Population ecology explains how and why the abundance, composition and distribution of a species change. It connects births, deaths and movement with food, habitat, competitors, predators, disease and environmental conditions. Its central task is to explain a population's trajectory through mechanisms that can be measured, rather than treating an increasing or decreasing count as an explanation in itself.
The gold standard is a clear population definition, an observation method that fits it, and a demographic explanation that survives comparison with alternatives. A stable total can hide poor reproduction sustained by immigration. A falling count can reflect lower detectability rather than fewer organisms. This guide works through those distinctions using original fictional examples, explicit calculations, practical diagnostics and a changed-context exercise. None of the numerical cases is a forecast for a named species.
Use the Gold Standard master to explore the collection, the Education guide to connect understanding with independent application, and the How X Works hub for other mechanisms. The broader Ecology root places population questions within communities and ecosystems.
Choose your population ecology route
Connect abundance with births, deaths, movement, life stages, observation and environmental change.
Route 1 · Chapters 1–3: Define populations, balances and rates
Route 2 · Chapters 4–6: Follow density, survival and life stages
Route 3 · Chapters 7–9: Interpret models, observations and variation
Route 4 · Chapters 10–11: Diagnose change and test mechanisms
Route 5 · Chapters 12–14: Transfer, diagnose and continue
Chapter index
Define populations, balances and rates · Chapters 1–3
Follow density, survival and life stages · Chapters 4–6
Interpret models, observations and variation · Chapters 7–9
Diagnose change and test mechanisms · Chapters 10–11
Transfer, diagnose and continue · Chapters 12–14
A population is not simply every organism appearing in a photograph. For an ecological investigation, specify the species, geographical boundary, period and biological units included. Breeding adults in a wetland, all life stages in a lake and seedlings emerging in a forest plot are different populations for measurement purposes. Their counts cannot be exchanged without changing the question.
Suppose a survey reports 120 frogs around a pond. Does this mean 120 individually recognised animals, 120 calling records or 120 sightings that might include repeat observations? Were juveniles included? Did the survey cover the whole breeding season or one evening? An apparently precise number becomes informative only when its meaning and collection method are clear.
Distinguish abundance, density and distribution
Abundance is the number of individuals in the defined population. Density relates that number to an area or volume. Distribution describes where individuals occur. A population can retain the same total abundance while becoming concentrated in a smaller usable area. Its density then rises even without population growth. Conversely, an expanding survey boundary can raise the count without any increase inside the original boundary.
Use a fictional example: 600 individuals occupy 30 hectares, giving 20 individuals per hectare. If usable habitat contracts to 20 hectares while the total initially remains 600, density becomes 30 per hectare. Reporting the higher density as recovery would reverse the likely interpretation. The organisms now occupy less habitat, and subsequent crowding effects remain to be measured.
For plants, the counting unit can be especially consequential because stems, rooted shoots and genetic individuals need not coincide. The Plant Population Biology root develops those distinctions. The broader lesson applies everywhere: decide what an individual and a sampling unit mean before selecting a model or comparing sites.
For a population with a consistent boundary, the accounting identity is ending abundance = starting abundance + births − deaths + immigration − emigration. These are event counts over the same interval. The identity does not explain why events occurred, but it identifies the processes that any complete explanation must address.
Consider a fictional island population starting with 500 animals. During a year, 140 are born, 110 die, 30 arrive and 60 leave. Ending abundance is 500 + 140 − 110 + 30 − 60 = 500. The unchanged total conceals a natural increase of 30 offset by net outward movement of 30. Calling this population reproductively stagnant would be wrong.
Stable totals can have different causes
A second population also starts and ends at 500, but has 80 births, 120 deaths, 60 immigrants and 20 emigrants. Its local birth–death deficit is 40, balanced by net immigration of 40. These two populations have identical net change but different dependencies. A movement barrier could affect their future trajectories differently.
The balance must avoid double counting. If an animal is born outside the study area and later enters, it contributes immigration to the local accounting, not a local birth as well. If births are counted at hatching but the abundance census includes only adults, a direct addition of all hatchlings to adult abundance is invalid. Survival and maturation must connect the stages.
An open population exchanges individuals with its surroundings. A closed-population assumption excludes specified exchanges during a specified interval; it is not a permanent property of an island, fence or map boundary. A pond can appear geographically discrete while adults move among neighbouring ponds. State whether closure concerns births and deaths, movement, or both, and explain why the chosen interval makes the assumption plausible.
If a population grows from 200 to 242 individuals over two years, the total proportional increase is 21%. The constant annual multiplier that would produce that result is the square root of 242/200, or 1.10. Under that constant-multiplier description, the population increases by 10% per year. Dividing 21% by two gives an arithmetic summary that does not reproduce the compounded trajectory.
Write the discrete model as N at the next census = λN at the current census. With constant λ, abundance after t intervals is N₀λᵗ. A multiplier above one indicates growth, below one decline and equal to one no change, within that model. The interval must accompany λ: a multiplier of 1.10 per month differs greatly from 1.10 per year.
Translate between continuous and discrete descriptions
For continuous exponential growth, N(t) = N₀ exp(rt), where r is the instantaneous per-capita rate. For an interval of length Δt, λ = exp(rΔt), so r = ln(λ)/Δt. With λ = 1.10 over one year, r is approximately 0.0953 per year, not exactly 0.10. The distinction becomes more important as changes or intervals grow larger.
The implied doubling time at this constant annual multiplier is ln(2)/ln(1.10), approximately 7.27 years. That is a conditional calculation, not a promise that the population will double. Resources, weather, stage composition and movement may change before then. An exponential model is often a useful short-term benchmark precisely because its assumptions are simple enough to inspect.
Do not estimate an intrinsic biological growth capacity from an observed local multiplier without considering immigration, density and composition. The observed increase could include arrivals or reflect an unusually favourable year. The Mathematical Biology root offers a wider route into model construction, units and the distinction between a mathematical consequence and an empirical prediction.
Density dependence concerns how demographic performance changes with population density or abundance under a specified model. If food becomes harder to obtain as competitors increase, growth or reproduction may fall. If finding a mate is difficult when individuals are sparse, performance may instead improve as density rises over a low-density range. The sign and mechanism both matter.
The continuous logistic model is dN/dt = rN(1 − N/K). Here r is a low-density per-capita growth parameter and K is the positive equilibrium abundance under fixed model conditions. The per-capita rate is r(1 − N/K), while the total rate includes the factor N. These are different quantities even though they arise from the same equation.
Compare per-capita and total change
Choose r = 0.4 per year and K = 1,000 individuals. At N = 100, per-capita growth is 0.36 per year and total growth is 36 individuals per year. At N = 500, per-capita growth is lower, 0.20 per year, but total growth is higher, 100 individuals per year. Lower individual performance does not necessarily mean fewer individuals added in total.
At N = 900, the corresponding rates are 0.04 per year and 36 individuals per year. In this model, total growth peaks at N = K/2. That algebraic result is useful for understanding the curve, but it is not sufficient evidence for a real harvesting policy. The model omits age structure, uncertainty, movement, delays and changing environmental capacity.
Treat K as a conditional model parameter rather than a permanent number stamped onto a habitat. A change in resources, disease, competitors or usable space can change the relationship between abundance and demographic rates. A 2019 primary study of density effects using encounter data illustrates why estimating density dependence requires an explicit statistical model, not merely a downward-looking scatterplot.
CHAPTER 5 OF 14
5. Follow cohorts and distinguish survival from persistence at a site
A cohort is a group defined by a shared starting event, such as birth during a particular season. Following a cohort separates individual survival from the changing composition of the whole population. A high total count can coexist with poor survival if recruitment is strong, while a stable older cohort can coexist with almost no recruitment.
Suppose 100 newly emerged insects are individually identifiable. By the next scheduled census, 70 are alive; by the following census, 42 remain alive. Survival from the start to the first census is 0.70. Conditional survival from the first to the second is 42/70 = 0.60. Survival across both intervals is 0.70 × 0.60 = 0.42, matching 42/100.
Do not mix conditional and cumulative probabilities
Subtracting 0.60 from 0.70 does not give the fraction of the original cohort dying in the second interval. There were 28 deaths among the 70 survivors, or 40% of that interval's starting group and 28% of the original group. Each is a valid quantity with a different denominator. Life-table interpretation depends on maintaining those distinctions.
In field studies, an individual that is not seen again may be alive elsewhere or alive but undetected. A return rate is therefore not automatically a survival probability. Models of marked individuals must account for observation and movement according to their design. The USGS account of capture–recapture methodology explains how repeated encounters support estimates of abundance, survival and transitions despite imperfect detection.
A 2025 USGS-associated study of joint recapture and resighting designs examined how sampling design affects survival estimation when animals move. Its relevance is methodological: adding observations outside a capture site is useful only when their spatial coverage and model assumptions are appropriate. Calling a model sophisticated does not establish that its observation design separates survival from movement.
For example, if observers revisit only the easiest nearby locations, dispersers travelling farther may remain systematically unobserved. More repeated visits to those same locations can increase effort without resolving that spatial gap. Diagnose the missing state or movement pathway before assuming that collecting more of the same data will solve the problem.
A practical repair is to draw the sequence of states: alive in the study area, alive outside it, dead and observed or unobserved conditional on location. Then ask which distinctions the data can identify. Adding a sophisticated model cannot recover information that the sampling design never made available without additional assumptions or evidence.
Age and life stage affect survival, reproduction and movement. A population dominated by breeding adults can produce a different immediate outcome from one dominated by juveniles, even when total abundance is identical. A total-only growth model compresses this structure; sometimes that approximation is useful, and sometimes it hides the process of interest.
Consider a fictional female-only, two-stage annual census with juveniles J and adults A. Define the next year's juveniles as J' = 1.2A. Define adults as A' = 0.3J + 0.7A. The first coefficient means an adult at this census contributes an expected 1.2 juveniles present at the next census. It is not simply the number of eggs laid: any survival needed to reach that census is already included.
Calculate two populations with the same initial total
Population X has 80 juveniles and 20 adults. Its next census contains 24 juveniles and 24 + 14 = 38 adults, totalling 62. Population Y has 20 juveniles and 80 adults. Its next census contains 96 juveniles and 6 + 56 = 62 adults, totalling 158. Both started at 100, and both used identical demographic coefficients.
The difference is initial composition. The calculation does not prove that X has permanently worse habitat or that Y will grow forever. It shows that a one-year total multiplier can be strongly affected by stage structure. Repeated projection changes that structure, so extrapolating either first-year multiplier indefinitely would contradict the stage model used to generate it.
The matrix representation is simply a compact way to organise these contributions: the first row contains 0 and 1.2; the second contains 0.3 and 0.7. Leslie's 1945 paper is an important early population-matrix reference. The example here is a deliberately simplified stage model; a biological application must justify the states, census timing and transitions rather than borrowing its coefficients.
For the two-stage model, a constant long-run multiplier λ satisfies λ² − 0.7λ − 0.36 = 0. The positive root is approximately 1.045. Under the fixed model and its convergence conditions, stage composition approaches a stable proportion and annual multiplication approaches that value. This does not erase the sharp initial differences between the populations in the previous chapter.
The long-run juvenile-to-adult ratio follows λJ = 1.2A, giving J/A = 1.2/λ ≈ 1.148. A population beginning near that ratio behaves differently at first from one containing almost entirely adults. The mathematical distinction is between transient behaviour while composition changes and asymptotic behaviour after the composition settles in the constant model.
Check the calculation biologically
The multiplier above one means growth under the specified coefficients, not unlimited growth under all ecological conditions. If increasing abundance reduces recruitment, the coefficients cannot remain constant indefinitely. If a drought reduces juvenile survival, a model fitted to favourable years may give the wrong sign of change. Report the conditions under which the multiplier was obtained.
Sensitivity analysis asks how a model output changes when a parameter changes. It is useful for identifying influential processes, but influence is not the same as practical control. A demographic rate can have high mathematical sensitivity and be nearly impossible to alter. Another may have lower sensitivity but be more feasible to improve. Decisions need the likely achievable change and its uncertainty, not a ranking of derivatives alone.
Where body size or another continuous trait is more informative than a few stage classes, follow the specialist guide to integral projection models and kernel uncertainty. More detailed models can represent additional biology, but they also require more evidence. A finer classification should earn its place by improving the answer to a real question.
Field counts are produced by an observation process. Even if the underlying population is unchanged, different weather, vegetation, observer effort or behaviour can change how many organisms are detected. A useful conceptual starting point is that an expected count depends on both abundance and detection probability. The exact observation model depends on the sampling design.
Imagine a true population of 400 animals. If each has a 0.50 probability of being detected in a standard survey, the expected count is 200. In a later period, detection falls to 0.30 while abundance remains 400, giving an expected count of 120. The expected observed decline is 40%, despite no change in true abundance. Random variation would also affect realised counts.
A simple capture–recapture illustration
In a fictional closed population, suppose 50 individuals are marked and released. A later sample contains 40 individuals, of which 10 are marked. If marked animals are well mixed and capture probabilities meet the simple model's assumptions, the observed marked fraction 10/40 estimates the population's marked fraction 50/N. Solving gives N ≈ 200.
This elementary estimator is a teaching illustration, not a recommendation to capture wildlife. Marks must remain identifiable, marking must not alter relevant capture or survival processes, and births, deaths or movement must not invalidate closure. Unequal catchability and a small number of recaptures can make the estimate unreliable. Real studies use designs and estimators appropriate to these complications.
Different observation methods expose different assumptions. The existing guide to removal sampling, depletion curves and closure explains another approach. A declining catch across passes may reflect depletion, but changing effort or capture probability can produce misleading patterns. Before interpreting a trend, ask how the method turns living organisms into recorded data.
Demographic stochasticity arises from chance outcomes for individuals, such as which individuals survive or reproduce. Environmental variation changes conditions that affect many individuals, often together. Both can influence population trajectories, but their consequences and scaling differ. A larger population can average over some independent individual outcomes while still being exposed to the same severe environmental event.
Suppose each of 10 individuals independently survives an interval with probability 0.8. Expected survivors are eight, but the realised number need not be eight. Under this binomial teaching model, the standard deviation is the square root of 10 × 0.8 × 0.2, approximately 1.26. With 1,000 individuals, expected survivors are 800 and the standard deviation is approximately 12.65. Absolute variation is larger, but variation relative to the mean is smaller.
Multiplication makes the order of averaging important
Consider annual population multipliers alternating between 1.5 and 0.5. Their arithmetic mean is one, but the two-year product is 0.75. A population of 100 becomes 150 and then 75. The equivalent constant annual multiplier is √0.75 ≈ 0.866. Averaging percentage changes and expecting cancellation would miss the decline.
This is an original deterministic alternating-year calculation. Lewontin and Cohen's 1969 paper develops the wider distinction between arithmetic and geometric growth in randomly varying environments. Applying a stochastic result requires its assumptions; real sequences can have autocorrelation, density dependence and changing stage structure.
Be precise about what is varying. Increasing variation in population multipliers has a different interpretation from increasing variation in temperature, because growth may respond nonlinearly to temperature. The primary study on population effects of climate variation explains why environmental variation cannot always be translated into a single universal statement about population growth. Trace the environmental driver through the demographic response before drawing the conclusion.
A decline can result from lower reproduction, lower survival, outward movement, changed observation or combinations of them. Start by proposing competing explanations that imply different evidence. A general statement that the habitat is stressed is not yet a demographic diagnosis. Identify the affected stage, process and period.
For a fictional bird population, suppose adult counts decline after a dry year. One hypothesis is that adults died during the dry season. Another is that breeding failed, producing fewer new recruits for the next census. A third is that adults moved to wetter sites. A fourth is that survey conditions reduced detection. Counts alone may not distinguish these explanations.
Build a measurement that could change your mind
Adult encounter histories can inform survival-related questions, but movement and detection still need attention. Nest monitoring addresses parts of reproduction, but a successful nest does not guarantee survival to the population census. Surveys of neighbouring sites address redistribution. Repeated visits under recorded conditions help assess observation changes. Each data stream should correspond to a proposed mechanism.
Time lags also matter. A poor breeding year may affect adult abundance only after a delay, depending on maturation. A population with long-lived adults can remain visually impressive while recruitment has already failed. Conversely, a strong juvenile cohort may not yet contribute to breeding. Align the timing of environmental exposure with the affected life stage.
The Metapopulation Ecology root follows dispersal, local loss and regional persistence. It helps distinguish a local decline from a regional decline. Local and regional questions are both legitimate, but a change in the study boundary must not silently change the claim being evaluated.
A model can reproduce an observed trajectory for the wrong reason. Several combinations of births, deaths and movement can generate the same totals. A flexible curve can fit abundance without identifying the ecological cause. Good model assessment therefore checks both the predicted pattern and the evidence for the processes used to produce it.
If resource limitation is proposed, specify which resource, how its availability changes with density and which demographic rate responds. If predators are proposed, separate encounter rates, mortality and compensatory responses elsewhere in the life cycle. The Resource Competition root connects resource depletion to organism performance and population outcomes.
Use comparisons with a credible alternative
Suppose an intervention site grows from 100 to 130 counted individuals while a comparable untreated site grows from 100 to 120. The intervention site's raw increase is 30. A simple difference in changes is 30 − 20 = 10. This arithmetic contrast removes the shared observed increase only under assumptions about how the untreated site represents the intervention site's alternative trajectory.
The number ten is not automatically a causal effect. Sites may differ in habitat, initial stage composition or detection; individuals may move between them; an intervention may be applied precisely where recovery was already expected. Replication, allocation, baseline records and process measurements determine how convincing the comparison becomes. State those features before presenting a precise effect estimate.
Separate calibration from validation. Data used to choose parameters are not an independent test of predictive performance. Holdout periods, additional sites or prospective predictions can provide stronger checks when they match the intended use. A model that predicts next year's count may still fail under a new drought regime. The relevant test is the one corresponding to the proposed decision, not whichever test produces the most reassuring graph.
Return to the two-stage model with J' = 1.2A and A' = 0.3J + 0.7A. Its positive long-run multiplier is about 1.045. Suppose a changed environment raises the contribution of juveniles per adult to 1.5 but lowers adult survival to 0.5. Juvenile maturation survival remains 0.3. Does the improved recruitment coefficient guarantee population growth?
Before calculating, write the new equations: J' = 1.5A and A' = 0.3J + 0.5A. This prevents the attractive improvement in one parameter from concealing deterioration in another. The positive multiplier now satisfies λ² − 0.5λ − 0.45 = 0, giving λ ≈ 0.966. Under the fixed changed model, the long-run prediction is decline.
Find the recruitment threshold
Replace the recruitment coefficient by F while keeping juvenile maturation at 0.3 and adult survival at 0.5. At the boundary λ = 1, the characteristic equation becomes 1 − 0.5 − 0.3F = 0. Thus F = 0.5/0.3 ≈ 1.667. Recruitment of 1.5 is below this model threshold despite exceeding the original value of 1.2.
Now start with 50 juveniles and 50 adults under the changed model. The next census has 75 juveniles and 15 + 25 = 40 adults, totalling 115. The population increases in this first year even though the asymptotic multiplier is below one. This is not a contradiction: the initial stage composition differs from the stable composition of the changed model.
The exercise tests transfer across changed conditions. A correct answer must identify the new parameter combination, distinguish initial change from long-run growth, and keep the conditional model result separate from an empirical forecast. More offspring alone does not settle the question; survival, census timing and the stage composition of the population remain part of the explanation.
Does a large population necessarily have a healthy future?
No. Abundance is a current state, while future persistence depends on demographic processes and conditions. A large population can contain mostly ageing individuals with poor recruitment. It can also be concentrated into shrinking habitat. Assess the state and the mechanisms together rather than using one count as a certificate of security.
Does every population settle at carrying capacity?
No. A stable equilibrium belongs to a particular model and parameter range. Conditions can change, feedback can be delayed and interactions can create cycles or more complex behaviour. May's 1976 analysis shows that simple deterministic difference equations can generate varied dynamics. Irregular fluctuations therefore do not, by themselves, identify random environmental forcing.
Can a population grow even when reproduction is insufficient locally?
Yes, if immigration more than offsets the local deficit. That is why the demographic balance includes movement. But a single year's budget does not establish a permanent source–sink classification. Repeated evidence is needed to understand dependence, density effects and variation across years.
Is an accurate count better than a model-based estimate?
An actual complete census can be valuable, but many field counts are incomplete. A count with unknown detection can be less interpretable than an estimate with a justified observation model and clearly reported uncertainty. The important comparison concerns what each method measures, how errors arise and whether the result answers the ecological question.
Why do two studies report different growth rates?
Check their intervals, boundaries, stages, observation methods and definitions. One may report an annual multiplier, another a continuous per-capita rate, and another change in occupied sites. These are not interchangeable. Reconcile the quantities before deciding that the biological findings conflict.
What should be repaired first in a weak population explanation?
Locate the first unsupported link: the population definition, count interpretation, transition estimate or causal mechanism. More decimal places cannot repair an unknown denominator. A more elaborate model cannot automatically repair unmeasured movement. A useful correction makes the missing relationship explicit and identifies evidence capable of testing it.
Start with the demographic balance. Create two fictional populations with the same ending abundance but different births, deaths and movement, then explain how their vulnerabilities differ. Next, calculate growth over unequal intervals and convert between a multiplier and a continuous rate. Keep the original units beside every value so the calculation remains auditable.
Move to the stage model only after the balance is clear. Use a spreadsheet with separate cells for juveniles, adults and each coefficient. Project several years from contrasting initial compositions. Then change one demographic coefficient and ask whether the first-year total and long-run tendency move in the same direction. Do not overwrite the original assumptions when making the comparison.
For an evidence-reading exercise, select a published population study and write a short account of its species, boundary, census timing, observation process and proposed mechanism. Identify which parameters are measured directly, which are estimated statistically and which are assumed. Find one result that would make the explanation less convincing. That final step turns reading into an independent test of understanding.
Sources and connected study
Leslie, 1945 provides an early population-matrix foundation. Lewontin and Cohen, 1969 examines growth in randomly varying environments. May, 1976 examines complex behaviour in simple difference equations. These sources establish theoretical questions; the fictional calculations in this guide are independently constructed illustrations.
The USGS capture–recapture account explains observation and demographic estimation. The 2019 encounter-data study addresses density effects within a statistical model. The study of population effects of climate variation helps distinguish variation in environmental drivers from variation in demographic growth.
Continue with Plant Population Biology for plant-specific units and life cycles, integral projection models for continuous individual states, or Metapopulation Ecology for spatial persistence. A dependable population explanation shows what changed, which demographic processes could produce it, how observations constrain those processes and where the conclusion stops being supported.