An incremental method for building and evaluating agent-based models of systems of cities

Clémentine Cottineau (1, 2, 3) , Paul Chapron (3, 2) , Romain Reuillon (3, 2)

1 UP1 UFR08 - Université Paris 1 Panthéon-Sorbonne - UFR Géographie
2 GC (UMR_8504) - Géographie-cités
3 ISC-PIF - Institut des Systèmes Complexes - Paris Ile-de-France

This paper offers a feedback on a modeling experience with the MARIUS model. It aims at exposing a method of parsimonious modeling using intensive quantitative evaluation. By confronting simple mechanisms of interurban exchanges with Soviet data of urban evolution, we explore the capacity of two models (the most parsimonious and a complexified one) to satisfy the macro-goal of reproducing the system's structure as well as a realistic simulation of micro-dynamics. This confrontation relies on a historical database and is evaluated through a calibration process and a sensitivity analysis using distributed computation. We show that the most simple model of spatial interactions we could think of was unable to reproduce the actual evolution of Soviet urbanisation between 1959 and 1989. However, by complexifying the model's structure by only two mechanisms, we were able to find results in accordance with our evaluation goals.

Mots clés: ABM; Simulation; model-building; system of cities; Evaluation; Incremental

Domaines: Géographie; Méthodes et statistiques

Clémentine Cottineau, Paul Chapron, Romain Reuillon. An incremental method for building and evaluating agent-based models of systems of cities. 2014. ⟨halshs-01093426⟩