Phylogeny and species traits predict bird detectability

Submitted by editor on 26 September 2018. Get the paper!

by Péter Sólymos (solymos [at] ualberta [dot] ca)

Video abstract:

Birds use songs to attract potential mates and defend their territories against competing males. The same songs are utilized by ornithologists when counting birds to estimate their abundances. But counting birds is not always easy, most often we run into the problem of detection error. Imperfect detection is a result of two main processes: first, not all the birds care to sing during the 3 or 10 minutes of the average field survey; and second, even if they do sing we might not hear them because they are too far away.


These well-known issues veil the true number of birds present, and accounting for the missed birds is not too easy. This is especially true for rare species. The other issue is that it can be extremely difficult to validate any kind of result because detectability is such an elusive quantity. This is so because the number of birds counted in the field is only an index of the true abundance that we have no way of observing directly.


Our results recently published in Ecography can help in better estimating detectability of rare species, and in validating results based on models accounting for imperfect detection. We demonstrated that both components of the detection process -- the signalling behaviour expressed as singing rate, and transmission reliability as reflected in detection distance -- could be predicted by traits and phylogenetic relatedness.


Singing rate was best explained by the combination of phylogeny and trait variables. Long-distance migrants sang at higher rates than resident birds. The evolution of behavioural traits that determine singing rates appears more labile, and we suspect many different ecological processes at play, such as density-dependent singing behavior.


Detection distances were best explained by the traits and not by phylogenetic relatedness. Larger detection distances characterized species that have larger body mass, and as a result, louder songs at lower pitches. Migratory species and species associated with open habitats had larger detection distances, too. These findings suggest that detection distance is mostly constrained by anatomy and that potential song volume is limited by body size.


As it turns out, detectability is not some obscure quantity after all, but it is strongly correlated with the traits and the relatedness of species. This provides the means to validate detectability estimates and to utilize these relationships to account for observation error when analyzing rare and data deficient species. For example, phylogeny and traits might be used to define groups of species for which the combined detection probabilities can be estimated jointly, or estimates available for closely related species can serve as a prior, alleviating sample size limitations.


To illustrate that this is a real and acute problem, let me take a quick detour and draw attention to some other work I am involved in where such results will be extremely helpful. Estimating population sizes of species is paramount for conservation, and it requires large scale data sets. Large scale data sets are rare, and quite often we need to integrate disparate smaller data sets, as it was done as part of the Boreal Avian Modelling (BAM) project ( The project aims to further continental scale avian conservation through the integration and analysis of point count data collected across northern North America.


The approach we have developed for this task was described in a series of papers (Matsuoka et al. 2012, Sólymos et al. 2013, Sólymos et al. 2018), and as a by-product we got the estimates that we used in the current Ecography paper. This means, on one hand, that validation of the estimates is now officially done (ta-da!). On the other hand, we – and other organizations also in the business of population size estimation, like Partners in Flight – still have a handful of species for which detectability estimates do not exist or could use a boost. In this sense, there is a real need for the phylogeny and trait based approach we described in our paper. Achieving this goal is only made easier by the R package that comes with our paper (read more about that here and here).


In general, it cannot be determined whether phylogeny, trait variables, or the combination of those will provide more information. The answer will likely depend on the nature of the taxonomic group, evolutionary mechanisms, and the detection process at play. Our results will help in disentangling how complex ecological and evolutionary mechanisms have shaped different aspects of detectability in boreal birds. But more work is required to better understand how the phylogenetic and trait-related signals affect diversity and other community metrics, and how to best correct for observation error.