A modelling framework for simulating large geographical systems of agents and fields

Computer models mimicking processes in the real world can be used to increase our understanding of how environmental systems work, and to make predictions about future states of these systems. For example, flood forecasting models are routinely used to decide whether or not measures must be taken to counteract the effects of high water levels. Modelling frameworks contain pre-built functionality for developing models. This allows them to be developed in less time. The focus of this thesis is on alleviating two limitations of current frameworks. The first is their limited support for representing and manipulating different kinds of information about a geographical system in a single model, like the amount of precipitation, and the location of houses. A uniform way of representing information is presented, which can be used to better represent the state of a system. The second problem focused on in this thesis is the limitation of modelling frameworks to support creating models that use all the available hardware efficiently. A new framework is presented which distributes computations automatically over the available hardware in a single computer or in multiple nodes in a supercomputer. This allows larger models to be created, that finish executing in less time, without the model developer having to know how this works. The results presented in this thesis support framework developers in making informed decisions about how to make the next generation of modelling frameworks for simulating large and complex geographical systems.