capture-recapture

Counting wild animals directly is almost never possible: they move, they hide, and no survey sees all of them. Capture-recapture works around that. Researchers record a set of individuals, each one identifiable, then sample the same area again and note how many of the individuals they see are ones they have already recorded.

The ratio does the work. If most of the animals recorded in the second round are ones already known, the population is probably small and the survey is seeing most of it. If few repeat, there are likely many more animals out there that the survey has missed. From that overlap a statistical model estimates the total, including the animals never detected at all.

The animals do not have to be physically caught. The US Geological Survey describes marks ranging from aluminium leg bands on birds to camera trap photographs of animals identified by their natural markings, which is how tiger surveys work: every tiger's stripe pattern is unique, so a photograph is the mark.

Spatially explicit capture-recapture, the version used in Nepal's national tiger survey, adds where each animal was detected. Modelling location as well as identity lets the estimate account for animals whose range straddles the edge of the survey area, and for the fact that a camera only ever sees what walks past it. The result is a density estimate with a stated range of uncertainty rather than a single flat headcount.