Habitat conditions at beaver settlement sites: implications for beaver restoration projects
Abstract
Recognition that beavers are integral components of stream ecosystems has resulted in an increase in beaver-mediated habitat restoration projects. Beaver restoration projects are frequently implemented in degraded stream systems with little or no beaver activity. However, selection of restoration sites is often based on habitat suitability research comparing well-established beaver colonies to unoccupied stream sections or abandoned colonies. Because beavers dramatically alter areas they occupy, assessing habitat conditions at active colonies may over-emphasize habitat characteristics that are modified by beaver activity. During 2015–2017, we conducted beaver activity surveys on streams in the upper Missouri River watershed in southwest Montana, United States, to investigate habitat selection by beavers starting new colonies in novel areas. We compared new colony locations in unmodified stream segments to unsettled segments to evaluate conditions that promoted colonization. Newly settled stream segments had relatively low gradients (β ± SE = −0.72 ± 0.27), narrow channels (β = −1.31 ± 0.46), high channel complexity (β = 0.76 ± 0.42), high canopy cover of woody riparian vegetation (β = 0.56 ± 0.21), and low-lying areas directly adjacent to the stream (β = 0.36 ± 0.24), where β denotes covariate effect sizes. Habitat selection patterns differed between our new settlement site analysis and an analysis of occupied versus unoccupied stream segments, suggesting that assessing habitat suitability based on active colonies may result in misidentification of suitable site conditions for beaver restoration. Our research provides recommendations for beaver restoration practitioners to select restoration sites that will have the highest probability of successful colony establishment.
Implications for Practice
- In the western United States, beavers seeking new settlement sites may be attracted to areas where initial dams have the best chance of withstanding high-water events.
- Practitioners should evaluate suitability of restoration sites at multiple spatial scales and emphasize low-gradient stream segments, dense woody riparian vegetation, relatively narrow stream channels, wetland types corresponding to low-lying areas, and greater channel complexity in the form of side channels and tributaries.
- Construction of artificial beaver structures at restoration sites should encourage settlement by beavers and facilitate riparian restoration.
- Habitat conditions associated with settlement of novel areas (e.g. restoration sites) are different than those at occupied sites; restoration efforts should emphasize habitat conditions related to the establishment of new colonies in novel areas.
Introduction
The habitat-modifying activities of beavers (Castor spp.) are instrumental in the creation, expansion, and maintenance of healthy stream systems and associated riparian and wetland habitats (Naiman et al. 1988; Gurnell 1998; Rosell et al. 2005; Burchsted et al. 2010). Over the last half-century, land and wildlife managers have recognized the potential benefits of beaver-mediated habitat restoration (hereafter, “beaver restoration”; McKinstry & Anderson 1999; Macfarlane et al. 2014). Beaver restoration refers to a wide variety of stream and riparian restoration techniques that rely on or mimic beaver activity to accomplish restoration goals (Pollock et al. 2018). The effectiveness and efficiency of these techniques has led to a proliferation of beaver restoration projects in many areas throughout the United States (Pollock et al. 2014, 2018; Pilliod et al. 2017). Most beaver restoration projects aim to establish self-sustaining colonies of beavers, whose activities increase landscape-scale water storage, improve stream conditions, and enhance habitat for other species (Albert & Trimble 2000; Bouwes et al. 2016; Pollock et al. 2018).
Critical to the success of any beaver restoration project is the identification of restoration sites where beavers will have a high probability of establishing and maintaining colonies. Habitat assessments are therefore a necessary step prior to project implementation. If habitat is not suitable at restoration sites, beavers will either not occupy the site, or their colonies may be intermittent, and habitat restoration goals may not be realized. Identification of suitable habitat is particularly important if beavers are to be translocated to the restoration site, as beavers released into unsuitable habitat will quickly abandon the area or be killed by predators (McKinstry & Anderson 2002; Petro et al. 2015).
Previous studies have reported a variety of biotic and abiotic environmental conditions associated with beaver colonies, and authors often interpret those conditions as indicative of beaver habitat suitability (Howard & Larson 1985; Suzuki & McComb 1998; Pinto et al. 2009). Important habitat metrics from these studies are then included in formal evaluations of potential beaver restoration sites (Allen 1982; Vore 1993; Macfarlane et al. 2015; Pollock et al. 2018). Published habitat suitability studies that provide the foundation for selection of restoration sites have followed a general protocol of comparing habitat conditions at established colonies to random locations (Dieter & McCabe 1989), or unoccupied stream sections (Beier & Barrett 1987; Barnes & Mallik 1997; Suzuki & McComb 1998). Results of these studies confound our understanding of habitat selection by beavers in relation to restoration because habitat conditions at established colonies may not accurately represent suitable habitat for beavers starting new colonies in novel areas.
Habitat conditions at established beaver colonies may reflect settlement site selection patterns or, more likely, beaver-induced habitat manipulations. Therefore, the selection of appropriate restoration sites requires an understanding of habitat conditions that allow for colony establishment in areas that are relatively unmodified by beavers. “Unmodified” habitats refer to stream segments where there is no detectable sign of previous beaver occupancy, though historically most streams in the western United States were likely modified by beavers at some point (Muller-Schwarze 2011). In areas where most of the high-quality habitat is occupied, beavers may be forced to start new colonies in unmodified and presumably suboptimal habitat (hereafter referred to as “new settlement sites”; DeStefano et al. 2006; František et al. 2010; Scrafford et al. 2018). New settlement sites, characterized by suboptimal habitat and lack of current or recent beaver activity, provide a natural analog to the types of habitats beavers would be encouraged to colonize as part of beaver restoration projects (Bouwes et al. 2016; Pollock et al. 2018).
In order to identify habitat characteristics that promote settlement by beavers, we conducted beaver activity surveys along streams and rivers in southwest Montana, United States over a 3-year period. Our primary research objectives were to: (1) map stream sections that were generally unmodified by beavers; (2) identify new settlement sites in previously unmodified habitats; (3) model new settlement site selection relative to biotic and abiotic habitat conditions; (4) compare habitat selection patterns between new settlement sites and established beaver colonies; and (5) develop recommendations for identifying potential restoration sites with the highest probability of successful colony establishment.
Methods
Study Area
We conducted beaver activity surveys along streams and rivers in the headwaters of the Missouri River watershed within the Custer-Gallatin National Forest in southwest Montana, United States. The study area was primarily made up of public lands with limited cattle grazing and resource extraction. Beavers were abundant in the study area prior to Euro-American settlement but overharvesting in the 19th century resulted in near extirpation (Muller-Schwarze 2011). Beaver populations rebounded in the 1900s and have since recovered throughout the study area, although accurate estimates of densities are lacking. The climate in the study area consists of long, cold winters where beaver ponds are frozen over for approximately 4 months of the year (January–April). Spring run-off periods cause rivers and streams in the study area to flood up to and beyond bank-full levels. Summers are generally hot and dry followed by cooler but still dry autumns. Average annual precipitation during the study period was approximately 54 cm with most precipitation coming in the form of snow (U.S. Climate Data 2019).
The study area encompassed the upper Madison River drainage and the upper Gallatin River drainage (Fig. 1). Both rivers and many of their tributaries flow out of high-elevation mountain ranges in the Greater Yellowstone Area. Streams in the study area were generally small (1–10 m wide) and shallow enough that beavers build dams to impound water to a depth to cache food under ice in the winter. Streams were surrounded by willow-dominant riparian areas while uplands were a mix of conifer forests and mountain big sagebrush (Artemisia tridentata vasevana) grasslands. The most common forage for beavers was willow (Salix spp.), although aspen (Populus tremuloides), alder (Alnus incana), and cottonwood (Populus trichocarpa) were available in some areas. Recreational trapping of beavers was only allowed in the Madison drainage, and was highly regulated by the Montana Department of Fish, Wildlife, and Parks (MFWP). Overall beaver trapping in the study area was low enough in the preceding 30 years that trapping-related mortality was considered rare in the study area (De Caussin 2013; J. Cunningham, MFWP, personal communication).
Beaver Activity Surveys
Prior to conducting field work, we defined potential beaver habitat in the study area as stream sections ≥400 m in length with gradients less than 15% and a woody riparian vegetation zone that was wider than a single band of vegetation immediately adjacent to the stream channel (Allen 1982; Suzuki & McComb 1998; Muller-Schwarze 2011). We included all suitable stream lengths in the study area except for some stream sections within Yellowstone National Park where access was restricted. We digitized suitable streams using ArcMap 10.3.1 software (Environmental Systems Research Institute, Inc., Redlands, CA, U.S.A.) and aerial imagery from the National Agriculture Imagery Program (NAIP; 1-m resolution; USDA-FSA-APFO Aerial Photography Field Office, Salt Lake City, UT, U.S.A.). We then divided stream lines into stream sampling segments 400 m in length, based on an evaluation of the spatial extent of active beaver colonies in the study area (Ritter 2018). We observed that beaver activity associated with a colony typically extended 200 m upstream and downstream of the primary lodges and dams (Ritter 2018).
We conducted ground-based beaver activity surveys once per year during July–November, 2015–2017. We timed surveys to correspond with active periods of dam and lodge construction after water levels dropped in late summer after most dispersing beavers had chosen a settlement site for the year (Jackson 1990; Van Deelen 1991; Ritter 2018). We walked along streams and classified sampling segments based on the type, activity level (e.g. active or inactive), and spatial distribution of beaver sign in the segment. Types of beaver sign included lodges, bank dens, dams, caches, clippings, and castor mounds, as well as beaver tracks and scat (Muller-Schwarze 2011).
We classified sampling segments into five categories: active, abandoned, relic, unoccupied, or newly settled. An active sampling segment was dominated by recent beaver sign including major dams or lodges and was part of a well-established colony where significant habitat manipulations had already occurred. Abandoned sampling segments contained unused dams or lodges and had no fresh beaver sign indicating the site was relatively recently abandoned. Abandoned segments had dams that were still holding water or could be easily repaired, woody vegetation that had not yet regenerated from clipping by beavers, and lodges that could be easily inhabited with minor repairs. Relic sampling segments contained unused dams or lodges that would need to be rebuilt for beavers to occupy the segment, and the stream channel had returned to a pre-colony state. Nevertheless, more subtle and long-term habitat manipulations may have existed in relic segments. A sampling segment classified as unoccupied had no sign a beaver colony was ever well-established in the area or had been gone long enough that only transient sign remained. We defined newly settled sampling segments as those where one or more beavers were constructing a dam and lodge within segments that were classified as relic or unoccupied the year prior to settlement. We followed this protocol to minimize the potential impact of previous beaver structures on habitat selection (Smith 1997).
Habitat Data
We collected habitat data for each sampling segment using two common approaches for evaluating habitat suitability for beaver restoration projects. First, we collected data using remotely sensed imagery and a geographic information system (GIS) (hereafter “GIS-based habitat sampling”) to represent a broad-scale habitat suitability assessment. Second, we measured fine-scale habitat conditions that could not be obtained from remotely sensed data to represent a field-based evaluation of habitat suitability (hereafter “field-based habitat sampling”). Our data covered three major habitat components hypothesized to influence the selection of new settlement sites by beavers: stream geomorphology, vegetation, and wetland type (Tables S1 and S2). Stream geomorphology variables represented stream channel and geologic conditions that influence dam-building, lodge construction, and access to vegetation. Vegetation variables represented the quantity and distribution of available forage and construction resources. We used the Montana Natural Heritage Program's Wetland and Riparian Framework to estimate the amount of various wetland classifications (Cowardin et al. 2013), within 30-, 50-, and 100-m buffers around each sampling segment (Jenkins 1980; Belovsky 1984).
For GIS-based habitat sampling, we evaluated habitat conditions at each sampling segment using ArcMap 10.3.1 software, a 10-m U.S. geological survey Digital Elevation Model, and NAIP aerial imagery. For field-based habitat sampling, we gathered habitat measurements at 10 random points spaced at least 20 m apart within each 400-m sampling segment. We were unable to measure habitat conditions at all unsettled sampling segments in the field. Instead, we randomly selected unsettled sampling segments within the same stream as the newly settled sampling segments to act as paired sites. If the newly settled sampling segment was previously classified as relic, we chose a paired sampling segment that was also classified as relic to account for possible selection for habitat conditions that were altered by previous beaver occupancy.
Analysis
We followed a used:unused study design when evaluating settlement site selection by beavers where resource units (sampling segments) were classified as used (newly settled) or unused (unsettled) based on the presence and age of beaver sign in the segment (Manly et al. 2002). We are confident we accurately distinguished used versus unused sites because beavers show high site fidelity by late summer and are conspicuous when occupying an area due to construction and foraging activities. We defined used units in the analysis as a sampling segment that was settled in a given year but was classified as relic or unoccupied the year before. We defined an unused unit as a sampling segment that was classified as relic or unoccupied for at least 1 year of the study and was not occupied by beavers.
We used generalized linear mixed models with the logistic link function and a binomial error structure to examine the influence of habitat variables on the probability a sampling segment would be newly settled by beavers. Prior to fitting models of settlement site selection, we used Spearman-rank correlation coefficients (r) to test for multicollinearity among our suite of habitat variables. We did not include any pairs of habitat variables with |r| ≥ 0.60 in the same models. The importance of habitat variables to settlement site selection by beavers likely varies depending on intrinsic or environmental conditions specific to a particular stream system (McLoughlin et al. 2010). Additionally, habitat conditions of sampling segments within the same stream are correlated due to being close in space. Therefore, we included a random intercept effect of individual stream ID in all models to account for potential spatial autocorrelation among experimental units (lme4 package for R; R Version 3.3.2, www.r-project.org, accessed 28 Dec 2017; Boyce et al. 2002; Bolker et al. 2009; Bates et al. 2015).
We developed sets of candidate models representing a priori hypotheses of the influence of habitat variables on new settlement site selection by beavers. We used a tiered approach to model evaluation by grouping covariates into three broad categories of environmental conditions that we analyzed separately (i.e. stream geomorphology, vegetation, and wetland types). Prior to the development of multivariate models, we tested hypothesized pseudo-threshold relationships (Franklin et al. 2000) between the probability of settlement and floodplain width and riparian width, as well as hypothesized quadratic relationships between the probability of settlement and watershed size and distance to the nearest active colony. We ranked candidate models using Akaike's information criterion adjusted for small sample sizes (AICc; Burnham & Anderson 2002). We considered all models less than two AICc from the top model parsimonious, and were conservative in our interpretation of model selection results following the recommendations of Anderson and Burnham (2002) and Arnold (2010). We used all the variables included in the parsimonious models for each habitat category to build a full model representing new settlement site selection by beavers. To avoid over-fitting the final model, we limited the number of terms to four or fewer using a backwards elimination procedure. When multiple models were supported, we calculated model-averaged estimates of beta-coefficients (
) using R package AICcmodavg (Mazzerolle 2017), which were used to interpret the direction and magnitudes of habitat covariates on settlement site selection. We evaluated the relative influence of the random stream effect using a likelihood ratio test (Pinheiro & Bates 2000), and plotted predicted values of the top fixed effects separately for each level of the random effect to illustrate the variation in responses among streams (“ggplot2” package for R; Wickham 2016). We evaluated goodness-of-fit for the top settlement site selection model by calculating marginal R2 and conditional R2 (Nakagawa et al. 2013).
We collected field-based habitat data at a subset of the total stream segments and included 19 new settlement sites and paired unsettled sites (n = 19 pairs). Therefore, we analyzed these data separately using logistic regression to evaluate the influence of fine-scale habitat characteristics on the probability of settlement by beavers. We included a maximum of three independent habitat covariates in any one model and used backward selection to drop uninformative covariates. We compared competing models, including a null model, using AICc. Although we observed spatial clustering of new settlement sites in streams, the lack of multiple settlement sites in some streams prevented us from including a random stream intercept and evaluating spatial autocorrelation in beaver settlement.
To evaluate habitat conditions associated with well-established beaver colonies in the study area, we defined used sampling segments as those occupied by beavers for at least 2 years of the study, not including new settlement sites. Sampling segments that fell exclusively into the classification of active or abandoned during the 3-year study period were also classified as occupied. Unused sampling segments were those segments classified as abandoned, relic, unoccupied, newly settled, or some combination of these categories, for the entire study period. Due to a larger sample size of occupied sampling segments relative to newly settled segments, we included up to seven terms in our final occupied versus unoccupied habitat selection model. As with the GIS-based analysis, we included a random stream intercept effect.
Results
We conducted beaver activity surveys over 244 km of streams represented by 613 sampling segments in 27 streams within the study area. Of the 613 sampling segments, 370 were classified as relic or unoccupied for at least 1 year of the study and were therefore available to be newly settled by beavers. We identified a total of 27 new settlement sites during 2015–2017. The 27 new settlement sites resulted in 48 relic or unoccupied sampling segments being settled during the study period. Thirty (63%) of the newly settled sampling segments were previously relic segments and 18 (37%) were previously unoccupied segments. Considering all sampling segments in the study area, 31% (30/97) of available relic segments were settled, compared to 7% (18/273) of available unoccupied segments. For our occupied versus unoccupied analysis, we identified 229 occupied sampling segments and 384 unoccupied sampling segments.
GIS-Based Settlement Site Selection
Three nested models were supported by the data and included the effects of stream gradient, canopy cover of woody riparian vegetation, and the proportion of sparse-willow and waterbody wetland types within 30 m of the stream (Table S3). A null model had virtually no support (AICc wnull < 0.001). Model-averaged coefficient estimates (± SE) indicated the probability of settlement by beavers declined sharply with the proportion of the stream segment in the waterbody wetland type (β = −1.31 ± 0.46). The waterbody wetland type can be interpreted as an overall measurement of stream width for the sampling segment. Stream gradient also had a negative effect (β = −0.72 ± 0.27); on average, the odds of a beaver settling a site declined by 51% for every 1% increase in stream gradient, and stream segments with >5% gradients were never settled by dispersing beavers (Fig. 2). The probability of settlement increased with woody canopy cover (β = 0.56 ± 0.21) and the proportion of sparse-willow wetland type (β = 0.36 ± 0.24). The odds that a stream segment was settled increased by 46% for every one unit increase in woody canopy cover index and 43% for every 10% increase in the cover of the sparse-willow wetland type (Fig. 2).
A likelihood ratio test indicated strong support for a stream-level random effect (p < 0.001). Some streams in the study area had higher baseline probabilities of settlement by beavers, and the spread of intercepts around the predicted fixed effect suggested wide variation in new settlement dynamics across streams (Fig. 2). Higher conditional R2 (R2GLMM[c] = 0.52) relative to marginal R2 (R2GLMM[m] = 0.31) provided further support for a random stream effect.
Field-Based Settlement Site Selection
Of the 27 new settlement sites identified in the study area during 2015–2017, we sampled 19 in the field along with 19 paired unsettled sites. We did not conduct field surveys at the remaining eight settlement sites because beavers had already drastically manipulated fine-scale habitat conditions before discovery. Of the new settlement sites where habitat sampling took place, 11 were previously relic sites and eight were previously unoccupied.
To evaluate the effects of fine-scale environmental conditions on settlement site selection, we considered seven models that included additive combinations of three independent covariates measured within stream segments: channel complexity, average distance to preferred forage, and mean stream width:depth ratio. Four models were considered parsimonious (ΔAICc ≤ 2) and accounted for 78% of the support from the data (Table S3). Model-averaged coefficients (± SE) of the standardized habitat effects indicated the probability a stream segment was settled by beavers increased with increasing channel complexity (β = 0.76 ± 0.42). The probability of settlement decreased with width:depth ratio (β = −0.59 ± 0.43) and distance from the stream to preferred forage (β = −0.91 ± 0.72), but the 85% confidence intervals of the model-averaged effect sizes overlapped zero.
Occupied/Unoccupied Habitat Selection
To evaluate the influence of habitat conditions on the probability a stream segment was occupied by a well-established beaver colony, we considered 14 models with combinations of habitat variables representing stream geomorphology, vegetation, and wetland types. Three models were considered parsimonious and accounted for 93% of the support of the data (Table S4). The three models included pseudo-threshold terms for sinuosity and riparian width, and linear terms for willow height, number of secondary channels, woody riparian vegetation canopy cover index, and the proportion of the sparse-willow and willow-dominant wetland types (Fig. 2). Model-averaged coefficients (± SE) of the standardized habitat effects indicate the strongest predictor of occupied beaver habitat from the stream geomorphology category was a pseudo-threshold effect of sinuosity (β = 0.38 ± 0.15), while forage height class two (80–190 cm) was the top vegetation variable (β = 1.81 ± 0.67), and the best predictor of occupied habitat overall. The proportion of the willow-dominant wetland type was the strongest predictor from the wetland type category (β = 0.96 ± 0.18).
The top models from the GIS-based analyses of new settlement sites and occupied versus unoccupied habitats shared several key variables, including a positive effect of woody vegetation canopy cover index and the proportion of the sparse-willow wetland type and a negative effect of stream gradient. However, the negative effect of stream gradient was not included in our final model for the occupied versus unoccupied analysis. The occupied versus unoccupied analysis resulted in a model weighted heavily toward habitat variables related to woody vegetation, while the new settlement site model suggested greater emphasis on geomorphological conditions of the stream (Fig. 3).
Discussion
The settlement of relatively unmodified stream segments by beavers was uncommon in our study area, and only 13% of available relic and unoccupied sampling segments were colonized over a 3-year time period. The scarcity of new settlement sites may be due to the low availability of high-quality stream segments. Most stream segments with active colonies in our study area were characterized by conditions previously found to represent high-quality habitats (Howard & Larson 1985; Scrafford et al. 2018), suggesting the remaining habitat available for settlement was of lower quality. In an associated study, we found that many existing colonies had more than two adult beavers (Ritter 2018), signifying delayed dispersal in our study population (Smith 1997; Sun et al. 2000; Mayer et al. 2017a; Mayer et al. 2017b). While the ecological drivers of delayed dispersal are complex, often high population densities and low-quality habitat in unoccupied territories leads to delayed dispersal in territorial species (Koenig et al. 1992; Stenseth & Lidicker 1992).
Our data indicate new settlement sites in the study area were established in areas with narrower floodplains, steeper gradients, and lower amounts of preferred forage than active and abandoned beaver colonies (Howard & Larson 1985; Scrafford et al. 2018). More than half (56%) of the new settlement sites we documented were abandoned within 1 year, suggesting beavers were forced into suboptimal habitats where long-term colony success was more difficult. Previous studies show that beaver colonies established in suboptimal habitats are more dynamic, either moving around the stream over time or being abandoned altogether, because habitat is less stable and therefore difficult to occupy long term (Howard & Larson 1985; Nolet & Rosell 1994; Demmer & Beschta 2008; František et al. 2010; Scrafford et al. 2018).
Sampling segments classified as relic were settled at nearly twice the rate of unoccupied segments. Relic segments had higher habitat quality as defined by Howard and Larson (1985) and Scrafford et al. (2018), compared to unoccupied segments. However, it is unclear whether habitat conditions at relic segments were intrinsically more suitable or made more suitable as a result of habitat manipulations brought about by previous beaver occupancy. Relic lodges were frequently used while colonizing beavers were constructing new dams and lodges in relic segments (Ritter 2018), suggesting past beaver use likely influences settlement decisions by dispersing beavers. Notably, abandoned stream segments where dams and lodges were still in place were settled at an even higher rate than relic stream segments, providing further evidence that dispersing beavers target previously occupied habitats when seeking settlement sites. The availability of existing beaver infrastructure to transient beavers may therefore have implications for population demography and expansion because dispersing beavers are more vulnerable to predation than those residing in established colonies (McNew & Woolf 2005).
Beavers dispersing into novel areas within the upper Gallatin and Madison River drainages of Montana selected stream segments with geomorphological conditions that best facilitated stable and efficient dam construction. Among all the available unoccupied and relic stream segments, new settlement sites were located in lower gradient segments with narrower stream channels and low-lying areas adjacent to the stream with sparse woody riparian vegetation. Segments with these characteristics may allow beavers to construct stable dams that are more likely to withstand high seasonal flows. Lower gradient stream segments generally have less stream power which reduces pressure on new dams (McComb et al. 1990; Pollock et al. 2014), and result in a larger flooded area per unit of dam height. Narrower stream channels may be easier for beavers to dam and dams in narrow streams may be more likely to withstand high flows (Suzuki & McComb 1998; Pollock et al. 2014). However, the influence of stream width on dam construction and maintenance is almost certainly mediated by other factors including initial stream depth, discharge, and bank height (McComb et al. 1990; Pollock et al. 2014).
The selection for low-lying areas next to the stream by beavers may be related to lower banks that provide easy access to forage, but we found no evidence for an effect of bank height in our field-based habitat analysis. Low-lying areas also had sparse woody riparian vegetation, but beavers selecting new settlement sites in our study area preferred stream segments with dense woody riparian vegetation overall. Low-lying areas close to the stream channel may allow beavers to create ponds and backwaters with relatively small dams that then allow them to access foraging areas more efficiently (Pollock et al. 2011). Dense woody riparian vegetation not only offers initial lodge and dam construction material, but also helps in predator avoidance as beavers do not have to go far from the safety of lodges to forage and are better concealed. The selection for stream sections with dense forage resources by beavers is well-supported in the literature (Howard & Larson 1985; Barnes & Mallik 1997; DeStefano et al. 2006).
Channel complexity was the most important variable distinguishing settled from unsettled sites in our field-based habitat analysis. Channel complexity was measured based on the presence of and proximity to side channels, backwaters, and tributaries. These stream features increase the density of feeding areas directly adjacent to water within a colony and provide smaller channels of water that may be easier to dam. Areas with greater channel complexity also help dissipate flood waters which can destroy dams and lodges (Scrafford et al. 2018).
Our results are applicable to other mountainous stream systems in northwest North America, where streams experience annual flooding from snowmelt and beavers are limited by the availability of woody riparian vegetation. Furthermore, our results generally apply to streams and stream sections where beavers are absent because of habitat degradation and not because of geographic isolation from source colonies. There are stream systems in northwest North America that contain high-quality beaver habitat but are not occupied due to dispersing beavers being unable to reach these drainages. We specifically focused on situations where beavers are starting new colonies in suboptimal habitat as this is the most common situation in beaver restoration scenarios. Settlement site selection patterns may be different for beavers recolonizing high-quality habitat.
We found that our suite of geomorphological and biotic habitat conditions explained only 31% of the variation in settlement site selection. Inclusion of a random stream effect significantly improved model fit. The uncertainty around our estimates of fixed habitat effects is likely due to a small sample size of new settlement sites across the study area where many different stream types were represented, and spatial clustering of habitat conditions within streams. Beavers occupying different streams are likely limited by different habitat factors and settlement site selection is likely hierarchical and context dependent. For example, the amount of riparian vegetation may have little influence on settlement probability in large, low-elevation streams where the availability of woody forage is consistently high, but a large influence on settlement probability in smaller headwater streams where woody vegetation is limited. Unfortunately, we were unable to formally evaluate potential interactions and cross-scale relationships in habitat predictors due to a small number of new settlement sites. However, the large improvement in fit of models with random stream effects highlights the relative importance of local habitat conditions within streams.
We observed key differences in habitat selection patterns when we evaluated our new settlement site results against the more common occupied versus unoccupied approach to modeling beaver habitat selection. Results based on a comparison of occupied and unoccupied sites emphasized habitat variables related to vegetation as being the most important factors affecting whether a site was occupied. In contrast, models of new settlement site selection emphasized geomorphological conditions of the stream. Notably, every variable that was included in the top occupied versus unoccupied models can be manipulated by beaver activity. Beavers tend to promote willow regeneration and expansion through selective clipping of vegetation, deposition of sediment behind dams, and raised water tables (Pollock et al. 2018). Beaver colonies also flood new areas and force overbank flow which can increase the number of secondary channels around their colony (Gurnell 1998; Polvi & Wohl 2012). In the long term, beaver occupancy can reshape stream channels, eventually influencing stream sinuosity, gradient, and vegetation communities (Naiman et al. 1988; Westbrook et al. 2006; Polvi & Wohl 2013). Our original hypothesis that occupied versus unoccupied habitat suitability assessments may over-emphasize conditions that are manipulated by beaver occupancy was supported. While there were similarities between the two modeling approaches, the top variables in the new settlement site models may provide a better indication of habitat conditions that are important for beavers starting new colonies in novel areas.
Riparian restoration projects in headwater streams of the Rocky Mountains seeking to leverage the benefits of colonizing beavers should give priority to sites that have signs of previous beaver occupancy. If no previous beaver sign is found, projects should focus on stream segments where beavers have easy access to dense woody riparian vegetation and where initial dams will be most likely to withstand high-water events. Woody riparian vegetation should be available within 30 m of the stream bank and should have a canopy cover of at least 60%. Stream segments with gradients less than 3% are preferred by settling beavers; and sites selected for beaver-mediated restoration should be located in the lowest gradient stream segments available along the stream. Restoration segments should have relatively narrow stream channels and ideally encompass side channels or have these features available within 100 m of the site. The proximity of low-lying areas around the stream that can be easily flooded by a channel-spanning dam should improve the success of restorations, even if these low-lying areas have limited woody riparian vegetation. Though many restoration projects focus on incised streams, beaver dams will frequently blow out in high water if water cannot escape onto the floodplain. In such situations, structural support may be necessary to maintain dams long enough to help rebuild the floodplain (Bouwes et al. 2016). The construction of artificial dams and lodges at sites with suitable conditions should increase colonization by beavers and the overall success of beaver restoration projects.
Acknowledgments
This research was primarily funded by Northwestern Energy, facilitated by the Technical Advisory Committee. Additional funding was provided by the Montana Department of Fish, Wildlife, and Parks, the Montana Agricultural Experiment Station, and the U.S. Forest Service, Hebgen Lake Ranger District. M. Ebinger, N. Rayl, K. Flagg, and the MSU Statistical Consulting Program helped with statistical analyses and R programming. Field technicians and volunteers included K. Szcodronski, C. Langell, D. Howing, A. Micklewright, E. Krieger, L. Macon, S. Wells, A. Netter, M. Ebinger, M. Delehany, T. Sutton, and many others. We thank H. Burt, J. Cunningham, J. Ramsey, J. Gude, and L. Hanauska-Brown for logistic support. The manuscript was improved by comments from the Associate Editor and two anonymous reviewers.
