Automatic Roof Type Classification from Cityjson for Rule-Based Roof Structure’s Material Quantity Assessment

Abstract: Current city-scale material stock assessment relies on archetype-based estimation with limited geometric grounding or on city-to-BIM conversion workflows that are costly and sensitive to representation detail. Both hinder auditable roof typology and quantity derivation from open 3D city models. This paper proposes an interpretable CityJSON pipeline that merges roof meshes into planar patches, extracts roof metrics (area, projection, slopes, span), performs rule-based roof classification (flat/shed/gable/hip/complex), and computes rule-based roof-layer and timber quantities. On 65 labelled buildings, LoD2 achieves 0.785 accuracy (macro-F1 0.705);BIM comparison on six buildings supports feasibility. The contribution is a reproducible, explainable roof-to-quantities method with quantified LoD sensitivity.