Leveraging Disparate Parcel-Level Data to Improve Classification and Analysis of Urban Nonresidential Water Demand

Abstract

This study details a novel procedure for analyzing water demands in the nonresidential sector (i.e., commercial, industrial, and institutional users). Nonresidential customers are classified into “subsectors” based on economic, land-use, and property appraisal data sets and analyzed using a linear mixed-effects regression modeling framework, which controls for random fluctuations around mean monthly parcel-level water demand, for a 4-year study period in Austin, Texas. Classification of nonresidential customers can improve the explanatory power of statistical models over models without any classification (R2=0.635 and 0.431, respectively). Additional improvement is seen by explicitly using the economic, land-use, and property data on which subsectors are based (R2=0.773), at the cost of computational expense and added model complexity. Results indicate that the subsector classification provides the best explanation of variation in monthly water usage at the parcel level, followed by conditioned floor area and number of employees. These results can improve traditional water demand forecasting techniques for the nonresidential sector and reveal subsector-specific trends that might otherwise be obscured without classification of customers.