Consideration of stabilisation techniques is essential in the development of physical models if they are to faithfully represent processes over a wide range of scales. Careful application of these techniques can significantly increase flexibility of models, allowing the computational meshes used to discretise the underlying partial differential equations to become highly nonuniform and anisotropic, for example. This exibility enables a model to capture a wider range of phenomena and thus reduce the number of parameterisations required, bringing a physically more realistic solution. The next generation of fluid flow and radiation transport models employ unstructured meshes and anisotropic adaptive methods to gain a greater degree of flexibility. However these can introduce erroneous artefacts into the solution when, for example, a process becomes unresolvable due to an adaptive mesh change or advection into a coarser region of mesh in the domain. The suppression of these effects, caused by spatial and temporal variations in mesh size, is one of the key roles stabilisation can play. This thesis introduces new explicit and implicit stabilisation methods that have been developed for application in fluid and radiation transport modelling. With a focus on a consistent residual-free approach, two new frameworks for the development of implicit methods are presented. The first generates a family of higher-order Petrov-Galerkin methods, and the example developed is compared to standard schemes such as streamline upwind Petrov-Galerkin and Galerkin least squares in accurate modelling of tracer transport. The dissipation generated by this method forms the basis for a new explicit fourth-order subfilter scale eddy viscosity model for large eddy simulation. Dissipation focused more sharply on unresolved scales is shown to give improved results over standard turbulence models. The second, the inner element method, is derived from subgrid scale modelling concepts and, like the variational multiscale method and bubble enrichment techniques, explicitly aims to capture the important under-resolved fine scale information. It brings key advantages to the solution of the Navier-Stokes equations including the use of usually unstable velocity-pressure element pairs, a fully consistent mass matrix without the increase in degrees of freedom associated with discontinuous Galerkin methods and also avoids pressure filtering. All of which act to increase the flexibility and accuracy of a model. Supporting results are presented from an application of the methods to a wide range of problems, from simple one-dimensional examples to tracer and momentum transport in simulations such as the idealised Stommel gyre, the lid-driven cavity, lock-exchange, gravity current and backward-facing step. Significant accuracy improvements are demonstrated in challenging radiation transport benchmarks, such as advection across void regions, the scattering Maynard problem and demanding source-absorption cases. Evolution of a free surface is also investigated in the sloshing tank, transport of an equatorial Rossby soliton, wave propagation on an aquaplanet and tidal simulation of the Mediterranean Sea and global ocean. In combination with adaptive methods, stabilising techniques are key to the development of next generation models. In particular these ideas are critical in achieving the aim of extending models, such as the Imperial College Ocean Model, to the global scale. Publisher URL: http://hdl.handle.net/10044/1/5496