Generating LoD2 City Models Using a Hybrid-Driven Approach: A Case Study for New Brunswick Urban Environment
Today 55% of the world's population lives in urban areas, a proportion that is expected to increase to 68% by 2050 (UN, 2018). 3D city models can be used to prepare for the future city, enabling informed analysis and sustainable development. Based on the Open Geospatial Consortium (OGC) standard, i.e. cityGML, 3D city models can be produced in different levels of detail (LOD). CityGML-3 introduces five predefined LODs (LOD0-4), with LOD0 being a building footprint and LOD4 being a realistic model representing the exterior and interior of the buildings. Currently, LOD0 and LOD1 are available for most cities in developed countries while LOD2+ are superior for informed analysis in different applications such as disaster management and insurance. However, with the current status of knowledge and technology, the production, storage and maintenance of such models are very time-consuming and expensive. This paper presents an initial study for 3D city model generation with a focus on the urban structure of New Brunswick, Canada, which is an introductory part of a larger project for 3D city modelling and maintenance in Canada. This paper intended to explore existing off-the-shelf 3D city modelling products and check their accuracies. Furthermore, inspired by existing literature, we proposed a decision-tree-based methodology for LoD2 3D city model generation, which follows a combination of data-driven and model-driven approaches, i.e. a hybrid approach. We tested the quality of the final 3D models using different metrics such as overall accuracy, Kappa Coefficient, Root Mean Square Error (RMSE) and slope difference. Besides, we compared our results to two off-the-shelf products, namely Schematic Local Government City Engine LOD2 (SLGCE) modelling and OpenStreet Map City Engine (OSMCE) LOD2 modelling. The results showed that the proposed hybrid approach achieved higher accuracies using the mentioned metrics. This paper also discusses the pros and cons of the proposed method and offers insights for improving the results even further.