Suitability of geographically weighted regression in water demand prediction.

Abstract

The aim of this study was to evaluate the use Geographically Weighted Regression (GWR) in predicting water demand in Lusaka City. The specific objective was to analyse the spatial distribution of water demand in the city, examine spatial relationships between water demand and its predictors and to assess the suitability of Geographically Weighted Regression for predicting water demand. Data of water demand (dependent variable) in cubic meters per day was obtained from a study done in 2011 by Lusaka Water and Sewerage Company. Independent variables used were population, income, tertiary education attainment, property values, level of spatial development, irrigated hectarage and temperature. Geographic Information System (GIS), Ordinary Least Squares (OLS) and GWR regression models were used to analyse the distribution and correlation between water demand its predictors and to predict it for the year 2035. The results showed that overall water demand was generally highest in high density neighbourhoods and lowest in low density neighbourhoods. Whereas percapita water demand was generally highest in low density and lowest in high density neighbourhoods. Further, the study found that water demand was not significantly related to temperature, irrigated hectarage and percentage of tertiary education attainment and property values but was significantly and positively correlated to population, income and size of spatial development. With a prediction power of R2 = 0.86, it was concluded that GWR model is suitable for water demand prediction. Forward selection method was applied on the three significant variables and results indicated that population variable had the strongest influence on water demand followed by size of spatial development then income with 49, 37 and 14 percent contribution respectively. The study also concludes that GWR is an important and reliable tool for analysing and predicting water demand as it is able to account for spatial variations. It further concludes that since factors that influence water demand are spatially varying, institutions responsible for planning and water management can use GWR to localise demand management interventions based on each area’s sensitivity to water demand predictors. Keywords: Geographically Weighted Regression, Water demand, Relationship, Prediction