Author(s): Giulia Bonanno; Elisa Castro; Claudio Iuppa; Federico Roman; Carla Faraci; Luca Cavallaro
Linked Author(s):
Keywords: No Keywords
Abstract: This study proposes a two-step artificial neural network (ANN) framework for oil spill diffusion prediction in the Port of Augusta (Italy) based on tabular, non-temporal data. The proposed approach aims to provide a reliable and computationally efficient alternative to traditional numerical models, which are computationally expensive and therefore unsuitable for real-time or early-warning applications. The port domain was discretized into 105 spatial segments and 90 spill locations, generating 14040 scenarios per segment through systematic variation of eight environmental and spill-related parameters. The framework consists of a binary classification model that predicts oil arrival at each segment, followed by a regression ANN that estimates the oil arrival velocity. The classification model achieved an accuracy of 0.95 on unseen data, while the regression model attained a percentage global RMSE of 6%.
Year: 2026