effect of two-way coupling on the calculation of forest ... of two-way coupli… · effect of...
TRANSCRIPT
Effect of two-way coupling on the calculation of forest firespread: model development
A. M. G. LopesA,B, L. M. RibeiroA, D. X. ViegasA and J. R. RaposoA
AADAI (Associacao para o Desenvolvimento da Aerodinamica Indistrial), DEM (Department of
Mechanical Engineering), University of Coimbra, PT-3030 Coimbra, Portugal.BCorresponding author. Email: [email protected]
Abstract. The present work addresses the problem of how wind should be taken into account in fire spread simulations.The studywas based on the software systemFireStation, which incorporates a surface fire spreadmodel and a solver for the
fluid flow (Navier–Stokes) equations. The standard procedure takes the wind field computed from a single simulation inthe absence of fire, but this may not be the best option, especially for large fires. The two-way coupling method, however,considers the buoyancy effects caused by the fire heat release. Fire rate of spread is computed with the semi-empiricalRothermel model, which takes as input local terrain slope, fuels properties and wind speed and direction. Wind field is
obtained by solving themass, momentum and energy equations. Effects of turbulence on themean flow field are taken intoaccount with the k – e turbulence model. The calculation procedure consists of an interchange between the fire spreadmodel and the wind model through a dynamic interaction. The present work describes the first part of this research,
presenting the underlying models and a qualitative sensitivity analysis. It is shown that the update frequency for thedynamic interactionmarkedly influences the total calculation time. The best strategy for updating thewind field during thefire progression is presented. The dependence of results on mesh size is also described.
Additional keywords: fire spread simulation, fire–wind interaction, surface fire, wind simulation.
Received 19 March 2016, accepted 31 May 2017, published online 17 August 2017
Introduction
Prediction of forest fire spread has been a challenge toresearchers and managers for several decades, but, in spite ofthe various models and fire behaviour prediction systems that
have been developed as described by Sullivan (2009a, 2009b,2009c), there is not yet a commonly accepted fire behavioursimulator that can be applied in operational conditions for large
and complex fires. This is due to the complexity of the physicaland chemical processes that are involved in large-scale firesrequiring a large number of input parameters that are not easy toobtain and physical models that are not yet fully developed.
Physical models have been formulated mathematically andimplemented in numerical codes like Grishin (1994), Albini(1996), Linn et al. (2002), Sero-Guillaume and Margerit (2002)
and Mell et al. (2007). Nonetheless, practical applications ofthese models are not yet possible owing to large computationalrequirements and uncertainty regarding the description of key
physical processes. Evaluation of some of these models showederrors of up to an order of magnitude (e.g. Marino et al. 2012).
Simpler semi-empirical models to predict fire spread like the
one developed by Rothermel (1972, 1983) have been of moreeffective use in fire behaviour prediction both in scientific and inoperational work (Finney 1998; Andrews et al. 2008). In spite ofits limitations, like the fairly small number of fire spread
regimes that it considers and the limited range of conditions
for which it was developed and validated, Rothermel’s modeldeveloped for surface fire spread prediction is still one of mostwidely applied in fire simulators. In the present study, we shall
assume this scope of application as well, although other firespread modes like crown fires and spot fires can be added to thefire simulator using adequate models to link surface to crown
and spot fires.It is commonly accepted that the fuel bed properties, terrain
slope and wind velocity are the most important factors affectingfire behaviour, namely its rate of spread (ROS). Many studies
have shown the particular importance of wind velocity near thefire front (see Clark et al. 1996a, 1996b; Viegas 2004a; Linn andCunningham 2005) and the role of fire-induced wind in modi-
fying fire spread (see Viegas 2004a, 2006). The present workwill be dedicated mostly to this parameter and to its role on firebehaviour. Atmospheric wind flow is a boundary layer-type
turbulent flow of air that is modified by the local topography, thelevel of instability of the atmosphere and its interaction withroughness elements like vegetation, houses, rocks and other
ground-cover obstacles upwind from the fire front. In order toestimate fire behaviour correctly, it is necessary to predictchanges in wind velocity components induced by the presenceof the fire in the portion of the atmospheric boundary layer
CSIRO PUBLISHING
International Journal of Wildland Fire 2017, 26, 829–843
https://doi.org/10.1071/WF16045
Journal compilation � IAWF 2017 www.publish.csiro.au/journals/ijwf
surrounding the area of the fire. A forest fire affects local windconditions considerably. The wind patterns that would exist inthe absence of a fire are replaced by localised flow structures
resulting from the interaction between the atmospheric wind andthe convection induced by the flaming combustion front. Severalauthors have analysed the coupling between the fire and the
atmospheric flow (Clark et al. 2004; Viegas 2004b, 2006; Coenet al. 2013) but this has not yet been applied in fire simulators thatcan be used operationally.
Models for simulation of wind over complex topographymay be classified into two main categories: prognostic anddiagnostic models. Prognostic models operate at large scales,with mesh sizes typically larger than 2–4 km. They describe the
dynamics of the atmospheric boundary layer, including predictivecapabilities for the variables field, according to the diurnal cycle.Weather Research and Forecasting (WRF) (Coen et al. 2013) and
(Mesoscale Model (MEMO) (Flassak and Moussiopoulos 1988)fall into this category. As pointed out in Forthofer et al. (2014),owing to the poor mesh resolution, prognostic models are not able
to capture flow effects from topography with significant interestfor fire simulation.Nevertheless, operatingwith a large simulationdomain, prognostic models constitute a good source of input data
for finer-mesh simulations. Diagnostic models provide a steady-state simulation and may be divided into two types. Mass-consistent models (also known as kinematic models) take as inputa first guess of the velocity field obtained from interpolation of the
input data, and modify it so as to achieve a divergence-freesolution. The stability of the atmosphere and other thermal effectsare simulated by adjusting the horizontal and vertical velocity
components. The Nuatmos (Ross et al. 1988) and WindNinja(Forthofer et al. 2009) models fall into this category. Kinematicmodels require very little computational effort.As shown inLopes
(2003) and in Forthofer et al. (2014), they provide rather goodsolutions upwind and above terrain features, but often fail on thelee side, especially if flow separation is present. Navier–Stokessolvers, such as WindWizard (Forthofer et al. 2014) and Wind-
Station (Lopes 2003), representing the second type of diagnosticmodel, constitute a more realistic approach, solving for mass andmomentum (Navier–Stokes equations) conservation, and often
include some form of turbulence closure as well. Application ofsuch models to atmospheric flows has been successfully done inthe past, with the Askervein experiment representing one of best-
known benchmark cases (Raithby and Stubley 1987; Castro et al.2003; Lopes 2003; Forthofer et al. 2014).
Several authors have coupled fire spread algorithms with
weather models (e.g. Clark et al. 2004, Coen et al. 2013). TheWRF-Fire model (Coen et al. 2013) is based on the WRF
software, which is a prognostic weather model developed forlarge-spatial-scale simulations. The WRF-Fire version includes
a model for coupling fire and wind. Coen et al. (2013) provide arather detailed description of this model, with application to atest case on flat terrain. WRF is configured in LES (Large Eddy
Simulation) mode, in order to be applicable to smaller scales(,40-mmesh spacing). On the simulations reported in the latterwork, zero constant slope should be considered, in order to use
periodic boundary conditions. It is not clear how themodelwouldperform in a real complex terrain situation with a real firesituation and what the limits are as concerns spatial resolutionand associated numerical errors and computational effort. The
WRF model has also been used for providing initial conditionsand boundary conditions for theCAWFE (CoupledAtmosphere–Wildland Fire Environment) predictionmodel (Coen 2013). This
model was applied to a specific wildland fire (Coen and Riggan2014) and, according to the authors, the main distinguishing firefeatures were correctly depicted by the simulation. Numerical
weather prediction models, such as WRF, may also be usedfor improving the accuracy of simpler kinematic modelsthrough a downscaling process, as demonstrated in the work
by Wagenbrenner et al. (2016).The present study describes an evolution of the fire simulator
(FireStation) presented in Lopes et al. (2002), now incorporat-ing a wind calculation module that takes into account feedback
from the convective flow induced by the fire on the atmosphericflow and its effect on the spread of the fire. This wind calculationengine is based on the original WindStation model (Lopes
2003), including additional features such as, among others, theinclusion of the energy equation and more flexibility in theassignment of incident velocity profiles. The present paper
describes the underlying physical and numerical models, aswell as simplified simulation scenarios. The fundamental equa-tions for fluid flow and turbulence modelling are presented,
along with a brief description of the numerical methodology fortheir solution. Next, models for fire spread are described,including fire ROS, fire growth and interaction between windand fire. A qualitative sensitivity analysis of this methodology is
undertaken, where the influence of the update frequency for thedynamic interaction between the fire and the wind is studied.The influence of mesh resolution for wind calculation is also
analysed.
Theory background
Wind field modelling
Governing equations
In the present approach, the Boussinesq approximation isadopted, considering that density is constant except in its
contribution to buoyancy forces. As the rate of change inboundary conditions due to fire is very low, the flow is solvedusing a steady-state approach. In fact, air flow velocity is much
greater than fire spread velocity. This means that, during thetime needed for a fluid parcel to travel through the fire-affectedarea, the heat release rate distribution at the ground due to fireremains essentially unchanged – thus, the fluid flow may be
regarded as behaving in a steady-state manner with respect tofire.
The Navier–Stokes equations represent momentum conser-
vation for a fluid flow. Considering the simplifications justdescribed, the formulation in a Cartesian coordinate system (seee.g. White 2011, p. 247), for a steady-state condition and
constant density, is:
@
@xjruiuj� �
¼ @
@xjG@ui@xj
� �þ @
@xjG@uj@xi
� �� @p
@xiþ Bi ð1Þ
where mi ¼ (u, v, w), xi ¼ (x, y, z) for i ¼ (1, 2, 3), r is the fluiddensity, p is pressure, and G ¼ meff ¼ m þ mt is the diffusion
coefficient for momentum, i.e. the effective viscosity, which is a
830 Int. J. Wildland Fire A. M. G. Lopes et al.
function of the turbulence variables, as described below. Buoy-ancy forces are present only for the vertical momentum compo-nent (i 5 3) and are given by:
B3 ¼ rref gb T � yð Þ ð2Þ
whereb is the thermal expansion coefficient, g is gravity,T is the
temperature and y is the temperature corresponding to the dryadiabatic lapse rate (�108C km�1). Thus, in the absence of otherthermal effects, the neutral atmosphere will produce no buoy-
ancy forces.Mass conservation is described by the continuity equation,
for which the steady-state formulation is:
@
@xiruið Þ ¼ 0 ð3Þ
The temperature field, necessary for the computation of
buoyancy forces, is obtained through the solution of the energyequation:
@
@xircpuiT� �
¼ @
@xiG@T
@xi
� �þ ST ð4Þ
In this case, the diffusion coefficient G is obtained as:
G ¼ mPr
þ mPrt
� �cp ð5Þ
with cp as the air specific heat and Pr and Prt as the laminar and
the turbulent Prandtl numbers respectively. The source term STin Eqn 4 accounts for energy exchanged with the ground, whichincludes heat release by the fire. Although not considered in thepresent work, this term may include modelling other heat
sources through raster mapping to take into account, for exam-ple, solar radiation heating. Alternatively, a specified tempera-ture field may be imposed at the ground.
As turbulence effects on fluid flowmay occur on a very largerange of time and length scales, obtaining the solution of thetransport equations for the instantaneous variables field is not
feasible for most practical applications, including atmosphericflows, where length scales may range from centimetres up tokilometres. As awork-around, eddy viscosity turbulencemodels
simulate the effects of turbulence on the mean flow field byincreasing the effective viscosity. One the most widely used isthe k – e turbulence model, proposed by Launder and Spalding(1972, 1974). The turbulence kinetic energy (k), which is a
measure of the flow velocity fluctuations, is related to turbu-lence intensity TI as:
TI ¼
ffiffiffiffiffi23k
qV
) k ¼ 3
2TIVð Þ2 ð6Þ
whereV is the local average velocitymagnitude. The dissipation
rate of turbulence kinetic energy (e) is related to the dissipationlength scale L as follows:
L ¼ Cmk3=2
e) e ¼ Cm
k3=2
Lð7Þ
Turbulence kinetic energy and its dissipation rate are used to
compute turbulence viscosity with the following equation:
mt ¼ Cmrk2
eð8Þ
In the original formulation of the k – emodel, the ‘constant’Cm
was adjusted to the value Cm ¼ 0.09, based on laboratoryexperiments for flow over a smooth plate. As stated in Raithbyand Stubley (1987), this value is to be corrected for the case ofatmospheric flows. Data analysis from several atmospheric stud-
ies byPanofsky andDutton (1984) providedCm¼ 0.033 for awiderange of surface conditions and scales. This is the value adopted inthe present work. The turbulence kinetic energy and its dissipation
rate are described by the following transport equations:
@
@xjrujk� �
¼ @
@xjmþ mt
sk
� �@k
@xj
� �þ P1 þ G � re ð9Þ
@
@xjruje� �
¼ @
@xjmþ mt
se
� �@e@xj
� �þ ek
C1P1 þ C1C3G � C2reð Þ
ð10Þ
where sk and se are turbulence Prandtl numbers. The term P1
represents the production rate of k as the result of the velocity
gradients:
P1 ¼ �ru0iu
0j
@ui@xj
¼ mt@ui@xj
þ @uj@xi
� �@ui@xj
ð11Þ
The term G (Markatos et al. 1982; Henkes et al. 1991)represents the production or destruction of turbulence due to
buoyancy forces and has the following formulation:
G ¼ �bgmtst
@T
@zð12Þ
The model constants are:
sk ¼ 1:0; se ¼ 1:3; C1 ¼ 1:45; C2 ¼ 1:9; C3 ¼ tanhwffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u2 þ v2p
ð13Þ
Effects of buoyancy on turbulence dissipation are less
understood than the effects on turbulence kinetic energy andthere is no consensual formulation for this. The term C3, asformulated above (Henkes et al. 1991), cuts buoyancy effects on
dissipation for horizontal shear layers and progressively allowstheir influence as the velocity vector approaches the verticaldirection. The remaining constants are given in Launder andSpalding (1972, 1974).
Boundary conditions and initialisation
At the ground boundary, a logarithmic law is assumed forvelocity variation with height, z, above the ground:
V
V�¼ 1
kln
z
z0
� �ð14Þ
Two-way coupling: model development Int. J. Wildland Fire 831
where V is the velocity magnitude, k ¼ 0.41 is the von Karmanconstant and z0 is known as the roughness parameter, which is
taken as a fraction f of a typical roughness height (h0) through theequation z0 ¼ fh0. Values for f range from 0.03 to 0.25,depending on the layout and shape of the roughness elements
(ground cover elements such as vegetation). A conservativevalue of f ¼ 0.15 is recommended by Plate (1982) for mostnatural surfaces. The ratio z/z0must always be higher than unity,
which means that the height of the first calculation level (centreof the control volume adjacent to the ground) should be higherthan the roughness parameter z0. The friction velocity V* is ameasure of the ground shear stress, tw ¼ rV 2
� and provides the
boundary condition for the momentum equations:
tw ¼ r Vk= lnz
z0
� �� �2
ð15Þ
Eqn 15 is evaluated at the first computational node near theground.
Boundary conditions for the turbulence quantities are obtainedbased on the hypothesis of local equilibrium. It can be provedthat, based on this assumption, turbulence kinetic energy is givenby (Raithby and Stubley 1987; Castro et al. 2003):
k ¼ V 2�ffiffiffiffiffiffiCm
p ð16Þ
which constitutes the boundary condition for k. The dissipationrate near the ground is given by:
e ¼ V 3�
kdð17Þ
with d as the distance of the first node to the ground.
For the energy equation, the ground boundary condition is theheat flux rate exchange with the ground. If a local fire exists, thepropagation model provides this value, to be included in the termST, inEqn 4. This is computedwithEqn 30, to be presented below.
If no fire exists, an adiabatic boundary is a reasonable assumption.At downstream and lateral boundaries, a null gradient condi-
tion is imposed for all variables (zero first derivative). One
should be aware that boundaries are always subject to someuncertainty; consequently, the wind calculation domain shouldalways be larger than the area of interest, in order to minimise
the effect of these boundaries on the region of interest.For the top boundary, null gradients are assigned to all
variables.Initial values must be assigned to the whole 3-D wind
calculation domain – this includes velocity magnitude anddirection, as well as turbulence quantities and temperature.Boundaries where inflow is verified are considered as inlet
boundaries, with conditions that remain unchanged during thecalculation. Temperature values are obtained from a specifiedprofile or computed from a selected class of atmospheric
stability. For velocity and turbulence quantities, fully developedprofiles typical of atmospheric flows are internally available inthe software, such as the example shown in Fig. 1. Velocity field
initialisation may also be done based on data available frommeteorological stations. In this case, wind speed and direction ateach meteorological station are employed to compute the shapeof the local boundary layer, using a power law equation of the
following form:
V=V1 ¼ z=dð Þ1=a ð18Þ
where VN is the velocity at the top of the boundary layer, d is theboundary layer height and a is an exponent characterising theboundary layer shape (Plate 1971). The profile, characterised by
the values VN, d and a, is determined from knowledge of thevelocity at two heights, and an estimated value of the boundary
(a) (b) (c)
1
10
100
1000
0 10 20
Z (
m)
V (m s�1)
1
10
100
1000
0 0.125 0.25 0.375 0.50
TI
1
10
100
1000
0 5 10 15 20 25
L (m)
Fig. 1. Example of fully developed profiles for wind field initialisation: (a) velocity; (b) turbulence intensity; (c) turbulence length scale.
832 Int. J. Wildland Fire A. M. G. Lopes et al.
layer height. It is assumed that wind direction is constant withheight at the station location. Values at each computational
mesh node are calculated through bi-linear interpolation fromeach available station location, using the inverse of the distanceas the weighting factor.
Numerical solution
The mesh where the flow solution takes place is of a
structured type (regular arrangement of parallel epipedic ele-ments), corresponding to a rectangular computational domain.Themesh isCartesianwith uniform spacing in the horizontal (x, y)
plane, running from west to east and from south to north, asexemplified in Fig. 2a. In order to properly describe the topo-graphic characteristics, a terrain-following (boundary-fitted) coor-
dinate system is adopted in the vertical direction, as depicted inFig. 2b. The mesh is generated algebraically, using a gradualvariable expansion factor in the vertical direction, computed so asto satisfy the specified number of vertical levels, the near-ground
first-level height and total domain height, ensuring a smoothtransition in the mesh spacing.
The transport equations are transformed from their original
Cartesian form into a generalised coordinate form using thechain rule:
@f@xi
¼ @x@xi
@f@x
þ @Z@xi
@f@Z
þ @x@xi
@f@x
¼ xxi@f@x
þ Zxi@f@Z
þ zxi@f@zð19Þ
where terms like xxi are the contravariant metrics of the
transformation, relating the computational coordinates (x, Z, z)and the physical coordinates xi (x1¼ x, x2¼ y, x3¼ z), andf is ageneric variable (u, v, w, k, e or T). After some mathematical
manipulation, the transport equation for a generic variable f is(see Lopes et al. 1995 for additional information):
@
@xjJrUjf� �
¼ @
@xjJ G gij
@f@xj
� �� �þ b ð20Þ
where Uj is a generic contravariant velocity, given by:
U1 ¼ xxuþ xyvþ xzw; U2 ¼ Zxuþ Zyvþ Zzw;
U3 ¼ zxuþ zyvþ zzwð21Þ
and J is the Jacobian of the transformation:
J ¼xx xZ xz
yx yZ yz
zx zZ zz
ð22Þ
where the quantities in the determinant are the co-variant
metrics, which are a function of the contravariant metrics. Inturn, b includes all terms not fitting in the diffusion term (crossderivatives, momentum sources due to temperature, pressure
gradients and other terms, depending on the particular concre-tisation for f). The generalised coordinate form of the massconservation equation is:
@
@xjJrUj
� �¼ 0 ð23Þ
Owing to fact that the mesh is Cartesian in the (x, y) plane and
the vertical mesh lines are straight, some of the metrics are null(e.g. xy, xz, xz). This allows some simplifications in the final formof the generalised equations, with benefits in terms of computa-
tional time and memory requirements. After discretisation andintegration, the equations are cast in the general algebraic form:
aPfP ¼Xnb
anbfnb þ b ð24Þ
This equation relates the value of the generic variable f(velocity components, turbulence quantities and temperature) at
location P with its neighbour (nb) values. Coefficients aP andanb account for diffusion and advection fluxes.
(a) (b)
10.
Height(m)
52.94.136.178.220.262.304.346.388.430.472.514.556.598.640.682.724.766.808.850.
Fig. 2. Example of mesh for wind calculation. (a) Top view, showing the Cartesian mesh component and terrain topography coloured according to local
height (area of 22.5 � 18.5 km). (b) Lateral view for a small region from (a), showing the terrain-following lines in a resampled mesh.
Two-way coupling: model development Int. J. Wildland Fire 833
A control volume approach is adopted for the integration ofthe transport equations. The SIMPLECalgorithm (VanDoormaaland Raithby 1984), which is a modification of the original
SIMPLE algorithm proposed by Patankar (1980), is employedas a means for the segregated solution of the primitive Cartesianvelocity components (u, v, w) and pressure. To account for the
non-staggered mesh arrangement, where all variables are locatedat the control volume (CV) centres, the Rhie–Chow interpolationprocedure (Rhie and Chow 1983) is implemented.
The iterative process is considered to be converged when allthe normalised residuals are lower than a predefined valueRconv.
Max Ru;Rv;Rw;Rm;Rk ;Re;RTð ÞoRconv ð25Þ
The total normalised residual for the transport equations isdetermined as follows:
Rf ¼Xall
aPfP �Pnb
anbunb þ b
uaP fmax � fminð Þ
0BB@1CCA=Xall
u ð26Þ
where (fmax –fmin) quantifies the amplitude of the variablef in
the calculation domain and u is the volume of each computa-tional cell.
Fire spread modelling
Fire rate of spread
The fire behaviour model implemented in the present work isbasedonRothermel’s surface fire spreadmodel (Rothermel 1972),
mainly owing to the latter’s well-established character and versa-tility of application. This is a semi-empirical model developed tobeusedunder awide rangeof conditions.Rothermel’smodel takes
as input fuel characteristics, wind speed and direction and terrainslope, and provides as output the fire ROS along the main spreaddirection. The fundamental equation of this model expresses an
energy balancewithin a unit volume of the fuel aheadof the flame.It illustrates the concept that the ROS is governed by the ratiobetween the rate of heating of the fuel and the energy required forbringing that same fuel to ignition. The fundamental equation in
the original proposal is modified to assume a vectorial character(Lopes et al. 1995), allowing for different directions betweenwindand slope:
~R ¼ Rb~1þ~jw þ~js
�) ~R ¼
Irxf ~1þ~jw þ~js
�rbe f Qi
ð27Þ
In Eqn 27, Rb is the rate of spread with zero wind speed andzero terrain slope, Ir is the reaction intensity, which representsheat release per unit area of the flame front (Wm�2) and xf is thepropagating flux ratio, i.e. the fraction of heat release that is
responsible for fuel heating and subsequent ignition. The sym-bol~1 represents a unit vectorwith the same direction as~jw þ~js,which are both vectorial quantities. The denominator of this
equation represents the heat sink, with rb as the bulk density,which is the mass of fuel per unit volume, ef is the effectiveheating number, quantifying the ratio between the bulk density
and the mass of fuel involved in the ignition process, andQi, theso-called heat of pre-ignition, is the heat required to bring a unitmass of fuel to ignition including the energy to dehydrate thefuel. The basic rate of spread Rb is affected by the presence of
wind and terrain slope through the factors~jw and~js respectively.Both these factors act directionally as a conditioning factor on theROS. As may be seen, each contribution (wind and slope) has its
own direction, which is added to the omnidirectional contributionof the zero-slope, zero-wind, basic ROS. Wind velocity forapplication of the previous equation should be taken at a vertical
distance from the ground dependent on the midflame height.Albini and Baughman (1979) considered the average valueresulting from the integration of the logarithmic law between
the vegetation top and the flame height as follows:
Vf
V1
¼1þ 0:36 hr=hf
� �ln z1þ 0:36hr
0:13hr
� lnhf =hr þ 0:36
0:13� 1
� �ð28Þ
where Vf is the computed average wind velocity value, V1 is thewind velocity computed at the first vertical level, hf stands forflame height and hr is the vegetation height.
All quantities in Eqn 27 are computed using fuel character-istics as defined by the fuel model concept (Albini 1976;Anderson 1982). These characteristics include fuel load, fuel
particle heat content, surface-to-volume ratio or fuel bed depth.The environment characteristics are also taken into account withparameters related to fuel moisture. Most of fuel properties are
supplied for different fuel size classes and dead–live classifica-tion. Because different size classes are allowed, fuel load,
(a) (b)
Fig. 3. Fire shape predicted by two ellipse models. The black dot
represents the ignition point: (a) double-ellipse model; (b) single ellipse
model.
Burning region
Smouldering region after firefront passage
Non-burning region
Fig. 4. Representation of coarse (thicker lines) and fine meshes used in
wind and fire calculation respectively. Burning and smouldering regions are
represented by shaded areas. In this example, there are 25 fire cells to each
wind cell.
834 Int. J. Wildland Fire A. M. G. Lopes et al.
surface-to-volume ratio andmoisture content are averaged usingthe total surface area as the weighting factor, following thesuggestion of Rothermel (1972). Dead and live components are
treated separately, each being used to compute its own reactionintensity. The final value for the reaction intensity is obtained asthe sum of the corresponding values for dead and live fuels.
Apart from fire ROS, Rothermel’s model provides severaloutput quantities concerning fire behaviour, such as firelineintensity (Wm�1), flame length (m), residence time (s) and rate
of heat release, also referred as reaction intensity Ir (W m�2).This last variable describes the fire energy source for the wind–fire interaction calculation. Residence time multiplied by thefire ROS allows determination of the flame front depth (m),
which, along with the fire perimeter, defines the burning region.
When a flame extinguishes, which occurs after the local burningtime exceeds the residence time, an exponential decay isassigned to Ir, corresponding to the smouldering phase. This is
computed according to the following equation:
Ird tð Þ ¼ fdIre�t=td ð29Þ
where t represents time, Ird is the reaction intensity during thedecay phase, fd is an initial reaction intensity cut-off due to flame
extinguishment and td is a relaxation timecharacteristic of the fuel.
Fire shape
Although Rothermel’s model describes fire ROS along themaximum spread direction, it does not provide information
8.788.347.907.457.016.576.135.685.244.804.353.913.473.022.582.141.701.250.810.37
Wind speed(m s�1)
(a) (b)
(c)
(d)
34.9737.9840.9843.9946.9950.00
31.9728.9625.9622.9519.94
Temp.(�C)
Fig. 5. Velocity field 10 m above ground and burned area shape. (a) No fire situation; (b) 450 ha of burned area; (c) 1000 ha
burned area; (d) vertical plane visualisation along the fire central line depicted in (c). Domain height is 2000 m.
Two-way coupling: model development Int. J. Wildland Fire 835
about ROS along other directions. This information is, never-theless, necessary to compute fire shape. Several models existfor fire shape, mostly based on ellipses whose geometrical
parameters are given as a function of the wind speed (e.g.Green et al. 1983). The effect of topography is modelledthrough the addition of a slope-equivalent wind speed that
would produce the same effect as the actual slope on the fireROS. The ‘effective’ wind speed, to be considered on the fireshape model, is obtained through the vectorial summation of
the actual wind velocity and the equivalent wind velocity dueto slope.
The model proposed by Anderson (1983), describing fireshape through a double ellipse, as depicted in Fig. 3a, represents,
according to the authors, the best practical realistic approxim-ation for observed fire shapes under a variety of circumstances.This model, nevertheless, does not cope with zero-wind situa-
tions. In the present implementation, when the local equivalentwind speed at midflame height falls below 0.2 m s�1, the doubleellipse formulation is replaced by a single ellipse (Fig. 3b), as
described in Alexander (1985).
Fire propagation
The local model for fire ROS, together with the globalmodel for fire shape, defines the physical behaviour of the fire.As input parameters are usually space-dependent owing to
terrain topography and non-uniform fuels distribution, not tomention time dependence due to changing atmospheric condi-tions, simulations must be carried out on a step-by-step basis in
order to account for the abovementioned spatial and temporalvariations. In terms of implementation, the fire time evolutionmay be regarded either as vector- or as raster-based. In the
former approach, the application of Huygens’ principle(Richards 1990) leads to the definition of the new fire frontas the envelope of small ellipses computed locally from theprevious fire front. For the raster approach, the one adopted in
the present work, the calculation area is divided into rectangu-lar- or other-shaped cells, where conditions are assumed aslocally uniform. Following Dijkstra’s dynamic programming
algorithm (Kourtz and O’Regan 1971), for each cell along thefire front, a contagion process is computed by assigningignition instants for a predefined set of neighbour cells.
Examination of all potential ignition candidates allows theidentification of the next ignited cell – the correspondingignition instant makes the solution advance in time. Thus, the
time progression is not constant, depending on the ignitioninstant of each cell.
Interaction between wind and fire
Wind and fire calculations are carried out with differentmeshes. Each computational cell for wind calculation (wind
cell) is defined by joining together several finer cells (fire cells)where fire propagation is calculated. Owing to the differentresolution of wind and fire meshes, the fire front may occupy a
fraction of a (coarse) wind cell, as exemplified in Fig. 4. Thetotal heat release rate per unit volume SF at the cells adjacent tothe ground, to be included in the source term ST in the energyconservation Eqn 4, is computed based on the reaction intensity
given by the fire spreadmodel in each fine cell located inside thecoarse cell as follows:
SF ¼Xntn¼1
IrnAn
!=v ð30Þ
where v is the coarse cell volume, An is the surface area of eachof the total nt fine cells located inside the coarse cell and Irnrepresents the reaction intensity Ir due to the fire in each fine celln. Total heat released at each coarse cell is used to compute anew temperature distribution within the air, by solving Eqn 4,
which affects buoyancy and turbulence production or destruc-tion, through the source terms in Eqns 1, 9 and 10.
The calculation procedure consists in an alternation between
the fire spread calculation and the wind calculation. After thefire area increases by a certain amount, the new heat release ratedistribution at the ground is passed into the wind calculationmodule. After convergence of the fluid flow and energy equa-
tions, the updated wind field is used as input to continue the firespread calculation.
(a)
(b)
0
100
200
300
400
500
600
700
800
900
1000
0 50 100 150 200 250 300
Are
a (h
a)
Static
Updt 250 (ha)
Updt 150 (ha)
Updt 100 (ha)
Updt 50 (ha)
0
100
200
300
400
500
600
700
800
900
1000
0 50 100 150 200 250 300
Time (min)
Updt 50 (ha); 3x
Updt 50 (ha); 2x
Updt 50 (ha)
Fig. 6. Influence of wind update criterion on fire area as function of time:
(a) update (Updt) for constant area increase; (b) update for constant area
fraction increase.
836 Int. J. Wildland Fire A. M. G. Lopes et al.
Results
Herein, we present a sensitivity analysis of the effect of somenumerical parameters on the calculation of the interaction
between wind and fire. These include the criterion for windupdate and the resolution adopted for the wind field calculation.In the companion paper (A. M. G. Lopes, M. G. Cruz and D. X.
Viegas, unpubl. data), application of the models described to alaboratory fire and to a large-scale real fire are reported.
Simulations were carried out for a digital terrain model incentral Portugal, as depicted in Fig. 2a, encompassing an area
of 22.5� 18.5 km (41 625 ha). The topography is fairly smoothwith a few isolated hills – terrain height is between 40 and500m above sea level, with localisedmaximum inclinations of
308. The size of the fire mesh for fire simulation is 50 m in bothdirections. For fire spread, an idealised uniform fuel bed was
adopted, with a load of 0.7 kg m�2 for 1-h fuels, 0.3 kg m�2 for10-h fuels and 0.6 kg m�2 for 100-h fuels, with fuel depth
0.6 m (the same value was adopted as terrain roughness forwind calculation). Surface area-to-volume ratio is 6000 m�1
and moisture content is set constant at 10% for dead fuels and
60% for live fuels. Reaction intensity cut-off and relaxationtime (fd and td in Eqn 29) are 0.5 and 30 min respectively. Theincident wind field is from the north, with a velocity of
15 m s�1 at the boundary layer height of 500 m above surface,with 1/a ¼ 0.28 (cf. Eqn 18). This corresponds to a windvelocity of 5 m s�1 at 10 m from the ground. The verticaltemperature profile is set to the adiabatic profile, thus corre-
sponding to neutral stability conditions. The domain top islocated at an altitude of 2000 m. Along the vertical direction,15 to 30 mesh levels were employed, with the first calculation
node at a distance of 2 m from the ground. On the horizontalplane, different resolutions were tested, ranging from 750 to50 m.
Fig. 5a–c shows visualisations of the wind field 10 m aboveground and fire shape at different stages. From these figures,one can easily see the influence of topography on wind, as well
as the air in-draft towards the fire region due to the convectiveplume from the fire. This is clearly visualised in Fig. 5d, whereinformation on the air temperature field is included in a lateralview. In this representation, temperature is corrected for alti-
tude, thus corresponding to potential temperature. Computedmaximum temperatures in the fire region are,240 and 1808C atvertical distances of 5 and 10 m above ground respectively.
Results depicted in Fig. 5 were obtained using a fire areagrowth of 150 ha as wind update criterion, i.e. wind field is
Updt 50 (ha)
Updt 100 (ha)
Updt 150 (ha)
Updt 250 (ha)
Static
(b)
(a)
0
2
4
6
8
10
12
0 50 100 150 200 250 300
Tim
e (m
in)
Area (ha)
Updt 50 (ha); 3x
Updt 50 (ha); 2x
Updt 50 (ha)
Fig. 7. Influence of wind update criterion on: (a) calculation time; (b) fire size and shape 4 h after ignition. Gridlines in the graph are
1000 m apart. Red dot represents ignition location.
Table 1. Calculation domain and mesh data
Horizontal
resolution (m)
Horizontal
domain size (km)
Total number
of nodes
1250 22.5� 18.5 20� 16� 15¼ 4800
750 22.5� 18.5 31� 26� 15¼ 12 090
500 22.5� 18.5 46� 38� 15¼ 26 220
250 22.5� 18.5 91� 76� 15¼ 103 740
150 22.5� 18.5 151� 126� 20¼ 380 520
100 8.6� 9.4 87� 95� 20¼ 165 300
50 8.6� 9.4 173� 189� 20¼ 653 940
Two-way coupling: model development Int. J. Wildland Fire 837
updatedwhen the fire area increases 150 ha since the last update.
A spatial resolution of 500 m for the wind mesh with 15 verticallevels was used. The influence of these parameters will now beinvestigated.
Study on dynamic interaction update frequency
For the first set of tests, to be presented next, wind is updated
every time the fire area increases by a specific constant valuesince the last wind update. Fig. 6a shows the fire area evolutionwith time for different values of fire area increase. Theseresults are compared with the ‘static’ case (no two-way cou-
pling), for which the initial wind field, computed with no heattransfer effects, is considered at all stages of fire growth. Thefirst noticeable characteristic is that fire spread is much slower
when dynamic interaction is taken into account. This behaviourcan be anticipated, because the buoyancy induced by the firecreates an inflow wind towards the fire region, which is
opposed to fire growth direction, a clearly evident feature inFig. 5b, c. The effect of wind update frequency influences firedevelopment mainly in the early stages after fire start. As
Fig. 6a shows, as soon as thewind is first updated, the change in
the wind magnitude and direction causes the fire ROS todecrease. Further updates do not change the wind by as large anamount as the first update. This suggests that it is not that useful
to update the wind based on relatively small amounts of firearea increase – a fractional increase would be a wiser criterion,mainly in terms of computational effort. In Fig. 6b, the results
obtained using various values of fractional increase are shown,starting from the first wind update (Updt) at 50 ha. Thus, forexample, for the case ‘Updt[50ha];2�’, the wind is updated for
fire areas of 50, 100, 200, 400 and 800 ha, which leads to fivewind calculation runs, instead of 20 runs for a constant areaincrease of 50 ha. As may be seen, even for the 3� case, firegrowth does not differ markedly from the constant 50-ha sit-
uation in the simulated temporal region. This reflects, never-theless, on computational time, because wind calculation isresponsible for more than 99% of the computational effort, as
may be appreciated from Fig. 7a. This emphasises the conve-nience of using a non-constant area increase as a wind updatecriterion. Fig. 7b, in turn, represents the fire shape 4 h after
ignition, for several update modes, including the static (nointeraction) case.
Study on wind calculation resolution
The choice of the spatial resolution adopted for the wind cal-culation mesh should be done with care. Higher mesh resolu-
tions reduce the numerical discretisation errors of the equations.In the case of simulations for topography, the mesh size alsodictates how correctly terrain features are described. It is a
general rule that each geometrical feature should be meshedwith at least 10 nodes. Nevertheless, owing to the fractaldimension of terrain characteristics, this criterion is virtually
impossible to satisfy and, thus, truly mesh-independent resultsare not obtainable. In the present work, several spatial resolu-tions in the horizontal plane for the wind mesh were adopted,ranging from 50 m, which corresponds to the fire mesh resolu-
tion for fire spread calculation, up to 1250 m. For the coarserwind meshes, a larger domain was considered (depicted inFig. 2a), in order to keep boundaries sufficiently far from fire in
terms of calculation nodes. A constant number of 15 verticallevels was considered in all situations. Table 1 synthesisespertinent data for the different wind meshes. For these simula-
tions, the dynamic interaction was updated starting at 50 ha ofburned area, and using an increment factor of 2�.
Results concerning rate of fire area growth are presented in
Fig. 8a. It may be seen that a considerable dependence onhorizontal resolution exists, though, as expected and desirable, itis attenuated for finer wind meshes; in fact, it may be stated thatresults for 100- and 50-m resolution are reasonably similar.
Fig. 8b shows that computation time increases in a strong non-linear manner with mesh resolution – in the interpretation of thisgraphic, one should keep in mind that the two finer-mesh data
sets correspond to a smaller domain.Results for fire shape 4 h after ignition are depicted in Fig. 9.
Results for the three coarsest meshes show a marked depen-
dency on mesh spacing. However, fire shapes for the two finermeshes (100 and 50 m) are rather similar. Visualisation of thewind flow, framed on the whole wind calculation domain for the50-m mesh, is presented in Fig. 10. For better visualisation, a
(b)
(a)
0
100
200
300
400
500
600
700
800
900
1000
Are
a (h
a)
Time (min)
Mesh size 1250 (m)Mesh size 750 (m)Mesh size 500 (m)Mesh size 250 (m)Mesh size 150 (m)Mesh size 100 (m)Mesh size 50 (m)
0.1
1
10
100
1000
0 50 100 150 200 250 300 350
0 200 400 600 800 1000 1200 1400
Tim
e (m
in)
Mesh size (m)
Large domain
Small domain
Fig. 8. Influence of mesh resolution for wind calculation: (a) fire area as
function of time; (b) computation time.
838 Int. J. Wildland Fire A. M. G. Lopes et al.
sampling of one out of four vectors was performed in eachdirection, so the wind field is presented on a 200-m-spacedmesh. One may note the strong in-draft all around the fire
perimeter. Fig. 11a, b shows a visualisation in a vertical planeperpendicular to themainwind direction (transect represented inFig. 10), with colour contours for temperature and velocity
magnitude.As the burning region assumes a doughnut-like shape,the central part, characterised by a lower heat release rate, issubject to air down-draft. The buoyant region on the flaming
front, however, produces a rising plume, with a vertical windcomponent that is even stronger than the horizontal component,creating a large vortex surrounding the fire. In this figure, one
may identify a typical dimension of ,500 m for the buoyantplume. Taking into consideration that a mesh size of 50 m wasfound to be appropriate for this case, this is consistent with ameshing requirement of 10 nodes for typical flow features.
Dependence of the results on the number of vertical levels ismuch weaker, as may be appreciated from Fig. 12, where allresults were computed using a 100-m horizontal resolution,
keeping the first vertical level 2 m above ground.
Conclusions
We propose a two-way coupling method for fire behaviourprediction, where the buoyancy effects caused by the fire heatrelease are fully simulated. The present paper describes the
underlying models for wind field and fire spread calculation.The wind field calculation consists of a Navier–Stokes solver
(a) (b)
Mesh size 250 (m)
Mesh size 500 (m)
Mesh size 750 (m)
Mesh size 1250 (m)
Mesh size 50 (m)
Mesh size 100 (m)
Mesh size 150 (m)
Mesh size 250 (m)
Fig. 9. Fire shape 4 h after ignition. Gridlines in the graph are 1000 m apart. Red dot represents ignition
location. (a) Results for coarser meshes; (b) results for finer meshes.
20.0019.0518.1017.1516.2015.2514.3013.3512.4011.4510.509.558.607.656.705.754.803.852.901.951.00
200 m Wind speed(m s�1)
Fig. 10. Fire shape and wind field 10m above ground computed on a 50-m
grid. Framing line corresponds to calculation domain boundaries. For ease of
visualisation, wind vectors are displayed on a 200-m grid. Fire region is
coloured according to rate of heat release. White line represents the transect
used in Fig. 11.
Two-way coupling: model development Int. J. Wildland Fire 839
applied to a boundary fitted mesh. For simplicity, the mesh isCartesian in the horizontal plane and terrain-following in thevertical direction. This considerably simplifies the calculation
procedure, leading to quite modest calculation times for aNavier–Stokes solver. Fire spread is computed with Rothermel’smodel for surface fire, with a cell-based approach, where fire
progression takes place through a contagion process. Fire shapeis given by ellipse-type models. The wind field is computed on amesh made by agglomerating cells from the fire calculation
mesh. The reaction intensity given by the fire spread modelprovides the heat release rate to be used for computing buoyancyfor the wind calculation.
We conducted a sensitivity analysis studying the influence of
numerical parameters on the calculated wind field and fire shapefor a simplified scenario. It is shown that the two-way couplingis most important in the early stages of fire development, but its
importance decreases as the fire area gets larger. As the windcalculation represents themajority of the computational effort, itis advisable to rationalise the number of wind calculations. It is
shown that updating the wind field at constant values for the firearea increase does not lead to significant differences on resultswhen compared with an update based on an area fractionincrease of 2� or even 3�. Furthermore, calculation times are
reduced to 1/5�with the latter option. Studies of the influence ofwindmesh resolution clearly indicate that the number of vertical
Temp.(�C)
Wind speed(m s�1)
47.4851.4555.4359.4063.3867.35
43.5039.5335.5531.5827.6023.6319.65
22.5024.00
21.0019.5018.0016.5015.0013.5012.0010.509.00
6.007.50
500 m
600 m
(a)
(b)
Fig. 11. Vertical plane perpendicular to incident wind direction, passing though fire shape centre (transect depicted in
Fig. 10): (a) temperature field; (b) velocity magnitude field.
20 levels
15 levels
30 levels
Fig. 12. Dependence of fire shape on number of vertical levels, using a
100-m mesh.
840 Int. J. Wildland Fire A. M. G. Lopes et al.
calculation levels affects the solution less than the horizontalresolution. The classical concept of mesh-independent resultson computational fluid dynamics is difficult to apply to wind
over complex topography, owing to the fractal characteristics ofterrain shape. Results indicate that, at least for the simulatedcases, horizontal wind mesh spacing should not be larger than,
typically, 50 to 100 m.
Conflicts of interest
The authors declare no conflicts of interest.
References
Albini F (1976) Estimating wildfire behavior and effects. USDA Forest
Service, Intermountain Forest and Range Experimental Station, General
Technical Report GTR-INT-30. (Ogden, UT, USA)
Albini FA (1996) Iterative solution of the radiation transport equations
governing spread of fire in wildland fuel. Fizika Goreniya I Zvryva
32(5), 71–82.
Albini FA, Baughman RG (1979) Estimating windspeeds for predicting
wildland fire behavior. USDA Forest Service, Intermountain Forest and
Range Experimental Station, Research Paper INT-221. (Ogden, UT,
USA)
Alexander ME (1985) Estimating the length-to-breadth ratio of elliptical
forest fire patterns. In ‘Proceedings of the Eighth Conference on Fire and
Forest Meteorology’, 29 April–2 May 1985, Detroit, MI, USA,
pp. 287–304. (Society of American Foresters: Bethesda, MD, USA)
Anderson HE (1982) Aids to determining fuel models for estimating fire
behavior. USDA Forest Service, Intermountain Forest and Range
Experimental Station, General Technical Report INT-122. (Ogden,
UT, USA)
Anderson HE (1983) Predicting wind-driven fire size and shape. USDA
Forest Service, Intermountain Forest and Range Experimental Station,
Research Paper INT-305. (Ogden, UT, USA)
Andrews PL, Bevins CD, Seli RC (2008) BehavePlus fire modeling system,
version 4.0: user’s guide. USDA Forest Service, Rocky Mountain
Research Station, General Technical Report RMRS-GTR-106. (Ogden,
UT, USA).
Castro FA, Palma JMLM, Silva Lopes A (2003) Simulation of the
Askervein flow. Part 1: Reynolds averaged Navier–Stokes equations
(k – e turbulence model). Boundary-Layer Meteorology 107, 501–530.
doi:10.1023/A:1022818327584
Clark TL, Jenkins MA, Coen J, Packham D (1996a) A coupled atmo-
sphere–fire model: convective feedback on fire line dynamics. Journal
Table 2. Nomenclature
Variable Definition
ap, anb coefficients in the general discretised equation
Ac area of the coarse cell (m2)
An area of the fine cell (m2)
b generic source term
cp specific heat (J kg�1 K�1)
Bi buoyancy force per unit volume, along direction i
(Nm�3)
Cm,C1, C2, C3 constants for the turbulence model
d distance between the first node and the ground (m)
f ratio between the roughness parameter and the roughness
height
fd initial reaction intensity cut-off after flame
extinguishment
g gravity acceleration (m s�2)
G production or destruction of turbulence due to thermal
effects (kgm�1 s�3)
h0 physical roughness height (m)
hf flame height (m)
hr vegetation height (m)
Ir reaction intensity (Jm�2 s�1)
Ird reaction intensity during decay phase (Wm�2)
Irn reaction intensity in cell n (Wm�2)
J Jacobian (unitless)
k turbulence kinetic energy (m�2 s�2)
L dissipation length scale (m)
p pressure (Pa)
P1 production rate turbulence kinetic energy (kg m�1 s�3)
Pr, Prt Laminar and turbulent Prandtl number
Qi heat of pre-ignition (J kg�1)
R residual (unitless)
Rb rate of spread with zero wind speed and zero terrain slope
(m s-1)~R rate of spread (m s�1)
ST energy transfer rate with the ground (J s�1m�3)
SF heat release rate due to fire in each cell (J s�1m�3)
ST energy transfer rate with the ground (J s�1m�3)
t time (s)
td relaxation time (s)
TI turbulence intensity (unitless)
u, v, w velocity components along the x, y, z coordinates (m s�1)
ui generic velocity component (m s�1)
Ui generic contravariant velocity component (m s�1)
V wind velocity (m s�1)
VN wind velocity at the top of the boundary layer (m s�1)
V* shear velocity (m s�1)
(Continued)
Table 2. (Continued)
Variable Definition
V1 wind velocity at the first vertical level (m s�1)
Vf average wind velocity (m s�1)
x, y Cartesian coordinates on the horizontal plane (m)
z vertical coordinate (m)
z0 roughness parameter (unitless)
G generic diffusion coefficient
a exponent for the power law of the boundary layer profile
(unitless)
b thermal expansion coefficient (unitless)
d boundary layer height (m)
ef effective heating number (unitless)
e rate of dissipation of turbulence kinetic energy (m2 s�3)
f generic variable
m, mt laminar and turbulent dynamic viscosity (N sm�2)
xf propagation flux ratio (unitless)
y temperature for the vertical adiabatic lapse rate (K)
r density (kgm�3)
rb bulk density (kgm�3)
sk, se Prandtl number for turbulence kinetic energy and turbu-
lence dissipation rate (unitless)
x, Z, z computational coordinates (m)
~js slope factor for fire spread (unitless)
k von Karman constant (unitless)
u volume (m3)
~jw wind factor for fire spread (unitless)
tw ground shear stress
Two-way coupling: model development Int. J. Wildland Fire 841
of Applied Meteorology 35, 875–901. doi:10.1175/1520-0450(1996)
035,0875:ACAMCF.2.0.CO;2
Clark TL, Jenkins MA, Coen J, Packham D (1996b) A coupled atmo-
sphere–fire model: role of the convective Froude number and dynamic
fingering at the fire line. International Journal of Wildland Fire 6,
177–190. doi:10.1071/WF9960177
Clark TL, Coen J, Latham D (2004) Description of a coupled atmosphere–
fire model. International Journal of Wildland Fire 13, 49–64.
doi:10.1071/WF03043
Coen JL (2013) Modeling wildland fires: a description of the Coupled
Atmosphere–Wildland Fire Environment model (CAWFE). NCAR
Technical Note NCAR/TN-500þSTR. Available at http://nldr.library.
ucar.edu/repository/collections/TECH-NOTE-000-000-000-866 [Veri-
fied 12 July 2017]
Coen JL, Riggan PJ (2014) Simulation and thermal imaging of the 2006
Esperanza Wildfire in southern California: application of a coupled
weather–wildland firemodel. International Journal ofWildland Fire 23,
755–770. doi:10.1071/WF12194
Coen JL, Cameron M, Michalakes J, Patton EG, Riggan PJ, Kara MY
(2013) WRF-Fire: coupled weather–wildland fire modeling with the
weather research and forecasting model. Journal of Applied Meteorolo-
gy and Climatology 52, 16–38. doi:10.1175/JAMC-D-12-023.1
Finney MA (1998) FARSITE: Fire Area Simulator – model development
and evaluation. USDA Forest Service, Rocky Mountain Station,
Research Paper RMRS-RP-4. (Ogden, UT, USA)
Flassak T, Moussiopoulos N (1988) Direct solution of the Helmholtz
equation using Fourier analysis on scalar and vector computers. Envi-
ronmental Software 3, 12–16. doi:10.1016/0266-9838(88)90004-4
Forthofer J, Shannon K, Butler B (2009) Simulating diurnally driven slope
winds with WindNinja. In ‘Proceedings of 8th Symposium on Fire and
Forest Meteorological Society’, 13–15 October 2009, Kalispell, MT,
USA. (Eds TJ Brown, BE Potter) (American Meteorological Society:
Boston, MA, USA)
Forthofer JM, Forthofer AB, Butler BW, Wagenbrenner NS (2014) A
comparison of three approaches for simulating fine-scale surface winds
in support of wildland fire management. Part I. Model formulation and
comparison against measurements. International Journal of Wildland
Fire 23(7), 969–981. doi:10.1071/WF12089
Green DG, Gill AM, Noble IR (1983) Fire shapes and the adequacy of fire-
spread models. Ecological Modelling 20, 33–45. doi:10.1016/0304-
3800(83)90030-3
Grishin AM (1994) ‘Forest Fire Physics’. (Publishing House of TSU:
Tomsk, Russia)
Henkes RAWM, van der Flugt FF, Hoogendoorn CJ (1991) Natural
convection flow in a square cavity calculated with low-Reynolds-
number turbulence models. International Journal of Heat and Mass
Transfer 34, 377–388. doi:10.1016/0017-9310(91)90258-G
Kourtz PH, O’Regan WG (1971) A model for a small forest fire ... to
simulate burned and burning areas for use in a detection model. Forest
Science 17(2), 163–169.
Launder BE, Spalding DB (1972) ‘Mathematical Models of Turbulence.’
(Academic Press: New York, NY, USA)
Launder BE, Spalding DB (1974) The numerical computation of turbulent
flows. Computer Methods in Applied Mechanics and Engineering 3,
269–289. doi:10.1016/0045-7825(74)90029-2
Linn R, Cunningham P (2005) Numerical simulations of grass fires using a
coupled atmosphere–fire model: basic fire behavior and dependence on
wind speed. Journal of Geophysical Research 110, D13107.
doi:10.1029/2004JD005597
Linn R, Reisner J, Colman J, Winterkamp J (2002) Studying wildfire
behavior using FIRETEC. International Journal of Wildland Fire 11,
233–246. doi:10.1071/WF02007
Lopes AMG (2003) WindStation – a software for the simulation of
atmospheric flows over complex topography. Environmental Modelling
& Software 18(1), 81–96. doi:10.1016/S1364-8152(02)00024-5
Lopes AMG, Sousa ACM, Viegas DX (1995) Numerical simulation
of turbulent flow and fire propagation in complex topography. Numeri-
cal Heat Transfer Part A 27, 229–253. doi:10.1080/104077895
08913698
Lopes AMG, Cruz MG, Viegas DX (2002) FireStation – an integrated
software system for the numerical simulation of wind field and fire
spread on complex topography. Environmental Modelling & Software
17(3), 269–285. doi:10.1016/S1364-8152(01)00072-X
Marino E, Dupuy JL, Pimont F, Guijarro M, Hernando C, Linn R (2012)
Fuel bulk density and fuel moisture content effects on fire rate of spread:
a comparison between FIRETEC model predictions and experimental
results in shrub fuels. Journal of Fire Sciences 30(4), 277–299.
doi:10.1177/0734904111434286
Markatos NC, Malin MR, Cox G (1982) Mathematical modelling of
buoyancy-induced smoke flow in enclosures. International Journal of
Heat and Mass Transfer 25(1), 63–75. doi:10.1016/0017-9310(82)
90235-6
Mell W, Jenkins MA, Gould J, Cheney P (2007) A physics based approach
to modeling grassland fires. International Journal of Wildland Fire 16,
1–22. doi:10.1071/WF06002
Panofsky HA, Dutton JA (1984) ‘Atmospheric Turbulence Models and
Methods for Engineering Applications.’ (Wiley: New York, NY, USA)
Patankar SV (1980) ‘Numerical Heat Transfer and Fluid Flow.’ (Hemi-
sphere Publishing Corporation: Washington, DC, USA)
Plate EJ (1971) ‘Aerodynamic Characteristics of Atmospheric Boundary
Layers.’ (US Atomic Energy Commission: USA, National Technical
Information Service, US Department of Commerce: Springfield,
VA, USA)
Plate EJ (1982) ‘Engineering Meteorology.’ (Elsevier Scientific Publishing
Company: New York, NY, USA)
RaithbyGD, StubleyGD (1987) The Askervein Hill Project: a finite control
volume prediction of three-dimensional flows over the hill. Boundary-
Layer Meteorology 39, 247–267. doi:10.1007/BF00116121
Rhie CM, Chow WL (1983) Numerical study of the turbulent flow past an
airfoil with trailing edge separation. AIAA Journal 21(11), 1525–1532.
doi:10.2514/3.8284
Richards GD (1990) An elliptical growth model of forest fire fronts and its
numerical solution. International Journal for Numerical Methods in
Engineering 30, 1163–1179. doi:10.1002/NME.1620300606
Ross DG, Smith IN, Manins PC, Fox DG (1988) Diagnostic wind field
modeling for complex terrain: model development and testing. Journal
of Applied Meteorology 27, 785–796. doi:10.1175/1520-0450(1988)
027,0785:DWFMFC.2.0.CO;2
Rothermel RC (1972) A mathematical model for predicting fire spread in
wildland fuels. USDA Forest Service, Intermountain Forest and Range
Experiment Station, Research Paper INT-115 (Ogden, UT, USA).
Rothermel RC (1983) How to predict the spread and intensity of forest and
range fires. USDAForest Service, Intermountain Forest andRange Experi-
ment Station General Technical Report INT-143. (Ogden, UT, USA).
Sero-Guillaume O, Margerit J (2002) Modelling forest fires. Part I: A
complete set of equations derived by extended irreversible thermody-
namics. International Journal of Heat and Mass Transfer 45,
1705–1722. doi:10.1016/S0017-9310(01)00248-4
Sullivan AL (2009a)Wildland surface fire spreadmodelling, 1990–2007. 1.
Physical and quasi-physical models. International Journal of Wildland
Fire 18(4), 349–368. doi:10.1071/WF06143
Sullivan AL (2009b)Wildland surface fire spreadmodelling, 1990–2007. 3.
Simulation andmathematical analoguemodels. International Journal of
Wildland Fire 18(4), 387–403. doi:10.1071/WF06144
842 Int. J. Wildland Fire A. M. G. Lopes et al.
Sullivan AL (2009c) Wildland surface fire spreadmodelling, 1990–2007. 2.
Empirical and quasi-empirical models. International Journal of Wild-
land Fire 18, 369–386. doi:10.1071/WF06142
Van Doormaal JP, Raithby GD (1984) Enhancements of the simple method
for predicting incompressible fluid flows. Numerical Heat Transfer 7,
147–163. doi:10.1080/01495728408961817
Viegas DX (2004a) A mathematical model for forest fires blow-up.
Combustion Science and Technology 177(1), 27–51. doi:10.1080/
00102200590883624
Viegas DX (2004b) Slope and wind effects on fire propagation.
International Journal of Wildland Fire 13(2), 143–156. doi:10.1071/
WF03046
Viegas DX (2006) Parametric study of an eruptive fire behaviour model
International Journal of Wildland Fire 15(2), 169–177. doi:10.1071/
WF05050
Wagenbrenner NS, Forthofer JM, Lamb BK, Shannon KS, Butler BW
(2016) Downscaling surface wind predictions from numerical weather
prediction models in complex terrain with WindNinja. Atmospheric
Chemistry and Physics 16, 5229–5241. doi:10.5194/ACP-16-5229-2016
White FM (2011) ‘Fluid Mechanics’, 7th edn. (McGraw-Hill Book Compa-
ny: New York, NY, USA)
www.publish.csiro.au/journals/ijwf
Two-way coupling: model development Int. J. Wildland Fire 843