A Coastal Case Study on Application of Convolutional Neural Networks for Red Tide Algal Bloom Image Classification

Abstract

A notable type of harmful algal bloom (HAB) is red tide phenomena, which creates the problem in the coastal areas repeatedly, with ecological and socio-economic consequences. Even though recent developments in satellite remote sensing have enhanced the detection of the blooms, it is still necessary to have a systematic synthesis of the reported incidences to gain insight into the wider spatial and temporal patterns. This paper offers a case-study-based observation of the red tide happening in the coastal area in the form of secondary data gathered following the recent satellite images and the existing published literature. Instead of formulating a predictive model, the research chooses a descriptive analytic framework, according to which documented events of the bloom are formed in a systematic temporal and spatial classification. The reported cases of red tide were detected in various sources and aggregated seasonally and spatially, and then the frequency-based aggregation was done to come up with relative occurrence measures. The findings show high seasonality with 45% of all events reported during summer, 25% during spring, 20% during autumn, and 10% during winter. Spatial analysis shows that there is a high concentration in coastal areas, with 55 % of the events reported to be in the nearshore areas and 30% and 15 % in the inner and outer shelves, respectively. Observation platforms have been evaluated to reveal that the majority of the reports on red tide are made by satellite-based data (40% of recorded events are Sentinel-2, 30% VIIRS, 20% harmonized Landsat Sentinel products, and 10% in situ). This study also reviews recent CNN-based methods to show how satellite images are used for red tide detection and to clearly explain how the proposed method is different from existing approaches.