We propose a new approach for coordinating traffic flows in large cities that helps in reducing the travel time and carbon emissions from vehicles. We use the UPPAAL STRATEGO tool chain that leverages statistical model checking and machine learning for synthesizing optimal traffic coordination strategies. Our approach employs a hierarchical view of the city with two levels - individual traffic intersections and area controllers. While the choice of a phase at an intersection is decided locally, the phase threshold is decided at the level of an area consisting of several intersections. The algorithm and models that we report in this paper are a nontrivial generalization of previous approaches that used UPPAAL STRATEGO. This generalization allows scaling to large cities with several traffic intersections, with improved results.We compare our approach against other techniques including fixed-time and fully-actuated controllers. Experiments show that the it performs better in terms of waiting time and carbon emissions, especially in scenarios of changing traffic loads. Our approach also reduces overall and individual delays at intersections.