Building and exploring systems of cities models via high performance computing

We present a mix of tools and methods used to build a family of spatial agent based models (ABM) aiming to reproduce the evolution of some systems of cities.
We adopted an incremental modeling process to formalize alternative hypotheses about urbanisation dynamics and high performance computing techniques to assess their ability to reproduce observed patterns.
The MARIUS family of models focus on the evolution of the post-Soviet system of cities in the past 50 years and aims to reproduce the macro structuration of the system (e.g. hierarchization of cities sizes) by simulating the micro-level spatial interactions (e.g. exchanges of goods).
Targeting equifinality and parsimony, MARIUS models are incrementally built, using hierarchically ordered mechanisms, from the more general to the most specific, concerning interactions and environment relationships.
MARIUS is framed by the platform OpenMOLE, which supports advanced calibration features and distributed computing environment integration, allowing to investigate strategically the large space of models behaviors.
It is then possible not only to reveal the best parameter set to fit the data, via automatic calibration, but also to explore the possible models structure itself, by testing alternative mechanism and various levels of mechanisms generality and specificity.