Tropical cyclones (TCs), especially intense hurricanes, pose a significant threat to coastal areas due to the destructive forces of storm surges and wind-generated waves. These forces can cause severe coastal erosion and flooding within a few hours. The present study aims to address the limitations of computational cost of morphodynamic models and data scarcity by developing a surrogate model to better address the morphodynamic beach response to storms event. Few studies have focused on developing surrogate models specifically dedicated to temporal coastal beach erosion forecast. A great subset of TC is selected from a synthetic database, considering time-dependent evolution of the oceanographic variables. Secondly, a 2d XBeach model is used to compute the morphological responses to selected storms. Thirdly, we adapt and test an Artificial Neural Network-based surrogate model to mimic XBeach at low computational cost. Results are presented at the study site of Le Carbet Beach, Martinique, Lesser Antilles.