Understanding the relationship between urban morphology and solar potential in mixed-use neighborhoods using machine learning algorithms

Extant studies on urban morphology and solar energy potential have mainly focused on the overall urban (macro) level or the individual building (micro) level, mostly using traditional linear regression methods. This paper takes a typical mixed-use developed city, Adelaide, as an example and demonstrates a complete workflow (data collection, data pairing, model training, model interpretation, and model application) using machine learning algorithms to better understand the relationships between urban morphology and solar potential at an urban neighborhood (meso-level) scale. The artificial neural network model has the highest accuracy, and six urban morphology indicators (building density, plot ration, building height, building floors, variance of height, and variance of volume) are shown to be influential in predicting urban solar potential. The model also reflects the interaction between urban morphological indicators and their nonlinear impact. Futhermore, five typical morphological prototypes are identified. The average solar potential of Cluster 1 is 1402.53 kWh/m2/year, 4.3% higher than Cluster 3, 11.2% higher than Cluster 4, 21.2% higher than Cluster 2, and 36.8% higher than Cluster 5. This study provides a standard, simplified workflow in the early planning and design stage to assess the urban form and related land use for harvesting solar energy.