Home >  Methods >  Avian Data Analysis >  Roadside Detection Bias


 Avian surveys collected along roads (e.g. North American Breeding Bird Survey) result in a different count of birds when compared to equivalent surveys conducted away from roads (off-road surveys).

BAM found evidence for roadside bias in survey counts for many boreal songbird species. This indicates that bird detection changes when surveys are conducted along a road compared to surveys conducted within a habitat type such as a forest. BAM is interested in this issue because we use data from both roadside and off-road surveys in many of our analyses to maximize our available data

BAM found bias in 80% of the 85 species of boreal songbirds that were studied (after controlling for differences in the habitats sampled). Positive roadside bias is particularly common, meaning that surveys conducted from the roadside result in higher counts than those conducted off-road.

Thus, analyses using combinations of roadside and off-road survey data must control for roadside bias.

Roadside bias in avian survey was estimated for 84 species using species-specific generalized linear mixed models. Positive bias was most common (73% of species with significant bias) and occurs when surveys along roadsides result in higher counts than surveys in off-road areas.

One potential cause of positive roadside bias is that birds can be seen and heard at greater distances along roadway clearings than through vegetation in off-road areas. BAM is addressing this by using distance sampling to estimate differences in Effective Detection Radius (EDR) between roadside and off-road counts. This approach is proving useful because EDRs tends to be larger for roadside than off-road surveys, thereby balancing the positive bias in the roadside counts. BAM will use this difference in EDRs when combining roadside and off-road surveys in our future analyses of avian densities and population sizes across the boreal region.

BAM is also exploring whether the type of road the survey is conducted on influences the magnitude of roadside bias. For example, a survey conducted along a narrow dirt road might result in a similar count to a survey in an off-road area because the amount of vegetation cleared is minimal. Conversely, the larger area cleared for a wider paved road might be expected to result in lower roadside counts of forest dwelling species and higher counts of edge species. Thus, information on road type may prove quite useful in controlling for roadside bias in the survey counts.