Virtual Horizon Method: Fast shading calculations for UBEM using lidar data rasterization

Solar shading simulation is an essential first step for many urban building energy modeling (UBEM) and photovoltaic simulation tasks. This paper introduces a novel point cloud-based solar shading algorithm designed to overcome the efficiency and scalability limitations of current shading calculation methods used in urban building energy modeling.

Traditional methods, such as Energy Plus, use a polygon-clipping (Sutherland & Hodgman, 1974) or OpenGL-based pixel-counting (Hoover & Dogan, 2017; Jones et al., 2012) to determine a surface’s sunlit fraction, sky view, and horizon visibility and then combine these with the Perez Sky Model. At the same time, the Radiance Daysim or the two-phase method (Subramaniam, 2017) method offers full, octree-accelerated ray traces that also account for indirect contributions.

While these methods are widely considered state-of-the-art, they pose limitations in urban applications where modelers encounter large models with high polygon counts where computational overhead becomes unfeasible. In addition, modelers are often required to reconstruct the mesh geometry of buildings, terrain, and vegetation from Light Detection and Ranging (LiDAR) point clouds. This mesh reconstruction is complex and often requires a significant overhead.

We present the Virtual Horizon Method, which can directly leverage LiDAR point clouds and does not require reconstructing the mesh geometry of the urban context (Bognár et al., 2021). Instead, we rasterize the point cloud data into a height map with arbitrary resolution. We then query the heightmap for a virtual horizon line for any point of interest within the model.

We use the virtual horizon to determine the visibility of the horizon line, sky dome, solar disc, and circumsolar region, and we use the Perez Sky model to compute incident radiation at a given sensor point. Preliminary results show that the algorithm maintains high performance for high sensor density and large context model sizes without significantly sacrificing accuracy. We demonstrate that our method is orders of magnitude faster than a conventional radiation mapping workflow using Radiance and ClimateStudio while maintaining a mean accuracy of 91.69% for annual irradiance predictions, assuming a Lambertian surface model.