fire

Article
Assessing the Accuracy of the Wildland–Urban Interface Index in
Portuguese Rural Villages’ Context: A Case Study of the 2017
Pedrógão Grande Wildfire
Dener Silva 1 , Tiago Miguel Ferreira 2 and Hugo Rodrigues 1, *

1 RISCO, Department of Civil Engineering, University of Aveiro, 3810-193 Aveiro, Portugal;

dener.silveira@ua.pt
2 College of Arts, Technology and Environment, School of Engineering, University of the West of
England (UWE Bristol), Bristol BS16 1QY, UK; tiago.ferreira@uwe.ac.uk
* Correspondence: hrodrigues@ua.pt

Abstract: Climate change and human interventions can boost wildfires. Although naturally happen-
ing, massive events are becoming more frequent and severe. In Portugal’s mainland, many rural
settlements are populated mainly by older people, and uninhabited houses are frequently poorly
conserved. This combination leaves the Wildland–Urban Interface (WUI) dangerously exposed to the
fires. Pursuing the understanding of WUI areas, this study applies the Wildland–Urban Interface
Index (WUIX) assessment methodology to an area severely affected by the massive 2017 wildfire of
Pedrógão Grande, Leiria, Portugal. The primary objective of this study was to compare the results
from WUIX with the areas burned during the fire event. As a result, maps of WUI effect were gener-
ated, visually pointing to villages with higher severity compared to the others. A statistical analysis
was performed in three villages from the region to validate the results by comparing the accuracy
of the results obtained to the actual damages. The results point out a high correlation between the
WUIX and the real scenario despite the apparent challenges in determining the variations in different
types of fire effect. Finally, the WUIX results align with the data from the Pedrógão Grande wildfire,
showing that some are promising in conjunction with other wildfire indicators.

Keywords: wildfire; Wildland–Urban Interface Index; rural villages

Citation: Silva, D.; Ferreira, T.M.;
Rodrigues, H. Assessing the Accuracy
of the Wildland–Urban Interface Index
in Portuguese Rural Villages’ Context:
1. Introduction
A Case Study of the 2017 Pedrógão
Grande Wildfire. Fire 2024, 7, 90. The occurrence of wildfires has been increasing over the years [1]. Although wildfires
https://doi.org/10.3390/fire7030090 have a natural character, the interval between massive events is becoming shorter, with an
impressive number of very deadly events recorded in recent decades. While Portugal is
Academic Editor: Grant Williamson
relatively used to and prepared to face wildfires, frequent during its dry Mediterranean
Received: 8 February 2024 summers, the devastating wildfires of 2017 have profoundly changed how the country
Revised: 13 March 2024 coexists with these events. In that year, more than 100 people died as a consequence of two
Accepted: 14 March 2024 significant events that occurred in the central region of the country. The municipality of
Published: 15 March 2024 Pedrógão Grande was the epicentre of the biggest one, which alone took the lives of about
66 people [2–4].
Besides these human losses, the official statistics point out that the effect of this
devastating wildfire also exerted tremendous pressure on the surrounding environment
Copyright: © 2024 by the authors.
and rural traditional settlements. In this massive wildfire, up to 80% of the area consumed
Licensee MDPI, Basel, Switzerland.
This article is an open access article
by the fire was forestry, and the remaining hardly affected areas are primarily composed of
distributed under the terms and
medium and small villages [5,6].
conditions of the Creative Commons
Like in many other countries [7,8], small rural villages in Portugal have struggled
Attribution (CC BY) license (https:// to retain their populations [9], as these places are losing most of their young families to
creativecommons.org/licenses/by/ the cities in the pursuit of better work opportunities. Thus, the villages are frequently
4.0/). composed of older remaining people with lesser mobility conditions and a significant

Fire 2024, 7, 90. https://doi.org/10.3390/fire7030090 https://www.mdpi.com/journal/fire

Fire 2024, 7, 90 2 of 15

number of uninhabited and poorly conservated houses. This is a dual challenge for an
evacuation situation, as elderly residents may take a longer time to abandon these areas,
and uninhabited buildings tend to delay the detection of the event by the population [10].
In the case of the 2017 wildfires in Portugal, almost 50% of the unoccupied buildings—
representing around 12.5% of the entire building stock—were destroyed [11]. The combina-
tion of these factors, old populations, and a large volume of unoccupied buildings, with
the proximity of those settlements to the forest results in a high level of exposure, and that
is why it is critically important to assess these wildland–urban interfaces (WUI) and to
outline and implement effective protection measures [12].
The term WUI is not standardised globally, there are discrepancies regarding the size of
areas considered representative, the safe distance between forests and settlements, and the
combination of data to be considered when assessing it [13–15]. Beyond these questions, it is
recognised that WUI areas worldwide tend to increase due to the expansion of urbanisation
into forests and deficiencies in fuel management [16–18]. This trend dangerously exposes
more people to wildfires, who are often unprepared to deal with such events [19] or
incapable of doing so [18,20].
Considering these points, the task of identifying WUI areas has been pursued among
researchers across the globe. Huang and Jin [21] developed a deep-learning-based method-
ology for identifying the building footprint, vegetation cover and distance to remote areas
in California, USA, over eight years. The work is interesting as it uses only satellite images
available through a governmental agency as the primary data source. With these, the
findings of the study provided accurate results in measuring the evolution of WUI areas
over time. However, this technique is limited by the existing available data, which are
frequently suitable only for large-scale analysis due to the resolution of the imagery.
To encompass finer analysis on identifying precise WUI areas, several studies have
been carried out by combining different remote sensing techniques, such as LiDAR and
high-resolution aerial images. While LiDAR equipment is very powerful in detecting
the three-dimensional properties of vegetation and relief, high-resolution images are
more easily encountered on open-access databases, avoiding the need for fieldwork.
Alvaréz et al. [22] developed a study in Galicia, Spain, aimed at characterising existing
biomass, for instance, proportions of vegetation, canopy sizes and density. The study was
mainly focused on the collection of point clouds with UAV and automatic classification of
the point clouds for differentiating the vegetation. The technique proposed by the authors
proved relatively efficient in detecting canopies, although challenges were found when
dealing with more intricate vegetation patterns.
When developing a strategy for identifying the WUI areas, many considerations may
be made to understand the actual vulnerability of urbanisation to wildfires. The study
by Tampekis et al. [23] defines that far from just considering the hazard in mapping the
wildfire risk, the socio-ecological aspect also plays an essential role in the WUI areas.
In their theoretical work, the authors propose to join the effect of affected areas, socio-
ecological impacts, and the effects of decision variables on improving the resilience of the
urban–forest interfaces.
While the combination of aerial imagery with deep learning algorithms has nowadays
leveraged the precision in identifying buildings and vegetation in WUI areas [24,25],
interpreting the outcomes of these methods and the usual demand for large training
datasets may restrict their applicability of all areas. In this perspective, Pereira et al. [26]
proposed assessing the WUI areas in Portugal using urbanisation data provided by the
government. The study successfully mapped the large-scale WUI areas, but the authors
mentioned that a detailed delineation of these segments along the perimeter of urban
areas would enhance the methodology. Nevertheless, relying solely on the post-processed
governmental data may subject the methodology to updates and cause delays in the
monitoring processes.
Considering the complexity of defining a methodology to identify interface zones, it
often becomes necessary to adapt those to suit regional requirements, bearing in mind the
Fire 2024, 7, 90 3 of 15

specificities of the area. In developing the Wildland–Urban Interface Index (WUIX) method-
ology, Caballero [27] tested his index in the post-wildfire event of Mati, Greece [28], yielding
promising results in terms of correctly identifying exposure levels in a local/regional con-
text. However, no further validation was carried out to enable the application of this
method for the Mediterranean region as a whole [29–31], thus limiting the results when
applied to the Portuguese context, for example.
From this perspective, considering the fundamental high value of accurate infor-
mation about the WUI areas in villages, this paper seeks to provide a validation of the
WUIX methodology through its application to Pedrógão Grande, Portugal, and the subse-
quent comparison of its results with actual fire scenarios in three villages affected by the
abovementioned 2017 events [32]. Ultimately, this work aims to contribute to enhancing
the understanding of the WUI conditions of those villages at that time and to discuss
the suitability of the WUIX methodology to be applied to the Portuguese territory, ulti-
mately leading to more effective wildland–urban interface assessment and pointing out
mitigation strategies.

2. Methodological Framework
To achieve the objective of the WUI assessment of the Pedrógão Grande region, the
definition of the most affected villages was initially conducted. Through an examination
of the aerial imagery related to the affected area [32], ruins were identified, and based on
the places with a higher concentration of them, the most affected villages as well as the
extension of the damages were characterised.

2.1. Area Definition and Data Collection

As in a typical wildfire development, while the higher concentration of buildings
and urban infrastructure shielded larger urban areas by reducing the available fuel, the
surrounding villages within the Pedrógão Grande region were hardly affected by the event.
These villages are predominantly small, and their geographical proximity to wildland areas
resulted in the spread of vegetation and fuels into their urban environment. This scenario
creates ideal conditions for the rapid percolation of fire and affecting of the dwellings.
After a visual screening of the post-disaster satellite image, a region was identified as
gathering a significant concentration of buildings destroyed. The region is in the middle
of the triangle formed by the three major cities in the area: Pedrógão Grande, Figueiró
dos Vinhos, and Castanheira de Pera (Figure 1). In this area, besides a high concentration
of forest, bushlands, and diverse vegetation, generally set in a very rugged terrain, the
landscape is interspersed by numerous rural villages. These villages are primarily agrarian,
i.e., their economy is based mainly on agricultural resources, although industrial and
service buildings and zones were also identified.
Based on the collected information, it was determined that the most suitable choice
for the case study would be villages located within this area. Among the possibilities, the
villages of Figueira (Pedrógão Grande), as well as Sarzedas de São Pedro and Sarzedas de
Vasco (both from Castanheira de Pera), were selected based on two primary criteria: the
impact of the 2017 wildfires in those villages and their size.
Each of these villages was severely affected by the wildfires, yet they experienced
no complete destruction, largely due to the efforts of firefighters and property owners
who worked tirelessly to protect the dwellings and properties. The size of the villages
was another factor in this selection since larger villages would implicate a more complex
preparation process, involving a much higher number of polygons to be drawn without
any expected positive contribution to the results since larger villages tended to be more
protected by their urbanisation and fire combat. By focusing on these smaller and heavily
affected villages, this study aims to reflect the impact of wildfires on WUI areas in a
general sense.
Fire 2024, 7, 90 Fire 2024, 7, x FOR PEER REVIEW 4 of 15 4

Figure
Figure 1. Schematic map 1.
of Schematic mapvillages.
the case-study of the case-study villages.by
The areas affected The
theareas
2017 affected by the 2017 fires are
fires are highlighted
lighted
in red in the figure. in red in the figure.

Based on the
As a result of large-scale collected
natural information,
hazards, such as theit was
onedetermined
that occurred that
inthe most suitable c
Pedrógão
for the case study would be villages located within
Grande, Maxar Technologies [32] has provided high-resolution aerial imagery of thethis area. Among the possibilitie
af-
villages of Figueira (Pedrógão Grande), as well as Sarzedas
fected regions. This imagery, which is intended to serve as a dataset for conducting de São Pedro and Sarzed
Vasco (both
post-disaster damage from Castanheira
assessment, has enoughde Pera),
detail to were selected
identify based
the ruins and onpaths
two primary
of the criteria
fires. However, impact
due to of
thethe 2017 wildfires
processes in those villages
of orthorectification andpictures,
of the their size.
the exact location
Each differs
of the buildings slightly of thesefrom
villages
the was
imagesseverely
fromaffected by the wildfires,
2015 provided yet they experience
by the Directorate
General (DGT) complete destruction,
[33] that will be used largely due to the efforts
for the methodology, of firefighters
for which and property
more details will be owners
worked tirelessly to protect the dwellings and properties.
presented in the next section. Through this process, the observed ruins in the vicinity The size of the villages
of wa
other factor in this selection since larger villages would implicate
burnt vegetation were identified as caused by the fire, and to validate whether the5ruins a more complex pre
Fire 2024, 7, x FOR PEER REVIEW of 16
tion process,
were already existent by theinvolving
time theya were
muchcaptured
higher number
two yearsof polygons
before theto be drawnthe
disaster, without an
same manual identification procedure was also performed for the 2015 images (Figure 2).more prot
pected positive contribution to the results since larger villages tended to be
by their urbanisation and fire combat. By focusing on these smaller and heavily aff
villages, this study aims to reflect the impact of wildfires on WUI areas in a general se
As a result of large-scale natural hazards, such as the one that occurred in Pedr
Grande, Maxar Technologies [32] has provided high-resolution aerial imagery of th
fected regions. This imagery, which is intended to serve as a dataset for conducting
disaster damage assessment, has enough detail to identify the ruins and paths of the
However, due to the processes of orthorectification of the pictures, the exact locati
the buildings slightly differs from the images from 2015 provided by the Directorate
eral (DGT) [33] that will be used for the methodology, for which more details will be
sented in the next section. Through this process, the observed ruins in the vicinity of b
vegetation were identified as caused by the fire, and to validate whether the ruins
already existent by the time they were captured two years before the disaster, the
manual identification procedure was also performed for the 2015 images (Figure 2).

Figure
Figure 2.
2. Ruins
Ruins identified
identified in
in aerial
aerial imagery
imagery from
from the
the village
village of
of Figueira:
Figueira: (a)
(a) the
the DGT
DGT image
image from
from
2015, and (b) Maxar image from 2017.
2015, and (b) Maxar image from 2017.

After the villages’ selection, the polygons of the buildings were drawn through aerial
images from 2015 provided by the DGT (Figure 3). This ensures that the analysis of inter-
faces between buildings, urban infrastructures and vegetated areas will be accurate when
using this correspondent orthoimage. This was a critical stage because the analysis was
made based on the patterns of the pixels on the images, meaning that minor discrepancies
Fire 2024, 7, 90 Figure 2. Ruins identified in aerial imagery from the village of Figueira: (a) the DGT image5from
of 15
2015, and (b) Maxar image from 2017.

After
After the
the villages’
villages’ selection,
selection, the
the polygons
polygons of
of the
the buildings
buildings were
were drawn
drawn through
through aerial
aerial
images
images from 2015 provided by the DGT (Figure 3). This ensures that the analysis of
from 2015 provided by the DGT (Figure 3). This ensures that the analysis ofinter-
inter-
faces
faces between
between buildings,
buildings, urban
urban infrastructures
infrastructures and
and vegetated
vegetated areas
areas will
will be
be accurate
accurate when
when
using
using this correspondent orthoimage.
this correspondent orthoimage. ThisThis was
was aa critical
critical stage
stage because
because the analysis was
the analysis was
made based on the patterns of the pixels on the images, meaning that minor discrepancies
made based on the patterns of the pixels on the images, meaning that minor discrepancies
resulting
resulting from
from the
the construction
construction oror demolishing
demolishing ofof buildings,
buildings, as
as well
well as
as differences
differences in
in the
the
process of orthorectification and other interferences, can interfere on the inferences
process of orthorectification and other interferences, can interfere on the inferences made made
from
from the
the analysis.
analysis.

Figure 3. Map of the polygons from the three villages. Buildings are labelled as affected and non-
Figure 3. Map of the polygons from the three villages. Buildings are labelled as affected and
affected.
non-affected.

2.2. Background and Application of the Wildland–Urban Interface Index (WUIX)

Within the selection of the villages and the visual delineation of building polygons
in the 2015 DGT images, it was possible to apply the WUI assessment method [27]. This
process was facilitated by a Python script developed for data preparation, processing, and
implementation [34]. The script automates the process of collecting aerial images from the
database, extracts the vegetation by using well-established indices and the buildings from
the vectorial file, converts these data into inputs for applying the WUIX and converts these
back into georeferenced results to be analysed and the map of interface types1 .
To assess the WUI areas, the Wildland–Urban Interface Index (WUIX)
methodology [27] systematically approaches the zones provided. It evaluates the level
of exposure to wildfires when they result from the interplay between human settlements
and vegetation. This methodology resorts to aerial or satellite imagery combined with
georeferenced building polygons, information that, although it can be obtained from public
repositories or open access geodata sources, needs to be treated in a way to separate the
vegetation from the images and combined with building information.
The first step concerns the collection of aerial images. These aerial images are obtained
from national agencies and serve as the foundation for the analysis. The basic requirements
for the images are related to the resolution required to correctly identify the local elements
as well as the availability of infrared imagery to pair with the natural satellite image. This
element is necessary to manipulate the pixels on the images and identify the vegetation. To
 building polygons, information that, although it can be obtained from public repositories
or open access geodata sources, needs to be treated in a way to separate the vegetation
from the images and combined with building information.
The first step concerns the collection of aerial images. These aerial images are ob-
Fire 2024, 7, 90
tained from national agencies and serve as the foundation for the analysis. The basic re-
6 of 15
quirements for the images are related to the resolution required to correctly identify the
local elements as well as the availability of infrared imagery to pair with the natural sat-
ellite image. This element is necessary to manipulate the pixels on the images and identify
ensure the accuracy
the vegetation. of the
To ensure thesubsequent
accuracy ofprocedures
the subsequent andprocedures
of the comparison
and of the ofcompari-
the outputs
with the burnt areas from Maxar images, the images were collected
son of the outputs with the burnt areas from Maxar images, the images were collected based on georeferencing
calculated on a Python script.
based on georeferencing Onlyon
calculated orthophotos from 2015,
a Python script. Only which is the latest
orthophotos imagery
from 2015,
produced by the Portuguese Agency (DGT) with the required spectral
which is the latest imagery produced by the Portuguese Agency (DGT) with the required bands (R, G, B, and
NIR), were used in this research.
spectral bands (R, G, B, and NIR), were used in this research.
Following the acquisition
Following the acquisitionofofaerial
aerialimages,
images, thethe Python
Python script
script waswas
usedused to apply
to apply colour
colour
band extraction,unravelling
band extraction, unravelling thethe problem
problem of accurately
of accurately capturing
capturing the vegetation
the vegetation from thefrom
the orthoimages.
orthoimages. By applying
By applying two different
two different indices,indices,
such as such as the Normalised
the Normalised DifferenceDifference
Vege-
Vegetation
tation IndexIndex
(NDVI)(NDVI) and
and the the Green
Green Normalised
Normalised Difference
Difference Vegetation
Vegetation Index (GNDVI),
Index (GNDVI), the
the
amountamount of vegetation
of vegetation waswas estimated
estimated basedbased ondensity
on its its density
andand vitality.
vitality. Details
Details about about
the the
selection
selection of the most
mostappropriate
appropriateindexindexforforthe
the requirements
requirements of the
of the analysis
analysis andanddatadata
aboutabout
the
the vegetation can be
vegetation can befound
foundininGNDVI.
GNDVI.
After preprocessingand
After preprocessing andpreparing
preparingthe thedata
data about
about thethe vegetation
vegetation andand housing
housing for the
for the
application of WUIX (Figure 4), these will serve as an input for the methodology
application WUIX (Figure 4), these will serve as an input for the methodology to proceed to proceed
to the
to the core
core of the process
processby bydetermining
determiningand andlevelling
levelling zones with
zones elevated
with elevatedWUI WUIexposure
exposure
due
due to to the interaction between the built environment and the surrounding
interaction between the built environment and the surrounding vegetation. vegetation. SuchSuch
isis determined
determined herein
hereinusing
usingthe
theContinuity
Continuity andandFriction
Frictionindices andand
indices WUIX
WUIX proposed
proposed by [27].
by [27].

Fire 2024, 7, x FOR PEER REVIEW 7 of 16

Figure Detail
Figure 4.4.Detail of of
thethe identified
identified vegetation
vegetation and building
and building polygons
polygons in Figueira:
in Figueira: (a) aerial
(a) aerial image high-image
highlighting identified vegetation, and (b) replication of image (a) with highlighted
lighting identified vegetation, and (b) replication of image (a) with highlighted buildings. buildings.

In short,
In short,thethefirst
firststep
stepfor
forapplying
applyingthe the WUIX
WUIX approach
approach involves
involves creating
creating calculation
calculation
matrices:
matrices: the study area is divided into matrices with two levels, each having differentdifferent
the study area is divided into matrices with two levels, each having res-
resolutions, creating
olutions, creating a hierarchical
a hierarchical structure
structure (Figure(Figure 5). In
5). In this this scenario,
scenario, the innerthelevel
inner level is
is the
the
split of all the data into a matrix with fine cells, while the second level is formed by grouping by
split of all the data into a matrix with fine cells, while the second level is formed
grouping
them in a set them in a set of
of matrices matrices
where each of where
their each
large of their
cells haslarge cells N²
a number hasofafine cells. N²
number Fromof fine
cells. From
this, two keythis, twoare
metrics keydefined:
metricsRare defined:
f (fine Rf (fine
resolution), andresolution), and Rg (large resolution).
Rg (large resolution).

Figure 5.
Figure 5. Schematic
Schematicimage
imagerepresenting
representingthe matrix
the division.
matrix division.

The choice of N, the large resolution basis, is critical for interpreting the outputs.
Smaller N values result in finer results and less concentrated information. Larger N values,
on the other hand, may lead to misrepresentations due to a high number of fine cells being
diluted in a few large ones. The value of N should be appropriate for the analysis, such as
Fire 2024, 7, 90 7 of 15

The choice of N, the large resolution basis, is critical for interpreting the outputs.
Smaller N values result in finer results and less concentrated information. Larger N values,
on the other hand, may lead to misrepresentations due to a high number of fine cells being
diluted in a few large ones. The value of N should be appropriate for the analysis, such as
garden analysis (8–10 m) or interface configuration (50–100 m).
Continuity (Ci ) and Friction (Fi ) are the key coefficients of the WUIX and are com-
puted based on each fine-resolution cell. These coefficients are gathered by measuring the
proportion of neighbouring vegetation as well as the existence of a building polygon in
that cell and its adjacencies. For Ci , a value of 1.0 is added for each adjacent cell containing
vegetation and 0.5 for the corner cells. As for Fi , a similar process was performed for cells
containing overlap between vegetation and buildings, summing up 1.0 more value for
overlap within the cell itself. In summary, Ci ranges from 0.0 to 6.0, while Fi ranges from
0.0 to 7.0.
This matrix of values is then aggregated into large-resolution cells (Cg and Fg ) by
capturing the fine cells constrained inside each large cell and summing up its multiplication
by each Rf . Continuity and Friction coefficients are obtained through Equations (1) and (2).
N2
Cg = R f ∑ Ci (1)
i =1

N2
Fg = R f ∑ Fi (2)
i =1

Ultimately, the Wildland–Urban Interface Index (WUIX) (Wg ) for every Rg cell is
acquired by multiplying the large cell of continuity by friction, Equation (3). The final
WUIX is expressed in square metres (m2 ).

Wg = Cg × Fg (3)

To enhance the readability of the results, the values obtained for Rg can be converted
in two different ways. First, by comparing each indicator to the worst-case scenario, it is
possible to convert the scale of the values into a percentage. Although these indices became
comparable between the different grid sizes, they might be challenging to interpret since
results close to the worst-case scenario are not commonly obtained. To convert the outputs
into more meaningful representations, Continuity and Friction indices and WUIX can be
converted into specific values by area. This conversion allows for intuitive comparisons
and analyses.
Since this study involves conducting a comparison between the sizes of the grid (Rg ),
the most adequate conversion of WUIX values is the normalised one, explained above. For
further details about the conversion, please refer to Caballero (2019) [27]. Yet, following the
range recommended by the author, Rf was defined in 1 m. The N value was set in a range
from 5 to 65 m in intervals of 5 m.

2.3. Testing the WUIX Outputs

The georeferenced WUIX values and the data observed from the 2017 post-disaster
satellite imagery [32] were compared to validate the obtained results. This evaluation aimed
to understand the accuracy of WUIX methodology in identifying the vulnerable areas in
the Portuguese context of Pedrógão Grande, and to point out limitations on the process
of classifying areas that should be identified as burned or unburned when calculating
their WUIX.
To provide means for this comparison, buildings were further classified into three
groups based on the impact of the wildfire on them. This was carried out by analysing
the burned areas on the MAXAR satellite image. The categories are as follows: “No-affec”
for the buildings that, based on the satellite image, was clear to see were affected by the
fire; “Surro-affec” for the buildings whose surroundings were noticeably burned, although
Fire 2024, 7, 90 8 of 15

these might not have been burned down by the fire (it was inconclusive from the satellite
image); and “build_affec” for the buildings that were noticeably affected by the fire (often
in ruin or with apparent damages on the roof).
By updating the shapefile, it was possible to assign to each building both the WUIX
found based on the DGT orthophoto of 2015 and the assessed categories from the post-
event image after the massive wildfire in 2017. The process of assigning WUIX values
for each building is also based on a condition: since the definition of N value does not
necessarily match the size of the polygon, the highest WUIX value intersecting the building
was considered the relevant one. This way, the largest identified exposure is defined as the
standard value of the polygon.
From the comparison, it was expected to obtain lower values of WUIX for non-affected
areas and higher values for those affected directly and by the surroundings. This outcome
would support the idea of the correlation between WUIX and the ground truth. The process
of testing WUIX was performed through a Python script that imports the data and applies
statistical methods to assess the accuracy of WUIX2 , and the results of this analysis are
presented in the next section.

3. Results and Discussion

Before analysing the variations between the different types of fire effect, a correlation
analysis was performed on each N number, from 5 to 100, with increments of 5, to assess the
influence of this parameter on the normalised WUIX. To perform the correlation analysis,
the normality of the data was tested using two different statistical tests (see Table 1).

Table 1. Hypothesis testing on the normality of the data, p-value = 0.05.

Shapiro–Wilk Kolmogorov–Smirnov
Village
Fire 2024, 7, x FOR PEER REVIEW Skewness Kurtosis Normality
9 of 16
Statistic p-Value Statistic p-Value
Figueira 0.65 4.12 × 10−19 5.00 × 10−01 2.28 × 10−42 2.523 6.281 Rejected
Sarzedas de Vasco 0.835 5.94 × 10−09 5.27 × 10−01 2.13 × 10−25 1.36 1.087 Rejected
Sarzedas
Sarzedas dePedro
de São Vasco 0.835 1.19
0.868 5.94 −13
× ×1010 −09 5.27 10−01−01
5.00××10 2.13
6.23× × −25−57
1010 1.36
1.732 1.087
5.228 Rejected
Rejected
Sarzedas de São Pedro 0.768
0.868 1.19 × 10−13 5.00 × 10−01−01 6.23 × 10−57−121 1.732 5.228 Rejected
Entire sample 1.79 × 10−26 5.00 × 10 5.75 × 10 2.245 6.237 Rejected
Entire sample 0.768 1.79 × 10−26 5.00 × 10−01 5.75 × 10−121 2.245 6.237 Rejected

The table
The table shows
showsthat
thatininallallvillages
villages andand both
both tests,
tests, there
there waswas a failure
a failure to keep
to keep the the
hypothesis
hypothesis ofofparametric
parametric data.
data. These These
teststests confirmed
confirmed the hypothesis
the hypothesis suggestedsuggested by the
by the data
data that most
that most of theofWUIX
the WUIX
valuesvalues are close
are close to zero
to zero but not
but are are negative,
not negative, meaning
meaning that that
the the
distribution is
isskewed
skewedtotothe
theleft.
left.From
From thethe
non-parametric
non-parametric assumption,
assumption, a Spearman’s rankrank
a Spearman’s
correlation test
correlation testwas
wasperformed
performedbetweenbetween thethe
product
productof of
thetheWUIX
WUIX methodology
methodology and and
the the
actual scenario
scenariofor
foreach
eachvillage
villageandandthe theentire sample
entire sample(Figure
(Figure6). 6).

Figure 6. Correlation
Correlationbetween
betweenWUIX
WUIXand
andreal-case effect
real-case (both
effect normalised)
(both by N_value.
normalised) by N_value.

In order to implement it, the categories of fire effect were converted into two catego-
ries: “both buildings and the surroundings were affected”; and “non-affected”. In this
way, it was possible to obtain a better correlation between the higher WUIX values and
the affected buildings. Likewise, the correlation between the lower WUIX values and the
non-affected group was also improved. Additionally, to avoid misinterpretations, when-
Fire 2024, 7, 90 9 of 15

In order to implement it, the categories of fire effect were converted into two categories:
“both buildings and the surroundings were affected”; and “non-affected”. In this way, it
was possible to obtain a better correlation between the higher WUIX values and the affected
buildings. Likewise, the correlation between the lower WUIX values and the non-affected
group was also improved. Additionally, to avoid misinterpretations, whenever the WUIX
values were divergent by more than 2%, they were considered non-correlated.
The graph of Figure 6 shows that the correlation of the group “Entire sample” presents
a steady increase for N values ranging between 5 and 25 and peaks when N equals 40
(Spearman’s rank correlation coefficient = 0.80). For N values greater than 40, Spearman’s
rank correlation coefficient remains relatively constant, which means that the correlation is
no longer affected by the limitation on non-affected buildings, as explained above. This is
because the N value is negatively correlated with the WUIX, i.e., the greater the value of N,
the smaller the WUIX. Therefore, when N is equal to or greater than 45, no non-affected
building presents a WUIX value greater than 2.
When analysing the curves given in Figure 5 individually, it is possible to observe that
Sarzedas de Vasco presented a higher, almost constant correlation, which diverges from
the other curves. On the other hand, Figueira showed a lower correlation with the actual
scenario overall; this fact might be related to the existence of some isolated buildings that
were not captured by WUIX. The trends illustrated in Figure 5 may denote that for larger
grid sizes, the results of the WUIX are very attenuated, resulting in a loss of resolution.
From the application of the WUIX methodology, a set of different ways to show the
same results can be generated. Based on the previous correlation analysis, the normalised
WUIX and an N value equal to 40 were chosen to produce the outputs for the three villages
(Figure 7). The ruler in the map’s legend represents the data’s quartiles. As can be observed,
the values close to zero are illustrated in the figure in green colour. The buildings are
represented with red, yellow and green dots, standing for affected, surrounded and not
affected, respectively.
The values obtained for the normalised WUIX being so low, not surpassing 4%, can
be explained by the characteristics of these particular settlements. The tests carried out
by Caballero [27] were made in areas with a great amount of intermixed vegetation, with
trees and shrubs directly surrounding the buildings. That is not common in Portugal, as
was already observed in the Montesinho Natural Park’s reality [34], where the vegetation
surrounds the village as a whole instead of the buildings individually.
Summarising the fire effect attribute, an effect of 69.9% on the buildings directly or
indirectly by the surroundings was found (Table 2). Figueira was the village with higher
severity of the effects of fire, while Sarzedas de Vasco showed less damage. Among the
numerous reasons for the discrepancy, the distance from the forest and the dispersion of
the buildings may have contributed to it.

Table 2. Buildings and fire effect of the villages.

Fire Effect
Village No. of Buildings % Effect *
Building Affected Surrounding Affected Non-Affected
Figueira 181 74 73 34 81.2%
Sarzedas de São Pedro 244 52 105 87 64.3%
Sarzedas de Vasco 96 30 26 40 58.3%
Entire sample 521 156 204 161 69.9%
* Either the building or just the surroundings.

From Table 2, it is possible to notice that the category “Surrounding affected” includes
a significant percentage of buildings. In this category, it is assumed that many of these
buildings could be burned due to the fire; however, due to fire combat or other types
of intervention acting as a suppressor of the effect, the fire was deviated to the building
surroundings. There are also cases in which the fire damage was not evident on the
Fire 2024, 7, 90 10 of 15
Fire 2024, 7, x FOR PEER REVIEW 10 of 16

satellite image, and, therefore, this adds uncertainty to the analysis. However, since WUIX
quantifies the exposure
are represented to the
with red, wildfire,
yellow and given
and green dots, the previous
standing for considerations, the buildings
affected, surrounded and
with close surroundings
not affected, respectively.affected by the fire were considered affected for comparison.

Figure 7. Normalised WUIX values were obtained for the three villages using N = 40. (a) Sarzedas
Figure 7. Normalised WUIX values were obtained for the three villages using N = 40. (a) Sarzedas de
de São Pedro; (b) Sarzedas de Vasco; (c) Figueira.
São Pedro; (b) Sarzedas de Vasco; (c) Figueira.
Fire 2024, 7, 90 11 of 15

Yet, in Table 2, it is evident that each village had a different behaviour. While in
Sarzedas de Vasco, the fire effect is almost evenly distributed among the three categories,
the buildings and the surroundings in Figueira were more severely affected. In Sarzedas de
São Pedro, on the other hand, the surroundings were particularly affected. This is probably
due to the size of the village, which is significantly larger than Sarzedas de Vasco and
Figueira, how the fire was combated, and the greater dimension of non-vegetated areas.
To assess the distribution of WUIX for each building, the overlap of “build_affec”
and “surro_affec” performed for Spearman’s rank correlation was used here, defining
the category as “Affected”. The plot in Figure 8 shows the distribution of affected and
non-affected buildings, where greater WUIX values express generally larger friction areas
Fire 2024, 7, x FOR PEER REVIEW 12 of 16
between the vegetation and building and, consequently, higher chances of the building
being hit by the fire due to its higher exposure.

Figure
Figure 8.
8. Distribution
Distribution of
of WUIX
WUIX for
for each
each building
building regarding
regarding and
and effect.
effect.

From
From the analysis
analysis presented
presentedin inFigure
Figure7,7,ititisispossible
possibletoto observe
observe anan apparent
apparent disper-
dispersion
sion ofresults
of the the results for low
for low WUIX WUIX values
values (between
(between 0 and 0 and 1). For
1). For higher
higher WUIX WUIX values,
values, on
on the
the other
other hand,
hand, there
there is ais clearer
a clearer correlationbetween
correlation betweenthe theWUIX
WUIXand and the effect,
effect, as
as expected.
expected.
Specifically, Figueira
Specifically, Figueirapresents
presentsaabetter
bettercorrelation,
correlation,consistently
consistently showing
showing burnt buildings
burnt for
buildings
WUIX
for WUIXvalues above
values 1.0. This
above correlation
1.0. This is notissonot
correlation clear
so in the in
clear other
the two
othervillages, particularly
two villages, par-
for Sarzedas
ticularly de São Pedro,
for Sarzedas de São which
Pedro, presents severalseveral
which presents cases of non-burned
cases of non-burnedbuildings with
buildings
a high
with WUIX
a high WUIXvalue.
value.NoNo data areare
data considered
consideredrelated relatedtotothe
thelocal
local interference
interference with fire fire
protection, which can change the relation with the observed mismatch of results.
protection,
The Kruskal–Wallis
The Kruskal–Wallis test test was
was carried
carriedoutouton
onthe thethree
threevillages
villagesseparately
separatelyand andjoined
joinedto-
to-
gether to
gether to assess
assess the
the difference
difference between
between thethe groups.
groups. Table
Table 33shows
shows thethe results
results of
of the
theanalysis.
analysis.

Kruskal–Wallis results for the villages with N = 40 and a significance of 5%.

Table 3. Kruskal–Wallis
Table

WUIXMedian
WUIX Medianperper Group
Group Kruskal–Wallis
Kruskal–WallisTest
Test
no_affec
no_affec surro_affec
surro_affec build_affec
build_affec stat
stat p-Value
p-Value
Figueira 0.09 0.26 0.17 4.08 0.13
Figueira 0.09 0.26 0.17 4.08 0.13
Sarzedas
Sarzedas de Pedro
de São São Pedro 0.53 0.53 0.51
0.51 0.580.58 1.10
1.10 0.58
0.58
Sarzedas
Sarzedas de Vasco
de Vasco 0.96 0.96 0.79
0.79 1.251.25 2.36
2.36 0.31
0.31
Entire
Entire sample
sample 0.53 0.53 0.40
0.40 0.470.47 3.67
3.67 0.16
0.16

Despite the medians of WUIX for each type of effect leading to the conclusion that
Despite
they are the medians
correlated, of WUIX for each
the Kruskal–Wallis test type
doesof effect
not leading
entirely to thethe
confirm conclusion that
statistical they
signif-
are correlated, the Kruskal–Wallis test does not entirely confirm the statistical
icance of the results. Higher medians for buildings affected by the fire may indicate that significance
of the
the results.
results Higherwith
obtained medians for buildings
the WUIX affecteddo
methodology bypoint
the fire
outmay indicate
regions withthat the results
higher WUI
obtained with the WUIX methodology do point out regions with higher WUI exposure
exposure and, therefore, are not randomly assessed. However, the Kruskal–Wallis test did
not show statistical significance in any of the groups.
The effect identified on the medians of the groups, presented in Table 3, is also ob-
served when comparing affected and non-affected categories. Figure 9 displays the ten-
dency for fire effect with the increase in the WUIX value.
Fire 2024, 7, 90 12 of 15

and, therefore, are not randomly assessed. However, the Kruskal–Wallis test did not show
statistical significance in any of the groups.
Fire 2024, 7, x FOR PEER REVIEW The effect identified on the medians of the groups, presented in Table 3, is also observed
13 of 16
when comparing affected and non-affected categories. Figure 9 displays the tendency for
fire effect with the increase in the WUIX value.

Figure 9. Comparison of different ranges of WUIX by building fire effect.

Figure 9. Comparison of different ranges of WUIX by building fire effect.

Although
Although the the expected
expected result
result of
of this
this plot
plot is
is an
anincrease
increase in in the
theline
lineassociated
associatedwith
withthe
the
affected
affected buildings
buildingsand and the
the opposite
opposite behaviour
behaviour in in the
the non-affected
non-affected ones, ones, the
the undefinition
undefinition
of
of the
the results
results inin Sarzedas
Sarzedas de de São
São Pedro
Pedro ee de de Vasco
Vasco for for the
the lower
lower values
values of of WUIX
WUIX caused
caused
this
this disturbance in the initial ranges. The last range of the plot was excluded because itit
disturbance in the initial ranges. The last range of the plot was excluded because
was
was affected
affected byby two
two outliers
outliers inin Sarzedas
Sarzedas de de Vasco
Vasco and,and, therefore,
therefore, does
does not
not represent
represent the
the
tendency
tendency of of the
the whole
whole data.
data.
Aiming
Aiming to to check
check the
the differences
differences between
between the the three
three types
types of of fire
fire effect
effect individually
individually
and
and toto perform
perform aa more
more robust
robust comparison,
comparison, aa post post hoc
hoc Dunn’s
Dunn’s testtest was
was performed,
performed, which
which
gave
gaveaaperspective
perspectiveofofwhich
whichgroupgroup is is
pointedly
pointedly distinct
distinct from thethe
from others withwith
others the same 5%
the same
significance
5% significanceas theasKruskal–Wallis
the Kruskal–Wallistest. Table
test. 4Table
presents the test the
4 presents results.
test From theFrom
results. analysis
the
analysis
of of theitresults,
the results, it is possible
is possible to observeto observe
that, forthat, for the
the case whencasethe
when the sample
entire entire sample was
was com-
combined,
bined, the hypothesis
the null null hypothesis was only
was only rejected
rejected for theforcomparison
the comparison between between “no _affec”
“no _affec” and
and “surro_affec”.
“surro_affec”.

Table 4.
Table Results from
4. Results from post
post hoc
hoc Dunn’s
Dunn’s test
test combined
combined with
with Cohen’s d for effect
effect size.

Figueira
Figueira S. Pedro
S. São São Pedro S. S. Vasco
Vasco Entire Sample
Entire Sample
Group
Group 1 1 Group
Group 2 2
Cohen’s dCohen’s d
p-adj p-adj Cohen’s
Cohen’s d d
p-adj p-adj Cohen’s
Cohen’s d d p-adj
p-adj Cohen’s d d p-adj
Cohen’s p-adj
build_affec no_affec 0.47 0.06 0.11 0.71 0.12 0.19 0.03 0.47
build_affec no_affec 0.47 0.06 0.11 0.71 0.12 0.19 0.03 0.47
build_affec surro_affec
build_affec surro_affec 0.22 0.220.18 0.18 0.20 0.20 0.64 0.64 0.35 0.35 0.180.18 0.24
0.24 0.36
0.36
no_affec surro_affec
no_affec surro_affec
−0.27 −0.270.37 0.37 0.11 0.11 0.29 0.29 0.17 0.17 0.860.86 0.21
0.21 0.05
0.05

From this analysis, it can be inferred that “build_affec” and “no_affec” tend to be
From this analysis, it can be inferred that “build_affec” and “no_affec” tend to be
different in Figueira, admitting a slightly higher significance than the defined 5% level.
different in Figueira, admitting a slightly higher significance than the defined 5% level. The
The Cohen’s d coefficient upholds the analysis by revealing a moderate effect on the dif-
Cohen’s d coefficient upholds the analysis by revealing a moderate effect on the differenti-
ferentiation. This result supports the claim that the higher fire exposure detected by the
ation. This result supports the claim that the higher fire exposure detected by the WUIX
WUIX correlates with the records of buildings affected during the Pedrógão Grande wild-
correlates with the records of buildings affected during the Pedrógão Grande wildfire.
fire. At the same time, for the global analysis, the difference between these two groups
Atobserved;
is not the same this
time,fact
formight
the global analysis,
be related theeffect
to the difference
of thebetween these two
other villages groups
on this is
assess-
not observed; this fact might be related to the effect of the other villages on this assess-
ment, showing the non-possibility of merging the results for the Kruskal–Wallis test and
subsequent post hoc analysis.
Nevertheless, in the comparison between “build_affec” and “surro_affec”, a greater
difference is identified than that observed for “build_affec” and “no_affec” for the “Entire
sample” group, which is not consistent since the lowest match between the first
Fire 2024, 7, 90 13 of 15

ment, showing the non-possibility of merging the results for the Kruskal–Wallis test and
subsequent post hoc analysis.
Nevertheless, in the comparison between “build_affec” and “surro_affec”, a greater
difference is identified than that observed for “build_affec” and “no_affec” for the “Entire
sample” group, which is not consistent since the lowest match between the first comparison
group was expected. This fact may be attributed to the higher uncertainty of these data,
and lower WUIX value range, causing a bigger dispersion of values in this analysis. On
the other hand, “no_affec” and “surro_affec” show concordance with the results when the
entire sample is assessed.
In summary, the trend shown in Figure 9 reveals that up to a normalised WUIX value
of 1, the categories present the same behaviour. Consequently, it is challenging to draw
some conclusion on whether the methodology determines what happened to the actual
scenario or not. Looking into the Kruskal–Wallis analysis in Table 3 and the distribution
of the values in Figure 8, it is clear that the results obtained for Figueira align better with
what would be expected. In contrast, the results obtained for the other two villages present
a higher dispersion, preventing one from drawing any solid conclusions.

4. Conclusions
The present study aimed to test the accuracy of the Wildland–Urban Interface (WUI)
methodology in the Portuguese rural context. The methodology was applied in three
villages located in the centre of the country, in a region that was severely affected by
the 2017 Pedrógão Grande wildfire. The depopulation and abandonment of these rural
settlements may have magnified their vulnerability, and due to their geographical isolation,
the assessment of these areas is challenging and potentially onerous. Therefore, remote
sensing technologies can open possibilities that, until very recently, were unavailable.
With that in mind, the investigation was started with the study of the size of the
analysis grid, through which it was possible to identify the ideal size for the analysis grid,
i.e., the size for which the correlation between WUIX values and the actual fire effect on the
buildings stabilised at the highest level. This result suggested that the most satisfactory
size of the grid is about 40 m, and the subsequent analysis was applied to this size.
From the results of the WUIX, the Kruskal–Wallis test and post hoc Dunn’s test were
applied to assess the significance of the differences between the WUIX values and the
real scenario of fire effect. While the results suggest a relationship between higher WUIX
values and fire effect, particularly in Figueira, the complexity of adapting this methodology
directly to Pedrógão Grande’s wildfire context is apparent. This is primarily because factors
related to fire combat and environmental variables that influence the development of the
fire are not included in the WUIX methodology.
One of shortcomings observed in the analysis reported herein is directly related to the
‘surro-affec’ category. When this category was grouped with the affected buildings, the
result was a significant correlation between WUIX results and the actual scenario. However,
when this category was disaggregated for subsequent tests, a greater dispersion of the
results was observed. This suggests that a clarification of this classification is required,
potentially through on-site inspections. The two villages with fewer buildings in this
category presented better results in the analysis, especially in the case of Figueira, where
the higher number of buildings seemed to attenuate the issue.
In general, this study contributes to the field of wildfire risk assessment with new
information, serving as an ally to more precise fuel management while emphasising the
importance of tailoring methodologies to address local specificities. Furthermore, although
Spearman’s rank correlation test suggests a high correlation between the real case and the
WUIX values, the tests for the differentiation between the groups did not show complete
concordance with what is expected from the method, suggesting that further analysis must
be conducted. By exploring further, the vegetation assessment could yield solid results.
As an additional factor, the impact of fire combat on the wildfire event might have added
Fire 2024, 7, 90 14 of 15

a layer of uncertainty to the results for lower WUIX levels and could serve as a future
enhancement of the methodology.

Author Contributions: Conceptualisation, D.S., T.M.F. and H.R.; methodology, D.S., T.M.F. and
H.R.; software, D.S.; validation, D.S., T.M.F. and H.R.; formal analysis, D.S.; investigation, D.S.,
T.M.F. and H.R.; writing—original draft preparation, D.S.; writing—review and editing, T.M.F. and
H.R.; supervision, T.M.F. and H.R. All authors have read and agreed to the published version of
the manuscript.
Funding: This work was financially supported by Project MTS/BRB/0086/2020—“Sustainability-
led approaches for the rehabilitation and revitalisation of the cultural built heritage of Montesinho
Natural Park” funded by national funds (PIDDAC) through FCT/MCTES. The first and third authors
acknowledge the support by the Foundation for Science and Technology (FCT)–Aveiro Research
Centre for Risks and Sustainability in Construction (RISCO), Universidade de Aveiro, Portugal
(FCT/UIDB/ECI/04450/2020).
Informed Consent Statement: Not applicable.
Data Availability Statement: The Python scripts developed in the scope of this work are available
at: https://github.com/dener-silv2/WUIX_implementation.git; https://github.com/dener-silv2
/WUIX_validation.git, accessed on 15 January 2024.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Ostertag, S.; Rice, M.; Qu, J.J. Investigating Spatiotemporal Trends of Large Wildfires in California (1950–2020). Adv. Cartogr.
GIScience ICA 2023, 4, 1–9. [CrossRef]
2. Viegas, D.X.; Ribeiro, C.; Almeida, M.; Pinto, P.; Ribeiro, L.M.; Silva, Á. Field and Laboratory Analysis of the Junction Fire Process
in the Catastrophic Fire of Pedrógão Grande in June 2017. Int. J. Wildland Fire 2023, 32, 951–967. [CrossRef]
3. Couto, F.T.; Iakunin, M.; Salgado, R.; Pinto, P.; Viegas, T.; Pinty, J.P. Lightning Modelling for the Research of Forest Fire Ignition in
Portugal. Atmos. Res. 2020, 242, 104993. [CrossRef]
4. Andrade, C.; Bugalho, L. Multi-Indices Diagnosis of the Conditions That Led to the Two 2017 Major Wildfires in Portugal. Fire
2023, 6, 56. [CrossRef]
5. Assembleia da República. Comissão Técnica Independente Análise e Apuramento dos Factos Relativos aos Incêndios Que Ocorreram em
Pedrogão Grande, Castanheira de Pera, Ansião, Alvaiázere, Figueiró dos Vinhos, Arganil, Góis, Penela, Pampilhosa da Serra, Oleiros e Sertã,
Entre 17 e 24 de Junho de 2017; Assembleia da República: Lisbon, Portugal, 2017.
6. Comissão de Coordenação e Desenvolvimento Regional do Centro—CCDRC. Relatório de Incêndios na Região Centro; Comissão de
Coordenação e Desenvolvimento Regional do Centro: Coimbra, Portugal, 2017.
7. Whittaker, J.; Handmer, J.; Mercer, D. Vulnerability to Bushfires in Rural Australia: A Case Study from East Gippsland, Victoria.
J. Rural. Stud. 2012, 28, 161–173. [CrossRef]
8. Pausas, J.G.; Fernández-Muñoz, S. Fire Regime Changes in the Western Mediterranean Basin: From Fuel-Limited to Drought-
Driven Fire Regime. Clim. Chang. 2012, 110, 215–226. [CrossRef]
9. Shakesby, R.A. Post-Wildfire Soil Erosion in the Mediterranean: Review and Future Research Directions. Earth Sci. Rev. 2011,
105, 71–100. [CrossRef]
10. Garner, J.M.; Iwasko, W.C.; Jewel, T.D.; Charboneau, B.R.; Dodd, A.A.; Zontos, K.M. A Multihazard Assessment of Age-Related
Weather Vulnerabilities. Weather. Clim. Soc. 2020, 12, 367–386. [CrossRef]
11. Ribeiro, L.M.; Rodrigues, A.; Lucas, D.; Viegas, D.X. The Impact on Structures of the Pedrógão Grande Fire Complex in June 2017
(Portugal). Fire 2020, 3, 57. [CrossRef]
12. Roos, C.I.; Swetnam, T.W.; Ferguson, T.J.; Liebmann, M.J.; Loehman, R.A.; Welch, J.R.; Margolis, E.Q.; Guiterman, C.H.;
Hockaday, W.C.; Aiuvalasit, M.J.; et al. Native American Fire Management at an Ancient Wildland–Urban Interface in the
Southwest United States. Proc. Natl. Acad. Sci. USA 2021, 118, 2018733118. [CrossRef]
13. Kumar, M.; Li, S.; Nguyen, P.; Banerjee, T. Examining the Existing Definitions of Wildland-Urban Interface for California. Ecosphere
2022, 13, e4306. [CrossRef]
14. Tikotzki, I.; Bar-Massada, A.; Levin, N. A Geographically Flexible Approach for Mapping the Wildland-Urban Interface Integrating
Fire Activity Data. Front. Env. Sci. 2023, 11, 1231490. [CrossRef]
15. Gonzalez, S.; Ghermandi, L. How to Define the Wildland-Urban Interface? Methods and Limitations: Towards a Unified Protocol.
Front. Env. Sci. 2023, 11, 1284631. [CrossRef]
16. Schug, F.; Bar-Massada, A.; Carlson, A.R.; Cox, H.; Hawbaker, T.J.; Helmers, D.; Hostert, P.; Kaim, D.; Kasraee, N.K.;
Martinuzzi, S.; et al. The Global Wildland–Urban Interface. Nature 2023, 621, 94–99. [CrossRef]
17. Godoy, M.M.; Martinuzzi, S.; Masera, P.; Defossé, G.E. Forty Years of Wildland Urban Interface Growth and Its Relation With
Wildfires in Central-Western Chubut, Argentina. Front. For. Glob. Chang. 2022, 5, 850543. [CrossRef]
Fire 2024, 7, 90 15 of 15

18. Bar-Massada, A.; Alcasena, F.; Schug, F.; Radeloff, V.C. The Wildland—Urban Interface in Europe: Spatial Patterns and Associa-
tions with Socioeconomic and Demographic Variables. Landsc. Urban. Plan. 2023, 235, 104759. [CrossRef]
19. Vaiciulyte, S.; Hulse, L.M.; Galea, E.R.; Veeraswamy, A. Exploring ‘Wait and See’ Responses in French and Australian WUI
Wildfire Emergencies. Saf. Sci. 2022, 155, 105866. [CrossRef]
20. Rodrigues, A.; Santiago, A.; Laím, L.; Viegas, D.X.; Zêzere, J.L. Rural Fires—Causes of Human Losses in the 2017 Fires in Portugal.
Appl. Sci. 2022, 12, 12561. [CrossRef]
21. Huang, Y.; Jin, Y. Aerial Imagery-Based Building Footprint Detection with an Integrated Deep Learning Framework: Applications
for Fine Scale Wildland–Urban Interface Mapping. Remote Sens. 2022, 14, 3622. [CrossRef]
22. Fernández-álvarez, M.; Armesto, J.; Picos, J. LiDAR-Based Wildfire Prevention in WUI: The Automatic Detection, Measurement
and Evaluation of Forest Fuels. Forests 2019, 10, 148. [CrossRef]
23. Tampekis, S.; Sakellariou, S.; Palaiologou, P.; Arabatzis, G.; Kantartzis, A.; Malesios, C.; Stergiadou, A.; Fafalis, D.; Tsiaras, E.
Building Wildland–Urban Interface Zone Resilience through Performance-Based Wildfire Engineering. A Holistic Theoretical
Framework. EuroMediterr J. Env. Integr. 2023, 8, 675–689. [CrossRef]
24. Miranda, A.; Carrasco, J.; González, M.; Pais, C.; Lara, A.; Altamirano, A.; Weintraub, A.; Syphard, A.D. Evidence-Based Mapping
of the Wildland-Urban Interface to Better Identify Human Communities Threatened by Wildfires. Environ. Res. Lett. 2020,
15, 094069. [CrossRef]
25. Chen, D.Y.; Peng, L.; Li, W.C.; Wang, Y. Da Building Extraction and Number Statistics in WUI Areas Based on UNet Structure and
Ensemble Learning. Remote Sens. 2021, 13, 1172. [CrossRef]
26. Pereira, J.M.C.; Alexandre, P.M.; Campagnolo, M.L.; Bar-Massada, A.; Radeloff, V.C.; Silva, P.C. Defining and Mapping the
Wildland-Urban Interface in Portugal. In Advances in Forest Fire Research 2018; Imprensa da Universidade de Coimbra: Coimbra,
Portugal, 2018; pp. 742–749.
27. Caballero, D. Utilización, Interpretación y Limitaciones del Índice de Interfaz WUIX; Cadernos Técnico: Madrid, Spain, 2019.
28. Efthimiou, N.; Psomiadis, E.; Panagos, P. Fire Severity and Soil Erosion Susceptibility Mapping Using Multi-Temporal Earth
Observation Data: The Case of Mati Fatal Wildfire in Eastern Attica, Greece. Catena 2020, 187, 104320. [CrossRef]
29. Sismanis, M.; Chadoulis, R.T.; Manakos, I.; Drosou, A. An Unsupervised Burned Area Mapping Approach Using Sentinel-2
Images. Land 2023, 12, 379. [CrossRef]
30. Pinto, M.M.; Trigo, R.M.; Trigo, I.F.; Dacamara, C.C. A Practical Method for High-Resolution Burned Area Monitoring Using
Sentinel-2 and Viirs. Remote Sens. 2021, 13, 1608. [CrossRef]
31. Fernández-García, V.; Beltrán-Marcos, D.; Fernández-Guisuraga, J.M.; Marcos, E.; Calvo, L. Predicting Potential Wildfire Severity
across Southern Europe with Global Data Sources. Sci. Total Environ. 2022, 829, 154729. [CrossRef]
32. Maxar Technologies MAXAR—Open Data Program. Available online: https://opendata.digitalglobe.com/events/portugal-
wildfires/post-event/2017-06-21/103001006A9A6B00/103001006A9A6B00.tif (accessed on 19 September 2023).
33. Directorate General for Territory (DGT) Orthophotos 50 cm—Portugal Mainland—2015. Available online: https://
eurogeographics.org/member/directorate-general-for-territory/ (accessed on 15 January 2024).
34. Silva, D.S. Da Avaliação da Exposição aos Incêndios Rurais na Interface Urbano-Florestal de Aldeias do Parque Natural de
Montesinho. Master’s Thesis, University of Aveiro, Aveiro, Portugal, 2023.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.