Extending methods for assessing fuel hazard in temperate Australia to enhance data quality and consistency

Description of the existing assessment method

The existing assessment method in this study followed the OFHAG fourth edition (Hines et al. 2010) (Table 1). The OFHAG was designed for use predominantly in eucalypt forests, woodlands and shrublands in temperate Australia, and focuses on fine fuel arrangement as one of the main determinants of wildfire behaviour (Hines et al. 2010). In the OFHAG, fine live and dead fuel (<3 and <6 mm in diameter, respectively) are assessed across four strata: surface, near-surface, elevated and bark (Supplementary Appendix S1). Each fuel stratum is assigned a hazard score (low, moderate, high, very high, extreme) based on the combination of fuel properties described below and by using the hazard tables in the guide. Higher hazard scores are associated with increases in elements of fire behaviour where surface fuels give an indication of forward rate of spread, near-surface and elevated fuels indicate forward rate of spread and flame height, and bark fuels indicate spotting potential. The OFHAG does not consider midstorey and overstorey canopy, or duff (duff is largely absent in fire-prone vegetation in temperate Australia). Assessments are over a circular area of ~10-m radius for surface, near-surface and elevated fuel, and 20-m radius for bark fuel.

In the OFHAG, surface fuel consists of dead leaves, twigs and bark lying in a predominantly horizontal orientation on mineral soil. Surface fuel is assessed by measuring depth (from the top of the litter bed to the mineral soil underneath) and cover. Depth is measured with a fuel depth gauge (a 15-cm circular disk with a ruler through a slot in its centre; Supplementary Appendix S1) and is the average of five measurements per plot, with the measurement locations at the discretion of the assessor. Cover is estimated visually across the 10-m radius area.

Near-surface fuels are live and dead vegetation such as graminoids, forbs, small shrubs and juvenile woody plants. This stratum is vertically continuous and effectively in touch with the ground. It will be consumed in a surface fire. To assess near-surface fuel, assessors visually estimate the cover and percentage dead across the 10-m radius plot.

Elevated fuel includes taller shrubs and sapling trees. This fuel stratum is mainly upright in orientation, and can contain suspended leaves, bark or twigs. There is a clear gap between the elevated and surface and/or near-surface fuels, meaning a surface fire could pass underneath the elevated strata. Elevated fuel is assessed in much the same way as near-surface fuel, by visually estimating the cover and percentage dead across the 10-m radius plot. The guide also describes how easily a person could walk through the vegetation as an additional measure of elevated fuel density.

Bark fuel consists of bark on both the trunk and branches from the ground to canopy. Bark fuel is divided into three main types that characterise the trees of southern Australia: stringybark, ribbon or candle barks, and ‘other’ barks, which include smooth, platy, papery and coarsely fibrous barks. Bark is assessed by visually estimating the percentage char and determining the degree of bark attachment (i.e. hand pressure needed to remove bark). An assessment is done for each bark type when the type represents more than 10% of the trees within a 20-m radius of the plot centre.

Development of an extended assessment method

We extended the OFHAG to improve the consistency of estimates (Table 1, Supplementary Appendix S1). First, we formally delineated the plot boundary by using two 50-m transects (north–south, west–east, intersecting at the 25-m point) to define a circular plot of 0.20 ha.

Second, to ensure coverage of the full plot area, we instructed assessors to take fuel measurements systemically at 2-m intervals along each transect, providing a total of 52 measurement points per plot for each of the surface, near-surface and elevated fuel strata. Surface fuel cover and depth were captured at the exact point of intersection with the transect as per the point-intercept method (Elzinga 1998). Near-surface and elevated fuel cover and percentage dead were assessed in a projected 50-cm radius cylinder (centred at the transect point) to enable a larger proportion of the total plot to be assessed (~0.004 ha). Additionally, near-surface top height and elevated top and bottom heights were measured (Supplementary Appendix S1), but not analysed in this study.

Third, to ensure consistency of definition of fuel strata, we developed a decision tree (Fig. 1) to help with the delineation of the near-surface and elevated strata – distinguishing near-surface from elevated fuel has been anecdotally noted as a consistent source of confusion for assessors in the OFHAG.

Fig. 1.

Decision tree developed in this study to reduce ambiguity in distinguishing between near-surface and elevated fuel strata.

Fourth, in addition to assessing bark structure as per the OFHAG, we estimated the quantity of bark fuel across the full 25-m radius plot. This was achieved by using a plot stem count and measuring the diameter at breast height (DBH; 1.3 m from ground) of the most representative stem (per bark type). The stem count and DBH were used to calculate total tree basal area to provide an estimate of the amount of bark (i.e. surface area of bark) by type. In comparison, the OFHAG does not consider the number of stems or stem size, only the percentage of trees exhibiting each of the bark types. For example, in the existing method, a plot with three large stringybark trees would have a higher bark hazard than a plot with three small stringybark trees, whereas the visual OFHAG would treat both plots equally.

All measurements were summarised per plot and then used to determine the fuel hazard score per fuel stratum.

To reduce ambiguity in converting fuel properties to hazard scores, we refined the OFHAG hazard tables (Fig. 2, Appendix 1). Fig. 2 highlights the source of some of the ambiguities in the existing OFHAG for surface, near-surface and elevated fuel. As an example, hazard score assignment in the elevated stratum relies on set combinations of percentage plant cover and percentage dead; however, the ranges of each often overlap among hazard categories, making it unclear which final hazard score should be assigned. The tables also contain qualitative fuel strata descriptors, which potentially add further assessor bias to the assessment method. These include ill-defined descriptions of vegetation flammability and fuel thickness and descriptors about the ease with which an assessor can walk through the forest. Further research is needed to properly incorporate measures of flammability, leaf thickness and vegetation density into assessments of fuel hazard. For our extended method, we omitted these qualitative metrics. We thus extended the surface, near-surface, elevated and bark hazard tables to ensure that the measurement range of individual fuel properties did not overlap among hazards categories and that the combination of two fuel properties resulted in a clearly defined hazard score by strata and, where possible, used parts of the tables to only include quantitative measures of fuel. The conversion of fuel properties to hazard score occurred during the data analysis stage. These refined tables are herein referred to as the extended hazard tables (Fig. 2, Appendix 1).

Fig. 2.

Existing and extended hazard tables developed in this study: (a) existing surface; (b) extended surface; (c) existing near-surface; (d) extended near-surface; (e) existing elevated; (f) extended elevated; (g) existing stringybark; and (h) extended stringybark fuel strata. The existing hazard tables are a simplified version of those in the OFHAG fourth edn. (Hines et al. 2010). Red numbering highlights ambiguity where fuel property values overlap multiple hazard categories. Fuel hazard categories are L, low; M, moderate; H, high; VH, very high; and E, extreme. ^AThe stringybark hazard table is shown as a bark fuel example; see Appendix 1 for the ribbon or candle and ‘other’ bark hazard tables.

Field test to evaluate fuel assessment methods

We established 69 plots in south-western Victoria, Australia (Fig. 3), spanning 272 km east–west (141.09°–144.19°E), 184 km north–south (37.00°–38.66°S), and 360 m in altitude (0–360 m above sea level (asl)) (Gallant et al. 2011). South-western Victoria has a temperate climate characterised by mild to warm dry summers (Bureau of Meteorology 2005) with the average annual rainfall across the plots ranging from 574 to 1188 mm year⁻¹ (Fick and Hijmans 2017). Plots were selected to represent the range in a fuel type known locally as Woodland Heath (Fig. 4). Fire managers in this region have identified Woodland Heath as a priority for further research because of its widespread occurrence on public land and its high biodiversity value (Gazzard et al. 2020). It is often targeted for fuel reduction and ecological burning, and therefore the ability to accurately quantify fuel properties in this fuel type is of high importance.

Fig. 3.

Map of site locations across the south-west of Victoria, Australia.

Fig. 4.

Woodland heath fuel types in the south-west of Victoria, Australia. Images depict different woodland structures with (a) higher density of canopy species, presence of near-surface fuels and absence of elevated fuels; (b) higher density of canopy species, presence of shrubby elevated fuels; and (c) sparser density of canopy species and presence of near-surface and shrubby elevated fuels. Photographer credit: Chris Medlin.

Our plots in the Woodland Heath fuel type include multiple vegetation types that are classified locally as Heathy Herb-rich Woodland, Heathy Woodland and Damp Heathy Woodland (Department of Environment Land Water and Planning 2021). These vegetation types commonly include an overstorey dominated by mixes of fire-tolerant eucalypts that are 10–15 m in height and a typical projective foliage cover of 10–20%. The eucalypts predominantly include stringybark trees (e.g. Eucalyptus obliqua (messmate stringybark), E. baxteri (brown stringybark)) and/or ribbon bark trees (e.g. E. viminalis (rough-barked manna-gum), E. ovata (swamp gum)) and/or ‘other’ bark trees (e.g. E. willisii (Jimmy’s shining peppermint)). Midstorey and understorey shrub species typically have narrow or ericoid leaves and have a foliage cover of 15–40% for medium shrubs and 15–20% for small shrubs. Medium shrub species commonly include Banksia marginata (silver banksia), Epacris impressa (common heath), Leptospermum myrsinoides (heath tea-tree), Leptospermum continentale (prickly tea-tree) and Correa reflexa (common correa). These shrubs often persist over a layer of grasses and graminoids, or a dense layer of bracken (Pteridium esculentum) (Department of Environment Land Water and Planning 2021). These vegetation types are grouped into the Woodland Heath fuel type because they exhibit similar fire behaviour (Kington and Tolhurst 2019).

Plots were selected using stratified random sampling to capture differing vegetation structures by encompassing a range of time since fire age classes (0.5–30 years since fire), fire severities (understorey burnt, canopy scorch and canopy burnt), and an aridity gradient (aridity index 1.7–3.4; Nyman et al. 2015) (Fig. 3). We omitted patches of vegetation smaller than 1.5 ha and ensured plots were located at least 10 m from roads and strategic fuel breaks to reduce edge effects. Plots within the same stratification unit were at least 1 km apart. A single set of assessments was completed at each plot by two assessors (n = 69). A plot centre co-ordinate was randomly selected for each plot and assessors were required to visually check that the plot’s vegetation and fire history matched expectations.

To enable comparisons between the existing and extended methods, assessors began by assessing the fuel in each plot using the existing method (outlined in the OFHAG). This was done from the plot centre, before assessors then undertook the extended assessment along the plot transects (Table 1, Supplementary Appendix S1). The existing assessment was completed first so that the assessment was not influenced by data collected for the extended method.

Assessors comprised local fire managers, with two members being lead assessors. At least one of these two members assessed every study site. A field-based training day was run to train staff in the extended field method and the lead assessors completed 2 days of field surveys together to standardise their assessment techniques. Assessments occurred over a 12-month period from April 2021 to April 2022.

Data analysis

Comparisons of the existing vs extended assessment methods were done in three stages: (1) a comparison of fuel properties, which focused on surface cover and depth, near-surface cover and percentage dead, and elevated cover and percentage dead; (2) a comparison of hazard tables by using fuel property data from the existing assessment method and comparing scores for those data derived from the existing hazard tables with those derived from the extended hazard tables for the surface, near-surface and elevated strata; and (3) an overall comparison of hazard scores derived from the full implementation of the existing vs extended assessment methods for all fuel strata (including bark). Fuel hazard scores (low, moderate, high, very high or extreme) were converted to integers from 1 to 5 and treated as pseudo-continuous.

Bland–Altman plots were used for the comparison of fuel properties between assessment methods. Bland–Altman plots are a diagnostic tool for assessing the agreement between two methods of measurement by plotting the difference of two paired measurements against their mean (Twomey 2006; Giavarina 2015; Lehnert 2015). The Bland–Altman plots display the bias between the two methods by plotting the mean bias (normal difference distribution) or median bias (non-normal distribution) of the difference data and their limits of agreement (±1.96 s.d. for mean difference and 2.5th and 97.5th percentiles for median difference; Twomey 2006). Difference data were checked for normality using histograms, qq-plots and the Shapiro–Wilk test (Giavarina 2015) and results were used to determine whether mean or median bias was appropriate. Mean bias was used for surface cover, surface depth and near-surface percentage dead. Median bias was used for near-surface cover, elevated cover and elevated percentage dead.

One-to-one plots and Spearman’s correlation coefficient were used to assess the agreement between fuel hazard scores. Cohen’s kappa statistic (k ranging from −1 to +1) was used as an index of inter-rater agreement between two hazard scores (Cohen 1968; Gamer et al. 2019). These data were used to calculate misclassification errors between assessment methods by calculating absolute difference between raters (Gamer et al. 2019) as a difference of only one hazard score (e.g. low hazard vs a moderate hazard score), or as a difference of two or more hazard scores (e.g. low vs a high or very high hazard score). A difference of two or more hazard scores was considered misclassified.

The R statistical programming language, Version 4.1.0 (R Core Team 2016), was used for data processing and analysis, including the packages BlandAltmanLeh (Lehnert 2015), cowplot (Wilke 2020), irr (Gamer et al. 2019), scales (Wickham and Seidel 2022) and tidyverse (Wickham et al. 2019).

Comparison of fuel properties measured with the existing and extended fuel assessment methods

The existing assessment method estimated a mean of 6.5% less surface cover and a median of 11% more near-surface cover than our extended assessment method (Fig. 5a, c). Surface fuel depth and near-surface percentage dead showed relatively strong average agreement between the two assessment methods (mean bias within 3.5%; Fig. 5b, d). However, individual plot differences for surface fuel depth were highly variable, sometimes exceeding 10 mm (Fig. 5b). For elevated fuel, the existing assessment method estimated a median of 9.8% more cover and 5.3% less dead fuel than the extended assessment method (Fig. 5e, f). Disagreement between the two methods for elevated cover increased with increasing cover up to a maximum of 85% difference (Fig. 5e).

Fig. 5.

Bland–Altman plot of differences between existing and extended assessment measurements for (a) surface cover, (b) surface depth, (c) near-surface cover, (d) near-surface percentage dead, (e) elevated cover, and (f) elevated percentage dead. Solid red, thick line represents mean or median bias and dashed lines represent the limits of agreement for the bias.

Comparison of the existing and extended hazard tables using data from the existing method

Fuel hazard scores derived using the existing vs extended hazard tables (i.e. both based on fuel property input data from the existing method) were strongly associated for the surface and elevated strata (Spearman’s correlation coefficient r = 0.79 and 0.81 respectively; Fig. 6a, c), and moderately associated for the near-surface stratum (r = 0.62; Fig. 6b). Despite the high correlation, there was an apparent bias for elevated fuel, with the existing hazard table producing scores that were up to three hazard scores lower than the extended hazard table (Fig. 6c). For near-surface hazard, there was considerable scatter around the 1:1 line, with scores varying by up to three classes (Fig. 6b).

Fig. 6.

Comparison of fuel hazard scores derived using the existing and extended hazard tables (both based on fuel property input data from the existing method) for (a) surface fuel hazard, (b) near-surface fuel hazard, and (c) elevated fuel hazard. Spearman’s correlation coefficient (r) is displayed. Data points are in black and have been jittered to show overlap. The blue line is a 1:1 line.

Surface fuels had the lowest misclassification rate, with misclassifications restricted to just one hazard score difference between the existing and extended hazard tables, which occurred 30% of the time (Fig. 7). The greatest misclassification rate was for near-surface fuel, with misclassification of two or more hazard scores occurring for 5% of plots, and a misclassification of one hazard score occurring for 50% of plots. Misclassification of the elevated fuel by two or more hazard scores also occurred for 5% of plots, while misclassification by one hazard score occurred for ~40% of the plots (Fig. 7).

Fig. 7.

Misclassification errors between the existing and extended fuel hazard tables as calculated using Cohen’s kappa statistic (k) (Cohen 1968). Differences of two or more hazard scores are labelled as misclassified.

Comparison of fuel hazard scores derived from the full implementation of the existing and extended assessment methods

Fuel hazard scores derived from the full implementation of the existing and extended assessment methods were most strongly associated for bark (Spearman’s correlation coefficient r = 0.79; Fig. 8d), and moderately associated for the near-surface stratum (r = 0.52; Fig. 8b). The methods had fair agreement for surface hazard (r = 0.40; Fig. 8b) and poor agreement for elevated fuel hazard (r = −0.10; Fig. 8c), with the existing assessment method generally generating higher surface and elevated hazard scores compared with the extended method. Estimates of elevated hazard were up to four scores higher when determined using the existing compared with the extended assessment method (Fig. 8c).

Fig. 8.

One-to-one plots comparing the existing and extended fuel hazard scores derived from the full implementation of each method for (a) surface fuel hazard, (b) near-surface fuel hazard, (c) elevated, and (d) bark fuel hazard. Spearman’s correlation coefficient (r) is displayed. Data points are in black and have been jittered to show overlap. The blue line is a 1:1 line.

Bark fuel had the lowest misclassification rate, with misclassifications restricted to just one hazard score difference between the two assessments, which occurred approximately 35% of the time (Fig. 9). The greatest misclassification rate (70%) was in the elevated fuel, with a misclassification of two or more hazard scores occurring for 30% of plots, and a misclassification of one hazard score occurring for ~40% of plots. Misclassification of the surface and near-surface fuel occurred in approximately half the plots, with ~10–15% of these being misclassifications of two or more scores.

Fig. 9.

Misclassification errors between the existing and extended fuel hazard scores from the full implementation of each assessment method as calculated using Cohen’s kappa statistic (k) (Cohen 1968). Differences of two or more hazard scores are labelled as misclassified.

Comparison of fuel properties measured with the existing and extended fuel assessment methods

The discrepancy in visual cover estimates between the whole-plot existing method and the small-quadrat based extended method for near-surface and elevated fuel possibly reflects differences in the assessor’s field of view across a plot. In the existing method, assessors have an oblique vantage point when assessing fuel strata from a single fixed location near the plot’s centre. Features within view, such as gaps between vegetation, are effectively combined across different scales and with different degrees of accuracy, and this means that the spatial patchiness of the fuel cannot be properly viewed (Unwin 1975). Near-surface and elevated fuels likely appear more horizontally and vertically continuous from a single viewing location, making it more difficult to distinguish between strata in a consistent manner for the whole plot and leading to the perception of higher cover estimates. Additionally, it becomes more difficult to accurately measure cover as the size of an assessment area (i.e. the plot) increases (Elzinga 1998; Caratti 2006). Our extended assessment allowed assessors to systematically consider multiple small areas across an entire plot, allowing better visual distinction between gaps and a comparatively less biased assessment of fuel cover variation in space. The use of the decision tree (Fig. 1) at each of these points also potentially increased assessor consistency when distinguishing near-surface from elevated fuels, although we note that the separation of these fuel strata remains at least partly subjective.

The magnitude of difference in cover estimates between methods is likely to have implications for predicting fire behaviour. As elevated and near-surface cover increase, so too does the quantity and horizontal connectivity of the fuel, allowing easier flame propagation through and between fuel layers (Hines et al. 2010; Cruz 2021). Increasing involvement of these fuels in the flaming fire front results in increased fire rates of spread, intensity and the potential for crown fire (McArthur 1967; Cruz et al. 2021). Consequently, over- or underestimates of near-surface and elevated cover could lead to inaccurate predictions of fire impact (e.g. final fire footprint). Fire behaviour prediction systems calibrated using fuel inputs derived from the existing approach may need re-calibration if the extended approach is used to derive fuel inputs (e.g. the fire spread models within PHOENIX Rapidfire are calibrated against case study fires and use fuel hazard curves derived from the existing approach; Kington and Tolhurst 2019). Additionally, a 10% difference in near-surface and elevated cover between wet and dry eucalypt forest has been found to contribute to substantial differences in the near-ground microclimate, where less cover resulted in drier surface conditions and more days where surface fuels were available to burn (Pickering et al. 2021). Therefore, overestimation of cover could lead to underestimations of surface fuel bed availability (fuels are drier than the assessment indicates).

For surface fuel measurements, both assessments employed the same method for measuring surface depth but with different numbers of measurements (5 vs 52 respectively) and an ad hoc versus systematic sampling strategy. Nonetheless, the methods were in close agreement with a small mean bias. This reflects the smoothing of overall differences rather than the absence of differences between the two methods (i.e. there was substantial variability in estimated means at the site level but no directional bias) (Hamilton and Stamey 2007). This means that an increase in the number of points measured across the plot did not substantially improve measures of average surface depth in these vegetation types; however, it is likely that data confidence is stronger for the larger sample size. The extended method could be simplified by reducing the number of points assessed for fuel depth, marginally reducing the overall assessment time; however, the numbers of points by confidence level requires further research.

Comparison of the existing and extended hazard tables using data from the existing method

Variability in the hazard scores (e.g. differences of up to three hazard scores for near-surface and elevated fuel) derived using the existing vs extended hazard tables (with the same fuel property input data) likely highlight the subjective nature of the existing hazard tables. Clear benefits of using the extended tables include improved consistency in hazard score assignment and more flexible data processing, allowed for by the distinct fuel property categories. For instance, the use of terrestrial laser scanners for gathering precise fuel data (e.g. cover) is becoming an attractive complementary data collection technique (Wallace et al. 2020, 2022) and the extended tables provide a defined pathway for an automated system to translate such data to fuel hazard scores. Post-processing fuel property data could also allow assessors more time to collect quality fuel property information in the field rather than spending time deriving on-site hazard scores. This would also encourage the storage of the granular fuel data, resulting in a valuable data pool for further research. Additionally, the extended tables can be easily updated, and data re-analysed with advances in our understanding of how fuel properties contribute to fire behaviour.

Comparison of fuel hazard scores derived from the full implementation of the existing and extended assessment methods

There were large differences in fuel hazard (up to three hazard scores) between the existing and extended assessment methods when the detailed data (i.e. fuel properties) were combined into broad hazard scores per strata (i.e. low, moderate, high, very high and extreme). This highlights that initial differences between existing and extended fuel properties (less than one hazard score; Figs 2, 5) were exacerbated by the hazard tables and highlights the importance of reducing subjectivity at this stage. Differences in assessment outcomes could have far-reaching consequences when there is a requirement for accurate data. For instance, hazard assessments have been used to quantify long-term impacts of fuel management (Martorano et al. 2021; Pickering et al. 2022) and to develop fire behaviour models (Cheney et al. 2012). Inaccurate detection of spatial and temporal changes in fuel could propagate uncertainty in determining the effectiveness of fuel management and when making fire spread predictions. Furthermore, substantive over- or underestimation of fire spread predictions across a landscape could lead to erroneous communication of fire risk, which has the potential to undermine operational decision making and erode public confidence in emergency warnings (Mileti and Peek 2000; Ripberger et al. 2015; Benali et al. 2016).

Inconsistency in the existing assessment has been attributed to imprecision in visually estimating fuel properties and different interpretations of the hazard scores between teams of assessors (Watson et al. 2012; Spits et al. 2017). Extensive field training helps reduce variability among assessors (Lendon and Lamacraft 1976; Kelly et al. 2011). Our assessors attended a training day where the repeated assessment of each fuel element (at multiple points in the extended method) allowed for intense discussion of the definitions of each fuel stratum, allowing deeper thinking about the differences between fuel strata, with one assessor suggesting that this enhanced their visual estimation of fuel properties in the existing method. Additionally, field assessors in this study were experienced fire practitioners, and the same assessors were used across most plots, which likely enhanced their ability to apply the existing assessment consistently. Larger differences between the existing and extended methods might have occurred if there had been more variability in assessor knowledge and training.

The inclusion of tree basal area in our extended method aimed to improve the representation of plot-level bark fuel quantity. However, there was minimal advantage in this inclusion for bark hazard assessment in the Woodland Heath fuel type (misclassification of only one hazard class occurring 35% of the time; Fig. 9), which suggests the existing assessment of bark is adequate for determining bark hazard in this fuel type. This finding should be tested across a wider range of vegetation types with differing ratios of bark types and stem densities before its exclusion as an ancillary measure for hazard in the extended method. Furthermore, the collection of such data is likely useful for purposes beyond fuel hazard. For instance, information about bark quantity potentially gives insight about firebrand density and ease of containment when conducting a prescribed burn.

Implications

Broad visual assessments (i.e. the existing method) provide a time-efficient method for collecting data, but at the expense of data detail and quality. The extended method provided multiple ways to reduce subjectivity (although we note the need for further work to confirm repeatability) but took three times as long to implement in the field, making it inappropriate in some contexts, e.g. when assessing fuel in the path of a wildfire. However, it provides a good alternative when consistency is a priority and additional time can be afforded – for example, in monitoring and research programs for detecting changes in fuel over time or between treatments. In essence, the choice of method for fuel assessment should be aligned with the purpose of the assessment (Duff et al. 2017).

The extended method could also be used to improve understanding of the existing method and the consistency of its application. For example, it could be used within training programs for field assessors. Field assessors using the extended method in our study reported having a better understanding of the distinction between near-surface and elevated fuel strata, potentially making their assessments more consistent beyond the training period. In addition, some aspects of the extended method could be integrated into the existing method, such as systematically moving along real or imagined transects to enable estimates of cover to better represent conditions across the whole plot.

The extended method offers greater data flexibility, which is advantageous in the context of monitoring and research. Direct measures of fuel properties captured as part of the extended method (e.g. percentage cover, fuel height) could be used to parameterise existing (Cruz et al. 2015) or new fire behaviour models. Additional metrics could be added with relative ease where they are needed to achieve a purpose, e.g. to parameterise a particular fire behaviour model. The extended method will make it simpler to integrate fuel data derived from remote sensing (e.g. via terrestrial laser scanners) into future fuel hazard assessments by removing subjectivity from the hazard score tables. The direct measures of fuel would also likely be more harmonious with co-assessment of measurements to represent other vegetation values including habitat (e.g. Rode et al. 2013) and carbon stocks (e.g. Bennett et al. 2014), and so the extended method lends itself to better integration of fuel assessments with other ecological surveys.