A new analytical framework for Maxent species distribution models unveils complex spatiotemporal suitability patterns for two migratory seedeaters (Aves: Sporophila) of conservation concern

Abstract

Maxent species distribution models (SDMs) for rare, migratory taxa can be hindered by methodological and analytical issues. To overcome these problems, we developed an analytical framework to build SDMs for key stages of the annual lifecycle of the Marsh Seedeater and the Black-bellied Seedeater, two Neotropical grassland bird species of conservation concern. Occurrence data were compiled from multiple sources and ten environmental variables were used as predictors. We built SDMs with different spatial partitions, feature classes, and regularization multipliers, and devised a procedure to select models with high discrimination ability, low overfitting, statistically significant performance metrics, and high biological realism. From a total of 992 SDMs, 22 fully met the selection criteria for both species. These selected SDMs were equally or less overfit than non-selected ones and were mostly trained with the checkerboard 2 and n − 1 jackknife partitions. For each modeling scenario, we projected and interpreted composite models that emphasized areas of both consensus and divergence across individual predictions, rather than choosing a single model from the set of final SDMs. In the breeding season, areas of highest suitability occurred in restricted, disjunct sectors of the Campos and Pampas grasslands for the Marsh Seedeater and the highland grasslands of southern Brazil for the Black-bellied Seedeater. In the winter, high suitability for the Marsh Seedeater occurred in the western and northern Cerrado and the Pantanal, while in the spring migration, high suitability for the Black-bellied Seedeater was mostly concentrated in the southeastern portion of the Cerrado. The restricted breeding distribution of the Black-bellied Seedeater suggests that its conservation status should be reviewed. Both species shift their climatic niches and track their habitats throughout the year, responding to a few topographic, land-use, and climatic variables that represent different niche components. By addressing major modeling issues in a three-step model selection procedure, we were able to project precise and biologically interpretable SDMs with the level of complexity required for each modeling scenario. Our framework is readily replicable and can be used to unravel intricate spatial suitability patterns of organisms whose distribution undergoes temporal shifts throughout the year.