Validating a Chaos Based Anoxic Event Prediction Method for Estuaries

This paper presents and experiments with a new validation dataset using a method that models anoxia in estuaries. The proposed method incorporates nonlinear time series characterization, time delay methods, unsupervised techniques for intrinsic dimensionality estimation, and manifold learning to predict anoxic events. Characterizing dissolved oxygen time series with features derived from nonlinear time series analysis effectively identified families of series exhibiting differences in their dynamics. The built models showed high correlations and relatively low errors, thus validating the approach on these additional data. However, further work is needed to develop more advanced models capable of operating with time-ordered events.