Fire Damage Assessment in Sardinia: the use of ALOS/PALSAR data for post fire effects management

Abstract

Fires in the Sardinia Island are one of the most important environmental factors controlling
the ecosystem´s function and structure. The evaluation of fire effects by means of remote
sensing is economically and practically the best way to assess fire damage, before going
to the field. The use of alternative techniques for fire effects assessment is needed, in
particular to characterize the biomass loss at the regional level. Radar remotely sensed data
can provide great advantages with respect to optical sensors. The paper is devoted to show
the results obtained by applying a semi-automatic algorithm to the images of the L-band
SAR sensor PALSAR, on board of the ALOS satellite, for the estimate of the burned area.
To assess the quality of the estimate, the radar based results have been compared with those
obtained from optical data and ground based information.

Full text

233
European J ournal of Remote Sensi ng - 2012, 45: 233-241
doi: 10.5721/EuJRS20124521
Fire Damage Assessment in Sardinia: the use of
ALOS/PALSAR data for post re effects management
Nicolas Mari1,2*, Giovanni Laneve3, Enrico Cadau4 and Ximena Porcasi1
1 Instituto Gulich - Comisión Nacional de Actividades Espaciales, Centro Espacial Teólo Tabanera,
Córdoba, Argentina
2 Instituto de Clima y Agua - Instituto de Nacional Tecnología Agropecuaria (CIRN-INTA) Hurlingham,
Buenos Aires, Argentina
3 Universita di Roma “La Sapienza”, Centro de Ricerca Progetto San Marco, Roma, Italy
4 Serco S.p.A. Frascati (RM), Italy
*Corresponding author, e-mail address: nmari@cnia.inta.gov.ar
Abstract
Fires in the Sardinia Island are one of the most important environmental factors controlling
the ecosystem´s function and structure. The evaluation of re effects by means of remote
sensing is economically and practically the best way to assess re damage, before going
to the eld. The use of alternative techniques for re effects assessment is needed, in
particular to characterize the biomass loss at the regional level. Radar remotely sensed data
can provide great advantages with respect to optical sensors. The paper is devoted to show
the results obtained by applying a semi-automatic algorithm to the images of the L-band
SAR sensor PALSAR, on board of the ALOS satellite, for the estimate of the burned area.
To assess the quality of the estimate, the radar based results have been compared with those
obtained from optical data and ground based information.
Keywords: SAR, Forest Fires, burned area, Mediterranean, optical data.
Introduction
During year 2009, the Island of Sardinia, suffered 684 res of important size, affecting 37104
ha of which 12270 ha were wooded vegetation, as reported by the JRC 2009. According to
historical data, the island is burned every year with important impacts in the loss of wooded
vegetation. It is thought that longer dry weather periods, can negatively contribute to the
biomass loss, as fuels gets drier and more susceptible to burn [Bajocco et al., 2007]. Related
to this, there is a growing concern in the conservation of wooded areas as sinks of carbon
and, on the other hand, to avoid high frequency res, as they contribute to the release of extra
greenhouse gas emissions to the atmosphere. Therefore the estimation of burned areas and
the quantication of biomass loss is a critical parameter to understand how res contribute

Mari et al. Fire Damage as sessment in Sardinia i sland with SAR data.
234
to the consumption of different vegetation types, and how this is related to the loss of carbon
stocks and its release to the atmosphere. Some research papers reported the utility of Synthetic
Aperture Radar (SAR) data for providing information on patterns of disturbance by detecting
re scars [Bourgeau-Chavez et al., 1997; Cahoon et al., 1994, Kasischke et al., 1993; Kasischke
et al., 1997], and for studying re effects in forested areas [Tanase et al., 2010]. The ability to to
detect burned areas with SAR, as for any ecological process, will depend rstly on dening
the optimal system parameters, including microwave frequency, polarization, incidence angle,
resolution, and sampling frequency [Kasischke et al., 1997]. These characteristics will determine
the type of backscatter mechanisms over forested terrains, and related to these, the type of effects
produced by the re disturbance. The ALOS/PALSAR sensor is a L-band microwave frequency
(23cm) radar which has a deep penetration capability under the canopy, interacting with large
branches, tree stems, and even has some interaction with the ground [Le Toan et al., 1992]. It is
expected that the effect of re on the reduction of the canopy structure, can produce a decrease of
the backscatter signal, depending on the polarization mode. Optimal polarization in burned area
detection was found to be the HV cross-polarized conguration (Transmit horizontal, receive
vertical) [Keiko et al., 2009], since exhibits good sensitivity to biomass, and being least affected
by forest types and ground conditions [Le Toan et al., 2004]. The incidence observation angle
of the radar beam will also affect on the proportion of the backscatter signal, as different angles
result in exposure of different branch structures and orientations. Therefore, the overall pattern
of the post-re canopy structure, and the optimal SAR system parameters, will determine the
interpretation of the associated re effects. For Mediterranean pine forest, Tanase reported good
agreements for L-band and HV polarization to burn severity assessment. In this work, we intend
to demonstrate the ability of ALOS/PALSAR data to estimate the effect of re over different
Mediterranean vegetation types and land covers, and to develop a methodology for burned area
estimation from SAR data, validated by using traditional optical burned area estimation methods
and ground truth data.
Material and Methods
Study Area
The Island of Sardinia is characterized by Mediterranean climate with hot and dry summers,
and a rainy season concentrated during winter and spring seasons. The heterogeneity of
topographic forms comprises hilly regions and interior valleys. Along the coast and river
valleys, vegetation is dominated by sclerophyllous shrubs, thermo Mediterranean Quercus
ilex forests, and agricultural lands. Inland areas are characterized by forest stands combined
with pastures and shrublands. The principal forest formations include mesomediterranean
Quercus ilex and Quercus suber forests. At higher elevations the sclerophyllous oak forests
merge with broadleaved forests of Quercus congesta and Quercus ichnusae [Bajoco et
al., 2007]. In the Sardinia Island wild res are one of the most important environmental
controls affecting the ecosystems functioning and structure, being responsible for the actual
shape of landscapes and vegetation forms. Nevertheless, changes in the natural re regime
have led to a higher frequency and intensity of res, representing a factor of degradation,
with signicant impacts on forest stands. In this work we analyzed a re occurred the 23d
of July of 2009 in Sardinia island, in the locality of Pardu along the municipalities of Pau,
Villaurbana, Usellus and Villa Verde (lat: 39 49 35.3 N lon: 8 48 02.8 E). The re affected
a total of 2242 ha. as reported by the reports of the CFVA (Corpo Forestale e di Vigilanza

235
European J ournal of Remote Sensi ng - 2012, 45: 233-241
Ambientale) of the Regione Autonoma della Sardegna, from which 50% of the area was
wooded vegetation (Fig. 1).
Figure 1 - The island of Sardinia and the Study area with detail of the burned perimeter. ALOS
image represented in RGB: HH, HV, HH.
SAR data
The SAR data used in this study corresponds to the Advanced Land Observing Satellite
(ALOS) Phased Array Type L-band SAR (PALSAR) (Fig. 2). Images were acquired in the
pre and post re dates of the re event, the 22 of June 2009 and 22 of September 2009
respectively. The pre-re date is 31 days prior to the re, and the post re image was acquired
61 days after the event. ALOS images were calibrated by using NEST software* in order to
retrieve σ0 (dB), and then co-registered on a common reference (Geographic WGS-84).
SAR -Statistical Analysis
Mean dB values of pre and post re sites were extracted for the different vegetation types
affected by re. Statistical analysis was made separately for each of both HH and HV
polarization modes. Negative differences in dB between the pre and post re events for
each vegetation types were interpreted to be related to re effect (Student’s Test).
SAR-Burned Area estimation
The analysis was made using a subset of the entire image around the burned perimeter (The
ground truth perimeter was obtained from Regione Autonoma della Sardegna). We calculated
the difference between pre and post re imagery only for HV polarization mode (∆HV). The
resulting image was ltered with an “Enhanced Frost” adaptive lter with an 11x11 window size,
with the purpose of isolating positive values related to higher damage levels. Subsequent positive
values were selected as burned and transformed to a binary map (re /non re) (Fig. 2).
Optical data
We use an optical SPOT-4 image of the post re event (27/07/2009). A simple threshold
methodology was developed to retrieve the burned perimeter based on Normalized Difference
Vegetation Index (NDVI), Normalized Burn Ratio (NBR) and Near Infrared (NIR) band 3
as main inputs (Fig. 3). Optimal thresholds were obtained from visual interpretation for the
3 inputs independently, avoiding commission errors as much as possible.

Mari et al. Fire Damage as sessment in Sardinia i sland with SAR data.
236
Figure 2 - ALOS/PALSAR data, Pre-re event at top left, Post-re event, top right, Difference in
HV polarization, down left, and the Burned area estimated from positive threshold values from the
∆HV SAR data.
Complementary data
Burned vegetation types were recognized with the Corine Land Cover 2006 map**.
Data sample extraction was assessed with an Ortofoto image mosaic obtained on line
by a WMS server. Four Vegetation types and land covers were identified to be affected
by the Pardu fire: Coniferous forest, Natural Grasslands, Sclerophyllous vegetation
and Agricultural areas.
Results and Conclusions
HH Polarization exhibits a similar and continuous effect of increased backscatter over the
different vegetation types for the post re event (Fig. 4b). It has been proved that the effect
of the increased backscattering response for the post re event with HH polarization is
attributed to local dielectric properties of the scatterers, with higher soil moisture content
[Tanase et al., 2010]. Nevertheless, for the HV polarization it is observed mayor variability
in the response of backscatter signals for the different vegetation types, with a more evident
reduction for the woody vegetation types (Fig. 4a).
Assuming dB data is normally distributed; sclerophyllous vegetation presented the greater
reduction in mean dB values (Mean reduction= 3.4dB, n=1412, p< 0.001), suggesting a
more structural damage in comparison to the other vegetation types.

237
European J ournal of Remote Sensi ng - 2012, 45: 233-241
Figure 3 - SPOT4 Optical data, Normalized Burn Ratio (NBR) top left, Normalized Difference
Vegetation Index (NDVI) top right, Near Infrared (NIR) down left, and the Burned area estimated
from thresholds of the indices and spectral NIR data.
Figure 4a - Interactions between pre and post re SAR data with HV cross polarized mode.

Mari et al. Fire Damage as sessment in Sardinia i sland with SAR data.
238
Figure 4b - Interactions between pre and post re SAR data with HH polarized mode.
Coniferous forest also showed a signicant reduction in mean dB values (Mean reduction=
1.2dB, n=951, P<0.001), but with a less mean difference in comparison to Sclerophyllous
vegetation (Tab. 1).
Table 1 - Statistical Analysis for the pre and post re events (Student’s Test), considering the effect
of different polarizations over diverse vegetation types.
Polarization
HH HV
Post-re Pre-re Post-re Pre-re
Vegetation
Types Mean dB S.D Mean dB S.D Mean dB S.D Mean dB S.D
Agricoltural
Areas -13.66897 3.032 -15.131 1.909 *** -21.267 2.996 -21.882 3.419
Coniferous
Forest -1081376 1.854 -11.465 2.206 -17.858 2.412 -16.619 2.616 ***
Natural
Grasslands -14.64136 3.279 -15.235 3.173 *** -22.179 2.752 -22.276 3.031
Sclerophylous
Vegetation 12.81717 1.740 -13.098 1.724 *** -21.084 1.857 -17.687 1.883 ***
Natural Grasslands were insensitive to the signal of HV polarization, meaning that there
is probably no utility in using HV polarization mode to verify re effects on herbaceous
types for these typical Mediterranean regions. Nevertheless it either could be also possible,
that grasslands recovery is faster after 60 days, and structural damage is almost repaired.
Agricultural Areas presented a similar increasing behavior as seen with the HH polarization

239
European J ournal of Remote Sensi ng - 2012, 45: 233-241
mode, remaining less clear what are the principal interactions involved. The burned area
estimation methodology was based on the fact that HV polarization resulted to be more
sensitive to re effects in wooded vegetation. The difference in HV (Pre-Post) evidenced a
spatial correlation with the wooded vegetation types, with higher dark dB values related to
structural damage (Fig. 3). The application of the enhanced frost adaptive lter was done
in a trial and error basis, reaching the 11x11 window which showed the better separability
between burned/unburned values (Fig. 3). Results show an agreement between 70 and 80%
of the burned area for Coniferous and Sclerophyllous vegetation as compared to the ground
truth perimeter. The Agricultural and Grassland areas presented a low agreement between
25-35% (Fig. 5). Other studies in tropical forest showed similar results for Agricultural
mixed areas [Hoekman, 1999]. Omission errors were evaluated for each vegetation type,
according to their relative proportion inside the ground truth perimeter: Agricultural
Areas had the higher omission error (0.74) followed by Coniferous Forest (0.65). Natural
Grasslands (0.27) and Sclerophyllous types (0.21). There were no commission errors
according to the information of the ground truth perimeter. It is conclusive that the shrubby
vegetation type can be more exposed to re ames, with the consequent total destruction
of the structural components of the canopy, more suitable to be detected with SAR data
(See example in Figure 6). The less reduction of mean dB values for Coniferous forest in
comparison with the sclerophyllous types may be related to the height of the trees, which
are in less contact with re ames. These results indicate the possibility of using ALOS/
PALSAR L-HV imagery for re Damage Assessment in forested Mediterranean Regions,
especially under challenging weather conditions. A mixed optical-SAR data methodology
could be explored for better discriminations of all types of vegetation structures.
Figure 5 - Total agreement for the different vegetation types from optical reectance data and SAR
data.

Mari et al. Fire Damage as sessment in Sardinia i sland with SAR data.
240
Figure 6 - Burned Vegetation Types from Monti Arci area.
Acknowledgements
This work has been nanced by Comision Nacional de Actividades Espaciales (CONAE)
and the Agenzia Spaciale Italiana (ASI) in a convening with Centro di Riccerca Projetto
San Marco (CEPSM). The PALSAR data and Spot data were obtained as a courtesy of the
European Space agency (ESA). The on-ground pictures have been taken by Dario Secci.
References
Bajocco S., Ricotta C. (2007) - Evidence of selective burning in Sardinia (Italy): which
land-cover classes do wildres prefer? Landscape Ecol (2008) 23: 241-248. doi: http://
dx.doi.org/10.1007/s10980-007-9176-5.
Bourgeau-Chavez, L. L., Harrell, P. A., Kasischke, E. S., French, N. H. F. (1997) - The
detection and mapping of Alaskan wildres using a spaceborne imaging radar system.
International Journal of Remote Sensing, 18: 355-373. doi: http://dx.doi.org/10.1080/0
14311697219114.
Cahoon D.R., Stocks B.J., Levine J.S., Cofer W.R., Pierson J.M. (1994) - Satellite Analysis
of the Severe 1987 Forest-Fires in Northern China and Southeastern Siberia. J. Geophys.
Res.-Atmos., 99(D9): 18627-18638. doi: http://dx.doi.org/10.1029/94JD01024.
Hoekman D. (1999) - Monitoring tropical forests using Synthetic Aperture Radar. Invited

241
European J ournal of Remote Sensi ng - 2012, 45: 233-241
paper presented at the International MOFEC-Tropenbos and NWO Workshop: The
balance between biodiversity conservation and sustainable use of tropical rain forests,
Balikpapan, Indonesia Dec. 6-8 1999.
Kasischke E.S., Melack J.M., Dobson M.C. (1997) - The use of imaging radars for
ecological applications - a review. Remote Sensing of Environment 59: 141-156. doi:
http://dx.doi.org/10.1016/S0034-4257(96)00148-4.
Kasischke E., French N., Harrell P., Christensen Jr. N., Ustin S., Barry D. (1993) - Monitoring
of wildres in Boreal Forests using large area AVHRR NDVI composite image data.
Remote Sensing of Environment, 45 (1): 61-71, ISSN 0034-4257. doi: http://dx.doi.
org/10.1016/0034-4257(93)90082-9.
Keiko Masanobu S., Osamu I., Kazuo I. (2009) - Detecting an area affected by forest res
using ALOS PALSAR. 3rd ALOS Joint PI Symposium, Nov. 2009, Kona, Hawaii.
Laneve G., Castronuovo M. M., Cadau E.G. (2006) - Continuous Monitoring of Forest
Fires in Mediterranean Area Using MSG. Transactions on Geoscience and Remote
Sensing, IEEE, Oct. 2006. 44:(10) Part 1.
Laneve G., Cadau E., Ferrucci F., Rongo R., Guarino A., Fortunato G., Him B., Di Bartola C.,
Iavarone L., Loizzo R. (2012) - SIGRI - an Integrated System for Detecting, Monitoring,
Characterizing Forest Fires and Assessing damage by LEO-GEO Data. Italian Journal
of Remote Sensing, 44: 19 - 25. doi: http://dx.doi.org/10.5721/ItJRS20124412.
Le Toan T., Beaudoin A., Riom J., Guyon D. (1992) - Relating forest biomass to SAR data.
Geoscience and Remote Sensing, IEEE Transactions on, Mar 1992. 30 (2): 403-411 doi:
http://dx.doi.org/10.1109/36.134089.
Le Toan (2004) - Relating radar remote sensing of biomass to modeling of forest carbon budgets.
Climatic Change 67: 379-402. doi: http://dx.doi.org/10.1007/s10584-004-3155-5.
Tanase M., Santoro M., de la Riva J., Pérez-Cabello F., Le Toan T. (2010) - Sensitivity of X-,
C-, and L-Band SAR Backscatter to Burn Severity in Mediterranean Pine Forests. IEEE
Transactions of Geoscience and Remote Sensing. 48(10). doi: http://dx.doi.org/10.1109/
TGRS.2010.2049653.
Received 03/10/2011, accepted 16/03/2012