SPATIAL SIMULATION OF F-IRESPREAD: TWO DIFFERENT APPROACHES APPLIED TO TAPADA DE IMAFRA, CW PORTUGAL 0s LOUREIRO', Pauln FERNANDES', Agostinho LOPES', Duncan H E A ~ ~ ~ F I Eand L DFrancisco ~ RE GO^ Departamento Florestal - UTAD (Portugal); I World in a Box, Finland; o de Ecologia Aplicada "Prof, Baeta Neves" - ISA. (Portugal) Email:
[email protected];
[email protected]
*>-z
'we attempt in this study to simulate fire spread on the landscape &pTqa& de Mapa, using two applications. In this work we, in a sirnple t&, try to test the use of two applications: FARSITE, fire area hulator - version 3.0 (Finney, 1997) and Landlord, spatial modelling - version 1.2 (Heathfield, 1999), using available information and z&eloping the requercd inputs, $t&
'I'BE STUDY AREA
1
T C V N ~~I / eMufra is a public land eslate with 534 ha, located in CW Portugal, north of Lisbon.
The soil occupation is varied, being constituted by pure or mixed stands of Pirills pinaster, PIUIJ.P pineu, Quercus stlber and Eucrrl~ptzlsglohzrlus, and riparian vegetation areas along the main water lines. A significant area is occupied by shrublands (Erica sp., [flex sp.) and rangelands for wildlife feeding The estate i s surrounded by a high wall which serves as an effective fuel break.
LANDLORD, SPATIAL MODELLING SYSTEM P R S I O N 1.2) Landlord is an component-based, open-architecture spatial modelling system It's implemented as Windows software. It looks a bit like a GIs, but is q uile different. In Landlord, maps are objects, not tiles. objects reside in map collections. Map coIlections interact with spatial processes, which take one or more maps as inputs and tnake changes to one or more maps as output. All of these components are organised and configured within workspaces which can be saved and reopened. (Heathfield, 1999) Landlord was developed in the frame of ModMed 1 project
MODMED LANDSCAPE FlRF, MODELLING eathfield, 1997)
The model attempts to predict the rasters affected by fire. The scape is represented using a raster rnapping system, which is a grid asters with variables describing the current state of each raster. In delling fire at tlie landscape level, we are simply trying to predict the nt of spread of a fire once it has started. The job of the landscape fire el is to come up with a list of rasters which have been fire-aected the time the fire goes out The fact that we are only interested in dicting the extent of spread makes our requirements, and therefore ur approach, slightly different from most other fire models, which are usually concerned with predicting the rate of spread. The model treats the fire event as instantaneous. Although it is obvious that an actual fire event takes at least some time to occur, the modd treats fire as an instantaneous event. Wind speed and direction are constant fbr the duration of the simulation. The sequence in which the pattern is developed is not intended to correspond to any likely sequence of fire pattern development in an actual fire. The sole objective of the simulation is to predict the damage pattern which results when the fire has gone out, The model treats fwe as a contagious process. In order to predict the extent of a fire's spread, we run a simulation. The simulation is iterative, which means that a set of calculations are performed over and over again to produce a result. At the first iteration, when the fire begins, the fire starts in one or more rasters. Each raster which is alight (a 'source' raster) has one 'opportunity' to ignite any adjacent rasters which are not alight ('target' rasters). Depending on the conditions, the model calculates a probabilty value (0-1) for the spread of fire from source raster to each target raster - the contagion probability. A random number (0-1) is generated, and if this is less than the probability calculated, then the target raster catches fire. At the end of each iteration, the source rasters stop being on fire and the newly alight target rasters become the new source rasters. As long as there are still rasters alight, the simulation continues by beginning another iteration. When there are no rasters left to burn, the simulation finishes, and a list of burned rasters has been collected. This approachbhandles fire as a contagious process.
There are several factors which are used to calculate the probability of contagion:
1
Slope
2.
Wind (speed and direction)
3.
Fuel horizontal continuity
4.
Fuel moisture content
A factor value is calculated for each, and then each factor value is weighted according to the model configuration. (For example, it is possible to increase the relative influence of wind against the influence of the other factors.) Applying the ModMed fire rnodel to some historical fire data from Tapada da Mafra
On the 23rd August 1975, a large fire burned for three days, starting at 15:OO. Each night the fire went out and re-ignited in the next day as the humidity fell.
We assumed:
+ + +
W e were modelling a surface level (the litter and shnib layers) tire contagion, using horizontal continuity of the lower levels of vegetation for modelling the spread of fire. The moisture content would be sitnilar across the whole Iandscape and the land use patterns for 1975 were the same in 1974. The same mean wind speed was used for the three days; wind direction was determined by the valley of the major water stream (Figure 1)
THE SOURCE DATA AND THEIR PREPARATION
Four map files (resolution 12.5 rn) were used: a map of the outline of the Tapada; an elevation map for Mafra and the surrounding area; L
L
a map showing 20 land use types for Mafra and the surrounding area in 1974; a map showing the outlines of the fire-affected area.
In each case, data were available for a rectangle which included the (non-rectangular) area of the Tapada. However, we only used data from within the Tapada because the estate is surrouhded by a high wall which serves as an effective fire break. This unnatural situation helped to simplify the modelling. j
The original maps were stored as shape files (Arcview version 3 2).They were imported to Idrisi to produce raster grid maps.
ESTIhIATlNG THE FUEL HORLZONTAL CONTTNUllI' (%,) We assumed that we were modelling fire coritagion for surface fuels The fire might affect the tree canopies, but the spread of thc fire would be in litter and shrub layers. So, we were only interested in the horizontal continuity of the lower levels of vegetation for tnodelling the spread of fire To translate the land use classes to file1 horizontal continuity values, we prepared a table which contained estimates of he1 horizontal continuity for each land use type classes. Then we used a hand-made program to translate the land use classes to fuel horizontal continuity values. The resulting rnap is shown below (Figure 2) *
Figure 2
L
TXMATING THE FVEL MOISTURE (%) After experimenting several combinations, considering the imity to water lines or the aspect, we eventually decided to use a constant value of 10% for the wole area (Loureiro, 2000).
ENPT TO S3[MULATE - FIRST RESULTS We set up workspace using the ModMed Fire process model. ran the simulation several times by setting fire to a small area araund (201,155), and found that the model typically predicted a spread in a re easterly direction than actually happened. A typical result is shown
SOME CONCLUSIONS The model typically predicted spread in a bigger area than it actually happened (fire fighting effect)
I
If we turn wind direction to near 360°, a large area would burn westward uphill, with good availability of fuel Landlord' allows, with simple data, to define the most dangerous areas and prioritise them for management actions The next step will be to adjust the fuel continuity maps to land use in 2000, according to the custom fuel models developed.
APPLYING FARSITE USING CUSTOM FUEL MODELS DEVELOPED FOR TAPADA DE MAFRA
I
The objectives of the study were to determine the areas within Tapada de Mafia with a higher potential of summer fire hazard, enabling. a subsequent test of different lnanagernent strategies designed to minimize the consequences of a wildfire. The existing Tapada de Mafra GIs was used to generate the raster data themes: elevation, slope, aspect and fud models. A field survey of the mapped vegetation patches (land use map of 1995) was carried (summer of 2009, correcting their boundaries where necessary, and assigning custom he1 models developed with the BEHAVE system (l3urgan and Rothemel, 1984). Data to develop the fuel models was obtained from indirect sampling procedures, namely line transects, and was subsequently used to estimate loading by fuel type, condition and size-class; published information was the source for the remaining fuel inputs. 21 custom h e 1 models resulted from this process (Table I). Tree canopy caracteristics, incluing cover, were considered, at this time, spatially constant.
, ;
Table 1 - Custom fuel models
No.
Hazard class
PNB-iVlATO MUITO ALTO
14
Very high
NB+PNM-MATOS ALTOS
15
Very high
NB+PNM-RIATO DE ALTURG R / ~ D J A 16
Moderate
PNB-MAT0 BAULO
17
Low
PNM-RAMOS MORTOS NO SOLO
18
LOW
PNM-SEM VEGETACAO ARBUSTIVA
19
LOW
EU-MAT0 MUITO ALTO
20
Very high
EU-MAT0 DE ALTURA JYI~DIA
21
Moderate
EU-MAT0 RASTEIRO E DISPERSO
22
Low
MATO MUIT0 ALTO
23
Very high A
MATO ALTO
24
l3gh
MATO DE ALTURA &DIA
25
Moderate
FETOS DE ALTURA IM&DIA
26
Moderate
MATO BAIXO
27
Low
FD-MAT0 MUITQ ALTO
28
Very high
FD-FETOS DE ALTURA MDIA
29
Low
FD-MAT0 BAMO
30
Low
FD-SEM VEGETACAO ARBUSTIVA
31
LOW
PASTAGEM-HERBACEASMI~DIAS
32
LOW
FR-JUNCOS E FETOS
34
LOW
This initial approach uses weather data collected by a meteorological station located within the Tapada area, for the period of 10 lo 17 July 2000, that should reflect the typical summer situation. The following tables present the fuel, weather and wind inputs for the Farsite simulations Figures 4 to 6 present a fire hazard [nap fur Tapada de Ma>n and results of the FARSTTI? nrn.
CUSTOM FIJEL FILE FMnd 1 1.08 10CIH LiveB LiveW LiveW SAV Depth XtMois t DHt LHt
IHSAV 1,iveEISAV
WEATliER FILE Mollth Day Precip Hour1 Humid2 Elevation
Hour2
Temp1 Temp2
METRIC 7
10
2
2300
1500
14
7
11
0
600
1500
13
7
12
0
630
1600
12
7
13
0
330
1430
15
7
14
0
800
1500
14
7
15
0
2400
1300
17
7
16
0
700
1530
14
WIND FILE Month Day Hour Speed Direction CloudCover METRIC 7
10
0
21
23
0
7
10
100
18
23
0
7
10
200
13
0
0
'
P
Humid1
Figure 4 - Fire hazard map for Tapada de Mafia
Figure 5 - FARSITE simulation result
Figure 6 - Fire line intensity map, exported from Farsite and imported to ' ' ArcView CIS.
References Heathfield, Duncan and Francisco Rego. 1997. ModMed Landscape Fire Modelling. Online in http ://www.worldinabox.co.uk/ModMed/FireModel/index. htm Loureiro, Carlos and Duncan Heathfield. 2000. Using the fire model for Mafra. Online in http://www.worldinabox.co.uk/ModMed/Mafra/MafraTest.html Heathfield, . Duncan. 1999. About Landlord. Online in http : / / ~ . w o r l d i n a b o xco.uk/landLord/brieflntroduc . tion.html
,
Finney, Mark. 1997.FARSITE- Fire Area Simulator, version 3.0. Users guide and Technical Documentation. Systems for Enviranmental Management. Missoula
Burgan, R. E. and R C. Rothemel. 1984. Behave: fire behavior prediction and fuel modeling system. USDA, Far. Serv. Gen. Tech. Rep. INT-167. Intemt. For. And Range Exp. Stn.Ogden, UTAH.
