Identification of the influential factor for the bloom of Alexandrium catenella: Machine learning and numerical simulation approach

Abstract

Harmful algal blooms (HABs) occur globally in freshwater, brackish, and marine environments. Numerou s organizations have attempted to monitor HABs in order to sec ure the safety of marine environments using operative monitoring programs. To analyze the monitoring data, numerical simulation and machine learning approaches were proven to be useful for environmental management. In this study, we applied numerical simulation and machine learning to identify the influential factor for Alexandrium catenella bloom. Our study site was located at the eastern coast of Geoje Island in Republic of Kor ea. This area was a hot spot for HABs that encountere d a high frequency of blooms. The monitoring data used as input forthe models were collected between February 2017 to June 2018. Random forest (RF) was applied for determining the influential factor of the bloom while the environmental fluid dynamic code (EFDC) analyzed the mechanism of oceanic physics. The trained RF model predicted the dynamics of Alexandrium catenella, and the sens itivity analysis identified the uncertainty of the influential factor. Meanwhile, the EFDC model provided the data for the ocean current and retention time. The trained model showed a good agreement with the observed Alexandrium catenella cells, yielding a coefficient of determination of 0.55 between simulated and observed values . Sens itivity analysis revealed that the influential factors for A