Society benefits considerably from large scale industrial activities. However, these activities can have undesirable side effects which must be adequately controlled and monitored; one motivation for this thesis is risk assessment. Industrial processes often involve large amounts of hazardous, liquified gases. Accidental release of such substances poses a threat to nearby populations and the assessment of this risk is carried out in the UK by the Health and Safety Executive.

Accidentally released gases are often denser than air and this thesis addresses the physics of dense gas dispersion. As dense gas clouds tend to adopt low-lying configurations, a shallow layer model is indicated for simulating dense gas dispersion: a cloud is described in terms of depth averaged quantities such as contaminant concentration. If vertical accelerations are small, a hydrostatic pressure distribution is appropriate and the shallow water equations may be used.

This thesis presents a shallow layer model for dense gas dispersion. The model is time dependent and two dimensional. This type of model has been comparatively neglected up to the present but is useful because it is capable of simulating the effect of complex terrain such as valleys and mountain ranges.

The special physics of the leading edge is handled by augmenting the shallow water equations: extra terms are added that account for the interaction between the ambient fluid and the dense layer. In this manner the front Froude number may be fixed.

The model has a number of free parameters, which have to be empirically determined. The parameters used were either chosen on theoretical grounds, or taken from earlier work.

A computational model has been developed to simulate the mathematical model. This model differs from previous work in being time dependent and fully two dimensional. The model uses the flux correction scheme of Zalesak, generalized to account for complex terrain. The computational model is checked against a number of theoretical results.

The model is then used to simulate the large scale dense gas dispersion field trials carried out at Thorney Island (instantaneous and continuous), and its predictions are compared with the experimental results. In general, there is no evidence to suggest that changing the entrainment parameters would give better agreement. Some assessment of the sensitivity of the model to the free parameters is made.

Model predictions are shown to agree broadly with a number of integral models whose parameters were based on these experiments; cloud averaged concentrations as a function of downwind distance and time were considered.

Model predictions are then compared with laboratory-scale experiments in which dense gas was released over different slopes in a calm ambient. Instantaneous and continuous releases were considered

A case study of a real hazard site, using the present model, is presented.