Machine learning techniques applied to urban metabolism

Cities alone account for three quarters of global CO2-eq energy-related emissions and two thirds of global energy use. In 2050, the number of inhabitants will increase to 67% in urban areas and reach 6.34 billion humans, which... [ view full abstract ]

Cities alone account for three quarters of global CO₂-eq energy-related emissions and two thirds of global energy use. In 2050, the number of inhabitants will increase to 67% in urban areas and reach 6.34 billion humans, which will further increase the environmental footprint of cities. Through urban areas, the humanity puts considerable pressure on the biosphere by extracting large amounts of resources and generating considerable quantities of waste. Urban metabolism, by accounting all flows entering and exiting urban areas, measures their environmental effect in a comprehensive way. Currently urban metabolism studies are however often restricted to a descriptive purpose and provide limited means to understand the relationship existing between an urban area and its resulting metabolic flows. Understanding these relations is necessary to manage, forecast and possibly mitigate future urban environmental pressures. Indeed, without taking into account the complexity of these relations, the identification of mitigation levers and the resulting policy-making are likely to provide fragmentary or even inadequate results. It is thus important to extend urban metabolism assessments with sufficient information on the urban system causing these flows. To do so, the increasing amount of data available to describe cities provides possibly gives an opportunity to explore the intricate aspects between a city and its metabolic flows. In this research we show how machine learning can be applied to large datasets, describing on the one hand metabolic flows and on the other hand urban systems. This could in turn help to overcome the black box effect of current urban metabolism studies by building more explicit and quantified relations between the urban and metabolic characteristics. Machine learning techniques have shown great success in retrieving (sometimes hidden) main variables of complex evolving systems that are useful for analytic and predictive purposes. The “urban-data” exploited in this paper are a set of indicators describing its demographic, socioeconomic and built environment state . The “metabolic-data” account a selection of metabolic flows of urban systems. We exemplify this approach through the case study of urban systems in Belgium. To show the potential capabilities of these techniques to overcome urban metabolism’ black box model (1) we briefly review the usefulness of machines learning techniques and summarize urban metabolism research question that could be addressed by them; (2) we present  the case study and its data availability ; (3) we show how relations between urban systems and metabolic assessments may be extracted through machine learning processes. With these results we hope to significantly contribute to central research questions regarding the operability of the urban metabolism approach such as the identification of drivers.