부산대학교 도서관

Harmful algal blooms represent significant ecolog-ical disturbances, characterized by the rapid accumulation of detrimental algae in water bodies. Such events, particularly red tides, have increasingly impacted marine ecosystems in areas such as the Ariake Sea and Yatsushiro Sea in Kumamoto, Southern Japan. This study investigates the temporal dynamics of red tide events in these regions and proposes an innovative early detection methodology that leverages satellite imagery through deep learning techniques. The primary aim is to accurately forecast the occurrence of this event within a predictive window of a few days. This paper details the development of the early detection framework, discusses the challenges faced during its implementation, and explores how machine learning can be further refined using time-series satellite data for environmental monitoring. The models, tested across two distinct Japanese regions, achieved a prediction accuracy of up to 89 % for impending red tide events, demonstrating the potential of this approach for timely and effective coastal management.