Abstract: With the gradual opening of the Arctic Passage, attention is increasing on the assessment of risk associated with sailing across the Arctic. In addressing the problem of the lack of visibility data in the Arctic region, this study proposed a technique for generating gridded visibility data by fusing reasoning results and measured visibility data. Through determination of the factors influencing visibility, the generation of sample data sets, comparison of the fitting effects between an artificial neural network and a Bayesian network and revision of the inferred results, an inference model of gridded visibility data was constructed based on an artificial neural network and the reasoning and interpolation results of visibility were compared. In comparison with the interpolation technique, the results showed the inference model produced values that were more accurate. Therefore, the developed technique could provide important reference material for risk assessment studies of the Arctic Passage.

Key words: Arctic Passage, artificial neural network, visibility, gridded data