Machine Learning Approach to Predicting Foredune Morphometrics

The resiliency of a barrier island, its ability to return to form and ecological function after storms, is important for our understanding of sea level rise and changes in storm activity. Foredune elevation, width, and extent are important factors that determine the response of a barrier island, though the underlying controls of these alongshore variations are not fully understood. Topographic complexity provides insight to the complex nature of processes and sediment supply and is critical for assessing how barrier islands recover and evolve through time. The purpose of this research is to generate a model to identify the position of foredune morphometrics, including the dune crest, dune toe, and dune heel, using a deep neural network. Sample data from Padre National Seashore is presented. Variables derived from a digital elevation model and vegetation are used as inputs to train the model using a portion of the study site, while the remainder of the site is used to test the model. Model results are compared to beach-dune morphometrics extracted using an automated relative relief approach and demonstrate it is possible to extract these metrics using a non-subjective approach. The machine learning methodology outlined here is feasible and useful for identifying dune crest, dune toe, and dune heel morphometries from different coastal variables, and nonlinearities between these variables can be used to explain alongshore variation in these metrics. This machine learning paradigm has the potential to advance our understanding of barrier island geomorphology and resiliency.