Forest Ecology and Management 446 (2019) 257–271

Contents lists available at ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier.com/locate/foreco

Modeling Pinus pinaster forest structure after a large wildfire using remote T
sensing data at high spatial resolution
José Manuel Fernández-Guisuraga , Susana Suárez-Seoane, Leonor Calvo
⁎

Area of Ecology. Department of Biodiversity and Environmental Management, Faculty of Biological and Environmental Sciences, University of León, 24071 León, Spain

ARTICLE INFO ABSTRACT

Keywords: In the Mediterranean Basin, wildland fires are major drivers of forest ecosystem dynamics. In the current context
Image texture of global change, these fires are becoming more severe and recurrent because of climatic conditions, land use
Pine population changes and invasive species. In areas affected by mega-fires (burned area > 10,000 ha), the patterns of re-
Post-fire decision-making generation may be heterogeneous due to local variations in fire regime, community composition and environ-
Spectral index
mental features. The goal of this study was to analyze the post-fire structure of both Pinus pinaster Aiton.
Understory community
WorldView-2
seedlings population and understory community in a Mediterranean fire-prone ecosystem at short-term by
means of high spatial resolution satellite imagery within the perimeter of a full stand replacing mega-fire that
burned around 12,000 ha of a Pinus pinaster forest in NW Spain. We established 234 field plots of 2 × 2 meters to
cover four recurrence-severity scenarios. In each plot, we sampled 15 vegetation structural variables at both pine
seedlings population and understory community levels. From the WorldView-2 satellite imagery, we obtained
three sets of spectral variables (reflectance, spectral indices and image textures) that were used as predictors of
vegetation recovery in generalized linear models. At population level, the number and cover of pine seedlings
were successfully modeled with spectral indices and textural information (normalized root mean square error of
16% and 17%, respectively). At understory community level, woody species cover was correlated with first order
textures (normalized root mean square error of 9%). Other understory structure variables (height and richness of
woody species, percentage of bare soil, necromass and leaves) were predicted with an error lower than 20%. The
predictive capacity of the models was similar for all recurrence-severity scenarios. Our results highlight the
usefulness of spectral indices and textural data at high spatial resolution in the analysis of post-fire recovery in
large and heterogeneous burnt areas. Given the accuracy and predictive capacity of the models obtained in this
study, high spatial resolution satellite imagery together with field data provide useful information in post-fire
decision making in fire prone ecosystems.

1. Introduction Lovreglio, 2004), resin (Soliño et al., 2018) or mushrooms (Taye et al.,
2016), but also regulating services such as soil protection, biodiversity
In the Mediterranean Basin, forest fires are major drivers of eco- support and cultural ecosystem services such as recreational benefits
system productivity, composition and dynamics (Lozano et al., 2008; (Leone and Lovreglio, 2004). On the other hand, the understory com-
Sagra et al., 2018). In this region, forest fires are increasingly larger and munity also plays a key role in the provision of regulating ecosystem
are becoming more severe and recurrent, mainly due to climatic factors services in fire-prone pine ecosystems (Gonzalez et al., 2013) such as
and land use changes (Chuvieco et al., 2010; Moreira et al., 2011; protection against soil erosion processes (Shakesby, 2011; Vieira et al.,
Álvarez et al., 2012; Lecina-Diaz et al., 2014; Quintano et al., 2015), but 2018). Therefore, rapid and effective post-fire regeneration evaluation
also to alien species introduction (Pausas and Keeley, 2014). Further- is considered key for both the dominant tree species population and the
more, large forest fires imply a loss of the ecosystem services provided understory community of these ecosystems (Fernández-Guisuraga et al.,
by forests (Calvo et al., 2015), which finally may affect human health 2019) in order to recover their services supply (Leverkus et al., 2018;
(Van Drooge et al., 2016). Schmeer et al., 2018). The evaluation of other post-fire variables such
In Mediterranean fire-prone ecosystems, pine populations not only as the dead plant material cover (i.e., coarse and fine woody debris), is
supply provisioning ecosystems services like timber (Leone and also of crucial importance for land management decision-making

⁎
Corresponding author.
E-mail address: jofeg@unileon.es (J.M. Fernández-Guisuraga).

https://doi.org/10.1016/j.foreco.2019.05.028
Received 4 March 2019; Received in revised form 10 May 2019; Accepted 11 May 2019
Available online 22 May 2019
0378-1127/ © 2019 Elsevier B.V. All rights reserved.
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

(Pesonen et al., 2008; Joyce et al., 2019). Dead plant material has a key population and understory community structure in a Mediterranean
role in the probability of occurrence of a new fire in a previously burned fire-prone ecosystem after the occurrence of a stand replacing mega-
area (Sullivan et al., 2018) but, on the other hand, dead plant material fire, by means of high spatial resolution satellite imagery and an ex-
increase rain interception following fire and behave as a sediment trap tensive field sampling campaign. Specifically, we try to: (1) Identify the
(Slesak et al., 2015). vegetation structure variables that best correlate with spectral in-
Nevertheless, fire effects and vegetation recovery are not homo- formation under different scenarios of fire recurrence and severity, at
geneous across the landscape, due to spatial variations in fire regime the level of both the dominant tree species population and the un-
parameters, species composition and environmental factors (Beaty and derstory community. (2) Determine the reflectance bands, spectral in-
Taylor, 2001). Fire severity and recurrence are the fire regime para- dices and textural features derived from WorldView-2 multispectral
meters that most affect post-fire regeneration (González-De Vega et al., satellite imagery that best explain post-fire regeneration, which would
2016; Tessler et al., 2016; Taboada et al., 2017; Fernández-García et al., therefore constitute a useful management tool in large and hetero-
2018a). Thus, their explicit consideration in the assessment of spatial geneous burned areas. We expect that within the perimeter of a stand
variations in post-fire regeneration is of vital importance for planning replacing mega-fire where salvage logging was conducted following
strategies of ecosystem restoration in areas affected by large fires (Ruíz- fire, the recovery models of the Pinus pinaster population will perform
Gallardo et al., 2004; Moreira et al., 2009; Solans-Vila and Barbosa, better in scenarios of low fire recurrence and severity. The pine seedling
2010). recruitment on these scenarios would be significant due to the massive
Because of the increase in the number of large forest fires, post-fire seed storage in the canopy seed bank during the previous fire-free
regeneration patterns in fire-prone pine ecosystems are increasingly period (Tapias et al., 2004; Calvo et al., 2008; Calvo et al., 2016).
heterogeneous (Fernández-García et al., 2018a), so that using ex- Moreover, obligate seeder shrubs with a slow recovery rate are pro-
clusively field-work monitoring is unfeasible (Chuvieco and Kasischke, moted under low fire recurrence scenarios and enhance seedling re-
2007). In this sense, remote sensing techniques (RST) are recognized as cruitment through facilitative interactions (Pausas and Vallejo, 1999;
indispensable tools to evaluate vegetation regeneration patterns after Taboada et al., 2017). In contrast, recurrent fires with short return in-
fire, particularly across large areas (Lozano et al., 2012; Veraverbeke terval will favor resprouter shrubs with a fast post-fire recovery rate
et al., 2012; Fernández-Manso et al., 2016). However, the application of (Calvo et al., 2012; Taboada et al., 2018), turning facilitative interac-
RST is still a challenge in the case of heterogeneous mega-fires (burned tions into competitive ones (Taboada et al., 2017). Therefore, under fire
area > 10,000 ha; Stephens et al., 2014) with high variation in fire regime scenarios with a high pine recruitment, the seedlings reflectance
regime parameters (Schoennagel et al., 2008; Chu et al., 2016). contribution would presumably be large enough so that their spectral
Among the products derived from satellite imagery, spectral indices response can be detected with the presence of surrounding vegetation.
are the most frequently employed in post-fire vegetation regeneration Furthermore, we expect the advantage of texture products, over re-
analysis (Marchetti et al., 1995; Diaz-Delgado et al., 2003; Clemente flectance values or spectral indices, in post-fire regeneration modeling
et al., 2009; Cuevas-Gonzalez et al., 2009; Lozano et al., 2010; Solans- under high spatial and fire regime heterogeneity (Gu et al., 2013)
Vila and Barbosa, 2010; Lozano et al., 2012; Veraverbeke et al., 2012; within the fire perimeter because the textural features allow for quan-
Chu et al., 2016). Other studies, such as that conducted by Pleniou and tifying spatial relationships in reflectance values (Pu and Cheng, 2015).
Koutsias (2013), have focused on the evaluation of reflectance bands as The flowchart of this study is schematized in Fig. 1.
predictors of post-fire regeneration, instead of grouping the reflectance
values into spectral indices. Additionally, some authors propose the use 2. Materials and methods
of satellite imagery texture analysis as an indicator of vegetation
structural parameters allowing for quantifying spatial variability within 2.1. Study area. Definition of different scenarios of recurrence and severity
a defined area (Kayitakire et al., 2006; Song and Dickinson, 2008;
Eckert, 2012; Viedma et al., 2012; Wood et al., 2012; Gu et al., 2013; The study area (Fig. 2) is located in Sierra del Teleno (NW Spain),
Bastin et al., 2014; Pu and Cheng, 2015). Texture analysis is particu- within the perimeter of a full stand replacing mega-fire (11,602 ha) that
larly relevant with respect to other products derived from satellite occurred in August 2012 and affected a forest dominated by Pinus pi-
imagery in areas of high heterogeneity in ground cover, since visible naster. The burned area sites at an average altitude of 1063 m.a.s.l. and
structures are not only related to discrete values of image pixels, but is relief is typically Appalachian, with quartzite banks outcropping in
also to the spatial variation of these values between adjacent pixels (Gu the peaks forming prominent crests. The north of the area is made of
et al., 2013). wide valleys with moderate slopes, while the southernmost zone is
Despite the increasing use of satellite products in post-fire recovery mainly composed of sedimentary plains and terraces. This is an
analysis, most of those studies are constrained by coarse resolution Atlantic-Mediterranean transition area, with a mean annual precipita-
sensors to characterize post-fire environments with a high spatial het- tion of 650 mm, a mean annual temperature of 10 °C and less than two
erogeneity (Meng et al., 2018). In this regard, the development of high months of summer drought. The soils are predominantly acidic (pH
spatial resolution satellite imagery, such as QuickBird or WorldView, 4.5–5.5), with sandy texture, being clay soils relegated to the qua-
provides an improved performance in the monitoring of vegetation ternary deposits (Santamaría, 2015). Forests dominated by Pinus pina-
structure in heterogeneous post-fire landscapes (Viedma et al., 2012; ster typically covered the quartzite slopes and tertiary soils of the sur-
Jung et al., 2013; Chu et al., 2016; McKenna et al., 2018; Meng et al., rounding plains and are particularly subject to the occurrence of large
2018; Fernández-Guisuraga et al., 2019). However, to our knowledge, forest fires (Calvo et al., 2008). Therefore, the population of Pinus pi-
there are no studies evaluating the post-fire forest structure both at the naster in this region shows adaptive characteristics to the prevailing fire
dominant tree species population and the shrub community levels by regime, such as higher production of serotinous cones, thicker bark and
means of high spatial resolution multispectral imagery alone (i.e. earlier flowering age than other Spanish populations (Tapias et al.,
without the combined use of optical satellite imagery and active sen- 2004; Alvarez et al., 2007). Quercus pyrenaica Willd. stands, more re-
sors), under the influence of the fire regime parameters and a high sistant to forest fires, are present in valley bottoms with deeper soils
spatial fine-scale heterogeneity. This approach could have broad sci- (Santamaría, 2015). The pine canopy was totally consumed by fire and
entific implications for land management decision-making in Medi- the burned stands were salvaged logged in the two years immediately
terranean fire-prone pine ecosystems since it is essential to separate the after the fire (Taboada et al., 2017; Taboada et al., 2018). Vegetation
dominant tree species recovery and the understory recovery (García- following fire mainly consists of Pinus pinaster regeneration stands in a
Morote et al., 2017; Meng et al., 2018). seedling growth stage, being the understory plant community domi-
The main goal of this study was to analyze the Pinus pinaster Aiton. nated by species such as Halimium lasianthum subsp. alyssoides (Lam.)

258
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Large wildfire (11602ha)

RECURRENCE- SEVERITY WorldView-2 satellite
15 vegetation variables sampled in 234 field plots Spectral and textural predictors

Regeneration spatial modeling

Understory community Pinus pinaster population

(9 structure variables modeled) (6 structure variables modeled)

Fig. 1. Conceptual framework of the present study.

Before the fire of 2012, the area had suffered another large fire in
1998 of around 3000 ha. Therefore, within the perimeter of the 2012
fire, we defined two fire recurrence scenarios: low recurrence (one fire
in the last 15 years) and high recurrence (two fires in the last 15 years).
Three months after the wildfire, we estimated the burn severity through
the CBI (Composite Burn Index; Key and Benson, 2006) over 54 field
plots of 30 × 30 m randomly distributed using the protocol described in
Fernández-García et al. (2018b). Additionally, the dNBR (differenced
Normalized Burn Ratio) index (Key, 2006) was derived from Landsat
pre-fire and post-fire images (September 20th, 2011 - September 6th,
2012). We established two severity scenarios within the perimeter of
the fire of 2012 based on a dNBR threshold computed from the CBI
(Miller and Thode, 2007; Quintano et al., 2015): low severity
(dNBR ≤ 573) and high severity (dNBR > 573). Overlaying the re-
currence and severity classes, we identified four recurrence-severity
scenarios (low recurrence-low severity, low recurrence-high severity,
high recurrence-low severity and high recurrence-high severity) (Fig. 2)
to subsequently model post-fire regeneration with regard to the fire
regime. Within the fire perimeter, we selected a 3000 ha study frame-
work where the four scenarios of recurrence and severity were re-
presented (Fig. 2) to collect field data for further model calibration and
validation.

2.2. Field data

In summer of 2015, a set of 234 plots of 2 × 2 m (resolution of

WorldView-2) were randomly established in the field. The number of
plots for each recurrence-severity scenario was proportional to the re-
lative area occupied by each scenario within the fire perimeter
(Table 1). The center of each plot was georeferenced by a sub-meter
Fig. 2. Study framework within the perimeter of the Sierra del Teleno mega-fire
accuracy GPS (Spectra Precision MobileMapper 20) in post-processing
which occurred in 2012, and the considered recurrence-severity scenarios.
mode. The guaranteed accuracy after post-processing the coordinates
was greater than 0.4 m. Within each plot, we measured 15 regeneration
structure variables of both the population of the dominant tree species
Greuter, Pterospartum tridentatum (L.) Willk. and Erica australis L. (Calvo
(Pinus pinaster) and the understory community with a great potential in
et al., 2008). Three years after the fire, the mean pine density in the
post-fire management decision-making (Table 2). All cover variables
study area was 2.25 ± 0.15 seedlings/m2, the mean percent cover of
were estimated using the methodology of visual percentage cover
pine seedlings per square meter was 4.30 ± 0.26 and their mean
(Calvo et al., 2008).
height was 22.71 cm ± 0.71 (Fernández-García et al., 2019).
The values of the predictors were extracted from the pixels of

259
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Table 1
Number of experimental plots used for model calibration and validation across recurrence-severity scenarios.
Recurrence-Severity scenarios Low recurrence & low severity Low recurrence & high severity High recurrence & low severity High recurrence & high severity

Model calibration 24 60 24 48
Model validation 12 30 12 24
Total plots 36 90 36 72

Table 2
Field regeneration variables measured in 2 × 2 m plots.
Category Variable Code Unit of measurement

Pinus pinaster population No. of living seedlings Pn number

No. of dead seedlings dPn number
Average height of living seedlings Ph cm
Maximum height of living seedlings Pmaxh cm
Minimum height of living seedlings Pminh cm
Cover of seedlings Pcov %

Understory community Cover of woody species Wcov %

Maximum height of woody species Wmaxh cm
Richness of woody species Wrich number
Cover of herbaceous species Hcov %
Richness of herbaceous species Hrich number
Cover of leaves and thin branches less than 2 cm Lcov %
Cover of lying necromass greater than 2 cm Ncov %
Cover of stones St %
Percentage of bare soil S %

WorldView-2 image matching the 2 × 2 m plots where field measure-

ments were made. 66% of the experimental plots were used to calibrate
the regeneration models, while the remaining 33% were used for va-
lidation (Table 1).

2.3. WorldView-2 image data

The WorldView-2 image was acquired on 16th of June 2015 at

10:34:02 UTC, with a cloud cover of 8.1%. Sensors on-board
WorldView-2 capture data in 11-bit format over a panchromatic (Pan,
450–800 nm) and eight multispectral bands (MS, 400–1040 nm), with a
spatial resolution of 0.5 m and 2.0 m, respectively. This satellite image
was selected due to its high spatial resolution, presumably able to
discriminate between spatial structures given the scale of variability
observed in the field. Furthermore, WorldView-2 is the first satellite to
offer eight multispectral bands on the visible and infrared region of the
spectrum (Ni et al., 2015).

2.3.1. Image processing

The image was orthorectified to subpixel accuracy (root mean
square error = 0.58 m) using rational polynomial coefficients provided Fig. 3. Workflow of WorldView-2 image processing (RPC = rational poly-
with the image, a digital surface model with a resolution of 0.5 m and nomial coefficients; GCP = ground control points).
ground control points taken in the field. The radiometric calibration of
the image to apparent surface reflectance was conducted by the at- 2.3.2. Spectral predictors: Reflectance values, spectral indices and image
mospheric correction algorithm FLAASH, the acronym for Fast Line-of- textures
sight Atmospheric Analysis of Spectral Hypercubes (Vermote et al., 1997; Three types of products were obtained from the WorldView-2 pro-
Adler-Golden et al., 1999; Berk et al., 1999; Matthew et al., 2003). cessed image to be used as predictors of vegetation recovery: (1) surface
Visibility, column water vapor amount and some ancillary data for at- reflectance values, (2) spectral indices and (3) first and second order
mospheric correction of the scene were obtained from the image me- textures.
tadata and from the State Meteorology Agency of Spain (AEMET). The We accounted for the reflectance values of the eight bands of the
workflow of the image processing is shown in Fig. 3. imagery: 1-coastal blue (400–450 nm), 2-blue (450–510 nm), 3-green

260
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Table 3
Spectral indices used in this study derived from WorldView-2 reflectance bands (B1–B8).
Index Formula Reference

Leaf pigments
Anthocyanin Reflectance Index 2 (ARI) B7[(1/B3)-(1/B6)] Gitelson et al. (2001)
Carotenoid Reflectance Index (CRI) (1/B3)-(1/B6) Gitelson et al. (2002a)
Modified Chlorophyll Absorption Ratio Index (MCARI) [(B6-B5)-0.2(B6-B3)](B6/B5) Daughtry et al. (2000)
Plant Senescence Reflectance Index (PSRI) (B5-B3)/B6 Merzlyak et al. (1999)
Structure Intensitive Pigment Index (SIPI) (B7-B1)/(B7-B5) Penuelas et al. (1995)
Transformed Chlorophyll Absorption Reflectance Index (TCARI) 3[(B6-B5)-0.2(B6-B3)(B6/B5)] Haboudane et al. (2004)

Detection of burned areas and soil

Burnt Area Index (BAI) 1/[(0.1 − B5)2+(0.06 − B8)2] Chuvieco et al. (2002)
WorldView Soil Index (WSI) (B3 − B4)/(B3 + B4) Wolf (2010)

Plant health
Atmospherically Resistant Vegetation Index (ARVI) (B7-2B5 + B2)/(B7 + 2B5-B2) Kaufman and Tanre (1992)
Difference Vegetation Index (DVI) B7-B5 Tucker (1979)
Far Red to Red Index (FRRI) B6/B5 Barry et al. (2008)
Green Atmospherically Resistant Index (GARI) [B7-(B3-1.7B2 + 1.7B5)]/ [B7+(B3-1.7B2 + 1.7B5)] Gitelson et al. (1996)
Green Difference Vegetation Index (GDVI) B7-B3 Sripada et al. (2006)
Green Normalized Difference Vegetation Index (GNDVI) (B7-B3)/(B7 + B3) Gitelson and Merzlyak (1998)
Green Ratio Vegetation Index (GRVI) B7/B3 Sripada et al. (2006)
Infrared Percentage Vegetation Index (IPVI) B7/(B7 + B5) Crippen (1990)
Leaf Area Index (LAI) 3.618[(B7-B5)/(B7 + 6B5-7.5B2 + 1)]-0.118 Boegh et al. (2002)
Modified Non-Linear Index (MNLI) [(B72-B5)1.5]/(B72 + B5 + 0.5) Yang et al. (2008)
Modified Simple Ratio (MSR) [(B7/B5)-1]/[sqrt(B7/B5) + 1] Chen (1996)
Modified Triangular Vegetation Index (MTVI) 1.2[1.2(B7-B3)-2.5(B5-B3)] Haboudane et al. (2004)
Normalized Difference Vegetation Index (NDVI) (B7-B5)/(B7 + B5) Rouse et al. (1973)
Optimized Soil Adjusted Vegetation Index (OSAVI) [1.5(B7-B5)]/(B7 + B5 + 0.16) Rondeaux et al. (1996)
Red Green Ratio Index (RGI) B5/B3 Gamon and Surfus (1999)
Renormalized Difference Vegetation Index (RDVI) (B7-B5)/[sqrt(B7 + B5)] Roujean and Breon (1995)
Transformed Difference Vegetation Index (TDVI) sqrt[0.5+(B7-B5)/(B7 + B5)] Bannari et al. (2002)
Visible Atmospherically Resistant Index (VARI) (B3-B5)/(B3 + B5-B2) Gitelson et al. (2002b)
WorldView Improved Vegetative Index (WIVI) (B8-B5)/(B8 + B5) Wolf (2010)

(510–580 nm), 4-yellow (585–625 nm), 5-red (630–690 nm), 6-red regeneration (Table 2) and spectral predictors (Tables 3 and 4) was
edge (705–745), 7-NIR1 (770–895 nm) and 8-NIR2 (860–1040 nm). analyzed using multivariate Generalized Linear Models (GLMs;
A total amount of 27 spectral indices were calculated from the ap- McCullagh and Nelder, 1989). For each biotic variable, we tested five
parent surface reflectance image and clustered into three groups: in- sets of predictors: reflectance values, spectral indices, first order tex-
dices related to leaf pigments, detection indices of burned areas and soil tures, second order textures and ensemble ([reflectance + spectral in-
and plant health indices (Table 3). These three sets of spectral indices dices + textures]). We calibrated separate models for each recurrence-
are expected to account for the structure of the Pinus pinaster population severity scenario and for the entire study framework. Therefore, we ran
and understory community during the first successional stages after the a total number of 375 models (15 biotic variables × 5 sets of spectral
fire. predictors × 5 frameworks). For modelling biotic variables related to
For each of the eight bands of WorldView-2 reflectance image, we vegetation height, we assumed a normal error distribution and an
calculated five first order (1rd) textures (occurrence statistic measure- identity link function (linear model). For count variables, a Poisson
ments: data range, mean, variance, entropy and skewness) and eight distribution of error with a logarithmic link function (Zeileis et al.,
second order (2rd) textures (co-occurrence statistic measurements: 2008) was used; in the case of overdispersion occurrence, a quasi-
mean, variance homogeneity, contrast, dissimilarity, entropy, angular Poisson distribution type with a logarithmic link function (Ver-Hoef
second moment and correlation) (Table 4). First order texture mea- and Boveng, 2007) was carried out. For response variables measured as
surements are statistics calculated from the original pixel values in a a percentage, we used a quasi-Poisson type error distribution with a
moving window. The second order texture measurements are calculated logarithmic link function, according to Martin et al. (2005).
using the Gray Level Co-Occurrence Matrix (GLCM), which describes In order to detect multicollinearity problems within each set of
the probability that the values of each pair of pixels concur in a given predictors, bivariate Pearson correlations and VIF values were eval-
direction and distance (Haralick et al., 1973). In all texture analysis, the uated. Correlation analyses allowed for discriminating strongly corre-
moving window size was 3 × 3 pixels since this window size will cap- lated groups of predictors (rPearson > 0.7) (Engler et al., 2013). Within
ture the heterogeneity of the reflectance values in the variability range each group, we only preserved for subsequent analysis the predictor
of the cover pattern observed in the field (Gu et al., 2013). Chen et al. that explained the greatest proportion of variance for a given response
(2004) also highlighted that, in heterogeneous landscapes, a small variable in univariate generalized linear models (GLMs) (McCullagh
window size should be chosen. Second order textures were calculated and Nelder, 1989). The uncorrelated predictors that were used in the
for the four spatial directions (0°, 45°, 90° and 135°, represented in this calibration of post-fire regeneration models are shown in Table SM.1
case by the Cartesian coordinates [1,0], [1,−1], [0,−1] and (Supplementary Material). Additionally, we checked that the variance
[−1,−1]). The directionally invariant texture measures were subse- inflation factor (VIF) of each predictor included in the models was
quently obtained by calculating the mean of the co-occurrence mea- lower than four (Zuur et al., 2010). Models were calibrated in a back-
surements in the four directions (Zhang and Xie, 2012). wards stepwise regression procedure (Draper and Smith, 1998). The
overall fit of the models was evaluated as the proportion of deviance
explained by the predictors (McFadden, 1974) and the predictive power
2.4. Statistical analysis as the root mean square error (RMSE, Eq. (1)) normalized from the
maximum (ymax) and minimum (ymin) value of observations for each
The relationship between biotic variables accounting for post-fire

261
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Table 4
First and second order textures extracted for each surface reflectance bands. On first order textures, pi is the relative frequency of the i gray levels in the window, X is
the pixel gray level and N is the number of gray levels. On second order textures, p(i,j) represents the value in i,j cell of the co-occurrence matrix and N is the number
of gray levels used in the analysis.
Texture Descriptiona, b
Formulac

First order
Data range Difference of the maximum and minimum pixel value in the moving window. max {X } min {X }
Mean Local mean of pixels in the moving window. N 1
i = 0 ipi = SM
Variance Local variance of pixels in the moving window. N 1
SM ) 2pi = SD2
i = 0 (i
Entropy Shannon diversity index. N 1
i = 0 pi ( lnpi )
Skewness Lack of symmetry around the mean of the moving window pixels. 1 N 1
[ i = 0 (i SM )3pi ]
3
SD

Second order
Mean Local mean of GLCM window. N 1
ip (i, j) = µ
i, j = 0
Variance Local variance of GLCM window. N 1
p (i , j )(i µ)2
i, j = 0
Homogeneity Measurement of homogeneous pixel values through image. The values range between 0 and 1. N 1 p (i, j )
i, j = 0 1 + (i, j)2

Contrast Measurement of the local variation in the image. Opposed to homogeneity. N 1

p (i , j )(i j )2
i, j = 0
Dissimilarity Similar to the contrast, measured as the absolute value of grayscale differences. N 1
p (i , j )|i j|
i, j = 0
Entropy Shannon diversity index. The values range between 0 and the natural logarithm of the GLCM window size. N 1
p (i , j )( lnp (i, j))
i, j = 0
Angular second moment Measure of the image homogeneity. The values range between 0 and 1. N 1
p (i , j )2
i, j = 0
Correlation Measurement of linear dependencies of the image gray levels. The values range between −1 and 1. N 1 (i µi )(j µj )
i, j = 0 p (i , j )
2 2
i j

a
Haralick et al. (1973).**
b
Wood et al. (2012).
c
Pu and Cheng (2015).

response variable (nRMSE, Eq. (2)). For its part, band 6 (red edge) was related with the cover of leaves,
n
lying necromass and stones (Table 7). Among the spectral indices,
1 Anthocyanin Reflectance Index 2 significantly explained the structure
RMSE = (Å·i yi )2
n i=1 (1) variables related to the regeneration of the Pinus pinaster population
and the understory community. The Burnt Area Index and the World-
nRMSE =
RMSE
× 100 View Soil Index were the best predictors of the cover of leaves, lying
ymax ymin (2) necromass and bare soil at community level (Table 7). In all models
analyzed, at least one variable within the set of first and second order
All statistical analyses were performed using R (Core Team, 2015).
textures was selected as a significant predictor of post-fire vegetation
structure. Among first order textures, the mean and data range of dif-
3. Results ferent bands had the highest level of significance. For the second order
textures, the mean also gained special relevance, along with entropy
The highest recovery of the Pinus pinaster population structure and homogeneity (Table 7). None of the set of predictors explained the
variables (density, cover and height of pine seedlings) was observed for number of dead pine seedlings or the minimum height of living pine
the low fire recurrence scenario. Fire severity was inversely related to seedlings.
the population recovery in both fire recurrence scenarios. The opposite For the recurrence-severity scenarios, the overall fit and predictive
pattern was observed for the understory community structure variables, capacity of the regeneration models ran at population level were similar
which exhibited greater recovery under high fire recurrence than for to those found for the entire study framework, except for the high re-
the low recurrence scenario. There was no clear pattern for the influ- currence and severity scenario (Fig. 5). In the latter case, the nRMSE
ence of fire severity in the community structure recovery (Table 5). rose to 52% for the number (Pn), 41% for the cover (Pcov), 62% for
The highest model fit (Table 6) and predictive capacity (Fig. 4) was average height (Ph) and 45% for maximum height (Pmaxh) of living
found, for the whole study framework, in either the models including seedlings. The prediction of community level variables (except necro-
texture predictors or the ensemble model (spectral indices + textures). mass cover -Ncov-) under the different scenarios presented the same
Pinus pinaster structure was modeled both in terms of number (Pn) performance in terms of fit and predictive capacity than for the entire
(deviance = 34% and nRMSE = 16%) and cover (Pcov) of pine seed- study framework. For Ncov, the predictive power of the model was
lings (deviance = 39% and nRMSE = 17%) from the second order significantly lower in the scenarios defined by a high recurrence
textures and the ensemble set (spectral indices + first order textures), (nRMSE higher than 40%).
respectively. At the understory community level, the biotic variable The recovery models for the entire study framework that achieved
most successfully modeled was the cover of woody species (Wcov) the highest predictive capacity (nRMSE around 20% or lower) are
(deviance = 58% − 63% and nRMSE = 9% − 11% for the considered shown in Table 8. These models explained the variation corresponding
set of predictors), with all sets of predictors having a similar perfor- to the dominant tree species regeneration (Pn and Pcov), understory
mance. The overall fit and nRMSE of the other understory community woody species regeneration (Wcov, Wmaxh and Wrich), necromass
variables were lower than 45% and 27%, respectively. cover (Ncov) and percentage of bare soil (S). The spatial outputs of the
None of the reflectance bands explained the recovery of the Pinus best performing models at population and understory community levels
pinaster population. However, bands 1 (coastal blue) and 7 (NIR-1) (Pn and the Wcov) are shown in Figs. 6 and 7 (see Figures SM.1 to SM.5
were significantly correlated with the understory community structure.

262
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Table 5
Mean ± standard deviation of Pinus pinaster population and understory community structure variables under the fire recurrence (R) and Severity (S) scenarios. The
variable codes correspond to that shown in Table 2.
lowR-lowS lowR-highS highR-lowS highR-highS

Pn (number) 5.35 ± 5.18 3.13 ± 4.55 0.73 ± 1.05 0.34 ± 0.90

dPn (number) 0.33 ± 0.66 0.24 ± 0.81 0.04 ± 0.14 0.00 ± 0.00
Ph (cm) 18.80 ± 9.15 23.36 ± 14.10 17.20 ± 9.78 28.62 ± 19.25
Pmaxh (cm) 32.88 ± 18.14 34.45 ± 23.23 13.16 ± 16.56 13.08 ± 22.38
Pminh (cm) 7.69 ± 4.13 10.91 ± 10.26 6.08 ± 7.68 8.77 ± 16.28
Pcov (%) 8.01 ± 7.85 6.72 ± 7.91 1.16 ± 1.68 0.97 ± 2.27
Wcov (%) 44.96 ± 14.79 41.29 ± 14.67 50.69 ± 14.21 50.19 ± 15.87
Wmaxh (cm) 68.55 ± 17.79 73.02 ± 18.26 86.77 ± 33.58 82.33 ± 22.70
Wrich (number) 2.41 ± 1.02 2.45 ± 0.91 3.22 ± 0.59 2.97 ± 0.50
Hcov (%) 3.84 ± 9.32 2.06 ± 2.54 1.06 ± 1.68 0.64 ± 1.12
Hrich (number) 1.31 ± 1.32 1.41 ± 1.31 0.50 ± 0.72 0.28 ± 0.42
Lcov (%) 17.28 ± 12.39 10.96 ± 12.45 5.46 ± 11.77 0.75 ± 1.52
Ncov (%) 10.06 ± 10.57 12.98 ± 10.06 3.57 ± 5.46 3.72 ± 3.78
St (%) 17.65 ± 11.79 13.80 ± 14.28 23.42 ± 14.49 25.86 ± 14.08
S (%) 12.88 ± 8.82 24.35 ± 15.58 20.88 ± 10.66 20.15 ± 11.38

Table 6
Explained deviance by the regeneration models run for the entire study framework (see Table 2 for the codes of the response variables).
Category Response variable Set of predictors

Reflectance Indices 1rd textures 2rd textures Ensemble

Pinus pinaster population Pn ─ 26.66% 27.85% 26.79% 33.63%

dPn ─ ─ ─ ─ ─
Ph ─ 25.97% 29.21% 27.59% ─
Pmaxh ─ 40.82% 13.13% 12.13% 16.49%
Pminh ─ ─ ─ ─ ─
Pcov ─ ─ 27.08% 38.63% ─

Understory community Wcov 62.84% 60.80% 63.02% 61.51% 57.90%

Wmaxh 27.88% 38.15% 33.04% 42.88% 44.73%
Wrich 23.07% 24.39% 27.75% 33.09% 36.41%
Hcov ─ ─ 8.63% 14.80% ─
Hrich ─ ─ ─ ─ ─
Lcov 17.60% 21.83% 22.23% 35.98% 38.83%
Ncov 15.46% 34.95% 23.12% 29.67% 39.24%
St 4.66% 3.46% 10.68% 14.20% ─
S ─ 33.07% ─ ─ 35.48%

Fig. 4. Normalized root mean square error (nRMSE) of the regeneration models calibrated for the entire study framework (see Table 2 for the codes of the response
variables).

263
 J.M. Fernández-Guisuraga, et al.

Table 7
Spectral predictors selected in the regeneration models ran for the entire study framework. The response variable code corresponds to that shown in Table 2. The abbreviations of the spectral indices correspond to those
shown in Table 3. For the first order (1rd) and second order (2rd) textures, CON = contrast, COR = correlation, DIS = dissimilarity, DR = data range, ENT = entropy, HOM = homogeneity, MEAN = mean and
VAR = variance.
Category Response Reflect. Spectral indices First order textures Second order textures Ensemble model
variable

Pinus pinaster population Pn ─ ARI***, PSRI*, B7(DR***) B7(ENT**) ARI**, PSRI*, 1rd[B4(ENT*), B7(DR**)]
WSI*
dPn ─ ─ ─ ─ ─
Ph ─ MTVI*, PRI*, B4(DR*,MEAN**) B8(MEAN**) ─
PSRI*
Pmaxh ─ PSRI* B4(DR**) B7(DR***) B7(HOM***) B8(MEAN*) MTVI*, 2rd[B3(HOM**), B7(HOM***), B8(COR*)]
Pminh ─ ─ ─ ─ ─
Pcov ─ ─ B8(DR**) B3(MEAN***,ENT**) B6(MEAN***) ─

264
Understory community Wcov B1***, B7*** ARI***, WSI* B3(DR*) B5(MEAN***) B1(MEAN***) B2(COR*) B3(ENT*) B7(MEAN***) GDVI***, 1rd[B3(DR*), B5(MEAN***), B8(DR**)]
B8(DR**,MEAN***)
Wmaxh B1*** ARI***, PSRI** B5(MEAN***) B1(ENT*) B4(MEAN***) B6(MEAN**) B7(COR*) ARI***, PSRI***, 1rd[B5(MEAN***)]
Wrich B1***, B7*** ARI***, BAI*** B1(MEAN***) B8(MEAN***) B1(MEAN**) B8(MEAN***,HOM**) ARI**, BAI***, 2rd[B3(HOM***),B8(CON**)]
Hcov ─ ─ B3(DR**) B8(DR**) B3(DIS***) B6(DIS***) ─
Hrich ─ ─ ─ ─ ─
Lcov B6** BAI***, FRRI*, B6(MEAN***,ENT*) B2(ENT***) B7(HOM***) FRRI*, GEMI*, 2rd[B2(ENT***), B7(HOM***)]
WSI*
Ncov B6** BAI*** B4(MEAN**,ENT*) B7(MEAN***) B2(ENT*) B4(COR*) B5(ENT*) B6(DIS***,COR*) BAI***, 2rd[B2(ENT*), B4(COR*), B6(DIS**,COR*), B7(ENT*),
B7(MEAN***) B8(COR*)]
St B6* FRRI*, GEMI* B6(DR*,MEAN***) B6(MEAN***,ENT*,COR*)
S ─ WSI** ─ ─ WSI**, 1rd[B5(DR*)]

* Significant variable at p < 0.05.

** Significant variable at p < 0.01.
*** Significant variable at p < 0.001.
Forest Ecology and Management 446 (2019) 257–271
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Fig. 5. Normalized root mean square error (nRMSE) of the regeneration models ran for the scenarios of recurrence and severity (see Table 2 for response variable
codes).

of the Supplementary Material for the maps of remaining models). From variables, which are valuable indicators of the structure and func-
these model spatial outputs, priority areas to apply post-fire manage- tioning of both Pinus pinaster population (i.e., number and cover of
ment strategies could be identified. living seedlings) and the understory community (i.e., cover, maximum
height and richness of woody species, necromass cover and percentage
4. Discussion of bare soil) have been successfully modeled under different fire re-
gimes (recurrence and severity) on the basis of texture predictors,
Spectral products derived from satellite imagery at high spatial re- which has clear implications for land management in areas affected by
solution, such as WorldView-2, represent a useful tool for a rapid as- mega-fires. The assessment of post-fire recovery at different ecosystem
sessment of the vegetation regeneration after the occurrence of large levels is a key approach to avoid harmful environmental processes such
forest fires in Mediterranean fire prone ecosystems. Key biotic as post-fire erosion, the loss of ecosystem services supplies or the

265
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Best performing post-fire recovery models for the entire study framework. The + or − sign indicates a direct or inverse relationship between the predictor and the response variable. The response variable code
corresponds to that shown in Table 2. The abbreviations of the spectral indices correspond to those shown in Table 3. For the first order (1rd) and second order (2rd) textures, CON = contrast, COR = correlation, occurrence of a new forest fire in the same area. For instance, within

B7(COR) (-)*
our study area, it would be a priority to evaluate the dominant tree
species and the understory community structure recovery within the
high and low fire recurrence scenarios, respectively, given their low
regeneration observed under those conditions. The methodology pro-

B6(MEAN) (-)***
posed in this study may be applied to other high spatial resolution re-
mote sensors available for land managers, which channels cover similar

B6(MEAN)
regions of the electromagnetic spectrum.

(+)**
Pinus pinaster seedlings presents a different spectral response in re-
lation to the surrounding understory vegetation, especially in the NIR

B8(CON) (+)**
region (Viedma et al., 2012) (Figure SM.6 of the Supplementary
B3(ENT) (-)**

B4(MEAN)
Material), given its geometry and leaf characteristics (Rautiainen,
2005). These spectral differences along with the regeneration ability of
(-)***

the pine population allowed for satisfactory modeling of the variables

corresponding to the number and cover of Pinus pinaster seedlings for
B3(HOM) (-)***

the entire study framework. However, for the case of the high recur-
B1(ENT) (+)*
2rd Textures

rence and severity scenario, where the number of seedlings was very
B3(MEAN)

small and situated under a dense shrub canopy, neither spectral in-
(+)***

formation nor textures could explain the population recovery due to the
small contribution of the seedlings to the overall reflectance of the plots
(Viedma et al., 2012).
The Pinus pinaster seedling prediction model was calibrated with
B8(MEAN)

Anthocyanin Reflectance Index 2 and Plant Senescence Reflectance

(+)***

Index, along with the first order textures corresponding to the yellow
region entropy and NIR-1 data range. The two spectral indices were
B8(DR) (-)**

negatively correlated with the number of pine seedlings, since these

indices measure plant stress from anthocyanins and carotenoids de-
tection (Merzlyak et al., 1999; Gitelson et al., 2001). Significant an-
thocyanins and carotenoids accumulation in this phenological stage is
induced, for example, as a result of stress caused by increased solar
B7(DR) (+)**

radiation received by the plant, presence of wounds or competitive

B5(MEAN)
DIS = dissimilarity, DR = data range, ENT = entropy, HOM = homogeneity, MEAN = mean and VAR = variance.

effects (Lee and Graham, 1986). These environmental stress conditions

(-)***

are characteristic after the occurrence of a forest fire and a higher plant
stress intensity caused by the competitive effects of resprouter un-
B3(DR) (+)*

B5(DR) (+)*
1rd Textures

B4(ENT) (-)*

derstory species could adversely affect the recruitment of Pinus pinaster

seedlings (Calvo et al., 2008; Taboada et al., 2017). Similarly, the en-
tropy texture of the yellow region presents a negative correlation with
the number of living pine seedlings. This fact can be explained because
BAI (-)***

this region can be used to detect the characteristic vegetation “yel-

PSRI (-)*

lowness” under certain conditions, such as plant stress (Immitzer et al.,

Set of predictors

Spectral indices

2012), and therefore a greater diversity of reflectance values in this

region of the spectrum could present an inverse relationship with the
BAI (+)***
ARI (+)**

WSI (-)**
ARI (-)**

number of Pinus pinaster seedlings. The direct correlation between the

data range in the NIR-1 region and the number of pine seedlings could
be explained because reflectance in this region varies substantially for
nRMSE

0.1645
0.1747

0.0906

0.1690

0.1929
0.2006
0.2157

different types of vegetation cover, such as conifers and understory

woody species, this band being really effective in the discrimination of
different vegetation types (DigitalGlobe, 2010; Immitzer et al., 2012;
Deviance (D2)

Viedma et al., 2012). The Pinus pinaster seedling prediction model re-
present an important base for land management decision-making in
0.3363
0.3863

0.6302

0.4288

0.3641
0.3495
0.3307

fire-prone pine systems given its predictive capacity. This tool may
assist land managers in identifying areas where external intervention
could be necessary to facilitate the regeneration of the dominant tree
species and recover their ecosystem services supply.
*** Significant variable at p < 0.001.
Response
variable

Wmaxh

The cover, height and richness of woody species (in addition to the
** Significant variable at p < 0.01.
* Significant variable at p < 0.05.
Wrich
Wcov

Ncov
Pcov

percentage of bare soil, necromass and leaves) were the post-fire re-
Pn

generation variables of the understory community with the best pre-

Model

dictive capacity. Despite the heterogeneity of land cover still present in

the study area, the successful modeling of the understory community
1
2

5
6
7

variables could be explained by the high regeneration ability of most of

Understory community

the understory species (Calvo et al., 2008) without the competitive

effects of trees, along with the high spatial resolution of the WorldView-
population
Pinus pinaster

2 image. Wood et al. (2012) reported that the use of satellite imagery at
Category

lower spatial resolution than that provided by WorldView-2 would not

Table 8

allow the identification of land cover changes given the spatial varia-
bility of these heterogeneous systems. In addition, the use of moderate

266
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Fig. 6. Spatial output of the understory woody species cover model developed for the entire study framework (see Table 8). White areas were covered by clouds or
their projected shadows in the raw WorldView-2 image and, therefore, were masked at the processing stage.

spatial resolution sensors, such as Landsat, involve a sub-pixel mixing 2001). The correlation of the NIR-2 data range with vegetation cover is
question (Veraverbeke et al., 2012) and would require the application inverse, since a wider range of reflectance values in this region in-
of sub-pixel image analysis techniques (Quintano et al., 2012; dicates the presence of other covers along with the vegetation. The
Fernández-Manso et al., 2016). model output could be used as a tool by land managers to define po-
The model of the woody understory species cover calibrated from tential areas of post-fire erosion with low vegetation recovery (Storey
first-order textures presents a predictive error of only 9%, both in the et al., 2016) and to detect areas of high fuel continuity in case of the
entire study framework and under different recurrence-severity sce- occurrence of a new wildfire.
narios. This is a relatively low error considering the difficulty in dis- The richness of woody species was successfully modeled in terms of
criminating covers from multispectral bands comparing to hyperspec- the homogeneity and contrast of green and NIR-2 bands, respectively.
tral sensors in large systems with high spatial heterogeneity (Yue et al., Vegetation patches with high species richness are more heterogeneous
2012). The model predictors with the highest significance are the mean than patches with a lower richness. Therefore, the spectral homogeneity
reflectance of both red and NIR-2 regions. The mean red reflectance is of the green band had an inverse correlation with species richness,
inversely correlated with vegetation cover since pigments from photo- while the correlation of NIR-2 band contrast was direct. Viedma et al.
synthetically active vegetation absorb a large part of this wavelength (2012) found the same patterns between species richness and textural
radiation (Elvidge and Chen, 1995). Meanwhile, the mean of NIR-2 features in their study at several spatial scales with the QuickBird sa-
region has a direct correlation with the vegetation cover since the in- tellite.
ternal structures of leaves actively reflect this radiation (Slaton et al., The maximum height of the woody species model calibrated from

267
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Fig. 7. Spatial output of the number of Pinus pinaster living seedlings model built for the entire study framework (see Table 8). White areas were covered by clouds or
their projected shadows in the raw WorldView-2 image and, therefore, were masked at the processing stage.

image texture data had a better predictive capacity than when cali- Other spatially explicit models of the understory community, such
brated from spectral data alone, as demonstrated by Ozdemir and as necromass and bare soil cover, offer a relevant ecological basis for
Karnieli (2011). The texture of the satellite imagery will be more uni- post-fire decision management in situations of soil loss erosion and
form when the vegetation consists in short and small individuals accumulation of dead biomass, particularly in Mediterranean systems
(Petrou et al., 2012; Petrou et al., 2015), which could explain the direct (Shakesby, 2011). The WorldView Soil Index contribution to the model
and the inverse relationship of the coastal blue band entropy and the of bare soil must be highlighted. Due to the lack of a shortwave infrared
NIR-1 band correlation, respectively, with the maximum height of the (SWIR) band in the sensor on-board WorldView-2 satellite, the differ-
woody species. For its part, the prediction error of the model was in line ence in reflectivity between the green and yellow region of the spec-
with the study conducted by Kayitakire et al. (2006), in which the top trum allows this index to detect the bare soil signal (DigitalGlobe,
height of the canopy was modeled on the basis of texture features ex- 2010). Whelan (1995) pointed out that the dead plant material (ne-
tracted from high spatial resolution satellite imagery. Although the cromass) presents high flammability and, thus, its removal from the
accepted fact that optical sensors can only estimate vegetation height previously burnt area must be a post-fire management priority. We
up to the canopy closure (Donoghue and Watt, 2006), this concern was identified a high correlation of dead plant material with the Burnt Area
not present in our study area because of the pine population seedlings Index due to the high charcoal content of these materials and the ability
and the understory vegetation structure consists on a unique vertical of this index to detect its signal (Chuvieco et al., 2002).
layer. It should be noted that first and second order textures were

268
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

particularly relevant in all post-fire recovery models because of the Appendix A. Supplementary material
great spatial variation of the vegetation structure in the study area.
Gómez et al. (2011) and Pu and Cheng (2015) highlighted that the Supplementary data to this article can be found online at https://
textures derived from the visible region of the spectrum offer a better doi.org/10.1016/j.foreco.2019.05.028.
model performance than those calculated from the red edge and near-
infrared region to predict parameters such as vegetation cover or ve- References
getation height. However, several textures from both regions have of-
fered a similar performance in the present study. Adler-Golden, S.M., Matthew, M.W., Bernstein, L.S., Levine, R.Y., Berk, A., Richtsmeier,
The use of active sensors such as LiDAR together with optical sa- S.C., Acharya, P.K., Anderson, G.P., Felde, G., Gardner, J., Hike, M., Jeong, L.S.,
Pukall, B., Mello, J., Ratkowski, A., Burke, H.H., 1999. Atmospheric correction for
tellite imagery could improve the estimation of forest structural para- shortwave spectral imagery based on MODTRAN4. SPIE Proc. 3753, 61–69.
meters over ecosystems with complex vegetation vertical structure Álvarez, A., Gracia, M., Vayreda, J., Retana, J., 2012. Patterns of fuel types and crown fire
(Clawges et al., 2008). However, in a full stand replacing post-fire en- potential in Pinus halepensis forest in the Western Mediterranean Basin. For. Ecol.
Manage. 270, 282–290.
vironment with heterogeneous horizontal vegetation structure, high Alvarez, R., Valbuena, L., Calvo, L., 2007. Effect of high temperatures on seed germina-
spatial resolution satellite imagery alone has proven to be in this study tion and seedling survival in three pine species (Pinus pinaster, P. sylvestris and P.
a useful tool in the estimation of post-fire vegetation structure. More- nigra). Int. J. Wildland Fire 16, 63–70.
Bannari, A., Asalhi, H., Teillet, P., 2002. Transformed Difference Vegetation Index (TDVI)
over, LiDAR data are not widely accessible for land managers as op-
for Vegetation Cover Mapping. Proceedings of the Geoscience and Remote Sensing
posed to satellite imagery (Wood et al., 2012) and their costs could be Symposium.
significantly higher (Donoghue and Watt, 2006; Petrou et al., 2015; Barry, K.M., Stone, C., Mohammed, C.L., 2008. Crown-scale evaluation of spectral indices
for defoliated and discoloured eucalypts. Int. J. Remote Sens. 29 (1), 47–69.
Fassnacht et al., 2016).
Bastin, J.F., Barbier, N., Couteron, P., Adams, B., Shapiro, A., Bogaert, J., De Cannière, C.,
2014. Aboveground biomass mapping of African forest mosaics using canopy texture
5. Conclusions analysis: toward a regional approach. Ecol. Appl. 24, 1984–2001.
Beaty, R.M., Taylor, A.H., 2001. Spatial and temporal variation of fire regimes in a mixed
conifer forest landscape, Southern Cascades, California, USA. J. Biogeogr. 28 (8),
1. The highest recovery of the Pinus pinaster population occurred in the 955–966.
low fire recurrence scenario, while the understory community ex- Berk, A., Anderson, G.P., Bernstein, L.S., Acharya, P.K., Dothe, H., Matthew, M.W., Adler-
hibited a greatest recovery under high fire recurrence. Fire severity Golden, S.M., Chetwyn, J.H., Richtsmeier, S.C., Pukall, B., Allred, C.L., Jeong, L.S.,
Hoke, M.L., 1999. MODTRAN4 radiative transfer modeling for atmospheric correc-
was only related to the Pinus pinaster population recovery, exhibiting tion. SPIE Proc. 3756, 348–353.
an inverse relationship. Boegh, E., Soegaard, H., Broge, N., Hasager, C., Jensen, N., Schelde, K., Thomsen, A.,
2. The spatial resolution of the WorldView-2 multispectral imagery 2002. airborne multi-spectral data for quantifying leaf area index, nitrogen con-
centration and photosynthetic efficiency in agriculture. Remote Sens. Environ. 81,
was appropriate for the quantitative analysis of the post-fire vege- 179–193.
tation structure in fire-prone Mediterranean ecosystems dominated Calvo, L., Baeza, J., Marcos, E., Santana, V., Papanastasis, V.P., 2012. Post-Fire man-
by Pinus pinaster. agement of shrublands. Springer Science+Business Media B.V., Dordrecht, pp.
293–319.
3. At the Pinus pinaster population level, the number and cover of
Calvo, L., Santalla, S., Valbuena, L., Marcos, E., Tárrega, R., Luis-Calabuig, E., 2008. Post-
seedlings were the best-modeled recovery structure variables. fire natural regeneration of a Pinus pinaster forest in NW Spain. Plan Ecol. 197,
However, Pinus pinaster population structure models require a 81–90.
Calvo, L., Huerta, S., Marcos, E., Calvo-Fernández, J., Taboada, A., 2015. The role of
seedling abundance large enough in the field plots to contribute at a
prescribed fire in the provision of regulating ecosystem services of Spanish heath-
large extent to the overall plot reflectance. For the understory lands. Ecol. Quest. 21, 71–73.
community, the woody species cover model presented the best Calvo, L., Hernández, V., Valbuena, L., Taboada, A., 2016. Provenance and seed mass
predictive capacity. determine seed tolerance to high temperatures associated to forest fires in Pinus pi-
naster. Ann. Forest Sci. 73, 381–391.
4. Prediction models of post-fire vegetation recovery had the same Chen, J., 1996. Evaluation of vegetation indices and modified simple ratio for boreal
performance in general terms both for the entire study framework applications. Canad. J. Remote Sens. 22, 229–242.
and for each recurrence-severity scenario. Chen, D., Stow, D.A., Gong, P., 2004. Examining the effect of spatial resolution and
texture window size on classification accuracy: an urban environment case. Int. J.
5. The combination of spectral indices, together with first or second Remote Sens. 25, 1–16.
order textures, or textures alone as predictors, provided the best Chu, T., Guo, X., Takeda, K., 2016. Remote sensing approach to detect post-fire vegetation
results in all recovery models. Image texture from visible and in- regrowth in Siberian boreal larch forest. Ecol. Ind. 62, 32–46.
Chuvieco, E., Pilar-Martin, M., Palacios, A., 2002. Assessment of different spectral indices
frared region of the spectrum offered a similar performance in this in the red-near-infrared spectral domain for burned land discrimination. Int. J.
study. Remote Sens. 23, 5103–5110.
6. The approach proposed in the present study could represent a fun- Chuvieco, E., Aguado, I., Yebra, M., Nieto, H., Salas, J., Martín, M.P., Vilar, L., Martínez,
J., Martín, S., Ibarra, P., de la Riva, J., Baeza, J., Rodríguez, F., Molina, J.R., Herrera,
damental tool for planning post-fire restoration actions given its M.A., Zamora, R., 2010. Development of a framework for fire risk assessment using
potential to identify areas where external intervention could be remote sensing and geographic information system technologies. Ecol. Model. 221
necessary to promote the recovery of provisioning and regulating (1), 46–58.
Chuvieco, E., Kasischke, E.S., 2007. Remote sensing information for fire management and
ecosystem services supply.
fire effects assessment. J. Geophys. Res. 112, G01S90.
Clawges, R., Vierling, K., Vierling, L., Rowell, E., 2008. The use of airborne lidar to assess
Acknowledgments avian species diversity, density, and occurrence in a pine/aspen forest. Remote Sens.
Environ. 112, 2064–2073.
Clemente, R., Navarro-Cerrillo, R., Gitas, I., 2009. Monitoring post-fire regeneration in
This study was financially supported by the Spanish Ministry of Mediterranean ecosystems by employing multitemporal satellite imagery. Int. J.
Economy and Competitiveness, and the European Regional Wildland Fire 18, 648–658.
Development Fund (ERDF), in the framework of the GESFIRE R Core Team (2015). R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.
(AGL2013-48189-C2-1-R) and FIRESEVES (AGL2017-86075-C2-1-R) org/.
projects; and by the Regional Government of Castilla and León in the Crippen, R., 1990. Calculating the vegetation index faster. Remote Sens. Environ. 34,
framework of the FIRECYL (LE033U14) and SEFIRECYL (LE001P17) 71–73.
Cuevas-Gonzalez, M., Gerard, F., Baltzer, H., Riano, D., 2009. Analysing forest recovery
projects. J.M. Fernández-Guisuraga is supported by a predoctoral fel- after wildfire disturbance in boreal Siberia using remotely sensed vegetation indices.
lowship from the Spanish Ministry of Education, Spain (FPU16/03070).

269
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Glob. Change Biol. 15, 561–577. on image texture analysis and IKONOS-2 imagery. Remote Sens. Environ. 102,
Daughtry, C.S.T., Walthall, C.L., Kim, M.S., Colstoun, E.B., McMurtrey, J.E., 2000. 390–401.
Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. Key, C.H., 2006. Ecological and sampling constraints on defining landscape fire severity.
Rem. Sens. Environ. 74, 229–239. Fire Ecol. 2, 34–59.
Diaz-Delgado, R., Lloret, F., Pons, X., 2003. Influence of fire severity on plant regenera- Key, C.H., Benson, N.C., 2006. Landscape assessment (LA) sampling and analysis
tion through remote sensing imagery. Int. J. Remote Sens. 24, 1751–1763. methods. USDA Forest Serv. Gene. Tech. Rep (RMRS-GTR-164-CD).
DigitalGlobe, 2010. < http://global.digitalglobe.com > , [Accessed in: 07-01-2016]. Lecina-Diaz, J., Alvarez, A., Retana, J., 2014. Extreme fire severity patterns in topo-
Donoghue, D.N.M., Watt, P.J., 2006. Using LiDAR to compare forest height estimates graphic, convective and wind-driven historical wildfires of mediterranean pine for-
from IKONOS and Landsat ETM+ data in Sitka spruce plantation forests. Int. J. ests. PLoS ONE 9 (1), e85127.
Remote Sens. 27, 2161–2175. Lee, D.W., Graham, R., 1986. Leaf optical properties of rainforest sun and extreme shade
Draper, N.R., Smith, H., 1998. Applied Regression Analysis. Wiley-Intersci. 736, pp. plants. Am. J. Bot. 73, 1100–1108.
Eckert, S., 2012. Improved forest biomass and carbon estimations using texture measures Leone, V., Lovreglio, R., 2004. Conservation of Mediterranean pine woodlands: scenarios
from WorldView-2 satellite data. Rem. Sens. 4, 810–829. and legislative tools. Plant Ecol. 171, 221–235.
Elvidge, C.D., Chen, Z., 1995. Comparison of broad-band and narrow-band red and near- Leverkus, A.B., Rey Benayas, J.M., Castro, J., Boucher, D., Brewer, S., Collins, B.M.,
infrared vegetation indices. Remote Sens. Environ. 54 (1), 38–48. Donato, D., Fraver, S., Kishchuk, B.E., Lee, E.-J., Lindenmayer, D.B., Lingua, E.,
Engler, R., Waser, L.T., Zimmermann, N.E., Schaub, M., Berdos, S., Ginzler, C., Psomas, Macdonald, E., Marzano, R., Rhoades, C.C., Royo, A., Thorn, S., Wagenbrenner, J.W.,
A., 2013. Combining ensemble modeling and remote sensing for mapping individual Waldron, K., Wohlgemuth, T., Gustafsson, L., 2018. Salvage logging effects on reg-
tree species at high spatial resolution. For. Ecol. Manage. 310, 64–73. ulating and supporting ecosystem services - a systematic map. Can. J. For. Res. 48,
Fassnacht, F.E., Latifi, H., Stereńczak, K., Modzelewska, A., Lefsky, M., Waser, L.T., 983–1000.
Straub, C., Ghosh, A., 2016. Review of studies on tree species classification from Lozano, F.J., Suárez-Seoane, S., Kelly, M., Luis-Calabuig, E., 2008. A multi-scale approach
remotely sensed data. Remote Sens. Environ. 186, 64–87. for modeling fire occurrence probability using satellite data and classification trees: a
Fernández-García, V., Quintano, C., Taboada, A., Marcos, E., Calvo, L., Fernández-Manso, case study in a mountainous Mediterranean region. Remote Sens. Environ. 112,
A., 2018a. Remote sensing applied to the study of fire regime attributes and their 708–719.
influence on post-fire greenness recovery in pine ecosystems. Rem. Sens. 10, 733. Lozano, F.J., Suárez-Seoane, S., De Luis, E., 2010. Effects of wildfires on environmental
Fernández-García, V., Santamarta, M., Fernández-Manso, A., Quintano, C., Marcos, E., variability: a comparative analysis using different spectral indices, patch metrics and
Calvo, L., 2018b. Burn severity metrics in fire-prone pine ecosystems along a climatic thematic resolutions. Landscape Ecol. 25 (5), 697–710.
gradient using Landsat imagery. Remote Sens. Environ. 206, 205–217. Lozano, F.J., Suárez-Seoane, S., de Luis-Calabuig, E., 2012. Does fire regime affect both
Fernández-García, V., Fulé, P.Z., Marcos, E., Calvo, L., 2019. The role of fire frequency temporal patterns and drivers of vegetation recovery in a resilient Mediterranean
and severity on the regeneration of Mediterranean serotinous pines under different landscape? a remote sensing approach at two observation levels. Int. J. Wildland Fire
environmental conditions. For. Ecol. Manage. 444, 59–68. 21 (6), 666–679.
Fernández-Guisuraga, J.M., Calvo, L., Fernández-García, V., Marcos-Porras, E., Taboada, Marchetti, M., Ricotta, C., Volpe, F., 1995. A qualitative approach to the mapping of post-
A., Suárez-Seoane, S., 2019. Efficiency of remote sensing tools for post-fire man- fire regrowth in Mediterranean vegetation with Landsat TM data. Int. J. Remote Sens.
agement along a climatic gradient. For. Ecol. Manage. 433, 553–562. 16, 2487–2494.
Fernández-Manso, A., Quintano, C., Roberts, D.A., 2016. Burn severity influence on post- Martin, T.G., Wintle, B.A., Rhodes, J.R., Kuhnert, P.M., Field, S.A., Low-Choy, S.J., Tyre,
fire vegetation cover resilience from landsat MESMA fraction images time series in A.J., Possingham, H.P., 2005. Zero tolerance ecology: improving ecological inference
Mediterranean forest ecosystems. Remote Sens. Environ. 184, 112–123. by modelling the source of zero observations. Ecol. Lett. 8, 1235–1246.
Gamon, J., Surfus, J., 1999. Assessing leaf pigment content and activity with a re- Matthew, M., Adler-Golden, S., Berk, A., Felde, G., Anderson, G., Gorodetzky, D.,
flectometer. New Phytol. 143, 105–117. Paswaters, S., Shippert, M., 2003. Atmospheric correction of spectral imagery: eva-
García-Morote, F.A., Martínez-García, E., Andrés-Abellán, M., Caballero, E.R., Miettinen, luation of the FLAASH algorithm with AVIRIS data. SPIE Proc. 5093, 474–482.
H., López-Serrano, F.R., 2017. Direct seeding of Pinus halepensis Mill. for recovery of McCullagh, P., Nelder, J.A., 1989. Generalized Linear Models. In: Chapman and Hall, pp.
burned semi-arid forests: implications for post-fire management for improving. Nat. 532.
Regener. For. 8, 353. McFadden, D., 1974. Conditional Logit Analysis of Qualitative Choice Behavior. In:
Gitelson, A., Merzlyak, M., 1998. Remote sensing of chlorophyll concentration in higher Zarembka, P. (Ed.), Frontiers in Economics. Academic Press, New York, pp. 105–142.
plant leaves. Adv. Space Res. 22, 689–692. McKenna, P., Phinn, S., Erskine, P.D., 2018. Fire severity and vegetation recovery on mine
Gitelson, A., Kaufman, Y., Merzylak, M., 1996. Use of a green channel in remote sensing site rehabilitation using WorldView-3 imagery. Fire 1, 22.
of global vegetation from EOS-MODIS. Remote Sens. Environ. 58, 289–298. Meng, R., Wu, J., Zhao, F., Cook, B.D., Hanavan, R.P., Serbin, S.P., 2018. Measuring short-
Gitelson, A., Merzlyak, M., Chivkunova, O., 2001. Optical properties and nondestructive term post-fire forest recovery across a burn severity gradient in a mixed pine-oak
estimation of anthocyanin content in plant leaves. Photochem. Photobiol. 71, 38–45. forest using multi-sensor remote sensing techniques. Remote Sens. Environ. 210,
Gitelson, A.A., Stark, R., Grits, U., Rundquist, D., Kaufman, Y., Derry, D., 2002b. 282–296.
Vegetation and soil lines in visible spectral space: a concept and technique for remote Merzlyak, M., Gitelson, A., Chivkunova, O., Rakitin, V., 1999. Non-destructive optical
estimation of vegetation fraction. Int. J. Remote Sens. 23, 2537–2562. detection of pigment changes during leaf senescence and fruit ripening. Physiol.
Gitelson, A., Zur, Y., Chivkunova, O., Merzlyak, M., 2002a. Assessing carotenoid content Plant. 106 (1), 135–141.
in plant leaves with reflectance spectroscopy. Photochem. Photobiol. 75 (3), Miller, J.D., Thode, A.E., 2007. Quantifying burn severity in a heterogeneous landscape
272–281. with a relative version of the delta normalized burn ratio (dNBR). Remote Sens.
Gómez, C., Wulder, M.A., Montes, F., Delgado, J.A., 2011. Forest structural diversity Environ. 109, 66–80.
characterization in Mediterranean pines of central Spain with QuickBird-2 imagery Moreira, F., Catry, F., Lopes, T., Bugalho, M.N., Rego, F., 2009. Comparing survival and
and canonical correlation analysis. Canad. J. Remote Sens. 37, 628–642. size of resprouts and planted trees for post-fire forest restoration in central Portugal.
Gonzalez, M., Augusto, L., Gallet-Budynek, A., Xue, J., Yauschew-Raguenes, N., Guyon, Ecol. Eng. 35 (5), 870–873.
D., Trichet, P., Delerue, F., Niollet, S., Andreasson, F., Achat, D.L., Bakker, M.R., Moreira, F., Viedma, O., Arianoutsou, M., Curt, T., Koutsias, N., Rigolot, E., Barbati, A.,
2013. Contribution of understory species to total ecosystem aboveground and be- Corona, P., Vaz, P., Xanthopoulos, G., Mouillot, F., Bilgili, E., 2011. Landscape-
lowground biomass in temperate Pinus pinaster Ait. Forests. Forest Ecol. Manage. 289, wildfire interactions in southern Europe: implications for landscape management. J.
38–47. Environ. Manage. 92 (10), 2389–2402.
González-De Vega, S., De las Heras, J., Moya, D., 2016. Resilience of Mediterranean Ni, N., Chen, N., Chen, J., Liu, M., 2015. Integrating WorldView-2 imagery and terrestrial
terrestrial ecosystems and fire severity in semiarid areas: responses of Aleppo pine LiDAR point clouds to extract dyke swarm geometry: implications for magma em-
forests in the short, mid and long term. Sci. Total Environ. 573, 1171–1177. placement mechanisms. J. Volcanol. Geoth. Res. 310, 1–11.
Gu, Z., Ju, W., Li, L., Li, D., Liu, Y., Fan, W., 2013. Using vegetation indices and texture Ozdemir, I., Karnieli, A., 2011. Predicting forest structural parameters using the image
measures to estimate vegetation fractional coverage (VFC) of planted and natural texture derived from WorldView-2 multispectral imagery in a dryland forest, Israel.
forests in Nanjing city, China. Adv. Space Res. 51, 1186–1194. Int. J. Appl. Earth Obs. Geoinf. 13, 701–710.
Haboudane, D., Miller, J.R., Pattey, E., Zarco-Tejada, P.J., Strachan, I.B., 2004. Pausas, J.G., Keeley, J.E., 2014. Abrupt climate-independent fire regime changes.
Hyperspectral vegetation indices and novel algorithms for predicting green LAI of Ecosystems 17, 1109–1120.
crop canopies: modeling and validation in the context of precision agriculture. Pausas, J.G., Vallejo, V.R., 1999. The role of fire in European Mediterranean Ecosystems.
Remote Sens. Environ. 90, 337–352. In: Chuvieco, E. (Ed.), Remote Sensing of Large Wildfires in the European
Haralick, R.M., Shanmugam, K., Dinstein, I.H., 1973. Textural features for image classi- Mediterranean Basin. Springer-Verlag, Berlin, pp. 3–16.
fication. IEEE Transac. Syst., Man Cybern. 3 (6), 610–621. Penuelas, J., Baret, F., Filella, I., 1995. Semi-empirical indices to assess carotenoids/
Immitzer, M., Atzberger, C., Koukal, T., 2012. Tree species classification with random chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31 (2), 221–230.
forest using very high spatial resolution 8-band WorldView-2 satellite data. Remote Pesonen, A., Maltamo, M., Eerikäinen, K., Packalèn, P., 2008. Airborne laser scanning-
Sensing 4, 2661–2693. based prediction of coarse woody debris volumes in a conservation area. For. Ecol.
Joyce, M.J., Erb, J.D., Sampson, B.A., Moen, R.A., 2019. Detection of coarse woody debris Manage. 255, 3288–3296.
using airborne light detection and ranging (LiDAR). For. Ecol. Manage. 433, Petrou, Z., Tarantino, C., Adamo, M., Blonda, P., Petrou, M., 2012. Estimation of vege-
678–689. tation height through satellite image texture analysis. Int. Arch. Photogram., Remote
Jung, M., Tautenhahn, S., Wirth, C., Kattge, J., 2013. Estimating basal area of spruce and Sens. Spat. Inform. Sci. XXXIX-B8, 321–326.
fir in post-fire residual stands in Central Siberia using quickbird, feature selection, Petrou, Z., Manakos, I., Stathaki, T., Mücher, C., Adamo, M., 2015. Discrimination of
and random forests. Procedia Comput. Sci. 18, 2386–2395. vegetation height categories with passive satellite sensor imagery using texture
Kaufman, Y., Tanre, D., 1992. Atmospherically resistant vegetation index (ARVI) for EOS- analysis. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 8, 1442–1455.
MODIS. IEEE Trans. Geosci. Remote Sens. 30, 261–270. Pleniou, M., Koutsias, N., 2013. Sensitivity of spectral reflectance values to different burn
Kayitakire, F., Hamel, C., Defourny, P., 2006. Retrieving forest structure variables based and vegetation ratios: a multi-scale approach applied in a fire affected area. ISPRS J.

270
J.M. Fernández-Guisuraga, et al. Forest Ecology and Management 446 (2019) 257–271

Photogram. Remote Sens. 79, 199–210. Landsat imagery. Remote Sens. Environ. 183, 53–64.
Pu, R., Cheng, J., 2015. Mapping forest leaf area index using reflectance and textural Sullivan, A.L., Surawski, N.C., Crawford, D., Hurley, R.J., Volkova, L., Weston, C.J.,
information derived from WorldView-2 imagery in a mixed natural forest area in Meyer, C.P., 2018. Effect of woody debris on the rate of spread of surface fires in
Florida, US. Int. J. Appl. Earth Obs. Geoinf. 42, 11–23. forest fuels in a combustion wind tunnel. For. Ecol. Manage. 424, 236–245.
Quintano, C., Fernández-Manso, A., Shimabukuro, Y.E., Pereira, G., 2012. Spectral un- Taboada, A., Tárrega, R., Marcos, E., Valbuena, L., Suárez-Seoane, S., Calvo, L., 2017. Fire
mixing: a review. Int. J. Remote Sens. 33, 5307–5340. recurrence and emergency post-fire management influence seedling recruitment and
Quintano, C., Fernández-Manso, A., Calvo, L., Marcos, E., Valbuena, L., 2015. Land sur- growth by altering plant interactions in fire-prone ecosystems. For. Ecol. Manage.
face temperature as potential indicator of burn severity in forest Mediterranean 402, 63–75.
ecosystems. Int. J. Appl. Earth Obs. Geoinf. 36, 1–12. Taboada, A., Fernández-García, V., Marcos, E., Calvo, L., 2018. Interactions between large
Rautiainen, M., 2005. The spectral signature of coniferous forests: the role of stand high-severity fires and salvage logging on a short return interval reduce the regrowth
structure and leaf area index. In: Doctoral Thesis. Universidad de Helsinki, Finlandia, of fire-prone serotinous forests. For. Ecol. Manage. 414, 54–63.
pp. 54. Tapias, R., Climent, J., Pardos, J.A., Gil, L., 2004. Life histories of Mediterranean pines.
Rondeaux, G., Steven, M., Baret, F., 1996. Optimization of soil-adjusted vegetation in- Plant Ecol. 171, 53–68.
dices. Remote Sens. Environ. 55, 95–107. Taye, Z.M., Martínez-Peña, F., Bonet, J.A., de Aragón, J.M., de-Miguel, S., 2016.
Roujean, J., Breon, F., 1995. Estimating PAR absorbed by vegetation from bidirectional Meteorological conditions and site characteristics driving edible mushroom produc-
reflectance measurements. Remote Sens. Environ. 51, 375–384. tion in Pinus pinaster forests of Central Spain. Fungal Ecol. 23, 30–41.
Rouse, J., Haas, R., Schell, J., Deering, D., 1973. Monitoring Vegetation Systems in the Tessler, N., Sapir, Y., Wittenberg, L., Greenbaum, N., 2016. Recovery of Mediterranean
Great Plains with ERTS. Third ERTS Symposium. vegetation after recurrent forest fires: insight from the 2010 forest fire on Mount
Ruíz-Gallardo, J.R., Castaño, S., Calera, A., 2004. Application of remote sensing and GIS Carmel, Israel. Land Degrad. Dev. 27, 1424–1431.
to locate priority intervention areas after wildland fires in Mediterranean systems: a Tucker, C., 1979. Red and photographic infrared linear combinations for monitoring
case study from southeastern Spain. Int. J. Wildland Fire 13, 241–252. vegetation. Remote Sens. Environ. 8, 127–150.
Sagra, J., Ferrandis, P., Plaza-Álvarez, P.A., Lucas-Borja, M.E., González-Romero, J., Van Drooge, B.L., Sicard, M., Stohl, A., Fontal, M., Bravo, N., Muñoz, A., Lange, D.,
Alfaro-Sánchez, R., De las Heras, J., Moya, D., 2018. Regeneration of Pinus pinaster Fernández, P., Grimalt, J.O., 2016. Detection and simulation of wildfire smoke im-
Aiton after prescribed fires: response to burn timing and biogeographical seed pro- pacting a Mediterranean urban atmosphere. Atmos. Pollut. Res. 7, 494–502.
venance across a climatic gradient. Sci. Total Environ. 637, 1550–1558. Veraverbeke, S., Gitas, I., Katagis, T., Polychronaki, A., Somers, B., Goossens, R., 2012.
Santamaría, J.E., 2015. El pino Pinaster de la Sierra del Teleno: crecimiento, producción y Assessing post-fire vegetation recovery using red–near infrared vegetation indices:
selvicultura. In: Doctoral Thesis. Universidad de León, España, pp. 156. accounting for background and vegetation variability. ISPRS J. Photogramm. Remote
Schmeer, S.R., Kampf, S.K., MacDonald, L.H., Hewitt, J., Wilson, C., 2018. Empirical Sens. 68, 28–39.
models of annual post-fire erosion on mulched and unmulched hillslopes. CATENA Ver-Hoef, J.M., Boveng, P.L., 2007. Quasi-poisson vs. negative binomial regression: how
163, 276–287. should we model overdispersed count data? Ecology 88, 2766–2772.
Schoennagel, T., Smithwick, A.H., Turner, M.G., 2008. Landscape heterogeneity fol- Vermote, E.F., Saleous, N., Justice, C.O., Kaufman, Y.J., Privette, J.L., Remer, L., Roger,
lowing large fires: insights from Yellowstone National Park, USA. Int. J. Wildland Fire J.C., Tanre, D., 1997. Atmospheric correction of visible to middle-infrared EOS-
17, 742–753. MODIS data over land surfaces: background, operational algorithm and validation. J.
Shakesby, R.A., 2011. Post-wildfire soil erosion in the Mediterranean: review and future Geophys. Res. 102, 17131–17141.
research directions. Earth Sci. Rev. 105, 71–100. Viedma, O., Torres, I., Pérez, B., Moreno, J.M., 2012. Modeling plant species richness
Slaton, M.R., Hunt Jr, E.R., Smith, W.K., 2001. Estimating near-infrared leaf reflectance using reflectance and texture data derived from QuickBird in a recently burned area
from leaf structural characteristics. Am. J. Bot. 88 (2), 278–284. of Central Spain. Remote Sens. Environ. 119, 208–221.
Slesak, R.A., Schoenholtz, S.H., Evans, D., 2015. Hillslope erosion two and three years Vieira, D.C.S., Malvar, M.C., Martins, M.A.S., Serpa, D., Keizer, J.J., 2018. Key factors
after wildfire, skyline salvage logging, and site preparation in southern Oregon, USA. controlling the post-fire hydrological and erosive response at micro-plot scale in a
For. Ecol. Manage. 342, 1–7. recently burned Mediterranean forest. Geomorphology 319, 161–173.
Solans-Vila, J.P., Barbosa, P., 2010. Post-fire vegetation regrowth detection in the Deiva Whelan, R.J., 1995. The ecology of fire. Cambridge University Press, pp. 346 pp..
Marina region (Liguria-Italy) using Landsat TM and ETM+ data. Ecol. Model. 221, Wolf, A., 2010. Using WorldView 2 Vis-NIR MSI Imagery to Support Land Mapping and
75–84. Feature Extraction Using Normalized Difference Index Ratios. Unpublished report.
Soliño, M., Yu, T., Alía, R., Auñón, F., Bravo-Oviedo, A., Chambel, M.R., de Miguel, J., Del Wood, E.M., Pidgeon, A.M., Radeloff, V.C., Keuler, N.S., 2012. Image texture as a re-
Río, M., Justes, A., Martínez-Jauregui, M., Montero, G., Mutke, S., Ruiz-Peinado, R., motely sensed measure of vegetation structure. Remote Sens. Environ. 121, 516–526.
García Del Barrio, J.M., 2018. Resin-tapped pine forests in Spain: ecological diversity Yang, Z., Willis, P., Mueller, R., 2008. Impact of Band-Ratio Enhanced AWIFS Image to
and economic valuation. Sci. Total Environ. 625, 1146–1155. Crop Classification Accuracy. Proceedings of the Pecora 17 Remote Sensing
Song, C., Dickinson, M.B., 2008. Extracting forest canopy structure from spatial in- Symposium.
formation of high resolution optical imagery: tree crown size versus leaf area index. Yue, Y.M., Wang, K.L., Zhang, B., Jiao, Q.J., Liu, B., Zhang, M.Y., 2012. Remote sensing of
Int. J. Remote Sens. 29, 5605–5622. fractional cover of vegetation and exposed bedrock for karst rocky desertification
Sripada, R.P., Heiniger, R.W., White, J.G., Meijer, A.D., 2006. Aerial color infrared assessment. Procedia Environ. Sci. 13, 847–853.
photography for determining early in-season nitrogen requirements in corn. Agron. J. Zeileis, A., Kleiber, C., Jackman, S., 2008. Regression models for count data in R. J. Stat.
98, 968–977. Softw. 27, 1–25.
Stephens, S.L., Burrows, N., Buyantuyev, A., Gray, R.W., Keane, R.E., Kubian, R., Liu, S.R., Zhang, C., Xie, Z., 2012. Combining object-based texture measures with a neural network
Seijo, F., Shu, L.F., Tolhurst, K.G., van Wagtendonk, J.W., 2014. Temperate and for vegetation mapping in the Everglades from hyperspectral imagery. Remote Sens.
boreal forest mega-fires: characteristics and challenges. Front. Ecol. Environ. 12, Environ. 124, 310–320.
115–122. Zuur, A.F., Ieno, E.N., Elphick, C.S., 2010. A protocol for data exploration to avoid
Storey, A., Stow, D.A., O’Leary, J.F., 2016. Assessing postfire recovery of chamise cha- common statistical problems. Methods Ecol. Evol. 1, 3–14.
parral using multi-temporal spectral vegetation index trajectories derived from

271