A multiple process univariate model for the prediction of chlorophyll-a concentration in river systems

Abstract

The concentration of chlorophyll-a (Chl-a) in river systems is dependent on various hydrometric and biochemical factors, including an intricate array of corresponding growth and extinction mechanisms. This complex and interactive assortment of factors makes prediction of algal blooms difficult. This paper introduces an innovative time-series model structure that predicts Chl-a concentration in inland waters. To improve the prediction accuracy of existing models, we assume that the predicting variable is determined by multiple and independent drivers. An enhanced stochastic model, namely a multiple process univariate model (MPUM), is developed to address the impacts of the distinct mechanisms associated with each regulator (e.g., hydrometeorological factors and anthropogenic activities). Observations of the algae concentration at 16 weirs along four major river systems in South Korea are used to model Chl-a concentration. Comparisons between traditional models and the proposed method demonstrate the strengths of the MPUM, both in making predictions and in the parsimony of the model structure. The robustness of the developed model was further validated by modeling algae concentration before and after river-flow regulation procedures.

Aquatic ecosystem assessment and management

“…Their results showed that rainfall, river flow, water temperature and nutrient concentration were the primary factors affecting cyanobacterial blooms in the study river system, suggesting that the successful control of cyanobacterial blooms requires integrated watershed management, accommodating the control and management of meteo-hydrological-physicochemical factors. Kim (2016) introduced an innovative time-series model to predict the chlorophyll-a concentration in Korean rivers. A multiple process univariate model, which is an enhanced stochastic model, was developed to address the effects of distinct mechanisms associated with hydrometeorological factors and anthropogenic activities.…”

Predicting Daily River Chlorophyll Concentrations at a Continental Scale

“…Modeling chlorophyll dynamics in rivers can be challenging due to non‐linear relationships between chlorophyll and its drivers, as well as covariation or interaction among drivers. Multiple approaches have been used to model chlorophyll in rivers including process‐based (Billen et al., 1994; Everbecq, 2001; Pathak et al., 2021), traditional statistical (S. Kim, 2016; K. B. Kim et al., 2020), and machine learning (Cho et al., 2018; Jeong et al., 2001; Park et al., 2022). Although model objectives of predictive accuracy and process understanding are not mutually exclusive (Duarte et al., 2003), each modeling approach varies in its emphasis on these two components and has strengths and weaknesses for broad‐scale use.…”

“…Modeling chlorophyll dynamics in rivers can be challenging due to non-linear relationships between chlorophyll and its drivers, as well as covariation or interaction among drivers. Multiple approaches have been used to model chlorophyll in rivers including process-based (Billen et al, 1994;Everbecq, 2001;Pathak et al, 2021), traditional statistical (S. Kim, 2016;K. B. Kim et al, 2020), and machine learning (Cho et al, 2018;Jeong et al, 2001;Park et al, 2022).…”

A Remote Sensing and Machine Learning-Based Approach to Forecast the Onset of Harmful Algal Bloom

“…The growth of HABs was successfully predicted using historical data and ecological informatics and applying DL methods [62]. The DL methods were also used to predict algal growth in rivers [63][64][65][66], lakes [67,68], and coastal areas [69,70]. However, these methods function as black boxes, and the neuron connections, their weights, and different layers cannot be associated much with the concept of the physical problem at hand [71].…”