Mapping fire risk in the Model Forest of Urbión (Spain) based on airborne LiDAR measurements

Abstract

The present study sets a methodological framework to combine LiDAR derived data with fire behaviour models in order to assess fire risk at landscape level for forest management and planning. Two forest areas of the Model Forest in Urbión, Soria (Central Spain) were analyzed, covering 992.7 ha and 221.7 ha. The modelling phase was based in 160 field sample plots as ground data, and the LiDAR data had a density of first returns of 2 pulses/m 2 , which were used to construct 13 models for stand variables (e.g. basal area, stem volume, branch biomass). The coefficients of determination ranged from 0.167 for shrub cover, to 0.906 for dominant height. The modelled variables were used for a classification of fuel types compatible with the continuous data. The simulation phase was performed using the spatialized data on FlamMap in order to assess the potential fire behaviour resulting across the whole landscape for four scenarios of moisture and wind conditions. The results showed maps of fire intensity and prob-ability of fire occurrence, based on the simulation of 500 random ignition points, which allowed the anal-ysis of the spatial relation between the initial state and allocation of forest resources and their risk of fire. The methodology proposed, as well as the results of this research are directly applicable for operational forest planning at landscape level. Ó 2012 Elsevier B.V. All rights reserved.

Full text

Mapping fire risk in the Model Forest of Urbión (Spain) based on airborne LiDAR
measurements
José-Ramón González-Olabarria1,*, Francisco Rodríguez2, Alfredo Fernández-Landa3,4,
Blas Mola-Yudego5,6
(González-Olabarria, J.R., Rodríguez, F., Fernández-Landa, A., Mola-Yudego, B)
1 Centre Tecnològic Forestal de Catalunya. Ctra St. Llorenc de Munys, Km 2. ES-
25280, Solsona, Spain. Phone: +34-973-481752, Fax: +34-973-481392, e-mail:
jr.gonzalez@ctfc.es
2 Cesefor Foundation, Pol. Ind. Las Casas, calle C, Parcela 4, ES-42005 Soria, Spain
3 Agresta Sociedad Cooperativa, C/ Numancia 1, ES-42001 Soria, Spain
4 Universidad Politécnica de Madrid, Ciudad Universitaria, ES-28040 Madrid, Spain
5 University of Eastern Finland, School of Forest Sciences, PO Box 111, FI-80101
Joensuu, Finland
6 Department of Crop Production Ecology, Swedish University of Agricultural Sciences
(SLU), P.O. Box 7016, S-750 07 Uppsala, Sweden.
*Corresponding author

Research highlights
-Fire behaviour and occurrence are estimated with defined scenarios in a forest area
-We present a methodology that combines forest fire simulators and LiDAR data
-It provides 13 models for stand level variables and its spatialization
-A set of rules to classify fuel types compatible with the continuous data is presented
-The methodology is applicable for operational planning purposes at landscape level.

ABSTRACT
The present study sets a methodological framework to combine LiDAR derived data
with fire behaviour models in order to assess fire risk at landscape level for forest
management and planning. Two forest areas of the Model Forest in Urbión, Soria
(Central Spain) were analysed, covering 992.7 ha and 221.7 ha. The modelling phase
was based in 160 field sample plots as ground data, and the LiDAR data had a density
of first returns of 2 pulses/m2, which were used to construct 13 models for stand
variables (e.g. basal area, stem volume, branch biomass…). The coefficients of
determination ranged from 0.167 for shrub cover, to 0.906 for dominant height. The
modelled variables were used for a classification of fuel types compatible with the
continuous data. The simulation phase was performed using the spatizalized data on
FlamMap in order to assess the potential fire behaviour resulting across the whole
landscape for four scenarios of moisture and wind conditions. The results showed maps
of fire intensity and probability of fire occurrence, based on the simulation of 500
random ignition points, which allowed the analysis of the spatial relation between the
initial state and allocation of forest resources and their risk of fire. The methodology
proposed, as well as the results of this research are directly applicable for operational
forest planning at landscape level.
Keywords: Airborne LiDAR, Forest inventory, Fire Risk Assessment, Mediterranean
Model Forest

1. Introduction
The inclusion of fire risk into the planning of forest management is a recurrent research
topic since the 1980´s (e.g. Van Wagner, 1983; Reed and Errico, 1986). Wildfires have
an obvious effect on the outcomes of forest management through post-fire tree mortality
or value depreciation of surviving trees. At the same time, however, forest management
has the potential of modifying fire behaviour by changing the quantity and spatial
arrangement of forest fuels (Agee and Skinner, 2005; Peterson et al., 2005; Finney et
al., 2007). The way of considering the risk of fire into the process of planning forest
management has evolved from non-spatial approaches where the effect of fire was
defined as deterministic or as a stochastic quantity of timber losses, to the more recent
approaches were fire behaviour and its (spatially explicit) components are being
considered in order to assess the extent of fire induced damage and the influence of fuel
modification on fire behaviour (Bettinger, 2010).
In this sense, assessing adequately the current state of the forest is one of the first steps
required for planning the management of a forest area when the risk of fire is
considered. The assessment requires collecting precise data of the amount and
distribution of desirable resources and at the same time, estimates of the potential threat
that fire means to those resources. However, to estimate the risk of fire over a forest
landscape it must be considered that fire is a spatially explicit event, and varies its
behaviour depending on site-specific fuel conditions and the spatial arrangement of
different fuels (Finney, 2001). This spatial dimension, that entails information about the
state of the forest and the potential behaviour of fire across the landscape, it is required
in order to choose the most effective fuel treatments, in terms of type and allocation,

that would reduce the negative impact of fire on the forest (Agee et al., 2000; Finney et
al., 2007).
However, mapping fuel and forest stocking characteristics at a broad spatial scale is
often not feasible based on direct field measurements. But during the last decades, the
use of diverse types of remote sensors has become popular to acquire information about
the continuous distribution of fuel (Chuvieco and Congalton, 1989, Arroyo et al., 2008)
and forest resources at landscape level (Kirby and Hall, 1980; Boyd and Danson.,
2005), and among them, Light Detection and Ranging (LiDAR) has gained special
relevance. Airborne LiDAR can provide accurate information about the forest structure
in three dimensions over large areas (Andersen et al., 2005), and this information can be
used in the assessment of forest resources (Nelson et al., 1988; Næsset, 2002; 2004;
Holmgren et al., 2003; Hollaus et al., 2004; Hall et al., 2005; Maltamo et al., 2005) as
well as for modelling and mapping forest fuels. In fact, recent studies have
demonstrated the usefulness of LiDAR derived data e.g.: modelling surface fuels (Riaño
et al., 2007; Mutlu et al., 2008a), modelling canopy related variables to predict crown
fire activity (Riaño et al., 2003, 2004; Andersen et al., 2005; Morsdorf et al., 2004), or
both (Skowroski et al., 2007; Mutlu et al., 2008b).
A potential application of the fuel maps generated from LiDAR data would be their use
as inputs for fire simulation models, which can simulate the potential fire behaviour
across a large area. The combination of LiDAR derived information with these type of
simulation models, essentially FARSITE (Finney, 1998, 2004), has been referred to be a
significant improvement for planning fuel management operations (Riaño et al., 2003,
2004; Morsdorf et al., 2004; Andersen et al., 2005; Skowroski et al., 2007; Mutlu et al.,

2008a), even when a explicit application to estimate fire behaviour parameters on a
landscape level has only seldom been taken (Mutlu et al., 2008b). Additionally, the
potential value of combining spatially continuous inventory of forest resources and
information on fire behaviour, both derived from LiDAR data, has not been sufficiently
explored in order to integrate the spatial component of fire in the context of landscape-
level planning of forest management (Bettinger, 2009; Kim, 2009, Gonzalez-Olabarria
and Pukkala, 2011).
In this context, the present study aims at providing a methodology that implements
LiDAR derived information, forest resources inventory, understory and canopy fuel
modelling, and fire behaviour simulation models in order to assess fire risk at landscape
level using spatially continuous information for forest management purposes in the
region of Soria (Central Spain).
2. Material and Methods
2.1. Description of the area of study
Two forest areas located within the Model Forest of Urbión in Soria (Central Spain)
were selected for the analysis, corresponding to the public utility forest numbers
MUP89 and MUP76, covering 992.7 ha and 221.7 ha, respectively. The Model Forest
selected is the most extensive forest mass of the Iberian Peninsula, with a forested area
of over 100 000 ha, and it combines diverse forest types and mountain pastures,
enclosed within the administrative borders of 35 municipalities (Segur, 2009). The
forest lands of Urbión are included in both the Mediterranean and Ibero-American
Commented [m1]: “as being” deja la frase sin sentido, no? mira
a ver

Model Forest Regions in the International Model Forest Network (Besseau et al., 2002).
The forest management of the two studied areas has being oriented towards the
production of a sustainable flow of certified timber, and the revenues and work derived
from the forest are shared by a large proportion of the local community.
In both cases, the forest lands are mainly dominated by pure stands of Pinus Nigra and
P. Pinaster, with a limited presence of Juniperus sp. The areas are located around 1000
- 1200 meters above sea level, and are generally influenced by Mediterranean climate,
although the high altitude and continentality of the region results in colder and longer
winters than most of Spain. As in most of the Mediterranean forests, wildfires are
considered an important risk for the production of timber, and should therefore be
considered in the forest management planning. For example, the studied area MUP89
was affected by a fire in 1970.
2.2. Description field data
In the studied area, 160 squared field sample plots of 500 m2, systematically distributed
in a square grid of approximately 250 x 250 m, were set as ground data, in order to
develop models for converting LiDAR data into forest information. Measurements of
dbh (diameter at breast height), height and canopy base height were recorded from
every tree in the field plots. Measurements of tree and shrub cover, and an estimation of
the mean shrub height were recorded at plot level. Finally, additional plot level
variables were calculated from the measurements using existing models: timber volume
using the models of Rodriguez et al. (2012); and biomass distribution in leaves,

branches, stem and roots, using the allometric functions of Ruiz-Peinado et al. (2011)
(Table 1).
TABLE 1
2.3. Description data from LiDAR
The LiDAR data were provided by Stereocarto S.L. The data was based in 20 linear
flights of a Cesnna 402-C airplane, equipped with an ALS60 II LiDAR sensor. The
LiDAR system provided a density of first returns of 2 pulses / m2, and an overall
quantity of 4 height bins per first return. The LiDAR raw data was treated with the
FUSION system (McGaughey and Carson, 2003), from the Remote Sensing
Applications Center (USDA) in order to obtain structured statistical information about
the laser returns. A predefined height of 2 meters was used as a threshold to model
mature trees and understory vegetation. The LiDAR data at was aggregated in data sub-
sets using a square lattice of 500 m2 , matching the reference ground data (field sample
plots), for computing metrics at that spatial scale. From the pulses returned from the
aboveground vegetation (>2 meters from the ground), the metrics of the LiDAR pulses
generated by FUSION, and afterwards used as predictors in the models, corresponded to
different percentiles (1st, 5th, 10th , 20th, 25th, 30th, 40th, 50th, 60th, 70th, 75th, 80th, 90th,
95th, 99th) height and intensity of the pulses, the mean height and intensity values of
pulses, the interquartile range of the intensity values, and an estimation of the forest
canopy cover derived from the ratio between the number of first returns above 2 meters
and the total number of first returns. Additionally, an approximation of the shrub cover

was defined as the ratio between the number of first returns between 0.3 and 2 meters
and the number of first returns bellow 2 meters.
2.4 Integral methodology applied
The methodology applied aimed to provide a broad range of information about the state
of the forest at landscape level, from conventional data of forest resources to fire
behaviour and fire risk estimates, covering the whole spatial continuum of the studied
areas. The working methods consisted in the following steps (Fig. 1): 1) Acquisition of
LiDAR based data through flights over the study areas; 2) Implementation of field
inventories to be used as ground data and transformation of the raw LiDAR data into
plot size metrics (defined by 500 m2) using the FUSION software; 3) Combination of
the LiDAR based metrics and the ground data in order to develop models to predict
stand variables (stocking, fuel types, canopy characteristics…), that at the same time are
inputs for a classification of fuel types; 4) Spatializing the resulting information across
the whole landscape; 5) Implementation of the map layers in the FlamMap fire
behaviour model (Finney, 2006), including e.g.: topography, canopy cover, bulk
density, canopy base height, stand height and the defined fuel types.
FIGURE 1
2.3. Modelling stand variables

The data derived from the LiDAR measurements and the on field measured data was
used to construct models for predicting stand variables. The dependent variables
considered were: stand density (N), basal area (G), total volume (V), average quadratic
diameter (Dq), dominant height (H0), Lorey’s height (hL), height to the first branch
(hB), leaves biomass (FB), stem biomass (SB), branch biomass (BB), root biomass
(RB), canopy cover of the undercover vegetation (BC), and height of undercover
vegetation (hB). The variables used as predictors were constructed based on the
calculated LiDAR pulses’ metrics generated using the FUSION system.
The criteria used for constructing the models aimed at a high predictive power, based on
unbiased predictions and avoiding variable redundancy or higher co-linearity. Each
independent variable included in the models had to be significant at the 0.05 level. A
maximum of three variables was allowed for each model in order to avoid excessive
complexity. Different combinations of the variables as well as different model
alternatives were tested systematically, and the best performing model was finally
selected. The selected relationships between ground-based characteristics of the sample
plots (dependent variables) and laser-derived metrics were defined using nonlinear
regression analysis. Finally, an n-way cross-validation of each equation was performed
and the RMSE and the model efficiency (MEF, equivalent to the R2 of the fitting phase)
were calculated from the residuals.
2.4. Fuel type definition
Besides the stand variables, we proceeded to develop a relationship between the stand
characteristics and the standard Northern Forest Fire Laboratory (NFFL) fuel types
(Anderson, 1982). The fuel types were defined based on: an interpretation of the

description of the NFFL fuel types, a comparison of potential stand level variables and
the NFFL fuel types recorded in 431 plots dominated by P. nigra and P. sylvestris,
extracted from the 3rd Spanish National Forest Inventory for the same area (DGCN,
2006). The variables selected for the classification were based on the previously
constructed models derived from LiDAR data. The process of relating the stand
variables and the fuel types was undertaken by using the Chi-squares Automatic
Interaction (CHAID) (Kass 1980), decision tree technique in SPSS 19. The CHAID
technique is able to select the splitting criterions of the predictors (stand level variables)
for better classifying in between multiple dependent responses (fuel types). The best
suited splitting criteria for classifying a forest land regarding its fuel type were
transformed into a set of classification rules for an easy application into the whole area
studied.
In order to validate the process and to assess the predictive capability, the same set of
rules were applied to the 431 plots used for defining the classification criteria, and the
predicted fuel types were compared to the ones observed in field.
2.5. Generating fire scenarios in FlamMap
The FlamMap fire mapping and analysis system (Finney, 2006) is a computer program
that integrates a FARSITE defined landscape (it requires layers for: elevation, slope,
aspect, fuel model, canopy cover, canopy base height, stand height and canopy bulk
density), a defined weather scenario (wind direction and wind speed) and fuel moisture
conditions. It is then able to compute fire behaviour characteristics for all the points in
the landscape, including flame length, fireline intensity and crown fire activity (Fig. 2).
FlamMap, has not a temporal component, and does not simulate the growth and

evolution of individual fires, although it is able to simulate a number of random fires in
order to estimate the probability of a point in the landscape to get burned. This last
characteristic of FlamMap, together with the relatively small time computational
requirements, makes the system appropriate for planning forest management under the
risk of fire (Gonzalez-Olabarria and Pukkala, 2011).
The layers defining the FARSITE landscapes required to operate FlamMap were
obtained from LiDAR data, by applying the constructed models for predicting the stand
level variables or by using the predicted stand variables for selecting the accompanying
NFFL fuel models. For the required weather and fuel moisture information, we
considered four scenarios: the combination of two alternative fuel moistures and two
alternative wind speeds, based on our experience in the area. The fuel moistures defined
were: 1) normal expected conditions in Soria during typical summer midday and 2)
conditions expected in dry Mediterranean conditions. The wind speeds defined were: 1)
mild wind (16 km h-1) and 2) a strong wind (32 km h-1). The wind was considered to be
constant over the whole area, direction south-west (210º).
FIGURE 2
Finally, once the models were developed, the next step was to spatialize the predictions
across the area, using the LiDAR available, in order to map and visualize variables that
represent the state of the forest across the whole landscape.
3. Results

The modelling of the stand level variables resulted in 13 models, with coefficients of
determination from 0.167 for BC, to 0.906 for Ho (Table 2). The RMSE obtained in the
cross-validation phase was on average 1.05 times higher, ranging from 1.02 to 1.30
times, than those obtained in the fitting phase. The coefficients of determination
obtained in the fitting phase were on average 1.07 times higher, than those obtained in
the fitting phase. All parameter estimates included in the models were significant. No
significant biases were observed when examined as a function of the predicted variable
and predictors of the model.
TABLE 2
The rules defined for classifying the forest into their corresponding fuel types (Table 3),
relied on cutting values of stand variables such as shrub and forest cover, the shrub and
tree height (HL), and the canopy biomass (FB + BB). The rules were found to classify
correctly 39% of fuels of the 431 NFI plots. However, when the fuel types were
grouped according their fire behaviour characteristics in pasture driven fuels (1, 2),
shrub driven fuels (5, 6, 7), and tree driven fuels (8, 9), the level of agreement between
the predicted and observed fuel type groups raised to 64.2% (Table 4).
TABLE 3
TABLE 4
The models allowed the spatialization of the stand level variables and the pre-defined
fuel types. In order to simplify the presentation of the results (Fig. 3), the maps
represent predictions for the whole area of the MUP89 (992.7 ha).

FIGURE 3
The potential fire behaviour was assessed by running FlamMap for the four scenarios
(alternative moisture and wind conditions) in the MUP89 forest. The predictions
resulting from the application of FlamMap are independent on the simulation of a single
fire event, and therefore are defined by the characteristics of each of the stands (map
cells of 500 m2), which allows to map fire related variables across the whole landscape
(Fig. 4) and evaluate the share of landscape surface belonging to a defined “fire
behaviour class” (Table 4).
In addition, FlamMap was also used to calculate the probability of fire occurrence,
obtained through the simulation of a number of fire events which served as a basis to
evaluate how many times a fire can potentially occur in a stand. We applied 500
ignitions randomly distributed across the landscape to identify areas more prone to be
affected by fire (Fig. 4). Finally, the results of the process are summarized in Table 4.
FIGURE 4
TABLE 5
4. Discussion
This study presents a methodology that combines the acquisition of LiDAR data and
fire behaviour simulators in order to provide geo-referenced and spatially continuous
information of forest resources and potential fire behaviour. The type of information

provided through this methodology can be a valid tool for forest management purposes,
since, presented in maps, such information is a basis for the analysis of the spatial
relation between the initial state and allocation of forest resources, in the one hand, and
the risk that fire entails for those resources, in the other. Additionally, the availability of
information about the state of the forest, instead of merely parameters that influence fire
behaviour, brings the possibility of integrating the management of forest fuels within
the timber management process. This is an important advantage for planning fire
prevention measures over longer periods of time, especially if they are included into
forest management planning at landscape-level (González-Olabarria and Pukkala,
2011).
For the implementation of the methodology, the simulator FlamMap was chosen to
study the fire behaviour, as it presents some interesting characteristics: it provides fire
hazard information that can be easily applied to the studied landscape, it can estimate
fire occurrence probabilities for all the forest stands based on multiple fire ignitions and,
at the same time, the extent of the fires caused by those ignitions, and, finally, the
outputs provided are sound indicators of potential fire damage on trees (e.g. fire
intensity, flame length, crown fire activity). These indicators can be used in post-fire
tree mortality models which, combined with the initial state of the forest, provide
estimates of tree mortality and timber losses (Peterson and Ryan, 1986; Hely and
Bergeron, 2003; Rigollot, 2004; Fernandes et al., 2008), as has already been done for
planning forest management (Bettinger, 2009; Kim et al., 2009). The use of FlamMap
for forest planning purposes has the additional advantage (when compared to other
simulators e.g. FARSITE), that delivers an overall idea of the risk of fire over the whole
landscape under different climatic conditions, instead of simulating the growth of

specific fire events. This is especially important when there is limited historic
information on recurrent fires, or when there is no information available about the
conditions leading into their occurrence and impact.
Concerning the assessment of the initial state of the forest, the data retrieval used for the
models was based on tree and stand level variables derived from LiDAR-based
inventory (Falkowski et al., 2010). The results of this study can therefore be used for
short-term operational planning and are a promising stepping stone for improving forest
planning systems that optimize forest management at landscape level. However, when
the information is to be applied for medium to long term forest planning (i.e. tactical
and strategic forest planning) it would require to predict the evolution of both forest
resources and fire related variables, subject to the proposed forest management
operations (Gonzalez and Pukkala, 2011). This is an aspect not considered in the study,
as in order to estimate the evolution of the forest resources over a planning period, it
would be necessary to apply tree growth models adapted to the local conditions, in
terms of composition of species and site characteristics, being at the same time adapted
for responding to the management strategies to be considered in the planning process.
On the other hand, the assessment of the fire related variables would require models that
relate the future evolution of the forest to the dynamics of its associated understory
vegetation (Coll et al., 2010) and dead fuel build-up (Reinhardt and Crookston, 2003).
Finally, specific limitations of the study are the relatively small area used and the
absence of data available on the fire regime of the area, which entails the impossibility
of predicting the probability of fire occurrence based on a strong analysis of the
probabilities of fire ignition. Another aspect to be consider are the lower coefficients of

determination resulting from the models predicting understory characteristics (shrub
cover and shrub height). This was an expected result, possibly due to the interception of
a large number of laser pulses by the upper tree canopy. However, the derived models
provided satisfactory cross-validation results, and the modelled effects of the forest
canopy cover on the predicted variables are consistent with the role that light
interception by trees plays in controlling the structure of the shrub strata in previous
studies (Coll et al., 2010). Nevertheless, due to the importance that shrub strata play in
defining fuel types, additional efforts are required in future research, in order to better
model the understory shrub structure,
Concerning the application of the methodology in other zones, it must be stressed that
the area studied was characterized by the dominance of carefully managed even-aged
forest, which is not considered to be the most hazardous regarding the risk of fire
(Gonzalez et al., 2006, 2007). Therefore, future studies can be oriented to validate the
methodology in landscapes dominated by other types of forest, with special focus in
uneven-aged forest structures, in order to identify potential weaknesses, and to propose
modifications to the methodology, when required.
Acknowledgements
Jose-Ramon González-Olabarria thanks the Ramon y Cajal program of the Spanish
Ministry for Economy for financial support and EFIMED, Cesefor and the
Mediterranean Model Forest Network for granting the additional support required for
visiting and working on the Model forest of Urbión. This study is conducted within the
CLAVE project “Parametrización y cartografía de especies vegetales en espacios
naturales de alto valor ecológico aplicando tecnologías de teledetección” co-funded by
Commented [m2]: mira a ver ahora

the Ministry for Economy and the private enterprises Agresta S. Coop. and Stereocarto
S.L. The authors greatly appreciate the technical and field support of David Lasala
Sanchez and the technicians of the Environment Department from Castilla y Leon.

References
Agee, J.K., Skinner, C.N., 2005. Basic principles of fuel reduction treatments. For.
Ecol. Manage. 211, 83–96.
Andersen, H.E., McGaughey, R. J., Reutebuch, S, E., 2005. Estimating forest canopy
fuel parameters using LIDAR data. Rem. Sens. Environ. 94, 441-449.
Anderson, H.E., 1982. Aids to determining fuel models for estimating fire behavior.
General Technical Report INT-122, United States Department of Agriculture, Forest
Service, Intermountain Forest and Range Experiment Station, Ogden, UT. 26 p.
Andrews, P.L., 1986. BEHAVE: Fire behavior prediction and fuel modeling system.
Burn subsystem. Part 1. Rep. No. GTR INT-194. USDA Forest Service,
Intermountain Forest and Range Experiment Station, Ogden, UT.
Arroyo, L., Pascual, C., Manzanera. J., 2008. Fire models and methods to map fuel
types: The role of remote sensing. For. Ecol. Manage. 256, 1239-1252.
Besseau, P., Dansou, K., Johnson, F., 2002. The International Model Forest Network
(IMFN): Elements of Success. Forest. Chron. 78(5): 648-654.
Bettinger, P., 2009. A prototype method for integrating spatially-referenced wildfires
into a tactical forest planning model. Res. J. Forest. 3, 8–22.
Bettinger, P., 2010. An overview of methods for incorporating wildfires into forest
planning models. Math. Comput. For. Nat. Res. Sci. 2, 43–52.
Boyd, D.S., Danson, F.M., 2005. Satellite remote sensing of forest resources: three
decades of research development. Prog. Phys. Geo. 29, 1 1-26.
Coll, L., González-Olabarria, J.R., Mola-Yudego, B., Pukkala, T., Messier, C., 2011.
Predicting understory maximum shrubs cover using altitude and overstory basal area
in different Mediterranean forest. Eur. J. For. Res. 130, 55-65.
DGCN. 2006. Tercer Inventario Forestal Nacional (2002–2004) Castilla y León: Soria,
Ministerio de Medio Ambiente, Madrid.
Falkowski, M.J., Hudak, A. T., Crookston, N.L., Gessler, P.E., Uebler, E.H., Smith,
A.M.S., 2010. Landscape-scale parameterization of a tree-level forest growth model:
a k-nearest neighbor imputation approach incorporating LiDAR data. Can. J. For.
Res. 40, 184-199.
Fernandes, P.M., Vega, J.A., Jiménez, E., Rigolot, E. 2008. Fire resistance of European
pines. For. Ecol. Manage. 256, 246-255.

Finney, M.A., 1998. FARSITE: Fire Area Simulator Model development and
evaluation. Research Paper RMRS-RP-4. USDA Forest Service, Rocky Mountain
Research Station, Ogden, UT, p. 47.
Finney, M. A., 2006. An overview of FlamMap fire modeling capabilities. In: Fuels
management—how to measure success: conference proceedings. 2006 March 28-30;
Portland, Oregon. Proceedings RMRS-P-41. Fort Collins, CO: U.S. Department of
Agriculture, Forest Service, Rocky Mountain Research Station: 213-220.
Finney, M.A., Seli, R.C., McHugh, C.W., Ager, A.A., Bahro, B., Agee, J.K., 2007.
Simulation of long-term landscape-level fuel treatment effects on large wildfires. Int.
J. Wildland Fire. 16, 712–727.
González, J.R., Palahí, M., Trasobares, A. and Pukkala, T. 2006. A fire probability
model for forest stands in Catalonia (north-east Spain). Ann. For. Sci. 63, 169-176.
González, J.R., Trasobares, A., Palahí, M., Pukkala, T., 2007. Predicting stand damage
and tree survival in burned forests in Catalonia (North-East Spain). Ann. For. Sci.
64, 733-742.
González-Olabarria, J.R., Pukkala, T., 2011. Integrating fire risk considerations in
landscape level forest planning. For. Ecol. Manage. 261, 278-297.
Hall, S.A., Burke, I.C., Box, D.O., Kaufmann. M.R., Stoker. J.M., 2005. Estimating
stand structure using discrete-return LiDAR: an example from low density, fire prone
ponderosa pine forests. For. Ecol. Manage. 208, 189-209.
Hély, C., Flannigan, M., Bergeron, Y., 2003. Modeling tree mortality following wildfire
in the southeastern Canadian mixed-wood boreal forest, For. Sci. 49, 566–576.
Hollaus, M., Wagner, W., Maier, B., Schadauer, K., 2007. Airborne laser scanning of
forest stem volume in a mountainous environment. Sensors. 7, 1559-1577.
Holmgren, J., Nilsson, M., Olsson, H., 2003. Estimation of tree height and stem volume
on plots using airborne laser scanning. For. Sci. 49(3), 419-428.
Kass , G. V., 1980. An exploratory technique for investigating large quantities of
categorical data . Appl. Stat. 29 , 119 – 127.
Kim, Y.H., Bettinger, P., Finney, M., 2009.Spatial optimization of the pattern of fuel
management activities and subsequent effects on simulated wildfires .Eur. J. Oper.
Res. 197, 253–265.

Kirby, C.L., Hall R.J., (Compilers). 1980. Practical applications of remote sensing to
timber inventory. Proceedings of a workshop held September 26-28, 1979 in
Edmonton, Alberta 169p
Maltamo, M., Packalen, P., Yu, X., Eerikainen, K., Hyyppa, J., Pitkanen, J., 2005.
Indentifying and quantifying structural characteristics of heterogeneous boreal
forests using laser scanner data. For. Ecol. Manage. 216, 41-50.
McGaughey, R. J., Carson, W.W., 2003. Fusing LIDAR data, photographs, and other
data using 2D and 3D visualization techniques. Proceedings of Terrain Data:
Applications and Visualization–Making the Connection, October 28–30, 2003 (pp.
16−24).
Mutlu, M., Popescu, C.S., Stripling, C., Spencer, T., 2008a. Mapping surface fuel
models using LiDAR and multispectral data fusion for fire behavior. Rem. Sens.
Environ. 112 (1), 274–285.
Mutlu, M., Popescu, S.C., Zhao, K., 2008b. Sensitivity analysis of fire behavior
modeling with LIDAR-derived surface fuel maps. For. Ecol. Manage. 256 (3), 289-
294.
Næsset, E., 2002. Predicting forest stand characteristics with airborne laser using a
practical two-stage procedure and field data. Rem. Sens. Environ. 80, 88-99.
Næsset, E., 2004. Practical large-scale forest stand inventory using a small-footprint
airborne scanning laser. Scand. J. For.Res. 19, 164-179.
Nelson, R.F., Krabill, W., Tonelli, J., 1988. Estimating forest biomass and volume
using airborne laser data. Rem. Sens. Environ. 24 (2), 247–267.
Peterson, D.L., Johnson, M.C., Agee, J.K., Jain, T.B., McKenzie, D., Reinhardt, E.D.,
2005. Forest structure and fire hazard in dry forests of the Western United States.
Gen. Tech. Rep. PNW-GTR-628. U.S. Department of Agriculture, Forest Service,
Pacific Northwest Research Station, Portland, OR.
Peterson, D.L., Ryan, K.C., 1986. Modeling postfire conifer mortality for long-range
planning. Environ. Manage. 10, 797–808.
Reed, W.J., Errico, D., 1986. Optimal harvest scheduling at the forest level in the
presence of the risk of fire. Can. J. For. Res. 16, 266 - 278.

Reinhardt E, Crookston NL (2003) The Fire and Fuels Extension to the Forest
Vegetation Simulator. Gen. Tech. Rep. RMRS-GTR- 116. USDA Forest Service,
Rocky Mountain Research Station, Ogden, UT, 209 p
Riaño, D., Meier, E., Allgöwer, B., Chuvieco, E., Ustin, S.L., 2003. Modeling airborne
laser scanning data for the spatial generation of critical forest parameters in fire
behavior modeling. Rem. Sens. Environ. 86, 177–186.
Riaño, D., Chuvieco, E., Condes, S., González-Matesanz, J., Ustin, S.L., 2004.
Generation of crown bulk density for Pinus sylvestris L. from LiDAR. Rem. Sens.
Environ. 92, 345–352.
Riaño, D., Chuvieco, E., Ustin, S.L., Salas, F.J., Rodríguez-Perez, J.R., Ribeiro, L.M.,
Viegas, D.X., Moreno, J.F., Fernández, H., 2007. Estimation of shrub height for fuel-
type mapping combining airborne LiDAR and simultaneous color infrared ortho
imaging. Int. J. Wildland Fire. 16, 341–348.
Rigolot, E., 2004. Predicting postfire mortality of Pinus halepensis Mill. and Pinus
pinea L. Plant Ecol. 171, 139–151.
Rodríguez, F. Lizarralde, I. Broto, M. Bravo, F. 2012. Merchantable volume equations
for the main species in the Spanish plateau. Manuscript.
Ruiz-Peinado, R., del Rio, M., Montero, G., 2011. New models for estimating the
carbon sink capacity of Spanish softwood species. For. Systems. 20 (1), 176-188.
Segur, M., 2009. Los bosques modelo y el modelo de Urbión. Montes. 98, 96-99.
Skowronski, N., Clark, K., Nelson, R., Hom, J., Patterson, M., 2007. Remotely sensed
measurements of forest structure and fuel loads in the Pinelands of New Jersey. Rem.
Sens. Environ. 108, 123–129.
Van Wagner, C.E., 1983. Simulating the effect of forest fire on long term annual timber
supply. Can. J. For. Res. 13, 451-457.
Wagner, W., 1993. Prediction of crown fire behavior in two stands of jack pine. Can. J.
Forest Res. 23, 442–449.

Table 1. Summary of the field plots used as ground data (N=160)
Parameter (unit)
Mean
Min
Max
N (stems ha-1)
702
60
2000
G (m2 ha-1)
30.7
1.3
72.1
Dg (cm)
24.0
13.6
43.7
Ho (m)
15.7
8.1
27.0
hL (m)
13.4
6.6
24.6
CBH (m)
3.9
0.1
12.4
V (m3 ha-1)
210.1
2.9
640.0
FB (t ha-1)
2.1
0.1
5.2
SB(t ha-1)
10.2
0.5
29.0
BB(t ha-1)
1.4
0.1
4.4
RB(t ha-1)
3.1
0.1
8.2
BC (%)
28.7
1.0
90.0
hB (m)
0.9
0.5
2.0
N: stem number; G: basal area; Dg: quadratic mean diameter; Ho: dominant height; hL: Lorey’s mean
height; CBH: Crown base height; V: stem volume; FB: Foliar biomass; SB: Stem biomass; BB: Branch
biomass; RB: Root biomass; BC: Shrub Cover; hB: Mean Shrub Height;

Table2. Models, predictors, and model’s assessment for the variables at stand level.
Fitting phase
Cross-validation
Dependent
variable
Predictive model
R2
RMSE
CN
R2
RMSE
N
181.76+(LFCC2.5417)*(LH_95-1.6616)*(LI_10-
0.1933)
0.640
206.98
39.2
0.609
215.80
G
(LFCC0.4183)*(LH_200.7545)*(LI_050.1089)
0.809
6.89
17.5
0.796
7.05
Dq
4.89*(LFCC-
0.1909)*(LH_Mean0.7997)*(LI_250.1702);
0.752
3.33
36.6
0.731
3.46
Ho
2.96*(LH_900.8468)*(LI_95-0.1070);
0.906
1.17
91.2
0.896
1.20
hL
1.13*(LH_900.9612)
0.900
1.17
21.5
0.892
1.19
CBH
0.1657*(LH_401.3879);
0.563
1.72
17.6
0.542
1.77
V
(LFCC0.3002)*(LH_Mean1.6721)*(LI_050.1757)
0.871
52.46
29.3
0.858
54.12
FB
(LFCC0.7767)*(LH_400.7363)*(LI_Mean-0.4757);
0.690
5.66
45.2
0.674
5.79
SB
(LFCC0.6711)*(LH_401.1725)*(LI_Mean-0.2296)
0.811
25.44
48.8
0.796
26.19
BB
(LFCC0.5634)*(LH_951.3488)*(LI_IQ-0.8220)
0.593
5.45
42.2
0.574
5.57
RB
0.0583*(LFCC0.9588)*(LH_300.9037)*(LI_050.1639);
0.864
6.55
93.6
0.852
6.74
BC
(LFCC0.6288)*(LBC0.3540)
0.167
21.97
6.1
0.136
22.38
hB
0.1732*(LFCC0.3154)*(LBC0.2119);
0.284
0.37
34.0
0.194
0.48
N: number of stems; G: basal area; Dq: quadratic mean diameter; Ho: dominant height; hL: Lorey’s mean
height; CBH: Crown base height; V: stem volume; FB: Foliar biomass; SB: Stem biomass; BB: Branch
biomass; RB: Root biomass; BC: Shrub Cover; hB : Mean Shrub Height; LFCC: LiDAR Forest Canopy
Cover; LBC: LiDAR Shrub Cover; LH_20. LH_30. LH_40. LH_90. and LH_95: corresponding to the
20th, 30th, 40th, 90th and 95th percentile height of the canopy pulses respectively (in meters); LH_Mean:
arithmetic mean height of canopy pulses (in meters); LI_05. LI_10. LI_25. LI_95: corresponding to the
5th, 10th, 25th, and 95th percentile intensities of the canopy pulses; LI_Mean: arithmetic mean intensity of
canopy pulses; LI_IQ= interquartile intensity range of canopy pulses ;
Commented [MSOFFICE3]: Alguno de estos no es el MEF que
esta en el texto???

Table 3. Description of standard Anderson´s (1982) fuel models. and the stand
variables used to allocate them; FCC: Forest Canopy Cover; BCC: Shrub Cover; hL:
Lorey Height; hB: Mean Shrub Height; W: Crown biomass (FB + BB).
Fuel Model
Typical Fuel Complex
Canopy and/or Shrub
cover (%)
Heigh (m)
Biomass
(t/ha)
Grass and grass-dominated
1
Short grass
(FCC+BC)<33
--
--
2
Timber (grass and
understory)
33<(FCC+BC) < 50
hB< 0.4
--
3
Tall grass
--
--
--
Chaparral and shrub fields
4
Chaparral
FCC >66
BC > 33
hL<7
hB>1.5
5
Brush
FCC > 25
BC >30
(FCC+BC) > 50
hB< 0.4
6
Dormant brush. hardwood
slash
FCC < 25
BC >30
(FCC+BC) > 50
hB< 0.4
7
Southern rough
FCC > 25
BC > 30
hB> 0.4
hL>7
Timber litter
8
Closed timber litter
FCC >50
BC < 30
hL>7
W > 45
9
Hardwood litter
FCC >50
BC < 30
hL>7
W < 45
10
Timber (litter and
understory)
--
--
--
Slash
11
Light logging slash
--
--
--
12
Medium logging slash
--
--
--
13
Heavy logging slash
--
--
--

Table 4. Predicted (based on the rules defined in Table 3),and observed fuel types for
431 plots of the national forest inventory.
Fuel type predicted
1
2
5
6
7
8
9
Total
(N)
Fuel Matching
(%)
Fuel group
matching (%)
Fuel type observed in field
1
0
0
0
0
2
0
0
2
0.0
2
1
1
2
0
4
0
4
12
8.3
14.2
4
0
0
6
0
10
0
4
20
0.0
0
5
4
2
30
2
26
1
8
73
41.1
6
0
0
12
0
39
0
9
60
0.0
76.9
7
5
0
7
0
74
2
26
114
64.9
8
2
0
1
0
11
5
18
37
13.5
9
5
1
13
0
18
4
58
99
58.6
62.5
10
0
0
0
0
2
1
0
3
0.0
11
1
0
4
0
3
0
2
10
0.0
0
12
0
0
0
0
1
0
0
1
0.0
Total
18
4
75
2
190
13
129
431
39.0
64.2

Table 5. Share of the MUP89 forest landscape (in ha) within a fire behaviour class.
Total area: 992.7 ha.
SCENARIO
FIRE CLASS
Fuel
Moisture
Wind
(km / h)
Crown fire activity
No fire
Surface Fire
Passive Crown Fire
Active Crown Fire
Normal
16.0
47.0
870.9
53.0
21.8
Normal
32.0
47.0
811.8
46.8
87.1
Dry
16.0
47.0
820.9
77.4
47.4
Dry
32.0
47.0
719.25
55.6
170.9
Heat per unit of area (KJ/m2)
<= 25000
25001 - 50000
50001 - 100000
> 100000
Normal
16.0
946.1
22.3
15.5
8.9
Normal
32.0
905.4
49.8
26.5
11.1
Dry
16.0
836.25
99.1
43.1
14.3
Dry
32.0
837.0
178.5
108.3
4.2
Burn probabilities
0
0.0001 - 0.0049
0.0050 - 0.0099
> 0.010
Normal
16.0
602.5
287.9
80.9
21.5
Normal
32.0
485.0
299.1
153.1
55.6
Dry
16.0
501.3
302.9
137.7
50.8
Dry
32.0
437.0
205.7
172.0
178.1
Flame length (m)
< 0.5
0.5 - 1
1.1 - 2
> 2
Normal
16.0
185.9
543.8
201.5
61.6
Normal
32.0
60.35
399.7
406.4
126.3
Dry
16.0
87.7
468.7
323.0
113.5
Dry
32.0
47.0
286.0
425.2
253.1

Figures’ Captions
Figure 1. Flowchart of the methods followed during the study
Figure 2: Input layers required to run FlamMap: a) elevation b) slope c) aspect d) fuel
model e) canopy cover f) canopy base height g) canopy bulk density and selected output
layers obtained from FlamMap: h) crown fire activity i) Flame length j) heat/Fireline
intensity k) burn probabilities
Figure 3. Mapping of some of the LiDAR derived variables in the MUP89 forest. From
left to right and top to bottom: a) Tree canopy cover; b) Stand dominant height (Ho); c)
Basal area; d) Marketable timber volume; e) Tree crown bulk density and f) NFFL Fuel
model.
Figure 4: Example of fire behaviour variables obtained through FlamMap simulation
for the “extreme scenario” (dry fuel moisture and wind speed of 32Km/h). A: Flame
length in meters; B: Fire type (0 equals to no fire. 1 equals to surface fire. 2 equals to
passive crown fire. and 3 equals to active crown fire); C: Probabilities of fire
occurrence.