Abstract Questions: Does urbanization promote biotic differentiation or homogenization of swamp plant communities? What is the contribution of natives and exotics to swamp response to urbanization? Location: Quebec City, Canada. Methods: Plant communities of 34 swamps located in low, moderately or highly urbanized landscapes were sampled, and species classified into three exclusive groups: native wetland, native upland and exotic plants. Urbanization influence on the richness of each plant group was assessed using mixed models. Between-site compositional similarities were calculated to identify variations in beta diversity with urbanization level using tests for homogeneity in multivariate dispersion. Beta diversity was further partitioned into species replacement and richness difference for each plant group. Finally, the relationships of ten environmental variables representing soil water saturation and microtopography with plant assemblages were determined by Redundancy Analysis. Results: Although the richness of exotics increased with urbanization intensity, revealing increasing propagule pressure, it remained six to 27 times lower compared to natives, whose richness remained stable with urbanization. On the other hand, beta diversity increased with urbanization, with higher dissimilarities in species composition between highly urbanized swamps than between low urbanized ones. This pattern resulted from high species replacement among natives, while richness difference mainly contributed to exotic beta diversity. Changes in plant assemblages were mostly associated with bryophyte cover and soil drainage and red mottle size, suggesting that hydrological conditions likely acted as a strong driver of swamp plant community response to urbanization. Conclusions: Swamp plant communities experienced biotic differentiation with increasing urbanization. This differentiation pattern likely was linked to the unpredictable effect of urbanization on hydrological regimes, which promoted high native turnover while limiting exotic spread. Long term monitoring is recommended to ensure that exotics do not outcompete natives through time. Designing sustainable cities requires a greater understanding of the multifaceted effect of urbanization on biodiversity. Methods Study area The study was conducted in the Quebec City metropolitan area (46°48'52"N 71°12'28"W; hereafter referred to as Quebec City), the seventh most populous urban area in Canada (569 717 inhabitants; Statistics Canada, 2016). Across this 548 km2 territory, 50% of land use consists of remnants of natural habitats, 39% of built-up areas and 11% of agricultural lands. Built-up areas, which have increased by 79% in the last 35 years (Nazarnia et al., 2016), correspond to residential (24%), industrial/commercial (5.5%), road networks and mining areas (5%), and vacant lots (4.5%; Cimon-Morin and Poulin, 2018). Yet, nearly 4 921 ha of wetlands (8% of the landscape) are still present across Quebec City metropolitan area, including 2 394 ha of swamps (Beaulieu et al., 2014) Site selection Sites were selected based on a map of Quebec City that situates wetlands larger than 0.3 ha according to seven classes identified by photointerpretation (bog, fen, forested peatland, marsh, swamp, wet meadow and shallow water). Among them, 102 swamps were retained according to the following criteria: 1) an area ranging from 1 to 6 ha, to avoid biases due to size effect; 2) a distance of at least 300 meters between sites; 3) a balance between riparian and isolated swamps (i.e., not directly connected to a permanent watercourse); and 4) a surrounding landscape not dominated by agricultural fields. These 102 swamps were then visited, to exclude bush-dominated, degraded and misclassified swamps (such as forested peatlands). Thirty-four swamps met all criteria. For each swamp, we then characterized landscape composition in a 100 m buffer zone using nine land use categories (Appendix S1) obtained from photointerpretation in QGIS 3.0.0 (QGIS Development Team, 2018). In the surrounding of the sampled swamps, urbanization had mostly taken place from the 1960s to the 1980s (Raimbault, 2019). Land use composition was then used to group swamps according to three levels of landscape urbanization based on the optimum of a non-hierarchical k-means clustering (Legendre and Legendre, 2012). This clustering approach allowed us to account for different land-use classes simultaneously, and therefore better represent the complexity of urbanization process which is hardly synthesized by a single continuous landscape variable (Grimm et al., 2008). Nine low, 14 intermediate and 11 highly urbanized swamps were identified (i.e., swamps respectively located in surrounding landscapes with low, intermediate and high urbanization levels). These urbanization levels increased with decreasing cover of forests and wetlands in the landscape surrounding each site (87% cover at low urbanization level, 57% at intermediate urbanization level and 25% at high urbanization level), and with increasing cover of impervious surfaces including residential and commercial areas, industrial sites, highways and secondary roads (8% cover at low urbanization level, 23% at intermediate urbanization level and 50% at high urbanization level; Appendix S1-2). Vegetation surveys Vascular plant communities in the 34 selected swamps were sampled during the summer of 2016 (end of June-beginning of September). Two to five sampling plots each measuring 400-m2 (20 × 20 m) were established per swamp, depending on its size, in order to uniformize sampling intensity per swamp area, for a total of 92 plots (i.e., 25, 38 and 29 plots sampled in swamps respectively corresponding to low, intermediate and high urbanization level) . These plots were randomly positioned within each swamp while respecting a 30-meter distance between plots to limit spatial autocorrelation and a 25-meter distance from the edge to avoid edge effect (Alignier et al., 2014). In each plot, the cover of each plant species was visually estimated using seven classes: <1%, 1–5%, 6–10%, 11–25%, 26–50%, 51–75%, 76–100%. Nomenclature follows VASCAN (Brouillet et al., 2019). Plant cover was averaged at the site scale for analyses. A preliminary analysis detected no significant correlation between swamp area and species richness (r = -0.06; P = 0.74), nor between sampling area and species richness (r = 0.18; P = 0.31), indicating that the sampling method did not induce species-area bias. Species groups To better determine the impacts of urbanization level on plant communities, all inventoried species were classified into three mutually exclusive plant groups: native wetland, native upland and exotic plants (Appendix S3). We first distinguished between species based on their origin (native or exotic to the Quebec province) following VASCAN (Brouillet et al., 2019). Then, all native species were sorted based on their habitat preference (wetland or upland species) following Bazoge et al. (2014) and the PLANT database (USDA, 2019). “Obligate” and “facultative wetland” were classified as wetland species (i.e., specialist plants preferentially occurring in wetlands), and “facultative,” “facultative upland” and “upland” as upland species (i.e., generalist plants equally occurring in wetland and terrestrial habitats as well as plants occurring preferentially in terrestrial habitats). Only two species, Lythrum salicaria and Lysimachia nummularia, were both exotics and wetland species, but neither was frequent (present in 15% and 9% of the sites, respectively) or abundant (<1% of cover in each site for both species) and they were thus classified as exotics exclusively. Environmental variables Ten environmental variables were evaluated in each plot. Soil texture and drainage were evaluated using a semi-quantitative scales ranging from 1 (sand) to 12 (clay) for texture and from 0 (excessive) to 6 (very bad) for drainage (Saucier, 1994). The size (1: < 5 mm; 2: 5-15 mm; 3: > 15 mm), depth (cm) and abundance (1: <2%; 2: 2-20%; 3: > 20%) of soil mottles as well as the thickness of humus or peat (cm) were quantified as proxies of water table depth and near-surface water saturation, given that humus degrades more rapidly in aerobic conditions (Zoltai and Vitt, 1995; Mitsch and Gosselink, 2015). No significant correlation between soil mottle abundance and sampling date was detected (r = 0.25; P = 0.16), suggesting that our sampling design did not induce biases in soil conditions, and evidencing that soil mottles are relatively stable through time as previously reported (Vepraskas and Craft, 2016). Similarly, no significant correlation was detected between soil type (organic vs. non-organic) and soil mottle abundance (r = -0.27; P = 0.12). Microtopographic variation was assessed using a four-class index based on the elevation difference between pits and mounds (0: flat, 1: <0.5 m, 2: 0.5-1 m, 3: more than 1 m of amplitude). The cover of bryophytes (largely dominated by Sphagnum spp.), vernal pools and bare ground surfaces was additionally estimated using the same classes as for plant cover to approximate hydric conditions at soil surface (Goguen and Arp, 2017).

Columns B to JS indicate species importance values. Columns JT to KD indicate urbanization level and environmental variables.