Marine Pollution Bulletin 161 (2020) 111731 Available online 30 October 2020 0025-326X/© 2020 Elsevier Ltd. All rights reserved. A real time data driven algal bloom risk forecast system for mariculture management Jiuhao Guo a , Yahong Dong b , Joseph H.W. Lee c, * a Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology, Hong Kong, China b School of Electromechanical Engineering, Qingdao University of Science and Technology, Qingdao, China c Department of Civil and Environmental Engineering and Institute for Advanced Study, The Hong Kong University of Science and Technology, Hong Kong, China A R T I C L E INFO Keywords: Eutrophication Harmful algal blooms Fisheries management Red tide Stratifcation Real-time forecast Water quality prediction Dissolved oxygen Chlorophyll Artifcial neural network Data assimilation Risk management ABSTRACT In eutrophic coastal waters, harmful algal blooms (HAB) often occur and present challenges to environmental and fsheries management. Despite decades of research on HAB early warning systems, the feld validation of algal bloom forecast models have received scant attention. We propose a daily algal bloom risk forecast system based on: (i) a vertical stability theory verifed against 191 past algal bloom events; and (ii) a data-driven artifcial neural network (ANN) model that assimilates high frequency data to predict sea surface temperature (SST), vertical temperature and salinity differential with an accuracy of 0.35 o C, 0.51 o C, and 0.58 psu respec- tively. The model does not rely on past chlorophyll measurements and has been validated against extensive feld data. Operational forecasts are illustrated for representative algal bloom events at a marine fsh farm in Tolo Harbour, Hong Kong. The robust model can assist with traditional onsite monitoring as well as artifcial- intelligence (AI) based methods. 1. Introduction In sub-tropical coastal waters around Hong Kong, algal blooms (the rapid growth of microscopic phytoplankton) are often observed. Harmful effects caused by algal blooms – including dissolved oxygen depletion, fsh kills, shellfsh poisoning, and beach closures – have been widely reported (AFCD, 2016; Wong, 2003). In April 1998, a devastating red tide initiated in Mirs Bay resulted in the worst fsh kill in Hong Kong’s history - over 80% (3400 t) of fsh stocks in HKSAR were wiped out, with an estimated loss of over HK$312 million. More recently, a severe algal bloom event started in Tolo Harbour in December 2015 and lasted for two months, resulting in fsh kills and threats of spreading to other Hong Kong waters. Despite signifcant upgrades of the water pollution control infrastructure over the past two decades, around 10 to 20 algal blooms still occur every year (Fig. 1) and present challenges to fsheries management. The dynamics of harmful algal blooms (HAB) is complex and poorly understood. The use of cell counts and Chlorophyll-a concentration as indicators of algal biomass is widely accepted. The relation of algal and dissolved oxygen (DO) dynamics has also been elucidated through water quality modeling and feld observation (Lee et al., 1991; Lee and Lee, 1995). Traditionally, fsheries and environmental management are based on monitoring of the above key indicators at typically sparse (monthly) intervals. In Hong Kong, the Agriculture, Fisheries and Con- servation Department (AFCD) has been carrying out weekly phyto- plankton sampling and biweekly water quality (nutrient) monitoring at a few key fsh culture zones (FCZ) since 1998. However, on-site feld surveys and manual enumeration of cell counts and species identifca- tion are laborious, and the sampling frequency is insuffcient to capture the highly dynamic variation of algal blooms. Since around 2015, real- time water quality sensors have been increasingly installed in key FCZs to continuously monitor key parameters (salinity, water temper- ature, DO, and chlorophyll fuorescence) at 2 depths and 10-minute sampling frequency. Nevertheless a systematic analysis of this set of high frequency data has thus far not been carried out. Over the past three decades, process-based mathematical models of eutrophication which describe the general ecological response of phytoplankton to environmental conditions have been extensively re- ported (Thomann and Mueller, 1987). However, the use of two- or three- dimensional deterministic models has been limited mainly to prediction * Corresponding author. E-mail address: jhwlee@ust.hk (J.H.W. Lee). Contents lists available at ScienceDirect Marine Pollution Bulletin journal homepage: www.elsevier.com/locate/marpolbul https://doi.org/10.1016/j.marpolbul.2020.111731 Received 4 June 2020; Received in revised form 27 September 2020; Accepted 30 September 2020