Space heating with gas as the primary fuel is a dominant contributor of residential energy consumption in UK and hence important for achieving UK's 2050 carbon targets. Whilst most cities in the UK have reduced their emissions per capita, there is a large variation of domestic gas consumption within cities. Past research suggests that the variations in gas consumption are as much as function of the quality of built environment, as it is of wider socio-economic-demographic features of households. Their combined influence on variations of residential heating consumption is however not well understood. This paper proposes a novel two-layer clustering framework to address this gap. The proposed framework constitutes of Gaussian Mixture Models and Hierarchical clustering and is illustrated through the analysis of residential gas consumption across London. Results show eight clusters of London Lower Super Output Areas (LSOAs) explained along 18 dimensions that include built-environment, social-economic and demographic information. These 8 clusters are used as variables to further cluster four groups of local authorities. The distinct gas consumption related properties of the resulting clusters explain the sources of variations in gas consumption across clusters and indicate possible directions for future fine-tuning of local policies.