Bird dataset
Bird abundance data was obtained from eBird, which is an online repository based on amateurs’ records. Observations are reviewed by experts and are then aggregated to produce abundance estimates (Sullivan et al. , 2009). We compiled all records falling within continental United States including Alaska, and excluded those i) with georeferencing errors, ii) not identified to the species level, iii) without abundance information available. We used a second filtering process to further homogenize the quality of observations. First, to reduce variation in birds’ detectability, we only kept records with less than 5 hours of observation, less than 5 km of traveled distance, and 10 or fewer observers (as recommended in Strimas-Mackey et al., 2020). Second, we only considered data ranging from 1969 to 2016, which is the greatest recording effort period (Appendix S2) for which we have environmental data available (see below). Third, we used BirdLife species distribution maps (BirdLife International and Handbook of the Birds of the World, 2016) to exclude sightings of birds recorded outside their home and breeding areas (for example, in migratory routes). Fourth, we only kept those species that presented at least 150 records to minimize type I error in QR (Cade et al. , 2005). Lastly, we estimated relative abundances as the species’ local abundance divided by the minutes of observation (i.e., abundance weighted by sample effort) and excluded outlier records (i.e., relative abundance values 100 times greater than the observed value in the 75th quartile). Our final dataset contained 49,835 records belonging to 114 bird species (an average of 437 records by species, with a minimum of 152 records for the speciesPasserculus sandwichensis and a maximum of 1,758 records for the species Zenaida macroura ).