In the event, no collision occurred and no violent avoidance manoeuvre was necessary. The decisions of both ships A and B to alter course to carry out their overtaking manoeuvres (rather than maintaining their original - collision - courses) resulted in the collision probability ceasing to exist.

　

 Had the relative positions of all four vessels been different at the start of the run, a different outcome may have resulted, but the way in which the simulation model resolved this particular multiple encounter situation is interesting, if not instructive.

　

　

 Decisions relating to coast proximity and a nearby encounter are similar to those relating to course changes discussed in Section 4.2.1. It could be argued that the route structure used in the program should take vessels clear of any coastline. While this is true to an extent, giving each vessel some degree of autonomy allows it to move outside the (fictitious) route boundaries when making an avoidance manoeuvre. It could therefore move close to a nearby coast.

　

　

　

 The decision-making process is therefore the same as that used in the encounter/navigation dilemma, but this time with more weighting placed on the rudder angle used to avoid the coast. At present this decision is not implemented in the model, but it is always possible to use the manual navigation option for a situation of particular importance.

　

 The key output of the Dymitri model consists of "encounters" which occur when vessels within the model activity take steps (change of course and/or speed) to avoid another vessel. These events are not collision events in themselves but the initiating events for a potential collision. The likelihood of an incident then occurring embraces a wide series of human factors associated with perception of the vessel masters, the decisions taken, bridge team co-ordination, and the interaction of both vessels as a collision is averted, or not. The complexity of modelling these elements has been addressed previously.

　

 The likelihood of an incident actually occurring has been developed for the Dymitri model on the basis of "Active" encounters, where the nature of the encounter (crossing, head-on, or over-taking) is factored in the initial stages. Analysis and "expert group consensus reported by the US Coastguard [ref. 4], has proposed that different types of encounters should have different relative weightings and the following values have been adopted within the model during recent studies to develop active encounters, from the total number of incidents logged:

　

・0.05 for overtaking encounters;

・0.65 for crossing encounters, and

・0.30 for head on encounters.

　

 Figure 6 illustrates a comparison of total active encounters and historic incident records for five recent projects addressed within Hong Kong waters. The assembly of the data points on a straight line (with a correlation factor R>0.99) illustrates an excellent correlation between model output of active encounters and historic collision experience over a wide range of traffic levels in the busy waters of Hong Kong.

　

 Model output may be generated in a flexible manner to develop the total number of anticipated collisions for different waterspaces, vessel types and time-periods. However the search for "acceptability" of differing collision, injury and fatality rates requires a benchmark for assessment. Marine traffic risks may best, for comparative purposes, be expressed in terms of Potential Loss of Life (PLL) where the frequency of incidents and likely number of casualties is developed for any given waterspace with respect to the population at risk, which for most marine collision incidents is the volume of crew and passengers passing through the waterspaces. Both these parameters can be readily developed from the model output and comparison made against specific risk criteria.

　

 While documents such as the IMO Interim Guideline (Reference 5), UK's Port Safety Code (Reference 6), and IALA (Reference 7) all support the use of Quantitative Risk Assessment techniques, there are no international guidelines on the acceptability of different PLL for marine transport operations. In recent studies comparison has been made against local Hong Kong criteria, (Reference 8) which has appeared well suited to discriminating between risks which are "Acceptable", "Unacceptable", or must be reduced to "As Low As Reasonably Practicable" (ALARP). However, we await the development of an international standard to provide further guidance on this issue.

　

 While other factors not included in the model (i.e. visibility, currents, navigator aptitude and interaction) undoubtedly impact the likelihood of an incident given an initiating event, the fact that good accuracy can be achieved without including these elements suggests strongly that they are of a lower importance than those aspects included in the model.

　

 It is clear that use of a dynamic traffic simulation model solves some problems of similitude, but raises many others.

　

 Experience gained through using the model in a variety of studies suggests that good predictions of collisions are obtained in the port of Hong Kong, where traffic levels are high. This would suggest that the model provides a sufficiently realistic simulation of the present situation and gives some confidence in its ability to predict the future. It will in future be interesting to apply the model to less congested waterspaces to identify if the existing level of accuracy is maintained.

　

 While apparently of high accuracy for congested waterspaces, where traffic "mass flow" dominates the questions raised in the development of the Dymitri model are nonetheless intriguing. They hark back to human factor issues, raised at many MARSIM conferences in the past when the development of a "pilot" model were being considered (Reference 9 for example). The modelling of how various shiphandling decisions are made is difficult, not least because of human variability and the need for "real world" data. Reference 10 was an early attempt to measure this important matter in which mariner's decision times were studied. It is clear however that simple deterministic decision rules, while useful, can only go so far and an approach using fuzzy logic would be preferable. This in turn would lead to computational problems, a not insignificant consideration if upwards of 10,000 vessels are being dealt with at a given time.

　

 It may therefore be concluded that, although the use of dynamic traffic simulation models, with some elements of autonomy given to each vessel, is in its infancy, it has significant value as a tool for operational research in marine traffic. Further developments are likely to be in the area of decision-making, outlined above, with the inclusion of more realistic manoeuvring models a possibility for some studies.