The Relationship Between Population Dynamics and Habitat Use of Terrestrial Carnivores

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/260001103

The relationship between population dynamics and habitat use of terrestrial carnivores

Thesis · January 2014

CITATIONS READS 0 278

1 author:

Garth Mowat Ministry of Forests, Lands, Natural Resource Operations and Rural Development

101 PUBLICATIONS 2,235 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Statistical Population Reconstruction for large mammals in British Columbia View project

Understanding landscape and human effects on wolverine abundance, distribution and connectivity in the Central and Southern Canadian Rockies ecosystem View project

All content following this page was uploaded by Garth Mowat on 03 February 2014.

The user has requested enhancement of the downloaded file. Norwegian University of Life Sciences

Department of Ecology and Natural Resource Management

Doctor Philosophiae (Dr.Philos.) Thesis 2014:01

The relationship between population dynamics and habitat use in terrestrial carnivores

Forholdet mellom populasjonsdynamikk og habitatbruk hos terrestre rovdyr

Garth Mowat

The relationship between population dynamics and habitat use in terrestrial carnivores

Forholdet mellom populasjonsdynamikk og habitatbruk hos terrestre rovdyr

Doctor Philosophiae (Dr. philos.) Thesis

Garth Mowat

Department of Ecology and Resource Management Norwegian University of Life Sciences

Ås 2014

Thesis number 2014:1 ISSN 0809-6392 ISBN 978-82-575-1160-9 Contents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

Abstract

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

List of papers

Case study 1: Canada lynx

3DSHU, 6ORXJK%*DQG*0RZDW/\Q[SRSXODWLRQG\QDPLFVLQDQXQWUDSSHGUHIXJLXP -:LOGO0DQDJH

3DSHU,, 0RZDW*DQG%*6ORXJK+DELWDWSUHIHUHQFHVRI&DQDGDO\Q[WKURXJKDF\FOH LQVQRZVKRHKDUHDEXQGDQFH&DQDGLDQ-RXUQDORI=RRORJ\

Case study 2: Short-tailed weasel and American marten

3DSHU,,, 0RZDW*&6KXUJRWDQG.*3RROH8VLQJWUDFNSODWHVDQGUHPRWHFDPHUDVWR GHWHFWPDUWHQDQGVKRUWWDLOHGZHDVHOVLQFRDVWDOFHGDUKHPORFNIRUHVWV1RUWKZHVWHUQ 1DWXUDOLVW

3DSHU,9 0RZDW*DQG'3DHWNDX(VWLPDWLQJPDUWHQSRSXODWLRQVL]HXVLQJKDLUFDSWXUH DQGJHQHWLFWDJJLQJ:LOGOLIH%LRORJ\

3DSHU9 0RZDW*DQG.*3RROH+DELWDWDVVRFLDWLRQVRIVKRUWWDLOHGZHDVHOVLQZLQWHU 1RUWKZHVW6FLHQFH

3DSHU9, 0RZDW*:LQWHUKDELWDWDVVRFLDWLRQVRI$PHULFDQPDUWHQVLQLQWHULRUZHWEHOW IRUHVWV:LOGOLIH%LRORJ\

Case study 3: Grizzly bear

3DSHU9,, 0RZDW*'&+HDUG'56HLS.*3RROH*6WHQKRXVHDQG':3DHWNDX *UL]]O\DQGEODFNEHDUGHQVLWLHVLQWKHLQWHULRUPRXQWDLQVRI1RUWK$PHULFD :LOGOLIH%LRORJ\

3DSHU9,,, 1DPV92*0RZDWDQG0$3DQLDQ'HWHUPLQLQJWKHVSDWLDOVFDOHIRU FRQVHUYDWLRQSXUSRVHVDQH[DPSOHZLWKJUL]]O\EHDUV%LRORJLFDO&RQVHUYDWLRQ

3DSHU,; 0RZDW*DQG'&+HDUG0DMRUFRPSRQHQWVRIJUL]]O\EHDUGLHWDFURVV1RUWK $PHULFD&DQDGLDQ-RXUQDORI=RRORJ\

3DSHU; 0RZDW*'&+HDUGDQG&-6FKZDU],Q3UHS3UHGLFWLQJJUL]]O\EHDUGHQVLW\LQ 1RUWK$PHULFD

Introduction

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

'HVFULSWLYHKDELWDWVWXGLHVKDYHOLPLWHGYDOXHIRUSUHGLFWLRQDQGHYHQFRPSDULVRQV DPRQJVWXGLHVDUHGDQJHURXVEHFDXVHGLIIHUHQFHVLQVFDOHRUDYDLODELOLW\FDQJUHDWO\ LQIOXHQFHUHVXOWV %R\FHDQG0DF'RQDOG0\VWHUXGDQG,PV%H\HUHWDO 6LPLODUO\WKHVHW\SHVRIVWXGLHVDOORZOLPLWHGRSSRUWXQLW\WRWHVWJHQHUDO K\SRWKHVHVDERXWIXQFWLRQDOUHODWLRQVKLSVEHWZHHQDQDQLPDODQGLWVHQYLURQPHQW *DLOODUGHWDO EHFDXVHKDELWDWVUDUHO\LQGH[DVLQJOHUHVRXUFHGURZQLQJIXQFWLRQDO DVSHFWVLQDPD]HRIFRUUHODWLRQVDPRQJYDULDEOHVDQGHIIHFWV *DUVKHOLV

%XWWKHSURFHVVRIKDELWDWFKRLFHLVDQDGDSWLYHSURFHVVZKLFKFDQEHWHVWHGLQWHUPVRI WKHLQIOXHQFHRIFKRLFHRQILWQHVV *DLOODUGHWDO 'HPRJUDSKLFSDUDPHWHUVVXPWR \LHOGSRSXODWLRQJURZWKDQGWKLVSDUDPHWHULVFRPPRQO\VWXGLHGWRLQYHVWLJDWHWKH SRSXODWLRQG\QDPLFSURFHVV 6LEO\HWDO 3RSXODWLRQJURZWKFDQEHGHFRPSRVHGWR DOORZWKHLQIOXHQFHRIHDFKOLPLWLQJIDFWRUWREHH[DPLQHG 6LEO\DQG+RQH 2QH FDQFRPSDUHILWQHVVDPRQJLQGLYLGXDOVWRLQYHVWLJDWHWKHLQIOXHQFHRIKDELWDWRQILWQHVV 0F/RXJKOLQHWDO EXWWKLVUHTXLUHVORQJWHUPGDWDVHWVWKDWLQFOXGHYDULDWLRQVLQ KDELWDWDQGWKHVHGDWDVHWVDUHUDUH *DLOODUGHWDO 'DWDVHWVWKDWPHDVXUHPRYHPHQW UHSURGXFWLRQDQGVXUYLYDORILQGLYLGXDOVDFURVVKDELWDWVDUHPRUHFRPPRQDQGWKHUHDUH PRUHFRPSDULVRQVRILQGLFHVRIILWQHVVDPRQJKDELWDWVLQWKHOLWHUDWXUH *DLOODUGHWDO 7KHFROOHFWLRQRIGHPRJUDSKLFGDWDDWYHU\EURDGVFDOHVLVGLIILFXOWDQGPRVW UHVHDUFKHUVKDYHXVHGGHQVLW\DVWKHPHWULFWRFRPSDUHKDELWDWTXDOLW\ VHH0RVVHUHWDO IRUDQH[DPSOH )RUFRPSDULVRQRIKDELWDWTXDOLW\DPRQJPDPPDOSRSXODWLRQV SRSXODWLRQGHQVLW\PD\EHDEHWWHUPHWULFWKDQJURZWKQRWZLWKVWDQGLQJWKHFRPSOLFDWLRQV LPSRVHGE\PHWDSRSXODWLRQG\QDPLFV 9DQ+RUQH

5HPRWHPHWKRGVRIVDPSOLQJDQLPDOVHVSHFLDOO\WKRVHLQYROYLQJJHQHWLFLGHQWLILFDWLRQRI LQGLYLGXDOVKDYHDOORZHGUHVHDUFKHUVWRVDPSOHFDUQLYRUHVDWPXFKODUJHUVFDOHVWKDQ SUHYLRXVO\,WLVSRVVLEOHWRVDPSOHDUHDVWKDWHQFRPSDVVHQWLUHSRSXODWLRQVRIWKHVWXG\ VSHFLHVZKLOHRQO\VDPSOLQJWKHKDELWDWFKRLFHRIDQLQGLYLGXDODIHZWLPHVGXULQJWKH VWXG\7KHIRFXVRIVDPSOLQJLVKHQFHWXUQHGDZD\IURPLQGLYLGXDOVDVLQUDGLRWHOHPHWU\ EDVHGVWXGLHVWRZDUGVSRSXODWLRQVVXFKWKDWLQFUHDVLQJWKHVDPSOHVL]HPHDQVVDPSOLQJ

PRUHLQGLYLGXDOV$WWKHVHVFDOHVFKRLFHLVQRORQJHUVLPSO\DPHDVXUHRIEHKDYLRXUEXW LVDOVRLQIOXHQFHGE\GLIIHUHQFHVLQGHQVLW\DPRQJHFRV\VWHPV$WWKHEURDGHVWVFDOHWKH UDQJHRIWKHVSHFLHVVHOHFWLRQLVOLNHO\WRPHDVXUHWKHSK\VLFDOOLPLWDWLRQRIWKHVSHFLHV

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

Objectives 7KHREMHFWLYHVRIWKLVWKHVLVZHUHWR

'HVFULEHWKHSRSXODWLRQG\QDPLFVRI&DQDGDO\Q[ Lynx canadensis DQGFRPSDUH WKLVWRFKDQJHVLQWKHLUXVHRIKDELWDWWKURXJKWKHF\FOHLQO\Q[DQGVQRZVKRHKDUH Lepus americanus QXPEHUV 3DSHUV, ,, 'HVFULEHWKHIRUHVWVWUXFWXUHDQGHFRORJLFDODWWULEXWHVWKDWLQIOXHQFHWKHKDELWDW VHOHFWLRQDQGGHQVLW\RI$PHULFDQPDUWHQ Martes Americana DQGVKRUWWDLOHG ZHDVHOV Mustela erminea 3DSHUV,,,9, 8QGHUVWDQGWKHIXQFWLRQDOFDXVHVRIYDULDWLRQLQJUL]]O\EHDU Ursus arctos GHQVLW\LQRUGHUWRSUHGLFWGHQVLW\DFURVVWKHSURYLQFHRI%ULWLVK&ROXPELD &DQDGD 3DSHUV9,,; &RQVLGHUKRZKDELWDWVHOHFWLRQVWXGLHVFRQWULEXWHWRWKHXQGHUVWDQGLQJRI SRSXODWLRQSURFHVVDVWKH\UHODWHWRLQGLYLGXDOVSHFLHV

Methods

Case study 1: Canada lynx

:HVWXGLHG&DQDGDO\Q[XVLQJUDGLRWHOHPHWU\IRUD\HDUSHULRGLQVRXWKZHVW

OLPLWHGRXUVSDWLDODFFXUDF\WRVHYHUDOKXQGUHGOLQHDOPHWHUVVRZHZHUHPHDVXULQJO\Q[ VHOHFWLRQIRUIRUHVWVWDQGVDQGQRWGLIIHUHQWSDWFKHVRIIRUHVW:HFRXOGQRWGHWHFW PRYHPHQWURXWHVQRUEHKDYLRXUDOFKDQJHVEHORZWKHVWDQGOHYHO%HFDXVHWKHVHPHWKRGV UHTXLUHLQWHQVHILHOGHIIRUWRXUVWXG\DUHDZDVUHODWLYHO\VPDOOFRPSDUHGWRO\Q[KRPH UDQJHVL]HLWZDVWKHVL]HRIDERXWPDOHKRPHUDQJHVZKHQO\Q[ZHUHDEXQGDQWDQG RQO\KRPHUDQJHVZKHQO\Q[ZHUHDWF\FOLFORZV+DUHDEXQGDQFHZDVLQGH[HGXVLQJ SHOOHWFRXQWVZKLFKKDGEHHQWHVWHGDJDLQVWNQRZQGHQVLWLHVLQQHDUE\.OXDQH

Case study 2: Short-tailed weasel and American marten

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

Case study 3: Grizzly bears

:HWHVWHGDQGUHILQHGPHWKRGVIRUHVWLPDWLQJSRSXODWLRQVL]HRIJUL]]O\EHDUVXVLQJ JHQHWLFVDPSOLQJLQPDQ\VWXG\DUHDVDFURVVZHVWHUQ&DQDGD*HQHUDOPHWKRGVLQFOXGHG V\VWHPDWLFVDPSOLQJRIEHDUVXVLQJYDULRXVKDLUUHPRYDOPHWKRGVDFURVVDQXPEHURI GLVFUHWHVDPSOLQJVHVVLRQVWRDGKHUHWRUHTXLUHPHQWVIRUFORVHGPDUNUHFDSWXUHDQDO\VHV +DLUVDPSOHVZHUHWKHQDVVLJQHGDQLQGLYLGXDOLGHQWLW\EDVHGRQPLFURVDWHOOLWH JHQRW\SLQJ0DUNUHFDSWXUHDQDO\VLVZDVXVHGWRHVWLPDWHSRSXODWLRQVL]H0RYHPHQW UDWHVZHUHDOVRPHDVXUHGEXWGHWHFWLRQVRIEHDUVDWPXOWLSOHVLWHVZHUHXVXDOO\IHZ 6WXG\DUHDVIRUSDSHUVLQWKLVWKHVLVUDQJHIURPVRXWKZHVW$OEHUWDWRFHQWUDO%ULWLVK &ROXPELD %& LQZHVWHUQ&DQDGD%HWZHHQDQGZHDSSOLHGWKHVHPHWKRGVLQ DQXPEHURIRWKHUVWXG\DUHDVLQQRUWKHUQ%& 3URFWRUHWDO 7KHVHGDWDDUHQRW SUHVHQWHGLQGHWDLOEXWDUHXVHGLQ3DSHU;IRUH[WUDSRODWLRQRISRSXODWLRQGHQVLW\ 7HFKQLFDODGYDQFHVLQFOXGHGWKHWHVWLQJRIYDULRXVPHWKRGVRIDWWUDFWLQJDQGUHPRYLQJ

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

7KHDSSOLFDWLRQRIJHQHWLFEDVHGVXUYH\PHWKRGVPRVWO\GULYHQE\FRQVHUYDWLRQJRDOV RIWHQLQFOXGLQJWKHQHHGIRUHVWLPDWHVRISRSXODWLRQVL]HJHQHUDWHGVSDWLDOGDWDDERXW JUL]]O\EHDUVGLVWULEXWLRQDFURVVODUJHDUHDV)RURQHVWXG\DUHDLQVRXWKHUQ%&ZHXVHGD *,6WRVXPPDUL]HKDELWDWYDULDEOHVDFURVVGLIIHUHQWUDGLXVZLQGRZVDURXQGWKHKDLU VDPSOLQJVWDWLRQV:HWKHQXVHGUHJUHVVLRQPHWKRGVWRUHODWHWKHSUHVHQFHRIEHDUVWR WKHVHKDELWDWYDULDEOHVLQDPXOWLYDULDWHIUDPHZRUNLQRUGHUWRH[DPLQHKDELWDWVHOHFWLRQ DQGWKHLQIOXHQFHRIVFDOHRQWKHVHUHVXOWV

:HZHUHGULYHQE\WKHQHHGVIRUPDQDJLQJWKHLPSDFWRIKXQWLQJRQJUL]]O\EHDUVWR GHULYHSRSXODWLRQHVWLPDWHVIRUDOODUHDVRIWKHSURYLQFHQRWMXVWWKRVHZKLFKZHKDG VXUYH\HG*LYHQWKHJUHDWFRVWRIWKHDERYHVXUYH\VRQO\DVPDOOSDUWRIWKHSURYLQFHKDV EHHQVXUYH\HGWRGDWH6RPHIRUPRIH[WUDSRODWLRQZDVQHFHVVDU\EHFDXVHWKHKXQWLQJ UHJLPHLQ%&LVEDVHGRQDTXRWDV\VWHPDQGFDOFXODWLQJTXRWDVUHTXLUHVDSUHGLFWLRQRI SRSXODWLRQVL]HIRUHDFKDUHDWKDWLVKXQWHG7KHDPRXQWRIVDOPRQ Oncorhynchus spp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

:HVXPPDUL]HGDOOHVWLPDWHVRIJUL]]O\EHDUGHQVLW\DYDLODEOHDQGXVHGD*,6WRH[WUDFW YDULDEOHVWKDWZHIHOWZHUHIXQFWLRQDOO\UHODWHGWRGHQVLW\:HWKHQXVHGPXOWLSOHOLQHDU UHJUHVVLRQWRUHODWHEHDUGHQVLW\WRDVHULHVRISUHGLFWRUYDULDEOHVLQFOXGLQJWKHGLHW YDULDEOHVZHPDSSHGDERYH:HZHLJKWHGHDFKPHDVXUHRIGHQVLW\EDVHGRQWKH PHDVXUHGRUDVVXPHGSUHFLVLRQDQGXVHG$,&WRKHOSVHOHFWDEHVWILWPRGHO)LWZDVPRUH LPSRUWDQWWKDQIXQFWLRQEHFDXVHWKHJRDOZDVWRSUHGLFWGHQVLW\PRVWSUHFLVHO\DQGQRW QHFHVVDULO\WRXQGHUVWDQGWKHUHODWLRQVKLSEHWZHHQGHQVLW\DQGWKHHQYLURQPHQW 3UHGLFWLYHDFFXUDF\ZDVDOVRLPSRUWDQWDQGZHH[FOXGHGRYHUO\FRPSOH[PRGHOVRQWKH ORJLFWKDWWKH\PD\JHQHUDWHVSXULRXVSUHGLFWLRQVLQSDUWVRIWKHPXOWLYDULDWHVSDFH $UQROG :HXVHGWKHILQDOPRGHOWRSUHGLFWJUL]]O\GHQVLW\DQGSRSXODWLRQVL]HIRU PDQ\DUHDVRI1RUWK$PHULFDDQGSUHVHQWHGWKHVHDVDOWHUQDWLYHSRSXODWLRQSUHGLFWLRQV IRUVRPHRIWKHVHDUHDV)LQDOO\ZHXVHGWKHSRSXODWLRQSUHGLFWLRQVIURPRXUPRGHOWRVHW PRUWDOLW\OLPLWVIRUJUL]]O\EHDUVWKURXJKRXWWKH3URYLQFHRI%&

Summary of results

Case study 1: Canada lynx

Lynx population dynamics (Paper I) /\Q[LQERUHDO1RUWK$PHULFDJRWKURXJKF\FOHVLQDEXQGDQFHWKDWRFFXURYHUDERXWWHQ \HDUV .HLWK ,QRXUVWXG\DUHDLQVRXWKZHVW

Lynx habitat use through a snowshoe hare population cycle (Paper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

Case study 2: Short-tailed weasel and American marten

Habitat associations of marten and short-tailed weasels (Paper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

Habitat associations of weasels (Paper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

Habitat associations of marten (Paper V) 0DUWHQFKRLFHRIKDELWDWVZDVDOVRGLIIHUHQWDFURVVVFDOHVVHOHFWLRQIRUVWDQGVWUXFWXUH ZDVJUHDWHUDWWKHILQHUVFDOHRIDQDO\VLVZKHUHDVVHOHFWLRQIRUVWDQGDQGHFRV\VWHPW\SHV ZDVJUHDWHUDWWKHEURDGHUVFDOH0DUWHQGHWHFWLRQZDVJUHDWHULQZHWWHUVWDQGVDQG HFRV\VWHPV,QFRQWUDVWWRZHDVHOVPDUWHQSUHIHUUHGKLJKHUFURZQFORVXUHDQGROGHU VWDQGV$QGOLNHZHDVHOVWKHIXQFWLRQDOYDULDEOHLQIOXHQFLQJEHKDYLRXUDSSHDUHGWREH FURZQFORVXUHDQGQRWVWDQGDJHEHFDXVHWKHWZRYDULDEOHVZHUHFRUUHODWHGDQGVHOHFWLRQ ZDVVWURQJHUDQGPRUHFRQVLVWHQWIRUFURZQFORVXUH6HOHFWLRQIRUVWDQGDJHZDVJUHDWHU LQWKHGULHUVWXG\DUHDZKHUHYDULDWLRQLQFURZQFORVXUHDFURVVVWDQGDJHZDVJUHDWHU 0DUWHQDUHRIWHQFRQVLGHUHGROGIRUHVWGHSHQGHQWEXWZHIRXQGPDUWHQLQDOODJHFODVVHV RIIRUHVWZLWKVRPHFURZQFORVXUH0DUWHQDEXQGDQFHDSSHDUVWREHIXQFWLRQDOO\OLQNHGWR HFRV\VWHPSURGXFWLYLW\EHFDXVHGHWHFWLRQZDVKLJKHULQZHWWHUHFRV\VWHPVHYHQZKHQ RWKHUIDFWRUVZHUHFRQWUROOHGIRU

Marten population size (Paper VI) :HSUHVHQWHGDVHWRIPHWKRGVWKDWFRXOGEHXVHGWRPHDVXUHSRSXODWLRQVL]HIRUVPDOO FDUQLYRUHVDOWKRXJKWKH\ZLOOQHHGWREHDGDSWHGIRUHDFKVSHFLHVDQGSODFH2XU PHWKRGRORJ\FDQEHFRQVLGHUHGDVWDUWLQJSODFHWRFRQVLGHUWKLVSUREOHP.H\ FRQVLGHUDWLRQVZLOOEHPD[LPL]LQJWKHGHWHFWLRQRIDQLPDOVWKDWDSSURDFKWKHWUDSVDQG WUDSUHVSRQVH

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

3RSXODWLRQGHQVLW\ZDVPDUWHQVNPRQWKH6HONLUN0RXQWDLQVWXG\DUHDDQG SRSXODWLRQGHQVLW\ZDVQRWHVWLPDWHGIRUWKH3XUFHOO0RXQWDLQVWXG\DUHDEXWGLIIHUHQW LQGLYLGXDOVZHUHLGHQWLILHGIURPUDQGRPO\FKRVHQPDUWHQVDPSOHVRIWRWDO VDPSOHV7KLVUHVXOWVXJJHVWVUHFDSWXUHVZHUHIHZLQWKHHQWLUHGDWDVHWDQGWKDW SRSXODWLRQVL]HZDVDWOHDVWGRXEOHWKHQXPEHURIVDPSOHVFROOHFWHG'HQVLW\LQWKH 3XUFHOO0RXQWDLQVWXG\DUHDZDVOLNHO\URXJKO\VLPLODUWRWKH6HONLUN0RXQWDLQV

Case study 3: Grizzly bears

Grizzly and black bear densities (Paper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

Habitat selection of grizzly bears (Paper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

Major components of grizzly bear diet (Paper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

DUHDVZKHUHFDULERXZHUHDEXQGDQWJDUQHUHGPRUHWKDQKDOIWKHLUQXWULHQWVIURP WHUUHVWULDOPHDWSUHVXPDEO\FDULERX$UHDVZLWKKLJKXQJXODWHGHQVLWLHVDUHGULHUDQG RIIHUSRRUSODQWIRRGVWREHDUV7KHDUFWLFLVSDUWLFXODUO\SRRULQSODQWIRRGV2XUUHVXOWV ZHUHFRQVLVWHQWZLWKSUHYLRXVVXJJHVWLRQVWKDWEHDUVDUHRSSRUWXQLVWLFIHHGHUVDQGWKDW WKH\IRFXVWKHLUIRUDJLQJRQVDOPRQZKHUHLWLVDYDLODEOH 6FKZDUW]HWDO

:HVKRZHGWKDWIHPDOHEHDUVUHDFKHGWKHLUPDWXUHERG\VL]HDWDERXWWKHDJHWKH\ILUVW EHFDPHSUHJQDQW \UV DQGPDOHVDW\UV0DOHVFRQWLQXHGWRJURZVORZO\WKURXJKRXW PRVWRIWKHLUDGXOWOLIHZKLOHIHPDOHERG\VL]HLQFUHDVHGOLWWOH6NXOOOHQJWKDQGZLGWK LQFUHDVHGZLWKPHDWLQWKHGLHWDQGWKHUHODWLRQVKLSEHWZHHQVDOPRQDQGVL]HZDVPXFK VWURQJHUWKDQWKDWZLWKWHUUHVWULDOPHDW

Predicting grizzly bear density in North America (Paper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

7KRXJKQRWIXQFWLRQDOLQQDWXUHRXUPRGHOVVXJJHVWWKDWJUL]]O\EHDUGHQVLW\LVUHODWHGWR SODQWSURGXFWLYLW\2XUUHVXOWVDOVRVXSSRUWWKHK\SRWKHVLVWKDWRSHQYHJHWDWLRQW\SHVDQG \RXQJVHUDOIRUHVWVSURYLGHJUHDWHUIRUDJHIRUEHDUVDQGUHVXOWVLQKLJKHUGHQVLW\ $OWKRXJKRXUUHVXOWVGLGQRWH[FOXGHWKHLQIOXHQFHRIEODFNEHDUFRPSHWLWLRQRQJUL]]O\ EHDUGHQVLW\WKH\GLGQRWSURYLGHGLUHFWVXSSRUWIRUWKLVK\SRWKHVLVEHFDXVHRIFROLQHDULW\ LQVRPHRIRXULQGHSHQGHQWYDULDEOHV$VH[SHFWHGGHQVLW\LQFUHDVHGZLWKVDOPRQLQWKH

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

Discussion

Case study 1: Canada lynx /\Q[GHFOLQHGE\DQRUGHURIPDJQLWXGHRQRXUVWXG\DUHD 3DSHU, DQGVLPLODUF\FOHVLQ GHQVLW\KDYHEHHQQRWHGLQRWKHUSDUWVRIERUHDO1RUWK$PHULFD 0RZDWHWDO HWDO ,Q.OXDQHZKLFKLVVHYHUDOKXQGUHGNPIURPRXUVWXG\DUHD WKHDPSOLWXGHRIWKHVDPHKDUHF\FOHZHVWXGLHGZDVDERXWIROG .UHEVHWDOE &DQDGDO\Q[DUHNQRZQWRUHO\KHDYLO\RQVQRZVKRHKDUHVDVSUH\2WKHUVSHFLHVVXFKDV UHGVT 5RWKHWDO EXWO\Q[DUHUHOLDQWRQKDUHVHYHQGXULQJSHULRGVDQGSODFHVRIPRGHVW

VXUSULVLQJO\O\Q[VHOHFWIRUKDELWDWVZLWKKLJKKDUHDEXQGDQFHDVZHDQGRWKHUVIRXQGLQ

DQ\VLJQLILFDQWFKDQJHLQVHOHFWLRQDPRQJIRUHVWVWDQGVLQRXUVWXG\DUHD 3DSHU,, VLPLODUWRUHVXOWVIURPWKHVRXWKZHVW1RUWKZHVW7HUULWRULHV 3RROHHWDO XVLQJ VLPLODUPHWKRGVDQGVFDOHVRIDQDO\VLV7KH.OXDQHVWXG\VKRZHGWKDWERWKKDUHVDQG O\Q[FKRRVHGHQVHUIRUHVWGXULQJSHULRGVRIORZDEXQGDQFHXVLQJDILQHUVFDOH

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

,QFRQFOXVLRQRXUVWXG\DQGWZRRWKHUVQHDUE\WKDWXVHGVLPLODUPHWKRGVDQGVFDOHVRI H[DPLQDWLRQVXJJHVWHGWKDWWKHH[DPLQDWLRQRIIRUHVWVWDQGVHOHFWLRQE\O\Q[DFURVVWKH DUHDRIDIHZLQGLYLGXDOKRPHUDQJHVGHVFULEHGEHKDYLRXUDOGHFLVLRQVPDGHE\LQGLYLGXDO O\Q[ZLWKUHVSHFWWRSUH\FKRLFH'HWDLOHGUHFRUGLQJRIPRYHPHQWVPD\DOVRGHVFULEH EHKDYLRXUDOGHFLVLRQVDURXQGKXQWLQJVXFFHVVDVZDVGRFXPHQWHGDWWKHQHDUE\.OXDQH

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

Case study 2: Short-tailed weasel and marten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

Case study 3: Grizzly bears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

,QFRQFOXVLRQWKHKDELWDWSUHIHUHQFHVRIJUL]]O\EHDUVDSSHDUPRUHYDULDEOHWKDQO\Q[RU WKHVPDOOPXVWHOLGVZHVWXGLHGEXWWKHUHGRQRWDSSHDUWREHFOHDUOLQNVEHWZHHQWKH W\SHVRIKDELWDWFKRVHQE\EHDUVDQGSRSXODWLRQG\QDPLFVRUGHQVLW\,QIDFWVRPHEHDUV

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

Perspectives or Management implications

6RPHGHILQLWLRQVRIKDELWDWVHOHFWLRQVXJJHVWWKDWVHOHFWLRQLVWKHEHKDYLRXUDOSURFHVVRU FKRLFHPDGHE\LQGLYLGXDOV .UHEV%H\HUHWDO DQGWKXVH[FOXGHVWKH LQIOXHQFHRISRSXODWLRQGHQVLW\RQWKHRXWFRPH7KHIDFWRUVWKDWOLPLWRUUHJXODWHGHQVLW\ DUHGHHSHUWKDQWKHEHKDYLRXUDOGHFLVLRQVLQGLYLGXDOVPDNHDERXWZKHUHWRIRUDJHORFDWH WKHLUQHVWFKRRVHWRUHVWHWF7KXVXQGHUWKLVGHILQLWLRQWKHLQYHVWLJDWLRQRIKDELWDW VHOHFWLRQLVOLPLWHGWRWKHVFDOHDWZKLFKEHKDYLRXUDOSURFHVVHVRIDJURXSRILQGLYLGXDO DQLPDOVOLPLWWKHLUGLVWULEXWLRQDFURVVWKHODQGVFDSH

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

&RQWLQHQW 5HJLRQ /DQGVFDSH 6WDQG 6LWH

'LVWULEXWLRQ $EXQGDQFH 5HJLRQDOOLIHKLVWRU\,QGLYLGXDOEHKDYLRXU RUUDQJH GHPLFEHKDYLRXU GLHW SUHGDWRUDYRLGDQFH LHVHDVRQDOGLHWVKLIWV )LJ+DELWDWVHOHFWLRQVWXGLHVPHDVXUHYDULRXVSRSXODWLRQDQGEHKDYLRXUDOSURFHVVHV GHSHQGLQJRQWKHVFDOHRIWKHH[DPLQDWLRQ$WWKHEURDGHVWVFDOHWKH\PHDVXUHWKHIRUFHV OLPLWLQJWKHUDQJHRIWKHVSHFLHVZKHUHDVDWWKHILQHVWVFDOHWKH\PHDVXUHEHKDYLRXUDO GHFLVLRQVRILQGLYLGXDODQLPDOV

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

'LIIHUHQWHFRORJLFDOTXHVWLRQVUHTXLUHGLIIHUHQWVWXG\GHVLJQVWRLPSURYHRXWFRPHV7R PHDVXUHWKHLQIOXHQFHRIEHKDYLRXURQVHOHFWLRQP\VXJJHVWLRQVDUH $QDO\]HGDWDIURPHDFKLQGLYLGXDOVHSDUDWHO\LQWKHDQDO\VLV

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

Acknowledgements

,ZRXOGOLNHWRWKDQNWKHFRDXWKRUVRQDOOWKHSDSHUVDQGWKHYHU\PDQ\SHRSOHZKR KHOSHGXVZLWKWKHVHSURMHFWVRYHUWKH\HDUV6SHFLILFDOO\,ZRXOGOLNHWRWKDQN%ULDQ 6ORXJK5HQH5LYDUGDQG6WDQ%RXWLQIRUDOOWKHLUKHOSDQGVXSSRUWRQWKH6QDIXO\Q[ SURMHFW$ELJWKDQNVWR'DUF\)HDU'DYH6WDQOH\DQG&RUE\6KXUJRWIRUWKHLUZRUNRQ

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

References

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p FRPSDUDWLYHIHHGLQJ HFRORJ\LQDQLVODQGDQGPDLQODQGSRSXODWLRQRI6SDLQ=HLWVFKULIWIU6lXJHWLHUNXQGH &XPEHUODQG5('HPSVH\-$)RUEHV*- 6KRXOGGLHWEHEDVHGRQELRPDVV" ,PSRUWDQFHRIODUJHUSUH\WRWKH$PHULFDQPDUWHQ:LOGOLIH6RFLHW\%XOOHWLQ (IIRUG0 'HQVLW\HVWLPDWLRQLQOLYHWUDSSLQJVWXGLHV2LNRV *DLOODUG-0+HEEOHZKLWH0/RLVRQ$HWDO +DELWDW SHUIRUPDQFHUHODWLRQVKLSV ILQGLQJWKHULJKWPHWULFDWDJLYHQVSDWLDOVFDOH3KLORVRSKLFDO7UDQVDFWLRQVRIWKH 5R\DO6RFLHW\%%LRORJLFDO6FLHQFHV *DUVKHOLV'/ 'HOXVLRQVLQKDELWDWHYDOXDWLRQPHDVXULQJXVHVHOHFWLRQDQG LPSRUWDQFH,Q)XOOHU7.%RLWDQL/ HGV 5HVHDUFKWHFKQLTXHVLQDQLPDOHFRORJ\

FRQWURYHUVLHVDQGFRQVHTXHQFHV&ROXPELD8QLYHUVLW\3UHVV1HZ

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

$ODVND,Q5XJJLHUR/)$XEU\.%%XVNLUN6:HWDO HGV 7KHVFLHQWLILFEDVLVIRU O\Q[FRQVHUYDWLRQ7KH8QLYHUVLW\3UHVVRI&RORUDGR5056*75:::SS 0RZDW*6ORXJK%* 6RPHREVHUYDWLRQVRQWKHQDWXUDOKLVWRU\DQGEHKDYLRXURI WKH&DQDGDO\Q[/\Q[FDQDGHQVLV&DQDGLDQ)LHOG1DWXUDOLVW 0RZDW*6WUREHFN& (VWLPDWLQJSRSXODWLRQVL]HRIJUL]]O\EHDUVXVLQJKDLU FDSWXUH'1$SURILOLQJDQGPDUNUHFDSWXUHDQDO\VLV7KH-RXUQDORI:LOGOLIH 0DQDJHPHQW 0XUUD\'/5RWK-'(OOVZRUWK(HWDO (VWLPDWLQJORZGHQVLW\VQRZVKRHKDUH SRSXODWLRQVXVLQJIHFDOSHOOHWFRXQWV&DQDGLDQ-RXUQDORI=RRORJ\ 0\VWHUXG$,PV5$ 5HODWLQJSRSXODWLRQVWRKDELWDWV7UHQGVLQ(FRORJ\DQG (YROXWLRQ 1DJRUVHQ':0RUULVRQ.))RUVEHUJ-( :LQWHUGLHWRI9DQFRXYHU,VODQGPDUWHQ 0DUWHVDPHULFDQD &DQDGLDQ-RXUQDORI=RRORJ\ 1LHOVHQ6(%R\FH066WHQKRXVH*% *UL]]O\EHDUVDQGIRUHVWU\,6HOHFWLRQRI FOHDUFXWVE\JUL]]O\EHDUVLQZHVWFHQWUDO$OEHUWD&DQDGD)RUHVW(FRORJ\DQG 0DQDJHPHQW 1LHOVHQ6(%R\FH066WHQKRXVH*% 'HYHORSPHQWDQGWHVWLQJRI SKHQRORJLFDOO\GULYHQJUL]]O\EHDUKDELWDWPRGHOV(FRVFLHQFH

%RRQVWUD5 HGV (FRV\VWHPG\QDPLFVRIWKHERUHDOIRUHVWWKH.OXDQH3URMHFW 2[IRUG8QLYHUVLW\3UHVV1HZ

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

3URFWRU0)0F/HOODQ%1%RXODQJHU-HWDO (FRORJLFDOLQYHVWLJDWLRQVRIJUL]]O\ EHDUVLQ&DQDGDXVLQJ'1$IURPKDLUDUHYLHZRIPHWKRGVDQGSURJUHVV 8UVXV 5LFNOHIV5 (FRORJ\6HFRQG(GLWLRQ&KLURQ3UHVV1HZ

DFDGHP\RIVFLHQFHV%LDORZLH]D3RODQGSS

Individual papers

PAPER I

PAPER II

1736

Habitat preference of Canada lynx through a cycle in snowshoe hare abundance

Garth Mowat and Brian Slough

Abstract: We assessed habitat preference of a lynx (Lynx canadensis) population through 8 years of a snowshoe hare (Lepus americanus) cycle. Seventy-four percent of our southern Yukon study area was approximately 30-year-old regenerating forest resulting from a large wildfire. The study area was not trapped and lynx density was very high compared with other populations in North America. Contrary to our prediction, there was no discernable shift in habitat preference through the hare cycle; however, our habitat types were coarsely mapped and our radiolocations relatively inaccurate. Lynx may have altered their habitat preferences at finer scales (for patches <2 ha). Lynx showed strong preference for regenerating habitats over mature white spruce (Picea glauca) and alpine–subalpine. Lodgepole pine (Pinus contorta) dominated regenerating stands were preferred over spruce–willow (Salix spp.) stands of equal age. Ri- parian willow stands were also preferred over mature spruce forest and alpine. Lynx used riparian willow stands more in winter, but we detected no other shifts in habitat preference between snow-free and winter periods. We did not detect any difference in habitat preference between sexes. Independent juveniles made greater use of mature forest and perhaps riparian willow than adults, but no other difference in preference between the two age groups was noted. Lynx preference for regenerating habitat over mature forest suggests that burns will benefit lynx, especially if the regenerating community is pine dominated. Logging will only likely provide similar benefits if a dense pine understory results, which is unlikely in intensively managed stands. The suppression of forest fires in recent decades may have contributed to the decline of lynx numbers in the south of their range.

Résumé : Nous avons déterminé le choix d’habitat chez une population de lynx (Lynx canadensis) au cours des 8 an- nées d’un cycle d’abondance du lièvre d’Amérique (Lepus americanus). Soixante-quatorze pour cent de la région d’étude dans le sud du Yukon est couverte d’une forêt en régénération à la suite d’un important feu de forêt ilyaune trentaine d’années. Il n’y a pas eu de trappage dans la région et la densité des lynx y est particulièrement élevée, par comparaison aux autres populations de l’Amérique du Nord. Contrairement à nos prédictions, il ne s’est pas produit de changement apparent dans les choix d’habitat au cours du cycle d’abondance des lièvres; cependant, notre cartographie des types d’habitats était grossière et nos déterminations par radio des emplacements des animaux relativement impréci- ses. Il se peut que les lynx aient modifié leur choix d’habitat à des échelles plus fines (parcelles de <2 ha). Les lynx préfèrent de beaucoup les habitats en régénération aux pessières matures (Picea glauca) et aux zones subalpines et al- pines. Ils choisissent les parcelles en régénération dominées par le pin vrillé (Pinus contorta) de préférence aux parcelles de même âge dominées par l’épinette et les saules (Salix spp.). Ils préfèrent aussi les parcelles riveraines de saules à la pessière mature et à la zone alpine. Les lynx utilisent plus les zones riveraines de saules en hiver et c’est le seul changement d’habitat observé entre la période hivernale et la période sans neige. Il n’y a aucune différence de choix d’habitat entre les sexes. Les jeunes indépendants utilisent davantage les forêts matures et peut-être aussi les saulaies que les adultes et c’est la seule différence observée entre les choix d’habitat chez ces deux groupes d’âge. La préfé- rence des lynx pour les habitats en régénération plutôt que pour la forêt mature laisse croire que les feux de forêt sont bénéfiques aux lynx, particulièrement lorsque la communauté en régénération est dominée par les pins. La coupe fores- tière ne fournit les mêmes avantages que si un sous-bois dense de pins se développe, ce qui est improbable dans les forêts fortement aménagées. La suppression des feux de forêt au cours des dernières décennies peut avoir contribué au déclin des densités de lynx dans la partie australe de leur répartition géographique. [Traduit par la Rédaction] Mowat and Slough 1745

Received 19 March 2003. Accepted 11 September 2003. Published on the NRC Research Press Web site at http://cjz.nrc.ca on 25 November 2003. G. Mowat1,2 and B. Slough.3 Fish and Wildlife Branch, Yukon Department of Renewable Resources, P.O. Box 2703, Whitehorse, YT Y1A 2C6, Canada. 1Corresponding author (e-mail: [email protected]). 2Present address: Aurora Wildlife Research, R.R. 1, Site 14, Comp 8, Crescent Valley, BC V0G 1H0, Canada. 3Present address: 35 Cronkhite Road, Whitehorse, YT Y1A 5S9, Canada.

Introduction ever, O’Donoghue et al. (2001), in southwest Yukon, found that lynx used stands with greater overstory during the cycli- During the last century, the southern distribution of Can- cal decline in hare numbers, as did hares (Murray et al. ada lynx (Lynx canadensis) has contracted, precipitating 1994; O’Donoghue et al. 1998). During the subsequent low considerable conservation concern (Ruggiero et al. 2000; phase, lynx moved to areas of lower cover that may have Poole 2001). The causes of this decline are uncertain. One partly been driven by a switch to hunting red squirrels hypothesis suggests that the reduction in large forest fires, (Tamiascius hudsonicus), which were a much greater part of due to fire suppression, has reduced the amount of available the lynx’s diet during the low phase (O’Donoghue et al. habitat (Koehler 1990). Alternative hypotheses suggest com- 2001). Lynx also spent more time foraging in experimental petition with other terrestrial predators, roads and human areas where hares were fed supplemental food and hare disturbance, reduced habitat quality, increased habitat frag- numbers were higher than in the surrounding landscape mentation, and reduction in connectivity with northern popu- (Ward and Krebs 1985; O’Donoghue et al. 1998). lations as causes for the contraction of lynx distribution The objectives of this study were as follows. (i)Weexam- (Ruggiero et al. 2000). Trapping may also have contributed ine lynx habitat use across the low, increasing, peak, and de- to the decline in some cases (Poole 2001). We examine habi- cline phases of the snowshoe hare cycle. We predict that tat preference of lynx in a recently burned landscape across lynx will utilize more open habitats when hares are abundant large differences in cyclic lynx density to help resolve the because hares colonize more open habitats when numbers importance of fire to lynx. are greater and that lynx will focus their use on those habi- Lynx populations in the north go through periods of in- tats with greatest hare abundance during the decline and low crease and decline as they follow the numbers of their main phases. (ii) We compare habitat selection between winter prey, the snowshoe hare (Lepus americanus). This cycle in and the snow-free season. We expect lynx to use open habi- abundance is of lower amplitude, but still apparent, in the tats more in summer, including habitats that have greater de- southern lynx range (McKelvey et al. 2000). Recent genetic ciduous or herbaceous cover. (iii) We investigate habitat work on lynx throughout their North American range shows selection among sexes and broad age classes (yearlings and low structuring of populations (Schwartz et al. 2002; Rueness adults). Body size differences among sexes and age classes, et al. 2003). This suggests regular and significant movement and the fact females are accompanied by their young much of breeding-age animals, which has been demonstrated by of the year, may influence habitat preference. (iv)Weexam- field studies (Mowat et al. 2000). Indeed, influxes of lynx ine the influence of stand structure and dominant plant spe- into populations at the southern edge of their range may cies on lynx habitat preference. Specifically, we expect have been commonplace historically (McKelvey et al. 2000). understory closure to influence habitat preference more than The connectivity and genetic similarity of North American dominant species or overstory closure. lynx populations suggest that lynx behaviour in the north of their range, especially during periods of hare scarcity, may Methods be relevant to lynx population maintenance further south. Lynx occur in boreal and montane forests throughout Study area North America (McCord and Cardoza 1982) and their distri- Field studies were conducted between December 1986 bution is contiguous with that of the snowshoe hare (Mowat and June 1994 on a 304-km2 area (Slough and Mowat 1996) et al. 2000). At finer scales, lynx select habitats where hares located approximately 100 km southeast of Whitehorse, are most abundant (Aubry et al. 2000; Mowat et al. 2000; Yukon, Canada (60°15′N, 135°20′W). The area is on the O’Donoghue et al. 2001). These are generally dense early Teslin plateau (Bostock 1965) of the Lake Laberge eco- seral coniferous stands, although mature stands with dense region (Oswald and Brown 1986) and is characterized by understory are also used (Koehler 1990; Mowat et al. 2000). rolling mountains dissected by meandering creek valleys. El- Mature forests may also be used for denning and as move- evations range from 800 to 1950 m above sea level. A large ment corridors (Saunders 1963; Parker 1981; Koehler 1990). forest fire burned more than 70% of the study area in 1958; Deciduous-dominated stands with well-developed under- regenerating stands are the matrix habitat in our study area stories can also support moderate numbers of hares (Koehler and the surrounding landscape. Regenerating shrubs and 1990; Hodges 2000b) and lynx. Hares appear to select for trees are predominantly lodgepole pine (Pinus contorta), white denser understory cover, while browse abundance, overstory spruce (Picea glauca), trembling aspen (Populus tremulo- closure, and stand age are less important (Hodges 2000a, ides), subalpine fir (Abies lasiocarpa), and willows (Salix 2000b); this is likely also true for lynx. Hares use deciduous spp.). Residual patches of mature timber cover 9% of the cover and open stands more in summer (Hodges 2000a, area, permanent water 4%, and alpine tundra 10% (Table 1). 2000b). Lynx, too, have broader diets and use a wider range of habitats in summer (Mowat et al. 2000; O’Donoghue et Habitats al. 2001), and hence, we may expect greater use of decidu- We interpreted vegetation classes from 1986 1 : 40 000 ous stands and areas with less understory in summer. scale aerial photographs and checked these using an en- Hodges (2000a, 2000b) reported that there was no clear hanced LandSat TM image and during a low-level flight on shift in hare habitat use through the hare cycle, although 28 September 1994. Existing plant community descriptions more open areas may receive greater use as hare density in- (Davies et al. 1983; Oswald and Brown 1986) and personal creases (Fuller and Heisey 1986). Poole et al. (1996) con- knowledge of the area assisted with the classification. We cluded habitat use of lynx did not differ between the sexes or mapped 19 vegetation classes based on dominant species, among years in the southwest Northwest Territories. How- overstory height (for pine only), and stand age. Immature

Table 1. Structural measures for seven habitats in our study area in southwest Yukon. Proportion of Understory Understory Stem Canopy Understory Overstory Habitat area (%) densitya densityb densitya closureb rankc rankc Alpined 10.8 Variable Average is low Near zero 1 1 Mature pine 0.2 21 13 600 36 2 4 Mature sprucee 9.6 66 65 3 700 32 3 5 Riparian willow 5.2 97 91 131 000 18 3 2 Mixed immature pine 23.3 46 29 4 2 Mixed immature spruce 38.7 60 7 4 3 Dense immature pine 12.2 89 41 100 5 1 aFrom Major (1989). bFrom G. Mowat, J. Staniforth, and V. Loewen (unpublished data). cThe subjective rank of average density for the mapped habitat classes (1 is lowest, 5 is highest). dNo measurements were done in this habitat; comments are based on personal observation. eThese measures are from samples in small unburned exclusions and are likely higher than in larger continuous stands. stands were initiated 30–35 years before our study, while J. Staniforth, and V. Loewen (Yukon Territorial Government, mature stands were not burned in 1958 and most were much unpublished data) sampled vegetation structure to examine older than 35 years. We plotted vegetation polygons on 1 : hare habitat use in a number of the above habitats using 50 000 scale topographic maps and digitized and imported identical methods. Most habitats had high understory density the polygons into a GIS software (ARC/INFO) (ESRI, except mature pine and alpine (Table 1), although the under- Redlands, Calif.). We combined the 19 vegetation classes story density for mature spruce was likely much lower in our into seven habitats that we deemed were most relevant to mapped polygons because the data in Table 1 were all col- lynx and hares (Table 1). The resulting 165 habitat polygons lected in small unburned spruce stands within the burned averaged 169 ha in size (SD = 301, range 2–2770 ha). matrix. Because only five sites were sampled in each habi- The seven relevant habitats are based on stand age and tat, and results were not necessarily indicative for our seven dominant species in the overstory (Table 1): (1) alpine and habitats, we ranked understory and overstory closure for subalpine habitats are found above 1220–1370 m above sea each mapped habitat based on field data, sample site loca- level and have sparse willow and shrub birch (Betula spp.) tion, and personal knowledge of the study area (Table 1). cover and few or no trees, (2) mature pine habitats are open mature pine stands, occasionally with an aspen component Snowshoe hare abundance in the overstory, but little spruce, and understory is limited, We counted snowshoe hare fecal pellets each June on per- (3) mature spruce stands are spruce dominated but also con- manent 0.155-m2 quadrats (5.08 cm × 305 cm in size) placed tain aspen, pine, and subalpine fur, and understory is often along 11 transects. Twenty to 50 quadrats were spaced 9.14 m willow and is more abundant than in mature pine stands, apart along each transect; 360 quadrants were sampled to (4) riparian willow habitats are climax stands of willow and June 1989 and 480 thereafter. Each quadrat was visually as- open sedge (Carex spp.) meadows associated with riparian signed one of the above seven habitats. Yearly hare densities zones, and overstory is nearly absent and other woody plant were presented in Slough and Mowat (1996) based on meth- species are rare, (5) mixed immature pine stands also con- ods described in Krebs et al. (1987) using the above data. In tain small patches of immature spruce, willow, aspen, and this paper, we calculate an index of hare abundance (mean veteran mature trees that were not killed by the 1958 fire, pellets per quadrat) among habitats and do not attempt to (6) mixed immature spruce also supports a mixture of wil- calculate hare density. Krebs et al. (1987 and 2001) have low and mature spruce trees, and (7) dense immature pine shown a linear relationship between pellet numbers and hare stands are monoculture pine with few veteran trees or com- density (r = 0.76) in a nearby boreal environment (but see peting shrubs. Murray et al. 2002). Because of the resolution of the aerial photographs and These data give only a rough measure of hare abundance Landsat mapping, the smallest polygons that we could map among our seven habitats because hare abundance was mea- were >2 ha and approximately 100 m across at the narrowest sured on 10 m long quadrants, a much finer scale than the point. The majority of the small or thin polygons were ripar- scale on which habitats were mapped. No pellet transects ian willow. Mixed immature habitats contained small mature were done in mapped mature forest polygons. Quadrats stands and willow that were too small to map at the 1 : classed as mature spruce or pine were small unburned inclu- 50 000 scale. Dense immature pine polygons were less sions, often only tens of metres across. Because of their mixed than the mixed immature pine and spruce classes. close proximity to regenerating habitat, these inclusions Few mature pine stands remained following the 1958 fire probably received greater use by hares than larger homoge- and mature spruce forest included little immature habitat. nous mature stands. None of the pellet quadrats classed as The riparian willow class supported few other woody plant riparian willow were in large valley bottoms; all were in species, but understory closure was highly variable because small riparian or seepage zones in upslope areas within the small open willow swales, sedge meadows, and small lake or burned matrix. Larger riparian willow areas probably had swamps were included in this class. more variable hare numbers than depicted by these data; We summarized the vegetation structure of our habitats numbers in these small riparian willow stands were probably based on previous work. Major (1989) and G. Mowat, more similar to those in the surrounding matrix.

Fig. 1. Cyclic phases of snowshoe hare (Lepus americanus) and lynx (Lynx canadensis) abundance from 1986 to 1994 in a 304-km2 area of south-central Yukon. Hare abundance was indexed by the mean number of pellets counted on 360–480 quadrats distributed systematically on 11 transects throughout the study area. Lynx density is taken from Slough and Mowat (1996, table 4). Error bars are 95% confidence intervals.

Lynx live capture and telemetry mothers to this point (Mowat et al. 1996) and are hence not Livetrapping for lynx was conducted annually from De- choosing habitat independently. We collected 4808 locations cember to April 1986–1994. Trapping effort was extensive from 103 lynx and used 3904 locations from 100 lynx for and distributed throughout the study area. All animals were population-level habitat analysis. handled in accordance with guidelines set out by the Cana- dian Council on Animal Care (1984). We captured 284 lynx Data analysis (including kittens <1 month of age), and 103 lynx (58 males We measured relative selection among habitats by calcu- and 45 females) were fitted with collar-mounted radiotrans- lating the selection index, w, which is analogous to other ra- mitters. Roughly 60% of all resident lynx were radiocollared tio measures of preference such as the forage ratio or Ivlev’s during the study and many of the uncollared individuals index (Manly et al. 1993). We combined data for all lynx be- were dependent kittens that were monitored via snow- cause we were interested in habitat selection by the lynx tracking (Mowat et al. 1996; Slough and Mowat 1996). Col- population resident on the study area. We combined mature lared lynx were located approximately weekly from a fixed- pine and mature spruce into a mature forest category be- wing aircraft and, more rarely, from the ground. Locations cause mature pine was too rare to analyse for separately. Se- were obtained during daylight hours, which varied from 7 to lection for each category was calculated as the proportion of 20 h per day from winter to summer. The mean lineal loca- all locations in habitat i divided by the proportion of habitat tion error was 282 m for 80 aerial test locations (SD = 208) i available in the study area (Manly et al. 1993, p. 40). Se- and the area of the 95% error polygon was 52 ha around lection is indicated by values from 1 to ∞ and avoidance by each location (Slough and Mowat 1996), several times values from 0 to 1. However, selection indices are best used smaller than the mean habitat polygon size. to infer preference among the habitats included in the analy- We removed capture locations because these were not dis- sis; the size of the index has no meaning except relative to tributed randomly with respect to habitat types. We removed the other habitats in the sample. Binomial confidence inter- locations of denning females between 20 May and 31 July vals were calculated using the program in Krebs (1999). To because during this period, females spent a great deal of reduce the risk of a Type I error when making multiple com- time near the den site (Mowat et al. 1996; Mowat and parisons, α was reduced to 0.008 based on n = 6 habitats us- Slough 1997), as do Eurasian lynx (Lynx lynx) (Schmidt ing the Bonferroni correction procedure (Manly et al. 1993, 1998). Even when denning females are away from their dens, p. 35). We assessed differences in habitat preference based they are probably not choosing habitats independently across on overlap in the confidence intervals of w. their home ranges; they likely select habitats closer to their den sites. We removed locations of adult males from 1 to Results 31 March because they were actively courting at this time and may not select habitat based on food but rather based on Snowshoe hare abundance the location of females (Mowat and Slough 1997; Breiten- The abundance of snowshoe hare pellets changed 16-fold moser et al. 1993). We also removed locations of kittens up from peak to low, while lynx abundance changed 25-fold to 31 March because they are mostly located with their over the same period (Fig. 1). The actual change in abun-

Fig. 2. Cyclic changes in snowshoe hare abundance across five habitats in south-central Yukon, 1986–1994. Habitats are described in the Methods. Hare abundance was indexed by the mean number of pellets counted on 360–480 quadrats distributed systematically on 11 transects throughout the study area. Error bars are 95% confidence intervals.

dance of hares may have been larger because the relation- over immature spruce (see combined data in Fig. 4). Prefer- ship between pellet numbers and abundance may not be ence for riparian willow was similar to that for immature linear at low densities (Murray et al. 2002). Hare and lynx pine habitats, and all of these habitats were preferred over densities were as high as any reported in North America mature forest and alpine. Alpine areas were segregated from (Slough and Mowat 1996). other habitats by elevation and were very open; hence, lynx Pellet transects were located to index hare abundance were not likely to travel through alpine areas as they might across the study area, not among habitats; hence, we have no have done for other habitats such as mature forest or riparian data for two of our seven habitats. Hare abundance was willow. Most locations in alpine were near the polygon higher in the dense immature pine class during all phases of boundary, where we observed the greatest numbers of hares. the cycle, including the low (Fig. 2). Generally, other habi- Riparian willow was used more in winter when the sub- tats had similar numbers of hares during the low phase. Hare strate was frozen than during summer (Fig. 4a). No other densities in mature pine and mixed immature pine peaked at seasonal preferences were apparent. Yearlings used mature roughly half the numbers of the other three habitats, while forest and perhaps riparian willow more than adults (Fig. 4b). numbers in riparian willow and mature spruce increased to Habitat preference was similar between sexes (Fig. 4c). nearly that found in dense immature pine. Hare numbers Lynx habitat selection was related to understory density as were likely very low in the alpine. Any sign of hares that we indexed by our assigned ranks, but selection was not related observed in alpine was in the subalpine forest and shrub to overstory density (Fig. 5). We also assigned hare abun- zones that are found along the alpine–forest boundary. The dance using a subjective rank and these ranks were identical majority of the alpine zone was covered in very low vegeta- to the understory rank except that abundance in immature tion and supported no hares. spruce may be lower than suggested by the understory den- These data are weak measures of hare abundance among sity. Our combined subjective and empirical information habitats because the level of sampling was inadequate to suggests that understory, hare density, and lynx habitat use measure the variation in each habitat (n = 1–4 transects) and are positively correlated, while overstory is not necessarily transects were confined to the burned matrix. Hares were related to any of the three measures. likely more abundant in mixed immature pine than indicated by these data because our sample sites were dryer than typi- Discussion cal. Abundance in mature spruce and mature pine was likely lower than the data indicate because our sites were all in In contrast with our prediction, we did not detect changes small unburned exclusions within the burned matrix. Data in lynx habitat preference among phases of the hare cycle. for dense immature pine and riparian willow likely represent We expected lynx use of dense immature pine to increase those found in the mapped habitats across the study area. during the decline and low phases because hare numbers declined to very low levels in other habitats during the low Lynx habitat preference phase. And we expected lynx to use open habitats less dur- Lynx habitat preference did not change through the hare ing the decline and low phases. In a review of snowshoe cycle; all confidence intervals overlap among phases for all hare ecology in North America, Hodges (2000a, 2000b) six habitats (Fig. 3). Immature pine habitats were preferred found little evidence that hares change habitat preferences

Fig. 3. Lynx habitat preference across a cycle in snowshoe hare abundance in south-central Yukon, 1986–1994. Preference is measured by w and 95% binomial confidence intervals are given with α adjusted to 0.008 (Krebs 1999).

during a cycle. Wirsing et al. (2002) found that predation cial structure. Populations that are heavily trapped and there- rates on hares were similar across habitats types and before less constrained by social structure may demonstrate tween resident and dispersing hares in a landscape with low larger scale changes in habitat use through a hare cycle. hare abundance in Idaho, which does not support the hy- While habitat preferences by lynx varied little through the pothesis that hares survive in refugia during periods of low hare cycle, preference for immature habitats that contained abundance. Using similar field methods and scales of analy- pine was obvious (Figs. 3 and 4). Immature pine habitats sis to the present work, Poole et al. (1996) did not detect a were preferred over immature spruce of the same age. Im- change in habitat preference by lynx through a portion of a mature pine trees were heavily browsed in our study area, hare cycle in a study area of largely regenerating forest in and we assumed that pine was the preferred browse for the southern Northwest Territories. In contrast, lynx were hares. Koehler (1990) found that pine stands were the most found to move to denser habitat, which supported more commonly used habitat by lynx in Washington and that 20- hares, in southwest Yukon during the decline phase in south- year-old pine stands supported the greatest number of hares. west Yukon (Murray et al. 1994; O’Donoghue et al. 1998, Hares ate largely pine branch tips, stems, and bark and 2001). These authors measured use of as small as 20 m wide avoided alder in young pine stands (Koehler 1990). Both im- habitat patches, while this study and Poole et al. (1996) mature and mature spruce habitats contained mostly willow measured use of at least 100 m wide patches. We conclude understory and our observations suggest that hares were eat- that variation in habitat preference at the mesoscale level ing largely willow in spruce habitats. Pine habitats were usu- through the hare cycle was minimal on our study area and ally too dry to contain much willow. We assume that hare this conclusion is supported by Poole et al. (1996). Like numbers were higher in immature pine than in immature Poole et al. (1996), we studied an untrapped population spruce habitats but we have no supporting data. where most habitats were favourable for lynx; only the ma- Riparian willow habitats were preferred equally to the im- ture forest and alpine had open understories. Our observa- mature coniferous habitats. Hare abundance was highly vari- tion does not preclude lynx showing variable preference in able in this class because it included sedge meadows, open more heterogeneous landscapes or at finer scales of analysis. willow swales, and dense stands of mature willow. We ob- Our habitat mapping was coarse and lynx may have shown served high numbers of hares in dense willow stands along variable selection for small patches (<2 ha) of willow or creeks and lake edges and we suspect that it was these dense pine that were present in all of the habitats that lynx stands that attracted lynx to this habitat. Meadows and wil- preferred. Our conclusion suggests that although fine-scale low swales received little hare use in winter. changes in habitat use may occur through the hare cycle, Most mature forest on our study area was dominated by perhaps based on changes in hunting strategy (O’Donoghue spruce with both a dense overstory and a relatively sparse et al. 2001), at larger scales, resident lynx use the landscape understory. Understory densities in mapped mature spruce in a similar fashion, perhaps owing to the constraints of so- polygons were probably less than indicated in Table 1 be-

Fig. 4. Lynx habitat preference between seasons (a), ages (b), and sexes (c) in south-central Yukon, 1986–1994. Preference is measured by w and 95% binomial confidence intervals are given with α adjusted to 0.008 (Krebs 1999).

Fig. 5. Relationship between vegetation overstory and understory and lynx habitat selection. Both understory and overstory were assigned a general index value (1 is lowest, 5 is highest) based on field data, sample locations, and our knowledge of the study area.

cause sample sites were in small mature inclusions in the served by Major (1989), who worked 20 km west of our burned matrix. These stands were often partially disturbed study area. and subsequently invaded by willow. In interior Alaska, We did not find that age or sex markedly affected habitat lynx were observed to select for mature stands within a re- preference of this population. Yearlings used mature forest cent burn, but they did not select for mature stands in a ma- and willow slightly more than adults, which may reflect the ture forest matrix (Kesterson 1988; Staples 1995). Our fact that they spent more time exploring potential home results concur with the latter observation. We suspect that ranges and these two habitats were natural travel paths. Wil- lynx use of pine-dominated mature stands would be even low was located along waterways and hence offered low- less than that of mature spruce because hare numbers are of- elevation routes to nearby areas and both riparian willow and ten low in pure pine stands (Major 1989, Fig. 2) because of mature forest were relatively open and were easier to travel the lack of understory (Table 1). through than the regenerating matrix. These modest differ- Lynx used the alpine–subalpine boundary area throughout ences in preference between age classes are perhaps not sur- the hare cycle but rarely used the more open upslope areas. prising for a habitat specialist such as the lynx. Poole et al. We observed patches of willow – subalpine spruce that con- (1996) did not find a difference in habitat selection between tained hare tracks in alpine habitat, but we do not know sexes in lynx in the southern Northwest Territories. We con- whether lynx hunted hares or other prey species, such as clude that there was no variation in habitat preference be- ptarmigan (Lagopus spp.), in this habitat. tween sexes and between subadults and adults at the Contrary to our prediction for greater use of open habitats mesoscale level in a burned forest matrix. These observa- in summer, the only difference that we detected in habitat tions were made in a largely untrapped study area where preference between seasons was for greater use of riparian turnover of adults was low. Also, data were collected from willow during winter. There are three possible explanations resident individuals, most of whom occupied stationary for this observation. First, much of the ground in this habitat home ranges, although overlap was common, especially dur- is saturated by water and lynx may prefer to use this habitat ing the high phase (Slough and Mowat 1996) and among re- when the ground is frozen. Second, visibility is greater dur- lated individuals (Mowat and Slough 1997). We may expect ing winter, after the leaves have fallen, and this may give greater difference in habitat use between sexes or age classes lynx an advantage given that horizontal cover can be ex- in populations that are well below the carrying capacity be- tremely dense in closed willow stands (Table 1). The third cause individuals presumably show less fidelity to their home explanation is based on human influence. We used approxi- ranges because social pressures are lower. mately 100 km of snowmobile trails during livetrapping and Vegetation structure appears to influence lynx habitat lynx regularly travelled on these trails. Our trails traversed preference. Understory density was high for all of the most willow polygons much more than other habitats and the ten- preferred habitats (Fig. 5, Table 1), but the immature spruce dency for lynx to use our trails may have positively biased class also had high understory closure and was less preferred our measure of lynx winter use of riparian willow. We con- than the immature pine and willow classes. Hare density was clude that variation in lynx habitat preference between sea- tightly related to increasing understory in Idaho (r2 = 0.97; sons is small at the mesoscale level in a burned forest Wirsing et al. 2002). The dominant species of the understory matrix. Finer scale changes in use may have occurred, as ob- also influenced habitat quality for lynx, and probably hares,

© 2003 NRC Canada 1744 Can. J. Zool. Vol. 81, 2003 on our study area. Notwithstanding the preference for pine Acknowledgements over spruce, our data support the importance of understory to hares (Keith 1990) and underline the ultimate importance The Yukon Department of Renewable Resources funded of understory to lynx as well. Overstory closure was not this study. We thank D. Larsen and V. Loewen for continued positively related to lynx habitat choice because the three support. We thank R.M.P. Ward, G.T. Hunter, K.R. Frankish, preferred habitats, the two immature pine classes and ripar- R. Rivard, B. Gilroy, and the many volunteer field assistants ian willow, all had low overstory closure (Table 1). Wirsing for technical contributions to the study. K. McKenna classi- et al. (2002) showed that hare abundance was not related to fied and mapped vegetation types and G. Perrier performed overstory in a series of low-density hare populations in GIS analyses. V. Loewen and J. Staniforth helped with habi- Idaho. These observations concur with the suggestion of tat mapping. Pilot D. Denison is acknowledged for his skill- Keith (1990) that overstory is not an important factor in hare ful flying and expertise in aerial telemetry and trapper habitat quality. T. Hall for sharing his trapline. M. O’Donoghue, K.G. Poole, We located lynx during the daylight hours when lynx and two anonymous reviewers provided constructive com- were more likely to be bedding than hunting, at least in win- ments on manuscript drafts. ter. Lynx bed in the habitats that they hunt in (Major 1989), but they may avoid bedding in more open habitats (Murray et al. 1995). Our telemetry data may underestimate lynx use References in open habitats such as mature forest, alpine, and possibly riparian willow, which contained open meadows and willow Aubry, K.B., Koehler, G.M., and Squires, J.R. 2000. Ecology of swales. We suspect that this potential bias is small because Canada lynx in the southern boreal forests. In Ecology and con- habitat polygons were large and lynx would probably have servation of lynx in the United States. Chap. 13. Edited by L.F. found a bedding site in the polygon that they were hunting Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Krebs, in, except perhaps in alpine. K.S. McKelvey, and J.R. Squires. University Press of Colorado, Boulder, Colo. pp. 373–396. Our data demonstrate the importance of regenerating stands Bostock, H.S. 1965. Physiography of the Canadian Cordillera, with to lynx, especially those with a pine component. Eighty-six special reference to the area north of the fifty-fifth parallel. Can. percent of all lynx locations were in regenerating habitat. Dep. Mines Resour. Bureau Geol. Topogr. Geol. Surv. Memo. Over 70% of our study area was 30- to 35-year-old regener- 247. ating forest, half of which contained pine in the understory. Breitenmoser, U., Kavczensky, P., Dötterer, M., Breitenmoser- These habitats contained among the highest hare and lynx Würsten, C., Capt, S., Bernhart, F., and Liberek, M. 1993. Spa- densities reported in North America (Slough and Mowat tial organization and recruitment of lynx in a re-introduced pop- 1996). Further, a relatively large proportion of our study area ulation in the Swiss Jura Mountains. J. Zool. (Lond.) 231: 449– (5%) was riparian willow that supported high hare numbers, 464. especially in the densest stands. Willow was the only climax Canadian Council on Animal Care. 1984. Guide to the care and use habitat with high hare numbers, and of the 14% of lynx lo- of experimental animals. Canadian Council on Animal Care, cations in nonburned habitats, 45% were in willow. Willow Ottawa, Ont. was strongly preferred over both other unburned habitats Davies, D., Kennedy, C.E., and McKenna, K. 1983. Resource in- (Fig. 2). Moderate lynx and hare numbers have been noted ventory — Southern Lakes. Yukon Department of Renewable in mature forests but only those that have relatively open Resources, Land Planning Branch, Whitehorse, Yukon. canopies and well-developed understories (Mowat et al. Fuller, T.K., and Heisey, D.M. 1986. Density-related changes in 2000). Our data suggest that continuous closed-canopy ma- winter distribution of snowshoe hares in northcentral Minnesota. ture forest would support very low densities of lynx. J. Wildl. Manag. 50: 261–264. Forest fires create regenerating landscapes that are often Hodges, K.E. 2000a. The ecology of snowshoe hares in northern pine dominated, and both regenerating habitat and pine ben- boreal forests. In Ecology and conservation of lynx in the efit lynx. As a result, suppression of fires must negatively af- United States. Chap. 6. Edited by L.F. Ruggiero, K.B. Aubry, fect lynx populations. Logging and disease outbreaks in S.W. Buskirk, G.M. Koehler, C.J. Krebs, K.S. McKelvey, and mature forests will also create regenerating environments J.R. Squires. University Press of Colorado, Boulder, Colo. pp. 117–162. that may benefit lynx. But habitat quality for lynx will de- Hodges, K.E. 2000b. Ecology of snowshoe hares in southern boreal pend on understory density and possibly the prevalence of and montane forests. In Ecology and conservation of lynx in the pine in the regenerating community. Planting lodgepole pine United States. Chap. 7. Edited by L.F. Ruggiero, K.B. Aubry, on cut areas may result in greater hare and lynx numbers S.W. Buskirk, G.M. Koehler, C.J. Krebs, K.S. McKelvey, and compared with other conifer species; however, hare brows- J.R. Squires. University Press of Colorado, Boulder, Colo. ing may impede juvenile pine regeneration. Understory den- pp. 163–206. sities are unlikely to attain the values observed in Table 1 in Keith, L.B. 1990. Dynamics of snowshoe hare populations. In Cur- managed stands; therefore, we conclude that regenerating rent mammalogy. Edited by H.H. Genoways. Plenum Press, New clearcuts will rarely support equal lynx densities to naturally York, N.Y. pp. 119–195. regenerating burns. Given the similarity in preference that Kesterson, M.B. 1988. Lynx home range and spatial organization we observed across large changes in lynx density, these con- in relation to population density and prey abundance. M.S. the- clusions may apply across much of the lynx’s range, at least sis, University of Alaska, Fairbanks, Alaska. in western North America (Koehler 1990; Aubry et al. Koehler, G.M. 1990. Population and habitat characteristics of lynx 2000). Fire suppression may have contributed to the decline and snowshoe hares in north central Washington. Can. J. Zool. in lynx numbers in the south of their western range. 68: 565–567.

Krebs, C.J. 1999. Ecological methodology. 2nd ed. Benjamin/ O’Donoghue, M., Boutin, S., Murray, D.L., Krebs, C.J., Hofer, Cummings, Redwood City, Calif. E.J., Breitenmoser, U., Breitenmoser-Wüersten, C., Zuleta, G., Krebs, C.J., Gilbert, B.S., Boutin, S., and Boonstra, R. 1987. Esti- Doyle, C., and Nams, V.O. 2001. Coyotes and lynx. In Ecosys- mation of snowshoe hare population density from turd transects. tem dynamics in the boreal forest: the Kluane Project. Chap. 13. Can. J. Zool. 65: 565–567. Edited by C.J. Krebs, S. Boutin, and R. Boonstra. Oxford Uni- Krebs, C.J., Boonstra, R., Nams, V., O’Donoghue, M., Hodges, K., versity Press, New York, N.Y. pp. 275–323. and Boutin, S. 2001. Estimating snowshoe hare population den- Oswald, E.T., and Brown, B.N. 1986. Forest communities in Lake sity from pellet plots: a further evaluation. Can. J. Zool. 79: 1–4. Laberge ecoregion, Yukon Territory. Can. For. Serv. Pac. For. Major, A.R. 1989. Lynx predation ecology and habitat use in the Res. Cent. Inf. Rep. BC-X-282. Yukon Territory, Canada. M.Sc. thesis, State University of New Parker, G.R. 1981. Winter habitat use and hunting activities of lynx York, Syracuse, N.Y. (Lynx canadensis) on Cape Breton Island, Nova Scotia. In Manly, B., Mcdonald, L., and Thomas, D. 1993. Resource selec- Proceedings of the Worldwide Furbearer Conference, Frostburg, tion by animals: statistical design and analysis for field studies. Md., 3–11 August 1980. Edited by J.A. Chapman and D. Chapman and Hall, London, U.K. Pursley. Worldwide Furbearer Conference, Inc., Frostburg, Md. McCord, C.M., and Cardoza, J.E. 1982. Bobcat and lynx. In Wild pp. 221–248. mammals of North America. Edited by J.A. Chapman and G.A. Poole, K.G. 2001. COSEWIC status report on Canada lynx (Lynx Feldhamer. Johns Hopkins University Press, Baltimore, Md. canadensis). Committee on the Status of Endangered Wildlife in pp. 728–766. Canada, Ottawa, Ont. Poole, K.G., Wakelyn, L.A., and Nicklen, P.N. 1996. Habitat selec- McKelvey, K.S., Aubry, K.B., and Ortega, Y.K. 2000. History and tion by lynx in the Northwest Territories. Can. J. Zool. 74: 845– distribution of lynx in the contiguous Unites States. In Ecology 850. and conservation of lynx in the United States. Chap. 8. Edited Rueness, E.K., Stenseth, N.C., O’Donoghue, M., Boutin, S., by L.F. Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Ellegren, H., and Jakobsen, K.S. 2003. Ecological and genetic Krebs, K.S. McKelvey, and J.R. Squires. University Press of spatial structuring in the Canadian lynx. Nature (Lond.), 425: Colorado, Boulder, Colo. pp. 207–264. 69–72. Mowat, G., and Slough, B.G. 1997. Some observations on behav- Ruggiero, L.F., Aubry, K.B., Buskirk, S.W., Koehler, G.M., Krebs, iour and natural history of Canada lynx. Can. Field-Nat. 112: C.J., McKelvey, K.S., and Squires, J.R. (Editors). 2000. Ecology 32–36. and conservation of lynx in the United States. University Press Mowat, G., Slough, B.G., and Boutin, S. 1996. Lynx recruitment of Colorado, Boulder, Colo. during a snowshoe hare population decline in southwest Yukon. Saunders, J.K. 1963. Movements and activities of the lynx in New- J. Wildl. Manag. 60: 441–452. foundland. J. Wildl. Manag. 27: 390–400. Mowat, G., Poole, K.G., and O’Donoghue, M. 2000. The ecology Schmidt, K. 1998. Maternal behaviour and juvenile dispersal in the of lynx in northern Canada and Alaska. In Ecology and conser- Eurasian lynx. Acta Theriol. 43: 391–408. vation of lynx in the United States. Chap. 13. Edited by L.F. Schwartz, M.K., Mills, L.S., McKelvey, K.S., Ruggiero, L.F., and Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Krebs, Allendorf, F.W. 2002. DNA reveals high dispersal synchronizing K.S. McKelvey, and J.R. Squires. University Press of Colorado, the population dynamics of Canada lynx. Nature (Lond.), 415: Boulder, Colo. pp. 265–306. 520–521. Murray, D.L., Boutin, S., and O’Donoghue, M. 1994. Winter habi- Slough, B.G., and Mowat, G. 1996. Lynx population dynamics in tat selection by lynx and coyotes in relation to snowshoe hare an untrapped refugium. J. Wildl. Manag. 60: 946–961. abundance. Can. J. Zool. 72: 1444–1451. Staples, W.R. 1995. Lynx and coyote diet and habitat relationships Murray, D.L., Boutin, S., O’Donoghue, M., and Nams, V.O. 1995. during a low hare population on the Kenai Peninsula, Alaska. Hunting behaviour of a sympatric felid and canid in relation to M.S. thesis, University of Alaska, Fairbanks, Alaska. vegetative cover. Anim. Behav. 50: 1203–1210. Ward, R.M.P., and Krebs, C.J. 1985. Behavioural responses of lynx Murray, D.L., Roth, J.D., Ellsworth, E., Wirsing, A.J., and Steury, to declining snowshoe hare abundance. Can. J. Zool. 63: 2817– T.D. 2002. Estimating low-density snowshoe hare populations 2824. using fecal pellet counts. Can. J. Zool. 80: 771–781. Wirsing, A.J., Steury, T.D., and Murray, D.L. 2002. A demographic O’Donoghue, M., Boutin, S., Krebs, C.J., Murray, D.L., and Hofer, analysis of a southern snowshoe hare population in a fragmented E.J. 1998. Behavioural responses of coyotes and lynx to the habitat: evaluating the refugium model. Can. J. Zool. 80: 169– snowshoe hare cycle. Oikos, 82: 169–183. 177.

PAPER IV

Estimating marten Martes americana population size using hair capture and genetic tagging

Garth Mowat & David Paetkau

Mowat, G. & Paetkau, D. 2002: Estimating marten Martes americana population size using hair capture and genetic tagging. - Wildl. Biol. 8: 201-209.

We tested non-invasive genetic methods for estimating the abundance of marten Martes americana using baited glue-patch traps to pull hair samples from individual animals. We divided our 800-km2 study area into 3 H 3 km cells and put one hair trap in each cell. We trapped 309 sites for an average of 15 days each between 15 January and 14 March 1997. Based on tracks in snow and hair morphology, we captured hair from marten, red squirrels Tamiasciurus hudsonicus, flying squirrels Glaucomys sabrinus, short or long-tailed weasels Mustela erminea and M. frenata, and several unidentified mouse and vole species. Of 309 sites, 58% collected a marten hair sample while 8% of sites removed weasel hair. When roots were embedded in adhesive, a xylene wash was used to remove them before extracting DNA. All marten samples were genotyped at six microsatellite loci to identify individuals. Xylene-washed samples yielded similar genotyping success to samples that had never been exposed to xylene, and genotyping success increased with the number of hairs in the sample. Genetic data allowed 139 samples to be assigned to 88 individual marten, constituting 124 capture events during the four trapping sessions. The population estimate for our study area was 213 (95% CI: 148-348) and the average capture probability was 0.15. The density of marten in our study area was 0.33/km2 when inhospitable habitat was removed from the calculation. We believe hair sampling and genetic analysis could be used to measure population distribution, trend and size for marten, and perhaps also for other carnivores.

Key words: Canada, density, hair removal, mark-recapture, marten, Martes americana, microsatellite genotyping

Garth Mowat, Aurora Wildlife Research, RR 1, Site 14, Comp 8, Crescent Valley, B.C., V0G 1H0, Canada - e-mail: [email protected] David Paetkau, Wildlife Genetics International, Box 274, Nelson, B.C., V1L 5P9, Canada - e-mail: [email protected]

Corresponding author: Garth Mowat

Received 22 October 2001, accepted 15 January 2002

Associate Editor: Paolo Cavallini

Monitoring the abundance of carnivores is problemat- estimate population size (Raphael 1994, Zielinski & Ku- ic; track plate, remote camera, and track count methods cera 1995). Detection methods are well suited for com- have been developed for marten and many small car- paring changes in distribution, but they may lack pow- nivores; however, these methods detect the target spe- er to detect trends in abundance (Strayer 1999). Intensive cies at a site but do not allow the identification of indi- sampling may be required to detect even gross changes viduals. Hence, detection methods can be used to mea- in abundance (Zielinski & Stauffer 1996). Also, using sure relative abundance and population trend but not to detection data to examine population trend assumes

© WILDLIFE BIOLOGY · 8:3 (2002) 201 that detection probabilities are similar among surveys relatives (Woods, Paetkau, Lewis, McLellan, Proctor & (Harris 1986). Variation in sampling techniques, weath- Strobeck 1999). It is also possible to use genetic marker and food abundance among surveys may make this ers to determine gender and species (Taberlet, Mattock, assumption questionable. Using estimates of abundance Dubois-Paganon & Bouvet 1993, Foran, Minta & Hei- may allow the detection of more subtle trends in abun- nemeyer 1997b). dance, and there is no assumption that capture proba- The goal of our study was to test the efficacy of non- bilities are equal among surveys. invasive DNA sampling for estimating population size The development of microsatellite markers for ana- of marten Martes americana. We were also interested lyzing patterns of genetic variation is recent (Snow & in finding out whether the field methods presented by Parker 1998). These markers are highly variable and (Foran, Crooks & Minta 1997a) worked for weasels Mu- therefore well suited to comparing genetic variation stela erminea and M. frenata and other similar-sized among individuals. Microsatellite analysis, using poly- mammals. Our long-term goal was to use marten, and erase chain reaction (PCR), can be performed using rela- perhaps weasels, as focal species in monitoring forest tively small DNA samples, allowing the use of non-inva- biodiversity (sic McLaren, Thompson & Baker 1998). sive collection techniques which typically yield small amounts of low quality DNA. Modern equipment allows the repeatable analysis of large numbers of samples, mak- Study area ing microsatellite markers particularly useful for consistently assigning identity to samples from unknown Our study area covered 797 km2 of the central Selkirk individuals. Match statistics can be used to ensure that Mountains in southeastern British Columbia (Fig. 1). the available genetic data are able to resolve individu- In this area, cedar-hemlock forests are found below als even when the study population contains many close approximately 1,400 m a.s.l., where western hemlock Tsuga heterophyla, western red cedar Thuja plicata, Douglas fir Pseudot- suga menziesii and spruce hybrids Picea spp. are the dominant canopy species. Engelmann spruce Picea engelmannii - subalpine fir Abies lasiocarpa forests are found between roughly 1,400 and 2,300 m a.s.l., where these two species dominate the canopy, although many early seral stands are dominated by lodgepole pine Pinus contorta. Extensive areas of alpine tundra are found above ca 2,300 m. The area is mountainous with little flat ground, scattered rock outcrops and cliffs, steep-sided watercourses and lakes, and few well-developed riparian areas. The study area is in the interior wetbelt and receives approximately 80 cm of precipitation per year (Environment Canada, New Den- ver weather station, climate normals). Marten and short and long-tailed weasels occur throughout both forest zones, but fisher Martes pennanti, which have morphologically similar hair to marten, are absent from this region (Cowan & Guiguet 1965, Gibilisco 1994). Figure 1. Grid cells and trap locations for the marten study area in southeastern British Columbia. Sites that detected marten are black circles, those that did not are open circles. Solid lines depict movements for individuals that were recaptured at a different site.

202 © WILDLIFE BIOLOGY · 8:3 (2002) Table 1. Hair capture results using glue-patch traps to sample individual marten in southeast British Columbia during January-March 1997.

Marten Weasel Marten Number Mean detections detections captures Session Start End of sites duration N (%) N (%) N 1 15 January 27 January 71 13 33 (46) 7 (10) 19 2 27 January 10 January 77 13 45 (58) 3 (4) 29 3 10 February 23 February 77 15 48 (62) 9 (12) 42 4 23 February 14 March 84 18 54 (64) 5 (6) 34 Total 309 15 180 (58) 24(8) 124

Methods with a piece of clean flagging tape and stored it in a sealed plastic bag. All samples were stored in a freez- We divided the study area into 3 H 3 km cells (see Fig. er. We recorded observations of tracks in snow at all sites 1). Study area boundaries were largely along heights of to aid in species identification and to record the presence land between major watersheds. Most boundaries were of a target species that was not detected in the hair trap. at high altitudes and above the treeline. These bound- Hair samples were sorted to species based on morphol- aries probably provided geographic closure to the res- ogy and colour. Weasels had short white hair, although ident marten population, especially in winter. There a few samples contained longer black hairs, presumably were 86 cells in the study area, and we tried to set one from the tail. Red squirrels Tamiasciurus hudsonicus and marten capture site in each cell during each trapping ses- flying squirrels Glaucomys sabrinus had short red- sion. However, some cells with very poor access or no brown to gray hair with little difference between the length marten habitat (i.e. no forest) were not trapped. In the of the guard hair and under fur. Mink Mustela vison hair first session, 15 cells were not trapped and during the is usually darker than marten hair, but the two species second and third sessions nine cells were missed; dur- could be confused based on hair morphology alone. ing the final session, all but two cells were trapped Mink were not abundant in our study area due to the because we used a helicopter for access (Table 1). After scarcity of riparian habitat. approximately 14 days, each capture site was changed We removed hair from glue patches by washing the to another location in the same cell. We trapped four 14- entire patch in xylene until the glue was soft enough to day sessions during January-March 1997. easily remove hairs from the patch (C. Strobeck, pers. The field personnel were instructed to install sites in comm.). In many cases we were able to cut enough places they felt would be most likely to catch marten roots from hair that was suspended beyond the glue while attempting to space sites at least 1 km apart. Both patch to avoid the xylene wash. We normally extracted principal technicians had previous experience trapping hair from only one of the glue patches at a site, although marten. We used baited glue traps to remove hair from on rare occasions we took hair from several patches to animals. Traps were fashioned after the design of Foran increase the number of hairs put into the extraction. et al. (1997a). They consisted of two pieces of wood (2 H Entire hairs were put in 1.5 ml microfuge tubes and DNA 14 H 60 cm), screwed together at the edge of each board was extracted using QIAamp™ DNA Mini Kits (QIA- to form a triangle, and were attached to trees with wood GEN Inc., Santa Cruz, California), eluting in a final screws. Inside the triangle formed by the trap, four volume of 200 μl. We used 15 hairs per sample when pos- patches (1.5 H 5 cm) of Stick-em™ (Woodstream Trap sible, but fewer hairs were used when necessary. We Corporation, Ithaca, NY, USA) were attached with screened 12 marten microsatellite markers: Ma1, Ma2, thumb tacks about one third of the way in from the trap Ma3, Ma4, Ma7, Ma8, Ma10, Ma11, Ma14, Ma15, opening. Stick-em™ is a mouse trap based on a card- Ma18, Ma19 (Davis & Strobeck 1998) and chose a board-backed, peanut butter-laced industrial adhesive. suite of six markers (Ma1, Ma2, Ma8, Ma10, Ma18, In the middle of the wooden triangle, a piece of rotten Ma19) based on demonstrated variability and amplifi- chicken was attached to a fence staple with wire. We also cation strength with small quantities of DNA. We designed applied commercial marten lure inside the trap. Above new primers for some loci to improve amplification the trap we attached a film canister with a cotton ball characteristics and to enable all loci to be analyzed in a soaked in fish oil (rendered fish) inside. Traps were single lane (Table 2). PCR conditions were as described attached to trees vertically and the top part of the trap by Davis & Strobeck (1998) using a volume of 15 μl con- was protected by a small wooden porch. taining 10 μl of template DNA. PCR products were ana- When a glue patch had hair on it, we covered the patch lyzed under standard conditions using an Applied Bio-

© WILDLIFE BIOLOGY · 8:3 (2002) 203 Table 2. Primer sequences, observed allele sizes, probability of identity and heterozygosity for the six microsatellite loci used to genotype marten samples in this study.

Size range Locus Label1 Labelled primer Unlabelled primer (base pairs) Psib2 H3 Ma1 FAM TTA TGC GTC TCT GTT GCG GTA GAA TAC AGG 177-196 0.445 0.738 TGT CA4 CTT GTT Ma2 TET ACC CAT GAA TAA TGT AAA TTG ATA GAA ATT 147-157 0.432 0.738 CTT AT ACT GGT GTT4 Ma8 TET GTT TTC TAA TGT TTC CAG TGG TTG ACT ACA 116-132 0.523 0.575 GTG TG AGA AA Ma10 HEX TCT TTT CCT CTC CCT GGT GCC CCA TAT TGA 163-171 0.358 0.825 CTT CC CTA TT Ma18 FAM TAC TCA GTG GGG AAT TGG GTG GGT GTA TTT 146-161 0.383 0.750 CTG CT GTG TAT Ma19 FAM GAT CAT TTG GTA TTT AAG GCT TAT GGA TAC 201-210 0.590 0.488 GTC TTT C CAC AT 1 One primer from each pair was 5’ labelled with a fluorescent dye group (Applied Biosystems). 2 The mean probability, across 80 individual marten for which we had complete 6-locus genotypes, that a full sibling would have the same genotype (the basis of our match declarations). A lower value indicates greater power to genetically resolve individuals. 3 Mean observed heterozygosity across the same 80 individuals. 4 These primers were designed based on the sequences of Davis & Strobeck (1998): Genbank accession numbers AF075137 and AF07138. systems 310 automated sequencer, and genotypes were In addition to restricting the analysis to samples with scored with the aid of Genotyper software (Applied Bio- 4-locus data or better, we calculated the probability systems). that a full sibling would have the same genotype (Psib; Davis & Strobeck (1998) provided information on Woods et al. 1999) for each pair of matching samples, amplification patterns (presence of product, size, vari- limiting the calculation to only those loci that were ability) in mink for the markers we used. To provide sim- complete for both samples under consideration. We re- ilar information on short-tailed weasels, we ran two quired that this probability be less than 0.05 to assign known short-tailed weasel samples as well as four sam- a sample to an individual. ples identified as weasel using tracks and hair morphol- We used the mark-recapture models available in the ogy. We assumed that squirrel samples would not pro- program CAPTURE (Otis, Burnham, White & Andersen duce scorable PCR products in the size range observed 1978, White, Andersen, Burnham & Otis 1982) for for marten. A likelihood of occurrence (Prandom; Woods estimation of population size. Model selection was et al. 1999) was calculated for each genotype to identify based on our knowledge of marten biology and move- extremely rare genotypes. These sources of information ments and the goodness-of-fit tests available in CAP- allowed us to confirm species for the samples we geno- TURE. We used Pearson correlation to assess the rela- typed. tionship between the number of hairs in a sample and Genotyping was carried out in several distinct phases. the number of loci that could be scored. We used logis- An initial attempt was made to obtain genotypes for tic regression to test the relationship between the num- every extracted sample. Samples producing scorable ber of hairs in a sample and the probability of scoring products for fewer than three loci in this initial screen a 6-locus genotype. We used the logit function and were excluded from subsequent analyses. Samples that tested assumptions regarding residuals. All analysis produced data for 3-5 loci were then subjected to a was done using SAS 6.12 (SAS Institute Inc., Cary, NC, second round of PCR and scoring for those loci that could USA). not be scored on the first screen. After the second attempt, samples with genotypes at less than four loci were excluded. We wrote a computer program to assign samples Results with identical genotypes to individuals, and to identify pairs of genotypes that differed at only one locus. These In November and December 1996, we tested hair re- highly similar pairs of genotypes were checked for moval methods near Nelson, British Columbia. We found scoring or typing errors. Similar genotypes that could that it was easier to attach traps on trees vertically and not be explained by these errors were then subjected to that this placement generated hair samples more con- another round of PCR and scoring for the loci where the sistently than horizontal placement. We experimented differences were observed (see also Poole, Mowat & Fear with a number of glue patch placements and found 2001). that we got larger hair samples if the patches were high

204 © WILDLIFE BIOLOGY · 8:3 (2002) up in the V of the trap and placed to capture hair mid- discarded because there may have been allelic dropout, way down the back of the animal, i.e. midway between but the DNA sample was used up before the suspi- the center of the trap and the end. Upper and lower hair cious genotype could be confirmed. After correcting these patches were equally successful at removing hair. For errors, all pairs of genotypes in the data set differed at marten, 92% of upper patches removed hair vs 87% of two or more loci. Given that we checked every case lower patches (N = 180 sites that detected marten). where a single error could have occurred and how rare For weasel, 96% of upper patches removed hair while these events were (4/139), we believe that the probability 100% of lower patches removed hair (N = 24). Snow, and of having errors at two loci in the same genotype is neg- especially rain, caused the paper backing on the back of ligible, and that each of the 88 different genotypes in the patches to soften, occasionally rendering the patch- our data set corresponded to a unique individual. es useless. Using large live trees and wooden porches The xylene treatment appeared to have little effect on above traps minimized this problem. genotyping success, with 83% (N = 101) of clipped sam- Based on tracks in snow and hair morphology, we cap- ples and 75% (N = 69) of xylene extracted samples gen- tured hair from marten, red squirrels, flying squirrels, erating 4-locus genotypes or better. The number of short or long-tailed weasels, and several unidentified roots in a sample was positively correlated with the num- mouse and vole species. Several glue patches obvious- ber of scorable loci for both clipped (r = 0.39, N = 99, ly had hair from more than one species. However, except P < 0.0001) and xylene treated (r = 0.32, N = 66, P = for the two weasel species and between marten and 0.01) samples. Logistic regression showed a significant mink, hair morphology was so different for the species relationship between the number of roots in a sample in our study area that they could be separated visually. and the event of scoring a complete 6-locus genotype Of 309 sites, 58% removed a marten hair sample while (P2 = 12.9, df = 1, P = 0.0003; Fig. 2). We also ran the 8% of sites produced weasel samples (see Table 1). Based logistic analysis for clipped and xylene treated samples on tracks in snow, 14 sites were approached by weasels separately. The logit probabilities and the shape of the but did not capture hair. Likewise, 15 sites approached logit curves were similar for both extraction methods, by marten failed to collect samples. The trap failure rate, so we present only the combined analysis. calculated as the proportion of visited sites that did not The 139 marten samples identified were all from dif- remove a hair sample, was much higher for weasels ferent sites and corresponded to 124 capture events during (37%) than for marten (8%). the four trapping sessions (see Table 1). The number of Marten were detected throughout the study area (see capture events per session is less than the total number Fig. 1). The linear fashion in which sites were distrib- of captures because several marten were detected at uted through the study area is largely a result of the very multiple sites within a session. Individual marten were mountainous terrain and the fact that we relied heavi- captured at as many as eight different sites, and multi- ly on the road network for access. Individual marten were ple captures of an individual within a session occurred detected at different sites 36 times, and the average 16 times during the study. movement distance was 2.6 km (range: 0.5-8.9 km). Movements appeared to be confined to low-elevation forested areas except for one movement in the north end 1.00 of the study area, where an individual probably crossed 0.90 through 1 km of alpine habitat to access the adjacent 0.80 drainage (see Fig. 1). 0.70 We extracted DNA from 180 marten hair samples. Of 0.60 these, 139 produced enough genetic data to be assigned 0.50 to individuals. Of those samples that could not be as- 0.40 signed to individuals, two were excluded because P 0.30 sib 0.20

was >0.05, and the remainder because <4 loci could be PROBABILITY OF SCORING SIX LOCI 0.10 scored. Two samples with 4-locus genotypes, and 21 0.00 samples with 5-locus genotypes, were assigned to indi- 0 5 10 15 20 viduals. Two scoring errors and two cases of allelic NUMBER OF HAIR ROOTS dropout (Taberlet, Griffin, Goossens, Questiau, Manceau, Escaravage, Waits & Bouvet 1996, Gagneux, Boesch Figure 2. Relationship between the number of hairs or roots put into a DNA sample and the probability of generating a six-locus genotype & Woodruff 1997) were detected by checking pairs of as calculated using logistic regression. Error bars give 95% confidence similar genotypes. One additional sample had to be intervals.

45 1 Marten density in our area was low 40 and most similar to other areas where 35 0.8 food abundance was low (Weckworth & Hawley 1962, Thompson & Colgan 30 0.6 1987), or where there was signifi- 25 cant trapping mortality (Payer 1999), 20 0.4 or to areas which had been clear-cut

15 CAPTURES in the preceding 35 years (Soutiere

NUMBER OF MARTEN 10 Captures 1979, Thompson & Colgan 1987,

0.2 PROPORTION OF NEW Payer 1999). The only other study of 5 % New captures marten population size done in tem- 0 0 perate western forest did not present 27 Jan 10 Feb 23 Feb 14 Mar densities. We calculated approximate TRAPPING SESSION densities from the measures of population size given in Weckworth & Figure 3. Number of marten captured and the proportion of new marten captured in each of Hawley (1962) using the study area the four trapping sessions in our study area in southern British Columbia during January-March size of ~15.5 km2 cited in Hawley & 1997. Dates are the last day of each trapping session. Newby (1957). Winter densities varied between 1.9 and 0.8 marten/km2 The goodness-of-fit tests for capture heterogeneity in in a mixture of forest types typical of the Rocky Moun- CAPTURE strongly suggested that capture probabili- tain west slopes. It is likely that some previous density ties were not equal among sessions (P2 = 12.99, df = 3, estimates are biased high by 'edge effect' (White et al. P = 0.005; Fig. 3), nor among individuals (P2 = 5.64, df = 1982), or may not be typical because of the tendency to 1, P = 0.02); however, there was only weak evidence for select good quality habitat as study sites (Smallwood & trap response behaviour (P2 = 4.66, df = 2, P = 0.10). Schonewald 1996). Our density estimate is a landscape-

The model selection routine in CAPTURE suggested Mtbh scale estimate for an area that has many recent logging (1.0) or Mo (0.96); Mth was selected at 0.88. Based on blocks, little trapping, no human residents, many logging these results, we selected Chao’s model Mth because it roads and a mixture of forest types from low elevation accommodates the predominant forms of capture vari- cedar-hemlock forests to upper elevation subalpine park- ation in our data. There is no model which accommo- land. dates all three forms of capture variation (Mtbh), and Mo We demonstrated that non-invasive sampling and does not accommodate any form of capture variation and genetic analysis can be used to measure distribution and is therefore unrealistic. estimate population size for areas large enough to be rel- The population estimate for our study area was 213 evant to marten population management or landscape- (95% CI: 148-348), and the average capture probabili- scale forest planning. This study area encompassed the ty was 0.15. The overall density of marten in our study greater part of three registered traplines (areas of pub- area was 0.27/km2 (CI: 0.19-0.44). However, this study lic land for which the licensee holds the exclusive right area was large and included considerable inhospitable to trap furbearing animals) and one entire forest license habitat for marten. Digital 1:20,000 scale forest cover area. In our study area 12 marten were trapped during mapping was available for the entire area, and we ex- the winter we worked there; this is <6% of the winter cluded all areas which were classified as rock, ice, water population. Marten were readily detected in our traps or alpine. No land affected by humans was removed be- as demonstrated by the low number of instances when cause little land was alienated by human settlement, animals had approached a trap but did not leave a hair but roads were plentiful. Virtually all area included in the sample (8%). Further, marten did not seem to be deterred density calculation would have succeeded to forest in- from re-entering traps as demonstrated by the relative- cluding open sub-alpine forest at high elevations. When ly high recapture rate. The method may work for oth- inhospitable area was removed, using a GIS, marten den- er carnivores as demonstrated by our ability to collect sity was 0.33/km2 (CI: 0.23-0.55). weasel hair, although modifications are needed to adjust the size of the trap to the smaller species. We screened for species based on tracks at the site,

206 © WILDLIFE BIOLOGY · 8:3 (2002) hair morphology and microsatellite results. While we multiple individuals on a single patch by carefully select- believe these efforts were sufficient in our study area, an ing clumps of hair, if adequate sample remains. As not- objective species test similar to the one presented by Foran ed by Foran et al. (1997a), restricting the trap in some et al. (1997b) or Woods et al. (1999) would likely be ne- way such that only one individual could enter would cessary in areas where mink or fisher are common. The reduce the risk of mixed samples during DNA analysis. data we obtained for the two known weasel samples and Shorter intervals between trap checks should also the four samples identified as weasel in the field were reduce the number of mixed samples. However, trap consistent in terms of size range, which loci amplified, check intervals must be long enough to ensure reason- and which loci were variable. No loci produced products able capture success. Trap-check interval was the most in the same size range as marten samples. Similarly, Davis important variable affecting detection success in a de- & Strobeck (1998) provided data which indicated that tailed study done by Zielinski, Truex, Ogan & Busse (1997) a mink genotype could not be confused with a marten using track plates to detect fisher and marten. Up to a genotype. In addition, none of the 88 genotypes we re- point, the longer traps remain out the greater capture corded were unusually rare. Combined with our confi- success one can expect for the same field effort. Mowat, dence in field species assignments, we feel certain that Shurgot & Poole (2000) found that a 2-week trap session the 139 samples included in our analysis are from increased track plate detection success by 18% over a 1- marten. In areas where hair samples can be visually week session for marten in coastal British Columbia. We sorted to exclude most non-target individuals, it may be checked hair traps weekly during our prototype testing cheaper to simply use microsatellite results to exclude and found that detection success increased through the the few non-target species remaining, rather than to run four weeks we trapped (G. Mowat, unpubl. data). a species test on all samples. Our population estimate was imprecise and perhaps The xylene wash did not appear to reduce the quali- biased because capture probabilities varied. Subsequent ty or quantity of DNA in a sample compared to clipped workers can probably increase precision by increasing samples, similar to chloroform (Foran et al. 1997a). trap density and minimizing capture variation. And, geno- However, unlike Foran et al. (1997a) we removed entire typing success can probably be increased by combining hairs from the xylene wash solution. We demonstrated samples from several glue patches at a site. Capture success that the probability of scoring a genotype for a sample varied among sessions (see Fig. 1), which is common is related to the number of roots available for the extrac- (White et al. 1982). Longer sessions, which offer similar tion. Similar results have been shown for marmots mean weather, and shorter study duration (i.e. fewer Marmota marmota and bears Ursus arctos (Gossens, capture sessions) may help minimize time variation in Waits & Taberlet 1998; G. Mowat, unpubl. data). We a hair sampling study where traps must remain set for achieved the greatest success with 20 hairs in the extrac- multiple days to ensure reasonable capture success. Lab tion, but the relationship in Figure 2 did not asymptote. failures may also add to variation in captures among The probability of scoring six loci increased from 0.78 sessions (Mowat & Strobeck 2000) which will be reduced to 0.88 when the number of hairs was increased from 15 as field and genetic analysis methods improve. to 20. Genotyping failures were numerous (41 of 180 Heterogeneity variation, defined by Otis et al. (1978) samples) and 38% of our samples had <10 hairs in as random variation among individuals, is also common them. Future workers may be able to increase genotyp- in field data (Conner & Labisky 1985, McCullough & ing success by combining hair from several glue patch- Hirth 1988, Corn & Conroy 1998), and was detected in es and putting more hairs in a sample. We avoided com- our data. Low trap density can cause heterogeneity; bining samples because we felt that it would increase the hence, Otis et al. (1978) recommended four traps per risk of mixing individuals in a sample. However, on three home range per session. This trap density would be dif- occasions we combined hairs from more than one pad ficult to achieve in large-scale studies such as ours, but none appeared to contain DNA from more than one but moving traps each session should increase effective individual (we expect mixed samples to show three or trap density and reduce heterogeneity (Pollock, Nichols, four alleles at one or more loci given the variability of Brownie & Hines 1990). Targeting preferred habitat the markers used). Nor did any of the 136 single sam- should also increase effective trap density. ples analyzed appear to be from different individuals. Marten may also have shown a positive behaviour Marten may not have entered previously visited traps ('trap-happy') response, probably because they received because bait was no longer present. Mixed samples a meat reward when they entered a trap. This form of may be more common in areas where the target species capture variation can cause negative bias in models that occurs at higher density. It may be possible to separate do not accommodate this variation explicitly, and behav-

© WILDLIFE BIOLOGY · 8:3 (2002) 207 iour models often require large sample sizes and deliv- nivores: American marten, fisher, lynx, and wolverine in the er poor precision (Boulanger & Krebs 1996). Given that western United States. USDA Forest Service GTR-RM-254, both time and heterogeneity variation are common in Rocky Mountain Forest and Range Experiment Station, Fort field data, it is advisable for field biologists to minimize Collins, Colorado, USA, pp 7-37. Conner, M.C. & Labisky, R.F. 1985: Evaluation of radioiso- behaviour response because no model can accommo- tope tagging for estimating abundance of raccoon popula- date all three forms of variation explicitly. Avoiding any tions. - Journal of Wildlife Management 49: 326-332. form of reward at a capture site should eliminate pos- Corn, J.L. & Conroy, M.J. 1998: Estimation of density of mon- itive behaviour response. However, it is not clear that gooses with capture-recapture and distance sampling. - marten and weasels could be enticed into hair traps Journal of Mammalogy 79: 1009-1015. with a scent bait such as those used for bears (Mowat Cowan, I. McT. & Guiguet, C.J. 1965: The mammals of & Strobeck 2000). Access to individual cells was often British Columbia, 3rd edition. - Handbook No. 11, British restricted to relatively short sections along one or two Columbia Provincial Museum, Victoria, British Columbia, roads on our study area, and so we had trouble moving Canada, 414 pp. traps more than 1 km between sessions. Short trap Davis, C. & Strobeck, C. 1998: Isolation, variability, and movement distances may have allowed previously cross-species amplification of polymorphic microsatellite loci in the family Mustelidae. - Molecular Ecology 7: 1776-1778. caught marten to successfully seek out our traps in fol- Foran, D.S., Crooks, K.C. & Minta S.C. 1997a: DNA-based lowing sessions. Greater movement of sites between ses- analysis of hair to identify species and individuals for pop- sions leading to more even trap coverage is preferred. ulation research and monitoring. - Wildlife Society Bulletin Geographic closure can lead to overestimates of pop- 25: 840-847. ulation size (White et al. 1982). In most cases, marten Foran, D.S., Minta, S.C. & Heinemeyer, K.S. 1997b: Species would have had to traverse alpine areas to cross the study identification from scat: an unambiguous genetic method. area boundary, and the movements of animals docu- - Wildlife Society Bulletin 25: 835-840. mented in this work suggest that this was rare. Also, the Gagneux, P., Boesch, C. & Woodruff, D.S. 1997: Microsatellite effect of closure bias declines as the ratio of study area scoring errors associated with noninvasive genotyping size to home range size increases (White et al. 1982). based on nuclear DNA amplified from shed hair. - Molecular Mean home-range size is likely to be less than 10 km2 Ecology 6: 861-868. Garshelis, D.L. 1992: Mark-recapture density estimation for in our area (Buskirk & Ruggiero 1994) suggesting that animals with large home ranges. - In: McCullough, D.R. the ratio of study area to home range size is at least 80 & Barrett, R.H. (Eds.); Wildlife 2001: populations. Elsevier times, which is much larger than many contemporary Press Science, London, pp. 1098-1111. mark-recapture studies (Garshelis 1992). Geographic clo- Gibilisco, C.J. 1994: Distributional dynamics of modern sure probably did not cause an important bias in our Martes in North America. - In: Buskirk, S.W., Harestad, A.S., study. Raphael, M.G. & Powell, R.A. (Eds.); Martens, Sables, and Fishers biology and conservation. Cornell University Press, Acknowledgements - funding for this inventory was provid- Ithaca, New York, USA, pp. 59-71. ed by Forest Renewal British Columbia and Slocan Forest Gossens, B., Waits, L.P. & Taberlet, P.T. 1998: Plucked hair Products, Slocan. We thank P. Cutts, M. Panian, C. Strobeck, samples as a source of DNA: reliability of dinucleotide C. Davis, D. Underwood and K.G. Poole. D. Fear, D. Stanley, microsatellite genotyping. - Molecular Ecology 7: 1237-1241. S. Petrovic, M. Petrovic, P. Cutts and C. Shurgot helped with Harris, R.B. 1986: Reliability of trend lines obtained from vari- fieldwork. M. Paradon, M. Watt, K. Stalker and C. Kyle assisted with DNA analysis. Val Johnson and Dave Pritchard did able counts. - Journal of Wildlife Management 50: 165-171. the GIS mapping and data extraction. Steve Minta and Kim Hawley, V.D. & Newby, F.E. 1957: Marten home ranges and Heinemeyer kindly shared their knowledge about hair removal population fluctuations in Montana. - Journal of Mammalogy of marten. Thanks to I. Thompson, C. Kyle, R. Weir and W. 38: 174-184. Duckworth for their reviews. McCullough, D.R. & Hirth, D.H. 1988: Evaluation of the Peter- sen-Lincoln estimator for a white-tailed deer population. - Journal of Wildlife Management 52: 534-544. McLaren, M.A., Thompson, I.D. & Baker, J.A. 1998: Selection References of vertebrate wildlife indicators for monitoring sustainable forest management in Ontario. - The Forestry Chronicle Boulanger, J. & Krebs, C.J. 1996: Robustness of capture-recap- 74: 241-247. ture estimators to sample biases in a cyclic snowshoe hare Mowat, G., Shurgot, C. & Poole, K.G. 2000: Using track plates population. - Journal of Applied Ecology 33: 530-542. and remote cameras to detect marten and short-tailed wea- Buskirk, S.W. & Ruggiero, L.F. 1994: Marten. - In: Ruggiero, sels in coastal cedar hemlock forests. - Northwestern Naturalist L.F., Aubry, K.B., Buskirk, S.W., Lyon, L.J. & Zielinski, 81: 113-121. W.J. (Eds.); The scientific basis for conserving forest car- Mowat, G. & Strobeck, C. 2000: Estimating population size 208 © WILDLIFE BIOLOGY · 8:3 (2002) of grizzly bears using hair capture, DNA profiling, and mark- 1993: Sexing free-ranging brown bears, Ursus arctos, using recapture analysis. - Journal of Wildlife Management 64: hairs collected in the field. - Molecular Ecology 2: 399-403. 183-193. Taberlet, P., Griffin, S., Goossens, B., Questiau, S., Manceau, Otis, D.L., Burnham, K.P., White, G.C. & Andersen, D.P. 1978: V., Escaravage, N., Waits, L.P. & Bouvet, J. 1996: Reliable Statistical inference from capture data on closed animal pop- genotyping of samples with very low DNA quantities using ulations. - Wildlife Monographs 62, 133 pp. PCR. - Nucleic Acids Research 24: 3189-3194. Payer, D.C. 1999: Influences of timber harvesting and trap- Thompson, I.D. & Colgan, P.W. 1987: Numerical responses ping on habitat selection and demographic characteristics of martens to a food shortage in northcentral Ontario. - Jour- of marten. - PhD dissertation, University of Maine, Orono, nal of Wildlife Management 51: 824-835. USA, 298 pp. Weckworth, R.P. & Hawley, V.D. 1962: Marten food habits Pollock, K.H., Nichols, J.D., Brownie, C. & Hines, J.E. 1990: and population fluctuations in Montana. - Journal of Wildlife Statistical inference for capture-recapture experiments. - Management 26: 55-74. Wildlife Monographs 107, 97 pp. White, G.C., Andersen, D.R., Burnham, K.P. & Otis, D.L. 1982: Poole, K.G., Mowat, G. & Fear, D.A. 2001: DNA-based Capture-recapture and removal methods for sampling population estimate for grizzly bears in northeastern British closed populations. - Los Alamos National Laboratory Columbia, Canada. - Wildlife Biology 7: 65-75. LA-8787-NERP, Los Alamos, New Mexico, USA., 235 pp. Raphael, M.G. 1994: Techniques for monitoring populations Woods, J.G., Paetkau, D., Lewis, D., McLellan, B.N., Proctor, of fishers and American Martens. - In: Buskirk, S.W., Ha- M. & Strobeck, C. 1999: Genetic tagging free ranging black restad, A.S., Raphael, M.G. & Powell, R.A. (Eds.); Martens, and brown bears. - Wildlife Society Bulletin 27: 616-627. Sables, and Fishers biology and conservation. Cornell Zielinski, W.J. & Kucera, T.E. (Eds.) 1995: American marten, University Press, Ithaca, New York, USA, pp. 224-240. fisher, lynx and wolverine: survey methods for their detec- Smallwood, K.S. & Schonewald, C. 1996: Scaling population tion. - USDA Forest Service, PSW-GTR-157, Pacific South- density and spatial pattern for terrestrial, mammalian car- west Research Station, Albany, California, USA, 163 pp. nivores. - Oecologia 105: 329-335. Zielinski, W.J. & Stauffer, H.B. 1996: Monitoring Martes popu- Snow, A.A. & Parker, P.G. (Eds.) 1998: Molecular markers for lations in California: survey design and power analysis. - population biology. - Ecology, Special feature 79: 359-425. Ecological Applications 6: 1254-1267. Soutiere, E.C. 1979: Effects of timber harvesting on marten Zielinski, W.J., Truex, R.L., Ogan, C.V. & Busse, K. 1997: in Maine. - Journal of Wildlife Management 43: 850-860. Detection surveys for fishers and martens in California, 1989- Strayer, D.L. 1999: Statistical power of presence-absence 1994: summary and interpretations. - In: Proulx, G., Bryant, data to detect population declines. - Conservation Biology H.M. & Woodard, P.M. (Eds.); Martes: taxonomy, ecolo- 13: 1034-1038. gy, techniques, and management. The Provincial Museum Taberlet, P., Mattock, H., Dubois-Paganon, C. & Bouvet, J. of Alberta, Edmonton, Alberta, Canada, pp. 372-392.

PAPER V

Garth Mowat1, Aurora Wildlife Research, RR 1, Site 14, Comp 8, Crescent Valley, British Columbia, V0G 1H0 CANADA and

Kim G. Poole, Aurora Wildlife Research, 2305 Annable Rd., Nelson, British Columbia, V1L 6K4 CANADA

Habitat Associations of Short-tailed Weasels in Winter

Abstract We sampled short-tailed weasel (Mustela erminea) presence in three study areas in the Selkirk and Purcell mountains of southeast British Columbia using hair removal traps and snow tracking.We extracted measures of forest cover and ecosystem type from digital resource databases to investigate habitat associations of weasels.We summarized use at the sample location (site scale) and in 500 m radius windows (home range scale) to investigate the effect of scale on our ﬁndings.Short-tailed weasels were detected in all forests surveyed.These covered the range from open dry Douglas-ﬁr (Pseudotsuga menziesii) forests to dense wet western redcedar (Thuja plicata)—western hemlock (Tsuga heterophylla) forests, and varied in structure from very recent clear-cuts to mature stands >300 yr in age.At the site scale, weasels were detected more often in younger stands with incomplete crown closure. At the larger scale, weasels used home ranges that had less mean crown closure, but no trend was observed across stand age.We conclude that weasels in montane forests prefer open habitats and that stand age is not functionally related to habitat preference. The greater use of younger stands at the site scale can be explained by the correlation between crown closure and age.Weasels occurred in all the ecosystems sampled, including sub-alpine parkland, but appeared to be more abundant in wetter ecosystems, perhaps due to the greater primary productivity of these ecosystems.

Introduction Study Areas Short-tailed weasels inhabit forested ecosystems Our Selkirk study area covered 797 km2 of the throughout the circumpolar region (King 1983) central Selkirk Mountains(Fig. 1).Three biogeo- and are found throughout British Columbia (BC), climatic zones occurred in this area: interior cedar- except for some of the smaller coastal islands hemlock (ICH), Engelmann spruce (Picea engel- (Cowan and Guiguet 1965).Short-tailed weasels mannii)-subalpine fir (Abies lasiocarpa)(ESSF) (hereafter, weasels) use early successional habitat and alpine tundra (AT) (Meidinger and Pojar or forest edges, including meadows, marshes, and 1991; Braumandl and Curran 1992).In this area, riparian woodlands (Simms 1979; King 1983; ICH forests are found below about 1400 m above Fagerstone 1987).Weasels are rare on the islands sea level, where western hemlock (Tsuga hetero- of coastal BC.In these wet environments they phylla), western redcedar (Thuja plicata), Doug- are found in early seral or edge habitats but not las-fir (Pseudotsuga menziesii) and spruce hybrids in continuous forest (Mowat et al. 2000; Reid et (Picea spp.) are the dominant overstory species. al. 2000). ESSF forests are found between roughly 1400 No data are available on weasel habitat use in and 2300 m, where these two species dominate montane areas in western Canada.We examined the canopy, although many early seral stands are weasel habitat associations in three study areas in dominated by lodgepole pine (Pinus contorta). southeastern BC to describe weasel use of forested Extensive areas of AT are found above about 2300 environments during winter.We examined habitat m.We did not sample in alpine tundra but we did use across stand age and crown closure because sample in subalpine parkland forests. these variables index succession and structure The Purcell study area was a 1512 km2 region which may have general relationships with habitat in the central Purcell Mountains (Fig. 1).The di- quality for weasels. versity of ecosystems was greater in the Purcell study area than the Selkirk study area.Along with the three zones described above, two dryer 1Author to whom correspondence should be addressed. zones occurred in this area: interior Douglas-fir E-mail: [email protected] (IDF) and montane spruce (MS).The IDF zone

28 Mowat and Poole Northwest Science, Vol. 79, No. 1, 2005

© 2005 by the Northwest Scientific Association. All rights reserved. Figure 1. The location of the Selkirk, Purcell and Creston study areas in southeast British Columbia. was dominated by Douglas-fir, lodgepole pine, ponderosa pine (Pinus ponderosa), western larch, and western larch (Larix occidentalis) stands. western white pine (P. monticola), and paper birch Engelmann spruce and lodgepole pine were the (Betula papyrifera) were more common than dominant canopy species in the MS ecosystem. cedar-hemlock stands in the Creston area. The Creston study area (650 km2) was located Extensive logging has occurred throughout in the southern Purcell Mountains and, like the all three areas, most within the past 30 years.Fur central Purcell area, was transitional between moist trappers were active in all three study areas during and dry climatic regions (Fig. 1).All of the Creston the winters we sampled but did not target weasels. area we sampled was located within ICH dry and Based on fur returns, six weasels were killed by ICH moist subzones.ESSF and, in some areas, AT trappers on the Purcell area during the winter zones occurred above the ICH zone, but were not we worked, compared to 115 American martens sampled.Due to extensive fires early in the 1900s, (Martes americana).We suspect a similarly low mixed seral stands of lodgepole pine, Douglas-fir, number of weasels were killed in the Selkirk and

Weasel Habitat Associations 29 Creston areas.We assume the number of weasels in 194 cells between 31 January and 26 March killed by fur trappers did not affect the outcome 2001; one cell was sampled twice in different of our analyses. locations.Each site was active for an average of 15.0 days (SD = 2.63) and was not visited again Methods until it was removed.Martens usually remove the bait from the trap during their first visit, and Field Methods another individual or species, such as a weasel, Weasel detection data were collected ancillary to was rarely detected after a marten had visited a other research objectives.The primary objective for site (Mowat and Paetkau 2002; Mowat, In Press). sampling the Selkirk area was to estimate marten In the Selkirk and Purcell areas sites were located density (Mowat and Paetkau 2002).The primary >1 km apart to minimize multiple detections of the objective in the Purcell area was to measure habitat same individual (Zielinski and Kucera 1995). associations of martens (Mowat, In Press), and While sampling for marten, Mowat and Paetkau in the Creston area our primary objective was to (2002) found that detection success for hair traps examine ungulate habitat selection (Poole and in the Selkirk Mountains was poorer for weasels Mowat, In Press). than for martens, and we tried to improve detec- In the Selkirk and Purcell study areas we tion success for weasels in the subsequent Purcell collected detection data for weasels at hair sites study.We attached a 4 x 4 x 15 cm block of wood using baited glue traps to remove hair from ani- to the trap tree parallel to the bait to enable easier mals.Hair traps were fashioned after the design access to the bait and glue patches for weasels. presented by Foran et al. (1997).Snow tracking We identified species detected in hair traps was also used to assess the presence of a target based on hair morphology and tracks at a site species within a 10 m radius of a hair trap, both (Mowat and Paetkau 2002).Weasels have short during the setting and removal of traps.Sampling white hair, while both red squirrels (Taimiasciurus was distributed systematically across both study hudsonicus) and northern flying squirrels (Glau- areas.The Selkirk study area was divided into 86 comys sabrinus) have short red-brown to gray 9-km2 square cells and each cell was sampled on hair with little difference between the length of four separate occasions for 14.6 days on average the guard and underfur.Fisher (Martes pennanti) (SD = 3.14) between 15 January and 14 March did not occur in the Selkirk or Purcell study areas 1997 to collect data for mark-recapture analysis. (Cowan and Guiguet 1965; Gibilisco 1994), and The trap was usually moved after each trapping during winter mink (Mustela vison) were rare in occasion.Traps were installed in places we felt the upland areas we sampled (Mowat and Paetkau would be most likely to catch martens within the 2002).Visual identification was not certain for cell, thus habitat was not sampled randomly or 57 samples from the Selkirk study area and 27 in proportion to availability.Sites were mostly set samples from the Purcell study; these were sent in patches of trees to maximize marten detection to a commercial genetics lab (Wildlife Genetics but a broad variety of sites from single trees in International, Nelson, Canada) for species testing. recently logged areas to forest >300 yr old were All 22 samples that were confirmed as weasel by sampled. genetic testing were short-tailed weasel; no long- We divided the Purcell study area into 5-km2 tailed weasel (M. frenata) were detected. cells and randomly located one sample site in each We counted tracks in snow along transects to cell.If there was a road within 500 m of a cell then measure weasel distribution in the Creston area we restricted the trap location to within 500 m of between 12 February and 29 March 2002.We ran the road.If no roads existed within 500 m of the transects directly upslope from the valley bottom. cell, then the location was randomly located.In We spaced 62 transects at 3-km intervals along practice many sites were too difficult or danger- 189 km of valley bottom.The upper elevation limit ous to access and these sites were moved closer to for transects (generally 1000 to 1200 m) was set accessible roads.Cells with no road within 500 m a priori to reach above the highest elevations were accessed using a helicopter.In that case we that we expected deer (Odocoileus spp.) and elk set the site in the closest landing location to the (Cervus elaphus) to use in late winter (K. Poole sample point below treeline.We set one hair trap and G. Mowat, unpublished data).We measured

30 Mowat and Poole distance along each transect using a hip-chain range scale.The 500 m window size was based and obtained global positioning system (GPS) on home range size estimates for male weasels locations at 100-m intervals along each transect. from the nearest study area (Lisgo 1999).We did For each 100-m segment, we recorded the number not calculate mean measures in 500 m windows of weasel tracks, but analyzed the data based on for Creston because tracks were sampled along the presence or absence of weasel tracks in each transects and segments within 500 m were auto- segment.Weasels were occasionally detected in correlated.Spearman rank correlation was used to sequential segments.Transects were surveyed 1 to measure the relationship between crown closure 8 d after snowfall (x– = 4.1, SD = 2.0). and stand age.

Statistical Analysis Results We derived overstory crown closure and stand After sites that detected martens were removed, age from BC Forest Cover data which maps weasels were detected at 34 of 133 sites (26%) forest structure and floristics based on overstory in the Selkirks, 19 of 97 sites (20%) in the Pur- species at a scale of 1:20,000.This mapping is cells, and 62 of 1,201 (5%) track segments near based on interpretation of 1:15,000 scale black Creston.Marten were also detected at 11 sites and white air photos and ground truthing plots. where weasels were detected in both the Selkirk Biogeoclimatic subzones were digitally mapped and Purcell study areas.Hair traps were not ef- at 1:50,000 scale (Meidinger and Pojar 1991; ficient at detecting weasels.In the Selkirks 14 of Braumandl and Curran 1992). 34 weasel detections were based on tracks near To analyze habitat associations we combined the site while 10 of 19 detections were based on stand age and overstory closure into categories that tracks in the Purcell study area.The proportion were comparable among study areas.The smallest of approaches within 10 m of a site that failed to sample size in any category for the Purcell and generate a weasel hair sample was similar between Selkirk data was 5, all other categories had n > the Selkirk and Purcell areas (41% versus 40%, 10. The mean sample size within categories at respectively; Mowat and Paetkau 2002).This sug- sites was 28.5 (SD = 13.7) and 32.1 (SD = 15.3) gests the step-up block added to the traps in the in 500 m windows.For Creston the lowest sample Purcells did not improve detection of weasels.Most in a category was 56 while the mean was 300 (SD detection failures resulted because the weasel did = 236).We divided the percent use by the percent not climb the tree to the trap. available to generate a relative measure of detec- We detected weasels in all forest types from tion success that is analogous to the forage ratio W open dry Douglas-fir stands to moist closed ce- (Krebs 1989).This simple analysis approach was dar-hemlock stands and from valley bottoms used because sampling was unlikely to be random to sub-alpine parkland.Weasels were detected or independent, and grouping the samples into more often at sites with low crown closure (Fig. categories would reduce the influence of random 2), and this observation was maintained at the error and outliers.We deleted all hair sites that broader scale of analysis (Fig. 3).Stands with detected only martens because the potential for <10% crown closure were used more than more trap interference was great given the much greater closed stands in all study areas.There was greater abundance of martens.Martens were detected at use of younger stands at the site scale (Fig. 4) over half the sites sampled hence removing these but not at the broader scale (Fig. 5).Stand age sites greatly reduced sample size and may have and crown closure at sites was correlated on all biased the sample against forested sites.Weasels three study area (r > 0.62, P < 0.001 for all study were detected at some sites where marten were areas) but less so in 500 m radius windows (r = also detected because hair traps detected both spe- 0.29, P = 0.004 for Purcells and not significant cies (rare) or weasels were detected by tracks in for Selkirk).Detection rates were similar in ICH snow (more common).No sites were deleted from and ESSF zones in the Selkirks (Fig. 6), greater the Creston track data.In the Selkirk and Purcell in the ESSF over the IDF and MS in the Purcells, areas we analyzed stand age and crown closure and greater in the ICH moist than the ICH dry at at two scales: points to measure use at the site Creston (Fig. 6).The ICH and ESSF had similar or foraging scale, and within a 500 m radius (78 rainfall in the Selkirks while the ESSF was wetter ha) circular window to measure use at the home than the IDF and MS in the Purcells (Braumundl

Weasel Habitat Associations 31 Figure 2. Habitat use of weasels at the site scale across increasing crown closure for the Selkirk, Purcell and Creston study areas.Use is indexed by percent use divided by percent available (W) and error bars are 95% conﬁdence intervals.

Figure 3. Habitat use of weasels when crown closure is averaged in a 500 m radius of the sample site (home range scale) for the Selkirk and Purcell study areas.Use is indexed by percent use divided by percent available (W) and error bars are 95% conﬁdence intervals.

32 Mowat and Poole Figure 4. Habitat use of weasels at the site scale across increasing stand age for the Selkirk, Purcell and Creston study areas.Use is indexed by percent use divided by percent available (W) and error bars are 95% conﬁdence intervals.

Figure 5. Habitat use of weasels when stand age is averaged in a 500 m radius of the sample site (home range scale) for the Selkirk and Purcell study areas.Use is indexed by percent use divided by percent available (W) and error bars are 95% conﬁdence intervals.

Weasel Habitat Associations 33 Figure 6. Habitat use of weasels among ecosystems for the Selkirk, Purcell and Creston study areas.Use is indexed by percent use divided by percent available (W) and error bars are 95% confidence intervals. and Curran 1992).Weasels were detected more and here too selection was most often for more often in wet ecosystems. open habitats.In dense coastal forests weasels are usually detected in small openings along Discussion roadways, watercourses, or in grassy cut-blocks (Mowat et al. 2000; Reid et al. 2000).All areas We detected short-tailed weasels in all forests we surveyed contained at least small openings surveyed, which varied in structure from very from roads, watercourses, or avalanche chutes, recent clear-cuts to mature stands >300 yr in age. and perhaps weasels made use of these small, Short-tailed weasels living in forested environ- unmapped openings when they resided in densely ments appear to prefer open areas across much forested landscapes.Use of riparian communities of their range (King 1983; Fagerstone 1987).Our at fine scales may be due to the presence of open results provide broad support for this general association because we sampled much larger areas habitats, which are interspersed among a gener- that included greater physical and ecological ally closed canopy matrix (Mowat et al. 2000; variation than other workers.Based on our analysis Reid et al. 2000). within 500 m of hair sites, weasels also appeared At the site scale, weasels were also detected to choose home ranges with lower mean overstory more often in younger stands in all three areas, than the surrounding landscape.Simms (1979) although use of the oldest stands was similar to noted that short-tailed weasels choose home ranges younger stands in the Selkirk study area.There with substantial amounts of early successional was no association across stand age within 500 m habitat.Short-tailed weasels were detected more windows suggesting that the finer scale trend was often in thinned mid-seral stands than un-thinned based on the correlation between stand age and stands in western Washington, possibly due to the crown closure, and not some more fundamental greater understory in the thinned stands (Wilson relationship with stand age.We suggest that the and Carey 1996).Lisgo (1999) and O’Donoghue fundamental habitat association is the preference et al. (2001) found weasels in all but the driest for open stands and landscapes, not the preference forest types in the boreal forests they sampled, for younger trees in the overstory.

34 Mowat and Poole What is the functional relationship between Our hair-traps provided weak measures of short-tailed weasels and open stands?Short-tailed occurrence for weasels because detection success weasels in the circumpolar north are small and prey was poor; this short-coming was partially overcome mainly on voles and mice (King 1989).Voles, other by track detection near hair traps.Variation in than Clethrionomys, are most abundant in open detection success among sites was likely random, grassy sites (Johnson and Johnson 1982), therefore, which may have reduced the power of our com- weasel use of open sites may be explained by the parisons but was unlikely to cause directional bias. abundance of their principal prey.In northern Al- Similarly, autocorrelation was unlikely to cause a berta, Lisgo (1999) found male weasels ate more major bias in the Creston data because the 1,201 squirrels than did females, and males used more track segments (sample units) were located on 62 closed canopy forest than females, who hunted systematic transects and weasels were detected mainly voles.This partitioning of resources may on 62 segments on 29 different transects. explain why weasels are often found in both open We did not sample availability representatively and closed environments in the same landscape, in any study area, but our samples covered the notwithstanding greater use of open environs.A range of possible variation for both crown closure second explanation for the preference for open and stand age in all three study areas.Further, the sites is based on competition.Martens are likely large size of our study areas and the systematic the main competitor of weasels in most of the nature of our samples conferred considerable in- forested ecosystems they occupy and martens dependence among samples. We sampled habitat appeared to be more abundant in all three study use among very different ecosystems, across large scales, using different methods, and our results areas.Detection success for marten was 69% in among study areas concur.The non-random nature the Selkirks, 63% in the Purcells, and 8% for the of our site locations negates our ability to make Creston area which was 1.6 to 3.2 times higher specific inferences about habitat selection within than for weasels.Red-backed voles (Clethrionomys study areas, such as building a predictive map of spp.) and red squirrels are generally more important habitat quality.We suggest that the non-random foods than other voles for martens and both prey nature of our sample locations did not bias detec- species occur in forested sites (Buskirk and Rug- tion across the range of our analysis variables, nor giero 1994).Martens avoided young, open canopy negate general comparisons across the range of stands in all three study areas (Mowat, In Press; variation sampled. Poole and Mowat, unpubl. data) and competition between martens and weasels may exclude the Acknowledgements smaller short-tailed weasel from closed canopy habitat. Funding for this work was provided by Forest Renewal British Columbia, Slocan Forest Products, Weasels occurred in all the ecosystems Tembec Inc., and the Arrow IFPA.We thank K. sampled, but like martens (Mowat, In Press), Stuart-Smith, P. Frasca, G. Richardson, R. Ser- they appeared to be more abundant in the wetter rouya, P. Cutts, M. Panian, C. Strobeck, C. Davis ecosystems in a study area.Several authors have and D. Underwood for logistical and administra- noted that weasels prefer moist or riparian sites tive support.D. Stanley, D. Fear, M.-A. Beaucher, within their home ranges, even in very wet envi- S. Petrovic, M. Petrovic, P. Cutts and C. Shurgot ronments (Doyle 1990; Lisgo 1999; Mowat et al. helped with fieldwork.D. Paetkau, M. Paradon, 2000; Reid et al. 2000), but this does not explain M. Watt, K. Stalker, C. Kyle, and J. Weldon as- our observations.Our ecosystems were hundreds sisted with DNA analysis.M. Buchanan and D. of km2 in size and we suggest the greater abun- Pritchard did GIS mapping and data extraction. dance of weasels in wetter ecosystems was due S. Minta and K. Heinemeyer kindly shared their to greater ecosystem productivity, not behavioral knowledge about hair removal of martens.Thanks preferences for structure, floristics, or the need to to V. Nams, I. Adams and R. Weir for their reviews drink (Fagerstone 1987). of the manuscript.

Weasel Habitat Associations 35 Literature Cited Krebs, C. J. 1989. Ecological Methodology. Harper and Row, New York, USA. Braumandl, T. F., and M. P. Curran, editors. 1992. A field guide Lisgo, K. A. 1999. Ecology of the short-tailed weasel in mixed- for site identification for the Nelson Forest Region. wood boreal forest of Alberta. M.Sc. thesis, University Land Management Handbook No. 20, B.C. Ministry of British Columbia, Vancouver, Canada. of Forests, Victoria, British Columbia, Canada. Meidinger, D., and J. Pojar. 1991. Ecosystems of British Buskirk, S. W., and L. F. Ruggiero. 1994. Marten. Pages 7-37 Columbia. Special Report Series 6, B.C. Ministry In L. F. Ruggiero, K. B. Aubry, S. W. Buskirk, L. J. of Forests, Research Branch, Victoria, B.C., Canada. Lyon, and W. J. Zielinski (editors). The scientific basis Mowat, G., C. Shurgot, and K. G. Poole. 2000. Using track for conserving forest carnivores: American marten, plates and remote cameras to detect marten and short- fisher, lynx, and wolverine in the western United tailed weasels in coastal cedar hemlock forests. North- States. USDA Forest Service GTR-RM-254. Rocky western Naturalist 81:113-121. Mountain Forest and Range Experiment Station, Fort Mowat, G., and D. Paetkau. 2002. Estimating marten popu- Collins, Colorado, USA. lation size using hair capture and genetic tagging in Cowan, I. M., and C. J. Guiguet. 1965. The mammals of southeast British Columbia. Wildlife Biology 8:201- British Columbia, 3rd edition. Handbook No. 11, Brit- 209. ish Columbia Provincial Museum, Victoria, British Mowat, G. In press. Winter habitat associations of American Columbia, Canada. martens Martes americana in interior wet-belt forests. Doyle, A. T. 1990. Use of riparian and upland habitats by small Wildlife Biology. mammals. Journal of Mammalogy 71:14-23. O’Donoghue, M., S. Boutin, E. J. Hofer, and R. Boonstra. Fagerstone, K. A. 1987. Black-footed ferret, long-tailed 2001. Other mammalian predators. Pages 324-338 In C. J. Krebs, S. Boutin, and R. Boonstra (editors). weasel, short-tailed weasel, and least weasel. Pages Ecosystem dynamics of the boreal forest: the Kluane 548-574 In M. Novak, J. A. Baker, M. E. Obbard, and project. University of Oxford Press, New York. B. Malloch (editors). Wild furbearer conservation and Poole, K. G. and G. Mowat.In press.Winter habitat relation- management in North America. Ontario Ministry of ships of deer and elk in the temperate interior moun- Natural Resources, Toronto, Ontario, Canada. tains of British Columbia. Wildlife Society Bulletin. Foran, D. S., K. C. Crooks, and S. C. Minta. 1997. DNA-based Reid, D. G., F. L. Waterhouse, P. E. F. Buck, A. E. Derocher, analysis of hair to identify species and individuals for R. Bettner, and C. D. French. 2000. Inventory of Queen population research and monitoring. Wildlife Society Charlotte Islands ermine. Pages 393-406 In L. Darling Bulletin 25:840-847. (editor). At risk: proceedings of a conference on the Gibilisco, C. J. 1994. Distributional dynamics of modern Mar- biology and management of species and habitats at tes. Pages 283-296 In S. W. Buskirk, A. S. Harestad, risk. B.C. Ministry of Environment, Lands and Parks, M. G. Raphael, and R. A. Powell (editors). Martens, Victoria, Vol. 1. sables, and fishers; biology and conservation. Cornell Simms, D. A. 1979. Studies of an ermine population in southern University Press, Ithaca, New York. Ontario. Canadian Journal of Zoology 57:824-832. Johnson, M. L., and S. Johnson. 1982. Voles. Pages 326-354 Wilson, T. D., and A. R. Carey. 1996. Observations of weasels In J. A. Chapman, and G. A. Feldhamer (editors). in second-growth Douglas-fir forests in the Puget Wild mammals of North America: biology, manage- Trough, Washington. Northwestern Naturalist 77:35- ment, and economics. John Hopkins University Press, 39. Baltimore, New Jersey. Zielinski, W. J., and T. E. Kucera, editors. 1995. American King, C. M. 1983. Mustela erminea. Mammalian Species marten, fisher, lynx and wolverine: survey methods No. 195. for their detection. USDA Forest Service, PSW-GTR- King, C. M. 1989. The natural history of weasels. Cornell 157, Pacific Southwest Research Station, Albany, University Press, Ithaca, New York. California, USA.

Received 29 July 2004 Accepted for publication 6 March 2005

36 Mowat and Poole

PAPER VI

Winter habitat associations of American martens Martes americana in interior wet-belt forests

Garth Mowat

Mowat, G. 2006: Winter habitat associations of American martens Martes americana in interior wet-belt forests. - Wildl. Biol. 12: 51-61.

I systematically sampled American marten Martes americana presence in two large study areas in the Selkirk and Purcell Mountains of southwest Canada using hair removal traps and tracks in snow. Both study areas were mostly forested and contained a broad cross-section of stand ages including abundant early seral and mature forest. I extracted measures of forest structure and dominant tree species, climax ecosystem types and human use from digital resource databases and used multiple logistic regression to model habitat selection of martens. I summarized data in windows of 100 m to 10 km in radius around each sample location to investigate the effect of varying data resolution on habitat selection. Marten detection at hair sites was positively related to temperature and trap duration and negatively related to snowfall while the trap was set. Martens were detected in all habitats sampled including recently logged areas, regenerating stands, dry Douglas-fir Pseudosuga menziesii forest and subalpine parkland. Overall selection was mildly greater using mean habitat values in 100 m and 2 km radius windows for both study areas. Martens selected for greater crown closure and older stands at the finer resolution; no selection for forest structure was detected at the larger resolution except that martens selected against increased overstory heterogeneity as measured by the standard deviation of crown closure (within the window). Martens preferred coniferous stands over deciduous dominated stands and were more abundant in wetter than in dryer ecosystems. Selection for ecosystems and stand types was stronger in the larger window size. At the intensity sampled in this study, neither road density nor logging appeared to affect marten habitat selection when I accounted for variation in ecosystems and stand structure. This study examined habitat selection at relatively coarse scales; stronger associations with forest structure may be expected at finer scales. In addition, roads or logging may influence habitat selection below the scale of my analysis.

Key words: British Columbia, Canada, habitat selection, Martes americana, spatial scale

Garth Mowat*, Aurora Wildlife Research, RR 1, Site 14, Comp 8, Crescent Valley, BC, V0G 1H0, Canada - email: [email protected]

* Present address: BC Ministry of Environment, Suite 401, 333 Victoria St. Nelson BC, V1L 4K3, Canada

Received 29 December 2003, accepted 11 October 2004

Associate Editor: Piero Genovesi

© WILDLIFE BIOLOGY · 12:1 (2006) 51 Martens Martes americana are found in coniferous and tribution of martens. I was also interested in the relation- mixed coniferous forests throughout North America and ship between distribution and more static variables such are typically considered denizens of mature forest stands as climax ecosystems which index broad-scale produc- (Strickland & Douglas 1987). Buskirk & Powell (1994) tivity. I selected independent variables based on previ- suggested that the physical structure of the stand is more ously established, or hypothesized, habitat relationships important than tree species composition, although other and chose to do my analysis at two levels of resolution authors have suggested that spruce and fir stands are because I felt that individual behaviour would influence favoured by martens (Clark et al. 1987, Buskirk & Powell selection at fine resolution, and variation in density would 1994). Huggard (1999) found that martens used wetter be more influential at coarser resolution. My goal was ecosystems more than xeric ones in high elevation Engel- to test the association between various habitat measures mann spruce Picea engelmannii-subalpine fir Abies lasio- and marten occurrence and to explain my results in terms carpa forest, and Buskirk & Ruggiero (1994) suggest- of their influence on fitness or density. ed that martens avoid dryer forests and ecosystems. Martens avoid openings, especially in winter, and select stands with higher cover (Clark et al. 1987). Structure Study areas on the ground such as woody debris provides access to the sub-nivean environment and sheltered resting places The Selkirk study area covered 797 km2 of the central (Buskirk & Ruggiero 1994, Coffin et al. 1997). Horizontal Selkirk Mountains in southeastern British Columbia structure may also provide increased habitat quality for (BC). Biogeoclimatic (BEC) zones, each with distinct voles, the principal prey of martens in most areas (Clark vegetation and soils, assume climate is the principle et al. 1987, Buskirk & Ruggiero 1994). The direct effect factor influencing ecosystem development (Meidinger of human alterations to forests on marten habitat quali- & Pojar 1991). BEC mapping is intended to map var- ty, beyond removal of the canopy, is unclear; the frag- ious scales of climax ecosystems which are assumed mentation of forest stands caused by settlement, logging to be relatively stable. Different plant species associa- and road building appears to have a negative effect on tions may dominate various seral stages but the plant martens (Chapin et al. 1998, Hargis et al. 1999). community will succeed to a stable climax ecosystem. What is the causal relationship between marten habi- Three BEC zones occurred in the Selkirk study area: tat selection and fitness? Overstory species and stand interior cedar-hemlock (ICH), Engelmann spruce-sub- age have no clear relationship with fitness. Stand struc- alpine fir (ESSF) and alpine tundra (AT). These zones ture may directly affect predation risk and hence fitness are divided into subzones based on variation in rain- (Buskirk & Powell 1994). Wetter ecosystems have great- fall and growing season. In this area, cedar-hemlock er primary productivity which directly influences food forests occur < 1,400 m a.s.l., where western hemlock abundance. All the above variables may be correlated Tsuga heterophyla, western redcedar Thuja plicata, to prey abundance. Human impacts to forests affect their Douglas-fir Pseudosuga menziesii and spruce hybrids structure and floristics and therefore predation risk and Picea spp. are the dominant overstory species. Engel- probably prey abundance; local human influences on mann spruce-subalpine fir forests are found between primary productivity are less likely. Additionally, the roughly 1,400 and 2,300 m a.s.l., where Engelmann scale or resolution of analysis may reflect the influence spruce and subalpine fir dominate the canopy, although of factors that operate at different scales (May 1994, many early seral stands are dominated by lodgepole pine Powell 1994). For example, fine scales of analysis may Pinus contorta. Extensive areas of alpine tundra are measure behavioural decisions related to forage prefer- found above about 2,300 m. We did not sample in the ence or predator avoidance by individuals, while coarse alpine tundra, but we did sample in the treed portion of scales of analysis likely measure differences in species the subalpine zone. density among sample units. The effect of scale can be The Purcell study area was a 1,059 km2 region in the investigated by examining habitat relationships across a Central Purcell Mountains of southeast BC. The diver- range of scales (Chapin et al. 1997, Meyer et al. 1998) or sity of ecosystems was greater in the Purcell study area selecting scales a priori (Bowers et al. 1996, Pedlar et al. than the Selkirk study area. Along with the three zones 1997). In this study I selected scales based on the strength described above, two dryer zones occurred in this area: of the relationship between habitat and response vari- interior Douglas-fir (IDF) and montane spruce (MS). ables (Nams et al. 2005). The IDF zone was dominated by Douglas-fir, lodgepole I wanted to know if broad-scale anthropogenic changes pine, and western larch Larix occidentalis stands. Engel- in forest structure and floristics were related to the dis- mann spruce and lodgepole pine were the dominant can-

52 © WILDLIFE BIOLOGY · 12:1 (2006) opy species in the MS ecosystem. The north and west of the road. If there was no road within 500 m of the cell, side of each study area were moist environments that then the location was randomly located. In practice many graded to much dryer forests in the southeast of the study sites were too difficult or dangerous to access, and these area. sites were moved closer to accessible roads. Cells with Logging occurred throughout both areas, mostly with- no road within 500 m were accessed using a helicopter. in the last 30 years, however significant amounts of mature In that case I placed the site 50-100 m from the closest forest remained in both areas. Road networks were exten- landing site to the sample point. In both study areas, I sive, all main drainages had all-weather roads and only attempted to locate sites > 1 km apart to minimize mul- a couple of side drainages in each area contained no road. tiple detections of the same individual (Zielinski & Other human impacts included back-country recreation Kucera 1995). I set one hair trap in each of 194 cells (including snowmobiling and alpine skiing), mining and between 31 January and 26 March 2001; one cell was hydroelectric lines. There were < 10 year-round human sampled twice in different locations. Each site was active residents living on either study area. Private land with for 15.0 days (SD = 2.63) and not visited again until it rural housing bordered the lower elevation, eastern edge was removed. Martens usually removed the bait from of both study areas. the trap during their first visit making it unlikely the trap would detect another individual or species. In contrast, short-tailed weasels Mustela erminea were sometimes Methods unable to remove the bait and so martens may be detected at a site which has already detected weasels. Because Field methods weasels often removed the bait and deterred a marten I collected detection data using baited glue traps to from entering a trap, I removed sites that detected a wea- remove hair from animals. I assessed the presence of sel from the database but not those where both marten martens in a 100-m radius of a capture site, using snow and weasel were detected. tracking, both during the setting and removal of sites. I identified species detected in hair traps based on hair Hair traps were fashioned after the design described by morphology and tracks at the site (Mowat & Paetkau Foran et al. (1997). Sampling was distributed systemat- 2002). Weasels have short white hair during winter, ically across both study areas. The Selkirk study area whereas both red squirrels Taimiasciurus hudsonicus and was divided into 9-km2 square cells and each cell was flying squirrels Glaucomys sabrinus have short red-brown sampled on four separate occasions for 14.6 days on to gray hair with little difference between the length of average (SD = 3.14) between 15 January and 14 March the guard and underfur. Fishers Martes pennanti did not 1997. Trap sites were usually moved after each trapping occur in either study area (Cowan & Guiguet 1965, occasion (Mowat & Paetkau 2002), and repeated trap- Gibilisco 1994) and, during winter, minks Mustela vison ping events at the same site were not used in habitat anal- were rare in the upland areas we sampled (Mowat & yses. Because the primary objective for the Selkirk study Paetkau 2002). Visual identification was not certain for was to estimate marten population size, field personnel 57 samples from the Selkirk study area and 27 samples were instructed to install sites where they felt it would be from the Purcell study; these I sent to a commercial most likely to detect martens within the cell, to increase genetics lab (Wildlife Genetics International, Nelson, capture success. Sites were mostly set in patches of BC, Canada) for species testing. Of these samples, six mature forest, thus habitat was not sampled randomly visually classified as weasel were marten (18%), and two or in proportion to availability. Further details on the samples classified as marten were weasel (4%). When methods for the Selkirk area are described in Mowat & the classification was uncertain, all weasel and all mar- Paetkau (2002). ten samples were genetically tested, therefore errors in The primary study goal in the Purcell study area was species assignment were likely to be few. All 22 wea- to examine the influence of commercial forestry on mar- sel samples were short-tailed weasel; no long-tailed wea- ten distribution and greater effort was made to locate sels M. frenata were detected. sample sites randomly. I divided the Purcell study area I interviewed fur trappers working in both areas and into 265 5-km2 square cells and used a GIS to random- asked them how many and where they trapped marten. ly locate one sample site in each cell. Cells located along Twelve martens of an estimated population of 225 (Mo- the study area boundary which were < 5 km2, or those wat & Paetkau 2002) were taken on the Selkirk study which were predominantly glacier or alpine tundra were area during the November-December (1996) previous not sampled. If there was a road in or within < 500 m of to my work. On the Purcell study area, 115 martens and a cell then I restricted the trap location to within 500 m six weasels (species unknown) were trapped on the study

© WILDLIFE BIOLOGY · 12:1 (2006) 53 area during November-December 2000, 2-3 months pre- discrete sizes around sample sites (Table 1). Window vious to my fieldwork. I felt that the number of marten sizes varied from 0 m (point extraction) to 10 km in radi- killed on the Selkirk study area were too few to measur- us. The systematic distribution of sample points, and the ably affect the outcome of habitat analyses. For the fact that habitat values were averaged across windows Purcell area, I scored trapping effort by fur trappers as means that habitat measures, though not strictly random- a binary variable. Any cell where a trapper killed a mar- ly chosen, probably reflect all habitat available across ten was considered trapped, all other cells were consid- the study area when windows are several km in radius ered untrapped. Kyle et al. (2003) genotyped 60 sam- (Nams et al. 2005), especially in the Purcell area. I sum- ples taken from different sites across the two month sam- marized habitat selection across stand age by dividing pling period and all 60 samples were different individ- the percent use of a category by the percent available; uals. This suggests that considerably more than 60 mar- this relative measure of selection is termed W by Manly tens were present on the Purcell area during our work. et al. (1993). I used logistic regression to test the relationship be- Statistical analysis tween various habitat variables and the probability of I derived habitat information from GIS databases. BC detecting a marten. I used the logit function and tested Forest Cover data map forest structure and floristics assumptions regarding residuals. I tested for relation- based on overstory species at a scale of 1:20,000. This ships among variables using Spearman rank correlation mapping system differs from BEC mapping because the analysis to avoid including highly correlated variables primary focus is to map stands based on stand type (over- in the same model. I compared model fit using Akaike’s story species associations) and current vegetation. These Information Criteria (AIC) values and the rescaled r2, stands may change rapidly across time while the eco- and classification accuracy using c. AIC is a relative systems mapped in the BEC mapping are assumed to be measure of model fit which is discounted as the number static. Seral stage changes within BEC ecosystems often of parameters in the model increases; reductions in AIC involve a change in stand type hence stand types mapped of two or more are considered significant improvements in Forest Cover mapping are loosely nested within BEC in model fit and parsimony (Burnham & Andersen 1998). zones and subzones. Forest Cover mapping is based on An increase in AIC suggests the model is less parsimoni- interpretation of 1:15,000 scale black and white air pho- ous than the global model. AIC weights are also given tos and ground plots for truthing (Resources Inventory for comparisons of multiple models. The rescaled r2 mea- Branch 1995). Structural values are averages for visu- sures the proportional reduction in the log-likelihood ally mapped polygons. Polygons are usually > 2 ha in measure and is analogous to the familiar r2 of linear size and average in the tens of hectares, depending on regression (Menard 1995). Overall classification accu- forest complexity. racy of the data in the model is measured by c; c of 1.0 I investigated the effect of habitat resolution by cal- means that all observations in the data set were correct- culating mean habitat values in circular windows of eight ly classified by the model. I included variables that indexed parameters that may affect detection success such as temperature, precipita- Table 1. Habitat measures and sources of data. Areas were transformed to a proportion of the window area when data were rescaled. tion, trapping effort and the length of time a trap was All habitat information came from public access databases and set, to reduce the influence of measurement errors. I measures were extracted using a GIS. BEC = biogeoclimatic zones measured the effect of resolution of habitat data by com- which are an ecosystem classification system designed to map climax ecosystems regardless of current vegetation cover (Meidinger paring the fit of a global model, which included forest & Pojar 1991). structure, stand types and ecosystems as explanatory variables, and variables that controlled for measurement Habitat Variables Units Source error, for each window size. I used SASTM Version 8 Ecosystem type (up to 4) % in window BC digital BEC mapping Stand age Mean years BC Forest Cover Mapping (SAS Institute Inc., Cary, NC, USA) software for micro- Crown closure Mean percent BC Forest Cover Mapping computers for data manipulation and analysis. Crown closure Standard deviation BC Forest Cover Mapping Douglas-fir and larch Mean % in overstory BC Forest Cover Mapping Spruce and subalpine fir Mean % in overstory BC Forest Cover Mapping Cedar and hemlock Mean % in overstory BC Forest Cover Mapping Results Pine Mean % in overstory BC Forest Cover Mapping Deciduous Mean % in overstory BC Forest Cover Mapping For marten habitat analysis in the Selkirk area I removed Length of roads Meters BC Forest Cover Mapping 1) one site for which I could not identify the hair sam- Logged land Proportion of area BC Forest Cover Mapping ple, 2) 19 sites that detected weasels in the hair trap, and

54 © WILDLIFE BIOLOGY · 12:1 (2006) Table 2. Multivariate logistic regression analysis of marten habitat selection in the Selkirk Mountains with habitat data summarized in 100 m radius circular windows. All models include the variables temperature, snowfall and trap duration to account for their influence on

detection success. i = AIC weight.

Model no Variables in model AIC Parameters i Corrected R c 1 Subzones, stand types & structure (global model) 306.0 1.1 14 0.218 0.16 0.71 2 Stand types and structure 313.1 8.2 11 0.006 0.10 0.66 3 Subzones and structure 311.2 6.3 9 0.016 0.09 0.65 4 Subzones and stand types 305.6 0.7 12 0.266 0.15 0.70 5 Subzones, stand types & crown closure 307.5 2.6 13 0.103 0.15 0.70 6 Subzones, stand types & stand age 304.9 0.6 13 0.378 0.16 0.71 7 Covariates only 311.8 6.9 4 0.012 0.03 0.58 Human use variables 8 Model 6 & SE of crown closure 306.3 1.4 14 0.16 0.72 9 Model 6 & roads 305.8 0.9 14 0.17 0.72 10 Model 6 & logging < 10 years old 305.9 1.0 14 0.17 0.71 11 Model 6 & logging < 20 years old 306.4 1.5 14 0.16 0.71 12 Model 6 & all logging 306.5 1.6 14 0.16 0.71 Measurement errors (covariates) 13 Model 6 minus duration 303.6 -1.3 12 0.16 0.71 14 Model 6 minus temperature 305.7 0.8 12 0.15 0.70 15 Model 6 minus snow 305.2 0.3 12 0.15 0.70

3) 33 sites that had been previously hair-trapped. Of the structure were all related to marten habitat selection at remaining 257 sites, 177 (69%) detected marten based the 100 m window size as can be seen by the increase on hair traps (166) and tracks near the site (11). Marten in AIC and decrease in r2 when these variables were detection rate was positively related to mean tempera- removed (models 2-5) and compared to the global mod- ture and negatively related to snowfall, but weakly relat- el (model 1) in Table 2 . Only crown closure was not ed to trap duration (compare models 13-15 to model 6 related to marten habitat selection at this resolution as

in Table 2 ). demonstrated by the reduction in AIC in model 6 with

¡¢£¡¤¥ ¥ ¦ ¥ §¡ ¨© ¦ © ¦ ¡© § ¨§¡ In the Purcell area I removed 15 sites which detected weasels; martens were detected at 117 of 187 sites (63%) AIC weight for model 6 than for model 1. Stand types based on hair traps (105) and snow tracks (12). Marten had similar fit to BEC subzones (compare models 2 and detection was related to trapping effort by fur trappers, 3 in Table 2). At the larger window size, BEC subzone mean temperature and trap duration, while there was lit- contributed less to model fit than stand type, and both tle support for the influence of snowfall on detection suc- crown closure and stand age contributed little to the rela- cess (compare models 12-15 to model 1 in Table 4). tionship (Table 3). None of the human use variables Marten hair traps performed equally well in the Purcell and Selkirk study areas. The proportion of hair trap approaches (based on negative hair trap results and tracks in snow near the trap) that failed to detect a marten was similar between study areas (8% in the Selkirks vs 3% 0.3 ) 2 in the Purcells; Mowat & Paetkau 2002). Temperatures Purcell were at times much colder during the Purcell work (as Selkirk low as -28°C), yet the glue patches worked well. 0.2 The relationship between explanatory variables and (r FIT MODEL marten detection varied little with resolution (the global model which included all independent variables was 0.1 used for this analysis), although it was greatest when 0 100 200 500 1000 2000 5000 10000 data were summarized for 100 m and 2,000 m radius WINDOW RADIUS (m) windows (Fig. 1). Therefore I did all further analyses at these two window sizes to attempt to separate habitat Figure 1. Relationship between marten presence at hair trap sites and related behavioural preferences of individuals and dif- mean habitat values across window size for the Selkirk and Purcell study areas. Model fit is measured by the corrected r2. All descriptive ferences in species density across the study area. habitat variables are included in each model; i.e. the global model In the Selkirk mountains, ecosystem, stand type and from Tables 2-5.

© WILDLIFE BIOLOGY · 12:1 (2006) 55 Table 3. Multivariate logistic regression analysis of marten habitat selection in the Selkirk Mountains with habitat data summarized in 2,000 m radius circular windows. All models include the variables temperature, snowfall and trap duration to account for their influence

on detection success. i = AIC weight.

$ 2 !" #

Model no Variables in model AIC Parameters i Corrected R c 1 Subzones, stand types & structure (global model) 305.2 3.6 14 0.079 0.17 0.71 2 Stand types and structure 306.0 4.4 11 0.053 0.13 0.69 3 Subzones and structure 310.3 8.7 9 0.006 0.09 0.65 4 Subzones & stand types 301.6 0.6 12 0.477 0.16 0.71 5 Subzones, stand types & crown closure 303.6 2.6 13 0.176 0.16 0.71 6 Subzones, stand types & stand age 303.3 1.7 13 0.204 0.17 0.71 7 Covariates only 310.8 9.2 4 0.005 0.03 0.59 Human use variables 8 Model 4 & SE of crown closure 303.6 2.6 13 0.16 0.71 9 Model 4 & roads 300.5 -1.1 13 0.18 0.72 10 Model 4 & logging < 10 years old 302.3 0.7 13 0.17 0.71 improved model fit above the best fit model at either closure were related to selection (see Table 4) howev- resolution (see Tables 2 and 3). There was a mild im- er, at the larger window size structure only mildly im-

provement in fit when roads were added to the model at proved model fit (see Table 5). I retained the structure

¤¥¦¢ ¥ §¨© § ¨ ¢ ¢ ¢ ¤ £¢ ¡¢ £ variables to test the human use variables at the larger the relationship with roads was positive and likely relat- window size because I wanted to account for the varia- ed to greater selection for wetter ecosystems or lower tion in structure (see Tables 4 and 5). Like the Selkirk elevations (where roads are often located) than selection data, these data did not support the hypothesis that roads for roads by martens. or logging negatively affected marten presence when The most striking difference for the Purcell Mountains other habitat factors were accounted for (see Tables 4 was that model fit (r2) was greater than for the Selkirk and 5). The addition of the standard error of crown clo- Mountains. Structure, stand type and BEC subzone were sure to the global model increased model fit (see Table all related to marten preference. There was little differ- 5: model 8) which suggests that greater fragmentation ence between the improvement in fit generated by stand of the overstory leads to a reduction in marten presence type or BEC subzone (Tables 4 and 5), probably because even when differences in overstory are accounted for. these variables were strongly correlated with one anoth- Although the difference in model fit between the two er. At the smaller window size both stand age and crown resolutions I analyzed was small (see Fig. 1), the pre-

Table 4. Multivariate logistic regression analysis of marten habitat selection in the Purcell Mountains with habitat data summarized in100 m radius circular windows. All models include the variables temperature, snowfall, trapping effort and trap duration to account for their

influence on detection success. i = AIC weight.

$ 2 !" # Model no Variables in model AIC Parameters i Corrected R c 1 Subzones, stand types & structure (global model) 227.6 0.6 14 0.133 0.31 0.79 2 Stand types and structure 226.1 -1.5 11 0.282 0.28 0.77 3 Subzones and structure 225.0 -2.6 10 0.488 0.28 0.77 4 Subzones and stand types 240.1 12.5 12 0.000 0.21 0.73 5 Subzones, stand types & crown closure 231.4 3.8 13 0.020 0.27 0.76 6 Subzones, stand types & stand age 228.7 1.1 13 0.077 0.29 0.77 7 Covariates only 242.9 15.3 5 0.000 0.10 0.67 Human use variables 8 Model 1 & SE of crown closure 228.5 0.9 15 0.31 0.79 9 Model 1 & roads 226.1 -1.5 15 0.33 0.80 10 Model 1 & logging < 10 years old 228.9 1.3 15 0.31 0.79 11 Model 1 & logging < 20 years old 229.5 1.9 15 0.31 0.79 12 Model 1 & all logging 229.5 1.9 15 0.31 0.79 Measurement errors (covariates) 13 Model 1 minus trapping 230.3 2.7 13 0.28 0.77 14 Model 1 minus duration 230.2 2.6 13 0.28 0.77 15 Model 1 minus temperature 231.1 3.5 13 0.28 0.77 16 Model 1 minus snow 226.5 -1.1 13 0.30 0.79

56 © WILDLIFE BIOLOGY · 12:1 (2006) Table 5. Multivariate logistic regression analysis of marten habitat selection in the Purcell Mountains with habitat data summarized in 2,000 m radius circular windows. All models include the variables temperature, snowfall, trapping effort and trap duration to account for their

influence on detection success. i = AIC weight.

¥ 2 ¢£ ¤

Model no Variables in model AIC ¡ Parameters i Corrected R c 1 Subzones, stand types & structure (global model) 228.9 0 14 0.195 0.30 0.79 2 Stand types and structure 231.7 2.8 11 0.048 0.25 0.76 3 Subzones and structure 233.8 4.9 10 0.017 0.22 0.74 4 Subzones & stand types 229.1 0.2 12 0.177 0.28 0.77 5 Subzones, stand types & crown closure 230.6 1.7 14 0.083 0.28 0.77 6 Subzones, stand types & stand age 227.1 -1.8 13 0.480 0.30 0.78 7 Covariates only 242.9 14 5 0.000 0.10 0.68 Human use variables 8 Model 1 & SD of crown closure 224.9 -4 15 0.33 0.80 9 Model 1 & roads 229.8 0.9 15 0.31 0.79 10 Model 1 & logging < 10 years old 230.7 1.8 15 0.30 0.79 11 Model 1 & logging < 20 years old 230.9 2 15 0.30 0.79 12 Model 1 & all logging 230.9 2 15 0.30 0.79 dictive power of each group of independent variables relatively high and even, compared to the Purcells, where changed across window sizes. Selection for stand struc- there was large variation in rainfall within the study area ture was greater at the smaller window size in both areas (Fig. 2). The greatest avoidance was for stands < 10 years while selection for stand types and ecosystems was great- of age in the Purcells; this was not seen in the Selkirks er at the larger window size. because few sites were set in stands < 10 years of age. In In the Selkirk area, martens preferred spruce-fir, cedar- summary, martens were detected more often in wetter hemlock and pine stands to Douglas-fir-larch and decid- ecosystems, and within ecosystems they preferred wetter uous dominated stands. Selection for the ICH and ESSF coniferous stands. At the finer resolution, they avoided zones was similar. In the Purcell Mountains, martens younger stands with less overstory. preferred coniferous stands over deciduous dominated stands and wetter ecosystems were preferred over dryer ones. In all cases where stand structure variables gen- Discussion erated measurable predictive power, martens selected for greater crown closure and older stands. Selection for I demonstrate that climax ecosystems and stand types stand age was weaker in the Selkirks, where rainfall was are more important in describing marten habitat selection at broader scales while stand age and crown closure (which are correlated with one another) are more important at the finer scale. The finer resolution analysis was 2 still relatively coarse and was unlikely to detect behav-

Purcell Selkirk ioural decisions regarding resting or denning sites. Greater selection for forest structure may be expected 1.5 at finer scales of resolution than analyzed here. Eco- systems and stand types are related to primary produc-

1 tivity and ultimately affect food abundance and population density. Martens have broad diets and select food from all seral stages (Buskirk & Ruggiero 1994, Clark RELATIVE SELECTION (w) RELATIVE 0.5 et al. 1987, Cumberland et al. 2001); therefore, coarse measures of ecosystem productivity such as rainfall likely affect marten density in a general fashion. My data 0 0 1-10 11-30 31-80 81-140 141-250 >250 suggest that wetter ecosystems and stand types support STAND AGE (years) greater marten numbers. Stand structure affects predation risk; presumably greater cover reduces predation Figure 2. Marten selection for stand age (in years) at hair detection sites risk (Buskirk & Ruggiero 1994). Therefore, cover may for the Selkirk and Purcell study areas in southeastern British Columbia. affect local movements and hence foraging opportuni- Selection is measured by percent use divided by percent available. Stands with age zero are non-commercially forested stands with no or ties (Lofroth 1993, Buskirk & Powell 1994, Krohn et al. sparse overstory. Error bars indicate 95% confidence intervals. 1997). Cover may also influence prey abundance (Coffin

© WILDLIFE BIOLOGY · 12:1 (2006) 57 et al. 1997). I suggest that forest overstory influences esis that broad-scale fragmentation of the forest canopy individual fitness by increasing predation risk when (as measured by the standard deviation of overstory clo- overstory is low. sure) negatively affects marten habitat (Thompson & My work was in winter during the period of maximum Harestad 1994:359). Several detailed analyses have re- snow depth. Preferences for structure may be relaxed in cently demonstrated the negative effect of forest frag- summer when deciduous vegetation has leaves. The dif- mentation on martens (Chapin et al. 1998, Hargis et al. ferences in abundance among ecosystems occurred at 1999, Potvin et al. 2000) although not all researchers large scales and are unlikely to vary among seasons. controlled for habitat loss (as caused by forest removal) Similar to Krohn et al. (1997) in California and Chapin explicitly in their analyses. I found that broad-scale frag- et al. (1997) in Maine, I did not find strong selection mentation of the canopy reduced habitat selection even among stand types when I controlled for variation due when I explicitly controlled for habitat loss by includ- to structure and ecosystem type. Nor was selection for ing stand age and crown closure in the model. As forest stand types consistent between areas. Like Lofroth fragmentation increases, some suitable habitat fragments (1993), I observed consistent selection against decidu- become difficult to access, and hence are not used (Cha- ous stands. I observed selection against Douglas-fir pin et al. 1998), presumably because martens are reluc- stands in the Selkirk area but not in the Purcell area. tant to travel through open habitat to access the remain- Selection against Douglas-fir-larch stands in the Selkirks ing patches. is not surprising because these were the driest stand types Thompson & Harestad (1994) suggested that all ser- in the ICH and ESSF ecosystems. Douglas-fir and larch al stages younger than and including mature forest will were more abundant in the Purcell area and occurred in receive less use by martens than old growth. My data do all ecosystems. Selection against the dryer Douglas-fir- not support this suggestion because I detected martens larch stands in the Selkirks was probably based on mois- in all stand ages even open cut-blocks, and selection ture regime rather than on overstory species. My obser- across stand age was relatively weak (see Fig. 2). The vations, like those of Chapin et al. (1997), support the relatively weak influence of stand age, or its correlate suggestion by Buskirk & Powell (1994) that tree spe- crown closure, suggests that the influence of seral stage cies composition is less important to martens than the on marten habitat quality was not so profound in the for- physical structure of the stand. ests I worked in. Baker (1992) and Mowat et al. (2000) Selection against deciduous stands may also be driv- found that only the most open cut-blocks (< 10 years en by lack of physical structure because crown closure old) received markedly less use by martens than older in BC forest cover mapping is measured in summer and stands in coastal British Columbia. Lofroth (1993) dem- is therefore overestimated in deciduous stands in win- onstrated that only stands with < 20% overstory closure ter. The consistent selection against deciduous stands, received less marten use than more closed stands in a even when the variation in structure is accounted for, wet subboreal region of central British Columbia. Bow- may be due to inaccuracy in the structural data during man & Robitaille (1997) found that marten used second winter, rather than to selection against a stand type per growth spruce/fir stands in Ontario and that martens se. Payer & Harrison (2003) point out that martens occur selected for greater structure in mid-seral stands, as they in almost pure deciduous stands in parts of their range. do in mature stands. Payer & Harrison (2003) showed Poole et al. (2004) have shown that a medium density that martens used second-growth conifer and deciduous resident population of marten exists in predominantly stands in Maine and again, they selected for greater struc- deciduous forest in northeastern BC. ture in regenerating areas. Canopy closure is achieved My data do not support the hypothesis that human use rapidly in wet forests and the above observations sug- of a landscape, as indexed by roads and logging, reduce gest that regenerating stands can reach the stage where habitat quality for martens beyond the change in forest overstory structure is no longer limiting in about 10 structure caused by logging. My analysis was a rather years. In addition, logging methods used in wetter areas coarse-scale test of this hypothesis in an area with rela- often leave large volumes of debris in the logged area tively low road and logging density; human use of both (Mowat et al. 2000, McCleary & Mowat 2002), hence areas was low, especially in winter. Robitaille & Aubry short-term limitation of downed wood is unlikely (Payer (2000) presented data that suggested that road traffic & Harrison 2003). Further, Cumberland et al. (2001) sug- caused martens to avoid forest in close proximity to roads, gested that martens may depend on snowshoe hares Lepus but they did not measure whether this avoidance might americanus, grouse and red squirrels for a much greater reduce habitat quality at a larger scale. portion of their diet than previously assumed. Hares and Results from the Purcell study area support the hypoth- grouse both use closed regenerating stands during win-

58 © WILDLIFE BIOLOGY · 12:1 (2006) ter (Krebs et al. 2001). I suggest that marten habitat use Zielinski & Stauffer (1996) demonstrated that trap dura- is less associated with forest age in wet forests than in tion was the major variable affecting trap success using dryer forests. baited track plates for martens and fishers. Duration did The methods used in my study present several limita- not affect detection success during the Selkirk fieldwork tions. First, selection for structure at the fine resolution because there was little variation in duration (Mowat & in the Selkirk area may have been weak because we did Paetkau 2002) compared to the Purcell work. The total not sample in many forest openings. Alternatively, we snowfall while the trap was active had little relationship may have observed reduced selection for structure be- to trap success in the Purcells where snowfall averaged cause the Selkirk area was relatively wet, and regener- 8 cm and was never > 10 cm during a trapping episode. ation was likely faster than on much of the Purcell study In the Selkirks, where snowfall averaged 38 cm per epi- area. The influence of stand age and crown closure may sode and was > 1 m during some episodes, this variable be small in wet areas because vegetation closure recov- negatively influenced detection success. In both areas ers rapidly. The systematic nature of the sampling on temperature was the most important covariate affecting both areas means that stand types and ecosystems were trap success. I know of no other studies demonstrating likely sampled across the entire range of variation. Also, this effect with carnivores, but fur trappers have often stands mapped as non-forested are given zero values for noted the influence of temperature on trapping success. age and overstory but are often sparsely treed. These White et al. (1982) pointed out that temperature can be stands do not support commercial forests based on the an important variable affecting trap success for small definition of the mapping system. Martens used these mammals. sparsely treed stands which confounded the relationship between stand age and marten presence, especially for Acknowledgements - funding for this work was provided by the Purcell area (see Fig. 2). The influence of non-for- Forest Renewal British Columbia and Slocan Forest Products, Tembec, and the Arrow IFPA. I thank K. Stuart-Smith, D. ested areas on regression results was likely to have been Stanley, P. Frasca, G. Richardson, R. Serrouya, P. Cutts, M. small because these areas were few and their influence Panian, C. Strobeck, C. Davis and D. Underwood. D. Fear, was reduced when habitat values were averaged across S. Petrovic, M. Petrovic, P. Cutts and C. Shurgot helped with windows. fieldwork. D. Paetkau, M. Paradon, M. Watt, K. Stalker and C. Other mapping errors may have influenced the rela- Kyle assisted with DNA analysis. M. Buchanan, D. Pritchard and K. Poole did GIS mapping and data extraction. S. Minta tionship among independent variables. Forest age is and K. Heinemeyer kindly shared their knowledge about hair updated regularly following the original mapping or the removal of marten. Thanks to V. Nams, I. Adams, K. Poole, harvest of an area. Crown closure is updated less often P. Genovesi and several anonymous referees for their reviews and, while this variable probably changes little in a of the manuscript. mature forest over 20-30 years (the age of mapping I used), major changes occur in young stands. For young stands, age is probably a better index of crown closure References in my data. My data for the Purcell area suggest little Baker, J.M. 1992: Habitat use and spatial organization of pine use of stands < 10 years of age, however, the ultimate marten on southern Vancouver Island, British Columbia. - limiting factor may be lack of overstory. Finally, the for- M.Sc. thesis, Simon Fraser University, Vancouver, Canada, est mapping data are averaged across hectares and bait 119 pp. sites attract animals from tens and perhaps hundreds of Bowers, M.A., Gregario, K., Brame, C.J., Matter, S.F. & meters. My data measure marten association with stand Dooley, J.L.J. 1996: Use of space and habitats by meadow structure, not site level structural features. Other inac- voles at the home range, patch and landscape scales. - Oeco- curacies in the GIS data such as misclassification of tree logia (Berlin) 105: 107-115. species and spatial errors in polygon boundaries likely Bowman, J.C. & Robitaille, J-F. 1997: Winter habitat use of created noise in the habitat data and reduced model fit; American martens Martes americana within second-growth bias was less likely. forest in Ontario, Canada. - Wildlife Biology 3: 97-105. Weasels may have reduced marten detection in the sites Burnham, K.P. & Andersen, D.R. 1998: Model selection and inference: a practical information-theoretic approach. - where the two species co-exist, but the number of sites Springer-Verlag, New York, USA, 488 pp. where I detected weasels was relatively small. Weasels Buskirk, S.W. & Powell, R.A. 1994: Habitat ecology of fish- were found in all forest types, but occurred more in open ers and American martens. - In: Buskirk, S.W., Harestad, sites and stands than martens (Mowat & Poole 2005). A.S., Raphael, M.G. & Powell, R.A. (Eds.); Martens, sables, I demonstrated that trap duration, temperature and and fishers; biology and conservation. Cornell University snowfall can affect trap success in a winter environment. Press, Ithaca, New York, 283-296 pp.

© WILDLIFE BIOLOGY · 12:1 (2006) 59 Buskirk, S.W. & Ruggiero, L.F. 1994: Marten. - In: Ruggiero, Lofroth, E.C. 1993: Scale dependent analyses of habitat selec- L.F., Aubry, K.B., Buskirk, S.W., Lyon, L.J. & Zielinski, tion by marten in the Sub-Boreal Spruce Biogeoclimatic W.J. (Eds.); The scientific basis for conserving forest car- Zone. - M.Sc. thesis, Simon Fraser University, Burnaby, nivores: American marten, fisher, lynx, and wolverine in B.C., 109 pp. the western United States. USDA Forest Service GTR-RM- Manly, B., McDonald, L. & Thomas, D. 1993: Resource selec- 254, Rocky Mountain Forest and Range Experiment Station, tion by animals: statistical design and analysis for field stud- Fort Collins, Colorado, USA, pp. 7-37. ies. - Chapman and Hall, London, UK, 177 pp. Chapin, T., Harrison, D. & Katnik, D.D. 1998: Influence of May, R.M. 1994: The effect of spatial scale on ecological ques- landscape pattern on habitat use by American marten in an tions and answers. - In: Edwards, P.J., May, R.M. & Webb, industrial forest. - Conservation Biology 12: 1327-1337. N.R. (Eds.); Large-scale ecology and conservation biology. Chapin, T., Harrison, D. & Phillips, D. 1997: Seasonal habi- Blackwell Scientific Publications, Oxford, UK, pp. 1-17. tat selection by marten in an untrapped forest preserve. - McCleary, K. & Mowat, G. 2002: Using forest structural diver- Journal of Wildlife Management 61: 707-717. sity to inventory habitat diversity of forest-dwelling wild- Clark, T.W., Anderson, E., Douglas, C. & Strickland, M. 1987: life in the West Kootenay region of British Columbia. - BC Martes Americana. - Mammalian Species Account No. 289, Journal of Ecosystems and Management 2(2): 1-13. 8 pp. Meidinger, D. & Pojar, J. 1991: Ecosystems of British Colum- Coffin, K.W., Kujala, Q.J., Douglass, R.J. & Irby, L.R. 1997: bia. - Special Report Series 6, B.C. Ministry of Forests, Interactions among marten prey availability, vulnerability, Research Branch, Victoria, B.C., Canada, 330 pp. and habitat structure. - In: Proulx, G., Bryant, H.N. & Wood- Menard, S. 1995: Applied Logistic Regression. - Sage Universi- ard, P.M. (Eds.); Martes: taxonomy, ecology, techniques, ty Paper Series on Quantitative Applications in Social and management. The Provincial Museum of Alberta, Ed- Sciences, 07-106, Thousand Oaks, CA, USA, 98 pp. monton, pp. 372-392. Meyer, J.S., Irwin, L.L. & Boyce, M.S. 1998: Influence of Cowan, I. McT. & Guiguet, C.J. 1965: The mammals of British habitat abundance and fragmentation on northern spotted Columbia, 3rd edition. - British Columbia Provincial owls in western Oregon. - Wildlife Monographs No. 139, Museum, Handbook No. 11, Victoria, British Columbia, Ca- 50 pp. nada, 414 pp. Mowat, G. & Paetkau, D. 2002: Estimating martens Martes Cumberland, R.E., Dempsey, J.A. & Forbes, G.J. 2001: Should americana population size using hair capture and genetic diet be based on biomass? Importance of larger prey to the tagging in southeast British Columbia. - Wildlife Biology American marten. - Wildlife Society Bulletin 29: 1125-1130. 8: 201-209. Foran, D.S., Crooks, K.C. & Minta, S.C. 1997: DNA-based Mowat, G. & Poole, K.G. 2005: Habitat associations of short- analysis of hair to identify species and individuals for pop- tailed weasels in winter. - Northwest Science 79: 27-35. ulation research and monitoring. - Wildlife Society Bulletin Mowat, G., Shurgot, C. & Poole. K.G. 2000: Using track plates 25: 840-847. and remote cameras to detect marten and short-tailed wea- Gibilisco, C.J. 1994: Distributional dynamics of modern sels in coastal cedar hemlock forests. - Northwestern Natura- Martes. - In: Buskirk, S.W., Harestad, A.S., Raphael, M.G. list 81: 113-121. & Powell, R.A. (Eds.); Martens, sables, and fishers; biolo- Nams, V.O., Mowat, G. & Panian, M.A. 2005: Determining gy and conservation. Cornell University Press, Ithaca, New the spatial scale for conservation purposes - an example with York, pp. 283-296. grizzly bears. - Biological Conservation 128: 109-119. Hargis, C.D., Bissonette, J.A. & Turner, D.L. 1999: The influ- Payer, D.C. & Harrison, D.J. 2003: Influence of forest struc- ence of forest fragmentation and landscape pattern on Ameri- ture on habitat use by American marten in an industrial for- can martens. - Journal of Applied Ecology 36: 157-172. est. - Forest Ecology and Management 179: 145-156. Huggard, D. 1999: Marten use of different harvesting treat- Pedlar, J., Fahrig, L. & Merriam, H. 1997: Raccoon habitat ments in high-elevation forest at Sicamous creek. - B.C. use at 2 spatial scales. - Journal of Wildlife Management Ministry of Forests, Research Report 17, 17 pp. 61: 102-112. Krebs, C.J., Boutin, S. & Boonstra, R. (Eds.); 2001: Ecosystem Poole, K.G., Porter, A.D., de Vries, A., Maundrell, C., Grindal, dynamics of the boreal forest: the Kluane project. - Oxford S.D. & Cassady St. Clair, C. 2004: Suitability of young University Press, New York, 511 pp. deciduous-dominated forest for American marten and the Krohn, W.B., Zielinski, W.J. & Boone, R.B. 1997: Relations effects of forest removal. - Canadian Journal of Zoology 82: among fishers, snow, and martens in California: results from 423-435. small-scale spatial comparisons. - In: Proulx, G., Bryant, Potvin, F., Belanger, L. & Lowell, K. 2000: Marten habitat H.N. & Woodard, P.M. (Eds.); Martes: taxonomy, ecology, selection in a clearcut boreal landscape. - Conservation Biol- techniques, and management. The Provincial Museum of ogy 14: 844-857. Alberta, Edmonton, pp. 372-392. Powell, R.A. 1994: Effects of scale on habitat selection and Kyle, C., Davis, C.S. & Strobeck, C. 2003: Genetic homoge- foraging behavior of fishers in winter. - Journal of Mammal- neity of Canadian mainland marten populations underscores ogy 75: 349-356. the distinctiveness of Newfoundland pine martens. - Cana- Resources Inventory Branch 1995: Relational data dictionary dian Journal of Zoology 81: 57-66. (RDD) 2.0. - Ministry of Forest, Victoria, B.C., Canada.

60 © WILDLIFE BIOLOGY · 12:1 (2006) Available at http://srmwww.gov.bc.ca/tib/reports/datadic- servation. Cornell University Press, Ithaca, New York, pp. tionary/rddzip.zip. 355-367. Robitaille, J-F. & Aubry, K. 2000: Occurrence and activity of White, G.C., Andersen, D.R., Burnham, K.P. & Otis, D.L. American martens in relation to roads and other routes. - 1982: Capture-recapture and removal methods for sampling Acta Theriologica 45: 137-143. closed populations. - Los Alamos National Laboratory LA- Strickland, M.A. & Douglas, C.W. 1987: Marten: - In: Novak, 8787-NERP, Los Alamos, New Mexico, USA, 235 pp. M., Baker, J.A., Obbard, M.E. & Malloch, B. (Eds.); Wild Zielinski, W.J. & Kucera, T.E. (Eds.) 1995: American mar- furbearer conservation and management in North America. ten, fisher, lynx and wolverine: survey methods for their Ontario Ministry of Natural Resources, Toronto, Ontario, detection. - USDA Forest Service, PSW-GTR-157, Pacific Canada, pp. 530-547. Southwest Research Station, Albany, California, USA, 163 Thompson, I.D. & Harestad, A.S. 1994: Effects of logging on pp. American martens and models for habitat management. - Zielinski, W.J. & Stauffer, H.B. 1996: Monitoring Martes pop- In: Buskirk, S.W., Harestad, A.S., Raphael, M.G. & Powell, ulations in California: survey design and power analysis. - R.A. (Eds.); Martens, sables, and fishers; biology and con- Ecological Applications 6: 1254-1267.

PAPER VII

Grizzly Ursus arctos and black bear U. americanus densities in the interior mountains of North America

Garth Mowat, Douglas C. Heard, Dale R. Seip, Kim G. Poole, Gord Stenhouse & David W. Paetkau

Mowat, G., Heard, D.C., Seip, D.R., Poole, K.G., Stenhouse, G. & Paetkau, D.W. 2005: Grizzly Ursus arctos and black bear U. americanus densities in the interior mountains of North America. - Wildl. Biol. 11: 31-48.

We collected hair samples from bears and used microsatellite genotyping to identify individual bears on three study areas near the Canadian Rocky Mountains. We estimated density of grizzly bears Ursus arctos in eight different ecosystems across five study areas, including the reanalysis of two previously published data sets. We also estimated black bear U. americanus density for two ecosystems in one study area. Grizzly bear density was lowest in boreal and subboreal plateau areas, moderate in the Rocky Mountain east slopes and highest in the Rocky Mountain west slopes. Presumably these gross differences are related to ecosystem productivity. In the Rocky Mountain west slopes, grizzly bear density was lower in populations that were partially isolated from the continuous bear population to the north. Presumably, these differences have more to do with human impacts on habitat and survival than ecosystem productivity, because productivity in partially isolated areas was similar to productivity in adjacent continuous populations. We show that large differences in bear density occur down to the ecoregion scale; broader ecosystem classes such as Banci’s (1991) grizzly bear zones, ecoprovinces or ecozones would include areas with major differences in density and are therefore too coarse a scale at which to predict grizzly bear density. There appears to be little movement across ecoregion boundaries further suggesting that this may be an appropriate scale at which to extrapolate density. Differences in density across finer- scale ecosystems are likely due to seasonal movements and not population level differences in density. Average bear movements were longer in less productive ecosystems. Female grizzly bears did not appear to leave their home ranges to fish for salmon Oncorhynchus spp., and extra-territorial movements by males appeared to be rare, in both ecosystems which supported spawning salmon.

Key words: carrying capacity, closure, ecosystems, movements, population size, Ursus americanus, Ursus arctos

Garth Mowat, Aurora Wildlife Research, RR 1, Site 14, Comp 8, Crescent Valley, BC V0G 1H0, Canada - e-mail: [email protected] Douglas C. Heard, British Columbia Ministry of Water, Land, and Air Protection, Fish and Wildlife Branch, Omineca Subregion, 1011 4th Ave., Prince George, BC V2L 3H9, Canada - e-mail: [email protected] Dale R. Seip, British Columbia Ministry of Forests, 1011 4th Ave., Prince George, BC V2L 3H9, Canada - e-mail: [email protected] Kim G. Poole, Aurora Wildlife Research, 2305 Annable Rd., Nelson, BC V1L 6K4, Canada - e-mail: [email protected] Gord Stenhouse, Alberta Sustainable Resource Development, Fish and Wildlife Division, Box 6330, Hinton, Alberta T7V 1X6, Canada - e-mail: gordon.sten [email protected]

Corresponding author: Garth Mowat

Received 30 December 2002, accepted 3 February 2004

Associate Editor: Joel Berger

The conservation of grizzly bears Ursus arctos is a eral DNA-based inventories, reviewed other studies high profile wildlife management issue in North Ame- of grizzly bear density in interior North America, and rica. The issue encompasses both concern about land use compared density among ecosystems. practices, and the impact of hunting and other human The objectives of our study were to estimate popu- caused mortality. Estimates of grizzly bear abundance lation density of grizzly bears in five interior ecosystems were few until the recent development of genetic tag- in three study areas and estimate black bear density in ging techniques (Taberlet et al. 1997, Woods et al. 1999). two interior ecosystems in one study area. For two ear- DNA-based inventories have been applied in many lier studies (Mowat & Strobeck 2000, Poole et al. 2001), areas of western Canada to facilitate harvest man- we recalculated density in order to generate estimates agement decisions. However, these inventories are for four additional ecosystems. We wanted to com- expensive, and they are unlikely to be applied in all areas pare grizzly bear abundance among ecosystems in and where grizzly bears occur. In contrast, there is much less around the Rocky Mountains. We assess which scale of conservation concern for black bear U. americanus ecosystem mapping delineates measurable differences populations in western North America, although the spe- in grizzly bear populations. This information is useful cies is harvested in large numbers, and no empirically in assessing conservation status (Banci 1991) and for derived estimates of abundance exist for the species in extrapolating density to predict harvest. the provinces of British Columbia or Alberta. Estimates of density have direct use for harvest management in both British Columbia and Alberta, especially Methods for grizzly bears, which are managed under a quota system. Human caused mortality of adult bears is the pri- Study areas mary factor limiting grizzly bear populations in west- Yellowhead ern North America (McLellan et al. 1999); hence, con- This 5,352-km2 area contains two distinctly different hab- trolling harvest and other human-caused mortality is a itats separated by an ecozone boundary. The moun- major conservation objective. Precise estimates of tainous habitat in Jasper National Park (ranging to abundance are required because allowable harvests 3,000 m a.s.l.) is typical of the Rocky Mountain east from bear populations are very small (Taylor 1994). In slopes and is the largest portion of the study area. To both British Columbia and Alberta, managers use habi- the east, the topography is rolling and the habitat more tat based population extrapolations to estimate carrying typical of the boreal region to the north and east (Frank- capacity for grizzly bears and then adjust these numbers lin et al. 2001). The Yellowhead study area contains a subjectively based on human impacts on habitat suit- variety of habitats including alpine and subalpine mead- ability (Fuhr & Demarchi 1990, Nagy & Gunson 1990). ows, wet meadow complexes, and forests dominated by However, estimates of carrying capacity are unavailable conifers in the west with a greater deciduous compo- for many of the ecosystems where grizzly bears occur. nent further east of the Rocky Mountains. No salmon Banci (1991) identified 14 grizzly bear zones in west- Oncorhynchus spp. are available to bears in this area. ern Canada based on climate and landform (Wiken People live throughout the study area, both in and out- 1986), which presumably reflect variation in both pri- side the park, although human residents (and use) are mary productivity (which we equate to food in this much fewer in the park relative to the public land to the paper) and life history strategies for bears. Finer scales east. Human use includes forestry, coal mining, oil and of ecosystem classification are available, couched with- gas exploration and development, transportation corri- in the same system; from coarsest to finest they are: eco- dors, trapping and commercial outfitting outside the park. zones, ecoprovinces and ecoregions (Wiken et al. 1996). In the park human use includes one major transporta- We treat the idea that bear density varies among ecosys- tion corridor and a single paved road; other use is main- tems as a hypothesis. To this end, we conducted sev- ly non-motorized recreation.

32 © WILDLIFE BIOLOGY · 11:1 (2005) and varies greatly in extent among drainages; the headwaters of most watersheds are undeveloped wilder- ness. There are no permanent residents in the mountains. Herrick Creek, to the south, has fall chinook salmon O. tshawytcha spawning runs. Spawning fish reach natal streams in mid-August or September and are available in < 10% of the mountain study area. Total numbers of spawning adults enumerated within Herrick Creek and associated tributaries have ranged from approximately 500 to 4,000 fish annually. In fall 2000, 679 fish were counted, and the run size was estimated at 2,000-2,500 fish (B. Toth, unpubl. data).

Bowron Figure 1. Location of the study areas (outlined in heavy black). PP = Prophet plateau, PM = The Bowron River study area covers Prophet mountains, HR = Parsnip mountains, FB = Parsnip plateau, BR = Bowron River, 2 YH = Yellowhead, NCS = northcentral Selkirk Mountains and SCS = southcentral Selkirk 2,494 km and is a broad valley that Mountains. Parks are stippled gray, highways are thin black lines and provincial and inter- separates a more rugged mountainous national boundaries are thicker black lines. Cities are black. area in the east from less rugged topography in the west; however, no ecoregion boundaries occur in the area. Low elevations are covered with subboreal forest, while moderate elevations are predomi- Parsnip nantly subalpine forest. Only very small patches of The Parsnip River study area covers 9,452 km2 and is alpine exist in the western portion of the study area, while divided into plateau and mountain areas for analysis larger and more continuous patches occur in the east due based on an ecoregion boundary (Demarchi et al. 1990; to the more mountainous topography. Likewise, ava- Fig. 1). In the 3,016-km2 plateau area, elevations range lanche chutes are rare in the eastern portion of the from about 800 to 1,100 m a.s.l. In the 6,436-km2 study area, but common in the west. Most of the area mountainous area elevations range from valley bot- is covered by an extensive road system although roads toms at about 800 m to peaks of 2,700 m. in the steep eastern portion are few and often impass- The plateau portion of the study area primarily sup- able. Extensive logging occurred in the subboreal zone ports subboreal conifer forests, and lakes and wetlands during the 1980s in an effort to control an insect outbreak. are abundant. Industrial development is extensive in- The center of the study area contained a 640-km2 clear- cluding a major highway and a network of forest roads cut which was largely continuous. and scattered human settlement. A substantial propor- There are both chinook and sockeye salmon O. ner- tion of the plateau forests has been logged over the past ka runs in the Bowron River. Total spawning escape- 30 years and supports new and regenerating cutblocks. ment of chinook within the Bowron system was esti- In the mountainous portion of the study area, subalpine mated at 5,636 for 2001, with an estimated mean annu- conifer forests dominate at lower elevations. Large al escapement of 6,168 for the previous 10-year peri- avalanche chutes measuring in the square kilometers are od (D. Michie, unpubl. data). Sockeye salmon enter the common. Rolling meadows and open basins are com- Bowron River slightly later than the chinook and the mon in the subalpine. Higher elevations contain a com- majority migrate directly through the mainstem and bination of barren rock, ice or alpine tundra communities. spawn south of the study area in Bowron Lakes Provin- Industrial development in the mountains is less than cial Park. Total spawning escapement of sockeye with- on the plateau, although all major and some minor in the Bowron River in 2001 was estimated to be 5,842 watersheds have logging roads along at least part of the (N. Todd, unpubl. data) with a mean annual escapement valley bottom. Logging is generally at lower elevations of 8,990 for the previous 10-year period (1987-1996).

© WILDLIFE BIOLOGY · 11:1 (2005) 33 Prophet River al. (1999) and Mowat & Strobeck (2000). Briefly, we The Prophet River study area is located in the boreal used a systematic grid design to distribute hair capture region of northeastern British Columbia and is described sampling effort across each study area. One hair removal by Poole et al. (2001). We divided the study area into site was set in each cell, and sites were moved within mountains and plateau areas for population estimation each cell for each subsequent trapping session. Sites were based on an ecoprovince division (Poole et al. 2001). baited with rotten blood, fish or meat, and hair samples were removed from bears as they passed by the single Central Selkirk Mountains strand of barbed wire surrounding the site. A novel The Selkirk Mountains study area is located in the scent such as beaver castor, catnip, skunk essence or fen- temperate mountains of southeastern British Columbia nel oil was added during each recapture session to re- and is described by Mowat & Strobeck (2000). We di- duce the chance that bears would become disinterest- vided the area north-south for population estimation ed in bait sites in later trapping sessions. based on evidence from Proctor et al. (2002) who used In the Yellowhead area, we trapped three hair capture genetic methods to show that there was little north-south sessions between 19 May and 9 July 1999 (Table 1). movement of bears in the study area, and Mowat & Intensive live capture effort using both aerial darting and Strobeck (2000) who found greater detection success in leg snares was conducted before and during our hair re- the north than in the south. The majority of the 9,866- moval work. Bears were attracted to open areas (to fa- km2 study area is in a single ecoregion; a sliver of area cilitate aerial darting) and snare sites with large baits. along the southern boundary fell within a dryer ecore- Grizzly bears were live captured between 28 April and gion. 20 June 1999 with much greater effort during April to mid-May. Live capture effort was carefully distributed Hunting and recreational activity across the study area, and home range data were avail- Recreational use in all study areas varied seasonally and able for a number of bears (Nielsen et al. 2002; J. Bou- included hunting, fishing, camping, and all-terrain vehi- langer & G. Stenhouse, unpubl. data.). cle use, except in Jasper Park in the Yellowhead study In the Parsnip study area, we trapped four hair cap- area. Grizzly bear hunting was controlled by quota in ture sessions between 30 May and 2 August 2000 (see both Alberta and British Columbia. Hunter effort is Table 1). We conducted additional hair sampling at controlled by limiting the number of hunters through a 10 sites along three salmon spawning streams in the lottery system. Hunters were encouraged to shoot males mountains along the southern boundary, between 16 and the shooting of a bear accompanied by another August and 4 September 2000. River sites were checked bear was prohibited to protect females with young. every 4-5 days and were not moved between sessions. Black bears were hunted during spring and fall with a River sites differed from upland sites in that they were bag limit of two per hunter. All grizzly bears killed (in- not baited; 1-3 strands of wire were strung across sand cluding problem bears killed by government employ- bars or trails to remove hair from passing bears. In the ees) were submitted to government wildlife staff for data north end of the Parsnip area, grizzly bears had been radio and sample collection. collared and tracked for three years previous to our study (Ciarniello et al. 2001). Using aerial darting and Field methods leg-snare capture techniques, 37 grizzly bears and 23 We followed the field methods outlined by Woods et black bears had been live captured previous to our

Table 1. Study areas, field methods and sampling design used in the five studies reported on in this paper.

Study area Year of Cell size No of Session Study area size (km2) survey (km2) sessions1 duration (days) Sample dates Bait South Central Selkirks 5226 1996 64 5 10 19 June-29Aug. Meat & fish oil North Central Selkirks 4640 1996 64 5 10 19 June-29Aug. Meat & fish oil Prophet plateau 5413 1998 81 5 12 25 May-21 Aug. Blood & fish oil Prophet mountains 3114 1998 81 5 12 25 May-21 Aug. Blood & fish oil Yellowhead 5352 1999 81 42 14 19 May-29 July Blood & fish oil Parsnip plateau 3016 2000 64 4 14 30 May-22 Aug. Blood & fish oil Parsnip mountains 64363 2000 64 4 14 30 May-22 Aug. Blood & fish oil Bowron River 2494 2001 64 3 14 2 Aug.-22 Sept. Meat, blood & fish oil 1 Sites were moved among sessions in all studies except for sites along the Bowron River (N = 27 of 159 sites sampled). 2 The first session was live trapping and subsequent sessions were hair capture. 3 The study area size for black bears was 3,636 km2.

34 © WILDLIFE BIOLOGY · 11:1 (2005) fieldwork. During the period immediately preceding our We attempted to increase genotyping success and hair removal work, three grizzly bears and 10 black bears decrease typing errors by using at least 10 guard hairs were live captured. in a sample. If < 5 guard hairs were available, then we In the Bowron area, we trapped three 14-day hair cap- added up to 30 underfur hairs to the sample. Roots were ture sessions between 2 August and 22 September 2001 clipped from guard hairs, but entire underfur hairs were (see Table 1). These sites were systematically distribut- used for extraction. Occasionally, samples from adja- ed throughout the study area, and we refer to these as cent barbs were combined if both samples had few upland sites. Concurrently, we trapped five one-week hairs. No roots were available for some river sites and sessions along the river between 4 August and 21 Sep- in this case we put all available hair into the extraction. tember 2001. River sites were placed along bear trav- We did not extract adjacent samples, or samples sepa- el routes adjacent to reaches which were known spawn- rated by a single barb, because they are usually from the ing areas for chinook salmon. At a few reaches no ade- same bear (J. Boulanger & S. Himmer, unpubl. data). quate trail set could be established so we built a bait site We preferred to analyze every third sample in a group and baited it with blood and fish oil. of consecutive samples, but this varied depending on Field methods for the Prophet River and Central Sel- sample quality. We analyzed at least one sample from kirk Mountains study areas were similar to the above each group of samples at a site unless there were more and are described in Mowat & Strobeck (2000) and Poole than eight groups, in which case we usually chose the et al. (2001) and summarized in Table 1. best samples from about eight groups. We analyzed all samples from the Yellowhead area. Samples with hair Genetic analysis from more than one individual have 3-4 alleles at one DNA analysis was conducted by University of Alberta or more loci (Paetkau 2003) and were excluded. (Edmonton, Alberta, Canada) and Wildlife Genetics In- We declared two samples to be from the same bear ternational (Nelson, British Columbia, Canada) fol- when the genotype they had in common (i.e. excluding lowing methods described in Woods et al. (1999) and the loci that were incomplete for either animal) had a P Paetkau (2003). Species was confirmed using a length value for the sibling match test of < 0.05 (Woods et al. polymorphism in the mitochondrial DNA (mtDNA) 1999). Genotyping errors were minimized by follow- control region (Paetkau & Strobeck 1996). Alternatively, ing the error testing procedures outlined in Woods et al. species assignment was based on the results of the mi- (1999) for the Yellowhead study and Paetkau (2003) for crosatellite marker G10J, where black bears have ex- the Parsnip and Bowron areas. Once the process of clusively odd-numbered alleles and brown bears have identifying individuals was complete, one sample from exclusive even-numbered alleles, at least in our study each individual was selected for gender determination. areas. This method is more efficient than a mtDNA-based Gender was identified using a length polymorphism in species test because it also provides information on in- the Amelogenin gene that differentiates the X and Y dividual identity. Furthermore, G10J amplifies in a chromosomes (Ennis & Gallagher 1994). Primers were greater proportion of samples than any other nuclear modified to discourage the amplification of human marker that we used, so eliminating samples that fail to DNA (D. Paetkau, unpubl. data). amplify with this marker is an efficient way to eliminate samples that have no prospect of producing multi- Statistical analysis locus, nuclear genotypes. We used the mark-recapture models in the software pro- In order to identify individual bears, each grizzly grams CAPTURE and MARK to test for capture vari- bear sample was analysed at six microsatellite markers ation and estimate population size (Otis et al. 1978, White (G1A, G10B, G1D, G10J, G10L and G10M) using pri- & Burnham 1999). Because most collared grizzly bears mers and methods described by Paetkau et al. (1998). had been darted from the air rather than snared, it is For the Parsnip and Bowron areas, another six markers unlikely that their previous live capture caused those indi- (G10C, G10H, G10P, G10X, MU50 and MU59; Paetkau viduals to avoid baited hair traps. A positive behavioural et al. 1998, Taberlet et al. 1997) were screened on a small response was unlikely for hair captured bears because number of samples, and rejected due to lower variabil- there was no available food reward at a site, and traps ity or less robust amplification. For black bears we were moved each session. Likewise, we tried to reduce screened 12 microsatellite markers (G10C, G10H, negative behavioural response (trap apathy) by adding G10J, G10L, G10X, G1A, G10B, G1D, G10M, G10P, novel scents during each recapture session. Therefore, MU50 and MU59) and used the five former markers for we did not consider behaviour models during model individual identity for the reasons given above. selection. Time variation is easily detected in the absence

© WILDLIFE BIOLOGY · 11:1 (2005) 35 of a behavioural response except when sample sizes are the grid to attempt to enclose the entire 'trappable area' small. Heterogeneity of capture probabilities among indi- or superpopulation (Dice 1938, Kendall 1999). Average viduals is more difficult to detect and can cause sig- spring-early summer home range size was taken from nificant bias. We examined the tests in CAPTURE for the literature or local unpublished data (e.g. Ciarniello evidence of heterogeneity, while considering that these et al. 2001). We calculated the radius of a circle equal tests have low power when sample sizes are small (Otis to the average home range sizes. We then extended the et al. 1978, Boulanger & Krebs 1996). We also tested effective census area beyond the perimeter of the study for capture variation among sexes using program MARK area by this distance to correct density estimates for clo- when sample size permitted. If heterogeneity was sug- sure bias (Dice 1938; see White et al. 1982 for a dis- gested we used one of the three heterogeneity models cussion of various similar methods). in CAPTURE for analysis. Data were often too sparse Another way to approach the problem of closure is to test for capture variation conclusively and usually, to weight each animal captured by a measure of its in these cases, there was inadequate data to utilize a het- residency on the study area as determined by radio erogeneity model with confidence (Manning et al. telemetry (Kenward et al. 1981). In the Yellowhead area, 1995). The only models available for these small data there were sufficient locations (mean N = 198, range: sets were time varying models, which would generate 49-689) to estimate residency for 27 bears during 1999- conservative results if heterogeneity was indeed pres- 2002. These 27 bears spent 89.3% (SE = 3.8) of their ent (Otis et al. 1978, Manning et al. 1995). time on the study area during the period of sampling. We used Chao’s time and heterogeneity model when We multiplied the point estimate and confidence inter- sample sizes were small because these models are vals by 0.893, which equates to the time based correc- designed for sparse data (Chao 1989, Chao & Jeng tion given by Kenward et al. (1981). This simplistic cor- 1992) and, the Jackknife heterogeneity model is nega- rection factor does not incorporate the sampling error tively biased with low sample size (Otis et al. 1978). We or the variability in measuring residency among animals. used Darroch’s time model when sample sizes were Finally, we used a simplification of the method pre- large. When both time and heterogeneity variation were sented by Boulanger & McLellan (2001) to correct for indicated we used Chao’s time and heterogeneity mod- closure based on the average distance each individual el. When model choice was uncertain, we used the sim- was detected from the edge of the grid. We calculated ulation routine in CAPTURE to compare the perfor- core population correction estimates by excluding a mance of the candidate models to aid model selection subset of bears based on their distance from the edge (Mowat & Strobeck 2000). Inhospitable habitat (rock, of the study area using 1-km interval increments (Bou- ice and lakes) was subtracted from study area size to cal- langer & McLellan 2001). We then looked for a decline culate all densities. and leveling off in density with distance from edge, and chose the distance from edge based on where density Closure bias began to level off. Population estimates are typically biased upwards when We used the core correction method for the Parsnip, the assumption of geographic closure is not met (White Bowron and Prophet study areas. Boulanger & McLellan et al. 1982). We assessed the possibility of closure bias (2001: Fig. 4) present similar calculations for the com- by visually examining capture locations with respect to bined Prophet River data. We did separate estimates for the study area boundary. In addition we examined the each sex because we felt each sex cohort was likely to closure test result generated in the program CAPTURE demonstrate different closure bias. Where possible, we while recognizing the test is prone to Type 2 errors in apply several closure correction methods for each area the presence of time and behaviour variation (Otis et al. to allow comparison among methods. 1978, White et al. 1982). Thirdly, we examined the relationship between the capture distance from the edge of the grid and density for signs of lack of closure (Bou- Results langer & McLellan 2001). Finally, we subjectively assessed the possible size of closure bias by comparing Yellowhead likely home range sizes to the study area size. We captured 41 grizzly bears 51 different times during We used three different methods to correct for clo- the three hair capture sessions. Live-capture crews cap- sure bias. The boundary strip correction was based on tured 23 different grizzly bears 24 times; one live-cap- the idea that the area trapped is larger than the outer peri- tured bear moved off the study area, and one bear died meter of the grid. In this case we increased the size of during handling. We captured 14 grizzlies during both

36 © WILDLIFE BIOLOGY · 11:1 (2005) Table 2. Capture success and population estimation results for the five study areas reported on in this paper.

No of No of Average capture Model Naïve Closure Study area bears caught captures probability used population size correction method1

South Central Selkirks 38 45 0.09 Mth-Chao 97 (61- 192) None North Central Selkirks 74 91 0.08 Mth-Chao 223 (138- 416) None Prophet plateau 32 42 0.13 Mt-Chao 63 (42- 122) Core Prophet Mtns 67 100 0.21 Mt-Darroch 96 (80- 122) Core Yellowhead 48 71 0.16 Mth-Chao 107 (71- 198) Radio Parsnip plateau, grizzly 21 24 0.12 Mt-Chao 50 (29- 122) Boundary strip Parsnip Mtns, grizzly males 76 106 0.22 Mo 120 (97- 156) Core Parsnip Mtns, grizzly females 140 193 0.22 Mt-Darroch 221 (188- 268) Core Parsnip plateau, black bear 194 216 0.06 Mth-Chao 892 (565-1509) Core Parsnip Mtns, black bears 85 90 0.08 Mt-Chao 363 (200- 758) None Bowron River 53 74 0.32 Mt-Darroc 76 (63- 104) None 1 see Methods for a description of each closure correction method. live capture and hair removal work. Most live captures Parsnip occurred before we began hair removal work, although DNA fingerprinting resulted in identification of 239 griz- live capture effort continued, at lower intensity, into our zly bears and 275 black bears (see Table 2). The griz- second hair removal session. Of the 49 bears captured zly bear sex ratio for the plateau and the mountains com- on the study area during summer 1999, 29 were female bined was 35M:65F (N = 237). The sex ratio of DNA- and 20 were male (41M:59F). Detection success at identified black bears was 45M:55F (N = 194) on the hair sites was higher in the more mountainous eco- plateau and 41M:59F (N = 85) in the mountains. zone to the west (22%; N = 162) than in the flat subboreal ecoregion to the east (16%; N = 38). Bear movements To increase sample size, we combined the live cap- During the Parsnip study, 86 grizzly bears were detect- ture and hair capture data sets for analysis by creating ed more than once (up to four times). Within the moun- a fourth capture session; the new session began 28 tains, the average movement by males was over twice April and ran until the beginning of our first hair removal that for females (Table 4). No recaptures of males session (19 May). When corrected for closure using radio occurred on the plateau. Only one grizzly bear crossed telemetry data, the population estimate was 96 bears the boundary between the mountain and plateau regions, (95% confidence interval (CI): 63-177), which gener- a male that moved from well inside the mountains to well ates a density estimate of 17.9 bears/1,000 km2 (CI: 12- inside the plateau and back again during the study. 32; Table 2). Home ranges were very large in this area, One long-distance movement was detected; a male and the boundary strip method generated a much low- bear that moved 111 km from the Sukunka River val- er density (Table 3). ley in the northeastern corner of the study area to the

Table 3. Grizzly and black bear naïve and closure corrected densities estimated by genetic sampling and mark-recapture analysis for 11 populations in western Canada during 1996-2001. Closure correction methods are described in Methods. Densities in italics were deemed most appropriate for the study area. The closure test examines the null hypothesis that the population described by the data set was closed (Otis et al. 1978).

Naïve Boundary strip Core corrected Radio telemetry Closure test Study area density corrected density density corrected density result South Central Selkirks 19 (12-537) 13 (8-326) n.a. n.a. 0.13 North Central Selkirks 48 (30-590) 36 (22-368) n.a. n.a. 0.08 Prophet plateau 11 (7-518) 7 (5-314) 10 (127-316) n.a. 0.60 Prophet Mtns 31 (26-539) 19 (16-324) 23 (120-329) n.a. 0.02 Yellowhead 20 (13-537) 9 (6-316) n.a. 18 (12-33) 0.04 Parsnip plateau, grizzly 17 (10-540) 12 (7-328)1 No evidence of closure bias n.a. 0.50 9 (5-323)2 Parsnip Mtns, grizzly 54 (46-568) 43 (37-354) 49 (143-459) n.a. 0.39 Parsnip plateau, black bear 296 (187-500) 208 (132-352) 257 (173-458)3 212 (156-405)4 n.a. 0.40 Parsnip Mtns, black bears 100 (55-210) 70 (38-145) 73 (143-149) n.a. 0.62 Bowron River 31 (26-542) 22 (19-332) No evidence of closure bias n.a. 0.71 1 We subjectively reduced the buffer length based on movement information and topography. 2 This estimate is based on a continuous buffer. 3 This estimate was based on a biological decision to use a 1-km buffer. 4 For this estimate we selected the distance from edge based on the apparent decline in edge effect (3 km for females and 5 km for males; Fig. 2) as described in Methods.

© WILDLIFE BIOLOGY · 11:1 (2005) 37 Table 4. Distance moved (in km) between capture locations for ed boundary strip resulted in a 33% higher density grizzly and black bears by ecoregion and sex in the Parsnip river area of British Columbia in summer 2000. over a continuous buffer. We believe the uncorrected population density is most accurate for the plateau Ecoregion Species Sex N Mean SE Range (16.6 bears/1,000 km2; CI: 10-40). Mountains Grizzly k a 45 13.1 2.38 1.6 -110.7 Grizzly m 75 5.3 0.57 0.8 - 38.1 We identified 216 different grizzly bears in the moun- Plateau Grizzly k a 0 tain portion of the Parsnip study area during hair sam- Grizzly m 3 4.6 1.92 1.1 - 7.7 Mountains Black k b 7 8.1 1.33 4.6 - 12.9 pling (see Table 2). We calculated separate popula- Black m 5 4.3 0.40 3.1 - 5.4 tion estimates for males and females in order to reduce Plateau Black k 16 4.6 0.50 2.2 - 8.3 Black m 6 2.7 0.49 1.6 - 4.8 capture variation and allow for more realistic correction a In addition, one male grizzly bear was detected moving from the for closure bias. During hair sampling, eight of 12 radio- mountains to the plateau (34.0 km), and back to the mountains (28.6 collared grizzlies were detected, giving an overall detec- km). b One male black bear that moved 40 km across the mountains to the tion probability of 0.67, which is similar to the overall edge of the plateau was excluded because we felt it was not a ter- detection probability of hair sampled bears (0.63). ritorial movement. The boundary strip correction method resulted in reductions of 15 and 29% for female and male grizzly bears McGregor River in the southern end of the study area in the mountains, respectively, compared to the naïve between sessions 1 and 3. density estimates (see Table 3). Mountain grizzly bear Black bear movements in the Parsnip-Herrick study population size declined measurably using a 1-km inte- were shorter than those of grizzlies, and as for grizzlies, rior boundary to correct for closure bias for females and females did not move as far as males (see Table 4). One a 2-km boundary for males, and then changed little male black bear moved 40 km across the entire moun- using smaller core populations (Fig. 2B). We used the tainous portion of the study area. Three black bears were 1-km core population estimate to calculate density for detected moving across the mountain-plateau boundary, females and the 2-km estimate to calculate density for although all three movements were within 2.4 km of the males, resulting in a density of 49 bears/1,000 km2 boundary. (CI: 43-59), a reduction of 7 and 12% from the naïve density estimates for female and male grizzly bears, re- Population estimation for Parsnip grizzly bears spectively. We believe the core corrected density is We captured 21 grizzly bears during hair sampling on the most accurate of the two closure correction meth- the plateau. During our study, three radio-collared bears ods. spent > 50% of their time on the plateau portion, hence the total number of grizzly bears known to be on the Population estimation for Parsnip black bears plateau during our study was 24 (see Table 2). We caught 194 black bears 216 times on the plateau (see Few bears were captured near the boundary on the Table 2). Changes in population estimates with increas- plateau, and core correction and the closure test in ing distance from the study area boundary suggest the CAPTURE did not demonstrate any evidence of closure use of the 3-km core estimate for female black bears and bias (Fig. 2A, see Table 3). Closure bias may have the 5-km estimate for males on the plateau (Fig. 2C). been small, or the small sample size may have rendered However, this means closure bias of 22% for females the bias undetectable. Total captures were only 14 for and 55% for males, which is highly unlikely given the the smallest core area, 7 km from the boundary. We felt large study area (see Table 3), and relatively small the core closure correction technique was unreliable for home ranges of black bears in wet ecosystems (Pelton the plateau grizzly bears due to the small sample size. 1983, Kolenosky & Strathearn 1987). Given the small Our only option to correct for closure bias for grizzly number of recaptures in this data set, stochastic events bears on the plateau was to use the boundary strip may have lead to declines in the estimates with distance method (see Table 3). To estimate corrected male and from the study area boundary. Therefore, we used the female density on the plateau, we assumed that the naïve 1-km buffer and combined sexes to estimate population estimate of 50 bears had the same 35M:65F ratio that size for the plateau as 753 black bears (CI: 506-1,344), we recorded over the entire grid area. We did not which yields a density of 257 bears/1,000 km2 (CI: buffer the sampling grid at the boundary between the 173-458). plateau and mountain portions of the grid because both We captured 85 black bears 90 times during the four radio telemetry and DNA data indicated almost no capture sessions in the mountains (see Table 2). We com- movement of bears across this boundary during the bined sessions 3 and 4 to estimate population size be- June-July period (see Table 3). This subjectively correct- cause capture success was extremely low in the last ses-

38 © WILDLIFE BIOLOGY · 11:1 (2005) 70 A) 250 B)

60 200 50 females 150 40 males sexes combined 30 100

20 50 10

0 0 024613570 12345 678910

700 C) 400 D)

600 350 300 500 250 400 males 200 females sexes combined 300 150 200 100

100 50

0 0 1234567 0 1 243 5

90 males E) 120 F) males NUMBER OF GRIZZLY BEARS NUMBER OF GRIZZLY 80 females females 70 sexes combined 100 sexes combined 60 80 50 40 60

30 40 20 20 10 0 0 0123 45 6 78 910 02341785 6910

DISTANCE FROM EDGE (km) G) 60

males Figure 2. Core-corrected population estimates for: A) Parsnip plat- 40 eau, grizzly bears; B) Parsnip mountain, grizzly bears; C) Par- females snip plateau, black bears; D) Parsnip mountain, black bears; E) 30 Prophet plateau, grizzly bears; F) Prophet mountains, grizzly bears; 20 G) Bowron river. Each estimate is calculated by removing bears whose mean capture distance is within the stated distance from 10 the edge of the study area boundary, and then extrapolating the estimate of the reduced area to the full study area (Boulanger & 0 McLellan 2001). 0 1357

DISTANCE FROM EDGE (km)

© WILDLIFE BIOLOGY · 11:1 (2005) 39 sion and time variation was obvious. Core-corrected esti- Table 5. Direction and distance moved (in km) between capture locations for grizzly bears in the Bowron River drainage fall in 2001. mates generated a sharp initial decline in population size by excluding bears within 1 km of the border; the esti- Type of movement Sex N Mean SE Range mates were roughly equal for the next four core calcu- Upland to river k 4 10.4 2.7 2.5 - 13.5 m 9 9.4 1.3 2.5 - 13.4 lations (Fig. 2D). Using the 1-km core corrected esti- Along river k 3 8.4 3.8 4.6 - 16 mate suggested a closure bias of 28%, and here again m 11 5.9 1.3 0.7 - 14 Within upland k 2 9.0 1.7 7.3 - 10.6 the low numbers of recaptures makes this number sus- m 7 5.1 1.4 1.6 - 12.2 pect. The boundary strip method generated similar reductions in density (see Table 3). We elected not to correct for closure because recaptures were few; black upland sites, 2) river sites and 3) upland and river sites bear density was 100 bears/1,000 km2 (CI: 55-210) in combined. Sessions 2-3 and 4-5 were combined for the mountains. river sites in order to align the data with the three upland sessions (see Table 2). There were 47 captures of Fall sampling along salmon streams 41 bears in the upland, 34 captures of 26 bears along the During fall hair sampling, grizzly bear hair was collected river, and 74 captures of 53 bears in the combined at seven of 10 sites. Black bear hair was detected at only data. The use of distance from edge to exclude bears one site, which did not detect a grizzly bear. We iden- caught near the study periphery did not suggest any tified 15 grizzly bears (9 M, 6 F), and four bears were degree of closure violation (Fig. 2G), a result that is not recaptured within the fall samplings, all on the same surprising because few bears were caught near the study creeks as originally detected. At four sites we removed area boundary. grizzly bear hair, but we were unable to identify an indi- Population size for the upland was 109 (CI: 67-221) vidual. Hair samples from the fall were generally of low- and 47 (CI: 33-85) for the river. The population estimate er quality than the summer samples; more hairs were using the combined data was 76 (CI: 63-104; see Table ^ required to score all six loci. We combined the third and 2). The estimate for females ( N = 49; CI: 41-70) was ^ fourth capture sessions to calculate grizzly bear popu- double that for males ( N = 25; CI: 21-39), and the com- lation size along these creeks because only one bear was bined density of grizzly bears in the study area was 30.9 captured during the last session. We estimated that 21 bears/1,000 km2 (range: 25.7-42.3). Based on the move- grizzly bears (CI: 16-44) used the 47 km of creek that ment data we assumed our river sites were trapping bears we trapped in fall (5 bears/10 km) which is only a that lived in the two rows of cells (roughly 16 km) on small portion of the bears resident in the nearby area. each side of the river. The estimated density for this area Of the 15 fall-captured bears, 11 (6 M, 5 F) had been using the river sites only was 27.2 bears/1,000 km2 detected previously. The distance between their last (range: 19.1-49.2). There was about six grizzly bears per summer capture location and their fall capture location 10 km of river, a similar number to that observed along − was greater for males (× = 18.6 km, range: 4.0-48.3 km) Herrick Creek in the Parsnip fall sampling, about 100 − than for females (× = 4.8 km, range 0.3-7.5 km). Two km further north. Boundary strip closure correction male grizzly bears made long-distance movements of suggested a much greater reduction in density (see Ta- 35 and 48 km. ble 3).

Bowron Closure correction for Prophet River grizzly Grizzly bear movements bears We detected 36 grizzly bear movements and grouped For the boreal plains portion of the Prophet River study these based on whether the movement was within the area, there was a small decline in the estimated core-cor- upland, along the river, or between the upland and the rected population size of grizzly bears up to 2 km inside river in either direction (Table 5). No movements > 13.5 the study area, but the population estimates increased km were detected from upland to the river (N = 13) even using data from 3 to 9 km inside the boundary (Fig. 2E). though 12 grizzly bears moved < 13.5 km from the up- We used the 2-km corrected estimate to estimate popu- land to the river, which was 24% of all grizzly bears lation size for the boreal plains because this is where the known to be in the study area. This pattern of movement initial decline stopped. Population size was estimated suggests that only local residents used salmon streams. at 53 (CI: 39-89) and density was 10 bears/1,000 km2 (CI: 7-16); virtually identical to the density presented Grizzly bear population size and density by Poole et al. (2001). Mark-recapture databases were generated from: 1) Using the core population correction for the moun-

40 © WILDLIFE BIOLOGY · 11:1 (2005) Table 6. Grizzly bear densities in and near the Rocky Mountains of North America.

Density Study area (bears/1,000 km2) 95% CI size (km2) Reference Rocky Mountain west slopes and adjacent ranges Flathead Valley, B.C. 46-80 130 McLellan (1989) Flathead Valley 48 30-92 3233 Boulanger et al. (2002) Parsnip mountains, B.C. 49 43-59 6436 This study Northern Central Selkirk Mountains, B.C. 48 30-90 4640 Recalculated from Mowat & Strobeck (2000) Glacier National Park, Montana 32 1044 Martinka (1974) Bowron River, B.C. 31 26-42 2494 This study Central Purcell Mountains, B.C. 25 22-40 1650 Boulanger et al. (2002) Central Columbia Mountains, B.C. 26 13-39 4096 Boulanger et al. (2002) Rocky Mountain east slopes Northern Canadian Rocky Mountains, B.C. 23 20-29 3114 Recalculated from Poole et al. (2001) Jasper-Cadomin, Alberta 18 12-33 5352 This study Kananaskis, Alberta 16 254 Wielgus & Bunnell (1994) Southwest Alberta 15 12-20 5030 Mowat & Strobeck (2000) Discontinuous populations in the Rocky Mountain west slopes South Selkirk Mountains, B.C. 23 235 Wielgus et al. (1994) Southern Central Selkirk Mountains, B.C. 19 12-37 5226 Recalculated from Mowat & Strobeck (2000) Swan Mountains, Montana 16 1457 Mace & Waller (1998) South Selkirk Mountains, WA, ID 14 100 Wielgus et al. (1994) Boreal and subboreal plains Parsnip plateau, B.C. 17 10-40 3016 This study Prophet River boreal plains, B.C. 10 7-13 5518 Recalculated from Poole et al. (2001) tainous portion of the Prophet River study area, there Mountains (LeFranc et al. 1987; Table 6). Densities are was a sizable decline in population size up to 6 km inside highest in the Rocky Mountain west slopes, followed the study area and estimates increased from 6 to 9 km by the east slopes, with the lowest densities observed inside the boundary (Fig. 2F). We used the 6-km cor- in the boreal and subboreal plains on either side of the rected estimate to estimate grizzly bear population size mountains. Small study areas are often placed in high- for the Prophet mountains because this is where the ini- er quality habitats than surrounding areas (Smallwood tial decline stopped. We combined two sexes for estima- & Schonewald 1996); therefore some of the densities tion because sample sizes were very low for individual reported for small study areas are likely higher than sexes with data 6 km in from the boundary. The final would be found over a larger surrounding landscape. estimate was 70 grizzly bears (CI: 63-91), which gen- Densities of coastal bear populations are usually much erated a density of 22.5 bears/1,000 km2 (CI: 20-29), 22% higher than observed in the continental interior (Mac- lower than the density presented by Poole et al. (2001). Hutchon et al. 1993, Miller et al. 1997), although Bou- langer et al. (2002) reported a density similar to the Central Selkirk Mountains grizzly bear Rocky Mountain west slopes for a coastal area of south- densities west British Columbia. Densities in the boreal and tun- Grizzly bear density in the southern half of the Central dra regions of Alaska are highly variable; some areas have Selkirk Mountains was 19 bears/1,000 km2 (CI: 12-37) densities similar to the Rocky Mountain west slopes, and 48 bears/1,000 km2 (CI: 30-90) in the north. Again while others are lower than documented in the subbo- the boundary strip closure correction suggested much real plains (Miller et al. 1997). lower densities, but Mowat & Strobeck (2000) suggested Bear density is moderate to low in all the discontin- the outer boundary was largely closed to bear movement, uous populations that have been studied along the and Proctor et al. (2002) demonstrated that there was southern edge of the bear’s range (LeFranc et al. 1987; little movement, among subpopulations within the see Table 6). Densities in these discontinuous popula- study area. There were very few bear captures near tions are likely limited by the effects of habitat loss and the north-south boundary. ongoing human impacts, not lower habitat productivity. All four densities from discontinuous populations in Table 6 are from areas in the west slopes of the Rocky Discussion Mountains and all have habitat qualities similar to adjacent areas to the north. For example, grizzly density in Grizzly bear density the north part of the Central Selkirk Mountains, where There are marked differences in grizzly bear density the grizzly bear distribution is continuous with popu- among biomes that are broadly defined by the Rocky lations to the north and east, was approximately dou-

© WILDLIFE BIOLOGY · 11:1 (2005) 41 ble that in the south despite similar habitat. Capture data We could not compare finer levels of ecosystem clas- and genetic analysis have shown that the southern half sification with our data because in most cases they of the Central Selkirk population is partially isolated from occur at finer resolution than our density data. For surrounding populations (Mowat & Strobeck 2000, Proc- example, ecoregions are further divided into ecodistricts tor et al. 2002). Habitat quality may be secondary to (termed ecosections in British Columbia; Demarchi et human impact in predicting bear density in the south- al. 1990), but our study areas often contained 3-6 dif- ern discontinuous portion of the grizzly bear’s distri- ferent ecodistricts making any comparison of density ten- bution. uous. Further, our movement data suggested that indi- Further north along the interior mountains, bear den- vidual bears move among ecodistricts (G. Mowat, un- sities are consistently higher on the windward side of publ. data), and it is more appropriate to assign densi- the continental divide than in the rain shadow on the lee- ty to units that include the year-round life history of the ward side to the east, likely a result of the reduced rain- population (Miller et al. 1997). Ecodistrict scale effects fall and hence reduced vegetative production in the to bear density are probably best measured by stand lev- rain shadow of the large interior mountains (Hamer & el effects on habitat quality, rather than broad scale cor- Herrero 1987, Hamer et al. 1991). The single study area relates to density. with estimated bear density > 20 bears/1,000 km2 on the east side of the continental divide is in the Prophet Black bear density River area of the Northern Rocky Mountains where rain- The pattern of black bear densities is opposite to that of fall is higher than on the Rocky Mountain east slopes grizzly bears, with estimated black bear density about further south. Grizzly densities are low in the boreal and three times greater in the Parsnip plateau than in the subboreal plains in British Columbia and Alberta (Le- mountains. Even in the mountains, black bears were Franc et al. 1987, this study). Many of these areas have mainly captured in valley bottoms. The low number of also been heavily impacted by humans, which may captures in the mountains in the third and fourth trap- partially explain the consistently low densities. However, ping sessions resulted from our tendency to move hair the boreal plains of the Prophet River were only mild- capture sites higher in elevation through the study. ly impacted by people (Poole et al. 2001), and densi- Black bears were rarely detected near the treeline, pos- ties were low here too. sibly due to competitive exclusion by grizzly bears. Griz- At what scale do landscape factors affect bear den- zly bears also appeared to exclude black bears from near sity? This question is important because density extrap- the Herrick salmon streams when salmon were running. olations are often used to predict harvest levels and using This pattern has also been observed along coastal salm- the correct scale for extrapolation will reduce errors in on streams (Miller et al. 1997, Jacoby et al. 1999). Con- harvest management. Large differences in bear densi- versely, black bears were detected regularly along the ty occur right down the ecoregion level of classification. Bowron River where a larger number of salmon were This is demonstrated by the large differences in densi- spread across a much longer stretch of river. ty among the adjacent Parsnip mountain and plateau Little research has been done on black bear popula- areas, where density varied > 4-fold on either side of an tion characteristics in British Columbia or Alberta. Den- ecoregion boundary (see Table 6). Bear densities also sities varied from about 320 to 800 bears/1,000 km2 in differed across coarser ecosystem levels, as shown by a small area of high quality in northeast Alberta, an area the difference in density across the ecoprovince bound- most comparable to our plateau study area (Young & ary in Prophet and the differences in detection success Ruff 1982). Given the reported home range sizes and between ecozones in the Yellowhead study area. Broader- the presence of three dumps on the study area, actual scale ecosystem classes such as Banci’s (1991) grizzly density was likely similar to, or less than, our Parsnip Plat- bear zones, ecoprovinces or ecozones (also termed eco- eau area because these authors did not correct for lack domains; Bailey 1997) encompass areas with major of closure, and this bias would likely have been large. differences in density and are clearly too coarse a lev- Jonkel & Cowan (1971) reported densities 2-5 times high- el at which to predict or extrapolate grizzly bear den- er than what we document in the Parsnip Mountains for sity. All three of these mapping scales combine areas an area in northern Montana which is most similar to of Rocky Mountain west and east slopes or subboreal our mountain study area. Again, the actual density of (or boreal) plains, where densities can vary > 5-fold. In the greater landscape was likely much lower given the contrast, ecoregions map areas of relatively homogenous study area size, bear home range sizes and the fact that topography and climate and appear to better group no correction was made for closure. Miller et al. (1997) areas of similar bear density. estimated black bear density to be 89 bears/1,000 km2

42 © WILDLIFE BIOLOGY · 11:1 (2005) (CI: 77-103) in a relatively flat, partially forested area Mattson & Reinhart (1995) documented grizzly bears of interior Alaska. This is much lower than the black bear feeding on spawning cutthroat trout Oncorhynchus density we found on the Parsnip plateau; however, black clarki in the Yellowstone ecosystem. They show that bears in their study area existed in creek bottoms and females made greater use of streams and that most forested strips, similar to our mountain population. bears that had > 5% of their home ranges touching a Also, grizzly bear density was considerably higher in spawning stream used trout. Some females may have the Alaska study area than in the plateau portion of our made extra-territorial movements to streams. In interi- area. Miller et al. (1997) reported black bear densities or Alaska, Miller et al. (1997) found that < 40% of res- similar to our plateau estimate in two areas of coastal ident bears moved to a nearby salmon stream; however, Alaska. Kolenosky & Strathearn (1987) reported den- bear use of salmon in this interior Alaska stream was sities of 200-1,300 black bears/1,000 km2 in the Pacific high. These authors also studied two areas on the north- Northwest, with the highest density reported for coastal ern coastal plains of Alaska with periodically abundant Washington. Our data support the suggestion by Miller salmon, but bears used these streams little. Most bears et al. (1997) that black bears occur at higher densities in coastal ecosystems move to streams when salmon are than grizzlies in the interior, and that black bears are most running (MacHutchon et al. 1993, Schoen et al. 1994); abundant where grizzly bears are rare or absent. however, Schoen & Beier (1990) documented that 14% of the bears they had radio-collared never moved to Movements streams during the salmon season. Often the bears that Movement between mountainous and plateau ecosys- did not use salmon streams were females with young tems in spring and early summer appears to be rare, cubs. Even bears that used salmon streams were locat- which is consistent with movements of radio-collared ed away from streams, usually in the alpine, 15% of the bears in this area (Ciarniello et al. 2001) and results in time during the salmon season. Using radio isotope the Rocky Mountains further north (Poole et al. 2001). analysis, Hilderbrand et al. (1996) also demonstrated that Females did not move as far as males, which is expect- most coastal bears rely heavily on salmon in the fall, ed given their smaller home ranges (Nagy & Harold- while a few bears do not use salmon at all. In a following son 1989, Ciarniello et al. 2001), and is also consistent paper, Hilderbrand et al. (1999) demonstrated that salmon with results from the Northern Rocky Mountains (Poole are important to bears at both the individual and popu- et al. 2001). Black bear movements were smaller than lation level because meat in the diet was correlated those of grizzly bears, which suggests that they have with body size, litter size and density. smaller home ranges. Converse to grizzly bears, black bear movements in the mountains were greater than on Closure the plateau, which suggests that black bear ranges are The boundary strip method generated more conserva- smaller in the flatter plateau area. tive estimates of density than the radio telemetry or core- Female grizzly bears did not appear to leave their home closure corrections. We subjectively adjusted the bound- ranges to catch salmon along the Herrick or Bowron ary strip calculations for the Parsnip areas based on our rivers. Given the relatively small home ranges of bears observation of essentially no bear movements along the in the Parsnip mountains, it seems likely that two males plateau-mountains boundary and the presence of glaciers made extra-territorial movements to fish for salmon in along portions of the boundary. The population estimate Herrick Creek; long-distance movements of males may would have been even more conservative had we not have gone undetected in the Bowron because our sam- made these adjustments. Indiscriminate application of ple size was small. Movement distances to the river for a boundary correction factor could greatly overesti- Bowron females were double those observed in the mate closure bias if portions of a boundary are indeed Parsnip mountains, suggesting that female home ranges closed or uninhabited by the study species (Boutin in the relatively flat part of the Bowron mainstem are 1984). Perhaps residency is heterogeneous along study larger than in the very rugged Parsnip mountains to the area boundaries because only a portion of all bears are east. Ciarniello et al. (2001) found that bears living on sampled, and application of a continuous boundary the flatter plateau portion of their study area had much strip overestimates the amount of permeable boundary. larger home ranges than bears residing in the moun- If this is true, then this bias is likely to be greater with tainous portion. Miller et al. (1997: Table 2) also found lower capture probabilities, because individuals along that males moved farther than females to fish in an in- the boundary would be more lightly sampled. terior salmon stream, and movement distances and In contrast, the modest reduction in predicted popu- density were similar to those reported in this paper. lation size for the Parsnip mountains, using concentric

© WILDLIFE BIOLOGY · 11:1 (2005) 43 increases in distance from edge, suggests that lack of Poole et al. (2001) demonstrated that the majority of closure only biased population estimates within about the bears in the Prophet study area resided in the moun- 1 km from the study area boundary. Corrected estimates tains; however, the centre of the study area was in the for buffers 2-7 km inside the study area were similar, boreal plains (Fig. 3). Capture probabilities were low- so the choice of the buffer distance had little effect on er on the boreal plains, which likely explains why Bou- the corrected density. Buffers of > 7 km predicted low- langer & McLellan (2001: Figs. 1 and 3) found a reduc- er population sizes, especially for females; these effects tion in fidelity and capture probabilities at large distances are more likely due to the reduction in sample size than from the edge. At about 20 km from the edge, bears closure bias. caught on the plateau became the majority in the mark- Boulanger & McLellan (2001) suggested closure- recapture sample and at 25 km from the edge, only one correction of 17-25% below naïve estimates for the mountain caught bear remained in the sample (see Fig. entire Prophet River area. They used a core correction 3). This demonstrates the sensitivity of the core correction distance of 10 km to estimate density for the entire study technique to variation in the distribution and capture area. This distance appears overly conservative when probabilities of individuals in the sample. compared to the home ranges observed for bears living In all analyses we attempted to calculate separate in the Parsnip mountains where home range diameters closure corrections for each sex because we expected averaged about 6 km for females and 15 km for males less closure bias for females due to their smaller home (Ciarniello et al. 2001). ranges. Closure bias generally influenced female estimates

Figure 3. Sites at which grizzly bears were detected () and sites at which grizzly bears were not detected () in relation to 5-km wide strips from the study area boundary for the Prophet River grizzly bear DNA inventory (Poole et al. 2001).

44 © WILDLIFE BIOLOGY · 11:1 (2005) closer to the boundary, although the differences were often that were closer to the center of the study area. Indiscrimi- subtle. Conversely, the reduction in sample size that nant application of all three of the closure correction results from dividing and excluding samples reduces methods discussed here could result in a less accurate power and could lead to spurious results. Heterogeneity estimate of density than the naïve estimate. It appears models are more sensitive to sample size than simple that the boundary strip method often overcorrects for clo- time models (Otis et al. 1978, Manning et al. 1995, Ken- sure, at least with low capture probabilities. Subjective dall 1999). The jackknife model in particular tends to reduction of the buffer length based on topography and be biased low when recaptures are few (Otis et al. population distribution can reduce this bias. The core cor- 1978). Sequentially reducing sample size could yield rection method may also overcorrect for closure when reduced estimates due to model bias that could be mis- sample sizes are small or capture distribution and suc- taken for closure bias. The core population correction cess are spatially heterogeneous. The radio-telemetry method appears to be a more accurate method to cor- method can generate biased corrections for closure bias rect for closure than the boundary strip method, but its if the radio-collared sample is not randomly drawn from application will be limited by sample size, as is the the study population. We had radio-collared individu- case for the similar nested grid method (Otis et al. 1978). als for both Parsnip study areas but were unable to use The core correction method, like the boundary strip, could them for this reason. Correcting for closure bias should also generate large errors if applied indiscriminately. be based on a detailed knowledge of the study area, cap- The radio-telemetry method can also provide accu- ture data, species biology and careful examination of the rate estimates of density, but a large and representative appropriateness of each technique. sample of bears must be collared during the study (Eber- hardt 1990, Garshelis 1992). This is difficult not only DNA-based inventory along salmon streams because collaring is expensive but also because live cap- Large numbers of bears can be sampled with relative- ture effort and success are rarely random with respect ly little effort along salmon streams in the interior. We to the residency of bears. generated a precise estimate of the number of bears using Kendall (1999) argued that closure violation would approximately 40 lineal kilometers of river (tributaries add heterogeneity to capture probabilities, necessitat- not included) with about 18 crew days (two people) of ing the use of models that accommodate this form of cap- effort and the use of a truck and a boat. A much less pre- ture variation. We suspect individuals with very low cap- cise estimate of the number of bears in the upland was ture probabilities (who presumably have only a small por- generated with 24 crew days of effort and 35 hours of tion of their home range on the study area) are few helicopter time and much greater truck costs. By our when the grid is large relative to home range size, which crude estimates the upland sampling covered roughly explains why we rarely detected heterogeneity in our data. one third more area than the river sampling for rough- However, when the mean capture probability is low, dif- ly four times the cost. ferences between the mean probability and those of edge The problem with sampling along streams during bears may be small, making it difficult to detect the het- salmon runs is calculating bear density. We spent con- erogeneity caused by edge bears, especially when sam- siderable effort attempting to document movement dis- ple size is low. It seems likely that bears that have only tances and still our results for males were weak, because a very small portion of their home range on the study area we caught few males overall and none on the western have capture probabilities so low that, if they are unde- periphery of our study area. Even in areas where bears tected during the study, they are not accounted for in the are more abundant and the population distribution more population estimate; which is perhaps the desirable re- homogenous, it will be difficult to get robust measures sult if the goal is to estimate mean population size for of bear movement distances. Field costs to measure only the area studied. movement distances could be reduced if sampling were Closure bias should be considered for all estimates of restricted to only those upland areas beyond the distance population density. In a mark-recapture setting statis- bears are known to move to the river. In the Bowron tical tests may help detect lack of closure, but test re- study for example, it was unnecessary to sample the up- sults should not replace knowledge of the study area. land within 10 km of the river to determine if bears were For example, closure test results for the Central Selkirk moving beyond their normal home ranges. Mountains study areas suggested lack of closure, but this result was likely due to the poor capture success along the study area boundary which declined through the Acknowledgements - this project was supported by Forest Re- study, because we moved our sites to higher elevations newal British Columbia, Canadian Forest Products, Corporate

© WILDLIFE BIOLOGY · 11:1 (2005) 45 Lands Information Base, Peace/Williston Fish and Wildlife 56 pp. Available at http://web.unbc.ca/parsnip-grizzly/ Compensation Program, British Columbia Ministry of Water, progress.htm Lands and Air Protection, and British Columbia Ministry of Demarchi, D.A., Marsh, R.D., Harcombe, A.P. & Lea, E.C. Forests, Foothill Model Forest, Alberta Sustainable Resource 1990: The Environment. - In: Campbell, R.W., Dawe, N.K., Development, Jasper National Park, Weldwood of Canada, McTaggart-Cowan, I., Cooper, J.M., Kaiser G.W. & Mc- Hinton Fish and Game Association and Alberta Conservation Nall, M.C.E. (Eds.); The Birds of British Columbia. Volume Organization. We acknowledge the many sponsors of the Foothill Model Forest Grizzly Bear Project. We thank A. de 1. Royal British Columbia Museum, pp 55-144. Vries, K. Deschamps, P. Patterson, B. Toth, G. Watts, G. Mer- Dice, L.R. 1938: Some census methods for mammals. - Jour- cer, B. Logan, T. Enzol and R. Munro for administrative nal of Wildlife Management 2: 119-130. support and advice. Field assistance was ably carried out by Eberhardt, L.L. 1990: Using radio-telemetry for mark-recap- D. Fear, E. Jones, J. Paczkowski, C. Pharness, D. Stanley, M. ture studies with edge effect. - Journal of Applied Ecology Wolowicz, T. Bochmann, J. Yarmouth, G. Watts, G. Mercer, 27: 259-271. K. McClery, B. Renkas, G. Slatter, R. Zroback, J. Dunford, Ennis, S. & Gallagher, T.F. 1994: PCR based sex determination T. Sorenson, C. Found and J. Weaver. We thank P. Rooney, assay in cattle based on the bovine Amelogenin Locus. - J. Bell, P. Bock, R. Diston, R. Klassen and R. McIntyre for Animal Genetics 25: 425-427. flying under often trying conditions. G. Haines, D. Pritchard, Franklin, S.E., Stenhouse, G.B., Hansen, M.J., Popplewell, J. Dugas and S. Sharpe provided GIS support. J. Bonneville, K. Stalker and the staff at Wildlife Genetics International per- C.C., Dechka, J.A. & Peddle, D.R. 2001: An integrated deci- formed the DNA analysis. We appreciate the support of the per- sion tree approach (ISTA) to mapping landcover using sonnel at the Anzac and Herrick camps and Camp Friendship satellite remote sensing in support of grizzly bear habitat at Tsitniz Lake. S. Himmer kindly advised us regarding hair analysis in the Alberta yellowhead ecosystem. - Canadian sorting and sampling along salmon streams, and L. Ciarniello Journal of Remote Sensing 27: 579-592. summarized and provided her data on bear home range sizes. Fuhr, B. & Demarchi, D.A. 1990: A methodology for grizzly bear habitat assessment in British Columbia. - B.C. Ministry of Environment, Wildlife Bulletin No. B-67, Vic- toria, B.C., Canada, 28 pp. Garshelis, D.L. 1992: Mark-recapture density estimation for References animals with large home ranges. - In: McCullough, D.R. & Barrett, R.H. (Eds.); Wildlife 2001: populations. Elsevier Bailey, R.G. 1997: Ecoregions of North America. - United Press Science, London, pp. 1098-1111. States Department of Agriculture, Forest Service, Washing- Hamer, D. & Herrero, S. 1987: Grizzly bear food and habi- ton, D.C. tat in the Front Ranges of Banff National Park, Alberta. - Banci, V. 1991: Updated status report on the grizzly bear in International Conference on Bear Research and Management Canada. - Committee on the Status of Endangered Wildlife 7: 199-214. in Canada, Ottawa, Ontario, 171 pp. Hamer, D., Herrero, S. & Brady, K. 1991: Food and habitat Boulanger, J. & Krebs, C.J. 1996: Robustness of capture-recap- used by grizzly bears along the continental divide in Water- ture estimators to sample biases in a cyclic snowshoe hare ton Lakes National Park, Alberta. - Canadian Field Naturalist population. - Journal of Applied Ecology 33: 530-542. 105: 325-329. Boulanger, J. & McLellan, B.N. 2001: Closure violation bias Hilderbrand, G.V., Farley, S.D., Robbins, C.T., Hanley, T.A., in DNA based mark-recapture population estimates of Titus, K. & Servheen, C. 1996: Use of stable isotopes to grizzly bears. - Canadian Journal of Zoology 79: 642-651. determine the diets of living and extinct bears. - Canadian Boulanger, J., White, G.C., McLellan, B.N., Woods, J., Journal of Zoology 74: 2080-2088. Proctor, M. & Himmer, S. 2002: A meta-analysis of griz- Hilderbrand, G.V., Schwartz, C.C., Robbins, C.T., Jacoby, zly bear DNA mark-recapture projects in British Columbia, M.E., Hanley, T.A., Arthur, S.M. & Servheen, C. 1999: The Canada. - Ursus 13: 137-152. importance of meat, particularly salmon, to body size, Boutin, S. 1984: Home range size and methods of estimating population productivity, and conservation of North American snowshoe hare densities. - Acta Zoologica Fennica 171: 275- brown bears. - Canadian Journal of Zoology 77: 132-138. 278. Jacoby, M.E., Hilderbrand, G.V., Servheen, C., Schwartz, C.C., Chao, A. 1989: Estimating animal abundance with capture fre- Arthur, S.M., Hanley, T.A., Robbins, C.T. & Michener, R. quency data. - Biometrics 29: 79-100. 1999: Trophic relations of brown and black bears in sev- Chao, A. & Jeng, S.L. 1992: Estimating population size for eral western North American ecosystems. - Journal of Wild- capture-recapture data when capture probabilities vary by life Management 63: 921-929. time and individual animal. - Biometrics 48: 201-212. Jonkel, C.J. & Cowan, I.M. 1971: The black bear in the Ciarniello, L.M., Paczkowski, J., Heard, D., Ross, I. & Seip, spruce-fir forest. - Wildlife Monographs No. 27, 57 pp. D. 2001: Parsnip grizzly bear population and habitat pro- Kendall, W.L. 1999: Robustness of closed capture-recapture ject, 2000 progress report. - Unpublished report for Canadian methods to violations of the closure assumption. - Ecology Forest Products Ltd., Chetwynd, British Columbia, Canada, 80: 2517-2525.

46 © WILDLIFE BIOLOGY · 11:1 (2005) Kenward, R.E., Marcström, V. & Karlbom, M. 1981: Goshawk Nielsen, S.E., Boyce, M.S., Stenhouse, G.B. & Munro, winter ecology in Swedish pheasant habitats. - Journal of R.H.M. 2002: Modeling grizzly bear habitats in the Yellow- Wildlife Management 45: 397-408. head ecosystem: taking autocorrelation seriously. - Ursus Kolenosky, G.B. & Strathearn, S.M. 1987: Black bear. - In: 13: 45-56. Novak, M., Baker, J.A., Obbard, M.E. & Malloch, B. Otis, D.L., Burnham, K.P., White, G.C. & Andersen, D.P. (Eds.); Wild furbearer management and conservation in 1978: Statistical inference from capture data on closed North America. Ontario Trappers Association, North Bay, animal populations. - Wildlife Monographs 62, 135 pp. Ontario, Canada, pp. 443-454 . Paetkau, D. 2003: An empirical exploration of data quality in LeFranc, M.N., Moss, M.B., Patnode, K.A. & Sugg, W.C. DNA-based population inventories. - Molecular Ecology (Eds.) 1987: Grizzly bear compendium. - Interagency Griz- 12: 1375-1387. zly Bear Committee, Washington, D.C., USA, 540 pp. Paetkau, D.W. & Strobeck, C. 1996: Mitochondrial DNA and Mace, R.D. & Waller, J.S. 1998: Demography and popula- the phylogeography of Newfoundland black bears. - Cana- tion trend of grizzly bears in the Swan Mountains, Montana. dian Journal of Zoology 74: 192-196. - Conservation Biology 12: 1005-1016. Paetkau, D.W., Shields, G.F. & Strobeck, C. 1998: Gene MacHutchon, A.G., Himmer, S. & Bryden, C.A. 1993: flow between insular, coastal and interior populations of Khutzeymateen Valley grizzly bear study: Final Report. - brown bears in Alaska. - Molecular Ecology 7: 1283-1292. Ministry of Environment, Lands, and Parks. Report No. R- Pelton, M.R. 1983: Black Bears. - In: Chapman, J.A. & Feld- 25, Victoria, British Columbia, Canada, 105 pp. hammer, G.A. (Eds.); Wild mammals of North America: Manning, T., Edge, W.D. & Wolff, J.O. 1995: Evaluating pop- biology, management, economics. John Hopkins university ulation-size estimators: an empirical approach. - Journal of Press, Baltimore, USA, pp. 504-514. Mammalogy 76: 1149-1158. Poole, K.G., Mowat, G. & Fear, D.A. 2001: DNA-based pop- Martinka, C.J. 1974: Population characteristics of grizzly ulation estimate for grizzly bears in northeastern British Co- bears in Glacier National Park, Montana. - Journal of Mam- lumbia, Canada. - Wildlife Biology 7: 65-75. malogy 55: 21-29. Proctor, M.F., McLellan, B.N. & Strobeck, C. 2002: Population Mattson, D.J. & Reinhart, D.P. 1995: Influences of cutthroat fragmentation of grizzly bears in southeastern British Co- trout (Oncorhynchus clarki) on behaviour and reproduction lumbia, Canada. - Ursus 13: 153-160. of Yellowstone grizzly bears (Ursus arctos), 1975-1989. - Schoen, J.W. & Beier, L.R. 1990: Brown bear habitat pref- Canadian Journal of Zoology 73: 2072-2079. erences and relationships to logging and mining in Southeast McLellan, B.N. 1989: Dynamics of a grizzly bear population Alaska. - Research Final Report 4.17, Alaska Department during a period of industrial resource extraction. I. Density of Fish and Game, Anchorage, USA, 90 pp. and age-sex composition. - Canadian Journal of Zoology Schoen, J.W., Flynn, R.W., Suring, L.H., Titus, K. & Beier, 67: 1856-1860. L.R. 1994: Habitat-capability model for brown bear in McLellan, B.N., Hovey, F.W., Mace, R.D., Woods, J.G., southeast Alaska. - International Conference on Bear Re- Carney, D.W., Gibeau, M.L., Wakkinen, W.L. & Kasworm, search and Management 9: 327-337. W.F. 1999: Rates and causes of mortality of grizzly bears Smallwood, K.S. & Schonewald, C. 1996: Scaling population in the interior mountains of British Columbia, Alberta, density and spatial pattern for terrestrial, mammalian car- Montana, Washington, and Idaho. - Journal of Wildlife Man- nivores. - Oecologia 105: 329-335. agement 63: 911-920. Taberlet, P., Camarra, J-J., Griffin, S., Uhres, E., Hanotte, O., Miller, S.D., White, G.C., Sellers, R.A., Reynolds, H.V., Waits, L.P., Dubois-Paganon, C., Burke, T. & Bouvet, J. Schoen, J.W., Titus, K., Barnes, V.G., Jr., Smith, R.B., 1997: Noninvasive genetic tracking of the endangered Nelson, R.R., Ballard, W.B. & Schwartz, C.C. 1997: Brown Pyrenean brown bear population. - Molecular Ecology 6: and black bear density estimation in Alaska using radio- 869-876. telemetry and replicated mark-resight techniques. - Wildlife Taylor, M. (Ed.) 1994: Density-dependent population regulation Monograph 133, 55 pp. of black, brown, and polar bears. - International Conference Mowat, G. & Strobeck, C. 2000: Estimating population size on Bear Research and Management Monograph Series of grizzly bears using hair capture, DNA profiling, and mark- No. 3, 43 pp. recapture analysis. - Journal of Wildlife Management 64: Wielgus, R.B. & Bunnell, F.L. 1994: Dynamics of a small, 183-193. hunted brown bear population in southwestern Alberta, Nagy, J.A.S. & Haroldson, M.A. 1989: Comparisons of some Canada. - Biological Conservation 67: 161-166. home range and population parameters among 4 grizzly bear Wielgus, R.B., Bunnell, F.L., Wakkinen, W.L. & Zager, P.E. populations in Canada. - International Conference on Bear 1994: Population dynamics of Selkirk mountain grizzly Research and Management 8: 227-235. bears. - Journal of Wildlife Management 58: 266-272. Nagy, J.R. & Gunson, J.R. 1990: Management plan for griz- White, G.C., Andersen, D.R., Burnham, K.P. & Otis, D.L. zly bears on Alberta. - Wildlife Management Planning 1982: Capture-recapture and removal methods for sampling Series Number 2, Alberta Forestry, Lands and Wildlife, Fish closed populations. - Los Alamos National Laboratory and Wildlife Division, Edmonton, Alberta, Canada, 164 pp. LA-8787-NERP, Los Alamos, New Mexico, USA, 235 pp.

© WILDLIFE BIOLOGY · 11:1 (2005) 47 White, G.C. & Burnham, K.P. 1999: Program MARK: Sur- Council on Ecological Areas, Occasional Paper No. 14, Otta- vival estimation from populations of marked animals. - Bird wa, Ontario, 95 pp. Study 46 (Suppl.): 120-138. Woods, J.G., Paetkau, D., Lewis, D., McLellan, B.N., Proctor, Wiken, E.B. (Compiler) 1986: Terrestrial ecozones of Canada. M. & Strobeck, C. 1999: Genetic tagging free ranging black - Ecological Land Classification Series No. 19. Environment and brown bears. - Wildlife Society Bulletin 27: 616-627. Canada, Hull, Quebec, 26 pp. Young, B.F. & Ruff, R.L. 1982: Population dynamics and Wiken, E.B., Gauthier, D., Marshall, I.B., Lawton, K. & Hir- movements of black bears in east-central Alberta. - Journal vonen, H. 1996: A perspective on Canada’s ecosystems: an of Wildlife Management 46: 845-860. overview of terrestrial and marine ecozones. - Canadian

48 © WILDLIFE BIOLOGY · 11:1 (2005) PAPER VIII

BIOLOGICAL CONSERVATION 128 (2006) 109– 119

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/biocon

Determining the spatial scale for conservation purposes – an example with grizzly bears

Vilis O. Namsa,*, Garth Mowatb, Michael A. Panianc aDepartment of Environmental Sciences, Nova Scotia Agricultural College, Box 550, Truro, NS, Canada B2N 5E3 bAurora Wildlife Research, RR 1, Site 14, Comp 8, Crescent Valley, BC, Canada V0G 1H0 cWildlife Branch, Ministry of Environment, Lands and Parks, 401-333 Victoria St. Nelson, BC, Canada V1L 4K3

ARTICLE INFO ABSTRACT

Article history: This study suggests procedures for determining the spatial scale for conservation guide- Received 7 June 2004 lines for animals, giving an illustration with an analysis of grizzly bear habitat selection. Received in revised form Bear densities were sampled by identifying hairs at bait stations in British Columbia. Hab- 14 February 2005 itat variables were measured using remote sensing. Spatial scale was changed by varying Accepted 20 September 2005 the window size over which the variables were averaged. First, the spatial pattern of bears Available online 2 November 2005 was studied, measuring the patchiness in bear densities at a variety of spatial scales, by calculating the correlation in bear densities between adjacent windows. This was repeated for Keywords: the habitat variables. Finally, the overall interaction between bears and habitats was ana- Ursus arctos lysed, measuring the strength of habitat selection at different spatial scales. There are Spatial scale three domains of scale: at 2–4 km, bears and habitats are patchy, at 5–10 km, bears select Habitat selection for habitats, and at 40+ km, habitats are patchy and bears select for habitats. At scales of Heterogeneity 40+ km, bears selected for: (i) higher slopes, or (ii) higher slopes, and some combination Patchiness of more avalanche chutes, fewer roads and trees, higher elevations, and less logged land. Hierarchical selection Within 15 km areas, bears selected for 6 km areas that are either at higher elevations, or at higher elevations and had fewer trees. The relationship of conservation guidelines at different spatial scales should be determined by measuring and comparing hierarchical to non-hierarchical selection. The scales that bears select for habitats roughly correspond to the scales used in present grizzly bear conservation plans in British Columbia. 2005 Elsevier Ltd. All rights reserved.

1. Introduction (i) at which scales are populations most patchy, (ii) at which scales are habitats most patchy, and (iii) at which scales do One of the cornerstones to animal conservation is under- animals select habitats most strongly? Habitat selection stud- standing the relationship between animals and habitat. It is ies that incorporate spatial scale typically address the last difficult to study this experimentally, especially for large ani- question only. mals, and thus we often measure habitat selection, assuming Such studies have to make several choices. The first is that it implies conservation value. One complication is that which spatial scales to use for measuring habitat – whether animals may not view habitats at the same spatial scales that to use biologically relevant scales (Loyn et al., 2001; Luck, biologists do. Thus a key question is what is the appropriate 2002a) or arbitrarily chosen scales at equidistant spatial inter- spatial scale for conservation guidelines. A first step to vals (Mitchell et al., 2001; McGrath et al., 2003). The second is addressing this would be to answer the following questions: which scales to use for measuring animal use-whether to use

* Corresponding author: Tel.: +1 902 893 6607; fax: +1 902 893 1404. E-mail address: [email protected] (V.O. Nams). 0006-3207/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.biocon.2005.09.020 110 BIOLOGICAL CONSERVATION 128 (2006) 109– 119

just one fixed spatial scale (Mitchell et al., 2001; Hatten and concentrations such as natural burns (Zager et al., 1983), Paradzick, 2003), or to vary animal scale with habitat scale logged areas (Zager et al., 1983; Waller and Mace, 1997), and (McLoughlin et al., 2002). The final choice is how to relate upper elevation forests (Hamer et al., 1991; Mace et al., 1996). the habitat variables at different scales – whether to combine Most of these studies have analysed grizzly bear habitat selec- habitat variables from all spatial scales into one relationship tion for the purpose of directing conservation effort, but they (Apps et al., 2004), whether to produce a uniquely different have focused on predicting bear abundance, not which are model for each scale (Hatten and Paradzick, 2003), or whether the most appropriate scales to use in conservation guidelines. to use all variables at each scale in order to compare strength These studies do not give a scientific basis for choice of of selection across scales (McGrath et al., 2003). scale for conservation guidelines. For example, in British We suggest that the following choices be made. First, habi- Columbia, residence of 25% of the North American grizzly bear tats should be measured at many spatial scales, spread out population (Banci et al., 1995), the spatial scales used in con- equidistantly over a wide spatial range. They should not be re- servation plans have been based on planning needs and ex- stricted to biologically relevant scales because it is often difficult pert perceptions of pertinent scales of importance for to measure these. For example, one common scale that has grizzlies (Kootenay Inter-Agency Management Committee, been used is home range size (McLoughlin et al., 2002; Zabel 1997), not on the results of specific scientific studies. The pur- et al., 2003). However some animals use multi-scaled hierarchi- pose of our study is to measure the appropriate spatial scale cal home ranges, and some animals do not even have asymp- for conservation guidelines of grizzly bears. We do this by find- totic home ranges (Gautestad and Mysterud, 1995). It takes ing at which scales grizzly bear populations are most patchy, little extra effort to analyse additional spatial scales – the largest at which scales are their habitats most patchy, and at which effort is gathering the data and initially setting up the analysis scales do grizzly bears respond to their habitats most strongly. algorithms. Furthermore, it is important to both measure and We then analyse habitat selection parameters at those scales. compare, hierarchical versus non-hierarchical selection. Second, animal use should be analysed at the same scales 2. Methods that habitat data are. A biological reason for doing this, is if an animal perceives a certain area of land as one piece of habitat, 2.1. Field methods then you would expect a similar use throughout that area. Differences in use throughout that area would be due to other Our study area was located in south-eastern British Columbia, factors, not the type of habitat. A statistical reason, is that Canada (51 Lat, 117 Long), and was dominated by the Selkirk variation among animal use among small areas is composed Mountains, with many peaks exceeding 2400 m. The majority of two parts: variation within large areas, and variation of the area was forested. Western red cedar (Thuja plicata) and among large areas – but we are only interested in variation western hemlock (Tsuga heterophylla) typically characterised among large areas. The variation within large areas would de- low elevation forests, while Englemann spruce (Picea engle- crease the power of our tests. manni) and subalpine fir (Abies lasiocarpa) forests were domi- Third, our measure of selection should be comparable nant above about 1400 m. Treeline was at 2000 m. among spatial scales. Thus all variables should be included Habitat use data were collected as part of a population in each model at each scale, and precision and accuracy should estimation study that used hair capture and DNA profiling not vary with scale. For example, if habitat and animal data as the individual mark in a mark-recapture design (Woods gathered at small spatial scales are averaged for the larger et al., 1999; Mowat and Strobeck, 2000). This sampling method scales, then the number of samples used to calculated the has been used to estimate population size for several large means in each window increases with window size. This areas of British Columbia and Alberta (Woods et al., 1999; Mo- decreases sampling error, increasing precision. Furthermore, wat and Strobeck, 2000; Poole et al., 2001). This technique sampling errors in independent variables create a bias in linear samples a large number of different bears over thousands of models (Neter et al., 1996), and the bias would change with spa- square kilometres, and although the sampling design is lar- tial scale. These effects of sampling errors should be removed gely chosen around the constraints of population estimation, in the analysis. These issues have not been discussed before. it is also useful in examining habitat selection. We illustrate these issues in an analysis of spatial scale of The bear population was sampled between June 19 and habitat use of grizzly bears (Ursus arctos). Grizzly bears select August 14 of 1996. Mean density of bears was 0.0266 (95% habitats at various spatial scales, but it is not known which c.i. of 0.0227–0.0317) bears/km2 (Mowat and Strobeck, 2000). scales are most important, and which should be used in con- The 9866 km2 study area was partitioned into 8 · 8 km cells servation. At landscape scales, larger than individual home with every second cell sampled in a checker board fashion. ranges, bears avoid roads (Archibald et al., 1987; Apps et al., One capture site was installed in each of 76 cells for approxi- 2004), select against forest (Mace and Waller, 1997a), but for mately 10 days. Capture sites were baited with rotten meat, the amount of deciduous vegetation (Mace et al., 1999). Within and barbed wire was used to catch hair from bears. DNA home ranges, bears still avoid roads and human habitation profiling of the hairs was used to identify individuals and spe- (Clevenger et al., 1992, 1997), but habitat use changes through- cies. Each site was visited once at the end of the 10-day trap- out the season. During spring and early summer, bears in ping session and removed. The next site was installed mountainous areas use riparian areas and forest openings elsewhere in the same cell with the stipulation that new sites that are free of snow (McLellan and Hovey, 1993) and ava- be at least 1 km from all previous sites. Further details on field lanche chutes (McLellan and Hovey, 2001). As berries begin and lab methods can be found in Mowat and Strobeck (2000) to ripen in mid summer, bears move to areas with high berry and Woods et al. (1999). The animals were handled in accor- BIOLOGICAL CONSERVATION 128 (2006) 109– 119 111

dance with the principles and guidelines of the Canadian means that they respond to habitat type in a 4 · 4kmarea Council on Animal Care. (i.e. a 4 km window) rather than to a small block of habitat 4 km away (i.e. a lag distance of 4 km). Thus, whenever we re- 2.2. Habitat variables fer to a window size, or a spatial scale, we will refer to the length of one side. We chose habitat variables (Table 1) that were readily avail- We calculated window means differently for large and able from existing databases, that were accessible and inex- small scales. For larger spatial scales (3–40 km) we used pensive, that would directly relate to future GIS modelling square windows spaced 1/4 window-width apart, covering endeavours, and that were perceived to be important to griz- the whole study site. The overlapping of the windows mini- zly bears. An ArcInfo Geographic Information System (GIS) mised variation due to window placement. For smaller scales was used to convert data from each variable to a grid with a (0.5–7 km) we used round windows centred on bear sampling 50 · 50 m cell size. These values were then averaged to pro- sites. These windows were placed this way in order to avoid duce habitat values for windows of different sizes in our anal- missing data, because most small randomly located windows yses. The averaging also caused the resulting habitat contained no bear sampling sites, and those that did had usu- variables to be normally distributed. ally one, located anywhere inside the window. The overlap in ranges of spatial scales between the large and small scales (an 2.3. Overall strategy of analysis overlap in the range of 3–7 km) gave a check that the two techniques were giving similar results. Our interest was in analysing grizzly bear habitat use at dif- The analyses used all windows that were completely in the ferent spatial scales. We did this in several steps. First, we study area. To avoid violating the assumption of independence studied the spatial pattern of grizzly bear densities by them- caused by the overlapping of windows, we used an effective selves, measuring the patchiness in bear densities at a variety sample size based on the number of non-overlapping win- of spatial scales. Then we did the same thing for the habitat dows in the study site. For the larger scales this calculation variables. Then we studied the overall interaction between was the number of windows divided by 16, but for the smaller grizzly bears and habitats, measuring the strength of habitat scales this calculation involved an iterative process where a selection at a variety of spatial scales. Finally, using those spa- window was selected, then all windows overlapping with that tial scales at which grizzly bears show the strongest selection, one were removed, then the next closest non-overlapping we did a more detailed analysis of habitat selection. window was chosen, etc, until all windows were gone. This whole process was carried out 10 times, starting with different 2.4. Spatial scales initial windows. At each window size the analysis incorporated patterns At each spatial scale of interest we defined windows of a fixed over the whole range of spatial scales, from that size up to size, distributed throughout the study site. We calculated the whole study area. For example, bears may select for spe- average values for bear density and habitat variables within cific forage at a small scale, prefer more open coverage at a each window, and varied spatial scale by varying the window medium scale, and avoid human disturbance at a large scale. size over which the variables were averaged. This differs from All of these behaviours affect bear density at a small the methods traditionally used in geospatial statistical analy- scale, and in order to isolate the effects at one specific spatial ses (Rossi et al., 1992), such as variograms and correlograms. scale, we need to remove the effects of the larger scales. There, one calculates various statistics using windows of Thus we also carried out analyses where we removed the fixed sizes, but varying the lag distances between windows. effects of larger scales, by calculating deviations from means Instead, we varied window size. We did this because it better of larger windows. This is analogous to partitioning the sums- describes what we mean when we consider how animals re- of-squares in an analysis of variance (Sokal and Rohlf, 1995; spond to habitats at different spatial scales. For example, if Neter et al., 1996). Our aim was to have a 2· range for each bears respond to habitats at a spatial scale of 4 km, this spatial scale – for example, values in 1 km windows were

Table 1 – Habitat parameters analysed and the sources of the data

Habitat parameters Units Source

Elevation Meters TRIMa Slope Percentage TRIM Distance to water Meters TRIM Distance to roadsb Meters TRIM Distance to human development Meters TRIM cultural, and private land Logged land Proportion of area B.C. Forest Cover Mapping and Forest Development Plans Treed land Proportion of area Landsat unsupervised classiﬁcation Avalanche chutes Proportion of area Landsat unsupervised classiﬁcation and Forest Cover Mapping

In the analysis, the negative of the distance parameter was used. The areas were transformed to a proportion of the window area. a TRIM represents ‘‘Terrain Resource Information Management’’ (Balser, 1987), a public access GIS data source. b This includes all roads, but most roads in the study site were secondary logging roads 112 BIOLOGICAL CONSERVATION 128 (2006) 109– 119

subtracted from the means of 2 km windows (for notation, used the numbers of independent pairs of windows (non- this was called the 1–2 km spatial scale) – this removed all overlapping), not the total numbers of pairs (overlapping). patterns at spatial scales larger than 2 km. For a measure of bear density we used the proportion of This use of deviations from a local mean has a couple of sites visited by grizzly bears within the window. benefits. First, it models the hierarchical manner in which it has been suggested (Johnson, 1980; Hutto, 1985; Sallabanks, 2.6. Causes of bear patchiness 1993) that animals select habitats – that animals first select home ranges, and then select habitats to use within those We also tested whether patchiness in bear populations was home ranges. In other studies using radio-tracking data, hier- caused by the same individual bears being found throughout archical selection has been analysed by first measuring home each patch. We did this by estimating the similarity in indi- range selection, by comparing home range habitats to habi- viduals between pairs of windows – i.e., to see if the same tats in the overall study area; then by measuring habitat use bears tended to use the same patch. For example, a peak in within home ranges, by comparing habitat use of individual similarity at small scales but not the larger ones would sug- animals to habitats available in each home range (Aebischer gest that the small-scale patchiness is due to grouping, but et al., 1993; Sallabanks, 1993). that the larger scale patchiness is not. The second benefit is that using local deviations mini- To test for this, we used the identifications of individual mises the effects of autocorrelations – since deviations from bears (by DNA profiling) and calculated a variation of JaccardÕs a local mean are less autocorrelated than deviations from Coefficient of species similarity (typically used in comparing an overall mean. communities; Krebs, 1999): One potential problem when analysing habitat use at dif- # bears common to both ferent spatial scales is that sampling error changes with spa- S ¼ . ð1Þ # of individual bears in both tial scale, since the number of samples used to calculated the means per window, increase with window size. Sampling er- This similarity measure ranges from 0, when completely dif- rors typically create a bias in linear models, and the bias ferent bears are found in each window, to 1, when exactly the changes with spatial scale. We removed the effects of sam- same bears are found in both windows. pling errors (Laws, 1997; Appendix A), thus eliminating this potential problem. 2.7. Spatial patterns in habitats

2.5. Spatial patterns in bear density We measured the scales of patchiness in habitats in a similar manner as the bear densities, using the same setup of pairs of In order to find the scales of patchiness in bear densities, we windows. This was done for each habitat variable separately, calculated the correlation in bear densities between adjacent and all together. As a measure of overall correlation between windows. If there is patchiness at only one spatial scale, then habitats of adjacent windows, we used the square root of the correlation should be positive when windows are smaller mean of correlations squared (we did not use canonical corre- than patches, negative when windows are approximately lation because estimates of it are inflated by low sample sizes, patch size, and zero when windows are larger than patch size. and in our situation, sample size changes with spatial scale). However if there is patchiness over a range of scales, then Thus heterogeneity is shown by a peak and drop in correla- there would be positive correlations over that range. This tion as spatial scale increases. analysis is similar to correlograms, as used in geospatial statistical analyses (Rossi et al., 1992), except that we varied win- 2.8. Habitat selection dow size instead of lag distance. ‘‘Scale’’ refers to the width of two windows, since correlation would peak when two win- We analysed the overall interaction between grizzly bears and dows are inside a patch. habitats by measuring the strength of habitat selection at a We carried out the procedures differently for large and variety of spatial scales. Unlike the previous analyses, at each small scales, since we had to use different types of window- spatial scale, each data point contains measurements of both ing procedures. For larger spatial scales (4–80 km), with the bears and habitats from one window, and we want to know square windows, we used vertical or horizontal pairs, centred how they correspond. Thus, rather than comparing adjacent anywhere on the study site. For smaller scales (1.5–20 km), windows, we analysed the relationship between bear density with the round windows centred on bear sampling sites, we and habitat within the same windows (of different varying used pairs of windows in any orientation. All possible pairs window sizes). ‘‘Scale’’ refers to the width of one window. of windows that were completely in the study site, and that We used general linear models in which our independent contained bear sampling sites, were used. variables were the habitats of the windows, and the response Note that patchiness in sampling sites does not affect our was bear density. In order to isolate the effects of the different results. This is because only windows that contained bear habitat variables we carried out an interactive step-wise sampling sites were used, and because our measure of bear regression, and at each step considered the relationships density was unbiased with respect to the numbers of sites. among the habitat variables themselves. The strength of the In order to estimate confidence intervals in the correla- overall relationship between bears and habitat, as modelled tions we used a bootstrapping technique (Efron and by the regression, was measured by a multiple correlation, Tibshirani, 1991). To avoid violating the assumption of with error estimates of it by the Fisher z and Hotelling z* independence, for the size of the bootstrapped samples we transformations (Sokal and Rohlf, 1995; Neter et al., 1996). BIOLOGICAL CONSERVATION 128 (2006) 109– 119 113

In order to see how these variables relate to each other, we considered their ÔToleranceÕ. This measures how independent one variable is from the others together and is estimated by 0.05 1 R2 of that variable against the rest of the habitat variables (Pedhazur, 1982). Tolerance ranges from 0, meaning that the variable is completely correlated with a linear combination of the others, to 1, meaning that is completely unrelated to all the others. The computer programs to select the pairs of windows and Proportion similar individuals calculate variance–covariance matrices were written in Math- 0.00 ematica (Wolfram, 1999), with the resulting matrices analysed 1 2 3 4 5 6 7 8 910 20 40 60 80 with the statistical package SYSTAT (Wilkinson et al., 1999). Spatial scale (km) Fig. 2 – Similarity in bear distributions between pairs of 3. Results windows, at varying distances. Values represent the proportion of bears that are the same between pairs of We monitored hair snares in 76 cells, approximately 5 times windows;a1represents all the same bears, and a 0 each, between June 19 and August 14, 1996, and retrieved hair completely different ones. Spatial scale is the width of a from 109 individual bears 134 times. Of the 381 sites, 23% suc- pair of adjacent windows. Dashed lines represent 90% cessfully snared grizzly bear hair. As many as 5 bears left hair confidence intervals. at a site; some multiple visits were females with cubs (data presented in Mowat and Strobeck, 2000). areas. At smaller scales the similarity is even lower – 0.02 of 3.1. Spatial patterns in bear density the bears found in areas 1 km apart are common to both areas. Thus the patchiness in bear numbers is not because Correlations in bear density between pairs of windows of the same individual bears using sites close together, nor decreased from 3 to 4 km (Fig. 1), suggesting that there is due to a grouping behaviour (e.g., family groups at small patchiness in bear density at small spatial scales (size 3– scales, or sub-populations at larger scales). 4 km). The confidence interval is too broad to tell if there are any significant patterns at larger spatial scales. 3.2. Spatial patterns in habitats The patchiness in bear distributions at small scales is not due to the spatial distribution of sampling sites, because the Correlations in habitat variables between pairs of windows analysis compares bear density only for pairs of windows that decreased from 2 to 4 km, then increased at 20 km (Fig. 3). contain sampling sites. This suggests that there is patchiness in habitat variables at We tested if it might be due to individual bears using sites small (2–4 km) and intermediate-large spatial scales (20– that are close together. The similarity in individual bears 80+ km). between pairs of habitat windows (Fig. 2) shows two important features. First, we do not see the peak at the small scale 3.3. Bear habitat selection that we saw in the bear patchiness results (Fig. 1). Second, similarities are extremely low at all scales – the largest is at Bears select habitat variables differently, depending on spatial a scale of 80 km, where it is still only 0.04. Thus only 0.04 of scale (Fig. 4), but overall, bears select most strongly for habi- the bears found in areas 40 km apart are common to both tats at two distinct spatial scales: 4–8 and 40+ km (Fig. 5), with selection being strongest at the large scale of 40+ km. We will thus concentrate our more detailed analysis at these two

1.0 spatial scales. Bias due to selection of bear sampling sites is minimal (Appendix B). Note that Fig. 5 shows selection with effects at larger scales removed. For example, the peak at 6 km means that 0.5 within 15 · 15 km windows (approximately 2·), bears select

Correlation for areas of size 6 · 6 km – i.e., hierarchical selection. In order to answer the question of which scale best predicts bear 0.0 over the whole study area 13452 6 78910 20 40 60 80 distribution and habitat relationships , this analysis was also carried out without removing effects at larger spatial scales (Fig. 6). The increased correlation at large Spatial scale (km) scales shows that windows of size 40 · 40 km are most useful Fig. 1 – Patchiness in bear density over a range of spatial in predicting bear density from habitat information, over the scales. The values represent correlations in bear density whole study area. between adjacent windows. Spatial scale is the width of a pair of adjacent windows. The dashed lines represent 90% 3.3.1. Large spatial scale – 40+ km conﬁdence intervals. The larger the correlation the stronger Elevation, slope, roads, logged, trees, and avalanche chutes all the patchiness at that scale. Patchiness peaks at 3 km. individually correlate with bear density (Table 2), and many 114 BIOLOGICAL CONSERVATION 128 (2006) 109– 119

1.0

0.5

0.5 Correlation Correlation

0.0 1 2 345678910 20 40 60 80 0.0 1345678910203042 0 Spatial scale (km) Spatial scale (km) Fig. 3 – Patchiness in habitats at various spatial scales. The Fig. 5 – Localised habitat selection at various spatial scales. values represent correlations in all habitat variables It is localised, or hierarchical, in that deviations from a local between adjacent windows. Spatial scale is the width of a mean of about 2· window size, instead of from the global pair of adjacent windows. Dashed lines represent 90% mean, are used. The values represent correlations between conﬁdence intervals. The larger the correlation the stronger grizzly density and all habitat variables within windows of the patchiness at that scale. The left line in each graph that size. Spatial scale is the width a window. The larger the shows data from windows centred at bear sampling sites, correlation the stronger the selection at that scale. Habitat while the right line shows data from windows in a grid selection peaks at 4–8 and 40+ km. The values represent pattern. Note the increased habitat patchiness at small and correlations between grizzly density and all habitat large scales. variables within windows of that size. Dashed lines are 90% conﬁdence intervals. Left set of lines is based on windows centred on bear sampling sites, and right set is from correlate with each other (Table 3). Thus we carried out an regularly spaced windows. interactive step-wise regression, and at each step considered the relationships among the habitat variables themselves. Water has the highest tolerance, 0.47 (Table 2), showing these interrelated variables also correlate strongly with bear that it is the most independent of the rest of the habitat vari- density (Table 2). ables. The other seven have very low tolerances, showing that When we control for the effect of slope (the variable with they are related to each other, with elevation, slope, logged, the strongest relationship to bear density; Table 2), then the trees and avalanche chutes being the most related. Six of relationships between bear density and each of the other hab-

Elevation Slope Water Avalanches 0.5

0.0

-0.5

Trees Roads Human Logged 0.5 0.5 Correlation

0.0 0.0

-0.5 -0.5

1 5 10 20 3040 1 5 10 20 3040 1 5 10 20 3040 1 5 10 20 3040 Spatial scale (km)

Fig. 4 – Localised habitat selection at various spatial scales-individual habitat variables. It is localised in that deviations from a local mean of about 2· window size, instead of from the global mean, are used. The values represent correlations between grizzly density and each habitat variable within windows of that size. Spatial scale is the width a window. Dashed lines are 90% conﬁdence intervals. The larger the correlation the stronger the selection at that scale. The left line in each graph shows data from windows centred at bear sampling sites, while the right line shows data from windows in a grid pattern. A positive correlation shows selection for that habitat variable and a negative correlation shows selection against it. BIOLOGICAL CONSERVATION 128 (2006) 109– 119 115

1.0 the other hand, if we control for all of the other ﬁve variables together (elevation, roads, logged, trees, avalanche chutes), then the partial correlation for bear density vs slope still remains signiﬁcant. This means that the relationship between bear density and slope is more than just due to slope being 0.5 correlated with these other habitat factors that bears might

Correlation select. We can conclude that over the whole study site, bears select for 40 km areas on the basis of:

0.0 12 3 4 5 6 7 8 910 20 30 40 (i) slope-bears prefer higher slopes, or Spatial scale (km) (ii) slope, and some combination of more avalanche chutes, fewer roads and trees, higher elevations, and less logged Fig. 6 – Habitat selection over the whole study site, at land. various spatial scales. This shows which scale best predicts bear distribution:habitat relationships over the whole study site. Values represent correlations between grizzly density 3.3.2. Intermediate spatial scale and all habitat variables within windows of that size. The analysis at this scale used means taken from 6 · 6km Spatial scale is the width a window. The larger the windows. In addition, in order to remove effects from larger correlation the stronger the selection at that scale. Habitat scales, deviations were taken from 15 · 15 km windows rather selection is highest at 40+ km, thus windows of size 40+ km than from global means. Thus this measured hierarchical are most useful in predicting bear distribution and habitat selection. relationships. Dashed lines are 90% confidence intervals. Bear density is most correlated with elevation and trees Left set of lines is based on windows centred on bear (Table 2), and also significantly correlated (but less so) with sampling sites, and right set is from regularly spaced avalanche chutes, logged, and humans. The smaller windows. tolerances for elevation and trees show that these variables are related to each other, but the rest of the variables are quite independent of each other. itat variables in that group of six disappear. This shows that Controlling for elevation, the significant relationship be- none of them has any effect on bear density that is statisti- tween bears and trees disappears. But controlling for trees, cally different from, or larger than, the effect of slope. On then there is still a significant relationship between bears

Table 2 – Correlations of individual habitat variables with bear density, at the 40+ and 6–15 km spatial scales Habitat 40+ km scale 6–15 km range of spatial scales Correlation P-valuea Toleranceb Correlation P-valuea Toleranceb

Elevation 0.62 0.01 0.08 0.37 0.001 0.39 Slope 0.88 0.00 0.07 0.04 0.57 0.77 Water 0.12 0.66 0.47 0.07 0.32 0.89 Roads 0.70 0.003 0.17 0.12 0.10 0.69 Human 0.43 0.10 0.22 0.14 0.06 0.78 Logged 0.52 0.04 0.13 0.15 0.05 0.80 Trees 0.72 0.002 0.08 0.30 0.001 0.42 Avalanche chutes 0.84 0.001 0.11 0.16 0.04 0.73

a P-value measures the signiﬁcance level for a non-zero correlation. b Tolerance measures how independent that variable is from all of the rest of them.

Table 3 – Correlations of individual habitat variables against each other, at the 40+ km spatial scale Slope Elevation Water Roads Human Logged Tree Avalanche chutes

Slope 1 Elevation 0.83 1 Water 0.00 0.30 1 Roads 0.75 0.79 0.18 1 Human 0.55 0.66 0.19 0.78 1 Logged 0.65 0.84 0.47 0.68 0.46 1 Tree 0.60 0.68 0.27 0.76 0.56 0.64 1 Avalanche chutes 0.83 0.59 0.05 0.72 0.62 0.37 0.52 1 116 BIOLOGICAL CONSERVATION 128 (2006) 109– 119

and elevation. Thus within 15 km areas, bears select for 6 km ies of the spatial scale aspects of animal:habitat relationships areas that are at either: (i) at higher elevations, or (ii) at higher have either used naturally occurring divisions of spatial scale elevations and have fewer trees. We cannot tell which it is, or (e.g. Diffendorfer et al., 1995; Mace et al., 1996; Chapin et al., if the second, how much bears select for elevation vs trees. 1997), or have arbitrarily chosen a few specific spatial scales Bear density is not related to the other seven variables. Over- to study (e.g. Bowers et al., 1996; Pedlar et al., 1997; Timoney, all, bears select for habitats much less strongly at this scale 1999). For example, Apps et al. (2004) developed predictive than at the 40+ km scale (an R of 0.38 vs 0.88 at 40+ km). functions of grizzly bear distribution north of our study site, testing 3 specific spatial scales (diameters of 5, 13 and 4. Discussion 22 km). However we found that habitat selection increased sharply at larger scales than they tested (at 40+ km; Fig. 5). 4.1. Spatial scale Perhaps Apps et al. (2004) would have found stronger or different predictive equations had they tested a larger spatial scale. Processes in ecological systems change with spatial scale, but Second, the study should allow comparisons of strength of they do not always change smoothly. Wiens (1989) proposed selection across the different spatial scales. This is difficult that there are regions (called ÔdomainsÕ) of scale, within which when different types of data are used – for example, some patterns do not change, and these domains are separated by studies measure habitats selection by using locations of indi- sharp transitions as different processes become important vidual animals at a small scale, vs whole home ranges at a at larger scales vs smaller scales. In our study we saw pat- large scale. Even when the data is collected in the same terns at three distinct spatial scales. Since the relationships way, the analysis must ensure that results can be compared between bears and habitat are qualitatively different in each – we had to remove the effects of sampling variation (Appen- of these regions, this suggests that these are distinct do- dix A) because they depended on spatial scale. mains, and that therefore different aspects of bear biology Third, measuring habitat patchiness helps in assessing the are important in each. applicability of predictive models of distribution, to other However, we see an interesting and unexpected feature in study sites. The best way to test them is to use independent our results. Bear densities and habitats are patchy at all spa- data sets (e.g. Luck, 2002b) – but often these are not available. tial scales ranging from 1 to 80 km, and there is significant In our study, the question is, do grizzly bears in other areas non-hierarchical habitat selection at all scales. There are var- select habitat at the same spatial scales? The underlying bio- ious ways this could occur. logical question is whether selection at these scales is inher- First, different categories of bears (e.g., ages, sexes, indi- ent, or is it a result of the spatial distribution of habitats? The viduals) could respond at different spatial scales. This is pos- habitat patchiness results helps us to answer this. Bears se- sible, because males and females have different home range lected most strongly at scales of 6 and 40 km, however habi- sizes, and females with small cubs select habitat in order to tats were patchy only at the 40 km scale. Thus selection at avoid males (Nagy and Haroldson, 1989; Mace and Waller, 6 km was likely not caused by habitat spatial distribution, 1997b), likely because strange males will kill cubs (Swenson but at 40 km it might have been. et al., 1997). Finally, studies include both hierarchical and non-hierar- Second, bears could change their scale of patchiness over chical selection. For example, grizzly bears selected for areas time. This too is possible, because grizzly bears tend to be of width 6 km that are at higher elevations than the surround- confined to snow-free greened-up areas early in spring, with ing 15 km, but bears did not show a peak in non-hierarchical their choices increasing as the season progresses (Jonkel, selection at that scale. Other animals have also been shown 1987; McLellan and Hovey, 1993); they also increase home to select hierarchically: caribou select types of patches range size when mating (Dahle and Swenson, 2003). depending on the type of landscape they are in (Johnson Finally, the idea of the domains being distinct, might be et al., 2004) and eagle owls select nest sites after selecting too simplistic-while there are some aspects of bear:habitat home range sites (Martinez et al., 2003). Other animals have biology that are more important at some spatial scales, there also been shown to select non-hierachically: mule deer se- might be other aspects that change continuously with spatial lected for patch size at small scales but not at larger scale scale. That is, that grizzly bears might actually view their (Kie et al., 2002). Vaughan and Ormerod (2003) stressed that world at a continuum of spatial scales. Analogously, Hoffman distribution models for conservation should include spatial and Wiens (2004) discovered that tenebrionid beetle species hierarchies. However no previous study has measured both richness and some habitats variables showed positive covari- types of scalar selection, nor compared them. ance at all spatial scales. Perhaps the distinctness of domains It is important to both measure and compare, hierarchical of scale needs more tests. versus non-hierarchical selection, for two reasons. First, ignoring hierarchical selection might lose important informa- 4.2. Conservation applications tion. Grizzly bears in our study did not show non-hierarchical selection at 6 km – one might conclude that selection at this We highlight several useful principles for evaluating the scale is unimportant, if hierarchical selection was ignored. choice of scale in forest-use guidelines. First, it is important Second, the type of selection determines how to relate to identify the main domains using the behaviour of the ani- conservation guidelines at different spatial scales. If non- mals themselves, before carrying out more detailed analyses hierarchical selection is most important, then guidelines at in those domains. When doing this a wide range of spatial the smaller scale can be independent of where the smaller scales should be used. This is typically not done – many stud- units are located. However, if animals select hierarchically BIOLOGICAL CONSERVATION 128 (2006) 109– 119 117

then conservation guidelines should be defined in the context changes with spatial scale (more samples taken from larger of the larger scale. For example, within 40 km areas, bears se- windows), causing the bias to change with scale. For example, lect for areas of locally high altitude. Guidelines for habitat the measure of bear density, the proportion of sites visited by preservation based on altitude should thus be based on the bears, is more variable at smaller scales because there are altitude of surrounding areas. This is an important topic that fewer sites in smaller windows. This might decrease the esti- has not been previously addressed. mated correlation between windows. The bias can be adjusted if the observation error can be 4.3. Grizzly bear conservation calculated separately (Rayner, 1985). We removed the effect of observational error by an adaptation of the methods of Present grizzly bear conservation plans in British Columbia Laws (1997). consider guidelines at the stand level (2500 ha on the average) Suppose for windows of a certain size, we calculate mean and multiple landscape unit level (200,000 ha on the average), bear densities in each window based on nB samples, and but the choice of scale has been based on planning needs and calculate habitat values based on nH samples. expert perceptions of pertinent scales of importance for griz- Suppose Bi = mean bear density in the ith window; rA2 ¼ zlies (Kootenay Inter-Agency Management Committee, 1997). B variance in bear density among windows; rW 2 ¼ Grizzly bears in our study area selected for habitats at two B variance in bear density among samples within a distinct spatial scales: at 4–8 km (1600–6400 ha) and 40+ km window; r (160,000+ ha), which roughly correspond to the ‘‘stand’’ and ABH = covariance of bear densities and habitat values among ‘‘multiple landscape unit’’ levels used in conservation plans. windows; then the variance among mean bear densities is gi- It is clear that our large scale results support the rele- ven by vance of the ‘‘multiple landscape unit’’ level in conservation, r 2 ð Þ¼r 2 þ WB ð Þ but it is less clear how our intermediate scale results relate Var Bi AB 2 nB to the ‘‘stand’’ level. This is because there are two different types of results at the intermediate spatial scale. Windows and a similar relationship exists for the habitat means. This of size 4–8 km are most useful in predicting bear density shows that the variance among mean bear densities is within 15 · 15 km areas (e.g. Fig. 5), but not over the whole inflated by the observational error (the second term). Of study site (e.g. Fig. 6). It is not clear which of these types course, the more samples we have in each window (larger of analyses is most useful for conservation guidelines, and nB), the smaller the effect of observational error. more research is needed to clarify the situation. It would However, the covariance among bear and habitat values is be useful for future studies to explore the nature of hierar- given by chical selection. For example, we measured selection within CovðBi; HiÞ¼rABH. ð3Þ a2· width area. With a larger sample size, it would be useful to vary the sizes of larger areas – and do this for a varying This means that the covariance among bear and habitats is sizes of core areas. not affected by the observational error. Consequently, the esti- An analytical approach emphasising scale as in this study mated correlation between bears and habitats is biased may provide justification for choosing appropriate conserva- downwards, because the variances are inflated but the covari- tion scales and serve to identify habitat parameters most sui- ance is not. If we now increase spatial scale, the numbers of ted for consideration at those scales. samples per window increases, decreasing the effect of observational error, resulting in an increase in the correlation – and Acknowledgements this increase is simply due to the effects of scale on sampling error, not due to any biological relationship. This study was funded by Forest Renewal BC. Parks Canada We can remove the effect of the observational error by kindly supported the DNA analysis. We thank P. Cutts, B. first estimating it; in each window, the observational error Hughson, J. Woods, M. Proctor, D. Paetkau, B. McLellan, K. term is the standard error squared. For an overall estimate Stalker, D. Fear, K. Robbins, S. Hawes, B. Park, A. Walker, M. of the term, we used the mean of the standard errors within Marello, S. Petrovic, M. Robbins, M. Robbins, J. Ivans, A. Chris- each window. We then subtracted this from the overall esti- tie, K. Shave, C. Davis, J. Bonneville and R. McClery. D. Wil- mated variance, to get an estimate for the variance among liams, D. Wassick, M. Homis, and K. Shier flew helicopters windows. We did this for bear densities and all habitat for field crews. We thank G. Smith, P. Rainville and V. Johnson variables. for preparing data and maps. We also thank Don Reid for Note that this procedure does not transform the means, reviewing the manuscript. but the variance–covariance matrix. We then used the transformed variance–covariance matrix in subsequent analyses. Appendix A. Removing the effects of observation error Appendix B. Hair capture station habitats

At each spatial scale, our measures of animal density and One potential problem in our analysis at smaller spatial scales habitat vary due to two components: natural variability was that the bear sites were placed nonrandomly within each through space, and observational error. Using type I linear 8 · 8 cm cell. They were placed where the technicians thought models in such a situation can create a bias (Sokal and Rohlf, bears were most likely to be found (this was done because the 1995; Neter et al., 1996). Even worse, the observational error primary focus in the study was estimating bear density (Mowat 118 BIOLOGICAL CONSERVATION 128 (2006) 109– 119

and Strobeck, 2000), while measuring habitat use was second- Chapin, T., Harrison, D., Phillips, D., 1997. Seasonal habitat ary). Thus it is important to know how this affects our analysis. selection by marten in an untrapped forest preserve. Journal of For each spatial scale sampled, we measured overall habi- Wildlife Management 61, 707–717. Clevenger, A.P., Purroy, F.J., Campos, M.A., 1997. Habitat tat differences between bear sampling sites and the study assessment of a relict brown bear (Ursus arctos) population in area, as follows: northern Spain. Biological Conservation 80, 17–22. sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X 2 Clevenger, A.P., Purroy, F.J., Pelton, M.R., 1992. Brown bear (Ursus 1 b s i i ; ð4Þ arctos 1) habitat use in the Cantabrian Mountains, Spain. n si Mammalia 56, 203–214. Dahle, B., Swenson, J.E., 2003. Seasonal range size in relation to where n is the number of habitat variables (here, n = 9); bi, reproductive strategies in brown bears Ursus arctos. Journal of mean of the ith habitat variable at bear sampling sites; s , i Animal Ecology 72, 660–667. mean of the ith habitat variable over the whole study area. Diffendorfer, J.E., Gaines, M.S., Holt, R.D., 1995. Habitat This is the mean Euclidean distance between the habitats fragmentation and movements of three small mammals of the bear sampling sites and the whole study site, but stan- (Sigmodon, Microtus, and Peromyscus). Ecology 76, 827–839. dardised in order to combine habitats with very different Efron, B., Tibshirani, R., 1991. Statistical data analysis in the scales. The measure ranges from 0 to 1. computer age. Science 253, 390–395. Gautestad, A.O., Mysterud, I., 1995. The home range ghost. Oikos Habitats of bear sampling sites differed most from the 74, 195–204. whole study site at a spatial scale of 1 km, with the effect Hamer, D., Herrero, S., Brady, K., 1991. Food and habitat used by decreasing at larger scales. The bear sites primarily differed grizzly bears along the continental divide in Waterton Lakes in that the slope was shallower, there was more water, more National Park, Alberta. Canadian Field Naturalist 105, of the area was logged, and there were more avalanche 325–329. chutes. Habitats in windows of size 4 km, or larger, around Hatten, J.R., Paradzick, C.E., 2003. A multiscaled model of each bear site were effectively the same as those found southwestern willow flycatcher breeding habitat. Journal of Wildlife Management 67, 774–788. throughout the study site. Thus this biased placement should Hoffman, A.L., Wiens, J.A., 2004. Scaling of the tenebrionid beetle have little effect at scales 4 km or larger. community and its environment on the Colorado shortgrass The biased placement at scales below 4 km strengthened steppe. Ecology 85, 629–636. some of our results. The biased analysis effectively uses a Hutto, R.L., 1985. Habitat selection by nonbreeding migratory land subset of more preferred habitats, and within those, com- birds. In: Cody, M.L. (Ed.), Habitat Selection in Birds. Academic pares those habitats that bears used vs did not use. This Press, Orlando, Florida, pp. 455–476. would make it more difficult to detect habitat selection – Johnson, C.J., Seip, D.R., Boyce, M.S., 2004. A quantitative approach to conservation planning: using resource selection making our results more conservative. We kept the smaller functions to map the distribution of mountain caribou at spatial scales in our analysis because while the bias made multiple spatial scales. Journal of Applied Ecology 41, 238–251. the reasons for nonsignificant habitat selection at small spa- Johnson, D.H., 1980. The comparison of usage and availability tial scales less certain, it made the significant patchiness in measurements for evaluating resource preference. Ecology 61, bear density more certain. 65–71. Jonkel, C.J., 1987. Brown bears. In: Novak, M., Baker, J.A., Obbard, M.E., Malloch, B. (Eds.), Wild Furbearer Management and Conservation in North America. Ministry of Natural Resources, REFERENCES Ontario, pp. 456–473. Kie, J.G., Bowyer, R.T., Nicholson, M.C., Boroski, B.B., Loft, E.R., 2002. Landscape heterogeneity at differing scales: Effects on Aebischer, N.J., Robertson, P.A., Kenward, R.E., 1993. spatial distribution of mule deer. Ecology 83, 530–544. Compositional analysis of habitat use from animal Kootenay Inter-Agency Management Committee, 1997. Grizzly radio-tracking data. Ecology 74, 1313–1325. Bear Management Guidelines. In: Kootenay Boundary Land Apps, C.D., Mclellan, B.N., Woods, J.G., Proctor, M.F., 2004. Use Plan Implementation Strategy (Chapters 3, 4). Ministry of Estimating grizzly bear distribution and abundance relative to Environment, Lands and Parks, British Columbia. habitat and human influence. Journal of Wildlife Management Krebs, C.J., 1999. Ecological Methodology, second ed. Harper & 68, 138–152. Row, New York. Archibald, W.R., Ellis, R., Hamilton, A.N., Responses of grizzly Laws, E., 1997. Mathematical Methods for Oceanographers. John bears to logging truck traffic in the Kimsquit River valley, B.C., Wiley & Sons, New York. 1987. International Conference on Bear Research and Loyn, R., McNabb, E., Volodina, L., Willig, R., 2001. Modelling Management 7, 251–257. landscape distributions of large forest owls as applied to Balser, R., 1987. Terrain resource information management managing forests in north-east Victoria, Australia. Biological (TRIM): a basis for geographic information systems Conservation 97, 361–376. applications in the province of British Columbia. In: 5th Luck, G.W., 2002a. The habitat requirements of the rufous Annual Northwest Conference on Surveying and Mapping, treecreeper (Climacteris rufa). 1. Preferential habitat use June 14–17. demonstrated at multiple spatial scales. Biological Banci, V., Demarchi, D.A., Archibald, W.R., 1995. Evaluation of the Conservation 105, 383–394. population status of grizzly bears in Canada. International Luck, G.W., 2002b. The habitat requirements of the rufous Conference on Bear Research and Management 9, 129–142. treecreeper (Climacteris rufa). 2. Validating predictive habitat Bowers, M.A., Gregario, K., Brame, C.J., Matter, S.F., Dooley, J.L.J., models. Biological Conservation 105, 395–403. 1996. Use of space and habitats by meadow voles at the Mace, R.D., Waller, J.S., 1997a. Grizzly bear habitat selection in the home range, patch and landscape scales. Oecologia 105, Swan Mountains, Montana. Journal of Wildlife Management 107–115. 61, 1032–1039. BIOLOGICAL CONSERVATION 128 (2006) 109– 119 119

Mace, R.D., Waller, J.S., 1997b. Spatial and temporal interaction of Pedlar, J., Fahrig, L., Merriam, H., 1997. Raccoon habitat use at 2 male and female grizzly bears in north-western Montana. spatial scales. Journal of Wildlife Management 61, 102–112. Journal of Wildlife Management 61, 39–52. Poole, K., Mowat, G., Fear, D., 2001. DNA-based population Mace, R.D., Waller, J.S., Manley, T.L., Ake, K., Wittinger, W.T., 1999. estimate for grizzly bears in northeastern British Columbia, Landscape evaluation of grizzly bear habitat in western Canada. Wildlife Biology 7, 105–115. Montana. Conservation Biology 13, 367–377. Rayner, J., 1985. Linear relations in biomechanics: the statistics of Mace, R.D., Waller, J.S., Manley, T.L., Lyon, L.J., Zuuring, H., 1996. scaling functions. Journal of Zoology (London) 206, 415–439. Relationships among grizzly bears, roads, and habitat in the Rossi, R.E., Mulla, D.J., Journel, A.G., Franz, E.H., 1992. Swan Mountains, Montana. Journal of Applied Ecology 33, Geostatistical tools for modeling and interpreting ecological 1395–1404. spatial dependence. Ecological Monographs 62, 277–314. Martinez, J.A., Serrano, D., Zuberogoitia, I., 2003. Predictive Sallabanks, R., 1993. Hierarchical mechanisms of fruit selection models of habitat preferences for the Eurasian eagle owl Bubo by an avian frugivore. Ecology 74, 1326–1336. bubo: a multiscale approach. Ecography 26, 21–28. Sokal, R.R., Rohlf, F.J., 1995. Biometry: The Principles and Practice McGrath, M.T., DeStefano, S., Riggs, R.A., Irwin, L.L., Roloff, G.J., of Statistics in Biological Research, third ed. Freeman, New 2003. Spatially explicit influences on northern goshawk York. nesting habitat in the interior Pacific Northwest. Wildlife Swenson, J.E., Sandegren, F., Soderberg, A., Bjarvall, A., Franzen, Monographs 154, 1–63. R., Wabakken, P., 1997. Infanticide caused by hunting of male McLellan, B.N., Hovey F.W., 1993. Development and preliminary bears. Nature 386, 450–451. results of partial-cut timber harvesting in a riparian area to Timoney, K., 1999. The habitat of nesting whooping cranes. maintain grizzly bear spring habitat values. In: Morgan, K.H., Biological Conservation 89, 189–197. Lashmar, M.A. (Eds.), Fraser River Action Plan, Riparian Habitat Vaughan, I.P., Ormerod, S.J., 2003. Improving the Quality of Management and Research, May 4–5, Kamloops, BC Distribution Models for Conservation by Addressing Environment Canada, pp. 704–712. Shortcomings in the Field Collection of Training Data. McLellan, B.N., Hovey, F.W., 2001. Habitats selected by grizzly Conservation Biology 17, 1601–1611. bears in a multiple use landscape. Journal of Wildlife Waller, J.S., Mace, R.D., 1997. Grizzly bear habitat selection in the Management 65, 92–99. Swan Mountains, Montana. Journal of Wildlife Management McLoughlin, P.D., Case, R.L., Gau, R.J., Cluff, D.H., Mulders, R., 64, 1032–1040. Messier, F., 2002. Hierarchical habitat selection by Wiens, J.A., 1989. Spatial scaling in ecology. Functional Ecology 3, barren-ground grizzly bears in the central Canadian Arctic. 385–397. Oecologia 132, 102–108. Wilkinson, L., Hill, M.A., Welna, J.P., Birkenbeuel, G.K., 1999. Mitchell, M., Lancia, R., Gerwin, J., 2001. Using landscape-level SYSTAT for Windows: Statistics, Version 9. SYSTAT Inc., data to predict the distribution of birds on a managed forest: Evanston, IL. Effects of scale. Ecological Applications 11, 1692–1708. Wolfram, S., 1999. Mathematica, fourth ed. Cambridge University Mowat, G., Strobeck, C., 2000. Estimating population size of grizzly Press, Cambridge. bears using hair capture, DNA profiling, and mark-recapture Woods, J.G., Paetkau, D., Lewis, D., McLellan, B.N., Proctor, M., analysis. Journal of Wildlife Management 64, 183–193. Strobek, C., 1999. Genetic tagging free ranging black and brown Nagy, J.R., Haroldson, M.A., 1989. Comparisons of some home bears. Wildlife Society Bulletin 27, 616–627. range and population parameters among four grizzly bear Zabel, C.J., Dunk, J.R., Stauffer, H.B., Roberts, L.M., Mulder, B.S., populations in Canada. International Conference on Bear Wright, A., 2003. Northern Spotted Owl habitat models for Research and Management 8, 227–235. research and management application in Calfornia (USA). Neter, J., Kutner, M.H., Nachtsheim, C.J., Wasserman, W., 1996. Ecological Applications 13, 1027–1040. Applied Linear Statistical Models, fourth ed. Irwin, Chicago. Zager, P., Jonkel, C., Habeck, J., 1983. Logging and wildfire Pedhazur, E.J., 1982. Multiple Regression in Behavioral Research: influence on grizzly bear habitat in northwestern Montana. Explanation and Prediction, second ed. Holt, Rinehart & International Conference on Bear Research and Management Winston, New York. 5, 124–132.

PAPER IX

473

Major components of grizzly bear diet across North America

Garth Mowat and Douglas C. Heard

Abstract: We measured stable carbon and nitrogen isotope ratios in guard hair of 81 populations of grizzly bears (Ursus arctos L., 1758) across North America and used mixing models to assign diet fractions of salmon, meat derived from terrestrial sources, kokanee (Oncorhynchus nerka (Walbaum in Artedi, 1792)), and plants. In addition, we examined the relationship between skull size and diet of bears killed by people inBritish Columbia. The majority of carbon and nitrogen assimilated by most coastal grizzly bear populations was derived from salmon, while interior populations usually derived a much smaller fraction of their nutrients from salmon, even in areas with relatively large salmon runs. Terrestrial prey was a large part of the diet where ungulates were abundant, with the highest fractions observed in the central Arctic, where caribou (Rangifer tarandus (L., 1758)) were very abundant. Bears in some boreal areas, where moose (Alces alces (L., 1758)) were abundant, also ate a lot of meat. Bears in dryer areas with low snowfall tended to have relatively high meat diet fractions, presumably because ungulates are more abundant in such environments. Kokanee were an important food in central British Columbia. In areas where meat was more than about a third of the diet, males and females had similar meat diet fractions, but where meat was a smaller portion of the diet, males usually had higher meat diet fractions than females.Females reached 95% of their average adult skull length by 5 years of age, while males took 8 years. Skull width of male grizzly bears increased throughout life, while this trend was slight in females. Skull size increased with the amount of salmon in the diet, but the influence of terrestrial meat on sizewas inconclusive. We suggest that the amount of salmon in the diet is functionally related to fitness ingrizzly bears.

Résumé : Nous avons mesuré les rapports des isotopes stables de carbone et d’azote dans des poils de garde provenant de 81 populations de grizzlis(Ursus arctos L., 1758) largement réparties en Amérique du Nord et nous avons utilisé des modèles de mélange pour attribuer les fractions du régime alimentaire correspondant au saumon, à la viande d’origine terrestre, au kokani (Oncorhynchus nerka (Walbaum in Artedi, 1792)) et aux plantes. De plus, nous avons examiné la relation entre la taille du crâne et le régime alimentaire chez des ours tués par les humains en Colombie- Britannique. La plus grande partie du carbone et de l’azote assimilés par la plupart des populations côtières de grizzlis provient du saumon, alors que, chez les populations de l’intérieur des terres, une fraction beaucoup plus faible des nu- triments provient du saumon, même dans les régions où les montaisons de saumons sont importantes. Les proies terres- tres forment une partie considérable du régime là où les ongulés sont abondants et les proportions les plus importantes s’observent dans la région arctique centrale où les caribous (Rangifer tarandus (L., 1758)) sont très nombreux. Les ours de certaines régions boréales riches en orignaux (Alces alces (L., 1758)) consomment aussi beaucoup de viande. Les ours qui habitent les régions plus sèches aux précipitations de neige réduites tendent à avoir des fractions importantes de viande dans leur régime, probablement parce que les ongulés sont généralement plus nombreux dans ces envi- ronnements. Les kokanis sont un élément significatifdurégime dans le centre de la Colombie-Britannique. Dans les régions où la viande constitue plus d’environ le tiers du régime, les mâles et les femelles ont des régimes de composition semblable; cependant, là où la viande représente une plus petite fraction du régime, les mâles ont généralement une fraction de viande plus élevée dans leur régime que les femelles. Les femelles atteignent 95 % de leur longueur moyenne de crâne vers l’âge de cinq ans, alors que les mâles n’y arrivent qu’à huit ans. La largeur du crâne des grizz- lismâles augmente tout au cours de la vie, alors que cette tendance est peu marquée chez les femelles. La taille du crâne augmente en fonction de la quantité de saumon dans le régime, maisl’effet de la viande d’origine terrestre sur la taille n’est pas clair. Nous croyons que la quantité de saumon dans le régime chez les grizzlis est fonctionnellement reliée à la fitness. [Traduit par la Rédaction] Mowat and Heard 489

Received 28 June 2005. Accepted 27 January 2006. Published on the NRC Research Press Web site at http://cjz.nrc.ca on 28 March 2006. G. Mowat.1,2 Aurora Wildlife Research, RR 1, Site 14, Comp. 8, Crescent Valley, BC V0G 1H0, Canada. D.C. Heard. BC Ministry of Environment, 4051-18th Avenue, Prince George, BC V2N 1B3, Canada. 1Corresponding author (e-mail: [email protected]). 2Present address: BC Ministry of Environment, Kootenay Region, 401-333 Victoria Street, Nelson, BC V1L 4K3, Canada.

Introduction tive season of temperate-dwelling bears (Hilderbrand et al. 1996;Felicetti et al. 2003b). The components of an animal’s The amount of meat inagrizzly bear’s(Ursus arctos L., diet are estimated by comparing the isotope ratios of diet 1758) diet is related to a bear’sfitness in the short term items to that of the consumer usingamixing model (Phillips through its influence on the bear’s nutritional state and over and Gregg 2001). These models assume that mass is con- the long term through its influence on the bear’s body size served as it moves through trophic levels and that the iso- (Hilderbrand et al. 1999a, 1999b). Choosing a meat-based tope ratio of the consumer’stissue isalinear combination of diet may increase fitness insituations where larger bears diet isotope ratios. have greater fitness than smaller conspecifics. But when pro- We collected grizzly bear hair samples from the field and teinavailability is unpredictable, being larger may have a fit- from other researchers and analyzed them for stable carbon ness cost. When proteinavailability is low, smaller bears are and nitrogen isotope ratios (i.e., δ13C and δ15N). We com- better able to meet theirmaintenance requirements by feed- bined those data with previously published data to describe ing on plants (Welch et al. 1997; Rode et al. 2001).Even bear diet. Our first objective was to describe the pattern of when proteinavailability is predictable enough for many marine and terrestrially derived meat in the diet of grizzly bears to have a largely protein-based diet, females with bears across North America. Second, we compared bear size young may be able to increase their inclusive fitness by with the amount of marine and terrestrial meat in the diet to feeding primarily on plants when they can acquire enough test the hypothesis that size increases with the amount of nutrients for hibernation without having to expose their meat in the diet. We predicted that bears would achieve adult young to potentially predatory males on salmon spawning sizeatsimilar ages and we compared size and bear age to streams (Ben-Davidetal. 2004). It also appears that protein test this prediction and to facilitate the examination of diet has a seasonal influence on size whereby bears that eat high- and size. proteindiets in spring and summer primarily add muscle mass, while high-proteindiets eaten in fall are largely stored Methods as lipids (Hilderbrand et al. 1999b). Berries are largely carbohydrates and when eaten in the fall, they allow bears to Isotope analysis add the fat required for denning, although larger bears are We analyzed the stable nitrogen and carbon isotope com- less able to maintain mass on a pure berry diet (Welch et al. position of grizzly bear hair from 81 areas across the current 1997). The addition of higher protein foods to a berry diet continuous range of the species in North America. We as- increases muscle mass gain (Rode and Robbins 2000; sumed that our diet estimate was an average of the bears’ an- Felicetti et al. 2003a). nual diet because we used only whole guard hairs. We used Protein in the diet is also important to grizzly bears at the data from the published and unpublished literature (Table population level because it is related to litter size and popu- S12) and from 1242 samples that we obtained and prepared lation density (Hilderbrand et al. 1999a). We wanted to ourselves. Three groups of samples from the literature con- expand on Hilderbrand et al.’s work (1999a)onthediet– sisted of hair and bone (Jacoby et al. 1999), but since isotopic density relationshipby(i) considering the current continuous signatures for two of these groups were similar to concur- range of grizzly bears, (ii) considering the application of the rently published hair data for the same area (Hilderbrand et relationship to the prediction of carrying capacity, (iii) eval- al. 1999a), we used combined hair and bone data to increase uating how different protein sources influence the sizeofin- sample sizes. Most samples were collected by other re- dividual bears, and (iv) increasing our understanding of the searchers from clearly laid out study areas. However, some functional relationship between fitness and carrying capac- samples came from mandatory submissions of human-killed ity. bears from Alaska, British Columbia, Northwest Territories, Diet estimates based on scat analyses (i) underestimate and Nunavut. We grouped samples by areas with similar life meat in the diet (Robbins et al. 2004), (ii) cannot always be histories to generate a mean signature for each local popula- assigned to species, (iii) can only rarely be assigned to an in- tion. No δ13C measurements were available for four study ar- dividual, thus precluding comparisons between diet and indi- eas but because no salmon were found in those areas, diet vidual fitness measures, (iv) do not measure assimilated could be determined from δ15N values alone (see below). We nutrients, and (v) would be difficult to carry out at the conti- excluded cubs but did not exclude bears killed by wildlife nental scale because of the effort required to collect repre- control officers because (on average) these bears did not sentative scat samples. Analysis of stable isotopes in bear have higher isotopicsignatures than other bears from the hair samples is less influenced by these problems. Stable same area, in contrast to the findings of Hobson et al. isotopes measure the assimilated nutrients for the elements (2000). examined. Representative hair samples are easily obtained Hair was cleaned by soaking itfor2hina2:1 from both passive hair snags, such as those routinely used chloroform–methanol solution;it was then rinsed indistilled during DNA-based inventories, and the inspection of hunter- water and airdried. One or more hairs totaling1mg(0.8– killed bears. Hair can be assigned to the correct species and 1.2 mg) were put intoatin cup and analyzed commercially individual by analyzing DNA extracted from the root of the at University of California, Davis. Measurement error, varia- hair (Paetkau 2003). Guard hairs are grown between late tion among repeated measures of hairs from within the same spring and fall, thus integrating the diet over much of the ac- sample, and variation among hairs taken from different areas

2 Supplementary data for this article are available on the journal Web site (http://cjz.nrc.ca) or may be purchased from the Depository of Un- published Data, Document Delivery, CISTI, National Research Council Canada, Building M-55, 1200 Montreal Road, Ottawa, ON K1A 0R6, Canada. DUD 5004.For more information on obtaining material, refer to http://cisti-icist.nrc-cnrc.gc.ca/irm/unpub_e.shtml.

© 2006 NRC Canada Mowat and Heard 475 of the body were not large enough to mask variation among supported by work on foxes, seals, and horses (Hobson et al. bears or among years. 1996; Roth and Hobson 2000; Ayliffe et al. 2004). Two control samples were analyzed after every 12 hair We calculated fractions for diet components using7dif- samples, and measurement error was 0.05 (SD) for δ13C and ferent methods. Methods 1–4 were used for systems where 0.12 for δ15N(n = 103), which was lower than the variation there were only two main components in the diet. We used a among repeated samples. Variation insignatures among simple, single-isotope linear mixing model similar to eq. 3 hairs from the same hair sample — hairs that presumably in Hobson et al. (2000) but based on δ15N values instead of came from similar locations on the same bear — was δ13C values. The main places where this approach was used SD13C =0.22 and SD15N =0.52 (n = 66). Variation insigna- were coastal areas, where salmon and plants were the two tures among samples collected on different dates withina maindiet components (method 1, n = 19), and much of the year, and presumably from somewhat different locations on eastern (method 2, n = 17), northern (method 3, n = 8), and the same bear, was higher (SD13C =0.60, SD15N =0.70, n = central (method 4, n = 13) range of grizzly bears, where 51). We analyzed paired samples from the back and the ungulates and plants were the maindiet components. rump of eight live-captured bears from the Yukon north Methods 5 and 6 were used where a population was slope, and variation among signatures was small (SD13C = thought to have a diet composed of three components. Here, 0.11, SD15N =0.33, n = 16). Variation among years was sim- we used the program Isoerror (Phillips and Gregg 2001) to δ13 δ15 ilar to other measurement errors (SD13C =0.67, SD15N = calculate diet fractions from the observed C and Nval- 1.21, n = 14) except in the Owikeno Sound, where there was ues if the mean for the population fell in the Euclidean space amajor decline in salmon abundance (Boulanger et al. 2004; of the diet components (method 6, n = 12). If the mean iso- SD13C =1.44, SD15N =2.64, n = 33). tope measures for the population fell outside the Euclidean We derived a population mean by weighting the mean for space, we used method 5 (n = 6), after Hilderbrand et al. each sex. Males often consume more meat than females (1996): (Jacoby et al. 1999; Hobson et al. 2000; Ben-Davidetal. [1] M =(δ13C – δ13C )/(δ13C – δ13C ) 2004) but rarely make up more than 40% of the population bear terr salmon terr (McLellan et al. 1999; McLoughlinetal. 2003; Schwartz where M is the fraction of a bear’sdiet consisting of salmon et al. 2003). We weighted the female isotopicsignature to δ13 δ13 or other marine-derived food, Cbear is the C value mea- make up 60% of the population’sdiet estimate for those δ13 δ13 sured in bear hair, Cterr is the average C value in all ter- populations for which we had sex data for all samples. In restrial plant and animal foods, increased for fractionation four areas where we did not have complete sex data, we δ13 (we used the generalized plant endpoint), and Csalmon is used an unweighted mean. In 10 cases the population mean the average δ13C value for salmon, also corrected for frac- was based on data for only one sex (usually female). tionation. We assumed salmon was the main marine-derived food of grizzly bears; other potential foods such as eulachon Estimation of diet (Thaleichthys pacificus(Richardson, 1836)) and other fishes, We estimated the contributions of four diet components — seals, whales, and invertebrates have similar isotopicsigna- plants, marine-derived nutrients (primarily salmon), terrestrial tures (Kelly 2000; Kurle and Worthy 2001; Ben-Davidetal. meat (primarily ungulates), and landlocked kokanee salmon 2004). (Oncorhynchus nerka (Walbaum in Artedi, 1792)) — to griz- All isotopicdiet endpoints except the plant endpoint were zly bear diets by comparing stable carbon and nitrogen iso- derived from appropriate data in the literature. Terrestrial tope ratios of bear hair and potential diet types. Stable isotope meat signatures were derived for broad areas inwhich we ratios in consumers are higher than those in theirdiet be- assumed bears would use similar ungulate prey (Table 1). δ13 cause organisms preferentially assimilate the heavier isotope We determined Cplant by inserting –23 (SD = 0.35, n = 91) δ13 or respire the lighter isotope (Hilderbrand et al. 1996). We for Cbear into eq. 1 above. This value was the mean δ13 accounted for the preferential assimilation (fractionation) of Cbear for four populations in the west slopes of the Rocky 15N by correcting (i.e., increasing) the measured isotope ra- Mountains (Parsnip Mountains, Columbia Mountains, and tios of potential diet groups according to the equation deter- central and southern Selkirk Mountains) that are known to mined by Felicetti et al. (2003b). The equation documents a eat little terrestrial meat (Hobson et al. 2000; Ciarniello et δ13 decrease in 15N fractionation with a decrease in the 15N/14N al. 2003) and have no salmon available. The mean Cval- δ15 ratioofthediet. We did not use their equation for δ13C ues for bears that had N values < 3.7‰ were similar because the fit was weaker and the equation implied that among various regions of the continent (Table 2). We used δ15 at δ13C values greater than –18.7‰, the consumer’stissue the mean N value from the above areas, plus 1 SD to al- would be depleted in 13C(i.e.,havealowerδ13C value than low for variation among areas and individuals, to identify theirdiet). This seemed unlikely, based on other investiga- bears that ate little meat. tions of fractionation of carbon inhair (Hobson et al. 1996; For method 5 we calculated the amount of terrestrial meat Roth and Hobson 2000; Ayliffeetal. 2004). Instead, we in- in the diet using eq. 2 inHilderbrand et al. (1996): creased δ13C values of terrestrial and freshwater diet compo- ⎛ δδδ15NNN−−M 15 15 ⎞ nents by 2‰ and that of the marine component (salmon) by [2] T = ⎜ bear salmon plant ⎟ ()1 − M 1‰, after Ben-Davidetal. (2004). We reduced all hair δ13C ⎜ δδ15− 15 ⎟ ⎝ NNterr plant ⎠ values by 1.5‰ because hair is enriched in 13C compared with blood and muscle and the above fractionation values where T is the fraction of a bear’sdiet consisting of terres- δ15 were for blood or muscle. Ben-Davidetal. (2004) suggested trial meat, Nbear is the isotope ratio measured in bear hair δ15 correcting hair by 1–2‰ compared with blood, which is (reduced for trophic fractionation), Nplant is the average © 2006 NRC Canada 476 Can. J. Zool. Vol. 84, 2006

Table 1. Observed isotope ratios (δ13C and δ15N) and isotopic endpoints (Δδ13C and Δδ15N) used to calculate relative measures of major diet components of grizzly bears (Ursus arctos) across western North America.

Food type δ13C (‰) Δδ13C SD δ15N (‰) Δδ15N SD n Sources Terrestrial meat Arctic and subartic tundra –22.3 –20.31 3.18.0 1 57 Barnett 1994; Gau 1998; Ben-Davidetal. 2001; B. Milakovic, personal communication Boreal –23.5 –21.51 2.17.11 35Szepanski et al. 1999; Ben-Davidetal. 2001; Kielland 2001; B. Milakovic, personal communication North coast –24.8 –22.81 2.57.5 1 107 Jacoby et al. 1999; Szepanski et al. 1999; Ben-Davidetal. 2004 Temperate mountains –25.3 –23.31 3.48.3 1 70 Jacoby et al. 1999; Hobson et al. 2000; Felicetti et al. 2005; B. Milakovic, personal communication Fish Kokanee –31.0 –29.01.56.711.21.2 3 Tom Johnston, BC Ministry of Water, Lands and Air Protection, Vancouver, British Columbia Anadromous salmon –19.9 –18.91 12.516.3 1 338 Bilby et al. 1996; Hildebrand et al. 1996; Jacoby et al. 1999; Satterfield and Finney 2002; Ben-Davidetal. 2004 Plants Generalized plant baseline –26.6 –24.62 –2.82.8 3 200 Derived using data from areas where there islittle meat in the diet; see text Note: Δδ13C and Δδ15N values have been adjusted for trophic-level fractionation.

Table 2. Mean isotope values for bears with δ15N<3.7‰, which is the mean value for bears in four study areas where bears were known to eat little meat, plus one standard deviation (see text), infive ecozones of western North America (Commission for Environmental Cooperation 1997). Ecozone δ13C (‰) SD δ15N (‰) SD n Temperate mountains –23.00.37 2.80.63 142 Sub-boreal mountains –23.80.42 3.44 0.42 9 Pacific coastal mountains –23.30.61 2.80.82 19 Boreal –22.80.63 3.10.41 89 Arctic –22.30.53 3.50.14 3 Note: The mean δ13C values represent carbon values for bears that eat little meat in these ecosystems. δ15N values are given to demonstrate the potential variation in meat consumption among bears in each sample group.

δ15 isotope ratio of plant foods eaten by bears, Nterr is the av- sible solutions, we assumed a diet component fraction was erage isotope ratio of terrestrial animal foods eaten by bears, zero when the mean estimate was <1% and used a simpler δ15 and Nsalmon is the average isotope ratio for salmon (Ta- analytical method. Our calculations of diet fractions are δ15 δ15 ble 1). We determined Nplant by inserting 2.85 for Nbear based on generalized food baselines and should not be inter- into eq. 2 above and setting T to zero. This value was the preted as accurate measures of assimilated carbon or nitro- δ15 mean Nbear for the four bear populations described above gen; we present relative measures of protein in the diet for that are known to eat little terrestrial meat. We did not use comparison across the continental range of grizzly bears. the absolute minimum stable carbon or nitrogen isotope ra- More accurate measures of diet can be developed using re- tios to calculate plant baselines because we did not want to gional isotopic baselines and with the addition of further diet base our endpoint derivations on single extreme values. components and isotope markers (e.g., Felicetti et al. 2003b; When the population was known to have four major diet Ben-Davidetal. 2004). components, we used the IsoSource model approach advo- cated by Phillips and Gregg (2003) (method 7). Thisap- Mapping diet proach was used for 12 populations in central British We mapped the geographic center of each population as a Columbia that were known or presumed to eat spawning point in a geographic information system and used the trian- kokanee, Pacific salmon, plants, and terrestrial meat. gular irregular network (TIN) procedure to interpolate a sur- To increase the accuracy of our results, especially when face of mean diet components across western North using method 7, which often resulted inawide range of pos- America. We added six dummy points to the data set to in-

© 2006 NRC Canada Mowat and Heard 477 crease the size of the interpolated area such that the surface We used simple bivariate regression to determine the rela- reached the edge of the continent. Dummy points were tionship between diet and age. In these cases we excluded added only within the area of a population unit for which we samples where the diet fraction of interest was zero; we as- had isotope data, hence these points essentially extended the sumed bears withadiet fraction of zero did not have access diet measures to the edge of an area where diet had been to the food type, although we acknowledge that this assump- measured. The TIN surface is based on simple linear relation is probably not always true. Residuals were normally tionships among three adjacent points and does not smooth distributed for all regressions. We used partial correlations to or detect spatial trend. Smoothed maps are cleaner and may compare the fitofindividual variables in the regression at times be more accurate, but we felt our data were too het- (Tabachnick and Fidell 1996). erogeneously dispersed for us to use an algebraic smoothing procedure. In addition, TINs clearly demonstrate spatial Results weaknesses or incongruities in the point data. Isotopic signatures Body size and diet δ13C and δ15N values were not linearly related (Fig. 1). Skull size and specimen age have been recorded by gov- There is modest variation in δ13C among most bears with ernment inspectors for almost all bears killed inBritish Co- δ15N values below 5‰, suggesting that bears that eat mostly lumbiasince 1975. Skulls were measured when unskinned, plants have similar carbon signatures. But some bears with fresh skinned, or skinned, boiled, and cleaned. Based on no access to salmon have increased δ13C values similar to data from the entire data set (n = 3840 males and 2018 fe- those seen for bears that eat salmon. Bears from the central males), which combines all ages except cubs, boiled and Arctic in particular show enriched δ13C values (Fig. 1). cleaned skulls averaged 5 mm shorter than fresh skinned Other non-salmon-eating bears with enriched δ13C values skulls for both males and females. Unskinned skulls aver- were usually from northern portions of the continent, where aged 11 and 13 mm longer than fresh skinned skulls for ungulate prey species have higher δ13C values than prey fur- males and females, respectively. Width of boiled skulls aver- ther south (Table 1). Bears with moderate δ15N values and aged 5 mm less for males and 3 mm less for females than low δ13C values were consuming kokanee. width of fresh skinned skulls. Unskinned skulls averaged 9 In general, mean carbon and nitrogen isotope ratios of and 11 mm wider than fresh skinned skulls for males and fe- populations were linearly related, which resulted in long, males, respectively. We used these means to estimate a fresh thin Euclidean spaces inmixing diagrams when plants, ter- skinned skull measurement for boiled and unskinned skulls. restrial meat, and salmon were potential diet components Age was measured using the cementum annuli method and (Fig. S12). The power to assign terrestrial meat was poor in included a fraction of the year based on the month the bear these cases because there was no clear separation of end was killed. Most bears were killed in spring or fall. members. Kokanee had a unique signature relative to other Growth of skull size was described by fitting the skull food classes, which gave good power to assignadiet frac- size measurement (length or width) to age with a Gompertz tion for this food. growth equation (Zullinger et al. 1984) through a three- parameter, nonlinear model: Diet −− [3] LA= e−e KY() I In coastal areas, where bears eat mostly salmon, males and females often have similar isotopicsignatures (Table where Y is age in years, L is skull length or width inmilli- S12). Where salmon is a smaller portion of the diet, females metres, A is asymptotic skull length or width, K is a growth often eat less salmon than males, and where salmon isa constant (age–1), and I is the inflection point. We used much smaller part of the diet, females generally consume STATISTICA® Version 6.0 (StatSoft Inc., Tulsa, Oklahoma) much less salmon than males. Males often had the highest to fit a model with three unknowns (K, I, and A)using itera- individual δ15N values inanygiven study area. tion. We used data from 255 hunter-killed bears from British Salmon was detected in the diet of bears as far north as Columbia (169 males, 86 females) for which we had both Kotzebue Sound. The highest salmon diet fractions were on diet and size measures to determine the effects of age and the Alaska Peninsula and parts of the mainland in southeast diet on body size. We excluded cubs and yearlings and seven Alaska and on the north coast of British Columbia. Bears in outliers that appeared to have mistakes in the age, sex, or Owikeno Sound and Glacier Bay, Alaska, had lower salmon size data.Even though the Gompertz growth function was a diet fractions than bears in all other areas of the Pacific biologically meaningful size at age relationship, a logisticre- coast. Salmon isotopicsignals were rarely detected in the in- lationship(l = a + blog(age)) also represented a statistical fit terior of the continent. The highest signals were detected in to the data (e.g., for male width, rGompertz =0.74 vs. rlogistic = the Quesnel Lake and Wells Gray Park regions of central 0.74), meaning that we could use log(age) to linearize skull British Columbia(Fig. 2; digital map data are in Supplemen- length and width measures and carry out multiple linear re- tary material2).Enriched δ13Csignals in bears from the cen- gression analysisusing log(age) as an independent variable tral Arctic generated marine components comprising >50% to account for the influence of age on size. Linear regression of the diet, which is not supported by fieldwork (Gau et al. was used to examine the relationship between diet and size; 2002; MacHutchon and Wellwood 2003 and citations salmon and terrestrial meat diet fractions, log(age), and sex therein). Therefore, we excluded the salmon endpoint from were included in the regressions. Salmon and kokanee diet Arcticmixing models, although this may have biased other fractions were summed and inserted in place of salmon to diet estimates. Hence, our analysis does not examine the po- assess whether kokanee helped explain growth patterns. tential for Arctic bears to feed on marine foods such as ana-

Fig.1.Stable isotope ratios in guard hairofgrizzly bears (Ursus arctos) from Alaska, Yukon, Northwest Territories, Nunavut, British Columbia, Alberta, and Montana. ᭹, data from bears that had no access to salmon; ᭝, data from bears from treeless areas of North- west Territories and Nunavut; ᭿, data from bears that may have had access to salmon. -15

-16 no salmon central Canadian Arctic -17 may have salmon -18

-19

C -20

δ -21

-22

-23

-24

-25 024681012141618 δ15N

Fig.2.Interpolation of the amount of salmon in the diet of grizzly bears across North America. The map is based on 81 data points (circles; Table 3) and 6 redundant points added to the edge of some population areas to expand the final map size. We used the triangulated irregular network procedure in ArcView® 3.2(Environmental Systems Research Institute, Inc., Redlands, California) to derive the map.

dromous Arctic char (Salvelinus alpinus (L., 1758)), whales, able and terrestrial meat consumption was higher than else- or seals. where, nitrogen signatures were similar between sexes. Ni- In Arctic areas, where barren-ground caribou were avail- trogen signatures were also similar between sexes in three

Fig.3.Interpolation of the amount of terrestrially derived meat in the diet of grizzly bears across North America. The map is based on 81 data points (circles; Table 3) and 6 redundant points added to the edge of some population areas to expand the final map size. We used the triangulated irregular network procedure in ArcView® 3.2(Environmental Systems Research Institute, Inc.)toderive the map.

boreal populations in northern British Columbia (Cassiar, Diet fractions averaged 17% (SD = 27) for salmon, 5% Spatsizi, and Muskwa), where terrestrial meat consumption (SD = 8) for kokanee, and 24% (SD = 21) for terrestrial contributed >40% of assimilated nitrogen. Where terrestrial meat for 255 hunter-killed bears from British Columbia. meat consumption was lower, males tended to have higher Skull length increased with increasing amounts of salmon 2 2 nitrogen signatures than females (Table S1 ). Terrestrial and terrestrial meat in the diet (F[4,250] =76.8, R =0.55, P < 2 meat consumption was near zero in the west slopes of the 0.001), as did skull width (F[4,249] =90.7, R =0.59, P < Rocky Mountains north to the Hart Ranges in east central 0.001). Adding kokanee to the salmon measure mildly British Columbia(Fig. 3; digital map data are in Supplemen- improved the fit over salmon alone for skull length (F[4,250] = 2 2 tary material ). Dryer areas to the east and west showed 80.3, R =0.56, P <0.001) and width (F[4,249] =93.1, measurable terrestrial meat consumption. Most boreal areas, R2 =0.60, P <0.001). In all four above models the slope for such as the interior of northern British Columbia and Alaska, salmon was 28%–71% greater than that for terrestrial meat, showed higher terrestrial meat consumption, although we while the error of the slope estimate was 21%–22% greater sampled few populations from these areas. Arctic regions for terrestrial meat.Further, partial correlations for salmon that supported barren-ground caribou consistently showed or fish were at least 2.5× larger than those for terrestrial the highest terrestrial meat consumption. meat. These data suggest that fish (largely salmon) influ- Kokanee was a detectable part of the diet in many areas in ences size to a greater extent than terrestrial meat, but confi- central British Columbia. Average predicted contributions to dence intervals of the slopes for fish and meat overlapped in assimilated carbon and nitrogen varied from 3% to 28% all four models above. Indeed, simpler models, where all among populations, suggesting that kokanee provided an im- three sources of meat were summed, fit the data as well as 2 2 portant contribution to the diet in some areas (Fig. S2 ). Pre- the above models for both length (F[3,251] = 105.2, R = 2 dicted contributions to diets of individual bears were as high 0.56, P <0.001) and width (F[3,250] = 124.1, R =0.60, P < as 36%. 0.001;Fig. S32). Where salmon was >30% of the diet, the contribution of Body size and diet salmon to the diet increased with age for males, but where Skull lengths reached 95% of their asymptoticsizeatage salmon was a smaller fraction of the diet, there appeared to 8 for males and age 5 for females; skull widths increased un- be no relationship between diet and age for either sex til 14 and 8 years for males and females, respectively (Fig. 5a). The amount of terrestrial meat in the diet did not (Fig. 4). change with age (n = 73 and 146, P >0.11;Fig. 5b), while

Fig.4.Mean skull size and 95% confidence intervals by age for hunter-killed grizzly bears from British Columbia, 1975–2005 (nmales = 3840, nfemales = 2018). We combined all ages >24 years for males and >21 years for females. Curves were generated with the raw data using a Gompertz growth equation; see Methods for details. To plot the graph, ages were rounded to the nearest whole number of years; hence, age depicts the number of summers each bear lived. Lines indicate the age when 95% of the asymptoticsizewas −−−0.273(Age ( 3.15)) −−−0.383(Age ( 3.14)) reached. Male skull length = 375e−e (R =0.62), female skull length = 329e−e (R =0.44), male skull −−−0.170(Age ( 3.24)) −−−0.244(Age ( 3.66)) width = 236e−e (R =0.74), female skull width = 195e−e (R =0.60).

the amount of kokanee in the diet decreased with age for fractions within sexes may indicate the degree to which 2 both males (F[1,63] =3.5, R =0.05, n = 65, P =0.07) and females and females use the same streams (Nevin 2003; 2 males (F[1,29] =14.2, R =0.33, n = 31, P <0.001;Fig. 5c). Gende and Quinn 2004) in response to both within-sex social interactions and competition between the sexes (Gende Discussion and Quinn 2004). Female bears with cubs are hypothesized to avoid salmon Salmon make major contributions to grizzly bear diet streams to minimize the risk of infanticide from males (Ben- throughout the Pacific coast of North America (Hilderbrand Davidetal. 2004). Admiralty and Chichagoff islands, where et al. 1999a). In our study, salmon was <33% of the diet in some females have already been shown to avoid salmon only three coastal areas: Owikeno Sound in southwest Brit- streams, were the only two coastal study areas where fe- ish Columbia (Boulanger et al. 2004), Glacier Bay in south- males consumed less salmon than males. But the data we east Alaska (TaniaLewis, US National Parks Service, present do not test the hypothesis well because most of our Glacier Bay, Alaska), and Kotzebue Sound along the Bering coastal data came from hunter-killed bears. Because hunters Strait(Miller et al. 1997), and each of these areas had rela- are prohibited from killing females accompanied by cubs, tively few salmon available. The fact that males and females our sample may not accurately document female avoidance had similar salmon diet fractions in coastal areas suggests of salmon streams. Three coastal samples were drawn from that salmon was abundant and dispersed enough to be avail- live-captured bears, which presumably better reflect all fe- able to most of the adult population, although some females males in the sample, and males and females had similar ni- may choose to avoid streams to minimize the risk of infanti- trogen signatures in all three areas. However, our sample cide (Ben-Davidetal. 2004). The variance in salmon diet sizes for females were 6–7 for these three areas, while the

Fig.5.Relationship between grizzly bear age and (a) the proportion of the diet consisting of salmon, (b) the proportion of the diet consisting of terrestrial meat, and (c) the proportion of the diet consisting of kokanee. Diet was calculated based on isotope analysisofentire guard hairs and several different linear mixing models (see Methods). Only samples with salmon, meat, and kokanee diet fractions > 0 were included in(a), (b), and (c), respectively. sample sizes for females from Admiralty and Chichagoff islands were 31–33. Thus, our samples were likely too small or biased to detect females that had chosen to avoid salmon streams to protect their cubs. In areas where salmon was <33% of the diet, males regularly had higher nitrogen signatures than females, which suggests that resources are less widespread or abundant and that males are able to monopolize the salmon resource (see Gende and Quinn 2004 for a review of the influence of social dominance on foraging). In contrast, followingamajor decline in salmon returns inOwikeno Sound, bears consumed less salmon than elsewhere along the coast, but both sexes had similar nitrogen signatures. The pattern of consumption of terrestrial meat was similar between sexes. In most areas of the Rocky Mountain west slopes, ungulates are not abundant (Shackleton 1999); nitrogen signatures were similar between the sexes and consistently suggested that little terrestrial meat was consumed (this study, Hobson et al. 2000). Males are likely unable to monopolize terrestrial meat resources in interior areas except perhaps where gut piles from hunter-killed ungulates are very common. The pattern of terrestrial meat consumption between sexes would suggest that at high ungulate densities, males and females encounter and exploit ungulates at similar rates. At lower ungulate densities the resource is more clumped and males spend more time actively hunting, are more likely to be able to defend carcasses until they are consumed, or encounter ungulates more often than females because of their larger home ranges. Other authors have shown that male grizzly bears consume the greater portion of a meat resource that occurs in relatively small patches.For example, male bears consumed more spawning cutthroat trout (Oncorhynchus clarkii bouvieri (Jordan and Gilbert, 1883)) than females, and spawning fish are confined to a portion of the small streams in the Greater Yellowstone Ecosystem (Felicetti et al. 2004). Jacoby et al. (1999) showed that male bears typically had higher meat consumption than females in areas of the continental Midwestern United States; these bears fed on presumably predated ungulates and scavenged road-killed ungulates. In contrast, bears that fed heavily on cattle had similar terrestrial meat intakes among sex and age classes. The highest terrestrial meat diet fractions were consumed by grizzly bears in Arctic areas where caribou were abundant. Similarly, moose were abundant in areas of Alaska and British Columbia where terrestrial meat diet fractions were also high (Miller et al. 1997; Hilderbrand et al. 1999b; in wet areas, where forests are dense and ungulates are not Shackleton 1999). Ungulates are abundant along the Rocky abundant. Our data suggest a weak negative relationshipbe- Mountain east slopes and in parts of the central interior of tween climate moisture and the fraction of the diet that con- British Columbia, and terrestrial meat fractions were modest sists of terrestrial meat. The functional relationship is to high in these areas. Terrestrial meat fractions were lowest probably between bears and ungulates, with ungulate num-

© 2006 NRC Canada 482 Can. J. Zool. Vol. 84, 2006 bers being higher where there is less snow (Kelsall and greater extent than the season inwhich salmon is consumed. Telfer 1973; Crete 1976; Thomas and Toweill 1982). In boreal areas, where bears feed largely on moose, calves Kokanee are sporadically abundant in parts of interior are the main prey and are killed in spring (Ballard et al. British Columbia and appear to be important contributors 1981; Gasaway et al. 1992). In areas where hunting is com- to diet in some areas. Diet estimates are based on three mon, gut piles may be a large portion of ingested meat kokanee samples from central British Columbia;if kokanee (Haroldson et al. 2004), and these are consumed in fall ex- isotope ratios are more similar to published data from Colo- cept in the Arctic, where hunting often occurs in spring and rado (Johnson et al. 2002), then kokanee diet fractions may late summer. Bears in the Arctic consistently showed the be overestimated by as much as 100%. Simulations using highest meat fractions in theirdiet, and presumably caribou endpoints from Johnson et al. (2002) suggest that salmon are consumed during all seasons (Reynolds and Garner diet fractions were little influenced by inaccuracy in the 1987; Gau et al. 2002; MacHutchon and Wellwood 2003). kokanee endpoint, whereas terrestrial meat would have been The variation in the season of terrestrial meat use may gen- slightly underestimated. Kokanee were important to certain erate greater variation in the relationship between meat in individual bears: 75 of 272 bears for which we calculated in- the diet and bear size at the continental scale. dividual diet had >5% kokanee in the diet and 10 bears were Arctic-dwelling bears are not larger than other bears that predicted to consume >25% kokanee. Kokanee were con- eat much less meat (Rausch 1963; Kingsley et al. 1988);in- sumed in some areas where anadromous salmon were also deed, some populations appear measurably smaller (Fergu- available, and kokanee spawn at similar times to chum son and McLoughlin 2000; Schwartz et al. 2003) and (Oncorhynchus keta (Walbaum in Artedi, 1792)) and densities are very low. Perhaps bears without access to sockeye (O. nerka) but are much smaller than anadromous salmon are smaller because of the high variation in ungulate salmon. Presumably some bears chose to fish kokanee rather abundance in space and time, which demands that bears than move to salmon streams. Our data suggest that younger never get so large that theirsize precludes theirability to bears fished kokanee (Fig. 5c) and then, when they got older gain and maintain mass feeding on vegetation. Alternatively, and larger, caught more salmon (Fig. 5a). Spawning cut- vegetation quality may also explain a portion of the variation throat trout appear to be important food for a few individual ingrizzly bear size across their range (Robbins et al. 2004). male bears in the Greater Yellowstone Ecosystem but are not We conclude that the amount of salmon in the diet is currently a significant resource at the population scale functionally related to the sizeofindividual bears but that (Felicetti et al. 2004). there is a weaker relationship between size and the amount What is the functional role of diet in bear population dy- of terrestrial meat in the diet. Plant quality also influences namics? Bears that eat meat grow larger, and size is related body size in bears (Robbins et al. 2004). Ultimately, the to population density, productivity, and other vital popula- amount of salmon or meat in the diet influences population tion rates (Stringham 1990; Hilderbrand et al. 1999a;Fergu- density at the continental scale (Hilderbrand et al. 1999a), son and McLoughlin 2000). The amount of meat consumed and perhaps at finer scales as well. Density was higher in is related to both skull length, which isanindex of subadult two western Arcticgrizzly bear populations with access to nutrition and growth, and skull width, which indexes growth abundant caribou than in a population with no such access during the subadult and adult phases of a bear’slife. How- (Reynolds and Garner 1987). ever, the influence of terrestrial meat on body size was only There are methodological limitations that may influence weakly supported by our data. the above conclusions. Guard hair indexes diet for a single Rausch (1963) suggested a clinal variation in bear size year, and the relationship between size and salmon fraction across North America based on topographical gradients but would suggest that yearly measures of salmon assimilation gave no functional explanation for that trend. His data roughly index the importance of salmon for the life of the (Fig. 2 in Rausch 1963) also support our evidence that bear bear. Consumption of terrestrial meat may be more variable size is related to the amount of salmon in the diet: bear pop- among years and hence the signal derived duringasingle ulations he studied that had no access to salmon had mean year may not index lifetime consumption of meat well. This skull lengths of 324–349 mm; those with some access to possibility seems slight, at the population level at least, be- salmon had skull lengths ranging from 346 to 366 mm; and cause the trend in meat consumption was consistent across coastal populations with access to abundant salmon had the continent; the signal from guard hairs generally docu- skull lengths ranging from 361 to 440 mm.Further, Rausch mented that bears ate more meat where ungulates were more (1963) suggested a clinal increase insize among coastal abundant. Also, our results are generally consistent with diet populations from south to northwest, which roughly follows estimates made using other methods (Reynolds and Garner our predicted increase in salmon diet fraction for these pop- 1987; McLellan and Hovey 1995; Mattson 1997; Gau et ulations. He further suggested clines insize for interior pop- al. 2002; MacHutchon and Wellwood 2003 and citations ulations, which are explained by the presence of salmon in therein), although, as expected, isotope analysis suggested south central Alaska and the Kotzebue Sound region of that meat is a larger portion of assimilated protein than tradi- northwest Alaska. tional analysis methods indicate. Nutritional ecology studies suggest that bears that eat Our analysis of bear size and diet provides several in- meat in spring tend to add lean body mass, whereas in fall sights regarding bear life history. The observation that the excess protein is converted to fat (Hilderbrand et al. 1999b; salmon diet fraction increases with age for males but not Felicetti et al. 2003a). The season inwhich terrestrial meat necessarily for females emphasizes the difference inlife- is consumed likely varies among individuals and areas to a history constraints between the sexes. Males grow larger

© 2006 NRC Canada Mowat and Heard 483 with age, which allows them to dominate resources both maining bears in their sample from the Rocky Mountain within and among species. The much larger body sizeof west slopes of British Columbia, whereas our general model males reduces theirability to gain and maintain mass feed- predicted no meat in the diet of that population. This result ing on vegetation (Welch et al. 1997; Rode et al. 2001). This was expected because we used the carbon and nitrogen sig- limitation may synergistically encourage male bears to mo- natures from four study areas inthis region to calculate the nopolize salmon resources. In contrast, females are smaller plant endpoints. In essence, this set the terrestrial meat con- and are better able to maintain their body mass on vegeta- sumption for these four areas to zero. tion alone, and some females even elect to forage away from We used data from the same 12 populations studied by salmon concentrations, presumably to reduce the risk of in- Hilderbrand et al. (1999a) and our salmon fractions were fanticide by males (Ben-Davidetal. 1994). Choosing to similar except on Admiralty Island, where our estimate was avoid spawning areas does not appear to be a lifelong strat- lower and more similar to results reported by Ben-Davidet egy because females that elect not to feed on salmon for a al. (2004). Our estimates of terrestrial meat fractions were single year are not smaller than those that do (Ben-Davidet higher than those of Jacoby et al. (1999) and Hilderbrand et al. 2004). Our limited data suggest considerable variation in al. (1999a) in all areas except coastal regions, where salmon yearly consumption of salmon among adult female bears were abundant, because here we did not consider meat a (compare the SD of mean isotope values between sexes in possible diet source in our models (except for the Kenai Table S12), which is consistent with a portion of females Peninsula, where moose are more abundant than they are in avoiding salmon feeding areas. As suggested earlier, other coastal areas; Hilderbrand et al. 1999b). Our assump- kokanee may be used more by younger bears as an alterna- tion would appear reasonable because these authors found tive to fishing for salmon where conflict with larger and little evidence for use of terrestrial meat in coastal areas. older bears is more likely. Our estimates of terrestrial meat consumption in interior Our data on age and skull size were collected throughout areas are higher than the estimates of these earlier authors British Columbia from areas with large differences in bear because we used a lower plant baseline that was specificto density and resource availability, but our conclusions appear bears. to be general to bears across North America. Skull dimen- Guard hairs provide a reasonable integration of yearly diet sions of bears from both the Arctic(Kingsley et al. 1988) because they are grown over a period of up to 5 months and the Alaska Peninsula (Glenn 1980) also showed that (Felicetti et al. 2003b). In addition, they contain both protein bear skull length asymptotes at around 8 years for males and and a stable inner core of lipids; hence, hair should integrate 5 years for females and that width continues to grow the assimilation of protein, carbohydrates, and fat. Hair can throughout most of the bear’slife. be collected noninvasively across broad areas, making large- How accurate are our predictions of diet? The distribution scale collections of hair much easier than collections of of our data is spotty, and the north and south ends of the blood or other tissue. We have shown that variation in range are least covered. Many populations in the south are repeated measures of stable isotope ratios among different not hunted, so fewer samples were collected there. We had hairs is not trivial but neither is it large enough to mask few samples north of the Alaska Peninsula, so our maps are continental-scale variation indiet.Future researchers may based on large extrapolations north of Anchorage in Alaska. minimize sample variance by analyzing several hairs from The coastal portion of Alaska north of Bristol Bay was the same bear (Ben-Davidetal. (2004) analyzed at least two mapped using data from two sample populations near Nome from each bear) or by using only spring- or fall-collected and Noatak. Our map indexes salmon consumption only hair so that hair growth is complete and documents the en- crudely inthis area. Similarly, inter-population variation in tire year’sdiet (as in Felicetti et al. 2004). meat consumption is poorly captured throughout the north The larger problems in our analyses likely revolve around by our map, given the paucity of samples we had for such a fractionation, especially of carbon, and the derivation of large area. endpoints.Fractionation studies have tended to compare The salmon diet fraction for one study area was likely bi- blood components with diet because blood components ased low.For the Susitna Valley area of interior Alaska we equilibrate to the diet faster than an entire guard hair does had data only from females, and because this was an area (Hilderbrand et al. 1996;Felicetti et al. 2003b) and are with moderate terrestrial meat consumption and low salmon therefore easier to study in captive animals. But carbon consumption, males likely had higher signatures than fe- shows different fractionation inhair than in plasma or blood males. Our salmon diet fraction was estimated to be zero but (Hobson et al. 1996; Roth and Hobson 2000; Sponheimer et the field biologists observed considerable use of a salmon al. 2003) and the fractionation relationship between diet and spawning area in a portion of the study area (Miller et al. consumer δ13C values is not simple (Hilderbrand et al. 1996; 1997). Therefore, the estimate for the Susitna Valley area is Ben-David and Schell 2001;Felicetti et al. 2003b; Ben- likely biased low owing to the lack of males in the sample. Davidetal. 2004).Felicetti et al. (2003b) argued against us- We had data for a single sex innine other areas, including ing carbon isotopes to make inferences about diet because of the Kenai Peninsula, but the differences indiet between the high variation in fractionation resulting from the com- sexes were likely small in these areas (Hilderbrand et al. plexities of carbon metabolism. Clearly, more controlled 1999b). studies of carbon fractionation in bear hair are needed. Using local plant and animal endpoints, Hobson et al. Fractionation of nitrogen appears to be similar inhair and (2000) predicted very small meat fractions for 46% of male other tissues (Hobson et al. 1996; Roth and Hobson 2000). and 18% of female bears and no meat in the diet of the re- The δ15N fractionation curve presented by Felicetti et al.

© 2006 NRC Canada 484 Can. J. Zool. Vol. 84, 2006 Year(s) ons across North i ng b i ght i 1 1982–2000 method We ly bear populat izz 3 9 1 1967–1996 6 1 1994–2002 5 4 Before 1998 2 1 1993–1995 07 1 16 1977–1996 1995–2003 4 4 1 Before 1998 2 1997 91 1 1 1 1995–2003 1999 1995–2003 2 1 1998 5 12 1 2003–2004 2000–2003 50 1 1 1997 2002–2003 1 1 2002 9 120 1997 1 1 1998 1996 4 1 2003 2 1 1995–2003 5 17 1 1994–1998 1995–2003 6 19 1995–2003 15 1 1995–2003 1995–2003 46 3 3 2003–2004 2003–2004 8 41 Before 1998 1 2002–2003 3 4 Before 1998 4 56 1 2003–2004 2003–2004 13 4 1 Before 1998 1989–1991 0 1 1995–2003 N ...... 9 15 δ (‰) on for gr 94 44 712 012 413 94 010 13 73 03 64 84 811 14 2 613 96 712 24 83 47 912 413 25 05 63 93 912 53 66 22 45 95 57 94 i ...... C 13 (‰) δ on i a od of sample collect Calculat method i 05 6 –23 15 7 –23 16 7 –22 15 713 –22 710 6 –21 –22 ...... ng method, and per i ght 32 3 –21 21 2 3 33 2 4 26 3 –22 4318 411 –22 41807 237 0 29 –23 –22 2 2 4 –23 –22 –22 3311 4 0 –23 12 232 –22 5 –19 21 0 60210703 3 2 –21 2 2 –22 –22 4 14 4 –22 16 0 40 0 22 0 0029 2 5 –23 –22 i ...... et i c means, we 04 0 56 0 09 0 8254 1 –19 163 –20 1 –19 7928717373 1 106 1 1 1 0 –17 –21 –17 –18 17 –17 0 31 1 –20 12 0 48 1 –20 78 1 –16 69 1 –18 i ...... sotop i on of d i ghted 68 0 79 0 6770 0 0 7412 0 0 54 0 18 0 5746 0 82 0 89 0 7793 0 637137 0 0 0 0 0 217229 0 27 0 27 0 67 0 0 0 63 0 30 0 8869 0 0 59 0 52 0 40799397 0 0 0 0 2286 0 0 68 0 31 0 00 0 i ...... Proport tude Plants Salmon Meat Kokanee 92 0 80 0 86 0 50 0 9042 0 0 90 0 94 0 60 0 50 0 9890 0 0 400035 0 50 0 0 50 0 0 006080 0 50 0 90 0 55 0 0 0 10 0 40 0 7007 0 0 90 0 44 0 00 0 9250 0 40 0 0 8600 0 0 40 0 26 0 07 1 i ...... on method, we i et calculat i tude Long i 70 –113 86 –142 40 –115 50 –121 1023 –139 –150 60 –130 24 –121 30 –126 53 –158 00 –135 2100 –150 –114 5000 –118 20 –117 34 –125 –154 757050 –164 80 –161 50 –154 67 –155 –161 –122 10 –127 40 –128 2589 –114 –137 20 –130 02 –134 36 –115 68 –129 20 –109 4000 –116 –117 74 –138 20 –126 20 –125 41 –117 ...... 48 69 60 63 58 fe i ne 57 i k ldl i ns 51 i et components, d i i or d rk 50 j i onal Park, ka 57 i i onal W i Ma a Mounta . i i ca k69 eymateen 54 i a – Lower St ralty 58 i i ar 59 er Bayer Nat 58 ck 54 i c Nat z i nenet–Yahki 55 48 i i i i i i chagoff 58 i ziz ntonngcome–Wakeman 53 51 Montana Refuge nlay–Osp i i d lk Valley 50 able 3. lathead 49 Amer T Study areaAdm Alaska GMU 10Alaska GMU 23Alaska GMU 9BAlaska GMU 9CAlaska GMU 9D Lat Alberta 54 Arct 66 60 Bab 58 Banff 55 Bathurst InletBlack LakeBowron 57 Bulkley LakesCab Cass 66 Central PurcellCentral Selk 51 56 Ch 54 Columb 53 CranberryDenal 50 E E Fi F Glac 55 Glac GranbyHerr H HylandIvvav Katma Kena Khut 49 K Kluane Park Central 59 60

© 2006 NRC Canada Mowat and Heard 485 Year(s) ng b i ght i 5 1995–2003 method We c 9 4 1 1995–2003 1 1 1995–2003 7 1 Before 1998 93 1 18 1995–2003 1 2001–2003 5 43 1998–1999 3 Before 1998 1995–2003 9 10 1 2003 1995–2003 0 13 4 2004–2005 Before 1998 77 1 108 1995–2003 1 1 1995–2003 1995–2003 1995–2003 2 17 1 1995–2003 2001 2 1 2003 6 1 1995–2003 62 1 10 1995–2003 1996 1 2003 44 1 3 1995–2003 1995–2003 9 17 1 1995–2003 3 1 1995–2003 5 1995–2003 19 1997 1 2000 60 1 1 1995–2003 1995–2003 1 1 2003 2 1 1995–2003 6 1 1995–2003 8 1 1995–2003 6 1 2002–2003 4 1 1995–2003 N ...... 9 15 δ (‰) c 38 74 59 54 73 56 95 25 75 25 56 54 211 74 14 08 86 5 37 46 33 64 93 12 94 711 54 34 22 94 111 34 55 85 48 85 98 34 26 ...... C 13 (‰) δ on i a Calculat method 2615 6 6 –23 –23 03 7 –22 18 728 7 –22 –23 07 7 –19 06 7 –22 07 6 –23 1510 7 7 –21 –22 08 7 –22 ...... 24 2 –22 491125 0 23 4 0 452 –22 48 –22 3 3 –21 –21 17 0 6134 3 4 –21 –21 29 0 1516 0 5 –21 43 4 –23 22 0 15 0 09 4 –22 3217 21500 0 –23 22 2 2 –22 –23 2 –22 1437 6 402 –22 –22 2 –23 5011 23613 0 0 –22 5 –21 14 532 –20 00 4 0 –23 ...... et i 06 0 12 0 3706 0 0 54 0 4229 104 0 –20 1 –21 51 1 –19 06 0 67 1 –19 222205 0 0 0 39 0 23 0 3053 1 1 –21 –19 64 1 –19 ...... on of d i 74 0 76 0 41 0 4849 0 0 17 0 5851 0 89506271 0 0 0 48 0 75 0 52 0 0 0 49 0 3979 0 66 0 33 0 0 672671 0 0 0 57 0 9147 0 0 6868 0 76 0 8500 0 78 0 0 70 0 47 0 0 63 0 98 0 50 0 68 0 36 0 ...... Proport tude Plants Salmon Meat Kokanee 00 0 50 0 50 0 3020 0 0 65 0 4060 0 6194 0 0 0 2590 0 0 8540 0 00 0 0 74 0 8027 0 0 48 0 90 0 385080 0 0 0 60 0 0050 0 0 6084 0 0 402299 0 0 1 8080 0 0 30 0 00 0 90 0 40 0 00 0 50 0 i ...... tude Long i 30 –122 50 –123 20 –129 7060 –132 –124 00 –129 5080 –127 5300 –127 –125 –123 0075 –125 –126 9138 –148 00 –120 –126 68 –152 8014 –112 –121 65 –148 54 –128 198080 –133 –126 –128 70 –128 00 –116 2030 –139 –137 7064 –122 –120 0046 –118 –116 0080 –113 –136 65 –125 50 –114 00 –130 80 –121 50 –119 20 –131 ...... 51 61 57 59 . ns 65 ) i ns 54 i or nued i ns 48 nland 56 i i i rks 49 i r52 i i n i e Mounta cont ght Bute 50 tna 62 ( i i zi i p Mounta p Plateau 54 i i izi neca 56 on 55 keno 51 i i i d Coastd Sus 52 Plateau i i able 3 T Study areaKluane Park NorthKluane Park SouthKugluktukLower NassMacGregor 61 Macken 60 Lat M M Muskwa 66 Nahann 55 Nat 54 North coastNorthwest AlbertaOm Ow Parsn 58 Parsn 57 Prudhoe Bay 54 Quesnel Lake NorthRobsonRockySahtu 52 Southeast ma Shuswap 69 MonasheeSouth PurcellsSouth Selk Southern Inter 53 51 Spats 57 Stewart 67 49 Southwest AlbertaSwan mounta TakuTatshensh Tatuk 49 Terror LakeToba Kn 56 TulsequahTweedsmu Upper Skeena – Nass 58 57 53 56 59

(2003b) shows good fit, and all other carnivore data (Hob- son et al. 1996; Roth and Hobson 2000; Kurle 2002) that we

N to cal- plotted on this relationship also fit closely (and see Ben- 15 δ David and Schell 2001:Fig. 3). The ultimate cause of the re- Year(s) lationship between Δδ15N and the δ15N ratio of the diet may

4, used be based on diet quality, which is the percentage of absorbed ; protein that is retained (Robbins et al. 2005). on mean based on ng b i i We derived isotopic endpoints for diet components from . ght i the literature and applied these over wide areas.For our deri- method We vation of plant endpoints we assumed that the mean diet of 5, populat

; bears in four of our study areas included no meat.Further, we extrapolated this plant endpoint to the entire western por- et components 3 1 1995–2003 37 1 4 1995–2003 Before 1998 N i . . .

4 1 2004 tion of the continent. Generally, we have faith in the end- . 15 (‰) δ

al tundra meat and plants points we derived from the literature because all except the i . kokanee endpoint were based on several measures. The end- th four d 64 35 14 94 i . . . . = 96) C point for kokanee was the weakest because it was based on n 13 δ (‰) three different fish from a single area of interior British Co- lumbia and was lower than values for kokanee from Colo- 4‰ ( . rado (Johnson et al. 2002). Any inaccuracy inthis endpoint on

i would likely have caused an overestimate of the importance a on mean based on females only i N=10 13 N to calculate terrestr δ

15 of kokanee in the diet. The C endpoint we derived for 15 δ δ 7, used Isosource w

; plants was similar to the average cited for C3 plants (Cerling Calculat method and Harris 1999) and similar to the plant endpoint developed 4, populat 3, used ;

9‰ and for the Kenai Peninsula using ungulates as surrogates for . ; vegetarian bears (Jacoby et al. 1999). The similarity in carbon signatures among bears with low δ15N values suggests et components i C = –20 that the carbon endpoint we developed for plants was robust. 13 δ Neither the δ13C values for vegetarian bears nor our derived ungulate endpoints suggest a decrease in δ13C values with th three d i increasing latitude (Hobson et al. 1999). The ungulate end- 13 36 4 –22 49 3 –23 0036 5 3 –22 –21 points we derived suggest an enrichment in Cwith increas- . . . . ing latitude, although this may have more to do with

al temperate meat and plants differences indiet among regions and species than with a i more fundamental continental-scale effect. et ng by sex, used mean of all samples 6, used Isoerror w i Perhaps the largest potential source of bias in our esti- i 17 0 . ; means for 1999–2001 were

ght mates of diet was the lack of separation of carbon and nitro- i : gen endpoints for plants, ungulates, and salmon in Euclidean (1996)

on of d 2 i . space (Fig. S1 ). The distinctive carbon signature for s area i

3, no we kokanee generated clear separation of end members. The ; 64 0 83 0 51 0 64 0 N to calculate terrestr . . . . poor Euclidean separation of our diet groups meant we had 15 Proport δ on of th

i to use simpler models than we preferred because isotopic

on mean signatures of bear hair were sometimes outside the mixing i lderbrand et al i

2, used polygon. In most coastal areas we assumed terrestrial meat ; nH i tude Plants Salmon Meat Kokanee 69 0 00 0 50 0 65 0

i use was zero and we used a simple nitrogen mixing model to . . . .

ght Inlet port estimate the salmon and plant fractions. In the Arctic, en- i riched carbon ratios generated measurable salmon fractions and smaller terrestrial meat components (often zero). To avoidthis untenable result, we set the salmon fraction to 5, used models

; zero for all Arctic areas, which may have resulted inan tude Long i 80 –134 30 –120 00 –138 90 –160 . . . . overestimate of the terrestrial meat fraction. In six cases we could not make simplifying assumptions because bears were known to eat both ungulates and salmon;in these cases we sotope data from the Kn i used the hierarchical model of Hilderbrand et al. (1996) to

. estimate the three diet fractions. Also, we used the plant ) endpoint for the terrestrial carbon endpoint in these calcula- ghted as 60% and males as 40% for populat i tions because we had no way of knowing what the signature al boreal meat and plants N to calculate salmon and plant components i was for the plant and terrestrial meat components combined. 15 δ concluded .

( This too may have resulted in an overestimate of the salmon n (2003) presented i component, although bias was likely small because the plant te Pass 59 i 1, used 1, females we Nev and animal endpoints were similar and much lower than that a b c able 3 males only T Study areaWells GrayWestern Brooks RangeWh Yukon north slope 68 Lat 52 69 culate terrestr of salmon. Any overestimate of the salmon component

© 2006 NRC Canada Mowat and Heard 487 would have resulted in an underestimate of terrestrial meat Hilderbrand and M. Dumond helped to improve the manu- consumption because the latter fraction is based on the frac- script.Funding was provided by the BC Ministry of Water, tion of the nitrogen signature that is unaccounted for by Land and Air Protection and the Government of Yukon, De- salmon (Hilderbrand et al. 1996).For these reasons we used partment of Renewable Resources. Hilderbrand et al.’s methods only for the six cases when a simpler model was not tenable and the bear signature was outside the Euclidean space of the diet endpoints. References There is error associated with diet estimates regardless of Ayliffe, L.K., Cerling, T.E., Robinson, T., West, A.G., Sponheimer, the calculation method used. We did not emphasize preci- M., Passey, B.H., Hammer, J., Roeder, B., Dearing, M.D., and sion because we were more interested in the spatial variation Ehleringer, J.R. 2004. Turnover of carbon isotopes intailhair indiet. All our estimates of diet fractions (Table 3) should and breath CO2 of horses fed an isotopically varied diet. be treated as approximate. Oecologia, 139: 11–22. Absorption and routing of specific molecules are general Ballard, W.B., Spraker, T.H., and Taylor, K.P. 1981. Causes of neo- problems with the use of isotope analysistoassign diet natal moose calf mortality in southcentral Alaska. J. Wildl. (Phillips and Koch 2002; Robbins et al. 2002). Direct rout- Manag. 45: 335–342. ing of fat from meat sources to stored lipids may cause un- Barnett, B.A. 1994. Carbon and nitrogen isotope ratios of caribou derestimates of the importance of meat if the diet tissue tissues, vascular plants, and lichens from northern Alaska. M.Sc. analyzed contains little fat (or if the fat is removed). We sus- thesis, University of Alaska, Fairbanks, Alaska. pect that thisbias is reduced when hair is used instead of Ben-David, M., and Schell, D.M. 2001. Mixing models in analyses blood or muscle tissue because guard hair should integrate of diet using multiple stable isotopes: a response. Oecologia, metabolite levels of both exogenous and endogenous carbon 127: 180–184. and nitrogen sources over most of the non-denning season, Ben-David, M., Shochat, E., and Adams, L.G. 2001. Utility of sta- and hair is made up of stable portions of both protein and ble isotope analysis in studying foraging ecology of herbivores: lipids. examples from moose and caribou. Alces, 37: 421–434. Our diet data support Hilderbrand et al.’s (1999a) conclu- Ben-David, M.,Titus, K., and Beier, L. 2004. Consumption of salmon sion that meat isanimportant resource for grizzly bear pop- by Alaskan brown bears: a trade-off between nutritional require- ulations. This conclusion was presumably based on the ments and the risk of infanticide? Oecologia, 138: 465–474. observation that salmon provide the majority of assimilated Bilby, R.E., Fransen, B.R., and Bisson, P.A. 1996. Incorporation of nutrients in coastal areas. We demonstrate that this observa- nitrogen and carbon from spawning coho salmon in the trophic system of small streams: evidence from stable isotopes. Can. J. tion iswidespread and that terrestrial meat may provide the Fish. Aquat. Sci. 53: 164–173. majority of assimilated resources in some ecosystems. We Boulanger, J.,Himmer, S., and Swan, C. 2004. Monitoring of griz- also provide a functional link between meat in the diet and zly bear population trends and demography using DNA mark– fitness via body size. Bears can continue to grow well into recapture methods in the Owikeno Lake area of British Colum- their adult life, and the consumption of meat causes an in- bia. Can. J. Zool. 82: 1267–1277. crease in both subadult and adult body size. We conclude Cerling, T.E., and Harris, J.M. 1999. Carbon isotope fractionation that spawning salmon are an important resource for grizzly between diet and bioapatite in ungulate mammals and implica- bear populations, but evidence for the functional importance tions for ecological and paleoecological studies. Oecologia, 120: of terrestrial meat is inconclusive. We suggest that further 347–363. work may demonstrate that salmon has a greater influence Ciarniello, L.M.,Seip, D., and Heard, D. 2003. Parsnipgrizzly on body size and perhaps population-level parameters than bear population and habitat project.Final Report, ParsnipGriz- terrestrial meat. Terrestrial meat sources are relatively secure zly Bear Project, BC Ministry of Environment, Prince George, but global climate change (Welch et al. 1998), hydroelectric B.C. Available from http://web.unbc.ca/parsnip-grizzly/. dams (Hilderbrand et al. 1996), overfishing, and the alien- Commission for Environmental Cooperation. 1997.Ecological re- ation of spawning habitat all jeopardize salmon conservation gions of North America: toward a common perspective. Commis- (e.g., Boulanger et al. 2004) and, by extension, pose a threat sion for Environmental Cooperation, Montréal, Que. Available to some of the highest density grizzly populations in North from http://www.cec.org/files/pdf/BIODIVERSITY/eco-eng_EN.pdf. America (Miller et al. 1997; Hilderbrand et al. 2004). Crete, M. 1976. Importance of winter climate in the decline of deer harvest in Quebec. Can.Field-Nat. 90: 404–409. Felicetti,L.A., Robbins, C.T., and Shipley, L.A. 2003a. Dietary Acknowledgements protein content alters energy expenditure of the mass gain in grizzly bears (Ursus arctos horribilis). Physiol. Biochem. Zool. We thank M. Dumond, R. Gau, R. Popko, A. Veitch, R. 76: 256–261. Maraj,G. Hilderbrand, S. Herrero, J.L. Weaver, G. Felicetti,L.A., Schwartz,C.C., Rye, R.O., Haroldson, M.A., Gun- MacHutchon, T. Lewis, M. Austin, K. Dimert, T. Smith, ther, K.A.,Phillips, D.L., and Robbins, C.T. 2003b. Useofsul- N.T. Johnston, D. Moyles, G. Stenhouse, R. Munro, M. fur and nitrogen stable isotopes to determine the importance of Proctor, B. Milakovic, M. Ben-David, K. Titus, R.Flynn, L. whitebark pine nuts to Yellowstone grizzly bears. Can. J. Zool. Ciarniello, R. Mace, M. Demarchi,R. Watt, R. Quinlan, D. 81: 763–770. Seip, and S. Himmer for providing samples or data. K. Felicetti,L.A., Schwartz,C.C., Rye, R.O., Gunther, K.A., Crock, Parker, M. Shook, J. Ayotte, E. Jones, C. Robertson, and B. J.G., Haroldson, M.A.,Waits, L., and Robbins, C.T. 2004. Use Heard helped with sample preparation and analysis. Thanks of naturally occurring mercury to determine the importance of to S. Arnt, K. Bray, and S. Miller for help with literature. D. cutthroat trout to Yellowstone grizzly bears. Can. J. Zool. 82: Pritchard helped with GIS mapping. Comments by G. 493–501.

Felicetti,L.A., Robbins, C.T., Herrero, S., and Pinto, M. 2005. Jacoby, M.E.,Hilderbrand, G.V., Servheen, C., Schwartz,C.C.,Ar- Diet of some eastern slopes grizzly bear project bears as deter- thur, S.M., Hanley, T.A., Robbins, C.T., and Michener, R. 1999. mined by stable isotope analysis. In Demography, ecology and Trophic relations of brown and black bears in several western management of grizzly bears in and around Banff National Park North American ecosystems. J. Wildl. Manag. 63: 921–929. and Kanaskis Country: the final report of the Eastern Slopes Johnson, B.M., Matinez,P.J., and Stockwell, J.D. 2002. Tracking Grizzly Bear Project. Edited by S. Herrero. University of Cal- trophic interactions in coldwater reservoirs using naturally oc- gary, Calgary, Alta. pp. 141–142. curring stable isotopes. Trans. Am.Fish. Soc. 131: 1–13. Ferguson, S.H., and McLoughlin, P.D. 2000.Effect of energy Kelly, J.F. 2000. Stable isotopes of carbon and nitrogen in the availability, seasonality, and geographic range on brown bear study of avian and mammalian trophic ecology. Can. J. Zool. life history.Ecography, 23: 193–200. 78: 1–27. Gasaway, W.C., Bortje, R.D., Grangaard, D.V., Kellyhouse, D.G., Kelsall, J.P., and Telfer, E.S. 1973. Biogeography of moose with Stephenson, R.O., and Larsen, D.G. 1992. The role of predation particular reference to western North America. Nat. Can. (Que.), inlimiting moose at low densities in Alaska and Yukon and im- 101: 117–130. plications for conservation. Wildl. Monogr. No. 120. Kielland, K. 2001. Stable isotope signatures of moose in relation to Gau, R.J. 1998.Food habitats, body condition, and habitat of the seasonal forage composition: a hypothesis. Alces, 37: 329–337. barren-ground grizzly bear. M.Sc. thesis, University of Sas- Kingsley, M.C.S., Nagy, J.A., and Reynolds, H.V. 1988. Growth katchewan, Saskatoon, Sask. length and weight of northern brown bears: differences between Gau, R.J., Case, R., Penner, D.F., and McLoughlin, P.D. 2002. sexes and populations. Can. J. Zool. 66: 981–986. Feeding patterns of barren-ground grizzly bears in the central Koch, P.L., and Phillips, D.L. 2002. Incorporating concentration Canadian Arctic. Arctic, 55: 339–344. dependence in stable isotope mixing models: a reply to Robbins Gende, S.M., and Quinn, T.P. 2004. The relative importance of et al. (2002). Oecologia, 133: 14–18. prey density and social dominance in determining energy intake Kurle, C.M. 2002. Stable-isotope ratios of blood components from by bears feeding on Pacific salmon. Can. J. Zool. 82: 75–85. captive northern fur seals (Callorhinus urinus) and theirdiet: Glenn, L.P. 1980. Morphometric characteristics of brown bears on applications for studying the foraging ecology of wild otarids. the central Alaska Peninsula. In Bears — TheirBiology and Can. J. Zool. 80: 902–909. Management: Proceedings of the 4th International Conference Kurle, C.M., and Worthy, G.A.J. 2001. Stable isotope assessment on Bear Research and Management, Kalispell, Mont., 21– of temporal and geographicdifferences in feeding ecology of Edited by 24 February 1977. C.J. Martinka and K.L. McArthur. northern fur seals (Callorhinus ursinus) and their prey. Bear Biology Association, Tonto Basin, Ariz. pp. 311–319. Oecologia, 126: 254–265. [Available from Terry D. White, Department of Forestry, Wild- MacHutchon, A.G., and Wellwood, D.W. 2003. Grizzly bear food life, and Fisheries, The University of Tennessee, P.O. Box 1071, habits in the northern Yukon, Canada. Ursus, 14: 225–235. Knoxville, TN 37901-1071, USA.] Mattson, D.J. 1997. Use of ungulates by Yellowstone grizzly bears Haroldson, M.A., Schwartz,C.C., Cherry, S., and Moody, D.S. Ursus arctos. Biol. Conserv. 81: 161–177. 2004. Possible effects of elk harvest on fall distribution of grizzly bears in the Greater Yellowstone Ecosystem. J. Wildl. McLellan, B.N., and Hovey, F.W. 1995. The diet of grizzly bears in Manag. 68: 129–137. the Flathead River drainage of southeastern British Columbia. 73 Hilderbrand, G.V., Farley, S.D., Robbins, C.T., Hanley, T.A.,Titus, Can. J. Zool. : 704–712. K., and Servheen, C. 1996. Use of stable isotopes to determine McLellan, B.N., Hovey, F.W., Mace, R.D., Woods, J.G., Carney, diets of living and extinct bears. Can. J. Zool. 74: 2080–2088. D.W.,Gibeau, M.L., Wakkinen, W.L., and Kasworm, W.F. 1999. Hilderbrand, G.V., Schwartz,C.C., Robbins, C.T., Jacoby, M.E., Rates and causes of mortality of grizzly bears in the interior Hanley, T.A., Arthur, S.M., and Servheen, C. 1999a. The impor- mountains of British Columbia, Alberta, Montana, Washington, tance of meat, particularly salmon, to body size, population pro- and Idaho. J. Wildl. Manag. 63: 911–920. ductivity, and conservation of North American brown bears. McLoughlin, P.D., Taylor, M.K., Cluff, H.D., Gau, R.J., Mulders, Can. J. Zool. 77: 132–138. R., Case, R.L., Boutin, S., and Messier, F. 2003. Demography of Hilderbrand, G.V., Jenkins, S.G., Schwartz,C.C., Hanley, T.A., and barren-ground grizzly bears. Can. J. Zool. 81: 294–301. Robbins. C.T. 1999b.Effect of seasonal differences indietary Miller, S.D.,White, G.C., Sellers, R.A., Reynolds, H.V., Schoen, meat intake on changes in body mass and composition inwild J.W.,Titus, K., Barnes, V.G.,Jr.,Smith, R.B., Nelson, R.R., and captive brown bears. Can. J. Zool. 77: 1623–1630. Ballard, W.B., and Schwartz,C.C. 1997. Brown and black bear Hilderbrand, G.V., Farley, S.D., Schwartz,C.C., and Robbins, C.T. density estimation in Alaska using radiotelemetry and replicated 2004. Importance of salmon to wildlife:implications for inte- mark–resight techniques. Wildl. Monogr. No. 133. grated management. Ursus, 15: 1–9. Nevin, O.T. 2003. The influence of prey abundance and risk- Hobson, K.A., Schell, D.M., Renouf, D., and Noseworthy, E. 1996. sensitive behavioral change on individual access to high-energy Stable carbon and nitrogen isotopic fractionation between diet food (salmon):impacts on the density and viability of bear pop- and tissues of captive seals:implications for dietary reconstruc- ulations. Ph.D. thesis, Utah State University, Logan, Utah. tions involving marine mammals. Can. J.Fish. Aquat. Sci. 53: Paetkau, D. 2003. An empirical exploration of data quality in 528–533. DNA-based population inventories. Mol.Ecol. 12: 1375–1387. Hobson, K.A., Wassenaar, L.I., and Taylor, O.R. 1999. Stable iso- Phillips, D.L., and Gregg, J.W. 2001. Uncertainty in source parti- topes (δD and δ13C) are geographic indicators of natal origins of tioning using stable isotopes. Oecologia, 127: 171–179. monarch butterflies in eastern North America. Oecologia, 120: Phillips, D.L., and Gregg, J.W. 2003. Source partitioning using sta- 397–404. ble isotopes: coping with too many sources. Oecologia, 136: Hobson, K.A., McLellan, B.N., and Woods, J.G. 2000. Using sta- 261–269. ble carbon (δ13C) and nitrogen (δ15N) isotopes to infer trophic Phillips, D.L., and Koch, P.L. 2002. Incorporating concentration relationships among black and grizzly bears in the upper Colum- dependence in stable isotope mixing models. Oecologia, 130: biaRiver basin, British Columbia. Can. J. Zool. 78: 1332–1339. 114–125.

Rausch, R. 1963. Geographicvariation insize in North American Shackleton, D. 1999. Hoofed mammals of British Columbia. brown bears, Ursus arctos L.,asindicated by condylobasal Vol. 3. The mammals of British Columbia. Royal British Co- length. Can. J. Zool. 41: 33–45. lumbia Museum Handbook. UBC Press, Vancouver, B.C. Reynolds, H.V., and Garner, G.W. 1987. Patterns of grizzly bear Sponheimer, M., Robinson, T.,Ayliffe, L., Passey, B., Roeder, B., predation on caribou in northern Alaska. In Bears — TheirBiol- Shipley, L., Lopez, E., Cerling, T., Dearing, D., and Ehleringer, ogy and Management: Proceedings of the 7th International Con- J. 2003. An experimental study of carbon-isotope fractionation ference on Bear Research and Management, Williamsburg, Va., between diet, hair, and feces of mammalian herbivores. Can. J. 21–26 February 1986, and Plitvice Lakes, Yugoslavia, 2–5 Zool. 81: 871–876. March 1986. Edited by P. Zager. International Association for Stringham, S.F. 1990. Grizzly bear reproductive rate relative to Bear Research and Management, Washington, D.C. pp. 59–67. body size. In Bears — TheirBiology and Management: Pro- [Available from Terry D. White, Department of Forestry, Wild- ceedings of the 8th International Conference on Bear Research life, and Fisheries, The University of Tennessee, P.O. Box 1071, and Management, Victoria, B.C., 20–25 February 1989. Edited Knoxville, TN 37901-1071, USA.] by L.M. Darling and W.R. Archibald. International Association Robbins, C.T.,Hilderbrand, G.V., and Farley, S.D. 2002. Incorpo- for Bear Research and Management, Washington, D.C. pp. 433– rating concentration dependence in stable isotope mixing models: 443. [Available from Terry D. White, Department of Forestry, a response to Phillips and Koch (2002). Oecologia, 133: 10–13. Wildlife, and Fisheries, The University of Tennessee, P.O. Box 1071, Robbins, C.T., Schwartz,C.C., and Felicetti,L.A. 2004. Nutri- Knoxville, TN 37901-1071, USA.] tional ecology of ursids: areview of newer methods and man- Szepanski,M.M., Ben-David, M., and Van Ballenberge, V. 1999. agement implications. Ursus, 15: 161–171. Assessment of anadromous salmon resources in the diet of Al- Robbins, C.T., Felicetti,L.A., and Sponheimer, M. 2005. The ef- exander Archipelago wolf using isotope analysis. Oecologia, fect of dietary protein quality on nitrogen isotope discrimination 120: 327–335. in mammals and birds. Oecologia, 15: 534–540. Tabachnick, B.G., and Fidell, L.S. 1996. Using multivariate statis- Rode, K.D., and Robbins, C.T. 2000. Why bears consume mixed tics. Harper Collins, New York. diets during fruit abundance. Can. J. Zool. 78: 640–645. Thomas, J.W., and Toweill, D.E. (Editors). 1982.Elk of North Rode, K.D., Robbins, C.T., and Shipley, L.A. 2001. Constraints on America: ecology and management. Stackpole Books, Harris- herbivory by grizzly bears. Oecologia, 128: 62–71. burg, Pa. Roth, J.D., and Hobson, K.A. 2000. Stable carbon and nitrogen Welch, C.A.,Keay,J., Kendall, K.C., and Robbins, C.T. 1997. isotopic fractionation between diet and tissue of captive red fox: Constraints on frugivory by bears.Ecology, 78: 1105–1119. implications for dietary reconstruction. Can. J. Zool. 78: 848– Welch, D.W., Ishida, Y., and Nagasawa, K. 1998. Thermal limits 852. and ocean migrations of sockeye salmon: long-term consequences Satterfield, F.R., and Finney, B.P. 2002. Stable isotope analysisof of global warming. Can. J.Fish. Aquat. Sci. 55: 937–948. Pacific salmon:insight into trophic status and oceanographic Zullinger, E.M.,Ricklefs, R.E., Redford, K.H., and Mace, G.M. conditions over the last 30 years. Prog. Oceanogr. 53: 231–246. 1984.Fitting sigmoidal equations to mammalian growth curves. Schwartz,C.C.,Miller, S.D., and Haroldson, M.A. 2003. Grizzly bear. J. Mammal. 65: 607–636. In Wild mammals of North America. Edited by G.A.Feldhamer, B.C. Thompson, and J.A. Chapman. The John Hopkins University Press, Baltimore, Md.

PAPER X

1 Garth Mowat, Douglas C. Heard and Carl J. Schwarz

3 Predicting grizzly bear density in North America

5 G. Mowat, BC Fish, Wildlife and Habitat Division, Suite 401, 333 Victoria St., Nelson,

6 British

7 Columbia, V1L 4K3, Canada.

9 D.C. Heard, BC Fish, Wildlife and Habitat Division, 4051 - 18th Avenue, Prince George,

10 British Columbia V2N 1B3, Canada.

12 C. J. Schwarz, Dept. of Statistics and Actuarial Sciences, Room SC K10545

13 8888 University Drive, Simon Fraser University, Burnaby, British Columbia V5A 1S6,

14 Canada.

16 Draft date: December 31, 2011; since published as Mowat et al. 2013 PLOS ONE 82757

18 Corresponding author:

19 G. Mowat

20 BC Fish, Wildlife and Habitat Division

21 Suite 401, 333 Victoria St., Nelson

22 British Columbia, V1L 4K3, Canada.

23 Tel. 011 250 354-6341 FAX 011 250 354-6332

1 24 email: [email protected]

2 26 Abstract Conservation of grizzly bears is often controversial and the balance of

27 disagreement is focused on the estimates of density that are used to calculate allowable

28 kill. Thanks to the development of aerial survey and genetic identification techniques,

29 many recent estimates of grizzly bear density are now available. But, grizzly bears are

30 hunted across vast areas and field-based estimates will never be available for more than a

31 small portion of the hunted populations. Current methods of predicting density to areas of

32 management interest are subjective and untested. Objective methods have been proposed

33 (e.g., using RSF’s) but these statistical models are so dependent on results from

34 individual study areas that the models do not generalize well. We built Tobit regression

35 models to relate grizzly bear density to ultimate measures of ecosystem productivity and

36 mortality for interior and coastal ecosystems. Coastal ecosystems were those areas where

37 bear density was strongly linked to salmon abundance and black bears were present. We

38 had too few data to build a model for areas where salmon were abundant but black bears

39 were absent. We used 89 measures of grizzly bear density in interior ecosystems of

40 which 14 were currently known to be unoccupied by grizzly bears. In coastal areas where

41 black bears were present we used 17 measures of density including 2 unoccupied areas.

42 Our best model for coastal areas included a negative relationship with tree cover and

43 positive relationships with the salmon diet proportion and topographic ruggedness which

44 was correlated with precipitation. Our best interior model included 2 variables that

45 indexed terrestrial productivity, 1 describing vegetation cover, 2 indices of diet, and 2

46 indices of human use of the landscape. We used our models to predict current density and

47 population sizes across Canada and present these as alternatives to current population

48 estimates. Our models predict a fewer grizzly bears in BC but a higher number of bears in

3 49 Canada than that provided in the latest status review. These predictions can be used to

50 assess population status, set mortality limits, and for conservation planning but because

51 our predictions are static, they cannot be used to assess population trend. In addition to

52 the prediction provided by our model, managers should also consider other information

53 [e.g., local knowledge about distribution and movements, major food sources such as

54 salmon, hunter success, age and sex ratios of hunter kills, the frequency of problem bear

55 occurrences and the precision of our model’s predictions] before deciding on a final

56 population prediction for an area.

58 Keywords abundance, carrying capacity, density, grizzly bear, population size,

59 predictive modeling, Ursus arctos,

60 61 Introduction

62 63 Grizzly bear hunting is controversial because of peoples’ conflicting values and interests.

64 Sarewitz (2004) argued that progress toward solving environmental controversies will

65 need to come primarily from political processes rather than from scientific research. He

66 regards as old-fashioned, the idea that scientific facts and theories will help settle disputes

67 or build the appropriate foundations for guiding environmental policy and suggests that

68 the role of science is to collect information to support the implementation of policies

69 determined through political processes. Where grizzly bear hunting opportunity is

70 managed by a quota system a maximum allowable kill rate is prescribed (the policy)

71 which is then applied to estimates of population size (the scientific information). This

72 paper describes an approach to predict grizzly bear densities that can be used support,

73 monitor and evaluate policy decisions.

4 74 In the last 15 years, improvements in the development of aerial survey (Miller et

75 al. 1997) and genetic identification techniques (Woods et al. 1999) have led to a

76 proliferation of grizzly bear density estimates (Proctor et al. 2010). However, due to the

77 high cost and the vast areas involved, field-based density estimates have been and likely

78 will continue to be infrequent and restricted to a small subset of hunted populations.

79 Two approaches for predicting grizzly bear density have been proposed where

80 field-based estimates are unavailable. 1. Measures of grizzly bear abundance can be

81 generated by assigning densities based on expert opinion regarding the value of landcover

82 attributes, supported by, or in conjunction with, field estimates derived in similar

83 ecosystems (Fuhr and Demarchi 1991). No expert-based models to date have estimated

84 confidence limits for the resulting density estimates so evaluating conservation risk was

85 subjective. Expert models have not considered fundamental concepts, such as whether

86 microsite attributes sum or scale up to provide an indication of landscape scale density

87 (Freckleton et al. 2005). In spite of their shortcomings, expert-based models have been

88 used in all jurisdictions in Canada where grizzly bear hunting is allowed (Nagy and

89 Gunson 1990, Fuhr and Demarchi 1991; Hamilton et al. 2004, McLoughlin 2011)

90 2. Resource selection function models can be used to predict the absolute or

91 relative probability of occurrence (Boyce and Waller 2003, Apps et al 2004) and possibly

92 density (Nielsen et al 2005) within small ecological units. While occurrence models may

93 be statistically explanatory and objective, considerable subjectivity may be required when

94 deciding how and where to apply them (e.g., Boyce and Waller 2003). Models vary with

95 the local availability of resources and behaviors related to regional life history or human

96 influence, and they do not generalize well to other landscapes (Boyce and MacDonald

5 97 1999; Mysterud and Ims 1999; Apps et al. 2004, Johnson et al. 2004, Ciarniello et al.

98 2007, Beyer et al. 2010). This is a considerable problem given that grizzly bears occupy a

99 wide range of environments and have many different life history strategies.

100 For the greatest general application, predictive abundance models must be

101 underpinned by an understanding of the functional processes affecting density, use direct

102 measures of resource abundance and, apply to all environments and life history strategies

103 (Carbone and Pettorelli 2009). Recent work has quantified relationships between

104 abundance and landscape scale measures of environmental attributes for white-tailed deer

105 (Crete 1999), carnivores (Carbone & Gittleman 2002), tassel-eared squirrels (Prather et

106 al. 2006), coyotes (Kays et al. 2008), kanagroos (Ritchie et al. 2008) and people (Beck

107 and Sieber 2010). Most of these models included one or more measures of food indexed

108 either by a direct measure of the resource such as tree basal area for squirrels or remote

109 measures such as temperature, precipitation and soil type for humans.

110 We considered both bottom–up (food supply) and top-down (competition and

111 predation) influences on grizzly bear abundance. Grizzly bears are omnivores and their

112 reliance on animal protein varies greatly across their range (Mattson et al. 1991,

113 McLellan and Hovey 1995, Hilderbrand et al. 1999). The single largest meat source in

114 their diet is spawning salmon and all areas of very high bear density have large numbers

115 of salmon over a large portion of the non-denning season (Miller et al. 1997, Hilderbrand

116 et al. 1999). In the continental interior, grizzly bears eat insects, rodents, freshwater fish,

117 and ungulates where available and supplement their diet with plants (Mattson et al. 1991,

118 McLellan and Hovey 1995, Hilderbrand et al. 1999). Bears prefer highly digestible plant

119 species and parts because they have a single gut and are relatively inefficient at digesting

6 120 plant matter (Rode and Robbins 2000, Rode et al. 2001). Most preferred plant species can

121 be considered hydrophilic and bears in the interior concentrate their foraging in moist

122 sites (McLellan and Hovey 2001, McLoughlin et al. 2002). The exception being when

123 they are digging for corms of various preferred plants. Fruits of several shrubs are highly

124 preferred in late summer and fall and, along with meat, are the basis for the fat deposition

125 required for winter hibernation (Robbins et al 2004). The abundance of grizzly bear

126 foods is not only related to ecosystem productivity but also, for bears that live in forested

127 environments, to successional stage, (Neilsen et al. 2004).

128 Competition may limit grizzly bear density where they are sympatric with black

129 bears (Herrero 1972, Mattson et al. 2005). Black bears and grizzly bears have similar

130 digestive and foraging efficiencies (Pritchard and Robbins 1990, Welch et al. 1997, Rode

131 et al. 2001) and any competitive advantage of black bears over grizzly bears feeding on

132 plants and fruits is based largely on the smaller body size of black bears (Welch et al.

133 1997, Rode et al. 2001). Where grizzly bears rely on meat they often appear to exclude

134 black bears from the meat source (Miller et al. 1997, Jacoby et al. 1999, Mowat et al.

135 2005), probably because the meat source is clumped and therefore defendable, unlike

136 plant and fruit supplies. Areas that are largely devoid of trees do not support black bears

137 where their range overlaps that of grizzly bears (Herrero 1972, 1978, Mattson 1991,

138 Miller et al. 1997, Mowat et al. 2005). Presumably grizzly bears exclude black bears from

139 large open areas because black bears cannot seek refuge from grizzly bear aggression by

140 climbing trees (Herrero 1972, 1978, Gunther et al. 2002).

141 Grizzly bears have no significant predators (Swenson et al. 1997, McLellan

142 2005) but social factors may limit bear numbers. For example, if large males are present,

7 143 female bears may avoid important feeding areas to minimize the chance of male bears

144 killing their accompanying offspring (Ben-David et al. 2004). Several different

145 hypotheses have been proposed regarding social regulation in grizzly bears (McLellan

146 2005), however a recent analysis of field data for 4 sites did not support the importance

147 of any form of social regulation on density (Miller et al. 2003).

148 The presence of humans and our activities limits grizzly numbers by direct

149 mortality, habitat loss, and displacement due to disturbance. Mortality obviously reduces

150 density temporarily but the relationship between mortality rate and density is complex

151 due to the effects of density-dependence and age ratios on vital rates. Habitat loss, and

152 environmental change that completely precludes occupancy by grizzly bears, obviously

153 reduces density. Disturbance has been shown to reduce grizzly bear density at fine scales

154 (e.g., along road corridors, Ciarniello et al. 1998) but the link between disturbance and

155 landscape scale population density, although largely accepted by practitioners (Schwartz

156 et al. 2003), has never been demonstrated. Human density may index the above 3 factors

157 but the functional link to bear density is unclear.

158 In this paper we modeled the relationship between existing grizzly bear density

159 estimates and potential limiting factors. We then use those relationships to predict grizzly

160 bear density across a large and varied portion of their Canadian range and demonstrate

161 how the use of these predictions for setting hunting quotas and evaluating past levels of

162 mortality can support the process of developing grizzly bear conservation policy.

163

164 Materials and methods

165 166 Model development

8 167

168 Based on previous research we felt that the following factors may functionally influence

169 grizzly bear density at the population scale: plant productivity, vegetation type, fish and

170 meat availability, scramble competition with black bears, human disturbance, and human-

171 caused mortality. Based on this functional model we began by assembling and deriving

172 indices for these factors while attempting to choose measures that are not highly

173 correlated with other factors to avoid collinearity (Table 1). Variables that described food

174 limitation were: salmon availability, the proportion of salmon or ungulates and other non-

175 plant foods in the diet, remote sensing measures of plant productivity, and the proportion

176 of the landscape covered by trees and herbaceous vegetation. Competition with black

177 bears may be best described by measures of black bear abundance, but these were so few

178 that we used tree cover or simple presence or absence of black bears as surrogates for

179 more direct measures of competition.

180 The only mortality factor we considered was direct killing by people. Although it

181 has been mandatory to report all human-caused deaths in all jurisdictions in North

182 America since the 1970’s, a substantial proportion of that mortality goes unreported

183 (McLellen at al. 1999). The direct effects of humans cause habitat loss in urban areas and

184 is accounted for in the remote sensing vegetation measures above. We used human and

185 livestock census information as measures of both the loss of habitat effectiveness due

186 behavioral decisions of bears to avoid areas frequented by humans and the impact of

187 unrecorded human-caused mortality. Other surrogates, such as road density or the

188 number or proportion of problem grizzly bears killed, were difficult to standardize across

189 jurisdictions and were more temporally variable. We did not consider predation, diseases,

9 190 parasites or social limitation because there is no evidence that those were general limiting

191 factors. Hence our three over-arching hypotheses were based on food limitation, scramble

192 competition with black bears and, human limits to density via mortality, habitat loss, and

193 decreased habitat effectiveness.

194

195 The dependent variable - meta analysis of grizzly bear abundance 196

197 We critically reviewed estimates of grizzly bear population size or density in the

198 published and unpublished literature. We were interested in estimates for landscapes

199 large enough to represent a grizzly bear population affected largely by births and deaths

200 rather than immigration and emigration so we only used data where study area size was

201 >800 km2 (range 789-22,875 km2) or approximately 10 female home ranges, and

202 contained at least 15 resident bears ( xˆ = 80, range 15-765). We used all population

203 estimates that met the above criteria and where we judged the authors had done enough

204 field sampling to generate an estimate that was indicative of the ecosystem. We noted

205 whether authors accounted for bears that were not detected during fieldwork, generally

206 this meant the use of mark-recapture analysis but we accepted more subjective

207 assessments for some intensive census studies. We also noted whether authors had

208 accounted for closure bias which is the positive bias caused by movement during the

209 study as well as the inclusion of individuals that only partly resided in the study area. We

210 indexed the accuracy of each population estimate by standardizing the confidence

211 interval (CI) width as a percent of the point estimate. We arbitrarily doubled the width of

212 the reported upper confidence limit (UCL) if authors did not consider incomplete

213 detection or the lower confidence limit (LCL) if authors did not account for closure bias.

10 214 Where no measure of precision was given we assigned CIs based on survey effort. We

215 arbitrarily assigned the LCI as 50% of the point estimate, if the authors considered

216 closure bias then we reduced the LCI to 25% of the estimate. We doubled the point

217 estimate to index the UCI unless the authors accounted for incomplete detection

218 explicitly in which case we reduced the UCI 50% of the point estimate. The subjectively

219 derived confidence limits were larger than most probability based limits and they

220 accounted for the fact that the minimum number of animals on a study is usually more

221 accurately estimated than the maximum, especially using census methods.

222 We selected 16 study areas in areas that were historically occupied by grizzly

223 bears but are currently not occupied by grizzly bears. These areas were all adjacent to

224 occupied areas and there was no known barrier to dispersal. We chose these areas to

225 represent the range of contemporary forces that work to exclude grizzly bears from parts

226 of their range. We selected areas of similar size to other studies in those ecosystems and

227 derived measures for independent variables as for occupied study areas. Though the

228 density was zero based on local knowledge we assigned upper CIs based on trapping

229 results, non-hunter kills and the recent record of bear sightings in the study area.

230 We revised all grizzly bear density estimates by removing the area of water, rock

231 and bare ground because we considered these unsuitable to bears.

232

233 Independent variables - derivation of surrogates for limiting factors

234

235 We derived average annual precipitation, average annual temperature, NDVI and

236 evapotranspiration to index plant productivity from freely available spatial databases

11 237 (Table 2). Study area boundaries were digitized and mean values for each variable were

238 calculated for each study area excluding areas of open water and barren areas which was

239 up to 35% of the study area (mean = 6.5%). For five study areas outside of our

240 precipitation map we used data from the nearest Environment Canada long-term weather

241 records.

242 Recent efforts to predict world agricultural production and the impact of global

243 warming have led to the development of numerous indices of plant growth and biomass

244 (Kozak et al. 2008). Several composite variables of factors that limit plant productivity

245 are mapped across large geographic areas. Evapotranspiration (AET) is a measure of the

246 water balance and energy available in an environment and is related to primary plant

247 productivity, species diversity and ungulate biomass (Rosenzweig 1968, Currie 1991,

248 Crete 1999). AET is a composite index of the two most limiting resources to

249 photosynthesis, water and solar radiational energy (Rosenzweig 1968). The normalized

250 differential vegetation index (NDVI) is a measure of plant vigor and can be equated to

251 above ground vegetation productivity and biomass. It is derived from a complicated set of

252 algorithms based on remote sensing data (Bannari et al. 1995). The NDVI we used was

253 based on MODIS data averaged across a single year. NDVI is a simplification of the

254 greenness index which has been used to describe grizzly bear habitat selection (Mace et

255 al. 1999). AET and NDVI combine the effects of soils, water and energy into a single

256 index and are based on similar or identical data. These, and other remote sensing based

257 measures of plant production, are often tightly correlated (the correlation between AET

258 and NDVI in our interior data was r = 0.73, N = 90).

12 259 We also derived a measure of topographic slope termed ruggedness (Riley et al

260 1999) which we included as a covariate to index the increased land area associated with

261 sloped areas. This variable was correlated to precipitation (r = 0.73, N = 90) but no other

262 productivity variables (r < 0.28, N = 90).

263 Landcover has been assigned globally into 3 structural classes: herbaceous (which

264 includes shrubs <2 m in height), trees >2 m tall, and barren (500 m pixels; Hansen et al.

265 2003; Table 1). MODIS data was used to assign percent cover values for each pixel for

266 forest, herb-shrub and barren and these three values totaled to one (Hansen et al. 2003).

267 We surmised that in areas where the two bear species are sympatric, grizzly bears would

268 monopolize the resources in open portions of the landscape and we used the proportion of

269 each study area which was tree covered to index inter-specific interaction while

270 recognizing that the lack of tree canopy may also increase the vegetation resources

271 available to bears. We calculated the proportion of forest in the vegetated portion of each

272 study area by summing the proportion of pixels that were rated as >25% forest. We

273 assumed that black bears would have adequate escape trees in areas that were >25% tree

274 covered. Conversely we derived a measure of herb-shrub cover to index vegetative forage

275 under the general assumption forage is more abundant in non-forested areas. We

276 calculated the proportion of herb-shrub area by summing the pixels with >50% herb-

277 shrub cover. This cut-off value assumed each pixel was dominated by herb-shrub cover.

278 Pixels that were rated as 100% barren were excluded from spatial calculations (using a

279 GIS mask); these areas were mostly rock and ice.

280 We used the fraction of salmon and animal tissue in the diet of bear populations

281 as surrogates for salmon and terrestrial meat availability. Diet fractions may not be

13 282 linearly related to resource availability but deriving measures of salmon and meat

283 availability across large areas was not feasible. Salmon and terrestrial meat components

284 of the diet were predicted using isotope analysis of grizzly bear hair collected from each

285 study area (Mowat and Heard 2006). Where we did not have hair samples to estimate diet

286 we calculated mean values in each study area from spatial raster-based maps built from a

287 continent wide diet dataset (Mowat and Heard 2006, digital data available in SOM).

288 Unoccupied areas were necessarily assigned diet fractions in this way. We based diet

289 fractions on direct feeding observations for two study areas (Nome and Midsu) because

290 the model predictions suggested no salmon in diet and this was known to be incorrect

291 (Miller et al. 1997). For one area (Tweedsmuir) we recalculated diet using just the

292 samples collected in the study area because the map-derived values included areas where

293 salmon were not available and were likely inaccurate. We also applied this value to the

294 nearby Kimsquit-Dean area because the mapped diet was likely biased low due to a

295 drastic decline in salmon numbers in the nearby Owikeno Sound (Boulanger et al. 2004

296 and unpubl. data). Kokanee (Oncorhynchus nerka), a landlocked form of sockeye salmon

297 (O. nerka), were considered part of the salmon component of the diet. Kokanee diet

298 fractions presented in Mowat and Heard (2006) were reduced by half based on further

299 data collected from kokanee in central British Columbia (D. Heard, unpubl. data). We

300 also derived a categorical variable indexing the importance of salmon where 0 meant no

301 salmon available, 1 for areas with little salmon such as some interior areas and 2 for those

302 areas where bears were considered to derive most of their resources from salmon.

303 Human and livestock density was used to index human displacement, disturbance

304 and unreported bear mortality. We tested log transformations of these variables because

14 305 the influence was expected to be nonlinear (Mattson and Merrill 2002). The number of

306 people and, cattle and sheep were calculated from polygon-based census data for the US

307 (2000) and Canada (2001). Count data was used to calculate density for each census unit

308 and density was used to calculate the number of resident people and livestock in each

309 study area. In order to minimize the number of variables we added the human, cattle and

310 sheep counts together to produce a single composite variable; this index combined both

311 the various human impacts with the threat of direct mortality posed by livestock grazing.

312 Human caused mortality necessarily reduced bear density and was entered

313 directly and as a squared term to account for the non-linear influence of recent mortality

314 on the standing population. We estimated the number of bears killed by people from

315 government databases or published accounts. Counts of all legally killed bears have been

316 recorded since at least the mid 1970s for all the jurisdictions in this study. These data

317 represent a minimum number of bears killed by people. We calculated the annual kill rate

318 (number bears killed/bear population estimate) over the 10 years previous to each density

319 estimate. Unoccupied areas were assigned mean kill rates as an attempt to infer a

320 reasonable value for these areas if bears had been present..

321

322

323 Statistical analysis 324 325

326 We used Principal Components Analysis (PCA; Tabachnick and Fidell 1996) to contrast

327 >3 variables that were surrogates for the same limiting factor. Principal component

15 328 scores were considered but not used because of the difficulty in interpreting the resulting

329 regression equation (Guthery 2008).

330 Ordinary least squares regression cannot be used directly to fit these models

331 because of the inclusion of study areas where no grizzly bears were located. For these

332 points, the assumption of normally distributed errors about the regression line is violated.

333 Consequently, Tobit regression (Tobin, 1958) was used. In this model, a latent (hidden)

334 variable (Y * ) follows the ordinary linear model:

335 Y* XE H 336

337 However, only the max(0, Y * ) can be observed, i.e. it is impossible to observe negative

338 densities.

339 Maximum likelihood was used to fit the Tobit model (Proc QLIM, SAS 9.2).

340 Models were fit where every study area was given equal weight and where study areas

341 were weighted by the inverse of the relative CIs. The weighted Tobit regression used the

342 inverse of the relative CIs to downweight the influence of study areas with poor estimates

343 of density.

344 We constructed an a-priori suite of models based on expert judgment including

345 variables thought to be influential for grizzly bear densities. Additional models were

346 added to the initial model set where potential predictors were dropped or transformed.

347 Potential models included transformed values for human and livestock density to

348 approximate the known form of the relationship based on previous research as described

349 earlier. We also included indicator variables that controlled for 3 cases we considered

350 outliers based on initial screening of the data. Two study areas had high and presumably

351 unsustainable mortality rates, and one unoccupied area had very high human density. We

16 352 used the small sample corrected AICc to compare model fit and ranked models using AIC

353 weights (Burnham and Andersen 2002) and we examined top ranked models for the

354 presence of uninformative variables (Arnold 2010). We investigated the leverage of each

355 record in the top fitting models using influence plots based on simple regression to check

356 for individual study areas that may account for the inclusion of a predictor in the model.

357 Residuals from the top fitting models were examined for outliers and distribution

358 (Tabachnick and Fidell 1996).

359 We used a leave-one-out jack-knife procedure to estimate the increase in

360 prediction error that might occur when the model is used to predict densities outside of

361 the set used to build the model. Each individual study area was sequentially dropped

362 from the analysis, the remaining data were used to fit the model being examined, this

363 fitted model with sample size n-1 was then used to predict density for the study area held

364 out.

365

366 Results

367 368 We derived 118 estimates of grizzly bear density including 16 estimates for areas

369 that were currently unoccupied by grizzly bears. This total included 2 repeated

370 inventories for 6 study areas that were carried out in different years. The number and

371 precision of those inventories was greatest in the southeast portion of the range of grizzly

372 bears and lowest in Yukon and the Northwest Territories (Fig. 1).

373 In coastal areas, where salmon were abundant (i.e., >27% of diet), grizzly bear

374 density estimates were up to an order of magnitude higher than densities in interior areas

375 (Fig. 2). In addition, coastal grizzly bear density estimates were much higher in areas

17 376 where black bears are absent than where they were present (Table 3). The one exception

377 was the Kuskowim delta in westcentral Alaska. This area had a modest density and the

378 diet contained more terrestrially derived meat (presumably caribou) than salmon or

379 vegetation. Not only did coastal areas where black bears were present have a much lower

380 grizzly bear density but the proportion of salmon in the grizzly bear diet was also more

381 variable (Table 4). We concluded the availability of salmon led to fundamentally

382 different ecological relationships in coastal areas, so we built separate models for coastal

383 and interior areas, based on an arbitrary cut-off of 20% salmon in the diet. We further

384 separated the coastal data into those areas where black bears were sympatric with grizzly

385 bears and those where black bears were absent. We did not attempt to build a predictive

386 model for coastal areas where black bears were absent because there were no areas like

387 this in BC.

388 In interior areas density varied from 2.5 to 65 grizzly bears/1000 km2 and there

389 was a broad range of values for most independent variables (Table 3). Human and

390 livestock density was higher for unoccupied areas, presumably because humans were the

391 ultimate cause of the extirpation in many of these areas. Expectedly, there were no

392 coastal areas with low precipitation (Table 4).

393 In interior areas, the first eigenvector of a PCA suggested all 5 vegetation

394 productivity variables were correlated and that ruggedness and NDVI contrasted in the

395 second eigenvector. These results suggested ruggedness and NDVI should be included in

396 the multivariate analysis but that any of the five productivity variables could substitute

397 for one another. For this reason we decided to include all five variables in the analysis. A

398 second PCA with the vegetation type variables also demonstrated strong correlation

18 399 among 4 of the 5 variables, while the fifth variable, herb100, was nearly invariant across

400 the data. These results suggested that any one of the vegetation type variables could be

401 used to index vegetation cover. We chose herb50 because we felt it would index bear

402 food with the most sensitivity. We also included tree25 in the global model to index black

403 bear competition which was supported by comparing black bear presence across

404 vegetation cover. Although the vegetation cover variables were strongly correlated the

405 presence of black bears was most clearly separated across the tree25 variable (Fig. 3). A

406 third PCA for the human influence variables showed essentially a single component that

407 was an average of all 5 variables. Both human and livestock density showed a threshold

408 with density which suggested a non-linear relationship. We excluded the 2 variables that

409 measured human and livestock density surrounding each study area because there was no

410 indication that density trended to zero at high human or livestock density (Mattson and

411 Merrill 2002).

412 There was very little variation in salmon in diet in interior areas. The composite

413 of fish and, fish and terrestrial meat in the diet was dominated by the meat component.

414 For this reason the food composite variables were not considered further in the analysis.

415 In coastal areas with black bears, a PCA for the vegetation productivity variables

416 yielded similar results to the interior dataset and we considered all five variables for the

417 same reasons as above. We also chose to include herb50 and tree25 because these two

418 variables may contrast herbaceous food abundance with competition with black bears.

419 Human density was the only human influence variable that had substantial variation

420 across the dataset and was the only variable we included. Diet was dominated by salmon

421 and this was the only diet variable considered (Table 4).

19 422 423 Tobit regression for interior areas 424 425 Based on biological considerations and the above investigation of the data we included

426 the following variables in our global regression: precipitation, NDVI, AET temperature,

427 ruggedness, herb50, tree25, salmon-in-diet or salmon-presence, meat-in-diet, human

428 density, livestock density, or, human + livestock density, human-caused mortality, and

429 (human-caused mortality)2. Log transformations were also compared for the human-

430 influence variables. When we compared the effect of salmon-in-diet versus the salmon-

431 presence variable, only the salmon-presence variable was consistently in the top models

432 and we therefore excluded the salmon-in-diet variable.

433 When we compared the prediction strength of human and livestock density versus

434 the single composite variable. The composite variable was not included in the top

435 competing models. A Log transformation did not improve the fit of the composite

436 variable. This suggests the form of the relationship between human density and bear

437 density differed from that of livestock density and bear density so we dropped the

438 summed variable, human plus livestock density .

439 One unoccupied area had double the human density of the next lowest value and

440 was considered an outlier. This study area (Thompson) and the two other study areas

441 (Upper Susitna and Swan Hills) where the number of grizzly bears killed by people was

442 very high were modeled using indicator variables in order to test their influence on fit. No

443 models that included these indicator variables were included in the top models suggesting

444 these cases did not unduly leverage the analysis (Table 5). Salmon-presence and meat-in-

445 diet were usually in the top models and the regression coefficient for the salmon variable

446 was positive while, surprisingly, that of the meat variable was negative.

20 447 The weighted Tobit models gave highest AIC weight (18%) to the global model

448 but this model had 19 parameters while the next model had similar wieght (16%) and an

449 AIC value that was only 0.3 higher than the global model. We chose to exclude the global

450 model on the basis that it contained noise parameters (Arnold 2010) and considered

451 Model 2 to be our best model (Table 5). Model 2 included: annual precipitation (sign

452 positive), annualized NDVI (+), average annual AET (+), the proportion of pixels in the

453 study area with more than 50% herb-shrub coverage (+), log human density (-), livestock

454 density (-) and ruggedness (+). The top 3 models had similar weight and differed by the

455 inclusion of tree cover (-) and the proportion of terrestrial meat (-) in the diet and the

456 exclusion of NDVI. Rankings of models were relatively consistent between the weighted

457 and unweighted analysis and we present only the weighted analysis here.

458 Residual plots, the difference between the observed study area densities and the

459 density predicted for those areas using Model 2, showed no evidence that the regression

460 assumptions were not met (Tabachnik and Fidell 1996) . There was considerable

461 variation in the residuals but the largest residuals were associated with the observed

462 densities with the widest CIs (Fig. 4). For computational ease, ordinary least squares was

463 used to compute approximate partial regression plots (leverage plots, Cook and

464 Weisburg, 1982). These showed no serious outliers. The standard error of all predictions

465 were approximately 10.5 bears/1000 km2. Most CIs of the observed densities overlapped

466 the 1:1 line (Fig. 4). Because the study areas used in the model fitting come from a wide

467 geographical area, no investigation or modeling of the spatial structure of the residuals

468 was performed; model errors were assumed to be independent among study areas. For 4

469 of the 90 study areas, the observed estimate fell outside the approximate 95% prediction

21 470 interval for that site but in all cases, the CI for the observed estimate overlapped the CI

471 for the regression (Fig. 4).

472 The mean model error for all Jackknife runs was compared to the mean error of

473 the model with all data included. For the top 10 interior models, the mean squared

474 prediction error was 9-13% greater using the jackknife procedure compared to the fit

475 using all of the data. Model 1 the global model, which we excluded, was 12% higher

476 while Model 2, our preferred model, was 8% higher (Table 7 in SM). The increase in

477 prediction error for the interior data set was modest and indicates that predictions for

478 study areas outside those used in fitting the model, should be reliable. Model averaged

479 predictions and predictions from Model 2 were similar and further justified considering it

480 our best model (Fig 5 SM).

481

482 Tobit regression for coastal areas 483 484 485 Based on biological considerations and the above investigation of the data we included

486 the following 7 variables in our global regression: precipitation, NDVI, AET,

487 temperature, ruggedness, herb50, tree25, salmon-in-diet, and human density or Log

488 human density. The top model in the weighted Tobit analysis had 7 variables and the

489 second model had 6 variables. We excluded both models based on the sample size to

490 parameter ratio, the Jackknife analysis results (Table 8 SM) and, because the third model

491 included only 3 variables and the AIC value was only 2.2 higher than the top model

492 suggesting the other 4 variables in the top model were largely uninformative (Arnold

493 2010). Our preferred model included the proportion of pixels in the study area with more

494 than 25% tree cover (-),the proportion of salmon in the diet (+), and ruggedness (+).

22 495 Rankings of models were relatively consistent between the weighted and unweighted

496 analysis and we present only the weighted analysis here.

497 The jackknife procedure was also used to compare among coastal models. The

498 mean square prediction error from the jackknife models varied from 30-112% greater

499 than that of the fitted model; it was 30% higher for Model 3 our preferred model (Table 2

500 SM). Because the coastal model is based on a relatively small number of data points

501 compared to the number of predictors, Models 1 and 2 may have been fitting artifacts of

502 the coastal study areas.

503 Residual plots using Model 3 showed no evidence of lack of fit. For

504 computational ease, ordinary least squares was used to compute approximate partial

505 regression plots (leverage plots, Cook and Weisburg, 1982). These showed no serious

506 outliers. Plots of predicted versus observed densities showed modest variation of

507 residuals (Fig. 6). Model averaged predictions and Model 3 predictions were similar

508 further justifying our use of Model 3 to predict density (Fig. 7 SM).

509

510 Model predictions

511

512 Our models can be used to predict grizzly bear density for any area for which data exist

513 for the input variables and those predictions can be added to derive a population number

514 for a larger area. We could not compute estimates of the uncertainty in composite

515 predictions (i.e. for the total over several study areas) because predictions for study areas

516 that are geographically adjacent with similar covariate sets are unlikely independent.

517 Combining the uncertainties of the individual predictions will underestimate the

23 518 uncertainty of the total. Without further information about the spatial structure of

519 predictions from neighboring geographical areas, it is unclear how to compute an

520 appropriate measure of uncertainty for the total over multiple study areas. This is an area

521 for further research.

522 We predicted grizzly bear densities in for Canada by summing the individual

523 predictions within wildlife management units (Table 10). We used the interior Model 2

524 for all areas except coastal British Columbia where were used coastal Model 3 (Table

525 12SM)

526 Predicted grizzly bear densities in the Northwest Territories and Nunavut varied

527 from zero in southcentral NWT to >30 bears/1000 km2 in parts of western arctic coast,

528 densities decline to the east and are much lower east of the Mackenzie river. In Yukon,

529 the model predicted densities that vary from 0-31 bears/1000 km2 (Table 11). The lowest

530 density was for the Labarge unit surrounding the city of Whitehorse; the only place in the

531 territory that had high human density. The Kluane area of southwest Yukon had higher

532 precipitation than the range of data in the model (234 cm). This prediction may be biased

533 high if the relationship with density is not linear but rather levels off beyond the range of

534 the input data (Table 3). The average kill rate among population units was 1.3% but was

535 <1% when calculated for the entire territory. Analysis at the finer scale points to the

536 Larbarge unit which has about 3 kills/year and predicted density of zero. This area may a

537 habitat sink for grizzly bears. Similarly, the Dezadeash and Arkell units have mortality

538 rates >5% and may be of conservation concern.

539 Predicted densities for BC varied from 0 to 58 bears/1000 km2. The highest

540 coastal densities were in northern areas where salmon consumption was high and tree

24 541 cover was low. Interior areas with high density occurred throughout the province but

542 were always rugged areas with high rainfall and few people. Many units in the province

543 were predicted to have low density and, while this was often associated with high human

544 density, predicted densities in flat areas with low rainfall and low herb-shrub cover were

545 also low. The predicted mortality rate was 2.6% / yr with most mortality from hunter kills

546 (mean=286/yr) but problem bear kills, and road and rail collisions, (mean=55/yr)

547 comprised a greater proportion of deaths in units with low predicted bear populations or

548 relatively high human density (Table 12 SM). As in Yukon, our predictions point to a

549 number of areas of conservation concern.

550

551 Discussion

552

553 Model Tests

554

555 We lacked the data to systematically test our models independently so our testing is

556 confined to comparisons with other approaches and examining areas of known low

557 density. Boyce and Waller (2003) used RSF models from previous research in the

558 Yellowstone and Swan mountains areas of northern and southern Montana to predict

559 population abundance in the Bitterroot mountains of central Idaho. This area is currently

560 not occupied by grizzly bears so the number was meant to assess recovery potential. They

561 had models for 3 seasons and the lowest predicted abundance was 321 bears while their

562 highest seasonal prediction was 484 bears in spring. Our interior model predicted density

563 to be 30 bears/1000 km2 in the southern zone, 31 in the northern and alternate zones (SD

25 564 was 10.6 for all three predictions) for a total population of 657 bears (approximate 95%

565 CI = 211-1103). Boyce and Waller (2003) may have underestimated populations size

566 because they assumed that the population estimate from the two reference study areas

567 were unbiased estimates of equilibrium density whereas, more recent surveys have

568 suggested that bear numbers have increased in both areas since their research was done

569 (Schwartz et al. 2006, Kendall et al 2009).

570 Mattson and Merrill (2004) provided an independent prediction of grizzly

571 numbers in the Cabinat-Yahk Recovery Area in Montana and Idaho based on study area

572 scale modeling of habitat capability and the depressive effects of human use. This area

573 was one of our model areas and the 2004 population was estimated at 44 bears

574 (Wakkinen and Kasworm 2004) while Mattson and Merrill’s model predicted the area

575 could support 123 bears and our interior model predicted 130 bears. The predicted

576 population sizes may be greater than the 2004 field-based estimate because the

577 population is recovering from very low numbers and human-caused mortality appears to

578 be limiting recovery (Wakkinen and Kasworm 2004).

579

580 Predictions for extirpated zones and depressed populations as model tests

581

582 There were 14 study areas that are not known to support grizzly populations in the

583 interior dataset. The predicted density was zero for 8 of these areas and >4 for the 6 other

584 areas. Two areas had predicted density of >14 bears/ km2. Grizzly bears do not occur in

585 the southern boreal regions of NWT and Nunuvut and northern Alberta and

586 Saskatchewan (McLoughlin 2011). This extirpated area is presumably grizzly bear-free

26 587 naturally because human density is very low. Predicted densities were not always low for

588 the extirpated area. Four unoccupied study areas in this area had low human density and

589 predicted densities were 0, 0, 9, & 21 (SE 10.6 - 10.8). Predicted grizzly density for

590 ecoregions in this area varied from 0 to 38 bears/1000 km2. The boreal portion of this

591 extirpated area was predicted to have low bear density, usually zero. However, the

592 parkland and grassland areas south of the boreal zone were predicted to have zero bears

593 throughout Alberta but very high densities were predicted in the same ecoregions in

594 Saskatchewan. Human and livestock densities were much higher in Alberta, all other

595 input data were similar. These results demonstrate that our interior model does not index

596 the ecological factor or factors limiting distribution in all unoccupied areas well. Some

597 combination of limiting factors excludes grizzlies from central Canada and this result was

598 only correctly predicted by the interior model when human density or tree cover was

599 high.

600 We also used our preferred models to predict equilibrium densities in known

601 extirpated areas and zones with depressed populations which are designated as threatened

602 or endangered in BC and the lower 48 states (Servheen 1989, Hamilton et al 2004, Table

603 7). Areas where grizzly bears are currently extirpated in BC were all predicted to have

604 density <8/ km2 and <30 grizzly bears in the population unit. Most threatened units were

605 also predicted to have small populations but four units were predicted to have >150 bears

606 (South Chilcotin Ranges, Squamish-Lillooet, Toba-Butte and the North Cascades).

607 Recent inventory in the southern portion of the South Chilcotin Ranges unit suggests

608 current populations may be similar to predicted numbers and hence no longer threatened

609 (Apps et al. 2009). The Squamish and Toba populations are recovering from human over-

27 610 exploitation and the populations are likely lower than predicted by the model. The North

611 Cascades area in southwest BC and central Washington currently supports very few bears

612 (Romain-Bondi et al. 2004) and our model suggests the area may be capable of

613 supporting several hundred bears on the Canadian side alone (Table 9). Previous habitat-

614 based modeling suggested the Canadian portion of the North Cascades could support 293

615 bears (North Cascades Grizzly Bear Recovery Team 2001), which is similar to the 284

616 suggested by our model (Table 9).

617 Our interior model predicted grizzly densities between 19 and 35 bears/1000 km2

618 in the six recovery areas in the lower 48 United States. This resulted in population

619 predictions between 64 and 874 per study area which is many more bears than currently

620 occurs in two of these 6 areas (Table 9). As in Canada, again the biggest discrepancy was

621 in the North Cascades. Additionally, the Bitterroot Ranges are currently unoccupied yet

622 the interior model predicted 445 bears in this unit. Grizzly bears were extirpated, or

623 nearly so, by humans in both these areas and recovery is likely limited by demographic

624 failure not habitat characteristics.

625 The choice of prediction units did not alter total population predictions for BC or

626 Yukon greatly. The impacts were greater for BC because density was more variable

627 among units and, smaller units allowed the choice between the interior and coastal model

628 to be applied at local scales, which was most appropriate. There are limits to the size of

629 area to which the model can be applied; ideally prediction units would be similar in size

630 to the study areas. Practically, the user must remember that all independent measures are

631 means and prediction area values must lie within the range of variation of the data in the

632 model. Small study areas are most likely to exceed this range.

28 633 In conclusion, our interior model generated high densities for some low density

634 populations and unoccupied areas. Areas of low forest cover such as grassland, prairie

635 parkland and tundra appeared to be over-predicted and the reason for this is unclear.

636 Also, areas that are unoccupied or greatly reduced in density due to excessive human

637 mortality were over-predicted. This may be explained by the relatively few data for areas

638 where human-caused mortality was limiting. Human-density was the only variable in the

639 model that worked to predict mortality impacts and we expect the precision to do this is

640 poor, especially at high human densities. Both our models are likely to be most inaccurate

641 for populations that are strongly limited by human-caused mortality. The large population

642 predictions for areas like the North Cascades suggests that the reason these areas are

643 currently at low density, or not occupied, is demographic failure or physical isolation, not

644 habitat limitations.

645 646 647 Using the models to manage grizzly bear mortality

648

649 The models we developed, and the population sizes predicted from them, provide

650 information to support the implementation of grizzly bear management policies.

651 Population predictions could be used to calculate mortality limits or, to predict habitat

652 capability. Sustained yield management involves the trade-off between conservation risk

653 and benefits to society. Conservation risk can be minimized by policy, such as reducing

654 the maximum harvest rate, or by investment, such as by increasing inventory effort. In

655 1978 the Government of BC began to move from seasonal hunting restrictions to a quota

656 system to manage grizzly bear hunter kill. This change was effected in order to reduce

29 657 conservation risk and indeed there is evidence some grizzly bear populations increased

658 following to this policy change (Hovey and McLellan 1999, G. Mowat, unpublished data)

659 although other limiting factors may have changed as well. The paradox of these good

660 intentions is that a quota system requires population sizes for every hunted population

661 because annual allowable kill is usually calculated as a portion of the standing

662 population. Implementation of that policy required population estimates for all areas

663 occupied by grizzly bears, even those that had never been surveyed. This policy change

664 precipitated a large increase in inventory investment (Proctor et al. 2010) but population

665 predictions were still required for many parts of the Province. The prediction method we

666 describe here is a reasonable balance between accuracy and investment. Our work used

667 all currently available data, did not require expensive field testing, and provides

668 predictions for all areas of Canada.

669 Our study demonstrates the uncertainty in extrapolating animal densities, even for

670 species for which there is considerable inventory data and a good understanding of the

671 population biology of the species. Many areas were predicted to be unoccupied when

672 clearly this is not the case based on local knowledge or kill data, and vice-versa. This

673 presents 3 problems for wildlife managers; 1) because they will be forced to decide

674 between those conflicting data (i.e., whether to allow hunting), 2) if they allow hunting

675 then they will have to assign a density to the area either subjectively or using some other

676 method, and 3) the credibility of the modeling process will be reduced because it will be

677 clearly evident that the model is ‘wrong’. These nuances and decisions are regularly

678 confronted by wildlife managers and our example highlights the fact that removing

679 subjectivity from the decision making process is impossible, even for very well studied

30 680 species. Our study demonstrates the benefit of local knowledge, even for this highly data

681 driven management system. Fifteen percent of our data were from extirpated areas based

682 on local knowledge. And, local knowledge will be key information in making decisions

683 in the areas described above.

684 The population predictions were higher than current estimates for all Canadian

685 provinces except BC (McLoughlin 2011). Much of the discrepancy can be attributed to

686 the fact the model predicted relatively high densities in northern boreal and tundra areas,

687 including areas that are currently unoccupied. Densities predicted by the interior model in

688 the tundra portions of the north are almost certainly too high in the eastern Arctic. The

689 territorial population predictions could be more realistic if areas that are unoccupied, or

690 recently recruited by bears, are excluded from the prediction. Both our models predict

691 more variable densities than other modeling efforts in BC (Fuhr and Demarchi 1990,

692 Mowat et al. 2004) or Alberta (Popplewell et al. 2003). If this variation is real, then

693 earlier models were over predicting density in some areas. Indeed, 14 management units

694 in BC appear to have annual kill rates higher than that allowed in policy (6%). A further

695 14 units which had predicted densities of zero had >2 reported bear kills during the past 5

696 years. This level of kill is likely not just emigrant bears wandering into the unit and thus

697 suggests a resident population. Many of these units were predicted to have small

698 populations and mortality would need to be managed at larger scales for many of these

699 units, as is currently done in BC. Regardless, our model predictions have direct

700 conservation implications. Two of the latter 14 areas had >20 reported kills over the last 5

701 years demonstrating there is a resident population in the unit. Management responses to

702 this new information must occur on a unit by unit basis and incorporate other

31 703 information such as: local knowledge about distribution and movement among units;

704 major food sources such as salmon runs; hunter success; age and sex ratio of the past kill;

705 trend in kill numbers; and the trend and distribution of problem occurrences. At the

706 opposite scale, the mean kill rate among wildlife units in BC is 3.4% for those units that

707 were predicted to support grizzly bears which should be sustainable in most populations

708 (Schwartz et al. 2003).

709

710 Biological implications of the models

711

712 We demonstrate that grizzly bear density is related to general indices of resources in the

713 environment. Our results suggest that ultimate factors such as vegetation biomass and

714 productivity, vegetation structure, and protein abundance and availability influence

715 grizzly bear density across its North American range. We further show the degree to

716 which density can be reduced by human influences, other than hunting, while the

717 limiting effect of competition on density was equivocal. Other research has demonstrated

718 the link between forage abundance and population growth in black bear (Cote 2005,

719 Reynolds-Hogland et al. 2007) and grizzly bear populations (Hilderbrand et al. 1999,

720 Pease and Mattson 1999, Schwartz et al. 2006). The negative effect of humans on grizzly

721 bear population growth was also empirically demonstrated in the latter two papers.

722 Although the competitive effect of black bears on grizzly bear population density has

723 been suggested (Mattson et al. 2005) and potential individual and population level

724 effects described (Apps et al. 2006, Belant et al 2006), demonstrating this effect on

725 population density or growth remains elusive.

32 726 Our results suggest the plant portion of the bear diet is indexed by precipitation

727 and NDVI. Precipitation likely indexes plant productivity while NDVI is thought to index

728 plant biomass (Bannari et al. 1995). Ruggedness also appears related to density and

729 although it was correlated with precipitation (r = 0.73), the PCA analysis suggested these

730 two variables contrasted in the third eigenvector suggesting some level of independence.

731 Ruggedness may index the increased surface area associated with sloped areas which

732 should provide greater plant biomass and increased variation in plant phenology among

733 microclimatic sites. Density was negatively related to forest cover and positively related

734 to herb-shrub cover presumably because bear plant foods are more abundant in non-

735 forested areas. Habitat selection studies have demonstrated avoidance of forested areas by

736 grizzly bears (Nielsen et al. 2004, Nams et al. 2006). However an alternative, and not

737 exclusive hypothesis using our data, is that fewer trees may benefit grizzly bears by

738 reducing competition with black bears. We could not isolate these two effects in our

739 analysis.

740 Density increased as the amount of salmon in the diet increased in coastal areas.

741 Similar data supporting this observation was presented by Hilderbrand et al. (1999). Our

742 interior data also provides weak support for the generality of this observation because the

743 salmon presence variable appeared in 6 of the 10 top interior models and was always

744 positively related to density, even though there was only a small range of variation in the

745 actual values of this variable.

746 But diet fractions do not necessarily correlate directly with salmon abundance or

747 availability, especially where bears eat mostly salmon. In coastal areas with similar

748 salmon consumption (with one exception), grizzly bear density was almost an order of

33 749 magnitude higher where black bears were absent (Table 4). Many observers have

750 suggested that body size increases with dietary meat (reviewed in Mowat and Heard

751 2006) and while Mowat and Heard (2006) showed that body size was strongly related to

752 the amount of salmon in the diet, it was less correlated with terrestrial meat (see

753 Hilderbrand et al. 1999 for similar results). Terrestrial meat-in-diet had a negative slope

754 for all credible models in which it was included in our interior model results. Regardless

755 of the strength of the relationship between diet and body size, body size may be a better

756 index of food quality or abundance than population density. In Alaska, bears in hunted

757 areas were larger than bears in nearby ecologically similar areas that were not hunted

758 (Miller et al. 2003, see also Zedrosser et al. 2006). We conclude that salmon availability

759 influences grizzly bear abundance but this relationship may be more complex than a

760 simple linear relationship across the range of the species. Our data suggest a negative

761 relationship between density an terrestrial meat availability but this observation may be

762 due to colinearity with other variables.

763 Our results support those of Mattson and Merrill (2002) and many others who

764 report that human density negatively influences grizzly bear density and that grizzly bears

765 cannot exist at a human density >7/km2. Our data generally support this hypothesis; only

766 1 of 101 occupied areas had human density >7/ km2 and this study area straddled a

767 heavily settled valley but did not sample a great deal of the less settled mountains on

768 either side of the valley. Mattson and Merrill (2002) showed that the probability of

769 grizzly bear persistence was inversely related to cattle density; similar to our findings that

770 density was inversely related to livestock numbers.

34 771 Mattson and Merrill (2002) also found that grizzly bear extirpation was more

772 likely where salmon were present in the ecosystem; and that grizzly bear range in 1850

773 was negatively correlated with abundant salmon areas. The authors speculated that

774 competition for salmon with native North Americans excluded grizzly bears from key

775 salmon areas. Our coastal model suggests grizzly bear numbers are low in southern

776 coastal areas due to dense tree cover and lower rainfall (as indexed by ruggedness) than

777 further north on the coast. Historic analyses of diet have shown that grizzly bears on the

778 south Pacific coast, and even well into the interior along the Columbia river, ate largely

779 salmon (Hilderbrand et al. 1996). Reductions in the availability of salmon may have lead

780 to extirpation because these bears were likely too large to switch to vegetative food

781 (Welch et al. 1997, Rode et al. 2001). This observation is supported by our admittedly

782 simple coastal model which suggests that given similar forest cover, coastal bears with

783 less salmon available occur at lower densities. The high forest cover, moderate terrain

784 (and hence plant productivity), and variable salmon availability may have lead to low

785 historic numbers of grizzly bears along the south Pacific coast.

786 Reported mortality explained relatively little of the variation in density when

787 other factors were accounted for. Nor did our data suggest a non-linear relationship

788 between density and kill rate as would be expected based on current theories on

789 population growth (Caughley and Sinclair 1994). This was not too surprising given that

790 most kill rates in our dataset were low. Low kill rates would be unlikely to result in much

791 of a reduction in population size because grizzly bear populations are thought to show

792 high compensatory responses near ecological carrying capacity (Taylor et al. 1994).

793 Hunter kill may also have been compensated for by emigration. Lags in the impact of

35 794 mortality due to immigration would confound comparisons of instantaneous measures of

795 density and kill rate. Similarly, populations that were reduced due to human-caused

796 mortality but have not recovered for various demographic reasons would also confound

797 the instantaneous comparison of density and kill rate, as discussed earlier. Human or

798 livestock density may have accounted for some of the influence mortality had on density

799 but this was likely small because correlations were weak between these variables in both

800 datasets (r < 0.07). Human and livestock density may be correlated with unreported

801 human-caused mortality however.

802

803 Alternative methods for predicting grizzly bear density

804

805 Our approach to predicting density differs from most previous attempts in that previous

806 efforts have usually been based on bear distribution or movement data and applied a use

807 versus availability analysis approach. The models from this approach are often called

808 resource selection functions (Manly et al. 1993). This approach has been applied several

809 times using radiotelemetry data from individual grizzly bears (Boyce and Waller 2003) or

810 detection data (Popplewell et al. 2003, Apps et al. 2004, Mattson and Merrill 2004, Nams

811 et al. 2006). Although these studies used rigorous analysis techniques often accounting

812 for multiple scales and contrasting the importance of many different variables the model

813 structures appear unique to the landscape of origin and the authors were careful not to

814 extrapolate the models much beyond the original study area. The above studies

815 extrapolated density to areas roughly the size of one grizzly bear management unit in BC.

816 Many models of this scale would be needed to predict density for BC. All final models

36 817 included abstract landcover measures such as elevation or greeness that likely had

818 complex relationships with other variables and with density, makings these types of

819 models difficult to interpret in a functional sense.

820 In contrast we followed a more functional approach using measures of landscape

821 scale density as the dependent variable rather the presence or abundance of individuals at

822 a site. Density combines all the factors which influence population dynamics in a single

823 measure however it is variable and therefore population growth is a better metric when

824 testing for functional links between populations and environments (Sibly et al. 2003).

825 Density should be independent of factors such as individual behaviour and regional

826 specific life history strategies which influence the outcome of finer scale analyses and,

827 our analysis was unaffected by the relative abundance or availability of different

828 resources within a study area that can limit predictions of an RSF model (Beyer et al.

829 2010). Our independent variables reflect measures of the total abundance of resources in

830 a landscape roughly the size of a grizzly bear population. We consider the scale of our

831 approach more appropriate than behavior-based models because our dependent variable

832 was measured at a similar scale to which we hope to make predictions (Gaillard et al.

833 2010) and, our model incorporated data across the entire area we intended to make

834 predictions.

835

836 Model weaknesses and improvements

837

838 Perhaps the largest weakness of our coastal model is the salmon diet data. This was a key

839 variable but it was extracted from a diet surface constructed from 81 diet measures across

37 840 northwest North America, and many fewer along the coast. We augmented extracted data

841 with local diet data where it was available but this was sporadic. Perhaps the most

842 important influence of the diet data on outcome was in deciding whether prediction areas

843 were coastal or interior. The main information we used for this was the salmon diet data

844 and when this was not available we used local knowledge. Finer scale diet data should

845 improve results and their application.

846 The variables driving our model were largely static. A dynamic model would

847 require information on vegetation cover that is more detailed than what we found. A truly

848 dynamic model would require annual databases for each variable such that every variable

849 was an appropriate multi-annual mean pre-dating each survey. This would require vastly

850 more digital data and the elimination of much early bear density data because most

851 databases begin in the 1980's or later. An effort of this level should also attempt to

852 document human use of the landscape via a dynamic measure of road abundance and

853 distribution; this too would require annual road layers for the past and the future. For its

854 current application it is crucial that local practitioners of the model understand its

855 limitations so they understand where model predictions are least certain.

856 We were unable to index the availability of terrestrial meat and the correlation

857 between precipitation and terrestrial meat in the diet (r = -0.63) undermined our ability to

858 evaluate this factor using diet proportions. And the correlation of these two variables with

859 tree cover (rprecip = 0.33, rmeat = -0.35) negated a controlled evaluation of the importance

860 of competition with black bears (ignoring the question as to whether the tree cover

861 variable was a reasonable index of competition). The availability of salmon was not

862 correlated to other independent variables but salmon was not a major contributor to diet

38 863 in the interior. In coastal settings the salmon diet proportions were quite variable and did

864 appear sensitive to changes in salmon availability (Boulanger et al. 2004, Mowat and

865 Heard 2006).

866 We were unable, at the scale we worked at, to index key vegetative foods like

867 berries. Huckleberries are a key food in most wetter environs and this resource is linked

868 to natural burns (McLellan 2011). We could not find data to index huckleberry

869 abundance directly, nor could we find digital data that documented burn history at the

870 scale of the continent.

871 Density estimates may vary with the size of the area over which they were

872 measured (Mayor and Shaefer 2005, Lewontin and Levins 1989) but, we believe we

873 avoided this issue because we selected estimates based over large areas (Tables 3-4).

874 There were weak negative relationships between density and area (r < -0.59) in all 3

875 datasets but we believe this is explained by the need to have larger study areas in lower

876 density populations in order to meet sample size requirements. The data we used were not

877 subject to publication bias, a potential meta-analysis problem, because we used both

878 published and unpublished data.

879

880 Conclusions

881

882 Modeling density to determine general landscape scale limiting factors has rarely been

883 done and only twice, to our knowledge, have the models been used for predictions (Kays

884 et al. 2008, Beck and Sieber 2010). The availability of digital databases which index

885 many facets of an ecosystem’s productivity and, the increasing resolution of these data

39 886 (Kozak et al. 2008) suggest it may be possible to predict density using a similar approach

887 to that followed here for other species for which a sample of populations have been

888 surveyed. Other meta-analyses are possible such as the comparison of reproduction or

889 productivity across ecosystem productivity or survival rates across indices of predator

890 risk. The correction of density by removing barren area should facilitate general

891 comparisons among study areas and ecosystems. Finer scale habitat classifications over

892 large areas will improve our ability to test hypotheses and make predictions and, datasets

893 that are stored annually would allow for dynamic ecosystem modeling.

894 We chose a broad-scale modeling approach that was not influenced by individual

895 animal behavior or the habitat composition of the study area (Beyer et al. 2010). We did

896 this in the hopes the fitted models would have broad application. Our area of

897 management concern (BC) is nearly 1 million km2 and grizzly bears occupy a much

898 larger and more diverse area than that in Canada.

899 Model predictions varied as expected considering the precision of the models. The

900 static nature of the variables we used and the large natural variation in density means that

901 model fit, and hence precision, will always be low for this system. Large increases in

902 precision and hence prediction accuracy will only likely come with increased functional

903 understanding.

904

905

906 Acknowledgements This work was funded by BC Ministry of Environment, FRBC and

907 the British Columbia Grizzly Conservation Strategy. We thank the many scientists who

908 shared their data and advice especially A. D. Hamilton, S. D. Miller, R. Sellers, E.

40 909 Becker, R. Flynn, L. Van Daele, G. MacHutchon, C. Apps, C. Johnson, and C. Carroll. T.

910 Gaines and D. Pritchard skillfully did the GIS analysis. B McLellan, reviewed drafts of

911 the manuscript. We dedicate this paper to Tom Gaines who worked tirelessly on earlier

912 versions.

913

914 References

915

916 Apps CD, McLellan BN, Woods JG (2006) Landscape partitioning and spatial inferences 917 of competition between black and grizzly bears. Ecography 29:561–572 918 Apps CD, McLellan BN, Woods JG, Proctor MF (2004) Estimating grizzly bear 919 distribution and abundance relative to habitat and human influence. Journal of 920 Wildlife Management 68:138-152 921 Apps C, Paetkau D, Rochetta S, McLellan B, Hamilton A, Bateman B (2009) Grizzly 922 bear population abundance, distribution and connectivity across British Columbia’s 923 southern Coast Ranges. Aspen Wildlife Research and Ministry of Environment,, 924 Victoria, BC, Canada 925 Arnold TW (2010) Uninformative parameters and model selection using Akaike’s 926 Information Criterion. The Journal of Wildlife Management 74:1175–1178 927 Bannari A, Morin D, Bonn F, Huete A (1995) A review of vegetation indices. Remote 928 Sensing Reviews 13:95–120 929 Beck J, Sieber A (2010) Is the spatial distribution of mankind’s most basic economic 930 traits determined by climate and soil alone? PloS one 5:e10416 931 Belant JL, Kielland K, Follmann EH, Adams LG (2006) Interspecific resource 932 partitioning in sympatric ursids. Ecological Applications 16:2333–2343 933 Ben-David M, Titus K, Beier LR (2004) Consumption of salmon by Alaskan brown 934 bears: a trade-off between nutritional requirements and the risk of infanticide? 935 Oecologia 138:465-474 936 Beyer HL, Haydon DT, Morales JM, Frair JL, Hebblewhite M, Mitchell M, 937 Matthiopoulos J (2010) The interpretation of habitat preference metrics under use– 938 availability designs. Philosophical Transactions of the Royal Society B: Biological 939 Sciences 365:2245–2254 940 Boulanger J, Himmer S, Swan C (2004) Monitoring of grizzly bear population trends and 941 demography using DNA mark–recapture methods in the Owikeno Lake area of 942 British Columbia. Canadian Journal of Zoology 82:1267-1277 943 Boyce MS, McDonald LL (1999) Relating populations to habitats using resource 944 selection functions. Trends in Ecology & Evolution 14:268–272 945 Boyce MS, Waller JS (2003) Grizzly bears for the Bitterroot: predicting potential 946 abundance and distribution. Wildlife Society Bulletin 31:670-683

41 947 Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a 948 practical information-theoretic approach. Springer Verlag 949 Carbone C, Gittleman JL (2002) A common rule for the scaling of carnivore density. 950 Science 295:2273 951 Carbone C, Pettorelli N (2009) Testing relationships between energy and vertebrate 952 abundance. International Journal of Ecology 2009 953 Caughley G, Sinclair ARE (1994) Wildlife ecology and management. Wiley-Blackwell 954 Ciarniello LM, Boyce MS, Seip DR, Heard DC (2007) Grizzly bear habitat selection is 955 scale dependent. Ecological Applications 17:1424-1440 956 Cook RD, Weisberg S (1982) Residuals and influence in regression. Chapman and Hall 957 New York 958 Côté SD (2005) Extirpation of a Large Black Bear Population by Introduced White- 959 Tailed Deer. Conservation Biology 19:1668–1671 960 Creel S, Spong G, Creel N (2001) Interspecific competition and the population biology of 961 extinction-prone carnivores. In: CONSERVATION BIOLOGY SERIES- 962 CAMBRIDGE-.pp 35–60 963 Crête M (1999) The distribution of deer biomass in North America supports the 964 hypothesis of exploitation ecosystems. Ecology Letters 2:223–227 965 Currie DJ (1991) Energy and large-scale patterns of animal-and plant-species richness. 966 American Naturalist:27–49 967 Dahle B, Zedrosser A, Swenson J (2006) Correlates with body size and mass in yearling 968 brown bears (Ursus arctos). Journal of Zoology 269:273–283 969 Daly C, R. P. Neilson, D. L. Philips (1994) A statistical model for mapping 970 climatological precipitation over mountainous terrain. Journal of Applied 971 Meteorology 33:193-200 972 Ferguson SH, McLoughlin PD (2000) Effect of energy availability, seasonality, and 973 geographic range on brown bear life history. Ecography 23:193-200 974 Freckleton R, Gill J, Noble D, Watkinson A (2005) Large-scale population dynamics, 975 abundance–occupancy relationships and the scaling from local to regional population 976 size. Journal of Animal Ecology 74:353–364 977 Fuhr B, Demarchi DA (1990) A methodology for grizzly bear habitat assessment in 978 British Columbia. Ministry of Environment 979 Gaillard JM, Hebblewhite M, Loison A, Fuller M, Powell R, Basille M, Van Moorter B 980 (2010) Habitat–performance relationships: finding the right metric at a given spatial 981 scale. Philosophical Transactions of the Royal Society B: Biological Sciences 982 365:2255–2265 983 Gunther KA, Biel MJ, Anderson N, Waits L (2002) Probable grizzly bear predation on an 984 American black bear in Yellowstone National Park. Ursus 13:372–374 985 Guthery FS (2008) Statistical ritual versus knowledge accrual in wildlife science. The 986 Journal of Wildlife Management 72:1872–1875 987 Hamilton A (2008) Grizzly bear population estimate for British Columbia. British 988 Columbia Ministry of Environment, Victoria, BC 989 Hamilton A, Heard D, Austin M (2004) British Columbia grizzly bear (Ursus arctos) 990 population estimate 2004. British Columbia Ministry of Water, Land, and Air 991 Protection, Biodiversity Branch Victoria, BC

42 992 Hansen M, DeFries R, Townshend J, Carroll M, Dimiceli C, Sohlberg R (2003) Global 993 percent tree cover at a spatial resolution of 500 meters: First results of the MODIS 994 vegetation continuous fields algorithm. Earth Interactions 7:1–15 995 Herrero SM (1972) Aspects of evolution and adaptation in American black bears (Ursus 996 americanus Pallus) and brown and grizzly bears (U. arctos Linne) of North America. 997 International Conference on Bear Research and Management 2:221-231 998 Hilderbrand GV, Farley SD, Robbins CT, Hanley TA, Titus K, Servheen C (1996) Use of 999 stable isotopes to determine diets of living and extinct bears. Canadian Journal of 1000 Zoology 74:2080-2088 1001 Hilderbrand G, Schwartz C, Robbins C, Jacoby M, Hanley T, Arthur S, Servheen C 1002 (1999) The importance of meat, particularly salmon, to body size, population 1003 productivity, and conservation of North American brown bears. Canadian Journal of 1004 Zoology 77:132–138 1005 Hovey FW, McLellan BN (1996) Estimating population growth of grizzly bears from the 1006 Flathead River drainage using computer simulations of reproductive and survival 1007 rates. Canadian Journal of Zoology 74:1409-1416 1008 Jacoby ME, Hilderbrand GV, Servheen C, Schwartz CC, Arthur SM, Hanley TA, 1009 Robbins CT, Michener R (1999) Trophic relations of brown and black bears in 1010 several western North American ecosystems. The Journal of wildlife 1011 management:921–929 1012 Johnson CJ, Boyce MS, Mulders R, Gunn A, Gau RJ, Cluff HD, Case RL (2004) 1013 Quantifying patch distribution at multiple spatial scales: applications to wildlife- 1014 habitat models. Landscape Ecology:869-882 1015 Kays RW, Gompper ME, Ray JC (2008) Landscape ecology of eastern coyotes based on 1016 large-scale estimates of abundance. Ecological Applications 18:1014–1027 1017 Kendall KC, Stetz JB, Boulanger J, Macleod AC, Paetkau D, White GC (2009) 1018 Demography and genetic structure of a recovering grizzly bear population. The 1019 Journal of Wildlife Management 73:3–16 1020 Kozak KH, Graham CH, Wiens JJ (2008) Integrating GIS-based environmental data into 1021 evolutionary biology. Trends in Ecology & Evolution 23:141–148 1022 Lewontin R, Levins R (1989) On the characterization of density and resource availability. 1023 American Naturalist:513–524 1024 Linnell JDC, Strand O (2000) Interference interactions, co-existence and conservation of 1025 mammalian carnivores. Diversity and Distributions 6:169–176 1026 Mace RD, Waller JS, Manley TL, Ake K, Wittinger WT (1999) Landscape evaluation of 1027 grizzly bear habitat in western Montana. Conservation Biology 13:367-377 1028 Manly BF, McDonald L, Thomas DL, McDonald TL, Erickson WP (2010)Resource 1029 Selection by Animals: Statistical Design and Analysis for Field Studies, 2nd edn. 1030 Springer 1031 Mattson DJ, Merrill T (2002) Extirpations of grizzly bears in the contiguous United 1032 States, 1850-2000. Conservation Biology 16:1123-1136 1033 Mattson DJ, Merrill T (2004) A model-based appraisal of habitat conditions for grizzly 1034 bears in the Cabinet-Yaak region of Montana and Idaho. Ursus 15:78-91 1035 Mattson DJ, Blanchard BM, Knight RR (1991) Food habits of Yellowstone grizzly bears, 1036 1977-1987. Canadian Journal of Zoology 69:1619-1629

43 1037 Mattson DJ, Herrero SM, Merrill T (2005) Are black bears a factor in the restoration of 1038 North American grizzly bear populations? Ursus 16:11-30 1039 May RM (1994) The effects of spatial scale on ecological questions and answers. In: 1040 Large-scale ecology and conservation biology. Blackwell Scientific Publications, pp 1041 1–17 1042 Mayor SJ, Schaefer JA (2005) The many faces of population density. Oecologia 1043 145:275–280 1044 McLellan BN (2005) Sexually selected infanticide in grizzly bears: the effects of hunting 1045 on cub survival. Ursus 16:141-156 1046 McLellan BN (2011) Implications of a high-energy and low-protein diet on the body 1047 composition, fitness, and competitive abilities of black (Ursus americanus) and 1048 grizzly (Ursus arctos) bears. Canadian Journal of Zoology 89:546-558 1049 McLellan BN, Hovey FW (1995) The diet of grizzly bears in the Flathead River drainage 1050 of southeastern British Columbia. Canadian Journal of Zoology 73:704-712 1051 McLellan BN, Hovey FW (2001) Habitats selected by grizzly bears in a multiple use 1052 landscape. Journal of Wildlife Management 65:92-99 1053 McLellan BN, Hovey FW, Mace RD, Woods JG, Carney DW, Gibeau ML, Wakkinen 1054 WL, Kasworm WF (1999) Rates and causes of grizzly bear mortality in the interior 1055 mountains of British Columbia, Alberta, Montana, Washington, and Idaho. Journal of 1056 Wildlife Management 63:911-920 1057 McLoughlin PD (In Press) COSEWIC Status Report on Grizzly Bear Ursus arctos in 1058 Canada. 1059 McLoughlin PD, Case RL, Gau RJ, Cluff HD, Mulders R, Messier F (2002) Hierarchical 1060 habitat selection by barren-ground grizzly bears in the central Canadian Arctic. 1061 Oecologia 132:102-108 1062 Miller SD, Sellers RA, Keay JA (2003) Effects of hunting on brown bear cub survival 1063 and litter size in Alaska. Ursus 14:130-152 1064 Miller SD, White GC, Sellers RA, Reynolds HV, Schoen JW, Titus K, Barnes Jr VG, 1065 Smith RB, Nelson RR, Ballard WB, others (1997) Brown and black bear density 1066 estimation in Alaska using radiotelemetry and replicated mark-resight techniques. 1067 Wildlife Monographs:3–55 1068 Mowat G, Heard DC (2006) Major components of grizzly bear diet across North 1069 America. Canadian Journal of Zoology 84:473-489 1070 Mowat G, Heard D, Gaines T (2004) Predicting Grizzly Bear densities in BC using a 1071 multiple regression model. BC Ministry of Water, Land and Air Protection Prince 1072 George, BC 28pp 1073 Mowat G, Heard DC, Seip DR, Poole KG, Stenhouse G, Paetkau DW (2005) Grizzly 1074 Ursus arctos and black bear U. americanus densities in the interior mountains of 1075 North America. Wildlife Biology 11:31–48 1076 Mysterud, Ims (1999) Relating populations to habitats. Trends Ecol Evol (Amst) 14:489- 1077 490 1078 Nagy JA, Gunson JR (1990) Management plan for grizzly bears in Alberta. Alberta Fish 1079 and Wildlife Division, Edmonton, Alberta 1080 Nams VO, Mowat G, Panian MA (2006) Determining the spatial scale for conservation 1081 purposes – an example with grizzly bears. Biological Conservation 128:109-119

44 1082 Nielsen SE, Boyce MS, Stenhouse GB (2004) Grizzly bears and forestry I. Selection of 1083 clearcuts by grizzly bears in west-central Alberta, Canada. Forest Ecology and 1084 Management 199:51–65 1085 Nielsen SE, Johnson CJ, Heard DC, Boyce MS (2005) Can models of presence-absence 1086 be used to scale abundance? two case studies considering extremes in life history. 1087 Ecography 28:197–208 1088 Pease C, Mattson DJ (1999) Demography of the Yellowstone grizzly bear. Ecology 1089 80:957-975 1090 Popplewell C, Franklin SE, Stenhouse GB, Hall-Beyer M (2003) Using landscape 1091 structure to classify grizzly bear density in Alberta Yellowhead Ecosystem bear 1092 management units. Ursus 14:27-34 1093 Prather JW, Dodd NL, Dickson BG, Hampton HM, Xu Y, Aumack EN, Sisk TD (2006) 1094 Landscape models to predict the influence of forest structure on tassel-eared squirrel 1095 populations. Journal of Wildlife Management 70:723–731 1096 Pritchard GT, Robbins CT (1990) Digestive and metabolic efficiencies of grizzly and 1097 black bears. Canadian Journal of Zoology 68:1645–1651 1098 Proctor M, Boulanger J, Nielsen S, Servheen C, Kasworm W, Radandt T, Paetkau D 1099 (2007) Abundance and density of Central Purcell, South Purcell, Yahk, and South 1100 Selkirk Grizzly Bear Population Units in southeast British Columbia. BC Ministry of 1101 Environment, Nelson and Victoria, BC, Canada 1102 Proctor M, McLellan B, Boulanger J, Apps C, Stenhouse G, Paetkau D, Mowat G (2010) 1103 Ecological investigations of grizzly bears in Canada using DNA from hair, 1995– 1104 2005: a review of methods and progress. Ursus 21:169-188 1105 Rausch RL (1963) Geographic variation in size in North American brown bears, Ursus 1106 arctos L., as indicated by condylobasal length. Canadian Journal of Zoology 41:33–45 1107 Reynolds-Hogland MJ, Pacifici LB, Mitchell MS (2007) Linking resources with 1108 demography to understand resource limitation for bears. Journal of Applied Ecology 1109 44:1166–1175 1110 Riley SJ, DeGloria SD, Elliot R (1999) A terrain ruggedness index that quantifies 1111 topographic heterogeneity. Intermountain Journal of sciences 5:23–27 1112 Ritchie EG, Martin JK, Krockenberger AK, Garnett S, Johnson CN (2008) Large- 1113 herbivore distribution and abundance: intra-and interspecific niche variation in the 1114 tropics. Ecological monographs 78:105–122 1115 Robbins CT, Schwartz CC, Felicetti LA (2004) Nutritional ecology of ursids: a review of 1116 newer methods and management implications. Ursus 15:161–171 1117 Rode KD, Robbins CT (2000) Why bears consume mixed diets during fruit abundance. 1118 Canadian Journal of Zoology 78:1640-1645 1119 Rode KD, Robbins CT, Shipley LA (2001) Constraints on herbivory by grizzly bears. 1120 Oecologia 128:62–71 1121 Romain-Bondi KA, Wielgus RB, Waits L, Kasworm WF, Austin MA, Wakkinen W 1122 (2004) Density and population size estimates for North Cascade grizzly bears using 1123 DNA hair-sampling techniques. Biological Conservation 117:417-428 1124 Rosenzweig ML (1968) Net primary productivity of terrestrial communities: prediction 1125 from climatological data. American Naturalist:67–74 1126 Saether BE, Engen S, GrU00F8tan V, Fiedler W, Matthysen E, Visser ME, Wright J, 1127 MU00F8ller AP, Adriaensen F, Van Balen H, others (2007) The extended Moran

45 1128 effect and large-scale synchronous fluctuations in the size of great tit and blue tit 1129 populations. Journal of Animal Ecology 76:315–325 1130 Sarewitz D (2004) How science makes environmental controversies worse. 1131 Environmental Science & Policy 7:385–403 1132 Schwartz CC, Haroldson MA, White GC, Harris RB, Cherry S, Keating KA, Moody D, 1133 Servheen C (2006) Temporal, spatial, and environmental influences on the 1134 demographics of grizzly bears in the Greater Yellowstone Ecosystem. Wildlife 1135 Monographs 161:1-68 1136 Schwartz C, Sterling Miller, Mark A. Haroldson (2003) Grizzly bear. In: Wild mammals 1137 of North America: biology, management, and conservation, Second Edition. JHU 1138 Press, Baltimore, Maryland, USA, pp 556-586 1139 Servheen C (1990) The status and conservation of the bears of the world. International 1140 Conference on Bear Research and Management Monograph Series No. 2 1141 Sibly RM, Hone J, Clutton-Brock TH (2003) Wildlife Population Growth Rates. 1142 Cambridge University Press 1143 Sinclair ARE, C. J. Krebs (2003) Complex numerical responses to top-down and bottom- 1144 up processes in vertebrate populations. In: Wildlife Population Growth Rates. 1145 Cambridge Univ Press, pp 127-147 1146 Swenson JE, Sandegren F, Soderberg A, Bjarvall A, Franzen R, Wabakken P (1997) 1147 Infanticide caused by hunting of male bears. Nature 386:450-451 1148 Tabachnick B, Fidell L (1996)Using multivariate statistics. HarperCollins College 1149 Publishers, 3rd edn. Harper Collins Publishers, New York 1150 Taylor MK, Garshelis DL, McLellan BN, Derocher AE (1994) Density-dependent 1151 population regulation of black, brown, and polar bears. International Conference on 1152 Bear Research and Management 1153 Tobin J (1958) Estimation of relationships for limited dependent variables. 1154 Econometrica: Journal of the Econometric Society:24–36 1155 Wakkinen WL, Kasworm WF (2004) Demographics and population trends of grizzly 1156 bears in the Cabinet-Yaak and Selkirk Ecosystems of British Columbia, Idaho, 1157 Montana, and Washington. Ursus 15:65-75 1158 Welch CA, Keay J, Kendall KC, Robbins CT (1997) Constraints on frugivory by bears. 1159 Ecology 78:1105–1119 1160 Wilkinson LC, Stenhouse GB, Team AGBR (2008) Alberta grizzly bear recovery plan 1161 2008-2013. Alberta Sustainable Resource Development, Edmonton, Alberta 1162 Woods JG, Paetkau D, Lewis D, McLellan BN, Proctor M, Strobeck C (1999) Genetic 1163 tagging of free-ranging black and brown bears. Wildlife Society Bulletin:616–627 1164 Zedrosser A, Dahle B, Swenson JE (2006) Population density and food conditions 1165 determine adult female body size in brown bears. Journal of Mammalogy 87:510–518 1166

Table 1 Factors limiting grizzly bear density in interior ecosystems in North America and the variables we derived to index these factors. We digitized the study area boundary for each study area and calculated the average for each index using a GIS.

Plant productivity Vegetation Diet Competition Human human-caused type1 with black disturbance mortality bears annual tree10 salmon tree10 tree10 human density precipitation presence annual tree25 %salmon tree25 tree25 human+livestock temperature in diet density annualized NDVI herb50 %terrestrial presence livestock density human+livestock meat in density within diet 10km2 evapotranspiration herb75 human density human+livestock density within 50km2 ruggedness herb100 human+livestock mean recorded density human-caused mortality in past 10 years human+livestock density within 10km2 human+livestock density within 50km2

1This is the sum of all pixels with >the stated percentage of describe cover. For example herb50 = the proportion of the study with pixels rated as >50% herb/shrub.

2This is the mean human and livestock density (summed) for the area within 10 or 50 km of the study area boundary.

Table 2 Variables we extracted from digital databases for this analysis including a description of the original data used to build the GIS coverage and the spatial resolution

(all data were shifted to raster format).

Variable Underlying Resolution Source for data Reference

data (km) mean annual precipitation ground 4 PRISM Daly et al. 1994

station www.ocs.orst.edu/pri

weather data sm actual evapotranspiration ground 55 UNEP/DEWA/GRID-

(AET) station Geneva

weather data www.grid.unep.ch normalized differential AVHRR 1 EROS Data Center MTPE EOS Data vegetation index (NDVI) satellites Edcwww.cr.usgs.gov Products Handbook

1992-93 /landdaac/glcc Volume 1 ruggedness 1 EROS Data Center Riley et al. 1999

ftp://edcftp.cr.usgs.g

ov/ human and livestock US and Variable Statistics Canada numbers Canada but 50-100 US Censu Bureau

census data km outside www.statcan.ca

of cities www.census.gov/ge

o/www/tiger

Landcover: forest, water & MODIS 0.5 Global Landcover Hansen et al. 2003 barren (Vegetation satellites Facility, University of

Continuous Fields, VCF) Marland

Glcf.umiacs.umd.ed

u/data/modis/vcf

Table 3 Descriptive statistics for data used to build models to predict grizzly bear density in interior North America.

Variable Occupied areas Unoccupied areas

F SD Min Max n F SD Min Max n

Population size 92 96 15 765 76 0 0 0 0 14

Study area size 5110 4297 789 22875 76 6358 1629 3913 8959 14 (km2) barren (% of study 6.7 8.6 0.0 34.6 76 1.7 2.1 0.1 6.7 14 area) Density (barren 23.0 15.1 2.5 64.6 76 0.0 0.0 0.0 0.0 14 area removed) CI relative (% of 1.0 0.5 0.1 1.9 76 0.8 0.3 0.5 1.0 14 49 density) human-caused 3.7 3.9 0.0 20.1 76 01 0.0 0 0 14 mortality (%) Annual 84 40 16 199 76 52 7 43 68 14 Precipitation (cm) NDVI 116 15 77 137 76 129 5 115 136 14

AET 296 104 97 443 76 342 60 228 440 14

Average annual -4.0 5.0 -17 3.7 76 0.9 3.1 -5.0 4.7 14 temperature ruggedness 3.9 1.4 1.0 6.0 76 2.2 1.3 1.0 4.2 14

Trees (>%25 per 48.1 34.7 0.0 99.5 76 69.7 30.0 8.4 99.7 14 pixel) Herb-shrub (>%50 57.6 28.1 3.1 99.1 76 50.6 28.4 6.8 94.0 14

per pixel) Salmon (% of diet) 1.0% 3.0% 0.0% 14.0% 76 1.5% 3.6% 0.0% 10.2% 14

Kokanee (% of 0.7% 3.5% 0.0% 26.0% 76 0.6% 1.5% 0.0% 5.1% 14 diet) Meat (% of diet) 25.1% 15.5% 0.0% 58.2% 76 29.3% 13.9% 12.5% 48.1% 14

Human density 0.9 1.7 0.0 8.4 76 4.3 5.6 0.0 21.4 14 (humans/km2) Livestock density 1.7 5.5 0.0 39.4 76 11.5 16.9 0.0 53.2 14 (animals/km2)

1The kill rate in unoccupied areas was zero but we used the mean rate of 3.7 from occupied during analysis so that these areas did not

50 bias the distribution for this variable.

Table 4 Descriptive statistics for data used to predict grizzly bear density in coastal North

America. We separated areas where black bears were absent because there were large difference in density between these areas.

Variable Black bears present Black bears absent

F SD Min Max n F SD Min Max n

Population 81 88.6 0 352 17 619 450 102 1548 11 size Study area 3743 2858 431 9854 17 2829 2463 228 9163 11 size (km2) barren (% of 16.0 10.4 0.7 43.4 17 9.6 10.0 1.1 35.7 11 study area) Density 31.1 25.9 0 86.6 17 332 215 37 856 11 (barren area removed) CI relative (% 1.1 0.5 0.1 2.0 17 0.6 0.5 0.2 1.5 11 of density) human- 2.3 2.6 0.0 10 17 3.8 2.7 0 7.3 11 caused mortality (%) Annual 275 111 115 473 17 160 57 101 255 11 Precipitation (cm) NDVI 109 16 75 130 17 104 12 79 115 11

AET 370 58 263 474 17 321 30 239 351 11

Average 1.6 2.5 -2.6 6.7 17 1.0 2.0 - 2.6 11 annual 4.5 temperature Ruggedness 5.5 0.6 4.2 6.2 17 4.3 1.0 2.6 5.4 11

Trees (>%25 54.4 23.1 17.8 96.8 17 40 25 1 79 11 per pixel) Herb-shrub 44.9 16.6 8.8 71.6 17 68 21 36 98 11 (>%50 per pixel) Salmon (% of 41 22 0 78 17 59 15 28 82 11 diet) Kokanee (% 0.4 0.8 0.0 3 17 0 0 0 0 11 of diet) Meat (% of 2.3 3.9 0.0 13.4 17 3 11 0 36 11 diet)

Human 4.3 13.4 0.0 55.0 17 0.6 1.6 0.0 5.5 11 density (humans/km2) Livestock 1.1 4.3 0.0 17.9 17 0.01 0.00 0.0 0.02 11 density (animals/km2)

Table 5 Model selection results for study areas in interior North America (n = 90) relating grizzly density to variables that were hypothesized to be functionally related to density. The top ranked model was excluded from this list because it contained 8 uninformative variables.

Model number Model description AICc ǻAICc K AICc weight 2 Prcp_NDVI_AET_H50_LHum_Live_Rug 978.9 0.0 9 0.17 3 Prcp_NDVI_AET_H50_T25_LHum_Live_R 979.0 0.1 10 0.16 ug 4 Prcp_AET_H50_Meat_LHum_Live_Rug 979.1 0.3 9 0.15 5 Prcp_NDVI_AET_H50_T25_Meat_LHum_ 980.1 1.2 10 0.09 Live 6 Prcp_NDVI_AET_H50_T25_SP_Meat_LHu 980.2 1.4 12 0.08 m_Live_Rug 7 Prcp_NDVI_AET_H50_SP_Meat_LHum_Li 980.5 1.6 11 0.07 ve_Rug 8 Prcp_AET_H50_SP_Meat_LHum_Live_Ru 981.2 2.3 10 0.05 g 9 Prcp_AET_H50_T25_Meat_LHum_Live_R 981.5 2.6 10 0.05 ug 10 Prcp_NDVI_AET_H50_T25_SP_Meat_LHu 982.6 3.7 11 0.03 m_Live 11 Prcp_AET_H50_T25_SP_Meat_LHum_Live 982.8 3.9 12 0.02 _Harv_Rug

Table 6 Model selection results for study areas where grizzly and black bears were sympatric in coastal North America (n = 17) relating grizzly density to variables that were hypothesized to be functionally related to density. The two top ranked models were excluded from this list because they contained four and three uninformative variables.

Model number Model description AICc ǻAICc K AICc weight 3 T25_salmon_Rug 157.7 0.0 5 0.17 4 Prcp_NDVI_AET_salmon_Hu 157.8 0.1 8 0.16 m_Rug 5 Prcp_NDVI_AET_salmon_LH 158.0 0.3 8 0.15 um_Rug 6 Prcp_NDVI_Temp_H50_salmo 158.8 1.2 9 0.10 n_Hum_Rug 7 Prcp_NDVI_Temp_H50_salmo 159.9 2.2 9 0.06 n_LHum_Rug 8 Prcp_NDVI_Temp_H50_T25_s 160.2 2.6 9 0.05 almon_Rug 9 Prcp_NDVI_Temp_T25_salmo 160.7 3.1 8 0.04 n_Hum 10 Prcp_NDVI_H50_T25_salmon 160.9 3.3 9 0.03 _LHum_Rug 11 Prcp_NDVI_H50_T25_salmon 161.2 3.5 9 0.03 _Hum_Rug 12 NDVI_T25_salmon_Rug 161.2 3.5 6 0.03

Table 7 SM A comparison of the absolute model error and the mean error after removing a single observation iteratively using a Jackknife procedure for the interior data. The difference between the model error and the Jackknife mean error tests for case sensitivity and helps identify over-fit models which may not generalize well. We chose model 2 as our top model in part because it had the lowest difference between the model error and the mean Jackknife error.

Model Model description Absolute Mean AICc model Increase in rank model error Jackknife weight prediction model error error1

1 Prcp_NDVI_AET_ 740 805 0.165 9% H50_LHum_Live_R ug

2 Prcp_NDVI_AET_ 726 806 0.157 11% H50_T25_LHum_Li ve_Rug

3 Prcp_AET_H50_Me 741 804 0.145 9% at_LHum_Live_Rug

4 Prcp_NDVI_AET_ 707 789 0.09 12% H50_T25_Meat_LH um_Live

5 Prcp_NDVI_AET_ 711 808 0.084 14% H50_T25_SP_Meat _LHum_Live_Rug

6 Prcp_NDVI_AET_ 731 815 0.074 11% H50_SP_Meat_LHu m_Live_Rug

7 Prcp_AET_H50_SP 736 808 0.052 10% _Meat_LHum_Live _Rug

8 Prcp_AET_H50_T2 737 815 0.045 11% 5_Meat_LHum_Liv e_Rug

9 Prcp_NDVI_AET_ 705 797 0.026 13% H50_T25_SP_Meat _LHum_Live 10 Prcp_AET_H50_T2 714 808 0.023 13% 5_SP_Meat_LHum_ Live_Harv_Rug

1 This is the increase in prediction error of Jackknife model vs the full model.

Table 8 SM A comparison of the absolute model error and the mean error after removing a single observation iteratively using a Jackknife procedure for the coastal data. The difference between the model error and the Jackknife mean error tests for case sensitivity and helps identify over-fit models which may not generalize well. We chose model 3 as our top model in part because it had the lowest difference between the model error and the mean Jackknife error.

Model Model description Absolute Mean AICc model Increase in rank model error Jackknife weight prediction model error error1

1 Prcp_NDVI_AET_ 52 110 0.286 112% H50_T25_salmon_R ug

2 Prcp_NDVI_Temp_ 68 141 0.157 107% salmon_Hum_Rug

3 T25_salmon_Rug 111 144 0.097 30%

4 Prcp_NDVI_AET_s 72 133 0.09 85% almon_Hum_Rug

5 Prcp_NDVI_AET_s 72 142 0.083 97% almon_LHum_Rug

6 Prcp_NDVI_Temp_ 47 90 0.054 91% H50_salmon_Hum_ Rug

7 Prcp_NDVI_Temp_ 53 106 0.032 100% H50_salmon_LHum _Rug

8 Prcp_NDVI_Temp_ 54 107 0.027 98% H50_T25_salmon_R ug

9 Prcp_NDVI_Temp_ 72 131 0.021 82% T25_salmon_Hum 10 Prcp_NDVI_H50_T 57 115 0.019 102% 25_salmon_LHum_ Rug

1 This is the increase in prediction error of Jackknife model vs the full model.

Table 9 Extrapolated grizzly bear densities and population sizes for a selection of areas that are currently unoccupied, occupied at low densities, or are considered threatened populations in western North America. Current population estimates are taken from government sources in BC and the US and predicted population sizes were derived using our top coastal or interior model.

Population unit Current population size Predicted population size

Okanagan valley 0a 27

Thompson valley 0a 13

Caribou plateau 0a 0

Peace river agriculture zone <40a 29

North Cascades (Canada only) 23a 284

Garibaldi-Pitt 18a or 0b 40

Squamish-Lilloet 56a or 52b 180

Toba-Butt 75a or >106b 211

Stein-Nahatlach 61a or 23b 129

South Chilicotin Ranges 104a or >147b 257

Blackwater West Chilicotin 193a 24

Granby-Kettle 81a 88

South Selkirks 58c 85

Yahk 44c 12

Bitteroot RA 0d 445

Cabinet-Yahk RA 44 130

North Cascades RA <5 874

Northern Continental Divide RA 765e 641

South Selkirks RA 30-40 64

Yellowstone RA 600 567

aHamilton 2008. bApps 2009. cProctor et al. 2007. d C. Servheen, USFWS, Montana, pers. com. eKendall et al. 2009.

Table 10 A summary of predicted numbers of grizzly bears in Canada and National

Parks by province based on the coastal and interior models developed in this paper.

Province Current Predicted population Prediction Number in

population size from this paper units National Parks

projection

Alberta 841a 1250 ecoregions 396

British 16,014b 13,131 WMU’s 126

Columbia 13,974 GBPU’s

14,101 ecoregions

Nunuvut 1000c 8080 ecoregions 0

Northwest 5100c 16,771 ecoregions 835

Territories

Yukon 6300c 10,404 guide 465

territories

10,465 ecoregions afrom Wilkinson et al. 2008. bHamilton 2008. cfrom McLoughlin 2011.

Table 11 SM Extrapolated densities and population sizes for all management units in

Yukon using our best fit model for interior areas. Grizzly bears are not know to rely heavily on salmon in any of these areas although local feeding on salmon, arctic char and marine mammals have been documented. Recent kill rates are based on predicted densities.

NAME Area Predicted LCL UCL Populatio Human caused Kill rate density n size kill 1995-2004 Aishihik 22898 19.1 1.0 43.2 436 83 1.9% Anvil 9776 24.9 3.8 45.9 243 20 0.8% Arkell 7537 9.2 -12.729.3 69 41 5.9% Big Salmon 13115 17.3 -6.0 35.7 227 35 1.5% Bonnet Plume 10664 30.7 7.8 50.3 328 46 1.4% Cassiar 35572 27.1 3.2 44.9965 45 0.5% Dezedeash 4564 17.6 -2.4 39.480 48 6.0% Eagle Plains 22736 29.7 7.9 50.3 676 1 0.0% East Arctic 9372 19.5 -1.1 41.1 182 4 0.2% Frances 13149 10.9 -10.331.3 144 8 0.6% Glenlyon 9851 12.4 -6.5 35.4 122 35 2.9% Gold 36448 19.1 -0.2 41.8 696 36 0.5% Hart 17974 30.9 9.2 51.4556 31 0.6% Hyland 28359 16.1 -3.3 39.0 457 27 0.6% Klondike 9669 30.4 9.0 51.0294 31 1.1% Kluane 11239 25.6 3.4 55.7287 18 0.6% Knorr 10294 28.9 5.7 47.7298 1 0.0% Laberge 8308 0.0 -22.919.6 0 29 0.0% MacMillan 21781 30.4 6.4 48.0661 48 0.7% Nadaleen 8257 31.4 9.0 51.4259 46 1.8% Nisling 8883 21.8 3.9 46.9194 43 2.2% North Ogilvie 14863 30.5 9.6 51.9 453 40 0.9% Old Crow Flats 22926 21.0 -0.7 41.1 481 19 0.4% Pelly 15296 23.5 0.5 42.2359 39 1.1% Richardson 17862 28.8 4.1 46.1515 5 0.1% Ruby 2079 26.4 8.8 52.1 55 8 1.5% Southern Lakes 7282 8.5 -12.0 30.0 62 25 4.0% Stewart 19603 28.8 6.8 48.4565 60 1.1% West Arctic 13878 16.9 -1.9 40.5 234 1 0.0% West Ogilvie 14361 26.6 7.0 49.0 382 51 1.3% White 7311 16.4 -0.6 41.7 120 20 1.7% TOTAL 10,404 944

Table 12 SM Extrapolated densities and population sizes for all wildlife management units in British Columbia known to support

grizzly bears derived using our best fit coastal and interior model. The choice of which model to use was based on a survey of

biologists known to be familiar with each area. Recent kill rates are presented based on these predicted densities.

Man Predicte Predicte Predicte Predicte Predicte 5 yr 5 yr % % other % kill age d d SD d LCL d UCL d hunter other hunter kill ment density populati kill kill kill unit on size 1-14 14.6 8.91 -3.23 32.42 85 14 0 3.3% 0.0% 3.3% 1-15 24.2 8.27 7.70 40.77 141 18 2 2.6% 0.3% 2.8% 2-5 4.1 9.08 -14.02 22.30 12 0 0 0.0% 0.0% 0.0% 2-6 36.8 7.56 21.70 51.95 80 0 0 0.0% 0.0% 0.0% 2-7 8.9 8.36 -7.88 25.58 8 0 0 0.0% 0.0% 0.0% 2-8 0.0 9.04 -22.94 13.20 0 0 0 63 2-9 20.5 7.87 4.79 36.28 32 0 0 0.0% 0.0% 0.0% 2-10 14.1 8.05 -2.02 30.17 15 0 0 0.0% 0.0% 0.0% 2-11 40.9 10.85 19.20 62.59 109 0 0 0.0% 0.0% 0.0% 2-12 0.0 9.38 -21.13 16.39 0 0 0 2-13 46.6 7.64 31.32 61.86 108 0 1 0.0% 0.2% 0.2% 2-14 51.7 7.69 36.34 67.11 103 0 0 0.0% 0.0% 0.0% 2-15 47.4 7.58 32.24 62.58 113 0 0 0.0% 0.0% 0.0% 2-18 0.0 9.18 -25.64 11.06 0 0 0 2-19 0.0 9.43 -34.40 3.33 0 0 0 3-15 10.6 8.13 -5.64 26.90 21 0 1 0.0% 0.9% 0.9% 3-16 31.4 10.34 10.76 52.10 92 0 0 0.0% 0.0% 0.0% 3-32 16.8 10.32 -3.79 37.48 60 0 0 0.0% 0.0% 0.0% 3-33 22.5 10.36 1.75 43.17 45 0 0 0.0% 0.0% 0.0% 3-34 22.1 10.47 1.17 43.03 11 0 3 0.0% 5.5% 5.5% 3-35 35.7 10.46 14.81 56.64 30 4 2 2.7% 1.3% 4.0% 3-36 23.1 10.32 2.47 43.73 62 3 0 1.0% 0.0% 1.0% 3-37 8.7 10.33 -11.96 29.38 15 0 0 0.0% 0.0% 0.0% 3-38 2.9 10.43 -17.96 23.74 5 0 0 0.0% 0.0% 0.0%

3-39 0.0 10.39 -21.91 19.65 0 0 0 3-40 16.9 10.45 -4.04 37.77 28 0 0 0.0% 0.0% 0.0% 3-41 21.4 10.39 0.58 42.14 20 0 0 0.0% 0.0% 0.0% 3-42 33.1 10.46 12.15 54.00 49 1 0 0.4% 0.0% 0.4% 3-43 41.5 10.58 20.39 62.69 42 1 0 0.5% 0.0% 0.5% 3-44 46.4 10.58 25.27 67.58 75 3 1 0.8% 0.3% 1.1% 3-45 29.3 10.41 8.48 50.12 61 0 0 0.0% 0.0% 0.0% 3-46 26.2 10.23 5.77 46.69 75 0 0 0.0% 0.0% 0.0% 4-1 47.2 10.68 25.81 68.52 75 7 6 1.9% 1.6% 3.5% 4-2 32.3 10.31 11.70 52.94 40 8 0 4.0% 0.0% 4.0% 4-3 1.4 10.38 -19.38 22.15 2 0 4 0.0% 36.2% 36.2% 4-4 8.4 10.42 -12.41 29.28 9 0 1 0.0% 2.1% 2.1% 4-5 10.3 10.41 -10.51 31.12 9 0 1 0.0% 2.3% 2.3% 4-6 16.2 10.37 -4.53 36.96 36 1 1 0.5% 0.5% 1.1% 4-7 22.5 10.36 1.77 43.19 42 0 0 0.0% 0.0% 0.0% 4-8 20.1 10.36 -0.67 40.78 43 0 3 0.0% 1.4% 1.4% 4-9 10.4 10.43 -10.44 31.27 12 0 1 0.0% 1.6% 1.6% 64 4-14 13.0 10.32 -7.66 33.63 12 0 3 0.0% 5.2% 5.2% 4-15 22.3 10.31 1.65 42.90 38 0 1 0.0% 0.5% 0.5% 4-16 28.4 10.33 7.71 49.04 50 0 0 0.0% 0.0% 0.0% 4-17 32.5 10.32 11.90 53.19 54 8 0 2.9% 0.0% 2.9% 4-18 27.0 10.35 6.34 47.72 37 3 1 1.6% 0.5% 2.2% 4-19 36.6 10.39 15.79 57.36 52 1 0 0.4% 0.0% 0.4% 4-20 20.9 10.30 0.30 41.48 79 11 1 2.8% 0.3% 3.0% 4-21 24.2 10.24 3.69 44.64 32 8 1 5.1% 0.6% 5.7% 4-22 27.9 10.31 7.30 48.54 66 11 12 3.4% 3.7% 7.0% 4-23 24.2 10.42 3.33 44.99 81 31 17 7.7% 4.2% 11.9% 4-24 33.0 10.38 12.21 53.75 60 10 2 3.3% 0.7% 4.0% 4-25 22.9 10.29 2.30 43.45 70 10 1 2.8% 0.3% 3.1% 4-26 22.8 10.34 2.12 43.47 67 9 1 2.7% 0.3% 3.0% 4-27 39.1 10.33 18.42 59.75 54 14 0 5.2% 0.0% 5.2% 4-28 49.7 10.41 28.88 70.51 37 9 0 4.8% 0.0% 4.8% 4-29 47.2 10.42 26.41 68.07 38 7 0 3.7% 0.0% 3.7% 4-30 40.3 10.32 19.64 60.91 51 4 1 1.6% 0.4% 2.0%

4-31 36.0 10.32 15.39 56.69 46 7 0 3.1% 0.0% 3.1% 4-32 27.8 10.25 7.31 48.33 69 7 1 2.0% 0.3% 2.3% 4-33 49.9 10.52 28.86 70.96 105 13 0 2.5% 0.0% 2.5% 4-34 26.1 10.30 5.49 46.68 98 13 3 2.7% 0.6% 3.3% 4-35 18.2 10.33 -2.43 38.88 46 8 2 3.5% 0.9% 4.4% 4-36 25.9 10.30 5.26 46.47 92 20 2 4.3% 0.4% 4.8% 4-37 49.3 10.52 28.29 70.37 92 9 0 1.9% 0.0% 1.9% 4-38 47.7 10.64 26.39 68.97 173 32 0 3.7% 0.0% 3.7% 4-39 45.4 10.48 24.47 66.37 109 23 1 4.2% 0.2% 4.4% 4-40 36.7 10.40 15.94 57.53 77 16 0 4.2% 0.0% 4.2% 5-3 0.1 10.41 -20.70 20.94 0 0 0 0.0% 0.0% 0.0% 5-4 5.7 10.33 -14.90 26.40 43 0 5 0.0% 2.4% 2.4% 5-5 15.0 10.31 -5.67 35.58 79 0 0 0.0% 0.0% 0.0% 5-6 21.6 10.25 1.06 42.05 104 0 0 0.0% 0.0% 0.0% 5-7 23.3 7.97 7.35 39.23 142 0 4 0.0% 0.6% 0.6% 5-8 1.9 8.83 -15.72 19.59 15 21 0 27.6% 0.0% 27.6% 5-9 13.6 8.03 -2.48 29.63 102 12 3 2.4% 0.6% 2.9% 65 5-10 18.8 10.42 -2.09 39.61 63 0 0 0.0% 0.0% 0.0% 5-11 29.0 10.29 8.38 49.53 53 0 3 0.0% 1.1% 1.1% 5-12 0.0 10.45 -24.37 17.41 0 0 13 5-13 0.0 10.50 -28.27 13.72 0 0 0 5-15 23.1 10.28 2.51 43.63 209 0 2 0.0% 0.2% 0.2% 5-16 28.0 10.35 7.27 48.68 30 0 0 0.0% 0.0% 0.0% 6-1 4.6 10.53 -16.43 25.69 24 1 2 0.8% 1.6% 2.5% 6-2 11.9 10.58 -9.23 33.10 51 6 0 2.4% 0.0% 2.4% 6-3 2.3 8.87 -15.46 20.02 35 14 13 7.9% 7.4% 15.3% 6-4 0.0 10.43 -23.39 18.32 0 6 0 6-5 0.0 10.72 -36.05 6.82 0 1 0 6-6 0.0 10.57 -31.16 11.14 0 0 0 6-7 36.6 10.41 15.80 57.44 154 15 1 2.0% 0.1% 2.1% 6-8 1.4 10.34 -19.32 22.04 13 13 9 20.4% 14.1% 34.6% 6-9 17.0 10.43 -3.85 37.85 225 19 7 1.7% 0.6% 2.3% 6-10 58.2 8.00 42.15 74.15 74 0 0 0.0% 0.0% 0.0% 6-11 0.0 10.94 -43.67 0.08 0 17 3

6-14 20.3 8.31 3.66 36.91 180 6 13 0.7% 1.4% 2.1% 6-15 30.9 7.68 15.58 46.30 250 14 8 1.1% 0.6% 1.8% 6-16 7.5 9.08 -10.66 25.64 37 21 3 11.4% 1.6% 13.1% 6-17 46.4 10.61 25.23 67.66 461 37 1 1.6% 0.0% 1.6% 6-18 45.9 10.57 24.76 67.06 294 12 0 0.8% 0.0% 0.8% 6-19 31.3 10.32 10.64 51.91 234 12 0 1.0% 0.0% 1.0% 6-20 32.2 10.23 11.70 52.63 432 45 4 2.1% 0.2% 2.3% 6-21 398 31 0 1.6% 0.0% 1.6% 6-22 291 14 1 1.0% 0.1% 1.0% 6-23 15.5 10.23 -4.96 35.97 167 19 4 2.3% 0.5% 2.7% 6-24 19.3 10.23 -1.14 39.78 179 18 0 2.0% 0.0% 2.0% 6-25 15.5 10.22 -4.94 35.95 243 29 3 2.4% 0.2% 2.6% 6-26 460 47 0 2.0% 0.0% 2.0% 6-27 25.5 10.87 3.71 47.21 113 10 0 1.8% 0.0% 1.8% 6-28 29.1 10.76 7.56 50.59 71 12 0 3.4% 0.0% 3.4% 6-29 55.8 9.80 36.22 75.40 223 17 0 1.5% 0.0% 1.5% 6-30 21.0 10.22 0.57 41.46 140 13 4 1.9% 0.6% 2.4% 66 7-1 29.1 10.35 8.39 49.81 56 0 0 0.0% 0.0% 0.0% 7-2 30.7 10.33 10.06 51.37 92 17 0 3.7% 0.0% 3.7% 7-3 34.9 10.37 14.14 55.63 129 23 0 3.6% 0.0% 3.6% 7-4 26.4 10.33 5.72 47.02 60 7 0 2.3% 0.0% 2.3% 7-5 26.5 10.24 6.02 46.99 77 19 1 4.9% 0.3% 5.2% 7-6 22.8 10.37 2.02 43.51 46 8 2 3.5% 0.9% 4.4% 7-7 1.8 10.44 -19.09 22.68 5 12 5 45.3% 18.9% 64.2% 7-8 2.5 10.37 -18.25 23.24 5 0 0 0.0% 0.0% 0.0% 7-9 0.0 10.46 -24.40 17.45 0 0 1 7-10 0.0 10.51 -28.69 13.35 0 0 0 7-11 0.0 10.61 -31.51 10.92 0 0 1 7-12 0.0 10.60 -32.99 9.40 0 1 2 7-13 0.0 10.69 -37.51 5.23 0 0 4 7-14 0.0 10.70 -35.53 7.25 0 0 1 7-15 0.0 10.85 -37.61 5.81 0 0 6 7-16 9.5 10.44 -11.39 30.38 36 9 1 5.1% 0.6% 5.6% 7-17 23.8 10.35 3.11 44.53 64 13 1 4.1% 0.3% 4.4%

7-18 48.9 10.71 27.48 70.31 181 21 1 2.3% 0.1% 2.4% 7-19 18.1 10.47 -2.86 39.03 79 13 1 3.3% 0.3% 3.6% 7-20 0.0 10.47 -27.31 14.57 0 6 1 7-21 8.8 10.28 -11.76 29.35 59 14 2 4.8% 0.7% 5.4% 7-22 16.4 10.36 -4.36 37.06 68 6 1 1.8% 0.3% 2.1% 7-23 33.8 10.55 12.67 54.87 146 25 1 3.4% 0.1% 3.6% 7-24 0.0 10.48 -22.39 19.52 0 4 1 7-25 0.0 10.87 -37.81 5.68 0 4 0 7-26 0.0 10.50 -24.44 17.57 0 5 1 7-27 6.9 10.32 -13.76 27.53 37 17 1 9.3% 0.5% 9.8% 7-28 5.1 10.46 -15.78 26.07 39 8 0 4.1% 0.0% 4.1% 7-29 0.0 10.52 -25.11 16.96 0 4 0 7-30 5.6 10.49 -15.36 26.61 13 3 0 4.5% 0.0% 4.5% 7-31 10.8 10.34 -9.86 31.50 58 15 0 5.2% 0.0% 5.2% 7-35 0.0 10.44 -27.98 13.77 0 4 0 7-36 10.9 10.49 -10.07 31.91 36 6 3 3.3% 1.6% 4.9% 7-37 28.4 10.52 7.39 49.46 177 33 0 3.7% 0.0% 3.7% 67 7-38 24.0 10.42 3.16 44.85 308 32 1 2.1% 0.1% 2.1% 7-39 43.1 10.51 22.03 64.07 370 20 0 1.1% 0.0% 1.1% 7-40 29.3 10.28 8.70 49.82 216 12 1 1.1% 0.1% 1.2% 7-41 27.5 10.35 6.78 48.17 208 36 0 3.5% 0.0% 3.5% 7-42 20.7 10.51 -0.33 41.69 120 10 12 1.7% 2.0% 3.7% 7-43 8.2 10.47 -12.76 29.13 25 9 2 7.3% 1.6% 8.9% 7-44 0.0 10.44 -25.38 16.39 0 3 2 7-45 0.0 10.44 -24.87 16.91 0 5 1 7-46 0.0 10.54 -27.33 14.85 0 0 0 7-47 0.0 10.69 -25.79 16.95 0 1 1 7-48 0.0 10.50 -25.06 16.94 0 9 0 7-49 0.0 10.48 -27.58 14.34 0 18 2 7-50 22.3 10.45 1.41 43.19 137 11 16 1.6% 2.3% 3.9% 7-51 21.9 10.37 1.20 42.69 381 55 2 2.9% 0.1% 3.0% 7-52 26.6 10.39 5.82 47.40 459 51 2 2.2% 0.1% 2.3% 7-53 3.7 10.49 -17.24 24.73 37 14 1 7.6% 0.5% 8.2% 7-54 9.6 10.46 -11.35 30.50 71 16 2 4.5% 0.6% 5.1%

7-55 1.7 10.56 -19.45 22.81 30 0 0 0.0% 0.0% 0.0% 7-56 0.0 10.57 -22.60 19.67 0 0 0 7-57 21.0 10.43 0.11 41.84 52 7 3 2.7% 1.1% 3.8% 7-58 0.0 10.41 -21.02 20.64 0 0 2 8-3 7.1 10.53 -13.94 28.19 9 0 0 0.0% 0.0% 0.0% 8-4 6.8 10.45 -14.12 27.69 7 0 0 0.0% 0.0% 0.0% 8-13 14.3 10.29 -6.27 34.89 8 0 0 0.0% 0.0% 0.0% 8-14 9.6 10.35 -11.13 30.27 24 0 0 0.0% 0.0% 0.0% 8-15 11.5 10.36 -9.20 32.25 32 0 0 0.0% 0.0% 0.0% 8-23 12.8 10.32 -7.88 33.39 33 0 2 0.0% 1.2% 1.2% 8-24 20.7 10.31 0.08 41.32 26 0 2 0.0% 1.5% 1.5% 8-25 0.0 10.52 -22.21 19.88 0 0 0 TOTAL predicted number of grizzly 13157 bears 68

Fig. 1 The location of each study area for which we derived grizzly bear density across

North America. Areas that are currently unoccupied are shown in dark and others in grey.

Fig. 2 The relationship between brown bear density and mean annual precipitation for study areas where brown bears were allopatric (squares) and where black and brown bears were sympatric (diamonds). Open symbols denote coastal study areas where salmon was a major component of the diet; filled symbols show study areas where salmon were few. Unoccupied areas and one coastal area where brown bears were allopatric and very high density (856) are not shown.

Fig. 3 The relationship between vegetation cover and grizzly bear density in 87 study areas across western North America. These are interior sites where salmon is a minor component of the diet. Black bears appear to be excluded from areas where grizzlies are present and trees cover < about 20% of the study area and the proportion of the study with >25% tree cover appears to best describe this process.

Fig. 4 Observed versus predicted values of grizzly bear density (bears 1000/km2) using the best fit interior model described in Table 4. Data included 75 inventoried study areas and 14 unoccupied areas across the interior of western North America. Error bars are

95% confidence intervals for observed data derived from the survey results or, estimated subjectively based on survey methods (see Methods for detailed description). The cases with the largest residuals often have the greatest error and were hence weighted lower in the regression.

Fig. 5 SM Model averaged predicted densities compared to predictions from the top model in Table 4. Data included 75 inventoried study areas and 14 unoccupied areas across the interior of western North America.

Fig. 6 Observed versus predicted values of grizzly bear density (bears 1000/km2) using the best fit coastal model described in Table 5. Data included 15 inventoried study areas and 2 unoccupied areas across the interior of western North America. Error bars are 95% confidence intervals for observed data derived from the survey results or, estimated subjectively based on survey methods (see Methods for detailed description).

Fig. 7 SM Model averaged predicted densities compared to predictions from the top model in Table 5. Data included 15 inventoried study areas and 2 unoccupied areas across the interior of western North America.

Fig. 1 71

500

400 2

300 72 200 Brown bears/1000 km 100

0 0 100 200 300 400 500 Mean annual precipitation (cm) Fig. 2

80 b No black bears present

80 a No black bears present 2 Black bears present Black bears present

2 Few black bears Few black bears 60 60

40 40

20 20 Grizzly bears/1000 km Grizzly bears/1000 km 0 0 0 20406080100 020406080100 Tree cover (>10%) Tree cover (>25%) 73 80 c No black bears present 80 d No black bears present Black bears present Black bears present 2 2 Few black bears Few black bears 60 60

40 40

20 20 Grizzly bears/1000 km Grizzly bears/1000 km 0 0 0 20406080100 0 20406080100 Herb-shrub cover (>50%) Herb-shrub cover (>75%) Fig. 3 (a-d)

Fig. 4 74

Fig. 6 75

Fig. 5 SM 76

Fig. 7 SM 77

ISSN: 0809-6392 ISBN: 978-82-575-1160-9

Postboks 5003 Norwegian University NO-1432 Ås of Life Sciences 67 23 00 00 www.nmbu.no

View publication stats