Does drought advance the onset of 1 autumn leaf senescence in temperate 2 deciduous forest trees? 3

20  Severe droughts are expected to become more frequent and persistent. However, their effect on 21 autumn leaf senescence, a key process for deciduous trees and ecosystem functioning, is currently 22 unclear. We hypothesized that (I) severe drought advances the onset of autumn leaf senescence 23 in temperate deciduous trees and that (II) tree species show different dynamics of autumn leaf 24 senescence under drought. 25  We tested these hypotheses using a manipulative experiment on beech saplings and three years 26 of monitoring mature beech, birch and oak trees in Belgium. The autumn leaf senescence was 27 derived from the seasonal pattern of the chlorophyll content index and the loss of canopy 28 greenness using generalized additive models and piece-wise linear regressions. 29  Drought did not affect the onset of autumn leaf senescence in both saplings and mature trees, 30 even if the saplings showed a high mortality and the mature trees a high leaf mortality (due to 31 accelerated leaf senescence and early leaf abscission). We did not observe major differences 32 among species. 33  Synthesis: The timing of autumn leaf senescence appears conservative across years and species, 34 and even independent on drought stress. Therefore, to study autumn senescence, seasonal 35 chlorophyll dynamics and loss of canopy greenness should be considered separately. 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 55 hypotheses by subjecting young trees to drought stress in an experimental set-up and by 111 examining the effect of years with different drought intensities (2017, 2018 and 2019) on mature trees in 112 natural forest stands.


20
 Severe droughts are expected to become more frequent and persistent. However, their effect on 21 autumn leaf senescence, a key process for deciduous trees and ecosystem functioning, is currently 22 unclear. We hypothesized that (I) severe drought advances the onset of autumn leaf senescence 23 in temperate deciduous trees and that (II) tree species show different dynamics of autumn leaf 24 senescence under drought. 25  We tested these hypotheses using a manipulative experiment on beech saplings and three years 26 of monitoring mature beech, birch and oak trees in Belgium. The autumn leaf senescence was 27 derived from the seasonal pattern of the chlorophyll content index and the loss of canopy 28 greenness using generalized additive models and piece-wise linear regressions. 29  Drought did not affect the onset of autumn leaf senescence in both saplings and mature trees, 30 even if the saplings showed a high mortality and the mature trees a high leaf mortality (due to 31 accelerated leaf senescence and early leaf abscission). We did not observe major differences 32 among species. 33  Synthesis: The timing of autumn leaf senescence appears conservative across years and species, 34 and even independent on drought stress. Therefore, to study autumn senescence, seasonal 35 chlorophyll dynamics and loss of canopy greenness should be considered separately .  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61 1. Introduction 65 Autumn leaf senescence is a developmental stage of the leaf cells. The core function of this process is the 66 remobilization of nutrients and death is its consequence (Medawar, 1957;Keskitalo et al., 2005). Its 67 evolutionary purpose is likely stress resistance and, as such, the process dynamics are affected by different 68 forms of environmental stress (e.g. high temperatures, water logging) (Benbella and Paulsen, 1998;Leul 69 and Zhou, 1998;Munné-Bosch and Alegre, 2004). The process of autumn leaf senescence is highly 70 coordinated and characterized by a tight control over its timing. Furthermore, its most manifest feature, 71 the detoxification of chlorophyll, allows the degradation of leaf macromolecules and subsequent nutrient 72 remobilization -the essence of autumn leaf senescence- (Hörtensteiner and Feller, 2002;Munné-Bosch 73 and Alegre, 2004; Matile, 2000). In addition, chlorophyll degradation allows for the typical leaf coloration 74 during autumn. However, autumn leaf senescence is also an important process at the ecosystem scale 75 because it affects multiple ecological processes, such as trophic dynamics, tree growth or the exchange 76 of matter and energy between the ecosystem and atmosphere (Richardson et al., 2013). Droughts are expected to occur more frequently and become more intensive due to global warming and 89 changes in precipitation patterns ( Erlangen, Germany) to monitor the relative humidity and air temperature (Fig. 1, panel A and B). One pot 125 per glasshouse was also equipped with a soil moisture smart sensor (HOBO S-SMD-M005, Onset, MA, 126 USA) to monitor the soil water content (Fig. 1,  slow-release fertilizer (DCM ECO-XTRA 1) and 1.8 g of micro elements (DCM MICRO-MIX). Using the 134 relative humidity and air temperature data between 7 a.m. and 7 p.m., the vapor pressure deficit was 135 calculated for both treatments (see below) and the reference plots using the formulas of Buck (1981) (Eq. where e0 is the saturation vapor pressure (in kPa), T is the temperature (in °C), e is the actual vapor 144 pressure deficit (in kPa), RH is the relative humidity (in %) and VPD is the vapor pressure deficit (in kPa). 145 146 From planting until April, the saplings were all irrigated two to three times a week until the pots 147 overflowed. At the start of the treatment, in early May, we shielded all the glasshouses using polyethylene 148 film (200 µm thick) and irrigated the saplings only once a week with circa 1.5 liter of water. In addition, 149 we enhanced the drought in six glasshouses by raising the air temperature by three degrees compared to 150 the ambient air temperature (+3 °C). The idea was to simulate 'natural' drought conditions, which are 151 typically associated with warmer temperatures. The air temperature in the other six glasshouses followed 152 the ambient air temperature (+0 °C). During the treatment, the daily soil water content and the daily 153 relative humidity in the glasshouses with the +3 °C treatment were lower in comparison to the glasshouses 154 with the +0 °C treatment. The difference was around 0.02 m³/m³ for the soil water content and 20% for 155 the relative humidity (Fig. 1). The plan was to continue the treatment till the end of June but, due to the 156 significant mortality rate, we were obliged to alleviate the drought stress already from the 20 th of June. 157 https://doi.org/10.5194/bg-2020-337 Preprint. To indicate the magnitude of the droughts, we computed the rainfall deficit from 2017 to 2019 using data 192 on the relative humidity, solar radiation, wind speed, temperature and precipitation from the 193 meteorological station in Ukkel. Here, the meteorological records go back the longest in Belgium. The 194 rainfall deficit is computed on a daily basis by accumulating the daily potential evapotranspiration minus 195 the daily amount of precipitation. This was done in two ways: (I) per hydrological year, starting from a 196 zero deficit at the start of the hydrological year (1 st of April) and (II) continuous computation, so no restart 197 from 0 at the start of each hydrological year. The latter method has the benefit that the long-term effect 198 of accumulated droughts from successive years is accounted for. 199 200 The potential evapotranspiration was computed by means of the method of Bultot et al. (1983), which is 201 similar to the method of Penman (1948), but has parameters that are calibrated specifically for the local 202 Belgian conditions. Unlike for the rainfall deficit starting from a zero deficit, we accounted in the 203 calculation of the continuously computed rainfall deficit for the hydrological fraction in wet periods that 204 https://doi.org/10.5194/bg-2020-337 Preprint. Discussion started: 14 October 2020 c Author(s) 2020. CC BY 4.0 License.
does not contribute to building up ground water reserves. At the station of Ukkel, daily precipitation and 205 potential evapotranspiration data are available since more than 100 years. The precipitation data are 206 collected since 1898 on the same location, and is measured using the same instrument. For this study, the 207 data for the 100-year period 1901-2000 was considered as the reference period for the computation of 208 long-term statistics on the rainfall deficit. For the 16 mature trees in the two forests and from the end of July to the end of November, tree-climbers 221 collected leaves on eight occasions per year separated by two to three weeks. During each measurement 222 day, they collected five sun-leaves and five shade-leaves from each tree. Afterwards, the CCI was 223 immediately measured on the harvested leaves using the same chlorophyll content meter as described 224 above diameter cylinder, we collected samples of leaf tissue from the leaves of the mature trees for which we 230 also measured the CCI. After storage at -80 °C, the samples were grounded using glass beads and a 231 centrifuge. The result was dissolved in ethanol and the absorption of the solution was measured using a 232 spectrophotometer (Smart Spec Plus Spectrophotometer, Bio-Rad) at different wavelengths for 233 Chlorophyll a (662 nm) and chlorophyll b (644 nm). The chlorophyll concentrations could then be derived 234 from the absorption values using the formulas described in Holm (1954) and Vonwettstein (1957). 235 236 2.3. Tree mortality in the manipulative experiment 237 In this study, we only considered those trees that defoliated due to autumn leaf senescence. Other tree 238 saplings have died or defoliated completely due to accelerated leaf senescence during or just after the 239 drought period. Since chlorophyll degradation is a common feature of both senescence processes and 240 nutrient remobilization was only measured indirectly by CCI, we did not consider (I) tree saplings that 241 showed an early or abrupt defoliation (without gradual coloration) before the 18 th of August (n = 20) and 242 (II) tree saplings with constant CCI values lower than three, the limit at which the values of the CCI meter 243 can be interpreted, for the whole period from August to November (n = 18 To model the CCI of both our tree saplings and mature trees as a function of their covariates, Gaussian 262 GAMMs with the identity link function were used ( information criterion) and all factor-smooth interaction terms were smoothed using P-splines to address 267 the large gap in data (i.e. from November to June) between the yearly sampling periods. 268 269 For the CCI of the beech saplings, the fixed covariates were the treatment (categorical with three levels), 270 leaf place (categorical with three levels) and day of the year (continuous; model 1). The interaction term 271 was modelled as a factor-smooth interaction between the covariates day of the year and treatment. interaction term was modelled as a factor-smooth interaction between the covariates day of the 296 year and year. The dependency among observations of the same individual tree was incorporated 297 using individual tree as random intercept. 298

Model 3 299
Yij ~ Gaussian(µij, cst.) variability. Trees that did not show a clear breakpoint (13 in the manipulative experiment) were not 331 considered in the analysis. These trees did not show a different pattern of CCI or loss of canopy greenness 332 than the other trees (Fig. S2). 333 334 2.4.3. Comparing the onset of autumn leaf senescence among tree saplings exposed to different 335 drought treatments 336 We tested whether the beech saplings exposed to the three treatments in 2018 differed in their onset of 337 autumn leaf senescenceCCI using a linear model with the onset of autumn leaf senescenceCCI as response 338 https://doi.org/10.5194/bg-2020-337 Preprint. Discussion started: 14 October 2020 c Author(s) 2020. CC BY 4.0 License. where g is the identity link function, µij is the conditional mean, Yij is the jth observation of the response 362 variable in Species i, and i = 1,…, 3 and Speciesi is the random intercept. 363 The effect of the species on the onset of autumn leaf senescenceCCI and the onset of the loss of canopy 364 greenness was assessed using two linear mixed effect models with the onset of autumn leaf senescenceCCI 365 and the onset of the loss of canopy greenness from the mature beech, birch and oak trees as response 366 variable. The fixed covariate in these two models was the Species (categorical with three levels; model 6). 367 To incorporate the dependency among observations of the same year, we used Year as random intercept. 368

Model 6 369
Yij ~ Gaussian(µij, cst.) where g is the identity link function, µij is the conditional mean, Yij is the jth observation of the response 374 variable in Year i, and i = 1,…, 3 and Yeari is the random intercept. 375 The residuals of the models were approximately normally distributed and showed no heteroscedasticity 376 (tested using diagnostic plots   Table 2). In the +0 and 406 especially the +3 °C treatment, an abnormal CCI decline was observed in early August with only a partial 407 recovery later on. As a result, from the beginning of August until mid-September, the CCI values of the 408 beech saplings in the reference plots were significantly higher than the CCI values of the beech saplings in 409 the glasshouses. However, no significant difference was detected in the timing of the onset of autumn 410 leaf senescenceCCI among the beech saplings exposed to the three different treatments, as the mean onset 411 of autumn leaf senescenceCCI was between the 21 st (DOY = 260 ± 5) and 25 th (DOY = 264 ± 4) of September 412 (P = 0.7; Fig. S3). 413 414

Results
The loss of canopy greenness for the beech saplings showed a stable decline from early August until the 415 end of autumn ( Fig. 4; panel B & D; Table 2). Nevertheless, during September, the loss of canopy greenness 416 of the beech saplings in the reference plots was significantly higher than the loss of canopy greenness of 417 the beech saplings in the glasshouses with the +3 °C treatment. 418 419 The tree saplings in the glasshouses of both treatments were exposed to a high mortality with 14% and 420 26% of the tree saplings in the glasshouses with the +0 °C and +3 °C treatment, respectively, considered 421 'dead' along our criteria (see §2.3.). In the reference plots, no beech saplings died. The pattern in the CCI values for the mature beech, birch and oak trees seems consistent throughout the 426 years with stable values in summer and a rapid decline around late October (Fig. 5 -7; panel A & C; Table  427 2). We also observed no significant difference in the onset of autumn leaf senescenceCCI among the years 428 (P = 0.09) and species (P = 1). The mean onset of autumn leaf senescenceCCI among the years was from 429 the 8 th (DOY = 281 ± 6) to the 19th (DOY = 292 ± 6) of October ( Fig. S4; panel A), while the mean onset of 430 autumn leaf senescenceCCI among the species was around the 13 th of October (DOY = 286 ± 6; Fig. S4;  431 panel B). The CCI correlated linearly with the chlorophyll concentrations but the data showed more 432 variation in 2018 than 2017 (see Fig. S1). 433 434 The pattern in the loss of canopy greenness for the mature beech, birch and oak trees seemed less 435 consistent throughout the years ( However, all tree species differed significantly in their onset of the loss of canopy greenness across years 447 (P = 6 x 10 -9 ). Compared to birch (DOY = 268 ± 9; Fig. S5; panel B), the onset of the loss of canopy greenness 448 for beech was on average 16 days later (P = 1 x 10 -4 ; DOY = 284 ± 4), while for oak this was 30 days later 449 (P = 1 x 10 -4 ; DOY = 298 ± 4). The onset of the loss of canopy greenness for beech was also 14 days earlier 450 than that for oak (P = 7 x 10 -4 ). 451

452
Our results showed that the timing of the onset of autumn leaf senescence in both tree saplings and 453 mature trees was not significantly altered by severe drought stress induced by a decline in the soil 454 moisture, relative humidity, and an increase in the air temperature and vapor pressure deficit. Although the onset of autumn leaf senescence in both the tree saplings and the mature trees was not 498 advanced by drought stress, the onset of autumn leaf senescence in beech saplings was around 22 days 499 earlier than mature beech trees. Such difference could be due to the different growing conditions (pots 500 versus normal soil), environmental conditions at the different sites, the difference in the average leaf age 501 (tree saplings have an earlier bud-burst than mature trees) or the different ecophysiological response of 502 tree saplings and mature trees (e.g. tree saplings are more vulnerable than mature trees and therefore 503 are likely to use different functional strategies) ( (the species with the earliest onset of the loss of canopy greenness) has an indeterministic growth pattern, 520 which also means continuous leaf mortality. Second, the fact that oak (the species with the latest onset 521 of the loss of canopy greenness) has typically a second leaf flush, which might connect the difference 522 between beech and oak to differences in leaf longevity. 523

524
The different environmental conditions of three years (comprising a severe dry year and a severe warm 525 year) did not affect the timing of the onset of autumn leaf senescence in mature beech, birch and oak 526 forest trees in Belgium. This suggests that deciduous trees have a conservative strategy concerning the 527 timing of their senescence. Like our mature beech trees, beech saplings exposed to a drought also did not 528 show any advancement in their onset of autumn leaf senescence compared to beech saplings in normal 529 conditions. Although the drought stress did not affect the timing of the onset of autumn leaf senescence, 530 it is clear from our results that the drought stress did affect the mortality rate in tree saplings and the leaf 531 mortality in mature trees. 532