Nationally consistent mapping of wildland fuel types across Australia using satellite-derived vegetation structural data

The Bushfire Fuel Classification typology

The focus of this work is to generate layers of fundamental fuel structural attributes and then to convert these into fuel type layers using an existing fuel typology. Although any structurally based typology could be used, we used the BFC (Hollis et al. 2015; Cruz et al. 2018). This typology is a hierarchical classification of wildland fire fuels developed in close consultation with stakeholders to meet the requirements of Australian fire management applications (Hollis et al. 2015). At its basis, the classification integrates life form (e.g. trees, shrubs, grass), vegetation height and foliage cover characteristics, as per Specht (1970) and Carnahan (1990), to derive a structural representation of fuel types. The hierarchical approach comprises three distinct tiers that allow for descriptions of fuel types for a range of applications and data accuracies. Top-tier fuel types give a broad overview of the overarching fuel type (e.g. open forest, woodland, shrubland) but do not have enough detail for fire behaviour modelling in a number of fuel types. The mid-tier fuel types are nested within the top-tier group and provide information on understorey fuels, allowing the most appropriate fire behaviour model to be used in each fuel type. The bottom-tier fuel types were envisioned to provide detailed static fuel information; yet, as fuel is temporally dynamic, this bottom tier can be replaced by models of fuel structure dynamics (e.g. fuel load accumulation with time since the most recent fire (Watson 2011)).

Input variables

The approach to derive the fuel type layer relied as much as possible on the use of current remote sensing data. For selected fuel types for which remote sensing products did not offer enough discriminatory power, we relied on existing spatial layers relating to land use and vegetation type. Table 1 lists the input variables required for the rulesets used to generate the BFC fuel type layers. More detailed descriptions of the remote sensing-derived variables are given below. Variables in Table 1 that were obtained from existing spatial sources and used directly without further modification (noted in the Table 1 as ‘Used as per source’) are not further described here. Any input data not already in the Australian Albers GDA94 projection were reprojected and resampled with a 90 m grid resolution. The nearest neighbour resampling method was used when the original input resolution was higher than 90 m, and bilinear interpolation was used for lower resolutions.

Remote sensing

Remote sensing data from multiple space-borne sensors were used to characterise foliage cover and height characteristics, including the vertical arrangement of cover, and its partitioning by vegetation type: woody plants (trees and shrubs) and grasses. Remote sensing was also used to characterise specific arid-region grassland types, such as spinifex (Triodia sp.), as well as water bodies, bare ground and building footprints. The approach integrated data from numerous satellite-based sensors (Table 2).

Existing land use, vegetation and plantation data

Several available data layers were used to identify fuel types and features that could not be discriminated through available remote sensing data. The extent of industrial forest plantations was obtained from the National Plantations Inventory (NPI) (ABARES 2022) dataset. We used land use information from Australian Land Use and Management Classification System (ALUM) version 8 (ABARES 2021) to describe the extent of wetlands and of production agriculture (namely croplands, pasture and horticulture). Information from the National Vegetation Information System (NVIS) version 6 (DCCEEW 2020) major vegetation groups was used to describe the extent of sedgelands. Data currency for plantations is 2021, for land use is 2016 and for native vegetation (NVIS) is 2020.

Fuel type classification rules

Several rulesets were developed to derive the BFC fuel typologies from the underlying vegetation structural information. The rules are applied sequentially. For the top-tier BFC classes (Table 3), the ruleset aims to first classify native vegetation using data from vegetation height, and woody and grass cover. Classification of non-burnable (e.g. water, built-up, wetlands) and agricultural (e.g. horticulture and croplands) areas, which were mostly derived from existing land use databases, was conducted last. For open vegetation systems, such as woodlands and open shrublands, where the drivers of fire propagation are not related to the overstorey component, there is an interest to apply the BFC mid-tier classification that defines the type of understorey fuel. For these open systems, the understorey can be shrub, grass, spinifex or sparse (Sullivan et al. 2012). The BFC mid-tier ruleset is shown in Table 4 and is applied only to the pixels previously classified as woodland, low woodland or open shrubland.

Validation of fuel type classification

Ground truth data

Evaluation of the generated BFC fuel type layer was undertaken using existing field data obtained from several sources (Table 5 and Fig. 2). Field data for forests were largely obtained from published studies that reported vegetation type, plot location and, in some cases, forest structural variables such as overstorey cover and height. Overall, a total of eight studies containing forest information were selected for the evaluation, comprising 469 field plot locations. Not all these studies explicitly reported canopy height and cover, which are required to define the BFC type (as per Table 3). Some studies only provided environmental or ecological forest type descriptors (such as dry forest or wet sclerophyll forest). For such cases the height and/or cover of the forest had to be inferred from the given descriptions. Given the subjectivity introduced by inferring structure, we further separated the forest evaluation data into those reporting structure (n = 331) and those where structure was inferred (n = 138).

Table 5.Data sources used in the evaluation of the BFC fuel type layer.

Source	Number of observations	Vegetation type as described in study
Forest environment with reported forest structure
Burrows et al. (2023)	70	Tall open wet forest
Cawson et al. (2018)	98	Wet forest
Hollis (unpublished)A	125	Open forest
Furlaud et al. (2021a)	38	Tall wet forest
Forest environment with inferred forest structure
McCaw (2011)	67	Open forest
Furlaud et al. (2021b)	17	Tall wet forest
Slijepcevic et al. (2018)	7	Forest (wet, moist, dry)
Sparrow et al. (2021)	47	Forest – rainforest, wet, open
Non-forest environment
Sparrow et al. (2021)	282	Woodland, shrubland, grasslands

^A Study focused on Eucalyptus marginata and Corymbia calophylla-dominated forests of the southwest Western Australia.

Fig. 2.

Spatial distribution of validation sites by vegetation type. See Table 5 for data sources.

For non-forest sites, we used information from the Terrestrial Ecosystem Research Network (TERN) Ecosystem Surveillance AusPlots network (Sparrow et al. 2021). TERN AusPlots is a national network for native vegetation monitoring plots. We used the structural and floristic description of the vegetation, plus photographic records to classify each plot according to the most suitable BFC fuel type. Only plots with cover and height data were selected for evaluation, for a total of 282 locations.

Native vegetation structural properties

The first step of the process of classifying vegetation into structurally defined fuel types was to spatially estimate the underlying vegetation structural attributes. The derived woody cover profile defines a three-dimensional vegetation distribution (Fig. 3a), which, when combined with grass cover (Fig. 3b) and woody vegetation canopy height (Fig. 3c), allows the definition of the dominant fuel type and understorey fuels (Fig. 3d).

Fig. 3.

Variation of vegetation structural characteristics along a 55-km long transect (black line) across the Australia Capital Territory (boundary in yellow). (a) Woody vegetation cover vertical profile along the transect, grouped in three height segments: <5, 5–10 and >10 m; (b, c) grass cover and canopy height, respectively, along the transect and (d) major vegetation formations and urban areas around the ACT. The western end of the transect has coordinates 35°29.483′S and 148°43.088′E and with the eastern end being located at 35°29.444′S and 149°15.510′E.

Fig. 3d illustrates how the approach captures the variability in fuel type across a 55 km long transect in south-eastern Australia covering a gradient of productivity and land use, with open forests (characterised by high woody cover, tall vegetation and relatively low grass cover) in the west of the transect transitioning to more open woodlands and pastures in the east (relatively high grass cover and discontinuous woody vegetation patches).

For open fuel types, such as woodland and open shrublands, the analysis aimed to produce mid-tier fuel types that identify the understorey fuel responsible for fire propagation: shrubby, grassy or hummock grass. Fig. 4c–e illustrates the process and the data layers used in identifying the woodland mid-tier understorey type along a 487 km transect in southern Western Australia (Fig. 4a, b). The analysis identified areas with overstorey cover varying between 5 and 30% and height above 5 m as woodland/low woodland (as per Table 3 ruleset). Within these areas, the cover of understorey vegetation (e.g. shrub (0–5 m height) as defined by Specht and Specht (1999), grass or spinifex) are weighted (as per Table 4 ruleset) to define the relevant understorey vegetation (Fig. 4).

Fig. 4.

Variation in the understorey of woodland fuel types along a 490-km long transect (black line in b) in Western Australia. (a) shows the distribution of Woodland and Low woodland BFC top‐tier fuel types across Australia, with the rectangle delineating the region represented in panel (b). (b) illustrates Woodland and Low woodland BFC mid‐tier fuel types in southern Western Australia, corresponding to the area highlighted in panel (a). (c, d) depict the overstorey and understorey cover profiles across the transect, respectively. (e) displaye the grass/shrub cover ratio along the transect, and (f) shows the derived mid-tier woodland fuel types across transect (legends are shown below plot (a)). The western end of the transect has coordinates 31°14.095′S and 119°18.586′E with the eastern end being located at 31°2.115′S and 124°21.160′E.

A simplified BFC framework

In implementing the BFC framework of Cruz et al. (2018), several modifications of the original framework became necessary, resulting in the simplified BFC framework shown in Fig. 5. Simplifications of the framework were made by aggregating a number of similar fuel type classes into one class. This was done partially because the remote sensing methods used could not provide the precision needed to differentiate between the structural characteristics of some adjacent classes. For example, we could not identify heights of shrubs lower than 2 m, so low shrublands and shrublands were fused into a single fuel type – Shrublands. It was noticed that there was likely too much detail and an inherent redundancy in the original fuel type classification, meaning – for practical purposes – some vegetation differences did not warrant separation into distinct fuel types. For the top-tier classification, we now use four aggregate fuel types, namely Closed forest, Low woodland, Open shrubland and Shrubland; each previously comprised two fuel types (Fig. 5). The fuel type map is shown in Fig. 6.

Fig. 5.

The Bushfire Fuel Classification top-tier fuel types as used in this study. Four fuel types were aggregated in comparison with the original Cruz et al. (2018a) top-tier fuel type matrix.

Fig. 6.

Australia-wide fuel types generated using the Bushfire Fuel Classification fuel typology. This shows the simplified top-tier BFC fuel types.

Table 6 summarises the relative areas of BFC (top-tier) fuel types for each Australian state and territory. The result shows that the distribution of fuel types varies substantially between the states. The relatively small jurisdictions of Victoria, Tasmania and the ACT have the highest proportion of forests. In contrast, the large jurisdictions of Western Australia, Northern Territory, South Australia and Queensland have the largest proportion of arid fuel types, such as hummock grasslands and open/sparse grasslands. The areal distributions of the mid-tier BFC types are shown in Table 7.

The top-tier fuel typologies allow fire behaviour models to be assigned for all fuel types except the open woody vegetation systems, such as woodlands and open shrublands. In these open woody systems, the mid-tier BFC types provides the extra information required to allocate fuel behaviour models. However, although the current BFC fuel type mapping can be used to determine which fire spread model should be used at a given location (Appendix 3), it does not provide the fuel parameter values also required by the models. Further modelling is required to generate these parameters, which is commonly done using the fuel types as a framework.

Evaluation of fuel type classification

Forest fuel types

The forest field data were separated into sites that explicitly reported structure and those for which structure had to be inferred. For the former (n = 331), the BFC classification produced an overall accuracy of 89% (293 correctly classified cases out of 331), with study-specific accuracies varying between 81 and 95% (Fig. 7a). For these sites, misclassification of fuel types was more commonly associated with errors in estimating vegetation height (e.g. a tall open forest being classified as an open forest owingto underprediction of forest height) rather than cover (e.g. an open forest being classified as a woodland). Only 7 out of the 331 forest validation sites were misclassified as non-forest, and all of these were labelled as Woodland (which is only one cover class away from a forest type; see Fig. 5).

Fig. 7.

Contingency table for fuel type classification accuracy for forest data. (a) Studies with known forest cover and height. (b) Studies with unknown forest structure but with broad vegetation description given. The percentage number above each contingency table reports overall classification accuracy.

The overall accuracy for the sites where forest structure was assumed (n = 138) was 87% (120 cases classified correctly out of 138), ranging from 71 and 100% for individual studies (Fig. 7b). Specifically, the accuracy was 76% for tall forest (n = 21), 100% for closed forest (n = 2) and 89% for open forest (n = 115). This 89% accuracy for open forests contrasts with 94% accuracy when the evaluation is against data with measured forest structure rather than inferred, highlighting the uncertainty introduced when making assumptions about forest structure. It is noted that the Tall forest and Open forest categories included two BFC fuel types each (see description of evaluation methods), which may artificially increase the accuracy of the classification.

Non-forest fuel types

The fuel type classification algorithm classified correctly 71% of the TERN non-forest plots (n = 282). Most misclassifications were in the arid systems with open shrublands, shrublands and woodlands being mapped as grasslands and hummock grasslands (Fig. 8). Importantly, the errors tended to be largely within one fuel type class, e.g. underestimation of height leads to a woodland being predicted as open shrublands, or overestimation of cover leads to woodland being predicted as forest. Approximately 7% of the woodland observations were mapped as forest. The system showed good discrimination between hummock grasslands and grasslands. A combined evaluation metrics of forest and non-forest sites is shown in Fig. S1.

Fig. 8.

Contingency table for fuel type classification accuracy for non-forest data. The percentage number above the contingency table represents the overall classification accuracy.

The value of a remote-sensing approach to mapping fuel types

In Australian fire management reality, a practitioner aiming to estimate fire risk or behaviour potential for a given situation or location can be confronted with at least three distinct fuel classification schemes, with conflicting fuel types and fire behaviour potentials. The existence and adoption of nationally consistent and periodically updated fuel type data are needed for correspondingly consistent national-scale fire danger forecasting and local fire behaviour prediction and risk assessment. Accuracy in fuel type data is also critical, as errors due to misclassification of fuel type lead to the wrong fire spread model being applied, resulting in errors several times larger than errors associated with incorrect fire spread model fuel inputs.

The approach used herein to develop up-to-date, national fuel layers brings together distinct space-borne satellite sensor types (multispectral, LiDAR, radar) with pre-existing geospatial data. Our approach aimed to minimise the use of pre-existing information (usually pre-mapped, ecologically based vegetation types); such data were only used where specific floristics and growth forms could not be determined from remote sensed products (such as the Sedgeland fuel type), or for specific land uses (such as Horticulture).

The remote sensing basis of the fuel type layers ensures consistency across Australia, something that is not present in current vegetation and fuel type maps (e.g. Matthews et al. 2019), which contain drastic changes in fuel types across jurisdictional borders. Our modular approach also enables continuous updates that account for changes in fuel types due to events such as land clearing, tree and shrub encroachment into grasslands and the occurrence of high intensity, stand-replacing fires. For such continuous updates to be possible, all inputs need to be available at an annual time step. This is already true for woody and grass vegetation cover (e.g. Donohue and Renzullo 2025), burn area (e.g. Giglio et al. 2021) and burn severity (e.g. Gibson et al. 2020) mapping. The absence of annual height data is a current limitation.

The study results also highlights the value of combining multiple types of remotely sensed data to better describe fuel structure, especially the differentiation between grass, shrub and hummock (spinifex) understories in open woody systems. This is essential information for the accurate use of fire behaviour models (Cruz et al. 2015a). Such integration of diverse and new sensor technologies is an active area of research in the vegetation mapping community. Numerous recent studies have combined optical imagery, radar imagery and/or LiDAR data to generate improved fuel type maps, commonly with some machine learning component. For example, space-borne LiDAR data, such as from GEDI, have recently been utilised in mapping vegetation vertical profiles (Wang et al. 2024). Similarly, several studies have integrated vegetation phenological signatures from optical and radar satellite imagery with GEDI-based height information to estimate structural and floristic attributes (Schlund et al. 2023; Burns et al. 2024). Mihajlovski et al. (2023) demonstrated that combining airborne LiDAR with optical and SAR (Synthetic Aperture Radar) data significantly improves fuel classification in Mediterranean forests. At a regional scale, Chrysafis et al. (2023) employed fused Sentinel-1, Sentinel-2 and DEM (Digital Elevation Model) derivatives to classify fuel types, highlighting the value of SAR and topographic information. Earlier work by García et al. (2011) integrated multispectral imagery with airborne LiDAR using (Support Vector Machine) SVM to map fuel types with enhanced accuracy. Similarly, Domingo et al. (2020) showed that fusing Sentinel-2 and (Airborne Laser Scanner) ALS data improved post-fire fuel type classification in Mediterranean regions, emphasising the importance of structural and spectral information. Most of these models, however, are fuel type-specific or have been regionally trained, and so are difficult to incorporate into an Australian fuel type model where diverse fuel types exist and training data are sparse. The value of the method presented in our work is that it is generic within the Australian context (not restricted to any particular fuel type), requires no training or calibration, and represents one, nationally consistent method.

We intentionally mapped the primary vegetation structural attributes (from which fuel type classes can be derived), rather than mapping fuel types directly. Additionally, we also used a physically based ruleset for combining structural attributes. This approach, only made possible with remote sensing data, has four main advantages. The first is that any structurally based fuel type typology can be populated, as we have done here with the BFC layer. To populate structural-floristic typologies – such as the AFDRS – requires the addition of floristic data layers (which are far more challenging to derive using satellite imagery). Second, nuanced variations in classifications can be easily implemented for specific purposes – for example, the cover threshold that divides woodlands and open forests can be changed to better fit local needs and expectations. Third, fuel type maps can be updated to incorporate changes in structure caused by land use change (e.g. forest or woodland clearing for pasture or cropland) and disturbance–regrowth cycles. Fourth and finally, this approach, being based on physical rulesets, provides a methodological transparency useful for decoding model error sources. This contrasts with currently popular approaches that rely on various forms of machine learning, which mask links between variables. Such models are yet to routinely deliver more accurate results than statistically simpler and logical rule-based models (e.g. López-De-Castro et al. 2022; Alipour et al. 2023; Sismanis et al. 2023). The accuracy of such machine learning approaches is inherently limited by the quality and quantity of training data, and model applicability is generally limited to the spatial and temporal extents of the training data (Hengl et al. 2018; Nikparvar and Thill 2021).

The BFC fuel type layer

The BFC framework provides a fine level of spatial and structural detail, as demonstrated by the variation evident in fuel types along the transects in Figs 3 and 4. The approach of using remotely sensed structural data to populate the BFC layer means we inherently capture the landscape-scale variation in fuels caused by soils, topography, drainage, disturbances and land management. The ability to identify the understorey fuels responsible for fire propagation in woodland and shrubland systems ensures appropriate fire behaviour models are applied to these vegetation types. For example, fire spread predictions for a woodland with a shrub understorey will rely on the generic fire spread model of Anderson et al. (2015) for shrubland fuels, whereas for a woodland with a spinifex understorey, the most appropriate model is that developed by Burrows et al. (2018). Although no fuel characteristics are associated with the fuel types in the current description of the BFC, there is a link between BFC fuel types and associated fire behaviour models (see Appendix 3). This association determines which fuel characteristics need to be estimated to apply the relevant fire behaviour models.

Another strength of the BFC framework is that it is based on the widely used vegetation classification scheme of Specht (1970). This maximises the relevance of the fuel type layers – as well as the underlying structural layers – to other applications. One example is linking fire behaviour modelling with biodiversity and ecological modelling (Loudermilk et al. 2022). Another example relates to land surface parameterisations within Earth system models (ESMs). These generally represent vegetation using Euro-centric categories, such as deciduous forest and needle-leaf forest (e.g. Clark et al. 2011; Haverd et al. 2018). Being physically based, the BFC fuel types (or at least their underlying structural data) can provide the basis for improving vegetation parameterisations of the various ESMs that are used to predict the effects of climate change in Australia. This would provide greater consistency between fire and climate modelling systems, improving interoperability, and would arguably improve predictions of the multiple feedbacks between vegetation, fire and climate. One further strength of the BFC typology is that it is equivalent to the AS3959 fuel classification currently used by Australian fire agencies and private industry to guide the construction of buildings in bushfire-prone areas.

Herein, we applied a slightly revised version of the original BFC top-tier fuel types (see Fig. 5, Cruz et al. 2018). This was done for two main reasons: (i) the inability of current remote sensing products to discern differences in low woody vegetation heights; and (ii) because it was felt that there was some redundancy in a number of fuel types. In this revision (see Fig. 5), low closed forest and closed forest were combined into Closed forest, whereas low woodland and tall open shrublands formed Low woodland. Tall shrubland and tall closed shrublands were merged into Tall shrubland and an Other plantation class was added. Shrublands and low shrublands were integrated into Shrublands and the Sparse shrublands class was removed. Hummock grassland became a single class, and Grassland and Closed grassland were merged. We advocate that the new suggested fuel type definitions be adopted in future work.

Model error and future improvements

The remotely sensed height and foliage cover data used herein have known limitations. It is acknowledged that the GEDI product (Lang et al. 2023) is unable to provide estimates of woody vegetation height below 2 m, and there are substantial uncertainties in the estimation of height between 2 and 5 m. This limited the classification of shrub fuel types and low woodlands, and is one of the reasons why we pooled the Tall open shrubland and Low woodland into a single fuel type – Low woodland. Research into canopy height mapping is advancing quickly owing to the increasing availability of LiDAR data and machine learning techniques, and so this limitation is expected to ease in the near future.

There are three known limitations in the MODIS-based foliage cover data used herein. Firstly, the foliage cover of pure grasslands is systematically underestimated (Donohue and Renzullo 2025). Although this does not affect the ability to differentiate between woody and grass vegetation types, it means that some Grassland areas are likely to be incorrectly mapped as Open grasslands (Fig. 5) owing to this bias. Secondly, the nature of cover errors in sparsely vegetated systems means that even small absolute errors equate to large relative errors. This makes it difficult to accurately differentiate between grassy and woody fuel types when total woody foliage cover varies between 5 and 10%. Thirdly, estimates of woody foliage cover do not differentiate between canopy and understorey foliage. For this, we used the modelled vegetation height-cover profiles of Scarth et al. (2019), meaning that the accuracy of our BFC classification (especially the mid-tier classes) is partly determined by the accuracy of that height-cover product.

The evaluation results obtained for the BFC fuel typology against independent data, namely ~89% for native forest fuel types and 71% for other fuel types, are encouraging, especially when compared with other published results for the accuracies of large-scale fuel type maps (e.g. Rollins et al. 2004; Krasnow et al. 2009; Keane et al. 2013). It is important to note that the evaluation data focus on native vegetation, in particular forests and woodlands where the system is working best. We note that the lower and sparser the vegetation, the higher the uncertainty in the remote sensing data. The evaluation data do not adequately cover all fuel types, and so it is unclear how well the fuel type identification ruleset captures some of the fuel types, such as low shrubs. It is expected that the fuel types derived from NVIS and ALUM layers will have a higher level of uncertainty than that obtained for native vegetation fuel types mapped from remote sensing structural data.

A known limitation of our fuel type layers is that they comprise source data with varying spatial resolutions, ranging from 30 m up to 250 m for vegetation and 500 m for bare ground. Coarse-resolution satellite data are no less accurate per se than fine-resolution data for vegetation and fuel type mapping over homogeneously vegetated areas (in fact, fine-resolution imagery can become less accurate if the resolution is below the minimum mapping scales embedded in the conceptualisation of vegetation and fuel type classes). It is when the boundary locations between fuel types are required with high precision, such as within WUI areas, that the multi-scale nature of our products becomes limiting. It is also important when assessing fire risk for or planning fire operations around sensitive landscape features, such as fire-sensitive vegetation patches. Hence, a key enhancement of the method presented here is to generate foliage cover (and other cover-related variables) at 30 m resolution using a combination of Landsat and Sentinel-2 imagery.

Aside from addressing the above-mentioned limitations, arguably the next advance needed in Australian fuel type mapping relates to understorey structure. Here, we only mapped woodland and open shrubland understorey and only using four basic types – shrub, grass, hummock and ‘sparse’ (this, however, is a novel advance in itself). Ideally, understorey mapping would extend to forests and would be mapped in terms of the number of layers present, and their height and cover. These attributes can be mapped from airborne LiDAR (e.g. García et al. 2011 and Marino et al. 2016) – but, being from an airborne source, these data are only available over limited areas, especially in the Australian context. They can also be mapped from satellite LiDAR – but only at points across the continent. The greatest potential lies in a model that combines these two sources with other satellite imagery and environmental layers into a biophysical modelling framework (such as the ecohydrological equilibrium framework of Specht (1972)).

Comparison with Australia’s contemporary fuel type mapping

The fuel typology currently used operationally by the states and territories in Australia for fire danger rating is the AFDRS typology. This was initially developed from a combination of NVIS (for native vegetation) and land use (for non-native vegetation) spatial datasets for much of Australia with more detailed state and territory data used for specific fuel types, such as forests (Matthews et al. 2019). Hence, this is a structural-floristic typology. To date, floristics information (vegetation species and associations) cannot be routinely delineated from satellite imagery nationally, and so (for now), the AFDRS system is dependent on traditional state-level vegetation type mapping. This has a cascade of implications. One is that the quality of the fuel type mapping is largely determined by that of the underlying vegetation mapping – which is highly variable in its accuracy, consistency and currency at the national scale, as well as at the local scale over the more remote regions of Australia. As traditional mapping programs are expensive, neither the vegetation type maps themselves, nor the techniques used to derive them, are updated frequently. Similarly, until space-based detection of floristics becomes mainstream, advances in satellite-based vegetation mapping are difficult to incorporate into the AFDRS system. This means, among other things, that the advantages of the potentially high latency of satellite products cannot be realised within the system. If the fundamental fuel structural attributes could be perfectly quantified, there would be no need for fuel typologies to incorporate floristic information (as floristics serve only as proxies for the multiple effects of underlying structural attributes associated with specific vegetation and fuel assemblages). Ideally, Australia’s operational fuel typology needs to be flexible enough to accommodate potential future improvements in satellite-based (and airborne-based) vegetation structural mapping, which will potentially require a transition away from a floristic-based typology.