Evaluating the Impact of LiDAR-Derived Digital Surface Models on Rooftop Solar Potential Assessment Using GIS and Machine Learning

Abstract

Solar Cadastres are key for urban solar energy assessment. These cadastres are Geographic Information System (GIS) -based tools that map and evaluate the solar potential of rooftops and land parcels supporting renewable energy integration by enabling data-driven decisions for solar panel deployment. A critical component of accurate cadastre development is the use of Digital Surface Models (DSMs) generated from Light Detection and Ranging (LiDAR) point cloud data, which provide detailed rooftop geometry, shading, and elevation information. This study investigates how different interpolation techniques used to generate DSMs from LiDAR data impact rooftop solar potential estimation in Qatar. It contributes through: (1) a literature review on solar cadastres, GIS, DSMs, and interpolation methods, (2) a GIS-based case study evaluating the impact of four interpolation methods, and (3) a machine learning-based approach for DSM generation. The literature review examines solar cadastres, GIS technologies, and LiDAR-based DSM generation, with a focus on how interpolation techniques influence the accuracy and resolution of DEMs, DSMs, and DTMs. Emphasizing DSMs for their role in rooftop solar assessments, it compares traditional and non-traditional methods across studies to identify the most effective techniques for each model type. The review also explores emerging machine learning approaches for DSM rasterization, highlighting their potential to improve solar cadastre accuracy and urban solar mapping. The case study utilized LiDAR data (3.06 pts/m²) to generate DSMS using TIN, IDW, NN, and RBF-TPS interpolation methods at fixed resolution of 0.5m. These DSMs were combined with ground-measured solar irradiance data to estimate rooftop solar energy potential. TIN delivered the most accurate results, with rooftop area percentage errors of 5.671%, 11.192%, and 2.127% compared to Google Earth for three buildings, and annual solar potential of 1884.491 kWh/m², 1931.027 kWh/m², and 1873.752 kWh/m² respectively, closest to the real GHI values. IDW performed similarly, while NN and RBF-TPS showed reduced accuracy due to edge blurring and smoothing. A Gradient Boosting Regression (GBR) model was also implemented to generate DSMs from LiDAR data, achieving an R² score of 0.97. While numerically accurate, the model showed visual limitations such as smoothed rooftop edges and pixelation, highlighting the need for further refinement in ML-based DSM generation. Overall, the findings reinforce the importance of DSM quality in solar assessments and Solar Cadastre development. Optimizing interpolation methods—both traditional and machine learning-based—can enhance rooftop PV planning. Future work will focus on improving DSM resolution, shadow analysis, and machine learning integration for more robust urban solar mapping.

Date of Award	2025
Original language	American English
Awarding Institution	HBKU College of Science and Engineering