Wildfire simulation using paralleled cellular automaton

Simulation of wildfire using cellular automaton and used mpi4py to parallel the program.
Final Project for High Performance Computing and Parallel Computing
May.2018

If you like my work, feel free to click star or fork on the top-right corner for quick reference in the future.

Environment Specification

In order to achieve best compatibility, this program only needs very basic prerequisite.

1. Openmpi/3.0.0
2. Python/2.7.11 (and mpi4py package)

However, to generate a better visualization, this program needs additional packages.

1. numpy
2. matplotlib
   If the program failed to import these two packages, then it will automatically run in pure text mode.

Usage

Run the code:

Clone the code from github using:git clone https://github.com/XC-Li/Parallel_CellularAutomaton_Wildfire.git
After loading the module(openmpi and python), you can use mpirun -n [n] python Parallel_Wild_Fire.py to run this code. You can specify the number of workers in n if you have multiple CPU.

Easiest place to change

This program is carefully designed, so the places you can change is very few. The easiest part to change is surrounded by -----Quick Change-----. You can simply change:

Or you can do some custom changes on the environment initial functions:

Please don't change anything below the line Do not change anything below this line

Parallel Programming is the key to save time

Based on the experiment on GW's high-performance computing cluster colonial one, I found the execution time of this program is inversely proportional with the number of parallel workers.
And the execution time has a linear relation with the number of rows that each worker has to compute.
Which shows parallel computing can greatly reduce execution time when the problem is parallelable.

Interesting Findings

No special environment: boundary of fire is round(Visualize in matplotlab and pure text)

Different p_continue_burn: 0.1(Left) or 0.9(Right) affect the shape of boundary

Density is on: Fire is more likely to spread to high density area(Right 1/3)

vegetation is on : Fire is more likely to spread to Hallepo-pine area (Right 1/3)

altitude is on : Fire is more likely to spread to higher place(Right side)

wind is on: Fire is affected by the north wind(From North to South)

All of the environment factor is on: altitude and wind are the most powerful factor

Technical Detail

Cellular Automaton Model

The model I use here is based on the research paper: "A cellular automata model for forest fire spread prediction: The case of the wildfire that swept through Spetses Island in 1990"

Each cell has 4 states:
State = 1: The cell contains no forest fuel. This state may describe the cells corresponding to parts of the city with no vegetation, rural areas with no vegetation etc. We assume that cells that are in this state cannot be burned.
State = 2: The cell contains forest fuel that has not ignited.
State = 3: The cell contains forest fuel that is burning.
State = 4: The cell contained forest fuel that has been burned down.

These 4 states changes under these 4 rules:
Rule 1: IF state (i,j,t) = 1 THEN state (i,j,t + 1) = 1.
This rule implies that the state of a cell with no forest fuel (empty cell) remains the same and thus it cannot catch fire.
Rule 2: IF state (i,j,t) = 3 THEN state (i,j,t + 1) = 4.
This rule implies that a burning cell at the current time step will be burned down at the next time step.
Rule 3: IF state (i,j,t) = 4 THEN state (i,j,t + 1) = 4.
This rule implies that the state of an empty cell that has been burned down in the previous step stays the same.
Rule 4: IF state (i,j,t) = 3 THEN state (i ± 1, j ± 1, t + 1) = 3 with a probability pburn.
I changed Rule to IF state (i,j,t) = 3 THEN state (i,j,t + 1) = 4 with probability 1-p_continue_burn

And I used the following formulas to determine p_burn: p_burn=p_h(1+p_veg)(1+p_den)p_wp_s

Effect of the wind speed and direction
p_w = exp(c₁V)f_t,
f_t = exp(Vc₂(cos(θ)-1))
θ is the angle between the direction of the fire propagation and the direction of the wind.

Effect of the ground elevation
P_s = exp(aθ_s)
θ_s = tan-1[(E₁-E₂)/l]

Table for p_den

Table for p_veg

Table for other parameters

Implementation

I implemented the model in the research paper to the code, and this is the structure of the program.

Parallel Design

The idea of parallel with MPI(Message Passing Interface) is also simple:

1. Assign a part of the full grid to each worker.
2. Let the worker exchange the top and bottom row after each iteration.