Coastline and shoreline change assessment in sandy coasts based on machine learning models and high-resolution satellite images

Abstract

Changes to the coastline or shoreline arise from the water's dynamic interaction with the land surface, which is triggered by ocean currents, waves, and winds. Various methods have been proposed to identify and monitor coastlines and shorelines, but their outcomes are uncertain. This study proposes indicators for identifying coastlines and shorelines in the fields and on the remote sensing data. Different pixel- and object-based machine learning (ML) models were built to automatically interpret coastlines and shorelines from high-resolution remote sensing images and monitor coastal erosion in Vietnam. Two pixel-based models using Random Forest and SVM structures and eight object-based models using U-Net, and U-Net3+ structures were trained. All models were trained using the high-resolution images gathered using Google Earth Pro as input data. The U-Net achieves the most remarkable performance of 98% with a loss function of 0.16 when utilizing an input-image size of 512×512.

Object-based models have shown higher performance in analyzing coastlines and shorelines with linear and continuous structures than pixel-based models. Additionally, the coastline is appropriate to evaluate coastal erosion induced by the effect of sea-level rise during storms. At the same time, the shoreline is suited to observe seasonal tidal fluctuations or the instantaneous movements of current waves. Under the pressure of tourist development, the coasts in Danang and Quang Nam provinces have been eroded in the last 10 years. River and ocean currents also cause erosion in the southern Cua Dai estuary. In the future, the trained U-Net model can be used to monitor the changes in coastlines and shorelines worldwide.