Sampling in Aerial Survey

Abstract

procedures appropriate to aerial survey were compared firstly in terms of the statistical efficiency and secondly according to their operational practicality. without replacement is more precise than sampling with replacement, but it requires a standard of navigation often impossible to achieve in practice. Systematic sampling can lead to biased standard errors but that must be weighed against its practical advantages when animal distributions are mapped. Fragmented cluster sampling returns unbiased estimates but autocorrelation between units within clusters (transects) biases the standard errors. Except in special circumstances, transect sampling is more efficient than quadrat sampling. J. WILDL. MANAGE. 41(4):605-615 Aerial survey is the only practicable means of estimating the number of large animals inhabiting a large area. Although the estimate is usually inaccurate and often imprecise it answers a broad range of ecological and management questions to an acceptable level of approximation. Three previous papers (Caughley and Goddard 1972, Caughley 1974, Caughley et al. 1976) addressed the problem of inaccuracy, suggesting methods by which errors of measurement could be detected, measured and eliminated from the estimate. This paper is concerned with maximizing the precision of the estimate after the counts have been corrected for visibility bias. The strictly operational problems of planning and running an aerial survey are not covered here. They have been detailed by Norton-Griffiths (1975) and Caughley (1977). The analyses appropriate to aerial survey are to be found within sample survey theory, a well developed field with a copious literature. Cochran's (1963) book Sampling Techniques serves both as a compendium of its methods and as a tribute to its breadth and depth. In practice, survey statisticians usually restrict themselves to estimating parameters like (1) level of unemployment, (2) retail prices, (3) employment of scientists and engineers in industry, and (4) voting intentions. The sampling units appropriate to these would be (1) households, (2) shops, (3) companies and (4) persons, units in a cognate rather than a geographic sense. The sampling units of aerial survey are, in contrast, pieces of land whose size and shape are arbitrary. Although survey sampling theory implicitly embraces such a frame, the difference in kind between the 2 classes of units must be kept in mind when adapting to one the methods developed for the other. There is nothing new in this paper save some illustrative data. It expounds, yet again, the well established principles of survey statistics. Its modest contribution lies in the examination of these principles as they relate specifically to the special problems of aerial survey. A firm basis has already been laid by Jolly (1969) in a paper which, paradoxically, has been as influential in the field of aerial survey as is the difficulty of finding it in a library. The present paper can best be considered as a set of footnotes to Jolly's paper and Cochran's book. Its immediate stimulus was a private debate between M. Norton-Griffiths and myself over the efficacy of fragmented-cluster sampling in aerial survey, a question now happily settled in the former's favour by the investigation reported here. I am grateful to D. S. Robson, G. C. Grigg, G. M. Jolly, M. Norton-Griffiths and G. A. F. Seber for criticising a previous draft of this paper. The work was funded by the J. Wildl. Manage. 41 (4):1977 605 This content downloaded from 157.55.39.25 on Mon, 25 Jul 2016 04:10:53 UTC All use subject to http://about.jstor.org/terms 606 SAMPLING IN AERIAL SURVEY Caughley