Forest Fire Management Using Geospatial Information System

F. Hoseinali
Email: [email protected]

M.A. Rajabi
Email: [email protected]

Department of Geomatics Engineering, University of Tehran, Tehran, Iran
Tel: +98 21 800 8841, Fax: +98 21 800 8837

Abstract:
Forest Fire is a natural phenomenon. It is part of the nitrogen cycle and it helps forests to grow healthy. However, “wildfire” is a huge and out of control fire which destroys human wealth and therefore, it is kind of disaster. The very first strategy to defend the forests against wildfire is to avoid it. So fire risk maps are produced and even prescribed burning is done. Nevertheless, wildfire happens and the only choice in this case is to control it. Knowing the fire behavior, one can use forest fire simulation to predict and control the wildfire. There are many factors which play a role in a forest fire. The most important ones are wind velocity and direction, index of forest flammability, slope and aspect of the terrain. Slope is an efficient factor to determine the fire spread direction. Velocity of the wind plays a more serious role in fire spread, if its direction matches with the aspect. A grid-based Digital Terrain Model (DTM) is used to extract slope and aspect of the terrain. Also raster maps showing the index of flammability, velocity and direction of the wind are used. All of these information are put in a Geospatial Information System (GIS) to simulate and visualize the wildfire spread. It is shown that GIS is a very efficient tool to predict and control the wildfire.

Introduction
Nowadays application of Geospatial Information Systems (GIS) in disaster management has extended considerably and in some cases it is even unavoidable. Forest fire happens from time to time and this has given human a chance to observe and develop different models for forest fire behavior. GIS as a powerful tool for management of spatial information, has also proved its potential in forest fire management. There are different applications of GIS in forest fire management out of which the most important ones are hazard map production, forest fire simulation and resource management. Simulation by itself has a main role in the management of forest fire. GIS uses various information layers such as Digital Elevation Model (DEM) and index of flammability along with different models for the pur[ose of forest fire management. Models can be simple or complex. Simple models have few parameters and can be implemented even without GIS, however, because of their simplicity their results are not so reliable. On the other hand, very complex models which use detailed physical characteristics of fire are not reliable either as there is not either enough or up to date information on their parameters. So finding an optimum model that takes advantage of sufficient number of parameters while has an acceptable level of simplicity is very important.

Importance of forest fire simulation
Forest fire destroys jungles which are an important part of human environment and devastates human wealth and facilities, causes weather pollution and is dangerous for the people who live near forests. Just in 1994 the United States spent more than one million dollars to protect forests against fire while 33 people lost their lives due to forest fires [3]. Therefore, there are many attempts to predict and prevent wildfire. Simulation, as one of these attempts, tries to forecast the forest fire extension and prepare a plan to control and manage the fire through preventing its extension, extinguishing the fire and diminishing its destruction. Obviously, efficient simulation models can have a major role in protecting the natural resources as well as environment management [2].

Fire, a chemical reaction
Fire starts when three elements come together. These elements are flammable material, oxygen and primary energy to start the reaction between the flammable material and oxygen. Without even one of these elements fire will not start and when all of them come together, fire is unavoidable.

Forest Fire
Depending on the source of fire, forest fires are divided into three categories [4]:

1. Natural fire: Among different reasons for natural forest fire, the most important one is lightening strikes. For example, in the North American forests, lightening strikes is the major reason for natural forest fire.
2. Prescribed burning: In regions where fire has a high possibility to start, prescribed burnings are intentionally done to control and prevent wildfire.
3. The other type of forest fire is the one which starts by human. This kind of forest fire is the most frequent one in the world.

Forest fire is a natural phenomenon and is a part of nitrogen cycle. Its exclusion is an interference with the nature. However wildfire is a disaster and any discussion about forest fire is essentially related to this type of forest fire. As mentioned before, prescribed burning is a strategy to protect forests against wildfire [4].

Forest fire and Geospatial Information Systems
GIS has three segments:

1. Spatial segment
2. Attribute segment
3. A set of spatial relations called topology

GIS is a powerful spatial processing tool which is used to solve many complex problems. In the context of forest fire some applications of GIS are as following:

• GIS in fire risk/probability assessment
• GIS in prescribed burn planning
• GIS in preventing fire and its spread
• GIS in fire simulation
• GIS in post fire assessment and monitoring
• GIS in disaster management

Among them, fire simulation is the subject of this paper.

Simulation
Simulation of a phenomenon is the operation that consists of studying its behavior in situations generated by virtual data in order to better understand the behavior of the phenomenon in the real world [1].

Parameters of forest fire simulation
There are many different parameters which can be used in forest fire simulation and obviously different models take different set of parameters into consideration. Using too many parameters in a model doesn’t necessarily increase the precision of model and even sometimes makes the model very unreliable. In other word, a reliable model uses the most important and effective parameters. The most important parameters for forest fire simulation are:

1. Wind which both its velocity and direction are important. Velocity of wind affects velocity of fire spread and direction of wind affects direction of fire spread. Information about wind can be extract from meteorological data.
2. Vegetation coverage which determines index of flammability. The best tool to analyze vegetation coverage is remote sensing and Canada has used this tool for this purpose ninety’s [8]. Of course direct sampling is necessary to better determine vegetation coverage.
3. Topography which determines slope and aspect of the terrain. Digital Elevation Model is the best source for topographic data.
4. Fire starting point which is the initial point in fire modeling.

Evaluation of some models for forest fire simulation
Not all of the models for forest fire simulation are based on GIS. Among them the followings can be mentioned [6]:

• Concentric model: It is the simplest model for forest fire simulation. Parameters of this model are starting point of fire and wind velocity. This model uses concentric circles to show fire spread in time. The radius of the circles can be calculated by R=x a Vf T where, T is the time in hour, Vf is wind velocity in kilometer per hour, a is a unit less coefficient, x is a coefficient which falls between 0 and 1 and is related to index of flammability and finally R is the radius of circles in kilometer. This model not only has some technical problems but also experiments have shown that fire spread is never circular. Therefore, this model is not so much reliable. Figure 1-a shows an example of this model.
• Pseudo-conical Model: One of the most important reasons which makes concentric model not adequate is that it doesn’t take the direction of wind into account. In pseudo-conical model fire is assumed to spread in a flat conic shape which its vertex is located at the starting point of fire and its extension is along the direction of wind. Even though this model takes the direction of wind into account and is more accurate in compare to concentric model still does not sufficiently match with reality. Figure 1-b shows an example of this model.
  
  
  
  Figure 1-a: Concentric model, two circles show fire spread prediction after 1 and 2 hours.

  
  
  Figure 1-b: pseudo-conical model, two conics show fie spread prediction after 1 and 2 hours
  It can be seen that in the above mentioned models there is no particular information layer, thus these models are not based on GIS. However, the followings are some of the GIS-based models.

• Polygonal model: In this model it is assumed that fire spreads in a polygon shape for which there is no limitation in number of its vertexes. The higher the number of vertexes is the more accurate the model is. The goal of this model is to obtain the coordinates of vertexes in any direction with any angle. If q is the angle measured anticlockwise from x axes of a Cartesian coordinate system then the coordinates of each vertex can be calculated by:
  
  
  Where Yd, Xd are the coordinates of fire starting point, Vf is wind velocity in kilometer per hour, T is the time in hour, Dv is wind direction, Ic is the flammability index, Mt is a parameter for DEM, q is sampling angle, ai is a set of three coefficients that sum of them is equal to 1 and shows the relative effect of slope, flammability and wind, and finally X and Y are the coordinates of sampling vertex. Figure 2 shows this model where wind direction is toward north-west. In figure 2-1 the velocity of wind is more than its velocity in figure 2-2. This model considers fire starting point, wind velocity and direction, topography and flammability and takes advantage of a number of information layers. This model is more accurate in compare to concentric and pseudo-conical model; however, it has some problems. For example the value of sampling angles is not obvious, or the effect of DEM and flammability index is averaged in each direction which is not appropriate for long distances, or determination of correct value for is very difficult if not impossible and even if it is determined it will not be correct after any change in environmental situations [6].
  
  
  
  Figure 2-1: Fire spread after 1 and 2 hours predicted by polygonal model , when wind velocity is severe.

  
  
  Figure 2-2: Fire spread after 1 and 2 hours predicted by polygonal model , when wind velocity is not severe.

• Network model: This model uses a grid-based DEM and considers burning area as cells in a raster network. Each burning cell has a burning period and in this period emits heat to its adjacent cells (four cells which have common edge with it). Also each cell has a flammability degree and if a cell receives more energy than its degree of flammability, it will begin to burn. So the process of burning continues until no cell is left for burning. The importance of this model is that it considers individual cells and follows the process of burning. However, this model takes just few parameters which are not enough and important information such as topography, velocity and direction of wind are neglected [7].

Implemented model (Normal model)
By taking the fire characteristics into account this paper tries to suggest a flexible model which is a function of all of the important parameters of fire. Like the network model, this model simulates the fire cell by cell.

Brief definition
In this model fire life in a cell is assumed to follow a normal PDF curve. In other words, it means that at the start of burning the intensity of fire is low and then it becomes high when it reaches its maximum and then it reduces until it diminishes. Physical experiments show that this assumption is not far away from reality and deviations from normal curve can be ignored in this model. During fire life, a burning cell emits heat to its neighboring cells. Rate of heat transfer depends on some factors that will be discussed later. If the received heat by a cell is more than its degree of flammability, it begins to burn and this process goes go.

Input data
Main information layers in this model are: DEM and index of flammability. DEM is used in grid form. Index of flammability or combustibility is a map in which forest regions are classified by their combustibility characteristics in a quantitative manner. Such a classification has been done for the forests in Canada [8]. Wind velocity and direction are two other main parameters in this model. The combined effect of these two parameters on the fire spread is very complex. Other parameters and coefficients used in the model are described later on.

The algorithm
The algorithm based on which this model works is as following:

1. Capturing and setting parameters
2. Loading DEM and index of flammability
3. Calculating normal vector for the whole cells
4. Acquisition of fire starting point
5. Performing main fire loop up to the end of burning
6. Illustration of the fire spread prediction
7. Providing statistical information about fire

The main part of this algorithm is fire loop which work as following:
This loop becomes active when the fire starts and becomes inactive when the fire stops. The input for this loop is the fire starting point and as a function of time calculates the amount of heat for the burning cell according to the normal cure. Meanwhile, the heat of the burning cell is emitted to the adjacent cells and if the received heat is more than the degree of flammability, adjacent cells start to burn too. Cells adjacent to one or more burning cells receive heat continuously but do not absorb all of the received heat. Some of the received heat is continuously lost. Temperature of a burning cell, value of heat transfer to the adjacent cells and value of heat loss are all depended on some factors which are set by parameters. Fire loop continues till no burning cell exists. After the loop is over, the cells are mainly divided into two categories: burnt and not burnt.

Effective factors
The parameters considered in this model are:

• Wind: Wind increases flame of the burning cell and on the other hand increases life of fire in the cell. Also wind increases rate of heat transfer toward its direction and reduces rate of heat transfer against its direction. The more is the wind velocity, the more is its effect on fire.
• Topography: Slope which is usually extracted from a DEM has some effect on fire spread. As heat has a natural tendency to move upward, therefore, rate of heat transfer increases by an increase in the slope.
• Index of flammability: In this model these characteristics are divided into three classes. As mentioned before, temperature of a burning cell is assumed to follow a normal curve. But this curve is not unique. Standard deviation of this curve is the first parameter which is extracted from index of flammability. When the standard deviation of a cell increases, fire life for that cell increases too. Furthermore, the extracted values from normal curve are multiplied by a coefficient to achieve a logical value for its temperature or heat. This coefficient is the second parameter which is extracted from the index of flammability. The third quantitative value which is extracted from index of flammability is the degree of flammability.
• Regional temperature: This factor and moisture are from the most important factors which determine forest fire risk. At the time of forest fire, loss of heat in cells that receive the heat, is strongly depended on regional temperature. Therefore, in cold weather the domain of forest fires are more limited and speed of fire spread is much lower.
• Precipitation: This factor cools both burning cells and cells adjacent to the burning cells.
• Proximity: A burning cell has its most effect on its adjacent cells which have a common edge with it. However, adjacent cells with a common vertex receive heat as well. So these cells can not be neglected but the coefficient of heat capture for these cells is lower than the adjacent cells with a common edge.

Combined effects
The effect of above mentioned factors is more complex when they act together:

• Combined effect of wind and topography: As mentioned before, wind increases the flame of a burning cell and decreases resistance to fire. If the aspect of a cell is against the wind, the effect of wind is increased and if it is toward the wind, the effect of wind is decreased and even sometimes it can be neglected. Effect of the wind can be evaluated by determining the angle between normal vector (perpendicular to the surface) of a cell and vector of wind direction.
• Combined effect of wind, topography and temperature: Another effect of wind is to accelerate cooling. In the case of adjacent cells to the burning ones, wind increases the loss of received energy. If the aspect of a cell is against the wind direction, the cooling effect of the wind will increase and reduction of temperature makes this effect more severe.

Simulation
Normal model was implemented by Matlab 6.1. Simulated region is a 2 km x 2 km region with 20 m resolution cells. The index of flammability is simulated with four different regions which can be used with different DEM scenarios. Figure 3 shows the assumed index of flammability. According to this index, forest No.3 has a low degree of flammability, forest No.1 has a medium degree of flammability and degree of flammability for forest No.2 is a bit more than the forest No.1. Region of meadow has the lowest degree of combustibility.

This model has many coefficients and setting these coefficients properly has a great effect on the model performance. To set these coefficients, some experiments are done on fire. For instance, coefficient of the diagonal neighbors, coefficient of the effect of slope and some other coefficients are determined in this way. Due to the lack of sufficient, appropriate and up to date information about forest fires in Iran, a simulated forest with a scale of 1:100 is put on fire. Burning this simulated forest provides a good tool to assess the performance of the model. Figure 4 shows one of the simulations. In this figure the terrain has a constant slope of 13 percent downward, the wind with the azimuth of 150 degrees has a velocity of 15 kilometer per hour. Each color shows fire spread after one hour.



Figure 3: Index of flammability for simulated area. This simulation has three different forest regions and a meadow.

Figure 4: Simulated model, fire starting point has been shown. Each color shows fire spread after one hour.
Conclusions and recommendations
Concentric, pseudo-conical, polygonal and network models have obvious problems and are not compatible with normal model. Therefore, their results are not compatible either. For assessment of the normal model, information acquired from the simulated as well as forest fire records in developed countries are used. This comparison shows that normal model can be an acceptable approximation of fire spread. The degree of compatibility between the result of normal model and the simulated forest fire is about 70 percent. However, just comparison with a real forest fire can reveal compatibility and power of this model.

Fire is a complex and dynamic phenomenon and therefore, simulation of a forest fire is a real challenge. For example velocity and direction of the wind can vary continuously but real time determination of these variations is almost impossible. On the other hand, the combined effect of wind and topography can not be easily determined. Moreover, the behavior of wind in the valleys curl current of wind is not predictable.

More assessment on the relative effect of the fire parameters is one of the requirements of this model in the future. In case a real forest fire record exists, setting coefficients with artificial neural networks seems to be appropriate. Also because of the similar effect of fire in all directions, one can use a network with hexagonal cells instead of square ones.

References

1. Brimicombe A., (2003), GIS, Environmental Modeling and Engineering, Taylor & Francis
2. Wainwright J., Mulligan M., (2004), Environmental Modeling, John Wiley & Sons
3. Cunningham W.P., Woodworth Saigo B., (1999), Environmental Science, McGraw-Hill
4. Forest Fire in the American Southwest, https://forestfire.nau.edu/archives_june03.htm
5. Burrough A. and McDonnell R.A. (1998), Principles of Geographical Information Systems, Oxford University Press.
6. Ahmed Saidi, A. MISSOUMI, CNTS, (1999), The use of the GIS into the Forest Fire prediction The Simulation Model, https://www.cs.wright.edu/~bwang/course/ceg434634/pa1.pdf
7. Borlawsky T., (2000), Forest Fire Simulation using Percolation Theory, www.dbmi.columbia.edu/~tbb7001/projects.htm
8. Canadian Wildland Fire Information System (CWFIS), https://fire.cfs.nrcan.gc.ca