Subaerial beach profiles classification: An unsupervised deep learning approach

Abstract

Beach profiles vary significantly based on sediment supply and hydrodynamic, aeolian, and anthropogenic influences. Successful management of beach and dune resources is heavily based on understanding and classifying beach profile shapes based on their dominant features. As such, the highly dynamic shape of beach profiles has motivated researchers to classify them based on different criteria where the number of groups depend on the empirical basis and actual coastal environment. In the interest of classification, beside the theoretical and analytical approaches, nowadays, the application of artificial intelligence, especially, deep learning is well recognized. To this end, in this study, subaerial beach profiles are classified through unsupervised learning and cluster analysis. Accordingly, a convolutional neural network is designed that can search the entire dataset and categorizes beach profiles without any prior definitions. With the proposed method, 916 subaerial beach profiles from 2005 to 2018, representing six beaches from three different towns in Maine, USA were categorized into 2, 3, and 5 categories. The results showed that depending on the number of required categories, the proposed model was able to spot the most common features among all profiles in a category.