Comparative Analysis of Correlative Modeling Methods in Predicting North American Bird Abundance

Over the past decade, correlative ecological niche models (ENMs) have been increasingly used to infer species abundance through the estimation of continuous scores, which can be interpreted as environmental suitability. However, little attention has been drawn to the assumptions and conceptual differences between these methods and how they affect abundance predictions. Here, we address this gap by comparing the performance of nine ENM methods, including some recently proposed approaches, in predicting the abundance of North American birds using standardized count data. Relationships between abundance and environmental suitability were generally positive, with average correlation coefficients greater than 0.18. This implies that environmental suitability can partially explain abundance, as it is likely influenced by other factors such as biotic interactions and dispersal constraints. Although no method proved to be significantly superior, newer modeling approaches (e.g., MaxLike and Gaussian models) showed higher average correlations compared to older ones (e.g., Boosted Regression Trees and Random Forest), underscoring the comparable performance across modeling approaches. Results were consistent across taxonomic groups (e.g., Passerellidae), an indication that the observed patterns are robust within specific clades. Furthermore, no significant differences were found between models using complete distribution data versus breeding area data, or annual versus seasonal environmental variables, highlighting that simpler data inputs may be enough for abundance prediction in many contexts. Taken together, these findings reinforce the potential of ENMs to provide reliable insights into species abundance patterns.