Bias in Aerial Survey

Abstract

Aerial censuses of large mammals are inaccurate because the observer misses a significant number of animals on the transect. The accuracy deteriorates progressively with increasing width of transect, cruising speed, and altitude. Methods of eliminating bias by refining techniques are discussed and rejected; there seems to be no technical solution. An alternative strategy is to measure the bias and correct the estimates accordingly. A method is suggested for estimating bias during an aerial census, the subsequent analysis returning an unbiased estimate of density. No direct measure of true density is needed and little extra effort is involved over that required for a standard aerial survey. J. WILDL. MANAGE. 38(4):927-933 This paper examines the effect of visibility bias, discusses the means by which the bias arises, and suggests methods by which it might be eliminated from aerial survey estimates of density and population size. Aerial survey is, at best, a rough method of estimating the size of a population. Most efforts at refinement have been aimed at raising the precision of the estimate by combining impeccable survey design, high sampling intensity, intricate stratification, and powerful methods of analysis. This trend can be traced back to Siniff and Skoog's (1964) superb paper on an aerial census of caribou (Rangifer tarandus). Their use of stratified random sampling, with sampling effort allocated proportional to density, contrasted markedly with the crudity of previously reported surveys. Subsequently, Jolly's (1969a) paper on designs and analyses appropriate to aerial survey has encouraged a rigorous and disciplined application of the method. Recent papers following this lead have tended to treat the difficulties of estimating population size from the air largely as constituting a sampling problem, a survey being rated successful or otherwise according to the size of the estimate's standard error. Tacitly, the standard error was treated as a measure of the estimate's accuracy rather than of its repeatability. Underlying the preoccupation with precision there often lurked an implicit assumption that the estimate is free of bias, that the observers counted all animals on each sampled