Estimating and Understanding Urban Congestion

The growing trends towards urbanization have increased the pressure on urban transportation networks. Consequently, we have seen an increase in large-scale congestion, including its negative externalities, such as increases in travel time, fuel inefficiency, and excessive emissions. Fortunately, recent studies in the transportation field have developed elegant methods to model and quantify traffic dynamics in cities. In particular, the framework of the macroscopic fundamental diagram (MFD) has recently received substantial attention. Much like its link counterpart (fundamental diagram), the MFD relates the network average flow and density of traffic. This dissertation aims to standardize urban traffic measurement and disentangle urban congestion mechanisms from a static and dynamic perspective. After discussing the introductory background in chapter 1, we investigate different methods to account for the multitude of empirical issues when estimating MFDs in chapter 2. We first focus on potential data biases (e.g., placement bias). As a result, we developed a correction method and a resampling framework that reduces the uncertainty in empirical MFDs. Based thereon, we define a practice-friendly partitioning algorithm that allows for finding homogeneous traffic regions, which is key to many applications of the MFD. In turn, this allows us to find a new versatile function for the empirically observed MFDs in chapter 3. This function lays the cornerstone for the standardization of the empirical MFD. This process allows for us in chapter 4 to cross-compare the important critical point of the MFD in over 40 cities worldwide. The critical point describes when network-level congestion is reached. We find that only four variables describing traffic conflicts (road network density, network redundancy, intersection spacing, and bus production) explain roughly 90% of the observed variance. In a more dynamic approach, chapter 5 presents empirical evidence for the long-term repeatability of the observed MFD by using a time-series method. Over a year, we find the observed MFDs can be grouped into a relatively low number of six (Zurich) or eight (Lucerne) representative repeatable MFD shapes. Moreover, we show that the evolution of the heterogeneity in flow is a good predictor of the expected MFD's shape. Using only five roads, we can accurately predict the expected MFD shape early in the day. Chapter 6 provides evidence from a multi-city simulation study that shows how urban traffic networks follow the laws of a percolating system. Non-smooth transitions characterize such a phenomenon. We find that (shortly) after reaching the critical point of the MFD, some urban traffic networks also exceed a hypercritical point, where the whole network is heavily congested. This point characterizes the moment when congestion becomes truly widespread (i.e., the size of the largest cluster is on the order of the network). We conclude that this is a point not to exceed because the congestion recovery becomes much harder afterward. These findings have implications from an academic and from a practical point of view, which are discussed in chapter 7. First, our empirical investigation of the observed MFD defines a standardized way to monitor, predict, and model urban neighborhoods, simplifying, and unifying current approaches. Second, our analysis allows for assessing the first-order impact that changes to the network topology have on traffic performance. Third, understanding urban traffic as a percolating system provides new modeling perspectives