Multi‐scale environmental filters and niche partitioning govern the distributions of riparian vegetation guilds

Across landscapes, riparian plant communities assemble under varying levels of disturbance, environmental stress, and resource availability, leading to the development of distinct riparian life-history guilds over evolutionary timescales. Identifying the environmental filters that exert selective pressures on specific riparian vegetation guilds is a critical step in setting baseline expectations for how riparian vegetation may respond to environmental conditions anticipated under future global change scenarios. In this study, we ask: (1) What riparian plant guilds exist across the interior Columbia and upper Missouri River basins? (2) What environmental filters shape riparian guild distributions? (3) How does resource partitioning among guilds influence guild distributions and co-occurrence? Woody species composition was measured at 703 stream reaches and each species' morphological and functional attributes were extracted from a database in four categories: (1) life form, (2) persistence and growth, (3) reproduction, and (4) resource use. We clustered species into guilds by morphological characteristics and attributes related to environmental tolerances, modeling these guilds' distributions as a function of environmental filters—regional climate, watershed hydrogeomorphic characteristics, and stream channel form—and guild co-existence. We identified five guilds: (1) a tall, deeply rooted, long-lived, evergreen tree guild, (2) a xeric, disturbance tolerant shrub guild, (3) a hydrophytic, thicket-forming shrub guild, (4) a low-statured, shade-tolerant, understory shrub guild, and (5) a flood tolerant, mesoriparian shrub guild. Guilds were most strongly discriminated by species' rooting depth, canopy height and potential to resprout and grow following biomass-removing disturbance (e.g., flooding, fire). Hydro-climatic variables, including precipitation, watershed area, water table depth, and channel form attributes reflective of hydrologic regime, were predictors of guilds whose life history strategies had affinity or aversion to flooding, drought, and fluvial disturbance. Biotic interactions excluded guilds with divergent life history strategies and/or allowed for the co-occurrence of guilds that partition resources differently in the same environment. We conclude that the riparian guild framework provides insight into how disturbance and bioclimatic gradients shape riparian functional plant diversity across heterogeneous landscapes. Multiple environmental filters should be considered when the riparian response guild framework is to be used as a decision-support tool framework across large spatial extents.


Introduction 55
Riparian zones are globally threatened ecosystems due to widespread hydrologic alteration, 56 watershed degradation, and the introduction of novel disturbance regimes and biota (Patten 1998 To answer this question, riparian ecologists have suggested that by aggregating individual 71 species into groups based on common life history strategies, broad inference can be made about 72 the environmental drivers of riparian plant diversity and used to predict ecosystem change 73 (Merritt et al. 2009(Merritt et al. , 2010. This trait-based approach to riparian community assembly, riparian 74 vegetation guilding, or determination of riparian "flow-response guilds" sensu Merritt et al. 75 (2010), provides a framework to identify how functional vegetation guilds assemble across 76 environmental gradients that filter species and life history strategies from biological 77 communities. Environmental filtering, it its most simple form, assumes that as environmental 78 conditions change, specific life history strategies and traits will be selected for at a given 79 location, leading to the assembly of communities with morphological and physiological 80 tolerances suited to a given environment (Keddy 1992, Díaz et al. 1998). When the dominant 81 environmental filters that shape riparian biodiversity are known, then riparian guilds can be 82 probabilistically modeled to predict ecosystem change as environmental filters shift (Merritt et 83 al. 2009).While many environmental filters shape riparian plant communities (Hough-Snee et al.

Study Sites 155
We selected 703 low-order stream reaches within the interior Columbia and upper Missouri 156 River basins (Figure 1) for inclusion in the study. These reaches are part of an existing stream 157 monitoring program and were sampled under a spatially balanced, probabilistic sampling design 158 (Kershner et al. 2004). All reaches were low-gradient (≈ 3%) and occur within subwatersheds 159 (USGS 6 th order hydrologic unit code) with > 50% federal ownership upstream of the sampled cover was measured for all species in a lower vegetation layer (< 1m in height) and an upper 171 woody species layer (> 1m in height). Cover was estimated in classes: t5-15%, t15-25%, t25-172 38%, t38-50%, t50-75%, t75-95%, and t95-100%. Due to the possibility of overestimating 173 guild cover by using data from both layers or underestimating guild cover by only using one of 174 the layers, species presence and absence were derived. If a species was observed in either 175 vegetation layer, then it was classified as present at a site, otherwise it was classified as absent. riparian environment along a typical, low-order stream. Smaller, wadeable streams are exposed 182 to multiple stressors from fluvial (overbank flooding, erosion, deposition, etc) and terrestrial 183 processes (wildfire, grazing, forest fragmentation, etc.) as well as landscape processes (climate, 184 etc.). Accordingly, the plant attributes we selected for guilding aligned with multiple 185 environmental filters across the study landscape (Table 1)  We use the term morphological or functional "attribute" as opposed to "trait", because traits are 204 defined as empirically measured physiological and morphological parameters that change in 205 response to the physical environment, whereas many of our species attributes were 206 categorizations and not empirical measurements. It is worth noting that of the small number of 207 attributes selected for guilding here, many often covary with other traits. A limited number of 208 attributes (or when available, measured traits) may be used in such guilding providing the 209 advantage that a parsimonious set of traits may actually represent a family of traits (Duckworth 210 et al. 2000). For example, wood density is easy to measure yet represents a complex set of 211 physiological traits that are strongly correlated with water use efficiency in plants (Reich 2014). 212 213

Environmental metrics 214
Stream gradient, bankfull width, bank stability, channel sinuosity, bank angle, median particle 215 size, wood frequency, wetted width-depth ratio, residual pool depth, hydraulic radius, and 216 percent undercut banks were field measured at each reach using standardized protocols (Table 2; 217 (PIBO EM 2012b). We identified a 30m buffer surrounding each stream in GIS and calculated 218 the proportion of each buffer polygon that was grazed by livestock in the last 30-years using 219 USFS grazing allotment data. Because forest patchess serve as corridors for propagule dispersal 220 following disturbance and tree canopies shape understory light and humidity, we identified the watershed-scale filters were summarized for the watershed area upstream of each reach (Table  232 2). 233 234

Riparian guild identification 235
We identified riparian life history strategy guilds by clustering species based on their 236 morphological and physical attributes (Table 1). We calculated a distance matrix of species and 237 species' attributes using Gower's distance (Gower and Legendre 1986), which scales variables 238 between 0 and 1 and allows for the use of continuous and ordinal variables. We clustered species 239 based on this distance matrix using Ward's method and examined cluster results for three to ten 240 guilds, settling on a five-guild (cluster) solution. We visualized the resulting guilds and the 241 attributes that differentiated them using a three-dimensional principal coordinate analysis 242 PrePrints Filtering, niche partitioning, and riparian guilds A systematic approach was taken to model each guild's presence and absence across the study 261 region. Generalized linear models were fitted for each guild using environmental attributes as 262 predictors of guild presence and absence (binomial function; Table 2). Prior to model building 263 we removed environmental variables with correlations > 0.65 to avoid collinearity. We included 264 interaction terms for variables with spatial codependence including bank angle and buffer slope, 265 sinuosity and gradient, and bankfull width and wetted width to depth ratio. We used a systematic

Environmental gradients and guild distributions 306
Riparian guild assemblages occurred in 32 different combinations at the 703 study reaches, from 307 reaches with no woody riparian guilds present to reaches where all identified woody riparian 308 guilds were present ( Figure 4, Appendix F). A three-dimensional NMDS ordination solution of 309 guild assemblages converged after 17 tries (principal components rotation; Euclidean distance; 310 stress = 0.047, P = 0.009). The combinations of guilds that assembled at each reach and 311 individual guilds were strongly correlated to multiple environmental filters ( Figure 4, Table 4, 312 Appendix C, D). Buffer slope, reach elevation, sinuosity, stream gradient, buffer forest cover, The presence and absence of individual riparian guilds corresponded to many of the same 325 environmental filters that correlated to guild assemblages (Table 4). Generalized linear models 326 (GLMs) and conditional inference trees (CITs) showed that for most guilds, in addition to 327 environmental filtering effects from hydrologic and channel form attributes, the presence and 328 absence of other guilds were significant predictors of guild presence and absence (Table 4). The 329 final evergreen tree guild GLM contained numerous environmental filters and riparian guilds. 330 Hydrologic variables that negatively correlated to evergreen tree guild presence were watershed 331 area and average water table depth while the channel-form variables, sinuosity and buffer slope, 332 were also negatively correlated to conifer presence. Annual precipitation, wetted width-depth 333 ratio, buffer forest cover and the presence of the upland disturbance and understory shrub guilds 334 were positively correlated to evergreen tree guild presence ( Table 4) (Table 4, Appendix E). The upland disturbance guild's CIT showed that the presence of 346 the evergreen tree guild was a major predictor of upland disturbance guild presence behind 347 buffer slope. The final CIT successfully predicted upland disturbance guild presence at 71.6% of 348 reaches ( Figure 5). We found that the occurrence of these guilds can be predicted by multiple environmental 397 gradients that filter life history strategies from individual reaches. Our results build on previous 398 research that showed riparian forest regeneration strategies are tied to multiple environmental 399 gradients and biotic interactions (Sarr et al. 2011) and that functional guilds that respond to such 400 gradients are informative. The occurrence of each guild was strongly associated with 401 environmental conditions at landscape (e.g., elevation, precipitation and temperature), In addition to identifying environmental filters that predict guild distributions, we found 414 evidence for the coexistence of multiple guilds at the same reach. Individual guilds were 415 consistently found either to be complementary to or mutually exclusive with other guilds, 416 suggesting that in some cases guild's species differentially partition their niches within similar 417 environments. For example, the evergreen tree guild was positively associated with both the 418 upland disturbance guild and the canopy understory guild, likely because these guilds acquire 419 resources differently when co-existing in similar environments. The less disturbance adapted 420 evergreen tree guild is unlikely to occupy disturbed forest edges suitable for the upland 421 disturbance guild, and thus the two were often found together at a site (i.e., the two guilds 422 occupied different unique locations within a site, preventing competitive exclusion). The 423 understory shrub guild is positively associated with the evergreen tree guild because the tall, 424 mature overstory trees provide suitable shaded habitat for the shade-tolerant understory guild. The utility of the riparian guild framework is developing rapidly and its utility will improve as 464    was not a better predictor of guild presence or absence than random chance and is not presented 729