely to be synchronized with each other, either triggered by related environmental drivers or internally coordinated. The temporal correspondence of autumn senescence and dormancy induction on the one hand and the impact of the timing of dormancy induction on spring phenology on the other hand provide now a causal explanation for the correlation between autumn and spring phenology (Fu et al., 2014;Shen et al., 2020): Later leaf senescence indicates later dormancy induction, which causes later leaf-out. If the temporal correspondence of leaf senescence and dormancy induction is maintained under atypical environmental conditions, such as drought, remains to be subject to further investigations.

| No delayed spring phenology after warming in winter
During November and December, warming had no significant effect on spring leaf-out despite reduced accumulation of chilling units (Table 2). Empirical knowledge on the effective chilling temperature ranges is still limited (Chuine et al., 2016). Baumgarten et al. (2021) recently reported a wide range of effective chilling temperatures from −2 to +10°C, temperatures that were given under both of our treatments. Another nonexclusive explanation would be that chilling and forcing temperatures can both reduce the dormancy depth (Malyshev, 2020), since both processes, chilling and forcing, operate simultaneously in the bud (Harrington & Gould, 2015). In the control compartment, the plants could have experienced more chilling, whereas in warming compartment, they experienced more forcing, both of which drive dormancy release and eventually resulting in similar spring budburst dates. In any case, chilling temperature in November and December does not seem to be a limiting factor, even though the study site has mild winters and the specific study year was 3.4 warmer than the long-term mean (1981-2010, average over September-April, Table 1).
During meteorological winter, that is, January-March, warming caused an earlier leaf-out, both in early and in late flushing species. The advancing effect started already in January and increased continuously toward the time of flushing (Figure 1), reaching 2.5 days/°C in the month before leaf-out (averaged over all species).
Interestingly, Roberts et al. (2015) report the shift between delaying and advancing effect of warm temperatures also around New Year. By that time, the buds had experienced enough chilling to become increasingly sensitive to warm forcing temperatures, in other words, they gradually entered the state of ecodormancy. Since we could not find any delaying effect of warming on from November onward, we conclude that the actual chilling temperatures is of less importance for dormancy release than the chilling time, that is, the duration of the cold period (Baumgarten et al., 2021). If this is true for other species and regions than investigate in Roberts et al. (2015), Baumgarten et al. (2021) and our study remain subject to further investigations.
Chilling units are usually calculated from a fixed day onward, very often from September 1. In this way, chilling might actually capture the process of dormancy establishment as well, as a high amount of accumulated chilling at a given time in spring is likely to be related to an early onset of cold temperatures in autumn in the year before. By investigating the whole cold period separated in monthly intervals, we found an advancing effect of cold temperatures only at its beginning. This suggest that the timing of the start of the cold season mainly determines the chilling effect on spring phenology.

| CON CLUS ION
Our staggered monthly warming experiments indicate that the observed decline in phenological sensitivity of trees to spring warming can be explained by a delay of dormancy induction due to warmer autumn temperatures. Elevated temperatures during winter (i.e., January-March) per se do not delay spring phenology, but rather have an advancing effect. However, climate warming is accompanied by a reduction in the number of days cold enough for chilling.
A shorter period of winter rest due to later dormancy induction in response to warm autumns decreases the effectiveness of forcing temperatures in spring and herby delays spring phenology. These insights will help to improve future process-based modeling of leaf phenology and growing season length. Our findings also call for greater attention to the phenological processes in the autumn, in particular to the mechanisms driving dormancy induction and their sensitivity to temperature.

ACK N OWLED G M ENTS
We kindly thank Anna Weißig, Kathrin Hölscher, and Felix Nößler for their great help during data collection. We are grateful for the mani-

AUTH O R S CO NTR I B UTI O N
AVM and IB conceived the study; IB together with CM and NL conducted the experiment and collected the data; IB performed the statistical analysis with contributions from JK; all authors discussed the results; IB wrote the manuscript with support from JK and AM.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data will be made publicly available via dryad.