JHU Center for Advanced Modeling from NIH Directors Pioneer Award
The goal of this project is to create realistic disaster scenarios in a virtual city that can be used to model the behavior of humans interacting with one another and with their compromised built environment during and after major crises. We capture the effect of multiple hazards on the same virtual city, and model how […]
The goal of this project is to create realistic disaster scenarios in a virtual city that can be used to model the behavior of humans interacting with one another and with their compromised built environment during and after major crises. We capture the effect of multiple hazards on the same virtual city, and model how these various hazards and resulting levels of damage hinder emergency response. We also realistically model the behavior of humans in these environments, including the use of warning time and willingness to make pre-event preparations. Human behavior (such as contagious fear) may also generate noncompliance with disaster directives, producing congestion, impeding the flow disaster medical services, and undermining evacuation efforts. We build on the work we have been doing on the single structure scale and expand beyond the limited work we’ve done at the regional scale. Overall, the aim is to subject cognitively compelling software individuals to a range of disasters.
Five related lines of research will be pursued.
I. Earthquake Scenario:
To effectively model the flow of humans in damaged buildings and throughout an impacted region during and immediately following a seismic event, it is necessary to model the damage and potential loss of function of systems within the structure. We employ building-scale fragility functions to estimate probability of collapse and structural and nonstructural damage. The damage results are used as input to a subsequent loss analysis, which quantifies the building performance in terms of societal metrics, such as repair cost, loss of function, forcible building closures, productive work days foregone, and other economic and safety issues. The results from these damage and loss models will be used to study the impact of varying seismic hazard levels on the physical infrastructure to the virtual city, and will allow us to construct rules of behavior and delineate optimal evacuation routes-from inside the structure to outside the immediately affected region-for the agents.
II. Hurricane Scenario:
To effectively model the damage to a region due to a hurricane (include both wind and rain loading conditions), it is important to manipulate key features of the storm event in a probabilistic way. We are developing a wind model that can easily be used by our collaborators to create hurricane scenarios making landfall along the US coastline. The features of this hurricane model that can be directly manipulated include the landfall location, wind angle, radius of the storm, translational speed, internal pressure, asymmetry of the event, roughness of the boundary layer, rainfall rate, and rain duration. Hurricane Sandy has highlighted the importance of hurricane surge, and so this will be explored as a potential extension the hurricane module. As above, agent behavior will shape preparation and evacuation efforts. We will develop a measure of loss incurred for every day of evacuation delay. We expect this curve to be concave up, with very high costs associated with extreme delay—which can also produce congestion and consequent vulnerability.
III. Plume Scenario:
A Lagrangian buoyant plume model is being developed using Smoothed Particle Hydrodynamics. The plume will be located within an idealized urban environment such that the trajectory of the plume is advected by the wind, affected by turbulence, and interacts with the buildings. Various types of plumes and wind fields will be modeled. All the computation will be carried out on graphics cards (GPU computing). Buildings will have various permeabilities to the plume. In some cases, shelter-in-place is superior to evacuation. We will explore various mixes of interventions, under various assumptions about behavior—willingness to shelter in place, or conform to routing and carpooling directives.
IV. Traffic Modeling:
This module is being built through three parallel tasks. The first task is focused on source and destinations. All agents seeking to use a vehicle are identified with a source and destination, which depend on time of day and type of scenario. For instance, for the plume disaster, the source may be a parking lot or bus stop. The agent would then need to supply the transportation module with a destination (e.g., daycare to pick up children or simply a direction orthogonal or opposite to the apparent track of the plume). The second task 2 focuses on desired travel paths. This is simply the set of directions from the source to the destination, not unlike that supplied by a GPS. The third task will involve modeling traffic signaling and flow. Starting with a map of the city region of interest, signaling following standard practice (in duration, direction and inter-signal timing) will be programmed into the module. Once these tasks are completed, then a traffic flow analysis can be performed with agents following their trajectories while encountering traffic signals. Allowance will be made for agents who choose—due to fear, mis-information, or peer effects–to ignore the signals and deviate from their initial travel path. The city region boundaries will have reduced flow rates that reflect congestion that would arise from a mass exodus. This reduced rate would affect traffic within the city region where the detailed agent behavior can be observed.
V. Human Behavior Modeling:
Coarse models are being developed that produce a flow of agents exiting the building during or after the hazard event. The agents have goal-oriented behavior and make exiting the structures a priority. Outside of buildings, they also exhibit other behavioral characteristics common in disaster situations, such as herding, grouping, and rescuing behavior. The repertoire of behaviors will include features of the Epstein Agent. The building occupant agents will need to be positioned in two spaces: a vector space, which positions the agents within the building using a fixed (non-inertial) frame of reference, and a graph space, which positions the agents within the network of pathways within the building that lead to the exit passageways. The building dynamic behavior will be translated into neighborhoods for each agent. For the vector space neighborhoods, Euclidean objects with a structural integrity scalar will be used to describe the condition and any movement of the floors, walls, and other nonstructural components. For the graph space neighborhoods, the edges and vertices will be characterized by their possibly reduced ability to handle pedestrian traffic. Behavior once outside of buildings—mass panic and congestion—will be modeled at this, larger, spatial scale.
These activities will produce a novel integrated disaster modeling framework in which the advanced agents developed by Epstein can be subjected to a range of stresses and calibrated to replicate observed crisis behaviors.