Monitoring forest bird communities for impact assessment: The influence of sampling intensity and spatial scale

There are several applications for which one might monitor avian communities but one of the main ones is to evaluate the impact of an anthropogenic disturbance (e.g. logging). There has been relatively little attention paid to how sensitive such monitoring might be to sampling strategy, which refers to the temporal frequency and spatial extent over which the monitoring is conducted rather than the actual method by which the data are collected (i.e. strip transect, point counts, etc.). By using data collected during 92 visits spaced over three successive years at one site, 48 different sampling strategies (‘schedules’) are emulated by using resampling techniques to investigate the sensitivity of outcomes to transect size (3 classes, ≈17, ≈34, ≈50 ha) and sampling intensity (once-off; one or more times per season; one, two or three years). Sensitivity is gauged by reference to the accuracy of estimates of species richness and the maximum total density. A key point is that the analyses are to be based only on species recorded in the pre-impact phases of the monitoring; thus, losses of species or individuals of these species cannot be offset or obscured by increases in disturbance-tolerant or exotic species when judging whether there has been a deleterious impact. For the kind of forest and avian community considered here, single-visit surveys performed uniformly badly on every criterion, while results for the smallest spatial extents (≈17 ha) deviated greatly from outcomes for larger transects. A recommended sampling schedule derived from these analyses is two visits every season for two successive years at the largest spatial extent (i.e. eight visits at ≈50 ha). This would form the basis by which to assess ‘significant’ declines in richness or maximum total density, with the entire survey repeated prior to and following the putative impact. These assessments require the calculation of the magnitudes of declines that might arise by chance from sampling the unperturbed community and corresponding power estimates for specified changes.