Exploiting open data for CO2 estimation via artificial intelligence and eXplainable AI

Climate change is a challenge of global relevance that requires in-depth understanding and immediate response. Urban CO₂ emissions are one of the main causes of climate change, and their estimation is crucial for planning actions aimed at creating more sustainable cities. Currently, models for estimating CO₂ emissions mainly focus on traffic patterns, energy consumption, or a limited set of socio-economic factors, often overlooking the increasing role of urban services in today’s cities. Moreover, these data are rarely accessible, which limits their usefulness for policy design. In the present paper, a general CO₂ emission estimation approach, based on a large range of often accessible open data as predictors, is presented. Such open data are related to human activity regarding services and (socio-economic) conditions arising in a given (urban) area. The proposed model focuses on fine-scale prediction to better understand the dynamics of emissions via machine learning approaches, while taking into account an innovative study based on open data sources including city services. The resulting best models have been based on XgBoost and GCN (graph convolutional network). The outcomes provided better precision (MAPE in the order of 8%) with respect to the state-of-the-art solutions. The goal has been to understand how specific predictors can contribute to or mitigate CO₂ emissions in the observed area. To this end, the impact of several features has been analyzed in order to identify the related key factors influencing emissions. A formal study has been conducted to perform feature relevance analysis by using eXplainable AI (XAI) approach. The proposed model is useful to define targeted policies reducing the pollutant impact of cities, promote a more ecologically sustainable urban lifestyle and improve sustainable urban planning. Both solutions and models have been assessed and improved, so as to be more flexible by using some transfer learning techniques. This research and its related results have been produced and validated by exploiting the Snap4City framework for smart city, mobility and transport and data analytics on CN MOST, national center on sustainable mobility. In particular, such results deal with the metropolitan cities of Florence and Bologna.

© 2017 Elsevier Inc. All rights reserved.