Winter temperature changed next-summer counts of the same bird species only in states where the species stayed for winter, on 4,758 North American survey routes in 1967-2024

Question: does local winter temperature change the next-summer count of a bird species only where the species spends the winter? Data: USGS Breeding Bird Survey 2025 release (6,385,153 year-to-year changes in log(1+count), 373 land-bird species, 4,758 routes, 1967-2024), NOAA GHCN_CAMS December-February temperature anomalies from a route trend, and a new winter presence index for each species in each US state and Canadian province from GBIF records (December-February records divided by June records, corrected for seasonal reporting effort). Models had route x year and species x year fixed effects, so observer, survey-day and species-wide annual effects are removed; winter-absent species on the same survey are the negative control. Results: in 40 species that stay for winter in some states and leave other states, a +1 C winter was followed by +1.09% (95% CI 0.75 to 1.43) on routes where the species stays and +0.19% (-0.05 to 0.44) on routes it leaves; random-effects mean within-species difference +0.68% per C (0.22 to 1.14), 27 of 40 species positive, sign-flip p = 0.007. Over all species, winter-present species changed 0.80% per C more than winter-absent species (0.62 to 0.97); the route-level slope was 6.5 permutation SDs from zero (year-field permutation p = 0.0005). Winters colder than -2 SD gave -4.0% (-5.2 to -2.7). The slope was clear in the Central and East regions and in 1967-1985, 1986-2005 and 2006-2024, but uncertain in the West (0.30, -0.14 to 0.75) and Canada (0.19). About half of the change reversed in the next year and the third-year level effect was uncertain. Small species (< 20 g) and Carolina Wren (+8.6% per C) were most sensitive. This independently replicates three campaign studies that used species-level classes. Limitations: counts are an index without a detection model; the placebo (winter after the survey) gave -0.27% per C (permutation p = 0.08); state-clustered SEs are about 1.5 times smaller than permutation SDs; the GBIF index is state-level and mostly from recent records; monthly mean temperature does not measure snow or short cold spells.

Breeding Bird Survey · winter severity · weather extremes · winter presence · GBIF · within-species test · negative control · observation bias · resident birds · migratory birds · GHCN_CAMS · AVONET · fixed effects · permutation test · population dynamics · climate change · North America