Red tide caused severe impacts on marine fisheries, ecology, economy and human life safety. The formation mechanism of the red tide is rather complicated; thus, red tide prediction and forecasting have long been a research hotspot around the globe. This study collected ocean monitoring data before and after the occurrence of red tides in Xiamen sea area from 2009 to 2017. The Pearson correlation coefficient method was used to obtain the associated factors of red tide occurrence, including water temperature, saturated dissolved oxygen, dissolved oxygen, chlorophyll-aand potential of hydrogen. Then, we built a short-time red tide prediction model based on the combination of multiple feature factors. chlorophyll-a, dissolved oxygen, saturated dissolved oxygen, potential of hydrogen, water temperature, salinity, turbidity, wind speed, wind direction and Air pressure were used as the input variables, building a short-time prediction model based on the combination of multiple feature factors to forecast red tide in the next 6 h by using the monitoring data. The accuracy of different forecast models with different feature combinations was compared. Results show that the distinguishing factors which have the most significant influence on red tide prediction in Xiamen are chlorophyll-a, dissolved oxygen, saturated dissolved oxygen, potential of hydrogen, and water temperature. the convergence speed of the Gated Recurrence Unit (GRU) prediction model based on the main feature factor proposed in this paper was faster and obtained the expected result, and the accuracy rates of the buoys are above 92%. The research shows the feasibility to use GRU network model to predict the occurrence of red tide with multi-feature factors as input parameters. the paper provides an effective method for the red tide early warning in Xiamen sea area.
Keywords: Correlation; GRU Network; Neural network; Prediction model; Red tide.
Copyright © 2022. Published by Elsevier Ltd.