Abstract:

Emerging infections and environmental disruptions increasingly threaten wildlife and ecosystem health.
Free-ranging wildlife often serve as early indicators of ecological instability, making timely detection of
morbidity and mortality events critical for early warning. Yet, existing systems lack the analytical capacity for
real-time outbreak detection. We present WildAlert, an AI-driven early warning system that integrates fine tuned BERT-based natural language processing models with unsupervised anomaly detection framework to
identify unusual wildlife health events using real-time pre-diagnostic clinical data from wildlife rehabilitation
organizations. The NLP module achieved high accuracy across clinical classifications and circumstances of
admissions, enabling a pre-diagnostic syndromic surveillance framework. Retrospective validation
demonstrated that WildAlert anomalies frequently coincided with or preceded confirmed morbidity events,
including highly pathogenic avian influenza (HPAI), harmful algal bloom–associated toxicosis, cold-stunning
in sea turtles, mass stranding events, West Nile virus, and mycoplasmosis. WildAlert establishes the world’s
largest standardized, near real-time wildlife health surveillance system, transforming wildlife rehabilitation
clinical records into actionable intelligence capable of detecting anomalies across taxa and regions, often
before other surveillance methods. WildAlert provides a transferable analytical framework and scalable One
Health model linking biodiversity monitoring, zoonotic disease preparedness, and ecosystem-linked
environmental threats, with implications for conservation, public health and environmental hazard response.

Full Publication: https://www.biorxiv.org/content/biorxiv/early/2026/04/10/2026.04.07.716505.full.pdf