Burbot resource selection in small streams nearthe southern extent of the species range

Introduction

Many species of fish in stream ecosystems arethreatened by degradation of habitat by humandevelopment and landscape alteration (Wang et al.2000; Wheeler et al. 2005). As humans alter aquaticsystems many fish species are likely to face populationdeclines and range reductions (Neves & Angermeier1990; Rahel et al. 1999; Mattson & Angermeier2007). Declines of burbot Lota lota populationsthroughout much of their holarctic range have resultedin their inclusion on species of conservation concernlists (Kirchhofer & Hefti 1996; Paragamian & Bennett2008). Further, interest in the conservation of biodi-versity of fish has increased attention to burbotmanagement among fisheries scientists (Arndt &Hutchinson 2000; Fisher et al. 2000). This study was

prompted by the desire better to understand habitat useof burbot occurring in small streams near the sourthernextent of the species range in North America, osten-sibly threatened by limited range, suburban sprawl andthe predicted loss of coldwater streams in a changingclimate.

The conservation of aquatic habitat is a goal ofbiodiversity management because human effects onaquatic systems are often in conflict with speciespersistence (Abell et al. 2000; Labonne & Gaudin2006). Aquatic conservation is informed by identify-ing the physical habitat variables most influencing fishabundance in streams (Ahmadi-Nedushan et al. 2006;and references therein); it is important to understandthe relative influence of each habitat variable and thespatial scale over which each variable functions(Rabeni & Sowa 1996). Information about habitat

Ecology of Freshwater Fish 2009: 18: 234–246Printed in Malaysia Æ All rights reserved

� 2008 The AuthorsJournal compilation � 2008 Blackwell Munksgaard

ECOLOGY OFFRESHWATER FISH

Dixon CJ, Vokoun JC. Burbot resource selection in small streams near thesouthern extent of the species range.Ecology of Freshwater Fish 2009: 18: 234–246. � 2008 The Authors.Journal compilation � 2008 Blackwell Munksgaard

Abstract – Burbot Lota lota are of global conservation concernparticularly near the southern extent of the species range. Populations insmall streams lack quantitative descriptions of resource selection.Microhabitat and reach-scale habitat were surveyed at occupied andunoccupied locations. Data were analyzed with logistic regression todevelop resource selection function models which predicted the probabilityof occurrence and abundance categories for microhabitat and reach-scalehabitat, respectively. Information-theoretic model selection methods wereused in an exploratory context, and four competing models were multi-model averaged to create a microhabitat resource selection function. Thesum of Akaike’s Information Criterion weights across the four competingmodels indicated that substrate type, substrate embeddedness, and depthwere equally the most important microhabitat explanatory variables. At thereach scale, high correlation among explanatory variables precludedmultivariate model development. Stream-reach gradient was the mostsupported univariate model. Three additional models competed asimportant; mean summer temperature, sinuosity, and dominant substratetype. The created resource selection functions can be used in rankingpotential translocation sites, used in instream flow assessments, andprovide habitat selection information that can provide context for habitatmanagement decisions.

C. J. Dixon, J. C. VokounNatural Resources Management and Engineering,College of Agriculture and Natural Resources,University of Connecticut, Storrs, CT, USA

Key words: burbot Lota lota ; resource selectionfunction; logistic regression; habitat suitability;habitat use

J. C. Vokoun, Natural Resources Managementand Engineering, College of Agriculture andNatural Resources, 308 WB Young Building, 1376Storrs Road, University of Connecticut, Storrs, CT06269-4087, USA; e-mail: jason.vokoun@uconn.edu

Accepted for publication September 24, 2008

234 doi: 10.1111/j.1600-0633.2008.00341.x

selection can be informative to instream flow analysesintended to guide regulated river management andecosystem conservation (Tharme 2003). Further,translocation is a prominently used strategy in recov-ery programs for threatened fishes (Williams et al.1988), and identifying the suite of habitat variablesselected (and perhaps required) is an important step inpredicting the eventual success of a translocation site(Harig & Fausch 2002).

The spatiotemporal scales of analyses are importantcomponents of research designed to identify habitatcharacteristics influencing fish presence and abun-dance and to develop effective habitat restoration andconservation strategies (Frissell et al. 1986; Schlosser1991, 1995; Lewis et al. 1996; Rabeni & Sowa 1996;Mattingly & Galat 2002; Lowe et al. 2006).A hierarchal approach allows managers to identifyprocesses that may affect species distribution andabundance and that limit processes at smaller scales.

Our goal was to identify habitat conditions associ-ated with the presence of burbot in CT, USA. In CT,burbot have been consistently found only in smallstreams in one tributary watershed of the HousatonicRiver (Linsley 1844; Whitworth 1996). Other recordsof burbot in CT are inconsistent among locations andfew in number, and the species is listed as state-endangered (CTDEP 2004). Translocation of burbot tonew sites is a potential tool to increase the number ofpopulation occurrences (Dillen et al. 2008b) andlessen the threat of extirpation due to environmentalcatastrophes.

Specifically, we sought to identify through statisticalmodel selection a suite of habitat characteristics thatburbot select at the microhabitat and reach spatialscales to develop resource selection functions (RSF).An RSF is defined as any model that identifiesparameters of the environment that predict the prob-ability of use of a resource by an organism (Boyceet al. 2002). An RSF then, is a probabilistic form ofhabitat suitability criteria (Boyce et al. 2002; Ahmadi-Nedushan et al. 2006).

Methods

Study sites

In northwest CT USA the Housatonic River basintopography is characterized by rounded hills andmountains draped with glacial deposits, with steep-sided valleys cut into the hills (USEPA 2002). Studysites were selected and expanded from river segmentswhere burbot had been previously documented by CTDepartment of Environmental Protection fisheriesbiologists. The sites were located within the Hollen-beck River and Blackberry River tributary watersheds(Fig. S1). Two of these sites, the Hollenbeck ⁄Brown

and Lower Hollenbeck, were in the Hollenbeckwatershed, that drained an area of 112 km2. TheHollenbeck ⁄Brown site was approximately 2 km longand included the Hollenbeck River mainstem and thejunction with the lower portion of Brown Brook. TheHollenbeck River has �5 m bankfull width channeland 0.3 m mean pool depth. Brown Brook was �3 mwide and had 0.15 m mean pool depth. The Hollen-beck River streambed was characterized by sand inthe lower section of the site shifting to predominantlycobble and boulder at upstream higher elevations,with well-defined pool-riffle sequences. The BrownBrook channel was characterized by cobble andboulder substrate with step-pool sequences. Landsurrounding this site was primarily forested, with asingle agricultural field at the lower end. The LowerHollenbeck site was 3.7 km long and �5.5 m widelocated on the mainstem Hollenbeck River down-stream of the Hollenbeck ⁄Brown site. The channelwas highly sinuous and characterized by sandsubstrate with pool-run sequences lacking definedriffles: average depth was approximately 0.3 m. Largewoody debris was prevalent throughout the lower site.Land surrounding this site was primarily deciduousforest, with patches of conifer stands and occasionalagricultural fields.

The Blackberry site was located in the BlackberryRiver watershed and drained an area of 120 km2. Onlyone burbot has been recorded in the Blackberrywatershed, discovered during a fish kill. The Black-berry site was 5.4 km long and bounded upstream by asmall run-of-river dam (�1.8 m tall), a likely barrier toupstream fish movement. The channel was approxi-mately 7.0 m wide with a mean pool depth of 0.4 m.The Blackberry River streambed was characterized bysand, gravel, and cobble in the lower section shiftingto boulder and bedrock upstream near the dam. Pool-riffle sequences were well defined throughout the site.Land surrounding the site was predominately agricul-tural in the floodplain and forested on the mountain-sides.

Fish sampling

We electrofished the three study sites during July–September 2005. Burbot were collected in shallowareas (<0.6 m) with a model LR-24 pulsed-DCbackpack electrofisher (Smith-Root Inc., Vancouver,WA, USA). Electrofishing voltage ranged between 200and 350 volts based on measures of water temperature,conductivity, and specific conductivity of the waterwith a model 556 multimeter (YSI Inc., YellowSprings, OH, USA). Sampling was conducted withone person wearing the backpack and two people oneach side who waded the channel with dip nets tocapture fish. In the infrequent deeper areas of stream,

Burbot resource selection in small streams

235

a push-barge electrofishing raft (Smith-Root Inc.)powered by a 3600 watt gasoline generator was used.Electrofishing was methodical and thorough, probingall available habitat. Sampling proceeded upstreamand continued until a burbot was captured. Thelocation of capture was marked with a surveyor’sflag. Although the positions of fish prior to capturewere not known, burbot were not observed swimmingahead of the crew and were observed experiencinglimited electrotaxis and then narcosis from concealedlocations in the stream bottom because the water wasclear during all sampling. We also waded into the sites,located rocks under which burbot were present andlifted them before the burbot swam away. Althoughour methods are not quantitative and are without anestimate of absolute capture efficiency, we are confi-dent they revealed actual burbot locations duringdaytime. Water depths sampled ranged from 0.1 to0.9 m with a mean of 0.3 m. Captured burbot weremeasured to the nearest mm total length (TL), andweighed to the nearest g. We wore wetted wool glovesto maintain better control of the fish and reducehandling time and stress.

Water temperature

Water temperature was measured with HOBO� WaterTemp Pro temperature loggers (Onset ComputerCorp., Pocasset, MA, USA). Temperatures weresampled hourly in 15 locations in each stream, withmultiple loggers in each site, and at the upstream anddownstream boundaries. Loggers recorded streamtemperature (±0.02 �C) 24 times a day for a year,but only summer data (June–August) were used.

Microhabitat

Microhabitat was described at the time and locationof capture for each burbot. Depth, velocity, temper-ature, pH, dissolved oxygen, dominant substrate type,a visual estimate of embeddedness, and percentcanopy cover were measured. Velocity was measuredusing a Flo-Mate model 2000 portable flowmeter(Hach ⁄Marsh-McBirney Inc., Frederick, MD, USA).Dominant substrate type classes were determinedusing a modified Wentworth scale. Wentworth sub-strate size categories used were: fines (<0.25 mm);sand (0.25–2.0 mm); gravel (2–16 mm); pebble (16–64 mm); cobble (64–256 mm); boulder (>256 mm);and bedrock (after Mattingly & Galat 2002). Sub-strate embeddedness was visually estimated intocategories of low, intermediate, and high usingmethods adapted from Platts et al. (1983). Percentcanopy cover was measured with a spherical dens-iometer. Samplers faced each cardinal direction andcounted cells with canopy coverage. The mean of the

four measurements was used to characterize thecapture location.

Microhabitat measurements at capture locationsconstituted used habitat datasets to develop RSFmodels. Available habitat was collected at randomlocations within three channel units (pool, riffle, run)that were randomly-selected upstream or downstreamfrom the location of each capture. Microhabitatlocation within a channel unit was determined byvisualizing a 10 · 10 grid over the channel and thenmeasuring habitat in a random cell within that grid.Random microhabitats were measured with the sameprocedures as capture locations.

Reach-scale

A reach was defined as 3–5 pool-riffle sequences and alength of at least 20 bankfull channel widths (Arend1999). Reach-scale habitat characteristics were col-lected by surveying thalweg longitudinal profiles foreach sample reach and by summarizing the habitat inchannel units along the thalweg.

Thalweg longitudinal profiles were measured usinga total station (model TCR307; Leica GeosystemsAG, Heerbrugg, Switzerland) with methodologyadapted from Panfil & Jacobson (2001). Cross-sections were surveyed above riffle crests andthrough the deepest sections of pool channel units.Mean channel width of a reach was calculated fromthe bankfull widths of these cross-sections. Thenumber of discrete points surveyed along the streamthalweg was based on geomorphic complexity withthe objective of capturing important changes instreambed elevation and channel unit boundariesalong the thalweg; they averaged out to approxi-mately 2 survey points per longitudinal meter ofstreambed. Channel unit boundaries (e.g., pool, riffle,and run) along the thalweg were visually identified,surveyed, and used to calculate the percent of poolhabitat in each reach. Elevations of surveyed pointsat the upstream and downstream ends of each reachwere used to determine reach gradient.

Stream habitat used to characterize channel unitswere dominant substrate type, substrate embedded-ness, bank vegetation, a canopy-shade index, aninstream-cover index, and a tributary index. Domi-nant substrate type was assessed visually as thesubstrate size that was covering the largest percent-age of streambed in a channel unit, and observationswere combined to represent dominant substrate in areach. Substrate embeddedness was also visuallyestimated, and similarly combined to determine theembeddedness category with the largest percentage ofoccurrence and was used to represent a reachcondition. Criteria for classifying canopy-shadeindex, instream-cover index, and tributary index were

Dixon & Vokoun

236

categorical and modeled after Mattingly & Galat(2002), with four categories representing increasinglevels of presence of each respective reach-scalehabitat component and are reported in detail in Dixon(2007).

Mean residual pool depth, a discharge-independentmeasurement of pool depth defined as the differencebetween the lowest elevation in a pool and theelevation of the downstream riffle crest, required anobjective determination of the riffle crest (or geo-morphic equivalent) streambed elevation. The bedform differencing technique was used to measureelevation changes in the streambed that exceeded acalculated reach-specific tolerance value to determinethe riffle crest elevations within each reach (O’Neil &Abrahams 1987). We followed the methods describedby Arend (1999) for calculating bed form differenc-ing. Mean reach residual pool depth was thencalculated as the average of residual pool depthswithin a reach.

Mean reach temperature of the summer samplingperiod was calculated from the hourly temperaturedata. Reaches with a temperature logger in the reachwere assigned the mean of that logger. Mean reachtemperature of reaches without a logger were esti-mated using the mean of the two nearest loggers, oneupstream and one downstream of the reach.

Data analysis

An information-theoretic approach was used todevelop RSF that predict the presence of burbot usingthe presence versus availability dataset (Boyce et al.2002). Information-theoretic approaches rank compet-ing hypotheses (expressed as statistical models) anduse simple calculations to estimate a formal strength ofevidence for alternative hypotheses rather than statis-tical tests of null hypotheses and decisions based onP-values (Burnham & Anderson 1998; Anderson et al.2001; Anderson & Burnham 2002). Using thisapproach, evidence for alternative hypotheses includethe rank of each hypothesis (an estimate of the formallikelihood of each hypothesis), and a measure ofprecision that incorporates model selection uncer-tainty. Information-theoretic approaches use the set ofalternative models to make formal inference (Ander-son et al. 2000).

No quantitative data within the literature wasavailable on the habitat use of burbot in small streamsin North America (but see Dillen et al. (2008a) inFrance). We adapted the exploratory approach ofAmadio et al. (2005). Selected microhabitat-scale andreach-scale habitat characteristics were hypothesizedto affect occurrence based on the literature describingburbot ecology in lotic and lentic systems. A limitedset of habitat characteristics that were correlated to the

occurrence of burbot, but not highly intercorrelatedwere identified. Dependence among predictor vari-ables (including habitat characteristics) in observa-tional studies is commonplace (Burnham & Anderson1998). Therefore we retained habitat charactistics forsubsequent analysis that were intercorrelated at|r|<0.50 to reduce multicollinearity among models(Glanz & Slinker 1990; Graham 2003). All modelsubsets using the resulting set of habitat characteristicswere then created and fitted with logistic regressionusing PROC LOGISTIC in SAS (SAS Institute Inc.2003).

Support for candidate models was assessed usingthe small sample bias adjustment of Hurvich & Tsai(1989) to Akaike’s Information Criterion (AICc;Burnham & Anderson 1998; Amadio et al. 2005).The AICc weights (wi) were used to rank models. Themodel with the greatest wi was considered to best fitthe data (Burnham & Anderson 1998). Competingmodels were identified as models with wi ‡ 10% ofthe highest ranking model. Model selection uncer-tainty was addressed among competing models bymulti-model averaging and calculating model-aver-aged estimates of the coefficients and their standarderrors (Burnham & Anderson 1998). The wi ofpredictor variables (habitat characteristics) weresummed across competing models included in thecalculation of the multi-model averaged coefficients toestimate their relative importance (Burnham & Ander-son 1998). The variable depth and its squaredtransformation were both included in models in whichdepth appeared to improve model fit.

Polytomous logistic regression was used to predictthe probability of reach use based on a datasetcategorized into four ordinal responses representingthe number of burbot sampled within a reach (absent,low, moderate, high) at the reach scale and therefore,RSF models. Ordinal categories were determined byidentifying natural breaks in the data. Polytomouslogistic regression model diagnostics included theScore Test for Proportional Odds Assumption, Devi-ance and Pearson Goodness-of-fit statistics (Stokeset al. 2000). Separation of ordinal-category probabil-ities were calculated and are presented in the resultsfollowing conventions used by Harig & Fausch(2002). Model development progression was similarat both spatial scales, except to accommodate theordinal responses in the dependent variable (burbotpresence) at the reach scale.

Results

Summer electrofishing captured 201 burbot, 141 at theHollenbeck ⁄Brown site, 57 at the Lower Hollenbecksite, and 3 burbot at the Blackberry site. The meanlength of burbot was 191 mm and ranged from 84 to

Burbot resource selection in small streams

237

356 mm TL (Fig. 1). The mean weight was 45 g andranged from 4 to 240 g. Sixty percent of fish caughtwere from 160 to 200 mm. Large burbot (over500 mm) have never been recorded in the HousatonicRiver watershed. Age estimated from otoliths of 32fish revealed that fish were 1–5 years of age with amodal age of 2–3 years (Dixon 2007). Examiningsacrificed fish revealed that fish estimated to be 2 ormore years old were mature. Additional samples ofburbot in November 2005 revealed male and femalefish in this size range expressing gametes.

Microhabitat resource selection

Burbot sampled during the summer of 2005 in theLower Hollenbeck and Hollenbeck ⁄Brown sites wereused to build an RSF. Data collected in the BlackberryRiver site were not used because few burbot werecollected. Consequently, the use versus availabilitydata set used to develop RSF had 396 microhabitatdescriptions including an equal number of capturelocations and random locations.

Five habitat characteristics that were correlated(P < 0.05) with burbot presence, and not highlycorrelated with each other, were used in RSF devel-opment. They were; substrate type, substrate embedd-edness, channel unit type, velocity, and depth. Allunique subsets of these five were combined to propose31 logistic regression models. The model with thehighest wi value contained four of the five variables(Table 1) and three competing models had wi valueswithin 10% of the top-ranked model. These fourmodels were multi-model averaged to create themicrohabitat RSF (Table 2). The sum of wi valuesacross the four competing models indicated thatsubstrate, substrate embeddedness, and depth wereequally important microhabitat predictor variables,each had a summed wi of 0.98. These three variables

appeared in all the competing models. Channel unittype (i.e., pool, riffle, or run) had a summed wi of 0.57(Table 2), and occurred in the top and third rankedmodels. Velocity was the least influential variable(summed wi of 0.50), and occurred in the second andthird ranked models. Note that there is some inherentbias in the interpretation of the sum of AIC wi becausenot all variables were represented in all four candidatemodels.

The direction of effect indicated by competingmodels and the multi-model averaged RSF suggestedthat burbot selected pools over riffles and runs, and

0

5

10

15

20

25

30

35

40

45

51–6

0

71–8

0

91–1

00

111–

120

131–

140

151–

160

171–

180

191–

200

211–

220

231–

240

251–

260

271–

280

291–

300

311–

320

331–

340

351–

360

TL (mm)

Fre

quen

cy

Fig. 1. Length–frequency histogram of the 201 burbot captured atthe study sites.

Table 1. Microhabitat variables comprising competing logistic regressionmodels predicting occurrence of burbot.

Model K )2logeL AICc wi

Substrate, embeddedness,depth, channel unit

15 201.58 217.33 0.33

Substrate, embeddedness,depth, velocity

16 204.49 217.82 0.26

Substrate, embeddedness,depth, velocity, channel unit

13 201.03 218.00 0.24

Substrate, embeddedness,depth

12 206.74 218.88 0.15

Variable were; substrate type, substrate embeddedness, depth, channel unittype, and velocity. Models are ranked according to Akaike’s InformationCriterion corrected for small sample size (AICc), which was calculated fromthe number of estimated parameters (K), and the log likelihood ()2logeL).AIC weights (wi) were used to select the ‘best approximating’ models from aset of candidate exploratory models. Competing models with wi values thatwere 10% or greater than the maximum wi are included in the table.

Table 2. Explanatory microhabitat predictor variables included in competingmodels, multi-model averaged coefficients (standard error), and sums ofAkaike Information Criterion (AICc) weights summed across competingmodels (see Table 1).

Model variable Averaged coefficient Sum of AICc weights

Intercept )1.52 (0.17)Substrate 0.98

Fines N ⁄ ASand )0.19 (0.16)Gravel )0.35 (0.24)Pebble )0.22 (0.23)Cobble 0.28 (0.11)Boulder 0.73 (0.10)

Substrate embeddedness 0.98Low 0.27 (0.09)Intermediate N ⁄ AHigh )0.09 (0.15)

Depth 0.98Depth 3.36 (0.82)Depth2 )3.03 (0.90)

Channel unit 0.57Pool 0.08 (0.05)Riffle N ⁄ ARun 0.06 (0.05)

Velocity )0.44 (0.41) 0.50

N ⁄ A denotes category used as reference in coefficient determinations so noestimate is available.

Dixon & Vokoun

238

locations with the largest dominant substrates, lowestlevels of embeddedness, greatest water depths, andlowest water velocities. Use and availability fre-quency histograms were developed to provide addi-tional support for categorical habitat parameters(Fig. 2).

Reach-scale resource selection

The number of burbot captured ranged from 0–26 inthe reaches sampled (N = 24). Categorization ofreaches as absent, low, moderate, and high reflectedburbot numbers of 0, 1–5, 8–15, and 22–26,respectively (Fig. 3). Six of the reach-scale variableswere significantly correlated with burbot occurrence(|r|>0.50; P < 0.05); gradient, mean temperature,sinuosity, dominant substrate, canopy shade index,and instream cover index. High multicollinearityamong the variables was present and precludedmultivariable model development (Table 3).Therefore, the six univariate models were considereda candidate model set and AICc values and wi werecalculated. A top-ranked model was identifiedalong with three competing models with wi valueswithin 10% of that of the top-ranked model(Table 4).

The AICc values and respective wi of the fourcandidate models indicated that the model containinggradient was the most supported for predicting burbotabundance within reaches. The fitted polytomouslogistic regression model intercept and coefficient forgradient were used to develop probability functionsgiven by the following;

0

5

10

15

20

25

30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Reaches

Fre

quen

cy

HighModerateLowAbsent

Fig. 3. Frequency of burbot abundance instudy site stream reaches. Categorization ofreaches as absent, low, moderate, and highabundance based on capture numbers of 0,1–5, 8–15, and 22–26 burbot during ele-trofishing sampling.

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

Embeddedness

Channel unit

Substrate

Pool RunRiffle

Used

Available

Fines

Sand

Grave

l

Pebble

Cobble

Boulde

r

Bedro

ck

Fre

qu

ency

(%

)

Low Intermediate High

Fig. 2. Frequency of use and availability histograms of categoricalhabitat characteristics in the Lower Hollenbeck and Hollen-beck ⁄Brown samples sites.

Table 3. Pearson correlation coefficients (P-values) among reach-scale habitat characteristics correlated with burbot abundance.

Gradient Mean temp. Sinuosity Dominant substrate Canopy shade index Instream cover index

Gradient 1.0000 )0.4332 (0.0344) )0.4715 (0.0200) 0.8626 (0.0001) 0.6630 (0.0004) 0.3970 (0.0547)Mean temp. 1.0000 0.2317 0.2760 )0.6781 (0.0003) )0.2283 (0.2834) )0.4141 (0.0443)Sinuosity 1.0000 )0.4658 (0.0218) )0.4245 (0.0387) )0.3562 (0.0876)Dominant substrate 1.0000 0.5932 (0.0022) 0.3470 (0.0967)Canopy shade index 1.0000 0.3952 (0.0560)

Values from 24 stream reaches were used. Correlations significant at a = 0.05 are in bold.

Burbot resource selection in small streams

239

P ð‘‘absent’’Þ

¼ expð0:7394� 0:193� gradientÞ1þ expð0:7394� 0:193� gradientÞ ;

P ð‘‘lowþ absent’’Þ

¼ expð�1:3414� 0:193� gradientÞ1þ expð�1:3414� 0:193� gradientÞ ;

Pð‘‘moderateþ lowþ absent’’Þ

¼ expð�2:7727� 0:193� gradientÞ1þ expð�27727� 0:193� gradientÞ ;

where P = probability of a reach having burbotabundances of ‘absent’, ‘low and absent combined’or ‘moderate, low, and absent combined’. Isolatingthe probabilities for low, moderate, and highcategories was calculated by; P(low) = P(low +absent) ) P(absent), P(moderate) = P(moderate +low +absent) ) P(low + absent), and P(high) = 1 )P(moderate + low + absent), respectively. The sepa-rated probabilities for low, moderate, and high burbotabundance in stream reaches are plotted togetheragainst gradient (Fig. 4). Reaches with low, moderate,or high abundance of burbot were predicted to havesteeper gradients than reaches where burbot wereabsent.

The second-ranked polytomous logistic regressionmodel contained mean temperature and was given bythe following;

Pð‘‘absent’’Þ

¼ expð51:5822� 2:2945� temperatureÞ1þ expð51:5822� 2:2945� temperatureÞ ;

Pð‘‘lowþ absent’’Þ

¼ expð49:5075� 2:2945� temperatureÞ1þ expð49:5075� 2:2945� temperatureÞ ;

P ð‘‘moderateþ lowþ absent’’Þ

¼ expð48:2238� 2:2945� temperatureÞ1þ expð48:2238� 2:2945� temperatureÞ :

Separated probabilities were calculated the same asfor gradient. The probability plot indicated that thenumber of burbot within a reach was predicted todecrease as mean temperature increased (Fig. 4).

Table 4. Polytomous logistic regression models predicting occurrence ofburbot from reach-scale explanatory variables.

Model K )2LogeL AICc wi

Gradient 4 55.76 60.43 0.46Mean temperature 4 57.02 61.70 0.24Sinuosity 4 58.21 62.88 0.13Dominant substrate 6 54.76 63.26 0.11

Models are ranked according to Akaike’s Information Criterion corrected forsmall sample size (AICc), which was calculated from the number of estimatedparameters (K), the log likelihood ()2logeL). AIC weights (wi) were used toselect the ‘best approximating’ models from a set of candidate exploratorymodels. Competing models with wi values that were 10% or greater than themaximum wi are included in the table.

0

0.2

0.4

0.6

0.8

1

0.5 2.5 4.5 6.5 8.5 10.5

20.5

22.5

24.5

12.5

14.5

16.5

18.5

Gradient (m km–1)

AbsentLow ModerateHigh

20.3

20.5

20.9

20.7

20.1

21.3

21.5

21.9

21.7

21.1

22.3

22.5

22.9

22.7

22.1

0

0.2

0.4

0.6

0.8

1

Mean temperature (°C)

Pro

bab

ility

1 31.4 1.8 2.2 2.6

53.4 3.8 4.2 4.6

0

0.2

0.4

0.6

0.8

1

Sinuosity

Fig. 4. Probability plots of single-variable polytomous logisticregression models that predict burbot abundance levels in streamreaches. Plots show the predicted probabilities of occurrence forabundance categories of absent (0 burbot), low (1–5 burbot),moderate (8–15 burbot) and high (22–26 burbot) in terms of levelsof habitat predictor variables gradient, mean summer temperature,and stream channel sinuosity.

Dixon & Vokoun

240

The model containing sinuosity was also used todevelop probability functions and was given by thefollowing;

P ð‘‘absent’’Þ

¼ expð4:2032� 1:2564� sinuosityÞ1þ expð4:2032� 1:2564� sinuosityÞ ;

P ð‘‘lowþ absent’’Þ

¼ expð2:0599� 1:2564� sinuosityÞ1þ expð2:0599� 1:2564� sinuosityÞ ;

P ð‘‘moderateþ lowþ absent’’Þ

¼ expð0:9109� 1:2564� sinuosityÞ1þ expð0:9109� 1:2564� sinuosityÞ :

The probability plot indicated that the number ofburbot within a reach was predicted to decrease assinuosity increased (Fig. 4).

In the fourth-ranked model, dominant substratewas a categorical explanatory parameter; and aprobability plot similar to gradient, mean tempera-ture, and sinuosity could not be developed. Instead,frequency histograms of predicted abundance levelsof burbot within a reach were constructed based onsubstrate category. The fitted polytomous logisticregression model intercepts for each category ofdominant reach-level substrate were given by thefollowing;

P ð‘‘absent’’Þ

¼ expð2:3972� 1:4514� sandÞ1þ expð2:3972� 1:4514� sandÞ ;

P ð‘‘lowþ absent’’Þ

¼ expð0:2901� 1:4514� sandÞ1þ expð0:2901� 1:4514� sandÞ ;

P ð‘‘moderateþ lowþ absent’’Þ

¼ expð�1:2466� 1:4514� sandÞ1þ expð�1:2466� 1:4514� sandÞ :

Separating probabilities of low, moderate, and highabundances were conducted the same as for continu-ous variables. Burbot were predicted to be absent inreaches dominated by sand 62% of the time, and tohave low abundances 31%, moderate 5%, and high 2%(Fig. 5).

The model intercepts for cobble were given by thefollowing;

Pð‘‘absent’’Þ

¼ expð2:3972� 0:3824� cobbleÞ1þ expð2:3972� 0:3824� cobbleÞ ;

Pð‘‘lowþ absent’’Þ

¼ expð0:2901� 0:3824� cobbleÞ1þ expð0:2901� 0:3824� cobbleÞ ;

Pð‘‘moderateþ lowþ absent’’Þ

¼ expð�1:2466� 0:3824� cobbleÞ1þ expð�1:2466� 0:3824� cobbleÞ :

Burbot were predicted to be absent in reachesdominated by cobble 38% of the time, to have lowabundances 45%, moderate 12%, and high 5%(Fig. 5).

The model intercepts for boulder were given by thefollowing;

0

0.2

0.4

0.6

0.8

1Cobble

Pro

bab

ility

0

0.2

0.4

0.6

0.8

1Sand

0

0.2

0.4

0.6

0.8

1

Absent Low Moderate High

Boulder

Abundance of burbot

Fig. 5. Probabilities of burbot abundance categories among cate-gories of the habitat predictor variable dominant substrate. Reachessampled only contained dominant substrates of sand, cobble, andboulder.

Burbot resource selection in small streams

241

P ð‘‘absent’’Þ

¼ expð�2:3972� 1:8338� boulderÞ1þ expð�2:3972� 1:8338� boulderÞ ;

P ð‘‘lowþ absent’’Þ

¼ expð�0:2901� 1:8338� boulderÞ1þ expð�0:2901� 1:8338� boulderÞ ;

P ð‘‘moderateþ lowþ absent’’Þ

¼ expð1:2466� 1:8338� boulderÞ1þ expð1:2466� 1:8338� boulderÞ :

Burbot were predicted to be absent in reachesdominated by boulder <1% of the time, and have lowabundances <1%, moderate <1%, and high 99%(Fig. 5). The abundance of burbot within a reachwas predicted to increase as substrate size increases.

Discussion

Five microhabitat predictor variables were supportedas influencing burbot occurrence, the most importantbeing substrate type, substrate embeddedness, anddepth. Robins & Deubler (1955) reported that the mostimportant habitat feature for burbot was shelter ofsufficient size to completely hide. In lakes, shelteramong the interstices of large stones in the littoral zoneis more important than temperature preferences ofburbot during their first year of life (Hofmann &Fischer 2002). Cover might provide protection frompredators and refuge from current velocities for burbot(McMahon et al. 1996; Fisher 2000b). Laboratoryexperiments have shown that the unavailability ofrefuges stressed burbot, even when predators wereabsent (Fischer 2000a). Burbot used boulder sizesubstrate with low levels of embeddedness in greaterproportion than availability in the Hollenbeck drainagesample sites that may have optimized opportunitiesto inhabit shelter of sufficient size. Although largewoody debris was prevalent throughout samplessites, relatively few burbot used it at the time ofcapture.

The influence of depth on burbot presence was alsowell supported in the models. In general, burbotselected deeper locations, but the result was condi-tional on substrate size and embeddedness. Meanresidual pool depth of available pools was one-thirdgreater than the average pool depth at capturelocations, which suggests that burbot may select largesubstrate for cover over depth at the microhabitatscale. Depth selection might not be realized beyond a

threshold depth. Also, our microhabitat analysis wasbased on daylight sampling when burbot typically aresedentary (Carl 1995). This may also partly explainwhy velocity was the least important of the variablesincluded in candidate models, even though velocityand discharge affect burbot migration in rivers (Joneset al. 1974; Paragamian 2000). In the Ohre River inthe Czech Republic, burbot occupied relatively shal-low habitats with large substrate size and highvelocities during daylight hours, shifting to deeperareas with lower velocities and small substrate size atdusk (Slavık et al. 2005). Also, our sampling capturedpredominantly adult burbot, age 0-sized fish werenoticeably rare, and the resulting RSF and predictorvariables must be interpreted accordingly. Various lifestages require different types of physical habitats(Fischer 1999; Fisher 2000b). Future research shoulddescribe the habitat requirements of larval and juvenileburbot in streams because little information exists(Fisher 2000b). Several factors affected the probabilityof burbot occurrence at the reach-scale, with gradientbeing the most supported explanatory variable. Gra-dient has been a good predictor of fish distributions asobserved in changes in assemblage memberships atdifferent elevations in stream systems and littoralzones (Rahel & Hubert 1991; Fladung et al. 2003;Robinson & Rand 2005). Study reaches with agradient of 8 mÆkm)1 or greater contained a moderateor high abundance of burbot and accounted for 49% ofall burbot captured. Gradient was correlated withseveral other reach-scale features, including meantemperature, sinuosity, and dominant substrate. Inobservational studies such as this one there is a highprobability that some of the predictor variables will becorrelated (Burnham & Anderson 1998). For example,as mean gradient decreases longitudinally downstreamfrom the headwaters mean water temperature increasesas the influence of riparian vegetation declines,reducing shading within the channel (Rutherford et al.2004). Gradient may therefore be a surrogate forseveral other reach-scale features affecting burbotoccurrence in these small streams. In the La Bar Riverwatershed in France, burbot were similarly moreabundant in narrow upstream reaches than in the lowerreaches (Dillen et al. 2008a).

Mean water temperature was the second mostsupported reach-scale model in the candidate set.The probability plot suggested that habitat wasmaximized below a mean temperature of 21.5 �C.The trend illustrated by the model indicated thatburbot occupy reaches with cooler average tempera-tures which were found at higher elevations within thewatershed. Water temperature affects all biochemicalprocesses (Hardewig et al. 2004) and was consideredthe most important environmental factor controllingthe distribution of burbot in areas of suitable habitat in

Dixon & Vokoun

242

Pennsylvania rivers (Robins & Deubler 1955). Incontrast to the reach-scale, water temperature was notcorrelated with burbot presence at the microhabitat-scale. Water temperatures at capture locations rangedfrom 16.0- 26.8 �C which subjected 83% of all burbotto temperatures above the reported optimal thermalpreference of 15.6–18.2 �C (Scott & Crossman 1973).All reaches within the study sites exceeded mean dailytemperatures of 20 �C in July and August 2005.Burbot have an evolutionarly-adapted strategy tosurvive warm water temperatures by going into aquiescent state during the summer months andrestricting activities such as spawning to the winterseason when water temperatures are lower (Holkeret al. 2004). This project was not intended to evaluatediel activity levels of burbot during summer months,but future investigations could prove useful in under-standing survival mechanisms in these thermallyvariable watersheds.

Stream ecosystems may be viewed as hierarchalenvironments (Frissell et al. 1986) such that processesthat structure habitat at larger spatial scales maydirectly influence the structure of habitat at smallerscales. For example, gradient on the reach scaledirectly affects dominant substrate type at the micro-habitat scale through sediment transport mechanisms(Knighton 1998). These relationships can be altered byanthropogenic activities, as evidenced by two low-gradient reaches comprised of primarily sand substratewithin the Lower Hollenbeck study site whichcontained a moderate and high abundance of burbot.These reaches were both areas where meander bendsneared roadways and large boulders (rip-rap) had beenplaced within the stream to increase bank stability.

The information-theoretic approach is not oftenconsidered in exploratory data analyses, however, itproved to be a useful framework for this research eventhough little a priori knowledge was available on theresource selection of burbot in small streams. Theapproach allowed us to identify competing combina-tions of habitat features associated with burbot occur-rence and address model uncertainty throughmulti-model averaging. Our limited all-subsetsapproach can potentially stabilize the negative charac-teristics of multicollinearity among explanatoryvariables (Graham 2003). Future researchers now haveour RSFs to hypothesize upon when investigating loticpopulations. Additionally, the RSFs can be paired withhydrodynamic models of riverine habitat to performinstream flow assessments (Guay et al. 2000).Specifically this study probabilistically quantified thatat the microhabitat-scale burbot used channel units withlow current velocities, increased depth, and largesubstrate particles that have low levels of embedded-ness. At the reach-scale burbot used reaches associated

with higher gradients, relatively less sinuosity, coolermean water temperatures, and large substrate.

At the onset of the project we hoped to discovermany burbot in the Blackberry River, but only threefish were captured by thoroughly sampling 5.4 kmencompassing the fish kill zone where burbot werefirst observed in the watershed. Although we couldnot use these limited data in RSF model develop-ment, it provides an example of how results can beused to inform conservation and management deci-sions. Given lower gradient and higher watertemperature, our RSFs do not predict the Blackberrysite be as highly populated as the higher-gradient,cooler reaches in the Hollenbeck watershed. Thesereach-scale observations modify the similarity of thesites when only microhabitat is considered. Forexample, the mean gradient of the reaches at theBlackberry site was 4.3 mÆkm)1. The probability plotfor the gradient RSF predicts a 45 % chance ofobserving low abundance and only a 12% chance ofobserving high abundance. Mean temperature in theBlackberry site (21.8� C) was near the higher end ofthe range observed, and the RSF predicted again a45% chance of low abundance and a 14% chance ofhigh abundance. With this information, fisheriesbiologists may choose to turn attention elsewhere,such as higher gradient reaches further upstream inthe Blackberry watershed which may be bettercandidates for translocation. The microhabitat andreach RSFs we developed can be used to evaluatepotential translocation sites in other locales bycollecting the reduced suite of habitat measurementsto quantitatively evaluate sites.

Acknowledgements

This project was supported by the State Wildlife Grants Programand the Connecticut State Income Tax Checkoff, EndangeredSpecies Program. Additional support from the Department ofNatural Resources Management and Engineering, and theBishop-Carder Fellowship Fund is acknowledged. RobbetteSchmit, Dustin Edwards, Sam Bourret, Justin Wiggins, RalphTingley, Bradley Trumbo, Scott Porter, and Chuck Lavoieassisted with data collection. This manuscript was improvedwith comments from Yoichiro Kanno, and three anonymousreviewers.

References

Abell, R.A., Olson, D., Dinerstein, E., Hurley, P., Diggs, J.,Eichbaum, W., Walters, S., Wettengel, W., Allnutt, T.,Loucks, C. & Hedao, P. 2000. Freshwater ecoregions ofNorth America: a conservation assessment. Washington, DC,USA: Island Press.

Ahmadi-Nedushan, B., St-Hilaire, A., Berube, M., Robichaud,E., Thiemonge, N. & Bobee, B. 2006. A review of statistical

Burbot resource selection in small streams

243

methods for the evaluation of aquatic habitat suitabilitycriteria for instream flow assessment. River Research andApplications 22: 503–523.

Amadio, C., Hubert, W., Johnson, K., Oberlie, D. & Dufek, D.2005. Factors affecting the occurrence of saugers in small,high-elevation rivers near the western edge of the species’natural distribution. Transactions of the American FisheriesSociety 134: 160–171.

Anderson, D.R. & Burnham, K.P. 2002. Avoiding pitfalls whenusing information-theoretic methods. Journal of WildlifeManagement 66: 912–918.

Anderson, D.R., Burnham, K.P. & Thompson, W.L. 2000. Nullhypothesis testing: problems, prevalence, and an alternative.Journal of Wildlife Management 64: 912–923.

Anderson, D.R., Link, W.A., Johnson, D.H. & Burnham, K.P.2001. Suggestions for presenting the results of data analyses.Journal of Wildlife Management 65: 373–378.

Arend, K.K. 1999. Classification of streams and reaches. In:Bain, M.B. & Stevenson, N.J., eds. Aquatic habitat assess-ment: common methods. Bethesda, MD, USA: AmericanFisheries Society, pp. 57–74.

Arndt, S.K. & Hutchinson, J. 2000. Characteristics of burbotspawning in a tributary of Columbia Lake, British Columbia,over a four-year period. In: Paragamian, V.L. & Willis, D.W.,eds. Burbot: biology, ecology, and management. Bethesda,MD, USA: American Fisheries Society, Fisheries Manage-ment Section, Publication Number 1, pp. 46–60.

Boyce, M.S., Vernier, P.R., Nielsen, S.E. & Schmiegelow,F.K.A. 2002. Evaluating resource selection functions. Eco-logical Modeling 157: 281–300.

Burnham, K.P. & Anderson, D.R. 1998. Model selection andinference: a practical information-theoretical approach. NewYork: Springer-Verlag.

Carl, L.M. 1995. Sonic tracking of burbot in Lake Opeongo,Ontario. Transactions of the American Fisheries Society 124:77–83.

CTDEP (Connecticut Department of Environmental Protection)2004. Connecticut’s endangered, threatened and specialconcern species 2004. CT, USA: Connecticut Departmentof Environmental Protection, Hartford.

Dillen, A., Coeck, J. & Monnier, D. 2008a. Habitat use andseasonal migrations of burbot in lowland rivers in northFrance. In: Paragamian, V.L. & Bennet, D.H., eds. Burbot:ecology, management, and culture. Bethesda, MD, USA:American Fisheries Society, symposium 59, pp. 29–42.

Dillen, A., Vught, I., De Charleroy, D., Monnier, D. & Coeck, J.2008b. A preliminary evaluation of reintroductions of burbotin Flanders, Belgium. In: Paragamian, V.L. & Bennet, D.H.,eds. Burbot: ecology, management, and culture. Bethesda,MD, USA: American Fisheries Society, symposium 59, pp.79–183.

Dixon, C.J. 2007. Habitat use and population demographics ofstate-endangered burbot (Lota lota) in northwestern Connect-icut. M.S. Thesis. Storrs, CT: University of Connecticut.

Fischer, P. 1999. Otolith microstructure during the pelagic,settlement and benthic phases in burbot. Journal of FishBiology 54: 1231–1243.

Fischer, P. 2000a. An experimental test of metabolic andbehavioral responses of benthic fish species to different typesof substrate. Canadian Journal of Fisheries and AquaticSciences 57: 2336–2344.

Fisher, S.J. 2000b. Early life history observations of burbotutilizing two Missouri River backwaters. In: Paragamian,V.L. & Willis, D.W. eds. Burbot: biology, ecology, andmanagement. Bethesda, MD, USA: American FisheriesSociety, Fisheries Management Section, Publication Number1, pp. 96–101.

Fisher, S.J., Willis, D.W. & Pope, K.L. 2000. Burbot manage-ment: population condition and size structure assessmenttools and their application. In: Paragamian, V.L. & Willis,D.W., eds. Burbot: biology, ecology, and management.Bethesda, MD, USA: American Fisheries Society, FisheriesManagement Section, Publication Number 1, pp. 136–138.

Fladung, E., Scholten, M. & Thiel, R. 2003. Modeling thehabitat preferences of preadult and adult fishes on theshoreline of the large, lowland Elbe River. Journal of AppliedIchthyology 19: 303–314.

Frissell, C.A., Liss, W.J. & Hurley, M.D. 1986. A hierarchicalframework for stream habitat classification: viewing streamsin a watershed context. Environmental Management 10: 199–214.

Glanz, S.A. & Slinker, B.K. 1990. Primer of applied regressionand analysis of variance. New York: McGraw-Hill.

Graham, M.H. 2003. Confronting multicollinearity in ecolog-ical multiple regression. Ecology 84: 2809–2815.

Guay, J.C., Boisclair, D., Rioux, D., Leclerc, M., Lapointe, M.& Legendre, P. 2000. Development and validation ofnumerical models for juveniles of Atlantic salmon (Salmosalar). Canadian Journal of Fisheries and Aquatic Sciences57: 2065–2075.

Hardewig, I., Portner, H.O. & van Dijk, P. 2004. How does coldstenothermal gadoid Lota lota survive high water tempera-tures during summer? Journal of Comparative PhysiologyB-Biochemical Systemic and Environmental Physiology 174:149–156.

Harig, A.L. & Fausch, K.D. 2002. Minimum habitat require-ments for establishing translocated cutthroat trout popula-tions. Ecological Applications 12: 535–551.

Hofmann, N. & Fischer, P. 2002. Temperature preferences andcritical thermal limits of burbot: implications for habitatselection and ontogenetic habitat shift. Transactions of theAmerican Fisheries Society 131: 1164–1172.

Holker, F., Volkmann, S., Wolter, C., van Dijk, P.L.M. &Hardewig, I. 2004. Colonization of the freshwater environ-ment by a marine invader: how to cope with warm summertemperatures? Evolutionary Ecology Research 6: 1123–1144.

Hurvich, C.M. & Tsai, C.L. 1989. Regression and time seriesmodel selection in small samples. Biometrika 76: 297–307.

Jones, D.R., Kiceniuk, J.W. & Bamford, O.S. 1974. Evaluationof the swimming performance of several species of fish fromthe Mackenzie River. Journal of the Fisheries Research Boardof Canada 31: 1641–1647.

Kirchhofer, A. & Hefti, D. 1996. Conservation of endangeredfreshwater fish in Europe. Switzerland: Birkhauser Verlag,Basel.

Knighton, D. 1998. Fluvial forms and processes a newperspective. London, Great Britian: Arnold.

Labonne, J. & Gaudin, P. 2006. Modeling population viabilityin fragmented environments: contribution to the conservationof an endangered percid (Zingel asper). Canadian Journal ofFisheries and Aquatic Sciences 65: 650–659.

Dixon & Vokoun

244

Lewis, C.A., Lester, N.P., Bradshaw, A.D., Fitzgibbon, J.E.,Fuller, K., Hakanson, L. & Richards, C. 1996. Considerationsof scale in habitat conservation and restoration. CanadianJournal of Fisheries and Aquatic Sciences 53: 440–445.

Linsley, R.J.H. 1844. Catalogue of the fishes of Connecticut,arranged according to their families; prepared for the YaleNatural History Museum. American Journal of Science andArts 47: 55–80.

Lowe, W.H., Likens, G.E. & Power, M.E. 2006. Linking scalesin stream ecology. BioScience 56: 591–597.

Mattingly, H.T. & Galat, D.L. 2002. Distributional patterns ofthe threatened Niangua darter, Etheostoma nianguae, at threespatial scales, with implications for species conservation.Copeia 3: 573–585.

Mattson, K.M. & Angermeier, P.L. 2007. Integrating humanimpacts and ecological integrity into a risk-based protocol forconservation planning. Environmental Management 39: 125–138.

McMahon, T.E., Zale, A.V. & Orth, D.J. 1996. Aquatic habitatmeasurements. In: Murphy, B.R. & Willis, D.W., eds.Fisheries techniques, 2nd edn. Bethesda, MD: AmericanFisheries Society, pp. 83–120.

Neves, R.J. & Angermeier, P.L. 1990. Habitat alteration and itseffects on native fishes in the upper Tennessee River system,east-central U.S.A. Journal of Fish Biology 37: 45–52.

O’Neil, M.P. & Abrahams, A.D. 1987. Objective identifica-tion of pools and riffles. Water Resources Research 20: 921–926.

Panfil, M.S. & Jacobson, R.B. 2001. Relations among geology,physiography, land use, and stream habitat conditions in theBuffalo and Current River systems, Missouri and Arkansas.Springfield, VA: United States Geological Survey, NationalTechnical Information Service. Biological Science ReportUSGS ⁄BRD ⁄BSR-2001-0005.

Paragamian, V.L. 2000. The effects of variable discharges onburbot spawning migrations in the Kootenai River, Idaho,USA, and British Columbia, Canada. In: Paragamian, V.L. &Willis, D.W., eds. Burbot: biology, ecology, and manage-ment. Bethesda, MD, USA: American Fisheries Society,Fisheries Management Section, Publication Number 1, pp.111–123.

Paragamian, V.L. & Bennett, D.H. eds. 2008. Burbot: ecology,management, and culture. Bethesda, MD, USA: AmericanFisheries Society, symposium 59.

Platts, W.S., Megahan, W.F. & Minshall, G.W. 1983. Methodsfor evaluating stream, riparian, and biotic conditions. Ogden,UT: U.S. Forest Service, Intermountain Forest and RangeExperiment Station. General Technical Report INT-138.

Rabeni, C.F. & Sowa, S.P. 1996. Integrating biological realisminto habitat restoration and conservation strategies for smallstreams. Canadian Journal of Fisheries and Aquatic Sciences53: 252–259.

Rahel, F.J. & Hubert, W.A. 1991. Fish assemblages andhabitat gradients in a Rocky Mountain–Great Plains stream:biotic zonation and additive patterns of community change.Transactions of the American Fisheries Society 120: 319–332.

Rahel, F.J., Muth, R.T. & Carlson, C.A. 1999. Endangeredspecies management. In: Kohler, C.C. & Hubert, W.A.,eds. Inland fisheries management in North America, 2nd

edn. Bethesda, MD: American Fisheries Society, pp. 431–454.

Robins, C.R. & Deubler E.E., Jr 1955. The life history andsystematic status of the burbot, Lota lota lacustris (Walbaum),in the Susquehanna River system. New York State Museumand Science Service, Circular 39. Albany, NY:

Robinson, J.L. & Rand, P.S. 2005. Discontinuity in fishassemblages across an elevation gradient in a southernAppalachian watershed, USA. Ecology of Freshwater Fish14: 14–23.

Rutherford, J.C., Marsh, N.A., Davies, P.M. & Bunn, S.E.2004. Effects of patchy shade on stream water temperature:how quickly do small streams heat and cool? Marine andFreshwater Research 55: 737–748.

SAS Institute Inc. 2003. SAS ⁄STAT user’s guide, version 9,volumes 1-2. Cary, NC: SAS Institute Inc.

Schlosser, I.J. 1991. Stream fish ecology: a landscape perspec-tive. BioScience 41: 704–712.

Schlosser, I.J. 1995. Critical landscape attributes that influencefish population dynamics in headwater streams. Hydrobiolo-gia 303: 71–81.

Scott, W.B. & Crossman, E.J. 1973. Freshwater fishes ofCanada. Fisheries Research Board of Canada Bulletin 184:641–645.

Slavık, O., Bartos, L. & Mattas, D. 2005. Does streammorphology predict the home range size in burbot? Envi-ronmental Biology of Fishes 74: 89–98.

Stokes, M.E., Davis, C.S. & Koch, G.G. 2000. Categorical dataanalysis using the SAS system, 2nd edn. Cary, NC, USA:SAS Institute Inc.

Tharme, R.E. 2003. A global perspective on environmentalflow assessment: emerging trends in the development andapplication of environmental flow methodologies for rivers.River Research and Applications 19: 397–441.

USEPA (U. S. Environmental Protection Agency). 2002.Ecological characterization of the Housatonic River. USEPA,Contract No. DACW33-94-D-0009 ⁄ 032, Boston, MA.

Wang, L., Lyons, J., Kanehl, P., Bannerman, R. & Emmons, E.2000. Watershed urbanization and changes in fish commu-nities in southeastern Wisconsin streams. Journal of theAmerican Water Resources Association 36: 1173–1189.

Wheeler, A.P., Angermeier, P.L. & Rosenberger, A.E. 2005.Impacts of new highways and subsequent landscape urban-ization on stream habitat and biota. Reviews in FisheriesScience 13: 141–164.

Whitworth, W.R. 1996. Freshwater fishes of Connecticut, 2ndedn. State Geological and Natural History Survey Bulletin114. Hartford: Connecticut Department of EnvironmentalProtection.

Williams, J.E., Sada, D.W., Williams, C.E., Bennett, J.R.,Johnson, J.E., Marsh, P.C., McAllister, D.E., Pister, E.P.,Radant, R.D., Rinne, J.N., Stone, M.D., Ulmer, L. & Withers,D.L. 1988. American Fisheries Society guidelines forintroductions of threatened and endangered fishes. Fisheries13: 5–11.

Supporting information

Additional Supporting Information may be found inthe online version of this article.

Burbot resource selection in small streams

245

Fig. S1. Location of study sites in northwest CT thatwere selected based on locations where burbot wererecently found. The Blackberry site was located on theBlackberry River in the town of North Canaan. TheLower Hollenbeck and the Hollenbeck ⁄Brown siteswere located in the Hollenbeck River watershed inCanann.

Please note: Wiley-Blackwell are not responsiblefor the content or functionality of any supportingmaterials supplied by the authors. Any queries (otherthan missing material) should be directed to thecorresponding author for the article.

Dixon & Vokoun

246