Methodology based on fine spatial scale and preliminary clustering to improve multivariate linear regression analysis of domestic water consumption

Abstract

The increasing world population and the growing water demand generate a stress on water resources which become particularly scarce in arid urban areas. In this context, the present research aims at providing a better understanding of water consumption through a methodology that highlights the main factors involved in the domestic water demand in the city of Seville, Spain. To achieve this objective, water consumption was examined at the micro-scale level (census tract) for a better resolution and statistical power. Initially, fifteen predictor variables including sociodemographic factors and buildings characteristics were considered. A preliminary clustering of census tracts followed by a comprehensive multivariate linear regression analysis revealed that the average cadastral value (positive correlation) and the number of inhabitant per household (negative correlation) were the most significant predictors. Other significant variables included, for instance, the height of the buildings (positive correlation) and the residential density (negative correlation). In addition, a spatial analysis revealed a gradient of water consumption from higher values in the city centre to lower values at the periphery. These results could be used for urban planning and management purposes by the municipality and stakeholders in order to further lower the water footprint of the city of Seville.