BBS route trends in 2007–2024 do not decrease with local abundance for 335 North American bird species, unlike eBird model trends at the same places

Question: Johnston et al. (2025, Science) used eBird models and reported that bird populations decreased most quickly where species were most abundant (negative slopes for about 83% of 495 species). Does the independent North American Breeding Bird Survey (BBS) show the same pattern? Approach: We used the BBS 2025 release (2007–2021 and 2010–2024) for 335 species. For each species, we fit a Poisson model with route × observer intercepts, a first-year-observer effect, region-specific trends, and a route trend that changes with log10 local abundance. Standard errors are robust and clustered by route. Abundance came from BBS years that were not used for the trend (cross-fitting), from the eBird 2014 abundance values, from the previous decade, or from the first window years. A simulation on the real survey structure measured the bias of each design. Results: In the eBird estimates at cells within 20 km of BBS routes, the slope of trend on abundance was negative for 81% of species (mean −0.35 percentage points per year per tenfold abundance). In BBS cross-fitted data, it was negative for 38% (mean +0.66, 95% CI +0.45 to +0.86) in 2007–2021 and 45% (+0.31) in 2010–2024. BBS trends regressed on the eBird abundance values gave 41% negative (+0.43). The simulation shows that the cross-fitted design can detect the eBird slope (true −0.32 gave −0.57) and has a small bias toward it. Designs with abundance from before or from the start of the window gave negative slopes (72% and 93% of species). Simulations gave negative slopes of the same size when the true slope was zero, so these designs are not evidence. eBird cell trends predicted BBS route trends with a mean calibration slope of 0.33. Limitations: BBS covers roadsides only and gives count indices. Per-species estimates are noisy, so the conclusions are averages across species. Our eBird regressions are simplified versions of the published model. We did not identify why the two sources disagree.

Breeding Bird Survey · eBird · population trends · abundance · regression to the mean · cross-fitting · North American birds · replication · spatial heterogeneity · biodiversity monitoring