Causal discovery and analysis of global city carbon emissions based on data-driven and hybrid intelligence

Abstract

The unclear causal links of carbon emissions among global cities challenge policy development. This study develops two causal discovery algorithms to aid in this understanding. The first, scalable causal discovery, excels in unraveling complex causal relationships within extensive non-Euclidean networks encompassing thousands of nodes. The second, knowledge-injection causal discovery, merges expert expertise with artificial intelligence's data mining capabilities, employing a human-computer interaction approach for precise causal analysis. The proposed algorithms outperform leading causal discovery methods in the Granger causality test and causal structural consistency. This study investigates the emission causal networks across global cities and key international organizations, including the Organization for Economic Cooperation and Development, the Commonwealth, G20, the Belt and Road Initiative, and the Asia-Pacific Economic Cooperation. The analysis encompasses networks, countries, cities, and emission sources, providing insights for developing collaborative urban emission reduction policies. It underscores the tightly interconnected nature of the worldwide emission network, where the effects are rapidly disseminated. Furthermore, sub-networks reveal consistency and variability in their causal patterns, with core cities exerting significant influence over various dynamics. It is essential to leverage the unique structural characteristics inherent in each sub-network to enhance the effectiveness of coordinated emission reduction initiatives.