Data-based methods and algorithms for the analysis of sandbar behavior with exogenous variables

• Data-driven techniques were used to analyze sandbar data.
• Neural networks helped to detect nonlinear relations between sandbar and waves.
• It was found that sandbar behavior has memory.
• Empirical mode decomposition helps to visualize complex sandbar behavior.
• Memory and external waves produce an elastic effect in sandbar movement.

Sandbars are natural features generated in the nearshore zones by the interaction between the sea and the coast. The short-term processes that drive sandbar behavior are waves and sediment transport. The interaction between waves and the coast is highly nonlinear and, traditionally, process-based models (e.g., evolution models) have been used for modelling and analyzing sandbar behavior in the short term. However, medium- to long-term predictions are not always possible with these models due to their exponential error accumulation. Data-driven models emerge as an alternative to process-based models as they do not need insight on the physical knowledge of the model, but they extract knowledge from patterns found in the data. In this paper, we apply the data-driven techniques: EMD (empirical mode decomposition) method and ARNN (autoregressive neural networks) on sandbar and wave time series from the coast of Cartagena de Indias, Colombia. The former is used for analyzing the relationship between sandbar and wave conditions in a graphical way; and the latter is used for deriving nonlinear simple/partial cross/auto-correlation coefficients. Evidence of nonlinear dependencies is detected between the present state of sandbar location and the past states of wave conditions.