Contributions of winter and spring warming to the temporal shifts of leaf unfolding


 Changes in winter and spring temperatures have been widely used to explain the diverse responses of spring phenology to climate change. However, our understanding of their respective roles remain incomplete. Using >300,000 in situ observations of leaf unfolding date (LUD) in Europe, we show that the advancement of LUD since 1950 is due both to accelerated spring thermal accumulation and changes in winter chilling which explain 61% and 39% of the LUD shifts, respectively. Winter warming did not substantially retard the releasing of bud dormancy, but increased the thermal requirement to reach leaf unfolding. The increase of thermal requirement and decreased efficiency of spring warming on accelerating thermal accumulation partly explained the temporally (1950s-2010s) decreasing response of LUD to warming. Our study stresses the need to better assess the antagonistic and heterogeneous effects of winter and spring warming on leaf phenology, which is key to projection of future vegetation-climate feedbacks.


Abstract (150 words) Changes in winter and spring temperatures have been widely
used to explain the diverse responses of spring phenology to climate change. 17 However, our understanding of their respective roles remain incomplete. 18 Using >300,000 in situ observations of leaf unfolding date (LUD) in Europe, we show 19 that the advancement of LUD since 1950 is due both to accelerated spring thermal 20 accumulation and changes in winter chilling which explain 61% and 39% of the LUD 21 shifts, respectively. Winter warming did not substantially retard the releasing of bud   Here, we take a new step forward to advance our quantitative and mechanistic 68 understanding in ongoing changes in leaf phenology. Using long-term (1951- Fig. S2). The forcing stage starts as soon as this chilling requirement is met and leaf 91 unfolding occurs when the thermal accumulation during the forcing stage exceeds a 92 given threshold. This threshold, denoted by TA0, declines exponentially with the total 93 amount of chilling received during the whole pre-growing season (CHAtot, Fig. S2b), 94 defined as the period from the onset of chilling accumulation till the LUD. Therefore, 95 the temporal shifts in LUD are determined by the time when bud dormancy is released 96 (df0), which in turn depends on the chilling accumulation rate (CHr, Eq. 1), by the 97 thermal accumulation rate (Fr, Eq. 2), and by the required amount of thermal 98 accumulation (TA0, Eq. 3) which in turn depends on the total amount of chilling 99 (CHAtot, Eq. 4) received before the LUD (Fig. S2).

100
The Unified model with nine optimized model parameters (see Methods and Table   101 S2) captured well the observed LUD for each species at each site (Fig. S3). For the six     CHrave due to winter warming can induce changes in df0 (Fig. 2) and explain a part of 217 the temporal variation in ST (Fig. 4), we found very limited changes in CHrave and df0.

218
In fact, at many observation sites, df0 actually occurred slightly earlier and not later in 219 time in response to the winter warming ( Figs. 1 and S4).

220
For all species, the increase of forcing rate consistently explained more variation in ST 221 than did chilling rate and chilling accumulation taken alone (Fig. 4). Since dormancy 222 break date did not change significantly (Fig. 1) and the forcing rate increased 223 substantially, the duration of the forcing stage decreased which explained most of the 224 advancement in LUD. Therefore, the variation in ST was, to date, also strongly 225 dependent on the duration of the forcing stage and the average temperature during this 226 stage. For example, in Fig. 3, when the LUD advances from d0 to d0+6, the forcing 227 temperatures on days between d0+6 and d0 become useless to leaf unfolding (but still 228 useful to late leaf growing). The daily average forcing rate (Frave) from df0 to d0+3 in the year with +3 °C warming (Y0+3) has to be larger than the Frave from df0 to d0 in the 230 reference year Y0, and the increment of Frave from Y0+3 to the year with +6 °C warming 231 (Y0+6) should be even larger than the increment from year Y0 to Y0+3, due to the 232 further decreased DF. This explains why the temperature sensitivity of LUD is 233 significantly related to the DF (Fig. 4).

234
The weakening effect of spring warming on ST can be explained by the non-linear 235 shape of the relationship between daily forcing rate (Fr) and daily mean temperature 236 (Eq. 2). At most sites, this relationships indeed followed a sigmoid pattern (e.g.

237
forcing rate presented in Fig. S2a), rather than a nearly linear change pattern as the 238 widely used forcing metric-growing degree-days (defined as the difference between 239 daily mean temperature and a base temperature, e.g. Fig. S10d). Therefore, the effects 240 of warming on accelerating forcing accumulation increases as temperature moves 241 towards T50, the temperature inducing 50% of optimal growth in forcing rate and the 242 inflection point of the function, but decreases as temperature moves away from T50. temporal changes in ST at many sites (Fig. 4). Therefore, part of the remaining 263 variation might be explained by an effect of photoperiod. In addition, we found that 264 the relationships between temperature sensitivity of LUD and the duration of the 265 forcing stage and the forcing rate are likely nonlinear at many sites (e.g. Fig. S15). (1) 343 where Top is the optimal chilling temperature, and c1 and c2 are two calibration 344 coefficients (Table S2) (2) 353 where T50 is the mid-response temperature, which induces 50% of optimal growth in 354 forcing rate and is the inflection point of the function. c3 is a calibration coefficient.  Table S2. Root 372 mean square error (RMSE, Eq. 5) between simulated (LUDsim_i) and observed (LUDobs_i) LUDs was used as the objective function, and parameter valuess that 374 minimized the RMSE were regarded as optimal.
where n is the number of LUD records (years) for each species at each site.

377
Using the optimized parameters, the Unified model was then applied to estimate the  To estimate the trend in the temporal change of the temperature sensitivity of LUD, 440 we also conducted a reduced major axis regression for each species at each site with a Competing interests 462 The authors declare no competing interests.