Woodland caribou scientific review to identify critical habitat: chapter 20

Appendix 6.7

Spatial Population Viability Analysis Case Study

Carlos Carroll, Ph.D.



The overarching goal of the national recovery strategy for boreal caribou is to conserve and recover boreal caribou populations and their habitat; that is, prevent extirpation of local populations and maintain or enhance habitat condition to allow these populations to be self-sustaining (EC 2007). Thus the link between population viability and habitat amount and condition is an explicit part of the recovery goal. The question of “How much and what confi guration of habitat is enough to achieve the goal of self-sustaining (viable) populations?” links the process of delineation of critical habitat designation with an analytical approach or suite of methods known as population viability analysis (PVA).

Population viability analysis often involves the use of analytical models to provide quantitative estimates of extinction times and probabilities. Most recent review papers on PVA have judged these metrics less than robust to model and data uncertainty (McCarthy et al. 2003). This type of PVA has also been criticized for limited relevance to real-world conservation planning contexts, due to its emphasis on “small population paradigm” factors (e.g., inbreeding depression) rather than more pressing “declining population paradigm” factors (e.g., habitat loss) (Caughley 1994). Here we use a broader defi nition of PVA that includes a range of methodologies to integrate existing knowledge and models of varying complexity in a structured way. The most valuable output of such PVA is often a better understanding of how trends in species distribution at larger spatial and longer temporal scales are linked to landscape change (development) trends, in a way that is diffi cult to assess without some form of modeling. This allows PVA to be used as a tool to rank alternative management scenarios rather than assign absolute persistence probabilities. There are signifi cant challenges to application of such PVA modeling to boreal caribou, such as the species’ relatively complex local population dynamics. A variety of specifi c analytic methods can be used, with the most appropriate method for boreal caribou depending on factors such as the spatial scale of the question and the nature of available input data. The critical habitat science review has pursued four complementary analytical approaches: environmental niche analysis, rangewide meta-analysis of demography-habitat relationships, non-spatial (heuristic) PVA, and spatial PVA. Here we review initial results from the spatial PVA. The timeline of the critical habitat science review did not allow for completion of a full PVA study. These results instead serve as a proof of concept to assess the relevance of spatial PVA to the boreal caribou critical habitat analysis and recovery process. Although spatial PVA modeling methods are more complex, time-consuming, and require greater levels of input data than other methods, their potential to inform critical habitat designation and planning may justify their use as a complement to other, less data-intensive decision support tools. The major questions addressed in this report include:

  • Adequacy of spatial (habitat) data – What type and quality of spatial data are required for a PVA?
  • Adequacy of demographic data – Is possible to estimate demographic rates in different habitats with a level of accuracy suffi cient for a PVA?
  • Relevance of results – Do results from a spatial PVA inform recovery planning in ways not possible with other methods?
  • Integration with other tools – How are results from a spatial PVA best integrated in a decision support context with results from the other analysis methods used in the critical habitat science review (environmental niche analysis, meta-analysis, and heuristic PVA)?



The boreal caribou critical habitat science review participants chose spatial PVA as one of four methodologies to evaluate during the science review process. In common with the environmental niche analysis, the PVA incorporates spatial data. Spatial models are essential supports to critical habitat analysis in that they provide a broad-scale summary of landscape condition. In contrast to the environmental niche analysis, which addresses habitat primarily at the broadest spatial scale, the spatial PVA focuses on aspects of habitat such as forest type and distance from roads that act at an intermediate spatial scale corresponding to the extent of the local population.

Given this scale of analysis, several approaches of varying levels of complexity could be implemented. The same habitat data used as input to the spatial PVA could also be appropriately used to develop a “static” habitat model (e.g., habitat suitability index (HSI) or resource selection function (RSF)). However, even if such static models are used in place of a dynamic population model, a PVA-type process may be useful to help structure range-wide meta-analysis of habitat data and consideration of how the habitat relationships translate up spatial scales from habitat patches to landscapes and from short-term temporal fl uctuations to long-term trends and persistence thresholds.

The model used here (“HexSim”; (Schumaker et al. 2004; Schumaker in prep.)) is a spatiallyexplicit population model (SEPM; also termed an individual-based model) in which habitat quality affects individuals that are followed as they age, give birth, disperse, and die over time. Individuals may hold exclusive territories or live in social groups. To justify its additional complexity, a spatial PVA must provide insights not possible with a static habitat model. One benefi t of a SEPM is that it can help incorporate landscape processes into conservation planning and thus facilitate evaluation of the effects of alternate future scenarios. Planners must consider multiple future landscape scenarios due to uncertainties as to the effects of climate change, inherent uncertainty in ecosystem processes such as fi re, and alternate options for management processes that transform habitat.

Previous research applying SEPMs to threatened species recovery planning found that the models gave insights beyond those provided by static habitat models because they could assess area and connectivity effects (e.g., inter-population dynamics and source-sink dynamics) that strongly affected persistence of the species considered (Carroll et al. 2006). This may also be the case for boreal caribou. Alternatively, a caribou SEPM could provide similar conclusions to a simpler model such an HSI and thus the simpler model would be preferred. Or a caribou SEPM could potentially offer new insights but require spatial data or demographic parameters that are largely unavailable. Each of these three outcomes is likely true in different regions, and a case study such as the one described here can help planners assess when and where SEPM are an appropriate decision support tool. Even if the data in a particular portion of caribou range are inadequate for deriving SEPM-based predictions regarding quantitative persistence thresholds, SEPM may still be useful in a heuristic sense in offering insights as to emergent processes and effects of landscape condition and structure on caribou persistence.

Caribou SEPM can be expected to be more complex than those for species such as the spotted owl, where individuals defend exclusive territories. Because boreal caribou occur in social groups, local population dynamics should be added to the SEPM. Movement between seasonal habitats should be added to the model for local populations where this occurs. In addition, multi-species SEPM that can capture the interaction between predators and caribou, and indirectly with alternate prey species such as moose, should be possible and may reveal important insights. However, it is important to keep in mind a key guideline: what is the simplest model that effectively supports conservation planning, and what real-world complexities can be ignored in the model without qualitatively compromising results in terms of the questions at hand?

The spatial scale of the case studies presented here was determined somewhat opportunistically by the extent of the available spatial habitat data. Ideally, as was the case here, the spatial data used would encompass the larger landscape, rather than only areas currently occupied by caribou. This extent allows addressing questions such as “How does habitat condition in the larger landscape support or not support caribou occurrence?” But unlike methods that assess summary statistics on aggregate habitat amount within a local population’s range (e.g., the proportion of the landscape within a set buffer distance from roads), a SEPM also focuses on fi ner-scale habitat pattern and composition. At this scale, the model addresses “How does the arrangement of habitat patches within the extent of a local population infl uence its persistence and demography?”, e.g., by infl uencing withinrange movement and consequent exposure to predation.


Relationship with other components of science review

The four components of the critical habitat science review form a spatial and analytical hierarchy of methods. Their output shows less generality and more complexity (or “biological realism”) as one descends the hierarchy. Environmental niche analysis and range-wide meta-analysis can be seen as top-level methods, followed by the heuristic PVA, and fi nally the spatial PVA. Results from top-level analyses reveal overarching constraints on processes examined at lower levels. This perspective allows a synthesis of the four components. Lowerlevel results suggest factors missing from the top-level analyses, and in turn the top-level analyses suggest the extent to which conclusions from e.g., the spatial PVA results may lack generality to some portions of range.

Environmental niche analysis (ENA) characterizes the distribution of boreal caribou by examining which abiotic factors (climate and topography) characterize the distribution of observed locations. These models may be especially relevant in predicting potential effects of climate change. In a second stage of ENA, broad-scale biotic variables (land cover and human impact levels) are added to further refi ne the models. However, these variables, because they are the lowest common denominator of detail available range-wide, lack the fi ne-scale habitat data possible in the spatial PVA. The second range-wide approach is a meta-analysis of relationships between demography and habitat. Both of these approaches, in contrast to the spatial PVA, can produce broadly general conclusions as to what abiotic and biotic conditions permit boreal caribou occurrence and persistence. However, neither approaches are mechanistic, in that they do not address the biotic mechanisms by which e.g., climate limits distribution. The heuristic PVA, in contrast, uses non-spatial models to assess how population persistence is affected by aspects of boreal caribou life history and population structure (e.g., age structure, age-specifi c survival and fecundity, environmental stochasticity, breeding structure, and density dependence). Because such a non-spatial PVA has far fewer parameters and computational demands than a SEPM, the heuristic PVA can more exhaustively explore the plausible parameter space for population dynamics and assess sensitivity of model results to chosen parameters. The spatial PVA explores only a subset of this parameter space but adds consideration of landscape structure and individual movement.

The spatial PVA is linked to the meta-analysis component, in that results of the metaanalysis can be used to inform, and to some extent validate, PVA results. The PVA can help in interpreting results of the meta-analysis in that the PVA may offer heuristic insights as to the mechanisms by which the ability of an area to support caribou scales up spatially from the patch to landscape. Additionally, spatial PVA tools allow simulation of longer-term trends and scenarios to extrapolate the relationships drawn from the meta-analysis to future landscapes.

Comparison of the heuristic and spatial PVA results helps assess 1) to what degree the spatial PVA model’s behaviour is an artefact of particular assumptions as to parameters, 2) whether spatial effects produce qualitatively different results in terms of predictions of population persistence. An integrated assessment using the four approaches might begin with general conclusions as to what climatic conditions and broad-scale habitat characteristics are associated with boreal caribou occurrence (ENA) and persistence (meta-analysis), and refi ne these conclusions by assessment of what life history characteristics (heuristic PVA) and spatial population dynamics (minimum area requirements or dispersal limitation) may explain these patterns and further limit distribution and persistence.



Spatially-explicit population models (SEPM), like static HSI models, use input data on habitat factors that affect survival and fecundity of the species of concern. But SEPM then integrate additional information on characteristics such as demographic rates and dispersal behaviour. For example, social carnivores often require larger territories than solitary species of similar size, and may thus be more vulnerable to landscape fragmentation in a SEPM (Carroll et al. 2006). Unlike a simpler HSI model, a SEPM can provide insights on the effects of population size and connectivity on viability and identifying the locations of population sources and sinks.

HexSim, the SEPM used here, links the survival and fecundity of individual animals or groups to GIS data on mortality risk and habitat productivity (Schumaker et al. 2004, Schumaker in prep.). Individual territories or group ranges 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 expected in those habitat classes. Base survival and reproductive rates derived from published fi eld studies, are then supplied to the model as a population projection matrix. The model scales these base matrix values based on the mean of the habitat weights within each hexagon, with lower means translating into lower survival rates or reproductive output. Each individual in the population is tracked through a yearly cycle of survival, fecundity, and dispersal events. Environmental stochasticity can be incorporated by drawing each year’s base population matrix from a randomized set of matrices whose elements were drawn from a beta (survival) or normal (fecundity) distribution. Adult organisms are classifi ed as either territorial or fl oaters. Floaters must always search for available breeding sites or existing groups to join. Movement decisions can be parameterized in a variety of ways, with varying proportions of randomness, correlation (tendency to continue in the direction of the last step), and attraction to higher quality habitat (Schumaker et al. 2004). Because it is diffi cult to parameterize movement rules directly from fi eld data (but see Fryxell and Shuter 2008), it is important to assess the sensitivity of model results to a range of plausible movement parameters.

SEPM can produce a wide range of output in the form of both spatial data (maps) and summary statistics (e.g., population time series). This output can be used to assess an area in terms of the probability of occurrence of the species (similar to the output of a HSI model), the area’s demographic role (source or sink) as well as give population-level predictions of long-term persistence or extirpation.

Because absolute estimates of risk from a SEPM are suspect due to uncertainty in data and models, SEPM output should instead be used to rank candidate recovery strategies in terms of viability (or extinction risk) and distribution (range expansion or contraction).


Spatial Data

Two case study areas were selected opportunistically for the SEPM analysis based on data availability. The fi rst study area is located in northeastern Alberta on lands with forest tenure held by Alberta Pacifi c Forest Industries (ALPAC). This area encompasses the extent of the ESAR (eastside of Athabasca River) and WSAR (westside of Athabasca River) caribou herds (local populations). The area is predominantly a mixture of peatland and upland habitats with the predominant resource industries being timber extraction and oil and gas development. The second case study area is located in southeastern Manitoba, and encompasses the extent of the Owl Lake herd. The predominant resource industry in this area is timber extraction. Data for this study area was provided by the Eastern Manitoba Woodland Caribou Advisory Committee (EMWCAC). While the two case study areas obviously do not represent the full spectrum of landscape contexts encountered across the range of boreal caribou, they do show contrasts in habitat use and type of threats to population persistence. For example, a large expansion of linear features related to the energy sector is ongoing in the Alberta study area. The Manitoba case study allows examination of effects of timber harvest scenarios (as well as lower rates of expansion of linear features) on population persistence. Use of two contrasting case study areas allows more general assessment of what minimum level of habitat data (vegetation and linear features) is required for SEPM analysis.

In Alberta, data from the Alberta Vegetation Inventory (AVI) was classifi ed into high, medium, and low quality caribou habitat. High quality habitat was defi ned as pure stands of black spruce, pure stands of larch, and mixed stands of black spruce and larch. Medium quality habitat was defi ned as black spruce and larch dominated-stands mixed with tree species other than larch and black spruce.

Low quality habitat was defi ned as all remaining areas. A second habitat layer was created from data on linear features. Areas within 250m of a roads or seismic lines were considered reduced in habitat suitability based on previous research (Dyer et al. 1999). The spatial data for the Manitoba study area was received later than the Alberta data and time constraints permitted only initial evaluation of its suitability for SEPM modeling. It is anticipated that spatial data predicting summer and winter habitat suitability (HSI model) will be the key input to the SEPM. Data on linear features (roads and transmission) lines are also available and may be buffered as in the Alberta case study.



Survival rates were parameterized for the Alberta study area based on an expert workshop held with a subset of the Science Advisory Group (SAG) in Vancouver, BC, February 11-12, 2008. Rates were set to vary by habitat type and age class. Survival rates in high and medium quality habitat varied based on the proportion, averaged over a 10 km 2 moving window, of the area within 250m of linear disturbance. The equation for adult annual survival rate [S a] in high and medium quality habitat was S a = 0.98 – (proportion within buffer * 23)(Figure 1). The equation for calf annual survival rate [S c] in high and medium quality habitat was: S c = 0.50 – (proportion within buffer * 40). Adult annual survival rate in poor quality (upland) habitat was set to 0.65 irrespective of proportion of linear disturbance buffer. Calf annual survival rate in poor quality (upland) habitat was set to zero irrespective of proportion of linear disturbance buffer. Fecundity rate was set constant across habitats as 0.5 female offspring/female/year. A range of values for the parameter for maximum movement distance have been assessed. The base value used in the simulations shown here is 112 km (total path length, not total net displacement). All of the parameters used above would be subject to further review, revision, and sensitivity analysis in the course of a complete PVA study in order to produce a credible decision support tool.



This initial report focuses on qualitative patterns in the results because it is expected that quantitative predictions would change as initial exploratory simulations are subject to review and sensitivity analysis in a complete PVA study. In the initial simulations, areas of high predicted occupancy are relatively widespread across the Alberta study area when linear disturbance effects are not considered (Figure 2a). This may be conceptualized as representing a landscape state closer to historic (pre-development) condition. These areas are much reduced in extent under the simulations where survival rates are affected by linear disturbance buffer zones (Figure 2b). This may be conceptualized as assessing the current landscape condition. The ESAR herd is affected more heavily by linear disturbance than is the WSAR herd. According to our data, 63.0% of the ESAR range is within 250 meters of linear disturbance, versus 44.93% of the WSAR range. A comparison between the HexSim simulations with and without linear disturbance shows a decline in occupancy probability of 76.7% for the ESAR herd, versus 58.7% for the WSAR herd. Although neither local population has a high likelihood of extirpation (given no further habitat loss) in these initial simulations, more realistic assessment of persistence probabilities should await simulations that better incorporate group dynamics.

Occupancy rates shown above are drawn from the fi nal decade of 200 year simulations, averaged over 10 simulation runs. Although the simulations are 200 years in length, the landscape does not change in the current analysis. Therefore, predictions show the equilibrium “carrying capacity” of the current landscape, not the future persistence probabilities of the population given landscape change. Both stochastic landscape change, such as driven by fi re, and deterministic habitat trends, such as increases in linear disturbance, would alter current equilibrium carrying capacity. These aspects could be explored in future simulations.

Despite a static landscape, population levels show wide variation around carrying capacity. A plot of fi ve population time series drawn from the Alberta simulations with linear disturbance (Figure 2b) is shown in Figure 3. Relatively large population fl uctuations (~20%) over periods of several decades are evident although the longer-term trend is stable. These fl uctuations are driven by both demographic stochasticity and habitat pattern. The potential of caribou life history structure and demographic stochasticity in relative small populations to cause longterm fl uctuations should be evident in a non-spatial (heuristic) PVA model. However, a spatial model such as a SEPM allows habitat fragmentation and dispersal limitation to accentuate small population effects and increase the magnitude of fl uctuations. The larger population inhabiting the “historic” landscape (Figure 2a) shows fl uctuations of smaller magnitude due to both larger population size and lower levels of landscape fragmentation. The model output emphasizes that it is inherently challenging to interpret data from population monitoring programs for long-lived vertebrates, and SEPM simulations could be instructive in designing monitoring programs for more intact landscapes. However, deterministic habitat changes in the Alberta study area over the short-term will likely swamp the effects of demographic stochasticity.

Although HexSim simulations for the Manitoba study area were not possible within the timeframe of this study, the input habitat layers appear suitable for use in HexSim simulations. Figure 4 shows predictions from the EMWCAC HSI model (averaged over 100 km2 moving window) for the Manitoba study area for a) caribou summer habitat, and b) winter habitat, overlaid with linear features. Although HexSim does allow habitat value to change seasonally, there is relatively low contrast between winter and summer HSI values (correlation = 0.944). Although here the HSI values are averaged over a moving window to graphically display large-scale landscape pattern, the unaltered HSI values would be used as input to HexSim. Although density of linear features is much lower than in the Alberta study area, there is enough separation between blocks of high-quality habitat to suggest that a spatial model that incorporates effects of landscape structure may be informative.



The HexSim model has been previously used in population viability analyses for species where individuals hold exclusive territories (Carroll et al. 2003, Schumaker et al. 2004). Boreal caribou are the fi rst species with group, rather than individual, movement dynamics to which HexSim has been applied. The complexity of adapting the HexSim model to caribou life history and group dynamics has slowed initial progress in developing realistic simulations. However, despite these challenges, further effort invested in model development with HexSim is worthwhile due to the potential for HexSim to provide unique insights into the relationship between habitat and viability of boreal caribou populations.

Concurrently with the national critical habitat science review process, a spatial PVA of Ontarioboreal caribou populations has been developed (Fryxell and Shuter 2008). This work extends previous caribou simulation models (e.g., Lessard 2005) in several areas, notably by parameterizing movement paths from statistical analysis of detailed movement data rather than by conceptual models (e.g., attraction to high quality habitat). The model of Fryxell and Shuter (2008) is not fully spatial or individual-based as demographic rates experienced by caribou are based on an analytical wolf-moose-caribou predator-prey model. The model is highly suited for exploration of the general types of demographic parameters and landscape conditions that support caribou persistence and thus falls into an intermediate level of complexity between the heuristic non-spatial PVA and the HexSim model. In contrast, the strengths of the HexSim model are that it is fully individual-based, and thus can evaluate relationships that emerge from spatial interactions between caribou, their predators (e.g., wolves), and alternate prey species (e.g., moose). A “canned” software application such as HexSim inevitably lacks the fl exibility of a program developed specifi cally for a single species, but as a consequence offers the potential for greater standardization and comparability between study areas and between species than is possible with a custom-built program such as used in Fryxell and Shuter (2008).

Although a conclusive evaluation of the potential for SEPM as a decision support tool in the boreal caribou conservation planning process is not possible in this report, initial results do shed light on the four questions outlined in the introduction (adequacy of habitat and demography data, relevance compared to and integration with results from other methods). The spatial (habitat) data from the two case study areas appear adequate for conducting PVA simulations. However, although the habitat suitability model based on vegetation type and linear features generally matches observed caribou distribution in the Alberta study area, there are contrasts in some areas (high habitat quality with no herds observed) that needed to be further evaluated. The demographic data available for the Alberta study area also appear adequate for HexSim parameterization, as estimates of adult and calf survival by major habitat class in disturbed and undisturbed habitats can be derived from fi eld data. Suggested methods for integrating spatial PVA results with those from the environmental niche analysis, meta-analysis, and heuristic PVA have been described above. Although it is not yet possible to conclusively evaluate whether SEPM tools will inform recovery planning in ways not possible with other methods, the potential benefi ts justify further model development as described above.

The boreal caribou conservation planning process has at least three stages: 1) the nowcompleted critical habitat science review, 2) assessment of what constitutes effective protection, to be completed over the coming months, and 3) longer-term conservation planning efforts at the provincial and federal level. In the shorter term of the fi rst two stages of planning, it seems clear based on the challenges encountered so far in parameterizing the caribou HexSim model that the SEPM approach is best developed as a heuristic tool for illuminating area and connectivity effects in representative case study areas. This is due to limitations on available habitat data, but also as a strategy to concentrate effort on refi nement of the SEPM model before application to a large number of study areas. Although initial predictions can be developed from a SEPM at a relatively early stage in the modeling process, they should not be used in a decision support context until exhaustive sensitivity analysis has been completed. In the interim, static habitat (HSI or RSF) models (e.g., Sorenson et al. 2008) should be developed and used to track amount and quality of habitat at local and range-wide scales, and perhaps refi ned through consideration of landscape structure (core area size, etc.) in addition to habitat amount. These static models are a foundation for and complementary to SEPM model development.

Over the longer term (stage 3), SEPM seems a promising approach for addressing issues that have arisen during the critical habitat science review. This is because SEPM output directly addresses the relative risk to population persistence of alternate conservation strategies and thus what constitutes effective protection. By evaluating persistence under scenarios where habitat is maintained, enhanced, or decreased, SEPM output supports placing populations within a framework of range adequacy and resiliency as developed in the critical habitat science review process. SEPM are also currently the best tool for rigorously assessing the importance of intra- and interpopulation connectivity for persistence of boreal caribou, as in cases where large-scale industrial development may fragment habitat of formerly continuous populations.

The next steps in SEPM development for the two case study areas described here fall into several categories. Initially, the focus will be on parameter refi nement and sensitivity analysis under the current static landscape. Availability of seasonal HSI models as in Manitoba will allow the SEPM to incorporate seasonal ranges and movement between them. More complex population dynamics (e.g., Allee effects) will be incorporated in the simulations. Once a satisfactory parameter set for current landscapes has been developed, simulations will incorporate future scenarios, including threats from development and climate change, and simulation of landscape dynamics due to forest succession and fi re. The value of the SEPM analysis will be enhanced by continued interaction and integration of the spatial PVA with the other three facets of the science review (environmental niche analysis, meta-analysis, and non-spatial PVA).

Report a problem or mistake on this page
Please select all that apply:

Thank you for your help!

You will not receive a reply. For enquiries, contact us.

Date modified: