Project no. FP6-018505 Project Acronym FIRE PARADOX Project Title FIRE PARADOX: An Innovative Approach of Integrated Wildland Fire Management Regulating the Wildfire Problem by the Wise Use of Fire: Solving the Fire Paradox Instrument Integrated Project (IP) Thematic Priority Sustainable development, global change and ecosystems D 6.3-5 Requirements for WAsP adaptation Due date of deliverable: Month 18 Actual submission date: 31/03/2008 Start date of project: 1st March 2006 Duration: 48months Organisation name of lead contractor for this deliverable: P20 UNINA Revision (1000) Project co-funded by the European Commission within the Sixth Framework Programme (2002-2006) Dissemination Level PU Public PP Restricted to other programme participants (including the Commission Services) RE Restricted to a group specified by the consortium (including the Commission Services) CO Confidential, only for members of the consortium (including the Commission Services) X Authors: Duncan Heathfield, World in a Box Finland OY, Finland (sub-contracting P20: UNINA) Morten Nielsen, Risø National Laboratory, Denmark (sub-contracting P20: UNINA) Reference: Heathfield D., Nielsen M., 2008. Wind models for fire simulation modelling in Fire Paradox. Deliverable D6.3-5 of the Integrated project “Fire Paradox”, Project no. FP6018505, European Commission, 40 p. Executive summary: WAsP Engineering (WEng) offers a fast, basic wind flow model which can be used to provide most of the wind field predictions needed by the other parts of the Paradox fire model as currently envisaged. If Paradox wishes to provide ‘a European FARSITE’, then WEng is a good starting point. The FARSITE group is implementing a straightforward flow model (WindWizard) to generate input wind fields for their spread simulation, with no feedback from the fire to the wind. WEng has known limitations in complex terrain. A more advanced type of wind model might be necessary if these limitations are revealed to be significant. We can almost certainly provide such a model and drop it in to the Paradox system as a replacement if needed. In the meantime, we should devote some effort to evaluating the shortcomings of WEng in this respect, and comparing its output against other, more complex, flow models. If the simulations need to be run for low wind speed conditions, in which diurnal thermally-driven breezes are relatively important contributors to the predicted wind field, then we can use WEng in conjunction with another model to capture and predict these effects. This has been done before at Risø. So WEng seems an adequate first choice: one which can be extended and consolidated (even replaced) if necessary. Most of this work would fall within our existing experience and expertise. If the Paradox effort must provide a complete European alternative to the ongoing coupled fire-atmosphere modelling work at NCAR, then WEng cannot serve us. In that case, we must begin working with an atmosphere model capable of being coupled, which for us would represent a new research area rather than a re-application of existing technology. D6.3-5-18-1000 Page 2 of 47 WEng cannot be used to model the two-way interaction between the fire and atmosphere (a so-called coupled model), since it is simply not designed to work with intense thermal updrafts. However, we think that we could develop a simpler, implicit treatment of the thermal effects. This would be applied as corrections to the fire-free wind field vectors. We imagine that it could allow us to emulate satisfactorily some important phenomena which coupled fire-atmosphere models have been able to exhibit, but without the complexity and performance costs which have prevented coupled models becoming operational tools. D6.3-5-18-1000 Page 3 of 47 Table of Contents Requirements for WAsP adaptation ........................................................................... 6 Wind models for fire simulation modelling in Fire Paradox ........................................... 6 Context ............................................................................................................... 6 Who are we, and what is our role? ..................................................................... 6 Intended distribution ........................................................................................ 6 Summary............................................................................................................. 6 Contents ............................................................................................................. 4 Terms of reference ............................................................................................... 8 Criteria for model evaluation ................................................................................. 9 A summary list of the models considered ............................................................. 10 Winds and wind modelling .................................................................................. 11 Wind data – speed, direction, time and place .................................................... 11 Variation in the wind ....................................................................................... 11 Wind model terms .......................................................................................... 13 Wind model types ........................................................................................... 15 Wind and fire ..................................................................................................... 16 Wind phenomena relevant to fire events........................................................... 17 Wind variation over the hours of a fire ............................................................. 18 Flow within, over and around vegetation canopies ............................................. 19 Upstream vegetation and fire damage .............................................................. 19 Smoke and pollution dispersion........................................................................ 20 Fire’s thermal effects on the wind .................................................................... 20 Blowups ......................................................................................................... 21 Wind models and fire models .............................................................................. 21 Wind in existing fire models ............................................................................. 21 Coupled fire-atmosphere modelling .................................................................. 25 Models from the wind energy field ....................................................................... 28 WAsP and WEng............................................................................................. 28 Alternatives to WEng ...................................................................................... 33 Mesoscale models ........................................................................................... 34 A wind model for Paradox ................................................................................... 37 D6.3-5-18-1000 Page 4 of 47 Exploring the wind model’s requirements .......................................................... 37 Matching requirements and capabilities ............................................................ 38 How does WEng match the requirements? ........................................................ 38 Do we need a coupled model? ......................................................................... 40 How to proceed - recommendations .................................................................... 43 Appendices ........................................................................................................ 44 Appendix 1: Notes on technical implementation ................................................ 44 Appendix 2: Some remarks about accessibility .................................................. 45 Appendix 3: Notes on other work ..................................................................... 46 Appendix 4: The RimPuff model ....................................................................... 47 . D6.3-5-18-1000 Page 5 of 47 Requirements for WAsP adaptation Wind models for fire simulation modelling in Fire Paradox Context Work package 6.2 of the EU project Fire Paradox will lead to the development of some fire simulation software. This will be a “completely spatial fire growth tool”, which is “likely to create a European alternative [to] the North-American software FARSITE.” According to the design outline, the software eventually produced will include an integrated wind model. This will interact - directly or indirectly - with a high-resolution 3D fire simulation model. This document considers the likely requirements which the wind model will need to satisfy, suggests a model to use, and explains the choice. Who are we, and what is our role? Duncan Heathfield is a software developer with some experience of Mediterranean ecological modelling. For the last ten years, he has mostly been involved in the design and programming work around the WAsP family of wind energy modelling tools. Morten Nielsen is a scientist at the Risø National Laboratory Wind Energy Department, the makers of WAsP. He has been involved in research into a variety of wind phenomena, including atmospheric turbulence, boundary-layer meteorology, concentration fluctuations, dense gas dispersion, air flow in complex terrain, and wind energy resources. We have been sub-contracted by the Università degli Studi di Napoli Federico II (UNINA) team of the Fire Paradox project. Our task is to contribute a wind model to the eventual software, if we are confident that we can provide an appropriate and useful one. The first part of our task consists in the production of this document: an assessment of whether the model which we have in mind is adequate. Intended distribution This document is intended for Fire Paradox project internal use only. Summary WAsP Engineering (WEng) offers a fast, basic wind flow model which can be used to provide most of the wind field predictions needed by the other parts of the Paradox fire model as currently envisaged. If Paradox wishes to provide ‘a European FARSITE’, then WEng is a good starting point. The FARSITE group is implementing a straightforward flow model (WindWizard) to D 6.3-5-01000 Page 6 of 47 generate input wind fields for their spread simulation, with no feedback from the fire to the wind. WEng has known limitations in complex terrain. A more advanced type of wind model might be necessary if these limitations are revealed to be significant. We can almost certainly provide such a model and drop it in to the Paradox system as a replacement if needed. In the meantime, we should devote some effort to evaluating the shortcomings of WEng in this respect, and comparing its output against other, more complex, flow models. If the simulations need to be run for low wind speed conditions, in which diurnal thermally-driven breezes are relatively important contributors to the predicted wind field, then we can use WEng in conjunction with another model to capture and predict these effects. This has been done before at Risø. So WEng seems an adequate first choice: one which can be extended and consolidated (even replaced) if necessary. Most of this work would fall within our existing experience and expertise. If the Paradox effort must provide a complete European alternative to the ongoing coupled fire-atmosphere modelling work at NCAR, then WEng cannot serve us. In that case, we must begin working with an atmosphere model capable of being coupled, which for us would represent a new research area rather than a re-application of existing technology. WEng cannot be used to model the two-way interaction between the fire and atmosphere (a so-called coupled model), since it is simply not designed to work with intense thermal updrafts. However, we think that we could develop a simpler, implicit treatment of the thermal effects. This would be applied as corrections to the fire-free wind field vectors. We imagine that it could allow us to emulate satisfactorily some important phenomena which coupled fire-atmosphere models have been able to exhibit, but without the complexity and performance costs which have prevented coupled models becoming operational tools. D 6.3-5-01000 Page 7 of 47 Terms of reference A stated goal for the Fire Paradox simulation software is the inclusion of an integrated wind model. The WAsP Engineering program from Risø National Laboratory is mentioned explicitly in Chapter 8 of the detailed implementation plan as a likely candidate. That decision is contingent upon our making a proper assessment of alternatives, and that’s the purpose of this document. Our aim is to check whether WAsP Engineering (hereafter “WEng”) is an appropriate choice, to examine some alternatives and - if possible - to justify the decision. Our consideration of other models is mostly constrained to ones with which we and our colleagues are already familiar: the purpose here is not to perform an exhaustive survey of every wind model ever developed. This document should demonstrate that we have systematically considered the requirements of the project and ensured that the software which we plan to use will satisfy them. A secondary aim of the document is that it should reflect out role in the project: providing some kind of scientific ‘bridge’ between the intellectual worlds of southernEuropean wild fire modelling and that of Danish wind energy modelling. To that end, we have mixed in some general remarks and explanations to clarify some of the terms and concepts used. D 6.3-5-01000 Page 8 of 47 Criteria for model evaluation In evaluating models, it’s important to distinguish three types of criteria Capability: does the model output tell us what we want to know? Performance: does the model give good results in a useful time? Accessibility: how convenient is it for us to use the model? A model can be ill-suited to a particular application because it simply doesn’t provide the kind of results we need, given the input data which we can provide. That’s a capability mismatch. Alternatively, a model could be rejected because it runs too slowly, or its predictions are of inadequate quality. Performance inadequacies may be overcome by improving the algorithms, by modelling in more detail or by simply using faster computers, but a capability inadequacy is an architectural issue which requires redesign. We can use the capability criterion to identify candidate models, and then choose among them using performance requirements. Accessibility reflects the operational considerations which arise when trying actually to use the model. These consist in the intellectual, technical, financial and legal obstacles to integration. If two models are otherwise similar, then one which is familiar will quite sensibly be favoured over the other. The programming language and required platform may also constrain the use of a particular model which is otherwise a perfect match. Then again, it might be simply too expensive or even secret. Accessibility problems can usually be overcome by expending more programmer time and/or money. The question of input data requirements is difficult to categorise in this scheme. Onerous requirements may be considered as reducing a model’s capability, performance and accessibility. When planning to integrate a model with others into a unified system, though, the data input requirements become a matter of capability: if a model requires data which just aren’t available from elsewhere in the system, then it cannot work for us. If WAsP Engineering is indeed the best fit for Fire Paradox, then it needs to satisfy all three criteria. It’s immediately obvious that it enjoys a tremendous advantage in terms of accessibility: since we actually designed and developed the software, and we continue to work with the scientists who know the algorithms. D 6.3-5-01000 Page 9 of 47 A summary list of the models considered This document does not have references. In case the reader is interested in any of the models mentioned, here is a list with a useful starting point for finding out more about them. Flammap http://fire.org/index.php?option=content&task=category&sectionid=2&id=9 &Itemid=30 Prometheus http://www.firegrowthmodel.com/index.cfm Firestation http://www2.dem.uc.pt/antonio.gameiro/ficheiros/artigos/FStation.pdf WindStation http://www2.dem.uc.pt/antonio.gameiro/ficheiros/artigos/WindStation.pdf Geofogo http://geofogo.igeo.pt/ WAsP www.wasp.dk WEng www.waspengineering.dk AIRFIRE http://www.publish.csiro.au/?paper=WF02047 CAWFE http://www.mmm.ucar.edu/research/wildfire/afm/afm.html MEMO http://en.wikipedia.org/wiki/MEMO_model COAMPS http://www.nrlmry.navy.mil/coamps-web/web/home WRF http://www.wrf-model.org/index.php Fluent http://www.fluent.com/software/fluent/index.htm Wind Wizard http://www.firelab.org/index.php?option=com_content&task=view&id=274& Itemid=147 WindSim http://www.windsim.com/ Raptor http://www.windlabsystems.com/raptornl.htm Ventos http://paginas.fe.up.pt/ventos/ FARSITE http://www.firemodels.org/content/view/112/143/ CALMET http://www.src.com/calpuff/calpuff1.htm D 6.3-5-01000 Page 10 of 47 Winds and wind modelling Wind data – speed, direction, time and place Wind is – trivially - the movement of air. At any one place and time, the air will have velocity: a vector. We typically characterise this vector information separately as a speed and direction. For example “wind at 12 metres per second, from 34 degrees”. Note that we thereby disregard the vertical component of the direction vector. There are mechanical and anthropological reasons for this. Mechanically, the most common way of collecting wind data is to use separate devices for speed (a cup anemometer) and direction (a wind vane). Wind vanes don’t usually measure the vertical component. Other devices old and new (such as wind socks and sonic anemometers) do capture three dimensions of the vector information. Humans are also interested in direction and speed for different reasons. The general direction of the wind will usually be associated with other weather conditions, whereas the direction-independent speed of the wind will be immediately relevant to various effects (for example: chilling, drying, damage or kite lift). A wind has a time, and a place and a vector. These can be more-or-less instantaneous, as from a point measurement, or vaguer, covering a wider area and/or a longer period of time. Data measurement campaigns commonly record the mean wind speed and direction at some interval. These mean values are usually the averages, measured over a ten-minute period prior to the recording. We typically have much more information about variation over time than space, again because of mechanical constraints: we tend to leave a few instruments in the same place for a long time. Variation in the wind Wind varies over time and space. It’s this variation which wind models are designed to predict or capture in some way. The variation is usually discussed at several different characteristic temporal and spatial resolutions (scales). It is customary to describe the time variations as either turbulence or changing weather. D 6.3-5-01000 Page 11 of 47 Turbulence: short-term, local variation Very quick and local variations are referred to as small-scale turbulence (gusting for speed). A typical time scale would be a several seconds. These may have great significance for biological and physical processes, but are hard to measure and hard to model: they are often characterised statistically. It has been suggested that the spectrum of variations has minimum energy at a frequency corresponding to one hour. In practice this empirical result is not always reproduced, and it is debatable from a physical viewpoint, since the time scale of cloud convection, usually regarded as turbulence, is on the order of twenty minutes, whereas wind changes during the passage of a cold front are sometimes faster. In flow models the mean flow or weather is the predictable result of the boundary conditions and turbulence is the random variation. The mechanisms for turbulence production are the velocity shear and buoyant convection. These processes determine the size of the most energetic eddies, which are the height above the ground and the height of the convective boundary layer. Large eddies break into smaller and smaller ones in a cascading process until the energy is dissipated into heat at very fine scales. The turbulence signal at a given location depends on how fast the mean flow transports the turbulent eddies past the observer, and the turbulence spectrum is generally shifted to higher frequencies in stronger wind. Turbulence statistics are predicted by boundary-layer theory. The turbulence in flat homogeneous terrain depends on wind speed, observation height, surface roughness, and atmospheric stability. Turbulence in real terrain is modified by vortex stretching in accelerated flow over topography and adjustments to patterns of surface roughness and surface heat flux. Wakes behind obstacles or steep terrain are an additional source of turbulence. Diurnal patterns of variation There are several reasons for diurnal wind patterns. Solar radiation will warm the ground and nocturnal long-wave heat radiation results in a net heat loss, especially if the sky is clear with low humidity. These processes change the turbulence level by buoyant convection and thermal stratification, respectively, and changed turbulent friction modifies the wind profile. Thus, with a constant upper wind speed above a surface layer in flat terrain, the near-surface wind will be strongest during daytime. A cold dense surface layer develops during nights with clear sky, and in complex terrain this tends to flow in down-slope directions. If the upper wind is not strong, the cool air D 6.3-5-01000 Page 12 of 47 meets in valleys where it is guided in local down-valley directions. This flow is called katabatic wind and it may continue after sunrise until the pool of cold mountain air has been emptied. Conversely, anabatic wind is driven by a heated surface layer. The land surface has a lower heat capacity than bodies of water, and in coastal areas a temperature difference develops on sunny summer afternoons. The maritime air becomes denser that the warm inland air and this drives the sea-breeze circulation, where cold air moves onshore, rises over land, reverses direction, and sinks over the sea. The significance of diurnal variations depends on the season, both because of the variable day length and sun angle and because it depends on a relatively weak synoptic wind. Monsoon winds are similar to sea breeze but have an annual cycle and are of a much bigger scale. Large-scale variations: weather The winds which we consider at the meso- and micro- scale arise in the context of larger scale weather systems, which create the synoptic conditions. Driving the weather, at the coarsest resolution, we have climate, which is regional in spatial extent and describes many years of conditions. There are patterns and models for variations in climate but for the purposes of this survey, we are not concerned with the climate and weather modelling/forecasting tools. We’ll take synoptic wind as given (context) data, and limit our scope to a spatial resolution range of metres to tens of kilometres and a temporal resolution range of seconds to days: we’ll consider small-scale turbulence and mesoscale processes. Wind model terms Flow separation means that flow trajectories leave the surface. This happens when the pressure gradient associated with general deceleration after passing an obstacle is stronger than the inertia of the near-surface flow. Separation may also occur at steep upwind slopes or just behind sharp edges. The opposite phenomenon is called flow reattachment. Correct prediction of flow separation is easier to model if the separation point is triggered by an abrupt change in the topography, rather than if it depends on gradually changing pressure over smooth hill. D 6.3-5-01000 Page 13 of 47 Recirculation zones have near-surface wind opposite to the general flow direction. The name refers to two-dimensional flow in which stream lines of recirculation zones are closed. Stream lines in three-dimensional flows recirculation zones may have a screwdriver shape or run parallel to the separation line approximately perpendicular to the general flow direction. Wind shear describes the process where an air parcel gradually is deformed by vertical or lateral differences in the wind field. Strong wind shear implies that the wind changes quickly over a short distance (vertical or horizontal). The combination of wind shear and friction produces more turbulence. Wind shear also affects smoke dispersion. Non-hydrostatic flow equations include vertical acceleration and are of local importance for flow over steep topography. Otherwise, vertical motion is driven by the divergence of horizontal movement and the vertical pressure gradient balances the force of gravity. Navier-Stokes equations describe the motion of a fluid. The momentum budget is derived from Newton’s second law taking pressure gradients and viscous friction into account. Turbulence closure is a semi-empirical sub-model used to predict local turbulence in RANS models (see below). The key concepts are turbulent kinetic energy, turbulence dissipation rate, and turbulence length scale. Second-order turbulence closure involves a Reynolds-averaged turbulence budget and provides a better mean-flow solution. Terms involving third-order correlation of the perturbations are modelled empirically. Synoptic meteorology describes global weather systems and front generation. Moisture, heat, pressure, and the Coriolis Effect determine the flow. The atmosphere is monitored by satellites and radiosondes launched by weather balloons. Observations are taken over a wide area at nearly the same time by a global network of meteorological institutes. The term ‘synoptic’ refers to disturbances on the characteristic scale of weather maps. Mesoscale meteorology describes systems of intermediate size. The following sub classes are used: Meso-alpha 200-2000 km (squall lines, tropical cyclones) Meso-beta 20-200 km (sea breezes, lake effect, snow storms) Meso-gamma 2-20 km (thunderstorm convection, complex terrain flows) D 6.3-5-01000 Page 14 of 47 Numerical mesoscale models are often run in a nested mode, where high-resolution models covering a limited area use forecasts by low-resolution models covering a wider area as boundary conditions. Microscale meteorology describes local flow depending on local terrain, land cover, and buildings. Turbulence and boundary layers are modelled. Roughness length is an aerodynamic property of the surface, not a literal measurement of any physical property. It is a parameter affecting the shape of the vertical wind profile. It is generally used to describe the vegetation’s effect, but could also represent rocks or other roughness elements. Still water and ice have a roughness length close to zero. An open forest has something like 0.5 metres. Geostrophic wind is a theoretical balance between the pressure gradient and the Coriolis force, which results from the rotation of Earth. This wind will follow contours of the pressure field in a direction, which is clockwise around low pressure systems on the southern hemisphere and anticlockwise on the northern hemisphere. In most situations the geostrophic wind is a good approximation of upper-air flow, whereas surface friction will send winds spiralling around low-pressure centres in the atmospheric boundary layer. Other non-geostrophic corrections are introduced by curvature of pressure gradients and by horizontal temperature gradients. The Askervein project was a major field study which ran the early 1980s. Data were collected along three transects running across a hill on the island of Uist in 1982 and 1983. The topography is fairly simple, and the roughness length low and uniform. It provides a reference case for models simulating air flow in non-complex terrain. Wind model types Mass-consistent models predict flow fields by spatially separated meteorological measurements. This is done by interpolation followed by a divergence-correcting algorithm. The prediction is kinetically (though not necessarily physically) correct, as momentum equations are ignored. It can be used as an initial condition for an advanced flow model or in emergency preparedness system with limited computer power and a need for swift results. Interpolation is safest in areas surrounded by reference measurements. Linear flow models predict small perturbations from a reference solution. Non-linear terms are approximated by products of reference state and perturbation and the flow equations become linear. This allows superposition of solutions and it is often possible to simplify the problem to two dimensions and solve it for individual waves in Fourier D 6.3-5-01000 Page 15 of 47 domain. With no reflection at the upper boundary the perturbation-forced undulation at the lower surface has exponential decay in the vertical direction. Computational fluid dynamics (CFD) is a general term for numerical flow predicting the velocity in a computational grid. These models are capable of predicting flow separation and recirculation zones. Reynolds averaged Navier-Stokes (RANS) equations describe the mean motion of a turbulent fluid. Velocity and pressure in the Navier-Stokes equations are decomposed into mean and random perturbations, the equations are averaged, and terms resulting from correlated perturbations are interpreted as turbulent forces. The turbulent Reynolds stress is an important contribution and it is modelled by a turbulence closure model. The direct effect of the viscous friction on the mean flow is insignificant in welldeveloped turbulent flow. Large-eddy simulation (LES) is similar to RANS except that only fine-scale turbulence contributes to the Reynolds stresses while time-dependent large-scale motions are resolved. LES is needed for unsteady flow phenomena, but simulation is computationally expensive. Detached eddy simulation (DES) is a hybrid model which automatically switches from the LES approach in the outer flow to the RANS approach close to solid boundaries. The motivation is that the turbulent length scale decreases near solid boundaries creating problems for a pure LES solver with finite grid resolution. Wind and fire In almost every note about wild fire, the importance of wind conditions is emphasised. Also acknowledged is that the local wind conditions at a given time may differ significantly from the general background wind or that observed and recorded at some other place. The temporal and spatial variation in wind often seems to be the main explanatory factor in narrative accounts of how actual fire events unfolded. This emphasis has not apparently been reflected in the research effort to date. It seems that until recently, fire simulation models took a rather simplistic approach to the wind conditions, despite the universal agreement about their complexity and importance. Indeed, it was the perception that there has been something of a wind-modelling ‘deficit’ in mainstream fire models which led to our involvement in this project. D 6.3-5-01000 Page 16 of 47 Wind phenomena relevant to fire events A crude characterisation of a fire event is that it starts, then spreads and then stops. It can stop by being actively extinguished, or by just not spreading and burning out. A finer series of stages is usually defined, but this will be adequate for our immediate purpose. Wind can interact with these stages in different ways. Wind and fire extinction In general, the significant effect of the wind on the extinguishing task is indirect, via the spread being slowed or stopped. It’s possible to imagine some ways in which wind might directly affect efforts to extinguish a fire, but these would be rather contrived (for example making waves on a water source too high for aircraft to refill by skimming, or affecting the horizontal drift of water and foam dumped from aircraft). A possible exception is the deliberate use of backfires, lighted in front of an advancing fire to stop the spread. These are sometimes lighted close enough to the main fire be drawn towards it by the thermal uplift, in which case the airflow created by the fire itself is exploited. Wind and fire ignition Humans are responsible for most fire ignitions in Europe. Wind is significant in at least two ways. Most obviously, the integration of the weather conditions leading up to a fire will affect the fuel dryness. Since wind speed varies over space, it could be expected to contribute to some corresponding spatial variation in fuel moisture. The direction of the wind is relevant here only inasmuch as it affects the spatial speed distribution. A less obvious significance sources reaching the fuel. moved somehow from the depends on the wind speed of the wind is that it may affect the likelihood of ignition A cigarette tossed carelessly from a car window must be road to the fuel beside it: the probability of this outcome and direction at the point of discard. An increase in the wind speed could result in smouldering fires being re-ignited, but it’s unclear whether this should be treated as a new ignition, or the continuation of earlier spread. D 6.3-5-01000 Page 17 of 47 Generally, wind has a minor role in fire ignition, at least in comparison to its role in the spread of fire. Wind and fire spread Another crude characterisation will allow us to consider fire spread as happening in two ways: by a moving fire front and by spotting. For each of these, wind is very significant. For spotting, the updraft from the fire and the driving wind will be responsible for lifting the fire brands up and away. Thereafter they’ll be transported way before falling again. The detachment of firebrands is presumably related to small-scale turbulence, both firedriven and ambient. The transport can be understood only by knowing the different horizontal wind speeds and directions obtaining at different heights. The deposition of brands will likely be related to the spatial pattern of wind speeds (as for snow, seeds and blown sand). The fire front is generally driven forward in the direction of the wind, at a rate related to the wind speed. A spatial model to predict fire growth therefore requires a spatially distributed prediction of the winds over the land surface. The wind speed and direction change with height. The vertical wind speed profile is usually logarithmic, but may have some irregularities and kinks, especially where upstream roughness changes are affecting the flow. The angle of a fire’s flame is presumably sensitive to the vertical speed profile. The 10 minute mean wind speed measured or predicted for a given location includes some variation in speed and direction (gustiness). Depending on the upwind conditions, two predicted mean speeds could be associated with quite different speed/direction distributions. These short term variations in direction are unlikely to have a significant effect on the overall direction of the fire spread, but they may contribute to the determination of the spread shape. Small-scale speed variation is more obviously important. Where a fire advances by jumping and consolidation, the overall rate of advance may be better predicted with reference to the small scale gusting than the mean speed alone. Wind variation over the hours of a fire Over tens of minutes, changes in the synoptic wind will affect the wind field experienced nearer the ground. Depending on the terrain shape (orography) a fairly small change in upper wind direction may result in a radical change in the wind direction and speed at the fire front. These dramatic changes in direction are among the most difficult and D 6.3-5-01000 Page 18 of 47 dangerous phenomena with which fire fighters must contend: from attacking the fire front side-on, they may suddenly find themselves directly in the path of the fire. Throughout the day, thermally-driven wind phenomena will arise and interact with the background wind conditions. These include land/sea breezes, katabatic winds and anabatic winds. The significance of these winds is inversely related to the strength of the background wind. In Europe, most problem fires are associated with fairly strong winds, so it follows that the importance of the diurnal winds may be reduced in fire simulation work. Flow within, over and around vegetation canopies At low resolution, vegetation can be represented as ’green slime’: it’s enough to characterise vegetated areas using some aggregated aerodynamic properties such as the roughness length and the displacement height. As the resolution of representation increases, or if we are interested in the wind flow just above or over the vegetatation itself, this characteristation is becomes uncomfortably crude. There are several complicating factors. In relatively uniform forest, where there is a fairly continuous canopy profile (leaf and stem density changing with height), then a corresponding complex wind profile may develop, with a sub-canopy flow quite distinct from that obtaining above the upper canopy. This is relevant to the well-known crown/ground fire phenonmena, where fire may spread along the ground or only in the tree crowns without immediately jumping to the other strata. Spatial variation in stand structure makes for notoriously complex air flow patterns. One well-studied example is the pattern of turbulence at the edge of a forest. Another is the wind regime which develops in forest clearings. Yet further complications are provided by the dynamic nature of vegetation. Deciduous canopies change density with season, which affects their aerodynamic properties. All canopies change shape when exposed to strong winds (generally reducing their drag.) Upstream vegetation and fire damage The wind speed and turbulence at a site will be significantly affected by the aerodynamic roughness of the surface over which it has arrived. We can expect that wind ‘driving’ a fire will mostly be arriving over just-burned ground. The roughness length of the burned area will be different from the unburned fuel area. Generally, the roughness would be expected to be lower, but it’s possible to contrive some imagined circumstances where the reverse is true. What’s more, the ground may be still hot from the fire itself, or hot D 6.3-5-01000 Page 19 of 47 because ash and soot has lowered the albedo. This extra hotness will be patchy, and will complicate the wind field further, with a patchwork of varying buoyancy. This is one way in which the fire itself can affect the wind which propels it. Smoke and pollution dispersion Smoke and pollutants can be carried high and far, affecting aircraft and causing environmental health problems. The distribution and deposition pattern is a product of the prevailing local and regional winds. Fire’s thermal effects on the wind A fire can create significant disruption to the ambient air flow. The significance of fireinduced wind phenomena is generally inversely related to the strength of the ambient wind. If a fire becomes large, it may create its own wind: a convection column caused by the intense thermal uplift will draw air in at ground level from all directions. In some circumstances the wind direction just in front of the advancing fire may be reversed relative to the ambient flow direction. The powerful uplift can also generate intense turbulence at the edges of the fire line. Such fires are described as ‘plume-dominated’, as opposed to ‘wind-driven’. The pattern of a fire’s spread is determined by which of these (the convection plume or the wind) is dominant. Sardinian fire-fighters apparently make extensive use of backfires in suppression activity. This technique depends on the fire-induced airflow to carry the backfire in the right direction. Large convection columns are apparently prone to unpredictable collapse (‘column collapse’) with consequent strong downdrafts. This may follow the fire moving over an area with less fuel. Several other dramatic and dangerous fire-wind phenomena are recognised. For example, a fire burning across the valley bottom has been reported to act as an ‘air dam’ restraining an evening’s katabatic wind for several hours. As the fire intensity lessened through the night, the dam collapsed, allowing a rapid down-valley rush of air. Another example is that eddies in the ambient air flow meeting a convection column can give rise to the development of fire whirls: vortices of flame which create locally intense winds and uplift. D 6.3-5-01000 Page 20 of 47 An alarming and much-debated phenomenon known as “area ignition” is possibly explained by a build-up of volatile gases and their sudden, simultaneous combustion. This may be linked to negative wind shear. Blowups The term ‘blowup’ is used to describe a sudden, surprising acceleration of a fire’s spread and increase in intensity. These are responsible for most fire-fighter fatalities. Viegas argues convincingly that we should describe this as ‘eruptive’ fire behaviour. These events are certainly not entirely wind-driven phenomena. Indeed, some argue that they can be explained best by considering only aspect, slope and fuel conditions. This is contentious, however, and the importance of blowups demands that we make some mention of them here. Changes in the wind conditions may help to trigger a blowup, by bringing several factors into ‘alignment’. In the famous South Canyon disaster in 1994, a weather-driven change in the direction of the upper wind, coupled with the topography combined to change the spread pattern of the fire. It eventually reached a place where the wind was blowing more strongly in a different direction. Understanding the anatomy of the South Canyon blowup seems impossible without reference to the complex wind field. Wind models and fire models Wind in existing fire models What do existing fire models do about wind? Until fairly recently, it seems that a fair answer would be ‘Not much’. There are two likely explanations for this: data availability and computational capability. Wildfires - of course - occur in wild areas, where meteorological stations are absent or rare. Even an observation or series of observations is available from one or more points in the fire area, the computational challenge of predicting a distributed wind field for the rest of the area has presumably been too intimidating. The early FARSITE software apparently worked with a simple speed and direction applied globally: unaffected by the terrain. It seems that research activity in this area has accelerated over the last decade; this is no doubt mostly due to wind flow models becoming more accessible and running more quickly on modern computers. Data availability has also improved. D 6.3-5-01000 Page 21 of 47 Let us briefly consider a few modern fire models, and examine their interaction with winds. Prometheus Prometheus is a landscape fire simulation package vaguely equivalent to FlamMap from the USDA (see below), produced by a group in Canada. It can be freely downloaded and runs on a Windows PC. This software does not have an internal wind flow model. Instead it accepts different kinds of wind inputs which must be externally provided. Various types of wind input data are accepted, from simple whole-area all-time wind to a time-series of grid maps describing speed and direction. The authors do not recommend using any particular flow model, but by opening up their program to receive ASCII grid files, they’ve made it flexible. The grids for input are in Arc-View *.asc format. The duration (in simulation time) for each grid is unambiguously specified when the grids are provided. It’s not clear how the direction data are used in the fire propagation model. There is an internal model for diurnal wind variations, but this is apparently applied only to speed. Geofogo Geofogo is an application of TGIS dedicated to modelling wildfires. (TGIS is a simulationorientated GIS developed by a team of Portuguese researchers). Like Prometheus, there is no integrated wind flow model, though there are hints that the wind speed may be adjusted by the terrain. The wind direction is uniform over the whole modelled area. A series of climate records are provided to the simulation. According to the software documentation, it’s possible to feed the model a time series of wind speed and direction maps as IDRISI *.img files like Prometheus, but the downloadable version of the software does not apparently support this. D 6.3-5-01000 Page 22 of 47 WindStation WindStation is the wind model used to generate wind fields for the FireStation landscape fire simulator program. It uses two different wind models (users can choose which one): NUATMOS and CANYON. Capability: Both models accept an input terrain map (probably a grid), and wind observations from stations in the area, and (for CANYON) a single, global, roughness length. The output is an array of wind speeds and directions which can be mapped as grids. Performance: NUATMOS is a mass-consistent model. It produces results quickly (a matter of seconds.) CANYON is a 3D Navier-Stokes solver and runs more slowly. Even with a new computer, the calculations are too slow for real-time fire simulation. Output from both programs has been compared with the Askervein experiment, and some difficult-to-model Portuguese mountains. The CANYON model is apparently better at predicting speeds in the lee of hills, and in predicting direction changes. The researchers seem generally disappointed with the output from both models, though. Accessibility: This software was produced by a team at Coimbra, led by Antonio Lopez. It was written several years ago, but is not available for download. The models are FORTRAN PC console applications, and the program GUI is implemented in a relatively obscure commercial CAD application called MicroStation. This makes it impossible simply to install and run the software. Programmatic accessibility would presumably be high if one were allowed access to FORTRAN source code for the models: they could be easily repackaged. Remarks: A Windows version of the software is apparently under development and should be available in 2008. The new version should allow a map of varying roughness lengths derived from vegetation type information. The new version will apparently include thermal effects on the air flow, but it’s not yet clear how these will be modelled. Presumably, it would be possible to compare the CANYON output with the WindWizard (see below), because both have been applied to the Askervein campaign. D 6.3-5-01000 Page 23 of 47 Wind Wizard Wind Wizard is part of the FARSITE fire modelling suite, and its role is to generate input maps of wind which can be provided to the FARSITE or FlamMap landscape fire simulator program. In addition, pre-calculated wind field maps are expected to be of value to fire commanders in the field and for planning work. Capability: The model requires a grid-format digital elevation model (DEM) and an input wind. A uniform vegetation cover type is selected (tree/shrub/grass). The output is a grid of wind speeds and directions. It seems that the height of the output grid can be specified. FlamMap expects a wind speed grid of “20 foot open wind”. Performance: The program is a user-friendly wrapper around the Fluent CFD model, which is very computationally intensive. Solving each flow field for a landscape takes in the order of an hour. The output of Wind Wizard has been compared against the standard Askervein hill data, and found to perform quite well. Accessibility: This system is currently under development by Bret Butler and his team (USDA). It seems that it will soon be available for public download at a price of perhaps USD 1K, and will run on a Windows PC. Several documents are available describing how to use the program, and the underlying FLUENT model is widely known and wellunderstood. Programmatic accessibility would presumably be high, if equivalent rights to the FLUENT engine could be negotiated. Remarks: Fluent is a well-known model with many commercial applications. It is more commonly used for small-scale aerodynamic modelling work, rather than landscape modelling, but there’s no reason to think that it’s in any way inadequate for the purpose. In emulation of the WindStation work, the Wind Wizard team have compared the output against a home-baked mass-consistent model, and used the results of each as input for some fire simulation runs. The mass-consistent model was much faster, but gave less pleasing results. This simpler model has been publicly released as ‘WindNinja’. Getting good results from a CDF model usually requires careful work by an expert. Of particular importance is the specification of the calculation domain grid structure. The Wind Wizard program is to some extent protecting the end user from some of the complexity. D 6.3-5-01000 Page 24 of 47 Only a single roughness length can be specified for the whole area. This precludes the modelling of water bodies, and roughness transitions from grass to forest, and different stand structures. Wind grids as inputs and outputs Note that Geofogo, Prometheus, FlamMap and FireStation all accept wind grids (of speed and direction) as inputs. Wind Wizard and Wind Station produce them. In fact, the output wind grid maps from one of these models could readily be converted to be input to another. Coupled fire-atmosphere modelling The models described so far treat the wind solely as an input to the fire model. It’s a one-way effect and there’s no facility for the fire itself to affect the wind flow: there are no thermal effects on the wind. In places where diurnal or water-body winds contribute significantly to the wind regime, some kind of mesoscale model would be needed to capture their effects. Mesoscale models explicitly handle heat and moisture fluxes between the air and the surface. Normal mesoscale models are not designed to handle the intense thermal updrafts such as those generated by a large fire. A more complex approach is to model the fire and wind together, allowing each to affect each other. Such models are described as ‘coupled’ fire-atmosphere models. FIRETEC & FIRESTAR High-resolution, computationally intensive models such as FIRETEC (3D) and FIRESTAR (2D) use a full physical representation of the fire process in which heat flows and thermal effects are explicitly captured. FIRETEC is coupled with HIGRAD, a NavierStokes atmospheric model. These models are not intended for representing landscape scale fires, though. It seems that they are more immediately appropriate for scientific exploration of particular features of fire phenomena: a virtual lab within which investigations can be performed that would be impossible in field tests. CAWFE A team at NCAR has developed CAWFE (Coupled Atmosphere-Wildland Fire Environment model). It is particularly dedicated to simulating large wildfire events. This model is still D 6.3-5-01000 Page 25 of 47 experimental: the intention apparently is to explore and demonstrate how such coupled models could be applied. The atmosphere is modelled using the Clark-Hall model. The fire modelling is relatively simple (it uses spread rate equations and a fuel consumption model to calculate heat release). The BEHAVE model is used for spread rate, and BURNUP for fuel combustion. Capability: The simulator takes a fuel map, an orographic map and a regional weather forecast. It runs a simulation of the progress of a fire for about 48 hours. The wind modelling is connected to the fire spread model. The initial fuel moisture is must be given, and thereafter varies according to a simple daily pattern. Performance: It seems that recent runs of the model have been fast enough to support real-time fire simulation on modestly-specified computers. Accessibility: This is a research model (albeit well-described). There’s certainly nothing to download or use. It’s not clear what the platform is, but it sounds like a UNIX system. There’s mention of a mainframe computer being used. Remarks: Papers describing this coupled model are impressive, especially in offering some potential to simulate the development of situations dangerous to fire fighters. This approach places less emphasis on modelling the details of fuel and combustion and much more on the wind, which seems to match better the way that actual fire events are described and explained. Powerful thermal updrafts and their effects on the spread pattern of the simulated fire are apparently captured. It’s not clear whether column collapse is predicted using this model, but the areas of intense turbulence at the fire head flanks are reportedly generated. Operating a model of this complexity presumably requires some considerable skill in configuring the 3-D domain grids. There is great scope for analysing the performance of this landscapescale low-resolution model against the predictions of the more detailed HIGRAD/FIRETEC models. This will presumably help to inform the transformation of CAWFE into an operational tool. The wind conditions which are provided as inputs to the fire spread calculated at some 5m behind the fireline. This is because Rothermel/BEHAVE models expect to receive wind which is unaffected by fire. The actual horizontal wind speed at the fire front in a plume dominated is zero, and CAWFE apparently predicts this. D 6.3-5-01000 are the the fire Page 26 of 47 AIRFIRE A group of Portuguese researchers at Aveiro have developed a model to predict the effect of fires on air quality, by simulating the distribution of fire pollutants. Almost incidentally (it seems) they have implemented a coupled fire-atmosphere model which is in some ways equivalent to CAWFE. AIRFIRE is a landscape-scale fire simulation model in which the atmosphere is explicitly represented in a 3-level nested grid. Fire spread is simulated using the Rothermel model with a deterministic elliptical shape tendency (like FARSITE).The atmosphere is modelled using a mesoscale model called MEMO. The highest resolution grid of the atmospheric model is centred on the fire, and provides local wind predictions for driving the fire spread. The sensible heat generated by the fire in turn affects the atmosphere. Capability: The simulator takes a simple fuel map, some orographic information and some local wind data. The wind modelling is connected to the fire spread model. The main outputs are time series maps of CO2 and CO concentrations and wind fields. Byproduct outputs are presumably maps of fire extent and surface temperature. Performance: A 50*50 MEMO simulation run can proceed at about 20 times faster than real time, on a powerful computer. Accessibility: This model runs on a LINUX cluster. The model is described but not available for download. It consists of some FORTRAN source code files. Remarks: This model’s emphasis is strongly on the air quality prediction. The fire modelling is a necessary step in this process, but is not the main objective. The fire and atmosphere models are coupled, though only through heat fluxes. The CAWFE model also includes moisture. The MEMO model is not in the public domain, but can be acquired by negotiation D 6.3-5-01000 Page 27 of 47 Models from the wind energy field Having considered some modern landscape fire simulation models, let us now consider models from the world of wind energy which have equivalent - or compatible capability. WAsP and WEng The WAsP software is produced by the Wind Energy Department (VEA) of the Risø National Laboratory in Denmark. The software consists of various programs, of which two are of interest here: WAsP and WEng (WAsP Engineering). The software’s fundamental raison d’être is to serve the needs of consultants planning the installation of electricity-generating wind turbines around the world. The WAsP program has been available since the mid 1980s, and is the industry standard for wind energy prediction work. The WAsP program itself is mostly devoted to predicting the electrical output of a wind farm: it concentrates on the mean wind climate, characterised by the Weibull probability distribution of speeds. The WEng program helps to calculate what kind of turbine can safely be installed in a given location: it concentrates on extreme wind climates, characterised by the Gumbel (double exponential) probability distribution, and turbulence. Despite being mainly designed for wind energy consultants, the programs have a wider range of possible applications. Both WAsP and WEng work by taking a wind observed in one place and predicting the simultaneous wind conditions obtaining in other places in the same landscape. Both programs do this in a different way, but the result is nevertheless logically equivalent, and results should be compatible. In WAsP, the flow model and the climatological representation are very tightly bound together, whereas in WEng, the climatological work is built atop of a clearly separate flow model. It is possible to ‘trick’ WAsP to work with particular wind conditions, but for WEng, this is the normal mode of operation. It has a wide range of possible applications: predicting snow depth, forest storm damage and (as we explore in this document) the simulation of the spread of wild fires. D 6.3-5-01000 Page 28 of 47 WEng In WEng, one can set up a modelled situation (domain) with a map describing the orography (elevation) and surface roughness (roughness length). A wind (speed & direction) observed at a given place in the map (x,y,z) or for geostrophic height can be entered, and the program predicts the speed and direction of the wind at any other point in the domain. The program can do many other things too, but this essential predictive capacity is the function most immediately relevant here. Let us repeat the Capability/Performance/Accessibility arrangement which we applied to the others. (Basic) capability: WEng takes maps of surface roughness and orography, along with an input wind (geostrophic, or regional or measured). Output includes grids of speed and direction for any number of heights. Performance: The internal flow model is LINCOM, a linear flow model. It calculates the overall flow field in a matter of seconds, and then can derive grids for each height almost immediately. There are known problems in the prediction of flow in very complex terrain. Flow separation and recirculation in the lee of hill ridges is not captured. Where slope angles exceed 25%, the predictions become unreliable. Accessibility: The program is a commercially available Windows program which can be downloaded and used if licenced (EUR 3500). The models (and their known limitations) are very well documented. We (of course) have good access to the scientists who understand the models intimately. Programmatic accessibility is high, because the model and its GUI are structurally separate. A well-documented scripting interface exposes the model functionality. Running the model without the GUI requires a special licence negotiation, though. Remarks: A significant advantage of WEng for ecological modelling work is that the roughness length is provided as an input grid map, rather than a uniform global value. This allows the representation of water bodies (over which different boundary layers will develop), and changes of vegetation type and structure. Because the domain can be quite quickly recalculated, it’s possible to provide a series of changing roughness maps and to obtain a recalculated wind field almost immediately. D 6.3-5-01000 Page 29 of 47 Other capabilities of WEng Geostrophic wind inputs One particularly interesting feature of WEng is that it offers explicit support for transforming a geostrophic height wind observation or prediction and generating wind fields at ground level. This allows the use of synoptic wind forecasts or records, even in areas for which no field observations are available. For example, the NCAR reanalysis data can be used to obtain six-hourly wind predictions for almost anywhere on earth anytime in the last 30years. Dynamic surface roughness, and turbulence WEng can provide other outputs which are tailored to the needs of wind energy engineers, but which may have ecological application. The LINCOM model offers a range of velocity derivative grids which describe the rate of change of wind vectors. The friction velocity grid is also available. The roughness length of a water body is discovered iteratively, with strong winds inducing waves and in turn increasing the ‘dynamic’ roughness of the surface. A simple obstacle model can also be used, which predicts speed deficits in the lee of rectangular obstacles of varying porosity. A turbulence model is also integrated, which can predict the turbulence spectrum for a point in the domain, and even generate a simulated three-dimensional turbulent wind field. Climate analysis The climate analysis tools which are built into WEng are of potential interest for planning work. 50-year extreme winds from various directions can be calculated, extrapolated from a representative time series of data taken from a nearby location, or from synoptic wind history. Thermal effects There is a version of WEng’s LINCOM flow model which can handle gentle thermal effects. That version is called LINCOM-T (T for thermal). LINCOM-T is an extension of the WEng flow model taking weak buoyancy effects in complex terrain into account. The buoyancy is not incorporated as a model variable but as a temperature deviation near the surface estimated by the unperturbed wind speed and heat flux. Surface cooling will result in a dense surface layer and LINCOM-T will add a local down-slope wind component. Moderation of the temperature profile by the flow is neglected and the model is only expected to be valid for weak buoyancy-driven perturbations. The buoyancy effect of fires is probably beyond this limitation. D 6.3-5-01000 Page 30 of 47 Problems and limitations of WEng Vegetation WEng represents vegetation only as a roughness length. There’s no built-in facility whereby the depth of a forest can be represented. It’s possible to change the orographic map to indicate a forest’s displacement height, but this becomes an implicit representation. There is no attempt to model any within-canopy flows, and the complexities of abrupt forest edges are not handled. The challenge of complex terrain Perhaps the most significant shortcoming of WEng as a wind model for fire simulations is the problems LINCOM has with complex terrain. This is a common feature of linear flow models, and gives rise to much discussion. As more wind generating capacity is nowadays being installed in mountainous regions, the relevance of these complex terrain effects is ever-growing. Behind a hill or ridge, the flow can become detached from the terrain surface and recirculate: the speed and direction of wind predicted by a model like LINCOM will be quite misleading. For wind energy applications, the wind speed behind a ridge close to the ground is not usually vital to the overall predictions, but for fire simulations, these can be especially important situations. A turbine is unlikely to be placed there, but the question of what happens when a fire spreads up to the top of a ridge and begins descending the next slope is very relevant. The case of a simple ridge is a particularly well-identified ‘idealised’ problem which is easy to illustrate and explain, but landscapes are in reality more complex still. It’s generally true that an area with complex orography will tend to weaken the predictive quality of a linear flow model. CFD models may also struggle, but would theoretically be expected to produce better results. In practice, the application of CFD models to wind energy prediction is by no means clearly an improvement over the simpler linear models. Several studies have shown little or no improvement of the prediction quality for annual energy production (AEP). However, we should remember that the requirements and goals of AEP predictions and fire wind field predictions are entirely different. The compensating advantages which WAsP enjoys over CFD models may be irrelevant or absent to the fire modelling application. D 6.3-5-01000 Page 31 of 47 Linear speed predictions In WEng, the predicted speed and direction for a site will always have the same correction relative to the input wind, regardless of the speed. The minor exception is where a water body affects the flow, because of the iterative wave roughness solution described above. In practice, wind will behave differently at different speeds: a linear relationship is not correct. It’s really unclear how relevant this issue is to fire simulation, though, and it is of limited importance compared with the question of complex terrain. WAsP WEng’s ‘sister’ program - the original WAsP - has different algorithms at its core, but uses fundamentally the same kind of model and has the same kinds of limitations. Capability: The program takes a vector format map with orographic and roughness data and will calculate flow corrections for any point in the landscape. These can be arranged and exported as a grid. Performance: Calculating a grid of corrections in WAsP takes considerably longer than in WEng. For a normal size grid and map, though, this is still a matter of several minutes rather than several hours. WAsP suffers the same problems as WEng in complex terrain. Accessibility: Like WEng, WAsP is a commercially available program (EUR 3K). The GUI is perfectly separable from the underlying model, which can be automated using scripting or some other client. The performance of the model is very well understood and documented. Remarks: The WAsP model may be expected to produce a more accurate set of site corrections for a precise location (because there’s no interpolation: the grid is recentred and zoomed onto the site in question). This is not a relevant advantage when calculating a grid, but could be exploited to calculate for many points along a fire line. In this case, the calculation speed would be perhaps faster than in WEng, because the entire domain of (unnecessary) results is not calculated. WAsP takes a map of roughness length, but this is a polygon-vector map, not a grid. This makes it harder (but no means impossible) to update programmatically in response to changes to the surface caused by a fire. The growing importance of complex terrain sites for wind turbines has driven a lot of research to clarify exactly where the model will fail. A measure of terrain ‘ruggedness’ has been developed and found to be a good predictor of the magnitude of error. This technique has been successfully applied as a systematic D 6.3-5-01000 Page 32 of 47 correction to compensate for the shortcomings of the core model. This is a very general method, though, and cannot yet be used to make direction-specific corrections. While there are several other wind energy software packages, the most significant European products (WindPro & WindFarmer) actually use the WAsP flow model. Upcoming WEng/WAsP developments Enhancements to WEng’s performance are planned in the 24 months ahead. Better vegetation modelling and a new flow model are being prepared. In particular, there will soon be significant improvements to the representation of forest drag. The calculations may also get faster. These changes won’t extend the model capabilities, though. A new WAsP flow model which is much more like a CFD is under development. This might provide a useful alternative with better results in complex terrain. Again, the capability is not affected by this, but the performance changes. Predictions may improve at the cost of calculation speed. This ‘insider’ information about upcoming improvements to the WAsP software doesn’t detract from the capability or performance of the other models: there may be improvements underway for those too. The information is simply a reflection of the accessibility of WEng. Alternatives to WEng WindSim A relative newcomer to the market is called WindSim. It’s marketed explicitly as a wind energy program which works well in complex terrain. Capability: WindSim takes a grid map of orography and roughness, and works out the wind field for several directions. Thereafter, it’s apparently possible to extract a grid of wind predictions for a given direction. As well as simple roughness, there is some support for the explicit representation of within-canopy air as a part of the 3-D domain. Performance: WindSim is built on a RANS CFD model called Phoenics, with the kepsilon model used for turbulence closure. This is a computationally demanding task, iteratively performed, and may not always produce an answer. The simulation time is exponentially proportional to the number of cells in the domain. Users are recommended D 6.3-5-01000 Page 33 of 47 to construct a nested set of domains which concentrate calculations on the areas of interest. It’s difficult to interpret the calculation speed, but a matter of seconds or minutes for each direction on a reasonable size grid seems to be correct. Accessibility: WindSim is a commercial product, with an academic price of some EUR 3800. It’s not clear whether it would be possible to automate the model independently of the GUI, but there’s no sign of it in the documentation. As with WindWizard/Fluent, it would perhaps involve getting the Phoenics code and working from there. Remarks: WindSim does allow spatial variation of roughness length, like WEng. The emphasis of this software is in repackaging a complex CFD for a dedicated wind energy application. The program is carefully and attractively designed, and manages to simplify the process of running a complex CFD model as far as possible. But there are limits to how simple you can make a CFD model without making it practically useless: the user still needs to understand something about the domain specification to get reasonable results. This is not a tool for casual users: the difficult configuration decisions needed are not made automatically. Compare this with the WindWizard program, where some attempt has been made to make such decisions on behalf of the user. Other non-linear models There are several other pieces of software which are equivalent to WindSim: programs which attempt to bring more advanced flow modelling to bear on the problem of wind energy prediction. The programs RaptorNL and RaptorC from Windlab in Australia, and the Ventos program from the University of Porto are two examples. Such other programs are not considered here, because there are no relevant significant differences from WindSim, and they are no more accessible. Mesoscale models CAWFE and AIRFIRE (he two coupled models described above) each used an atmosphere model, rather than just a landscape-scale wind flow model. If we were to attempt to reproduce or emulate these coupled models, we would need to use something similar. A mesoscale type model would also be needed in order to simulate diurnal winds. D 6.3-5-01000 Page 34 of 47 There are a great many mesoscale models available. It may be useful here briefly to discuss some possible candidates. KAMM Mesoscale wind models are quite widely used in wind energy investigation. At Risø there is considerable experience with the KAMM model from Karlsruhe. Results from KAMM have been integrated with WAsP in several research projects, but the programs themselves cannot be integrated because KAMM runs on UNIX only. KAMM is also very computationally intensive (a matter of hours for a given input wind). COAMPS An alternative is COAMPS, from the US Navy. The code is freely available, but again it is a UNIX program. COAMPS is a particularly likely candidate for fire model integration because it was originally designed for modelling tropical storms, where strong thermal gradients are important. We made a simple test by introducing some extremely high surface temperatures to a COAMPS simulation, and found that at least the program did not crash. It seems that there was once a plan to integrate FIRETEC/HIGRAD with COAMPS, but perhaps that initiative was overtaken by the development of CAWFE. WRF The mesoscale model which is currently enjoying the most attention is called WRF (say ‘warf’). This is a next generation version of the widely-used MM5 mesoscale model. WRF development is led by NCAR and other American institutions, but attracts contributions from all over the world. As with KAMM and COAMPS, this is another UNIX model. The code is freely available and well documented. The project is very active and we can expect lots of development work over the years ahead. The CAWFE model may be adapted to use WRF in place of the Clarke-Hall atmosphere model. CALMET Most mesoscale models run on UNIX, but a Windows model is called CALMET is available from ASG in the US. It is part of a larger package called CALPUFF: an atmospheric dispersion program. The code can be downloaded, and there are primitive graphical user interface components available too. D 6.3-5-01000 Page 35 of 47 General remarks about mesoscale models’ accessibility The source code for these models is available, and it should be possible to run them on a well-specified LINUX machine. The models are properly documented and generally understood. In these respects they are quite accessible, but they are not off-the-shelf programs which can be installed and run by end users. There are many technical obstacles to their integration with a fire spread model, even if the necessary scientific decisions about how to do it can be made. CAWFE and AIRFIRE have shown that this can be done, but the result is a custom research program, rather than anything suitable for end-user distribution. Risoe also has an in-house hybrid LES/CFD model which may be applicable. D 6.3-5-01000 Page 36 of 47 A wind model for Paradox Exploring the wind model’s requirements What are the capabilities which the Fire Paradox simulation model will demand of its wind model? At this stage, that is not finally decided, but existing documents and ongoing design discussions allow us to be fairly confident about the likely requirements. We assume that the simulation context will provide us with information about the weather: a time series of conditions either input as ground measurements somewhere in the area, or as synoptic winds. We don’t need a weather model. Basic: spatial variation of wind field The general basic requirement will be a time series of information about local wind speed and direction at points on the fire front, with the operational area in the order of kilometres in extent. This spatial variation is calculated solely from the terrain effects. A vital requirements question is whether flow separation and recirculation need to be explicitly captured. If the spread of a fire at the top of a ridge is slowed or even stopped because of lee-side ground level winds blowing counter to the prevailing wind direction, then correct fire behaviour can only be simulated by a wind field which includes the reversal. The alternative is to impose some kind of deterministic slope rule (which may be inappropriate further away from the ridge top). Advanced: allowing thermal effects on the wind field Thermal effects include valley winds and see breezes, as well as the heat flux from the fire itself. Where fire is burning under conditions of low ambient wind speed, the relative importance of diurnal winds is increased. Though we might expect most of the critical and interesting fire events to be associated with strong winds, prescribed burning operations would of course be undertaken under non-extreme conditions. Without the fire feedback effects on the wind field, it will be impossible to capture the effectiveness of backfires as a suppression technique. If - as the CAWFE authors suggest - the typical elliptical shape of the fire is itself a product of the combination of ambient D 6.3-5-01000 Page 37 of 47 wind and fire thermal uplift, then a model which fails to account for this must resort to deterministic rules to produce realistic spread patterns. Matching requirements and capabilities The current generation of operational fire simulators (FARSITE, Geofogo, FireStation, etc) aim only to satisfy the basic requirement that a terrain-influenced, spatially distributed wind field be used to drive the fire spread. Such wind fields can be provided by WindWizard, WindStation, WAsP, WEng, WindSim and the other wind models discussed. Only the non-linear models would be able to predict recirculation zones. WEng and WindSim do enjoy a clear advantage in that they allow the definition of spatially varying roughness lengths, which seems so obviously relevant to realistic fire simulations that their omission from Wind Wizard and Wind Station are hard to understand. Spatially varying roughness is important for Fire Paradox for two reasons. First: the western-European landscape is generally fragmented, giving rise to spatially more diverse roughness lengths. Second: the burning fire will change the vegetation structure dramatically. None of these models attempts to model diurnal wind variation, though. All would need to be integrated with some kind of mesoscale model to generate such effects. The integration can be quite superficial, since the information needs to flow in one direction only. The mesoscale model can run as a larger context model within which the more detailed flow simulation is conducted: it calculates modifications to the prevailing input wind. In any case it is the result of these combined simulations which is provided as input to the fire. Any of the mesoscale models discussed above (and many other available models) could serve adequately to satisfy this requirement. To capture the fire feedback effects on the atmosphere, a more profound interaction must be achieved, with information flowing both ways between the fire and the wind atmosphere. What’s more, the atmosphere model must be able to cope correctly with the extremely high temperatures generated by the fire. The Clark-Hall model used in CAWFE, and the MEMO model used in AIRFIRE have been successfully used in this role, but we don’t yet know whether WRF, COAMPS, CALMET and others would work correctly. How does WEng match the requirements? In terms of the basic capability criterion, WEng/LINCOM is as suitable as the other flow models mentioned so far (WindWizard/Fluent, WindStation/CANYON, WindSim/Phoenics). One provides a terrain map as input and specifies a wind description of some kind, and the output can be a pair of grid maps of speed and D 6.3-5-01000 Page 38 of 47 direction over some or all of the area. It’s possible to imagine feeding output from all of these models into Prometheus, for example. WEng compared with a CFD model The basic capability of WEng (making grid of wind conditions for input to fire model) is equivalent to that offered by WindWizard and the others. In various ways it’s more powerful: it has a better representation of roughness, it allows geostrophic input winds, and it produces turbulence predictions. We know, however, that WEng will fail to produce good results in very complex terrain and will not predict recirculation zones in the lee of ridges. What remains to be established is how much better the WindWizard (Fluent) model is at predicting these complex terrain phenomena, and how significant they are for the eventual modelling outcome. The linear models in WEng and WAsP are very well understood, and it’s possible to say with some certainty where they will fail. A CFD model may under-perform less transparently. The more detailed models’ calculation speeds are apparently not so slow as to make them completely useless for real-time fire simulation work, especially where the roughness length is fixed: there’s no need or value in continuous recalculation of the wind field as the fire progresses, so the wind maps can be pre-calculated. It seems, though, that the improvements which more computationally expensive and complicated CFD models ‘ought’ theoretically to provide are not reliably delivered in practice. In the meantime, WEng provides us with an accessible and very quick way of generating terrain-influenced wind fields. Should it become clear that WEng flow model is inadequate and that a full CFD model is needed, then we could quite easily make a replacement because the inputs and outputs are equivalent. There is a CFD-type model landscape flow model under development at Risoe, and that would be one possible replacement. Another possibility is to implement some empirical post-processing which corrects the linear model output to give flow reversal in the lee of hilltops. The thresholds for separation are known, so this would be better than simple guesswork. The actual circulation of air at all heights in the lee of a ridge is not relevant to the surface fire spread: it’s the wind speed and direction at flame height which is of interest. This may not be true for firebrand modelling, of course. D 6.3-5-01000 Page 39 of 47 WEng compared with CAWFE We don’t have any mechanism in which the fire could affect the wind flow modelling in WEng, except indirectly by adjusting the roughness of the surface. Even if we could adjust LINCOM-T to give us some kind of extreme heat buoyancy, this would not result in a full atmosphere model with a column of hot air rising from the fire and creating local winds. If the objective of Paradox is to produce a coupled model like CAWFE, then WEng cannot serve us. Other models such as the mesoscale LES-KAM model or COAMPS should be investigated to see how well they could work with the intense thermal effects. Do we need a coupled model? The CAWFE model is clearly a much more ambitious attempt at capturing the full story about the wind and fire. Is it necessary? If ‘sometimes’, then when? Is it relevant for our work? CAWFE workers say that ignoring the coupling makes the modelling less accurate. There is nothing available ‘of the shelf’ at Risoe which could be integrated to make a coupled fire-atmosphere model. Developing something like that would be a research project, rather than an application of existing technology. European relevance The relative significance of convection winds increases with the size of the fire, but decreases with the ambient wind speed. In Europe, problematic wildfires are usually associated with strong winds. They are usually smaller than wildfires in the US, and occur in more fragmented landscapes. If the Paradox project is interested mainly to simulate the most difficult wildfires, then a coupled fire-atmosphere model is less important. If, on the other hand, the objective is to model the progress and possible escape of prescribed burns, then a coupled model may be essential in order to generate a meaningful representation of the wind conditions during the fire. Coupled model results cannot be de-coupled A coupled model is capable only of providing a ‘complete’ simulation of an event: there’s less facility for disaggregating the result. If it’s absolutely necessary to run a fully coupled simulation to get any meaningful results, then that is entirely appropriate, but how can this be used a-priori, other than an educational tool? Various likely scenarios can be simulated, but the results are essentially narrative. D 6.3-5-01000 Page 40 of 47 For example, if the role of the model is to give predictions for the next 24-hour period of a fire in the midst of a real event, then the inevitable uncertainty about contextual weather conditions and the intrinsic complexity of the model’s interactions will require that several runs are performed. How can these be synthesised safely into reliable prognoses? If the weather forecast tells us that the wind will change direction and speed up at sometime around mid-day, then how many model runs would be needed to provide a safe exploration of the effect? If we introduce the change at 11:00, 11:30, 12:00, 12:30, is that enough? What about 12:15? In such complex simulations there may be critical thresholds lurking in the input range. An alternative is to provide less integrated model results (for example two local wind field maps for the different speeds and directions), and allow the on-site experts to complete the modelling exercise implicitly in their heads as the actual event unfolds. These remarks are not in any way a criticism of the coupled approach: these are questions about how model results can be understood, or applied by commanders during fire events. Emulating a coupled model Our primary requirements do not actually call for a full 3-dimensional representation of the atmosphere above and around the fire. For surface spread, all that’s required is a correct prediction of the wind field near the ground surface at the fire front. (As with the recirculation question, it might be a different matter for distributing firebrands.) An intriguing possibility is to attempt a short-cut. We could try to adjust the fire-free wind field vectors with some corrections according to the modelled surface temperature pattern. The thermal updrafts would be implicitly modelled and the horizontal in-drafts could be emulated as corrections, rather than flows. We imagine that hot areas could affect the wind by acting as attractors to the vectors. Their attractive effect would be cumulative and distance-proportional. This would be simple to implement and would run extremely fast. This would give a thermally-corrected wind flow model, rather than a coupled wind-fire model. It would tell us nothing about the air movement above the surface - nothing about to where the hot air rises and nothing about where any downdrafts might develop. We would also get no indication of turbulence and eddies which may arise from the thermal effects. Still, though, these corrections might be expected to give rise to the vital fire behaviour phenomena which are exhibited by the coupled models. In particular, we would expect that at the fire front, the wind velocity could be nearly zero. The deflections of wind vectors around the fire head should also generate the elliptical fire shape as in CAWFE. D 6.3-5-01000 Page 41 of 47 We have no data from which to begin making the corrections: this would not even be an empirical post-process hack. We would need to revert to physical principles and generate some initial estimates. If the model begins to behave as expected, then we could perhaps use a fully coupled fire-atmosphere model to check and refine its behaviour. D 6.3-5-01000 Page 42 of 47 How to proceed - recommendations WEng seems like a good starting point. It’s accessible and fast. It could be immediately used to generate some maps for input to FlamMap and Prometheus. We can begin the Paradox simulation work using WEng. A good architectural design for the software will insulate the actual wind model completely, allowing us to try out different flow models and explore their performance in complex terrain. When WindWizard is actually released, we should obtain a copy and compare the results. It would be interesting also to compare the output from WEng against the main WAsP program. We can also canvas opinion among scientists at Risø: explaining the Paradox requirements and asking whether they think a CFD model would yield significant improvements. We could ask for a copy of the FORTRAN code used to implement CANYON in WindStation, and have a go at integrating that too. Our speculative ideas about applying post-hoc corrections to the basic wind field vectors can be easily explored, and should quickly give us an impression about whether they are likely to lead to a useful method. In the meantime, we should watch closely the application of NCAR’s coupled model CAWFE (perhaps trying to understand why WindWizard is being developed if the coupled approach is so much more interesting and realistic). It would be very useful to open a channel of communication and co-operation with the CAWFE researchers. In case we decide to proceed eventually to a coupled fire-atmosphere model, we should proceed with some background investigations about how this could be achieved. A sensible first step would be to do have discussions with the AIRFIRE researchers in Aviero. We could conduct some simple initial technical tests with the mesoscale models which might be used, to see how difficult it would be to harness them for this purpose. D 6.3-5-01000 Page 43 of 47 Appendices Appendix 1: Notes on technical implementation The Paradox fire model software architecture can be constructed so as to allow for the possibility of substituting different wind models quite easily. The most elegant design would include a wind field intermediary layer, to which the fire model could address queries about the wind conditions. This layer could be constructed so as to allow various different flow/atmosphere models to drive its answers. The answers might be calculated on demand, or cached internally from prior runs of a slow model, but this would be entirely opaque to the fire model. This alone would be enough to facilitate the switch to using a CFD model instead of a linear flow model, and would allow various tricks and hacks to be experimented with. If the same design incorporated some mechanism for the wind model to receive some information about the fire situation, then this would permit an atmosphere model to be coupled at a later date if desired. We should try to understand the information flows between the coupled parts of NCAR’s CAWFE model in order to anticipate our own possible future requirements. The wind model requires an orographic map and some kind of spatial surface roughness information (this could be a land use category map with an interpretation table). We could also offer a map of surface temperature. Some kind of time-series of meteorological information would be needed. In return, the fire model can request wind speed and direction information about any point in 3-D space over the modelled area for any time. If the model’s combustion is driven by FIRESTAR or FIRETEC running at high resolution, the input wind data requirements are rather modest: the within-canopy flow and local terrain effects are handled by those more detailed models. A simpler fire model may require more detailed input from the wind component. The SALTUS spotting model (which is also mentioned as a integration candidate for the Paradox work) also appears to have fairly simple wind input requirements, which would be easily satisfied by this arrangement. D 6.3-5-01000 Page 44 of 47 Appendix 2: Some remarks about accessibility In reviewing just a handful of models, it became clear that there’s a huge variation in the accessibility of the different models. It’s easy to imagine a scientifically brilliant model being virtually ignored if it fails to make itself genuinely accessible. I highlight this issue in the hope that it might inform and guide our work in the Paradox project: let’s make sure our work is accessible. The differences in accessibility arise in various ways: Can the software be downloaded and run? This is not just a question of web availability, but also a matter of which computing platforms are supported and the system requirements. Once is downloaded, some supplementary questions arise: o How difficult is it to install the software and work out what to do? o Is there a ‘getting started’ document which will give immediate insight into how to use the software and what’s needed? o Are there any useful and representative sample data files provided? o How hard is it to get the model to run on one’s own data? What are the data file formats? Are they well-known and well-documented? Is the origin of the underlying model clearly indicated? Are there readily-available documents describing the model, and any special features of the implementation, and clear, honest descriptions of the limitations of its applicability and performance? Has the model been used to perform simulations of well-known cases, so as to make comparison with other models more straightforward? Does useful information about the models come up readily from a web search? Is it obvious what keywords would be best to use? Are most of the links genuinely useful, or just to abstracts of published papers which are only available for paying subscribers? D 6.3-5-01000 Do the corresponding authors actually reply to emails? Page 45 of 47 Appendix 3: Notes on other work Landscape sensitivity analysis A WEng script is being written which will explore the sensitivity of the predicted wind field to changes in wind direction. The geostrophic wind direction is gradually adjusted, and the changes in the corresponding ground-level predicted wind field are evaluated to identify places where either the speed or direction change surprisingly or dramatically. The intention is to explore the possibility of producing a map of potential “wind surprises”, or perhaps warning notes for a selected area. This idea may be extended to look for places where the wind changes direction with height above the ground surface. Making use of turbulence for fire process modelling We should present the turbulence model output of WEng to fire scientists and see whether it offers any interesting or relevant predictions, and explore how they could be used. Fire-climate time series analysis Work is ongoing to explore the relationship between wind direction and fire danger index, so as to identify which wind directions are actually associated with the risk of severe wild fires. WEng can generate wind fields corresponding to these directions, and we suspect that those ‘danger maps’ would be useful in planning: constructing fire breaks and deciding on roadside clearances, etc. D 6.3-5-01000 Page 46 of 47 Appendix 4: The RimPuff model Another model which is widely used at Risoe is ‘RimPuff’, which is an atmospheric dispersion model. It’s been applied to foot-and-mouth disease simulations and nuclear accident modelling. There is no immediate relevance for the Paradox work, but it seems appropriate to draw attention to it here, in case it’s of interest at some point in the future. About RimPuff Atmospheric dispersion models could be used to predict transport and dilution of smoke. These models predict contaminant advection by the wind, growing plume width, and decaying centre-line concentration. Plume mixing is driven by turbulent motions. Simple models rely on an analogy with molecular diffusion defining an eddy-diffusivity coefficient, and, with sufficient simplification of the wind profile, the problem can be solved analytically. This concept is valid if the mixing process is the result of many small random movements, as in the case of vertical mixing in a near-surface plume. Eddydiffusivity is less successful in predicting the lateral spread of a plume as this is influenced by a small number of large eddies sweeping the entire plume from side to side rather than diluting it. The several-hour average plume profile could be modelled in this way, but it would be wider and have lower centre-line concentration than the instantaneous plume. The dimensions of the instantaneous plume could be predicted by a reduced eddy-diffusivity coefficient based on the high-end part of the turbulence spectrum above a frequency determined by the plume dimension. Random plume movements could be simulated as a stochastic process driven by the low-end part of the velocity spectrum. The Risø mesoscale puff model (RIMPUFF) and similar relies on a divide-and-conquer strategy, where the plume is split into multiple puffs. When the local wind profile is assumed uniform, the concentration profile of a puff will be Gaussian and grow with a rate depending on the high-end part of the spectrum. When the puff grows too big for the uniform-wind approximation to be valid, it is split into smaller ones. The wind and turbulence field is modelled by an external flow model and this makes the model much more flexible than Gaussian-plume models. RIMPUFF has extra features for special pollution problems, e.g. decaying radiation from nuclear isotopes, wet or dry deposition to the ground, initial plume rise from a hot release, and reflection at the ground and inversion height. It might be possible to model the transport of firebrands, which could be considered dense particles tending to fall out of a buoyant gas flow. D 6.3-5-01000 Page 47 of 47