Bias in transect counts of forest birds: An agent-based simulation model and an empirical assessment

Abstract

Transect counts are often used to estimate broad-scale densities of conspicuous organisms, notably birds. However, these counts are prone to numerous biases, which are difficult to disentangle in purely empirical studies due to observer-related and contextual uncertainty. To measure how different biases combine, we constructed a model that simulates observer movement across a theoretical landscape that is inhabited by birds moving within their circular territories. The model was parameterized based on data from Estonian forests where, as an additional field test, we conducted actual transect counts of bird assemblages that had been territory-mapped based on multiple visits. The simulations revealed that biases vary significantly among bird species. In dense populations, accurately locating detections can be a key issue that can produce either over- or underestimation when combined with observer speed. Counts of sparsely distributed, poorly or only seasonally detectable species appeared most challenging. Compared with these field errors, record interpretation had smaller effect on the density estimates. The test counts confirmed variable underestimation of the territory-mapped bird densities and a resulting underestimation of local species richness. We conclude that biases of single-visit transect counts cannot be easily corrected to reveal true densities of birds and should be considered as abundance indices. The capacity to detect trends in repeated counts is profoundly affected by changes in observer persona and may be sufficient in common species only. We encourage using agent-based models to analyze the behavior of researchers who collect ecological data as a tool to inform methodological standardization and researcher training.