Predicting shoreline changes along the California coast using deep learning techniques applied to remote sensing data

Abstract

The ability to predict changes in shorelines is critical for coastal planners, and requires large-scale monitoring programs that are currently not available over most regions. Satellite observations of shoreline changes promise global coverage, but the role of these data for predictions has not yet been determined. The abundance of field observations in California provides a unique opportunity to test the utility of satellite-derived estimates for predicting changes in shorelines. In this study, we use 20 years of shoreline change estimates from satellite imagery for California’s beaches and combine them with wave model outputs in a deep neural network (DNN) framework to estimate beach characteristics and predict future changes. We find good agreement between DNN estimates and field observations of beach slope and width. DNN predictions of 2050 shorelines suggest a mean retreat of 3 m in an intermediate emissions scenario, which will result in substantial losses of California’s beach areas.