The risk of birds' functional diversity loss with climate change uncovered using deep learning hurdle models

Climate change and sea-level rise (SLR) increasingly threaten coastal biodiversity, yet most predictive assessments focus on species distributions while overlooking changes in ecosystem functioning. We developed a predictive ecological indicator framework that integrates a generative Variational Gaussian Mixture Autoencoder (GMVAE) hurdle model with Trait Probability Density analysis to generate continuous, pixel-level functional diversity indicators from AI-predicted multispecies abundances. The framework extends conventional functional diversity assessment from discrete survey sites to spatially continuous estimates of functional richness, redundancy, evenness, and divergence under historical and future climate scenarios. Applied to Gulf of America bird communities, the model recovered species distributions in sparsely sampled regions and revealed that the greatest biodiversity losses occurred not along exposed shorelines but within landward coastal transition zones. These areas exhibited declining functional richness and increasing redundancy, indicating contraction of functional trait space and ecosystem homogenization. In contrast, protected coastal areas maintained high functional diversity and acted as potential climate refugia. Our framework provides a scalable approach for forecasting ecosystem functioning changes and supporting climate adaptation and conservation planning with functional trait diversity indicators.