A Method for the Analysis and Processing of Marine Collision Data

Abstract

This paper presents a description of a general method for the evaluation of marine collision risk. The approach differs considerably from existing methods in this area, in that it combines analytical modelling in a special way to analyse historical data. The method makes it possible to find the collision probability of a specific ship in any particular location, taking into account such location factors as area geometry, ship movements and speeds. It also constitutes a basis for worldwide data collection.In view of the growth of marine traffic and the increase in the transportation of hazardous materials over the past two decades, considerable efforts are being made towards the development of maritime risk evaluation methods, the main objective being the quantification of ship collision probability. The methods developed to date can roughly be subdivided into the following two groups: (a) kinematic models (see for instance refs. 1–6); (b) historical data analysis (see for instance refs. 7–17). It is a well-established fact that most collisions are due to human failures, and it is widely accepted that human behaviour cannot be described well by mathematical formulae. Kinematic models are therefore not recognized as suitable for estimating this probability. Historical data analysis estimates the collision probability as the ratio of the number of observed collisions to the number of ship movements. Probabilities estimated in this way are only correct in a probabilistic sense when the assumption is satisfied that every ship movement has the same probability of collision. This assumption is rarely fulfilled since most ships often pass each other without collision danger. In addition, this approach is often hampered by lack of data for a specific traffic area, while the application of worldwide data cannot be considered realistic. As far as we know, there does not exist a method which can transfer, in a statistically justifiable manner, marine traffic data from one area to another. Moreover, neither of the above-mentioned approaches is appropriate to analyse alternative routes in the same area, or under changing circumstances.