Vertical Evacuation Simulation

Disclaimer: This is a “toy” simulation, based on untested assumptions. It should not in any way be used in making real-life planning decisions.

Created with NetLogo; View/download model file: vevac.nlogo

What Is It?

This model simulates "vertical evacuation," a technique for bringing people to safety by having them "go up" in buildings, for instance, in flooding or tsunami situations. The model implements two behaviors: seeking to go up (modeling people who receive and understand the warning, know what to do and do it), and seeking to get outside (modeling people who want to get out for whatever reason).

The project demonstrates how individual behaviors may slow down overall movement, create traffic jams, and block access to safe zones.

How to Use It

Click on the SETUP button to set up the space and generate a set of people in the world.

The gray horizontal strip represents the street level, and the vertical white strips represent buildings (stairwells in buildings to be exact).

Solid arrows represent "uppers" (people seeking higher altitudes), hollow arrows represent "outers" (people seeking to reach ground level). The direction of the arrow shows the individual's direction of movement. The color of the arrow indicates the safety status of the individual (green being safe, yellow at rist and red unsafe), with respect to altitude.

Click on the GO button to start the simulation.

The plot on the left shows the number of people in each safety zone, at any given time.


All individuals will move forward until they are blocked by someone else, or they need to turn (e.g., to go up or get onto the street). All individuals slow down and speed up as the path ahead of them is blocked or clears. Each individual has a different maximum speed at which they can move.

If an individual is blocked, they will slow down or, if possible, "step around" the block. Individuals have a preference for stepping to the left if they are passing someone, but will choose to step to the right in some cases. They have a similar preference to step to the right if they are yielding to someone heading in the opposite direction. Note that, while they can step around one other individual, they are not intelligent enough to "go around" a group of people.

If they are blocked for too long, they will turn around and try in the opposite direction. Note that the street "wraps" around the world, so that they will move off one side of the view to reappear on the other.

An "upper" on the street will move forward until they are opposite a building, then turn and try to enter the building. They are not smart enough to head for the nearest building. Once they are in the building, they will not give up on trying to go up.

An "outer" in a building will not give up on moving downward until they are on the street. Once on the street, they will keep moving in a randomly chosen direction until blocked.

The model will stop either when no one is moving, or when the only individuals moving are outers on the street.


In addition to modeling simple movement and isolated decision-making, the model also demonstrates socially transimitted behavior. "Directors" are people who try to convince others to emulate their behavior, i.e., if they are "uppers," they will tell people to go up, and if they are "outers," they will tell people to get out. If a non-director is convinced by a director, they will change their behavior (if necessary) to match that of the director.

Directors can be heard only by their immediate neighors. Each director has a "credibility" factor, and each non-director has a "belief" factor, so that some directors are more convincing than others and some non-directors are more difficult to convince than others.

In the present simulation, "upper" directors are generally more convincing than "outer" directors.

Directors are identified by a different arrow shape. Individuals who have been converted are indicated by a heavier outline.

The right-hand plot shows the percentage of the population that exhibits the "upper" behavior.

Model Parameters

SAFE-HEIGHT controls the altitude that is considered safe (where individuals turn green).

STAIR-WIDTH controls the width of the buildings.

STREET-WIDTH controls the width of the street.

NUMBER-OF-BLOCKS controls the number of randomly placed obstacles in the street. These are represented as black squares.

NUMBER-OF-PEOPLE controls the total number of individuals in the world.

PERCENT-UPPERS controls the percentage of "uppers." The rest will seek to get outside.

DIRECTORS controls whether the director behavior is included in the simulation. Note that the number of directors is fixed at approximately 10% of the total population, selected at random from the population.

Things to Notice

Note how clots will form at building entrances and the middle of the building is often unoccupied.

Note how a phalange of uppers or outers can block anyone from moving in a building. The narrower the building is, the more likely this is to happen.

Note how directors convert others, and how the rate of conversion increases if a director is stuck in a traffic jam..

Note how the higher credibility of "upper" directors tends to make the number of "uppers" increase, but then, as the "upper" directors reach their own safe zones, the remaining "outer" directors may make it decrease.

Things to Try

Set the SAFE-HEIGHT to 0 and mentally "reverse" the meaning of the red and green colors. You are now looking at a building evacuation model, e.g., for a fire or an earthquake situation. Try setting the PERCENT-UPPERS to 0.

Set STAIR-WIDTH to 1. You now have the beginnings of a queuing model, e.g., an evacuation situation where individuals must pass through a checkpoint.

Change the behavior of individuals so that they "see" the nearest building (maybe allow the user to specify how close they have to be to "see" something).

Change some of the directors so that they do not evacuate themselves, but stay in place (possibly at key locations such as building entrances) to direct traffic. How would that change evacuation results?

Sample Experimental Results

The BehavoirSpace included in the model defines five experiments, that vary only in the percent-uppers and whether directors are on or off. Model parameters for the five conditions are shown in Table1.





































Table 1 Model settings for each condition.

Tables 2 and 3 show a set of results from running the experiments (running each condition five times), showing the average time until evacuation is “complete,” the number of people who reach safe zones, and the standard deviations for each of these measures. Note that the evacuation is considered “complete” when the number of people reaching safe zones has slowed to a trickle for a considerable length of time. That doesn’t mean that everyone who could be evacuated has been or that a large group might not evacuate later, just that the wait is too long to consider them evacuated safely.

The five experimental results are contrasted with an “optimal” condition that is based on pure calculation. It assumes that all individuals start at the midpoint between the two buildings, head directly toward the nearest building and go up. The calculation assumes that all individuals move at a constant speed (half as fast as the median of that used in the simulation), that individuals are launched in groups of four (two toward each building) at each time step. In other words, this is the fastest that people could possibly be assumed to evacuate.


Avg Time

Avg Safe

StD Time

StD Safe































Table 2 Results over five runs of each condition.

Table 3 Results over five runs of each condition.

If these results were confirmed by statistically valid experimentation (five runs is insufficient to achieve meaningful confidence levels, and further experimentation with varying population sizes is necessary), and the model were based on realistic assumptions, the following conclusions could be made:  

Note, again, that this model is not based on tested assumptions. Real-life planning and decision-making should not be based on these results or their interpretation.


Version 1.0


This model was developed at the Pacific Disaster Center, Kihei Maui HI, by NRC Research Associate Susanne Jul, PhD, to illustrate the potential for incorporating agent-based modeling in disaster management.

Copyright 2007 SJul. Rights to non-commercial use and development granted, with appropriate attribution.