Constructing three-dimensional urban models for Urban Building Energy Modeling using UAV oblique imagery

In sustainable urban development, Urban Building Energy Modeling (UBEM) and smart city initiatives depend on computable three-dimensional (3D) urban models. To enhance computational efficiency, research often uses Level of Detail (LOD) 1 simplified models for simulations. While LOD1 models simplify building geometry, obtaining accurate building height data typically requires substantial manual effort.

Balancing model simplification with simulation accuracy is crucial for reliable and efficient results. This study proposes an automated method using open-source electronic maps and UAV oblique imagery to address 3D modeling challenges caused by incomplete databases. We developed computable 3D urban models at various LODs, with the highest detail achieved at LOD2.1. EnergyPlus was used to examine the relationship between LOD and energy use.

The study follows five key steps: (1) Data Collection and Preprocessing—gather building footprints and UAV point clouds, refine point clouds to create voxels; (2) Roof Point Acquisition—extract and cluster roof points to determine building height; (3) Roof Type Identification—use machine learning to predict roof types; (4) Parametric Modeling and Simulation—construct databases and models at varying LODs for UBEM; (5) Results Comparison—evaluate modeling time, modeling accuracy, and energy use across LODs. This method demonstrates high accuracy, achieving a roof face identification accuracy of 92%, with building height errors within 1 meter and an average modeling time of just 2.5 seconds per building.

Additionally, the roof type prediction model based on point cloud features attained an impressive AUC of 0.93. Increasing LOD slightly extended the energy simulation time, with LOD2.1 increasing simulation time by 6.5 seconds per building compared to LOD1_H_mean. The baseline model and LOD1_H_mean’s Energy Use Intensity (EUI) exhibited a MAPE of 3.02% and an RMSE of 4.41 kWh/m², underscoring the importance of high-LOD models for accurate simulations. It was found that LOD has a greater impact on heating than cooling, and building height errors significantly affect cooling simulations.

These findings highlight the importance of accurate building height in hot regions for effective cooling and high-LOD modeling in cold regions, with assuming flat roofs where the type is uncertain helping to minimize errors. In summary, this study presents methods for automating the construction of high-LOD models for UBEM, also applicable to microclimate simulation and CIM.

It demonstrates the significant impact of high-LOD on simulation accuracy and provides solutions for various simulation needs and data limitations, advancing UBEM and smart city development.