Machine Learning Enables Rapid Assessment of Disease Vulnerability in a Threatened Cetacean Population

Abstract

Cetaceans (whales and dolphins) are important ecosystem sentinels but face growing threats from major disease-related mortality events expected to intensify under climate change. Because both environmental factors (temperature, salinity) and demographics (age, sex) influence health and disease risk, understanding these relationships is essential for effective management. Direct health assessments are challenging in cetaceans, but skin lesions can indicate active infection and tooth-rake marks reflect social stressors that increase transmission risk. Yet, traditional photographic analysis of these indicators is inefficient, creating processing bottlenecks that limit timely evaluation of population health. To address this gap, we applied machine learning to rapidly assess lesions and rake marks in Tamanends bottlenose dolphins (Tursiops erebennus) photographed in the Chesapeake Bay, a known hotspot for disease-related die-offs. This represents the first analysis of environmental and demographic contributions to dolphin health in this region. We found significant negative relationships between lesion prevalence and both temperature and salinity for some lesion types. Adult males also showed higher rake mark coverage than adult females and calves. These patterns suggest dolphins in colder, fresher waters may face elevated disease risk, while adult males may be particularly vulnerable to behavioral stress and related health consequences. Our findings are consistent with prior studies, lending validity to our machine learning models, while also revealing novel patterns of calf and male vulnerability in this threatened population. More broadly, our approach demonstrates the potential of automated image analysis to enable timely, non-invasive health assessments across cetacean populations in an era of rapid global change.