TRG Chapter 3: The Fire Grid
- James Walker
Table of contents
- Introduction
- Development philosophy of PHOENIX
- The Fire Grid
- Inputs
- Fire Behaviour
- Fire Perimeter Propagation
- Asset Impact
- Outputs
3. The Fire Grid
3.1 Purpose
The Fire Grid is central to PHOENIX in that it is used to manage and represent data. The Fire Grid function of PHOENIX is a means to capture input data and represent it in a gridded modular fashion that the various models can utilise. Additionally, it is a means to record and store the various outputs from PHOENIX.
3.2 Basis
A geospatial grid that is created using various input data layers provided by the user.
3.3 Inputs
- Fuel types;
- Wind reduction factors;
- Time since last fire;
- Topography;
- Assets and values;
- Road proximity; and
- Linear fuel disruptions.
3.4 User interactions
The user defines a grid resolution for the Fire Grid.
3.5 Description
Input data must be properly prepared in a GIS such as ESRI's ArcGIS or MapInfo (refer to Chapter 4: Inputs for more information). There is no fixed resolution requirement for the input data; however, a data resolution of 25 or 30 m is recommended for meaningful simulations.
PHOENIX reads, into computer memory, gridded data at the input data resolution (e.g. 30 x 30 m). The user defines a Fire Grid resolution for simulation calculations, as this improves processing speeds and makes the scale of the calculation more realistic. It is generally recommended this be between 120 – 210 m (e.g. 180 x 180 m) and be a multiple of the underlying input data. This cell size was found to strike a balance between data accuracy and computer processing efficiency and is consistent with the minimum fire width required to achieve a quasi-steady-state rate of spread (e.g. Cheney and Gould 1995).
3.5.1 Fire grid sampling
PHOENIX uses a sampling protocol to sample underlying input data for the purpose of running simulations.
Figure 6. 180 m fire grid in white over a 30 m resolution fuel type input layer. Figure 7. The 30 m resampling pattern shown as grey dots on a single 180 m fire grid cell.
In the case of Figure 6, a user-defined Fire Grid resolution of 180 m has been specified for simulation runs. However, the fuel type input data was provided as a 30 m grid, and the 180 m simulation grid and resolution of underlying data do not necessarily perfectly align. PHOENIX deals with this through a resampling protocol. Figure 7 is a closeup of one of the 180 m cells from Figure 6, and the grey dots illustrate how PHOENIX resamples underlying fuel type data. The sampling method varies depending on the cell attribute as illustrated in Table 2.
Table 2. Grid cell attributes and their sampling method
Cell Attribute | Fire grid cell sampling method | Description |
|---|---|---|
Elevation | Centroid | In metres derived from DEM |
Aspect | Centroid | In degrees derived from DEM |
Slope | Centroid | In degrees derived from DEM |
Grass load | Area weighted | Grass load in t/ha |
Surface load | Area weighted | Forest surface fuel load in t/ha |
Elevated load | Area weighted | Forest elevated fuel load in t/ha |
Bark load | Area weighted | Forest bark fuel load in t/ha |
Total fuel load | Area weighted | Total fuel load in t/ha |
Wind reduction factor | Area weighted | Fuel type specific 10 m to mid flame height (1.2 m) wind reduction factor |
Road proximity | Centroid | Distance from cell centroid to nearest road |
Disruption | Area weighted | Effective linear disruption width in metres |
Wind modifiers | Centroid | Terrain affected wind modifiers |
Centroid sampling simply takes the centroid value of the input data cell (e.g. 30 m cell) and averages it for the Fire Grid cell (e.g. 180 m cell).
The area-weighted method samples each 30 m cell. Where the Fire Grid (e.g. a 180 m cell) contains a mixture of 'grass' and 'woody' fuel types, the fire spread calculations will be an area-weighted average of the fuel types. For example, if 12 of the underlying 30 m cells are 'grass', then the relative area of grass will be 12/36 or 33%. Thus, grass will contribute 33% to the weighted average fire behaviour attributes. Refer to Section 6.1: Fire Perimeter Propagation for more information on how weighted averages for fuel are used.
3.5.2 Fire grid data retrieval
Prior to simulating fire spread, PHOENIX loads the Fire Grid tiles within a 10-tile buffer of the ignition point into memory (see Figure 8). Tiling of data provides for efficiency: large domains do not need to be pre-loaded into memory and can be loaded on-demand in a progressive manner as the simulation proceeds. PHOENIX computes fire spread using dynamic time steps; time steps are partially determined by the resolution of the spatial information being affected by the fire. High data resolutions will result in an increased number of computations, slowing the overall simulation process. Data resolutions that are higher than the uncertainty around parameter estimates will not necessarily provide any benefit; in fact, analysis times will be increased, however, there may be limited or no improvement in the overall result.
Figure 8. Data is retrieved in tiles of 10 x 10 Fire Grid cell as a fire spreads in the landscape. Red polygons show the fire perimeter and the varying shades of green indicate fuel loading with darker green representing more fuel. Additional cells are shown in this illustration because they have been impacted in some way (perhaps by embers).