Graph-based analysis has proven to be a good approach to study topological vulnerabilities of road networks through specific metrics, such as betweenness centrality (BC). Even though BC of unweighted, undirected graphs has been widely adopted to identify critical road segments and intersections, given the very high number of potentially highly-traversed paths flowing through them, congestion and vulnerability are strongly influenced also by static and dynamic context factors, such as road capacity, speed limits, travellers' behaviors, accidents, social gatherings and maintenance operations. In this paper, we focus on the analysis of BC on dynamically weighted graphs, used as a model of a road network and associated dynamic information (e.g. travel time). The aim is to discover correlations between the centrality metric and vehicle flows, both in space and in time. The analysis proves the existence of relevant spatio-temporal correlations that provide useful information about the characteristics of road networks and the behavior of drivers. In particular, we identify the existence of anti-correlations that point out forecasting properties of BC when computed on dynamic graphs.These properties justify the usage of the metric for the implementation of next-generation proactive, data-driven urban monitoring systems. These systems are expected to empower urban planners and traffic operators with novel intelligent solutions to reduce traffic congestion and vulnerability risks, therefore contributing to implement the vision of a more resilient and sustainable city.