Shoreline displacement assessment on the Pacific Coast of Colombia using numerical simulations, remote sensing and machine learning in a data-limited environment

Abstract

Understanding Shoreline Displacement (SLD) in data- limited coastal regions, is critical for risk-informed management. We evaluate an integrated workflow that couples wave-hydrodynamic modeling, tide-corrected CoastSat satellite shorelines, and machine learning (ML) to characterize SLD drivers in Tumaco Bay on Colombia's southern pacific coast. Using data from 2017 to 2018, various structured meshes and advection schemes are tested. A 100 m mesh with a cyclic advection scheme best reproduced mesotidal dynamics when forced with TPXO8 tidal model for boundary conditions and wind data provided by a meteorological station. Model performance evaluated against Waverys reanalysis and available observations showed good skill, supporting application where in situ data are sparse. Multidecadal (1993–2024) CoastSat shoreline positions for El Morro and El Bajito beaches indicate an overall accretional tendency but found erosion hotspots affecting densely settled and touristic sectors. Regional wave conditions feature significant wave height (H_s) of 0.3–1.1 m, peak wave period (T_p) of 4–18 s, and predominantly westward wave approach, Random Forest (RF) results identify sea level (SL) and mean wave propagation direction (Θm) as the leading contributors to observed SLD variability, whereas H_s and T_p are secondary under the area's moderate wave energy. The study demonstrates the need to monitor localized erosion despite net accretion and demonstrate that combining physics-based modeling, open satellite archives, and data-driven methods can yield policy-relevant coastal insights in data-scarce tropical estuarine environments.