Functional data-based estimation of coastal erosion and accretion at Playa Salguero using J-Net Dynamic

Coastal erosion is a critical environmental and management issue along the Colombian Caribbean, where nearly half of the shoreline presents erosive trends. This study assesses shoreline change at Playa Salguero, Santa Marta, from 2016 to 2025 using 58 PlanetScope NICFI images, bilateral filtering, shoreline extraction with J-Net Dynamic, and DSAS metrics (NSM, SCE, EPR, and LRR). Two observational periods were considered around groin construction in 2023. During 2016–2023, 88.52% of the transects showed erosion, with a mean NSM of -14.35 m and a maximum retreat of -24.02 m. During 2023–2025, erosion remained dominant but accretion increased to 37.30% of the transects and mean NSM became +2.01 m, indicating a temporal shift toward more accretional conditions during the post-intervention period. Shorelines were subsequently represented as functional data and projected one year ahead using universal functional kriging with trend. Leave-One-Date-Out cross-validation yielded average RMSE values of 6.33 m in X and 3.70 m in Y. This procedure evaluates internal temporal model performance conditional on the satellite-derived shorelines and does not constitute independent validation of shoreline-extraction accuracy. The mean model-based predictive standard deviation ranged from 15.2 to 19.6 coordinate units across the projected dates. These values represent kriging-based predictive uncertainty conditional on the satellite-derived shorelines and should not be interpreted as a complete shoreline-position uncertainty budget because shoreline extraction, georeferencing, spatial-resolution, water-level, and other observational uncertainties were not propagated through the model. Accordingly, the projected shoreline positions are interpreted as short-term model-based probabilistic scenarios rather than independently validated future shoreline positions. The post-2023 observations showed an increase in localized accretion that temporally coincided with the intervention period, while erosive behavior persisted in other sectors. The available observational design does not establish a causal effect of the groins. The methodological contribution of this study is the integration of ACM-based J-Net Dynamic shoreline extraction, conventional DSAS shoreline change metrics, functional representation of the complete shoreline geometry, and universal functional kriging within a unified image-based framework. The functional component complements conventional transect-based analysis by representing the shoreline as a continuous spatial object and by providing model-based predictive uncertainty.