Genetic programming for analysis and real-time prediction of coastal algal blooms

Abstract

Harmful algal blooms (HAB) have been widely reported and have become a serious environmental problem world wide due to its negative impacts to aquatic ecosystems, fisheries, and human health. A capability to predict the occurrence of algal blooms with an acceptable accuracy and lead-time would clearly be very beneficial to fisheries and environmental management. In this study, we present the first real-time modelling and prediction of algal blooms using a data driven evolutionary algorithm, Genetic Programming (GP). The daily prediction of the algal blooms is carried out at Kat O station in Hong Kong using 3 years of high frequency (two-hourly) chlorophyll fluorescence and related hydro-meteorological and water quality data. The results for the prediction of chlorophyll fluorescence, a measure of algal biomass, are within reasonable accuracy for a lead-time of up to 1 day. The results generally concur with those obtained with artificial neural network. As compared to traditional data-driven models, GP has the advantage of evolving an equation relating input and output variables. A detailed analysis of the results of the GP models shows that GP not only correctly identifies the key input variables in accordance with ecological reasoning, but also demonstrates the relationship between the auto-regressive nature of bloom dynamics and flushing time. This study shows GP to be a viable alternative for algal bloom modelling and prediction; the interpretation of the results is greatly facilitated by the analytical form of the evolved equations.