S.C.M. Oerlemans*, W. Nijland, T.D. Price

Utrecht University

*corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.


Subtidal sandbars are ubiquitous features in the nearshore zone of many sandy coasts, and unravelling their dynamics is crucial to the understanding of nearshore sediment pathways. Wave breaking and wave-driven currents constantly rearrange nearshore sediment into complex patterns leading to the development of sandbar morphology, ranging from shore-parallel ridges of sand to an alongshore alternation of shore-attached bars and rip-channels. Wright and Short (1984) created a widely used beach state classification scheme, in which they distinguish a total of six beach states with distinct sandbar configurations. Recognition and classification of these beach states is not trivial and hitherto involved manual classification or pre-defined image features. The tremendous progress in data-driven learning in image recognition over the past years has led to a first automated classification of single-barred beach states from video (Argus) imagery (Ellenson et al., 2020), using a convolutional neural network (CNN). We build upon this work to extend the classification of single-barred beach states to double-barred beaches. The objective of this study is to perform a multi-class classification of beach states for the inner and outer bar separately.


To make our CNN model we used the pretrained network ResNet50 with transfer learning from a natural image dataset, ImageNet. Our data consisted of labelled images from the single-barred beaches Narrabeen (Australia) and Duck (US), as used by Ellenson et al. (2020), complemented with over 9 years of daily images of the double-barred beach of the Gold Coast (Australia). We implemented various combinations of the data to train, test and validate the performance of each model for the detection of Wright and Short's (1984) beach states.


Adding the inner and outer bar data separately to the single-bar trained CNN increased model performance (up to an F1-score of 0.88). For the double-barred beach it mattered which of the two bars was used for training the model; training with outer (inner) bar data led to higher performance for the outer (inner) bar detection. During the NCK days, we will present our CNN model evaluation and analysis, as well as valuable insights that can be extended to the design of high performance detection and classification systems for other imaging tasks in the coastal domain.


Figure 1: CNN in action


Ellenson, A.N., Simmons, J.A., Wilson, G.W., Hesser, T.J., Splinter, K.D. (2020). Beach State Recognition Using Argus Imagery and Convolutional Neural Networks. Remote Sensing, 12, 3953. https://doi.org/10.3390/rs12233953
Wright, L.D., Short, A.D. (1984). Morphodynamic variability of surf zones and beaches: A synthesis. Marine Geology, 56, 93-118, https://doi.org/10.1016/0025-3227(84)90008-2

  TU Delft logo transp

nioz logo transp

UT Logo 2400 Sta Black EN

Deltares logo D blauw RGB footer

RWS EN transparant

uu logoengels rgb

ihe delft logo new transparant

TNO text transparant 

WUR RGB standard