A Bayesian structural model for predicting algal blooms

• DOI: http://doi.org/10.1002/for.2583
• Author: Jiayin Wang,Tao Liu,Xinyu Sun
• Published date: 01 December 2019
• Date: 01 December 2019

RESEARCH ARTICLE

A Bayesian structural model for predicting algal blooms

Xinyu Sun

1

| Tao Liu

1

| Jiayin Wang

2

1

School of Management, Key Lab of the

Ministry of Education for Process Control

and Efficiency Engineering, Xi'an Jiaotong

University, Xi'an, Shannxi, China

2

School of the Electronic and Information

Engineering, Xi'an Jiaotong University,

Xi'an, Shannxi, China

Correspondence

Xinyu Sun, School of Management, Key

Lab of the Ministry of Education for

Process Control and Efficiency

Engineering, Xi'an Jiaotong University,

Xi'an, Shannxi, China.

Email: xinyu.sun@xjtu.edu.cn

Funding information

Fundamental Research Funds for the Cen-

tral Universities of China, Grant/Award

Number: CXTD2017003; MOE Project of

Humanities and Social Sciences of China,

Grant/Award Number: 19YJE630002;

National Science Foundation of China,

Grant/Award Number: 31701150; Natural

Science Foundation of Shaanxi Province,

China, Grant/Award Number:

2017JM7009; Soft Science Research Pro-

gram of Shannxi Province, China, Grant/

Award Number: 2018KRZ005

Abstract

A Bayesian structural model with two components is proposed to forecast the

occurrence of algal blooms, multivariate mean‐reverting diffusion process

(MMRD), and a binary probit model with latent Markov regime‐switching pro-

cess (BPMRS). The model has three features: (a) forecast of the occurrence

probability of algal bloom is directly based on oceanographic parameters, not

the forecasting of special indicators in traditional approaches, such as phyto-

plankton or chlorophyll‐a; (b) augmentation of daily oceanographic parameters

from the data collected every 2 weeks is based on MMRD. The proposed

method solves the problem of unavailability of daily oceanographic parameters

in practice; (c) BPMRS captures the unobservable factors which affect algal

bloom occurrence and therefore improve forecast accuracy. We use panel data

collected in Tolo Harbour, Hong Kong, to validate the model. The model dem-

onstrates good forecasting for out‐of‐sample rolling forecasts, especially for

algal bloom appearing for a longer period, which severely damages fisheries

and the marine environment.

KEYWORDS

algal bloom, Bayesianestimation, binary probit,forecasting, latent Markov regime‐switching process,

mean‐reverting process

1|INTRODUCTION

A rapid increase in algae population in a water system, a

phenomenon called algal bloom or red tide, can cause a

large‐scale marine mortality event. These harmful algal

blooms (HAB) have direct negative impacts on human

well‐being, mainly through their impact on fisheries,

tourism, and recreation, as well as on human health

through exposure to biotoxins through inhalation, direct

contact, or through ingestion through the food chain

(Willis, Papathanasopoulou, Russel, & Artioli, 2018). To

control the damage of HAB to human beings and indus-

tries, forewarning and predicting the occurrence of

large‐scale HAB, which relies on a viable and efficient

indexing system, is critical. In this paper, we build a

Bayesian structural model for the prediction of the

occurrence of algal blooms in a water system and validate

the model using data collected from Tolo Harbour, Hong

Kong. The model demonstrates good out‐of‐sample pre-

diction performance.

One approach to predicting the occurrence of algal

blooms is mathematical modeling based on numerical

simulations. For example, Aleynik, Dale, Porter, and

Davidson (2016) provide a new high‐resolution hydrody-

namic coastal modeling system based on the finite‐

volume coastal ocean model and the weather research

forecast model in areas of complex coastline and topogra-

phy. Moreover, several others have examined the

response of phytoplankton to the flow field of Langmuir

cells (Watanabe & Harashima, 1986) or to two‐

dimensional, cross‐frontal circulation (Franks & Ander-

son, 1992). Yanagi et al. (1995) used the Euler–Lagrange

Received: 13 November 2018 Accepted: 12 February 2019

DOI: 10.1002/for.2583

Journal of Forecasting. 2019;38:788–802.© 2019 John Wiley & Sons, Ltd. wileyonlinelibrary.com/journal/for

788