1. Introduction
Wildfires can result in substantial social, economic and environmental impacts, and recovery activities may take many years. For example, an illegal campfire in California’s Garrapata State Park, in July 2016, ignited the most expensive fire in US history, costing more than 250 million US dollars [
1] Fires in Australia have resulted in mass house losses in the states of Victoria in 2009 [
2] and 2015 [
3], Western Australia in 2011, and New South Wales in 2013 [
4]. The total annual economic cost of bushfires in Victoria is estimated to be approximately 180 million Australian dollars [
5]. These costs have been forecast to double over the next 40 years to
$378 million [
6]. It is important to develop strategies that are able to reduce the risk of loss and thereby decrease the economic, environmental and social impacts of wildfire.
The occurrence and behavior of fires are driven by complex processes. Wildfires, and their associated management activities, have complex financial, social and environmental impacts. Here, we focus on fire behaviour alone, however recent research indicates [
7,
8,
9] that there is a need for improved quantitative information and tools in a wide range of management areas. There are a few national or multinational systems [
10,
11] providing basic fire behaviour information during wildfires, such as fire size, hotspots or burned area. There are individual attempts to improve this situation, predominantly in the collection of data post-fire. For example, the US National Institute of Standards and Technology (NIST) [
7] is developing wildland-urban interface (WUI) data collection methodology and first generation tools for improved risk assessment and risk mitigation in WUI communities at risk from wildfires.
Fire simulation systems [
8,
9,
12,
13] have been developed as part of management decision support systems and are vital to assessing fire risk to people and property. Most of these simulation tools are based on empirical fire forward rate of spread (FROS) models and do not necessarily emulate physical processes. Empirical FROS models were predominantly developed using observations of experimental fires burning in conditions that allow the fires to be safely managed. As a result, data representing the conditions under which damaging wildfires occur were rarely included. Indeed, current operational fire spread models assume that fires burn at an approximately constant (quasi-steady) rate of spread under a specific set of environmental conditions (e.g., Rothermel [
14], Canadian FBP system [
15], Cheney et al. [
16], CSIRO Grassland fire behaviour model [
17]). However, under extreme weather conditions there are emergent forms of fire behaviour that can rapidly change fire progression and intensity, including phenomena such as plume dominated spread, vortex structures and mass spotting events [
18,
19]. Consequently, simulation tools that utilise these FROS models are not able to emulate these dynamic wildfire behaviours.
Fire behaviour and management research cannot develop fully without better quantification of the various fire behaviour phenomena that occur under moderate and extreme weather conditions. To do so requires comprehensive and accurate data [
20]. Experimental research into intense fire behaviour cannot be undertaken as these fires cannot be safely managed; as a result, alternative sources of data are required and the only opportunity to collect information about fires under moderate and extreme conditions is to collect observations at wildfires as they occur. Case-study fires are commonly used in research [
2,
21,
22]. However, data is usually collated from various sources post event, hence data availability and quality is highly variable. There is currently no formal procedure for ensuring data collected during and post-fire is appropriate for meeting research requirements (consistent, accurate, correct and complete data). Without new data regarding wildfire behaviour, fire research, the future development of fire simulation tools and the associated decision support systems will be unable to improve significantly.
Fire information collected by management agencies varies by jurisdiction and fire size. In small fires, agencies may record simple details such as ignition location [
23,
24,
25,
26], final fire perimeter and fire area [
27]. For large fires that have substantial impacts, data may be extended to include fire severity [
28,
29,
30,
31], fire progression [
32,
33] and impact [
34,
35,
36]. However, much of this information is collected and collated post event. During fires there are many transient fire behaviour phenomena that cannot be easily reconstructed post event. These include spotting/fire storms, fire tornado/whirls, lateral vortices, junction zones (jump fires), eruptive fires, independent crown fires [
37], conflagrations, downbursts, and pyro-convective events [
18,
19], among others.
Information about fire behaviour is best collected as fires occur, however, there is currently no agreed set of standards or methodologies that define (a) what information needs to be collected during fires and (b) when collected, what data standards are appropriate [
27]. Data collected during a fire may be discarded if it is not required by an organisation. As a result, data that are saved will only be a subset of the information available during an incident.
In this paper we argue for a minimum standard of data collection during all wildfires. Doing this would enable fire behaviour phenomena to be documented and analysed. Furthermore, if such data collection were to be undertaken in a standardised manner across Australia or worldwide, it would enhance interagency collaboration, increase the research potential of datasets and make research findings more broadly relevant. To do this we first analyse the current information routinely collected during fire events for most states in Australia. Secondly, we provide an overview of some existing research methodologies that have the potential to be routinely used for observations during fires. Finally, we provide some initial recommendations of variables that would ideally be considered for routine collection during wildfires. While we focus primarily on Australian agencies, the recommendations are relevant for agencies worldwide.
2. Data Collection in Australia
Australia is a diverse continent with ecosystems ranging from tropical rainforests through to desert environments. Fires occur at varying intervals and intensities across the country [
38]. Land and fire management is the responsibility of state-level governments (which include six states and two territories). The industry body AFAC (The Australasian Fire and Emergency Service Authorities Council) endeavours to bring together fire and land management agencies across Australia and New Zealand to provide a co-ordinated response to fire and emergency management. To date, there has been no national policy developed focused on data collection and management during fires.
To understand what data are collected during wildfires, we approached representatives from all fire and land management agencies in Australia (
Table 1). Representatives of state agencies were contacted via email and telephone and asked to complete a guided survey (
Appendix A). There were multiple agencies from each state as fire management responsibilities are typically divided by land tenure. Specifically, we asked:
What information is collected and stored during fires? (
Table A1);
How frequently are data collected? (
Table A2); and
Does this information collection vary between fires under different conditions? (
Table A2)
Responses (
Appendix B) were received from Australian Capital Territory (ACT,
Table A3), New South Wales (NSW,
Table A4), Queensland (QLD,
Table A5), South Australia (SA,
Table A6), Victoria (VIC,
Table A7) and Western Australia (WA,
Table A8). No responses were received from Tasmania (TAS) and the Northern Territory (NT). Where multiple agencies responded from the same state, if at least one of the agencies in the state collects a certain type of data the variable was considered ‘collected’ by the state.
As fires are complex events and there are many sources of data, in the surveys we classified fire data into the broad types defined in
Table 2.
The responses in relation to the fire data were broken into three categories relating to incident size as determined by the Australasian Inter-service Incident Management System (AIIMS)/Incident Control System (ICS) system:
Small fire (Level 1)—characterised by being able to be controlled through local or initial response resources within a few hours of notification;
Medium fire (Level 2)—are more complex either in size, resources, risk or community impact. May require interagency response;
Large fire (Level 3)—are protracted, large and resource intensive. They may affect community assets and/or public infrastructure, and attract significant community, media and political interest.
We found that the amount of information collected increases with increasing fire size (
Figure 1). Basic information that is simple to collect such as ignition location, incident type and final perimeters are recorded by at least one agency in all states. Data types that are more complex to collect (such as fire perimeters) or have technological requirements (such as forward looking infrared (FLIR) camera-based data) are collected in fewer states. This is due in part to the differing technical capabilities of the states (for example, some states do not have access to aircraft with linescan and infrared equipment). There more detailed quantitative data (which is important for conduction analysis of fire behaviour) such as weather radar, progression isochrones, FLIR video, linescans, are generally only collected occasionally (
Figure 1b). Apart from fire sizes, it is unclear what stimulates the collection of such data. If these data are only collected from fires of a specific nature, it may result in biases that affect analysis and interpretation of the frequency of extreme fire behaviour.
When asked what kind of data should be collected routinely in the future, the highest number of responses, irrespective to fire size, were received for fire behaviour type, weather radar and local weather (
Figure 1c). From our surveys, we also identified that there is a high degree of variation in the way data are curated. While we were unable to conduct quantitative analysis, it is evident that it is stored in a variety of ways (e.g., hard copies, local servers, online data repositories). Databases are not shared between states and rarely between agencies within the same state, and information storage is not centralised; i.e., different categories of fire data may be stored in different systems or at different physical locations. For example in South Australia data are stored in an Incident database, logbooks, a fire behaviour analyst server, a Corporate GIS database, the Critical Resource Incident Information Management System Online Network (CRIIMSON), the SA Computer Aided Dispatch (SACAD) system, the Australasian Incident Reporting System (AIRS, and Incident Management Teams reports (IMTs). For access to each data source, separate permissions are typically required. Even if data are of high quality and correctly scoped, difficulty in access may hinder fire behaviour science.
3. Innovation in Data Collection
The management of information during active wildfires is an undoubtable challenge to managers. However, with recent technological developments, it is likely to become simpler to collect some information. There are a wide range of methods that have been developed in the research space that have not yet been adapted for operational use by fire management agencies. Research will always produce more methods than agencies will adopt, however methods that can be demonstrated to efficiently provide meaningful data are likely to be considered. For a new method to be adopted, ideally it should offer (1) a tangible immediate benefit to the agency utilising it; (2) a long-term benefit to the agency through improved decision support as a result of research outputs; (3) feasible implementation within the operational context and (4) ease of use. Researchers and agencies need to work more closely to identify such methodologies and develop strategies for data collection that ensure the quality of the data recorded while minimising cost and disruption to the agencies. In this section, we review a number of recent innovations that have the potential to assist with both management and science. Some of these are already in use in parts of Australia.
Perhaps the greatest recent advance in fire behaviour research are data derived from remote sensing before, during and after the fire. Remotely sensed data provide researchers a means to quantify patterns of variation in space and time. The utility of these data depends on the scale of application. Satellites and aircraft are the main sources of these data. Multi-temporal remote sensing techniques based on space and airborne sensors have been effectively employed to assess and monitor landscape change in a rapid and cost-effective manner [
40,
41]. Remotely sensed data have been used to detect active fires [
42,
43]; map fire extents scales [
44,
45,
46,
47]; estimate surface and crown fuel loading [
48,
49]; assess active fire behaviour [
50,
51,
52] and examine post-fire vegetation response [
53,
54].
Fire behaviour and measures of the fuel consumed have been quantified through the analysis of thermal infrared imagery [
55,
56,
57]. Infrared (IR) sensors and Infrared Line Scanning Systems on aircrafts allow land managers to detect actively burning areas, spot fires, estimate the energy radiated from the fire as it burns and to analyse fire behaviour. These approaches allow for the determination of key parameters of the fire, such as intensity, size, rate of spread, hazards and other factors relevant to suppression activities and logistics. Line Scanning Systems have been used for many years for fire mapping for firefighting purposes [
58]. However, to date the systematic use of them to collect fire behaviour data has been limited. When routinely collected, progression isochrones will significantly simplify the process of fire reconstruction and improve fire simulation tool validation. Mapped data will also provide an understanding of how spatial processes like climate, topography, and vegetation dynamics influence fire behaviour and regimes. Combining these data with information on fire behaviour type and evidence of “unusual” behaviour, such as extreme fire behaviour, is vital. Routinely collecting information about fire intensity, fire front depth, spotting ignitions and “unusual” fire behaviour will help to better understand fire behaviour and improve operational and physical models.
Another system in operational use for firefighting that has had limited adoption for systematic data collection is the use of low altitude IR fire observation. Operationally in Australia, aircraft use a single IR sensor which can detect fire fronts or hot spots and firebrands but not both. Most imaging techniques intended to detect the heat signature of fire are based on MWIR (Medium Wavelength Infrared) and TIR (Thermal Infrared) (TIR band includes spectrum from both MWIR and LWIR (mainly LWIR) spectral regions [
59]) sensors [
60]. Using a single IR sensor is problematic as the signal varies with emissivity, there is considerable incident energy and only a small fraction of the pixel may correspond to the fire. Using multi-spectral methods can solve this problem. For example, in the USA, the airborne fire data gathering is derived from multi-spectral data acquired by autonomous modular line-scanner sensors (AMS) operating in shortwave (SWIR), MWIR and LWIR spectral regions and providing enhanced dynamic range in support of active fire imaging [
60]. Also, by using a multispectral approach the fire radiative power, fire fractional area and temperature can be estimated [
61]. Furthermore, such systems can view through smoke, allowing the nature of ember generation and transport to be observed.
A relatively recent set of methods used in research but not yet in operational fire management is the 3D visualisation and measurement of wildfire smoke plumes and the atmosphere using LIDAR (LIght Detection And Ranging) [
62,
63,
64], SODAR (SOnic Detection And Ranging) [
65,
66] and RADAR (RAdio Detection and Ranging) [
63,
67,
68,
69,
70]. These methods extract vertical profiles of the smoke plumes, as well as record the movement of winds and hot gases from the fire. Such information is critical for scientists to understand fire behaviour—in particular the rapid acceleration that occurs with some fires as they become large. The intensity and evolution of convective plumes is critical in the understanding of lofting and spotting of firebrands, where plume structure begins to play an important role in how the firebrands are spatially distributed. A number of studies have also characterised smoke plume behaviour using information derived from satellite data [
71,
72,
73]. Information on smoke-plume heights and their dynamics and these related data will allow for improvements in smoke dispersion and air quality models.
Weather RADAR [
67,
68,
69,
70] and LIDAR [
62,
63,
64] have also been used for visualizing active fires in context of dynamic broad scale weather events, understanding plume formation and estimation of it characteristics. As weather RADARs are maintained over large parts of Australia as part of rain monitoring, they have very broad coverage and scan at a high frequency. Extreme fire weather features like sudden wind changes, the escalation of a plume into a pyrocumulonimbus (PyroCb) (or Cumulonimbus Flammagenitus (CbFg) according to the new International Cloud Atlas,
https://cloudatlas.wmo.int) or the advent of dry thunderstorms and associated lightning are all important events to be considered during a major bushfire event but are rarely captured using existing methods. Ground-based scanning systems such as RADAR can be considered an important auxiliary tool for detecting unauthorised burning and forest fires, adding significant value to the information for decision-making in monitoring, detecting and suppressing wildfires. An advantage of using weather RADAR to analyse fire is that the network is already in place and maintained for another purpose. Consequently, barriers to its adoption are low.
Remote sensing methods have provided a major step forward in data collection and understanding fire behaviour. Methods for collecting these data are also under constant development. Two major areas are worth highlighting. Firstly, as new satellites are launched the quality and quantity of data available will increase. In Australia, research and management have both used the Advanced Very High Resolution Radiometer (AVHRR) imagery and the Moderate Resolution Imaging Spectroradiometer (MODIS) on Terra (1999) and Aqua (2002) [
74]. The launch of the Japan Meteorological Agency (JMA) Himawari-8 satellite, with the 16-band Advanced Himawari Imager (AHI-8) onboard in October 2014 presents a significant opportunity to improve the timeliness of satellite fire detection across Australia. The near real-time availability of images, at a ten minute frequency, may also provide contextual information (background temperature) leading to improvements in the assessment of fire characteristics [
43]. Secondly, unmanned aerial vehicles (UAVs, commonly known as drones) as remote sensing platforms have the great potential to increase the efficiency of data acquisition, but their applications are still at an experimental stage [
75,
76,
77]. UAV remote sensing has low material and operational costs, flexible control of spatial and temporal resolution, high-intensity data collection, and a reduction of risk to crews. As the complexity of UAVs and sensors increases, so will our ability to capture high resolution spatial data at wildfires. An additional advantage is that they can be used in conditions that would be hazardous to human health; particularly around fast moving fires or where there is unstable weather.
Table 3 shows innovations for which adoption can bring immediate benefit to fire science and management.
4. Challenges of Data Collection and an Ideal Dataset
Experience of fire behaviour data collection from different fire agencies in Australia showed that there are a substantial amount of data across events. However, these data are often inconsistent and limit quantitative analysis of fire behaviour. For example, line scanning frequency varies significantly during a fire, from several times to only a single scan per day. Aircraft infrared video consists of fragments of fire front depending on the preferences of the pilot and does not show the fire front or spotting development. A further complication is the source of data. For example, weather records are often derived from weather stations, which are sparsely located across the landscape. Data from these stations can differ significantly from weather at the fire location, which may only be 10 km away. Also, mismatches in temporal and spatial resolution of data between multiple sources create challenges in aligning data for meaningful analysis.
Our paper is focused on measuring data for analysis of various fire behaviour phenomena—how to understand them and take them into account in operational and physical based models. The ideal dataset for these analyses would be recorded every 5–15 min and include information about fire progression (linescans and ground observations), infrared video in SWIR, MWIR and LWIR spectral regions of fire front and spot fires, RADAR and LIDAR measurements, high resolution satellite images (<100 m), photo and video of fire and plume development and ground weather observations. Such data would allow researchers to catch even short lifetime phenomena and dynamic effects, such as extreme fire behaviours which can have devastating consequences. Unfortunately, while this dataset is desirable it is unrealistic as it requires huge human and equipment resources, which are very limited during wildfires.
As a starting point we recommend a focus on particular categories (
Table 4). These categories are those that will provide the greatest information gains, with minimum additional resources. Our focus list is in relation to all types of fire behaviour, but particularly extreme fire behaviour—the phenomena that only occur at large scales and under severe conditions that cannot be safely replicated experimentally.
Any system or set of measures must be accompanied by the development of a robust data storage and sharing system. The development of such system could greatly reduce data discoverability issues for research and governmental inquires. The information needs for fire behavior research are not necessarily the same of those needed for managing the control of wildfires. Control requires information at high temporal frequencies, but does not necessarily require the degree of accuracy or precision required for research. However, intelligence gathering infrastructures are in place for fire control, and, at the very least, the information currently being collected could be archived in a way to make it suitable for future analysis. Much of the information currently gathered during a fire by a fire management agency is stored in some form, however only a small proportion is centralised and can be easily accessed. A centralised and/or standardised data storage approach would streamline this process and result in better management and research outcomes. Furthermore, consistency in data storage and management should result in improved data sharing between fire management agencies. From a research perspective this should allow for more comprehensive datasets to be developed, thereby increasing the application of research results. As a starting point, data could be collected in each state fire agency with the perspective to create a national data collection system. Also, all data formats should comply with the International System of Units (SI). Such a system would provide a number of challenges in terms of collation and management of the data, but it is beyond the scope of this paper to discuss these issues. Of the data sources in
Table 4, much of the information is already being collected (e.g., fire observations, line scans), so there is the potential for rapidly improving data available for research. While the information currently collected may not be of a suitable standard for research, integrating scientific data collection into existing systems is much more likely to be supported by managers, in contrast to demanding new data collection activities that compete for the resources being used for fire control.