OCli .311

OCli .311

Distribution and habitat use by juvenile Atlantic salmon (Salmo salar) at multiple spatial scales, and implications for habitat modelling and fish-habitat management.

by

«()Tammo Peter Bult

A thesis submitted to the School of Graduate Studies in partial fulfilment of the requirements for the degree of Doctor of

Philosophy

Department of Biology Memorial University of Newfoundland

St. John's,Newfoundland 1999

dankzijHan

Implicit assumptionsof micro-habitatmodelsarethat(I)habitatlimits populationlevels and small-scaleinformationon habitat selection behavioursof individuals can be used to manage populationsat largespatio-temporalscaJes (scale-up);(2) thesingle or few measurement scales usedinhabitatmodelsareappropriatefor identifYjngimportanthabitats;and {3) "better"

habitatsareeharaeterisedb yahi gherdensityo rfrequency-of-use, i.e.densitycanbeusedasan indicator of habitat quality.

(I)Basedonscope- and rate-diagramsfromfield-data and theoreticalscenarios ofmovement andmonality,l conciuded thatsalmonidhabitatmodelsope rate inthe contexlofprocessest hat maynotbeimponantto theproblemswewouldlike toaddress.I suggestedsurveydesigns thatallow problemsassociatedwith scaJe-upt o beovercome.

(2) I evaluateddistributionsofjuvenileAtlantiesalmon(Salmosal ar)overa range0fspatial scaJesbasedonastrearn-tankstudy {spatialsealeslemto3m)andfie lddata (spatialscalesl cmtol5m),todeterrninewhetherpatehinessoffishd istributionsorassociationswithdepth, water velocityand substratedependedon spatialscaJe,to deterrnir.escaJesmost appropriateto habitatmodels,and to comparemulti-scaleversus single-scalehabitatmodellingapproaehes.

Results indicatedassociations witheonspecifies, substrate,watervelocity and depth changed withspatialscale and direetionrelativetowater flow.Associations were most differentfromrandomatsmall spatial scales(ambit radius<50em).Bothstudies indicatedthatsi ngle- and multi-scalehabitatseleetionmodels wereequallyabletodeseribe fish densities atsmallspatialscales (ambit radius<4 m).Thefield-based study indicated thatsingle-andmulti-scalemodelsoftenfailed todeseribe fish densitiesat scales larger thanusedin the model(scale-up).

(3) I studieddensity-dependenthabitatuse byAtlanticsalmonparr basedon experimental riverineenciosuresand fielddata.Results fromthe experimentalstudyindicatedthat habitat use changedwith populationdensity.Resultsfromthe field-basedstudywere lessclearwith

some ofthe results suggestingdensity-dependentdistributionprocesses.lconcludedthat habitatselectionbysalmonparr wasdensity-dependentand highly variable.Changesin habitat usewith densitywere most likelydue to small-scalespacingbehaviourorte rritoriality.

I concludedthat quantitativemulti-scaleapproaches are importantto habitat rnodelling, identifiedimportantresearchquestions.presented some noveltechniquesforscalinganalyses and made suggestionstoimprove habitatmodellingand resourcemanagement.

Acknowledgements

Firstof all I would liketothank mythesissupervisor,Dr.R.L.Haedrich,for giving me the opportunity to work and studyin Newfoundland.[would like to thank my thesis-committee,Dr. R.I.Gibson, Dr.R.L.Haedrich, Dr.1.Heggenes and Dr. D.C.

Schneider for support and critical reviewof the work. I am grateful to Dr. S.C.Riley for introducing me into salmonid biologyand riverine field work.

Kerri Abbott, Hillary Baikie,Hanneke Bult,Petrus Bult,Lynn Bussy,Gavin andIiIl Crutcher,Keith Hillier,Lisa lanes, Terra Martin,DavidMethven, Shannon O'Reilly, Iohn Pike,Tina Pittman ,SteveRiley,Iennifer Robinson,SteveSutton,Tom Therriault,Erin Trowbridge,Bob Whalen and Matthew Windsorall providedvaluable support with the data-collection.I would liketo thank DavidScruton of the Department of Fisheries and Oceans in St.John's,Newfoundland, for providingthe field assistance of Tony Bowdring and IenniferMaddock.

A discussion with Dr. L.Zedelon autocorrelationtechniques wasthe source of inspiration for several of the computer programs I developed in the thesis. Parts of the manuscript were greatly improved by the comments from Dr.R.Cunjak,Dr.1.Ellis, Dr.I.Home,Dr.

I.A.Hutchings, Dr.D.Methven, Dr.1. Nestler, Dr.M.Simpson, Dr.M. Rodriguezand Dr. J.Wroblewski. This study was supported via a Natural Science and Engineering Research Council (NSERC) grant to Dr. R.L.Haedrich and the Canadian Centre for Fisheries Innovation.Additional financial support was provided by Petrus and Toos Bult, Eimerd and Annemarie Bult, and Wil and Piet Plieger.

CHAPTERI.HABITAT SELECTIONBEHAVIO URS INHABITATMODELLING AND nSH·

HABITATMANAGEMENT•••_.••._ _ 1

CHAPTER2:MORTALITYVERSUS SPATIALDYNAMIC S AT MULTIPLESCALES:

SCALED-RATE PLOTS FOR SALMONIDSANDIMPLICATIO NSFOR HABITAT

MODELLING _ 19

CHAPTER 3.MULTI-SCALEANALYSES OF HABITAT USE BY JUVENILE ATLANTIC

SALMON 43

-'n •

• ..

CHAPTER4:DENSITY-DEPENDENTHABITATSELECUONBYJUVENILEATLANTIC

SALMON 153

.. oa,

.. h.

~.

,", ..

"."';. "

ru. '"

..

'om.

4.2

.m _

4.2

TABLE2.1.SCENARIOSUSEDFORCALCULATINGCRIT1CALSCALESOF MOVEMENTVERSUS MORTALITY..•... 25 TABLE 3.2.1.ASSOCIATIONSOF FISH WITH WATERDEPTHS QUANTIFIEDBY MEANSOF

RELATIVECONTACT(RC) AT 5 CM AMBlTRADII. .

TABLE 3.2.2.ASSOCIATIONSOF FISH WITH WATERVELOCITIES(CM S·I) CLOSETO THE BOTTOM(WB)AND AT 60% DEPTH (W6),QUANTIFIEDBY MEANSOF RELATIVE

CONTACT(RC) AT 5 CM AMBITRADII 89

TABLE 3.2.3. ASSOCIATIONSOF FISH WITH WATERDEPTHS(CM) QUANTIFIEDBY MEANS

OF RELATIVE CONTACT(RCD) AT5 CMAMBITRADlI 91

TABLE 3.2.4. ASSOCIATIONSOF FISH WITH WATERVELOCITIES(CM S·,)AT 60% OF DEPTH.

;;ffuf~~~ra~~~f~~:~f:Ji

TABLE 3.2.8.DIFFERENCESINFISHBEHAVIOURSAMONGPREFERRED(RC>O)AND

::f*~~fik1§fJ~~~~~~:~~~~~~~~~~~~~

~~R~~~t~I~~::R~~~:;~.~~:.~~~.:.~.~·

^..

S.~~.~~

^..

~~:.IIO

::~~~R=~t~![a~i~:·::~:~:S::~:Z:==~~~

NORTHARMRIVER IN 1994AND1995 114 TABLE 3.3.5.SUMMARYOF ASSOCIATIONSOF 0+ SALMONWITHTHE VARIABLESDEPTH.

TAB~~~?~§:~~~f~:~:trJg~~~~:::=f.:E~1

AND WATERVELOCITYAS A FUNCTIONOF SCALE,QUANTIFIEDIN TERMS OF

TAB~~~~ik~~Fs1~~~~~:~:~!~~~~1~~~·~··~··134

TABffi1~~~!r~:f~~~EWN~~ui~~7l~ABLE~35

DEPTHANDWATERVELOCITYAS A FUNCTIONOF SCALE.QUANTIFIEDIN TERMS OF

~~~~CTr:~~~~~~.~~~~.~.~~=~.~~.~~:.~~~

..

S.~:::~.~

..

~~.137

TABLE 3.3.9.CORRELATION(R) BETWEENOBSERVEDFISH DENSITIES(% TOTAL POPULATIONM")ANDFISH DENSITIESGENERATEDBY SINGLE-(SS)ANDMULTl- SCALE (MS) HABITATSELECTIONMODELS.AT THE SPATIALSCALESOF POOL. RlFFLE

~~~~:s~~~~~:~~~~~~:) .. ~~ .. S.~~~~ .~~::.~) .. I.~

....143

TABLE4.1.1.SELECTIONOF POOLVERSUSRUNHABITAT.RIFFLEVERSUSRUNHABITATAND RIFFLEVERSUSPOOLHABITATBY PARRAS A FUNCTIONOF PARRDENSITYANDTIlE TEMPERATUREATREMOVALlNTIlREE EXPERlMENTALSECTIONS IN 1993AND 1994....164

::E~E:~~~=~~~3::~i7~~:~~~7

BROOKAND FRESHWATERRIVERIN 1984-1993 180 TABLE 4.2.2.DESCRIPTIONOF HABITATANDFISH DENSlTrES AT SAMPLINGSTATIONS IN

NORTIlEASTTREPASSEYBROOKAND FRESHWATERRIVER.•... ...•...•..181 TABLE 4.2.3.SUMMARYSTATISTICSDESCRIBING

s

(INTRASPECIFICSTATION-TO-RiVER

~~~~E~~~~~~~P~~i:~~;{~~~~;i~~t~~"::;a::~og~N

~~.~NA~~~S~~~6~;A~~v'E~~:iY.~~~~~

^..

S~~~~~.~~~~

¹⁹³

TABLE4.2.4.SUMMARYSTATISTICSDESCRffiINGTJ(INTERSPECIFICSTATION-TO-RIVER

~{~~~~~~[~t~~~i~~~~~;ia~£{~~dPrE~~

AND WATERVELOCITY.

TABLE4.2.5.ATLANTICSALMONAND BROOKTROUTorSTRmUTIONS IN NORTHEAST TREPASSEYRIVERAND FRESHWATERRIVERDESCRmED BY TAYLORPOWER PLOTS.196

TAB~~~S~~:~~~~~i~l~:~~~;;.a~~~~~E~~~;~~

LEVELOF TROUT AND PARRRESPECTIVELY•...•...••...•...•...•...•...•197

:~~~~~~~~~:

List of figures

f1GUREl.l.sCOPE-DlAGRAMILLUSTRATINGSCAlE-UPINHABITAT MODELLING..•... 9 f1GURE2.l.CRmCALSCALESOF1ElRITORIALf1sH...•.•...•...•.26 f1GURE2.2.CRmCALSCALESOFf1SHDISPLAYlNGDlURNALMOVEMENrS...••28 f1GURE2.3.CRmCALSCALESOFf1SHDlsFLAYlNGsEASONALMOVEMENrS 30 f1GUREH.CRmCAL SCALESOF1ElRITORIAL f1SHDISPLAYINGDlURNALANDSEASONAL MOVEMENrS...•...•..•...•...•... 31 f1GURE2.S.CRmCALSCALESOFTERRITORIALf1sHDlSPLAYlNGDlURNALANDSEASONAL

~'I'_1r(;~

:~~~~~~ir~~~;~ :~"

FIG~~~\~~~~~:·D~~~~ONS:·:::.::::::::::::::::::::::::::::::::::::::::::::::::::::~~

FIGURE3.2.2.ASSOCIATIONSOF FISHPOSITIONSIN THE STREAMTANK. QUANTIFIEDBY

=~~~A6i~~~:;:::~~~~.~.~.~~~~~ .. ~~~~.::..~.~~~.:::.~ .. ~:=~83

FIGURE3.2.3.ASSOCIATIONSOF FISHWITHCOBBLEIN THE STREAMTANK.QUANTIFIED BYMEANSOF RELATIVECONTACT(RC)AND RELATIVECONTACTEXHAUSTIVE

(RCEXl OVER A RANGE OF SPATIAL s CALES 8~

FIGURE3.2.4 ASSOCIATIONSOF FISHWITHDEPTHSINTHE STREAMTANK. QUANTIFIED

FIG~~~N~~~~~~~5~~~~~~~~~~~~~A~~:i~E~HE

...85 BOlTOM (FIGURESA)ANDAT60%OF DEPTH(FIGURE58) INTHE STREAMTANK.

~~.=~ . . ~~.~.~~=.~:..~~:::.~.:.~~~.~ .. ~~!..~.~.~.~~.=.~.:..~.:.~:::~ . ...

⁸⁷

f1GURE3.2.6 ASSOCIATIONS OFFISHWITHWATERVELOClTYCLASSES[0.SI.<S,lOI.<-Io.501 AND>50CMS4MEASUREDAT60%DEPTH(FIGURE3.2.6A-D RESPECTlVELy).

FIG~E~€;;i~~~~;~;~~s~:~~~~~~~~~€;~G:

::=~iSEE~:5~~c~l~::~::-:~~~·::::~=;:::::lIS

TROUT DISTRIBUTIONS(0+.>0+).AS OBSERVEDIN NORTHHARBOURRIVERIN199~

PREFACE

Understanding and predictingeffects of man on naturedependsonunderstandingthe relations between organisms and theirenvironment.To achievesuchan understanding, distributions of organisms are studiedrelativeto distributions of environmentalfeatures. From such study, inferences are made on the processes that underlie the observed distributions and the most important of these are expressed and linked together in a habitat model. Habitat modelssimplify,summarise and describe this understanding,and as such arevaluable to resource management and environmentalimpact assessments.

Inthisprefacelwillexplainwhyitisimportanttostudydistributionsandprocessesat multiple scales, and will show how this relates to habitat modelling. To begin, I will use two examples:one illustrating scale dependency of associations and0ne illustrating scale dependencyofprocesses.Fromtheseexamples,lwillintroducethemaintopics addressed in the thesis.

Scale dependency of associations: an example

The concepts involved in multi-scale habitat modelling are more readilyconveyedbyusing hypothetical examples close to daily experience.Suppose we would like to understand camping behaviour of people in order to design the best possible campground.To achieve such an understanding, we could study the distribution oftentsrelativetothedistribution of environmental features that are thought to be important, and we could then summarise this understanding in a "camping model"that describes the occurren ceoftentsrelativeto these environmental features.

We stan the project by making maps of the distribution of tents and thoseenvironmental features thought to be important in camping behaviour,e.g.theavailabilityofwater,the

flatnessof theterrain, and the availability of firewood.Themaps can be drawn at different resolutio nsor spatialscales.

Depending on the resolutio n ofthe maps we use, our resultswillvary.At a small spatial scale(lxl m),tentsarenegativelyasso ciated with the availability0fwater andwoodand positivelywith the flatness of the terrain,because campers do not puttheirtents in treesor inlakes but do put them on levelground.Atlarger spatialscal es (IOOOxlOOOm), however,tents arepos itively associated withthe availability of waterandwood, aswater is used foractivities suchas fishing,swimming or sailing,and woodisusedforcooking andcampfires.Becausesmall-scaleflat placesmayoften be foundwithin largerareasthat are generallysteeplysloping,such as mountains,theflatnessoftheterrainma ynotbea good predictor at larger spatial scales.Negative associat ionswith flatness mayevenbe found at theselargespatialscaleswhen mountainous areasaret he onespreferredfor recreationalactivitiessuchasclimbing orhiking.

These relationsmay be summarisedquantitatively in a camp ingmodeLUsingthis model, we then maytry to evaluate differentareas with respect to suitabilityfor camping.

However,dependingonthescaleofthismodel ,ourcondusions withrespecttosuitability willdiffer:a small-scalemodel maypredictthat desert plainsare suitableforcampinga s wood and waterare hardlypresentand the terrain is quitelevel;a large-scalemodel may predict thatmangroveswampsare suitablefor campingastrees and water are abundant.

Obviously,neitherone ofthese conclusionsis right,despitethefact that both models do giveavalid,al though incomplete,descriptionof how someone choos eswhere toplace a campsite.The problem is thatcamping behaviouroperates at multiplescales,whereas the modelsoperateoni yatthe scaleap propriatetotheoriginalresolution.

Thisexample show s thatassociations modelledatsmall scales may give oppositeresults from thosemodelled at large scales.This implies that acomparison of resultsfrom studies thatdifferinmeasurementscalecannotbedone·.,ithoutsomeunderstandingofhowscale

affects results, and that results fromstudies done usinginconsistent measurement scales cannot beinterpreted

Scale dependency of processes:an example

Distributions of organisms are theresult offour processes:mortality,movement, reproduction and growth.Ifone aims to describe the distributionof organisms it is helpful to have some idea of which ofthese processes are important and which are not.Research couldthenbedirectedatthemoreimportantprocesses,unimportantprocesses could be ignored and, from this, a simpler model could be made without sacrificing model efficiency.However,the relativeimportance of different processes varies with scale.This may be best explainedusing another example.

Suppose that we areinterested in the distribution of mice.Tobegin,wevisualisethe continent as a huge checkerboardwith mice scatteredrandomly across it.Whenthecells that compose the checkerboard are small (say Ix1 m), changes in the number of mice over short periods of time (sayI hour) in each cell are mainlyinfluenced by the wayin which mice run about,i.e.the distributionof mice at small space/timescales (I rn, I hour) is dominatedby movement.Bycontrast,when cells arelarge (say IOOxlOOkm) and times arelong(saylyear),thisverysamedistributionisdominatedbymortality,insteadof movement, as the chance an individual mouse will live and diewithin a single cell is larger than the chance it will move to a differentcell.

This example shows that small-scale processes may not be that relevant to describing distributions at larger scales:A model describing movement of mice at small space/time scales maynot adequately describe this verysame distributionatlarger space/time scales, becausethedistributionofmiceattheselargerscalesisdrivenbyreproduction and mortalityrather than movement.In addition,differentvariables maybedifferentially

important in their contribution to movement and reproduction/mortality.In other words, what is seen at larger scales may not be simply the summation of small-scale processes.

Scaling analyses

These two examples illustrate that, depending on the scale we use to study a system, our results and understanding of the system in terms of distributions ,associationsand processes may differ:The associations between tents and environ mentalfeatures,the models that summarised these associations, and the recommendations made based on these models were all scale-dependent (example I);different processes were perceived as being importantindeterrniningthedistributionofmice,dependingonthe scales used to study them (example 2).

Because observational resultsvarywithscaJe, it is important to consider explicitly the measurement scales one chooses in a study.Multi-scale analyses that explicitly evaluate distributions, associations and processes over a range of scalescanaid in determining which scales are most relevant in a particular problem.Consider the camping model: by studying the associations between tents and the availability ofwaterandwoodovera range of scales one could identify the several scales at which campingbehaviouroperates and then, with that knowledge, make the best model to answer the question "How far is one wilIing to travel from a tent site to gather wood or water?" Without multi-scale analyses,the choiceofa particular measurement scale for making the model could easily become purely personal and subjective. '

Multi-scale analyses could also help to understand how the structure and orientation of landscapeelements-thelandscapemosaic-affectsthesuitabilityofan area for camping, i.e.is it better for a campground to have a few large lakes ora lot ofs mailer ponds, and how does the distribution of smaller and largerlakesaffeet the suitability ofa terrain for camping? In addition, multi-scale approaches may act as a framework to incorporate

results obtained at differentscales and to evaluatethe validity0fextrapolating srnall-scale modelsin order to address problemsoperating at muchlargersca les, i.e.is it possibleto make inferenceson the suitabilityfor campingof very large areas, based on observations on the distributionof tents and environmentalfeatures withinsuch areas (scale-up)?

Multi-scale approaches in salmonid habitat modelling and thesis questions

Salmonidsare probablyamong the best studied fish speciesin the world.Habitat models that describe relations between the occurrence of salmonids and riverine habitats are widely used in impact analysesand instreamimprovement projects.Despite the considerableresearch effort that has gone into these models,associationsofsalmonidsand their habitats and the processes thatgovem salmoniddistributionshavehardlybeen studied usingexplicitquantitativemulti-scaleapproaches.

The choice of measurementscale is often based on the biologicalintuitionofthe researcher constrained by logistics.Forexample,previousworkhasshownthatsalmonids seleetpositionsin streamsbasedon theircompetitiveabilitiesand the profitabilityofpositions interrnsofpotentialnetenergyintakerateandpredationrisk,withprofitabilityofpositions beinglargelydeterrninedby the physicalhabitatin terrnsof cover,bottomtopographyand currentflow pattems. As such,the area withina streamis oftenregardedas a hierarchyof potentialpositions,rangingfrom inaccessibleto ideal,witheachfishchoosingthe most profitablepositionthatitsrankin the socialhierarchywillallow.Territoriality,small-scale spacingbehaviouror pre-emptiveexclusionare thus assumedto regulateuseofpreferred positionsand~pacewhich,ifinshortsupply,areassurnedto regulatepopulationdensity.

Thus,the physicalhabitatis regardedas a templatedeterrniningdistnbution patternsoffish.

Basedon this,use of availablehabitatby salmonidsis oftendescribedat smallspatialscales usingso-calledmicro-habitatmodellingapproaches(habitatsdescribedatscales<l m' ).But the intuitivewishto work at thisfinescalemayhaveto be changeddependingon the resolution of availablemapson riverinehabitatsor other logisticalconstraints,such as the time and

funding available for the study. The result is that measurement scales vary both among and within habitat modelling studies.

The fact that measurement scales vary constitutes a problem when interpreting,comparing and applying results from various studies.In particular, the scale-up from habitat model to management problem has hardlybeen evaluated quantitatively:Whatistherelevanceofa model that describes the distribution offish over small-scale habitats to the density offish in a much larger area, i.e.how relevant are small-scalemodels to large-scale problems?

Multi-scale analyses are needed that evaluatesalmonid distrib utions,associationsbetween salmonidsand their habitats, and the processes that govern salmoniddistributions.Critical questions are: At what scales are salmonids associated with their habitats? Do multi-scale analyses confirm the importance of scales as determined by other stu dies? Whatprocesses predominate at what spatio-temporal scales? Such studies could act as a framework to incorporate ideas from studies operating at different scales.

This thesis makes a start at multi-scale analysis ofsalmonid distributions.Processes important to salmon distributions were studiedover a range of spa tio-ternporal scales to determine which processes predominate at which space-time scales,as in the mouse distribution example, and to explore the problems associated wit h scale-up (Chapter 2).

Atlantic salmon distributions and associations between salmon and their habitats were studiedoverarangeofscales,asinthecampingexample,todeterminethescalesmost important to habitat modelling (Chapter 3)_Because use of habitats by salmonids is generally considered a result of competition for preferred habit ats,specialattentionwas given to effects of this process on the distribution of salmon (Cha pters3,4).

Chapter 1: Habitat selection behaviours in habitat modelling and fish-habitat management

1.1.Habitatmodels in resource management

Anunderstandingofhoworganismsaredistributingarnongavailablehab itatsiscru cialto managingnaturalpopulationsof animals.To achievesuch an understanding,distributionsof organismsare srudiedrelativeto distributionsof resourcesand conditionsthoughttobeof importance.Habitatmodelsaim at quantifyingrelationsbetweendistributionsoforganisms and habitats,and as suchare animportant part of resourcemanagement:Habitatmodelsare widelyusedfor a varietyofaquaticas wellas terrestrialspeciesand habitats(cf.Dueletal. 1996).

Implicitassumptionsof suchhabitatmodellingapproachesarethat(1) habitat limitspopulation levels;(2) "better"habitatsarecharaeterisedbyahigherdensityorfrequen cy-of-use, Le.

densitycan be usedas an indicatorof habitatquality; (3) habitat selectionisi mportantto distributionsof organisms,i.e.thesedistributionsare largelydrivenby habitatseleetion behaviours;and (4) habitatseleetionmodelsbasedon observationsofindividualsorsmall groupsof organismscan be used to addressproblemsat the populationleveI,Le.processesthat operateat smallspace-timescalesare importantto dynamicsat space-timescalesmuchlarger than those of the initialobservationsand small-scalehabitatseleetionmodelscan be used to predictor describedistributionsat largespace-timescales.

Itiswellknownthatassociationsbetweeno rganismsandtheirhabitatsvarywithscale(cf.

Wiens1973,Morris 1987A-C,Piatt1990,Syms1995,Poizatand Pont 1996)andt hat the relativeimportance of processesvarieswiihscale(Home and Schneider1994).Becauseof this,a scale-explicitapproachis neededtoidentifyimportant processes,variables,and scales.

Nevertheless,most habitat modellingstudiesusea singleor few measurementscalesand animplicit use ofscaling,despite an awareness oft hei mportanceo fs cale(cf Frissellet al.

1986,Minshall 1988,Imhof et al. l996,Lewis et al.l996,A1lan et al. 1997).The measurement scale chosenisoftennot theresultof aquanti tativemultiscaleapp roach, butisbased on the biologicalintuition oftheresearchercombined withlogistical constraints;the scale-up fromobservationto problemisintuitive,seldomly madeexplicit, and rarelyquantified.

In thisthesisI showtheimportance ofscaletohabitat modelsand resourcemanagement:1 developseveralnew scaling techniquesthatcanbeusedinhabitatselection and habitat modellingstudies. Thesetechniquesallow for aqu antitativeand scale-explicit assessmentof fish-habitat associationsand an evaluationoftheimportanceofhabitatselection tohabitat modelsand resourcemanagement.Basedonthesetechniques,Iinvestigatewhether possibilitiesexistforimproving habitatmodelsbyusingscale-explicitapproaches.Thethesis focuseson Atlanticsalmon(Salmo salar).TheideasIpresent,however,arenotrestrietedto managementof salmonpopulationsalone.

1.2. Habitat models in fisheries and tish-habitatmanagement

Habitat models are widelyappliedto riverinefish populations where theyfind usein stream habitatinvestigationsandin the resolutionof conflicts arisingfrom water allocation and hydropower development(Fauschet a!.1988,Reiseret al. 1989,Armour and Taylor 1991).Habitatmodelsare basicaIlydose-responserelations,with"habitat" as doseand

"habitat use"asresponse.The mathematicalformof these modelsmaybe multivariatemodels, frequency-of-useeu rves,preferenceeurves,orweighted-useable-areas,withexplanatory variables mostlyreferring to abiotichabitatco mponents (Orth and Maughanl 982,Fauschet al.1988).Variablesmostcommonly includedin fishhabitatmodelsare(I) drainage descriptors,such astotal streamlength,streamorder andstreamgradient,orchemical parameterssuchas conductivity (macro-scalevariables),(2)c hannel morphometry and flow

descriptors,such as discharge,streamwidth,mean water velocityand stream depth, or broad-scalefeatures such as pools, rifllesand runs(meso-scale variables),and (3) fish micro- habitatdescriptors,such as water depth, watervelocity, coverand substrate(micro-scale variables)(Fausch et al.1988).Variablesreferringto biologicalhabitat components,such as invertebrate drift or food availability,are seldomincluded, despitethe faet that food availability and drift concentrationsaffeetfish distributions(Jenkinsetal.1970,Griffith 1974, Gibsonand Galbraith1975,Wankowski 1981,Fausch 1984,Hughes and Oi1l1990,Hughes 1992A, 1992B). Thisfocus on physicalhabitatvariablesoriginatesfrom the faetthat othervariables are more difficultto measureand requirean often unrealistictime demand for data-gathering (Gore and Nestler 1988).Habitat modelsmust referto variablesthat can be affectedby managementaetions(Fauschetal. 1988).Decision-supportsystemsthat relyon habitat models,such as the instreamflowincremental methodology(lFIM/PHABSIM, Bovee 1982, 1986,Milhousetal.1989)often aimat relatingbioticvaluesin equivalentterms to those used to estimateother uses of availablewater (Gore and Nestler 1988).

Fishhabitatmodelscanbeclassifiedasmicro-,meso-ormacro-habitatmodels,dependingon the spatialresolutionor"scale"of the explanatoryvariables.Micro-habitatmodelsdescribethe distributionof individualfishover small-scalehabitatfeatures.Meso-andmacro-habitat modelsdescribefishdensitiesasafunetionofmediumtolarge-scalehabitatfeatures. The distinetionbetweenmicro- ,meso-andmacro-habitatmodelsisnotweUdefined.In this paper I willreferto micro-habitatmodelsas modelsbased on habitatfeatures smallerthan 1 m²,to meso-habitatmodelsas modelsbasedon habitat featuresrangingfrom 1 m'to 1000 m²,i.e.

one to severaltimes the width of the river,and to macro-habitatmodels as modelsbased on habitatfeatureslarger than 1000 m'(\argereach, tributary,or riverscales).

Exarnplesofsalmonid micro-habitatmodelscan be found in Shirvelland Morantz(1983), DeGraafandBain(1986),RaIeighetal.(1986),Morantzet al.(1987), Lambertand Hanson (1989),Heggenes(1990),Heggenesand SallVeit(1990),Heggenes(1991),Harrisetal.(I992) and Nehringand Anderson(1993). These modelsare generallyderivedfrom direet

observationsof individualfish,oftenobtainedby snorkellingorelectroshocking(BoveeI986).

ThespatialscalesoftheseobservationsareintherangelO-'tolm', dependingon the precisionof positiondeterminationand resolutionofhabitatobservations.The temporalscales of these observationsrangefromsecondsto severalminutes,dependingon the timespent observingindividualfish.At thesespatialand temporalresolutions,habitatuse willvary primarilydue to habitatselectionbehavioursand individualmovements.

Examplesofsalmonidmeso-and macro-habitatmodelscan be foundinBinnsand Eiserman (1979),Raleigh(1982),Bowlbyand Roff(1986),Lankaet al.(1987),Kozeland Hubert (I989A),BozekandHubert(l992),Amiro(l993),Gibsonetal.(1993) and Scrutonand Gibson(1993).Thesemodelsaregenerallybaseduponinformationon fishdensityand habitat in riversections.This informationis obtained byremoval-ormark-recaptureestimates,using electrofishingequipment,barrier-netsorseines.The spatialscalesof observationsthat underlie these modelsare usuallyin the rangeof 10' to 10^4m2.The temporalscalesrangefrom I sec to more thanseveralweeks,dependingon whetherobservationalunitswere blockedoff and hencedensities reflect an instantaneouspictureoffish densityat the observationalunit,or whetherdensitieswere monitoredover a periodof time,e.g.asin mark-recaptureestimates fromunclosedareas.Atthesespatialandtemporalresolutions,habitatusewillvaryduetoa complexmixtureof movementand mortality.Some studiesuse a combinationof micro-and meso-macro-habitatapproaches (cf Bozekand Rahel 1991)

The currentstate-of-the-anof habitatmodelswas developedlargelywithin the lasttwo decades,and habitatmodellingtechniquesare fast changing.Habitatmodelshavebeen developedsincetheI970's(Fauschetal.1988),althoughb iologistshavestudiedrelations betweenfishand theirhabitatsfor a lot longer. In particular,the PHABSIMcomponentof the InstrearnFlowIncrementalMethodology(Bovee1982),a micro-mese-habitatmodelling approach,is frequentlyused inwater allocationconflictsand hydropowerdevelopment(Orth 1987).Currentresearcheffortsfocuson the developmentoflocal modelsfor differentriver systemsorregions(e.g.DeGraafandBainI986,ScrutonandGibsonI993),oronan

evaluationofthespatio-temporalgeneralityof models(e.g. KozelandH ube111989B, HeggenesandSaltveit 1990,Bozekand Rahel 1992).In addition,effOI1S are madeto increase the descriptiveand predictivepower of modelsby addingmore and more detailand realism.

Examplesare a changeinfocustowardstwo and threedimensionaiflow models,the developmentof dynamichabitatmodelsthat addresschangesin habitatsand habitat requirementsovertime, the developmentof modelsoffish metabolismand driftfeeding,and thedeterrninationofmicro-habitatrequirementsofstrearninseets,an impoltantsourceoffood forfish(cf.Leclercetal.1996).

Habitatmodellingapproacheshavebeenwidelycriticised(cf.Orth and Maughan 1982,Van Home 1983,Mathuret a1.1985,Bleed 1987,Orth 1987,Fauschet ai.1988,Gore and Nestler 1988,Barinaga1996).In short,feweffortshavebeenmadeto test the predictivecapacityof modelswithindependent data. Thereis littleevidencethat fishesrespondto changesin model parameters.Modelsare oftenbasedon few data.Observationaldata on fishdensities.

individuaifishand habitatvariablesmaybe biased.Soundstatisticalproceduresare often overlooked.Methodsfor choosingthe best modelare poor.Fishdensityrnaynot be limitedby habitat,but by other factorssuchas exploitation.Variablesthataremorerealisticwithrespect to the biologyof the fish,suchas food availabilityand bioticinteractions,are oftenoverlooked.

Effeetsofflow alterationsmaytake manyyearsbeforethefulJimpacton habitatsandfishmay be recognised.whichlimitsthe possibilityto assessthesechanges.Habitatmodelsmostlyrefer to game animaisor other speciesthat are of interestto the generalpublic,butignoreother species.Temporaivariationsinhabitatandhabitatrequirementsareseldominciuded.Habitat modelsare often derivedfrom specificlocationsat specificmomentsin time.Most habitat modelsare basedon observationson habitatusein summer. Modelsrarelyinciudehabitatuse in winter.at night,duringhightlow ortlood conditions,orat placeswheresamplingisdiflicult.

From this. importanthabitatsor criticallifestagesmaybe overlooked.Fishdensitymaynot be a good indicatorof habitatquality.Synergisticeffeetsamongresourcesand/orconditionsare often ignored.

In spite of manyshortcomings,habitatmodellingapproachesare stillwidely useddue to their arguedefliciency,theirapparentsimplicityandcorrespondingease-of-use,andforlackof betteraitematives(GoreandNestlerI988) .

1,3.Development of effective habitat models

Three aspectsof modelsare of importance:realism,precisionandgeneralism(LevinsI966).

From this,the ultimatehabitatmodelwouldbe basedon functionalrelationsbetweenfishand habitat(realism),explaina largeportionof the observedvarianceand giverepeatableresults (precision),andbe applicable10differentaquaticsystemsat differentmomentsin time (generalism).Levins(1966)proposedthat at most two of these threeassetscan be attained.

ThisissupportedbyfindingsofFauschetai. (1988)whoconciudedfroma reviewofa large numberof habitatmodelsthat precisehabitatmodelsoftenstem fromrelativelyshort periods (one season)or fromsmallgeographicareas(singlestreamor watershed)and that precise modelsoften lackgenerality.In addition,itisimportantto note that a modelis a simplification of reality,i.e.simplicityis a model'svirtueand not necessarilyits weakness.Complexmodels are often implicitlyfavouredover simplerones,as morecomplexmodelsseeminglytake into accountmore of the processesthat are thoughtto beofimportance and fromthis,are assumed to mimicrealitybetterthan simplermodelsdo.However,modelcomplexitymay not necessarilybe positivelyassociatedwith modelrealismor precision.Whenaddingcomplexity tomodels,e.g.toinc reasemodelrealismorprecision,wehavetobalancepossiblebenefits with the associatedincreasein researchcosts.In addition,we haveto carefullyassessifthe complexityof the modelcorrespondsto areal understandingof the systemratherthanmerely supportingsomespeculation.lngeneral, an increasein modelcomplexitywillput a disproportionatedemandon the abilityof the researcherto understandanddescribemodel componentsand interrelationsand on modelvalidationefforts.Simplemodelsmaybe less effectivein describingobservedfishdistributions(explainedvariance10wer)than more complexmodels,but maybetterpredictdistributionswhenextrapolatedover spaceor time,i.e.

simplemodelsmaybe more robust (cf.Fauschetal.1988).

Therefore.itisimportanttoidentilYclearlytheobjectivesofhabilatmodelsbeforeundertaking fieldworkbyprioritisinggeneralism,realismand precision:modelsaimedatSlUdying fundamentalmechanismsgovemingfishdistributionsin a particularwatershedmayhaveto sacrificegeneralismforrealismand precision;modelsforfisheriesmanagementthatare tobe usedover widegeographicrangesmayhaveto sacrificeprecisionandlorrealismto anain generality.The most successfulapproachforfisheriesmanagementmaybe to developmodels that are realisticinthe firstplace. Habitatmodelsofdilferent complexityshouldbe compared to assesseffectsof modelcomplexityon modelgeneralismand precision.Habitatmodels shouldaimat descnbingthe mostimportantprocessesfirstbeforeincludingothers.Variables that mostincrease modelprecisionandlormodelgenerality,withthe leasteffectson model complexity,shouldbe includedfirst.A theoreticalframeworkis necessarythat clearlyIinks theoriesof distributionwithhabitatmodellingpraetices.Becauseof this,tooIs are neededthat a1lowfor(l) a prioritisationof distributionprocessesand (2) a prioritisationofvariablestobe includedin habitatmodels.TheselWoaspectswillbediscussedinthenextlWoseetions(l.4 andU).

1.4.Prioritisationofdistributionprocesses

The problemis to developan understanding of the possibilitiesand limitationsassociatedwith the use of small-scale observations of individual behaviours or density information on small groups offish, to dynamics at scales relevant to management problems (scale-up) ,which generally arise at time scales of years to decades and space scales of rivers orwatersheds(cf.

Imhofetal.I996,Richardsetai.1996).This scale-up can be quantified as the range of the problem relative to the resolution of the observations and can begraphically depicted ina so called"sco pe-diagram " as proposedby Schneidereta1.(1997).This approachis illustrated in Figure1.1for a generic micro-and meso-habitat modelling study,assuming a river0fSOkm lengthwithanaveragewidthoflOm. The lengths of the arrows connectingdata resolution and problem range indicate the degree ofscale-up or "scope" .

second month week day century

-I

^PROBLEM

.'.

survey alltransects ~

f!

/

i

transe~-;1".

/!L ,

^ns

j _·~:ation

_section

104

0

¹⁰³

l 1Q2

";10¹

~

100

~1Q-1

[10-²

~

^10-3 10-' 10-⁵

lD-310-^210-' 1001Q11Q21031041D510⁶¹⁰⁷ SpatialScale (m²)

Figure!.I.Scope-diagramillustratingscale-upin habitatmodelling.Thearrows conneetingobservation to problem relate to scale-upfromdireetobservationsof fish behavioursby snorkellinginmicro-habitatmodelling;the arrowsconneetingseetion to problemrelate to a meso-habitatmodellingapproachwhereriversectionswereblocked off andsampledbyelectrofishing(seetext).Doned arrows indicate assumed orintuitive scale-up.Solidarrowsindicate scale-upverifiable bystatistical inference.

Whensurveyingthis hypotheticalriverby snorkellingalongtransects(typicalmicro-habitat modellingapproach),the arrowconnecting"observation"and"transect"indicatesthe degree of scale-upfromindividualfishobservations(10sec,0.01m'')to individualtranseets(30 min;

20m").The arrowconneeting"transect" and"all transects"indicatesthe degreeof scale-up from individual transeetsto a total surveyconsistingof50transeets (50·30min,50·20 m²).

Transeetsaregenerallynot surveyedconcurrently,norare they positionedadjacently,Rather,

transects are positionedoverthe length oftheriver,and transectsarevisited overa period of severaldays to weeks.AJl transects combined,therefore.r epresentalarge r space-scale(river) and an"average"habitatuse ofa longertime-scale(one week; assuming thisis therimeneeded to visitallSOtransects).Thisscale-upis representedbythe arrow connecting"all transects"

and"survey". Manyhabitatmodelsare derivedfroma singlesurveydonein earlysummer, sinceflowconditionsinfilll,winterandspringoftenpreciudesarnpling.Therefore,anassumed or intuitivetemporalscale-upis done by usinginformationfroma singlesurveyas a basison whichto managefishpopulationsat timescalesrelevantto mostmanagementproblems: summer habitats are assumedto belimiting Thisis representedbythe arrowconnecting

"survey" and"problem"."Intuitive"in this contextcontrastswith the other scale-up routines(core ...visit ...survey)whichareverifiable usingstatisticalinference.

Asimilarscale-up isdoneformeso-habitatmod~ls.Intheexarnpleo fF igure l. l,fishdensities areassessedbyelectrofishingin 2SseClions ofSOmlengththat are blocked-offwithbarrier nets priorto removalof the fish.Assumingthat sectionswere sarnpledconsecutivelyovera periodof one month,the arrowconnecting"section" and"survey"indicatesthedegreeof scale-upfromobservationsatindividual riverseetionstothetotal survey(SOOm'persection;

2S*SOO=I2S00m'total;onemonthperiod).The arrow connecting"survey" and"problem", indicates theintuitivescaleup fromthissingle surveyto scalesrelevantto most management problems.

FromFigurel.litisobvious thatfor habitatmodellingsurveys,the degreeofscale-upisoften considerable.as the total area surveyedand thetotalarnountof timespentsurveyingis onlya fractionof the spatio-temporalrangeassociatedwithtypicalmanagementproblems(river, watershed;yearsto decades).!naddition,lhevariablesofhab itatmodelsaremeasuredat spatio-temporalscalesthat are muchsmallerthanare thoseofmostmanagementproblems, especiallyinthe contextof micro-habitatmodelling.

The relativeimpon ance of processes isknownto vary with spatio-temporalscales (Horne

and Schneider 1994).Because of this,small-scale behavioural processes that are imponant to habitat selection maynot necessarily be relevantat rhe larger scalesof our problemsandsmall-scalefish-habitatassociationsasdescribedinmicro-meso-habitat models may not necessarily beimpon ant to larger-scaledistributions.E.g.,small-scale habitatselectionmodels will not be elfectiveatdescribing dilferencesin fish densities among tributaries when these dilferencesare driven bydemographic processesinstead of habitatselection behaviours. Therefore,the scale-up from observationto problem will haveto be validated.Thisvalidation processis largely ignored in fishhabitatmodelling,in spite of the fact that problems associatedwith scale-up have been recognised(cf.lmhofet aI.1996). The reason for thisis that the collection of data needed for a quan titative evaluationofscale-upislabourintensive,aslargegeographicalareas will have to be sampled overlong periods of time.Another reason maybe a lack of knowledge of mathematical techniquesthat can be used forquantitativeevaluationsofscale-up.These problems will be further discussed in Chapter 2.

Fish are associated with theirenvironment overa range of spatial andtemporalscales.

This is because they react to their environment at a range of scales,because processes that alfect fish distributionsoperate over a range of scales and because of the propagation of effects from one scale to another.An exampleof behaviouroperating atmorethana singlespatial scale is the selection bysalmonidsfor specificholdingpos itions(small -scale) with relativelylow snout velocities in areas of highcurrent cond itions(larger scale) where drift is concentrated (Chapman and Bjomn 1969,Everestand Chapman 1972,Wankowski and Thorpe 1979,Fauschand White 1981).Anexampleofthepropagationofelfects across scales are the effects oftlood eventsand ice scour (small temp oral/largespatial scale events) on riverine fish populations (largespatio-temporal scaleeffect) (cf.Errnanet al. 1988, Fausch and Bramblett 1991,Pearsonsetal. 1992). Althoughflashtloodsmay have detrimental influenceson riverinesalmonid populationsat the timetheyoccur,the long-

tenneffectoffJashfJoodsmaybethatsuitablesubstratesaremaintained,thatfishspeciesare favoured that are adapted at re-colonisingandrninirnisingexposureofvulnerabIe life history stages(pearsonsetal.1992) ,orthatorganicmatterfromtheterrestrialenvironment is added to the riverine ecosystem, and thus actually sustains the standing stockofsalmonids.Another example of the propagation of effects across scaJes is the proces s of expansion and contraction,wherelarge-scaledistributionsareinfJuencedbysmall-scalehabitat selection processes (MacCallI990,Swain 1993, Marshall and Frank 1995).

Because organisms are associatedwiththeir environment at a rang eofspatialand temporal scales, a comprehensiveunderstanding of factors affecting the distributionand abundance offish can only be achieved bystudying factors affectin gfish distributionsat a range of scales,rather than one or even a few selected scales.From this,multi scale habitat models may be more effectivein describing associations0ffish with their habitats than single scale approaches."Effective" in this context refersto models with good descriptiveorpredictivecapacities,basedonaselectedandsmallnumber of variables and scales.

An example of this in the spatial domain is selectionby salmonids for specific holding positions with relatively low snout-water-velocities in areas0fhigh current conditions.A habitatmodel of this behaviour will indicate anegativeassociation withhigh water velocities at small spatial scales, but a positive association atlargerspatialscales.Asthere is no single "right"scale to describe this behaviour,a multi seale approach may be more appropriate for describing such behaviours.

Another example in the spatial domain is selection by salmon of spawning substrates.Ifsalmon need spawning substrates in patches ofacertain rninimum size, rivers that harbour spawning substrates in smaller patches only may not be suitable for spawning. Further, the relationship between spawning substrate patch size and suitability for spawning may not be linear for patchesexceedingthesca1esofreddseither. Currentspawningsurveysgenerallyoperateat

sca1esofriversections(severaltimestheriverwidth},largelybecausesurveysaredoneeither by helicopteror bya quick walk along the riverbank,i.e. measurementscalesare very much determinedbylogistics.However,the scale-mismatchbetweenthe sca1eof the surveyand the sca1eofredd-selection-behaviourmay lead to wrongfulpredictionson spawninghabitat quality andavailability,whichmayaffectsubsequentinstreamimprovementdecisions.

Anexamplein the temporaldomainis theinfluence of hightemperatureson mortalityof salmonids. Ifhigh temperaturepeaks occur at the scaleof hours, mortalitymay not be affected. However,ifhightemperatures occur at the scaleof days, no fishmay survive.

Currently,themajorityofhabitatmodellingapproachestendstooperateata single or few selected scales.Because of this,other important scales mayhav e been overlooked.In addition,by studying fish distributions and associationsoffish with theirhabitats over a range of scales, rather than a single or few selected scales,one may avoid a situation where measurement scales are chosen primarilyfrom an anthropocentric interpretation of fishbehavioursandlife-history(cf.KotiiarandWiensI990}.

Habitat modelling may greatlybenefit from a more explicituse of scale within the context of quantitative multiscale approaches.Thiswould involve an assessment of how patchiness offish distributions and habitats varies with scale and0fhowassociationsof fish with their habitats vary with scale. This would helpidentify important processes that affeetfish distributions and the scales at whichthey operate.The identification of scales at which fish distributions are most extreme, i.e.,scales at which patchinessis most different from random and variabilityis largest, and theidentification of scales of maximum association between fish and their habitats,may help to identify measurementscalesthat are most efficient to habitatmodels.

Multi scale approaches allow for an assessment of how the spatialand temporal heterogeneityof habitatswithina landscapeor landscapemosaicinfluencesspeciesoccurrence

and habitatuse(cf: Tumer1989). Fishhabitatmodelstend to focuson the effectsof habitat availabilityon habitatuse but tend toignore the effectsof the orientationand struetureof landscapeelements,especiallythoseusedwithinlFIMIPHABSIM.Systemsthat consistof similarhabitalSbutwherehabitatsoccuratdifferentpatchsizesorwherehabitatsare positioneddifferentlywithinthe landscapemayharbourdifferentspeciesand densitiesof organisms(cf. Riemanand Mclntyre1995). Examplesincludeeffectsof habitatfragmentation (cf.OehlerandLilVaitisI996),patchinessofresourcedistributionsanddispersionoforgarusms onspeciesoccurrence,communitystruetureandabundance,such as work byChamov (1976) and Parker and Stuart (1976)(MarginalValueTheorem),the HabitatTemplet,proposedby Southwood(1977),Grime's (1974, 1979)classificationof plantlifehistories,thedistinetion betweenrand Kseleetinghabitats(pianka 1970), and theoriesrelatingto islandbiogeography (MacAnhurandWilson 1967).

Becausespatio-temporalhabitatheterogeneityis of such importance10habitatquality(cf.

Wiens1976),measuresof habitatqualitythat includeheterogeneitymayperform berterthan thosewhichdo not,Severalauthorshavetriedto addressthis problemby classifyingstream habitatsat multiple(hierarchical)scales(e.g.Frisselletal.1986,Hawkinsetal.1993,lmhofet a1.1996)thatcanbeusedasaframeworkforevaluatingfish-habitatrelationsinriver restorationprojects.However,thescalesintheseclassificationsareoftenbasedonan anthropocentricinterpretationof processesand,becauseofthis,maylead to a situationwhere scalesand processesimponantto fishare overlooked.Multiscaleanalysesbasedon empirical studieswillbe neededto funheridentifyimponantprocessesand scales,and to evaluatethe relativeimponance of processeswithscale.Thesevariousaspectswillbe funherdiscussedin Chapter2 and Chapter3.

1.6.Density as indicator of habitat quality

So far,I havediscussed the studyof distributionpatternsasameans to identifyunderlying processes.However, the relation betweenpattern and processis notunidirec tionaland distributionprocesses may vary withdistributionpattern and density;processesinduce patterns and patterns determineprocesses:

Previouswork has shownthat salmonidsselectpositionsin streamsbasedon theircompetitive abilitiesand the profitabilityof positionsin terms of potentialnet energyintake rate and predationrisk,with profitabilityof positionsbeingmuchdeterminedbyth epbysicalhabitat in termsof cover,bottom topographyand currentflowpatterns(Fausch 1984,Hughesand Dill 1990,Hughes1992A,1992B,Grand 1997,Grandand Dill 1997).Assuch,the area withina streammaybe regardedas a hierarchyof potentialpositions,rangingfrominaccessibletoideal, with eachfishchoosingthe most profitablepositionthat its rankin the socialhierarchywill allow(FauschI984,HughesI992A). Territoriality,srnall-scalespacingbebaviourorpre- ernptiveexclusionwillthus regulateuse of preferredpositionsand space,if in short supply,will regulatepopulationdensity(Bohlin1977,GrantandKramer1990). From this,the physical habitatrnayberegardedasatemplatedeterminingdistributionpattemsoffish(Hughes 1992B).

Theseprocessessuggestthatsalrnoniddistributionsrnaybebestdescribedusingthe ideal- despoticdistributiontheoryofFretwell(1972). Thistheorydescribeshowanirnalsselecttheir habitatsassumingthattheyare"ideal "inknowingwhereprofitab ilityishighestbutwhere accessto resourcesare governedby territorialbehaviours.Whenorganismsdistributeideal despotic,the most desirablepositionswillbe occupiedfirst,followedbypositionsin progressivelylessdesirablehabitats.Becauseof this,the averagegainper individualmaydiffer andhabitatusernaychangewithdensity.·Fromthis,habitatrnodelsrnayvarywithpopulation density.

The ideal-free distribution theory (Fretwell and Lucas 1970) contrasts with this ideal-despotic theory in that access to resources is not restricted by competitive behaviours but all individuals are equal and "free" to move among patches without constraints or restrictions. When organismsdistnbuteidealfree,fitnessofindividualsdeclineswithdensityasindividualsoccupy the best habitats, the average gain per individual will stabilise to be equal in all habitats, and the fraction ofa population in each habitat should equal the fraction ofresourcesoccurringthere (cf. input matching; Parker 1974). When organisms distribute ideal freeamonghabitats and the rate of resource renewal in these habitats is not affected by organisms densityordistribution and all habitats are occupied at low population densities,then relativedensities in habitats do not vary with population density.

When distributionschange with density,habitat models are expect ed to change with density as well. As a consequence,managerialactions may vary with population level.

However, a quantitative evaluation of how impon ant density-dependent effects are relative to density-independent effects in shaping fishdistribut ions has not been done.

Because of this, it is not known ifor how much habitat models change with density.This will be further addressedin Chapter 4.

1.7. Conclusions, research questions and thesis outline

Fish distributions are the result of multiple processes operating at multipie scales.Fromthis, fish are associated with their environment at multiple scales. Because fish are associated with their environment over a range ofspatio-temporal scales, a comprehensive understanding of processes affecting fish distributions can only be achieved by studying associations0ffishwith habitats over a range of scales.Scaling analyses and theory can act as a framework that allows forconnectingresultsfromstudiesoperatingatdifferentspace-times~es.

Tools for fish-habitat management maybe most successfully developed within the framework ofrea1istichabitatmodels, Le.modelsthatarederivedprimarilyfrombiologicalknowledge

ratherthan from correlation aJone.AsfishdistributionsareultimatelytheresultofindividuaJ decisions, an understanding of habitat selectionbehaviourofindividuaJswill be important to fish habitat-management. Important research questions in this context are:(I) how do fish perceive and reaet to their environment;(2) is habitat use or density indicativeofhabitat quality;(3) to what extent are fish distributionsdrivenby habitat selection and to what extent by other processes;and (4) how can we extrapolateindividuaJfish behaviours to scaJes relevant to management problems?

The thesis focuses on questions 1,3 and 4in the context ofhabitatusebyjuveniIe Atlantic salmon in rivers.Habitatselection was defined asa process ofindividuaJschoosing arnong options (different habitats) based on some preference.A habitat in this context is a space where an organism lives, with "space"referring not onlyto area or volume butalso to the resources that maybe obtained and the conditions within this area orvolume.

I mostly aimed at achieving an understandingof"how" saJmonparrselecttheirhabitatrather than"what" they are selecting for, and of the implicationsof habitat seleetionbehavioursto habitat models.Habitat is described largelyin terrnsofsubstrate, water depth andwater velocity, as these are the variables most often included in habitat mod els of riverine fish species (Orth and Maughan 1982,Fausch et at. 1988, Heggenes 1990).My fieldwork (Chapters 3-4) concentrated on spatialanalyses operating at small to intermediary scales

«100m²),because these areimpo rtant to habitatselectionandhabitatmodellingand becauseoflogistics.

Inthis thesis,I firstevaJuated the scaJe-upin habitatmodelling from behaviouraJ0bservationto fish-habitat problem (Chapter 2).Next, I presented a new scaJing method that can be used in habitat selection and habitat modelling (Chapter 3.1),extended this technique using data from an experiment donein a stream tank (Chapter 3.2),and applied the techniques de velopedin Chapter 3.1and Chapter 3.2to a field-based study(Chapter 3.3).Chapter3islargelyfocused on effeets of habitat selection on distributionpatterns.By contrast,inchapter 3.2and chapter

4 I showed how distribution patterns may affect habitat selection processes.This was done by studying density-dependent habitat use, using a combinationofan experimental(Chapter 3.2,4.I)andobservational(Chapter4.2)approach.Inthelastchapter(Chapter5)I summarised the various studies and diseussed implications to habitat modelling and fish-habitat management. To facilitate readability, I organised the thesis such that chapters and study projects can be read separately.Because of this, the differentchapters may show some overlap.

The objectives of this thesis were (1) to illustrate howa variety0fnewlydeveloped scaling-techniques can be used in habitat modelling andbehaviouraI studies;(2) to evaluate limitationsof using information on small-scaleobservations and experiments to address problems at scales relevant to fish-habitat management;(3) to identity scales important to habitat modelsfor juvenile Atlantic salmon;(4) to formalise observed habitat selection behaviours that operate at multiple scalesinto explicitmultiscalehabitat selection models;(5) to study density-dependent habitat selection;and(6)tocompare explicitmultiscaleapproacheswithsingiescaleapproachesinregardto their ability to identity how fish seleet theirhabitats and in theirability to describ eand predict fish

I hypothesisedthat (1) multiscale approaches are better for unders tanding and describing fish distributions because habitat selection behavioursthemse!vesoperateatmultiple scales; (2) habitat use changes with density due to small-scale spacing behaviouror territoriality of individual fish; (3) multi scale habitat models perform better than single scale habitat models,especially when extrapolatingsmall-scalehabitatselectionbehaviours to density-predictions at larger spatialscales,i.e.observed andpredicteddistributionswill be more similar when using multi-scalehabitatmodels;(4) small-scale behavioural processes or small-scale fish-habitat associations will be limited forexplaininglargerscale distributionsor addressinglarge-scale habitat management problerns.

Chapter 2: Mortality versusspatial dynamics at multiple scales:

scaled-rate plots forsalmonidsand implications for habitat modelling

2.1. Scale-upinecological studies

Understanding howorganisms interactwith their natural environment is crucial to the management of natural populations.Toobtainthis understanding,man uses surveys,field and laboratoryexperimentsto studythe distributionsof organismsrelativeto environmental factors.The relative importance ofprocessesis known tovary with spatio-temporalscales (Home and Schneider1994).Consequently, processes that are important at the smallerscalesofexperimentsor mostfield observationsmay not necessarilybe important at the largerscales of ecologicalproblems.Developingthe ability to determinewhich processes predom inateat anyspaceandtimescaIe would greatly improve the efficiency of research and confidence inits generality.In tum,this should ideally lead to more effectiveenvironmentalmanagement.

Home and Schneider(1994) recentlyproposed a technique to evaluatetherelative importance of processesin a scale-explicitmanner. This method can also bean aid in scaling-upfrom experiments(i.e,extrapolating)to addressenvironmental problems at regionalor global scales (Schneideretal.1997).This techniquecompares demographic.

growth and kinematic rates via dimensionless ratios,whichare subsequentlyusedto indicate which processespredominateat a givenscale.Thisprocedureconsists of five steps:(1) state the quantityofinterest; (2) write a conservation equationinco rpo rating the sources of variability in the quantity';(3) form dimensionlessratios from the terms of the equation;(4) obtainvaiuesfrom the literature and calculate these dimensionlessratios for

"benchmark" spatio-temporalscales;(5) create a graphwith "temporal scale" and

• E.g.:numberof individuals=births~deaths+immigration- emigration

"spatial-scale" asYand X axes, respectively, and drawconto ur linesseparating spatio-temporal scales where denominatorand nominatorof ratesprevail.As this techniqueuses information from a limitednumber ofspatio-temporalscales (benchmark scales)with interpolation,I will furtherreferto thistechnique as the"benchmark"

approach.

Step1requires that the problem be defined usingquantitiessuch as biomass or countdata.

Theconservatio n equation(step2)ensures clo sure of the first moment (average) of the quantityofinterest.Formingall possibleratios(step 3)re-normalises the terms in the equation, i.e.the rate of change inthe numerator ismeasuredrelative to therate of change of the denominator .

Theadvantagesofthisapproacharethatallimponantprocessesareincludedandthat ratios arereadily obtainedfor literature values of component rates.A disadvantage is that interpolation between benchmarks is difficultbecausebenchmarksarefewinnumber.

Becauseofthis , rate -diagramsmayberough,approximate,anddependent on intuition.

Inthis chapter I extended the techniqueby using intensive compu tationratherthan hand-drawn linesbetween benchmarks , in an individual-basedLagra ngianapproachwith randomisation (Chapter 2.2).I illustrated thistechnique using severaltheoretical examplesfirst (Chapter 2.3).Next I developedrate-diagramsofmovementversus mortalityfor cutthroat-trout(Ol1carhy llclnls darla)and Atlantic salmon parr(Salma .<alar) from publisheddata (Saunders and Gee 1964,Heggenes et al. 1991) (C hapter 2.4).This combination ofexamples and real data was necessary because I found that detailed rate-diagramsare difficultto obtain from benchmarkscalesalone,panlyduetoscarcityof movementinfo rmation and partlydue to difficultiesassociated withi nterpolationfrom benchmark values.By first calculatingrate-diagrams fromrelatively simple co mputer -generatedmovement and mortalityscenarios and next combining thesewith rate-diagrams from observeddata,I was able to evaluatewhereinformation waslacki ng

and how this affects conclusions.Anadditionalobjective was to provide reference rate-diagrams for future studies.

2.2. Scaled-rate plots: method and calculations

Themodel simulates movement and mortality of individual organisms.Based on these FORTRAN-based simulations, critical scales are identified,i.e.space-time scales at which movement (M, year" ) equals mortality (D, year"; MID=R=1).Random numbers, needed for several of the analyses,are generated using the FORTRAN system-supplied random number generator,upgraded by the shuffle-routine as outlinedbyPressetal.

(1986).

Movement can be modelled along a transect (I D),in a plane (2D),or in a volume (3D).

For all three approaches,the main computational flow is similar. In this chapter the computational flow for the one dimensional transect application is presented.

(I)10⁴random locations were chosen along a transect (length=1000Ian)as initial positions offish.

(2) The transeet was subdividedinto consecutivebins of equal length(L).For this, a random location along the transect was chosen as a starting point.Next,Idetermined the section or bin in which each individual fish was positioned.

To avoid the problem of having sections cut-off by either the start0r the end of the transect,Iconneetedthese,leadingtoacircularorinfinitetranseet.This greatly facilitatedcomputationsandjudgednottoaffectconclusions,given the length of the transect. This was verified in additional analyses using longer andshortertransects.

(3) Movements and deaths of individualswere modelledfor a periodof time T(days), usinga random pointin theannuai cycleas a startingpoint.Afterthis periodof time (T), I determined the number oforganisms that had died (No),the number that were alive and stayedwithinthe originalsection(Ns),and the numberthatwere alive and moved from the originalsection(NM)within period T.From this,I determined whether Ns, exceededNo.

(4) These calculationswere performedfor a range(i) of sect ionlengths (Lx, x= l ,i).

(5) Calculatio ns1-4were repeated fora rangeU)of timeperiods(Tv, y=lj) ,each time usinga differentrandomtransect starting lo cation and adifferent random startingtime in the annual cycle.

(6) Calculations 1-5were repeated NRR times (Number Repeat Random isations;see Table 2.1).From these repetitions,I recorded the total number of caseswhe re NM exceededNo(=NCM) and the total number ofcases where No exceededNM(=NCo) forall space-time scales (L, T) involved .IfNCMexceededNCo,I concludedthat movementdominated mortality (R>I), i.e.the distributionwas driven bymovement rather than mortality.If NCoexceeded NCM,[ concluded that mortality dominated movement(R<I ).I determinedcriticalscales (R=l )using a subroutinethat compared NCoand NCMoverall spatialscales(Lx,x= l,i)for timescales (Tv,y=lj )separate.

Criticalscaleswere identifiedfrom a shift ofNCM>NCoat L,.,T,to NCM<NCoat L,._"T,_

Transect length, number of organisms and repetitions,and space- (L)andtime-(T)scales may vary with scenario. I decidedon the transect length, number of organisms and repetitions as outlinedinthe text above andin Table 2.1,as resultsdidnotchange in additionalanalyses that used longer transects,higher numbersof organismsand more

repetitions.

In general, I recommend calculating three movemen t/mortalityscenarios:(I) one describing movement and mortality in the best possiblemanner,givenavailable information, (2) one describing a low-movementlhigh-mortaJityscena rio, and(3) one describing a high-movementllow-mortalityscenario.Rate-diagrams ofthese three scenarios can then be compared to indicate the range of plausible ou tcomes.

2.3. Scaled-rateplols:examples

I calculated critical scales for 5movementscenarios offish distributed along the length of a river. These scenarios were chosen to represent a range of plausible out comes,with movement and mortalityranging from very low to very high, as described in the previous

(I) Territory (TER):Fish were modelled to occupyindividual territories.Fish never left their territories,but were free to reposition themselves withinindividual territories. Thiswas modelled by randomly repositioning fish within I meter of positionsmarking the centre of individual territories at each time step of the calculation."Territory"in this context does not refer to an area thatis defended and territories may overlap.

(2) Diumal movement(DM):To mimic diurnal movements within a home range, individual fish were modelled to move along the length of the river according to a sinefunctionwith anarnplitudeoflOO mand a wavelength of24 hours.

(3) Seasonal movement (SM):Individual fish were modelled to move along the length of the river,according to a sine fimction with an amplitude of 1000 m and a wavelengthofone

(4)Total-/(TSIN):Fishwere modelledto displayterritorial-,diumal-,and seasonal movementscombined:territorieswere occupied(I) and positionsmarkingthe centre of individual territorieswere relocatedbasedon the sinefunctionsof2-3.

(5) Total-2(TSQi:Fishwere modelledas for TSIN.However,insteadof sinewavesfor diurnaland seasonalmovements,squarewaveswereused.

Criticalscaleswere calculatedfor eachof these 5 behaviours, withthe relativerate of mortalitymodelledat 0.5 year"(TER-M50;OM-M50;SM-M50;TSIN-M50andTSQ-M50 respectively)and 0.75year" (TER-M75;OM-M75;SM-M75;TSIN-M75and TSQ-M75 respectively;see Table2.1).In addition,I calculatedcriticalscalesfor TER-M50;OM-M50;

SM-M50;TSIN-M50andTSQ-M50,with5% of the fishbeingrandomlyrelocatedwithin100 m of theirpositionsas determinedbyTER, OMand SM,for every24 hours (TER-M50/R;

OM-M50/R;SM-M50/R;TSIN-M50IRandTSQ-M501Rrespectively). I willexplainlaterin Chapter2.3 whyI chose this dispersionlevel. Estimatesof criticalscaleswere donefor spatio-temporalscalesrangingfrom3 hoursto 2 yearsand 1em to 100Ian.

Table2.1 summarisesthe scenarios.Figures2.1-2.5displaythe results.Thelinesinthese figuresconsistof allcriticalvaluesofR(i.e.,R=I),separatingspace-timescaleswhere movementdominates(R>I)fromspace-timescaleswheremortalitydominates(R<I).

• These movementscenarios were based on a combinationoffield experienceof the author and informationfromtheliterature(mostnotablySaundersandGeel964andHeggenesetaI.1991).

In addition,J.Hutchings(pers.comm.)confirmedthatthe movementscenariosweregenerally supportedby resultsfroma studyon brooktroutmovements in theCapeRacearea0f Newfoundland withthe exceptionof thediurnalmovements,whichwereprobablyoverestimated.

Table 2.1.Scenariosused for calculatingcriticalscalesof movementversus mortality.

Scenarioscompriseda combinationof movementand mortality(RMR,year") .Movement behavioursincluded:Territoriality(TER-),diurnal movement(OM-),seasonal movement (SM-) and randombehaviours(IR),as explainedin the text.For TSQ-*, diurnaland seasonal movement were modelledusingsquarewaves.For all otherscenarios,diurnaland seasonal movementswere modelledusingsinewaves. NRRrefers to the numberof repeat randomisationsusedtoestimate criticalscales.

TER S '0 RMR NRR

yes no no TER-M50 0.50 100

yes no no TER-M75 0.75 100

:tes no yes TER-M501R 0.50 100

5

yes no no DM-M50 0.50 500

no yes no DM-M75 0.75 500

6 no yes yes DM·M501R 0.50 500

7 no yes SM-M50 0.50 100

8 no no yes no SM-M75 0.75 100

9 no no yes yes SM-M501R 0.50 100

10 yes yes yes TSIN·M50 0.50 500

II yes yes yes no TSIN-M75 0.75 500

12 yes yes yes yes TSIN-M501R 0.50 500

13 yes yes TSQ·M50 0.50 2000

14 yes yes no TSQ-M75 0.75 2000

15 yes yes yes TSQ-M501R 0.50 2000

Figure2.1showstherate-diagrams for fishdisplayingterritorial behaviour(TER).The"jagged"

outline of the linesare the resultof theapproximationroutineused to detenninecriticalscales.lngeneral , movement dominatedat small space-time scales and mon ality dominatedatlargespace-time scales.Monalityalwaysdominated overmovernentat timescaleslarger thanoneyearforTER-M50and TER-M501Rand at time-scales larger lhan I83 days for TER-M75.

The reasonfor this is that50"/0of thefishdiedduring intervalsof one yearforTER-M50and TER-M501R (monality=O.5year·') and 50%

duringintervaJso f 183 daysfor TER-M75 (mon ality=O.75year").

During longer intervaJs,more than 50"/0of the fishdied and,becauseof this,monalityalwaysdominatedat thesetime-scales.During shoner spatial scale(ml

'00TER·MSO R<1

J\

'00TER·M7S ^R<1

Figure 2.I. Criticalscalesof territorialfish.Scenarios includeTER-M50 (top, 50"/0mortality peryear), TER-M75(middle,75%mortality peryear) and TER-M501R(bottom,50"/0mortalityper year+ randommovements).Territorialbehavioursare modeUedas explainedinthetext.(R=movement (year·l)versusmonaIity(year·').)

l - - ----'intervals,dominationofmo vement

over mortalitydependedon space-timescale.The differencein monalityofO.5year ·¹ (TER-M50)and 0.75 year"(TER-M75)resulted in aminorshiftto the leftand a majorshift down of the lineofcriticalvaluesfromTER-M50to TER-M75.Note that movementmay dominateover monaIityat spacescales much larger than thatofthesizeof individual

territories,especiallywhentime-scalesaresmall. TER-M501Ris largely detennined byrandom behaviours,withlinleinlluence of territoriality.

smalldegree due to the

spatial scale (m)

M ^{U ..IlSO R <1}

: R >1 /~

°0 ...175 R<1

.00

:~~ I

:;~, R<1

:~

_{\00 10' 10: H}3 HI' !O'}

Figure2.2.Critical scalesoffish displayingdiumal mo~ements.Scenariosinclude DM-M50(top,50%

mortalityper year).DM-M75 (middle,75% mortality peryear)andDM-M501R(bottom.50"IomortaIityper year+random movements).Diumal movementsare modelledusingsinewaves,as explainedinthetext, (R

=movernent(year")versus mortality (year").)

r - - - -- - - - -- - - ,Figure2.2showstherat~-diagrams for fishdisplayingdiumal movements(DM).The"jagged"

outlineof the linesare primarilydue todiumalmovementsandonl ytoa

approximationroutineused:no fish movedattime-scalesof 1,2,3•...

day,andmovement is maximalat time-scales of 0.5,1.5,2.5•...day.

Fromthis. diumal movements induced a characteristicregular patternwith a wavelengthof l day-scale.Thesmallwavelengthof this regular pattern,incombination with theresolutionoftheY-axis (temporal scale),makest he linesof criticaJvaJues(R=I)appearas a broad blackband.This isfurther illustratedinFigure 2.2by expandingpertions of thegraphs for temporalscalesof 95 to100days.

Asinthepreviousfigure,deminance of movementovermortality is

L - - - - 'restrietedtosmallerspace-time

scales. Again, we see a dominanceofm ertalityevermovementat time scales longer than one year forDM-M50,DM-M50IR and at timescaleslongerthan 183daysferDM-M75.The differencein mortalityof 05year"(DM-M50)and0.75year " (DM-M75),resultedin a

minor shifttotheleft anda majorshiftdown of the lineof criticalvaluesfromDM-M50 to DM-M75. The rate diagramofDM-M501Rislargelydeterminedbyrandombehavioursat space-tirnescalesleft ofthe band of critical values.Atiargerspace-timescales,diumal movementsdeterminetherat e-diagram.Animpression of this can also be obtainedby overlayingthe rate-diagramsofDM-M50andTER-M501R:movementdominated mortalityin DM-M501Rat scaleswhereeitherone or both ofthe rate diagramsofTER-M501Rand DM-M50lRindicatedthatR>I.Thelevel ofdispersionwaschosen such thatthis overlay-procedurecouldbe illustrated.