Species distribution model performance improves when habitat characterizations are centered on detected individuals instead of observers

Species distribution models (SDMs) link species occurrence to environmental characteristics to predict suitable habitats beyond known occurrences. The conventional procedure to fit SDMs for individual organisms detected at some distance away from observers is to characterize species’ associated habitat based on observer’s survey location. However, each surveyed individual may be detected in habitats distinct from those where observers are located. Here, we compared environmental variables centered on the observer and individual bird locations and the consequent effects on SDMs performance. We utilized remote sensing data on observer- and bird-locations to characterize habitat at three radii (pixel radius: 30-m; fixed radius: 100-m; species-specific effective detection radius). We trained Poisson boosted regression tree models for 105 bird species from structured professional surveys. We evaluated models’ predictability with Kendall’s rank correlation coefficient and used linear mixed-effect models to measure the effect of characterization locations and radii. Models based on bird locations exhibited a median increase of 22.9% in predictive performance, demonstrating higher Kendall’s rank correlation coefficients than those based on observer locations, leading to more reliable prediction maps. SDMs of habitat specialists and generalists performed better when habitat characterization was centered on bird instead of surveyor locations. A higher percentage of habitat specialists (72%) than generalists (55%) showed better model performance in bird-location than in observer-location models. Across radii, fixed radius generally performed better than species-specific effective and pixel radii. Our findings emphasize the importance of prioritizing habitat characterizations based on detected individuals’ locations to enhance model performance and improve species distribution predictions.