Abstract
Ecological forecasting in data-poor ecoregions remains a critical challenge for biodiversity conservation. Traditional modeling approaches require extensive time-series or observational data that are often unavailable for many threatened species and remote ecosystems. Here, we present Species Shadows, a novel framework leveraging Generative Adversarial Networks (GANs) to synthesize realistic species distribution and demographic trajectories. By training GANs on proxy datasets (e.g., related taxa, functional analogs, and environmental covariates), we generate high-dimensional synthetic data that illuminate possible extinction trajectories. Results suggest that GANs can effectively capture nonlinear extinction signals and project risk patterns in under-sampled regions. The framework promises a scalable alternative for conservation prioritization where empirical data are scarce.
