Climate zoning for buildings with spatial clustering: Moving beyond K-means

Abstract

Climate zoning for buildings (CZB) plays an important role in optimizing building design strategies and related energy consumption according to local climatic conditions. However, clustering-based CZB methods group data points by attribute similarity only, ignoring the spatial nature of climate data. This study looks into spatially constrained clustering (SCC) for CZB. Explicitly incorporating geographic relationships into partitioning, it facilitates zones that are both climatically meaningful and geographically homogeneous. For the first time in CZB research, a set of three specific SCC algorithms were applied for CZB. Their influence on formation of climate zones, geographical homogeneity, and energy misclassification was investigated and compared with K-means, which served as a baseline for comparison. Clustering performance was assessed using statistical indices (Calinski-Harabasz and Davies-Bouldin), two spatial autocorrelation (SAC) measures (local and global Moran's I), and climate zoning energy needs overlap expressed through mean overlap value. The results show that while K-means tends to optimize internal data partition, it usually leads to spatially dispersed zone boundaries, which are shown by the lowest global autocorrelation scores, with values often below 0.50. On the other hand, SCC explicitly enforcing geographical continuity (often at the expense of internal statistical cohesion), does not translate it into higher misclassification of energy-related metrics, making it well suited for CZB applications. The results suggest that spatial coherence is not just a cosmetic improvement, it is essential for CZB maps that can be effectively applied to guide building design, policy development, and energy efficiency strategies.