Conservation and Restoration

In 1995, managers released eight female pumas from Texas into southern Florida, where a small, inbred population of Florida panthers still lived. By 2010 panther numbers had about tripled and heterozygosity had doubled.

Was that a success, and how would you know? Answering takes three ideas: why small populations are fragile, how to put a number on extinction risk, and the difference between how many animals there are and how many pass on genes. Then you can weigh what each conservation tool has actually been shown to do, and who has a stake in the choice.

Why small populations are fragile

A large population absorbs bad luck. A small one can be ended by it. Mark Shaffer, writing in BioScience in 1981, sorted the chance events that threaten small populations into four kinds.

Demographic stochasticity is chance in individual fates: which animals happen to survive, and how many offspring, of which sex, each produces. Suppose, as an illustration, a population is down to its last breeding female and she has five young. The chance that all five are male is (1/2)^5 = 1/32, about 3%. In a population of thousands such runs of luck average out. In a population of ten they decide the outcome.

Environmental stochasticity is variation from year to year in weather, food and the populations of competitors, predators and disease that a population depends on. It hits every individual at once, so it does not average out with size the way demographic chance does. Catastrophes, such as fires, hurricanes and epidemics, are its extreme. The fourth kind is genetic: in a small population genetic drift (chance change in allele frequencies) erodes variation, and mating among relatives raises inbreeding, which exposes harmful recessive alleles and lowers survival and fertility. The arithmetic of drift and inbreeding belongs to population genetics: the short version is that loss of variation per generation scales with one over the effective size.

These forces feed each other. Fewer animals mean more inbreeding, which lowers survival, which means fewer animals. Michael Gilpin and Michael Soulé called this feedback an extinction vortex in 1986. The age- and stage-specific rates that drive these models come from life tables and the growth models of population growth.

MVP and PVA: risk as a probability

How many individuals are enough? Shaffer’s answer was that the question has no answer until you choose a standard. He defined a minimum viable population (MVP) as the smallest isolated population, in a given habitat, with a chosen probability of persisting for a chosen time. As a tentative example he suggested a 99% chance of persisting for 1,000 years. Change the probability, the horizon or the habitat, and the MVP changes.

A population viability analysis (PVA) is how such a probability is estimated. Think of it as a weather forecast for one population: you do not predict the future, you run it many times and report how often it rains. The model takes in survival and birth rates by age or stage, how much those rates vary between years, density dependence, catastrophe frequencies and, sometimes, inbreeding effects. It puts out the fraction of runs that go extinct, or fall below a chosen threshold, within the horizon.

Can such forecasts be trusted? In 2000, Barry Brook and colleagues fitted PVA models to the first half of each of 21 long-term ecological data sets and predicted the second half. They reported that predicted risks of decline matched observed outcomes closely, with no significant bias. That supports PVA within populations and spans like those tested. It does not make any single forecast exact, and because errors in the inputs affect every scenario alike, PVA is most useful for comparing management options against each other.

Worked problem, part A. These numbers are invented. For one hypothetical species in one habitat, a PVA is run 1,000 times from each of three starting sizes, over 100 years. Extinction means zero individuals by year 100. Assume the model’s vital rates and their variation are correct and the habitat stays as it is.

Starting size N0	Runs extinct	Extinction probability
20	120	120 / 1,000 = 12.0%
50	37	37 / 1,000 = 3.7%
100	8	8 / 1,000 = 0.8%

1. Probabilities. Divide extinct runs by total runs, as in the table: 12.0%, 3.7% and 0.8%.
2. Compare with a target. Suppose managers choose at most 5% extinction risk in 100 years. N0 = 20 fails. N0 = 50 passes, with persistence 1 - 0.037 = 96.3%. So 50 is the smallest tested size that meets the target. Sizes between 20 and 50 were never run, so the true threshold could lie anywhere in that range.
3. Check the Monte Carlo error. With 1,000 runs, the standard error of 3.7% is the square root of (0.037 x 0.963 / 1,000), about 0.006, so a rough 95% range is 2.5% to 4.9%. The answer survives, narrowly. A different target, horizon or model would give a different answer.

For comparison, the IUCN Red List can classify a taxon as Endangered under its criterion E when a quantitative analysis gives at least a 20% probability of extinction within 20 years or five generations, whichever is longer, up to 100 years.

Census size and effective size

The census size N is the number of individuals you count. The effective size Ne is the size of an idealised population, with equal sexes, random mating and equal family sizes, that would lose genetic variation at the same rate. Ne is usually smaller than N, because uneven sex ratios, unequal numbers of offspring and swings in population size all speed drift.

For an unequal sex ratio alone, with Nm breeding males and Nf breeding females:

Ne = 4 Nm Nf / (Nm + Nf)

Worked problem, part B. Suppose 50 adults breed: 10 males and 40 females. Assume discrete generations and that sex ratio is the only departure from the ideal.

1. Effective size. Ne = 4 x 10 x 40 / (10 + 40) = 1,600 / 50 = 32. So Ne/N = 32 / 50 = 0.64.
2. Convert a genetic guideline. In 2014 Frankham, Bradshaw and Brook recommended Ne of at least 100 to keep the loss of total fitness to 10% or less over five generations, and at least 1,000 for long-term evolutionary potential. At Ne/N = 0.64, Ne = 100 needs 100 / 0.64 = 156.25 adults. With whole animals at one male to four females, 32 males and 128 females (160 adults) give Ne = 4 x 32 x 128 / 160 = 102.4, which clears 100.

Breeding males N_m 10 Breeding females N_f 40 Target N_e 100 (Frankham and colleagues’ short-term guideline is 100; other values are what-ifs)

Census of breeders N

50

N_e = 4 N_mN_f / (N_m + N_f)

32

N_e / N

0.64

Adults needed for target (exact)

156.25

Smallest whole-animal group at this ratio that reaches the target

32 males + 128 females = 160 adults, N_e = 102.4

In this sketch: only the sex ratio lowers N_e; discrete generations, and other causes of low N_e are ignored; the sex ratio is held constant as the population grows.

The guideline is an effective size, not a head count: convert it with the N_e/N of the population in question. Equal numbers of each sex give N_e = N, and any imbalance gives less.

This assumes the ratio holds and ignores the other causes of low Ne, which in wild populations often push Ne/N much lower. It also shows why the older 50/500 rule, and Frankham’s 100/1,000, are about effective size: neither is a census target. A standardised median MVP of 4,169 individuals across 212 species (Traill and colleagues, 2007) summarises a literature and is not a target for any species either.

Interventions and their evidence

Each tool has its own kind of evidence, and the kinds differ in strength.

Protected areas. Across terrestrial sites worldwide, Gray and colleagues (2016) found sample richness 10.6% higher and abundance 14.5% higher inside protected areas than outside. Most of that advantage was explained by land-use differences between the sites. The comparison is observational: reserves are not placed at random.

Corridors. At the Savannah River Site in South Carolina, in restored longleaf pine savanna, Damschen and colleagues followed 239 plant species for 18 years in patches of equal area, some joined by corridors. Annual colonization rates were 5% higher and local extinction rates 2% lower in connected patches, and connected patches ended with about 200 species against 176. This is a randomized experiment in one landscape, the same logic as island biogeography and spatial populations, and it does not establish how corridors work everywhere.

Translocation. The IUCN separates reintroduction, releasing organisms within their native range where they have disappeared, from reinforcement, adding them to a population that remains. The panther release was reinforcement with genetic rescue as its aim. IUCN guidance asks for strong evidence that the threats that caused any earlier loss have been identified and removed or sufficiently reduced first.

Restoration. The Society for Ecological Restoration defines it as assisting the recovery of a degraded, damaged or destroyed ecosystem. In southern Costa Rica, Karen Holl and colleagues compared natural regeneration, applied nucleation (planting tree islands) and plantations at 13 sites. After about a decade, planting raised recruitment of animal-dispersed trees, yet tree-recruit composition in every treatment remained distinct from reference forest. A 2018 global analysis of 400 recovery studies found that ecosystems rarely recovered completely within the study periods. A 2024 meta-analysis of 665 trials with no-action comparisons found that conservation action improved biodiversity or slowed its decline in about two-thirds of cases.

Decisions with competing goals

Evidence says what an option is likely to do. It does not say which outcome matters most. For endangered sockeye salmon of Cultus Lake in British Columbia, Gregory and Long (2009) described a structured decision process that set out biological, social and economic objectives, the alternatives and their predicted consequences, so that the fishery agencies, First Nations and commercial fishers involved could see the tradeoffs explicitly. Climate change shifts the ground under all of these options, as global change shows, and the final lesson, the evidence workshop, practises weighing such studies together.

Design problem

A small, isolated wild cat population shows heart defects, kinked tails and low sperm quality, and its habitat is still shrinking. Managers propose bringing in individuals from another population of the same species. What would you measure, and what could the result show?

One way to do it

Before any release, record a baseline: counts, adult survival, kitten recruitment, heterozygosity at genetic markers, and the rates of each defect. After release, type every animal you can for ancestry, and track admixed and non-admixed individuals separately for survival, reproduction and defects, while continuing the population counts.

Then compare admixed with non-admixed animals living at the same time, and the period after release with the period before. Neither comparison is a randomised control, because there is only one population, so habitat, prey and road deaths change alongside ancestry. State that limit.

Weigh the risks too: outbreeding depression if the donor population is adapted to different conditions, and disease carried by the newcomers. Ask whether the threat that shrank the population remains; if the habitat keeps shrinking, a genetic gain may be temporary.

The Florida panther is the real case. After eight Texas females were released in 1995, Johnson and colleagues reported in 2010 that numbers had tripled and heterozygosity had doubled. A 2024 follow-up found the benefits persisted for five generations, but also that the population stabilised and declined during 2016 to 2020. Both results belong in the answer.