Forest Fire Tutorial

From AgentCubes
Revision as of 23:44, 24 August 2011 by imported>Andri
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
Forest fire.png


The forest fire simulation built with this tutorial enables you to explore how forest fires unravel by letting you set fires to virtual forests with different tree densities. This simulation can them be used for data collection and analysis in math units.

Creating the Project

Launch AgentCubes and create a new project (choose the location and name of your project).

Creating Agents

The simulation has a Tree agent, a Ground agent, a Controller agent, and a Start Here agent.

Forest Fire Agents.png
The Controller agent is used (later on) to synchronize the fire and signal the ground to generate forests of different densities
The Ground agent has one depiction and generates forests of different densities
The Start Here agent is used (later on) as a visual clue during experimentation and data collection for students to start the fire at the same location every time they run an experiment
The Tree agent has three depictions:
  • a healthy tree
  • a burning tree for when the tree is on fire, and
  • a burnt tree for then the fire completely goes out.

Creating a Forest

If you haven't created a new world, use the + button at the top of the project window to create one. Please note that later on we will be generating the forest automatically, but for now let's do this manually: Select your Ground agent in the Agents list and using the Volume Insert tool, lay out a layer of Ground agents in the background of your world. Then select your healthy Tree agent shape and using the Draw tool place a few on the Ground. Place one Tree_Burning, which will start the fire, somewhere in the middle of the world to create a forest like the one shown below.

Initial forest.png

This is a good time to save the world!

Programming the Tree

We need to program the Tree agent to catch on fire if it's next to a burning tree and burn completely when the fire dies out. Select the Tree agent in the Agents list and its behavior will appear in the Behavior view. Create two rules for the tree:, one to start the fire if it's a healthy tree with a certain probability, and one to stop it when the fire dies out with a certain probability:

Tree behavior - no controller.png
If I am a healthy tree and I am next to a burning tree, with a certain percent chance, I will change into a burning tree myself. This probability be a number (e.g. 80%) or can also be defined by a simulation property called Fire_Spreading_Probability.
If I am a burning tree, with certain percent chance i will turn into a completely burnt tree, to simulate the fire dying out. This probability can be a constant (e.g. 50%) or can be a "Fire_Die_Out_Probability" simulation property for easier experimentation.

Now try out your simulation by running it. Depending on the density of the forest you created, the fire spreads widely or dies out quickly.

But notice something: if you reset your world to its original saved state and hit the Step button to run one cycle of the simulation you will get something like this:

Initial forest - step once.png
What is wrong with this picture? You will notice that trees that are far away from the source (the initial burning tree) catch on fire in the very first step, which is not very realistic (especially given that we are not simulating wind). This happens because AgentCubes has a random, but explicit order in with it executes the behaviors of the agents in the world. The healthy trees that are in the execution list after the burning tree, will catch on fire (based on the probability) in the very first step, even if they are far away from the initial burning tree. If it happens that all the trees were placed on the list after the burning tree, all of them would catch on fire in the first step.

How can we fix this? We need to coordinate the behavior of the tree so that within the same simulation cycle, all the trees have a chance to determine if they should be burning first, and then actually perform the burning actions. For that, we can use a Controller agent to synchronize the tree behavior (make it parallel).

Programming the Controller & Tree to synchronize/parallelize tree burning

Controller while running.png
To synchronize the actions of trees, the Controller broadcasts to all trees a PERCEIVE message to determine if they are supposed to be burning, followed by an ACT message to actually start burning or have the fire die out.
Tree behavior.png
The tree does not do anything in its While-Running method.
The PERCEIVE method of the Tree determines whether a healthy tree should start burning - it does so if the healthy tree is next to at least one burning tree, with a certain probability (defined by the simulation property Fire_Spreading_Probability in this case). Please note that the PERCEIVE method does not actually set the tree on fire by turning it into a burning tree, but sets a variable/flag to indicate that the tree should be burning. This way, all the trees get a chance to determine whether they should be burning in parallel, avoiding the issue encountered in the previous section.
The ACT method of the tree actually carries out the burning: if the tree is healthy and its "burning" flag is set, it turns into the burning tree. Otherwise, if it's an already burning tree, with a certain probability (defined by the simulation property Fire_Die_Out_Probability), it turns into a burnt tree and increments the @burnt_trees counter (simulation property).
Starting a fire: when a tree is clicked with the mouse tool while holding the Control key, it turns into a burning tree.
The tree deletes itself when it receives the "delete" message. This will be used to regenerate forests of a certain density as described below.

Now try running your simulation again. Stepping through your fire simulation should look like this:


Regenerating the forest with a certain density

Controller mouse click.png

To regenerate the forest, we add a method to the Controller's behavior that is triggered with a mouse click. When the controller is clicked with in the world, it broadcasts to all the current trees in the forest to delete themselves, resets the total number of trees (@trees simulation property) and the number of burnt trees (@burnt_trees) to 0 and sends a "create_forest" message to the ground to generate the new forest.
Ground behavior.png
When the ground receives the "create_forest" message, it creates a new healthy tree on the same square the ground resides and increments the total number of tress (@trees) by one. This happens with a certain probability, that determines the density of the forest (@desired_density). We will later be changing this simulation property to experiment with setting fires to forests of varying densities.

Simulation Properties

Running your simulation will create the simulation properties used without your explicitly creating them. But especially for the input properties (e.g. Desired_Density), it's good to create them yourself so you can initialize it to the values you want - otherwise, the simulation automatically creates them with 0 as the default value. Pull up the Simulation Properties editor, using Tools->Simulation Properties and add the following properties:

Forest fire simulation properties.png
  • Burnt_trees: the number of burnt trees (set to 0 initially)
  • DESIRED_DENSITY: density your generated forests should have (set to 50% initially; this will change for experimentation)
  • PERCENT_BURNT: percentage of trees that got burnt (set to 0 initially - the simulation will calculate this)
  • Trees: the total number of trees the forest has (set to some number other than 0 initially - the simulation will set it to the right number when you regenerate the forest and you should save your properties with that number)
  • Fire_Spreading_Probability: the probability with which fire spreads from a burning tree to a healthy tree
  • Fire_Die_Out_Probability: the probability that the fire on a burning tree will die out, making the tree a completely burnt tree

Once you have your properties set, make sure you hit the Save button to save them.

Please note that we capitalized the DESIRED_DENSITY and the PERCENT_BURNT to make it easier for users to tell which simulations properties they need to manipulate or look at for their experimentation.

Updating the world to include Controller and Start_Here agents

Add a controller agent and a Start_Here agent to your world, save your world, and you are ready to start data collection for the Sampling and Bivariate Data Analysis unit.