Environmental Filtering and Spatial Processes in Urban Riparian Forests

1 1. Environmental filtering and spatial processes in 2 urban riparian forests

3 Marie-Hélène Brice, Stéphanie Pellerin & Monique Poulin

4 Brice, M.H. ([email protected])1,2, 5 Pellerin, S. (corresponding author, [email protected])1,2, 6 Poulin, M. ([email protected])2,3,

7 1Institut de recherche en biologie végétale, Université de Montréal and Jardin botanique 8 de Montréal, 4101 Sherbrooke Est, Montréal, Qc, Canada, H1X 2B2 9 2Québec Centre for Biodiversity Science, McGill University, 1205 Docteur-Penfield, 10 Montréal, Qc, Canada, H3A 1B1 11 3Department of Phytology, Université Laval, 2425 de l’Agriculture, Québec, Qc, Canada, 12 G1V 0A6

13 Running head: Urbanization effects on riparian forests

1 14 2. Abstract 15 Questions What are the spatial processes structuring plant trait composition in urban 16 riparian forest communities at different spatial scales? What are the relative role of local 17 conditions (including historical aspects), landscape context and spatial processes in the 18 community assembly of these forests? 19 Location Montréal, Québec, Canada 20 Methods Species plant composition was inventoried in 57 riparian forests located along a 21 gradient of urbanization. To analyse plant communities in terms of their trait 22 composition, community weighted means were calculated using eight functional traits. 23 Forests were characterized by local (physical features, hydrological regime and historical 24 disturbances) and landscape (surrounding land use) variables. Spatial processes 25 structuring communities were assessed using Moran's eigenvector maps and asymmetric 26 eigenvector maps. The relative importance of these three subsets (local, landscape and 27 spatial variables) on tree, shrub and herb functional composition was quantified by 28 variation partitioning using redundancy analyses. 29 Results Functional patterns in riparian forests resulted primarily from environmental 30 filtering (local and landscape variables). Local conditions, especially flood intensity, 31 exerted an overriding selection pressure on functional composition of riparian plant 32 communities. Urbanization seemed to act indirectly on trait patterns through the 33 alteration of hydrological disturbances caused by ongoing and historical land 34 transformation. Nevertheless, dispersal along rivers was also a significant structuring 35 force while overland dispersal was negligible. 36 Conclusions Our study highlights that under severe natural disturbance regimes, the 37 effect of natural filters outweighed the negative effects of urban filters. However, the 38 alteration of natural flooding processes by human activities is also a major mechanism 39 influencing plant trait composition in urban riparian communities as forests subjected to 40 reduced flooding intensity experienced greater effect of urbanization. The effects of 41 urbanization and of past land uses on plant communities were greater for trees than for 42 shrubs and herbs due to the high turnover rate of the latter. Finally, our results showed the 43 importance of dispersal along rivers for biodiversity even in fragmented urban 44 landscapes. 45 Keywords Asymmetric eigenvector maps; Community weighted means of traits; 46 Dispersal spatial processes; Ecological filters; Flood; Moran's eigenvector maps; Plant 47 functional traits; Riparian plant communities; Urbanization 48 Abbreviations 49 AEM = Asymmetric eigenvector maps 50 CWM = Community weighted means of traits 51 MEM = Moran’s eigenvector maps

2 52 Introduction 53 Urbanization is a driver of global change and a major threat to biodiversity worldwide. In 54 urban ecosystems, both human and natural filters, operating at multiple spatial and 55 temporal scales, influence the composition of communities through the selection of 56 certain species with specific traits (Williams et al. 2015). At a local scale, human filters 57 acting on species distribution include alteration of habitat conditions and horticultural 58 interventions, while at a regional scale habitat loss and fragmentation prevail (Williams et 59 al. 2008). Furthermore, past land uses that occurred decades or centuries ago, such as 60 agricultural practices prior to urbanization, can still influence current species distribution 61 (i.e., land-use legacy) through the modification of soil properties and the disruption of 62 natural disturbance regimes (Flinn & Vellend 2005; Lewis et al. 2006; Wulf & Kolk 63 2014). Hence, past and current land uses change the quality and spatial configuration of 64 habitats as well as the regional pool of species, which in turn alter natural environmental 65 filtering and dispersal-based processes of community assembly. Consequently, species 66 distribution in urban ecosystems is generally prevailingly structured by anthropogenic, 67 rather than by natural biotic and abiotic factors (Godefroid & Koedam 2007; Knapp et al. 68 2008). Yet, the relative importance of both urban and natural filters on riparian 69 communities has been overlooked until recently (e.g., Burton et al. 2009; Sung et al. 70 2011). 71 In riparian forests, floods constitute both a destructive and regenerating force that 72 strongly structures plant communities (Naiman et al. 2005). Riparian species have 73 developed numerous adaptations to waterlogging, fluctuating water levels and physical 74 stress from currents. For example, the production of aerenchyma and growth of 75 adventitious roots are traits that enable species to survive anoxia resulting from prolonged 76 waterlogging (Lytle & Poff 2004; Parent et al. 2008). In response to extreme fluctuations 77 ranging from water excess to water shortage, some plants persist as soil seed bank and 78 wait for suitable conditions for germination and growth (Gurnell et al. 2006; Moggridge 79 et al. 2009). Stress from floods and sediment erosion/deposition can favor plants with 80 flexible branches or capable of vegetative reproduction from ripped off fragments 81 (Karrenberg et al. 2002; Lytle & Poff 2004). In addition to strongly influencing local 82 conditions, rivers structure riparian communities by acting as an important upstream to 83 downstream dispersal vector (Moggridge et al. 2009; Nilsson et al. 2010), especially for 84 species with high seed buoyancy (Johansson et al. 1996). Urbanization can, however, 85 disrupt these natural hydrological processes. 86 Dispersal is an important mechanism of community assembly that structures 87 diversity and induces spatial patterns (Logue et al. 2011). While riparian forests are 88 highly connected through the river network (Nilsson et al. 2010), urban forests are highly 89 isolated by habitat fragmentation (Williams et al. 2008). These contrasting contexts 90 impede the extension of traditional predictions about terrestrial remnants to riparian 91 remnants in urban areas, but they provide an opportunity to study the importance of 92 different dispersal processes in community assembly. Although dispersal is expected to 93 mainly occur in a downstream direction (Nilsson et al. 2010), this directional process 94 may not be as effective in a fragmented landscape, and urban riparian communities may 95 also be influenced by other long-distance overland dispersal vectors, such as animal or 96 human (e.g., Knapp et al. 2009). In addition, hydraulic structures on rivers (e.g., dams 97 and dikes) disrupt upstream-downstream connectivity (Jansson et al. 2000; Merritt et al.

3 98 2010) as well as lateral connectivity, which can lead to a terrestrialization of riparian 99 communities by allowing upland species to invade the formerly dynamic riparian habitat 100 (Ward & Stanford 1995). Spatial modeling methods (Dray et al. 2006) can be used to 101 shed light on the pathways of dispersal processes in riparian plant communities and 102 assess the importance of spatial processes relative to local and landscape factors. 103 This study aims to investigate the effects of urbanization on riparian forest

104 communities. Identifying a specific filter as the cause of changes in species distribution is

105 challenging, since the effects of multiple filters often interact (Williams et al. 2008).

106 Because functional traits are an expression of the plant physiological responses to their

107 environment, their use can help to reveal the key filters acting on community assembly

108 and can thus give insight into the underlying mechanisms (Lavorel & Garnier 2002;

109 Mayfield et al. 2010). A trait-based approach also offers several powerful advantages

110 over approaches relying on species identity. First, traits are more directly connected to

111 and influenced by the environment because abiotic and biotic filters act on traits rather

112 than on species (Keddy 1992). Second, conclusions about traits give generality and

113 predictability, whereas species-based conclusions tend to provide highly peculiar rules

114 (McGill et al. 2006). Therefore, using a trait-based approach, we studied urban riparian

115 forests to investigate the independent and synergistic effects of natural and anthropogenic

116 filters on community composition, and to further unravel the role of these environmental

117 filters from that of spatial processes. Urban riparian forests are an ideal model system to

118 study such complex filters because of the strong structuring forces of urbanization and

119 flooding, and the expected role of the river as a dispersal corridor. More specifically, we

120 address the following questions: In the context of forests both fragmented by urbanization

121 and connected through the hydrological network, what are the spatial processes

122 structuring trait composition in riparian communities at different spatial scales? What are

123 the relative role of local conditions (including historical aspects), landscape context and

4 124 spatial processes in the community assembly of these forests? We hypothesize that (1)

125 while overland dispersal will be limited by fragmentation, river currents will connect

126 forests, more likely through hydrochorous dispersal, and create important directional

127 spatial structures; and (2) hydrological disturbances will exert the strongest selection

128 pressure, as numerous biological adaptations are needed for plant establishment and

129 survival in floodplains, while urbanization will have an indirect effect by disrupting the

130 natural flooding regime.

131 Methodology 132 Study area 133 The study was conducted in riparian forests of the Montréal urbanized region (Québec, 134 Canada; Fig. 1), Canada’s second most populous metropolitan region, with a population 135 of about 3.8 million inhabitants (www.statcan.gc.ca). The region was largely agricultural 136 prior to the 1960s, especially along the shores. Urban development then began slowly, 137 and from this period onwards farmland was converted to residential use. Currently, the 138 area is dominated by impervious surfaces (ca. 60% of the study area, of which 41% is 139 residential), but rural areas are still present (19%; Fig. 1, Appendix S1). 140 Site selection and sampling 141 Forests were selected using satellite imagery (1:5000) and field reconnaissance surveys, 142 according to size (area >0.4 ha) and vegetation structure (tree cover >80% and no turf). A 143 total of 95 forests met these criteria, and 57 of these were selected (Fig. 1, Appendix S2) 144 to represent a gradient of size, spatial position (from West to East), flood intensity (small 145 to large area flooded) and surrounding impervious surfaces (low to high). An effort has 146 been made to ensure that forest patches exist in all combinations of categories. 147 Within each selected forests, sampling plots were positioned along transects 148 perpendicular to the shoreline. Each forest contained two to four transects according to 149 the length of its shore to keep sampling effort proportionate to forest area (two transects 150 for sites between 50 and 300 m long (31 sites); three transects for sites between 301 and 151 500 m (11 sites); and four transects for sites > 500 m (15 sites)). On each transect a first 152 5 m × 20 m plot was located at the tree line along the shore (the length of the plot lying 153 along the shoreline), a second plot (10 m × 10 m) was established at 10 m from the first, 154 and a third one (10 m × 10 m) at 50 m from the second (total length of the 155 transect = 85 m). When the forest was narrower than 85 m (12 sites), only the two first 156 plots were sampled (transect length = 25 m). A total of 438 plots (from 4 to 12 per site) 157 were sampled in the 57 forests. 158 Field sampling was carried out from mid-July to mid-August 2013. Herbaceous 159 and shrub species cover in each plot was assessed using seven classes: <1%, 1–5%, 6– 160 25%, 26–50%, 51–75% and 76–100%. The diameter at breast height (DBH) of each tree 161 within each plot was estimated using the mid-range of four classes (10–20 cm; 21–30 cm; 162 31–40 cm; 41–50 cm) and measured directly for individuals > 50 cm. DBH values were

5 163 used to calculate the total tree basal area in each plot. 164 Environmental data sampling 165 Local variables 166 All variables (Table 1) were extracted from satellite images and maps of the region using 167 ArcGIS (version 10.1, ESRI, Redlands, CA, US). Each forest patch perimeter was 168 delineated, and the area and perimeter:area ratio were calculated. A digital elevation 169 model (1 m resolution; Montreal Metropolitan Community, unpubl.) was used to estimate 170 mean altitude and slope of each forest. The intensity of flooding was evaluated according 171 to the proportion of the forest that is usually inundated during spring. Where the 172 information was available, we estimated this area using the two-year flood limit given in 173 flood hazard maps (www.cehq.gouv.qc.ca/zones-inond/rapports-carto.htm). To complete 174 the missing information for remaining sites, we used the digital elevation model and raw 175 data of water elevation (Dubé 2006) to extrapolate the two-year flood limit. Finally, field 176 surveys during spring floods were conducted to evaluate the area of water present in each 177 site and to finalise the cartography of flood limit into three classes of flooding intensity 178 (from small to large area flooded; Table 1). 179 The recent history of the studied forests was characterized through interpretation 180 of aerial photos (1950–2010). Forest age was estimated as the time since the last major 181 disturbance. Past land use was categorized as either agricultural or other human uses, 182 which included residences (that no longer exist), and wasteland (industrial wastelands, 183 borrow pits or embankments). An index of forest area changes was compiled by 184 averaging the ratio of all previous forest areas for each decade (1950 to 2010) relative to 185 the current area. In general, forests were historically smaller than today, since most were 186 once cultivated or disturbed in whole or in part. 187 Landscape variables 188 To assess the effect of the surrounding urban matrix on forests, the landscape was 189 characterized in a 500 m buffer zone around each site (in accordance with Vallet et al. 190 2010) using 2012 land use maps (Montreal Metropolitan Community, unpubl.). For each 191 forest, we quantified the proportion of impervious surfaces (including transportation 192 infrastructures and residential, commercial and industrial areas), green spaces (all 193 vegetated areas including forests but excluding agricultural lands) and forests (areas 194 dominated by trees). To assess vegetation density and quality near the studied forests, a 195 Normalized Difference Vegetation Index (NDVI) was calculated using a SPOT satellite 196 image (20 m resolution; www.geobase.ca). All pixel values (ranging from -1 to +1) in the 197 buffer zone were averaged. Negative NDVI values indicate the absence or presence of 198 sparse unhealthy vegetation, while positive values correspond to the presence of healthy 199 vegetation. To include a more comprehensive measure of urbanization, we used the 200 Human Influence Index (HII; http://sedac.ciesin.columbia.edu/data/set/wildareas-v2-last- 201 of-the-wild-geographic), which takes into account multiple human factors presumed to 202 influence ecosystems (population density, roads, etc.). Using a map of the HII (1 km 203 resolution), each forest was assigned the average index value of the pixels in the buffer 204 zone. Finally, the edge-to-edge distance between each sampled forest and its nearest 205 neighbor (> 0.5 ha) was measured to assess the degree of isolation. 206 Functional traits 207 Each species was characterized according to eight functional traits (Table 2; see 208 Appendix S4 for raw data). Traits were selected to represent dispersal ability and

6 209 tolerance to hydrological and anthropogenic disturbances (based on reviews from Catford 210 & Jansson 2014 and Williams et al. 2015). The wetness index and the shade tolerance 211 (ecological tolerances sensu Violle et al. 2007) were used to serve as surrogates of many 212 other traits. The wetness index, which defines the estimated probability of a species to 213 occur in wetland habitats (Reed 1988), was selected to serve as a proxy of several traits 214 associated with capacity to cope with waterlogging (adventitious roots, aerenchyma, 215 anaerobic stress protein production, etc.) as they are often not available. Shade tolerance 216 is associated with a wide range of traits (photosynthetic rate, specific leaf area, relative 217 growth rate, etc.) and was used because light competition and interspecific differences in 218 shade tolerance are important determinants of forest structure and successional dynamics 219 (Valladeres & Niinemets 2008). Trait data were extracted from the primary literature and 220 published trait databases (Fitter & Peat 1994; Kleyer et al. 2008; Aubin et al. 2012 and 221 references therein; www.michiganflora.net). 222 Data analysis 223 Prior to analyses, species were separated into trees, shrubs and herbs, to form three 224 species-by-site matrices that were used individually. Separate analyses were performed 225 on each of the three plant forms because they are expected to respond differently; for 226 instance, longer lifespan allow woody species to persist long after a disturbance while 227 their structure makes them more resistant to disturbances. For each of the 57 forest 228 patches, species importance values (IV), a measure of the relative dominance of a species 229 in a community (Barbour 1987), were computed using all plots from a specific forest. For 230 trees, IVs were calculated as the mean of the relative density (number of individuals per 231 forest relative to the total number of individuals of all species), relative frequency 232 (number of plots per forest in which a species occurred relative to the total number of 233 occurrences of all species) and relative dominance (basal area coverage per forest relative 234 to the total basal area of all species). For herbs and shrubs, IVs were calculated as the 235 mean of the relative frequency and the relative dominance (cover of a species per forest 236 relative to the total cover of all species). 237 Mean functional traits for community 238 To assess the effects of environmental filters on functional composition, community 239 weighted means of traits (CWM) were used in subsequent analyses. CWM describe plant 240 communities in terms of their trait composition rather than their species composition 241 (e.g., Garnier et al. 2007; Lavorel et al. 2008). Compiled in a site-by-trait matrix, CWM 242 were obtained by the multiplication of the species-by-site matrix and the standardized 243 species-by-trait matrix (trait standardization ensures equal weight for all traits, regardless 244 of their units). To ensure biological relevance, traits with less than 10% of occurrence for 245 a specific plant form were discarded; hence, bird dispersal was not used for herbs and 246 epizoochory was not used for trees and shrubs. For continuous traits, CWM was the 247 average value of a trait for all species in the community, weighted by their relative 248 abundance. For ordinal and binary traits, CWM was the relative abundance of species for 249 each trait state. 250 Modeling spatial structure 251 Before analysing the independent and synergistic effects of natural and anthropogenic 252 filters on community composition, spatial patterns must be modeled to identify which 253 structures will be included in further analyses. To assess the relative importance of 254 nondirectional and directional spatial processes in trait distribution, we compared two

7 255 eigenvector-based spatial methods: distance-based Moran's eigenvector maps (dbMEM; 256 Dray et al. 2006) and asymmetric eigenvector maps (AEM; Blanchet et al. 2008a). MEM 257 analyses are meant to model symmetric spatial structures, which could emerge from 258 nondirectional plant dispersal (e.g., aggregation). In contrast, AEM are explicitly 259 designed to model directional patterns, which are expected in riparian ecosystems. 260 Contrary to AEM, MEM analyses must be carried out on detrended response variables 261 (Borcard & Legendre 2002). The significant linear trend in the direction of the 262 hypothesized spatial process (i.e., the broad-scale gradient created by the rivers) was 263 therefore removed prior to MEM calculation. 264 AEM and MEM eigenfunctions were computed using a connectivity matrix which 265 summarizes the connections between the studied forests. The connection diagram 266 (Appendix S5) was created based on interpretation of satellite imagery showing flow 267 direction. Both rivers, as well as prevailing summer winds, flow from west to east and 268 their flow is controlled by various dams, particularly during spring floods. Each river was 269 treated as independent from the other. Forests located on the north shore of the Lake of 270 Two Mountains were linked to those of des Mille Iles river, while forests on the south 271 shore to those of des Prairies river. To properly model the distinction between both rivers, 272 the AEM method of Blanchet et al. (2008a, 2011) was adapted (Appendix S5). 273 For MEM, the connectivity matrix was weighted using between-site overland 274 distances. For AEM, distances corresponded to watercourse distances. A measure of river 275 fragmentation associated with the presence of bridges and dams was also considered for 276 AEM. The weighting functions were: (1) no weight (binary connection); (2) a linear

277 function of the distances (f1 (dij) = 1 − (dij/max(dij))); (3) a concave-down function of the 2 278 distances (f2 (dij) = 1 − (dij/max(dij)) ); and (4) river fragmentation (f = 1 - (0.2*nb of dams + 279 0.1*nb of bridges)). The weight or combination of weights leading to the highest 280 explained variation in the functional composition data was retained for further analyses 281 (i.e., a concave-down function of distances and river fragmentation for trees, a linear 282 function of distances and river fragmentation for herbs, and river fragmentation for 283 shrubs; Appendix S6). 284 Only spatial functions modeling positive spatial correlation were used (selected 285 using Moran’s I coefficients of spatial correlation). For each model, spatial variables 286 were selected using a forward selection procedure (9999 permutations of residuals) based 287 on two stopping criteria, the alpha significance level and the adjusted R2 of the global 288 model including all explanatory variables of each subset, as proposed by Blanchet et al. 289 (2008b). AEM and MEM models were then compared through variation partitioning to 290 assess the directional and non-directional components of variation. Since AEM always 291 yield a larger explained variation (see Results), they were used to model spatial processes 292 in the subsequent analyses. Although eigenvector analysis does not model processes 293 directly, AEM vectors were mapped to visualize the spatial variations of community 294 composition and to make some hypothesis about their origins. 295 Partitioning environmental and spatial variation 296 The relative and unique importance of landscape context, local conditions and spatial 297 processes in the variation of plant trait distribution was assessed by partitioning the 298 variation explained by each subset using redundancy analyses (RDA; Borcard et al. 299 1992). First, to linearize the relationship between explanatory and response variables in 300 the RDA models, explanatory variables were transformed when necessary

8 301 (Appendix S7), after which they were standardized. Then, because we used mean trait 302 values and not a species abundance matrix, the RDA were calculated using the Euclidean 303 distance (Kleyer et al. 2012). To obtain parsimonious RDA models for each of the three 304 explanatory subsets, only variables selected using a forward selection procedure (9999 305 permutations of residuals) based on the two stopping criteria of Blanchet et al. (2008b) 306 were retained (Appendix S7). Finally, variation partitioning analyses were computed 307 from the parsimonious RDA for the three plant forms, and the significance of each 308 unique fraction was obtained by 9999 permutations. In the analysis on shrubs, two forests 309 were excluded, as their trait composition was highly divergent from that of other forests 310 due to the concomitant dominance of an uncommon wetland species (Cephalanthus 311 occidentalis) and the absence of an invasive one (Rhamnus cathartica). 312 All statistical analyses were performed in the R environment (version 3.0.1; R 313 Core Team, Vienna, AU). Canonical analyses, variation partitioning, and tests of 314 significance of the fractions were computed using the ‘‘vegan’’ package. AEM 315 eigenfunctions were created using the ‘‘AEM’’ package, while MEM were constructed 316 using the ‘‘PCNM’’ package. Forward selection was computed using the ‘‘packfor’’ 317 package. 318 Results 319 Directional vs non-directional spatial patterns 320 For tree and herb strata, only directional patterns were detected, explaining respectively 321 24% and 37% of the variation in functional composition (Fig. 2). For shrubs, both 322 directional and non-directional significant patterns were found, explaining respectively 323 26% and 10% of the variation. However, the pure fraction of directional patterns 324 prevailed (19%), while the non-directional patterns were nearly all confounded with the 325 directional patterns. 326 Scales of spatial variation 327 Spatial structures were observed at broad, medium (except for herbs) and fine scales, the 328 broad patterns usually being most important, as shown by their globally larger adjusted 329 R2 (Appendix S8). Plant trait variation was most strongly correlated to the spatial pattern 330 described by the AEM6 (trees), AEM7 (shrubs) and AEM4 (herbs) functions, which all 331 represented the distinction between the two rivers. The first AEM function (AEM1) of 332 the three plant form models represented a strong gradient along des Prairies river, while 333 the AEM6 of the shrub model represented a broad-scale gradient along des Mille Iles 334 river. While spatial patterns, from broad to fine-scale, were exclusively found along des 335 Prairies river for trees, they contributed equally to shrub trait variation along both rivers. 336 For herbs, trait variations were identified primarily along des Prairies river (AEM1, 3 and 337 13), while fine-scale patterns were minor along des Mille Iles river (AEM15). 338 Partitioning trait variation between environmental filtering and spatial processes 339 Local, landscape and spatial variables together explained about 40% of changes in the 340 functional composition of trees and shrubs and about 51% of herbs (Fig. 3 a–c). Trait 341 composition was primarily structured by local conditions (trees: 34%; shrubs: 23%; 342 herbs: 41%), and mostly by seasonal flood intensity (flood), as this variable alone had the 343 highest explanatory power among all variables (trees: 22%; shrubs: 12%; herbs: 28%; 344 Appendix S7). Most of the variance explained by the local subset was shared with 345 landscape, and particularly with spatial variables. The fraction shared by the three subsets 346 had a strikingly high adjusted R2 compared with individual fractions, suggesting that

9 347 local and landscape subsets together presented a strong spatial structure. Hence, although 348 landscape context alone was less important in structuring functional composition, its 349 influence was exerted through change in local conditions, especially for herbs. Still, the 350 significance of the pure directional spatial fraction indicates that a directional process, 351 other than the inherent spatial structure of the environment, acts on trait distribution. 352 These pure spatial patterns were as important as local conditions (about 8%) in 353 structuring shrub trait composition. 354 The first RDA axis of each plant form represents a flooding intensity gradient that 355 was relatively paralleled by an urban-to-rural gradient (Fig. 3 d–f). Forests associated 356 with rural areas (right side of the plots) were mostly associated with high vegetation 357 cover (represented by forest, green or NDVI variables) and were usually subjected to 358 more severe flooding than urban forests. On the other end of the gradient (left side of the 359 plots), urban forests were characterized by steep slopes, a high perimeter:area ratio 360 (PAR), as well as a large proportion of impervious surfaces and a frequent occurrence of 361 bridges in their vicinity. Along this gradient generally ranging from flooded rural forests 362 to drier urban forests, AEM1 (modeling a linear gradient along des Prairies river for all 363 plant groups) and AEM 6, 7 and 4 variables (modeling the distinction between both rivers 364 for trees, shrubs and herbs respectively) were opposed to flooding intensity indicating 365 that floods are more intense on des Mille Iles river and westward of des Prairies river. 366 As expected, flood intensity favored hydrochorous (hydro) wetland species 367 (wetness) for all plant forms. In highly flooded forests, often in less urbanized areas, trees 368 were taller (height) and mostly bird-dispersed, but also animal- (endozoo) and human- 369 dispersed (anthropo). On the contrary, in drier forests, often in urbanized areas, shrubs 370 were taller (height) and principally animal- or human-dispersed. These later forests were 371 dominated by annual herbs (lifespan) bearing wind-dispersed (anemo) or heavy seeds 372 (seedWg), which were, as for shrubs, principally animal- (endozoo and epizoo) and 373 human-dispersed. On the other hand, vegetative reproduction (veg) was favored in 374 flooded forests. 375 Secondary to the urban and flooding gradients, a successional gradient was 376 observed in the three ordinations. Following the historical variables related to human 377 activities (histo.human, which describes historically disturbed forests, and histo.area, 378 which describes historically large forests), this gradient ranged from recently disturbed to 379 old ‘pristine’ forests. This gradient contributed to the variance in the second axis and 380 appeared perpendicular to the flood gradient (parallel to the second axis) for trees, 381 crosswise for shrubs and almost parallel for herbs (almost parallel to the first axis). This 382 decreasing distinctiveness of the successional gradient among the three plant forms was 383 also depicted by the decreasing variance explained by the second axis. Along this 384 gradient, young, recently disturbed forests were dominated by human-dispersed 385 (anthropo) trees, shrubs and herbs, wind-dispersed (anemo) trees, and shrubs with a long- 386 lived seed bank (seedBk). In contrast, old ‘pristine’ forests were dominated by shade- 387 tolerant (shade) trees, shrubs and herbs, heavy-seeded (seedWg), animal-dispersed 388 (endozoo) and long-lived seed banking trees, and perennial (lifespan) herbs. 389 Discussion 390 The interplay between local, landscape and spatial filters 391 In this study, we showed that under severe natural disturbance regimes, the effect of 392 natural filters outweighed the negative effects of urban filters, since seasonal flood

10 393 intensity was the most important variable in explaining trait composition of riparian plant 394 communities. Nonetheless, the antagonism between flood and urbanization variables in 395 the ordinations, which gave rise to strong spatial structures responsible for the shared 396 fraction between the local, landscape and spatial subsets, indicated that urbanization also 397 played a significant role by indirectly influencing traits through alteration of the 398 hydrological regime. Urbanization also likely fostered a successional gradient in forests 399 with reduced flooding, since species traits, notably for trees, differed between young 400 disturbed and old ‘pristine’ forests. This result suggests that as flood intensity decreases, 401 its filtering effect is relaxed and other filters associated with succession induced by 402 human disturbances become more important in community assembly. The effect of 403 urbanization on riparian communities thus appeared to be twofold: a principal effect 404 through hydrological alterations and a secondary effect through initiation of succession. 405 Spatial structure and dispersal processes 406 Although environmental filtering, driven directly by local conditions and indirectly by 407 landscape modification, was predominant in riparian community assembly, dispersal 408 processes through the river network still influenced community structure. The role of 409 dispersal processes can be inferred from the pure spatial fraction and from the fact that 410 water dispersal by river (AEM) was more important than overland dispersal (MEM). The 411 importance of river connectivity was also illustrated by the fact that weights associated 412 with river fragmentation by dams and bridges were included in all AEM models. The 413 inclusion of these weights indicates that human constructions can act as barriers to the 414 movement of propagules of the upstream species pool, impair seed deposition on river 415 shores and severely alter hydrologic conditions, which may affect potential composition 416 of downstream communities (Jansson et al. 2000; Merritt & Wohl 2006; Merritt et al. 417 2010). 418 Plants dispersed predominantly through the river network, but the significant pure 419 spatial fraction indicated that they were still subjected to dispersal limitation processes 420 (Peres-Neto & Legendre 2010). This latter fraction was higher for herbs than for trees, 421 likely because herbs are generally weaker dispersers than trees, due to their low seed 422 releasing height (Garnier et al. 2013). In contrast to trees and herbs, shrubs were not 423 limited by distance (no function of distance was included in their AEM model) and were 424 influenced by non-directional spatial processes, modeled by MEM. While these results 425 suggest good dispersal potential, shrubs nonetheless had a relatively high pure spatial 426 fraction likely induced by other unstudied processes. For instance, as a consequence of 427 human preference, ornamental shrubs with fleshy fruits are overrepresented in urban 428 floras (Burton et al. 2009; Knapp et al. 2009; Huang et al. 2013). The influence of planted 429 ornamental species could have created an asymmetrical spatial pattern, similar to the one 430 expected to arise from river dispersal, included in the pure spatial fraction. Also, the 431 prevalence of human- and animal-dispersal, especially by birds, which have long-distance 432 potential (Knapp et al. 2009), could contribute to the ability of shrubs to establish in 433 isolated forests through overland dispersal. 434 Shift of plant traits along an urban–rural gradient 435 The main trait pattern found was a shift in species distribution from wetland 436 waterdispersed species dominant in flooded forests to upland species in drier ones often 437 located in more urbanized areas. This large-scale shift has been previously attributed 438 either to flow regulation that lessens the river influence on riparian habitats (Stromberg et

11 439 al. 2007; Catford et al. 2011) causing a terrestrialization process (Ward & Stanford 1995), 440 or to regional alteration of the flood regimes that induces a low water table in urban areas 441 (White & Greer 2006; Burton et al. 2009; Sung et al. 2011) reported as hydrologic 442 drought by Groffman et al. (2003). In light of our results, we suggest that urbanization, 443 and not flow regulation alone, leads to a general trend toward terrestrialization. This 444 process was noticeable mainly at the landscape-scale, but was also observed at a more 445 local-scale, since upland shrubs and lianas (the latter represented by the height and 446 endozoochory traits) were more abundant in forests adjacent to a bridge than in other 447 forests. The rock fill of adjacent banks during bridge construction, inherent to the process 448 of urbanization, likely disconnected the river from the forest (Blanton & Marcus 2009) 449 facilitating this local terrestrialization. However, because the wetness index was strongly 450 correlated with hydrochory, it is difficult to truly disentangle the filtering effect of local 451 flooding that act on physiological traits that provide flood tolerance (which we believe to 452 prevail) from those that allow dispersal by water. 453 Shift of plant traits along a human-induced successional gradient 454 In the studied sites, dispersal processes seemed to prevail in early-successional forests 455 (historically small and human-disturbed forests), as indicated by the abundance of highly 456 mobile tree species (dispersed by wind and humans). In contrast, late-successional 457 communities, represented by historically large forests located in a vegetated matrix, the 458 competition filter was likely more important and favored shade-tolerant tree species that 459 disperse locally (e.g., large nuts). This phenomenon suggests the presence of a 460 successional gradient induced by historical local human activities. The successional 461 gradient was more pronounced for trees than for shrubs, and almost indistinguishable for 462 herbs, suggesting that historical disturbances leave an ecological footprint (disturbance 463 legacy; Flinn & Vellend 2005; LaPaix & Freedman 2010; Johnson et al. 2014) on species 464 functional composition that likely fades over time proportionally to species longevity 465 (Kuussaari et al. 2009). Indeed, due to their high turnover rate, herbs show not only 466 faster, but also stronger responses to urban disturbances than woody species (Burton et al. 467 2009; Huang et al. 2013; Ramalho et al. 2014), which can persist long after their habitat 468 has been degraded. 469 In the study area, the most disturbed forest patches were mainly industrial 470 wastelands, borrow pits or embankments a few years ago. We suggest that their 471 composition, characterized by a high representation of anthropochores, could be the 472 result of a source-sink process driven by in situ and ex situ plantation of cultivated 473 species. In urbanized regions, cultivated plants are overrepresented in the species pool 474 (Williams et al. 2008) and act as a massive source of propagules that wields a high 475 pressure on nearby vacant habitats, which could be viewed as sinks. This source-sink 476 process mirrors findings in other studies, which reported that the types of surrounding 477 habitats exert a strong influence on species composition during succession in human- 478 disturbed habitats (Prach et al. 2001; Westermann et al. 2011). This hypothesis is further 479 reinforced by the shrub trait patterns identified here, where the observed successional 480 gradient was exclusively related to a single species, Rhamnus cathartica, a tall invasive 481 shrub from Eurasia which dominated the mid-story of most of the studied forests (68% of 482 the patch studied). Brought to North America as ornamental and agricultural hedgerows 483 (Knight et al. 2007), R. cathartica may have spread from these human-planted sources

12 484 and invaded the studied sites when they were still largely open and non-forested, while its 485 wide habitat tolerance (Knight et al. 2007) may have contributed to its success over time. 486 Conclusion 487 Overall, our results highlights that local conditions, especially flood intensity, exerted an 488 overriding selection pressure on riparian communities. The effect of the landscape was 489 principally indirect, acting on trait patterns through alteration of hydrological 490 disturbances. In addition to environmental filtering, dispersal along rivers was also an 491 important process structuring riparian forests. Hence, urbanization seems to alter riparian 492 community composition through changes in community assembly rules, primarily 493 through changes to both environmental and dispersal processes. While disturbances are 494 an inherent feature of riparian ecosystems, human-induced disturbances have long lasting 495 consequences, and impair natural ecological processes. Management strategies in urban 496 areas should therefore aim at restoring or preserving the natural processes critical for 497 maintaining the functional composition of floodplain vegetation and limit 498 terrestrialization. For instance, a first step is to avoid or remove any constructions or 499 modifications that disconnect the riparian habitat from the river (e.g., dikes, 500 embankments) along riverbanks. Identification of functional trait shifts can be useful for 501 identifying forests that have high conservation value, or degraded forests in need of 502 ecological restoration (Mayfield et al. 2006). Finally, because of the importance of water 503 dispersal, maintaining upstream-downstream connectivity may be effective for preserving 504 biodiversity in urban riparian forests. As dams and bridges are a necessary evil and thus 505 cannot be avoided, managers should focus on maintaining as much as possible a 506 continuous strip of vegetation along rivers. 507 Acknowledgements 508 This research received financial support from the Natural Sciences and Engineering 509 Research Council of Canada (scholarship to MH Brice, Discovery grant to S. Pellerin: 510 RGPIN-2014-05367 and M. Poulin RGPIN-2014-05663), the Fonds de recherche Nature 511 et technologies and the Quebec Centre for Biodiversity Science. We are grateful to all 512 municipalities and private landowners who allowed us to work in their forests. Our 513 thanks to A. Langlois, A. Richer, J. S. Mignot, S. Pasquet and Y. Tendland for field and 514 laboratory assistance, P. Legendre for statistical advices, K. Grislis for linguistic revision 515 and two anonymous reviewers made helpful comments on earlier drafts of the 516 manuscript. 517 References 518 Aubin, I., Messier, C., Gachet, S., Lawrence, K., McKenney, D., Arseneault, A., Bell,

519 W., De Grandpré, L., Shipley, B., Ricard, J.-P. & Munson, A.D. 2012. TOPIC -

520 Traits of Plants in Canada. Natural Resources Canada, Canadian Forest Service,

521 Sault Ste. Marie, Ontario, Canada.

522 Barbour, M.G., Burk, J.H. & Pitts, W.D. 1987. Terrestrial Plant Ecology. Benjamin/

523 Cummings, Menlo Park, CA.

13 524 Blanchet, F.G., Legendre, P. & Borcard, D. 2008a. Forward selection of explanatory

525 variables. Ecology 89, 2623–2632.

526 Blanchet F.G., Legendre P. & Borcard D. 2008b Modelling directional spatial processes

527 in ecological data. Ecological Modelling 215: 325–336.

528 Blanchet, F.G., Legendre, P., Maranger, R., Monti, D. & Pepin, P. 2011. Modelling the

529 effect of directional spatial ecological processes at different scales. Oecologia 166:

530 357–68.

531 Blanton, P. & Marcus, W.A. 2009. Railroads, roads and lateral disconnection in the river

532 landscapes of the continental United States. Geomorphology 112: 212–227.

533 Borcard, D. & Legendre, P. 2002. All-scale spatial analysis of ecological data by means

534 of principal coordinates of neighbour matrices. Ecological Modelling 153: 51–68.

535 Borcard, D., Legendre, P. & Drapeau, P. 1992. Partialling out the spatial component of

536 ecological variation. Ecology 73: 1045–1055.

537 Burton, M.L., Samuelson, L.J. & Mackenzie, M.D. 2009. Riparian woody plant traits

538 across an urban–rural land use gradient and implications for watershed function with

539 urbanization. Landscape and Urban Planning 90: 42–55.

540 Catford J.A., Downes, B.J., Gippel, C.J. & Vesk, P.A. 2011. Flow regulation reduces

541 native plant cover and facilitates exotic invasion in riparian wetlands. Journal of

542 Applied Ecology 48: 432–442.

543 Catford, J.A. & Jansson, R. 2014. Drowned, buried and carried away: effects of plant

544 traits on the distribution of native and alien species in riparian ecosystems. New

545 Phytologist 204: 19–36.

14 546 Dubé, S. 2006. Révision des cotes de crues, Rivière des Prairies, Villes de Charlemagne,

547 de Laval, de Montréal, de Repentigny et de Terrebonne. Centre d’expertise hydrique

548 du Québec [rapport CEHQ 13-002], Québec, Canada.

549 Dray, S., Legendre, P. & Peres-Neto, P.R. 2006. Spatial modelling: a comprehensive

550 framework for principal coordinate analysis of neighbour matrices (PCNM).

551 Ecological Modelling 196: 483–493.

552 Fitter, A.H. & Peat, H.J. 1994. The Ecological Flora Database. Journal of Ecology 82:

553 415-425.

554 Flinn, K.M. & Vellend, M. 2005. Recovery of forest plant communities in post-

555 agricultural landscapes. Frontiers in Ecology and the Environment 3: 243–250.

556 Garnier, E., Lavorel, S., Ansquer, P., Castro, H., Cruz, P., Dolezal, J., Eriksson, O.,

557 Fortunel, C., Freitas, H., (…) & Zarovali, M.P. 2007. Assessing the effects of land-

558 use change on plant traits, communities and ecosystem functioning in grasslands: A

559 standardized methodology and lessons from an application to 11 European sites.

560 Annals of Botany 99: 967–985.

561 Garnier, E., Lavorel, S., Poorter, H., Jaureguiberry, P., Cornwell, W.K., Craine, J.M.,

562 Gurvich, D.E., Urcelay, C., Veneklaas, E.J., (…) & Cornelissen, J.H.C. 2013. New

563 handbook for standardised measurement of plant functional traits worldwide.

564 Australian Journal of Botany 61: 167–234.

565 Godefroid, S. & Koedam, N. 2007. Urban plant species patterns are highly driven by

566 density and function of built-up areas. Landscape Ecology 22: 1227–1239.

15 567 Groffman, P.M., Bain, D.J., Band, L.E., Belt, K.T., Brush, G.S., Grove, J.M., Pouyat, R.

568 V, Yesilonis, I.C. & Zipperer, W.C. 2003. Down by the riverside: urban riparian

569 ecology. Frontiers in Ecology and the Environment 1: 315–321.

570 Gurnell, A.M., Boitsidis, A.J., Thompson, K. & Clifford, N.J. 2006. Seed bank, seed

571 dispersal and vegetation cover: colonization along a newly-created river channel.

572 Journal of Vegetation Science 17: 665–674.

573 Huang, L., Chen, H., Ren, H., Wang, J. & Guo, Q. 2013. Effect of urbanization on the

574 structure and functional traits of remnant subtropical evergreen broad-leaved forests

575 in South China. Environmental monitoring and assessment 185: 5003–5018.

576 Jansson, R., Nilsson, C. & Renöfält, B.M. 2000. Fragmentation of riparian floras in rivers

577 with multiple dams. Ecology 81: 899–903.

578 Johansson, M., Nilsson, C. & Nilsson, E. 1996. Do rivers function as corridors for plant

579 dispersal? Journal of Vegetation Science 7: 593–598.

580 Johnson, A.L., Tauzer, E.C., & Swan, C.M. 2014. Human legacies differentially organize

581 functional and phylogenetic diversity of urban herbaceous plant communities at

582 multiple spatial scales. Applied Vegetation Science 18: 513–527.

583 Karrenberg, S., Edwards, P.J. & Kollmann, J. 2002 The life history of Salicaceae living

584 in the active zone of floodplains. Freshwater Biology 47: 733–748.

585 Keddy, P.A. 1992. Assembly and response rules: two goals for predictive community

586 ecology. Journal of Vegetation Science 3: 157–164.

587 Kleyer, M., Bekker, R.M., Knevel, I.C., Bakker, J.P., Thompson, K., Sonnenschein, M.,

588 Poschlod, P., Van Groenendael, J.M., Klimeš, L., (…) & Peco, B. 2008. The LEDA

16 589 Traitbase: a database of life-history traits of the Northwest European flora. Journal

590 of Ecology 96: 1266–1274.

591 Kleyer, M., Dray, S., Bello, F., Lepš, J., Pakeman, R.J., Strauss, B., Thuiller, W. &

592 Lavorel, S. 2012. Assessing species and community functional responses to

593 environmental gradients: which multivariate methods? Journal of Vegetation

594 Science 23: 805–821.

595 Knapp, S., Kühn, I., Bakker, J.P., Kleyer, M., Klotz, S., Ozinga, W.A., Poschlod, P.,

596 Thompson, K., Thuiller, W. & Römermann, C. 2009. How species traits and affinity

597 to urban land use control large-scale species frequency. Diversity and Distributions

598 15: 533–546.

599 Knapp, S., Kühn, I., Wittig, R., Ozinga, W.A., Poschlod, P. & Klotz, S. 2008.

600 Urbanization causes shifts in species’ trait state frequencies. Preslia 80: 375–388.

601 Knight, K.S., Kurylo, J.S., Endress, A.G., Stewart, J.R. & Reich, P.B. 2007. Ecology and

602 ecosystem impacts of common buckthorn (Rhamnus cathartica): a review.

603 Biological Invasions 9: 925–937.

604 Kuussaari, M., Bommarco, R., Heikkinen, R.K., Helm, A., Krauss, J., Lindborg, R.,

605 Ockinger, E., Pärtel, M., Pino, J., (…) & Steffan-Dewenter, I. 2009. Extinction debt:

606 a challenge for biodiversity conservation. Trends in Ecology and Evolution 24: 564–

607 571.

608 LaPaix, R. & Freedman, B. 2010. Vegetation structure and composition within urban

609 parks of Halifax regional municipality, Nova Scotia, Canada. Landscape and Urban

610 Planning 98: 124–135.

17 611 Lavorel, S. & Garnier, E. 2002. Predicting changes in community composition and

612 ecosystem functioning from plant traits: revisiting the Holy Grail. Functional

613 Ecology 16: 545–556.

614 Lavorel, S., Grigulis, K., McIntyre, S., Williams, N.S.G., Garden, D., Dorrough, J.,

615 Berman, S., Quétier, F., Thébault, A. & Bonis, A. 2008. Assessing functional

616 diversity in the field - methodology matters! Functional Ecology 22: 134–147.

617 Lewis, D.B., Kaye, J.P., Gries, C., Kinzig, A.P. & Redman, C.L. 2006. Agrarian legacy

618 in soil nutrient pools of urbanizing arid lands. Global Change Biology 12: 703–709.

619 Logue, J.B., Mouquet, N., Peter, H. & Hillebrand, H. 2011. Empirical approaches to

620 metacommunities: a review and comparison with theory. Trends in Ecology and

621 Evolution 26: 482–491.

622 Lytle, D.A & Poff, N.L. 2004. Adaptation to natural flow regimes. Trends in Ecology

623 and Evolution 19: 94–100.

624 Mayfield, M.M., Ackerly, D. & Daily, G.C. 2006. The diversity and conservation of plant

625 reproductive and dispersal functional traits in human-dominated tropical landscapes.

626 Journal of Ecology 94: 522–536.

627 Mayfield, M.M., Bonser, S.P., Morgan, J.W., Aubin, I., McNamara, S. & Vesk, P.A.

628 2010. What does species richness tell us about functional trait diversity? Predictions

629 and evidence for responses of species and functional trait diversity to land-use

630 change. Global Ecology and Biogeography 19: 423–431.

631 McGill, B.J., Enquist, B.J., Weiher, E., & Westoby, M. 2006. Rebuilding community

632 ecology from functional traits. Trends in ecology & evolution 21: 178–185.

18 633 Merritt, D., Nilsson, C. & Jansson, R. 2010. Consequences of propagule dispersal and

634 river fragmentation for riparian plant community diversity and turnover. Ecological

635 Monographs 80: 609–626.

636 Merritt, D.M. & Wohl, E.E. 2006. Plant dispersal along rivers fragmented by dams. River

637 Research and Applications 22: 1–26.

638 Moggridge, H.L., Gurnell, A.M. & Mountford, J.O. 2009. Propagule input, transport and

639 deposition in riparian environments: the importance of connectivity for diversity.

640 Journal of Vegetation Science 20: 465–474.

641 Naiman, R.J., Décamps, H. & McClain, M.E. 2005. Riparia: Ecology, conservation, and

642 management of streamside communities. Elsevier Academic Press, London, UK.

643 Nilsson, C., Brown, R.L., Jansson, R. & Merritt, D.M. 2010. The role of hydrochory in

644 structuring riparian and wetland vegetation. Biological reviews of the Cambridge

645 Philosophical Society 85: 837–858.

646 Parent, C., Capelli, N., Berger, A., Crèvecoeur, M. & Dat, J.F. 2008. An overview of

647 plant responses to soil waterlogging. Plant Stress 2: 20–27.

648 Peres-Neto, P.R. & Legendre, P. 2010. Estimating and controlling for spatial structure in

649 the study of ecological communities. Global Ecology and Biogeography 19: 174–

650 184.

651 Prach, K., Pyšek, P. & Bastl, M. 2001. Spontaneous vegetation succession in human-

652 disturbed habitats: a pattern across seres. Applied Vegetation Science 4: 83–88.

653 Ramalho, C.E., Laliberté, E., Poot, P. & Hobbs, R.J. 2014. Complex effects of

654 fragmentation on remnant woodland plant communities of a rapidly urbanizing

655 biodiversity hotspot. Ecology 95: 2466–2478.

19 656 Reed, P.B. 1988. National list of plant species that occur in wetlands: national summary.

657 U.S. Department of the Interior, Fish and Wildlife Service [report 88-24],

658 Washington, DC, US.

659 Stromberg, J.C., Lite, S.J., Marler, R., Paradzick, C., Shafroth, P.B., Shorrock, D., White,

660 J.M. & White, M.S. 2007. Altered stream-flow regimes and invasive plant species:

661 the Tamarix case. Global Ecology and Biogeography 16: 381–393.

662 Sung, C.Y., Li, M.-H., Rogers, G.O., Volder, A. & Wang, Z. 2011. Investigating alien

663 plant invasion in urban riparian forests in a hot and semi-arid region. Landscape and

664 Urban Planning 100: 278–286.

665 Valladares, F. & Niinemets, Ü. 2008. Shade tolerance, a key plant feature of complex

666 nature and consequences. Annual Review of Ecology, Evolution, and Systematics,

667 39: 237–257.

668 Vallet, J., Daniel, H., Beaujouan, V., Rozé, F. & Pavoine, S. 2010. Using biological traits

669 to assess how urbanization filters plant species of small woodlands. Applied

670 Vegetation Science 13: 412–424.

671 Violle, C., Navas, M.-L., Vile D., Kazakou, E., Fortunel, C., Hummel, I. & Garnier, E.

672 2007. Let the concept of trait be functional! Oikos 116: 882–892.

673 Ward, J.V & Stanford, J.A. 1995. Ecological connectivity in alluvial river ecosystems

674 and its disruption by flow regulation. Regulated Rivers: Research and Management

675 11: 105–119.

676 Westermann, J.R., von der Lippe, M. & Kowarik, I. 2011. Seed traits, landscape and

677 environmental parameters as predictors of species occurrence in fragmented urban

678 railway habitats. Basic and Applied Ecology 12: 29–37.

20 679 White, M.D. & Greer, K.A. 2006. The effects of watershed urbanization on the stream

680 hydrology and riparian vegetation of Los Peñasquitos Creek, California. Landscape

681 and Urban Planning 74: 125–138.

682 Williams, N.S.G., Hahs, A.K. & Vesk, P.A. 2015. Urbanisation, plant traits and the

683 composition of urban floras. Perspectives in Plant Ecology, Evolution and

684 Systematics 17: 78–86.

685 Williams, N.S.G., Schwartz, M.W., Vesk, P.A., McCarthy, M.A., Hahs, A.K., Clemants,

686 S.E., Corlett, R.T., Duncan, R.P., Norton, B.A., Thompson, K. & McDonnell, M.J.

687 2008. A conceptual framework for predicting the effects of urban environments on

688 floras. Journal of Ecology 97: 4–9.

689 Wulf, M. & Kolk, J. 2014. Plant species richness of very small forests related to patch

690 configuration, quality, heterogeneity and history. Journal of Vegetation Science 25:

691 1267–1277.

692 Supporting Information 693 Additional Supporting Information may be found in the online version of this article: 694 Appendix S1 Map of residential density and agricultural area 695 Appendix S2 Criteria for patches selection 696 Appendix S3 Correlation coefficients between explanatory variables 697 Appendix S4 Species functional trait matrix for trees, shrubs and herbs 698 Appendix S5 Connectivity diagram and steps to create the connectivity matrix 699 Appendix S6 Comparison of weights for AEM analyses 700 Appendix S7 Selected explanatory variables and their transformation 701 Appendix S8 Maps of selected AEM functions.

21 702 Table 1. Description of all explanatory variables characterizing the 57 riparian forests of

703 the Montréal urbanized region (see Appendix S3 for correlations between variables).

Code Explanatory variables Units Local conditions area Forest area m2 PAR Perimeter: area ratio ratio flood Seasonal flood intensity 1: flooded area ≤ 20% 2: flooded area ]20, 50]% 3: flooded area > 50% altitude Forest mean altitude m slope Forest mean slope 1: slope ≤ 2 ° 2: slope ]2, 6] ° 3: slope > 6 ° age Time since the last major disturbance Years histo.area Mean proportion of historical area compared to current area % histo.agricul Historical agricultural land use 0-1 t histo.human Historical human land use (residential or wasteland) 0-1 Landscape context isolation Edge-to-edge distance to the nearest forest m HII Human Influence Index 0 −72 bridge Presence of a bridge in the vicinity 0-1 impervious Proportion of impervious surfaces in a 500-m buffer zone % rural Proportion of rural areas in a 500-m buffer zone % forest Proportion of forest cover in a 500-m buffer zone % green Proportion of vegetation in a 500-m buffer zone % NDVI Mean NDVI value in a 500-m buffer zone -1 −1 Spatial process XY Geographic coordinates of forest centroid used for spatial UTM eigenfunction analyses 704

22 705 Table 2. Description of the selected functional traits used in analyses. Functional traits Code Units/Scale Variable Maximum height height cm Quantitative Life cycle (herbs only) lifespan 0: annual Nominal 1: perennial Shade tolerance shade 1: intolerant Ordinal 2: mid-tolerant 3: tolerant Wetness index wetness 1: upland (almost never in wetlands) Ordinal 2: facultative upland (occasionally in wetlands) 3: facultative (equally likely in wetlands and uplands) 4: facultative wetland (usually in wetlands) 5: obligate wetland (almost always in wetlands) Seed weight seedWg ln(mg) Quantitative Seed bank seedBk 1: transient Ordinal 2: short-term persistent 3: long-term persistent Vegetative veg 0: not possible Binary reproduction 1: possible Dispersal vector endozoo endozoochore (internal animal dispersal) Binary bird bird dispersal epizoo epizoochore (external animal dispersal) hydro hydrochore (water dispersal) anemo anemochore (wind dispersal) anthropo anthropochore (human dispersal) other other (include unassisted, myrmechore, explosive)

23 707

708 Fig. 1. Location map of the 57 riparian forests (in black) sampled in the Montréal 709 urbanized region.

24 711 Fig. 2. Variation partitioning of trait composition (CWM) explained by the subsets of 712 selected AEM and MEM variables. Percentages in the partitioning are R2adj and P-values 713 that were obtained through permutation tests (n=9999) of individual fractions.

25 714

26 715 Fig. 3. Variation partitioning (a, b, c) and ordination diagrams (d, e, f) of CWM 716 explained by the subsets of local, landscape and spatial variables. Percentages in the 717 partitioning are R2adj and circle sizes are their approximation. P-values were obtained by 718 permutation tests of individual fractions. Grey vectors are the explanatory variables; 719 round end vectors are for local variables; arrow end vectors are for landscape variables; 720 vectors without an end mark are for spatial variables (AEM vectors were named by the 721 numbers corresponding to those of Appendix S8). Response variables (black crosses) 722 should be interpreted as vectors. See Table 2 for trait codes. The square terms of 723 polynomial transformed variables model the non-linear (quadratic) relationship between 724 the response and explanatory variables. 725