The beaver (Castor canadensis) is an ecologically important
keystone species capable of altering plant communities and creating
vital wetlands through herbivory and water impoundment (Broschart et.
al., 1989). Beavers are valued for their fur, but can be costly
nuisances when their foraging damages valuable trees and dams cause
flooding of roads and crops. Beaver populations have increased
dramatically in Illinois during the past 30 years, but they are not
distributed uniformly across the state and abundance varies regionally
(Hoffmeister, 1989). Previous research showed that aerial surveys can be
useful tools for comparing the relative abundance of this species in
some watersheds (Woolf et al., 2003). However, these surveys can be
problematic in stream habitats (which account for ~84% of all beaver
habitat in southern Illinois) because fluctuating water level and tree
canopy often obscure bank dens and food caches reducing the
detectability of beavers (Woolf et al., 2003).
We investigated whether beaver-habitat models could be useful
alternatives for predicting habitat quality and relative abundance of
beavers in Illinois' streams. Our objectives were to: (1) map and
quantify the distribution of beaver colonies, (2) test the efficacy of
two existing models for predicting relative abundance, the Habitat
Suitability Index (HSI) model and a Missouri habitat model, and (3)
develop a new logistic regression model to predict relative abundance of
colonies in Illinois' rivers.
This study was conducted on the Embarras River in east-central
Illinois, one of 9 major watersheds in Illinois. The river is typical of
many Illinois rivers, is moderate-sized with a low gradient, and drains
a large, flat watershed heavily impacted by agriculture. Water levels
can fluctuate dramatically in part because the agricultural drainage
systems move water quickly from crop fields. Over 50% of the river is
classified as "biologically significant", a designation
reserved for Illinois' highest quality streams (Wiggers, 1998).
Measuring habitat variables
Locations of all beaver colonies were mapped during November
2001--February 2002 when bank dens, food caches, and chewed trees were
most evident. The entire river was searched thoroughly during this
period by canoe and on foot. Most colonies were identifiable based on
the presence of dens in close proximity to food caches. When dens were
not visible, a colony was defined as a stream segment >300 m in
length with fresh sign (Robel et al., 1993). The location of each active
colony was recorded in Universal Transverse Mercator (UTM) coordinates
using a global positioning system (GPS) and marked on a 7.5-minute
United States Geological Survey (USGS) topographic map.
Variables that could influence the quantity and quality of beaver
habitat were selected a priori based on natural history and previous
habitat models. These were measured in 26 2.5-km segments of river
selected using a stratified-random scheme. First, the river was divided
into upper, middle, and lower divisions, then each division was divided
into 25-km sections, and two 2.5-km segments within each section were
randomly selected for sampling. The UTM coordinates delineating the
beginning and end of each segment were recorded from USGS maps and these
were located in the field using GPS. In each segment, the number of
colonies (dependent variable) and set of habitat variables (independent
variables) were quantified.
Between June and August, we sampled vegetation in each river
segment using five 100-m transects located perpendicular to the river at
500-m intervals on alternating banks. Sample points were established
along each transect at 10-m intervals and species composition, diameter,
and canopy cover of trees were estimated using the point-quarter
technique and a densiometer. Shrub cover and height were estimated using
the line-intercept method and a height pole (Cox, 2002). We measured
channel width, bank height, and composition (silt, sand, or clay) at
each transect. Width of the riparian zone, presence of agriculture
fields, and number of roads within 200 m of the river, were measured on
georectified aerial photographs. Stream gradients were extracted from
the Illinois Stream Identification System (ISIS) database developed by
the Illinois Department of Natural Resources (IDNR).
Testing habitat models
We first tested the U.S. Fish and Wildlife Service's habitat
suitability index (HSI riverine habitat model) for beavers (Allen,
1983). Variables used in the model include: stream gradient, average
water fluctuation (m), % canopy closure, % trees in the 2.5-15.2 cm
diameter at breast height (dbh) class, % shrub cover, shrub height, and
woody species composition within 200 m of the stream. We calculated the
HSI score for each of 26 segments using the mean of transect data.
Average water fluctuation was estimated based on minimum and maximum
flow rates reported by the USEPA at gauging stations in Camargo, Ste.
Marie, and Lawrenceville, IL. We used simple linear regression to test
the relationship between HSI habitat scores and the number of colonies
in each segment to test the efficacy of the HSI model for predicting
relative abundance of beavers.
Next, we tested a model developed to quantify habitat suitability
in the bottomland forests of Missouri (the Missouri model; Hallett and
Erickson, 1980). To our knowledge, this is the only model designed
specifically for use in the riverine habitats of the Midwest. However,
its validity had not been tested prior to our study. Variables used in
the model include: bank texture and slope, tree species composition and
dbh, number of important food plants, proximity of crop fields, and
presence of permanent water. The model provides the user with the option
of removing habitat characteristics not applicable to a site. Because
all of the river segments provided permanent water, we deleted the
latter variable from the model and adjusted scores accordingly (Hallett
and Erickson, 1980). Again, the relationships between habitat scores and
the number of colonies in each segment were tested using linear
Finally, we developed a new beaver habitat model using forward
logistic regression to determine which of 12 independent habitat
variables (Table 1) could be used to predict the presence/absence of
beavers in each 2.5-km stream segment. To avoid potential
multi-collinearity between variables in the model, we first conducted
Spearman correlation analyses and eliminated 3 variables that were
highly correlated (P < 0.05) with other more easily measured
variables. We set the threshold necessary for a variable to enter the
model at [less than or equal to] 0.15 so as not to exclude any that
might be biologically important to beavers. Each of the four variables
included in the final model were accompanied by a significant (P <
0.1) change in the F-value associated with the overall regression.
Spearman correlations and logistic regression analyses were performed
using SPSS software (SPSS Inc. Chicago, IL).
Density and spatial distribution of colonies
We located and mapped 125 colonies on the Embarras, a mean of 0.40
colonies/ km (Figure 1). Based on nearest-neighbor distances, colonies
tended to be uniformly distributed along the river, with a
disproportionate number occurring approximately 1-km apart ([X.sup.2] =
32.6; P < 0.01; Figure 2). The minimum distance between adjacent
colonies was 400 m.
Twenty of the 26 (76.9%) river segments contained active beaver
colonies; only six segments lacked colonies. Of the segments with
colonies, nine had a single colony and 11 contained two colonies. The
majority (97.6%) of colonies occupied bank dens; only three occupied
lodges and these were all located in the headwaters of the river.
Similarly, only two dams were found on the main channel of the river,
both in the upper reaches where the river was narrow and flow was low.
Of the habitat characteristics measured, only stream gradient,
correlated significantly with colony density (r = -0.440, P = 0.024).
The gradient was lowest in the upper reaches and near the river mouth
where densities tended to be high. In contrast, middle sections of the
river had the highest gradient and colonies were sparse. Several other
habitat parameters approached statistical significance, including
percentage of the river with low banks (r = 0.363, P = 0.068), riparian
width (r = 0.355, P = 0.075), shrub cover (r = 0.351, P = 0.079), and
canopy cover (r = 0.337, P = 0.092).
Testing existing habitat models
HSI scores ranged from 0.0 to 1.0, with a mean = 0.82 (SD = 0.28),
suggesting that the quality of beaver habitat varied considerably along
the river, but generally was good. The only segment with unsuitable
habitat (HSI = 0.0) had no beavers present and the segments with highest
colony densities had optimal habitat according to the model. However,
overall HSI scores did not correlate well with the number of colonies
([r.sup.2] = 0.111, P = 0.588). For example, four segments with
excellent habitat (HSI > 0.8) had no colonies, whereas six segments
with only moderate habitat (HSI ~ 0.5) contained high densities.
Habitat scores derived from the Missouri model correlated
significantly with colony density ([r.sup.2] = 0.578; P = 0.002). Scores
ranged from 43% to 71% (mean = 59%, SD = 6.7), again suggesting habitat
along much of the river was good. Segments with highest scores also had
the greatest number of colonies. Segments lacking colonies earned scores
ranging from 43% to 60%. Segments with lowest scores lacked a forested
riparian zone and provided little winter food after crops were
harvested. Variables that most influenced habitat scores on the Embarras
River were size class of trees and bank texture. Segments dominated by
large, mature trees or with sandy banks unsuitable for dens received low
Logistic regression model
We developed a regression model that retained four independent
variables: % riparian trees <15 cm dbh, riparian zone width, stream
gradient, and number of roads within 200 m. The resulting standardized
regression coefficients (Tree < 15 B = 0.075, RZW B = 0.030, Grad B =
-0.760, Roads B = -0.723, and a constant of -0.642) indicated the
relative importance of each variable to the model. Probability of a
stream segment being occupied by beavers increased with relative
abundance of small trees and wide riparian zones and decreased with
stream gradient and road density. The resulting model was a significant
predictor of the presence/absence of beavers ([r.sup.2] = 0.58, P =
0.014) and successfully predicted their presence/absence in 24 of 26
(92%) stream segments, including all 20 where beavers were present and
four of six segments where beavers were apparently absent.
Density and spatial distribution of colonies
With a mean of 0.40 colonies/km of stream, the Embarras River
provides good quality beaver habitat along most of its length. Robel et
al. (1993) found that rivers with good beaver habitat had densities of
0.12 to 1.40 colonies/km in Kansas and Semyonoff (1951 in Novak, 1987)
found mean densities of 1.5 colonies/km for rivers with good habitat,
0.5 colonies/km in moderate habitat, and 0.1-0.2 colonies/km in poor
Distribution of colonies varied along the length of the watershed,
reflecting changing environments along the river. Colony density was
highest in the headwaters characterized by slow moving water, a narrow
channel, and a broad floodplain that provided beavers with the
opportunity to build lodges, dams, and bank dens. Topography is very
flat and prone to flooding; consequently, farmers have removed some low
areas from crop production and these ephemeral wetlands provide refuge
and habitat during periods of flooding and drought. As flows and channel
width increase downriver, beavers build fewer dams and are less capable
of altering their local environment to create preferred habitat.
The Embarras watershed, like much of central Illinois, is dominated
by corn and soybean fields. Robel et al. (1993) found that beavers in
Kansas were as likely to forage on corn and sorghum as preferred trees
such as cottonwood and willows. Beavers in our study area fed on corn
and soybeans in the fall and corn stalks were evident in many food
caches. However, after crop harvests, the landscape changes and the use
of woody vegetation by beavers increased. Consequently, during winter,
availability, composition, and stem size of woody plants likely
influences habitat quality (Boyce, 1981). We found beavers to be most
abundant in stream segments where periodic flooding maintained
early-successional riparian forests dominated by small diameter trees.
Generally, as bank height and channel volume increased in lower
portions of the watershed, the proportion of large trees (>45 cm dbh)
increased and woody understory decreased, as did beaver density. Beavers
inhabiting the lower portion of the Embarras River have adapted to
fluctuating water levels. Trails from the river into cornfields and
foraging areas frequently extended up steep banks and den openings were
stacked vertically allowing use of different den openings depending on
Beavers are highly territorial and social interactions should lead
to a uniform dispersion in suitable habitat assuming resources are
evenly distributed (Davies, 1978). Uniform spacing of colonies along the
Embarras River, particularly in the middle and lower portions suggests
that territoriality, rather than resource limitation, is an important
factor influencing distribution. A greater proportion of colonies
occurred approximately one km apart than would be expected by chance.
This is consistent with reported home ranges of approximately 0.8 km on
streams (Nordstrom, 1972). Busher (1983) reported intercolony distances
ranged from 0.84-1.55 km in California streams.
Testing existing habitat models
Habitat models have proven to be useful tools for quantifying
habitat quality and relative abundance of beavers in streams and
wetlands throughout North America (Slough and Sadlier, 1977; Allen,
1983; Howard and Larson, 1985; Broschart et al., 1989). We tested
Allen's (1983) HSI model because it is widely used for
environmental impact assessments throughout the U.S. The model was not
developed specifically to predict beaver densities, Robel et al. (1993)
noted that HSI scores should be positively correlated with beaver
densities if the model is composed of key habitat variables. Robel et
al. (1993) and Stromayer (1999) reported poor performance for the HSI
model in the midwestern and eastern U.S., respectively. Model
limitations in these regions included its failure to incorporate local
plant species as high quality foods and narrow definitions of suitable
water quality and stream substrates.
The HSI model did not produce useful estimates of beaver density on
the Embarras River and probably is not useful for estimating relative
abundance of beavers in Illinois' watersheds. Correlations between
HSI scores and colony densities were low ([r.sup.2] = 0.111; P = 0.588)
in part because the model is based on characteristics more typical of
beaver habitat in the northern and western portions of the geographic
range, emphasizing winter foods and stream characteristics not typical
of Illinois. For example, the model fails to incorporate regional foods
such as corn, maple, and ash, and it defines suitable stream
characteristics too narrowly, particularly water levels and substrates
(Robel et al., 1993; Stromayer, 1999). On the Embarras River, beavers
have adapted to fluctuating water levels and steep banks, as long as
water depth is sufficient for travel and protection.
In contrast, the Missouri model proved well-suited for predicting
the quality of beaver habitat in Illinois. Model variables such as bank
characteristics, forest age and composition, important food plants, and
distance to cropland are appropriate descriptors of beaver habitat in
Illinois. The model captured the importance of tree species, size, and
regional food plants (including crops), to beavers in the Embarras River
watershed. Our results suggest that the Missouri model could be a useful
tool for evaluating the quality of beaver habitat in Illinois and
advances in remote sensing and GIS systems will allow future refinement
of statewide habitat maps.
Logistic regression model
Our final regression model was a significant predictor of beaver
presence, retaining variables that we believe represent important
habitat components of Illinois streams. Of 12 habitat characteristics
entered into the regression, four were retained in the final model. Two
(abundance of small trees and riparian zone width) were positively
associated with beaver colonies and two (stream gradient and proximity
of roads) were negatively associated.
Habitat models are most useful when they incorporate variables that
are easily quantifiable and produce results that can be clearly
interpreted (Hurley, 1984; Salwasser, 1984; Garshelis, 2000). Our
regression model is useful because variables are habitat features
important to beavers and can be derived from existing data sets
(watershed surveys and aerial photographs) eliminating the need to
measure them in the field.
Riparian trees provide important food for beavers during winter
when herbaceous vegetation is dormant and crops have been harvested.
Beavers forage on small diameter woody stems during winter and use these
to construct dens and dams. We frequently observed willows (Salix spp,),
maples (Acer saccharinum, A. saccharum), and green ash Fraxinus
pennsylvanicum) in food caches and dams. Consequently, it is not
surprising that availability of small trees and the extent of riparian
zone were important factors associated with beaver presence. Mature
riparian forests dominated by large trees appear less suitable for
beavers, perhaps because felling large trees is labor-intensive, their
shade reduces understory growth, and their presence suggests infrequent
flooding and scouring, disturbances that favor the early-successional
plants used by beavers.
Two variables (stream gradient and road proximity) were negatively
associated with beaver colonies. It is not surprising that beavers avoid
areas with high gradients. Typically, higher gradients result in a
higher flow which makes travel and transportation of food more
difficult, and destroys dams, dens, and food caches. We believe that
road proximity provides an indirect measurement of human activity along
the river. Although beavers often live in close proximity to humans,
previous research suggests that roads, railways, and land development
near waterways limit habitat quality (Slough and Sadleir, 1977).
In conclusion, our research suggests that the Embarras River
provides good quality beaver habitat along much of its extent. Although
the HSI model was not a useful predictor of relative abundance of
beavers on the river, both the Missouri model and our regression model
produced scores that were correlated with beaver abundance. We believe
this river is typical of many in Illinois and therefore these habitat
models are likely to be useful for evaluating habitat quality and
relative abundance of beavers in similar watersheds throughout the
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
This study was funded by the Federal Aid in Fish and Wildlife
Restoration Project W135-R, Illinois Department of Natural Resources
(IDNR), and the Department of Biological Sciences at Eastern Illinois
University (EIU). We thank Keith Valenti, Mike Douglas, and Nick Owens
for field assistance and Robert Bluett (IDNR) and Drs. Scott Meiners
(EIU) and Alan Woolf (SIUC) for technical assistance.
Allen, A. W. 1983. Habitat suitability index models: beaver. U. S.
Fish and Wildlife Service, FWS/OBS-82/10.30. 20 pp.
Boyce, M. S. 1981. Habitat ecology of an unexploited population of
beavers in interior Alaska. M.S. Thesis, University of Alaska,
Broschart, M. R., C. A. Johnston, and R. J. Naiman. 1989.
Predicting beaver colony density in boreal landscapes. Journal of
Wildlife Management 53:929-934.
Busher, P. E. 1983. Interactions between beavers in a montane
population in California. Acta Zoologica Fennica 174:109-110.
Cox, G. W. 2002. General ecology laboratory manual. 8th ed.
McGraw-Hill Co. Inc. New York, NY. 312 pp.
Davies, N. B. 1978. Ecological questions about territorial
behaviour. Pages 317-350 in J. R. Krebs and N. B. Davies (eds.).
Behavioural ecology: an evolutionary approach. Sinauer, Sunderland,
Mass. 494 p.
Garshelis, D. L. 2000. Delusions in habitat evaluation: measuring
use, selection, and importance. Pages 111-164 in L. Boitani and T. K.
Fuller (eds.). Research techniques in animal ecology. Columbia
University Press, New York, N.Y.
Hallett, D. L. and D. W. Erickson. 1980. Beaver (Castor
canadensis). Pages 99-105 in T. S. Baskett, D. A. Darrow, D. L. Hallett,
M. J. Armbruster, J. A. Ellis, B. F. Sparrowe, and P. A. Korte (eds.) A
handbook for terrestrial habitat evaluation in central Missouri. U. S.
Department of Interior, Fish and Wildlife Service Resource Publication
133. Washington, D. C.
Hoffmeister, D. 1989. Mammals of Illinois. University of Illinois
Press, Urbana and Chicago, IL. 348 pp.
Howard, R. J. and J. S. Larson. 1985. A stream habitat
classification system for beaver. Journal of Wildlife Management
Hurley, J. F. 1986. Development, testing, and application of
wildlife-habitat models--the manager's viewpoint. Pages 151-153 in
J. Verner, M. L. Morrison, and C. John Ralph (eds.). Wildlife 2000:
modeling habitat relationships of terrestrial vertebrates. The
University of Wisconsin Press, Madison, WI.
Nordstrom, W. R. 1972. Comparison of trapped and untrapped beaver
populations in New Brunswick. M.S. thesis, University of New Brunswick,
Fredricton. 104 pp.
Novak, M. 1987. Beaver. Pages 283-312 in Wild furbearer management
and conservation in North America. M. Novak, J. A. Baker, M. E. Obbard
and B. Malloch, (eds.) Ontario Trappers Association, North Bay, Ontario,
Robel, R. J., L. B. Fox, and K. E. Kemp. 1993. Relationship between
habitat suitability index values and ground counts of beaver colonies in
Kansas. Wildlife Society Bulletin 21:415-421.
Salwasser, H. 1986. Modeling habitat relationships of terrestrial
vertebrates--the manager's viewpoint. Pages 419-424 in J. Verner,
M. L. Morrison, and C. J. Ralph (eds.). Wildlife 2000: modeling habitat
relationships of terrestrial vertebrates. The University of Wisconsin
Press, Madison, WI.
Semyonoff, B. T. 1951. The river beaver in Archangel Province.
Pages 5-45 in Russian game report. Vol. 1. Can. Dep. North. Aff. and
Nat. Resour., Ottawa, Ont. (Transl. from Russian.)
Slough, B. G., and R. M. F. S. Sadlier. 1977. A land capability
classification system for beaver (Castor canadensis Kuhl). Canadian
Journal of Zoology 55:324-335.
Stromayer, K. A. K. 1999. Developing management alternatives to
promote beaver occupancy. Abstract of paper presented at the Virginia
Wildlife Society Meeting, Virginia Tech University, Blackburg, VA.
Wiggers, R. 1998. The Embarras River Basin: an inventory of the
region's resources. A project of the critical trends assessment
program, Illinois Department of Natural Resources, Springfield, IL.
Woolf, A., T. Nelson, and R. Bluett. 2003. Beaver relative
abundance in southern Illinois watersheds. Transactions of the Illinois
State Academy of Science 96:135-144.
Daniel R. Cox (1) and Thomas A. Nelson, Department of Biological
Sciences, Eastern Illinois University Charleston, IL 61920, USA
(1) Current address: Environmental Solutions and Innovations, Inc.
Cincinnati, OH, 45233, USA
Table 1. Twelve habitat variables tested for use in the
final logistic regression model for beaver habitat in the
Embarras River Watershed, central Illinois.
Variable Units Abbreviation
Canopy cover % CC
Riparian trees >45-cm dbh % Trees>45
Riparian trees <15-cm dbh % Trees<15
Shrub cover % SC
Shrub height m ShrubHt
Riparian zone width (mean) m RZW
Channel width (mean) m CW
Stream gradient % Grad
Bank height m BankHt
Stream sinuosity % Sinuous
No. roads within 200 m -- Roads
No. cropfields within 200 m -- Crops