Marine terraces record past sea levels and serve as strain markers to quantify vertical deformation from tectonic and climatic processes. Accurate mapping of these ephemeral features is essential but often limited by data quality and operator subjectivity. TerraceM-3 reduces both non-systematic and systematic mapping errors by integrating machine learning that replicates expert interpretation within standardized workflows. A new TerraceM-ICESat module enables global, vegetation-free, high-resolution mapping using ICESat-2 altimetry, including shallow offshore bathymetry. Tested in Peru and Algeria, TerraceM-3 reveals detailed coastal deformation patterns and advances research in tectonic geomorphology and coastal hazard assessment. Financial support: This study was supported by TANTA “Earthquakes and coastal deformation in subduction zones at continental scale” grant P2022-13-001 funded by the Carl-Zeiss-Stiftung; the Millennium Nucleus CYCLO “The Seismic Cycle Along Subduction Zones” grant NC160025 funded by the Millennium Scientific Initiative (ICM) of the Chilean Government; the Chilean National Fund for Development of Science and Technology (FONDECYT) grant 1150321; and the German Science Foundation (DFG) grant STR373/41-1.