The geographical range of Amur tigers in the RFE stretches south to north throughout the length of Primorski Krai (province) and into southern Khabarovski Krai for almost 1000 km (Fig. 1). This region, encompassing approximately 300 000 km2, is bounded by China to the west, North Korea to the south and the Sea of Japan to the east. The majority of the region is represented by the Sikhote-Alin mountains, a low (500–800 m a.s.l.) mountain range that parallels the Sea of Japan from Vladivostok in the south to the mouth of the Amur River in the north.
Figure 1. Study area for Amur tiger in the Russian Far East showing protected areas mentioned in the text (boundaries as of 2000) and tiger distribution based on the 1996 full range survey.
Download figure to PowerPoint
Tigers are restricted to forest-covered landscapes, which includes more than 70% of Primorski and southern Khabarovski Krais (210 000 km2). Typical tiger habitats are Korean pine Pinus koraiensis Sieb. & Zucc. broad-leaved forests. The majority of these forests have been selectively logged, and human activities, in association with fire, have resulted in conversion of many low altitude forests to secondary oak Quercus mongolica Fisch. and birch Betula costata Trautv. Betula lanata Regel and other birch species forests (Bogatov et al. 2000). Above 700–800 m, spruce Picea ajanensis Fisch.–fir Abies nephrolepis Trautv. ex Maxim. forests prevail in central Sikhote-Alin. This altitudinal transition zone to predominantly coniferous forest types decreases northwards until, at 47′20″ latitude, coniferous forests occur along the coastline.
The faunal complex of the region is also represented by a mixture of Asian and boreal life forms. The ungulate complex is represented by seven species, with red deer Cervus elaphus L., roe deer Capreolus pygargus Pallas and wild boar Sus scrofa L. being the most common throughout the Sikhote-Alin mountains but rare in higher altitude spruce–fir forests. Sika deer Cervus nippon Temminck are restricted to the southern half of the Sikhote-Alin mountains. Musk deer Mochus moschiferus L. and Manchurian moose Alces alces cameloides Milne-Edwards are associated with the conifer forests and are near the southern limits of their distribution in the central Sikhote-Alin mountains.
Detailed information on tiger ecology (estimates of survival rates, movement and territory size) were derived from radio-telemetry studies conducted in and around Sikhote-Alin biosphere zapovednik (SABZ) (Miquelle, Smirnov & Goodrich 2005), which is located in the central portion of tiger range and includes coastal and inland habitat on both slopes (east and west) of the Sikhote-Alin mountains. Survivorship was estimated using Cox proportional hazards model on 42 radio-collared tigers, and cause of mortality was derived from examination of remains in the field, pathology examinations of samples and (in the case of poaching events) information derived from local people. Home range sizes were derived from 14 and five adult resident females and males, respectively.
We developed RSF models of tiger distribution from data on the occurrence of tracks in snow on routes throughout the tiger range in the RFE surveyed in 1996. Standard Russian survey protocols used to estimate tiger distribution and abundance are based on two-staged track counts in winter. The entirety of tiger habitat (in 1996 134 621 km2) is divided into survey units (averaging 237 km2 in size in 1996; Matyushkin et al. 1999). A single hunter, trapper or conservation officer (all of whom spend weeks at a time in the forest) working within each unit is trained to record tiger tracks encountered over a 3-month period. At the end of this period, in mid-February, one or more survey routes (a total of 1795 in 1996) of at least 10 km are covered. Routes are placed non-randomly in survey units to maximize the probability of encountering tiger tracks. Variability in tiger track density estimates has been shown to increase with decreasing route length (Hayward et al. 2002), and a 10-km minimum was selected to reduce variability in tiger track density but also in acknowledgement of the logistic constraints on winter travel. The locations of tiger tracks were recorded on 1 : 100 000 maps. Although survey effort was poorly estimated in the first stage, it represents a much greater effort (multiple daily routes covered) than the second stage. Using a two-stage sampling design based on double coverage of transects within survey units within a single winter, radio-collared tigers were ‘captured’ 75% of the time in a single survey but 96% in double surveys (D. Miquelle, unpublished data). Thus the two-stage sampling protocol comes close to representing a true presence–absence design.
Habitat data evaluated as potential explanatory variables included both natural characteristics and human influences on the landscape. Habitat variables fell into four classes: vegetation, topographic (latitude-adjusted altitude, slope, transformed aspect (Beers, Dress & Wensel 1966), slope position), climatic (mean annual temperature and precipitation, mean January temperature and precipitation; Hijmans et al. 2004) and human impact (road density, interpolated population density, protected status and habitat effectiveness). Habitat effectiveness is a composite metric for relative mortality risk to large carnivores based on roads and human population (Merrill et al. 1999). Data layers, with the exception of climatic data (Hijmans et al. 2004), were compiled by TIGIS (Pacific Institute of Geography GIS Center, Vladivostok, Russia).
The scale of the input vector data layers was 1 : 500 000, while that of input raster data varied from 100 m (topographic) to 1 km (climatic). First, all data were resampled to 1 km to provide a consistent resolution. Secondly, habitat variables were evaluated at two spatial scales, that of the area within a 100-m buffer surrounding the survey routes, and a landscape scale within a 150-km2 moving window around each route. The moving-window analysis was equivalent to averaging habitat values within a buffer of 7 km, the mean daily travel distance of female tigers in this region. Because moving-window metrics tend to be strongly intercorrelated over differing window sizes at larger scales, the 150-km2 metrics probably also approximate those measured at the scale of a female tiger home range in the region (c. 400 km2; Goodrich et al. 2005a).
A set of potential explanatory variables was developed based on field knowledge, and used to predict the probability of detecting tiger tracks on a transect (a binary response variable). We identified an optimal RSF model based on information criteria (AIC; Akaike 1973) and model interpretability and generality (Taper 2004). The form of the univariate relationship (e.g. linear or quadratic) between tiger occurrence and individual variables was first evaluated with generalized additive models (GAM; Hastie 1993). The initial 52 vegetation types were collapsed into 11 types based on dominant cover type, and then grouped into five types based on similarity of model coefficients. Stepwise analysis using AIC (Venables & Ripley 1997) was then used to identify an optimal multivariate model. We validated the RSF model by assessing how well it predicted track detection in a separate data set of tiger track locations reported during chance encounters by people active in the forest (hunters, forest guards, conservation officers, etc.) collected across the entirety of tiger habitat throughout the winter of 1996 (all-winter data set; Matyushkin et al. 1999).
We assumed that adult tiger survival rates are largely driven by human-related mortality factors (Kerley et al. 2002; Goodrich et al. 2005b) and therefore grouped one set of RSF model variables thought to affect tiger survival (human impact-related variables such as road density, human population and habitat effectiveness). Other variables, including habitat type and climate parameters, are likely to be more closely related to the productivity and prey biomass of an area and thus the habitat potential or productivity in the absence of human influences, which we assumed to be correlated with tiger fecundity (Carroll et al. 2003; Naves et al. 2003). The division of variables in a distribution model into mortality- and non-mortality-related is defensible for tigers and other large carnivores for which most mortality is directly caused by human persecution (Naves et al. 2003; Goodrich et al. 2005b). Prey abundance is a more proximate influence on tiger fecundity than vegetation type and other bioclimatic factors (Karanth et al. 2004). Therefore we compared fecundity classes derived from the RSF model with those derived from models of prey biomass developed from prey track transect data collected as auxiliary data during the course of the 1996 simultaneous surveys for tiger (Matyushkin et al. 1999; C. Carroll & D. Miquelle, unpublished data). Prey data were available as summarized per management unit, rather by individual survey transect. Encounter rates with tracks of the four major prey species (red deer, roe deer, Sika deer and wild boar; Miquelle et al. 1996) were converted to prey density by means of the Formozov equation (Stephens et al. 2006) using mean daily travel distances as documented from the SABZ (Stephens et al. 2006). In order to extrapolate prey densities across the region, we then developed linear regression models of square-root transformed prey densities using the environmental variables listed above. We produced a composite ungulate biomass index as the sum of prey abundance by species multiplied by the mean biomass of adult females of that species (Miquelle, Smirnov & Goodrich 2005) and tiger prey electivity estimates (based on data from Miquelle et al. 1996).
We input the GIS data produced by the RSF models and prey analysis into an SEPM, PATCH, a female-only model designed for studying territorial vertebrates (Schumaker 1998). PATCH links the survival and fecundity of individual animals to GIS data on mortality risk and habitat productivity at the scale of an individual territory (Schumaker 1998). Territories are allocated by intersecting the GIS data with an array of hexagonal cells. The different habitat types in the GIS maps are assigned weights based on the relative levels of fecundity and survival rates expected in those habitat classes. Survival and reproduction rates, derived from published field studies, are then supplied to the model as a population projection matrix (Table 1) (Caswell 2001). The model scales the matrix values based on the mean of the habitat weights within each hexagon, with lower means translating into lower survival rates or reproductive output (Table 1). The continuous values in the RSF-derived GIS data were sliced into 10 equal-area (fecundity) or equal-interval (survival) ranked classes (Fig. 2a,b and Table 1). Habitat rankings were calibrated to demographic values by comparing the mean habitat rankings within intensive demographic study areas (i.e. over the composite home ranges of the SABZ study animals; Goodrich et al. 2005a) with the survival (Goodrich et al. 2005b) and fecundity (Kerley et al. 2003) rates from those same areas, as well as by comparisons with maximum and minimum demographic rates reported from long-term studies in other regions (Smith & McDougal 1991; Smith 1993). Because habitat within the SABZ population boundary showed a rank of 80% of the maximum value in the GIS fecundity model (Fig. 2b) and a rank of 95% of the maximum value in the GIS survival model (Fig. 2a), we set the maximum demographic rates in the base scenario (Table 1) as those reported for the SABZ population (Kerley et al. 2003; Goodrich et al. 2005b) divided by 0·8 and 0·95, respectively. However, we tested the sensitivity of model results to this assumption that SABZ demographic rates were exceeded in other areas of the region. Because a strong positive effect of protected area status on tiger survival and fecundity has been documented in field studies (Miquelle et al. 2005a) but may not be evident in regional-scale RSF models, we analysed the sensitivity of PATCH predictions to setting survival rates in strictly protected areas (zapovedniks) to the maximum ranking, and simultaneously increasing the fecundity ranking in zapovedniks by two ranks and within partially protected areas (zakazniks) by one rank.
Table 1. Demographic parameters used in the PATCH simulations. The scaling of demographic parameters by habitat rank is shown for the adult (year 3+) age class. Cub and subadult parameters also scale by the same proportions
|Home range size||360 km2|
|Maximum dispersal distance|| 54 km|
|First reproduction at age|| 3|
|Maximum female young per female per year|| 0·851|
|Maximum survival (female)|
|Year 2|| 0·747|
|Year 3+|| 0·863|
|Scaling of parameters by habitat rank|
|Adult fecundity||Adult survival|
Figure 2. Results from the resource selection function and least-cost path analysis for tiger in the Russian Far East. The continuous values in the resource selection function output were divided into 10 equal-area (fecundity) or equal-interval (survival) ranked classes. (a) Gradients in tiger survival rates as derived from the habitat effectiveness variable of the resource selection model. Protected areas are shown in crosshatch. (b) Gradients in tiger fecundity rates as derived from the remaining variables of the resource selection model: vegetation type, latitude-adjusted altitude and slope position. (c) Gradients in tiger fecundity rates as derived from prey biomass models developed from prey encounter rates on tiger survey transects. (d) Least-cost path network connecting major protected areas within the tiger range in the Russian Far East. The cost of paths in the network was based on the inverse of habitat effectiveness, thus paths sought to avoid roads and developed areas.
Download figure to PowerPoint
The PATCH simulations incorporate demographic stochasticity with a random number generator. In the case of survival, a uniform random number between zero and one is selected. An individual dies if this number is greater than the scaled survival probabilities produced from the habitat rankings (Table 1). A random number is also selected to force the number of offspring in a year to take on integer values. Environmental stochasticity is incorporated by drawing each year's base population matrix from a randomized set of matrices whose elements were drawn from a normal (fecundity) or beta (survival) distribution. In the scenarios incorporating environmental stochasticity, a coefficient of variation of 25% for fecundity and mortality was used. As no data were available on environmental stochasticity in the region's tiger population, this was a conservative estimate based on values used in previous population viability analyses (PVA) for large felids (Eizirik, Indrusiak & Johnson 2002).
Adult tigers are classified as either territorial or non-territorial (floaters). The movement of territorial individuals is governed by a site fidelity parameter, but floaters must always search for available breeding sites. Movement decisions use a directed random walk that combines varying proportions of randomness, correlation (tendency to continue in the direction of the last step), and attraction to higher quality habitat. However, there is no knowledge of habitat quality beyond the immediately adjacent territories. Floaters do not experience additional mortality risk because of dispersal but rather have yearly mortality rates based on the habitat class they occupy at the end of that year's dispersal path. Although simplified SEPM may be sensitive to variation in parameters such as dispersal distance and behaviour (Ruckelshaus, Hartway & Karieva 1997), PATCH results in most systems appear more sensitive to habitat ranking and demographic parameters. This is because of the presence of large patches with low extinction probability that stabilize metapopulations and reduce their sensitivity to dispersal (Carroll et al. 2004).
In order to evaluate the relative vulnerability of different portions of tiger range to several potential threats and policy options, we created five PATCH scenarios spanning a range from habitat degradation to restoration. These scenarios were as follows. (i) Current conditions. (ii) Increased poaching pressure in the landscape matrix (lands other than strictly or partially protected areas) as a result of relaxed enforcement or increased demand for tiger parts, expressed as a one rank decline in survival rate. (iii) Renewed logging of Korean pine Pinus koraiensis, a high-value timber species whose harvest is currently restricted. Logging would affect tigers primarily through increased road density. Increased logging and resultant increased road density and poaching pressure in the Korean pine forest type was expressed as a one rank decline in survival rate. (iv) Changes in management of zone 1 forests. Russian Forest Service lands are zoned into three categories, which largely dictate what types of exploitation can occur. Zone 1 forests, which are largely exempt from commercial logging, could approximate the current high value of zapovedniks as tiger habitat (Miquelle et al. 2005a) if road closures and increased law enforcement reduced access and poaching pressure. We set the survival rank of the larger blocks of zone 1 forests in Primorski Krai, as well as all zakazniks, to the maximum value (Table 1). This effectively increased the extent of the protected area network from its current 7·17% (21 400 km2) of the region (3·44% strictly protected and 3·73% partially protected) to 13·10% (39 200 km2). (v) A combination of scenarios (ii) and (iv) such that the expanded protected area network is surrounded by a more hostile landscape matrix.
Sensitivity analysis of the PATCH results involved comparison of four scenarios. (i) With (base scenario) and without environmental stochasticity. (ii) Without (base) and with enhanced fecundity and survival within strictly protected areas. (iii) With fecundity rankings derived from the tiger RSF model (base) vs. from models of prey abundance. (iv) Without (base) and with an alternate assumption that the demographic parameters documented in the SABZ population were the maximum rates shown by tigers in the region.
We adapted an approach that sets priority areas for conservation action based on their irreplaceability and vulnerability, in order to minimize the loss of options for conservation planning during an interim period where new reserves are being achieved in some areas while habitat loss is occurring elsewhere (Pressey & Taffs 2001). An area's irreplaceability is the relative contribution it makes to reaching a conservation goal, here species persistence (Pressey & Taffs 2001). We defined irreplaceability in this context as the relative value of an area as source habitat (Carroll et al. 2003). Vulnerability, the likelihood that a site's conservation value will be reduced over time, is measured here as the predicted decline in demographic value (lambda) between low threat (scenario i) and high threat (scenarios ii and iii) scenarios. We created a composite metric consisting of the sum of irreplaceability and vulnerability averaged over both scenario contrasts. This took the form Y= 2λ1– (λ2 + λ3)/2, where λi indicates lambda values from scenario i. This metric was assessed for each of the region's 32 administrative counties (Fig. 3d). Sites with high index values, and hence high irreplaceability and high vulnerability, are the highest priority sites for conservation (Pressey & Taffs 2001), and can be characterized in this context as ‘threatened source habitat’ (Carroll et al. 2003).
Figure 3. Results from a spatially explicit population model of the demography and distribution of Amur tiger. (a) PATCH results from scenario (i) (current conditions). Predicted lambda values for areas with greater than 50% predicted occupancy in Fig. 2a–c are shown as in the legend. Areas with less than 50% predicted occupancy are shown in white. (b) PATCH results from scenario (iii) (increased logging of Korean pine). (c) PATCH results from scenario (v) (expanded protected area system embedded in a landscape matrix experiencing higher poaching pressure). (d) Irreplaceability–vulnerability index values for administrative units within the tiger range in the Russian Far East. The irreplaceability–vulnerability index is derived by comparison of PATCH results from scenario (i) (low threat) with those from scenarios (ii) and (iii) (high threat).
Download figure to PowerPoint
In order to compare priority areas suggested by a SEPM-based analysis with those suggested by a LCP-based analysis, we created a simplified corridor network between the region's seven largest protected areas. This simplified analysis served as a conceptual tool for contrasting general aspects of SEPM-based and LCP-based planning rather than as a detailed conservation plan. The cost surface was based on the inverse of the habitat effectiveness metric used in the RSF and PATCH models. Therefore the LCP algorithm (Ray 2005) sought to link protected areas by routes that minimized encounters with roads and humans. Because human-induced mortality is the primary factor affecting the ability of habitat generalist large carnivores to survive the dispersal event, it is a major input to most LCP analyses for these species (Singleton, Gaines & Lehmkuhl 2004). However, a fully developed LCP analysis would also be likely to include additional variables such as altitude and habitat type (Wikramanayake et al. 2004).