Re-thinking the region

Abstract

In this thesis we present a computational framework, which allows simulating residential location choice. The specific application of the model focuses on representing populations under disaster risk. The aim of the tool is to enable public planning agencies to explore the synthesized households’ location choice under risk and different spatial, financial policy scenarios. The simulated urban system is represented as an agent-based model, where the households are the agents choosing between discrete options to relocate from one house to another. Their choice behavior is built upon a notion, that people make decisions based on regret. This is done with the help of Random Regret Minimization (RRM) model, allowing to capture varying levels of regret (profundity) and enabling incorporating multiple attributes of different dimensionality. Given the agent heterogeneity and the changing availability of the building stock, the choice sets are dynamic. Therefore, the traditional RRM approach would not return stable results: with every new option housing available it would have to be re-calibrated. As a solution, we propose re-interpreting the classical c{RRM} model by scaling the beta values by the choice set attribute variance. This allows to represent $eta$ as a unit-less preference weight, associated with a homogeneous population group.

We apply the framework to two different scale case-studies in an earthquake-prone area in Groningen province, The Netherlands. The data used for simulation includes several public and private spatial datasets, as well as aggregate level statistical data to synthesize the properties of the households. As a showcase, we applied the simulation assuming homogeneous and equal preference weights for 7 optimization criteria. These criteria relate to static properties of the building stock and household-related dependencies to the network (job, school locations). To further exhibit model usability within public sector agencies, we also apply the model on several financial policy scenarios. The output of which is captured on both aggregate and semi-disaggregate levels, allowing for interactive exploration of the effects of the proposed scenarios. The model outcomes correspond to the expectations set prior to simulation. It showcases convergence, anticipated optimization behavior and spatial patterns, corresponding to the building stock properties of the region.