By use of balanced experiments amenable to analysis of variance we explored the effects of several factors on the accuracy of aerial survey estimates of animal density. Speed, height above ground, transect width, and observers had significant effects, whereas time of day, fatigue of observers, and length of survey were less important. We tested the hypothesis that a regression of observed density on speed, height, and transect width could be extrapolated backward to estimate true density at zero values of these survey variables. The results were generally consistent with this expectation. The uses of this technique are outlined, with examples, in the context of correcting an observed density to an estimate of true density, of calibrating one observer against another, and of comparing the results from aerial surveys flown at different speeds and heights, and with different widths of transect. The field experiments utilized red kangaroos (Megaleia rufa) at unknown densities and domestic sheep at known densities. Laboratory experiments were performed on dots of known density projected onto a screen. Before the model is accepted as generally applicable it must be tested against several other species in a variety of habitats. J. WILDL. MANAGE. 40(2):290-300 During the early stages of World War II, Royal Air Force crews experienced great difficulty in hunting German submarines in the Bay of Biscay. Their flight path seldom passed near a surfaced submarine and even when it did they often failed to see it. This and related problems forced the birth of an infant branch of applied mathematics called operations research. It swiftly reduced the problem of hunting submarines to its theoretical essentials (Morse and Kimball 1960). A similar problem, amenable to similar treatment, is posed by the difficulty of estimating animal populations from the air. Over the last decade evidence has accumulated steadily to show that aerial surveys provide underestimates, often gross underestimates, of animal density. These data from a broad range of species and habitats were summarized by Caughley (1974). While that paper was in press, further evidence appeared (LeResche and Rausch 1974) showing that the accuracy of counting moose (Alces alces) from the air was influenced significantly by a number of factors. Experienced observers counted only 68 percent of moose on a quadrat; inexperienced observers counted 43 percent. This degree of error is completely unacceptable when the aim is to calculate population size as a prelude to estimating a sustained yield, but there are strong grounds for suspecting that the error cannot be eliminated by refining techniques, by elaborating the survey design, or by hoping that a technological solution such as greatly improved infrared sensing is just around the corner. The alternative is to accept that the best of observers, in the most favorable conditions of observation, will fail to observe many of the animals at which he is staring. This was the starting point of an approach introduced previously (Caughley 1974) in which partial regression analysis was suggested as a means of estimating true density. Since sightability declines with increasing speed, height above ground, and transect width, a regression of observed density on these variables should estimate true density as the y-intercept constant a. 1Study supported by the Australian Research Grants Committee. 290 J. Wildl. Manage. 40 (2) :1976 This content downloaded from 157.55.39.255 on Mon, 01 Aug 2016 06:11:50 UTC All use subject to http://about.jstor.org/terms EXPERIMENTS IN AERIAL SURVEY * Caughley et al. 291 Table 1. Experimental design for the field surveys. Area Transect Sampling surveyed Height Speed width unit Expt. (km2) (m) (km/h) (m) (0.5 km2) Species Observer 1 50 46 129 50 858 red kangaroos G.C., R.S. 91 161 100 858 183 193 200 858 2 70 46 161 100 864 red and grey G.C., D.S. 91 193 200 864 (Macropus giganteus) 183 193 200 864 kangaroos 3 16 46 161 100 576 sheep G.C., D.S. 91 193 200 576 183 193 200 576 The purpose of the present paper is to explore this hypothesis in greater depth by determining the relationship between observed density and the survey variables, and the extent to which these relationships change as viewing conditions change. Finally, we will show that factors can be calculated by partial regression to correct standard aerial survey estimates for visibility bias. This paper reports the results from four experiments in aerial survey. Since the design of each, and the need for the experiment itself, was dictated by the results of the previous experiment, these will be reported in chronological order. We are grateful to T. Mathews who piloted for the experiments; to G. Wilson for survey piloting; to R. Goldsby, E. McLaughlin, and N. Crisp for providing us with the facilities of their stations and for their hospitality; and to G. Courtice for assisting in the simulation experiment.
Read full abstract