This is the accepted version of the article:
Vergotti, Marina; Fernández-Martínez, Marcos; Kefauverce, S.C.; [et al.].
«Weather and trade-offs between growth and reproduction regulate fruit
production in European forests». Agricultural and forest meteorology, Vol. 279
(Dec. 2019), art. 107711. DOI 10.1016/j.agrformet.2019.107711
This version is avaible at https://ddd.uab.cat/record/218213
under the terms of the license
1 Weather and trade-offs between growth and reproduction
2 regulate fruit production in European forests
3
4 Authors: M.J. Vergotti*1,2,3, M. Fernández-Martínez4, S.C. Kefauver3,5, I.A. Janssens4, J.
5 Peñuelas1,2.
6 Affiliations:
1
7 CSIC, Global Ecology Unit, CREAF-CSIC-UAB, Bellaterra, 080193 Barcelona,
8 Catalonia, Spain
2
9 CREAF, Cerdanyola del Vallès, 08193 Barcelona, Catalonia, Spain
3
10 Integrative Crop Ecophysiology Group, Plant Physiology Section, Faculty of Biology,
11 University of Barcelona, 08028 Barcelona, Spain.
4
12 Centre of Excellence PLECO (Plant and Ecosystem), Department of Biology, University
13 of Antwerp, 2610 Wilrijk, Belgium.
5
14 AGROTECNIO (Center for Research in Agrotechnology), Av. Rovira Roure 191, 25198,
15 Lleida, Spain.
16
*
17 Correspondence to: Marina Vergotti, email: mj.vergotti@gmail.com
1
18 Abstract
19
20 Some tree species have a highly variable year-to-year pattern of reproduction which has
21 repercussions for the entire ecosystem. Links between meteorological variability, fruit
22 production and crown cover, and trade-offs between reproduction and vegetative growth,
23 remain elusive, despite a long history of research. We explored how meteorological
24 conditions determined variations in fruit production and crown cover and how remotely
25 sensed vegetation indices, such as the enhanced vegetation index (EVI), may be used
26 to characterize the fluctuations in fruit production. We used data for fruit production from
27 six European tree species (Abies alba, Picea abies, Pseudotsuga menziesii, Fagus
28 sylvatica, Quercus petraea and Q. robur) growing in monospecific stands, EVI and
29 seasonal meteorological variables (precipitation and temperature) for 2002-2010.
30 Weather accounted for fruit production better than EVI. Deciduous trees were more
31 responsive to weather than evergreens, most notably to different seasonal temperatures,
32 which was positively correlated mainly with crown cover and fruit production in deciduous
33 species. Our results also suggested different patterns of relationships between fruit
34 production, crown cover and weather, indicating different strategies of resource
35 management. These patterns indicated a possible internal trade-off in evergreens, with
36 resources allocated to either growth or reproduction. In contrast, in deciduous species
37 we found no evidence for such a trade-off between vegetative growth and reproduction.
38
39
40 Keywords: Variability, fruit production, EVI, crown cover, weather, trade-off.
1
41 1. Introduction
42
43 Seed and fruit production are key functions for both plants and ecosystems. They are
44 essential parts of the biological cycle of individual plants, because they allow them to
45 reproduce and thus transmit their genes to the next generation and disperse the species.
46 However, fruit production is variable from year to year, and some species might even
47 show an extremely erratic pattern of low and massive fruit production among years
48 (Espelta et al. 2008; Kelly et al. 2013). Fruiting variability will thus have an impact on the
49 entire ecosystem by affecting various processes. For example, a variable supply of
50 seeds and fruits will entail a bottom-up cascade of trophic responses by causing
51 fluctuations in the population densities of fruit and seed consumers. By extension, this
52 will affect the population densities of their predators and competitors (Ostfeld and
53 Keesing, 2000; Espelta et al., 2008, 2017; Bogdziewicz et al., 2016).
54
55 Determining the factors that drive fruit production has been the focus of many studies in
56 recent decades (Kelly 1994; Koenig and Knops 2000; Kelly and Sork 2002; Pérez-
57 Ramos et al. 2015; Fernández-Martínez et al. 2017b), and various hypotheses have
58 been proposed to account for fluctuations from year to year. The most commonly
59 accepted theories accounting for the proximal causes of fruit production involve tree
60 resources: the hypothesis of resource matching [also known as the hypothesis of
61 weather tracking (Kelly 1994; Kelly and Sork 2002; Pearse et al. 2016)], the hypothesis
62 of resource accumulation (Isagi et al. 1997; Satake and Iwasa 2000; Camarero et al.
63 2010; Fernández-Martínez et al. 2015; Pearse et al. 2016) and the hypothesis of
64 resource switching (Norton and Kelly 1988; Kelly and Sork 2002; Sánchez Humanes and
65 Espelta Morral 2011). These hypotheses stipulate that fruit production is driven by the
66 levels of resources in trees (e.g. carbohydrates and nutrients), which in turn are regulated
67 to some extent by meteorological conditions such as temperature or rainfall, even though
68 these conditions can also act as synchronizing cues (Kelly and Sork 2002; Sala et al.
69 2012; Ichie et al. 2013; Bogdziewicz et al. 2017; Fernández-Martínez et al. 2017a). The
70 resource-matching hypothesis proposes that fruit production would indirectly respond to
71 meteorological conditions by the availability of resources, with reproduction likely
72 positively correlated with growth. Years of good weather would thus favor
73 photosynthesis, so more resources would be available for both vegetative and secondary
74 growth and for reproduction (Kelly, 1994; Kelly and Sork, 2002; Fernández-Martínez et
75 al., 2015; Pearse et al., 2016). The resource-accumulation hypothesis, though, refers to
76 the accumulation of resources up to a threshold. Massive fruit production is then
77 triggered and thus the exhaustion of the accumulated resources when the threshold is
78 reached, usually decreasing vegetative and secondary growth as a trade-off (Pearl and
79 Verma, 1988; Kelly, 1994; Fernández-Martínez et al., 2015; Bogdziewicz et al., 2016).
80 This hypothesis implies that resource reserves would be severely depleted after an
81 episode of high fruit production, which could also entail a negative autocorrelation in
82 annual fruit production (Kelly, 1994; Fernández-Martínez et al., 2012, 2015). Finally,
83 resource switching hypothesis states that there is a trade-off between reproduction and
84 vegetative growth from year to year (Sánchez Humanes and Espelta Morral, 2011).
85
86 These hypotheses suggest that the resource levels should vary among trees. Previous
87 studies have inferred these changes by monitoring the dynamics of the crown cover
88 (Camarero et al. 2010; Fernández-Martínez et al. 2015). These studies tested this
89 inference using remotely sensed vegetation indices such as the normalized difference
90 vegetation index [NDVI, (Camarero et al. 2010)] and the enhanced vegetation index [EVI,
91 (Fernández-Martínez et al. 2015)]. These indices are useful in comparative studies of
92 spatial and temporal variations in the crown cover and, by extension, of variations in the
93 photosynthetic capacity of forests (Huete et al., 2002; Garbulsky et al. 2013; Fernández-
94 Martínez et al. 2019).
95
96 The aim of our study was thus to determine the influence of meteorological conditions
97 on fruit production and crown cover in monospecific forests of Abies alba, Picea abies,
98 Pseudotsuga menziesii, Fagus sylvatica, Quercus petraea and Q. robur distributed
99 across Europe. We also evaluated the ability of remotely sensed vegetation indices to
100 account for the temporal and spatial variability of fruit production in these forests. We
101 hypothesized that variations in crown cover, estimated using EVI, could reliably monitor
102 fruit production in forests. We also hypothesized that evergreen and deciduous trees
103 would have different strategies of resource management, indicated by their relationships
104 between fruit production, crown cover and weather.
105
106 We achieved these goals by first identifying the differences among species in the
107 average and interannual variability of fruit production and EVI. We then determined the
108 spatial relationship between the variations in fruit production and EVI and the
109 relationships between annual variations in fruit production, EVI and meteorological
110 variables. Finally, we determined the direct and indirect effects of meteorological and
111 EVI variables on fruit production.
112
113
114 2. Materials and methods
115
116 2.1 Data collection
117
118 2.1.1 Data for fruit production
119
120 We used data for fruit production from 76 European forests distributed over France,
121 Luxemburg and Germany (Figure 1). All sites were monospecific stands of deciduous
122 species: F. sylvatica (21 sites), Q. petraea (20 sites) or Q. robur (9 sites), and evergreen
123 species: A. alba (10 sites), P. menziesii (6 sites) or P. abies (10 sites). All species are
124 producers of dry fruits.
125
126 The data for fruit production were obtained from the ICP Forests database (International
127 Co-operative Program on Assessment and Monitoring of Air Pollution Effects on Forest,
128 operated under the UNECE Convention on Long-range Transboundary Air Pollution,
129 ). The data were gathered using the methods for litterfall
130 sampling described by ICP Forests (Ukonmaanaho et al., 2016). Litterfall was collected
131 using nets or litterbags distributed uniformly over the sampling areas to ensure that entire
132 plots were represented in the samples, not just the dominant trees; more collectors may
133 have been used for deciduous species, because their leaves are more affected by air
134 turbulence. The litter was then separated into fractions (leaves, fruits and branches). The
135 data contained records of fruit net primary production (fNPP) from 2002 to 2013 in units
136 of 𝑔𝐷𝑊 × 𝑚−2 × 𝑦 −1 that were then converted to 𝑔𝐶 × 𝑚−2 × 𝑦 −1 using the data for C
137 concentration provided by the same database and then used to calculate the average
138 fNPP per year for each site. The database contained many more sites, but only
139 monospecific forests with fNPP records for at least five years (the longest available
140 record was seven years) were selected to avoid artifacts in the results.
141
142
143 2.1.2 Crown-cover estimates (satellite data) and meteorological data
144
145 We preferentially used EVI rather than NDVI for estimating crown cover, because EVI is
146 more sensitive to structural variations of the crown and has provided good results in
147 previous studies (Fernández-Martínez et al., 2015). EVI time series for each forest were
148 obtained from the MOD13Q1 MODIS (Moderate Resolution Imaging Spectroradiometer,
149 ) product for 18 February 2000 to 16 November 2016,
150 with a spatial resolution of 250 × 250 m and a temporal resolution of 16 days. We only
151 used the central pixels of the forest stand in our analysis. Because of this, we were not
152 able to select single crowns.
153
154 Time series for precipitation and temperature were extracted from the meteorological
155 database of the MARS unit AGRI4CAST/JRC (), with
156 a spatial resolution of 0.25 × 0.25°, for 2002-2010, which provided monthly mean
157 temperatures and total precipitation.
158
159 We used monthly values for both the EVI and meteorological data to estimate average
160 values of EVI, temperature and precipitation per season for each year: October-
161 December for autumn of the year previous to fruit ripening, January-March for winter,
162 April-June for spring and July-September for summer. We then calculated the seasonal
163 means for each site using the yearly averages previously calculated. Finally, we
164 estimated the anomalies associated with EVI, temperature and precipitation for each
165 year by subtracting the seasonal mean per site to the yearly seasonal value. We only
166 used the values between 2002 and 2008 for the EVI data to maintain consistency
167 between the EVI and meteorological data. We decided to use only weather conditions
168 for the year of, or previous to fruit ripening based on previous studies suggesting that the
169 selected time frame is very important for explaining variations in fruit production (Kelly et
170 al. 2013; Fernández-Martínez et al. 2017a).
171
172
173 2.2 Statistical analyses
174
175 We first determined the differences in annual means and interannual variability in fNPP
176 and EVI among species by calculating the average EVI per year and its associated
177 standard error (SE) using the seasonal values of EVI in our data. We then calculated the
178 average EVI per site using the average annual EVI. The SE for EVI per site was
179 estimated by propagating the SE for average annual EVI, following the error-propagation
180 method:
181 𝛿𝐶 = √(𝛿𝐴)2 + (𝛿𝐵)2
182 Where δ is the standard error for variables A, B, and C. The same procedure was used
183 to calculate the average EVI per species, using the average EVI per site and propagating
184 its SE across sites. The fNPP data contained annual values, which were used to
185 calculate the average fNPP per site and its associated SE. We then calculated the
186 average fNPP per species using the average fNPP per site and propagating its SE.
187
188 The interannual variabilities of fNPP and EVI were assessed by calculating the
189 proportional variability index [PV, see section 1 in Supplementary Materials (method S1)
190 (Heath, 2006)] and the disparity index [D, see section 2 in Supplementary Materials
191 (method S2) (Martín-Vide 1986; Fernández-Martínez et al. 2017b, 2018)]. Both PV and
192 D provide a more robust assessment than the coefficient of variation (𝐶𝑉 = 𝑆𝐷 ×
193 𝑚𝑒𝑎𝑛−1) of the temporal variability, because they are much less sensitive to changes in
194 the mean of the time series, among other properties (Martín-Vide, 1986; Heath, 2006;
195 Fernández-Martínez et al., 2017a). PV and D are similar in concept, but D differs from
196 PV in that D takes into account the chronological order of the values of the time series,
197 but PV is insensitive to the order (Fernández-Martínez et al., 2017b). Estimating site and
198 species-specific variability allowed us to differentiate the behavior of fNPP and EVI
199 between sites and species. These indices for EVI were first estimated for the annual
200 average EVI using the seasonal EVI values. Annual values of these indices were then
201 averaged per site, and the associated SEs were calculated. Finally, the values of the
202 indices per species were estimated by calculating the mean of the average indices per
203 site and propagating its SE. PV and D for fNPP were first estimated for the average fNPP
204 per site using the annual fNPP values. The values of the indices per site were then
205 averaged per species, and their associated SEs were calculated.
206
207 We tested for differences among species by first performing analyses of variance
208 (ANOVAs) using generalized linear mixed models (GLMMs) from the “nlme” R package
209 (Pinheiro et al., 2018). Average fNPP and EVI (separately) per site were the response
210 variables, species was the explanatory variable and sampling site was the random factor.
211 We fitted linear models for PV and D in which the average indices per site were the
212 response variables and species was the explanatory variable. We then performed
213 Tukey’s honestly significant difference (HSD) tests at P < 0.05 for multiple comparisons
214 on each model. These analyses allowed us to determine whether species could be
215 grouped by reproductive behavior.
216
217 We also identified the spatial relationships between the variations in fNPP and EVI to
218 determine whether greener sites (i.e. those with higher EVIs) also had higher average
219 fruit production. We used average values per site and fitted a GLMM, with average fNPP
220 per site as the response variable, average EVI per site as the explanatory variable and
221 species as the random factor, using the “nlme” R package (Pinheiro et al. 2018). We also
222 similarly tested each species individually by fitting a linear model and correlating average
223 fNPP with the average EVI for each species separately.
224
225 We estimated the relationships between the annual variations of fNPP, EVI and
226 meteorological variables by fitting one GLMM per species, where fNPP was the response
227 variable, and anomalies of yearly seasonal averages of EVI, temperature and
228 precipitation (see subsection 2.1.2) were the explanatory variables (e.g.
229 𝐹𝑟𝑢𝑖𝑡 𝑁𝑃𝑃 𝑦𝑒𝑎𝑟 𝑖 ~ 𝐸𝑉𝐼 𝑎𝑢𝑡𝑢𝑚𝑛 𝑦𝑒𝑎𝑟 𝑖 − 1 + 𝐸𝑉𝐼 𝑤𝑖𝑛𝑡𝑒𝑟 𝑦𝑒𝑎𝑟 𝑖 + ⋯). Site was the
230 random factor. We applied a natural logarithm to the response variable in this model to
231 maintain a normal distribution of the residuals. Variables were selected using the
232 stepwise backward-forward method, starting from the full model (containing all possible
233 variables and interactions) and using the Bayesian information criterion (BIC). The
234 stepwise model selection used the “stepAIC” function of the “MASS” R package
235 (Venerables and Ripley, 2002). ΔBIC was then calculated as the difference between the
236 BICs of the full model and the model without the target variable as a measure of variable
237 importance. The collinearity of the variables was evaluated using the variance inflation
238 factor (VIF) and the “vif” function in the “car” R package (Fox and Weisberg, 2011): VIF
239 was approximately 1 for each variable, indicating no problems of multicollinearity. The
240 model results were visualized using partial-residual plots in the “visreg” R package
241 (Breheny and Burchett, 2017). The variances explained by the fixed factors (marginal
242 coefficient of determination, R2m) and both fixed and random factors (conditional
243 coefficient of determination, R2c) were calculated using the “rsquaredGLMM” function of
244 the “MuMIn” R package (Barton, 2018).
245
246 Finally, we determined the direct and indirect effects of the seasonal meteorological
247 variables and EVI on fNPP for each species by path analyses, using the method of
248 directed separation (d-separation). D-separation is a type of path analysis that uses
249 directed acyclic graphs (DAGs) and that specifies sets of variables (represented by
250 nodes in a DAG), for which independence is conditioned by a third variable (Pearl and
251 Verma, 1988; Voyer and Garamszegi, 2014). That is, two variables, A and B, may be
252 correlated using a third variable, C; A and B are d-separated by C if the knowledge of A,
253 having established C, does not provide more information of B. The independence of A
254 and B is thus conditioned by the nature of C. This method allowed us to fit path analyses
255 while taking into account site as the random factor. We then defined the paths. Seasonal
256 meteorological variables were defined as exogenous variables, i.e. they could potentially
257 have a direct effect on both fNPP and EVI and an indirect effect on fNPP through EVI.
258 fNPP and seasonal EVI variables were defined as endogenous variables, i.e. seasonal
259 EVI variables could potentially be directly affected by meteorological variables and
260 directly affect fNPP. The total effect of a variable on fNPP was calculated by adding its
261 direct effect on fNPP to its indirect effect on fNPP multiplied by the direct effect of the
262 intermediary variable on fNPP. That is, variable A directly affects variable B and C, and
263 variable B directly affects variable C; the total effect of A on C is due to the direct effect
264 of A on C plus the effect of A on B multiplied by the effect of B on C (i.e. the indirect effect
265 of A on C). These analyses used R script, which is available upon request.
266
267
268 3. Results
269
270 3.1 Differences in average fruit production and EVI and their variability between
271 species
272
273 Our analyses identified significant differences in fruit production between species (P <
274 0.001) when testing for differences between species. A. alba and P. abies had the
275 highest fNPPs (Figure 2). EVI was lower for A. alba and P. abies than the other species
276 (Figure 2, Table S1). PV and D had different patterns, with generally higher values for
277 the deciduous species, F. sylvatica, Q. petraea and Q. robur (Figure 2), i.e. the
278 deciduous species varied more year to year than the evergreen species, A. alba, P. abies
279 and P. menziesii. fNPP generally had higher variability indices than EVI, particularly for
280 D (Figure 2, Table S1). Both PV and D had more similar values for EVI than fNPP (Figure
281 2, Table S1).
282
283 When testing for spatial relationships between variations in fNPP and EVI, average fNPP
284 and EVI did not differ significantly among the sites, both when using all sites and using
285 each species individually (P > 0.05, for both linear models and GLMMs). Sites with more
286 variable EVIs were not correlated with sites with more variable fNPP (P > 0.05). The
287 spatial variability of EVI was therefore not linked to the spatial variability of fNPP in our
288 database. Therefore, it is not conclusive whether a lower or higher fruit production is
289 related to the site being greener.
290
291
292 3.2 Relationships among weather, EVI and fruit production
293
294 The path analyses identified a positive effect of winter temperature on spring EVI for all
295 species (Figure 3). Winter temperature also had a positive effect on winter EVI for all
296 three deciduous species (F. sylvatica, Q. petraea and Q. robur). Spring temperature also
297 had a positive effect on spring EVI for F. sylvatica and Q. petraea (Figure 3d and e) and
298 for the evergreen species P. abies. Spring temperature, however, negatively affected
299 spring EVI for P. menziesii (Figure 3b and c). Spring EVI for F. sylvatica was favored by
300 precipitation in winter (Figure 3d). Autumn EVI was affected negatively by autumn
301 precipitation for Q. petraea and positively by autumn temperature for Q. robur (Figure 3e
302 and f). Spring precipitation had a positive effect, and summer temperature had a negative
303 effect, on summer EVI for Q. robur (Figure 3f).
304
305 Winter temperature was the most recurrent meteorological variable associated with
306 fNPP. Winter temperature was associated negatively with fNPP for A. abla and P.
307 menziesii and positively for Q. petraea (Figure 3a, c, and e). fNPP for both oaks, Q.
308 petraea and Q. robur, responded similarly to the meteorological variables. Autumn
309 temperature had a positive effect on fNPP for both species (Figure 3e and f). fNPP was
310 also positively correlated with high summer temperature in both species. This result
311 differs from those for P. menziesii where fNPP was negatively correlated with high
312 summer temperature (Figure 4). fNPP for Q. robur was positively correlated with high
313 precipitation in summer and high temperatures in autumn (Table 1, Figure 4). F. sylvatica
314 and P. abies had similar associations: fNPP for both species was correlated negatively
315 with precipitation in spring and positively in winter (Figure 4). The GLMM analyses
316 identified some similarities among the evergreen species: fNPP for both A. alba and P.
317 abies was negatively correlated with winter temperature (Table 1, Figure 4).
318
319 fNPP was correlated with EVI for A. alba, F. sylvatica and Q. petraea (Table 1, Figure
320 5). The response for F. sylvatica was more similar to the responses for the evergreen
321 species than to the responses for the other deciduous species. fNPP was negatively
322 correlated with summer EVI for both A. alba and F. sylvatica. fNPP was correlated
323 positively with autumn EVI and negatively with spring EVI for F. sylvatica (Figure 5). This
324 outcome was similar to the outcome of the path analyses for P. menziesii: fNPP was
325 correlated positively with autumn EVI and negatively with spring EVI (Figure 3). The
326 response differed for Q. petraea: fNPP was positively correlated with spring EVI (Figure
327 5). fNPP was not significantly correlated with the EVI variables for P. abies, P. menziesii
328 or Q. robur (Table 1). Seasonal EVI (spring) directly affected fNPP in the path analyses
329 only for P. menziesii and Q. petraea (Figure 3c and e): positively for Q. petraea and
330 negatively for P. menziesii.
331
332 P. menziesii and Q. petraea were the only species with indirect effects of the
333 meteorological variables on fNPP through EVI in addition to the direct effects of the
334 meteorological variables. Spring EVI for both species was correlated with winter and
335 spring temperatures that in turn were correlated with fNPP. Winter temperature
336 negatively affected fNPP, with both negative direct and indirect effects through spring
337 EVI. Spring temperature did not have a direct effect on fNPP but was negatively
338 correlated with it through spring EVI, leading to a small positive total effect (Figure 3c,
339 Table 2). Winter and spring temperatures had a positive direct and indirect effect on
340 fNPP through spring EVI, leading to a positive total effect on fNPP (Figure 3e, Table 2),
341 in contrast to P. menziesii.
342
343
344 4. Discussion
345
346 4.1 Weather as a driver of crown cover and fruit production
347
348 Precipitation was positively correlated with crown cover of the following season for F.
349 sylvatica and Q. robur (winter precipitation for spring EVI and spring precipitation for
350 summer EVI, Figure 3d and f, respectively). Fruit production responded positively to
351 precipitation in the autumn previous to fruit ripening for Q. petraea and to precipitation in
352 winter for P. abies and F. sylvatica (Figure 4). In contrast, fruit production was negatively
353 correlated with spring precipitation for P. abies, F. sylvatica and A. alba (Figures 3 and
354 4). Precipitation generally has a positive effect on crown cover and fruit production,
355 because it can increase the photosynthetic capacity of trees and nutrient mineralization
356 in dry soils. Especially wet autumns and winters may increase fruit production by the
357 replenishment and accumulation of mineral resources. More resources would then be
358 available for fruit production in the next productive season [especially N (Sardans and
359 Peñuelas 2007; Smaill et al. 2011; Bogdziewicz et al. 2017)].
360
361 Crown cover in the summer was negatively correlated with summer temperature for Q.
362 robur. Fruit production for this species was also positively correlated with summer
363 precipitation (Figure 4). When combined with precipitation shortages, warm
364 temperatures, particularly soon after the end of dormancy and the start of the growing
365 season, will accelerate soil drying and thereby negatively affect photosynthetic activity
366 and slow vegetative growth, while respiration rates may increase (Martín-Benito et al.
367 2008; Adams et al. 2009; Carevic et al. 2010; Olivar et al. 2011; Estiarte and Peñuelas
368 2015). These effects would decrease the internal resources available for fruit production
369 and could also cause premature fruit abscission (Carevic et al. 2010; Fernández-
370 Martínez et al. 2012, 2015; Bogdziewicz et al. 2017).
371
372 Autumn temperature was positively correlated with crown cover in autumn for Q. robur
373 and with fruit production for both oak species (Figures 3 and 4). Temperature and
374 photoperiod are principal regulators of foliar senescence in autumn. A warm autumn
375 before fruit ripening would delay foliar senescence and abscission and provide a longer
376 period of photosynthetic activity, which would provide more time for trees to accumulate
377 resources and to reabsorb nutrients from leaves before they fall (Staaf and Stjernquist
378 1986; Doi and Takahashi 2008; Estiarte and Peñuelas 2015), increasing the availability
379 of nutrients for fruit production in the next growing season. These responses are
380 consistent with the resource-matching hypothesis (Kelly and Sork 2002; Pearse et al.
381 2016).
382
383 Winter temperature was positively correlated with winter crown cover for all deciduous
384 species, even though they do not have leaves during this season. The greenness
385 detected in winter with EVI may be due to the presence of evergreen shrubs and
386 perennial plants in the understory and not to the foliage of the deciduous trees. The
387 positive effect of winter temperature on winter crown cover may thus be an artifact, with
388 the trees exposed to longer and warmer autumns and springs. Winter temperature had
389 a positive effect on crown cover in spring for all species (Figure 3). Warm winters and
390 springs positively affect crown cover by advancing the period of foliar flushing, for
391 deciduous species, and generally extending the growing period (Estiarte and Peñuelas
392 2015). Warmer winters and springs could also allow an early onset of pollination,
393 extending the pollination period and thus potentially positively affecting fruit production
394 (Frenguelli and Bricchi 1998; Fernández-Martínez et al. 2012). Evergreen trees will lower
395 their metabolic rate to a minimum during winter, maintaining a minimal rate of primary
396 production. They are thus able to accumulate and store resources during unfavorable
397 winter conditions (Havranek and Tranquillini 1995; Falge et al. 2002; Pérez-Ramos et al.
398 2015). Shorter and warmer winters will shorten winter dormancy, with an increase in
399 metabolic activity and an earlier renewal of growth that will be indicated by spring crown
400 cover (Havranek and Tranquillini 1995), which would also imply higher respiration costs.
401 Fewer resources will therefore remain available for reproduction at the end of winter,
402 which would negatively affect reproduction rates (Havranek and Tranquillini 1995;
403 Fernández-Martínez et al. 2017a), as indicated by the results for P. menziesii and A.
404 alba, where winter temperature was negatively correlated with fruit production (Figures
405 3 and 4).
406
407
408 4.2 EVI and fruit production
409
410 Average fruit production was not significantly spatially correlated with EVI for any
411 species, perhaps due to the low spatial resolution of our crown-cover data. Other factors,
412 such as the heterogeneous meteorological conditions among the sampling sites, the
413 structure and age of the forests or nutrient availability, which are important determinants
414 of the spatial variability of fruit production, however, may have obscured a possible
415 connection (Fernández-Martínez et al. 2017b).
416
417 Autumn temperature had a positive effect on fruit production of the next growing season
418 for both oaks, Q. petraea and Q. robur (Figure 3). Fruit production was also positively
419 correlated with autumn crown cover for F. sylvatica (Table 2). As stated above,
420 temperature is a main regulator of foliar senescence in autumn, so warmer
421 meteorological conditions are often found to extend the growing period (Estiarte and
422 Peñuelas 2015). A greener crown cover in autumn would allow trees to store more
423 resources that could be invested in fruiting in the following growing season. The weather
424 had both direct and indirect effects on fruit production for Q. petraea through EVI. Both
425 winter and spring temperatures had a positive effect on spring EVI, which in turn was
426 positively correlated with fNPP (Figure 3), leading to a positive total effect from both
427 temperatures on fruit production (Table 1). Fruit production was also positively correlated
428 with spring EVI (Figure 5). Higher winter and spring temperatures could trigger an earlier
429 phenology, which in turn would increase photosynthesis and thus favor reproduction.
430 This scenario is consistent with the resource-matching hypothesis, where favorable
431 meteorological conditions would increase the accumulation of resources and their
432 subsequent allocation to both growth and reproduction (Kelly and Sork 2002; Fernández-
433 Martínez et al. 2012; Pearse et al. 2016).
434
435 P. menziesii, however, behaved differently. Winter temperature had an overall negative
436 effect on fruit production (Table 1) and a positive effect on spring crown cover, which
437 both negatively correlated with fruit production (Figure 3). This finding was described by
438 Kelly (1994) as “resource switching”, common in genera such as Abies, Picea,
439 Pseudotsuga and Fagus (Kelly and Sork 2002). Switching refers to the mechanism of
440 reallocating resources from vegetative growth or storage to reproduction (Kelly 1994;
441 Pearse et al. 2016), i.e. the internal trade-off between vegetative growth or storage and
442 reproduction (Sánchez Humanes and Espelta Morral 2011). Our results for P. menziesii
443 were similar to those for A. alba and F. sylvatica: winter and spring temperatures had a
444 direct negative effect on fNPP and a positive effect on spring EVI for A. alba and F.
445 sylvatica, respectively, with no obvious relationship between fNPP and spring EVI. fNPP
446 was also negatively correlated with summer EVI for both species and with spring EVI for
447 F. sylvatica (Table 2), suggesting that trees would prioritize growth and storage over
448 reproduction when meteorological conditions were favorable for photosynthesis. When
449 meteorological conditions deteriorated, the allocation of resources would shift to favor
450 reproduction (Kelly and Sork 2002; Pearse et al. 2016). These results are consistent with
451 those obtained for P. menziesii and with the resource-switching model. More conclusive
452 results, however, would be needed for confirmation. Greener P. abies crown covers in
453 spring were correlated with warm winters and springs, and fruit production was larger
454 after warm summers, i.e. meteorological conditions that may favor photosynthesis in
455 these species (Figure 3). The reproductive behavior of P. abies is consistent with the
456 resource-matching hypothesis, in contrast to the other evergreen species.
457
458 The relationships between the meteorological variables, crown cover and fruit production
459 could not be clearly distinguished between the evergreen and deciduous species. Some
460 patterns among some of the evergreens and F. sylvatica, however, recurred,
461 distinguishing them from the oaks. This similarity of F. sylvatica with the evergreens
462 could be due to the difference in distribution of the species. F. sylvatica and the
463 evergreens inhabit more humid and cooler zones than the oaks. Q. petraea, Q. robur
464 and P. abies had similar clear patterns of a resource-matching strategy. In contrast, A.
465 alba and P. menziesii favored more conservative strategies of resource management,
466 i.e. the resource-switching model. This strategy may include increased nutrient-use
467 efficiency, reducing nutrient loss by favoring the development of long-lasting tissues over
468 reproduction and only occasionally inversing the trade-off (Aerts 1995; Sánchez
469 Humanes and Espelta Morral 2011).
470
471
472 5. Conclusions
473
474 Crown cover was clearly influenced by meteorological conditions (especially temperature
475 and particularly winter temperature). Its responses, however, depended on the species,
476 particularly whether they were evergreen or deciduous. Fruit production was also
477 sensitive to environmental conditions, mainly winter and spring temperatures. Crown
478 cover and fruit production were correlated for A. alba, F. sylvatica and Q. petraea,
479 although crown cover directly affected fruit production only for P. menziesii and Q.
480 petraea.
481
482 The main differences observed between the species of this study were in the
483 relationships between fruit production, crown cover and meteorological conditions and in
484 the influence of seasonal meteorological conditions. Our results suggested that resource
485 matching may occur in both Quercus species and P. abies. We instead observed a trade-
486 off between growth and reproduction, following the resource-switching model, for A. alba,
487 P. menziesii and F. sylvatica. This response could be due to a strategy to increase
488 nutrient-use efficiency by prioritizing the production of long-lasting tissues. Though there
489 seemed to be reoccurring patterns, there was no obvious differences between evergreen
490 and deciduous trees. We would need a wider sample of species to conclude on whether
491 there is indeed a separation in the patterns used by each.
492
493 Finally, the use of EVI as the sole monitoring tool could not accurately assess fluctuations
494 in fruit production. Identifying species-specific relationships between EVI and fruit
495 production will be needed to predict the sizes of fruit crops. The variability of the weather
496 combined with EVI could be a useful proxy of the patterns of fruit production.
497
498
499 Acknowledgments
500
501 This research was supported by the European Research Council Synergy grant ERC-
502 2013-SyG 610028-IMBALANCE-P, the Spanish Government grant CGL2016-79835-P
503 and the Catalan Government project SGR 2017-1005. Marcos Fernández-Martínez is a
504 Postdoctoral Fellow of the Research Foundation – Flanders (FWO). We thank the ICP
505 Forest organization and collaborators for providing the litterfall data used in this study.
506 We thank Sara Raquel Marrafa for her collaboration in this study.
507 References
508
509 Adams HD, Guardiola-Claramonte M, Barron-Gafford GA, et al (2009) Temperature
510 sensitivity of drought-induced tree mortality portends increased regional die-off
511 under global-change-type drought. Proc Natl Acad Sci USA 106:7063–7066. doi:
512 10.1073/pnas.0901438106
513 Aerts R (1995) The advantages of being evergreen. Trends Ecol. Evol. 10:402–407
514 Barton K (2018) MuMIn: Multi-Model Inference. -R package version 1.40.4.
515 Bogdziewicz M, Fernández-Martínez M, Bonal R, et al (2017) The Moran effect and
516 environmental vetoes: phenological synchrony and drought drive seed production
517 in a Mediterranean oak. Proc R Soc B 284:20171784. doi:
518 10.1098/rspb.2017.1784
519 Bogdziewicz M, Zwolak R, Crone EE (2016) How do vertebrates respond to mast
520 seeding? Oikos. doi: 10.1111/oik.03012
521 Breheny P, Burchett W (2017) Visualization of Regression Models using visreg. R J
522 9:56–71
523 Camarero JJ, Albuixech J, López-Lozano R, et al (2010) An increase in canopy cover
524 leads to masting in Quercus ilex. Trees - Struct Funct 24:909–918. doi:
525 10.1007/s00468-010-0462-5
526 Carevic FS, Fernández M, Alejano R, et al (2010) Plant water relations and
527 edaphoclimatic conditions affecting acorn production in a holm oak (Quercus ilex
528 L. ssp. ballota) open woodland. Agrofor Syst 78:299–308. doi: 10.1007/s10457-
529 009-9245-7
530 Doi H, Takahashi M (2008) Latitudinal patterns in the phenological responses of leaf
531 colouring and leaf fall to climate change in Japan. Glob Ecol Biogeogr 17:556–
19
532 561. doi: 10.1111/j.1466-8238.2008.00398.x
533 Espelta JM, Arias-Leclaire H, Fernandez-Martinez M, et al (2017) Beyond predator
534 satiation: Masting but also the effects of rainfall stochasticity on weevils drive
535 acorn predation. Ecosphere 8:. doi: 10.1002/ecs2.1836
536 Espelta JM, Corté PS, Molowny-horas R, et al (2008) Masting mediated by summer
537 drought reduces acorn predation in Mediterranean oak forests. Ecology 89:805–
538 817
539 Estiarte M, Peñuelas J (2015) Alteration of the phenology of leaf senescence and fall in
540 winter deciduous species by climate change: Efects on nutrient proficiency. Glob
541 Chang Biol 21:1005–1017. doi: 10.1111/gcb.12804
542 Falge E, Tenhunen J, Baldocchi D, et al (2002) Phase and amplitude of ecosystem
543 carbon release and uptake potentials as derived from FLUXNET measurements.
544 Agric For Meteorol 113:75–95. doi: 10.1016/S0168-1923(02)00103-X
545 Fernández-Martínez M, Belmonte J, Maria Espelta J (2012) Masting in oaks:
546 Disentangling the effect of flowering phenology, airborne pollen load and drought.
547 Acta Oecologica 43:51–59. doi: 10.1016/j.actao.2012.05.006
548 Fernández-Martínez M, Garbulsky M, Peñuelas J, et al (2015) Temporal trends in the
549 enhanced vegetation index and spring weather predict seed production in
550 Mediterranean oaks. Plant Ecol 216:1061–1072. doi: 10.1007/s11258-015-0489-1
551 Fernández-Martínez M, Vicca S, Janssens IA, et al (2018) The consecutive disparity
552 index, D, as measure of temporal variability in ecological studies. Ecosphere 9:.
553 doi: 10.1002/ecs2.2527
554 Fernández-Martínez M, Vicca S, Janssens IA, et al (2017a) The North Atlantic
555 Oscillation synchronises fruit production in western European forests. Ecography
556 (Cop) 40:864–874. doi: 10.1111/ecog.02296
557 Fernández-Martínez M, Vicca S, Janssens IA, et al (2017b) The role of nutrients,
558 productivity and climate in determining tree fruit production in European forests.
559 New Phytol 213:669–679. doi: 10.1111/nph.14193
560 Fernández-Martínez M, Yu R, Gamon J, et al (2019) Monitoring Spatial and Temporal
561 Variabilities of Gross Primary Production Using MAIAC MODIS Data. Remote
562 Sens 11:874. doi: 10.3390/rs11070874
563 Fox J, Weisberg S (2011) An {R} Companioon to Applied Regression, Second. Sage,
564 Thousand Oaks CA
565 Frenguelli G, Bricchi E (1998) The use of the pheno-climatic model for forecasting the
566 pollination of some arboreal taxa. Aerobiologia (Bologna) 14:39–44. doi:
567 10.1007/BF02694593
568 Garbulsky MF, Peñuelas J, Ogaya R, Filella I (2013) Leaf and stand-level carbon
569 uptake of a Mediterranean forest estimated using the satellite-derived reflectance
570 indices EVI and PRI. Int J Remote Sens 34:1282–1296. doi:
571 10.1080/01431161.2012.718457
572 Havranek WM, Tranquillini W (1995) Physiological Processes during Winter Dormancy
573 and THeir Ecological Significance. In: Smith WK, Hinckley TM (eds)
574 Ecophysiology of Coniferous Forests. Academic Press Limited, San Diego, pp 95–
575 124
576 Heath JP (2006) Quantifying temporal variability in population abundances. Oikos
577 115:573–581
578 Huete A, Didan K, Miura H, et al (2002) Overview of the radiometric and biopyhsical
579 performance of the MODIS vegetation indices. Remote Sens Environ 83:195–213
580 Ichie T, Igarashi S, Yoshida S, et al (2013) Are stored carbohydrates necessary for
581 seed production in temperate deciduous trees? J Ecol 101:525–531. doi:
582 10.1111/1365-2745.12038
583 Isagi Y, Sugimura K, Sumida A, Ito H (1997) How Does Masting Happen and
584 Synchronize ? J Theor Biol 187:231–239. doi: 10.1006/jtbi.1997.0442
585 Kelly D (1994) The evolutionary ecology of mast seeding. Trends Ecol Evol 9:465–470.
586 doi: 10.1016/0169-5347(94)90310-7
587 Kelly D, Geldenhuis A, James A, et al (2013) Of mast and mean: Differential-
588 temperature cue makes mast seeding insensitive to climate change. Ecol Lett
589 16:90–98. doi: 10.1111/ele.12020
590 Kelly D, Sork VL (2002) Mast Seeding in Perennial Plants: Why, How, Where? Annu
591 Rev Ecol Syst 33:427–447. doi: 10.1146/annurev.ecolsys.33.020602.095433
592 Koenig WD, Knops JMH (2000) Patterns of Annual Seed Production by Northern
593 Hemisphere Trees: A Global Perspective. Am Nat. doi: 10.1086/303302
594 Martín-Benito D, Cherubini P, Del Río M, Cañellas I (2008) Growth response to climate
595 and drought in Pinus nigra Arn. trees of different crown classes. Trees - Struct
596 Funct 22:363–373. doi: 10.1007/s00468-007-0191-6
597 Martín-Vide J (1986) Notes per a la definició d’un índex de «desordre» en pluviometria.
598 Soc Catalana Geogr 7:89–96
599 Norton DA, Kelly D (1988) Mast Seeding Over 33 Years by Dacrydium cupressinum
600 Lamb. (rimu) (Podocarpaceae) in New Zealand: The Importance of Economies of
601 Scale. Funct Ecol 2:399. doi: 10.2307/2389413
602 Olivar J, Bogino S, Spiecker H, Bravo F (2011) Climate impact on growth dynamic and
603 intra-annual density fluctuations in Aleppo pine (Pinus halepensis) trees of
604 different crown classes. Dendrochronologia 30:35–47. doi:
605 10.1016/j.dendro.2011.06.001
606 Ostfeld RS, Keesing F (2000) Pulsed resources and community dynamics of
607 consumers in terrestrial ecosystems. Trends Ecol. Evol. 15:232–237
608 Pearl J, Verma T (1988) Causal Networks: Semantics and Expressiveness.
609 Probabalistic Reason Intell Syst 352–359
610 Pearse IS, Koenig WD, Kelly D (2016) Mechanisms of mast seeding: resources,
611 weather, cues, and selection. New Phytol 212:546–562. doi: 10.1111/nph.14114
612 Peñuelas J, Sardans J, Filella I, et al (2017) Impacts of Global Change on
613 Mediterranean Forests and Their Services. Forests 8:463. doi: 10.3390/F8120463
614 Pérez-Ramos IM, Padilla-Díaz CM, Koenig WD, Marañón T (2015) Environmental
615 drivers of mast-seeding in Mediterranean oak species: Does leaf habit matter? J
616 Ecol 103:691–700. doi: 10.1111/1365-2745.12400
617 Pinheiro J, Bates D, DebRoy S, et al (2018) nlme: linear and nonlinear mixed effects
618 models.-R package ver. 3.1-137
619 Sala A, Hopping K, McIntire EJB, et al (2012) Masting in whitebark pine (Pinus
620 albicaulis) depletes stored nutrients. New Phytol 196:189–199. doi:
621 10.1111/j.1469-8137.2012.04257.x
622 Sánchez Humanes B, Espelta Morral JM (2011) Acorn production and growth in
623 Mediterranean oaks trade-offs and the role of climate, modular organization and
624 phenology. Universitat Autònoma de Barcelona
625 Sardans J, Peñuelas J (2007) Drought changes phosphorus and potassium
626 accumulation patterns in an evergreen Mediterranean forest. Funct Ecol 21:191–
627 201. doi: 10.1111/j.1365-2435.2007.01247.x
628 Satake A, Iwasa Y (2000) Pollen Coupling of Forest Trees : Forming Synchronized and
629 Periodic Reproduction out of Chaos. J Theor Biol 203:63–84. doi:
630 10.1006/jtbi.1999.1066
631 Smaill SJ, Clinton PW, Allen RB, Davis MR (2011) Climate cues and resources interact
632 to determine seed production by a masting species. J Ecol 99:870–877. doi:
633 10.1111/j.1365-2745.2011.01803.x
634 Staaf H, Stjernquist I (1986) Seasonal dynamics, especially autumnal retranslocation,
635 of nitrogen and phosphorus in foliage of dominant and suppressed trees of beech,
636 fagus sylvatica. Scand J For Res 1:333–342. doi: 10.1080/02827588609382425
637 Ukonmaanaho L, Pitman R, Bastrup-Birk A, et al (2016) Sampling and analysis of
638 litterfall. In: ICP Forests Manual, vol. XII. ICP Forests, Hamburg, Germany:
639 Institute for World Forestry
640 Venerables WN, Ripley BD (2002) Modern Applied Statistics with S, Fourth. Springer,
641 New York
642 Voyer AG, Garamszegi LZ (2014) An introduction to phylogenetic path analysis. In:
643 Modern Phylogenetic Comparative Methods and their Application in Evolutionary
644 Biology. Springer Berlin Heidelberg, Berlin, Heidelberg, pp 201–229
645
646 Tables
647
648 Table 1. Summary of the GLMMs correlating fruit production with temperature (T) and
649 precipitation (P) and for Enhanced Vegetation Index (EVI) for the autumn of the year
650 previous to fruit ripening (au), winter (w), spring (sp) and summer (sm). Only the
651 standardized coefficients for the most relevant variables identified by the BIC are
652 presented. The variances explained by fixed (R2m) and random (R2c) factors are also
653 presented.
654 Significance levels: *: P < 0.05; **: P < 0.01; ***: P < 0.001
655
656 Table 2. Summary of standardized total effects in the path analyses associating weather
657 (T, temperature; P, precipitation) and Enhanced Vegetation Index (EVI) for each season
658 (w, winter; sp, spring; sm, summer; au, autumn) with annual fruit production. Only
659 significant (P < 0.05) values are shown.
660
661
662
25
663 Table 1
664
β ± SE ΔBIC R2m R2c
Pinaceae
Abies alba
Tw -0.382 ± 0.100 *** 7.14
EVIsm -0.282 ± 0.098 * 3.19
Model 0.218 0.339
Picea abies
Pw 0.203 ± 0.085 * 0.81
Psp -0.270 ± 0.086 ** 4.06
Model 0.066 0.624
Pseudotsuga menziesii
Tw -0.484 ± 0.125 *** 7.72
Tsm -0.432 ± 0.108 *** 8.89
Model 0.168 0.559
Fagaceae
Fagus sylvatica
Pw 0.344 ± 0.075 *** 15.01
Psp -0.170 ± 0.074 * 0.59
EVIau 0.203 ± 0.073 ** 3.02
EVIsp -0.276 ± 0.078 *** 7.44
EVIsm -0.247 ± 0.077 ** 5.46
Model 0.294 0.347
Quercus petraea
Tsm 0.254 ± 0.083 ** 4.33
EVIsp 0.479 ± 0.083 *** 25.08
Model 0.218 0.218
Quercus robur
Tau 0.442 ± 0.108 *** 11.57
Tsm 0.272 ± 0.118 * 1.28
Psm 0.353 ± 0.117 ** 4.98
Model 0.346 0.346
Significance levels: *: P < 0.05; **: P < 0.01; ***: P < 0.001
665
26
666 Table 2
667
Pinaceae Fagaceae
A. alba P.menziesii P. abies F. sylvatica Q. petraea Q. robur
Seasonal EVI
EVIau - 0.12 - - - -
EVIsp - -0.25 - - 0.45 -
Seasonal
temperature
Tau - - - - 0.5 0.34
Tw -0.33 -0.24 - - 0.51 -
Tsp - 0.08 - -0.35 0.44 -
Tsm - - 0.45 - - -
Seasonal
precipitation
Pau - - - - 0.2 -
Psp -0.24 - - - - -
668
669
27
670 Figures
671
672 Figure 1. Map of the distribution of the studied forests.
673
674 Figure 2. Box-and-whisker plots of the average fruit production (a), average Enhanced
675 Vegetation Index (EVI) (b), proportional variability index for both fruit production (c) and
676 EVI (d) and disparity index for both fruit production (e) and EVI (f) for each species. White
677 boxes represent evergreen species (Abies alba, Pseudotsuga menziesii and Picea
678 abies), and gray boxes represent deciduous species (Fagus sylvatica, Quercus petraea
679 and Q. robur). Different letters indicate significant differences among the species.
680 Average values and standard errors for each species are presented in Appendix Table
681 S1. Boxes hold 50% of the data, with the bold horizontal line in the box indicating the
682 median of the distribution. Whiskers hold the lower and higher 25% of the data, with the
683 end of each whisker indicating respectively the minimum and the maximum value of the
684 distribution. Dots outside the box and whiskers plot indicate outliers of the distribution.
685
686 Figure 3. Directed acyclic graphs of the d-separation path analysis for Abies alba (a),
687 Picea abies (b), Pseudotsuga menziesii (c), Fagus sylvatica (d), Quercus petraea (e)
688 and Q. robur (f). Each graph shows the direct effects of temperature (T), precipitation
689 (P), and Enhanced Vegetation Index (EVI) for winter (w), spring (sp), summer (sm) and
690 autumn (au) on annual fruit production (fNPP). The values of the effects are standardized
691 estimates for comparing the variables within each model. Only significant paths (P <
692 0.05) are shown. Total effects on fruit production are presented in Table 1.
693
694 Figure 4. Partial-residual plots of the models correlating fruit production (fNPP) with
695 temperature (T) and precipitation (P) for autumn (au), winter (w), spring (sp) and summer
696 (sm) for Abies alba (a), Pseudotsuga menziesii (b and c), Picea abies (d and e), Quercus
28
697 petraea (f), Q. robur (g and h) and Fagus sylvatica (i and j). Beta weights (β ± SE) and
698 the P values are indicated for each model. The blue lines represent the slope of the
699 relationship, and the gray shading indicates the 95% confidence intervals. Only the two
700 most relevant relationships between weather and fruit production are shown for Q. robur
701 (based on the BIC); see Table 2 for detailed information.
702
703 Figure 5. Partial-residual plots of the models correlating fruit production (fNPP) with
704 Enhanced Vegetation Index (EVI) for autumn (au), spring (sp) and summer (sm) for
705 Fagus sylvatica (a and b), Quercus petraea (c) and Abies alba (d). Beta weights (β ± SE)
706 and P values are indicated for each model. The blue lines represent the slopes of the
707 relationship, and the gray shading indicates the 95% confidence intervals. Only the two
708 most relevant relationships between EVI and fruit production are shown for F. sylvatica
709 (based on the BIC); see Table 2 for detailed information.
710 Figure 1.
711
712
30
713 Figure 2.
714
715
31
716 Figure 3.
717
718
32
719 Figure 4.
720
721
33
722 Figure 5.
723
724
34
725 Supplementary material
726
Table S1. Summary of average fruit production (fNPP) and Enhanced Vegetation Index (EVI) (means ± SEs) and the proportional variability (PV) and disparity (D)
indices. Means were estimated per year (between 2000 and 2008) for each site and species. Different letters within a column indicate significant differences among the
species (identified by Tukey's HSD test for multiple comparisons).
fNPP EVI
Species
Mean PV D Mean PV D
Pinaceae
Abies alba 20.29 ± 0.57ab 0.61 ± 0.04a 1.03 ± 0.12ab 0.34 ± 0.33a 0.24 ± 0.06a 0.28 ± 0.10a
Picea abies 30.01 ± 0.51b 0.54 ± 0.04a 0.88 ± 0.11a 0.32 ± 0.26a 0.21 ± 0.06a 0.24 ± 0.09a
Pseudotsuga menziesii 6.08 ± 0.56a 0.52 ± 0.07a 0.64 ± 0.10a 0.41 ± 0.32bc 0.23 ± 0.03a 0.28 ± 0.04a
Fagaceae
Fagus sylvatica 13.41 ± 0.99a 0.80 ± 0.01b 2.07 ± 0.17c 0.39 ± 0.99b 0.40 ± 0.06b 0.57 ± 0.12b
Quercus petraea 12.34 ± 1.01a 0.79 ± 0.01b 1.67 ± 0.13bc 0.42 ± 1.02bc 0.40 ± 0.06b 0.54 ± 0.11b
Quercus robur 9.51 ± 0.81a 0.81 ± 0.02b 1.69 ± 0.16bc 0.43 ± 0.66c 0.37 ± 0.05b 0.48 ± 0.07b
727
35
728 Method S1. The proportional variability (PV) index
729
730 PV is used to assess the variability of a time series by calculating the average difference
731 between all possible combinations of values in the time series, ignoring the chronological
732 order of the values. PV, however, differs from the coefficient of variation (CV) in that it is
733 independent from the mean. That is, PV does not estimate the deviation from the mean
734 of the series but compares all values with each other, so it is less sensitive to non-
735 normally distributed data.
736
737 PV is calculated as:
738
𝑛(𝑛 − 1) (𝐸𝑞. 𝐴1)
𝐶=
2
min(𝑝𝑖 , 𝑝𝑗 ) (𝐸𝑞. 𝐴2)
𝐷(𝑝) = 1 −
max(𝑝𝑖 , 𝑝𝑗 )
∑𝑝 𝐷(𝑝) (𝐸𝑞. 𝐴3)
𝑃𝑉 =
𝐶
739
740 The distribution of proportional differences, 𝐷(𝑝), is first calculated using all possible
741 combinations of values in our time series (𝐶; Eq. A1), with each p representing a possible
742 pair of values. The values are then compared by dividing the absolute proportional
743 difference in each pair by the maximum value of each pair (Eq. A2). Finally, PV is
744 obtained by dividing the sum of the values of 𝐷(𝑝) by the total number of possible
745 combinations in our time series (Eq. A3).
746
747
748 References:
749 Heath JP (2006) Quantifying temporal variability in population abundances. Oikos
750 115:573–581
36
751 Method S2. The consecutive disparity (D) index
752
753 D is a measure of the variability of a time series that accounts for the chronological order
754 of values in the series, which better assesses the interannual tendencies of the series.
755 In contrast to CV, D does not depend on the mean of the time series but entirely on its
756 chronological order.
757
758 D is calculated as:
759
𝑛−1 𝑝𝑖+1
𝐸= ∑ |𝑙𝑛 | (𝐸𝑞. 𝐴4)
𝑖=1 𝑝𝑖
𝐸
𝐷= (𝐸𝑞. 𝐴5)
𝑛−1
760
761
762 The disorder (E) of the time series is first assessed (Eq. A4), where 𝑝𝑖 is a value of a
763 series of length 𝑛. D is then calculated by dividing E by 𝑛 − 1 (Eq. A5), which compares
764 time series. If the time series contains a zero, a constant, 𝑘, may be added to avoid
765 division by zero (Eq. A6):
766
1 𝑛−1 𝑝𝑖+1 + 𝑘
𝐷= ·∑ |𝑙𝑛 | (𝐸𝑞. 𝐴6)
𝑛−1 𝑖=1 𝑝𝑖 + 𝑘
767
768
769 References:
770 Fernández-Martínez M, Vicca S, Janssens IA, et al (2018) The consecutive disparity
771 index, D, as measure of temporal variability in ecological studies. Ecosphere 9:
772 doi: 10.1002/ecs2.2527
37
773 Fernández-Martínez M, Vicca S, Janssens IA, et al. (2017b) The role of nutrients,
774 productivity and climate in determining tree fruit production in European forests.
775 New Phytol. 213:669–679. doi: 10.1111/nph.14193
776 Martín-Vide J (1986) Notes per a la definició d’un índex de «desordre» en pluviometria.
777 Soc. Catalana. Geogr. 7:89–96