Fig 1.
Map of study region showing relative positions of monitoring areas for project BirdSafe and Important Bird Area Landbird Monitoring Program.
Map of the greater Minneapolis-St. Paul metro area in Minnesota, USA. Project BirdSafe monitoring areas are indicated by shaded purple regions. The Mississippi River Twin Cities Important Bird Area (IBA) is outlined in green and the sites within the IBA that were monitored as part of the Landbird Monitoring project are indicated with solid green.
Fig 2.
Flowchart diagraming modeling process.
Modeling was completed in 3 stages; the first determined the best distribution to use for the dataset, the second determined the most parsimonious combination of phylogenetic taxa, and the third determined which natural history fixed effects were present in the most strongly supported models. All model selection was based on the information-theoretic approach, and used Akaike Information Criterion (AIC) values to determine the best model.
Table 1.
Table of the 4 taxonomic models tested with ΔAIC values as well as a null model.
Table 2.
Table of the forward model selection results with the base phylogeny model and the top models from the model selection process, based on minimum ΔAIC.
Fig 3.
Caterpillar plot of the 13 supercollider species random effects estimates ± shrinkage estimates within the top model: Abundance+Timing+(1|Family/Genus/Species).
Random effect estimates indicate the estimated level of risk for each species after accounting for the effects of Abundance and Timing as approximated by the model. Shrinkage estimates are a measure of variance for the random effect estimates. Species were classified as supercolliders when random effect estimates – shrinkage estimates > 0, indicating high confidence in a non-zero estimate of risk of collisions greater than expected by Abundance and Timing alone. Within the supercolliding species, there is a clear gradient from highest risk, White-throated Sparrow (Zonotrichia albicollis) and Tennessee Warbler (Leiothlypis peregrine), to lower risk, Nashville Warbler (Leiothlypis ruficapilla) and Least Flycatcher (Empidonax minimus).
Fig 4.
Scatterplots of correlations of previous species risk assessments with the species net vulnerabilities from our analysis.
P-value, Pearson product-moment correlation (r), and linear trendline shown for each graph. Relative abundance data for our analysis came from point counts conducted within 36 km of the location of collisions. Collision data came from daily monitoring of buildings along two highly urban routes within Minneapolis and St. Paul, MN, USA. (A) Species vulnerabilities calculated by Arnold and Zink [4] are weakly correlated with our species net vulnerabilities. Arnold and Zink used Partners in Flight estimates to determine relative abundances of species and used collision data from select buildings as well as communication towers for their analysis [4]. (B) Species risk assessment calculated by Loss et al. [2] was not correlated with our species net vulnerabilities. Loss et al. also used Breeding Bird Survey data to assess relative abundances of species but collision data came from building collisions in residential and urban areas throughout North America [2]. (C) Aggregate species risk levels (calculated by using our data and mimicking the Arnold and Zink analysis [4]) are strongly correlated with our species net vulnerabilities.