Computational Intelligence Approach for the Analysis of Estuary Hypoxia Phenomena

This paper presents a data driven study of dissolved oxygen times series collected in Atlantic Canada. The main motivation of presented work was to evaluate if machine learning techniques could help to understand and anticipate hypoxic episodes (low level of dissolved oxygen) in nutrient-impacted estuaries, a phenomenon that is exacerbated by increasing temperature expected to arise due to changes in climate. A major constraint of the analysis was limiting ourselves to a single variable, the dissolved oxygen time series. Data exploration using unsupervised machine learning methods revealed the existence of low dimensionality spaces containing most of the similarity relations between the dissolved oxygen series. In them, it is possible to distinguish three reasonably well defined classes, with bridge objects in between. Additional findings using supervised machine learning methods show that recurring neural networks, and in particular LSTM, may be capable of exploiting time series seasonality in predicting short horizon levels.