Operational Fuel Models Map for Atlantic Landscapes Using ALS and Sentinel-2 Images

doi:10.21203/rs.3.rs-2582490/v1

Download PDF

Research Article

Operational Fuel Models Map for Atlantic Landscapes Using ALS and Sentinel-2 Images

https://doi.org/10.21203/rs.3.rs-2582490/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 17 Oct, 2023

Read the published version in Fire Ecology →

You are reading this latest preprint version

Background:

In the new era of extreme wildfire events, new fire prevention and extinction strategies are emerging using software that simulates fire behavior. Having updated fuel models maps is critical in order to obtain reasonable simulations. Previous studies have proven that remote sensing is a key tool for obtaining these maps. However, there are many environments where remote sensing has not yet been evaluated in an operational context. One of these contexts are Atlantic environments. In this study, we describe a remote-sensing-data-based methodology for obtaining an operational fuel models map for an Atlantic-vegetation-covered area in Galicia (Northwestern Spain). We used Sentinel-2 images and ALS (Aerial Laser Scanner) data.

Results:

We have developed a methodology that allows to objectify the fuel models mapping for this type of environments since. For that we obtained the correspondences between the vegetation of the area and Rothermel fuel models. Additionally, since the methodology relies in remote sensing data, it allows us to obtain upgradable fuel models maps. For the study area, we obtained a map with high accuracy metrics. The accuracy of the supervised classifications involved in the mapping ranges between 70% and 100% (user’s and producer’s accuracies).

Conclusions:

The obtained methodology and the upgradable fuel models map will help to improve fire prevention and suppression strategies in Atlantic landscapes, aiding to shift towards more modern fire-simulation-based mitigation strategies.

fuel models map

Galicia

remote sensing

simulation software

wildfire

wildfire prevention.

Large wildfires are an increasingly frequent phenomenon across the globe 2018; Linley et al. 2022). The 2017 and 2022 wildfire seasons are examples of this trend (Nature, 2017; Rodrigues et al., 2023). In 2017, a total of 622,000 ha burned in Chile, Russia, the US, Canada, Greenland, and the Mediterranean (Nature, 2017). Similar examples, such as the recent wildfires recorded in California, have arisen in subsequent years (Cal Fire, 2022). The Australian 2019/20 wildfire season is another example: wildfires affected, either directly or indirectly, nearly 80% of Australian citizens (Biddle et al., 2020). These large wildfires have not only severe direct and indirect impacts on human lives (Nature, 2017; Tedim et al., 2018; Biddle et al., 2020; Cal Fire, 2022), but also significant ecological and socio-economic implications (Li et al., 2022; Taboada et al., 2021; Reilly et al., 2021; Robinne et al., 2018; Wang et al., 2021. According to different scientific records, these events represent a trend that is expected to continue in the future (Dong et al., 2022; Senade-Rivera et al., 2022; Flannigan et al., 2013).

Multiple wildfire events have shown that traditional fire extinction and prevention strategies are insufficient for containing large wildfires (McWethy et al., 2019; Moreira et al., 2020; Romero-Vivó et al., 2019; Quílez et al., 2020). Therefore, new approaches to fire suppression and prevention are emerging (Moreira et al., 2020; McWethy et al., 2019; Romero-Vivó et al., 2019; Quílez et al., 2020). In the Mediterranean basin, where large wildfires pose a great threat, the concept of strategic management areas has been proposed to address wildfire prevention (Costa et al., 2011; Romero-Vivó et al., 2019; Dupuy et al., 2020; Royé et al., 2019; Oliveira et al., 2018). For instance, in Spain, the recent law, "Real Decreto-ley 15/2022-August 1st ", mandates the identification and design of annually-updated strategic management areas throughout the country (RDL 15/2022).

Among the new approaches to wildfire prevention, simulation software is becoming paramount since it allows for the prediction of wildfire behavior in a certain area (Simons, 2013; Romero-Vivó et al., 2019; Moreira et al., 2020; Quílez et al., 2020; Iglesias et al., 2022). Wildfire simulation software uses several types of input data to determine rate-of-spread, intensity, flame length, and affected area, among other variables. The input data mainly include weather conditions, topographic information, and fuel models (Ellsworth et al., 2022; Rothermel, 1983). Fuel models consist of sets of defined vegetation types and structures that can lead to a significantly different wildfire behavior in the case that a fire should occur. These models are the primary input when it comes to the reliability of the simulations since fire behavior is highly dependent upon vegetation type, and on the availability as well as the vertical and horizontal distribution of biomass (Jain et al. 2020; Gale et al., 2021). Accurate and up-to-date fuel models maps of a study area are, thus, essential for effective wildfire prevention and the identification of strategic areas.

According to Simons et al. (2013) most of the existing simulation software relies on Rothermel fuel models (Rothermel, 1983), or on Scott and Burgan fuel models (Scott & Burgan, 2005). Traditionally, identifying the corresponding fuel model for the specific vegetation types present in an area has been done through field surveys (Anderson, 1982; Arroyo et al., 2008). However, using field surveys to obtain updated fuel models maps for large areas can be arduous and potentially not cost-effective (Gale et al. 2021).

In recent years, several methodologies based on remote sensing techniques have been presented in specialized scientific publications. Gale et al. (2021) recently published a review on this topic. They highlighted the existence of two main approaches: the use of remote sensing data for the characterization of different fuels and the use of remote sensing data in the automatic identification of predefined fuel models. Of the studies analyzed, only 30% used the latter approach (Gale et al., 2021). Some examples are Domingo et al. (2020) and Huesca et al. (2019). They used remote sensing data to identify, at the pixel level, models included in the Prometheus fire model, a model that was developed specifically for Mediterranean ecosystems (Arroyo et al., 2008). In order to use the obtained fuel maps operationally in simulation software, they designed a procedure to assimilate the Prometheus models to the Rothermel fuel models. Another example is the study of Marino et al. (2016) which developed a fuel models map for the Canary Islands. They used ALS data and Landsat images to create the map, using the Canary Islands Fuel Model (CIFM), a local adaptation of the Rothermel fuel models (Marino et al., 2016). Palomino and Rodríguez y Silva (2021), used LiDAR data to characterize, identify and map fuel models according to the UCO40 fuel models, thus adapting Scott and Burgan’s fuel models to a specific Mediterranean environment (Andalucía). An example of a different type of study is the work of Heisig et al. (2022) that, in central European forests, uses remote sensing, field work, and modeling techniques to develop a customized fuel map that can be used with the FlamMap software.

An important consideration inferred from the Gale et al. (2021) review is that most of the studies analyzed focus on Mediterranean forests and woodlands. These kinds of studies, as is pointed out by Cardil et al. (2021), should be performed in a variety of different environments so that wildfire simulation software can be used for operational decision-making in those areas as well (i.e., designing AEGs). For instance, Atlantic climates express certain singularities that require specific analysis: vegetation grows faster than in a Mediterranean climate leading to a faster accumulation of large fuel loads which can lead to extreme fire behavior (Arellano et al. 2017). This is the case in Galicia, one of the Spanish regions most prone to wildfires (de Diego et al., 2021). Various efforts have been made to identify and characterize the most common fuel models in Galicia; the most prominent is the fuels photo-guide elaborated by Lourizán CIF (fuels photo-guide from now on) (Arellano et al., 2017). Other research has assessed fuel variables through remote sensing; examples of such studies are: Alonso-Rego et al. (2021), Alonso-Rego et al. (2020), Fidalgo-González et al. (2019) and Arellano-Pérez et al. (2018). However, methodologies for producing updated fuel models maps are still need, since these maps are essential for wildfire simulation.

In this study, we describe a remote-sensing-data-based methodology for obtaining an operational fuel models map for an Atlantic-vegetation-covered area in Galicia (Northwestern Spain). We used Sentinel-2 images and ALS (Aerial Laser Scanner) data.

This study was conducted in the region of Galicia, Spain. This region represents only 6% of the total surface area of Spain, however it encompassed 22.51% of the total burned area in Spain between 2006 and 2015 (Gobierno de España, Ministerio de Agricultura, Pesca y Alimentación, 2019). This trend has continued in subsequent years (López-Rodriguez et al., 2021; Consellería Do Medio Rural, 2022) with several large wildfires and extensive burn areas reported (Rodrigues et al., 2023; Consellería Do Medio Rural, 2022). For this reason, improved wildfire prevention plans are needed. For this research a pilot area was selected, it is shown in Fig. 1.

Galicia has an Atlantic climate (Royé et al., 2016). According to the Spanish Forest Map, approximately 69% of the area studied is covered by forest or shrublands (MITECO, 2022). The main tree species present in the study area are Pinus pinaster, Pinus radiata, Eucalyptus globulus and Quercus robur, along with riparian forest tree species (Alnus glutinosa, Fraxinus excelsior, and Salix spp. among others) (MITECO, 2022). The main shrubs in the area are Ulex europaeus and Erica spp. Obtaining fuel models maps through remote sensing is especially challenging in this region due to the rapid rates of vegetation growth (Arellano et al. 2017), which, in conjunction with an active forestry sector, lead to highly dynamic land cover paradigms.

3.1.- Satellite Images

A time-series of Sentinel-2 images dating from the year 2019 was used in this study. Sentinel-2 is a constellation of two satellites (Sentinel-2A and 2B), each equipped with a Multispectral Instrument (MSI) (ESA, 2015). The MSIs can obtain multispectral images with 13 multispectral bands of different spatial resolutions, ranging from 10 m to 60 m. The bands’ specifications are presented in Table 1. Together, the satellites provide a revisit time of five days at the equator and two to three days at mid-latitudes.

Table 1

Sentinel-2 specifications.
Band	Central wavelength (nm)	Bandwidth (nm)	Spatial resolution (m)
B01 - Coastal aerosol	443	20	60
B02 – Blue	490	65	10
B03 – Green	560	35	10
B04 – Red	665	30	10
B05 – Near Infrared (NIR)	705	15	20
B06 – NIR	740	15	20
B07 – NIR	783	20	20
B08 – NIR	842	115	10
B8A – NIR narrow	865	20	20
B09 – Water vapor	945	20	60
10 – Shortwave Infrared (SWIR) (cirrus)	1375	30	60
B11 – SWIR	1610	90	20
B12 – SWIR	2190	180	20

Level-2A Sentinel-2 products were used. These products incorporate radiometric, geometric and atmospheric corrections (ESA, 2022). A total of 12 images was used in this study, one image per month. The images were download from the Copernicus Open Access Hub (ESA and Copernicus, 2023). For each month, the image with the least cloud cover was selected, with the stipulation that no image should have a cloud cover greater than 50%.

3.2.- Airborne Lidar Data

The LiDAR data were acquired in November of 2019. The point clouds have a mean spatial resolution of 7.2 points/m². The structure of the data consists of 7786 1 km x 1 km tiles that cover a total area of 7426.35 km² of coastal municipalities. The study area includes 378 tiles, corresponding to 6 different municipalities. These 378 tiles cover 366.25 km², which represents 1.2% of the total surface area of Galicia.

3.3.- Reference Data

Open-access high-resolution aerial orthorectified images were used as reference images for the land cover mapping verification process. They were obtained from the Spanish National Plan of Aerial Orthophotography, PNOA by its acronym in Spanish (MTMAU, 2021). In particular, two sets of images were used: a) images obtained in photogrammetric flights performed between the 30th of May and the 1st of September in 2017 (2017 PNOA); b) images obtained in photogrammetric flights performed between the 30th of May, 2020 and the 1st of September, 2020 (2020 PNOA). The 2017 PNOA images have a spatial resolution of 0.25 m and a georeferencing mean square error of ≤ 0.5 m (MTMAU, 2021). The 2020 PNOA images have a spatial resolution of 0.15 m and a georeferencing mean square error of ≤ 0.20 m (MTMAU, 2021). Both sets of images include 4 bands (Red, Blue, Green and NIR).

3.3.- Fuel Models

The fuel models that were used were the Rothermel fuel models (Rothermel, 1983) and the fuels photo-guide (Arellano et al., 2017). Table 2 presents a detailed description of the Rothermel fuel models, one of the most-commonly-used fuel models in the world (Ascoli et al., 2015; Vacchiano and Ascoli, 2015; Simons, 2013).

Table 2

Description of Rothermel (Rothermel, 1983) fuel models considered to be present in the study area.
Fuel Group	Code	Height (m)	Fuel load (Ton/ha)	Qualitative characteristics
Pastures	1	0.3	1–2	Dry, short grass that covers the ground. Shrubs or trees covering less than 1/3 of the surface.
	2	0,3 - <1	5–10	Grassy areas with shrubs or sparse trees that cover between 1/3 and 2/3 of the surface.
	3	> 1	4–6	Thick, tall grass (over 1 meter). Typical of savannas. Grain fields are representative of this model.
Shrubs	4	1–2	25–35	Dense shrubs or young trees. Horizontal and vertical continuity. Abundance of woody fuel.
	5	< 1	5–8	Dense, young shrubland.
	6	> 1	10–15	Shrubland similar to model 5 with remnants of broadleaf plants.
	7	< 2	10–15	Shrubland of highly flammable species less than 2 meters tall or the underbrush of pine forests.
Trees	8		10–12	Leaf litter in dense conifer or broadleaf forests (leaf litter forming a compact layer of less than 5 cm).
	9		10–12	Leaf litter in dense conifer or broadleaf forests, differing from model 8 in that it forms a spongy, loosely compact, lofty layer with a great deal of air in between the leaves. It is made up of long needles.
	10		30–35	Naturally-occurring woody debris, including large fallen wood due to storms, intense pests, or excessive forest maturity, with herbaceous vegetation and shrubbery growing among the woody debris.
	11		30–35	Lightly wooded or cleared forest with light debris (less than 7.5 cm in diameter), silvicultural or harvesting treatments, forming a loose, low-lying layer (about 30 cm high).
Trees	12		50–80	Predominance of debris over trees, heavier than in model 11, forming a continuous layer of greater height (up to 60 cm). More than half of the leaves are attached to the branches without being completely dry.
Forest Harvesting Residues	13		100–150	Large accumulations of heavy, bulky waste or harvesting waste (diameter > 7.5 cm) covering the entire ground.

The fuels photo-guide offers a sample graphic representation of the fuels present in Galicia and provides an original empirical fuel models classification for this region. For each fuel scheme it includes information about dominant species, mean vegetation heights and fuel loads, as well as a representative photograph. Additionally, it provides an estimation of fire behavior in each fuel scheme, including information such as the estimated speed of propagation and flame lengths for different wind speeds and terrain slopes under a dead fuel humidity scenario that represents moderate-low risk in Galicia. This fuel photo guide was elaborated using empirical models developed at the Lourizán CIF (a forestry research station). Table 3 summarizes the fuels photo-guide fuel types present in the study area.

Table 3

Description of Lourizán CIF (Arellano et al., 2017) fuel models considered to be present in the study area.
Fuel group	Code	Species	Average height (m)	Fuel load (Mg/ha)
Pastures	PI-01	Pseudarrhenatherum longifolium; Agrostis curtisii	0.22	7.3
	PI-02	Pseudarrhenatherum longifolium; Ulex europaeus	0.5	13.2
	Pa-01	Pteridium aquilinum	0.76	3.56
	Pa-02	Pteridium aquilinum	0.67	5.41
Shrubs	Eu-01	Erica umbellata; Pterospartum tridentatum	0.37	10.04
	Eu-05	Erica umbellata; Pterospartum tridentatum; Ulex breoganii	0.65	18.83
	Eu-06	Erica umbellata; Pterospartum tridentatum; Ulex breoganii	0.86	30.2
	Pt-01	Pterospartum tridentatum; Erica umbellata	0.65	14.1
	Ue-01	Ulex europaeus; Erica umbellata	0.42	15.61
	Ue-02	Ulex europaeus	0.67	18.52
	Ub-03	Ulex breoganii	0.41	27.08
	Ea-05	Erica australis; Ulex europaeus	3.26	18.9
	Es-01	Erica scoparia; Ulex breoganii	1.03	16.15
	Pt-05	Pterospartum tridentatum; Ulex europaeus	1.5	26.96
	Pt-06	Pterospartum tridentatum	1.51	28.53
	Pt-07	Pterospartum tridentatum; Pteridium aquilinum	1.95	35.48
	Ue-05	Ulex europaeus; Pteridium aquilinum	3	21.63
	Ue-08	Ulex europaeus	2.01	23.31
	Ue-10	Ulex europaeus	1.5	28.05
	Ub-02	Ulex breoganii	1.27	20.27
	Ub-04	Ulex breoganii; Pteridium aquilinum	1.37	33.41
	Cs-01	Cytisus scoparius; Pteridium aquilinum; Rubus ulmifolius	2.48	14.41
	Cs-03	Cytisus scoparius	4.17	21.89
Trees (deciduous and evergreen trees)	PpMB-02	Pinus pinaster (Monte Bravo); Understory: Erica umbellata; Ulex europaeus	0.5	13.62
	PpL-04	Pinus pinaster (Latizal); Understory: Ulex breoganii; Pteridium aquilinum	0.81	2.1
	PpL-09	Pinus pinaster (Latizal); Understory: Ulex europaeus; Daboecia cantabrica	1.21	12.34
	EgB-02	Eucalyptus globulus (Monte bravo); Understory: Brotes E. globulus; Pteridium aquilinum, Ulex europaeus	1.31	1.17
	EgL-08	Eucalyptus globulus (Latizal); Understory: Acacia melanoxylon	3.84	12.11
	QrL-04	Quercus robur (Latizal); Understory: Rubus ulmifolius, Pteridium aquilinum	0.53	2.8
	PpF-06	Pinus pinaster (Fustal); Understory: Pterospartum tridentatum, Erica umbellata	0.4	7.18
	PpF-05	Pinus pinaster (Fustal); Understory: Pteridium aquilinum	1.1	4.3
	EgL-03	Eucalyptus globulus (Latizal); Understory: Pteridium aquilinum, Ulex breoganii	0.78	4.5
	QrF-02	Quercus robur (Fustal); Understory: Pteridium aquilinum, Robus ulmifolius	0.28	1.66
	QrF-04	Quercus robur (Fustal); Understory: Ruscus aculeatus, Hedera helix	0.55	4.02
	BaL-01	Betula alba (Latizal); Understory: Pteridium aquilinum, Erica arborea	0.68	1.18
	EgL-09	Eucalyptus globulus (Latizal); Understory: Ulex europaeus, Pteridium aquilinum	1.14	12.84
	PpF-03	Pinus pinaster (Fustal); Understory: Pteridium aquilinum	0.71	1.59
	PpF-05	Pinus pinaster (Fustal); Understory: Pteridium aquilinum	1.1	4.3
	EgL-06	Eucalyptus globulus (Latizal); Understory: Erica cinérea, Erica umbellata	0.42	10.41
	EgF-03	Eucalyptus globulus (Fustal); Understory: Pteridium aquilinum	0.7	8.37
	EgF-05	Eucalyptus globulus (Fustal); Understory: Ulex micranthus, Daboecia cantabrica, Calluna vulgaris	0.43	11.64

The methodology aims to create an automatic process that uses remote sensing data to generate a complete, objective fuel map that can be used with wildfire prevention simulation software. The Rothermel fuel models were selected for this purpose since they are the most commonly used in simulation software (Simons et al., 2013). However, since the Rothermel Models are qualitative and were developed for a climate that is quite different from that of the study area, the correspondence between the vegetation structure and type and the Rothermel Models cannot be derived systematically using remote sensing. The fuels photo-guide was used for this purpose. It includes the most common vegetation types in Galicia, and they are characterized according to qualitative and quantitative parameters which could potentially be estimated for the study area using remote sensing data. Additionally, the guide provides wildfire behavior data for each model, which facilitates the identification of correspondences between the fuels photo-guide and the Rothermel fuel models. This process allows for the objective and systematic mapping of Rothermel fuel models using remote sensing data from the study area.

The methodology consists of five main steps: 1) identification of the vegetation types in the study area using Sentinel-2 images; 2) characterization of the vegetation height (Canopy Height Model, CHM) and structure (LiDAR metrics) for the study area using ALS data; 3) establishing correspondences between the identified vegetation types and structures and the fuel models from the fuels photo-guide; 4) establishing the correspondences between the fuels photo-guide models and Rothermel fuel models; 5) mapping of Rothermel fuel models. An overview of the method is shown in Fig. 2.

4.1.- Land Cover Mapping

According to the fuels photo-guide, the principle variable that determines the assignment of a fuel type to a plot or forest stand is its dominant species. Sentinel-2 images were used to obtain a land cover map of the study area specifying the principle species and groups of species. The target classes were defined by considering the vegetation in the study area, the aim of the study, and the authors’ previous experience in multispectral classification. These classes are detailed in Table 4. The map was created following the methodology described in Alonso et al. (2021). First, the images were downloaded and pre-processed to mask out the cloudy pixels from each image. This step was performed by applying the cloud mask provided by the Sentinel-2 Level-2A product.

Table 4

Legend of the land use map created using Sentinel-2 images.
Class	Description
Eucalyptus sp.	Adult individuals and young masses of Eucalyptus globulus.
Conifers	Adult individuals and young masses of conifers (mainly Pinus pinaster, Pinus radiata).
Broadleaves	Adult individuals and young masses of hardwoods (mainly Quercus robur and riparian species).
Crops and pastures	Crops or land covered by non-woody vegetation.
Shrubs	Areas of shrubs (mainly Ulex and Cytisus genera and ericaceae family).
Bare soil	Land covered by rocks and very small shrubs or non-anthropogenic, non-vegetated areas.
Anthropic	Built-up areas.
Water	Accumulations of water, both marine and continental.

Next, training data were collected through the photointerpretation of 2017 and 2020 PNOA images in order to perform a supervised classification. Polygons for each target class were manually delineated over the entire study region. Then, a supervised classification was performed for each image in the time series, which consisted of 12 images from 2019, one per month. The supervised classifications were performed using the random forest algorithm (Breiman, 2001). This step was carried out using the “randomForest” R package (Liaw and Wiener, 2002) with the default configuration parameters. The Sentinel-2 bands used in this step were the B02, B03, B04, B05, B06, B07, B08, B8A, B11, and B12 bands. All bands were resampled at 5 m using nearest neighborhood interpolation. Once single-date classifications were obtained, they were aggregated using the plurality voting decision criteria (Alonso et al., 2021; Lewiński et al., 2017). These decision criteria identify the class that most frequently occurs for each pixel throughout the time series and assigns that class to the pixel (Lewiński et al., 2017).

Finally, the map obtained was cross-verified. To do this, a random stratified set of points was distributed throughout the study area. The ground truth of these points was obtained through the photointerpretation of 2017 and 2020 PNOA images. Sentinel-2 images were used to provide supporting information, to address which PNOA image should ultimately be used to assign the land cover class. For example, if the 2017 PNOA image shows an area covered by conifer trees and the 2020 PNOA image shows that that same area is now covered by shrubs, Sentinel-2 images were used to confirm when the harvesting event took place, allowing us to determine if the land cover class for the year 2019 should be "Shrubs" or "Conifers". A confusion matrix was then constructed and accuracy metrics were calculated. The accuracy metrics calculated are as follows:

Overall Accuracy (OA): This value represents the proportion of correctly classified reference points. It is calculated by dividing the total number of correctly classified points by the number of reference points.
Producer’s Accuracy (PA): This value indicates how often real-world features are correctly represented on a classified map, or how likely it is that a given land cover of an area is correctly classified. It is the map accuracy. It can be computed for each category by dividing the number of correctly classified reference points for that category by the total number of reference points.
User’s Accuracy (UA): This metric represents the probability that the ground truth of a categorized pixel genuinely corresponds to that category. It represents the reliability of the map. It is calculated by dividing the total number of correctly categorized pixels in each category by the overall number of pixels classified in that category.
Kappa Index: This value compares the computed accuracy of the classification with the accuracy that would be computed randomly. It is determined by subtracting the accuracy that would be obtained through a random classification from the difference between the total accuracy (OA), divided by one minus the accuracy that would be obtained through a random classification.

4.2.- Structural Characterization Of Vegetation

According to the fuels photo-guide, the assignment of a fuel to a certain area is determined, not only by the species present in that area, but also by the height and structure of the vegetation. LiDAR data were used to characterize the vegetation height in the study area and to generate an approximation of the forest structure. The process consisted of two main steps: the first one, which includes noise removal, height noise removal, classification, and normalization, is oriented at preparing the data. The second step is the data processing; it includes canopy height modeling, computation of statistics and rasterization.

4.2.1.- Lidar Data Pre-processing

The first step of the LiDAR data pre-processing is noise removal, that is, the elimination of isolated points. This was performed using the “lasnoise” tool in the LAStools software (rapidlasso GmbH, 2014). Afterwards, the LiDAR data were classified in two classes: ground points and non-ground points. This was done with the “lasground” tool in the LAStools software (rapidlasso GmbH, 2014). The tool settings were adjusted to obtain an optimized performance. This step was needed for further normalization of the point cloud. The normalization was performed in order to transform the z coordinates. These coordinates, which originally represented the altitude of each point above the Earth’s reference surface (according to the official vertical Datum); after normalization, represent the height of each point above the ground. The normalization algorithm applied consisted of a nearest neighbourhood (KNN) interpolation with an inverse distance weighting (IDW). It was implemented using the "lidR" package in the R software (Roussel and Auty, 2021; Roussel et al., 2020).

The final pre-processing step was the vegetation height segmentation. Ground points and aerial noise were removed so that the data processing would focus solely on the vegetation points. For this step, all points below 0.5 m and above 60 m were removed from the point cloud.

4.2.2.- Lidar Data Processing

In order to estimate vegetation height and obtain information about the vertical structure of stands, several statistics were obtained from the normalized point cloud. Each was represented as a raster layer to facilitate its individual analysis, its interpretation using the reference images and its comparison with the other statistics.

First, a canopy height model (CHM) was obtained. The CHM is a raster layer where the digital value of each pixel corresponds with the height of the vegetation at that pixel. The vegetation height is estimated as the maximum value of the z coordinate, selected from among all of the points within the vertical prism defined by the pixel’s contour. These values were obtained using the “LidR” package available in the R software (Roussel and Auty, 2021; Roussel et al., 2020). The CHM was produced using a 5 m pixel size.

Additionally, a set of 36 statistics were calculated for the point clouds. Each of the statistics was computed using the normalized values of the z coordinates of the points; meaning that they ought to provide an approximation of the vertical structure of the point cloud. They were obtained and rasterized using the “LidR” package available in the R software (Roussel and Auty, 2021; Roussel et al., 2020). A raster layer with a spatial resolution of 5 meters was obtained for each metric. The complete list of metrics is detailed in Table 5. A detailed explanation of the metric “Cumulative percentage of returns in the X^th layer” can be found in Woods et al. (2008). In order to obtain these layers, the maximum height of the LiDAR data was divided into 10 intervals. The cumulative percentage of LiDAR returns was calculated for each interval. The final interval is excluded as the cumulative percentage of points in this interval is always equal to 1. Further information on the rest of the metrics can be found in Roussel et al. (2021).

Table 5

Metrics obtained from the normalized LiDAR point clouds.
ID	Metric	ID	Metric	ID	Metric	ID	Metric
B1	Maximum	B10	10th Percentile	B19	55th Percentile	B28	Cumulative percentage of returns in the 1st layer
B2	Mean	B11	15th Percentile	B20	60th Percentile	B29	Cumulative percentage of returns in the 2nd layer
B3	Standard deviation	B12	20th Percentile	B21	65th Percentile	B30	Cumulative percentage of returns in the 3rd layer
B4	Skewness	B13	25th Percentile	B22	70th Percentile	B31	Cumulative percentage of returns in the 4th layer
B5	Kurtosis	B14	30th Percentile	B23	75th Percentile	B32	Cumulative percentage of returns in the 5th layer
B6	Entropy	B15	35th Percentile	B24	80th Percentile	B33	Cumulative percentage of returns in the 6th layer
B7	Percentage of returns above the mean	B16	40th Percentile	B25	85th Percentile	B34	Cumulative percentage of returns in the 7th layer
B8	Percentage of returns above 2	B17	45th Percentile	B26	90th Percentile	B35	Cumulative percentage of returns in the 8th layer
B9	5th Percentile	B18	50th Percentile	B27	95th Percentile	B36	Cumulative percentage of returns in the 9th layer

Finally, horizontal sections were obtained to evaluate the number of points at different levels and the vertical continuity of the vegetation. Specifically, 7 vertical sections were considered. The height thresholds of the sections are summarized in Table 6. It should be noted that the first threshold, 0.5m, was fixed in order to discard points at ground level. The last threshold, 12 m, was established since points above that value correspond to high stand canopy. These statistics were transformed into raster layers with a spatial resolution of 5 m. In these layers the digital value of each pixel corresponds to the number of points within the prism defined by the cell contour and the minimum and maximum z values for the section in question. The rasters were obtained by using the “counter” switch that is available in the “lasgrid” tool in LAStools (rapidlasso GmbH, 2014).

Table 6

Height thresholds of the sections in which the point clouds were divided to evaluate the vertical continuity of the vegetation.
Section 1	Section 2	Section 3	Section 4	Section 5	Section 6	Section 7
0.5–2 m	2–4 m	4–6 m	6–8 m	8–10 m	10–12 m	> 12 m

4.3.- Definition Of Correspondence Between Remote Sensing Data And Fuel Models

The land cover map, the CHM, and the analysis of the vertical structure of the vegetation allowed for the assignation, at the pixel level, of the corresponding Lourizán CIF fuel models for the entire study area. These fuel models are specified in Table 7. An extra fuel model, fuel model 0, was added to assign to areas that do not correspond with any particular fuel model, typically built-up areas and bodies of water. Then, the fire behavior graphics of the identified Lourizán-CIF fuels were compared to the nomograms established by Rothermel; in this way, for the models present in the study area, the proper correspondences between the fuels photo-guide and the Rothermel fuel models were established. These correspondences are also specified in Table 7. All of the defined correspondences were verified through field work and visual inspection by an expert in fuel model evaluation.

Table 7

Fuel models correspondences between the fuels photo-guide, the Rothermel fuel models and the remote sensing classes.
Remote sensing classes	Lourizán CIF fuels photo-guide	Rothermel model
Bare soil	-	1
Crops and pastures	Pl-01, Pl-02, Pa-01, Pa-04	2
Shrubs greater than 1 m in height	Ea-05, Es-01, Pt-05, Pt-06, Pt-07, Ue-05, Ue-08, Ue-10, Ub-02, Ub-04, Cs-01, Cs-03	4
Shrubs less than 1 m in height	Eu-01, Eu-05, Eu-06, Pt-01, Ue-01, Ue-02, Ub-03	5
Forest areas with continuous vertical structure	PpMB-02, PpL-04, PpL-09, EgL-08, QrL-04	7
Forest areas with discontinuous vertical structure	PpF-03, PpF-05, PpF-06, EgL-05, EgL-06, EgF-03, EgF-05, QrF-02, QrF-04, BaL-01	10
Litter and other forestry remnants	-	12
Anthropogenic areas, bodies of water	-	0

The “bare soil” class from the land cover map was assigned to Rothermel fuel model 1. According to the Rothermel descriptions, fuel model 1 is driven by fine, dry, low grasslands with a light shrub load. Bare soils correspond to areas mainly covered by rocks, but they usually include small patches of scattered grasslands and shrubs. Therefore, as a principle of caution, bare soils were also included under fuel model 1. This particular landscape was not identified in the fuels photo-guide, therefore, this correspondence could not be cross-referenced.

Crops and pasture classes were assigned to Rothermel fuel model 2, since, according to the fuels photo-guide, the typical pastures species in the study area present a fire behavior typical of this model.

Shrubs were assigned to Rothermel fuel models 4 and 5 since the wildfire behavior associated with these classes was found to be similar to the behavior associated to shrubs in the fuels photo-guide. The height threshold for differentiating between the two models, following Rothermel’s indications, was 1 m.

Tree covered vegetation classes included in the land cover map were found to correspond to Rothermel fuel models 7 and 10. Rothermel fuel model 7 is associated to very flammable shrub species with heights of less than 2 meters or pine stands such as those found in the state of Florida. However, according to the fuels photo-guide, many tree stands in the study area can develop a fire behavior similar to the behavior described for this Rothermel fuel model. The main characteristic of these stands is that they present vertical continuity, therefore fuel model 7 was assigned to areas covered by trees that presented a continuous vertical structure. On the other hand, areas covered by trees with a discontinuous vertical structure were associated with Rothermel fuel model 10. A stand is considered discontinuous when the base height of the crown is 2 to 3 times the height of the understory.

Fuel models 8 and 9 have not been classified in this study since the difference between one and the other is the compaction of the mantle, a characteristic that can hardly be differentiated using the sentinel images and the LiDAR data implemented in this study. Areas that would correspond to those models were assigned to fuel model 10, as a principle of caution.

Finally, areas potentially covered by significant amounts of slash were assigned to Rothermel fuel model 12. Rothermel fuel model 11 was rare or found only in areas that were not large enough to study; i.e., within the minimum unit analyzed (a 5 m pixel).

4.4. Mapping Of Rothermel Fuel Models

Once the correspondences between the remote sensing data and the fuel models were defined, the next step was the generation of a map for each Rothermel fuel model. Different models involved different procedures, which are described below. The final step was the integration of all of the models into a single Rothermel fuel models map.

a) Models 0, 1 and 2

These three fuel models were obtained by establishing a direct relationship between the model and some of the classes from the land cover map. Specifically, all pixels classified as anthropogenic areas and bodies of water in the land cover map were assigned to fuel model 0. Bare soil pixels were assigned to fuel model 1 and Crops and pastures were assigned to fuel model 2.

b) Models 4 and 5

These models were identified through the land cover map and the LiDAR-derived CHM. Models 4 and 5 corresponded with shrubs: model 4 with tall shrubs and model 5 with short shrubs. Therefore, all the pixels covered by shrubs in the land cover map were classified as model 4 or 5. Furthermore, shrub pixels with a CHM value of greater or equal to 1 m were reclassified as fuel model 4. Shrub pixels with a CHM of less than 1 m were reclassified as fuel model 5. This procedure was performed using the “raster” R package (Hijmans, 2021).

c) Models 7, 10 and 12

Models 7 and 10 corresponded with areas covered by trees, with and without a continuous vertical structure, respectively. First, all of the pixels covered by trees (Eucalyptus sp., conifers and broadleaves) on the land cover map were selected.

Once the tree-covered areas of the study area were identified, the vertical continuity of the vegetation in these zones was evaluated. LiDAR metrics were needed for this purpose. Since the vertical structure of vegetation is a complex parameter, it cannot be identified by simply establishing threshold values, as was possible in previous cases. The automatic identification of Rothermel fuel models 7 and 10 was performed through supervised classification. Training and verification areas were defined for this purpose. A set of 90 regular polygons, 25 m x 25 m in size, was designed and distributed over the previously identified tree-covered areas. Half of the polygons represented model 7, and the other half, represented model 10. The identification of fuel models in these polygons was carried out by visually analyzing vertical sections of the point clouds for each training set; the 7 horizontal raster layers containing the total number of points per pixel, obtained as described in section 4.2.2, were used for support in this process. Pixels where all of the 7 raster layers had a high number of points were considered to correspond to stands with vertical continuity, and therefore assigned to fuel model 7. Conversely, pixels where the upper and lower layers presented a much greater number of points than the intermediate layers were considered to correspond to stands with vertical discontinuity, and therefore assigned to fuel model 10. The 90 regular polygons were randomly divided into training and verification data. A total of 60 polygons were used as training areas, encompassing a total of 1500 pixels.

Once the training data were obtained, the supervised classification was performed. The supervised classification was performed using the machine learning algorithm Random Forest (Breiman, 2001); it was applied using the “randomForest” package in R (Liaw and Wiener, 2002) with the pre-defined configuration parameters. The predictor variables, used in the implementation of the Random Forest predictive model, were the 36 metric rasters obtained from the LiDAR point cloud (see section 4.2.2). As a result, the random forest algorithm creates a model that can be applied to each individual pixel in the study area to predict its corresponding class, or in this case, fuel model. Afterwards, the model was applied to the selected pixels (pixels covered by a tree-related class), resulting in a fuel model map with fuel models 7 and 10. The different variables’ importance, based on their Mean decrease in Gini, was also calculated (Breiman and Cutler, 2004) in order to find out which variables were the most vital in predicting the fuel model.

Finally, a cross-verification was performed. For this, the verification dataset comprised of 30 regular polygons was used. A confusion matrix was then obtained and accuracy metrics were calculated. The same accuracy metrics were calculated as the ones presented in Section 4.2.2.

Fuel model 12 is associated with recently harvested areas. All pixels with tree land cover (either Eucalyptus sp., Conifers or Broadleaves) on the land use map but that have no data in the LiDAR metrics layers were assigned to this model. As explained previously, the land cover map is obtained by assigning the most frequent land cover throughout the year in question (in this case, 2019); therefore, harvested areas are classified as a tree-related cover. However, since the LiDAR point-cloud was obtained in November of 2019, areas where harvest events had taken place prior to the flight do not present enough returns to calculate the statistics due to the absence of trees or any other vertical structure.

5.1.- Land cover mapping

The land cover map obtained is presented in Fig. 3. In Table 8a summary of the areas occupied by each land use is presented. The cross verification of the map resulted in high accuracy metrics: an Overall Accuracy of 90%, a Kappa Index of 0.88 and a UA and a PA of between 80 and 100% (except in the case of the PA for shrubs, which was lower: 70%). These results are detailed in Table 9 where the obtained confusion matrix is presented. It might be noticed that cloudy pixels include areas permanently covered by the Sentinel-cloud cover, they include areas permanently covered by clouds but also some anthropic areas with very reflective materials such as metallic roofs.

Table 8

Occupied area by each land use in the obtained land cover map.
LAND COVER	AREA (ha)	PERCENTAGE (%)
Clouds	41.05	0.14
Eucalyptus sp.	4531.21	14.92
Conifers	4976.25	16.39
Broadleaves trees	2425.26	7.99
Crops and pastures	9240.79	30.43
Shrubs	3959.37	13.04
Bare Soil	3525.40	11.61
Anthropic	962.67	3.17
Water	704.10	2.32

Table 9

Confusion matrix of land cover classification. **Euc**: *Eucalyptus* sp., **Con**: Conifers, **Bro**: Broadleaves, **Cro**: Crops and pastures, **Shru**: Shrubs, **Bare**: Bare soil, **Ant**: Anthropogenic areas, **Wat**: Water, T: Total, UA: User’s Accuracy, PA: Producer´s Accuracy, OA: Overall Accuracy, KI: Kappa Index.
		Ground Truth
		Euc	Con	Bro	Cro	Shru	Bare	Ant	Wat	T	UA (%)
Classification	Euc	49	1	0	0	0	0	0	0	50	98
	Con	5	43	0	0	2	0	0	0	50	86
	Bro	2	1	40	0	7	0	0	0	50	80
	Cro	0	0	0	45	5	0	0	0	50	90
	Shru	1	3	0	0	46	0	0	0	50	92
	Bare	0	0	0	0	6	40	4	0	50	80
	Ant	0	0	0	0	0	3	45	2	50	90
	Wat	0	0	0	0	0	0	0	50	50	100
	T	57	48	40	45	60	43	49	52		OA (%)
	PA (%)	86	86	100	100	70	93	92	96	OA (%)	90
										KI	0.88

5.2.- Structure Of Vegetation

A detailed examples of the CHM is presented in Fig. 4. This Figure helps to visualize the coherence of the CHM. Lower values are obtained for shrub areas and higher ones to Eucalyptus sp. and coniferous.

Examples of the horizontal slices obtained and the number of points in each slice are presented in Figs. 5 and 6. Each example includes a vertical section of the LiDAR point cloud in the selected stand that helps to better understand the examples. Figure 5 corresponds to an example of a tree stand with vertical continuity while Fig. 6 corresponds to a stand with a discontinuous vertical structure. Both figures help to understand the relation within the number of points in each slice and the vertical continuity of the stand. Specially, the lower slices (2–4 m and 6–8 m) were there is a low amount of points in the second figure and higher number of points the first one.

Examples of the set of statistics obtained for fuel models 7 and 10 are presented in Figs. 7 and 8, respectively. Both figures together help to understand how the entropy and the percentage of points in the intermediate layers (4th to 6th layer) help to reflect the vertical structure of the stand. The fuel model 7 stand obtains higher values for all this metrics than the fuel model 10 stand.

5.3.- Correspondences With Rothermel Fuel Models

Field work enabled confirmation that bare soils corresponded to Rothermel fuel model 1. These soils include rocky areas but also rocky areas partially covered by grass or very small shrubs (see Fig. 9 (a)).

In relation to Rothermel fuel model 2, field work confirmed that crops and pastures in the study area do not usually exceed 1 m in height (see Fig. 9 (b)). Furthermore, there is an absence of large grain crops that would lead to fire behavior characteristic of Rothermel fuel model 3.

The shrubs in the study area were assigned to Rothermel fuel models 4 and 5, tall ones and short ones respectively. The most common shrubs in the study area are Ulex spp., Cytisus spp. and Erica spp. Forest areas that are covered by Cytisus spp. have mainly been assigned to model 4 (see Fig. 9 (c)), while Erica spp. have been assigned to model 5 (see Fig. 9 (d)), and Ulex spp. can be found in both models.

In this study, Rothermel fuel model 7 was assigned to areas covered by trees with a continuous vertical structure. These areas are mostly dense young eucalyptus or conifer plantations and stands with mixed species. Broadleaves are included in this category as when combined with shrubs or ferns. Figure 9 (e) shows an example of the model.

Conversely, fuel model 10 was assigned to areas covered by trees with a discontinuous vertical structure. These areas mainly correspond to single-species plantations where silviculture actions are commonly or periodically performed. They are usually eucalyptus or conifer plantations or broadleaves stands without a shrub layer. Figure 9 (f) shows an example of the model.

Fuel model 12 has been assigned to areas that have been identified as harvested areas with forest harvesting residues. An example of this model can be seen in Fig. 9 (g).

5.4.- Fuel Models Mapping

Figure 10 shows the fuel models map obtained for the study area. On Table 9a summary of the areas occupied by each fuel is shown. Since we had a high percentage of area covered by agricultural land fuel model 2 is one of the models more present in the study area. Regarding forest-land-related fuel model it highlights that there is a bigger share of models 4 and 7 than 5 and 10. Additionally, it seems that fuel models 10 and 7 do not present a mosaic structure and they tend to be concentrated. For example, there is a higher presence of fuel model 10 in the southeast that in the northeast were fuel model 7 prevails amongst the forest areas. Additionally, if this map is compared with the land cover map it could be said that fuel model 10 is more linked to areas covered by coniferous than to areas covered by broadleaves or eucalypts.

Table 9

Occupied area by each fuel model.
FUEL MODEL	AREA (ha)	PERCENTAGE (%)
Model 0 – Anthropogenic areas, water bodies	1708.86	5.63
Model 1 - Bare soil	3525.33	11.61
Model 2- Crops and pastures	9240.84	30.43
Model 4 - High shrubs	2818.49	9.28
Model 5 - Low shrubs	1140.10	3.75
Model 7 - Tree covered areas with continuous vertical structure	7179.21	23.64
Model 10- Tree covered areas with discontinuous vertical structure	4322.20	14.23
Model 12 - Harvested areas	431.59	1.42

The accuracies of the fuel models map for models 0, 1 and 2 are directly related to the accuracy of the land cover classes used for assigning the models. The accuracies of the fuel models map for fuel models 3, 4 and 12 are also related to the accuracy of the land cover classes used for assigning the models as well as to the precision of the CHM. The accuracies of fuel models 7 and 10 are related to the land cover classes used for assigning the models as well as to the performance of the random forest model built for mapping the models. The performance of the random forest model is presented in Table 10, which shows the confusion matrix performed and the accuracy metrics obtained. The accuracy metrics calculated were all quite close to 100%, and the Kappa Index was 0.9771.

Table 10

Confusion matrix of fuel models 7 and 10. The numbers correspond with the number of pixels analyzed.
		Ground truth
		Fuel model 7	Fuel model 10	TOTAL	UA (%)
Classification	Fuel model 7	750	17	767	98
	Fuel model 10	0	717	717	100
	TOTAL	750	734	1484	OA (%)
	PA (%)	100	98	OA (%)	98
				KI	0.977

Figure 11 shows a comparison of variables in terms of their importance, their Mean Decrease in Gini, for constructing the model. The four most valuable bands were the entropy and the cumulative percentage of returns in the 5th, 6th and 4th layers. All of these variables are closely related to the vertical distribution of the points in the point-cloud, therefore they were able to capture and reflect the vertical continuity or discontinuity of the forest.

The described methodology uses remote sensing data to obtain an updated operational fuel models map for an area characterized by an Atlantic climate. The map will be useful with most fire simulation software since most rely on Rothermel fuel models (Simons et al., 2013). The methodology includes the definition of objective criteria for systematically associating the Rothermel fuel models to the types of vegetation present in the study area. This is not a simple task since the Rothermel Models were described for a different climate region with quite different vegetation types and structures. This adaptation could lead to a standardized way of mapping fuel models in this area. This will improve this process making it less dependent of the criteria of each technician.

These types of adaptations are not needed when the fuel model behavior is available for the study area (Huesca et al., 2019; Domingo et al., 2020; Marino et al., 2016; Palomino and Rodríguez y Silva, 2021), as in Marino et al., (2016) for the Canary Islands or in Palomino and Rodríguez y Silva, (2021) for Andalucía. Various efforts are being made to establish the fire spread equations for Galicia (Arellano et al., 2017). However, until they are completed and made ready for use in wildfire simulation software, the current study provides reliable fire behavior simulations for the current fire-risk and weather scenarios.

In the present study, the land cover classification separates tree stands into three different classes; whereas the fuel models classification divide the three classes of tree stands into two fuel models: model 7 and 10. It may be noted that this three tree classes may present slightly different behavior in the face of wildfires (Arellano et al., 2017). Consequently, this distinction has been done to maximize the future benefit derived from remote sensing capabilities in case of different fire spread equations would be developed for each tree class in the future.

The proposed methodology is similar to the approach followed by Marino et al. 2016. In this research, they also use a land cover map derived from satellite imagery (Landsat-8) and ALS data to subsequently build a decision tree to obtain fuel models. However, the methodology they used to create the land cover map was quite different from the method used in this study. Like in this study, they perform a random forest supervised classification, but rather than using the digital values of spectral bands, like in the present study, they used spectral indices. Neither do they rely on multi-temporal data. The overall accuracy of the land cover of the present study is higher (9 percentage points higher) than the OA obtained by Marino et al. (2016), they remarked that the reliability of these kinds of methodologies is highly dependent upon the land cover map accuracies (Marino et al., 2016). However, since the target land cover map classes in this study greatly differ from those used by Marino et al. (2016), a comparison of the accuracies obtained in the two studies might not be fully appropriate.

The most remarkable difference in relation to previous studies is the procedure used for discriminating areas with vertical continuity from those with vertical discontinuity. Marino et al. (2016) and studies relying on Prometheus models (Arroyo et al., 2008; Huesca et al., 2019; Domingo et al., 2020) fix thresholds for height and percentages of points in the point clouds. However, the specific values to be applied are dependent on the LiDAR density, occlusions and the type of species present (García-Cimarras et al., 2023). As a result, the replicability of the method might be compromised. In this case, stands are differentiated through a random forest classification, whose results do not depend on local parameters. Furthermore, studies that rely on the Prometheus system depend greatly on the study of the canopy cover (Arroyo et al., 2008). This kind of analysis is not feasible when using a 5x5 m pixel.

An important advantage of our proposed methodology in relation to some other previous ones is that it does not rely on field data. For example, the methods presented by Palomino and Rodríguez y Silva (2021) or Heisig et al. (2022) might be difficult to apply on a broader scale since they are designed for pilot areas or for local conditions and therefore the models developed cannot easily be extrapolated to other conditions (Huesca et al., 2020; Palomino and Rodríguez y Silva, 2021; Heisig et al., 2022). Since field work is not need in our methodology, the method ought to be easily applied in the rest of Galicia or areas with similar conditions.

Another significant advantage of our proposed methodology is the spatial resolution of the final map. It has a spatial resolution of 5 m. In contrast, Huesca et al. (2019) produced a map at 30 m, Marino et al. (2016) at 25 m and Domingo et al. (2020) at 10 m. According to Taneja et al. (2021) different spatial resolutions of fuel models provide different results in fire simulations; they found that higher resolutions allowed for better prevention strategies through wildfire simulations. Nevertheless, spatial resolution is limited by the resolution of the ALS data. The commented studies used point clouds with a density of 0.5 pts/m² and 1 pt/m² (Huesca et al., 2019; Marino et al., 2016; Domingo et al., 2020), while the ALS resolution for this study was 7.2 pts/m².

Although this methodology is presented as a methodology for obtaining updated operational fuel maps, it is true that the dependence upon LiDAR data could mean that its updatability is compromised. Therefore, between one LiDAR data update and the next, it may be necessary to update the fuel models map using open-access satellite imagery as was done, for example, by DeCastro et al. (2022). A methodology that may prove valuable for this task is the methodology developed by Alonso et al. (2022) for detecting forest disturbances. However, this methodology would be insufficient for detecting changes between models that differ solely in the vertical structure of the vegetation (i.e. it can’t differentiate between model 7 and model 10). Having updated fuel models maps is of the utmost importance, and it would be advisable to increase the frequency at which open-access ALS data is available in Spain (currently it is updated every 7 to 11 years (MTMAU, 2021). Additionally, it is worth exploring in detail for how long model 12 can indeed be assigned to describe an area since harvesting slash is never permanent.

Bearing in mind the high-risk fire regime in this region, the maps produced would allow to study in detail the fuel models distributions and patterns in the study area. Also this map may be useful for the stakeholders in the area to enhance the fire prevention using new strategies based on the use of fire behavior simulation software (Simons, 2013; Romero-Vivó et al., 2019; Moreira et al. 2020; Quílez et al. 2020; Iglesias et al., 2022). This, in addition, would help them to fulfil the new legal requirements set forth by the Spanish government in terms of fire prevention (RDL 15/2022).

This study allowed for the creation of an updated fuel models map for an Atlantic landscape using remote sensing. The fuel map obtained is adapted to the Rothermel fuel models and therefore it can be used operationally with different fire simulation software. Additionally, it was based on the particular fuel types present in the study area and could serve as the basis for future fuel models designed specifically for this environment. This would help in the future replicability of the methodology and mark a step towards mapping customized fuel models.

The availability of these maps will allow stakeholders in this region to make a shift towards better suppression and fuel management actions and to address the new phenomena of ever-worsening fire scenarios. This is thanks to having better, more up-to-date fuel maps for use with fire behavior simulation software. Additionally, it would also allow them to fulfil the legal requirements, aimed at fire prevention, set forth by the Spanish government.

Author Contributions: All authors have read and agreed to the published version of the manuscript. Conceptualization, J.A., J.P.; methodology, J.P., J.A., L.A., S.A., T.R., D.M.; software, A.S., L.A.; validation, L.A., A.S., T.R., J.P.; formal analysis, L.A., A.S., T.R., J.P., J.A., D.M.; investigation, A.S., L.A.; resources, A.S., L.A.; data curation, S.A., L.A., C.B.; writing—original draft preparation, A.S., L.A.; writing—review and editing, A.S., L.A., J.A., J.P., D.M., T.R.; visualization, A.S., L.A.; supervision, J.A., J.P.; project administration, J.A., J.P.; funding acquisition, J.A., J.P.

Funding: This research is part of the “PALEOINTERFACE: STRATEGIC ELEMENT FOR THE PREVENTION OF FOREST FIRES. DEVELOPMENT OF MULTISPECTRAL AND 3D ANALYSIS METHODOLOGIES FOR INTEGRATED MANAGEMENT” project, funded by the competitive call RETOS 2019 of the Spanish Ministry of Sciences and Innovation, Grant code PID2019-111581RB-I00. This research is also funded by the Administration of Rural Areas of the Government of Galicia under Grant 2020CONVINVENTARIOFORESTALR002; and it is also supported by an FPU grant from the Spanish Ministry of Sciences and Innovation under Grant FPU19/02054.

Availability of data and materials: Not applicable

Ethics approval and consent to participate: This research was performed with the informed consent of research participants.

Consent for publication: The authors consent to publication.

Conflicts of Interest: The authors declare that they have no competing interest.

Alonso, Laura, Juan Picos, and Julia Armesto. “Forest Land Cover Mapping at a Regional Scale Using Multi-Temporal Sentinel-2 Imagery and RF Models.” Remote Sensing 13, no. 12 (2021): 2237. https://doi.org/10.3390/rs13122237.
Alonso-Rego, Cecilia, Stéfano Arellano-Pérez, Carlos Cabo, Celestino Ordoñez, Juan Gabriel Álvarez-González, Ramón Alberto Díaz-Varela, and Ana Daría Ruiz-González. “Estimating Fuel Loads and Structural Characteristics of Shrub Communities by Using Terrestrial Laser Scanning.” Remote Sensing 12, no. 22 (2020): 3704. https://doi.org/10.3390/rs12223704.
Alonso-Rego, Cecilia, Stéfano Arellano-Pérez, Juan Guerra-Hernández, Juan Alberto Molina-Valero, Adela Martínez-Calvo, César Pérez-Cruzado, Fernando Castedo-Dorado, Eduardo González-Ferreiro, Juan Gabriel Álvarez-González, and Ana Daría Ruiz-González. “Estimating Stand and Fire-Related Surface and Canopy Fuel Variables in Pine Stands Using Low-Density Airborne and Single-Scan Terrestrial Laser Scanning Data.” Remote Sensing 13, no. 24 (2021): 5170. https://doi.org/10.3390/rs13245170.
Anderson, Hal. E. Aids to Determining Fuel Models for Estimating Fire Behavior. 122. Vol. 122. US Department of Agriculture, Forest Service, Intermountain Forest and Range Experiment Station, 1982.
Arellano, Stéfano, José A. Vega, Ana D. Ruiz, Antonio Arellano, Juan G. Álvarez, Daniel J. Vega, and Elena Pérez. Foto-Guía De Combustibles Forestales De Galicia y Comportamiento Del Fuego Asociado. Santiago de Compostela, Spain: Andavira Editora, S. L, 2016.
Arroyo, Lara A., Cristina Pascual, and José A. Manzanera. “Fire Models and Methods to Map Fuel Types: The Role of Remote Sensing.” Forest Ecology and Management 256, no. 6 (2008): 1239–52. https://doi.org/10.1016/j.foreco.2008.06.048.
Ascoli, Davide, Giorgio Vacchiano, Renzo Motta, and Giovanni Bovio. “Building Rothermel Fire Behaviour Fuel Models by Genetic Algorithm Optimisation.” International Journal of Wildland Fire 24, no. 3 (2015): 317. https://doi.org/10.1071/wf14097.
Biddle, Nicholas; Ben Edwards; Diane Herz; Toni Makkai; Ian, McAllister. "ANU Poll 2020: Bushfires, The Environment, and Optimism for The Future" ADA Dataverse, V2 (2020) doi:10.26193/S1S9I9
Breiman, Leo. Machine Learning 45, no. 1 (2001): 5–32. https://doi.org/10.1023/a:1010933404324.
Breiman, Leo, and Adele Cutler. “Breiman and Cuttler’s Random Forest for Classification and Regression” Random forests - classification description. Accessed January 24, 2023. https://www.stat.berkeley.edu/~breiman/RandomForests/cc_home.htm#giniimp.
California Department of Forestry and Fire Protection (CAL FIRE). “Stats and Events.” Cal Fire Department of Forestry and Fire Protection. Accessed December 22, 2022. https://www.fire.ca.gov/stats-events/.
Consellería Do Medio Rural. Dirección Xeral de Defensa Do Monte. PLADIGA: Plan de Prevención e Defensa Contra os Incendios Forestais de Galicia. Santiago de Compostela, Spain, 2022.
Cardil, Adrián, Santiago Monedero, Gavin Schag, Sergio de-Miguel, Mario Tapia, Cathelijne R. Stoof, Carlos A. Silva, Midhun Mohan, Alba Cardil, and Joaquin Ramirez. “Fire Behavior Modeling for Operational Decision-Making.” Current Opinion in Environmental Science & Health 23 (2021): 100291. https://doi.org/10.1016/j.coesh.2021.100291.
Costa Alcubierre, Pau, Marc Castellnou Ribau, Asier Larrañaga Otxoa de Egileor, Marta Miralles Bover, and Paul Daniel Kraus. La prevención De Los Grandes Incendios Forestales Adaptada Al Incendio Tipo. Barcelona: Unitat Tècnica del GRAF, 2011.
de Diego, Jaime, Antonio Rúa, and Mercedes Fernández. “Vulnerability Variables and Their Effect on Wildfires in Galicia (Spain). A Panel Data Analysis.” Land 10, no. 10 (2021): 1004. https://doi.org/10.3390/land10101004.
DeCastro, Amy L., Timothy W. Juliano, Branko Kosović, Hamed Ebrahimian, and Jennifer K. Balch. “A Computationally Efficient Method for Updating Fuel Inputs for Wildfire Behavior Models Using Sentinel Imagery and Random Forest Classification.” Remote Sensing 14, no. 6 (2022): 1447. https://doi.org/10.3390/rs14061447.
Domingo, Darío, Juan de la Riva, María Lamelas, Alberto García-Martín, Paloma Ibarra, Maite Echeverría, and Raúl Hoffrén. “Fuel Type Classification Using Airborne Laser Scanning and Sentinel 2 Data in Mediterranean Forest Affected by Wildfires.” Remote Sensing 12, no. 21 (2020): 3660. https://doi.org/10.3390/rs12213660.
Dong, Chunyu, A. Park Williams, John T. Abatzoglou, Kairong Lin, Gregory S. Okin, Thomas W. Gillespie, Di Long, Yen-Heng Lin, Alex Hall, and Glen M. MacDonald. “The Season for Large Fires in Southern California Is Projected to Lengthen in a Changing Climate.” Communications Earth & Environment 3, no. 1 (2022). https://doi.org/10.1038/s43247-022-00344-6.
Dupuy, Jean-luc, Hélène Fargeon, Nicolas Martin-StPaul, François Pimont, Julien Ruffault, Mercedes Guijarro, Carmen Hernando, Javier Madrigal, and Paulo Fernandes. “Climate Change Impact on Future Wildfire Danger and Activity in Southern Europe: A Review.” Annals of Forest Science 77, no. 2 (2020). https://doi.org/10.1007/s13595-020-00933-5.
Ellsworth, L. M., B. A. Newingham, S. E. Shaff, C. L. Williams, E. K. Strand, M. Reeves, D. A. Pyke, E. W. Schupp, and J. C. Chambers. “Fuel Reduction Treatments Reduce Modeled Fire Intensity in the Sagebrush Steppe.” Ecosphere 13, no. 5 (2022). https://doi.org/10.1002/ecs2.4064.
ESA (European Space Agency) and Copernicus. “Copernicus Open Access Hub.” Accessed January 24, 2023. https://scihub.copernicus.eu/.
ESA (European Space Agency). “Sentinel-2 User Handbook. ESA Standard Document.” (2015). Accessed January 24, 2023. https://sentinels.copernicus.eu/web/sentinel/user-guides/document-library/-/asset_publisher/xlslt4309D5h/content/sentinel-2-user-handbook.
Fidalgo-González, L. A., S. Arellano-Pérez, J. G. Álvarez-González, F. Castedo-Dorado, A. D. Ruiz-González, and E. González-Ferreiro. “Estimación De La Distribución Vertical De Combustibles Finos Del Dosel De Copas En Masas De Pinus Sylvestris Empleando Datos Lidar De Baja Densidad.” Revista de Teledetección, no. 53 (2019): 1. https://doi.org/10.4995/raet.2019.11241.
Ferrer Palomino, Aurora, and Francisco Rodríguez Silva. “Fuel Modelling Characterisation Using Low-Density Lidar in the Mediterranean: An Application to a Natural Protected Area.” Forests 12, no. 8 (2021): 1011. https://doi.org/10.3390/f12081011.
Flannigan, Mike, Alan S. Cantin, William J. de Groot, Mike Wotton, Alison Newbery, and Lynn M. Gowman. “Global Wildland Fire Season Severity in the 21st Century.” Forest Ecology and Management 294 (2013): 54–61. https://doi.org/10.1016/j.foreco.2012.10.022.
García-Cimarras, Alba, José Antonio Manzanera, and Rubén Valbuena. “Lidar Scan Density and Spatial Resolution Effects on Vegetation Fuel Type Mapping.” Croatian journal of forest engineering 44, no. 1 (2022): 189–201. https://doi.org/10.5552/crojfe.2023.1689.
Gale, Matthew G., Geoffrey J. Cary, Albert I.J.M. Van Dijk, and Marta Yebra. “Forest Fire Fuel through the Lens of Remote Sensing: Review of Approaches, Challenges and Future Directions in the Remote Sensing of Biotic Determinants of Fire Behaviour.” Remote Sensing of Environment 255 (2021): 112282. https://doi.org/10.1016/j.rse.2020.112282.
Gobierno de España, Ministerio de Agricultura, Pesca y Alimentación. Los Incendios Forestales en España: Decenio 2006-2015. Antonio López Santalla, and Marta López García. Madrid: Ministerio de Agricultura, Pesca y Alimentación, 2019
Heisig, Johannes, Edward Olson, and Edzer Pebesma. “Predicting Wildfire Fuels and Hazard in a Central European Temperate Forest Using Active and Passive Remote Sensing.” Fire 5, no. 1 (2022): 29. https://doi.org/10.3390/fire5010029.
Hijmans, Robert J. (2021). raster: Geographic Data Analysis and Modeling. R package version 3.4-13. https://CRAN.R-project.org/package=raster
Huesca, Margarita, David Riaño, and Susan L. Ustin. “Spectral Mapping Methods Applied to LIDAR DATA: Application to Fuel Type Mapping.” International Journal of Applied Earth Observation and Geoinformation 74 (2019): 159–68. https://doi.org/10.1016/j.jag.2018.08.020.
Iglesias, Miguel Cánibe, Virgilio Hermoso, João C. Campos, Cláudia Carvalho-Santos, Paulo M. Fernandes, Teresa R. Freitas, João P. Honrado, et al. “Climate- and Fire-Smart Landscape Scenarios Call for Redesigning Protection Regimes to Achieve Multiple Management Goals.” Journal of Environmental Management 322 (2022): 116045. https://doi.org/10.1016/j.jenvman.2022.116045.
Jain, Piyush, Sean C.P. Coogan, Sriram Ganapathi Subramanian, Mark Crowley, Steve Taylor, and Mike D. Flannigan. “A Review of Machine Learning Applications in Wildfire Science and Management.” Environmental Reviews 28, no. 4 (2020): 478–505. https://doi.org/10.1139/er-2020-0019.
Lewinski, Stanislaw, Artur Nowakowski, Radek Malinowski, Marcin Rybicki, Ewa Kukawska, Radoslaw Malinowski and Michal Krupiński. “Aggregation of Sentinel-2 Time Series Classifications as a Solution for Multitemporal Analysis.” Image and Signal Processing for Remote Sensing XXIII, 2017. https://doi.org/10.1117/12.2277976.
Li, Zheng, Jay P. Angerer, and X. Ben Wu. “The Impacts of Wildfires of Different Burn Severities on Vegetation Structure across the Western United States Rangelands.” Science of The Total Environment 845 (2022): 157214. https://doi.org/10.1016/j.scitotenv.2022.157214.
Liaw, Andy; Matthew Wiener. “Classification and Regression by randomForest”. R New 2, no. 3 (2022): 18-22.
Linley, Grant D., Chris J. Jolly, Tim S. Doherty, William L. Geary, Dolors Armenteras, Claire M. Belcher, Rebecca Bliege Bird, et al. “What Do You Mean, ‘Megafire’?” Global Ecology and Biogeography 31, no. 10 (2022): 1906–22. https://doi.org/10.1111/geb.13499.
López-Rodríguez, Gervasio, Verónica Rodríguez-Vicente, and Manuel F. Marey-Pérez. “Study of Forest Productivity in the Occurrence of Forest Fires in Galicia (Spain).” Sustainability 13, no. 15 (2021): 8472. https://doi.org/10.3390/su13158472.
Romero-Vivó, Mario; José Luis Soriano; Miguel A. Botella; A. Cervera; Raúl Quílez Moraga; Emma Gorgonio Bonet; David Caballero Valero; Asier Larrañaga; Francisco rodríguez y Silva; Javier Blanco; Gema Ortega Rebolo; Rafaél Lopez del Río; Jorge Cantón; Álvaro Escrig del Valle; Jaime Baeza; Santiago Bateman, Gloria Romero; Sonia Monferrer and Fernando Pulido. “Definición y recomendaciones técnicas en el diseño de puntos estratégicos de gestión”, edited by Sociedad Española de Ciencias Forestales; Generalitat Valenciana-Consellería de Agricultura, Medio Ambiente, Cambio Climático y Desarrollo Rural, Valencia, 2019. http://secforestales.org/sites/default/files/archivos/documentopeg.pdf
MITECO (Ministerio para la Transición Ecológica y el Reto Demográfico). Mapa Forestal de España (MFE) de máxima actualidad. 2022. https://www.miteco.gob.es/es/cartografia-y-sig/ide/descargas/biodiversidad/mfe.aspx.
Marino, Eva, Pedro Ranz, José Luis Tomé, Miguel Ángel Noriega, Jessica Esteban, and Javier Madrigal. “Generation of High-Resolution Fuel Model Maps from Discrete Airborne Laser Scanner and Landsat-8 Oli: A Low-Cost and Highly Updated Methodology for Large Areas.” Remote Sensing of Environment 187 (2016): 267–80. https://doi.org/10.1016/j.rse.2016.10.020.
McWethy, David B., Tania Schoennagel, Philip E. Higuera, Meg Krawchuk, Brian J. Harvey, Elizabeth C. Metcalf, Courtney Schultz, et al. “Rethinking Resilience to Wildfire.” Nature Sustainability 2, no. 9 (2019): 797–804. https://doi.org/10.1038/s41893-019-0353-8.
Moreira, Francisco, Davide Ascoli, Hugh Safford, Mark A Adams, José M Moreno, José M Pereira, Filipe X Catry, et al. “Wildfire Management in Mediterranean-Type Regions: Paradigm Change Needed.” Environmental Research Letters 15, no. 1 (2020): 011001. https://doi.org/10.1088/1748-9326/ab541e.
MTMAU (Ministerio de Transporte Movilidad y Agenda Urbana) Plan Nacional de Ortofotografía Aérea (PNOA). 2021. Accessed January 24, 2023. https://pnoa.ign.es/
Nature. “Spreading like Wildfire.” Nature Climate Change 7, no. 11 (2017): 755–55. https://doi.org/10.1038/nclimate3432
Oliveira, Sandra, Fernando Félix, Adélia Nunes, Luciano Lourenço, Giovanni Laneve, and Ana Sebastián-López. “Mapping Wildfire Vulnerability in Mediterranean Europe. Testing a Stepwise Approach for Operational Purposes.” Journal of Environmental Management 206 (2018): 158–69. https://doi.org/10.1016/j.jenvman.2017.10.003.
Quílez, Raúl, Luz Valbuena, Jordi Vendrell, Kathleen Uytewaal, and Joaquín Ramirez. “Establishing Propagation Nodes as a Basis for Preventing Large Wildfires: The Proposed Methodology.” Frontiers in Forests and Global Change 3 (2020). https://doi.org/10.3389/ffgc.2020.548799
rapidlasso GmbH, "LAStools - efficient LiDAR processing software" (2014) (academic). Obtained from http://rapidlasso.com/LAStools
Real Decreto-ley 15/2022, de 1 de agosto, por el que se adoptan medidas urgentes en materia de incendios forestales. Boletín Oficial del Estado (BOE), 2 de agosto de 2022, núm. 184. Accessed February 2, 2023 https://www.boe.es/eli/es/rdl/2022/08/01/15/con
Reilly, Sean, Matthew L. Clark, Lisa Patrick Bentley, Corbin Matley, Elise Piazza, and Imma Oliveras Menor. “The Potential of Multispectral Imagery and 3D Point Clouds from Unoccupied Aerial Systems (UAS) for Monitoring Forest Structure and the Impacts of Wildfire in Mediterranean-Climate Forests.” Remote Sensing 13, no. 19 (2021): 3810. https://doi.org/10.3390/rs13193810.
Robinne, François-Nicolas, Kevin D. Bladon, Carol Miller, Marc-André Parisien, Jérôme Mathieu, and Mike D. Flannigan. “A Spatial Evaluation of Global Wildfire-Water Risks to Human and Natural Systems.” Science of The Total Environment 610-611 (2018): 1193–1206. https://doi.org/10.1016/j.scitotenv.2017.08.112.
Rodrigues, Marcos, Àngel Cunill Camprubí, Rodrigo Balaguer-Romano, Celso J. Coco Megía, Francisco Castañares, Julien Ruffault, Paulo M. Fernandes, and Víctor Resco de Dios. “Drivers and Implications of the Extreme 2022 Wildfire Season in Southwest Europe.” Science of The Total Environment 859 (2023): 160320. https://doi.org/10.1016/j.scitotenv.2022.160320.
Rothermel, Richard C. “How to Predict the Spread and Intensity of Forest and Range Fires,” 1983. https://doi.org/10.2737/int-gtr-143.
Roussel, Jean-Romain and David Auty (2021). Airborne LiDAR Data Manipulation and Visualization for Forestry Applications. R package version 3.1.4. https://cran.r-project.org/package=lidR
Roussel, Jean-Romain, David Auty, Nicholas C. Coops, Piotr Tompalski, Tristan R.H. Goodbody, Andrew Sánchez Meador, Jean-François Bourdon, Florian de Boissieu, and Alexis Achim. “LidR: An R Package for Analysis of Airborne Laser Scanning (ALS) Data.” Remote Sensing of Environment 251 (2020): 112061. https://doi.org/10.1016/j.rse.2020.112061.
Royé, Dominic, Fantina Tedim, Javier Martin‐Vide, Michele Salis, Jordi Vendrell, Raffaella Lovreglio, Christophe Bouillon, and Vittorio Leone. “Wildfire Burnt Area Patterns and Trends in Western Mediterranean Europe via the Application of a Concentration Index.” Land Degradation & Development 31, no. 3 (2019): 311–24. https://doi.org/10.1002/ldr.3450.
Royé, D., J. J. Taboada, A. Martí, and M. N. Lorenzo. “Winter Circulation Weather Types and Hospital Admissions for Respiratory Diseases in Galicia, Spain.” International Journal of Biometeorology 60, no. 4 (2016): 507–20. https://doi.org/10.1007/s00484-015-1047-1.
Scott, Joe H., and Robert E. Burgan. “Standard Fire Behavior Fuel Models: A Comprehensive Set for Use with Rothermel's Surface Fire Spread Model,” 2005. https://doi.org/10.2737/rmrs-gtr-153.
Senande-Rivera, Martín, Damián Insua-Costa, and Gonzalo Miguez-Macho. “Spatial and Temporal Expansion of Global Wildland Fire Activity in Response to Climate Change.” Nature Communications 13, no. 1 (2022). https://doi.org/10.1038/s41467-022-28835-2.
Simons, Nicole Elisabeth. “Improving decision making during wildland fire events” PhD diss., University of California, Santa Barbara, 2013.
Taboada, Angela, Paula García-Llamas, José Manuel Fernández-Guisuraga, and Leonor Calvo. “Wildfires Impact on Ecosystem Service Delivery in Fire-Prone Maritime Pine-Dominated Forests.” Ecosystem Services 50 (2021): 101334. https://doi.org/10.1016/j.ecoser.2021.101334.
Taneja, Ritu, James Hilton, Luke Wallace, Karin Reinke, and Simon Jones. “Effect of Fuel Spatial Resolution on Predictive Wildfire Models.” International Journal of Wildland Fire 30, no. 10 (2021): 776. https://doi.org/10.1071/wf20192.
Tedim, Fantina, Vittorio Leone, Malik Amraoui, Christophe Bouillon, Michael Coughlan, Giuseppe Delogu, Paulo Fernandes, et al. “Defining Extreme Wildfire Events: Difficulties, Challenges, and Impacts.” Fire 1, no. 1 (2018): 9. https://doi.org/10.3390/fire1010009.
Vacchiano, Giorgio, and Davide Ascoli. “An Implementation of the Rothermel Fire Spread Model in the R Programming Language.” Fire Technology 51, no. 3 (2015): 523–35. https://doi.org/10.1007/s10694-014-0405-6.
Wang, Daoping, Dabo Guan, Shupeng Zhu, Michael Mac Kinnon, Guannan Geng, Qiang Zhang, Heran Zheng, et al. “Economic Footprint of California Wildfires in 2018.” Nature Sustainability 4, no. 3 (2021): 252–60. https://doi.org/10.1038/s41893-020-00646-7.
Woods, M., K. Lim, and P. Treitz. “Predicting Forest Stand Variables from Lidar Data in the Great Lakes – St. Lawrence Forest of Ontario.” The Forestry Chronicle 84, no. 6 (2008): 827–39. https://doi.org/10.5558/tfc84827-6.

Download PDF

Journal Publication

published 17 Oct, 2023

Read the published version in Fire Ecology →

Reviewers invited by journal
11 Apr, 2023
Reviewers agreed at journal
19 Feb, 2023
Editor assigned by journal
15 Feb, 2023
First submitted to journal
13 Feb, 2023

You are reading this latest preprint version

Operational Fuel Models Map for Atlantic Landscapes Using ALS and Sentinel-2 Images

Status:

Journal Publication

Version 1

Abstract

Background:

Results:

Conclusions:

Figures

1. Background

2. study Area

3. Materials

3.1.- Satellite Images

3.2.- Airborne Lidar Data

3.3.- Reference Data

3.3.- Fuel Models

4. Methodology

4.1.- Land Cover Mapping

4.2.- Structural Characterization Of Vegetation

4.2.1.- Lidar Data Pre-processing

4.2.2.- Lidar Data Processing

4.3.- Definition Of Correspondence Between Remote Sensing Data And Fuel Models

4.4. Mapping Of Rothermel Fuel Models

5. Results

5.1.- Land cover mapping

5.2.- Structure Of Vegetation

5.3.- Correspondences With Rothermel Fuel Models

5.4.- Fuel Models Mapping

6. Discussion

7. Conclusion

Declarations

References

Status:

Journal Publication

Version 1