Evaluation of the propagation of oil spills in ports through Artificial Neural Networks

Assessing oil spill propagation in port terminals is crucial for environmental protection. The dynamics of an oil spill are related to several factors, including the type of oil, volume, spill points, prevailing hydrodynamics and environmental conditions, oil trajectory and fate, and the geographical zone of potential impact. A system that enables the rapid monitoring and prediction of oil spill propagation is crucial for addressing incidents quickly and effectively. Stochastic or numerical model approaches can be used. In this sense, implementing technologies that predict the port facilities involved in the propagation of an oil spill can improve future response capabilities to incidents. The primary objective of this work is to utilise Artificial Neural Networks (ANNs) to address oil spills and improve detection, prediction, and response capabilities, leveraging their ability to analyse large amounts of data and recognise complex patterns. As a case study, this research focuses on the Port of Augusta (Sicily, Italy), one of the most important ports in Italy. Results revealed that using too few training simulations (e.g., 1,000) led to underfitting, while overly complex networks with too many neurons per layer resulted in overfitting and reduced performance. The optimal balance was achieved with 15 neurons per hidden layer and 9,000 training samples, yielding low RMSE values on unseen data and performance comparable to that of larger datasets. This indicates that computational efficiency can be improved without sacrificing accuracy, highlighting the robustness of machine learning approaches for modelling complex coastal phenomena, such as oil spill dispersion.