Analysis of the influence of human errors on the occurrence of coastal ship accidents in different wave conditions using Bayesian Belief Networks

•A study is made to assess the human error contribution in ship accidents in different weather conditions.•A Bayesian Belief Network model is developed, which includes variables related to the different wave conditions.•The accident database of the Portuguese Maritime Authority is used, which includes records of 1997–2006.•Several significant wave height databases are used.•The results show high risk acceptance in fishing vessels and a low risk perception in recreational vessels. The paper describes a study aiming to assess the human error contribution in ship accidents in different weather conditions and the contribution that high significant wave heights have in the occurrence of certain accident typologies. To this aim, a Bayesian Belief Network model is developed, which includes variables related to the maritime accident but also to the different wave conditions. For the quantification of the conditional probability tables the maritime accident database of the Portuguese Maritime Authority is used, which includes 857 validated accidents registered in the period 1997–2006. Similarly, several significant wave height databases are used to populate the node correspondent to this variable. The importance of accurate estimation of the significant wave height is also discussed. To substantiate this discussion a comparison between hard evidence (ε) and a soft one (μ,σ) is performed for the significant wave height node of the BBN model. The application of different combinations of evidence in the model allows the identification of patterns of influence of the human error cause in comparison with others, namely with the sea and weather one. The results show one apparent high-risk acceptance within the crews of the fishing vessels and low risk perception in the recreational vessels. Based on the results, are provided recommendations to decrease the risk associated to specific probable causes.