High-Resolution Agent-Based Modeling of Campus Population Behaviors for Pandemic Response Planning

TL;DR

This paper reports a case study of an application of high-resolution agent-based modeling and simulation to pandemic response planning on a university campus and found that reducing the population density to 40-45% would be optimal and sufficient to suppress disease spreading on campus.

Abstract

This paper reports a case study of an application of high-resolution agent-based modeling and simulation to pandemic response planning on a university campus. In the summer of 2020, we were tasked with a COVID-19 pandemic response project to create a detailed behavioral simulation model of the entire campus population at Binghamton University. We conceptualized this problem as an agent migration process on a multilayer transportation network, in which each layer represented a different transportation mode. As no direct data were available about people's behaviors on campus, we collected as much indirect information as possible to inform the agents' behavioral rules. Each agent was assumed to move along the shortest path between two locations within each transportation layer and switch layers at a parking lot or a bus stop, along with several other behavioral assumptions. Using this model, we conducted simulations of the whole campus population behaviors on a typical weekday, involving more than 25,000 agents. We measured the frequency of close social contacts at each spatial location and identified several busy locations and corridors on campus that needed substantial behavioral intervention. Moreover, systematic simulations with varying population density revealed that the effect of population density reduction was nonlinear, and that reducing the population density to 40-45% would be optimal and sufficient to suppress disease spreading on campus. These results were reported to the university administration and utilized in the pandemic response planning, which led to successful outcomes.

Figures (6)

• Figure 1: The multilayer transportation network of the Binghamton University main campus reconstructed using the data manually extracted from Google Earth. The first layer of the network (thick lines) are roadways for vehicles, whereas the second layer (thin lines) are pedestrian pathways. Pink/orange/green/blue circles indicate locations of buildings/food places/residence halls/parking spaces, respectively.
• Figure 2: A snapshot of a sample simulation run. Tiny green/black dots represent student/employee agents, respectively, and small blue dots on roadways represent vehicles. Green disks in the central area represent buildings, with the disk size scaled according to the number of agents inside the building. Blue disks represent parking spaces, with the disk size scaled according to the number of cars parked there. The red-orange-ish heatmap displayed in the background shows the density of close social contacts occurring in each area (spatial diffusion and exponential decay operations were applied to enhance the visibility and interpretability).
• Figure 3: Comparison of the numbers of parking openings between the actual data and the simulation result using a calibrated model. (a) Heatmaps showing the number of parking openings along space (x-axis: parking lot ID) and time (y-axis: two-hour time points from 8:00am to 6:00pm). The top and bottom maps visualize the actual data and the simulated result, respectively, which show reasonable similarity. (b) Result of linear regression between the actual data and the simulated result (plotted in log-log scales, which makes the fitted linear line curvy). Each blue dot represents values for a specific space-time pair. The linear line fit shows high correlation ($R^2 = 0.768909$).
• Figure 4: Summary density map visualizing the cumulative frequencies of close social contacts for the whole campus population over the entire duration of simulation (a whole day). A "C"-shaped high-contact area is clearly visible, which connects the food court (top right of the "C") to the Lecture Hall (left of the "C") and then to the bus terminal (bottom right of the "C"). Another clearly visible dot inside the "C" corresponds to the Library Tower, which has a popular coffee kiosk on the ground floor.
• Figure 5: Summary density maps visualizing the cumulative frequencies of close social contacts with reduced population density. (a) 50% of normal population density. (b) 25% of normal population density.