Bias in density estimates from avian point-count surveys: Prospects for post-hoc corrections using calibration data

Abstract Point-count surveys are commonplace in avian monitoring and research but were initially designed to collect data on avian relative abundance rather than densities. However, the increasing realization that detection biases influence conclusions from point-count surveys has given rise to several statistical approaches to correcting such biases. Distance sampling allows for correction of biases in perceptibility and estimation of avian densities. A key assumption is that distance estimation is accurate, but experimental evidence suggests observation error is large, particularly for estimates based on acoustic detections. We had observers estimate distances to 128 singing birds and one mammal while other staff systematically tracked calling individuals and measured the distance from the observer. Log-log regression showed distance estimation errors and uncertainty increased with increasing distance from the observer. We modeled the relationship between true effective detection radius (EDR) and estimated EDR (including distance estimation error). Simulations showed that species with small EDRs (30 m) had densities underestimated by 23% on average (SD = 11), while densities for species with large (130 m) EDRs were biased upward by 59% on average (SD = 25), but these biases could be corrected post-hoc. Applying the same post-hoc corrections in species-habitat regression models for 4 species of boreal forest birds found that our post-hoc corrections increased EDRs by up to 26% and decreased estimated bird densities by up to 38% for highly detectable species, while effects on less detectable species were more subtle. We suggest collection of more known-distance data that may allow post-hoc corrections to reduce bias in avian density estimates from point-count surveys.