As cities keep growing worldwide, so does the demand for key resources such as electricity, gas, and water that residents consume. Meeting the demand for these resources can be challenging and it requires an understanding of the consumption patterns. In this study, we apply extreme gradient boosting to predict and analyze electricity, gas, and water consumption in large-scale buildings in New York City and use SHapley Additive exPlanation to interpret the results. For this, the New York City's local law 84 extensive dataset was merged with the Primary Land Use Tax Lot Output dataset as well as with other socio-economic datasets. Specifically, we developed and validated three models: electricity, gas, and water consumption. Overall, we find that electricity, gas, and water consumptions are highly interrelated, but the interrelationships are complex and not universal. The main factor influencing these interrelationships seems to be the technology used for space and water heating (i.e., electricity vs. gas). Building type also has a large impact on interrelationships (i.e., residential vs. nonresidential), especially between electricity and water. Moreover, we also find a nonlinear relationship between gas consumption and building intensity. The main results are summarized into seven major findings. Overall, this study contributes to the urban metabolism literature that ultimately aims to gain a fundamental understanding of how energy and resources are consumed in cities.
K E Y W O R D Sbuilding energy use, energy consumption, industrial ecology, machine learning, urban sustainability, water consumption
INTRODUCTIONAs the world population keeps growing, so does the demand for key energy and resources. Moreover, most of this growth is happening in cities that already account for over 67% of energy use worldwide (Hobley, 2019). As a result, providing the necessary resources and energy for a growing population has become one of the most significant challenges for engineers, planners, and policymakers that also need to account for local specificities (Mohareb, Derrible, & Peiravian, 2016).Building energy consumption alone represents a significant portion of total energy consumption and carbon emissions worldwide (Dalia, 2014).