Addressing sampling bias in counting forest birds: a West African case study

Abstract

No one bird survey technique is perfect. Either the assumptions made by a technique are difficult to meet in the field, or there are biases due to the observer, the birds themselves, the landscape, or the nature of the habitat. These include survey effort, time of day, time of year, edge effects, vegetation structure, and variation in detectability functions. Biases in methodologies are only a problem if they are systematic with respect to the variables that are being tested. If, however, biases are randomly distributed with respect to the variables under consideration, they cannot affect a positive result, although a negative result may occur because of the noise they may introduce into the analysis. Therefore, it is crucial to determine and minimise biases in any study. We use extensive line transect census data from a study of the effects of habitat fragmentation on forest birds in Nigeria to illustrate the range of possible biases and their effects, and then how these may be accounted for. Asymptote analysis showed that, although more bird species were recorded in larger fragments because transects in those fragments were longer, this bias had little overall effect as long as transects were longer than a few hundred meters. The number of bird species declined significantly and substantially between 07:00 and 11:00. The effects of edge on number of species was only significant for the first 200m from the edge. Vegetation variables did not vary significantly across sites, suggesting that variation in overall detectability could be ignored, but there was considerable variation in the detectability functions of different species. We recommend the following procedures for surveys of West African rain forests: a minimum transect length of 1 600m; sub-division of transects into shorter sections (100m–200m) along with vegetation measurements for each section to allow for the assessment of the effects of habitat variation; transects to start at least 200m from the edge of the forest; the use of speciesspecific detectability functions if they are to be used at all; and for all transects to reach their mid-point at the same time each day, preferably early in the morning.

Ostrich 2007, 78(2): 281–286