Deep Learning To Model The Complexity Of Algal Bloom

Abstract

Literature of studying algal growth has started to take advantages of data mining and machine learning methods, such as classification, clustering, regression, correlation analysis and principal component analysis. However, the performance of such methods might heavily rely on the data collectable for the studies sites. Moreover, some factors directly relate to algal growth, including hydrodynamics, weather and ecology, are notoriously difficult to model and predict. In this paper we present a study to model algal bloom using deep learning methods. It is assumed that algal bloom is the consequence of all factors that are more or less associated with the growth of algal. This offers a new way of thinking that even unknown factors or those factors far too complicated to model can still be inexplicitly represented by the deep learning models. We evaluate this new approach through our studies of algal bloom in the JinJi Lake, Suzhou, China. The experimental results are compared with the popular machine learning methods used in literature. It has been found that the deep learning method can achieve a better accuracy in comparison with other well applied machine learning methods.