Spatial interpolation of McArthur's Forest Fire Danger Index across Australia: Observational study☆
Introduction
Wildfires are one of the major natural hazards facing the Australian continent (Cheney, 1976, Harris et al., 2012). Chen (2004) rated wildfires as the third largest cause of building damage in Australia during the 20th century. Most of this damage was due to a few extreme wildfire events. The worst of these extreme events, and one of the worst natural disasters in Australian history, occurred in February 2009 when 173 people died and over 2000 houses were destroyed in the Victorian wildfires known as ‘Black Saturday’ (Teague et al., 2010, Cruz et al., 2012).
Detecting regions of high fire weather danger would allow planning and emergency authorities to warn the population of the elevated risk in the areas concerned and also help to mitigate the consequences of a wildfire if such an event were to occur. One of the most widely used indicators of fire weather danger in Australia at any given time is the McArthur Forest Fire Danger Index (FFDI) (Dowdy et al., 2009, Matthews, 2009). The FFDI is a function of drought factor and varies exponentially with temperature, humidity and wind speed (Noble et al., 1980).
Lucas (2010) developed historical fire weather data sets for Australia including calculation of FFDI. These data sets allow analysts to better understand the climatic conditions in which wildfires can develop. For a large country such as Australia, with a sparse observing network of meteorological sites, it is necessary to also consider the spatial distribution of this weather-based fire danger indicator. In this study we attempt to close the gap between station-based and spatial analysis by calculating observationally-based return period of extreme FFDI and presenting the spatial distribution of these return periods over the Australian continent using an advanced spatial interpolation technique. Return period calculated by employing extreme value distributions allow analysts to make inferences about the magnitude and frequency of extreme events beyond the range of years available in the observational data.
The study also shows that the best interpolation results for this fire-weather indicator can be obtained by using an algorithm which combines new interpolation techniques such as random forest with more conventional techniques such as inverse distance weighting. The interpolated maps using this combination showed physically plausible results and gave the smallest prediction errors as discussed in Section 5.3.
This study is important in the process of understanding the fire weather danger (utilising FFDI) faced by communities that are located some distance from the nearest meteorological observing stations. It is possible to undertake empirical calculations of the actual risk to these communities by comparing the FFDI with house loss information as discussed in Blanchi et al. (2010).
In practical applications the actual fire danger potential for a region depends on the type of vegetation in the region. In some cases an FFDI would not be the most appropriate index, for instance in regions dominated by grassland. In these cases the GFDI (Grassland Fire Danger Index) would be a more appropriate index to use. In more complex vegetation a combination of both, forest and grassland indexes weighted by a factor, could be a better indicator of fire danger. In this paper we use the FFDI to illustrate our methodology although grass rather than forest is the most common fuel type in Australia (Cheney and Sullivan, 2008).
Section snippets
Data sets available for this study
The main data sets used in this study are the fire weather data sets of 78 Australian stations compiled by Lucas (Lucas, 2010). These data sets contain records of daily weather variables which influence fire weather conditions such as temperature, relative humidity, drought factor, daily rainfall and wind speed and direction. After quality assessment and cleaning of these records Lucas calculated daily maximum FFDI and GFDI. These data sets cover the temporal range of June 1972 to June 2010.
The
Return Period of FFDI (RP)
Natural hazards can be quantified by using the Average Recurrence Interval (ARI) more commonly known as the Return Period (RP) of the natural phenomenon. If a given value of the phenomenon, termed ‘return level’, is exceeded with probability ‘p’ on average once a year, the RP corresponding to this return level is 1/p years (Coles, 2001). For instance, if the average annual probability of exceeding a gust wind speed of 45 m/s at some location is 0.002, we can say that the 500-year RP (1/0.002)
Spatial interpolation algorithms
RP of FFDI for a wide range of years (5–500) was calculated for the data of Lucas (2010) using the procedure described above. This produced discrete location data for a sparse network of observing stations. However, spatially continuous environmental information is often required by the government, industry, community and various organisations for environmental management and hazard mitigation. For our intended long-term application (fire weather danger assessment), measurements were often only
Calculation of RP curves using moments and maximum likelihood
Curves of RPs of FFDI for the 78 stations were calculated. The threshold to fit the GPD was found using the automatic threshold selection procedure developed by Sanabria and Cechet (2007). For 70 observing stations the GPD curve using the method of moments was higher than the maximum likelihood method. Eight stations, however, went against this trend. The station at Mt Gambier, SA, is one of those as shown in Fig. 2. However, the difference between the two curves was small (less than 1% for the
Examining the quality of FFDI data
Clarke et al. (2012) studied the quality of the data sets used in this project. In some stations wind speed measurements were based on visual estimates prior to the installation of anemometers. Other data sets have cumulative missing observations of more than one year. In general the FFDI calculation using these stations is considered of poor quality. After the low quality stations are eliminated, only 38 stations out of the 78 remain. In order to examine the impact of the quality of FFDI on
Spatial maps of seasonal and monthly FFDI
In the previous sections we presented a model for assessment of fire weather danger over the Australian continent. In practical applications it is convenient to look at maps of RP of FFDI with more detail. It is also important to look at the seasonal characteristics of FFDI. These two issues are addressed in the next sections.
For a more detailed assessment of fire weather danger we present maps of 50-year RP using a discrete scale which is very similar to the partitions utilised in the National
Conclusions and future work
Fire danger indices calculated at widely-spaced point locations are used by fire management agencies to assess fire weather conditions and issue public warnings. For the assessment of local hazard and risk, high resolution mapped information is becoming more available to consider community exposure and impact. The most widely used fire danger index in Australia is the McArthur Fire Forest Danger Index (FFDI). We have developed an efficient technique for the spatial mapping of long-term (return
Acknowledgement
We thank our Bushfire Cooperative Research Centre (Bushfire CRC) colleagues for their reviews and encouragement. We also thank the Geoscience Australia Product Development Team for their assistance in preparing the maps. This work is co-funded as part of the Bushfire CRC Understanding Risk Theme, Fire Impact & Risk Evaluation Decision Support Tool (FIRE-DST) project.
We want to express our appreciation to the anonymous reviewers for their constructive critique and thoughtful comments on this
References (48)
- et al.
Anatomy of a catastrophic wildfire: the black saturday Kilmore East fire in Victoria, Australia
For. Ecol. Manag.
(2012) Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall
J. Hydrol.
(2000)- et al.
Application of the Generalised Pareto Distribution to extreme value analysis in wind engineering
J. Wind Eng. Indus. Aerodyn.
(1999) - et al.
A review of comparative studies of spatial interpolation methods in environmental sciences: performance and impact factors
Ecol. Inform.
(2011) - et al.
Application of machine learning methods to spatial interpolation of environmental variables
Environ. Model. Softw.
(2011) - et al.
Can we improve the spatial predictions of seabed sediments? A case study of spatial interpolation of mud content across the southwest Australian margin
Cont. Shelf Res.
(2011) - et al.
Meteorological conditions and wildfire-related house loss in Australia
Int. J. Wildland Fire
(2010) Climate and Past Weather
(2012)Fire Weather Warnings
(2013)Relative Risks Ratings for Local Government Areas
(2004)
Busfire disaster in Australia, 1945–1975
Aust. For.
Grass Fires: Fuel, Weather and Fire Behaviour
Practical Geostatistics
Regional signatures of future fire weather over Eastern Australia from Global Climate Models
Int. J. of Wildland Fire
Changes in Australian fire weather between 1973 and 2010
Int. J. Climatol
An Introduction to Statistical Modeling of Extreme Values
A comparison of spatial interpolation techniques in temperature estimation
Australian Fire Weather as Represented by the McArthur Forest Fire Danger Index and the Canadian Forest Fire Weather Index
1-second SRTM Derived Digital Elevation Models User Guide
Analysing Seasonal to Interannual Extreme Weather and Climate Variability with the Extremes Toolkit
Extremes Toolkit (extRemes): Weather and Climate Applications of Extreme Value Statistics
The relationship between fire behaviour measures and community loss: an exploratory analysis for developing a bushfire severity scale
Natural Haz.
A Practical Guide to Geostatistical Mapping
Spatial interpolation of monthly mean climate data for China
Int. J. Climatol.
Cited by (43)
Application of artificial intelligence methods to model the effect of grass curing level on spread rate of fires
2024, Environmental Modelling and SoftwareIntegrated wildfire danger models and factors: A review
2023, Science of the Total EnvironmentArtificial intelligence and sustainability: Solutions to social and environmental challenges
2022, Artificial Intelligence and Data Science in Environmental SensingRVFR: Random vector forest regression model for integrated & enhanced approach in forest fires predictions
2021, Ecological InformaticsLiDAR-based Computational Fluid Dynamics heat transfer models for bushfire conditions
2021, International Journal of Disaster Risk ReductionCitation Excerpt :Various fire danger indices are used around the world to quantify the severity of bushfires and most of them share a common set of parameters. As an example, McArthur Forest Fire Danger Index (FFDI) is used in Australia, where it uses atmospheric temperature, relative humidity, wind speed and fuel load to calculate the severity of a fire [12,13]. Catastrophic fire conditions are indicated by FFDI values greater than 100 [12].
Machine learning in the Australian critical zone
2021, Data Science Applied to Sustainability Analysis
- ☆
© Commonwealth of Australia (Geoscience Australia) 2013. This publication is provided under a Creative Commons Attribution 3.0 Australia Licence with the permission of the CEO of Geoscience Australia.