Amphibian spatial distribution modeling, what are we missing? Comparing adult-only and tadpole models of habitat suitability

Introduction Species distribution models (SDMs) are essential tools for assessing the conservation status of species, because they provide crucial insights into their ecological requirements and potential geographic distribution. Most SDMs rely on the correlation of presence-only records with environmental predictors and are constrained by their assumed accessible area. However, many species exhibit complex life cycles in which developmental stages occupy distinct habitats and respond differently to environmental conditions. In anurans, most species have biphasic life cycles, with free living larvae that differ in their responses to environmental constraints from adults. Here, we evaluate how suitability estimates from SDMs may differ between larval and adult stages of Mesoamerican Hylid frogs. Methods We modeled the current and future distributions of ten Mesoamerican hylid frog species by integrating correlative models, microclimatic simulations, and physiological limits. Adult and tadpole distributions were estimated using MaxEnt under current and future climate projections. To incorporate larval constraints, we simulated seasonal pond availability across Mesoamerica using a mechanistic microhabitat model. Finally, adult and tadpole models were filtered based on critical thermal maxima (CTmax) thresholds. Results The ten examined species are projected to experience substantial range contractions —up to 30% under non-dispersal assumptions—, although some widespread species may increase their distribution when dispersion is modeled. As expected, tadpole-based models predict smaller suitable areas than adult-based models (on average only ~4% of the adult current ranges), and larval habitats appear more spatially fragmented. Discussion Species distribution projections suggested reductions in suitable areas under non-dispersal assumptions, with dispersal only partially offsetting those losses and CTmax filtering further reducing climatically suitable areas. Incorporating physiological limits revealed that correlative SDMs often overestimate suitability of habitat –especially for widespread or thermally sensitive species— and that adult- and tadpole-based models diverge strongly due to stage-specific ecological constraints. Tadpole distributions were tightly limited by the availability and thermal stability of temporary ponds, underscoring the need for larval data, improved sampling, and trait-based modeling. Although data scarcity may bias our results for tadpole models. Overall, integrating life-stage ecology, dispersal uncertainty, and thermal physiology is essential for generating realistic forecasts in order to responsibly apply SDMs in conservation planning.