City-scale Modeling


We interact with the transportation system in our daily life: regular commute trips, weekend excursions; in rare but critical situations such as disaster evacuations and relocations. In modern life, we are blessed with easy means of traffic, however, at the same time, plagued by issues such as congestion, emission/pollution and capacity bottleneck in evacuation situations.

Many of the issues related to the traffic system are largely unresolved. We do not understand, for example, how to curtail vehicular emissions given the current increasing level of travel demand, or how to carry out more efficient evacuation from wildfires or hurricanes given limited road capacity. To answer these questions, our team develops tools to capture the traffic patterns at large scale (usually at the city or regional scale) and applies them to test different intervention scenarios.

We have two endeavors in city- or regional-scale traffic simulations. In the first approach, static/quasi-dynamic traffic simulation tools are developed to capture the change in traffic states at sub-hour or hourly resolutions. Though less suited for capturing the dynamic traffic conditions in the real world (e.g., in traffic accidents, tracing vehicle positions), the static/quasi-dynamic simulation is capable of running fast to simulate long-term scenarios. The applications are mainly targeted at road degradation or city-scale vehicle emission analysis. Additionally, developments on dynamic mesoscopic traffic simulation tools are also being carried out with the aim of representing the second-level traffic dynamics in the real world. The mesoscopic nature of the model has the unique advantage of reflecting the individual decision-making process while incorporating macroscopic traffic observations. This allows us to use a data-driven approach to calibrate the model while capturing the dynamically changing traffic conditions such as in evacuation cases.

This image has an empty alt attribute; its file name is traffic_1-1.png
Cumulative traffic volume on individual roads, obtained from quasi-dynamic traffic simulation with 15-minute update interval. Serving these traffic are the roads, elevated highways, bridges, tunnels, etc.

Fire Evacuation

We have applied the dynamic traffic simulations to wildfire evacuation studies. Wildfires have become an increasingly serious problem in California and also many other places around the world. Due to the ferocity of the fire, evacuation often becomes the only viable strategy to ensure safety. Shortly after the Camp Fire in 2018, our team paid several site visits to the town of Paradise to study the preparation and communication of wildfire risks in this severely damaged town. The report based on our site visits can be found [here]. Now we are extending the study to other small communities, e.g., Bolinas, in California. Our research framework consists of three components. The base layer is a dynamic wildfire propagation simulation (FlamMap, wFDS), which informs later components by creating various wildfire propagation scenarios. Given the fire status, a dynamic traffic model is then adopted to simulate the evacuation process, which identifies slow-downs and bottlenecks as well as potential strategies to improve the evacuation efficiency (e.g., reducing evacuation demand through car sharing, counterflow, …). The last component in the framework models the communication process, either between informal groups (e.g., residents taking time to help vulnerable neighbors), between agencies and residents (e.g., sending emergency order) or between agencies (e.g., radio communication between the fire department and law enforcement). As we found from our Paradise study, the loss of communication or the lack of reliable protocol can greatly affect the evacuation process. By including the communication model, we aim to bring into consideration such setbacks (i.e., due to communication issues).

This image has an empty alt attribute; its file name is 2020-update_-evacuation-1024x1024.png

The wildfire evacuation simulation framework includes three components:
Level 1: Fire propagation simulation using statistical or physical-based methods. (1a). Landscape: Ager, Alan & Vaillant, Nicole & Finney, M.A.. (2011). (1b). Weather (1c). Wind (1d). FlamMap fire propagation
Level 2: Dynamic traffic simulation. (2a). Road network from OpenStreetMap (2b). Travel demand based on parcel data
Level 3: Agent-based communication process model built with local knowledge and information extracted from official plans. (3a). Evacuation game designed by Prof. Tom Maiorana to draw information on evacuation behaviors and local knowledge from residents (3b). Marin County Mutual Threat Zone (MTZ) Fire Response Plan (3c). Agent-based simulation model of the communication process between individuals in various groups and organizations


Water pipelines

The water pipeline is one of the “lifeline” infrastructures for cities. As water is indispensable for both citizens and industries in cities, understanding the hydraulic properties of fluid inside pipes is essential for us to understand the resilience of the city under natural or man-made hazards. 

Our research focuses on developing a high-performance city-scale water-pipeline hydraulic simulator with the following functionalities:

  • Quantify fluids characteristics (hydraulic head, fluid flow rate, etc. ) for given water pipe networks.
  • Quantify leaking status for broken pipes.
  • Ability to run large scale (million nodes-links) simulations
  • Compatibility with other simulation tools

The hydraulic simulator can be used as a subsystem for earthquake resilience simulation of a multi-layered infrastructure system. See the figure below for details. 

This image has an empty alt attribute; its file name is pipe_system.png
Framework of incorporating multi-layered infrastructure system (water pipeline and road traffic) in simulating earthquake resilience of the San Francisco Bay Area cities. As earthquakes damage pipes, water flooding and pipe repair operations may cause potential traffic delays.


Tunneling is a complex construction process that has a substantial level of risk due to the inherent uncertainties in it, e.g. unforeseen variation of the geological conditions along the tunnel alignment and the unpredicted tunneling-induced ground movements. Due to the uncertainties, tunnel design and construction have always been a problematic task. 

Despite the inherent uncertainties, during the tunneling process, a tunnel boring machine (TBM) has the capability to record thousands of sensor measurements in every short time period. These measurements produce a great amount of data that cannot be handled by the TBM operators intuitively, during the process. Furthermore, these data have inherent complexity: they are the products of interaction among the earth (geological conditions), machine (TBM), and human (operators). Consequently, it is not practically possible to interpret and take advantage of the data by only using traditional data analysis techniques. 

The advancement of artificial intelligence (AI) and machine learning (ML) techniques in the past few years opens extensive opportunities for managing and interpreting “big” data. This advancement enables the detection of hidden patterns within data and interpretation of complex data with high dimensionality. The advancement also promotes the expansion of data-driven prediction and automation processes. Considering these facts, it can be conjectured that the implementation of this technology on the massive tunneling data might be promising to advance the tunnel construction practices, as well as enhance construction safety and productivity.

This image has an empty alt attribute; its file name is ml_tunnel.png
Unsupervised learning to reduce the dimensionality of TBM operation data and clustering the geological conditions along the tunnel alignment.
Source of the tunnel image: Shimizu Corporation (