Towards a global assessment of sandy shorelines: Systematic extraction and validation of optical satellite-derived coastal indicators at various sites

Abstract

Satellite-based remote sensing offers unprecedented opportunities for monitoring coastal dynamics at large spatiotemporal scales through proxies such as shoreline positions. Recent advances in computational tools and the accessibility of publicly available satellite imagery have made it feasible to derive these metrics efficiently using user-defined parameters (e.g., band combinations, segmentation algorithms, and preprocessing steps). This abundance of data and tools, coupled with increasing computational power, suggests that global-scale monitoring of shoreline change is within reach. However, efforts to develop continental-to-global shoreline change datasets remain limited, partly due to challenges associated with varying environmental conditions and the lack of standardized algorithms. Yet, quantitative relationships between errors in the satellite-derived observations and environmental conditions remain unexplored. Also, the large body of available methods for satellite-derived waterline (SDW) extraction has merely been inter-compared. In this study, we built on previous works to develop a customizable open-access SDW extraction Python toolkit. We assess the influence of spectral band combination and segmentation method on the accuracy of waterline and shoreline positions derived from Sentinel-2 and Landsat optical imagery, comparing 20 SDW extraction methods in total. By evaluating the performances of these extraction methods across eight sandy coast sites using in-situ beach elevation profiles and FES2022 tide model outputs, we highlight the systematic gap between SDW performances obtained at microtidal and meso- to macrotidal environments. We found that SDW/SDS accuracy at microtidal sites is slightly influenced by the choice of band combination, while the accuracy at meso- to macrotidal sites is rather impacted by the choice of thresholding method. We also validated related satellite-derived metrics, such as long-term trends, interannual variability, and seasonal cycles of shoreline change, as well as beach slopes, and found that there are generally well captured by the top-performing SDW methods, with errors in long-term trend, interannual variability and seasonal cycle amplitude estimations around 0.8 m/yr, 2 m, and 2.5 m, respectively. Finally, we address the issue related with low signal-to-noise ratio in SDS time series, emphasizing the necessity to quantify and mitigate noise by aggregating data over time, and propose empirical laws quantifying errors in shoreline position time series as a function of the inverse beach slope and the tidal excursion, meaning tidal range normalized by the beach slope. These results provide a critical framework for first order estimation of the accuracy of satellite-derived shoreline positions derived at sites lacking validation data, and bring insightful materials to discuss methodological challenges related to large-scale, automated shoreline monitoring, highlighting strengths and limitations of Sentinel-2- and Landsat-derived observations, beyond the sole coastal monitoring field.