How Do Scientists Count Wildlife? Estimating What Cannot Be Counted
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
Nobody counts all the deer in a forest, because they move, hide and are not distinguishable from each other. What is done instead is to count a sample under conditions that permit an estimate of the whole, and the entire craft lies in accounting for the animals that were present and not seen, which is always most of them.
Mark and recapture
The oldest formal method rests on a proportion. Catch a number of animals, mark them and release them. Later, catch a second sample and count how many carry marks. If the marked animals have mixed evenly back into the population, the proportion marked in the second sample estimates the proportion marked in the whole, and since the number originally marked is known, the total follows. That is the Lincoln-Petersen estimator and it was used to estimate human populations before it was applied to animals. Its assumptions are strong and each one fails somewhere: the population must be closed with no births, deaths, arrivals or departures between samples; marks must not be lost; marking must not affect survival or catchability; and every animal must be equally likely to be caught. That last is the most frequently violated, since animals that were caught once are frequently either shy of traps afterwards or attracted to them by bait, and both biases distort the estimate in predictable directions. Later models relax these assumptions explicitly and estimate capture probability as a parameter.
The other main approaches
Different situations call for different designs, and each handles imperfect detection in its own way:
- •Distance sampling, where observers walk or fly transects and record the distance to each animal seen, which allows a detection function to be fitted showing how detectability falls with distance and then corrects for what was missed
- •Point counts, used for birds, recording everything heard or seen from fixed points for a fixed period
- •Occupancy modelling, which asks not how many but what proportion of sites are occupied, using repeat visits to estimate the probability of detecting a species that is present
- •Quadrat sampling for plants and sessile organisms, counting within randomly placed frames and scaling up
- •Capture-recapture using photographs of naturally marked individuals, which avoids handling entirely
- •Indirect counts of signs including nests, dung, tracks and calls, which require a conversion factor to animals that is itself estimated and uncertain
- •Genetic capture-recapture from hair or dung samples, where the individual identity comes from DNA rather than from a tag
Why detection is the whole problem
Every method confronts the same fact: a count is a product of how many animals are present and how likely each is to be detected, and a change in a raw count could be either. That matters enormously for monitoring, because a survey showing fewer birds this year than last may reflect a genuine decline or a windier survey day. Methods that estimate detection explicitly separate the two, which is why occupancy and distance sampling are preferred to raw counts wherever they are practical. Detection varies systematically with habitat, weather, observer experience, time of day, season and the animal's own behaviour, and it varies between species, which is why comparing raw counts across species is meaningless. The practical consequences appear in standards: consistent protocols, trained observers, repeat visits, recording covariates that affect detection, and reporting confidence intervals rather than single numbers, since an estimate without an interval conceals whether it means anything.
Where the numbers come from in practice
Published population figures for wild species vary enormously in reliability and the difference is frequently invisible in how they are reported. Well-studied species in accessible places have estimates from designed surveys with stated methods and intervals. Many figures for large mammals in remote areas rest on expert judgement, extrapolation from small study areas or old surveys repeated by citation rather than by fieldwork, and several widely quoted global figures have been traced back to a single unreferenced estimate. Trend data is generally more reliable than absolute numbers, since a consistently applied method detects change even if its absolute calibration is uncertain, which is why indices of abundance rather than counts underpin most conservation assessments. Long-running volunteer schemes, including breeding bird surveys and butterfly transects run by thousands of people to a fixed protocol, produce some of the best trend data available, which is a substantial argument for the value of organised amateur recording.
The takeaway
Wildlife is estimated rather than counted, because most individuals present are never seen. Mark and recapture infers a total from the proportion of marked animals in a second sample, and its assumption that everything is equally catchable is the one most often broken. Distance sampling fits how detectability falls with range, and occupancy modelling estimates the chance of detecting a species that is there. Trends are more reliable than absolute numbers, and several widely quoted figures trace to a single old estimate.