Thesis Defence: A Data-Driven Recalibration Methodology for Spatially Transferring an Agent-Based Integrated Urban Model

Ifratul Hoque, supervised by Dr. Mahmudur Fatmi, will defend their thesis titled “A Data-Driven Recalibration Methodology for Spatially Transferring an Agent-Based Integrated Urban Model” in partial fulfillment of the requirements for the degree of Master of Applied Science in Civil Engineering.

An abstract for Ifratul Hoque’s thesis is included below.

Defences are open to all members of the campus community as well as the general public. Please email mahmudur.fatmi@ubc.ca to receive the Zoom link for this defence.

---

Abstract

STELARS is a large-scale Integrated Urban Model that simulates land use, vehicle ownership, and transportation decisions within an agent-based framework. Vehicle ownership in STELARS is conceptualized as a two-stage process involving vehicle transaction and type choice, represented by six interrelated micro-models. Traditional model updating approaches in enhancing transferability have largely focused on recalibrating a single model at a time and have not been designed to handle large, interdependent systems composed of multiple micro-models. This limitation hinders the spatial transferability of such complex frameworks when applied to new geographic regions with different socioeconomic and behavioural characteristics. To address this challenge, this study proposes a data-driven recalibration framework that enables updating of all micro-models within the vehicle ownership module, thereby enhancing spatial transferability while preserving structural consistency. The proposed framework employs a Steady-State Elitist Genetic Algorithm assisted by a Random Forest surrogate model to determine recalibrated parameter sets. For the recalibration, initial parameter sets were generated using Latin Hypercube Sampling within ±100 percent of original calibrated values, and Mean Absolute Error was used as the objective function to minimize. The recalibration was conducted to transfer the vehicle ownership module from the Okanagan Region to the Greater Vancouver Area, British Columbia, utilizing aggregate data sources. The proposed strategy yielded promising results: the simulated household vehicle ownership distribution in the Greater Vancouver Area matched the observed values within a 3 percent discrepancy range. Additionally, 62 percent of the parameters in the vehicle type models after recalibration remained within the ±75 percent range of their original estimates, indicating that most behavioral patterns observed in the Okanagan model were largely retained in the transferred context. This stability across regions demonstrates the robustness of the original model specification and suggests that while regional differences exist, the overall behavioural structure of vehicle ownership decisions remains transferable. Overall, the proposed framework provides a scalable, resource-efficient, and data-driven approach for transferring complex urban models to data-scarce contexts, thereby broadening their applicability for policy and planning analyses.