Animal ConservationAnimal Conservation. Print ISSN 1367-9430

Is the re a fu ture for Amur tigers in a res to red tiger conservation landscape in N ortheast China?M. Hebbiewhite\ F. Zimmermann^ Z. Li^ D. G. M iqueiie \ M. Zhang®, H. Sun®, F. Morschei^ Z. Wu®, L. Sheng®, A. Purekhovsky® & Z. Chunquan^®1 Wildlife Biology Program, D e p a r tm e n t of E co sy s tem and C onserva t ion S c iences , College of Forestry and Conservation , University of M ontana , Missoula , MT, USA2 KORA, Murl, Switzerland3 N o r th ea s t Normal University, Jllln Province, Changchun , China4 Russ ia Program, Wildlife Conserva tion Society, Vladivostok, Russia5 N o r th ea s t Normal University, Heilongjiang Province, Harbin, China6 Heilongjiang A c ad e m y of Forestry, Wildlife Conservation Institute, Heilongjiang Province, Harbin, China7 World W id e Fund for Nature (WWF) Germany, Frankfurt, G erm any8 Jllln A c ad e m y of Forestry, Jllln Province, Changchun , China9 Russian Far E as te rn Branch, World W ide Fund for Nature (WWF), Vladivostok, Russia10 Beijing Office, World W id e Fund for Nature (WWF), Beijing, China

KeywordsSiberian tiger; connectivity; conse rva t ion planning; carnivore conserva tion; SIkhote-Alln; China; Russian Far East.

CorrespondenceM ark Hebblew hlte , Wildlife Biology Program, College of Fores try and Conservation , University of M ontana , Missoula , MT 59812, USA.Email; m a rk .hebb lew h l te @ um ontana .e du

Editor; Nathalie Pettorelll

A sso c ia te Editor; Nathal ie Pettorelll

Received 5 October 2011; accepted 2 April 2012

dol;10 .1 1 1 1/j. 1469-1795.201 2 .00552.x

AbstractThe future of w ild tigers is dire, and the G lobal Tiger In itiative’s (G TI) goal o f doubling tiger population size by the next year o f the tiger in 2022 will be chal­lenging. The G T I has identified 20 tiger conservation landscapes (TCL) within w hich recovery actions will be needed to achieve these goals. The A m ur tiger conservation landscape offers the best hope for tiger recovery in C hina where all o ther subspecies have m ost likely become extirpated. To prioritize recovery p lan ­ning w ithin this TCL, we used tiger occurrence d a ta from adjacent areas o f the R ussian F a r E ast to develop tw o em pirical models o f potential hab ita t th a t were then averaged w ith an expert-based hab ita t suitability m odel to identify potential tiger hab ita t in the C hangbaishan ecosystem in N ortheast China. We assessed the connectivity o f tiger h ab ita t patches using least-cost p a th analysis calibrated against know n tiger movem ents in the R ussian F a r E ast to identify priority tiger conservation areas (TCAs). U sing a hab itat-based population estim ation approach, we predicted th a t a potential o f 98 (83-112) adult tigers could occupy all TCA s in the C hangbaishan ecosystem. By com bining inform ation about hab ita t quality, connectivity and potential population size, we identified the three best TCA s to taling over 25 000 km^ o f potential hab ita t th a t could hold 79 (63-82) adult tigers. S trong recovery actions are needed to restore poten tia l tiger hab ita t to prom ote recovery o f A m ur tigers in C hina, including restoring ungulate popu la­tions, increasing tiger survival th rough im proved anti-poaching activities, land- use planning th a t reduces hum an access and agricultural lands in and adjacent to key TCA s, and m aintaining connectivity b o th w ithin and across in ternational boundaries. O ur approach will be useful in o ther TCLs to prioritize recovery actions to restore worldw ide tiger populations.

Introduction

W ild tiger Panthera tigris num bers have dram atically d ropped to less than 3200 in the w orld, because o f tiger poaching, poaching of their ungulate prey, and hab ita t destruction exacerbated by rapidly grow ing hum an popu la­tions and economies in A sia (D inerstein e t a l , 2007). Tigers face a dire future, and recovery will take com m itm ent o f w orld leaders and governm ents (W alston et al., 2010). A t

the 2010 St. Petersburg Tiger Sum m it hosted by Russia, tiger range countries com m itted to doubling the population of wild tigers by the next year o f the tiger in 2022 th rough the G lobal Tiger Initiative (W ikram anayake e t a l , 2011). To achieve this am bitious goal, a num ber o f large-scale tiger conservation landscapes (TCLs) were identified (W ikram a­nayake et al., 2011) th a t will need to be actively restored to ensure a viable fu ture for wild tigers. Identifying and p rio ri­tizing smaller-scale recovery areas and actions w ithin these

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o log ioa l S oo ie ty o f London

Restoring tiger conservation landscapes Hebblewhite etal.

large-scale TCLs are the next steps needed to actively recover tigers (W alston et a l , 2010; W ikram anayake e t a l , 2011).

Successful restoration o f TCLs is an enorm ous conserva­tion challenge, and recovery o f tigers in C hina is no exception (D inerstein et a l , 2007) where the South C hina P. t. amoy- ensis, Indochinese P. t. corbetti and Bengal tiger P.t. tigris (Luo, 2010) m ay already be effectively extinct. Recent surveys suggest th a t while there is no viable population of A m ur tigers P. t. altaica in N ortheast C hina, since 2002, there have been a t least 16 different im m igrating increasing reports o f tigers in N ortheastern C hina mainly along the R ussian border (Z hou et al., 2008) from the adjacent Russian population o f 430-500 tigers (M iqueiie et a l , 2006). D espite the ongoing threats o f tiger poaching, prey depletion and h ab ita t fragm entation to A m ur tigers in C hina, growing national governm ent support for tiger recovery, the presence o f large forested areas th rough eastern Jilin and Heilongjiang provinces, and dispersal from connected populations in R ussia provide a good foundation for tiger recovery.

To recover tigers in TCLs around the w orld, potential tiger hab ita t needs to be first identified and prioritized w ithin these landscapes (Smith, A hearn & M cdougal, 1998; C ianfrani e ta l., 2010; W alston e t a l , 2010; W ikra­m anayake et a l , 2011). C arnivore h ab ita t is conceptually a com bination o f sufficient prey density, biophysical and landcover resources, and low m ortality from hum an causes (M itchell & H ebblew hite, 2012). T iger hab ita t is generally considered as forested areas w ith high densities o f large ungulate prey, w ith protection from hum an-caused m or­tality (Sm ith e t a l , 1998; W ikram anayake e t a l , 2004; C arroll & M iqueiie, 2006). A fter identifying potential hab ita t, connectivity betw een h ab ita t patches needs to be assessed (Schadt e t a l , 2002; Linkie e t a l , 2006) and potential population size o f recovered tigers determ ined (Boyce & W aller, 2003) to help prioritize conservation in discrete h ab ita t patches. Land-use planning in identified tiger hab ita t is a vital step in the recovery process because it integrates tiger recovery actions w ith political and eco­nom ic developm ent agendas (W alston e t a l , 2010; W ikra­m anayake e t a l , 2011).

O ur objective was to develop an approach to identify priority tiger recovery areas w ithin a greater TCL. We focused on identifying politically and scientifically defensi­ble tiger recovery areas for A m ur tigers in N ortheast C hina by ( i) defining potential tiger h ab ita t for recovery using three different m ethods developed by different stake­holder team s (Loiselle et a l , 2003); (2) identifying connec­tivity between large patches o f potential tiger hab ita t to identify larger tiger conservation areas (TCAs); (3) esti­m ating potential tiger population size in priority conser­vation areas if full resto ration were to occur; and (4) p ri­oritizing conservation areas for recovery efforts in the N ortheast C hina using a com bination o f criteria including h ab ita t quality, connectivity and potential population size o f recovered tigers. W hile focused on the A m ur tiger, our approach shows prom ise for im plem enting the global tiger initiative’s conservation policies w ithin TCLs th roughou t tiger range and o ther endangered carnivores.

Methods

Study areaO ur study area was a 218 785-km^ portion o f the C hang­baishan ecosystem in the Jilin and Heilongjiang Provinces in N ortheast C h ina and Southw est Prim orye, Russia, and adjacent S ikhote-A lin M oun ta in ecosystem (Fig. i) in southern Prim orski K rai, Russia. While C hangbaishan and the S ikhote-A lin ecosystems have been considered a single T C L (W ikram anayake e t a l , 2011), tiger populations appear to be genetically distinct between them (H enry e t a l , 2009). B oth areas are m ountainous landscapes w ith average elevations from 800 to iOOO m (m ax 2500 m). The climate is tem perate continental m onsoonal w ith average precip ita­tion from 519 to 1336 m m and 20 to 50 cm average snow depth in w inter. The average tem perature in January is -19°C and the average tem perature in July is 20.5°C. Veg­etation is very diverse, ranging from tem perate to boreal, and is characterized by m ajor land cover types o f K orean pine Pinus koraiensis mixed w ith deciduous forests o f birch and oak, mixed coniferous forests a t higher elevations, alpine areas, meadow s, na tu ra l shrublands, coniferous p lan ­ta tion forests, and agricultural areas (see Li et a l , 2010 for m ore details). The m ajority o f forests have been logged, and com bined w ith hum an-induced fire, m any low-elevation forests have been converted to secondary deciduous forests. There are over 370 tow ns or larger settlem ents in the Chinese p a rt o f the C hangbaishan ecosystem w ith over i 1.7 m illion people, and 75 towns/cities w ith i m illion people in southern Prim orye, Russia. U ngulate species in approxi­m ate order o f im portance in the diet o f tigers (M iqueiie et a l , 1996), include red deer Cervus elaphus, wild boar Sus scrofa, sika deer Cervus nippon and Siberian roe deer Capreolus pygargus. The area also has a diversity o f sym- patric carnivores including the sole rem aining population of critically endangered F a r E astern leopards Panthera pardus orientalis, wolves Cants lupus, E urasian lynx L yn x lynx, A siatic black bear Ursus thibetanus and brow n bear Ursus arctos. Thus, successful tiger recovery in this landscape m ay ensure a future for m any o ther rare and endangered species in no rthern Chinese forested ecosystems (Hebblewhite et a l , 2011).

Tiger habitat modelingTo identify potential tiger hab ita t in C hina, we used a simple ensemble hab ita t modeling approach (A raujo & New, 2007; Thuiller e t a l , 2009) th a t averaged three hab ita t models (Fig. 2). We used the average o f three m odels because o f the uncertain ty inherent in all hab ita t models (Barry & Elith, 2006) and to facilitate collaborative approaches in conservation planning am ong three different stakeholders [Chinese governm ent. W orld W ide F u n d for N atu re (W W F) and Wildlife C onservation Society (WCS)] in N ortheast C hina (Loiselle e t a l , 2003). A veraging all three models increased buy-in from all three stakeholders instead of com peting m odels w hen their accuracy to predict tiger

A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o log ica l S o c ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

Russia

C h a n g c h u n * j

• • China

Sea of JapanCh ongjin

North Korea0 25 50 100 150 200

3 km

Legend

n Study area

■■ International border

City

— Main roads

Russian tiger survey Q Tiger absent

I Tiger present

Figure 1 Location o f th e s tudy a rea in th e Changbaishan e c o s y s t e m in N o r th ea s t China and t h e so u th e rn Sikhote-Alin e c o s y s t e m in th e Russian Far East, show ing locations of th e 2005 Russian w in te r tiger su rvey units w h e r e t igers w e r e and w e r e not p resen t .

h ab ita t in C hina was unknow n because of tiger absence. We developed two com plem entary data-driven empirical m odels based on tiger d a ta in the R ussian F a r E ast, which were then extrapolated to N ortheast China; environm ental niche factor analysis (E N FA , Elirzel e ta l., 2002) and resource selection functions (RSE, M anly e t al., 2002). The th ird m ethod used expert knowledge o f A m ur tiger and its h ab ita t requirem ents w ithin C hina to define an expert-based h ab ita t suitability m odel across R ussia and C hina following m ethods explained in m ore detail in X iaofeng e t al. (2011). We first describe the tiger d a ta used to develop empirical models, landscape covariates, and then the three hab ita t m odeling approaches.

Russian tiger surveys

We used tiger track d a ta collected during a range-wide survey in F ebruary and M arch 2005 across all tiger hab ita t in the R ussian E ar E ast using survey m ethods th a t are reported in detail elsewhere (C arroll & M iqueiie, 2006; M iqueiie et a l , 2006), so we only briefly review them here.

The southern Prim orye K rai po rtion o f the study area was divided into 486 sam pling units averaging 131 km^. W ithin each sam pling unit, an average o f 89 km of transects (to ta ­ling 11 473 km ) were surveyed by vehicle, snowmobile o r on foot/skis. We only used sample units w ith > 25 km o f survey effort to ensure detection probability was 1.0 (C arroll & M iqueiie, 2006, Elebblewhite, unpubl. data). The num ber and location o f 595 fresh (< 24 h) tiger tracks were used elsewhere to estim ate tiger abundance using snow tracking - density algorithm s developed in the R ussian E ar E ast (M iqueiie e t a l , 2006, Stephens e t a l , 2006).

Landscape covariates

We used geographic inform ation system (G IS) landscape variables (see Supporting Inform ation for m ore detail) though t to explain tiger hab ita t based on o ther studies (W ikram anayake e t a l , 2004; C arroll & M iqueiie, 2006; Linkie e ta l., 2006). These included biophysical resources including topographic variables (elevation, slope, aspect) from a 100-m resolution digital elevation model, and four

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zo o log ica l S o c ie ty o f London

Restoring tiger conservation landscapes Hebblewhite etal.

Model-averaged tiger habitat model

Habitat-based population estimate

Expert-based MSIhabitat suitability

index model

ENFA(environmental niche

factor analysis)

RSF(resource selection

function)

Habitat delineation (habitat, non­

habitat)

Least-cost path TCA connectivity

analysis

Tiger conservation landscape

identification

Figure 2 S chem a t ic for identification of Am ur tiger habita t in th e Changbaishan s tu d y area in N or theas t China.

key 30-m land cover com m unities, K orean pine mixed w ith deciduous forests, deciduous forests, coniferous forests plus o ther na tu ra l landscapes, and hum au-dom iuated land­scapes. We also used rem otely sensed measures o f net prim ary productivity (N PP) derived from the m oderate resolution im aging spectroradiom eter (M O D IS) satellite d a ta (i-k m resolution, M O D 17A 3 d a ta p roduct. R unning e ta l., 2004) as well as the percent o f the 2005 w inter (N ovem ber i to A pril 30). E ach M O D IS pixel was covered by snow as m easured by the M O D IS satellite (250-m reso­lution, M O D iO A product. H ail e ta l., 2002), F inally, we used hum an use d a ta a t a iOO-m resolution including hum an settlem ents classified as tow ns (< 20 000 people) or cities (> 20 000), and roads divided in to low-use roads (m ostly logging roads), secondary roads (unpaved or rarely used paved roads) w ith m oderate levels o f traffic and prim ary roads (highways and m ain access roads) w ith high traffic volumes. D espite the im portance o f ungulate prey as hab ita t for carnivores like tigers (K aran th e ta l., 2004), the absence of standardized and reliable prey da ta across C hina prevented inclusion o f prey density. However, we exam ined prey in hab ita t m odels for the R ussian portion o f the study area elsewhere (Li e t al., 2010; M itch­ell & Hebblew hite, 2012), w hich we re tu rn to in the discussion.

Ecological Niche Factor Analysis (ENFA)

E N F A (Hirzel et al., 2002; Basille et al., 2008) relates cov­ariates at the spatial location o f a species com pared w ith covariates available w ithin a study area to a reduced

num ber o f uncorrelated and standardized factors in a p ro ­cedure similar to a principal com ponents analysis. The first factor th a t is extracted is the m arginality, w hich m easures how species’ locations differ from the average conditions in the study area. The next factors explain species’ specializa­tion (a m easure of the niche breadth) by com paring w hat is used by the species w ith the available range o f environm en­tal conditions w ithin the study area (Hirzel et a l , 2002). We laid ou t a custom ary 2 x 2-km grid across Southern P rim o­rye in R ussia, and considered the 441 grid cells w ith > i tiger track as a ‘presence’ d a ta point. We com puted the analyses in B iom apper 4.0 (Hirzel e t a l , 2007) and included only uncorrelated (r < 0.7) covariates in the E N FA , aver­aged across the 2-km^ grid using a 5-km radius moving window (to approxim ate the spatial scale o f the sam pling u n it in the RSE, see later), and standardized d a ta before analysis. The m odel derived from d a ta in R ussia was then in terpolated/extrapolated to the C hangbaishan region using the tool ‘E x trapo la te’ in B iom apper 4.0 to in terpolate/ extrapolate the m odel using the harm onic m ean algorithm (Hirzel e t a l , 2006). We validated the E N F A using Boyce et al.’s (2002) k-folds cross-validation Spearm an rank co r­relation index.

RSF modeling

RSEs (M anly e t a l , 2002) were estim ated in R ussia by com ­paring resource covariates in used (n = 198) and unused (n = 288) survey units following a presence-absence (used- unused) design. We evaluated tiger selection a t the scale o f the survey un it using m ean covariate values (density of

A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o log ica l S o c ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

roads, % forest type 1, etc, Supporting In fo rm ation Table S I) for survey units using A rcG IS 9.3 (ESRI, R edlands, CA, U SA ) Z onal Statistics function. U sed and unused sam pling units were then contrasted w ith fixed-effects logistic regres­sion, w hich yields a true probability betw een 0 and 1 when detection probability is near 1.0 (M anly e ta l., 2002). To extrapolate results from the R SF developed in R ussia to the C hangbaishan ecosystem, we used a m oving window analy­sis to spatially scale variables using a circular moving window w ith a 6.6-km radius, equivalent to the 131-km^ survey unit. F irst, we screened variables for collinearity using a cut-off o f r = 0.7 and w ith variance inflation factors, and assessed nonlinear effects using quadratics (H osm er & Fem eshow , 2000). We then created a set o f models fo r which we conducted m odel selection using A kaike inform ation criteria. We tested for m odel goodness o f lit using the like­lihood ratio chi-squared test (H osm er & Fem eshow, 2000) and evaluated the predictive capacity o f the top m odel using pseudo-r^, the logistic regression diagnostic receiver opera t­ing curves (RO C) and classiflcation success. We evaluated the predictive capacity o f the R SF also using k-folds cross validation (Boyce e t al., 2002).

The expert habitat suitability model

The expert m odeling approach was described by X iaofeng e ta l. (2011) who identifled potential tiger hab ita t in a b roader area th an ju s t the C hangbaishan region in N o rth ­east China. The expert m odel was developed using an appli­cation o f rule-based h ab ita t suitability functions (independent o f the E N F A or R SF models) to existing large forest patches (> 500 km^) in A rcG IS 9.2 (ESR I, Redlands, CA, USA). Based on this initial constrain t, four G IS land­scape covariates described earlier were included in the expert model: land cover type, elevation, proxim ity to roads and a settlem ent disturbance factor (which included village density and a m ultiplier for large settlements). We then estim ated arb itrary cost functions for the four variables as predictors o f h ab ita t suitability for tigers, w ith the lower the cost, the higher the value o f the param eter for tiger (see X iaofeng et a l , 2011 and Supporting Inform ation Table S2 fo r details). F o r example, m edian elevations (400-800 m) were considered the m ost suitable for tigers (cost score o f 1), m iddle and upper elevations (2 0 0 ^ 0 0 m and 800-1500 m) were assigned a cost score o f 2, and lower (< 200 m m ostly populated by hum ans) and high elevations (> 1500 m) were considered the poorest h ab ita t (cost score o f 4) (Supporting In fo rm ation Table S2). M ixed-coniferous (K orean pine) w ith deciduous forests, and deciduous broadleaved forests were considered the best tiger hab ita t (cost = 1), pure deciduous stands ‘good ’ (cost = 2) hab itats , coniferous forests shrublands, o r w etlands as poo r hab itats (cost = 6), and hum an-dom inated land covers (i.e. agricultural, cities) ranked as not-suitable (cost = 15). Cells > 5 km from prim ary roads and > 3 km from secondary roads were ranked the highest suitability ( c o s t= l ) , 2-5 km from prim ary roads and 1-3 km from secondary roads as good h ab ita t (cost = 2) and areas close to roads (0-2 km from

prim ary roads, 0-1 km from secondary roads) as poor h ab ita t w ith a cost o f 6. F inally, village density was ranked as the highest quality (cost = 1 ) w hen < 2 villages/100 km^, good w hen 3-6, poor w hen 7-19, and unsuitable when < 1 0 km from a city o r 5 km o f a county center or 2 km from a tow n (Supporting Inform ation Table S2). These values were assigned to cells o f 200 m^ across a grid covering bo th the C hangbaishan ecosystem in C hina and S ikho te- A lin in Russia. The resu ltan t potential hab ita t suitability index was calculated as: tiger hab ita t value = elevation -i- land cover -i- road proxim ity -i- settlem ent density, w ith hab ita t suitability values ranging from 4 (m ost suitable) to 37 (unsuitable). We rescaled the expert m odel between 0 and 100 where 100 was the highest suitability, and 0 the lowest (see F i e t a l , 2010; X iaofeng et al., 2011).

Habitat model averaging

We used a simple ensemble m odeling approach (A raujo & New, 2007) th a t averaged all three h ab ita t models to describe tiger h ab ita t in C hina. To accom m odate different spatial resolution (grain size) from different models, we recalculated m odel projections a t the largest resolution of inputs, and then resam pled to 500 m^. We averaged models instead of weighting based on area under the curve or R O C scores because o f scant tiger validation d a ta w ithin C hina for optim um m odel evaluation. M oreover, p rio r to a hab ita t m odeling w orkshop in 2008, different stake­holders (Russia, C hina) were using different modeling approaches, w ith the potential for com petitive and con tra ­dictory results ham pering conservation. Therefore, model averaging was used as p art o f a collaborative stakeholder process including R ussian, Chinese and w estern modeling approaches th a t w ould ultim ately be m ore successful politically than com peting hab ita t m odels (Foiselle e ta l , 2003). To com pare the predictions o f the three modeling approaches against each other, we evaluated the correla­tion and linear regression between all three models from 10 000 random ly generated locations across the study area.

Potential numbers of tigers in the Changbaishan ecosystemWe used the hab itat-based population m ethod developed by Boyce & M cD onald (1999) to estim ate potential tiger po p u ­lation size w ithin each T C A (see later) to the averaged h ab ita t suitability model. G iven the estim ate for the num ber o f tigers {N, range) in R ussian (M iqueiie e t a l , 2006), and the to ta l h ab ita t suitability [i.e. sum o f all h ab ita t suitability scores w(x)i from 0 to 100 across all G IS pixels] quality in Russia, we estim ated the to ta l hab ita t suitability required per tiger and predicted the tiger population size in each TCA following:

(1 )N v

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zo o log ica l S o c ie ty o f London

Restoring tiger conservation landscapes Hebblewhite etal.

where ^ .w(x)^ is the sum m ed hab ita t suitability for each TCAj, and N is the tiger population estim ate for R ussia (known) and for TCAj [solved by rearranging equation (1)]. K ey assum ptions of this approach include: (1) the right covariates are m easured; (2) similar landscape configuration o f available hab ita ts and selection patterns occur in bo th R ussia and China; and (4) there exist similar relationships between population param eters and available hab ita t (Boyce & M cD onald , 1999; Boyce & W aller, 2003).

Identifying and prioritizing TCAsWe used a cu t-poin t value from the averaged A m ur tiger h ab ita t m odel to tu rn the continuous prediction o f tiger h ab ita t suitability into discrete tiger h ab ita t patches (i.e. h ab ita t vs. non-hab ita t, L iu et a l , 2005) th a t correctly clas­sified 85% of tiger tracks collected in Russia. Potential tiger h ab ita t patches were defined using the tool R egion-G roup in A rcG IS 9.2 (ESRI). N ext, connectivity of these patches was assessed using a least-cost approach w ith the C ostD is- tance T ool in A rcG is 9.2 (Chetkiewicz, C lair & Boyce, 2006; Z im m erm ann & Breitenm oser, 2007; Jan in e t a l , 2009). We used tiger h ab ita t patches as sources for least-cost modeling, and estim ated the m ovem ent cost surface between patches using an expert-based ‘fric tion’ model.

The expert-based cost surface was defined as the relative ‘cost’ to tiger m ovem ent on a scale o f 1 (high-quality hab ita t and connectivity) to 1000 (insurm ountable barrier) for each land cover and hum an covariate using expert opinion sim ilar to the expert-based m odel (see Supporting In fo rm a­tion Table S3). Villages (< 10 000 people) were buffered by 500 and 1000 m and the three larger cities (10 000-20 000; 20 000-50 000; 50 000-100 000) were buffered by 2, 3 and 5 km , respectively. All 200 x 200-m cells falling into the buffer around settlem ent types I - I I I were considered insur­m ountable barriers for tigers and their value was set to 1000 (high cost to m ovem ent). A value of 400 and 100 was given to cells falling into the 0-500 m and 500-1000 m distances from villages, respectively. Low-use roads were no t included because they do no t appear to lim it m ovem ent o f tigers. Secondary roads in R ussia were given a value o f 130 and prim ary roads a value o f 200. Because m ain roads in C hina are generally fenced and all road categories have higher traffic volum es com pared w ith Russia, secondary and prim ary roads were assigned values o f 200 and 800, respec­tively. L and cover types com m only used by tigers (M iqueiie e t a l , 1999) were given a friction value o f 1 (K orean pine mixed w ith deciduous forests, deciduous forests), whereas coniferous forests and o ther na tu ra l vegetative types (less preferred) were given a value of 10. H um an-dom inated lands were given a cost value o f 100.

To determ ine w hich hab ita t patches were connected in a single TCA , we defined a threshold o f the m axim um accu­m ulated costs fo r tigers m oving between adjacent quality h ab ita t patches (Z im m erm ann & Breitenm oser, 2007; Janin e t a l , 2009). We used knowledge o f tiger m ovem ent in the southern S ikhote-A lin M oun ta in ecosystem to calibrate the cost surface. In the Sikhote-A lin , tigers move regularly

between adjacent quality hab ita t patches and thus, all patches can be considered to be connected to each other form ing one single un it except for Southw est Prim orye (H enry e t a l , 2009). We set the threshold o f the accum u­lated cost grid so th a t all h ab ita t patches in R ussia were connected except for Southw est Prim orski K rai (e.g. Janin et a l , 2009). By applying the same threshold values to the C hangbaishan ecosystem, we defined connected hab ita t patches greater than 400 km^ (approxim ately 1 female home range, G oodrich et a l , 2010) as TCAs.

We prioritized TCA s for recovery using the following criteria recom m ended based on previous studies o f carn i­vore landscape conservation (W ikram anayake et a l , 2004; C arroll & M iqueiie, 2006): (1) distance from the closest source population in R ussia along the least-cost pa th between the closest source and the respective TCA; (2) area; (3) the potential tiger population size; (4) a fragm entation index calculated as the ratio o f the perim eter to the area m ultiplied by 1000; (5) w hether tigers were currently present based on num ber o f reports; and (6) level o f isolation, ranked according to the num ber of linkages between the top nine TCA s (see Li et a l , 2010).

Results

Habitat modeling

ENFAU sing the m arginality {M), the preferred h ab ita t o f tigers was identified as areas including a higher m ean slope, a h igher frequency o f deciduous forests, a greater distance from villages and large cities, a lower frequency o f hum an- dom inated landscapes, and a lower density o f prim ary and secondary roads th an available in the R ussian study area (m arginality scores; Table 1). A n overall tolerance value T (T = 1!specialization) o f 0.68 indicated th a t tigers were p re ­dom inately n o t hab ita t specialists. We identified three factors {M, S I -2; Table 1) th a t accounted for 54.2% o f the to ta l specialization. Tiger d istribution was restricted to a narrow range regarding the m ean N PP, the m ean density o f prim ary roads, and, to a lesser extent, the frequency of deciduous forests (specialization scores SI and S2; Table 1). The m ean k-folds rank correlation was relatively high (0.82), confirming good m odel cross-validation, bu t w ith a re la­tively large standard deviation (SD) o f 0.24, indicating vari­ation in m odel perform ance. Overall m arginality, M , o f the E N L A m odel applied to the calibration d a ta sets was 0.73, confirming th a t potential tiger h ab ita t in the C hangbaishan region differed from the R ussian E ar Last. However, when in terpolating and extrapolating the m odel to the Chinese portion o f the study area (Supporting Inform ation Lig. S2a), all h ab ita t suitability values were w ithin the range o f the factor values in the calibration area ±10% (allowable per­centage o f extrapolation) and could therefore be com puted. All 1802 extrapolated cells (3.2% of the area) in the Chinese po rtion o f the study area were located in highly hum an- dom inated areas and, consequently, their tiger hab ita t suit-

A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o log ica l S o c ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

Table 1 S u m m a ry of ENFA and RSF empirical habita t m o d e ls for Amur t igers in Sou the rn Primorye Krai, Russ ian Far East, tha t w e r e u s ed to p redic t potentia l tiger habita t in th e C hangba ishan e c o s y s t e m in N or theas t China

Landscape covariate

ENFA resuits“ RSE resuits“

M^2 3 .4 %

S1“2 1 .3 %

S2^9.5% Selectivity, |3“‘ SE

Mean slope + + + + 0 * * * _ d -Elevation N/A“ -0 . 0 0 1 4 0.00079*Deciduous forests + + + * * * * * 2.12 0.612**Distance to villages + + + 0 0 - -Distance to cities + + + * * * - -Snowcover SD + + 0 0 - 2 .1 7 0.888**Mixed Korean pine -deciduous forests + + * * * 1.78 0.722NPP + + ******** ***** 0.00267 0.00145*NPP - quadrat ic -1 .66E-07 1.01E-07*Coniferous forests + 0 * 1.15 0.532**Low-use road density 0 0 * - -Mean aspect 0 0 0 - -Mean snow cover - * * 0 - 2 .1 7 0.888**Aspect SD - 0 0 - -NPP SD - 0 0 - -

Hiiishade - 0 0 - -

Density of villages N/A^ - 6 4 .3 27 .43**Density of cities N/A^ - 2 6 4 . 6 100.30**Secondary roads - * 0 - -

Primary roads -* * * * * ******* - -

Human-dominated landscape. - * 0 - 9 . 0 0 2 .223*

“17 ENFA covar ia tes in rank order from positive to negat ive e f fec ts on t igers w ith marginality (M) and special izat ion s c o r e s (SI and S2). For the RSF model, be ta coeff ic ients indicate s e lec t ion if > 0 and avo idance if < 0, and are s h o w n with SEs and P v a iu e s (* * P < 0.05, *0.05 < P < 0.10). ‘“Positive va lues (+) indicate tha t t igers w e r e found in locations w ith higher than a v e rag e ceil values. Negative va lues (-) indicate th a t t igers w e r e found in locations with lower than a v erag e ceil values. The g rea te r th e n u m b e r of sym bols , th e higher th e correlat ion; 0 indica tes a very w e a k correlation.“Any n u m b e r > 0 m e a n s t h e s p e c ie s w a s found occupying a na rrow er range of va lues than available. The g re a te r th e n u m b e r of sym bols , th e na rrow er th e range; 0 indicates a very low specialization.^Was no t re ta ined b e c a u s e of collinearity.ENFA, environmenta l niche facto r analysis; NPP, n e t primary productivity; RSF, re sou rce selec t ion function; SD, s tandard deviation; SF, s tandard error.

ability values were zero. These results confirm ed th a t extrapolation from R ussia to C hina was appropriate.

ing hab ita t from non-hab ita t was 0.42, w hich resulted in 75% correct classification o f units.

RSF

Tigers selected areas w ith low densities o f cities and villages, interm ediate N PP, and interm ediate elevations, while avoid­ing areas w ith high snowfall (Table 1, Supporting In fo rm a­tion Fig. S2b). In term s o f land cover, tigers preferred deciduous forests, K orean pine, then coniferous/other na tu ra l land cover types, and strongly avoided hum an- dom inated areas (Table 1). The overall m odel was signifi­can t (likelihood ratio %̂ = 51.5, P = 0.0001), dem onstrated good m odel fit (H osm er & Fem eshow goodness o f fit test,

= 4.45, P = 0.77), and had reasonable measures o f model fit and validation w ith a R O C score o f 0.77, a p s e u d o - o f 0.15 and k-folds cross-validation spearm an rank correlation o f 0.712, suggestive o f lower cross-validation success, bu t w ith narrow er predictions (SD = 0.10) com pared w ith the E N FA . The optim al cu t-point p robability for discrim inat­

Expert habitat suitability index model

We developed cost functions for each of the four variables (X iaofeng et a/., 2011) and then sum m ed values o f all layers to develop an assessm ent o f potential tiger hab ita t, w ith scores o f each grid ranging from 4 to 37, w hich were then rescaled to 0 to 100.

Habitat model averaging

Overall, the three m odels were reasonably correlated w ith each other, bu t no t high enough to suggest using only one model. The pair-wise correlation coefficients betw een the E N F A and R SF m odel was r = 0.49; the E N F A and Expert model, r = 0.52; and the R SF and the E xpert model, r = 0.38. All three models showed similar positive correla­tions as predicted hab ita t quality im proves, b u t the R SF and

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zo o log ica l S o c ie ty o f London

Restoring tiger conservation landscapes Hebblewhite etal.

E N F A models (which were calibrated against know n tiger occurrence in Russia) tended to predict m ore low-quality h ab ita t th an the expert model. We used the rescaled (between i and 100) average o f ail three hab ita t m odels to represent tiger hab ita t suitability (Fig. 2).

Identifying TCAsN ine TCA s were identified from the cost-distance analyses after patches < 400 km^ were rem oved (Fig. 3, Table 2). Two TCA s, Fiunchun-W angqing (TCA i) and M ulin (TCA 4; Fig. 3) are shared w ith Russia: 78.7 and 40% of their area is located in C hina, respectively. C hangbaishan (TCA 2; Fig. 3) is likely shared w ith K orea, bu t lack o f cooperation lim ited our ability to include K orea in analyses. The size o f the TCA s ranged from 440 to 14 230 km^, and to taled 22% of the C hangaishan ecosystem, m ostly concentrated in m ountainous regions (Table 2). F iunchun-W angqing, C hangbaishan, South Z hangguangcailing and M ulin encom pass im portan t protected areas, w ith the p roportion

o f effectively protected area ranging from 4.7% (TCA 3) to 13.4% (TCA 2) (Table 2). All TCAs have low village densi­ties (range: 0-0.35 villages per 100 km^) com pared w ith the overall area in C hina (4.3 villages per 100 km^). Secondary road density was also m uch lower in TCA s (range: 2 .3 - 6.8 km/iOO km^) com pared w ith overall area in C hina (15.5 km/iOO km^) except for T C A 6 and T C A 9. Based on ail factors, including predicted population size o f tigers, we ranked the top four TCA s as the Fiunchun-W angqing com plex (TCA i), southern C hangbaishan (TCA 2), sou th­ern Z hangguangaiiing (TCA 3) and M ulin (TCA4).

Potential numbers of tigers in the Changbaishan ecosystemIn con trast to the geographic split between the Chinese (63%) and R ussian (37%) portions o f the C hangbaishan ecosystem, R ussia contained 56% of tiger hab ita t, confirm ­ing higher-quaiity hab ita t on the R ussian side o f the border (Fig. i). There was an estim ated 181 (range o f 160-203)

N

A

' • : \ *«» a ■ • ■ Legend• City

I ] Study areaI Russian-Chinese border

Main roads

Model average value

High : 1.0

0 25 50 100 150 200 250 300 350 400I km Low : 0

Figures Potential Am ur tiger habita t for th e C hangba ishan and Russ ian Far East s tu d y a re a s b a sed on th e e n s e m b le ave raged re sou rce se lec t ion func tion, environmenta l niche factor analysis and e xper t habita t model.

A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zoo log ioa l S oo ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

-gr

E XCT) CD

- E

rvjm'=^c£>0O(TiLnc£>

o o o r ^ o o o o o o o

o o o o

00 C£> C£> m rvj 'T— I I I (71 ^

O 'T— CO (71 LT1 LT1Ln m

(71 L n L n (71 rvj

I I I I I rv j (71 (71 L n (71

rvj rvj

(71 (71 (£> m (71

r v j c o m ' ^ m i - n o O L n o(£)(£)(£)(£)r\jcD ';:trnr^

rv j L n (71 rv j LnfN r n r n CD (-NJ r n CD

CD CD CD CD CD CD

CO

fN rn

rvj o rvj (£> m CD 00

00 Ln m ON ON

C71 fD ^ C? 5 E <5

r\jrn'=^Lnc£)r^0O(7i

• i=: O) c cCO CD

>- .pO

D ^Q. ^O (7)Q. OCD ^o VID to O 7n CO g

0 ) ^ 0 ■i=! .> O

m ^ 2 0 D .

_ o ^O o

J= ^ Q.

<CJ

to oD O )

CC

E S0 =3 h=^ 9- ^c O . to oCDE

~ o0

Eo

Q .

%0 to

CD IIJ Z JD 0 c. 1 ■1—’ o oo CD c

CD CD

CD CD to oQ_ X Q to

<C J

II c X o0 -P

I ?

QCO

adult/subadult tigers in the southern R ussian F a r E ast study area in 2005 (M iqueiie e ta l., 2006). This translates into roughly 445 km^ per tiger, reassuringly close to hom e range estim ates for female tigers (G oodrich e t al., 2010). We p re­dicted a to ta l o f 98 (83-112) tigers across the nine m ain tiger conservation units (Table 2), bu t the top four TCA s con­tained 79 (80%) of those tigers, and only three TCA s con­tained h ab ita t for > 10 tigers: H unchun, C hangbaishan and South Z hangguangcailing (Table 2). The estim ates for the transboundary TCA s (1 and 4) do no t include the num ber o f tigers present on the R ussian side o f the zone.

Evaluating connectivity betw een tiger habitatWe identified the 12 highest-ranked potential linkages (black lines labeled from A to L in Fig. 4, Table 3) between the nine TCAs. Their lengths ranged between 1 and 68 km and accum ulative costs varied sixfold betw een the lowest- and highest-cost corridors. Fiunchun-W angqing (TCA 1), the largest TCA , is connected to three adjacent TCAs: South Z hangguangcailing (TCA 3, 2-km connection), M ulin (TCA 4, 11 km) and C hangbaishan south (TCA 2, 64 km). We identified o ther linkage zones ranging in length from 1 to 68 km betw een o ther TCA s (Fig. 4), b u t rank the three m ost im portan t linkages as between T C A 1 and 4 (Fig. 4b, 11 km), between T C A 1 and 3 (Fig. 4a, 2 km), and T C A 3 and 6 (Fig. 4d, 11 km) based on relative costs and adjacency to source populations (Table 3).

DiscussionO ur T C A prioritization benefited scientifically and politi­cally from using three different h ab ita t m odeling approaches. D uring a 2008 h ab ita t m odeling w orkshop, different stake­holders (W W F, W CS, Chinese governm ent) were approach­ing h ab ita t m odeling from different statistical backgrounds (e.g. R SF, E N FA , Expert model). This created the potential for com peting or contradictory modeling approaches th a t could potentially derail conservation planning (Loiselle e ta l., 2003). Instead, we assum ed th a t all h ab ita t models have differential weaknesses and strengths, and th a t averag­ing across m odels w ould increase scientific rigor (W ilson e t a l , 2005; A rau jo & New, 2007). This was especially im por­tan t w hen no independent tiger validation d a ta existed w ithin N ortheast C hina (Barry & Elith, 2006). U ltim ately, only fu ture tiger recovery will test o f the accuracy o f ou r models (M ladenoff, Sickley & W ydeven, 1999). O ur approach increased political support for the identified TCA s, which were form ally adopted by the Chinese governm ent fo r tiger conservation planning (Li et al., 2010). M oreover, all three m odels captured similar well-known aspects o f tiger habitat: tigers avoided steep slopes or high elevations, strongly selected for K orean pine and deciduous forests, avoided high snow depth, avoided coniferous forests, and strongly avoided hum an villages and roads (Table 1, Sm ith e ta l , 1998; W ikram anayake et a/., 2004; C arroll & M iqueiie, 2006;

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o iog ioa i S oo ie ty o f London

Restoring tiger conservation landscapes Hebblewhite etal.

I t v i * ' « ( V

■"C'.2r■%*rT C A rS feLegend

Tiger conservation area (TCA)

□ R ussia-C hina border

Cities/towns

Main roads0 25 50 100 150 200

km

(b)

T'.'. A.' t

Primorski Krai

0 30 60 120 180 240■km

Legend

Main connections betweeen TPAs Least-cost paths

| _ _ | Russian-Chinese border

Accumulated cost gridHigh : 500000

Low : 0

Figure 4 T g e r habita t in N or theas t China show ing (a) th e nine tiger conse rva t ion a re a s (TCAs) identified from potential tiger habita t pa tches : H unch u n -W an g g in g (TCA 1), C hangba ishan (TCA 2), Sou the rn Zhangguangcail ing (TCA 3), Mulin (TCA 4), Huadian (TCA 5), Nor thern Zhang­guangcailing (TCA 6), Baishan Tanghua - J ian (TCA 7), Lushui-Dongjiang (TCA 8) and J ingyu-J iangyuan (TCA 9); and (b) leas t-cos t p a th s b e tw e e n TCAs, highlighting th e tw e lve primary linkage z o n es b e tw e e n TCAs (labeled A-L) in red, w ith line th ick n ess inversely corre la ted to cost .

10 A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zoo log ioa l S oo ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

T ab les Characte r is tics of 12 potent ial primary linkages connec t ing TCA in th e C hangba ishan landscape , listed th e s h o r t e s t to th e lo n g es t connec t ing d is tance

Name Linkage Length (km) Accumulative costs

H TCA 2 - 8 1 161 000A TCA 1-3 2 161 883G TCA 5 -8 3 162 897F TCA 2 -5 9 168 580B TCA 1 ^ 11 62 307D TCA 3 -6 11 228 474i TCA 5 -9 13 52 740J TCA 2 - 9 15 185 865E TCA 3 -5 18 240 411L TCA 7 -9 39 292 631C TCA 1-2 64 309 132K TCA 2 -7 68 309 261

N am e (letter) identif ies e ac h linkage in Fig. 4. Accumulat ive c o s t s uni ts a re a relative value, s e e te x t for details.TCA, tiger conse rva t ion a reas .

Linkie e ta l., 2006). Based on the statistical convergence betw een models, and the conservation planning benefits, we recom m end m odel averaging as a useful approach to p rio ri­tize tiger recovery areas w ithin o ther TCLs across A sia, and o ther recovering carnivore populations.

O ur results suggest th a t the top four TCA s represent a viable opportun ity fo r the Chinese governm ent to m eet its com m itm ents o f recovering A m ur tigers by the next year o f the tiger, 2022, under the G lobal Tiger initiative. W ithin these top four TCA s, there was hab ita t for a to ta l o f 72 adult A m ur tigers. W hile n o t a large potential population by itself, if connectivity is m aintained w ith the adjacent S ikhote-A lin population in R ussia (400-500 individuals, M iqueiie e ta l., 2006), the C hangbaishan-S ikhote-A lin m eta-population w ould be one of the largest wild tiger populations in the w orld (D inerstein e ta l., 2007; W alston e t a l , 2010). The connectivity between R ussia and C hina is also encouraging, as tigers are already dispersing success­fully from R ussia to C hina, emphasizing rein troduction of captive anim als is no t needed (H ayw ard & Sommers, 2009). Recovery of tigers in the C hangbaishan ecosystem will be contingent on m aintaining and im proving connectivity w ith R ussia and w ithin C hina, increasing tiger survival by reduc­ing tiger and ungulate poaching (K aran th e t a l , 2004; C hapron e t a l , 2008), and reducing fragm entation from incom patible hum an land uses in and around TCA s (Kerley e t al., 2002; C arroll & M iqueiie, 2006).

The highest priority T C A is the H unchun-W angqing area, w hich has the largest connected netw ork o f hab ita t patches, is contiguous w ith source populations in Russia, and has the largest potential num ber of tigers. Tigers are already present in H unchun and W angqing, m uch of which is in protected areas already, and recovery efforts such as rem oval o f snares are already underw ay (Yu e t a l , 2006). The C hangbaishan is the second largest TCA , and while it has poten tial to hold up to 24 adu lt anim als, tigers have no t been reported in the C hangbaishan for > 1 5 years, and the

least-cost distance from a source population (184 km) is very far. Southern Zhangguangcailing and M ulin, the rem aining two top-ranked TCA s, bo th have potential link­ages (2 and 11 km) to H unchun-W angqing , and M ulin is also connected to suitable hab ita t in Russia. E fforts should be m ade to ensure fu rther hab ita t loss does no t occur, and to m aintain or restore linkages (Chetkiewicz e t a l, 2006) w ith the H unchun-W angqing area th rough tiger-friendly ‘green’ infrastructure (Q uintero et a l , 2011) such as wildlife cross­ing structures (Clevenger & W altho, 2005) in identified m ovem ent corridors (Colchero et a l , 2010) across tran spo r­ta tion networks.

D espite these encouraging results, ou r A m ur tiger hab ita t m odel m ay overestim ate tiger hab ita t quality because o f the lack of inform ation about ungulate prey densities, one o f the key com ponents o f tiger hab ita t (K aran th e t a l , 2004). U sing tiger-based R SF models developed on the R ussian side o f the border, Li e ta l. (2010) showed th a t predictive capacity and perform ance was greatly im proved in R SF m odels th a t included spatial prey covariates fo r red deer, wild boar and roe deer (Li et a l , 2010). In supporting analy­ses w ithin the R ussian portion of the study area, the ungulate-R SF m odel predicted lower h ab ita t quality than an R SF based on ju s t G IS covariates (Supporting In fo rm a­tion, Fig. S3; see also M itchell & H ebblew hite, 2012). Sim ilar analyses done for the critically endangered F a r eastern leopard in Southw estern Prim orye K rai also confirm th a t including spatial prey density tends to predict lower h ab ita t quality than expected ju s t based on land cover-type covariates alone (H ebblewhite et al., 2011). This emphasizes the potential for overestim ating tiger h ab ita t quality in C hina if ungulate prey densities are lower than Russia, and the key role o f reducing poaching on ungulates and increas­ing ungulate densities will play in A m ur tiger recovery in C hina (C hapron et a l , 2008).

O ur m odel averaging approach to understand the hab ita t needs o f recovering carnivores or active restoration o f ca r­nivores (H ayw ard & Sommers, 2009) will help overcome reliance on a single m odeling m ethod. W ith recent advances in sophisticated species d istribution m odeling approaches (e.g. B IO M O D , Thuiller e t a l , 2009), carnivore ecologists will be able to construct robust ensemble models o f up to a dozen different m odeling approaches. M oreover, although our approach identified hab ita t for A m ur tiger under current conditions, a loom ing question for tiger and carn i­vore recovery in general will be the interacting effects o f changing hum an land use and climate change (Carroll, 2007).

Conservation recommendationsR ecent debate has centered on w hether tiger conservation should focus on critical ‘source sites’ (W alston e t a l , 2010) or across w ider landscapes (W ikram anayake et a l , 2011). F o r the A m ur tiger, dependent on the one hand on prey densities for high reproductive rates (C hapron et a l , 2008), and on the other, on large h ab ita t patches (G oodrich e t a l , 2010), bo th are critically needed. O ur results em phasize the im portance

A n im a l C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zo o log ica l S o c ie ty o f London 11

Restoring tiger conservation landscapes Hebblewhite etal.

o f restoring connected hab ita t first to p rom ote natu ra l dis­persal, survival and recolonization of A m ur tigers in N o rth ­east China. The priority should be to increase tiger hab ita t quality in the H unchun and adjacent TCA s, where dispersing tigers from R ussia are regularly appearing (Li et a l , 2006). Tiger hab ita t quality m ust be increased by reducing livestock density and thus, tiger-hum an conflicts (Yu e ta l., 2006), increasing survival rates o f tigers and their ungulate prey th rough rem oval o f snares (Yu e t a l , 2006), and reducing hum an activity th rough regional land-use planning sur­rounding TCA s (M iqueiie et a l , 2005). A lthough the chal­lenges are great, we are encouraged th a t local and national governm ents have recognized TCA s as a basis o f A m ur tiger conservation in C hina (Li et al., 2010). M aintain ing connec­tivity of TCA s w ithin C hina and across the C h ina-R ussian border will also be critical to recovery, and our linkage zone analysis focuses im m ediate conservation atten tion on several key linkages under th reat, b u t ensuring ‘source sites’ where breeding female tigers are secure in N ortheast C hina is a necessary first step tow ard recovery.

Acknowledgem entsFunding was provided for this study by W orld W ide F und for N atu re (W W F) G erm any, U S, U K and N etherlands, Wildlife C onservation Society (WCS) K O R A Switzerland, N ortheast N orm al U niversity, C hina, and the U niversity o f M ontana. We thank participants o f the M ay 2009 C hang­chun w orkshop and m em bers o f the planning com m ittee for valuable assistance and feedback, and constructive com ­m ents on previous versions o f this paper from two anony­m ous reviewers, N athalie Pettorelli and H ugh R obinson.

ReferencesA raujo, M.B. & New, M. (2007). Ensemble forecasting of

species distributions. Trends Ecol. Evol. 22, 42M7.Barry, S. & Elith, J. (2006). E rror and uncertainty in

habitat models. J. Appl. Ecol. 43, 413M23.Basille, M., Callenge, C., M arboutin, E., Anderson, R. &

Gaillard, J.M . (2008). Assessing habitat selection using m ultivariate statistics: some refinements o f the ecological niche factor analysis. Ecol. Modell. 211, 233-240.

Boyce, M.S. & M cDonald, E.E. (1999). Relating popula­tions to habitats using resource selection functions.Trends Ecol. Evol. 14, 268-272.

Boyce, M.S., Vernier, P .R ., Nielsen, S.E. & Sclimiegelow, F.K .A . (2002). Evaluating resource selection functions. Ecol. Modell. 157, 281-300.

Boyce, M.S. & W aller, J.S. (2003). Grizzly bears for the bit- terroot: predicting potential abundance and distribution. Wildl. Soc. Bull. 31, 670-683.

Carroll, C. (2007). Interacting effects o f climate change, landscape conversion, and harvest on carnivore popula­tions at the range margin: m arten and lynx in the N o rth ­ern Appalachians. Conserv. Biol. 21, 1092-1104.

Carroll, C. & Miqueiie, D .G . (2006). Spatial viability analysis o f A m ur tiger Panthera tigris sltaica in the Russian F ar East: the role o f protected areas and landscape m atrix in population persistence. J. Appl. Ecol. 43, 1056-1068.

C hapron, G., Miqueiie, D .G ., Eambert, A., Goodrich, J.M ., Eegendre, S. & Clobert, J. (2008). The im pact on tigers of poaching versus prey depletion. J. Appl. Ecol. 45, 1667-1674.

Chetkiewicz, C.E.B., Clair, St. C.C.S. & Boyce, M.S.(2006). C orridors for conservation: integrating pattern and process. Annu. Rev. Ecol. Evol. Syst. 37:317-342.

Cianfrani, C., Ee Eay, G ., Hirzel, A .H. & Eoy, A. (2010). D o habitat suitability models reliably predict the recovery areas o f threatened species? J. Appl. Ecol. 47, 421M30.

Clevenger, A.P. & W altho, N . (2005). Perform ance indices to identify attributes o f highway crossing structures facilitating movem ent o f large mammals. Biol. Conserv. 121, 453M64.

Colchero, F., Conde, D .A., M anterola, C., Chavez, C., Rivera, A. & Ceballos, G. (2010). Jaguars on the move: modeling movem ent to mitigate fragm entation from road expansion in the M ayan Forest. Anim. Conserv. 14, 158-166.

Dinerstein, E., Eoucks, C., W ikram anayake, E., Ginsberg, J., Sanderson, E., Seidensticker, J., Forrest, J., Bryja, G., Heydlauff, A., Klenzendorf, S., Eeimgruber, P., Mills, J., O ’brien, T.G., Shrestha, M., Simons, R. & Songer, M.(2007). The fate o f wild tigers. Bioscience 57, 508-514.

Goodrich, J., Miqueiie, D .G ., Smirnov, E., Kerley, E.E.,Quigley, H., H ornocker, M. & Uphyrkina, O. (2010). Spatial structure o f A m ur (Siberian) tigers {Panthera tigris altaica) in Sikhote-A lin Biosphere Zapovednik, Russia. J. Mammal. 91, 737-748.

Hall, D .K ., Riggs, G .A., Salomonson, Y.Y., D iGirolamo, N .E., Bayr, K .J. (2002). M ODIS snow-cover products. Remote Sensing o f Environment. 83, 181-194.

Hayward, G.W. & Sommers, M .J. (2009). Reintroduction o f top-order predators. Exeter: Wiley-Blackwell.

Hebblewhite, M., Miqueiie, D .G ., Aramilev, V.V., D iG iro­lamo, N .E., Bayr, K .J., (2002). Predicting potential habitat and population size for reintroduction of the F ar Eastern Eeopard in the Russian F ar East. Biol. Conserv. 144, 2403-2413.

Henry, P., Miqueiie, D., Sugimoto, T., M cCullough, D .R., Caccone, A. & Russello, M.A. (2009). In situ population structure and ex situ representation o f the endangered A m ur tiger. Mol. Ecol. 18, 3173-3184.

Hirzel, A .H ., Hausser, J., Chessel, D. & Perrin, N . (2002). Ecological-niche factor analysis: how to compute habitat- suitability m aps w ithout absence data? Ecology 83, 2027-2036.

Hirzel, A .H ., Hausser, J., Perrin, N ., Eay, G .E. &Braunisch, V. (2007). B iom apper 4.0 in D. O. E. A. E.

12 A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zoo iog ioa i S oo ie ty o f London

Hebblewhite etal. Restoring tiger conservation landscapes

laboratory of conservation biology. University o f Lausanne, Switzerland. U R L http://www2.unii.ch/ biomapper.

Hirzel, A .H., Lay, G .L., Heifer, V., Randin, C. & Guisan, A. (2006). Evaluating the ability o f habitat suitability models to predict species presences. Ecol. Modell. 199, 142-152.

Hosmer, D.W . & Lemeshow, S. (Eds) (2000). Applied logis­tic regression. New York: John Wiley and Sons.

Janin, A., Eena, J., Ray, N ., Deiacourt, C., Aiiemand, P. & Joiy, P. (2009). Assessing landscape connectivity with calibrated cost-distance modelling: predicting comm on toad distribution in a context o f spreading agriculture.J. Appl. Ecol. 46, 833-841.

K aranth , K .U ., Nichols, J.D ., K um ar, N .S., Eink, W.A. & Hines, J.E. (2004). Tigers and their prey: predicting car­nivore densities from prey abundance. Proc. N atl Acad. S e t U.S.A. 101, 4854^858.

Kerley, E.E., Goodrich, J.M ., Miqueiie, D .G ., Smirnov, E.N., Quigley, H.B. & H ornocker, N .G . (2002). Effects o f roads and hum an disturbance on A m ur tigers.Conserv. Biol. 16, 97-108.

Ei, B., Endi, Z., Zhenhua, Z. & Yu, E. (2006). Prelim inary m onitoring o f A m ur tiger population in Jilin H unchun N ational N ature Reserve. Acta Theriol. Sin. 28, 333-341.

Ei, Z., Zimmermann, F., Hebblewhite, M., Purekhovsky,A., Morschel, F., Chunquan, Z. & Miqueiie, D .G .(2010). Study on the potential tiger habitat in the Chang­baishan area, China. Beijing: China Forestry Publishing House.

Einkie, M., C hapron, G., M artyr, D .J., Holden, J. & Eeader-Williams, N . (2006). Assessing the viability of tiger subpopulations in a fragmented landscape. J. Appl. Ecol 43, 576-586.

Fin, C., Berry, P.M ., Dawson, T. & Pearson, R .G . (2005). Selecting thresholds o f occurrence in the prediction of species distributions. Ecography 28, 385-393.

Foiselle, B.A., Howell, C.A., G raham , C.H., Goerck, J.M., Brooks, T., Smith, K .G . & Williams, P.H . (2003). Avoiding pitfalls o f using species distribution models in conservation planning. Conserv. B io l 17, 1591-1600.

Euo, S.J. (2010). The status o f the tiger in China. Cat News Special Issue 10-13.

M anly, B .F.J., M cD onald, E .E., Thomas, D .E., M cDonald, T.E. & Erickson, W.P. (Eds) (2002). Resource selection by animals: statistical analysis and design fo r fie ld studies,2nd edn. Boston: Kiuwer.

Miqueiie, D ., Nikolaev, I., Goodrich, J., Eitvinov, B., Smirnov, E. & Suvorov, E. (2005). Searching for the coexistence recipe: a case study of conflicts between people and tigers in the Russian F a r East. In People and wildlife: conflict o f co-existence?: 305-322. W oodroffe, R., Thirgood, S. & Rabinowitz, A. (Eds). Cambridge: Cambridge University Press.

Miqueiie, D., Smirnov, E., Quigley, H ., H ornocker, M., Nikolaev, I. & M atyushkin, E. (1996). Food habits of A m ur tigers in Sikhote-A lin Zapovednik and the Russian Far East, and implications for conservation. W ildl Res.1, 138-147.

Miqueiie, D .G ., Pikunov, D .G ., Dunishenko, Y .M ., A ram i­lev, Y.Y., Nikolaev, I.G ., Abram ov, Y.K., Smirnov, E.N., Salkina, G ., Seryodkin, I.Y., G aponov, Y.Y., Fomenko, P.Y., Eitvinov, M .N., Kostyria, A.Y., Yudin, Y.G., K or- risko, Y.G. & M urzin, A.A. (2006). A survey of A m ur (Siberian) Tigers in the Russian F a r East, 2004-2005. Wildlife C onservation Society, W orld Wildlife Fund.

Miqueiie, D .G ., Smirnov, E.N., Merrill, T.W., Myslenkov, A .E., Quigley, H .G ., H ornocker, M .G. & Schleyer, B. (1999). H ierarchical spatial analysis o f A m ur tiger rela­tionships to hab itat and prey. In Riding the tiger: tiger conservation in human-dominated landscapes: 71-99. Christie, S., Jackson, P. & Seidensticker, J. (Eds). C am ­bridge: Cambridge University Press.

Mitchell, M.S. & Hebblewhite, M. (2012). Carnivore habitat ecology: integrating theory and application for conserva­tion. In Carnivore ecology and conservation: a handbook o f techniques: 218-255. Powell, R.A. & Boitani, E. (Eds). Oxford: O xford University Press.

M ladenoff, D .J., Sickley, T.A. & Wydeven, A.P. (1999). Predicting gray w olf landscape recolonization: logistic regression models vs. new field data. Ecol. Appl. 9, 3 7 ^ 4 .

Quintero, J.D ., Roca, R., M organ, A., M arthur, A. & Shi, X. (2011). Sm art green infrastructure in tiger range coun­tries: a multi-level approach. Sustainable Development Program , East Asia and Pacific Region, The W orld Bank and the G lobal Tiger Initiative, 80pp.

Running, S.W., N emani, R .R., Heinsch, F.A., Zhao, M.S., Reeves, M., H ashim oto, H. (2004). A continuous satellite-derived measure o f global terrestrial prim ary production. Bioscience 54, 547-560.

Schadt, S., K nauer, F., Kaczensky, P., Reviiia, E.,Wiegand, T. & Trepi, E. (2002). Rule-based assessment of suitable habitat and patch connectivity for the E ura­sian lynx. Ecol A ppl 12, 1469-1483.

Smith, J.E .D ., Ahearn, S.C. & M cdougal, C. (1998). Eand- scape analysis o f tiger distribution and habitat quality in Nepal. Conserv. B io l 12, 1338-1346.

Stephens, P.A., Zaumysiova, O.Y., Miqueiie, D .G ., Mysien- kov, A .I. & Hayward, G .D . (2006). Estimating popula­tion density from indirect sign: track counts and the Form ozov-M aiyshev-Pereieshin formula. Anim. Conserv. 9, 339-348.

Thuiller, W., Eafourcade, B., Engier, R., Araujo, M.B. (2009). B IO M OD - a platform for ensemble forecasting of species distrubutions. Ecography 32, 369-373.

W alston, J., Robinson, J.G ., Bennett, E.E., Breitenmoser, U ., D a Fonseca, G.A.B., Goodrich, J., Gum a, M., H unter, E., Johnson, A., K aranth , K .U ., Eeader- Wiiiiams, N ., M acK innon, K ., Miqueiie, D., Pattanavi-

A n im a i C o nse rva tio n •• (2012) © 2 01 2 The A u th o rs . A n im a l C o nse rva tio n © 201 2 The Z o o iog ioa i S oo ie ty o f London 13

Restoring tiger conservation landscapes Hebblewhite etal.

bool, A., Poole, C., Rabinowitz, A., Smith, J.L.D.,Stokes, E.J., Stuart, S.N., Vongkhamheng, C. & Wibisono, H .T. (2010). Bringing the tiger back from the brink-the six percent solution. P L oS Biol. 8, e 100048 5.

W ikram anayake, E., Dinerstein, E., Seidensticker, J., Eumpkin, S., Pandav, B., Shrestha, M., M ishra, Ei., Ballou, J., Johnsingh, A .J.T., Chestin, I., Sunarto, S., Thiniey, P., Thapa, K ., Jiang, G., Eiagupiiiay, S., Kafiey, Ei., M an Babu Pradhan, N ., Jigme, K., Teak, S., Cutter, P., A bdul Aziz, M. & Than, U. (201 i). A landscape- based conservation strategy to double the wild tiger population. Conserv. Lett. 4, 219-227.

W ikram anayake, E., Mcknight, M., Dinerstein, E., Joshi,A., Gurung, B. & Smith, D. (2004). Designing a conser­vation landscape for tigers in hum an-dom inated environ­ments. Conserv. Biol. 18, 839-844.

Wilson, K .A., W estphal, M .I., Possingham, Ei.P. & Elith, J. (2005). Sensitivity o f conservation planning to different approaches to using predicted species distribution data. Biol. Conserv. 122, 99-112.

Xiaofeng, E., Yi, Q., Diqiang, E., Shirong, E., Xiulei, W., Bo, W. & Chunquan, Z. (2011). H abitat evaluation of wild A m ur tiger {Panthera tigris altaica) and conserva­tion priority setting in N ortheastern China. J. Environ. Manage. 92, 31M2.

Yu, E., Endi, Z., Zhihong, E. & Xiaojie, C. (2006). A m ur tiger {Panthera tigris altaica) predation on livestock in H unchun N ature Reserve, Jilin, China. Acta Theriol. Sin. 26, 213-220.

Zhou, S., Sun, H., Zhang, M., Eu, X., Yang, J. & Ei, E.(2008). Regional distribution and population size fluctua­tion of wild A m ur tiger {Panthera tigris altaica) in H ei­longjiang Province. Acta Theriol. Sin. 28, 165-173.

Zimmermann, F. & Breitenmoser, U. (2007). Potential dis­tribution and population size o f the Eurasian lynx {Lynx lynx) in the Jura M ountains and possible corridors to adjacent ranges. Wildl. Biol. 13, 406M16.

Supporting inform ationA dditional Supporting In fo rm ation m ay be found in theonline version o f this article:

Figure S I. Southern portion o f Prim orski K rai area sur­veyed fo r A m ur tigers in the R ussian F a r E ast showing sam pling un it design used in resource selection function w ith sam pling units (polygons) where tigers were present (in red) o r absent (grey) were treated as a used-unused design to develop R S F models for extrapolation to the Chinese po rtion o f the C hangbaishan study area.Figure S2. (a) Potential tiger h ab ita t predicted by the E N F A m odel in the C hangbaishan landscape in N ortheast C hina and the southern R ussian F a r East. Ail ceils are show n (direct, in terpolated and extrapolated), (b) P re­dicted h ab ita t for the A m ur Tiger from a resource selection function (R SF) m odel in the C hangbaishan landscape in N ortheast C hina and the southern R ussian F a r East. M ajo r cities (> 50 000) and m ajor roads are shown, (c) Potential tiger hab ita t, as predicted by the expert m odel excluding da ta on prey densities. The higher the value (tow ards color green) the better the hab ita t potential.Figure S3. C om parison o f predictions o f the environm ental spatial covariates-oniy R S F m odel (G IS H ab ita t) and the same R SF m odel w ith covariates o f relative density o f the top three prey species for A m ur tigers in the southern portion o f their range in the R ussian F a r East.Table S I. P redictor variables included in the three com ple­m entary A m ur tiger h ab ita t modelling approaches. E nvi­ronm ental N iche F ac to r A nalysis (E N FA ), Resource Selection Function (R SF) m odelling, and the expert- opinion based model.Table S2. C ost allocations of five p redictor variables used in the E xpert M odel: the lower the cost, the higher the value in term s of hab ita t suitability for tigersTable S3. F riction values o f the environm ental variables for A m ur tiger h ab ita t connectivity m odeling based on expert opinion. Values ranging from 1 (easy to cross) to 1000 (impossible to cross). R F E = R ussian F a r East.

Please note: W iley-Blackwell are no t responsible for the content o r functionality o f any supporting m aterials sup­plied by the authors. A ny queries (other th an missing m ate­rial) should be directed to the corresponding au tho r for the article.

14 A n im a l C o nse rva tio n •• (2012) © 201 2 T h e A u th o rs . A n im a l C o nse rva tio n © 201 2 The Zoo iog ioa i S oo ie ty o f London