The decentralized transition of urban energy systems necessitates highly accurate spatial assessments of rooftop photovoltaic (PV) potential. Historically, municipal energy models have relied heavily on two-dimensional building footprints (Level of Detail 1, or LoD1). This geometric oversimplification inherently assumes that roofs are entirely flat and structurally unobstructed, completely ignoring critical variables such as roof inclination, orientation, and architectural micro-shading. Consequently, this leads to a systemic and severe overestimation of viable PV capacity. While advanced 3D city models (LoD2 and above) mitigate these spatial limitations by detailing distinct roof structures, their generation is traditionally monopolized by expensive, proprietary software ecosystems. Furthermore, workflows that fuse multi-modal data—such as combining optical satellite imagery with digital elevation models—frequently suffer from stereoscopic misalignment and geometric distortion.
To bridge this methodological gap, this study presents a rigorous, fully open-source spatial framework developed entirely within the QGIS Free and Open-Source Software (FOSS) ecosystem. Focusing on a high-density urban sector in Birmingham, United Kingdom, the research relies exclusively on open-access, high-density airborne LiDAR (Light Detection and Ranging) point cloud data (.las/.laz formats) sourced from the UK Department for Environment, Food and Rural Affairs (DEFRA). By isolating LiDAR as the sole primary spatial input and utilizing native FOSS algorithms, this methodology bypasses proprietary barriers, ensuring high reproducibility, scalability, and democratized access for municipal planners seeking to generate smart city digital twins.
The workflow commenced with the extraction of base 2D building footprints using native point cloud segmentation algorithms. Following footprint extraction, a Digital Terrain Model (DTM) and a Digital Surface Model (DSM) were derived to establish the Normalized Digital Surface Model (nDSM), representing absolute relative building heights. A critical methodological intervention occurred regarding the treatment of inherent LiDAR topological noise. Standard raster interpolation techniques, such as Gaussian smoothing and 'Fill NoData' tools, are routinely used to "repair" line-of-sight occlusions where architectural features fall into the airborne sensor's shadow to create idealized, aesthetically pleasing solid models. However, spatial validation revealed that these smoothing algorithms aggressively distorted true physical boundaries and artificially altered actual roof inclinations. Consequently, these algorithms did not work effectively. Preserving the raw physical noise and data voids was deemed scientifically necessary to maintain authentic topological realities, even though this produced highly jagged vector boundaries in subsequent processing phases.
The potential incoming global solar radiation was computed directly over the continuous, un-smoothed DSM using advanced algorithms integrated into SAGA GIS (System for Automated Geoscientific Analyses). Rather than assuming static solar exposure, SAGA GIS dynamically models diurnal and seasonal solar paths based on Birmingham's latitude. Crucially, it performs continuous ray-tracing and viewshed analysis across the raw DSM. This ensures that the gross radiation values inherently account for complex urban micro-shadings—such as shadows cast by adjacent taller buildings or intersecting roof ridges—long before any vectorization or segmentation occurs.
To transition the spatial data from monolithic LoD1 blocks into semantically meaningful LoD2 roof structures, the unaltered DSM was utilized to derive continuous Slope and Aspect raster layers. Then, pixels from the aspect raster were reclassified into four cardinal-directional planes. The 2D building footprints were subsequently intersected with these aspect boundaries, effectively slicing the structures along their structural roof ridges into distinct, multi-directional segments. Because the framework preserved raw LiDAR noise, this intersection produced a large number of topological slivers. To resolve this, a rigorous heuristic geometric filter was applied, pruning meaningless slivers and unviable obstacle spaces smaller than 5 m², thereby isolating only the continuous macro-roof planes suitable for PV installation. The culmination of this semantic segmentation provides a scalable, empirical mechanism for quantifying and exposing the severe discrepancy between LoD1 models and physical reality. The framework explicitly contrasts the gross continuous footprint area of each building against its net viable roof segments. By filtering for physical constraints and isolating only favorable solar orientations (South, East, and West facing planes), the spatial analysis demonstrates a drastic reduction in the assumed available area.
Finally, a systemic approach was adopted for energy estimation. To translate the gross solar radiation—allocated to the segmented polygons via Zonal Statistics—into practical energy yield, the study parameterized the output using a macro-scale 0.60 usability coefficient. This conservative parameter, which captures 60% of the calculated global solar radiation, was deliberately selected to account for real-world deployment constraints that remain unmapped by raw geometry. These physical and operational limitations include inter-row panel spacing to prevent self-shading, necessary maintenance clearances, edge setbacks for safety, and systemic inverter losses. While this heuristic coefficient provides a robust baseline for municipal-scale estimations, future iterations of this framework could achieve even higher precision and empirical validation by calibrating the solar yield against local pyranometer field measurements or high-resolution global solar indices. The ultimate output of this methodology is a granular, building-by-building spatial inventory maintained entirely within a desktop 3D QGIS environment. For every individual structure, the framework computes a three-tier metric: the total segmented roof area, the geometrically constrained and optimally oriented installable area, and the realistic potential energy yield (kWh/year), visually demarcating high-yield exposures from non-viable planes based strictly on empirical geometric realities.
While this open-source framework provides a scientifically robust, accurate methodology for mapping geometrically constrained urban solar supply, it currently addresses only the production side of the energy equation. The spatial mapping of building-specific electricity consumption has not yet been integrated. Therefore, the immediate future trajectory of this research aims to incorporate open-source Energy Performance Certificate (EPC) data. This integration will facilitate a comprehensive supply-demand matching analysis, enabling the precise identification of buildings in the Birmingham study area capable of achieving true energy autonomy, thereby bridging the gap between theoretical solar potential and actionable urban energy resilience.