eseArch orest F ire r A dvAnces in F

(1)

A ^dvAnces ⁱⁿ F ^orest F ^ire

r ^eseArch

DOMINGOS XAVIER VIEGAS

EDITOR

2014

(2)

(3)

(4)

Advances in Forest Fire Research

- 2014 -

Edited by

Domingos Xavier Viegas

ADAI/CEIF, University of Coimbra, Portugal

(5)

ADAI/CEIF

Associação para o Desenvolvimento da Aerodinâmica Industrial (ADAI) Centro de Estudos sobre Incêndios Florestais (CEIF)

Rua Pedro Hispano nº12 3030 - 289 Coimbra

Telf: +351 239 708580 | Fax: +351 239 708589 Coimbra 2014

This publication may not be reproduced in whole or in part, stored in a retrieval system or transmitted, in any form or by any means without the permission of the

Publisher, ADAI.

Composition Luís Mário Ribeiro

(6)

The following mesh sizes were used: 2.5 cm, 5 cm and 10 cm called respectively grid 1, grid 2 and grid 3 in the following. The reason is that away from the fire, the mesh size can be coarse and having mesh size greater than the sizes mentioned previously can provide computational time saving without altering the resolution of the fire region for which the grid has to be finer (less than 2.8, 4.6 and 6 cm for the 0.6 to 0.9 and 1.2 kg/m² fuel loads). For the sensitivity analysis, we will consider 5 positions of interest on the bench on combustion (Figure 4): point C is at the center of the table and points A, B, D and E are positioned 0.5 m aside. The global velocity and the velocity components of the flow will be considered as a function of height above these 5 positions. The interval (Fig 4) between each point along height depends on the mesh size. The total velocity (and deviation) with height are displayed in figure 5 above positions A and C as example. The results presented are representative of simulations obtained at location B, E and F. The better agreement between grid 1 and 2 was obtained at position A while position C is representative of the worst agreement between simulations results obtained with both grids. Figure 5 shows that the values of the total velocity obtained with grid 2 are near to the simulation of reference performed with grid 1. We observe also that grid 3 fails to reproduce the tendency of the variation of the velocity with height observed for grid 1 and 2. The small differences observed between the results obtained with grid 1 and grids 2 suggest that a mesh size of 2.5 cm is sufficient to get convergence in the case of a non reacting flow. Another point that deserves mention is the comparison of inflow at the vent of the domain. The mass balance was checked for the three grids, but we observe that the inflow is very different for grid 3. In that case, a greater amount of air comes into the domain and then flow out at the vent denoting an important recirculation that was very low for grid 1 and grid 2. All these results show that a mesh size of 10 cm is not capable to accurately represent the flow in the hood and in the laboratory.

Figure 5. Time averaged velocity and standard deviation versus height in the hood for the large calculation domain (laboratory) above position A and C for the different grids

The quality of a particular simulation is most directly tied to the grid resolution. In order to measure this quality, the fraction of unresolved kinetic energy called the measure of turbulence resolution (MTR) was examined (Mc Grattan et al., 2013b). The MTR is similar to the Pope (2004) criterion.

The MTR was calculated at each point above the positions of interests as depicted in figure 4. Then the time average of MTR was calculated at each point along height. And finally the mean of these values was calculated versus height above the five positions of interest. Figure 6 displays the results of the mean time averaged MTR along height for the different grid tested. According to Mc Grattan et al. (2013b), maintaining mean values of MTR near 0.2 provides satisfactory results (simulation results within experimental error bounds) for mean velocities and species concentrations in non reacting, buoyant plumes. Hence figure 6 shows that grid 1 and grid 2 are convenient for this criterion while grid 3 is not suitable.

(14)

Figure 6. Measure of mean turbulence resolution with standard deviation for the different grids.

Size of the calculation domain and time saving

The size of the calculation domain is an important factor that affect greatly the calculation time. The greater the domain is the larger the time of simulation is. As an example, the simulation time performed per day for the flow is given in table 1 for the mesh sizes of 2.5, 5 and 10 cm (grids 1, 2 and 3).

Different domains of calculation have been studied and the simulated results have been compared as previously, attempting to reach a compromise between the precision of the results and the simulation time. Two domains are discussed hereafter. The first one is a little bit larger than the hood (50 cm larger in all direction). It is called ‘LSHR and vicinity’. The second one is limited to the size of the hood (see control volume in figure 1). It is called LSHR.

Table 1. Calculation time for the different grids

Grid 1 Grid 2 Grid 3 Grid 4

Domain Laboratory Laboratory Laboratory LSHR

Mesh size (cm) 2.5 5 10 2 and 4

Number of mesh 11 489 280 1 436 160 179 520 2 707 200

Number of processor 14 1 1 14

Calculation time (sec/day) 13 29 883 72

In order to optimize the different requirements of the grid (in the litter and in the flame), the mesh size was fixed to 2 cm in the litter, above it and below it (under the bench). A coarse mesh size of 4 cm was used elsewhere in the calculation domain since as seen previously grid 2 (5 cm size) was enough convenient for the flow. We have chosen 4 cm since it is not allowed with FDS to align meshes with size of 2 cm and 5 cm (Mc Grattan et al., 2013a). The simulation of the flow done with these domains was compared to a simulation of reference performed with the large domain (see Figure 2) for which a mesh size of 2 cm was used in the hood and 4 cm elsewhere in the domain to be coherent with the mesh of the small domains.

(15)

Figure 7. Simulation of the velocity in the hood of the LSHR: comparison between the large calculation domain and the domains limited to the hood and to the hood vicinity: a) global velocity, b) component u, c) component v, d)

component w.

Figure 7 shows the time averaged velocity versus height above position A and the time averaged of the three components of the velocity u, v and w along respectively the axis Ox, Oy and Oz as depicted in Fig 2, for the three calculation domain. We observe that for the different positions along the height the three calculation domain provide similar results for the three components of velocity. The LSHR domain is however in better agreement with the large domain than the ‘LSHR with vicinity’ domain for component w. We can note that the values of components u is almost zero along the height and simulation results obtained with both domains can be considered in agreement for u. The main difference between the small domains and the large one is visible for component v which is overestimated with the small domains. However, we can observe that the tendency of components v which increases with height is rendered by the simulation performed with the small domains. Here also, the difference between the results of both small domains is negligible. Similar results were obtained along height for points located above the positions B, C, D and E. We can conclude that both small domains are equivalent to simulate the flow by comparison with the simulations performed with the large domain.

Ignition method

Concerning the ignition method, experimental fires were ignited by using a small amount of ethanol and a flame torch to ensure a fast and linear ignition of the fuel beds. The fires were ignited along the entire width of the fuel bed by using between 4 and 6 ml of alcohol. Figure 8 displays a zoom of figure 3 to focus on the HRR just at the ignition. 4 ml were used to ignite the fire tests with fuel load of 0.6 kg/m² while 6 ml were used to ignite fire tests with higher fuel load.

(16)

0 20 40 60 80 100 120 140

40 90 140

HR R ( kW )

Time (s) w = 1.2 kg/m2

w = 0.9 kg/m2 w = 0.6 kg/m2

Duration of ethanol combustion

Figure 8 –Heat release rate at the ignition for fuel load of 0.6, 0.9 and 1.2 kg/m²

A study has been performed to investigate the duration time of ethanol burning when it is spread on the combustion bench without the vegetation. We measured this time (see Figure 8) in order to determine the mass of needles burned with ethanol during the ignition, m_needles :

needles c

ethanol c ethanol t

needles

H

H V

dt q m

ig

,

, 0











^

where tig represents the duration of ignition, Vethanol is the volume of ethanol used to ignite the vegetation and

 H

_c_,_ethanol

 21300

kJ/L is the net heat of combustion of ethanol. The mass of needles involved in the ignition has been estimated to 20 g, 32 g and 35 g respectively for the fuel load of 0.6, 0.9 and 1.2 kg/m². The mass of needles used for 0.9 and 1.2 kg/m² are very close because the same quantity of ethanol was used for both fuel loads.

WFDS offers different options to simulate an ignition source that have been tested to reproduce the experimental ignition. Two different numerical methods to ignite the fuel bed have been compared in order to assess their ability to reproduce our experimental ignition in terms of heat release rate. The first method considers a linear strip of a given width located at the edge of the combustion bench for which the HRR per unit area is fixed during a time interval corresponding to the duration of the ignition. With this method and given the mass of fuel involved in ignition a strip of one meter long and 4 cm wide was considered. Although the calculated width was slightly lower than 4 cm, this dimension was chosen to be in agreement with the mesh size (2 cm) within the fuel bed. The HRR per unit area was set to 1100 kW/m² for the fuel load of 0.9 kg/m². The strip of ignition was applied during 18 s. It should be noted that the fuel bed length was shorten of 4 cm in order to avoid superimposing the needles and the strip that can be viewed as a burner located in front of the fuel bed. The second method consists of fixing the temperature of an ignitor at 1000°C. The ignitor is 4 cm wide and one meter long. Its height is equal to the fuel bed height. The ignitor is like a set of particle with porosity equal to the porosity of the fuel bed. It is located in front of the fuel bed. The simulated results of HRR for both methods of ignition were very similar. A point that deserves mention is the necessary time required “before to ignite the numerical fire”. Even though the time for the flow to reach a quasi- stationary regime is not a problem for experimental fires since the hood is turned on a long time before ignition, it is the case for numerical fires. Indeed, the numerical ignition has to take place once a “quasi-

(17)

stationary flow regime” is established. This time is about 60 s for the LSHR domain of calculation and for a fuel load of 0.9 kg/m². Ignition before this time (that is to say during the transient flow regime) might result in a perturbation of the fire line that subsequently affects the spreading of the fire front over time (see Figs. 9a and 9b). In that case, the perturbation of the ignition is due to the flow that is not stabilized and introduces some dissymmetry before a given time that depends on the configuration (height of the fuel bed,…).

Figure 9. Influence of ignition time on the fire front shape for a fuel load of 0.9 kg/m²: a) numerical ignitions performed after stabilisation of the flow, b) numerical ignition performed before stabilisation of the flow

Conclusion and prospects

Physical model of fire spread are very promising to represent the behavior of wildland fires at a scale useful for risk assessment and fire safety concerns. The aim of our ongoing study is twofold. We propose a set of experimental data that can be used to validate physical model based on thermodynamics quantity of interest. Secondly we propose a methodology to validate such models paying attention to the influence of the experimental conditions that can affect the comparison between numerical and experimental fires. We focused on two parameters which could have a major role in the behaviour of numerical fires: the hood extraction system and the ignition method. This paper represents the first step of our work. Further studies will be devoted to the following activity:

measurements of flow velocity within the hood will be done to assess the accuracy of WFDS to represent the flow with no fire. Secondly a full study will be performed to test WFDS against the set of data obtained from experiments conducted with the calorimeter for fire spreading under both slope and no slope condition for the different fuel loads: HRR, mass burning rate, radiant heat flux and exhaust gas composition. The role of the char during the process of spreading will be particularly investigated.

Acknowledgements

Part of this work was carried out in the scope of project PROTERINA-Due supported by the EU under the Axis 3 (Natural and Cultural Ressources) of the Operational Program Italia/France Maritime 2007- 2013, contract: G25I08000120007.

(18)

References

Babrauskas, V., Peacock, R.D. (1992) HRR. The single most important variable in fire hazard. Fire Safety Journal 18, 255-292.

Dupuy J.L. (1995) Slope and fuel load effects on fire behavior: laboratory experiments in pine needles fuel beds, International Journal of Wildland Fire, 5(3) 153-164

McGrattan K., Hostikka S., McDermott R., Floyd J., Weinschenk C., Overholt K. (2013a) Fire Dynamics Simulator Technical Reference Guide, volume 1 : Mathematical Model. Technical Report. NIST Special Publication, 1018-6, National Institute of Standards and Technology, Gaithersburg, Maryland, <http://fire.nist.gov/fds>.

McGrattan K., Hostikka S., McDermott R., Floyd J., Weinschenk C., Overholt K. (2013b) Fire Dynamics Simulator User’s Guide. Technical Report NIST Special Publication, 1019-6, National Institute of Standards and Technology, Gaithersburg, Maryland, <http://fire.nist.gov/fds>.

Mell W.E., Jenkins M.A., Gould J., Cheney P. (2007) A physics-based approach to modelling grassland fires. International Journal of Wildland Fire 16, 1-22.

Mell W.E., Maranghides A., McDermott R., Manzello S. (2009) Numerical simulation and experiments of burning Douglas fir trees. Combustion and Flame 156, 2023-2041.

Parker W.J. (1982) Calculations of the heat release rate by oxygen consumption for various applications, NBSIR 81–2427–1.

Pope S.B. (2004) Ten questions concerning the large-eddy simulation of turbulent flows. New Journal of Physics, 6, 1–24.

Porterie B., Consalvi J.L., Kaiss A., Loraud J.C. (2005) Predicting wildland fire behaviour and emissions using a fine-scale physical model. Numerical Heat Transfer, Part A 47, 571–591.

Quintiere J.G. (2006) Fundamentals of Fire Phenomena. (Eds John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, England)

Santoni P.A., Morandini F., Barboni T. (2011) Determination of fireline intensity by oxygen consumption calorimetry. Journal of Thermal Analysis and Calorimetry, 104(3), 1005-1015, Tihay V., Morandini F., Santoni P.A., Pere-Ramirez Y., Barboni T. (2012) Study of the influence of

fuel load and slope on a fire spreading across a bed of pine needles by using oxygen consumption calorimetry. Journal of Physics: Conference Series, 395, 012075

Viegas D.X. (2004) On the existence of a steady state regime for slope and wind driven fires.

International Journal of Wildland Fire, 13, 101-117

(19)

Overview of the 2013 FireFlux-II Grass Fire Field Experiment

CB Clementsâ, B Davisâ, D Setoâ, J Contezacâ, A Kochanski^b, J-B Fillipi^c, N Lareauâ, B Barboni^c, B Butler^d, S Krueger^b, R Ottmarê, R Vihnanekê, W E Heilman^f, J Flynn^g, M A Jenkins^b, J Mandel^h, C Teskeⁱ, D Jimenez^d, J O’Brien^j, and B. Lefer^g

a Fire Weather Research Laboratory, Department of Meteorology and Climate Science, San José State University, San José, CA, USA. craig.clements@sjsu.edu

b Department of Atmospheric Sciences, University of Utah, UT, USA

c University of Corsica, Corsica, France

d Fire Sciences Laboratory, Missoula, MT, USA

e Pacific Northwest Wildland Fire Sciences Laboratory, WA, USA

f USFS, Northern Research Station, Lansing, MI, USA

g University of Houston, Houston, TX USA

h University of Colorado, CO, USA

I University of Montana, MT, USA

j USFS Southern Research Station, GA USA

Abstract

In order to better understand the dynamics of fire-atmosphere interactions and the role of micrometeorology on fire behaviour the FireFlux campaign was conducted in 2006 on a coastal tall-grass prairie in southeast Texas, USA. The FireFlux campaign dataset has become the international standard for evaluating coupled fire- atmosphere model systems. While FireFlux is one of the most comprehensive field campaigns to date, the dataset does have some major limitations especially the lack of sufficient measurements of fire spread and fire behaviour properties. In order to overcome this, a new, more comprehensive field experiment, called FireFlux II, was conducted on 30 January 2013. This paper will address the experimental design and preliminary results.

The experiment was designed to allow an intense head fire to burn directly through an extensive instrumentation array including fixed 42-m and three 10-m micrometeorological towers (Figure 2). The fuels consist of a mixture of native grasses.Each tower was equipped with a variety of sensors, including 3D sonic anemometers, pressure sensors, heat flux radiometers, and an array of fine-wire thermocouples to measure plume temperatures. The experiment was carried out under red flag warning conditions with strong winds of 8 m s-1 and relative humidity of approximately 24%. Instrumentation also included a scanning Doppler wind lidar, microwave temperature profiler, radiosonde balloons for upper-air soundings, a full suite of air quality instrumentation located downwind, and multiple ground and tower mounted infrared and visible video cameras. In addition, the fire spread was monitored from the air using helicopter mounted infrared and visible video cameras. Because the experiment was designed to be conducted under a north wind, the timing of the experimental period only allowed for a northwest wind. This required the instrumentation array to be moved in order to document the fire spread and was a limitation to the experiment. Preliminary results showed that the fire spread rate was ~1.5-2.5 m s-1 for the head fire while the flanks spread at 0.7 m s-1. The surface pressure field indicated that a low- pressure region formed downwind of the advancing fire front. The observations from the 42-m tower show that the strongest fire-induced winds occur a the surface in the cross-wind direction.

Keywords: Grass fire, field experiment, fire-atmosphere interactions, micrometeorology, fire-induced winds

Introduction

Studies on the fine-scale structure of fire-atmosphere interactions and fire behaviour have been based mostly on numerical simulations using coupled fire-atmosphere models (Clark et al. 1996; Mell et al.

2009; Linn and Cunningham 2005; Kochanski et al. 2013; Fillipi et al. 2013) and few field campaigns (Cheney et al. 1999; Clements et al. 2007). To better understand the dynamics of fire-atmosphere

(20)

interactions and the role of micrometeorology on fire behaviour, the FireFlux campaign was conducted in 2006 on a coastal tall-grass prairie in southeast Texas, USA (Clements et al. 2007, 2008). The FireFlux campaign dataset has become an international standard for evaluating coupled fire- atmosphere model systems. While FireFlux is one of the most comprehensive field campaigns to date, the dataset does have major limitations, especially the lack of sufficient measurements of fire spread and fire behaviour properties. In order to overcome these limitations, a new and more comprehensive field experiment, called FireFlux II, was conducted on 30 January 2013. This paper will address the experimental design and preliminary results.

Experimental Design and Instruments

The FireFlux II (FF2) field campaign was conducted at same plot as FireFlux at the University of Houston Coastal Center in Texas, USA, on 30 January 2013. The experiment was designed to allow for an intense head fire to burn directly through an extensive instrumentation array that included four meteorological towers, one fixed 42-m tower and three 10-m towers (Figure 1). The towers were equipped with a variety of sensors (Table 1), including three-dimensional sonic anemometers, pressure sensors, heat flux radiometers, and an array of fine-wire thermocouples to measure plume temperatures. Also located within the prairie were two interspersed grids of 28 surface thermocouples, buried underground, 18 pressure sensors positioned ~3.0 m above the burn plot, and 8 fire behaviour sensor packages that measured flame temperature, heat fluxes and gas velocities ~1 m AGL.

A key platform and instrumentation suite deployed during FF2 was the California State University- Mobile Atmospheric Profiling System (CSU-MAPS) (Clements and Oliphant 2014). The CSU-MAPS includes a truck mounted scanning Doppler wind lidar and a microwave profiling radiometer used to continuously measure background and plume thermodynamic and kinematic properties throughout the experimental period. The Doppler lidar provided high-resolution measurements of smoke plume aerosol backscatter intensity and radial wind velocities across the plot and around the fire front. The lidar has a range gate resolution of 18 m and the temperature profiler has a vertical resolution that scales with height, with a finer resolution of 50 m within the boundary layer. The CSU-MAPS also includes a mobile, trailer-mounted 32 m meteorological tower that is equipped with 5 levels of 2-d sonic anemometers and thermistor/hygristor sensors. In addition, a radiosonde system was used for an in-situ upper-air sounding taken just before ignition. Emissions and air chemistry were measured with a full suite of gas and particle samplers located downwind of the experimental plot (Figure 1). Multiple ground and tower mounted infrared and visible video cameras were used for measuring fire behaviour properties. In addition, the fire spread was monitored from the air using helicopter mounted infrared and visible video cameras. Table 1 provides a detailed description of meteorological instruments used during the experiment.

(21)

Figure 1. A map of the experimental setup during FireFlux II conducted on 30 Jan 2013.

Site description

The experimental plot was a natural tall-grass, of 155 acres (0.63 km²) in size. The fuels consisted of a mixture of native grasses, including big bluestem (Andropogon gerardi), little bluestem (Schizachyrium scoparium), and long spike tridens (Tridens strictus). Fuel loading calculations were made from 20 destructive sampling plots. The average fuel loading for the experimental unit was 2.88 tons acre^-1.

The experiment was to be conducted under a north wind. However, synoptic conditions, as well as the timing of the experimental period, only allowed for a northwest wind as the best scenario. The decision was made to take advantage of the lesser than ideal conditions. WRF-SFire simulations (Mandel et al.

2012, Kochanski et al. 2013) were conducted to confirm the idealized fire spread under a northwest wind and based on these simulations, the instrumentation array was reconfigured just 24 hours prior to ignition. The reconfiguration was aimed to provide the most efficient experimental configuration for capturing the expected fire spread and maximize the number of instruments in the path of the fire.

(22)

Fuel moistures

Fuel moisture calculations were made from 20 different sampling plots. Three different sample types were collected 30 minutes prior to ignition: upper level grass (UL), lower level grass (LL), and forb (a herbaceous flowering plant). The boundary between the two grass layers was assessed based on the visual center of the mass height of the grass. Table 2 describes the moisture content percentages in detail.

Preliminary Results

Synoptic environment

The ignition occurred at 15:04 (CST) local time on 30 January 2013 and was associated with a post- frontal environment. The month of January was associated with above average precipitation and weak frontal systems, limiting ideal experimental conditions that required strong post-frontal northerly winds. Due to the excessive rain, soil conditions were wet with some regions of standing water within the experimental plot. A cold frontal passage the previous night created a strong northwesterly surface flow in excess of 8 m s-1, with gusts up to 12 m s-1, as well as a relative humidity of approximately 24% at the time of ignition (Figure 2). These conditions led to a red flag warning to be issued by the National Weather Service for the day of the experiment. A radiosonde launched 40 minutes prior to the ignition shows that the daytime boundary layer associated with a shallow superadiabatic surface layer and nearly adiabatic to slightly stable above the surface layer up to a capping inversion at nearly 2000 m aloft. The upper-level winds transition to strong westerly flow from the more northwesterly at the surface (Figure 3).

Preliminary spread analysis

The fire quickly spread from its ignition to the southeast side of the plot in 4 min while the flank continued to spread to the south end of the experimental plot. Using the ground based temperature loggers (Fig 4) to determine the fire front position, the fire spread rate was calculated to be ~1.5-2.5 m s^-1 for the head fire while the flanks spread at 0.7 m s^-1.

Figure 2. Surface conditions before, during, and after the burn, January 29-30, 2013. Ignition time is marked by the vertical dashed line.

(23)

Standards: Data will need to be standardised, with quality controls on types, units, resolutions, formats and metadata. Standardised methodologies would greatly enhance cross jurisdiction collaboration and allow for broader, more widely applicable studies.

Access: In addition to data collection protocols, for value to be extracted it is important to ensure that contributing agencies are able to access the information provided by others. This is likely to require licensing agreements and the establishment of repositories. There is the potential for privacy issues to manifest, so limitation of access for some datasets may be a consideration.

The degree of difference in fire data collection protocols within a single country can be demonstrated with a search of the Australian public geospatial data directories using the term ‘fire’. Table 1 indicates that there are orders of magnitude in difference between the numbers of records available for each state. Wildfires occur within all states of Australia.

Table 1. Number of spatial records discoverable using the term ‘fire’ on the 5/2/2014 in Australian state based spatial data directories

State Records available

Australian Capital Territory 48

New South Wales 42

Northern Territory 2

Queensland 65

South Australia 4

Tasmania 11

Victoria 169

Western Australia 149

The three areas of fire information that need to be unified are described briefly below.

Data standards

Of the information typically stored by land managers in relation to fire, a large proportion of the data relates to infrastructure and other relatively invariant landscape attributes. There are a number of datasets that are in common between various states (for example fire history and ignition location), however metadata indicates that there is potentially substantial variation in the collection standards both within and between agencies (Walsh et al. 2007). While there are some barriers to the complete standardisation of data collection due to scale, access and technological issues, the degree of difference between the state data collection methodologies means there are likely to be interpretation issues when attempting inter-state investigations. In addition, while many agencies collect information on fire perimeters, for it to be of practical use to science, accessory data such as fuel details, topography, local weather observations, assets at risk and infrastructure must also be made available. Inconsistencies between jurisdictions in accessory data can further compound errors due to inconsistent fire data. For example, while approaches to evaluating fuel have been converging, there is still no system that is consistently applied or is suitable for all jurisdictions and vegetation types in Australia (Gould and Cruz 2012).

Inconsistencies in data can result in severe issues with regards to its access. Differing scales, units and margins of error can make it difficult to determine equivalence between datasets and can confound analysis. As there is much in common in the vegetation, weather and fire behaviour between states in Australia, it would be expected that many of the findings in one jurisdiction would also be appropriate in another. However with differing information, cross-border collaboration and model application is constrained. A common issue with fire science is that due to the complexity, rapidity and infrequent

(24)

nature of wildfire incidents it is difficult to collect enough observational data to provide suitable replication for statistical analysis. Standardisation of data collection will facilitate the aggregation of data throughout not only Australia, but the world, providing leverage on data collected, reduce unnecessary duplication and allow robust conclusions to be reached sooner and with less expense.

Scope of data

Considering Australia as a case study, there are a small number of data sets that are consistently collected by the majority of states despite the fact that wildfires are an issue to varying degrees in them all (Walsh et al. 2007). We believe that of the data available, only a small portion of data relevant to fire research is currently being archived as ‘fire data’ and made available for future needs. While it is common to collect information on impacts such as final perimeters and the locations of house losses, other information pertaining to fire spread predictions, firefighter accomodation and suppression deployments is not available. Table 2 provides an indication of potential information that may be generated; however this list is not exhaustive. Very few of the information sources listed are routinely collected.

Table 2. Information sources potentially available for typical fire incidents. Information is grouped by research group and incident management system category (AIIMS ICS 2013)

Research Potential Information source AIIMS ICS category Routinely available?

Management / Social Incident type Incident Control

Science Escalation / de-escalation Incident Control

Emergency declarations Incident Control Control structures Incident Control

Ignition point / points Investigation Y

Ambulance callouts Investigation Hospitalisation by cause Investigation

Insurance claims Investigation

Fencing / stock losses Investigation Urban infrastructure Investigation

Management Science Vehicle types Logistics

Supplies Logistics

Catering Logistics

Accommodation Logistics

Medical Logistics

Communications Logistics

Facilities Logistics

Finances Logistics

Resourcing Logistics/Planning

Emergency calls Investigation

Operational Research Weather radar Intelligence

Satellite images Intelligence

Weather Forecasts Intelligence Aircraft GPS tracks Operations

Vehicle GPS tracks Operations

Response structures Operations

Deployments Planning

Situation reports Operations/IC control Fire Behaviour Progression isochrones Intelligence

Ground/air observations Intelligence

Line scans Intelligence

Suppression strategies Operations/Planning

Final perimeters Planning Y

Fire behaviour predictions Intelligence

(25)

Weather observations Intelligence

Fuel /fire history Planning Y

Objectives Planning

Media strategy Public information

All Recovery All

Post fire impacts All

A secondary issue relating to data scope relates to the type of fires that are evaluated in detail. Due to the coronial and reactive nature of large fire investigations, there is a greater likelihood of data being available for extreme events. Data that is consistently collected at these fires includes maps of the final fire perimeter, headfire rates of spread, post-fire aerial imagery, weather conditions, fuel properties and the locations of house losses and deaths. However for more minor fires, the quality of information is generally much poorer. While it may be argued that larger fires are more important to study due their disproportionate impact, a strong counter argument is that some of the smaller fires may have had the potential to be catastrophic but effective preparedness or response measures prevented that from eventuating. Correspondingly it could be argued that smaller fires that occur under severe weather conditions deserve greater attention rather than less.

As it stands currently, a bias of data collection towards particular fire types has the potential to confound research and lead to incorrect conclusions. For example, evidence suggests that the predominant mechanism of house loss under the most extreme fire conditions differs to that under less severe but more common conditions. In addition, fire suppression and fuel reduction are less likely to be seen to be effective under extreme conditions; however they have clear value in preventing incidents from escalating under less severe conditions.

An additional issue is information decay. While information from some sources is robust, providing for simple post-fire collation, other forms are more transient. Highly temporally specific data is common in fires –it is important to be able to track and understand fast moving fire fronts. However if detailed information is not recorded at the time of the fire, information is easily lost and there may be limited opportunity to reconstruct occurrences post fire. This means that management agencies need to develop pro-active data policies. There need to be processes in place to actively target and record information while it is available and before its quality decays. While there is likely to be some overhead cost to this information gathering, such data can lead to improved decision support tools which reduce the overall overheads of fire management. Examples of information that can be transient include the location and rate of spread of fire fronts, the location, number and activity of suppression resources, and the behaviour and activity of civilians in the fire zone. After a fire event, there is greater luxury to collect data from some sources at a slower rate, however it remains important to actively target and record information before it is lost.

The scope of information being collected during and after fires is currently very limited. If information collection can be expanded, there are a wide range of research fields that stand to benefit. While the nature of Australian Eucalyptus forests mean that embers can play a disproportionate role in fire spread than in other forest types throughout the world (Cruz et al. 2012), fire remains a physical process. As such, standardisation of the scope and quality of data would be of greater benefit if it could be on an international scale, enabling research projects to be more expansive and enabling conclusions to be more broadly applicable.

Availability of data

Even if data is of high quality and correctly scoped, fire research may still be hindered by lack of access. This results from two primary issues. The first is that can be difficult for anyone (both internal and external to management agencies) to determine what data is being stored and is available. While spatial data directories have made substantial progress towards transparency of fire information in some jurisdictions, not all information that is collected is formally stored in a manner that is discoverable through such systems (e.g. Table 2). Much of the information gathered during a fire by a

(26)

managing agency would be stored in some manner, however only a small proportion is centralised and can be accessed using a query focused on a particular fire. The discovery of such information can require substantial resources; coronial reconstructions of fires take significant time and money to reconstruct particular events, even though much of the necessary information would be generated and available at the time of the event. Currently, the data recorded during fires shows limited consideration to the likely post-fire reconstruction information needs.

A second issue relates to data infrastructure. States, and within states; agencies, currently independently manage their own information. Obtaining data for a large geographic area is likely to mean that access must be negotiated from a number of different bodies. Once data is available, variability in formats and resolutions mean that further processing is typically required to ensure information from various sources is compatible and equivalent. Both obtaining data access and data processing can pose a significant time burden on users. The development of tools to access shared information is a necessary step towards effective cooperation and collaboration on informational needs. One issue that would need to be overcome in relation to information sharing is what limits would need to be applied. Many countries and states now have legislated privacy laws. Some of the information collected during fires may full under the jurisdiction of such laws – undeniably information relating to fatalities and events at particular residences is going to be sensitive and require some limitations on sharing under specific circumstances.

More broadly, the development of systems to recognise, tag, store and share fire related information could greatly reduce data discoverability issues. Furthermore, appropriate data sharing between agencies would create economies of scale, and more comprehensive data would enable more robust conclusions to be reached. If methods, scope and storage of fire data were standardised, substantial benefits may be realised, including greater success in preventing impacts from major fires in the future.

Proposed process

The development of a fire information system is expected to be a complex process as it would require the commitment and collaboration of a large number of emergency service agencies in disparate jurisdictions. It would be necessary to develop any new framework as a multi-stage project.

 Phase one would be to audit the ‘status quo’ and determine in detail how information is being managed within agencies.

 Phase two would be to undertake a needs analysis, evaluating informational requirements of both agencies and scientists. This would need to be carefully planned to ensure that findings are broadly applicable and not merely convenient for a single country.

 Phase three would be to analyse information sources and end-user needs to address data

‘gaps’, making recommendations on data standards, scope and storage mechanisms. This must also include the potential for the recognition of new needs and the inclusion of novel information sources. As technology is continually improving, the future improvement of information collection standards and the inclusion of new sources must remain an option even when a data framework is operating.

 Phase four would focus on the establishing compatible systems between agencies with an overarching aim to develop a centralised system for discovering and accessing data. This would require some reciprocal understanding in relation to data access, licensing, privacy and public availability.

To be effective, this process should be cooperative. While a considered analysis of informational needs will have immediate tangible benefits for a single agency, the greatest benefit overall would be the potential for cross border cooperation and increased research leverage.

(27)

We suggest that initial steps towards establishing an effective fire information system should be:

 The allocation of a unique code to each reported fire ignition. Any information pertaining to a particular ignition should be associated with this code. To avoid the risk of duplicates, a composite code could be used such as country/state/year/fire_number. Such a code would assist in the development of databases for the archiving of data.

 The existing fire command and control system should be used as a template for structuring and storing fire information. This will ensure that all relevant aspects of a fire are considered.

Each function group (i.e Planning, Intelligence etc.) should be responsible for ensuring that information is collected at a suitable standard (Australasian Fire and Emergency Service Authorities Council 2013).

Following this it will be necessary for substantial communication between agencies. This would need to be a forum style process, where views and perspectives can be shared to ensure the proposed procedures and policies best meet the needs of all involved.

The development of any new standards and processes requires careful consideration. Poorly designed standards can ‘lock in’ unsatisfactory methods or result in competing standards where the interested parties have not come to full agreement (Monroe 2011).

Summary

A common issue with fire science is that due to the complexity, rapidity and rarity of fire incidents it is difficult to collect enough observational data to provide suitable replication for statistical analysis.

Standardisation of data collection will facilitate the aggregation of data throughout the world, providing leverage on data collected, reduce unnecessary duplication and allow robust conclusions to be reached sooner and with less expense.

Land and emergency response organisations are increasingly being expected to deliver scientifically defendable decisions and to demonstrate continuous improvement in management and resource use.

The limited availability of high quality data restricts the rate at which research can advance and predictive capacity can improve. It is imperative that the losses caused by severe fires are not in vain;

losses should be offset by efforts to maximise the information obtained, helping to prevent a repeat of such events in the future.

References

Australasian Fire and Emergency Service Authorities Council (2013) 'Australasian Inter-service Incident Management System (AIIMS) 4th Edition.' (Australasian Fire and Emergency Service Authorities Council: Melbourne, Australia)

Cary, GJ, Flannigan, MD, Keane, RE, Bradstock, RA, Davies, ID, Lenihan, JM, Li, C, Logan, KA, Parsons, RA (2009) Relative importance of fuel management, ignition management and weather for area burned: evidence from five landscape fire succession models. International Journal of Wildland Fire 18, 147-156.

Cheney, NP, Gould, JS, McCaw, WL, Anderson, WR (2012) Predicting fire behaviour in dry eucalypt forest in southern Australia. Forest Ecology and Management 280, 120-131.

Chow, JYJ, Regan, AC (2011) Resource location and relocation models with rolling horizon forecasting for wildland fire planning. INFOR 49, 31-43.

(28)

Cruz, MG, Sullivan, AL, Gould, JS, Sims, NC, Bannister, AJ, Hollis, JJ, Hurley, RJ (2012) Anatomy of a catastrophic wildfire: The Black Saturday Kilmore East fire in Victoria, Australia. Forest Ecology and Management

Donovan, GH, Noordijk, P (2005) Assessing the accuracy of wildland fire situation analysis (WFSA) fire size and suppression cost estimates. Journal of Forestry 103, 10-13.

Gebert, KM, Black, AE (2012) Effect of suppression strategies on federal wildland fire expenditures.

Journal of Forestry 110, 65-9.

Gould, J, Cruz, MG (2012) Australian Fuel Classification: Stage II.(Ed. G Featherstone.) (Australasian Fire and Emergency Services Council (AFAC) and the Commowealth Science and Industrial Research Organisation (CSIRO): East Melbourne, Australia). [Accessed

Harris, S, Anderson, W, Kilinc, M, Fogarty, L (2012) The relationship between fire behaviour measures and community loss: an exploratory analysis for developing a bushfire severity scale.

Natural Hazards 63, 391-415.

Kepert, JD, Fawcett, RJB, Tory, KJT, W. (2013) Applications of very high resolution atmospheric modelling for Bushfires. In 'AFAC13 Shaping Tomorrow Together. Melbourne, Australia'. (Ed. RP Thornton) (Bushfire Cooperative Research Centre)

McLoughlin, D (1985) A framework for intergrated emergency management. Public Administration Review 45, 165-172.

Monroe, RP, 2011. Standards. XKCD. http://xkcd.com/927/,

Penman, TD, Bradstock, RA, Price, O (2013) Modelling the determinants of ignition in the Sydney Basin, Australia: implications for future management. International Journal of Wildland Fire 22, 469-478.

Prestemon, JP, Abt, K, Gebert, K (2008) Suppression cost forecasts in advance of wildfire seasons.

Forest Science 54, 381-396.

Strauss, D, Bednar, L, Mees, R (1989) Do one percent of forest fires cause ninety-nine percent of the damage? Forest Science 35, 319-328.

Sullivan, AL (2009) Wildland surface fire spread modelling, 1990–2007. 1: Physical and quasi- physical models. International Journal of Wildland Fire 18, 349-368.

Walsh, DJ, Rumba, KE, Hoare, J, Parsons, M, Thackway, R (2007) Reporting fire in Australia’s forests and vegetation. Bureau of Rural Sciences, Canberra, Australia.