Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

10
Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain Mariano Garcı ´a * , Emilio Chuvieco Department of Geography, University of Alcala ´, Calle Colegios 2, 28801 Alcala ´ de Henares, Spain Received 31 July 2003; received in revised form 27 April 2004; accepted 29 April 2004 Abstract Early and detailed information regarding the location and extension of areas affected by forest fires is a critical issue for assessing their effects at several scales. Remote sensing is a valuable tool for burned area mapping, providing spatially explicit information on the scorched areas, even for remote regions. Various sensors have been used for burned land mapping in recent years, covering local (10 – 30 m pixel size) and global scales (500– 1000 m pixel size). Regional inventories require higher detail than global studies, but typically cover larger territories than local studies. For this purpose, the use of medium spatial resolution sensors (100 – 300 m pixel size) is highly advisable. Few studies have evaluated these sensors, and none have specifically assessed the performance of images acquired by the Argentinian Satellite for Scientific Applications-C/ Multispectral Medium Resolution Scanner (SAC-C/MMRS) for mapping burned areas, which is the main objective in this paper. Since information on calibration parameters was not available, raw data was converted into ground reflectance values after a ‘‘cross- calibration’’ method, using Landsat Enhanced Thematic Mapper (ETM+) data as reference. Estimated radiance values showed very high correlation with those for the ETM+ (r 2 >0.9). Several spectral indices were generated and tested. The Burned Area Index (BAI), derived from red and near infrared (NIR) reflectances showed the greatest sensitivity to discriminate burned areas from other land cover types, although the Normalized Difference Infrared Index (NDII), based on the Red and Short Wave Infrared (SWIR) bands also provided good discrimination capability. Image segmentation was performed by using a multi-threshold approach, based on the BAI and the NDII since these two spectral indices have the highest sensitivity to (1) discriminate burned areas and (2) to avoid confusion with water and cloud shadows. Burned perimeters delineated using a seeded region growing algorithm showed good agreement with the perimeters digitised on Landsat ETM+ data, providing accurate estimation of the burned areas compared to statistics provided by the Spanish forest fire authorities. In the first study area (Almerı ´a, south-eastern Spain), the perimeters showed a 90% agreement. In the second study site (Madrid), two fires were studied. The first one (Cadalso, south – west of Madrid) showed an agreement of 64%, and for the second one (Patones, north – east of Madrid), the agreement was 84%. The lower accuracy of the former was caused by the great spatial discontinuity of that fire, which includes many patches of unaffected vegetation within the scar perimeter. D 2004 Elsevier Inc. All rights reserved. Keywords: Burned area mapping; SAC-C/MMRS; Spectral indices 1. Introduction Despite the high incidence of fires in the Mediterra- nean basin, burned area statistics are heterogeneous, as a consequence of the different methodologies used in each country (Ve ´lez, 2000). Additionally, an adequate cartog- raphy of burned areas is not available in most countries, which makes accurate estimations of affected areas and post-fire management practices more cumbersome. Traditional methods of generating fire statistics were based on field surveys, which are costly and inaccurate for large fires. Large fires (over 100 ha) are nowadays mapped in most of Spain with Global Positioning System (GPS) techniques, normally from helicopters. This tech- nique greatly improves spatial analysis tasks in the areas affected, since a detailed perimeter is digitized. However, it does not account for small unburned patches within the fire perimeter; nor does it take into account fire severity levels. 0034-4257/$ - see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2004.04.011 * Corresponding author. Tel.: +34-9-1885-4482; fax: +34-9-1885- 4439. E-mail addresses: [email protected] (M. Garcı ´a), [email protected] (E. Chuvieco). www.elsevier.com/locate/rse Remote Sensing of Environment 92 (2004) 414 – 423

Transcript of Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Page 1: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

www.elsevier.com/locate/rse

Remote Sensing of Environment 92 (2004) 414–423

Assessment of the potential of SAC-C/MMRS imagery for mapping

burned areas in Spain

Mariano Garcıa*, Emilio Chuvieco

Department of Geography, University of Alcala, Calle Colegios 2, 28801 Alcala de Henares, Spain

Received 31 July 2003; received in revised form 27 April 2004; accepted 29 April 2004

Abstract

Early and detailed information regarding the location and extension of areas affected by forest fires is a critical issue for assessing their

effects at several scales. Remote sensing is a valuable tool for burned area mapping, providing spatially explicit information on the scorched

areas, even for remote regions.

Various sensors have been used for burned land mapping in recent years, covering local (10–30 m pixel size) and global scales (500–

1000 m pixel size). Regional inventories require higher detail than global studies, but typically cover larger territories than local studies. For

this purpose, the use of medium spatial resolution sensors (100–300 m pixel size) is highly advisable. Few studies have evaluated these

sensors, and none have specifically assessed the performance of images acquired by the Argentinian Satellite for Scientific Applications-C/

Multispectral Medium Resolution Scanner (SAC-C/MMRS) for mapping burned areas, which is the main objective in this paper.

Since information on calibration parameters was not available, raw data was converted into ground reflectance values after a ‘‘cross-

calibration’’ method, using Landsat Enhanced Thematic Mapper (ETM+) data as reference. Estimated radiance values showed very high

correlation with those for the ETM+ (r2>0.9). Several spectral indices were generated and tested. The Burned Area Index (BAI), derived from

red and near infrared (NIR) reflectances showed the greatest sensitivity to discriminate burned areas from other land cover types, although the

Normalized Difference Infrared Index (NDII), based on the Red and Short Wave Infrared (SWIR) bands also provided good discrimination

capability.

Image segmentation was performed by using a multi-threshold approach, based on the BAI and the NDII since these two spectral indices

have the highest sensitivity to (1) discriminate burned areas and (2) to avoid confusion with water and cloud shadows.

Burned perimeters delineated using a seeded region growing algorithm showed good agreement with the perimeters digitised on Landsat

ETM+ data, providing accurate estimation of the burned areas compared to statistics provided by the Spanish forest fire authorities. In the

first study area (Almerıa, south-eastern Spain), the perimeters showed a 90% agreement. In the second study site (Madrid), two fires were

studied. The first one (Cadalso, south–west of Madrid) showed an agreement of 64%, and for the second one (Patones, north–east of

Madrid), the agreement was 84%. The lower accuracy of the former was caused by the great spatial discontinuity of that fire, which includes

many patches of unaffected vegetation within the scar perimeter.

D 2004 Elsevier Inc. All rights reserved.

Keywords: Burned area mapping; SAC-C/MMRS; Spectral indices

1. Introduction

Despite the high incidence of fires in the Mediterra-

nean basin, burned area statistics are heterogeneous, as a

consequence of the different methodologies used in each

country (Velez, 2000). Additionally, an adequate cartog-

raphy of burned areas is not available in most countries,

0034-4257/$ - see front matter D 2004 Elsevier Inc. All rights reserved.

doi:10.1016/j.rse.2004.04.011

* Corresponding author. Tel.: +34-9-1885-4482; fax: +34-9-1885-

4439.

E-mail addresses: [email protected] (M. Garcıa),

[email protected] (E. Chuvieco).

which makes accurate estimations of affected areas and

post-fire management practices more cumbersome.

Traditional methods of generating fire statistics were

based on field surveys, which are costly and inaccurate for

large fires. Large fires (over 100 ha) are nowadays

mapped in most of Spain with Global Positioning System

(GPS) techniques, normally from helicopters. This tech-

nique greatly improves spatial analysis tasks in the areas

affected, since a detailed perimeter is digitized. However,

it does not account for small unburned patches within the

fire perimeter; nor does it take into account fire severity

levels.

Page 2: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Table 1

SAC-C/MMRS sensor technical specifications

Characteristics Value

SAC-C launch 21st November 2000

SAC-C orbit 705 km altitude, sun-synchronous

SAC-C orbit inclination (j) 98.21j from the equator

SAC-C equator crossing

time (local time)

10:15 a.m. descending node

SAC-C repeating cycle 16 days (7–9 days for the

MMRS sensor)

MMRS spatial resolution 175 m

MMRS swath 360 km

MMRS spectral bands blue 480–500 nm

green 540–560 nm

red 630–690 nm

NIR 795–835 nm

SWIR 1550–1700 nm

Source: http://www.conae.gov.ar/sac-c/misionsacc.html.

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423 415

Satellite remote sensing is a good alternative to the use of

GPS, since it provides spatially explicit, timely evaluations

of fire affected areas. The use of remote sensing images for

burned land mapping has greatly increased in recent years

and covers a wide range of sensors and techniques (Koutsias

et al., 1999; Pereira et al., 1997). Most of these attempts

have relied on low resolution sensors such as the National

Oceanic and Atmospheric Administration/Advanced Very

High Resolution Radiometer (NOAA/AVHRR) (Fraser et

al., 2000a; Martın & Chuvieco, 1993; Pereira, 1999), SPOT-

VEGETATION (Fraser et al., 2000b; Gregoire et al., 2003;

Stroppiana et al., 2002), the Along Track Scanning Radi-

ometer (ATSR) data (Eva & Lambin, 1998) and more

recently, the Moderate Resolution Imaging Spectroradiom-

eter (MODIS) imagery (Barbosa et al., 2001; Roy et al.,

2002). Within high resolution sensors, Landsat Thematic

Mapper (TM) data has been the most widely used (Koutsias

& Karteris, 2000; Koutsias et al., 2000; Miller & Yool,

2002). However, medium spatial resolution sensors (100–

300 m) have seldom been used (Salvador et al., 2000;

Vazquez et al., 2001). An exception of this comment is

the use of the Indian Remote Sensing Satellite’s Wide Field

of View Sensor (IRS-WiFS) for mapping burned areas in

Europe by the Joint Research Centre (http://natural-hazards.

jrc.it/fires/).

Despite the advantages of using remote sensed images

for burned area mapping mentioned above, several con-

straints should be considered such as the temporal resolution

of the sensor (related to the persistency of the carbon signal

or cloud coverage), and the spatial resolution, which estab-

lishes the size of the burns that may be detected. Finally, the

spectral resolution of the sensor, is also a critical factor to

detect burned areas accurately and avoid spectral confusion

with other land cover types presenting a similar spectral

response.

1.1. The SAC-C/MMRS sensor

The SAC-C satellite is the first Argentinian Earth Ob-

servation satellite. It is part of the so-called ‘‘morning

constellation’’, comprising Landsat 7, Earth Observing-1

(EO-1), TERRA and SAC-C platforms. The four satellites

have the same orbital parameters but with descending

equatorial crossing times at 10:00 a.m., 10:01 a.m., 10:30

a.m. and 10:15 a.m. (GMT), respectively. The SAC-C

satellite has three cameras on board: The Multispectral

Medium Resolution Scanner (MMRS), the High Resolution

Technological Camera (HRTC) and the High Sensitivity

Technological Camera (HSTC) (www.conae.gov.ar). The

MMRS sensor is a push-broom scanner with a spatial

resolution of 175 m and a swath width of 360 km. As for

the spectral characteristics of the sensor it has five bands,

ranging from the blue to the short wave infrared (SWIR)

(Table 1 provides more detailed information on the MMRS

sensor). The HRTC is a ‘‘push-broom’’ panchromatic sensor

(400–900 nm), with a spatial resolution of 35 m and a swath

of 90 km. The HSTC has a spatial resolution of 300 m, the

swath is 700 km and a spectral coverage between 450 and

850 nm. This camera operates during the night overpass

(22:30 local time).

Considering its medium spatial resolution and good

spectral coverage, the MMRS sensor was assumed to be a

good alternative to validate burned land maps generated

from MODIS or AVHRR images, and to improve the spatial

coverage of higher resolution sensors, such as Landsat-

ETM+ or SPOT-HRV. Although it provides lower temporal

resolution than IRS-WiFS (with a similar spatial resolution,

180 m), the temporal resolution of the SAC-C/MMRS (one

image every 7–9 days) can be considered good enough for

burned area mapping in Spain since fire effects are expected

to remain longer. The main advantage of SAC-C/MMRS

data over IRS-WiFS for mapping burned areas relies on the

SWIR region, not available in WiFS images, which has

proved to be very useful for the purpose of this study

(Chuvieco & Congalton, 1998; Stroppiana et al., 2002).

1.2. Objectives

The main objective of this study is to assess the potential

of the SAC-C/MMRS images for burned area mapping in

medium size countries, such as Spain. The applications of

this sensor are still limited. Initially, it was designed for the

study of terrestrial and marine ecosystems (land use, defor-

estation, coastal studies. http://www.conae.gov.ar/satelites/

sac-c/mmrs.html). Its potential for burned land mapping is

still to be explored, and form the basic objective of this

paper. The specific goals of this research were the following:

� to carry out the radiometric and geometric calibration of

the MMRS images;� to identify the sensitivity of the MMRS bands and

derived vegetation indices to discriminate burned areas

from other land covers;� to assess the performance of MMRS images for mapping

burned areas and providing burned are statistics.

Page 3: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423416

2. Methods

2.1. Study sites

Two regions were selected within Spain to carry out this

analysis (Fig. 1): Almerıa (located at the south–east of the

country) and Madrid. Within this second area, two areas

were considered, Cadalso (south–west of Madrid) and

Patones (north–east of Madrid). Both areas had been

affected by large fires during the summer of 2002. Almerıa

suffered a large shrub fire in early June, whereas Madrid had

two large fires in August, affecting both shrub–land and

woodland. These two areas present very diverse vegetation

and soil characteristics. Almerıa is located in the most arid

and dry fringe of Spain (mean rainfall < 250 mm/year), with

mild winter temperatures and high temperatures during the

summer (mean yearly temperature 20 jC). Therefore, veg-etation is sparse in most of the region, with a predominance

of sclerophylous shrubs. Soils are generally dark, with some

volcanic bedrocks near the coast, while limestone is more

common further inland.

In contrast, Madrid region has a higher rainfall (over 500

mm) and lower temperatures than in Almerıa, with frequent

freezing conditions. Vegetation consists mainly on conifer-

ous species (Pinus sylvestris, Pinaster pinea), as well as

evergreen oaks (Quercus ilex), although some deciduous

species (Quercus pyrenaica, Castanea sativa) also grow on

mountain slopes between 1000 and 2000 m. The central

Plateau is mainly cropland, with some corridors of decidu-

ous (Populus, Fraxinus) along the river banks. Shrubs are

mainly sclerophylous species, such as Retama spherocarpa,

Cistus ladanifer and Rosmarinus officinalis. Soils are main-

Fig. 1. Overview of the two study areas on a SAC-C imagery mosaic

ly light to medium in colour with a predominance of sands

from granite origin.

2.2. Image preprocessing

The MMRS images were received from the Argentinian

Space Institute (CONAE)’s receiving station located in

Cordoba. After analyzing the image archive, 2 images were

selected for each study site. One corresponded to a date

before the fire occurrence (2/06/2002 for Almerıa and 18/

07/2002 for Madrid) and another one corresponded to the

first date available after the fire occurrence (27/06/2002 and

21/08/2002 for Almerıa and Madrid, respectively).

The images were received with orbital control points,

which were used for the georeferencing process. However,

this correction yielded a RMS error of approximately 1 km,

which is not appropriate for an accurate inventory of burned

areas considering the spatial resolution of the sensor. There-

fore, a further processing was applied to improve the

georeferencing and multitemporal matching. The registra-

tion of the pre-fire images was improved by using control

points extracted from a vector layer of rivers, lakes and

reservoirs, whereas post-fire images were corrected against

the pre-fire ones. In doing so, a good matching between the

images used in the subsequent multitemporal comparison

was assured. In all cases second degree polynomial equa-

tions were applied, with an RMS error under 0.5 pixels, and

the nearest neighbour re-sampling method was used to

preserve the original image values. Next, subsets of the

original images were extracted to isolate the two study areas

(Fig. 2). In the Almerıa study site a window of 688� 761

pixels was extracted, whereas in the Madrid study site, an

of Spain. The administrative regions of Spain are superimposed.

Page 4: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Fig. 2. Subset of post-fire images over the selected study sites: (a) Almerıa, (b) Madrid.

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423 417

area of 754� 897 pixels was subset, which covered the two

fires in the region.

After performing the geometric correction the raw data

was converted to ground reflectance values, thus correcting

the data for sensor-related distortions (gains and offset),

solar irradiance, solar zenith angle and atmospheric effects.

Since the calibration parameters required to convert raw

digital values to radiance were not available, a ‘‘cross

calibration’’ method was performed, using a Landsat

ETM+ image (path/row: 201/033) as a reference for this

analysis. This choice was based on very close acquisition

time of both sensors, as well as the similar spectral cover-

age. Since the ETM+ image available was acquired on the

9th June 2002, a fifth SAC-C/MMRS of the same date and

covering the same area (central part of Spain) was used to

derive the calibrations parameters. This ensured almost

identical atmospheric conditions, as well as the same land

covers present in the scene. The calibration parameters used

to convert digital counts to radiance were obtained from

linear regression analysis, based on clearly identifiable

features (water, green vegetation and bare soil) in both

images. Due to the different spatial resolution of both

sensors Landsat ETM+ (30 m) and SAC-C/MMRS (175

m), the mean value within a 6� 6 window was extracted

from the ETM+ image, for each sample point on the SAC-

C/MMRS image. This analysis provided the calibration

parameters required to convert the DNs of the SAC-C/

MMRS image into radiance values by applying the formula:

Lk ¼ ao;k þ al;k*DNk

where Lk is the radiance in band k, ao,k and al,k are the

estimated calibration parameters, and DNk is the raw Digital

Number for band k.

Atmospheric correction was based on the improved

Chavez’s dark object approach (Chavez, 1996), since no

information on atmospheric profiles was available for the

study regions. Although this method constitutes an approx-

imation to ground reflectance calculation, and it does not

fully account for the absorption effects caused by aerosols,

the results obtained using this method were comparable to

those obtained by models using in-situ atmospheric infor-

mation (Chavez, 1996). The reflectance was then calculated

as:

qk ¼Ke�s*p*ðLk � Lk;mÞ

E0;k*cosh*sk

Where Ke– s is a factor that takes into account the variation

of the Earth–Sun distance throughout the year; Lk is the

radiance in band k, Lk,m is the minimum radiance in band k

(dark object, used for atmospheric dispersion); E0,k is the

irradiance value for band k; sk is the atmospheric trans-

mittance for band k, and h is the sun zenith angle, which

was computed from latitude, longitude and acquisition

time extracted from the image. Transmittance values used

were those suggested by Chavez (1996), based on the

same spectral wavelengths of MMRS and TM/ETM+

sensors.

2.3. Spectral indices

The use of vegetation indices for burned area mapping is

widely used due to the important spectral changes caused by

fire on vegetation covers. Traditional vegetation indices

such as the Normalized Difference Vegetation Index

(NDVI) show a severe decrease after fire occurrence, as a

result of the deterioration of leaf pigments and leaf structure

Page 5: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Table 2

Raw to radiance calibration coefficients and correlation between ETM and

SAC-C/MMRS radiance values

Band a0 a1 r2

1 (Blue) 0.166 0.586 0.95

2 (Green) � 22.68 1.456 0.96

3 (Red) � 12.239 0.967 0.98

4 (NIR) � 20.768 1.063 0.99

5 (SWIR) � 4.611 0.148 0.99

M. Garcıa, E. Chuvieco / Remote Sensing418

(Kasischke et al., 1993). However, since those vegetation

indices were designed for vigorous vegetation, they may not

be appropriate for burned land discrimination, because the

reflectance ranges of burned areas differ significantly from

that of photosynthetic vegetation (Pereira, 1999). Accord-

ingly, some authors have proposed indices specifically

designed for burned area discrimination (Chuvieco et al.,

2002; Martın, 1998; Patterson & Yool, 1998; Pereira, 1999;

Trigg & Flasse, 2000).

For this study the following indices were generated to

test their sensitivity for burned land discrimination:

� NDVI: since it has been extensively used for burned land

discrimination.� NDII: defined as:

NDII ¼ qNIR � qSWIR

qNIR þ qSWIR

developed by Hunt and Rock (1989). An analogous

index, the Short Wave Vegetation Index (SWVI) was

applied by Fraser et al. (2000b) to SPOT-VEGETATION

data, and to Landsat-TM images by Lopez Garcıa and

Caselles (1991), based on TM/ETM bands 4 and 7. Key

and Benson (1999) named this index as Normalized Burn

Ratio (NBR).� GEMI (Global Environmental Monitoring Index): devel-

oped by Pinty and Verstraete (1992). This non-linear

index was designed to minimize atmospheric and soil

background effects. This index is defined as:

GEMI ¼ gð1� 0:25gÞ � ðqRED � 0:125Þð1� qREDÞ

where

g ¼ ð2ðq2NIR � q2

REDÞ þ 1:5qNIR þ 0:5qREDÞðqNIR þ qRED þ 0:5Þ

This index was selected since it was found to be very

good for burned area mapping (Pereira, 1999).� BAI (Martın, 1998): defined as:

BAI ¼ 1

ðPcRED � qREDÞ2 þ ðPcNIR � qNIRÞ2

where PcRED and PcNIR are the convergence points of the

Red and Near Infrared bands. The values corresponding

to the convergence points were defined as 0.1 and 0.06,

based on literature and empirical analysis (Martın, 1998).

This index was specifically designed to discriminate

burned areas by enhancing the charcoal signal of the

affected areas. The value of the index increases as the

distance between a pixel and the convergence point of a

given band decreases (Martın, 1998). This index was

selected given its good performance in Mediterranean

environments (Chuvieco et al., 2002).

2.4. Spectral separability

In order to identify the most suitable bands and index

for our purpose, their discrimination ability for burned area

mapping was tested. Representative samples of points,

corresponding to burned and unburned areas, over the

post-fire images were collected. Unburned areas were

grouped into six general classes based on the Corine Land

Cover map and present in both study areas which include:

agriculture, water bodies, woodland, shrubland, open

spaces and artificial surfaces. Two additional covers,

clouds and cloud shadows, were included in the analysis

since they were present in the images. For the Almerıa

study area, 100 sample points were considered for the

burned class and over 780 sample points for the unburned

class (around 100 points for each category included in the

unburned class). For the Madrid study area the same

criterion was employed, but in this case 200 sample points

were collected for both burned areas (100 sample points

per area) and over 880 unburned sample points. A simple

statistic measurement of the separability was then carried

out applying the index:

M ¼ Alu � lbA=ðru þ rbÞ

(Swain & Davis, 1978) where lu is the mean value of the

unburned areas, lb is the mean value of the burned areas,

ru is the standard deviation of the unburned areas and rb

is the standard deviation of the burned areas. The higher

the M value the higher is the separability of the covers

considered. In general terms M>1 indicates good separa-

bility between the burned and the unburned categories, and

M < 1 indicates poor separability.

After identifying the band and index with the highest

separability, the simple difference between the post-fire

and the pre-fire images was calculated. Multitemporal

analysis of burned areas has been shown as more effective

than single post-fire data, since the former minimizes the

possible spectral confusion with other cover types that

have a seasonal trend (Pereira et al., 1997). A simple

difference was carried out to compare pre-fire and post-fire

images. This method has showed good results in previous

studies (Kasischke et al., 1993; Martın & Chuvieco, 1995)

and is more suitable than the normalized difference or

PCA for burned areas estimation.

of Environment 92 (2004) 414–423

Page 6: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Table 3

Separability values (M) for Almerıa and Madrid (Patones and Cadalso fires combined) for the post-fire image

Cover Almerıa (post-fire) Madrid (post-fire)

B1 B2 B3 B4 B5 NDVI NDII GEMI BAI B1 B2 B3 B4 B5 NDVI NDII GEMI BAI

Artificial surfaces 3.012 3.069 3.898 4.303 3.048 0.645 1.564 0.146 3.174 2.415 2.486 3.006 2.204 1.115 1.111 1.466 0.133 1.574

Agriculture 2.182 2.542 2.943 3.566 2.227 1.825 1.539 0.656 3.154 1.957 2.519 3.777 3.320 2.737 1.396 0.834 0.317 1.822

Woodland 2.038 0.501 0.851 4.536 0.369 3.969 3.291 4.444 2.745 0.102 0.750 0.445 2.563 0.480 1.481 1.523 2.422 1.572

Shrubland 0.107 1.651 2.298 3.608 1.249 2.395 3.029 2.549 2.877 0.367 0.863 0.699 1.741 0.827 0.835 0.821 1.578 1.397

Open spaces 2.232 3.027 3.330 4.375 1.633 1.612 1.030 1.824 2.822 1.005 1.418 1.181 1.622 0.992 0.599 0.512 1.494 1.329

Water 1.717 1.876 1.353 0.561 2.327 2.001 3.170 1.483 1.410 2.203 1.330 2.128 1.483 2.584 0.800 3.085 1.265 0.522

Clouds 6.033 2.464 2.962 2.608 4.354 1.884 2.489 1.287 3.405 13.668 6.241 8.466 4.437 4.349 1.215 3.552 3.706 1.943

Cloud shadows 0.434 0.016 0.106 1.796 0.906 3.044 3.846 3.305 1.673 3.058 2.618 2.613 0.798 2.134 2.498 3.418 0.285 0.250

Average m value 2.219 1.893 2.218 3.169 2.014 2.172 2.495 1.962 2.658 3.097 2.278 2.789 2.271 1.907 1.242 1.901 1.400 1.301

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423 419

2.5. Techniques for burned area mapping

Once the images were corrected, processed and the

spectral separability was tested, the next step was to map

the burned surfaces. This was carried out in two phases:

firstly discriminating burned areas, and secondly mapping

individual fires (Chuvieco et al., 2002). In phase one, the

aim was to discriminate ‘‘core’’ burned pixels, which

would afterwards identify each individual fire scar. There-

fore, the goal was to identify the pixels that were more

severely burned in each fire, at least three pixels per fire,

while avoiding confusion with any unburned areas. Fol-

lowing traditional accuracy assessment terminology, this

phase was designed to avoid commission errors (i.e.

unburned pixels labeled as burned), while leaving room

for omission errors (burned pixels not classified as such),

which would occur in those areas within a single fire that

were less severely affected. The image was then segmented

into two classes (burned/unburned), following a multi-band

and multi-temporal threshold approach, initially defined as

the mean minus one standard deviation of the most

sensitive spectral index for separating burned and un-

burned areas.

In the second phase, the ‘‘core’’ pixels previously iden-

tified as burned were used as seeds for a region growing

algorithm. Region growing algorithms are based on the

contextual relationships found in the vicinity of pixels

marked as seeds. The inclusion of neighbouring pixels into

a region defined by a seed is based on spectral and spatial

Table 4

Separability values (M) for Almerıa and Madrid (Patones and Cadalso fires comb

Cover Almerıa (temporal difference)

B1 B2 B3 B4 B5 NDVI NDII GEMI B

Artificial surfaces 0.300 0.405 0.531 1.343 0.533 2.561 0.973 1.418 1

Agriculture 0.397 1.041 0.976 2.264 0.954 2.962 0.697 1.003 2

Woodland 1.125 1.641 1.639 3.409 2.122 0.849 0.558 3.492 2

Shrubland 0.730 1.167 1.546 2.825 1.734 1.526 0.900 2.138 2

Open spaces 0.213 0.857 0.866 2.422 1.134 2.506 1.275 2.793 2

Water 0.116 0.243 0.048 0.521 0.097 0.971 0.610 0.477 1

Clouds 4.630 2.307 2.594 2.253 3.225 0.452 1.106 1.533 2

Cloud shadows 0.569 0.459 0.611 0.889 0.394 0.241 1.094 0.794 1

Average m value 1.010 1.015 1.101 1.991 1.274 1.509 0.902 1.706 1

distances, following a homogeneity criterion (Adams &

Bischof, 1994; Lira Chavez, 2002). Thus, initial sub-regions

(i.e, seed pixels) grow spatially as long as the homogeneity

criterion is satisfied. This second phase was designed to

refine the boundary of the burned areas, or in other words, it

tended to reduce omission errors, (tolerated in the first

phase) by analyzing the spatial context of ‘‘core’’ pixels

within each fire scar. Consequently, this method would

improve the assessment of the total area affected by each

fire, starting from the core pixels. The seeded region

growing algorithm was carried out using a PCI Geomatics

utility (PCI, 2003). The algorithm was computed for the

images of the BAI index, since it presented the best

separability between burned areas and other land covers.

The reason for undertaking burned areas mapping in two

phases is related to the great potential variability of spectral

characteristics within a scorched area. The main sources of

this variability are related to the type of species affected, the

degree of severity and the length of the interval between fire

extinction and image acquisition (Pereira, 1999). If thresh-

olds are applied to such a variety of spectral conditions,

omission or commission errors will be likely to occur. By

applying the two-step approach, at least the effect of the

various degrees of severity is removed, since only fully

‘‘consumed’’ pixels are being identified in the first phase,

while the second phase aims to classify the lower severity

levels within areas already identified as being scorched.

Since the first phase determines the number of burned areas

that will be detected, a critical issue that arises in this two-

ined) for the temporal difference

Madrid (temporal difference)

AI B1 B2 B3 B4 B5 NDVI NDII GEMI BAI

.041 0.059 0.750 0.665 1.891 0.612 0.773 0.713 1.168 1.581

.417 0.274 0.670 0.610 1.772 0.391 0.948 0.794 0.609 1.560

.381 0.019 0.562 0.398 2.734 0.610 0.902 1.185 3.132 1.818

.446 0.086 0.626 0.489 2.324 0.656 0.774 0.763 2.491 1.686

.356 0.476 1.052 0.924 1.897 1.144 0.163 0.073 1.149 1.612

.236 1.0615 1.027 1.015 0.383 0.558 1.267 1.681 0.053 0.247

.607 6.932 5.805 6.858 4.450 3.610 1.313 0.628 4.199 1.717

.463 0.418 0.750 0.545 1.371 0.682 0.393 0.969 1.532 0.158

.993 1.166 1.405 1.438 2.103 1.033 0.817 0.851 1.791 1.297

Page 7: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Fig. 3. Seed pixels obtained for each study site using a multi-threshold approach: (a) Almerıa, (b) Madrid. (The two fires in the second study site, Cadalso and

Patones, are enlarged).

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423420

step approach is the adequate selection of the thresholds to

extract the seed pixels, especially when low severity burning

occurs, and must provide a good compromise between

commission and omission errors.

3. Results and discussion

Table 2 shows the calibration parameters obtained from

the regression analysis used to convert SAC-C/MMRS raw

data to radiance values. Radiance values estimated using the

calibration parameters obtained showed very high correla-

Fig. 4. Burned areas outlined for each study site using a seed region growing algori

and Patones, are enlarged).

tion with the radiance values corresponding to the Landsat

ETM+. Correlation coefficients for each band, based on 100

sample points, are also shown in Table 2.

The results of the separability analysis between burned

and unburned pixels are presented in Tables 3 and 4.

Considering the post-fire image, the NIR band and the

BAI index had the highest mean separability in the Almerıa

study area, while the Blue and Red bands had the best

discrimination capability for the Madrid site. As for multi-

temporal differences, BAI and NIR showed the highest

mean values in the Almerıa study area, whereas NIR and

GEMI presented the best separability for Madrid. Indepen-

thm: (a) Almerıa, (b) Madrid (the two fires in the second study site, Cadalso

Page 8: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

Table 5

Comparison of burned areas (in hectares) obtained from different sources

Burn site Seeded region

growing

Visual

analysis

DGCN

statistics

Almerıa 1678.3 1552.8 1937.9

410.4 280.2 381.8

91.9 72.4 N/A

Madrid (Cadalso) 1286.3 1486.4 1984.3

Madrid (Patones) 814.3 973.6 974.7

Total 4281.2 4365.4 5278.7

Fig. 5. Differences between the perimeter provided by the DGCN and the

one obtained from the SAC-C/MMRS sensor.

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423 421

dent analysis of the two burns occurred in the Madrid study

site showed that in one of them (Cadalso) the BAI index

presented a higher discrimination ability than the GEMI in

the post fire image (BAI = 2.355 and GEMI = 1.848), while

for the temporal difference the separability of theses indices

was almost the same (BAI = 1.63 and GEMI = 1.732). Re-

garding the other fire in Madrid (Patones), the BAI showed

the best separability results for the post-fire image

(BAI = 5.557 and GEMI = 2.595) as well as for the temporal

difference (BAI = 2.339 and GEMI = 1.976) (Garcıa, 2003).

Differences found between the Almerıa and Madrid study

sites may be caused by the different characteristics of

burned areas, both related to the date of occurrence (late

spring the former, and mid August the latter). The different

features of the two fires which occurred in the Madrid site

should also be taken into account to explain the differences

in the separability results, since the Patones area is domi-

nated by shrubs on dark soils while Cadalso has Pinus trees

on bright soils.

It is noteworthy when considering the mean M value that

the unburned category includes very different types of

covers, and therefore, the M values can hinder specific

discrimination problems such as cloud shadows, that could

be clarified by using SWIR data (M = 2.134 for the second

study area), as reported by other authors (Koutsias et al.,

2000). The unexpected high meanM value for the blue band

in the Madrid site was due to the presence of clouds, which

have a very high separability from burned area for this band.

More detailed analysis of Tables 3 and 4 revealed that for

the Almerıa study area the BAI was the only index with M

values higher than 1 for all covers. A similar situation is

observed in the Madrid study area, except for the water and

cloud shadows with poor separability in the BAI values. In

these specific covers, the NDII of the post-fire image showed

the highest discrimination capability for the two study areas.

A factor influencing the M values for the BAI is its high

variability compared to the other indices, since its values

greatly increase as the pixel value reaches the convergence

points. This can explain the lower performance for some of

the covers of the BAI compared to other indices, as for

instance woodland (GEMIpost-fire = 4.444, BAIpost-fire =

2.745; GEMIpost-fire = 2.422, BAIpost-fire = 1.572, for the

Almerıa and Madrid areas respectively). Although as men-

tioned, it showed a good separability (M>1) for most of

the covers.

After analysing discrimination problems with specific

land covers, the second step in this project was to identify

‘‘core’’ burned areas using multiple thresholds. The multi-

temporal difference in BAI values and the post-fire NDII

index were used for discrimination purposes. The BAI

clarified most confusion between burned areas with vege-

tation and bare soil covers, while the NDII reduced prob-

lems with water and cloud shadows. The small confusion

that still remained after using the multi-threshold approach

was solved by applying clumping and sieving techniques,

removing polygons smaller than 3 pixels (Fig. 3). As

expected, many omission errors were obtained in the burned

pixels identification phase, a drawback that was necessary to

assure nonburned areas were identified as such.

From the ‘‘core’’ pixels identified in the first phase, the

area burned in each single fire scar was outlined using the

region growing algorithm. This shape refinement showed

good agreement with the perimeters obtained from the ETM+

imagery (Fig. 4). Differences observed correspond to the

boundaries and can be attributed to the spatial resolution of

the sensor, since border pixels relate to areas where there is a

higher mixture of burned and unburned vegetation, and

therefore the signal of the burned area is not strong enough.

In the Cadalso fire the burned area presented a high hetero-

geneity showing a great number of unburned patches that

were successfully discriminated. Patones also presented

small unburned patches, but were not discriminated since

they were too small for the spatial resolution of the sensor.

Table 5 shows the burned areas derived using the seeded

region growing algorithm and the areas obtained from the

digitised polygons compared to official Spanish Forestry

Service (DGCN) statistics.

Assuming the Spanish Forestry Service statistics (gener-

ated from GPS helicopter survey) as ground truth, the

seeded region growing algorithm under-estimated the

burned area by approximately 10% in the Almerıa study

site, whereas for Madrid, it under-estimated the burned area

by 35% and 16% for Cadalso and Patones respectively. The

Patones under-estimation could be a consequence of the

presence of boundary pixels where there is a higher mixture

Page 9: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423422

of burned and unburned pixels, and therefore the signal of

the burnt area is comparatively weak. The large under-

estimation in Cadalso may be explained by the fact that

the perimeter provided by the DGCN does not include the

small unburned patches within the affected area, which were

successfully detected with the MMRS sensor (Fig. 5). As

observed by fieldwork records, this seems to be a case of

over-estimations on behalf of the DGCN.

4. Conclusions

The SAC-C/MMRS sensor produced promising results

for both study areas and could be very suitable for burned

area mapping. It provided accurate estimations of the affect-

ed areas, compared with the official statistics provided by the

DGCN, visual identification in higher resolution satellite

data and field work. However, further work in needed to

assess its performance on a national scale. In this respect, its

low temporal resolution became a problem since it was

difficult to obtain a cloud-free coverage over large areas.

Out of the spectral indices used, the BAI showed a high

efficiency in both study sites, which are good representa-

tives of Mediterranean environments. However, it did not

help clarify the confusion caused by water bodies and cloud

shadows, which were the covers with the highest spectral

similarity to burned areas. Applying the NDII index, defined

within the NIR-SWIR bi-spectral space, helped to clarify

this confusion thus proving the convenience of including the

SWIR region for burned area mapping studies.

Acknowledgements

SAC-MMRS images were freely issued via a project

approved by the Argentinian Space Institute (Comision

Nacional de Actividades Espaciales, CONAE), whose

support in this research is greatly thanked. Additional

funding was obtained from the Spanish Forestry Service

(Spanish Ministry of Environment). Suggestions provided

by Patrick Vaughan were very valuable.

References

Adams, R., & Bischof, L. (1994). Seeded region growing. IEEE Trans-

actions on Pattern Analysis and Machine Intelligence, 16(6), 641–647.

Barbosa, P. M., San-Miguel-Ayanz, J., & Schmuck, G. (2001). Remote

sensing of forest fires in Southern Europe using IRS-WiFS and MODIS

data. SPIE Remote Sensing Symposium, Toulouse, France, 17–21 Sep-

tember 2001.

Chavez Jr., P. S. (1996). Image-based atmospheric corrections-revisited and

improved. Photogrammetric Engineering and Remote Sensing, 62(9),

1025–1036.

Chuvieco, E., & Congalton, R. G. (1998). Mapping and inventory of forest

fires from digital processing of TM data. Geocarto International, 4,

41–53.

Chuvieco, E., Martin, M. P., & Palacios, A. (2002). Assessment of different

spectral indices in the red-near-infrared spectral domain for burned land

discrimination. International Journal of Remote Sensing, 23(23),

5103–5110.

Eva, H., & Lambin, E. F. (1998). Burnt area mapping in Central Africa

using ATSR data. International Journal of Remote Sensing, 19(18),

3473–3497.

Fraser, R. H., Li, Z., & Cihlar, J. (2000a). Hotspot and NDVI Differencing

Synergy (HANDS): A new technique for burned area mapping over

boreal forest. Remote Sensing of Environment, 74, 362–376.

Fraser, R. H., Li, Z., & Landry, R. (2000b). SPOT vegetation for character-

ising boreal forest fires. International Journal of Remote Sensing,

21(18), 3525–3532.

Garcıa, M. (2003). Burnt area mapping in Spain using satellite remote

sensing, MSc Dissertation, School of Earth and Environmental Scien-

ces, The University of Greenwich.

Gregoire, J. -M., Tansey, K., & Silva, J. M. N. (2003). The GBA2000

initiative: Developing a global burned area database from SPOT-VEG-

ETATION imagery. International Journal of Remote Sensing, 24(6),

1369–1376.

Hunt, E. R., & Rock, B. N. (1989). Detection of changes in leaf water

content using near and middle-infrared reflectances. Remote Sensing of

Environment, 30, 43–54.

Kasischke, E. S., French, N. H. F., Harrell, P., Christensen, N. L., Ustin, S.

L., & Barry, D. (1993). Monitoring of wildfires in Boreal Forests using

large area AVHRR NDVI composite image data. Remote Sensing of

Environment, 45, 61–71.

Key, C. H., & Benson, N. C. (1999). The Normalized Burn Ratio, a Land-

sat TM radiometric index of burn severity incorporating multi-temporal

differencing, (http://nrmsc.usgs.gov/research/nbr.htm).

Koutsias, N., & Karteris, M. (2000). Burned area mapping using logistic

regression modeling of a single post-fire Landsat-5 Thematic Mapper

image. International Journal of Remote Sensing, 21(4), 673–687.

Koutsias, N., Karteris, M., & Chuvieco, E. (2000). The use of intensity-

hue-saturation transformation of Landsat-5 Thematic Mapper data for

burned land mapping. Photogrammetric Engineering and Remote Sens-

ing, 66, 829–839.

Koutsias, N., Karteris, M., Fernandez, A., Navarro, C., Jurado, J., Navarro,

R., & Lobo, A. (1999). Burnt land mapping at local scale. In E. Chuvieco

(Ed.), Remote sensing of large wildfires in the European Mediterranean

basin ( pp. 123–138). Berlin, Germany: Springer-Verlag.

Lira Chavez, J. (2002). Introduccion al tratamiento digital de imagenes,

Ciencia de la Computacion, Instituto Politecnico Nacional, Universidad

Nacional Autonoma de Mexico.

Lopez Garcıa, M. J., & Caselles, V. (1991). Mapping burns and natural

reforestation using Thematic Mapper data. Geocarto International, 1,

31–37.

Martın, M. P. (1998). Cartografıa e inventario de incendios forestales en la

Penınsula Iberica a partir de imagenes NOAA-AVHRR. Departamento

de Geografıa. Alcala de Henares, Universidad de Alcala.

Martın, M. P., & Chuvieco, E. (1993). Mapping and evaluation of burned

land from multitemporal analysis of AVHRR NDVI images. In P. J.

Kennedy, & M. Karteris (Eds.), International workshop: Satellite tech-

nology and GIS for Mediterranean forest mapping and fire management

( pp. 71–83). Thessaloniki, Greece: European Union Publication Office.

Martın, M. P., & Chuvieco, E. (1995). Mapping and evaluation of burned

land from multitemporal analysis of AVHRR NDVI images. EARSeL

Advances in Remote Sensing, 4(3), 7–13.

Miller, H. J., & Yool, S. R. (2002). Mapping forest post-fire canopy con-

sumption in several overstory types using multi-temporal Landsat TM

and ETM data. Remote Sensing of Environment, 82, 481–496.

Patterson, M. W., & Yool, S. R. (1998). Mapping fire-induced vegeta-

tion mortality using landsat Thematic Mapper data: A comparison of

linear transformation techniques. Remote Sensing of Environment, 65,

132–142.

Pereira, J. M. C. (1999). A comparative evaluation of NOAA/AVHRR

vegetation indexes for burned surface detection and mapping. IEEE

Transactions on Geoscience and Remote Sensing, 37(1), 217–226.

Page 10: Assessment of the potential of SAC-C/MMRS imagery for mapping burned areas in Spain

M. Garcıa, E. Chuvieco / Remote Sensing of Environment 92 (2004) 414–423 423

Pereira, J. M. C., Chuvieco, E., Beudoin, A., & Desbois, N. (1997).

Remote Sensing of burned areas: A review. In E. Chuvieco (Ed.), A

review of remote sensing methods for the study of large wildland fires

( pp. 127–184). Alcala de Henares, Madrid: Departamento de Geo-

grafıa, Universidad de Alcala.

Pinty, B., & Verstraete, M. M. (1992). GEMI: A non-linear index to mon-

itor global vegetation from satellites. Vegetatio, 101, 15–20.

Roy, D. P., Lewis, P. E., & Justice, C. O. (2002). Burned area mapping

using multi-temporal moderate spatial resolution data—a bi-directional

reflectance model-based expectation approach. Remote Sensing of En-

vironment, 83, 263–286.

Salvador, R., Valeriano, J., Pons, X., & Dıaz-Delgado, R. (2000). A semi-

automatic methodology to detect fire scars in shrubs and evergreen

forests with Landsat MSS time series. International Journal of Remote

Sensing, 21(4), 655–671.

Stroppiana, D., Pinnock, S., Pereira, J. M. C., & Gregorie, J. M. (2002).

Radiometric analysis of SPOT-VEGETATION images for burnt area

detection in Northern Australia. Remote Sensing of Environment, 82,

21–37.

Swain, P. H., & Davis, S. M. (1978). Remote sensing: The quantitative

approach. New York: McGraw-Hill, Chapter 3.

Trigg, S., & Flasse, S. (2000). Characterizing the spectral-temporal re-

sponse of burned savannah using in situ spectroradiometry and in-

frared thermometry. International Journal of Remote Sensing, 21(16),

3161–3168.

Vazquez, A., Cuevas, J. M., & Gonzalez-Alonso, F. (2001). Comparison of

the use of WiFS and LISS images to estimate the area burned in a large

forest fire. International Journal of Remote Sensing, 22(5), 901–907.

Velez, R. (2000). Los incendios forestales en la cuenca mediterranea. La

defensa contra incendios forestales. Fundamentos y experiencias. R.

Velez. Madrid: Mc Graw Hill, Chapter 3.