Nothing else matters? Tree characteristics have more effects than biogeoclimatic context on microhabitat diversity and occurrence: a nationwide analysis

Managing forests to preserve biodiversity requires a good knowledge not only of the factors driving its dynamics but also of the structural elements that actually support biodiversity. Tree-related microhabitats (cavities, cracks, conks of fungi) are tree-borne features that are reputed to support specific biodiversity for at least a part of species’ life cycles. While several studies have analysed the drivers of microhabitats number and occurrence at the tree scale, they remain limited to a few tree species located in relatively narrow biogeographical ranges. We used a nationwide database of forest reserves where microhabitats were inventoried on more than 22,000 trees. We analysed the effect of tree diameter and living status (alive or dead) on microhabitat number and occurrence per tree, taking into account biogeoclimatic variables and tree genus.We confirmed that larger trees and dead trees bore more microhabitats than their smaller or living counterparts did; we extended these results to a wider range of tree genera and ecological conditions. Contrary to expectations, the total number of microhabitat types per tree barely varied with tree genus – though we did find slightly higher accumulation levels for broadleaves than for conifers – nor did it vary with elevation or soil pH. We observed the same results for the occurrence of individual microhabitat types. However, accumulation levels with diameter and occurrence on dead trees were higher for microhabitats linked with wood decay processes (e.g. dead branches or woodpecker feeding holes) than for other, epixylic, microhabitats such as epiphytes (ivy, mosses and lichens).Promoting large living and dead trees of several tree species may be an interesting, and nearly universal, way to favour microhabitats and enhance the substrates needed to support specific biodiversity. In addition, a better understanding of microhabitat drivers and dynamics at the tree scale may help to better define their role as biodiversity indicators for large-scale monitoring.

used a nationwide database of forest reserves where microhabitats were inventoried on more 23 than 22,000 trees. We analysed the effect of tree diameter and living status (alive or dead) on 24 microhabitat number and occurrence per tree, taking into account biogeoclimatic variables and 25 tree genus. 26 We confirmed that larger trees and dead trees bore more microhabitats than their smaller or 27 living counterparts did; we extended these results to a wider range of tree genera and 28 ecological conditions. Contrary to expectations, the total number of microhabitat types per tree 29 barely varied with tree genusthough we did find slightly higher accumulation levels for 30 broadleaves than for conifersnor did it vary with elevation or soil pH. We observed the same 31 results for the occurrence of individual microhabitat types. However, accumulation levels with 32 diameter and occurrence on dead trees were higher for microhabitats linked with wood decay 33 processes (e.g. dead branches or woodpecker feeding holes) than for other, epixylic, 34 microhabitats such as epiphytes (ivy, mosses and lichens). 35 Introduction (83180 living and 19615 dead trees, snags or stumps). The forest reserves in the database 96 actually encompass three broad types of protection status. First, (i) strict forest reserves, where 97 harvesting has been abandoned for a variable timespan and (ii) special forest reserves, where 98 management is targeted towards specific biodiversity conservation measures (e.g. 99 preservation of ponds). These two types are owned and managed by the French National 100 Forest Service. The third type, nature reserves, on the other hand, where management varies 101 from abandonment to classic wood production, may be of various ownership types (state, local 102 authorities, private). It should be noted that no homogeneous data on management intensity 103 or time since last harvesting could be gathered at the plot level for all the reserves in the 104 database. However, Vuidot et al. [13] showed that management has a limited effect on 105 microhabitat number and occurrence at the tree level. We thus assumed that management 106 differences would not play a significant role at the tree scale and therefore, did not take 107 management type or intensity into account in our analyses (but see discussion). 108 109 Stand structure and microhabitat inventories 110 On each plot, we combined two sampling methods to characterise forest stand structure [20]. 111 For all living trees with a diameter at breast height (DBH) above 30 cm, we used a fixed angle 112 plot method to select the individuals comprised within a relascopic angle of 3%. Practically, 113 this meant that sampling distance was proportional to the apparent DBH of a tree. For example, 114 a tree with a DBH of 60 cm was included in the sample if it was within 20 m of the centre of the 115 plot. This particular technique allowed us to better account for larger trees at a small scale. All 116 other variables were measured on fixed-area plots. Within a fixed 10-m (314 m 2 ) radius around 117 the plot centre, we measured the diameter of all living trees and snags (standing dead trees 118 with a height > 1.30 m) from 7.5 to 30 cm DBH. Within a 20-m radius (1256 m 2 ), we recorded 119 all snags with a diameter > 30 cm. Whenever possible, we identified all trees, both alive and 120 dead, to species level. In the subsequent analyses, we grouped some tree species at the 121 genus level to have sufficient representation in terms of tree numbers. This resulted in the 122 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; 6 following groups: Ash, Beech, Chestnut, Fir, Hornbeam, Larch, Maple (90% sycamore maple), 123 Oak (80% sessile and pedunculate oaks combined, 15% oaks identified to the genus level 124 only, 5% other oaksmainly Mediterranean), Pine (64% Scots pine, 22% mountain pine), 125 Poplar and Spruce. We assumed that tree genus, rather than species, influenced the 126 relationships we were studying. Unidentified species were excluded from the analyses. 127 We visually inspected all selected standing trees for microhabitats and recorded their presence 128 on each tree. Observers attended a training session and were given a field guide with pictures 129 to help them better determine microhabitat types and detailed criteria to include in the 130 inventories. Although inventory methods have recently improved [5,21], we assumed that the 131 method we used limited any potential observer effect linked with microhabitat inventories [22]. 132 Different microhabitat typologies were used concomitantly during the inventories and 133 harmonization has been lacking since 2005. Consequently, we only retained data with a 134 homogeneous typology. We preferred this solution to grouping microhabitat types to avoid 135 coarser classification with too much degradation of the original dataset. 136 137

Data selection and biogeoclimatic variables extraction 138
First, we focused on the microhabitat typology that was used for the largest number of plots 139 and sites (Table 1). This reduced the dataset to 43 sites comprising 3165 plots ( Figure 1). 140 Second, the smallest trees (7.5 ≤ DBH ≤17.5 cm) accounted for 36% of the trees in the 141 database but were also the least likely to bear microhabitats [12,13]. We therefore excluded 142 this category from the dataset to avoid zero-inflation in the subsequent models. Third, previous 143 studies had shown that tree living status (i.e. living vs. dead trees, see below) is a major driver 144 of microhabitat occurrence and density [12,13]. To properly account for this variable in our 145 statistical models, we excluded all tree species/genera with less than 50 standing dead trees 146 or snags in the dataset (ie. Ash, Chestnut, Hornbeam, Larch, Maple, Poplar, see Table 2 and 147 Supplementary Material, Figure S1, for a calculation based on a larger subset of living trees). 148 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Statistical analyses 158
Following Zuur et al. [25], preliminary data exploration did not reveal any potential variation in 159 the relationship between microhabitat metrics and any of the biogeoclimatic variables 160 mentioned above, apart from pH and elevation. Elevation correlated strongly to tree species; 161 indeed, only Beech and Pine were distributed over the whole elevation gradient while the other 162 species were elevation-dependent. We therefore kept pH and elevation only in the analyses 163 described below. 164 We used DBH, living status (alive vs. dead) and genus (Beech, Fir, Oak, Pine and Spruce) as 165 explanatory variables and included second and third order interactions between DBH, living 166 status and genus in the models. We added elevation and pH as covariables, but only included 167 pH in the second order interactions. Since Beech and Pine were not strongly biased by 168 elevation, we added elevation in the second order interactions for these two genera in two 169 separate analysis. 170 To model the total number of microhabitat types per tree, we used generalised linear mixed 171 models (GLMMs, library glmmTMB, [26]) with a Poisson error distribution for count data and 172 plot identity nested within site as a random variable. We also modelled the occurrence of each 173 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; microhabitat type, but with a binomial error distribution for binary data. We tested differences 174 in microhabitat numbers and occurrences between living and dead trees with post-hoc multi-175

Number of microhabitat types per tree 185
Estimates for all single parameters except for soil pH were significant in the model, while 186 second and third order interactions were less often significant (see Supplementary Materials, 187 Table S1). All tree genera except Pine had higher microhabitat numbers on dead than on living 188 trees. Overall, the difference was the highest for Oak (22% more microhabitats on dead trees, 189 for a mean DBH of 44 cm, Table 3); the other genera had around 10-15% more microhabitats 190 on dead trees. Globally, the number of microhabitats per tree increased with tree diameter, 191 both for living and dead trees ( Figure 2). However, the accumulation of microhabitats with 192 diameter varied with genus: broadleaves (Beech and Oak) had higher accumulation levels 193 than conifers (Fir, Pine, Spruce); and according to living status: dead versus living trees (except 194 for Pine; Figure 2, Supplementary Materials, Table S2). These results were generally 195 consistent with those obtained with the analyses concerning a higher number of genera but for 196 living trees only ( Figure S1). Broadleaves (Ash,Beech,Chestnut,Hornbeam,Maple,Oak,197 Poplar) showed higher microhabitat accumulation rates than conifers (Fir,Larch and Spruce). 198 Only Pine showed accumulation rates comparable to broadleaves ( Figure S1). 199 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; Number of microhabitats increased significantly with elevation, but not with soil pH. However, 200 higher soil pH had a positive effect on the accumulation of microhabitats with DBH (the second 201 order interaction was significant), mostly on dead trees (Supplementary Materials, Table S1). 202 Still, the effects of elevation and soil pH remained small compared to those of DBH and living 203

status. 204
For Beech and Pine, the overall results converged with those of the complete model. Soil pH 205 and elevation only had significant effects in the interaction terms (Supplementary Materials: 206 Five microhabitats out of twenty generally occurred more frequently on standing deadwood 215 than on living trees, though this was not systematic for all genera or even for living status: trunk 216 cavities (except Fir), woodpecker feeding holes (Figure 3), rot (except Fir), conks of fungi, bark 217 characteristics and crown skeleton (except Fir, Table 3 and Supplementary Materials, Table  218 S4). We observed the strongest differences for woodpecker feeding holes: whatever the 219 genera, they virtually only occurred on standing dead trees (i.e. they were nearly absent from 220 living trees, Figure 3, Table 3). Conversely, injuries, dead branches whatever their size and 221 forks (broadleaves only) occurred more frequently on living trees. Magnitudes for microhabitats 222 more frequent on living trees were around 60% to 90% (Table 3). 223 For most microhabitats, the probability of occurrence increased with DBH both for living and 224 dead trees, with the remarkable exceptions of canopy cavities, woodpecker cavities and crown 225 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; skeletons (Supplementary Materials: Figure S2, Table S4). However, the magnitude of the 226 relation varied with tree genus and living status. For some microhabitat types, the increase in 227 probability of occurrence with DBH was stronger for dead than for living trees, e.g. +35% base 228 and trunk cavities on dead vs. +18% on living Beech, or +23 to +42% for woodpecker feeding 229 holes on dead vs. +0.2 to +3% on living trees (Table S2). Conversely, the increase in 230 probability of occurrence of small and medium dead branches was stronger for living trees 231 (e.g. +53% medium dead branches on living vs. 0.7% on dead Oak) and, to a lesser extent, 232 for mosses on Beech and Fir (+20% and +24% on living trees, vs. +9% and +16% on dead 233 trees, respectively). All other increments with DBH for living trees were smaller, generally 234 below 10%. Note that in some cases, due to the very limited number of occurrences for some 235 microhabitats on certain tree genera, the estimates proved unreliable (huge confidence 236 intervals, e.g. canopy cavities on Oak, Pine and Spruce, Supplementary Materials: Figure S2, 237 Table S4). 238 Elevation had an overall negative effect on microhabitat occurrence, except for trunk cavities, 239 lichens and forks. Conversely, soil pH tended to have a positive effect on microhabitat 240 occurrence, except for conks of fungi. More interestingly, increasing soil pH had a positive 241 effect on the accumulation of some microhabitats when coupled with DBH (indicated by a 242 significant interaction term), but a negative effect on occurrence on living trees (Supplementary 243 Materials: Table S4). All these significant effects exhibited widely varying levels of magnitude, 244 and in several cases, the estimates were rather imprecise (Supplementary Materials: Figure  245 S2, Table S4). 246 247

248
Numerous recent studies in a variety of contexts have shown that the number of microhabitats 249 per tree as well as the occurrence of some microhabitat types increase with tree diameter [11, 250 14, 16]; the studies also show higher occurrence levels on dead than on living trees [12,13]. 251 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; Our nationwide study based on a large database confirmed these relationships and extended 252 them to a larger range of tree genera under wider biogeographical conditions. Indeed, our 253 results include five tree genera for both living and dead trees and eleven genera when only 254 living trees were taken into account (Supplementary Materials: Figure S1). 255 256 Dead trees bear more microhabitats than living trees 257 Standing dead trees contribute significantly to the supply of microhabitats; overall, they bore 258 10 to 20% more microhabitats than their living counterparts in our dataset. Dead trees often 259 bear considerably more microhabitats than living trees when individual microhabitat types are 260 analysed (e.g. woodpecker feeding holes - Figure 3 or bark characteristics). Once dead, 261 standing trees are affected by decomposition processes that trigger microhabitat genesis [15]. 262 Standing dead trees also constitute privileged foraging grounds for a number of species [5,7,263 8], including woodpeckers [31,32]. In particular, insect larvae or ants that live under the bark 264 of more or less recently dead trees constitute a non-negligible part of some birds' diet [8, 33, 265 34]. Furthermore, as living trees also bear microhabitats, it seems logical that many of these 266 would persist when the tree dies and would continue to evolve, or possibly even condition the 267 presence of other microhabitats linked with the decaying process [15]. For example, injuries 268 caused by logging, branch break or treefall could begin to rot, then slowly evolve into decay 269 cavities [5,35]. These successional changes are likely to explain why these microhabitats 270 types are more numerous on dead trees. The only exceptions to this global pattern concerned 271 epiphytes and forks with accumulated organic matter, which both tend to be more numerous 272 on living trees. Ivy, mosses and lichen are likely to benefit from bark characteristics (e.g. pH, 273 [36]) occurring only on living trees. Epiphytes, especially slow-growing mosses and lichens, 274 require a relatively stable substrate to take root and develop [37]. Stability is lost when bark 275 loosens and falls off during tree senescence, and this could cause epiphytic abundance to 276 decrease. In a nutshell, decaying processes linked to the tree's death reveal a clear difference 277 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; between microhabitats that are linked to decay (i.e. saproxylic microhabitats, sensu [5]) and 278 those that are notor less so (i.e. epixylic microhabitats). 279 Previous studies comparing microhabitat numbers on living and dead trees almost all found 280 more microhabitats on dead trees (see [17]). However, the degree of difference varies across 281 studies, from 1.2 times as many microhabitats in Mediterranean forests [16] and twice as many 282 in five French forests [13] to four times as many on habitat trees in south-western Germany 283 [38]. Our results ranged from 1.1 to 1.2 times as many microhabitats on dead as on living trees, 284 which is of a slightly lower order of magnitude than previously reported. This surprising result 285 may be due to the fact that our study encompassed more species with a lower microhabitat 286 bearing potential (namely conifers). Yet, even for the same species analysed in previous 287 studies (e.g. Beech), the levels we observed were lower, and this despite the small effects of 288 pH and elevation; this finding seems to indicate that the effect is not due to biogeographical 289 variation. 290 291

Number and occurrence of microhabitats increase with tree diameter 292
We confirmed that both microhabitat number and occurrence increase with tree diameter but, 293 contrary to expectations ([11-13], but see [14]), tree genus had a limited effect on this 294 relationship, with only slightly higher microhabitat accumulation levels on broadleaves than on 295 conifers. Almost all microhabitat types taken individually showed the same increasing trends 296 with tree DBH, but there were considerable variations in magnitude. Larger (living) trees have 297 generally lived longer than smaller ones, and are consequently more likely to have suffered 298 damage during their lifespan due to meteorological events (storms, snowfall), natural hazards 299 (rockfalls) or use by different tree-and wood-dependent species (woodpeckers, beetles, fungi, 300 e.g. [13,39]). Doubling tree diameter (from 50 to 100 cm) roughly doubles the number of tree 301 microhabitats [13,17,18], though some studies have found multiples of up to four [38] or even 302 five times [12] in certain cases. Again, our results showed magnitudes below the lower end of 303 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; 13 this range (the multiplication coefficient ranged from 1.2 to 1.4). This may be because the 304 largest trees in our dataset were undoubtedly younger than those in the other studies, 305 especially in studies on near-natural or long-abandoned forests [12,13]. Indeed, since most of 306 our sites had been (more or less) recently managed, selective felling may have cause trees 307 with a given diameter to be younger than their counterparts in primeval forests, where 308 competition levels may be higher and cause slower growth rates. At the individual microhabitat 309 scale, dead branches were more likely to occur on large trees than on smaller trees; although 310 this result seems quite obvious, it had rarely been quantified before. Larger trees have more, 311 but also larger, branches likely to die from competition with neighbours, especially in 312 broadleaves [40]. Indeed, Oak and Beech were the genera that showed the highest large dead 313 branch accumulation rates with diameter in our analyses, while conifers had almost no large 314 dead branches. 315 Cavity birds and bats are reputed to prefer larger trees for nesting or roosting [41,42], since 316 thicker wood surrounding the cavity provides a better buffered and more stable microclimatic 317 conditions [43]. However, we did not confirm this relationship; the accumulation rates of 318 woodpecker cavities with tree diameter was very weak and non-significant. The supposed 319 relationship between tree diameter and woodpecker cavity occurrence seems hard to prove in 320 the context of temperate European forests, at least with data from censuses comparable to 321 ours (see [13] at the tree scale, or [44] at the stand scale); more targeted research focusing on 322 this specific relationship is probably needed [31,45]. Our results could also be linked to the 323 non-linear dynamics [11] of this particular microhabitat. Some cavities in living beech can close 324 back up when they are no longer used [pers. obs. Y.P.], and trees weakened by cavity digging 325 can break, e.g. [45]. Other microhabitats, for instance conks of fungi, may also show non-linear 326 dynamics linked with specific phenology [46]. In our study, the number and occurrence of 327 microhabitats also increased with diameter in standing dead trees, sometimes at a higher rate 328 than for living trees. The longer persistence of large dead trees compared to smaller ones [47] 329 may combine the effects of increased damage due to hazards and the natural decaying 330 processes described above. This probably explains the higher accumulation levels we 331 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; observed in many cases, especially for saproxylic microhabitats (e.g. rot, feeding holes, trunk 332 cavities). Once again, the only exception to this rule were the epiphytes: their probability of 333 occurrence tended to increase with tree diameter but very noisily, both for living and dead 334 trees. For such epiphytic organisms (ivy, mosses and lichens), larger scale processes and 335 biogeoclimatic context (e.g. soil fertility, precipitation) is probably more important than 336 individual tree characteristics [48]. This is suggested by the significant and rather strong effects 337 of pH and elevation in our analyses (Supplementary Materials, Table S4). 338 339

Limitations and research perspectives 340
Contrary to our expectations, we found a limited effect of biogeoclimatic variables on the 341 relationship between microhabitats, tree diameter and living status. However, some specific 342 interactions may exist, especially in the case of epiphytes [48], but that could not be evidenced 343 by our approach. In addition, it was rather difficult to disentangle the effects of tree genus from 344 those of the biogeoclimatic variables, since the distribution of most tree genera is driven largely 345 by climate -apart from Beech, and more marginally Pine, which occur over broad bioclimatic 346 gradients. However, even when we analysed Beech and Pine separately, we did not find any 347 effect of soil pH or elevation on the number of microhabitats, and only slight effects on 348 accumulation levels with diameter. These results need to be confirmed by further analyses 349 with larger, more carefully controlled biogeographical gradients. 350 Our data from forest reserves potentially reflect a larger anthropogenic gradient than classical 351 managed forests. Some of the reserves had not been harvested for several decades and 352 exhibited characteristics of over-mature forests (see e.g. [20], who analysed some of the 353 reserves included in this paper). On the other hand, their overall structure reflected relatively 354 recent management abandonmentif anysince the reserves were marked by probable 355 intensive use or previous harvesting over the past centuries, as is characteristic of western 356 European forests [49]. This is testified to in the dataset we analysed by the relatively rare 357 occurrence of dead standing trees, in particular those with a large diameter: standing dead 358 trees represented a mere 10% of the total dataset and very large individuals (DBH > 67.5cm) 359 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; only 1% (Table 2). As a consequence, despite the fact that we worked on an extended 360 management gradient ranging from managed forests to unmanaged strict reserves, some of 361 the elements characteristic of old-growth and over-mature forests were still lacking, especially 362 large dead trees [50]. This truncated our relationships and made them imprecise for the larger 363 diameter categories. Further research on the last remnant of old-growth primeval forests in 364 Europe [51,52] is therefore needed to bridge this gap and better understand microhabitat 365 dynamics over the whole lifespan of the tree. 366 Compared to recent developments [5,21], the microhabitat typology we used (Table 1) seems 367 rather coarse or imprecise. This may explain why we were not able to confirm some of the 368 effects mentioned in the literature; different microhabitats from a given group may have 369 different requirements and dynamics (e.g. cavities dug by the black woodpecker vs. other 370 woodpecker species). On the other hand, our descriptions allowed us to have enough 371 occurrences in each type to analyse the combined effects of diameter and genus for almost all 372 the microhabitat types in the typology. Our approach can be viewed as a compromise between 373 providing the necessary sample size for statistical analyses and the degree of refinement in 374 typology. The current developments mentioned above [5] will certainly help to homogenize 375 data in the near future and to build larger, shared databases on common, comparable grounds. 376 Despite a training session prior to the inventories, observer effects cannot be totally ruled-out. 377 Our censuses were mostly performed by non-specialists [22], contrary to the scientific studies 378 previously published, and this may have led to the relatively low magnitudes observed, with 379 the hypothesis that detection error is higher on one status (either dead or living trees) or one 380 type of tree (e.g. small trees, which can be overlooked to the benefit of larger individuals). Such 381 issues remain to be explored. 382 Finally, our models assumedunrealistically as it turns outthat microhabitat number would 383 increase exponentially with diameter. In fact, recent studies, as well as ecological theory (e.g. 384 species-area relationship), tend to show a saturated (e.g. logarithmic or sigmoid) relationship 385 between microhabitats and diameter. Models allowing for different link functionsprobably 386 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; within a Bayesian frameworkwill need to be tested to see whether they perform better than 387 the ones used here (see e.g. [11]). 388 389 Implications for forest management and biodiversity conservation 390 Large old trees are considered keystone small natural features in forest and agro-pastoral 391 landscapes because of their disproportionate importance for biodiversity relative to their size 392 [3]. This role for biodiversity is further enhanced by the 'smaller' natural features -393 microhabitatsthey bear [7]. In our large-scale analysis, we confirmed and extended results 394 previously observed only locally: most microhabitats occur on large trees, and even more on 395 dead ones than on living ones. This relationship seems true for several tree genera included 396 in this analysis, and across a large gradient of ecological conditions, with minor variations in 397 accumulation rates with soil pH and elevation. As a consequence, conserving and recruiting 398 large living and dead trees in daily forest management will enhance structural heterogeneity 399 at the stand scale [6,53], and favour a variety of tree-borne microhabitats, which could further 400 help to better conserve specific forest biodiversity [5,54]. Even though the diameter effect We are in debt to the reserve and forest managers who fed the database and made this study 412 possible. Without their commitment and implication on a daily basis, our results would not have 413 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

467
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

527
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

554
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.  CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/335836 doi: bioRxiv preprint first posted online May. 31, 2018; . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

581
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

587
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

593
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

596
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

601
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.     Figure S1: Relationship between total number of microhabitats per tree (N microhabitats) and Diameter at 613 Breast Height (DBH, cm) by genus and living status (living vs. dead standing trees). Lines represent estimates 614 from generalized mixed effect models with a Poisson error distribution. Ribbons show the 95% confidence 615 interval of the mean. For this representation, pH and elevation were held constant. 616 617 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.