Population viability analyses in COSEWIC status reports
Approved by COSEWIC in April 2010
The assessment process for determining a wildlife species status under COSEWIC allows the use of quantitative analyses. This is referred to IUCN Red List criterion E (IUCN Standards and Petitions Working Group 2008). This criterion can be used where a robust estimation of extinction risk can be calculated, for example, a forest dependent endemic species whose habitat is expected to be clear-cut within the next 20 years (Mace et al. 2008). However, the main quantitative analysis available for species status assessment is Population Viability Analysis (PVA). Broadly defined, PVA is the use of quantitative methods to predict the likely future status of a population or collection of populations of conservation concerns (Morris and Doak 2002). Within that definition, "future status" typically refers to whether the population (or total number of individuals across all populations) will be above some minimum size at different future times. Assigning concrete numbers to measures of future status is the aim of PVA.
As of 2009, criterion E has been used in a very small proportion of all COSEWIC assessments. However, because monitoring programs are likely to be put in place for different species at risk in the near future - through the implementation of legally required recovery strategies - PVAs are likely to become more commonly used, especially for reassessment. A study of the use of PVA in recovery plans for endangered species in the US concludes: "Even if insufficient data now exist to use PVA for a particular threatened or endangered species, routine monitoring could make the construction and parameterization of population models feasible in the future. In many cases, population viability analysts may be able to improve the appropriateness of monitoring schemes for modeling purposes without significantly increasing the cost of monitoring" (Morris et al. 2002).
The Need for Standards
For the results of PVA to inform the (re-) assessment of species, they need to be well communicated to COSEWIC members. The Guidelines (Standards and Petitions Working Group 2006) for using the IUCN Red list categories and criteria (versions 6.2, Dec. 2006) contains a clear and well-documented set of guidelines to apply criterion E. Under the section on "Documentation requirements", the Guidelines state:
- Any [...] assessment that relies on Criterion E should include:
- a document that describes the quantitative methods used, and
- all the data files that were used in the analysis.
The document and accompanying information should include enough detail to allow an evaluator to reconstruct the methods used and the results obtained.
- The documentation should include:
- a list of assumptions of the analysis
- explanations and justifications for these assumptions.
- documentation of the uncertainties in the data
All data used in estimation should be either referenced to a publication that is available in the public domain, or else be included with the listing documentation.
- Methods used in estimating model parameters and in incorporating uncertainties should be described in detail. Time units used for different model parameters and components should be consistent; the periods over which parameters are estimated should be specified.
Overview of PVA models
One of the challenge for reporting results of PVAs so that they can inform species status assessment is that the types of quantitative methods that can be called PVA are varied, both in terms of the complexity of the underlying models and the quantity of data needed to parameterize them (Burgman et al. 1993, Morris et al. 1999, Beissinger and McCullough 2002). Recent texts describe four types of PVA (Morris and Doak 2002; Lande et al. 2003): unstructured population models, structured population models, metapopulation models, and spatially-explicit population models.
1) Unstructured population models
The simplest class of PVA uses time series data on population size to parameterize a basic stochastic exponential growth model with year-to-year variability in the growth rate, and no density-dependence. This type of PVA requires data on both current population size and trends in population size over time, but does not require age- or stage-structured data or spatially explicit information. Note that any age or stage-structured model will converge to the stochastic exponential model, which is why that model is so fundamental.
One property of the basic density-independent growth model is that it can be approximated by a diffusion equation (Dennis et al. 1991). This approximation opens a toolbox of parameterization methods for linear models with normal error. It also provides analytical estimates of passage probabilities, i.e., the probability of crossing a particular threshold within a given time frame, for example quasi-extinction estimates to some critical population size >1 (Holmes 2004).
The diffusion model approximates many types of stochastic age structured population processes, as seen both from simulated and real data (Lande and Orzack 1988, Dennis et al. 1991, Holmes and Fagan 2002; Holmes 2004). There are, however, well-known cases where the diffusion approximation performs poorly, such as when year-to-year variability is high and populations are small (Ludwig 1999), when demographic stochasticity introduces strong nonlinearities (Wilcox and Possingham 2002), and when density dependence is extreme (Sabo et al. 2004).
There are at least four different maximum likelihood methods for estimating the parameters of the diffusion approximation (DA) model: the regression methods (Dennis et al. 1991), the runsum-slope methods (Holmes and Fagan 2002, Holmes 2001), the Kalman filters (Lindley 2003) and the Restricted Maximum Likelihood method (Staples et al. 2004).
The regression method assumes that all the observed year-to-year variability in the population growth rate is due to environmental variation. The other methods assume that some of the variability comes from other sources, such as observation errors, error due to age-structure cycling, etc. By estimating the parameters of a state-space model in which the true size of the population remains unknown, the other methods aim at estimating the variance in population growth due solely to environmental variation, which is critical in PVA (Holmes 2004).
One of the strengths of DA methods is that the statistical distributions of the estimated parameters are known. As a result, the uncertainty in the estimated risks can be calculated. This is often not the case for other PVA approaches, such as matrix models or individual-based simulations, where uncertainty in the estimated model parameters is often poorly known (Holmes 2004).
Variants of the basic stochastic growth model can incorporate density-dependence, autocorrelation in the environment, and large variation in population growth rates (see Morris and Doak 2002). With these more complex unstructured models, viability metrics are obtained by simulating a large number (1000 to 10,000) of population trajectories.
Unstructured stochastic growth models can use the same data used for assessing species under criteria A (declining populations). However, they have been designed to work with count data, so when the population time series are based on an abundance index (ex. catch per unit effort), then either a biologically significant population threshold appropriate for that index must be specified, or count data must be corrected for a unique catch effort.
Examples of PVAs using the DA methods are; for bird species: Dennis et al. (1991), Stacey and Taper (1992), Middleton and Nisbet (1997), Engen and Saether (2000); for mammal species: Dennis et al. (1991), Gerber et al. (1999), Oli et al. (2001); for reptile species: Snover and Heppel (2009); for insect species: Schultz and Hammond (2003).
2) Structured population models
Structured population models typically use life tables or projection matrix models, which track changes in the numbers of individuals in different stages (e.g., age or size categories) in a population (Morris et al. 2002). Stochastic population projections with matrices use a given description of the stochastic environment, such as independent and identically distributed sequences. These sequences can be used to select whole matrices, matrix elements, or components of matrix elements from either finite sets of observed values or from parametric distributions (Caswell 2001).
Matrix models can incorporate any of demographic stochasticity, catastrophes or bonanzas, density-dependence, spatial structure, and can be used to simulate interventions such as harvesting or population augmentation (reintroduction, head-starting, etc.). They can also allow more detailed analysis of critical life stages or demographic processes that are potential targets for management (Caswell 2001). They require data on stage-specific fecundity and mortality rates, and the current stage structure of the population to be used to predict viability.
Examples of PVAs using projection matrices are; for bird species: Lande (1988), Beissinger (1995); for reptile species: Crouse et al. (1987); for plant species : Menges (1990), Nantel et al. (1996), Menges and Dolan (1998), Gross et al. (1998), Bell et al. (2003), Garcia (2003); for fish species : Kareiva et al. (2000), Legault (2005).
Stochastic population projections can also be done using individual-based simulations. These can allow the inclusion of genetic information (such as pedigree) and be used also to model the effect of different reproduction systems. Examples of PVAs using individual-based simulations, for plant species, are Schwartz et al. (2000), Kirchner et al. (2006). The software VORTEX is a popular PVA modelling tool that uses individual-based simulations (Miller and Lacy 2005).
3) Metapopulation models
"Metapopulation" PVAs follows the fates of multiple subpopulations and attempt to determine whether the rate of establishment of new subpopulations through colonization is sufficiently high to counter the extinction of subpopulations, thus allowing the entire metapopulation to persist. Such PVAs require information on the number of subpopulations, trends in the number of subpopulations or the rate of subpopulation extinction, and the colonization rate, typically as reflected in patterns of dispersal. An example of a metapopulation PVA for a butterfly species is Hanski et al. (1996).
4) Spatially explicit population models
"Spatially explicit" PVAs is the most complex and data intensive type. It typically involves simulating the behaviour of individual organisms on detailed landscapes upon which the sizes and locations of suitable habitat patches are mapped. In addition to requiring information about birth and death rates of individuals within each patch, and their movement patterns, this type of PVA also requires data on the degree of isolation and fragmentation of suitable habitat patches. Examples of spatial PVAs that use RAMAS-GIS as the modeling software, for various taxa, can be found in Akçakaya et al. (2004). Note that the software VORTEX can also be used for spatially explicit population viability analysis (Miller and Lacy 2005).
Sensitivity analyses are processes to measure how sensitive or changeable is an indicator of population fate, such as growth rate or extinction risk, to particular changes in the model's input parameters, such as individual vital rates. Sensitivity analyses are an important component of PVAs because they help, among other things, to identify input parameters that are more influential (Burgman et al. 1993, McCarthy et al. 1995, Caswell 2001). This is critical as uncertainties associated with the more influential parameters can contribute to a large portion of the uncertainty of the model's predictions.
The documentation requirements related to Criterion E in the IUCN guidelines do not mention sensitivity analyses. However, the results of a PVA will be more transparent, likely easier to interpret, and therefore more useful for status assessment when presented to COSEWIC with a sensitivity analysis that covers, at least, the most uncertain input parameters.
A large array of mathematical methods, either analytical or statistical, can be used for measuring parameter influence on the predictions of deterministic and stochastic population models. These include elasticity analysis, conventional and relative sensitivity analyses, life table response experiments, life table simulation analysis, retrospective simulation to evaluate responses to alternative future scenarios, and the use of standardized regression coefficients or other statistics to measure relative influence (Burgman et al. 1993, de Kroon et al. 2000, Cross and Beissinger 2001, Rustigian et al. 2003, Fieberg and Jenkins 2005, Curtis and Naujokaitis-Lewis 2008). Among those methods, the logistic regression approach is well established as an acceptable methodology for sensitivity analysis of PVA models (Curtis and Naujokaitis-Lewis 2008).
Check-list for reporting PVA and population projections in status reports.
In view of the diversity of models used in PVA, report writers who want to present PVA in status report should consider the following check-list of input and output. Note that the input parameters and details of the model structure can be summarized in a set of tables that could be appended to report.
- Nature of the data collected and how they were collected and analysed.
- Type of model used: non-structured or structured by age, stage, sex, subpopulations (patches); individual-based or matrix-based; model of density-dependence and how it has been parameterized.
- If non-structured: time span and frequency of counts (weighted or partial); method used for estimating parameters of the diffusion approximation or other models (ex. Kalman filter).
- If structured: Random matrices (RM) or vital rate simulations (VRS). If VRS, method(s) for estimating (and correcting) vital rates' means and variances. Table of vital rates estimates (mean and variance); probability distribution(s) used to generate random vital rates in stochastic projections. Method used to simulate within-yr and between-yrs correlation (if any). Definition of projection matrix (for vital rate simulations). For both RM and VRS: method (Monte Carlo) used for simulating demographic stochasticity (if simulated).
- Starting number of individuals (per class for structured models)
- Extinction threshold (with weights per class for structured models)
- Time span of projection into the future 1
- Number of runs
- Method used to simulate perturbations (population augmentation or harvesting), if any.
- Software package(s) used to build and run models.
- Stochastic population growth rate (with confidence intervals)
- Graph of the quasi-Extinction time cumulative distribution function (CDF) 2, with confidence intervals (if possible)
- Sensitivity analysis
Akçakaya, H. R., M. A. Burgman, O. Kindvall, C.C. Wood, P. Sjögren-Gulve, J.S. Hatfield, M.M. McCarthy. 2004. Species Conservation and Management: Case studies, Oxford University Press, New York, USA
Beissinger, S. 1995. Modeling extinction in a periodic environment: everglades water levels and Snail Kite population viability. Ecological Applications 5:618-631.
Beissinger, S., and D. McCullough, editors. 2002. Population viability analysis. University of Chicago Press, Chicago, Illinois, USA.
Bell T. J., Bowles M.L. and A.K. McEachern. 2003. Projecting the success of plant population restoration with viability analysis. In C.A. Brigham and M.W. Schwartz (Eds). Population Viability in Plants. Springer-Verlag, Berlin Heidelberg.
Burgman, M. A., S. Ferson, and H. R. Akçakaya. 1993. Risk assessment in conservation biology. Chapman and Hall, London, UK.
Caswell, H. 2001. Matrix population models: construction, analysis and interpretation. Sinauer, Sunderland, Massachusetts, USA.
Cross, P. C., and S. R. Beissinger. 2001. Using logistic regression to analyze the sensitivity of PVA models: a comparison of methods based on African wild dog models. Conservation Biology 15:1335-1346.
Crouse, D. T., L. B. Crowder, and H. Caswell. 1987. A stage-based population model for loggerhead sea turtles and implications for conservation. Ecology 68: 1412-1423.
Curtis, J. M. R. and I. Naujokaitis-Lewis. 2008. Sensitivity of population viability to spatial and nonspatial parameters using GRIP. Ecological Applications 18: 1002-1013.
de Kroon, J., J. van Groenendael, and J. Ehrlen. 2000. Elasticities: a review of methods and model limitations. Ecology 81: 607-618.
Dennis, B., P. L. Munholland, and J. M. Scott. 1991. Estimation of growth and extinction parameters for endangered species. Ecological Monographs 61:115-143.
Engen, S. and B. Saether. 2000. Predicting the time to quasi-extinction for populations far below their carrying capacity. Journal of Theoretical Biology 205:649-658.
Fieberg, J., and K. J. Jenkins. 2005. Assessing uncertainty in ecological systems using global sensitivity analyses: a case example of simulated wolf reintroduction effects on elk. Ecological Modelling 187:259-280.
Garcia, M.B. 2003. Demographic viability of a relict population of the critically endangered plant Borderea chouardii. Conservation Biology 17: 1672-1680.
Gerber, L., D. DeMaster, and P. Kareiva. 1999. Grey whales and the value of monitoring data in implementing the U.S. Endangered Species Act. Conservation Biology 13:1215-1219.
Gross, K.G., J.R. Lockwood, C. Frost, W.F. Morris. 1998. Modeling controlled burning and trampling reduction for conservation of Hudsonia Montana. Conservation Biology 12: 1291-1302.
Hanski, I., A. Moilanen, T. Pakkala, and M. Kuussaari. 1996. The quantitative incidence function model and persistence of an endangered butterfly metapopulation. Conservation Biology 10:578-590.
Holmes, E. E. 2001. Estimating risks in declining populations with poor data. Proceedings of the National Academy of Sciences (USA) 98:5072-5077.
Holmes, E. E. 2004. Beyond theory to application and evaluation: diffusion approximations for population viability analysis. Ecological Applications 14(4): 1272-1293
Holmes, E. E., and W. F. Fagan. 2002. Validating population viability analysis for corrupted data sets. Ecology 83:2379-2386
IUCN Standards and Petitions Working Group. 2008. Guidelines for Using the IUCN Red List Categories and Criteria. Version 7.0. Prepared by the Standards and Petitions Working Group of the IUCN SSC Biodiversity Assessments Sub-Committee in August 2008. Downloadable from Guidelines for Using the IUCN Red List Categories and Criteria (pdf 1.27 MB)
Kareiva, P., M. Marvier, and M. McClure. 2000. Recovery and management options for spring/summer chinook salmon in the Columbia River Basin. Science 290:977-979.
Kendall B.E. 1998. Estimating the magnitude of environmental stochasticity in survivorship data. Ecological Applications 8, 184-193
Kirchner, F., A. Robert, and B. Colas. 2006. Modelling the dynamics of introduced populations in the narrow-endemic Centaurea corymbosa: a demo-genetic integration. Journal of Applied Ecology 43: 1011-1021.
Lamberson, R. H., K. McKelvey, B. R. Noon, and C. Voss. 1992. A dynamic analysis of northern spotted owl viability in a fragmented forest landscape. Conservation Biology 6: 505-512.
Lande, R. 1988. Demographic models of the northern spotted owl (Strix occidentalis caurina). Oecologia 75:601-607.
Lande, R., S. Engen, and B. Saether. 2003. Stochastic population models in ecology and conservation: an introduction. Oxford University Press, Oxford, UK.
Lande, R., and S. H. Orzack. 1988. Extinction dynamics of age-structured populations in a fluctuating environment. Proceedings of the National Academy of Sciences (USA) 85:7418-7421.
Legault, C. M. 2005. Population Viability Analysis of Atlantic Salmon in Maine, USA. Transactions of the American Fisheries Society 134:3, 549
Lindley, S. T. 2003. Estimation of population growth and extinction parameters from noisy data. Ecological Applications 13:806-813.
Liu, J. G., J. B. Dunning, and H. R. Pulliam. 1995. Potential effects of a forest management plan on Bachman's sparrows (Aimophila aestivalis)-linking a spatially explicit model with GIS. Conservation Biology 9:62-75.
Ludwig, D. 1999. Is it meaningful to estimate extinction probabilities? Ecology 80:298-310.
McCarthy, M. A., M. A. Burgman, and S. Ferson. 1995. Sensitivity analysis for models of population viability. Biological Conservation 73:93-100.
Mace, G.M., N.J. Collar, K.J. Gaston, C. Hilton-Taylor, H.R. Akcakaya, N. Leader-Williams, E.J. Milner-Gulland, and N. Stuart. 2008. Quantification of Extinction Risk: IUCN's System for Classifying Threatened Species. Conservation Biology 22:1424-1442.
Menges, E.S. 1990. Population viability analysis for an endangered plant. Conservation Biology 4:52-62.
Menges, E.S. and R.W. Dolan. 1998. Demographic viability of populations of Silene regia in Midwestern prairies: relationships with fire management, genetic variation, geographic location, population size, and isolation. Journal of Ecology 86: 63-78.
Middleton, D. A. J., and R. M. Nisbet. 1997. Population persistence time: estimates, models, and mechanisms. Ecological Applications 7:107-117.
Miller, P.S. and R.C. Lacy. 2005. VORTEX. A stochastic simulation of the simulation process. Version 9.50 user's manual. Conservation Breeding Specialist Group (IUCN/SSC). Apple Valley, Minnesota.
Morris, W.F., P.L. Bloch, B.R. Hudgens, L.C. Moyle, and J.R. Stinchcombe. 2002. Population viability analysis in endangered species recovery plans: past use and future improvements. Ecological Applications 12: 708-712.
Morris, W. F., and D. F. Doak. 2002. Quantitative conservation biology: theory and practice of population viability analysis. Sinauer Press, Sunderland, Massachusetts, USA.
Morris, W. F., M. Groom, D. Doak, P. Kareiva, J. Fieberg, L. Gerber, P. Murphy, and D. Thomson. 1999. A practical handbook for population viability analysis. The Nature Conservancy, Washington, D.C., USA.
Nantel, P., D. Gagnon, and A. Nault. 1996. Population viability analysis of American ginseng and wild leek harvested in stochastic environments. Conservation Biology 10:608- 621.
Nicholls, J.D., J.R. Sauer, K.H. Pollock, and J.B. Hestbeck. 1996. Estimating transition probabilities for stage-based population projection matrices using capture-recapture data. Ecology 73:306-312.
Oli, M.K., N.R. Holler and M.C. Wooten. 2001. Viability analysis of endangered Gulf Coast beach mice (Peromyscus polionotus) populations. Biological Conservation 97: 107-118.
Rustigian, H. L., M. V. Santelmann, and N. H. Schumaker. 2003. Assessing the potential impacts of alternative landscape designs on amphibian population dynamics. Landscape Ecology 18:65-81.
Sabo, J. L., E. E. Holmes, and P. Kareiva. 2004. Efficacy of simple viability models in ecological risk assessment: does density dependence matter? Ecology 85:328-341.
Schultz, C. B. and P.C. Hammond. 2003. Using Population Viability Analysis to Develop Recovery Criteria for Endangered Insects: Case Study of the Fender's Blue Butterfly. Conservation Biology 17:1372-1385,
Schwartz, M. W., S.M. Hermann, and P.J. van Mantgem. 2000. Population persistence in Florida Torreya: Comparing modeled projections of a declining coniferous tree. Conservation Biology 14: 1023-1033.
Snover, M. L. and S. S. Heppell. 2009 Application of diffusion approximation for risk assessments of sea turtle populations. Ecological Applications 19: 774-785
Stacey, P. B., and M. Taper. 1992. Environmental variation and the persistence of small populations. Ecological Applications 2:18-29.
Staples, D.F., M. L. Taper, and B. Dennis. 2004. Estimating population trend and process variation for PVA in the presence of sampling error. Ecology, 85(4):923-929
Wilcox, C., and H. Possingham. 2002. Do life history traits affect the accuracy of diffusion approximations for mean time to extinction? Ecological Applications 12:1163-1179.
(1) The time span must cover those used in criterion E: 20 yrs or 5 generations (whichever is longer, up to a maximum of 100 yrs) for Endangered; 100 yrs for Threatened.
(2) Quasi-extinction CDF gives the probability that the population will decline below a given threshold (> 1 individual) at or before a given future time; this is considered the single most useful way to present extinction risk information (Morris and Doak 2002).