data on grizzly bears might allow us to make reliable estimates of extinction over the
next 5 or 10 years, but certainly not over the next 100 years.
This problem is compounded when we have no idea about the reliability of popula-
tion estimates in any given year. This is obviously less of an issue in the rare cases
where every individual is recognizable. Most populations are known only from
samples taken from a small fraction of the inhabited range, leading to considerable
uncertainty in true abundance. Observation errors play a key role, because they con-
vey a false impression about the true magnitude of environmental and demographic
stochasticity, as well as biased estimates (usually downward) about the strength of
density dependence. As a consequence, even well-studied populations may yield biased
predictions of extinction risk (Ludwig 1999).
Data for well-studied populations illustrate that catastrophic climatic events play an
important role in causing population collapse. Such catastrophes tend to be difficult to
predict using Monte Carlo simulations, particularly when long-term census data are
lacking (Coulson et al. 2001b). Most importantly, application of PVAs is founded on
an underlying faith that conditions (e.g. climate, habitat availability, and human inter-
ference) will hold far into the future (Coulson et al. 2001b). For example, the Yellow-
stone grizzly bear data for 1959–82 provide a substantially higher risk of extinction than
that of the later demographic data, suggesting a change in environmental conditions.
On the basis of these quantitative weaknesses, some critics have claimed that popula-
tion viability analysis is virtually meaningless (Ludwig 1999; Coulson et al. 2001b).
Moreover, preoccupation with the stochastic dynamics of small populations ignores
the ecological, physical, and anthropomorphic causes of population decline (Caughley
1994; Harcourt 1995; Walsh et al. 2003). Other conservation biologists argue that,
while not infallible, PVA might still be quite useful as a means of comparing relative
extinction risk among populations or in various subpopulations of a single species,
or of assessing the relative risks associated with alternative management actions
(Lindenmayer and Possingham 1996; Brook et al. 2000; Morris and Doak 2002).
To resolve these different views, Brook et al. (2000) gathered demographic data
from 21 well-studied populations. They used the first half of the data for each time
series to parameterize PVA models, and then used the resulting PVA models to pre-
dict the outcome of the second half of each data set. They concluded that the risk
of decline closely matched predictions and that there was no significant bias in pre-
dictions. They also found few major differences in the quality of predictions of any
of the most common models that are commercially available to decision-makers. Coulson
et al. (2001b) countered that the 21 data sets considered by Brook et al. were unrep-
resentative of endangered species most likely to be candidates for PVA. Rare organ-
isms are, by their very nature, poorly understood. Nonetheless PVA is a widely accepted
tool for risk assessment by both field biologists and decision-makers.
Despite the preoccupation of most population viability analyses with stochastic
extinction processes, the most common cause of extinction is a critical change in the
organism’s environment. This is distinct from year-to-year fluctuation due to either
demographic or environmental stochasticity. We identify the new environment as the
driving variable responsible for the population’s decline and the population may be
driven to extinction by its action. A population seldom “dwindles to extinction.” It
is pushed. If you can identify the agent imparting the pressure and neutralize that
pressure you can save the population.
CONSERVATION IN THEORY 305
17.8 Extinction caused by environmental change
