(b) Stable hydrogen isotope analyses
We used stable isotope analysis of hydrogen (δ2H) in feathers and claws to infer breeding and overwintering latitudes, respectively (Hobson, 1999; Rubenstein & Hobson, 2004; Wunder, 2010; Hobson, Van Wilgenburg, Wassenaar, & Larson, 2012; Bowen, Liu, Vander Zanden, Zhao, & Takahashi, 2014). Whole feathers and claws were cleaned using a 2:1 chloroform: methanol solution to remove external oils and contaminants. Prior to analysis, we controlled for the effect of exchangeable hydrogen by forcing isotopic equilibration with a water vapor of known isotopic composition in a flow-through chamber system at 115°C (Schimmelmann, 1991; Sauer, Schimmelmann, Sessions, & Topalov, 2009). The hydrogen isotopic composition of feathers (δ2Hf) and claws (δ2Hc) was then analyzed using a thermal conversion elemental analyzer coupled with a ThermoFinnigan Delta Plus XP isotope ratio mass spectrometer at the Indiana University Stable Isotope Research Facility. The δ2H values are reported in standard per mil notation (‰) relative to VSMOW (Vienna Standard Mean Oceanic Water) using two reference materials: USGS77 (polyethylene powder) and hexatriacontane 2 (C36n-alkane 2). Analytical precision was ± 0.6 ‰ for δ2H values. We then calculated the isotopic composition of the non-exchangeable hydrogen per sample (Schimmelmann, 1991; Schimmelmann, Lewan, & Wintsch, 1999).
Based on feather and claw δ2H values, we performed Bayesian geographic assignments of individual birds during the breeding and overwintering months, respectively, using the assignR package in R (Ma & Bowen, 2019; R Core Team 2019). Growing season and monthly precipitation isoscape rasters from waterisotopes.org (Bowen & Revenaugh, 2003; Bowen, Wassenaar, & Hobson, 2005), monthly precipitation amount rasters from CHELSA (Karger et al., 2017a, b), and resident dark-eyed junco isotopic data (Hobson et al., 2012; Becker et al., 2019) were utilized to generate geographic assignment probability maps. Latitudinal estimates were extracted from cells within the top 10% highest posterior probability using the raster package (Hijmans et al., 2019).