Coverage from the count

How much of an area does a drone survey of moose need to fly? A rule you can work out before the first flight, checked against SLU Rapport 2026.2. Plot-based drone surveys of moose must decide how much of an area to fly. SLU Rapport 2026.2 counted every moose in two areas and showed, by resampling the counts, that the answer depends on how clumped the moose are. We write that dependence as one line of standard sampling theory, Cochran's sample size solved for coverage: the share to fly follows from the number of moose and from how much the counts vary between plots. Precision is bought with animals seen rather than land covered, so at a given clumping the hectares a survey needs level off as areas grow. With the report's own numbers the rule gives 19.0% for Alster and 55.9% for Fredrika, where the report found 20% and 50–60% by simulation. It sizes for the width the report's simulations measure; a survey reporting a bootstrap over its own plots prints an interval 1.51 times wider at Fredrika's coverage, unless it is corrected for the frame. Corrected, it has the width the rule sized for, but on plots taken at random it does not yet hold the true density 95% of the time. An interval that does is wider, so the coverage the rule gives is a floor. A first round of plots can measure the clumping, but a survey sized on that measurement falls short more often than not, and an upper bound only partly fixes it. How to bound the clumping of strongly clustered moose is open. Preprint, not peer reviewed. The zip holds the code that computes every figure in the paper (Python 3, standard library only, MIT licence) and the text of SLU Rapport 2026.2 (CC BY 4.0) that the checks quote; see its README. A plain-language summary is at sparnum.com/en/moose-survey/coverage.