Watch a Real-Time Simulation of How Well Social Distancing at the Beach Actually Works to Contain COVID-19

When Virginia Governor Ralph Northam announced the tentative reopening of Virginia Beach for Memorial Day, he warned antsy residents that, if they didn’t abide by new restrictions such as social distancing and limits on group activities to mitigate the risk of new COVID-19 infections, he wouldn’t hesitate to re-close the area. He’s not alone. Coastline municipalities around the country are now instituting or contemplating similar liberties. To test how feasible it is for public spaces to reopen to those restless from weeks of stay-at-home orders, we’ve built a simulation.

It uses the real-world coordinates of a two-block stretch on the southern tip of Virginia Beach covering 4.4 acres of sandy respite along a 600-foot coastline, and runs a real-time computational experiment in which each digital person acts according a randomly assigned behavioral model. Each person is visualized as a circle with a six-foot diameter, meaning any time two circles overlap (and turn red), social distancing has been violated. Depending on where you place the slider at the top, a certain percentage of beachgoers in the simulation will rigorously try to observe social distancing while others will bend the rules by a few feet.

In the default setting, 75% of the 200 simulated beachgoers aspire to maintain a minimum of a six-foot distance, while 25% are willing to fudge that distance down to between 3 and 5 feet. In this case, about 20% of our tiny friends head straight for the water, while 30% plop down to sunbathe and the remaining 50% wander up and down the beach. After one hour in simulated time, at least 1,000 violations of social distancing typically occur as people move about, according to many tests of the simulation, which involves a small amount of variation due to the random nature of where people chose to move.

This sort of exercise, in which each person in a simulated group has a personal set of rules and goals, is more than just a thought experiment. The success of any policy whose outcome is dependent on voluntary compliance, like the rules for Virginia Beach, is difficult to predict ahead of time because small, simple differences in how individuals behave can create enormously complex patterns when they interact with one another—what’s known as “emergent” patterns. While TIME’s implementation uses only a handful of variables, many researchers use this style of computer science as a powerful means to predict emergent patterns of group behavior in anything from crowds to traffic to ecosystems.

The emergence we see here suggests that at least a modest amount of interaction inside 6 feet is inevitable as public spaces reopen. Even when limiting the crowd to only 200 people and assuming 100% those beachgoers try to cooperate, the model suggests it’s basically assured that people get too close to each other. Repeated trials indicate that the random movement of walkers, often impeded by stationary sunbathers, combined with the inevitable collisions between people at the bottlenecks at the entry to the beach, results in an average of about 200 accidental violations of social distancing in one hour.

And that’s a highly ideal scenario, in which the crowd size remains unrealistically small and improbably obedient. As soon as you start adding more people or introducing even a small percentage of negligent actors, the number of collisions skyrockets. For 1,000 beachgoers at 75% compliance, there are typically more than 20,000 such run-ins over the 90 minutes (in simulation time) it takes for all of them to reach the beach.

As desperate as the U.S. is for some form of normalcy over the warm summer months, every model we ran suggests that, without general acceptance of a new reality in which public spaces will need to be less populated, the health risks could be significant.

As I wrote recently, even as new COVID-19 infections fall, Americans must reconsider the outdoors as an exhaustible resource that must be rationed. As this experiment tentatively indicates, even a fairly large amount of beachfront—4.4 acres is the equivalent of 3.3 football fields, not including the water—cannot tolerate more than a few hundred people without a dangerous degree of interactions that are too close for comfort, no matter how assiduous those people are trying to be. While Virginia Beach acknowledged this to some extent by limiting parking capacity to 50% over Memorial Day weekend, it could take far more drastic and undesirable limitations to make public spaces a place people can enjoy without an intolerable level of risk.

Methodology

This sort of experiment is known as “agent-based modeling,” in which each “participant” acts as an autonomous program that makes decisions based on its surrounding environment.

While there are several JavaScript libraries that have implemented some degree of agent-based-modeling, this simulation was written from scratch to accommodate the specific use-case. In each step (which occur several times a second), every agent seeks to move approximately 20 feet toward his or her desired destination—either the water or an available spot on the sand. Unless the agent plans to sunbathe, once she reaches her destination she chooses a new one, whether it’s a spot on the beach on a nearby location in the water for swimmers. Except, As each time an agent “moves,” she “calculates” whether the move would violate another agent’s space below her minimum acceptable degree of social distancing—from 3 to 6 feet, depending how closely she is following recommended guidelines—and, if so, attempts a different route until one is appropriately spaced, up to five times. If, after five tries, the only option is to move inside another agent’s 6-foot radius, a collision occurs. The interactive tallies that as one violation.

The simulation runs at about 40 times faster than reality and ends after either an hour or when every prospective person has reach the beach or water.

This style of computer science was popularized by the free software NetLogo, a sophisticated descendant of the original LOGO program, popular in grade schools, that involved giving instructions to a “turtle” that would create patterns on the screen. While this simulation was not independently coded in that software, the author consulted a textbook on its structure to aid in the software design.