Machine learning-based prediction of jellyfish blooms and their influence on coastal fisheries

Abstract

Jellyfish blooms predictably exacerbate the economic and ecological challenges coastal fisheries face globally. Effective fishery management relies heavily on predicting growth patterns alongside mitigating possible risks. This investigation initiates a framework utilizing machine learning to forecast the growth of jellyfish populations and their corresponding impact on coastal fisheries. The described system, JellyNet, is a convolutional neural network (CNN) that utilizes high-resolution remote-sensing satellite imagery captured by drones (UAVs). Jelly Net allows fisheries to act based on predictions, providing 6 to 8 hours of early detection and bloom event forecasting. A dataset derived from Croabh Haven, UK, and Pruth Bay, Canada, with 1,539 images, was annotated into two categories: 'Bloom present' and 'No bloom present,' which is essential for precise feature identification during bloom detection. Employing transfer learning featuring the VGG-16 architecture, JellyNet surpassed baseline models, achieving a pinnacle accuracy of 97.5%. Furthermore, the study analyzes the relationship between predicted bloom occurrences and subsequent changes in fish catch data, illustrating jellyfish blooms’ dominantly negative influence on productivity. This study reveals the mastery machine learning holds in predictive analysis and sustainable coastal fishery operations.