Other than the heartbreaking sights of folks in makeshift boats carrying whatever belongings they can manage, one of the recurring images of natural disasters like hurricanes are pictures of mile upon mile of bumper-to-bumper traffic along major freeways leading out of the affected area. Under these circumstances, it is no wonder that during Hurricane Harvey, officials like the mayor of Houston opted to not order evacuation of neighborhoods known to be subject to flooding (in a city known for its disdain of zoning). Nonetheless, evacuation of these areas would have prevented considerable hardship on the affected families, and reduced the need for the kind of heroic rescues in the aftermath of the hurricane.
Are all evacuations doomed to lead to bumper-to-bumper, patience-eroding, frustrating hypercongestion? The answer is no, and yes. No, because traffic engineers and planners have known for years how to design evacuation that optimize the ability of the system to deliver people to safety. Yes, because just as we have not managed to eliminate daily congestion on the nation’s freeway arteries in major metropolitan areas, it is unlikely that the solutions for effective evacuations would be implemented—that agencies would enforce what they entail, and that the general public would adhere to those instructions.
First, some physics. Highways, just like any channel, can move a certain number of vehicles per unit time—what traffic engineers might refer to as “capacity”, or queueing specialists call service rate. So if you have a certain number of people to evacuate from a given area, the best one could do is to move people at the maximum rate allowed by the sum of outgoing lanes (plus any lanes that may be reversed during the evacuation). Thus if you have 6 total lanes of freeway, and each is operating at its nominal capacity of 2000 vehicles per hour per lane, you can move at most 12,000 vehicles per hour. If you need to evacuate 120,000 people, that would take 10 hours.
Are all evacuations doomed to lead to bumper-to-bumper, patience-eroding, frustrating hypercongestion? The catch, though, is that this nominal capacity is just that—nominal. Actual values tend to be much lower over a sustained period of time. The most vexing phenomenon about congestion is that beyond a certain level, the service rate drops considerably the more people are trying to use the route. This “more is less” phenomenon is rooted in the physics of traffic, and the behavior of human drivers. The drop can be substantial, with one third being a typical estimate. So from 2000 vehicles per hour the service rate becomes closer to 1340 vehicles per hour, and the time required to evacuate those same 120,000 jumps by 50% to 15 hours.
This drop occurs when the system is overloaded—for example, when no-notice evacuations dump the entire population of an area onto a network. These are the worst possible scenarios for evacuation, leading to near-instantaneous clogging, and likely gridlock. Optimal strategies are all about when and where to load—i.e. staging of the evacuation process. Optimal staging makes sure that the system does not overload prematurely and keeps service levels high.
Furthermore, recent understandings in traffic physics have shown that the entire network of a given area can be characterized in a similar manner as its component highways—namely in terms of a maximum service rate that it can deliver under a particular loading pattern. The overall network experiences flow breakdown, with significant drops in the service rate, once the load imposed exceeds a certain threshold.
Another recent finding in network traffic theory has shown the existence of a phenomenon called hysteresis. This describes how congestion forms and then dissipates in a network, indicating considerable inefficiency in how the network serves traffic, even after the peak load subsides. Hence it is best to load the network at given rate during a certain period, and then stop for a few hours to allow it to clear, before loading it again.
Simulation studies conducted of the Long Island area under a hurricane of Sandy proportions showed that the optimal evacuation strategy consisted in loading the road network for about 12 to 16 hours, then taking a break for about 4 hours before starting all over again. Even under such optimal loading, the best one could do was to evacuate all of Long Island in 3 days. Without the optimal loading strategies, as under a no-notice evacuation, the resulting queues would have been impossible to handle, leading to severe gridlock and unfathomable delays.
So why do these solutions not get implemented? Several reasons stand in the way. For these approaches to work, it is essential to have advance preparation. Most areas focus on identifying evacuation routes and shelters, and not on staging and notification. Officials tend to delay the orders to evacuate, in the hope that the area might be spared by a capricious hurricane following an uncertain path. Most important, people tend to wait until the last minute possible to leave. Unfortunately, their notion of “last minute” is typically based on travel times during normal conditions, not under severe evacuation. Also, when faced with the reality of needing to evacuate, people start figuring out who they need to pick up, where to go, and so on, creating secondary traffic patterns that interfere with the general direction of the evacuation. Likewise, once the orders are given, many people want to leave when they want, not necessarily when it is optimal for the system as a whole to ensure the shortest possible clearance time and the least delay to all users.
For residents of hurricane-prone and flooding-prone areas, evacuation preparedness is a vital necessity, and should not be an after-thought. The underlying science and engineering tools are much more accurate at predicting what would happen during an evacuation than the weather forecasters are at predicting the path of the hurricane. It is time to put this knowledge to stage efficient evacuations that can get people to safety as soon as possible.