Climate suitability trends did not predict species differences in North American BBS bird population trends in 1981–2010 or 2011–2024 (315 species, models fitted on 1981–2010)

Question: Stephens et al. (2016, Science) reported that the climate suitability trend (CST) from species distribution models (SDMs) predicts bird population trends in 1980–2010. Does this hold out of sample in time? Data and method: North American Breeding Bird Survey (BBS) 1966–2024 standard runs (127,646 runs, 5,306 routes) and CRU TS 4.10 climate. For 315 landbird species, we fitted one SDM (spline logistic model of GDD5, coldest-month temperature and AET/PET) on 1981–2010 route presence (median spatial-block cross-validated AUC 0.94). We calculated the CST from annual climate. Population trends came from Poisson models with route-observer fixed effects. We compared CST and trends in 1981–2010, in 2011–2024 (out of sample), in 1968–1985, in 1997–2010 and in 15 rolling 14-year windows. We used family bootstraps and six robustness variants. Results: at species level, CST did not predict trends: Spearman rho = 0.08 (95% CI −0.07 to 0.20) in 1981–2010 and 0.04 (−0.07 to 0.17) in 2011–2024. All variants gave 0.00 to 0.08 in 2011–2024. Rolling-window rho was −0.08 to 0.18, and its interval was above zero in 3 of 15 windows. Within species, the slope of regional (BCR) trends on regional CST was 0.22 (0.07 to 0.36) in 2011–2024 with species and region effects. But it was 0.19 (−0.11 to 0.48) in 1981–2010 and −0.22 (−0.51 to 0.12) in 1968–1985, and its interval was above zero in 3 of 15 windows. The climate impact indicator increased by 0.36% per year in 1981–2010 and 0.18% per year in 2011–2024, but the species trend difference behind it had intervals that included zero. The 1981–2010 trend (rho = 0.32) and the species temperature index (rho = 0.16) predicted 2011–2024 trends better than CST (rho = 0.04). Limitation: this is an independent reimplementation (different climate data, SDM, CST aggregation and trend estimator from Stephens et al.), so method differences can contribute. The tests were for North America only, and the regional intervals do not include spatial correlation.

climate suitability trend · species distribution model · Breeding Bird Survey · population trends · out-of-sample test · climate impact indicator · North America · birds · replication