2007 SOUTHEASTERN NATURALIST 6(2):271–282 Castor canadensis (Beaver) Impoundment Associated with Geomorphology of Southeastern Streams Andrew F. Jakes1,2,3, Joel W. Snodgrass1,2,*, and Joanna Burger2 Abstract - We used a geographic information system (GIS) and logistic regression to investigate relationships between geomorphology and Castor canadensis (North American beaver) impoundment of lower-order, blackwater streams of a southeastern landscape. Using GIS, we divided streams into 632 500-m reaches and measured a set of geomorphic variables for each reach. Beavers were most likely to impound stream reaches crossed by roads with a gradient of 􀂧 0.6 to 1.2% and watershed sizes of 􀂧 2500 ha; reaches with watershed sizes < 􀂧 500 ha or > 5000 ha were almost completely avoided. Gradient and road crossings contributed little to discrimination among impounded and unimpounded reaches, suggesting these variables had relatively small influences on beaver impoundment when compared to stream size. Our results indicate that GIS and geomorphic variables can be used to model the impoundment of streams over larger areas (e.g., the proportion of third-order watersheds impounded), but are less accurate at predicting the impoundment of individual reaches. However, the temporal dynamics of impoundment creation and abandonment will need to be incorporated into region-specific models before they can be used in ecosystem integrity assessment. Introduction Before European colonization, the Castor canadensis Kuhl (North American beaver) population was estimated at between 60-400 million (Naiman et al. 1986, 1988), but beavers were extirpated in many areas of North America during the late nineteenth and early twentieth centuries (Jenkins and Busher 1979). Today, with trapping regulations, a relative absence of natural predators, and assistance from reintroduction programs, beavers are once again occupying many streams and lakes in North America and are of management concern (Larson and Gunson 1983). Beaver activities, such as foraging and dam building, influence ecosystem structure and function and the availability of resources in streams and adjacent terrestrial ecosystems (Naiman et al. 1988, 1991; Smith et al. 1991), thereby defining beavers as ecosystem engineers within lotic systems (Jones et al. 1994). These changes lead to increased biological diversity among fishes, invertebrates, and plants in watersheds occupied by beavers (McDowell and Naiman 1986, Snodgrass and Meffe 1998, Wright et al. 2003). On the other hand, beavers may damage naturally aesthetic and economically important 1Department of Biological Sciences, 8000 York Road, Towson University, Towson, MD 21252. 2Division of Life Sciences, and Consortium for Risk Evaluation with Stakeholder Participation, Environmental and Occupational Health Sciences Institute, Rutgers University, Piscataway, NJ 08854. 3Current address - Montana Fish, Wildlife, and Parks, 1420 East Sixth Avenue, Helena, MT 59620. *Corresponding author - jsnodgrass@towson.edu. 272 Southeastern Naturalist Vol. 6, No. 2 sites by cutting particular species of trees (Beier and Barrett 1987), flooding timber, and damaging human-made structures such as roads and bridges (Arner and Dubose 1980, Bullock and Arner 1985, Hill 1982, McKinstry and Anderson 1999). Efforts to manage beaver populations would benefit from the development of methods and models for predicting beaver impoundment of streams over larger spatial scales. Previous studies of habitat factors affecting beaver impoundment of streams compared field observations and manual map measurements among impounded and free-flowing stream reaches (Allen 1983, Barnes and Mallik 1997, Beier and Barrett 1987, Curtis and Jensen 2004, Howard and Larson 1985, McComb et al. 1990, Slough and Sadleir 1977, Suzuki and McComb 1998). Additionally, some authors have suggested that roads affect the distribution of beavers (McComb et al. 1990, Slough and Sadleir 1977), and models have been developed to predict the occurrence of beavers at roads based on habitat conditions at stream-road intersections and culvert characteristics (Curtis and Jensen 2004, Jensen et al. 2001). In general, these models are of use when assessing the habitat quality of a given reach; however, because these models rely on field and manual map measures that are labor-intensive to gather, it remains difficult to apply them to larger areas (such as an entire third-order watershed, but see Slough and Sadleir 1977). In this paper, we present an initial attempt to quantify and model relationships between beaver impoundment of streams and geomorphic conditions using a geographic information system (GIS), aerial photograph interpretation, and logistic regression. Our goals were: 1) to determine if GIS and measures of stream geomorphology could be used to model beaver impoundment of streams over third-order watersheds, and 2) investigate association of roads with beaver impoundment. We concentrate on geomorphic variables as determinants of the distribution of beaver impoundments along streams because: 1) they can be obtained from existing data or by remote sensing over large spatial scales, 2) they are not likely to be influenced by beaver impoundments (at least at the spatial scale we measured them), and 3) studies of beaver-created patch dynamics using GIS suggest watersheds have a limited amount of stream suitable for impoundment that is largely determined by geomorphology (Johnston and Naiman 1987, 1990a, 1990b; Remillard et al. 1987; Snodgrass 1997). Because the construction of roads often results in floodplain and stream channel constrictions, which might facilitate dam construction (Novak 1987), we included roads in our analysis to assess their relationship with dam-building activity. Study Area We investigated relationships between stream geomorphology and beaver impoundment at the Savannah River Site on the Upper Coastal Plain of South Carolina (hereafter referred to as “the site”). The site is an 80,000-ha Department of Energy nuclear production facility established in the early 1950s. When the site was established, a large protected area was created 2007 A.F. Jakes, J.W. Snodgrass, and J. Burger 273 to serve as a security buffer zone in which natural systems were allowed to recover. Human-made structures on the site make up < 10% of the total area, consisting of a few areas of concentrated production facilities, roads connecting these areas, and roads constructed for timber harvest. The blackwater streams of the site are low-gradient (< 3%), sand-bottom streams that drain primarily from the northeast portion of the site southwest to the Savannah River. Approximately 34 km of the 325 km of stream on the site have received thermal pollution from nuclear reactors at some time in the last 50 years, which has altered riparian vegetation along these reaches. These impacts are primarily confined to higher-ordered streams in the Fourmile Branch, Pen Branch, and Steel Creek watersheds. We restricted our analyses to streams that have not received thermal effluents, including Upper Three Runs, Lower Three Runs, Meyers Branch, and downstream sections of Fourmile Branch, Pen Branch, and Steel Creek. While no estimates of beaver population size exist for the site, studies of pond creation by beavers suggest a recovery from a few colonies in the early 1950s (when the site was established) to at least 37 colonies by 1992 (Snodgrass 1997). Management of the beaver population at the site began in 1983 and focused primarily on fatal trapping of beavers from colonies at road and railroad structures. Although these efforts appear to alter succession of vegetation following impoundment, they do not appear to affect the growth rate or size of ponds associated with trapped colonies (Snodgrass 1997). Therefore, because we focus here on the probability of a reach being impounded based on its geomorphology (not on succession of pond vegetation), management activities probably had little effect on our results. Methods Development of GIS coverages Our general approach to addressing habitat relationships of beaver disturbance involved treating streams as one-dimensional landscape structures (rather than two-dimensional areas) and relating the presence or absence of beaver impoundments along individual stream reaches to geomorphic conditions. To provide a base coverage, we extracted a line coverage of streams for the site from 1:24,000 scale digital line graph (DLG) hydrography data (National Cartographic Information Center, US Geological Survey 1987), and divided streams into 500-m reaches using the ArcHatch Extension of ArcView (Environmental Systems Research Institute, Inc., Redland, CA). In some instances, beaver ponds were located above the terminus of headwater streams or on terminal reaches < 500 m in length. In these cases, we extended terminal streams, based on 3-m topographic contours (1:24,000 scale DLG; National Cartographic Information Center, US Geological Survey 1987), so that all ponds were included in a complete 500-m reach. Additionally, breaks in streams between their upstream terminus and their confluence with other streams were connected based on the same contour coverage. This process resulted in 632 reaches with a mean length of 500 m (SE = 0.02), excluding shorter terminal segments. 274 Southeastern Naturalist Vol. 6, No. 2 To determine presence of beaver impoundments along each 500-m reach, we developed a GIS coverage of beaver ponds on all streams unaffected by site operations using infrared aerial photographs (1:15,840 scale) acquired in January 1998. Photographs were geographically referenced using previously developed coverages of roads, railroads, and pipelines for the site and the Image Warp Extension of ArcView (RMS error m = 3.75 m; n = 46). Potential beaver ponds or previously impounded reaches were located on aerial photographs and visited during the summer of 2000 to verify beaver activity. The presence of dams, lodges, scent mounds, and downed or standing dead trees were considered indicators of past or present beaver activity. We considered reaches occupied if all or a portion of the reach had been impounded during or before 1998, with the exception of reaches at the confluence of smaller tributaries with larger main channels. Smaller tributary reaches that had only their downstream areas flooded as a result of impoundment on the main channel were classified as unoccupied. Because of the recent recovery of the beaver population at the site, it is unlikely we misclassified reaches as unimpounded because of forest recovery following impoundment. We measured four geomorphic variables to assess local habitat characteristics within each 500-m reach: watershed size above each reach (i.e., stream size), stream gradient, floodplain width, and the presence of roads. We delineated watershed boundary above the downstream end of each 500-m reach based on 3-m contours. We used the number of contour lines intersected by each 500-m reach as an index of stream gradient. We recognized four ordinal categories of stream gradient: 0 to 1 intersections (< 􀂧 0.6 %); 2 intersections (> 􀂧 0.6 to 􀂧 1.2%); 3 intersections (> 􀂧 1.2 to 􀂧 1.8%); and > 3 intersections (> 􀂧 1.8 %). Floodplain widths were measured at the downstream end of each 500-m reach. A coverage of soil types for the site, sorted by hydrologic class, was utilized to decipher floodplains adjacent to streams. For reaches that had well defined floodplains along only a portion of their length (i.e., flowed from upland areas on to the floodplains of higher-order streams), floodplain width was measured at the downstream end of the upland area. Reaches located entirely within the floodplains of higher-ordered streams (< 1% of the reaches) were assigned an extreme value of 9999 m. The presence or absence of road crossings was noted for each reach and coded as a dummy variable (0 for absence and 1 for presence). Statistical analyses We used logistic regression following the methods outlined by Manly et al. (1993) and Boyce and McDonald (1999) to evaluate relationships between the probability of impoundment and geomorphic characteristics of each 500-m reach. We used the stepwise logistic regression procedure of SAS (SAS Institute 1989) and maximum-likelihood methods for estimating model parameters. Independent variables considered as potential predictors of the probability of impoundment included watershed size, floodplain width, gradient index, and presence/absence of road crossings. Gradient index and presence/absence of road crossing were coded as dummy variables because of 2007 A.F. Jakes, J.W. Snodgrass, and J. Burger 275 their discrete nature. We also included second-order polynomial terms for watershed size and floodplain width in the set of potential independent variables because we anticipated possible unimodal responses of beavers to these habitat characteristics. For example, beavers are not likely to impound streams with large or small watersheds because of intermittent surface water conditions and physical constraints on the amount of discharge that can be impounded, respectively. We set the level for entry or retention in the model at P 􀂔 0.05. Wald-tests were used to test the significance of individual variables. Because our data represent a census (rather than a sample) of all stream reaches unaffected by site operations, there was no need to correct the intercept term in the final relationship for disproportional sampling efforts (Manly et al. 1993). To investigate overall fit of the final model, fit in relation to individual observations, and the influence of individual observations on parameter estimates, we followed methods presented by Hosmer and Lemeshow (1989). The overall significance of the final relationship was tested using a G-statistic (Sokal and Rohlf 1995). We investigated deviation of individual observations from predicted responses and their influence on parameter estimates using plots of changes in deviance (􀂨D), Pearson chi-squared statistic (􀂨􀁲2), and the standardized differences in the parameter estimates (􀂨􀁠) resulting from deleting individual observations versus predicted probability of a reach being impounded. Large values of 􀂨D and 􀂨􀁲2 indicate poorly fit observations, and large values of 􀂨􀁠 indicate observations with undue influence on parameter estimates. We assessed model performance from two respects: 1) the ability to predict the impoundment of individual reaches (i.e., discriminatory ability); and 2) the ability of the model to predict the percentage of small watersheds impounded. To assess the discriminatory ability of the model, and the contribution of individual variables to discriminatory ability, we compared the full model to models containing only subsets of the independent variables included in the full model. We used a one-step approximation to estimate sensitivity (i.e., the ability of the model to correctly predict impoundment) and specificity (i.e., the ability of the model to correctly predict non-impoundment) for each model. We selected the cut point for classification to maximize sensitivity and specificity simultaneously (i.e., so they were nearly equal). We also calculated a c-statistic for each model; c ranges from 0.5 to 1.0, where 0.5 indicates a model with discriminatory ability no greater than random. To validate the model and assess the ability of the model to predict the percentage of small watersheds impounded, we withheld 100 randomly selected reaches from the model building process and compared the predicted percentage of stream impounded to the observed percentage for this set of reaches, where the predicted percentage of stream impounded was calculated as the mean probability of impoundment multiplied by 100. We assessed the contribution of individual variables to predictive power at the watershed scale in the same way we assessed contribution to discriminatory ability. 276 Southeastern Naturalist Vol. 6, No. 2 Results Of the 632 reaches assessed, 501 (79.3%) showed no signs of beaver impoundment and were classified as unoccupied, while 131 reaches (20.7%) showed evidence of present (81%) or past (19%) impoundment and were classified as impounded during or before 1998. Watershed size above reaches ranged from 15 to 51,374 ha (mean = 5056 ha; SD = 11,880). Floodplain widths for those reaches with defined floodplains ranged from 34 to 1507 m (mean = 209 m; SD = 230), with the majority of stream reaches having floodplain widths < 200 m (65%). Overall, stream reaches had relatively low gradients; 28.5%, 32.2%, 19.8%, and 19.5% of the reaches had gradients scores of 1, 2, 3, and 4, respectively. Twenty-two percent of stream reaches were crossed with roads. In general, there was an increase in floodplain width and a decrease in gradient from headwater stream reaches (with small watersheds) through downstream reaches (with larger watersheds). Stream reaches with watershed sizes > 1000 ha were all low gradient (i.e., gradient index was 1 or 2). Streams with a gradient index of 1 had wider floodplains than streams with higher gradients. The majority of streams with gradient indices of 3 or 4 had floodplain widths of < 600 m (Fig. 1) and were restricted to headwater streams (i.e., had watershed sizes of 500 ha or less). Roads were evenly distributed in regards to watershed size (no roads: mean = 4981 ha, SD = 11,603; roads: mean = 5322, SD = 12,854; t-test with correction for unequal variance: P = 0.778), floodplain widths (no roads: mean = 1146 m, SD = 2885; roads: mean = 970 m, SD = 2663; t-test with correction for unequal variance: P = 0.500) and gradient (chi-squared test of no relationship between road presence/absence and gradient index: P = 0.251). In general, beavers were more likely to impound moderate-sized (watershed size 1000 to 5000 ha), lower-gradient streams (gradient index of 1 or 2) that were crossed by roads (Fig. 1). Stepwise logistic regression retained watershed size, its second-order polynomial, gradient index, and the presence or absence of road crossings in the full model (Table 1). Overall, the model fit the data (RL 2 = 0.73, P < 0.001; Fig. 1). Inspection of plots of 􀂨􀁠 versus predicted probability of impoundment indicated no single observation had undue weight on parameter estimates (i.e., the estimates of parameters in the model were stable). However, plots of 􀂨D and 􀂨􀁲2 versus predicted probability of impoundment indicated that the model poorly predicted seven observations. These observations were all impoundments on small streams with high gradients (i.e., gradient indices of 3 or 4), and we interpret them as unusual, yet biologically reasonable, outcomes since impoundments on higher-gradient streams are likely to be more ephemeral (J.W. Snodgrass, unpubl. data). The highest probability of impoundment (􀂧 0.78) occurred among streams reaches crossed by roads with gradient indices of 2 (􀂧 0.6 to 1.2%) and watershed sizes of 􀂧 2500 ha (Fig. 1). Although reaches with gradient indices of 2 had higher probabilities of impoundment, they were not significantly different from reaches with gradient indices of 1. Reaches with watershed sizes < 􀂧 500 ha or > 􀂧 5000 ha, or with gradient indices of 3 or 4, had low probability of being 2007 A.F. Jakes, J.W. Snodgrass, and J. Burger 277 impounded. Although the model predicted higher probabilities of impoundment for reaches with watershed sizes of 􀂧 2500 ha and gradient indices of 3 or 4, these conditions did not occur in the study area, and most reaches (94%) with gradient indices of 3 or 4 had watershed sizes < 500 ha and predicted probabilities of impoundment < 0.24. The observed frequency of impoundment among all reaches with gradient indices of 3 or 4 was 0.08. Among reaches with gradient indices of 1 or 2, the presence of one or more road crossings increased the probability of impoundment by 􀂧 10% when watershed sizes were near 2500 ha (Fig. 1). While the full model fit the data well (i.e., predicted the percentage of streams impounded in relationship to watershed size, gradient, and road crossings), the discriminatory ability of the model was marginal. For the full model c = 0.84, sensitivity was 74%, specificity was 75%, and 75% of the reaches were correctly classified at a cut point of 0.23. Among the 100 reaches withheld from the model building process, 84% were correctly classified with 2 impounded reaches classified as unimpounded and 12 unimpounded reaches classified as impounded. The model was better at predicting the percentage of a watershed that would be impounded (measured as the mean probability of impoundment for all reaches in the watershed); of the 100 reaches withheld from the model building process 18.0% were impounded, and the model predicted 18.7% impoundment. Figure 1. Model predictions (lines) of impoundment of streams by beavers as a function of watershed size above stream reaches, presence or absence of roads crossings, and gradient for relatively low-gradient streams of South Carolina. Streams in gradient class 1 had gradients of 0 to 􀂧 0.6%, and streams in gradient class 2 had gradients of > 􀂧 0.6 to 1.2%. 278 Southeastern Naturalist Vol. 6, No. 2 Examination of reduced models indicated no interaction between any of the terms included in the full model, and watershed size was the most important variable in discriminating among impounded and free-flowing reaches. The reduced model containing only watershed size and its polynomial term (Table 1) correctly predicted 76% of the reaches as to beaver impoundment; sensitivity, specificity and c for the reduced model were 0.74, 0.76, and 0.83, respectively. Addition of other terms to the reduced model containing only watershed size did not improve the discriminatory ability of the model. The reduced model did predict slightly higher occurrence of impoundment at the watershed scale (19.7%) than was actually observed (18.0%). Addition of gradient or presence/absence of road crossing resulted in slight improvement of model prediction to 􀂧 18.7% impounded. Discussion Relationship to other studies In the southeastern landscape that we studied, beavers tended to impound streams with low gradients, intermediate watershed sizes, and with roads crossing them (as opposed to reaches without road crossings). However, stream size (as measured by watershed size) appeared to be the most important factor in determining impoundment probability among the geomorphic variables we considered. In general, our results are consistent with past studies and the view of beaver habitat preference as a tradeoff between maximizing pond size and physical constraints of impoundment of larger discharges (Johnston and Naiman 1990b). Watershed size, or its correlates (stream width, average annual water flux, stream depth, and cross-sectional area), have been reported as both positively (Beier and Barrett 1987, Howard and Larson 1985) and negatively (Barnes and Mallik 1997, Slough and Sadleir 1977) correlated with beaver impoundment, and Suzuki and McComb (1998) incorporated a unimodal response of beavers to stream width in their habitat suitability model. Therefore, although beavers do not directly use watershed size to select locations for dam construction, for Table 1. Results of stepwise logistic regression analysis of relationships among road crossings, stream gradient, watershed area, and the probability of beaver impoundment within 500-m stream reaches. The full model contains all of the independent variables and the reduced model contains only those variables contributing substantially to the discriminatory ability of the model. Model/Parameter DF Estimate SE 􀁲2 P Full intercept 1 -1.91600 0.3759 25.9769 < 0.001 Gradient (2 vs. 1) 1 0.22630 0.3083 0.5386 0.463 Gradient (3 vs. 1) 1 -0.72280 0.4232 2.9173 0.088 Gradient (4 vs. 1) 1 -1.72210 0.6173 7.7839 0.005 Road crossing 1 0.71550 2.89E-01 6.1211 0.013 Watershed size (ha) 1 0.00183 0.000367 24.8371 < 0.001 Watershed size (ha2) 1 -0.0000004 8.42E-08 19.2357 < 0.001 Reduced intercept 1 -2.36810 0.2059 132.2495 < 0.001 Watershed size (ha) 1 0.00255 0.000334 58.1679 < 0.001 Watershed size (ha2) 1 -0.0000005 8.21E-08 37.7888 < 0.001 2007 A.F. Jakes, J.W. Snodgrass, and J. Burger 279 modeling purposes, watershed size may act as a surrogate for more proximate geomorphic factors directly selected by beavers. The relationship between stream gradient and beaver impoundment in our study may reflect a lack of water availability in small streams rather than avoidance of relatively high gradient streams by beavers. In landscapes with relatively large amounts of topographic relief (i.e., some streams with gradients > 3 %), gradient may be an important factor in beaver habitat selection (Beier and Barrett 1987, Curtis and Jensen 2004, Howard and Larson 1985, McComb et al. 1990, Slough and Sadleir 1977, Suzuki and McComb 1998). However, in lower gradient streams (< 1.5% gradient) of topographically flat terrain, stream gradients vary little and are not related to beaver impoundment (Barnes and Mallik, 1998). In our study, streams with gradients > 1.5% (i.e., gradient indices of 3 or 4) were restricted to extreme headwaters and included in our analysis because they appeared on USGS hydrography data. These relatively high gradient streams were intermittent in most cases and may not have been impounded for lack of water. Like stream gradient, the presence or absence of road crossings on a reach was significantly related to the probability of impoundment. Our results do suggest that road crossings that partially obstruct stream flow and/ or a portion of the floodplain increase beaver habitat quality within the smaller streams preferred by beavers at the site. When culverts associated with roads are small and reduce channel area, resulting in slower flows upstream of the culvert, the energy required to dam the stream may be reduced, and beavers will locate dams at culverts (Jensen et al. 2001). Scope and limitations Our model predicted a maximum of 􀂧 80% impoundment under optimal geomorphic conditions, resulting in the selection of a cut point of 0.23 for classification purposes. By selecting a cut point lower than 0.50 we essentially emphasized misclassification of unimpounded reaches as impounded; of the 16 reaches that were misclassified among the validation data set, only 2 were incorrectly classified as unimpounded. The lack of impoundments on reaches that were apparently suitable may have resulted from failure to consider vegetative characteristics that may influence beaver impoundment in our study area. Although authors have often reported geomorphic characteristics as being most important in determining beaver dam location (Howard and Larson 1985, McComb et al. 1990), others have found the availability of forage and preferred dam building material to be important factors in determining dam location (Barnes and Mallik 1997, Curtis and Jensen 2004, Slough and Sadleir 1977). Because our study area did not include land-use practices that resulted in complete forest clearing along streams, a lack of woody vegetations along streams probably did not limit beaver impoundment. However, more subtle differences in forest characteristics such as stem size and forest composition may have influenced impoundment probabilities in our area. These characteristics will be difficult to incorporate into models designed to predict impoundment over larger scales, and may limit the application of our approach. 280 Southeastern Naturalist Vol. 6, No. 2 It is possible that beaver populations in our study area are not at carrying capacity. If the population is still expanding and dispersal is limited (i.e., all reaches are not readily accessable from existing colonies), some reaches may not have been impounded because of isolation from points of colonization (Bergerud and Miller 1977). Furthermore, access to some reaches may be limited due to the territorial behavior of beavers. Beaver colonies consisting of an adult male, an adult female, and several offspring maintain one to several dams and aggressively defend reaches of streams against dispersers (Hill 1982, Nolet and Rosell 1994). This territorial behavior often results in separation of territories by 100 m or more of unoccupied stream (Bergerud and Miller 1977). Johnston and Naiman (1990a) documented an asymptote of pond area over a 46-year period in a northern Minnesota landscape. While beaver impoundment of streams was evident in 1978 at our study site (Snodgrass 1997), sufficient data to assess equilibrium of beaver patch dynamics is lacking. Therefore, further model development will require assessment of population and impoundment dynamics over longer periods to assess the equilibrium assumptions of our modeling approach (Manly et al. 1993). The accuracy of the model depended on the scale of the prediction being made; at the individual-reach scale, accuracy was relatively low (86% correctly classified in the validation data set), but was higher at the watershed scale (18.0% predicted and 18.7% observed in the training data set). Therefore, use of GIS-based geomorphology models of beaver impoundment may be most appropriate at the watershed scale. For example, models could be used to assess the degree of ecological integrity exhibited by a watershed; if percent impoundment of streams is similar to that predicted by models, a watershed might be considered to have high ecological integrity from the standpoint of beaver disturbance. However, because positive influences of beaver impoundment on biological diversity are dependent on the temporal dynamics of pond creation and abandonment (Snodgrass and Meffe 1998), development of models that incorporate these dynamics will be required for the complete incorporation of beaver disturbance into ecological assessments. Finally, while our general approach to developing models of beaver disturbance of streams may be exportable to other areas, the specific relationships and parameter estimates are likely to vary greatly among regions with different geomorphology. For example, beavers in streams of the Coastal Range of Oregon preferred streams with gradients of 1 to 3% (Suzuki and McComb 1998). In contrast, beavers at our study site on the upper coastal plain of South Carolina preferred streams with gradients of < 0.6%. Therefore, models will need to be developed individually for regions differing in geomorphology (Barnes and Mallik 1997). Acknowledgments This project was funded by the Consortium for Risk Evaluation with Stakeholder Participation (CRESP; through the Department of Energy Cooperative Agreement, AIDE-FC01-95EW55084, DE-FG-26-00NT 40938), NIEHS Grant 2007 A.F. Jakes, J.W. Snodgrass, and J. Burger 281 ESO 5022, the Environmental and Occupational Health Sciences Institute; Financial Assistance Award No. DE-FC09-96SR18546 from the US Department of Energy to the University of Georgia Research Foundation; and a grant from the Towson University Graduate Student Association. We thank Deno Karapatakis and Helen Wiggins-Brown for assisting with GIS coverage development. Logistic support at the Savannah River Site was provided by Larry Bryan, William Hopkins, John Roe, and Gary Wein. We thank Don Forester, Karen Gaines, Scott Johnson, Brenda McComb, Marty Roberge, and two anonymous reviewers for providing helpful insights on earlier versions of the manuscript. Literature Cited Allen, A.W. 1983. Habitat suitability index models: Beaver. US Department of the Interior, Fish Wildlife Service, Washington, DC. FWS/OBS-82/10.30 Revised. 20 pp. Arner, D.H., and J.S. Dubose. 1980. The impact of the beaver on the environment and economics in the southeastern United States. Proceedings of International Wildlife Conferences 14:241–247. Barnes, D.M., and A.U. Mallik. 1997. Habitat factors influencing beaver dam establishment in a northern Ontario watershed. Journal of Wildlife Management 61:1371–1377. Beier, P., and R.H. Barrett. 1987. Beaver habitat use and impact in Truckee River Basin, California. Journal of Wildlife Management 51:794–799. Bergerud, A.T., and D.R. Miller. 1977. Population dynamics of Newfoundland beaver. Canadian Journal of Zoology 55:1480–1492. Boyce, M.S., and L.L. McDonald. 1999. Relating populations to habitats using resource selection functions. Trends in Ecology and Evolution 14:268–272. Bullock, J.F., and D.H. Arner. 1985. Beaver damage to nonimpounded timber in Mississippi. Southern Journal of Applied Forestry 9:137–140. Curtis, P.D., and P.G. Jensen. 2004. Habitat features affecting beaver occupancy along roadsides in New York State. Journal of Wildlife Management 68:278–287. Hill, E.P. 1982. Beaver. Pp. 256-281, In J.A. Chapman, and G.A. Feldhamer (Eds.). Wild Mammals of North America: Biology, Management, and Economics. Johns Hopkins University Press, Baltimore, MD. 1147 pp. Hosmer, D.W., and S. Lemeshow. 1989. Applied Logistic Regression. John Wiley and Sons, New York, NY. 373 pp. Howard, R.J., and J.S. Larson. 1985. A stream habitat classification systems for beaver. Journal of Wildlife Management 49:19–25. Jenkins, S.H., and P.E. Busher. 1979. Castor canadensis. Mammalian Species 120:1–8. Jensen, P.G, P.D. Curtis, M.E. Lehnert, and D.L. Hamelin. 2001. Habitat and structural factors influencing beaver interference with highway culverts. Wildlife Society Bulletin 29:654–664. Johnston, C.A., and R.J. Naiman. 1987. Boundary dynamics at the aquatic-terrestrial interface: The influence of beaver and geomorphology. Landscape Ecology 1:47–57. Johnston, C.A., and R.J. Naiman. 1990a. Aquatic patch creation in relation to beaver population trends. Ecology 71:1617–1621. Johnston, C.A., and R.J. Naiman. 1990b. The use of geographical information system to analyze long-term landscape alteration by beaver. Landscape Ecology 4:5–19. Jones, C.G., J.H. Lawton, and M. Shachak. 1994. Organisms as ecosystem engineers. Oikos 69:373–386. 282 Southeastern Naturalist Vol. 6, No. 2 Larson, J.S., and J.R. Gunson. 1983. Status of the beaver in North America. Acta Zoologica Fennica 174:91–93. Manly, B.F.J., L.L. McDonald, and D.L. Thomas. 1993. Resource Selection by Animals: Statistical Design and Analysis for Field Studies. Chapman and Hall, London, UK. 221 pp. McComb, W.C., J.R. Sedell, and T.D. Buchholz. 1990. Dam-site selection by beavers in a eastern Oregon basin. Great Basin Naturalist 50:273–281. McDowell, D.M., and R.J. Naiman. 1986. Structure and function of a benthic invertebrate stream community as influenced by beaver (Castor canadensis). Oecologia 68:481–489. McKinstry, M.C., and S.H. Anderson. 1999. Attitudes of private and public land managers in Wyoming, USA, toward beaver. Environmental Management 23:95–101. Naiman, R.J., J.M. Melillo, and J.E. Hobbie. 1986. Ecosystem alteration of boreal forest streams by (Castor canadensis). Ecology 67:1254–1269. Naiman, R.J., C.A. Johnston, and J.C. Kelley. 1988. Alteration of North American streams by beaver. BioScience 38:753–762. Naiman, R.J., T. Manning, and C.A. Johnston. 1991. Beaver population fluctuations and tropospheric methane emission in boreal wetlands. Biogeochemistry 12:1–15. Nolet, B.A., and F. Rosell. 1994. Territoriality and time budgets in beavers during sequential settlement. Canadian Journal of Zoology 72:1227–1237. Novak, M. 1987. Beaver. Pp. 282–313, In M. Novak, J.A. Baker, M.E. Obbard, and B. Malloch (Eds.). Wild Furbearer Management and Conservation in North America. Ashton-Potter, Concord, ON, Canada. 1150 pp. Remillard, M.M., G.K. Gruendling, and D.J. Boqucki. 1987. Disturbance by beaver (Castor canadensis Kuhl) and increased landscape heterogeneity. Pp. 103–123, In M.G. Turner (Ed.). Landscape Heterogeneity and Disturbance. Springer Verlag, New York, NY. 239 pp. SAS Institute, Inc., 1989. SAS/STAT User’s Guide, version 6. SAS Institute, Inc., Cary, NC. Slough, B.G., and R.M.S. Sadleir. 1977. A land capability classification system for beaver (Castor canadensis Kuhl). Canadian Journal of Zoology 55:1324–1335. Smith, M.E., C.T. Driscoll, B.J. Wyskowski, C.M. Brooks, and C.C. Cosentini. 1991. Modification of stream ecosystem structure and function by beaver in the Adirondack Mountains, New York. Canadian Journal of Zoology 69:55–61. Snodgrass, J.W. 1997. Temporal and spatial dynamics of beaver-created patches as influenced by management practices in a southeastern North American landscape. Journal of Applied Ecology 34:1043–1056. Snodgrass, J.W., and G.F. Meffe. 1998. Influence of beavers on stream fish assemblages: Effects of pond age and watershed position. Ecology 79:928–942. Sokal, R.R., and F.J. Rohlf. 1995. Biometry. Third Edition. W.H. Freeman and Company, New York, NY. 887 pp. Suzuki, N., and W.C. McComb. 1998. Habitat classification models for beaver (Castor canadensis) in the streams of the Central Oregon Coast Range. Northwest Science 72:102–110. US Geological Survey. 1987. Digital line graphs from 1:24,000-scale maps: Data users guide. US Geological Survey, Reston, VA. Wright, J.P., A.S. Flecker, and C.G. Jones. 2003. Local vs. landscape control on plant species richness in beaver meadows. Ecology 84:3162–3173.