Risk assessment of chemical release accident triggered by landslide using Bayesian network

This study evaluated the risk of 461,260,800 scenarios of chemical release accidents triggered by landslides. Several industrial accidents triggered by landslides have recently occurred in Japan; however, only a few studies have ana-lyzed the impact of landslide-triggered chemical release accidents on the surrounding areas. Bayesian networks (BNs) have recently been used in the risk assessment of natural hazardtriggered technological accidents (Natech) to quantify uncertainties and develop methods applicable to multiple scenarios. However, the scope of BN-based quanti-tative risk assessment is limited to the risk assessment of explosions triggered by earthquakes and lightning. We aimed to extend the BN-based risk analysis methodology and evaluate the risk and the effectiveness of the countermeasures for specific facility. A methodology was developed to assess human health risk in the surrounding areas when n-hexane was released and dispersed into the atmosphere due to a landslide. Risk assessment results showed that the societal risk (representing the relationship between frequency and number of people suffering from a particular harm) of the stor-age tank closest to the slope exceeded the Netherlands' criteria, which are the safest among the criteria in the United Kingdom, Hong Kong, Denmark, and the Netherlands. Limiting the storage rate reduced the probability of one or more fatalities by up to about 40% compared with the no countermeasure case and was a more effective countermeasure than using oil fences and absorbents. Diagnostic analyses quantitatively showed that the distance between the tank and slope was the main contributing factor. The catch basin parameter contributed to the reduction in the variance of the results compared to the storage rate. This finding indicated that physical measures, such as strengthening or deepening the catch basin, are essential for risk reduction. Our methods can be applied to other natural disasters for multiple scenarios by combining it with other models.