Data-driven components in a model of inner-shelf sorted bedforms: a new hybrid model

Abstract

Abstract. Numerical models rely on the parameterization of processes that often lack a deterministic description. In this contribution we demonstrate the applicability of using machine learning, a class of optimization tools from the discipline of computer science, to develop parameterizations when extensive data sets exist. We develop a new predictor for near-bed suspended sediment reference concentration under unbroken waves using genetic programming, a machine learning technique. We demonstrate that this newly developed parameterization performs as well or better than existing empirical predictors, depending on the chosen error metric. We add this new predictor into an established model for inner-shelf sorted bedforms. Additionally we incorporate a previously reported machine-learning-derived predictor for oscillatory flow ripples into the sorted bedform model. This new "hybrid" sorted bedform model, whereby machine learning components are integrated into a numerical model, demonstrates a method of incorporating observational data (filtered through a machine learning algorithm) directly into a numerical model. Results suggest that the new hybrid model is able to capture dynamics previously absent from the model -specifically, two observed pattern modes of sorted bedforms. Lastly we discuss the challenge of integrating data-driven components into morphodynamic models and the future of hybrid modeling.

Stability and bistability in a one‐dimensional model of coastal foredune height

“…Coupling theoretical components (3a) with machine learning parameterizations (3b) in this way is an example of a "hybrid" model [e.g., Krasnopolsky and Fox-Rabinovitz, 2006;Goldstein and Coco, 2015]. This contribution serves as a further example of the utility in using machine learning processes for constructing models of coastal environments [e.g., Goldstein et al, 2014;Limber et al, 2014].…”

“…The arrangement and value of constants, variables and mathematical operators are modified at each time step to develop candidate solutions (equations) that can reproduce the relationships present in the data. This technique has been used successfully in other coastal studies, [ Goldstein et al ., , ; Limber et al ., ; Tinoco et al ., ] and we refer the reader to these works and the initial work by Schmidt and Lipson [] for additional details of the technique.…”

Modeling sorting dynamics of cohesive and non-cohesive sediments on intertidal flats under the effect of tides and wind waves

“…However, most of the studies primarily focused on the sorting dynamics of sandy beaches, rivers, deltas, tidal basins, and inner continental shelves (e.g., Coco et al, 2007a,b;Frings, 2008;van der Wegen et al, 2011;Goldstein et al, 2011Goldstein et al, , 2014Viparelli et al, 2014), while intertidal flats received much less attention. To the authors' knowledge, there is currently no numerical study specifically addressing intertidal sediment sorting dynamics.…”

A machine learning approach for the prediction of settling velocity

“…We select training data using a maximum dissimilarity algorithm (MDA) [ Camus et al ., ] and use this subsample in a genetic programming (GP) routine. GP [ Koza , ] is a population‐based optimization routine that has been previously used to examine other earth surface systems [ Kitsikoudis et al ., ; Baptist et al ., ; Goldstein et al ., ]. In contrast to other ML approaches, GP does not require a set basis function and can use a wide range of functional relationships as building blocks for the final expression.…”