Assessing Fine Beam RADARSAT-1 Backscatter from a White Mangrove (Laguncularia Racemosa (Gaertner)) Canopy (Kovacs et al, 2006)
Ó Springer 2006
Wetlands Ecology and Management (2006) 14:401–408
DOI 10.1007/s11273-005-6237-x
-1
Assessing fine beam RADARSAT-1 backscatter from a white mangrove
(Laguncularia racemosa (Gaertner)) canopy
John M. Kovacs1,*, Casey V. Vandenberg2 and Francisco Flores-Verdugo3
1
Department of Geography, Nipissing University, North Bay, Ontario, P1B 8L7, Canada; 2Department of
Geography, University of Lethbridge, Lethbridge, Alberta, T1K 3M4, Canada; 3Instituto del Ciencias del Mar
y Limnologı´a, UNAM, 82000 Mazatla´n, Sinaloa, Mexico; *Author for correspondence (e-mail: johnmk@
nipissingu.ca; phone: +705-474-3461; fax: +705-474-1947)
Received 31 October 2005; accepted in revised form 21 December 2005
Key words: Backscatter coefficients, Biophysical parameters, LAI, Mangroves, Mexico, RADARSAT-1,
SAR
Abstract
To determine whether spaceborne Synthetic Aperture Radar (SAR), specifically fine beam RADARSAT-1
C-band, could be used to provide quantitative data on white mangrove (Laguncularia racemosa (Gaertner))
forests, backscatter coefficients (r°) were examined in relation to structural parameter data collected from
plots located in a mangrove forest of the Mexican Pacific. Significant coefficients of determination were
recorded between the backscatter coefficients and the logarithms of both Leaf Area Index (LAI) and mean
stem height at two incident angles for both the dry and wet seasons. The highest coefficients of determi-
nation for LAI (r2 = 0.60) and mean stem height (r2 = 0.72) were observed using the shallower ($40°) and
steeper ($47°) incident angles, respectively. No significant relationships were recorded between the back-
scatter coefficients and either stem density, basal area or mean DBH. Given the results of this investigation,
it is recommended that for cloud covered regions, fine beam RADARSAT-1 data could be used by resource
managers when they require a quick method for surveying structural damage to mangroves resulting from
both natural and anthropogenic causes.
parameters such as Leaf Area Index (LAI). The
Introduction
results of these investigations do suggest that, in
Numerous studies have been conducted in the use conjunction with field data, these satellite sensors
of remotely sensed data from satellites for moni- can provide sufficiently accurate estimations of
toring mangrove forests but few have examined many of these structural parameters. For example,
the use of these data for estimating quantitative Kovacs et al. (2005) recently mapped estimated
measures of these forested wetlands (Green et al. mangrove LAI at the species level using a combi-
1998). Consequently, there has been a recent nation of very high resolution IKONOS data and
interest (Jensen et al. 1991; Ramsey and Jensen in situ LAI collected with a LAI-2000 Plant Can-
1996; Green et al. 1997; Kovacs et al. 2004a) in opy Analyzer.
determining whether these sensors can be used as a Although the results of the aforementioned
method for extracting mangrove forest biophysical remote sensing studies on mangroves are quite
402
promising, the sensors they have examined are Materials and methods
limited to optical (passive) satellite platforms. In
many tropical and subtropical regions persistent Study area
cloud cover is the norm and thus the availability
The Agua Brava Lagoon is located along the
of useful optical data of many mangrove regions
Pacific Coast of Mexico within the State of Nayarit
is quite limited. One alternative to optical data is
(Figure 1). This mangrove forest belongs to the
Synthetic Aperture Radar (SAR). Unhampered
´
much greater Teacapan–Agua Brava Lagoon–
by cloud cover, SAR imagery can also provide
Estuarine System, which is considered one of the
unique information on the surface targets based
largest mangrove wetlands of the Pacific coast of
on the interaction with the active energy and the
the Americas (Flores-Verdugo et al. 1990). This
geometry and dielectric constants of the ground
system experiences sharp seasonal variation in
features imaged (Ulaby et al. 1986). Moreover,
precipitation, typical of the tropical sub-humid
due to characteristic interactions of SAR with
climatic zone, with distinct wet and dry seasons
persistently flooded forests and relatively flat
occurring in the months of June–October and
terrain, mangroves are considered ideal forest
November–May, respectively. Although consid-
canopies for SAR investigations (Hess et al.
ered a very important ecological reserve for
1990). As with other forested wetlands, the
numerous organisms (Flores-Verdugo et al. 1997)
backscatter from SAR is enhanced due to the
and an important local source of renewable prod-
smooth water below which enhances scattering
ucts (Kovacs 1999), recent investigations (Flores-
mechanisms within the forest components (e.g.
Verdugo et al. 1997; Kovacs 2000; Kovacs et al.
stems, branches). Although limited in number,
2001a, 2001b, 2004a, 2004b) have indicated that
recent studies (Mougin et al. 1999; Proisy et al.
the system is experiencing considerable degrada-
2000; Proisy et al. 2002) of a mangrove forest in
tion, resulting primarily from anthropogenic cau-
French Guiana have shown the utility of air-
ses. Consequently, a recent study of estimated
borne SAR for estimating structural parameters
mangrove LAI (Kovacs et al. 2005) has identified,
of mangrove forest. Their results, based on 12
based on state of degradation and species compo-
sample stands, revealed significant correlations
sition, four types of mangroves located within the
between radar backscatter coefficients and
Agua Brava Lagoon. These mangrove classes in-
numerous stand parameters (e.g. basal area, stem
clude dead white mangrove (Laguncularia race-
density) at various SAR frequencies and polar-
mosa (Gaertner)), poor condition white mangrove,
izations, including C-band HH polarization.
healthy condition white mangrove and healthy red
The purpose of this investigation is to determine
mangrove (Rhizophora mangle (L.)). The latter
whether C-band HH polarization data, captured
class, red mangrove, does not appear to be affected
from a satellite platform (RADARSAT-1 fine
but is limited in distribution to the main edge of the
beam), can provide similar success in estimating
Agua Brava Lagoon as well as along the numerous
structural parameter data of mangrove forest. In
tidal channels that extend further inland.
comparison to airborne SAR, spaceborne SAR
data may be more applicable for mangrove mon-
itoring systems as the data are available on a
Field data collection
repetitive basis and cover much larger areas. Un-
like the previous SAR studies, this investigation
Fourteen circular plots (0.03 ha) were laid out in
will also focus on potential differences associated
the Agua Brava Lagoon system during the month
with multiple incident angles and multitemporal
of May 2004. Within each plot, all trees of a DBH
data (i.e. seasonal acquisitions). Although using
greater than or equal to 2.5 cm were measured and
standard beam RADARSAT-1 and focusing on
the central location recorded at a sub-meter
non-mangrove forested wetlands, the results of
accuracy using a differential post-processing GPS
Townsend (2002) suggest that the accuracy of the
unit. The selection of the stands was chosen at
estimation of the mangrove biophysical parame-
random from homogeneous white mangrove areas
ters are likely to vary according to these two
that represented the three conditions that persist in
variables.
403
Figure 1. The Agua Brava Lagoon of the Mexican Pacific.
this basin mangrove forest (Kovacs et al. 2005). Remote sensing data acquisition & processing
Specifically, four of the plots were laid out in poor
Two fine beam mode RADARSAT-1 (C-band)
condition white mangrove stands and seven of the
scenes of the study area, representing two incident
plots in healthy condition sites. The remaining
angles, were acquired for both the dry and wet
three plots were laid out in dead tree stands. All
seasons. Specifically, the two band positions cho-
fourteen plots were located in areas where the
substrate is completely saturated or flooded. An sen, SAR F2Near and SAR F5Far, represent
incident angles of approximately 40° and 47°,
average stem height, an average DBH and both
respectively. At present RADARSAT-1 fine beam
stem density and basal area was determined for
has the highest spatial resolution for a SAR based
each location (Table 1). In addition, to confirm the
satellite platform with a spatial resolution of
homogeneity of the locations, average tree heights
approximately 8 m and a pixel spacing of 6.25 m.
and quick estimates of basal areas, calculated
Meteorological records from an adjacent town
using a forest prism (BAF2), were determined at a
were used to determine if any precipitation events
20 m distance from the centre of the plot in all four
cardinal directions. Using the technique described occurred prior to or during the SAR acquisitions.
For each scene, radar backscatter coefficients (r°)
by Kovacs et al. (2004a), estimated LAI maps for
were then recorded, in decibels, by using the ori-
the two seasons, dry and wet, were then derived
ginal brightness values, the georeference segments,
from an April 2004 and an October 2004 IKONOS
the orbital segments and arrays provided. All four
scene, respectively. As a result, using the estimated
of the scenes were then co-registered to a geo-
LAI maps, an average LAI value was then deter-
metrically corrected 2004 IKONOS scene. Using
mined for each plot, for both the dry and wet
the sub-meter GPS data collected, a series of
seasons.
404
Table 1. Biophysical parameter data for the Laguncularia racemosa dominated plots.
Basal area (m2/ha)
Plot # Vegetated Stem density (stems/ha) Mean DBH Mean stem LAI dry LAI wet
condition total (L.r/R.m.) total (L.r/R.m.) (cm) height (m) season season
1 dead 3633 9.2 5.5 0.5 0.20 0.01
2 4367 13.5 6.1 0.5 0.10 0.31
3 4533 17.5 6.7 1.2 0.01 0.01
4 poor 4633 15.6 6.1 3.7 1.10 1.41
5 5133 24.6 7.1 3.5 1.16 1.28
6 7067 19.9 5.3 2.6 0.87 1.81
7 2902 4.7 4.2 3.0 1.10 1.63
8 healthy 6833 24.3 6.1 2.5 2.00 3.36
9 5133 19.5 6.2 3.0 2.57 3.95
10 9767 29.5 5.7 5.5 2.02 3.44
11 3712 (3666/46) 16.0 (15.93/0.02) 6.5 4.5 2.27 3.66
12 3237 (3116/121) 11.2 (10.93/0.29) 6.0 7.0 2.78 4.12
13 4282 (4068/214) 11.1 (10.88/0.22) 5.3 3.5 2.57 4.10
14 4988 (4926/62) 8.6 (8.54/0.05) 4.2 4.5 2.38 4.06
L.r. = Laguncularia racemosa and R.m. = Rhizophora mangle.
masks, to approximate the area of sampling, were the mean RADARSAT-1 backscattering coeffi-
then created for each plot. Consequently, a mean cient (Table 2), only two parameters, LAI and
backscatter coefficient, in decibels, was then de- mean stem height, were identified as having a sig-
rived from the pixels located under the masks. The nificant relationship (p < 0.01) for both incident
masks were slightly larger than the actual plot angles and for both seasons. Specifically, lower
areas but given the homogeneity of the areas, mean backscatter coefficients were recorded from
determined from the IKONOS maps and prism white mangrove plots having lower LAI and lower
sweeps, this was deemed appropriate. The size of mean stem heights (Figure 2). It must be noted
the masks was selected to reduce radiometric res- that although both of these parameters were
olution errors originating from speckle for the identified separately as statistically significant they
homogeneous targets (Ulaby et al. 1986). Finally, are interrelated with one another and thus should
simple linear regression techniques were employed not be expected to respond separately or uniquely
to examine whether the radar backscatter coeffi- in relation to the total backscatter observed from
cients could be used to predict any of the bio- each plot. Specifically, according to Table 1, LAI
physical parameters. and mean stem height are highly correlated.
However, given the characteristic properties of the
radar wavelength selected (C-band), it is postu-
lated that LAI plays the dominant role in the
Results and discussion
contribution of the observed backscatter. Specifi-
Based on the linear coefficients of determination cally, it is commonly agreed (Leckie and Ranson
between the logarithm of the stand parameters and 1998) that the primary contributor to backscatter
Table 2. Linear coefficients of determination (r2) between the logarithm of structural parameters of white mangrove plots (n = 14)
and RADARSAT-1 fine beam backscattering coefficients (dB) at two incident angles (F2N=40°; F5F=47°). Boldface numbers
represent significant relationships at p < 0.01.
Parameter Dry season Wet season
F2N F5F F2N F5F
LAI 0.60 0.56 0.48 0.45
Mean stem height (m) 0.68 0.55 0.61 0.72
Stem density (stem/ha) 0.04 0.16 0.10 0.05
Basal area (m2/ha) 0.01 0.06 0.14 0.00
Mean DBH (cm) 0.06 0.02 0.01 0.13
405
Figure 2. Fine beam RADARSAT-1 backscattering coefficients, at two incident angles (F2N=40°; F5F=47°), vs. Leaf Area Index
(LAI) and mean stem height for 14 white mangrove plots located in a mangrove forest of the Mexican Pacific.
from forest canopies in the shorter wavelengths tributions. In contrast, the interactions of SAR
(e.g. K, X, C) is from the interactions with the with trunks and main branches are considered the
leaves and smaller canopy elements (i.e. volume main contributors of forest stand backscatter for
scattering). Consequently, little or no backscatter the longer SAR wavelengths (e.g. L and P). As a
should be expected from dead mangrove areas result, for such SAR wavelengths, the backscatter
(Figure 3) which contain few, if any, leaves (i.e. coefficients from the dead areas, where trunks
LAI $ 0), especially with the flooded regime of the remain, should be significant. This would be
mangroves which limits ground reflectance con- especially true in these flooded forests where the
406
Figure 3. Comparison of RADARSAT-1 F5F backscatter from a white mangrove forest with an enhanced optical IKONOS data false
colour composite (4, 3, 2). Note the large expanse of dead mangroves in the top portions of the images. Refer to Figure 1 for
geographic position.
double-bounce effect would result from the inter- SAR backscatter coefficients and other man-
actions between the standing water and the trunks grove structural parameters. For example, in these
which act as dihedral corner reflectors. studies, in addition to biomass, coefficients as high
Although LAI and the corresponding back- as 0.90 and 0.88 were recorded for basal area and
scatter coefficients were deemed significant, the mean DBH in relation to AIRSAR C-HH back-
data plotted in Figure 2 indicate that the discrep- scatter. The much higher spatial resolution of their
ancy between no leaf content (i.e. LAI $ 0) and SAR platform and, possibly, the greater range of
some leaf content (i.e. low LAI) is more apparent the parameters recorded in their study plots may
than between low LAI and high LAI for the RA- explain the discrepancy. For example, in this
DARSAT-1 data. This circumstance could explain study, although the LAI varied considerably
the lower coefficients of determination recorded among the sample plots, this was not the case for
for the wet season data, when the range of the LAI the basal areas and stem densities (Table 1).
increased dramatically. The limited ability to dis- Enhanced backscatter from flooded areas is
cern LAI values for the healthier mangroves may angle dependent and thus an important consider-
be due to saturation of the SAR signal occurring ation for mangrove areas. However, with regards
much earlier than that which might be expected to the examination of two incident angles, the
from longer SAR wavelengths (Dwivedi et al. results of this study indicate that the use of the
1999). Specifically, shorter wavelengths are known shallower incident angle (40°) only slightly im-
to be constrained in their ability to penetrate proves the relationship between LAI and the cor-
closed forest canopies. responding backscatter coefficients. At the fine
Unlike the previous work of others using air- beam mode, RADARSAT-1 data selection is
borne SAR (Mougin et al. 1999; Proisy et al. 2000; restricted to an incident angle range of approxi-
Proisy et al. 2002), the results of this study do mately 38°–47°. At the standard beam mode, the
not indicate significant relationships between range of incident angles available does improve
407
(24°–47°) but at the expense of a lower spatial anticipated that with the planned launch of
resolution (25 m). RADARSAT-2 the extraction of mangrove bio-
With regard to seasonality of the data, the physical parameter data from spaceborne SAR
coefficients of determination for LAI and SAR will improve significantly. The launch of RA-
backscatter coefficients were greater for the dry DARSAT-2 will ensure the continuity of all
season. LAI values for the plots did, however, vary existing RADARSAT-1 beam modes, which
quite dramatically as a result of the phenological would allow for long term change detection stud-
changes associated with the white mangroves of ies. Moreover, RADARSAT-2 will provide new
this study area. In this investigation no precipita- capabilities for improving mangrove forest struc-
tion events were recorded three days prior to or tural parameter extraction including the addition
during the acquisition of the wet season imagery. of an ultra fine beam mode (3 m resolution) and
Consequently, it was not possible to determine quad-polarization.
whether surface moisture, which increases the
dielectric constant of the leaves, would have
altered the relationship between LAI and the
Acknowledgements
backscatter coefficients.
J.M. Kovacs wishes to acknowledge financial sup-
port (grant # 249496-02) of the Natural Sciences
Conclusion
and Engineering Research Council of Canada. The
Canada Space Agency provided the RADARSAT-
The results of this investigation suggest that
1 data to J.M. Kovacs and F. Flores-Verdugo as
spaceborne SAR, specifically fine beam RADAR-
part of the Data for Research Use (DRU) program
SAT-1, can be used to some extent for extracting
(project # 02-04). The authors would also like to
biophysical parameter data from mangrove for-
extend their thanks to Lance P. Aspden, Francisco
ests. Specifically, in this investigation it was pos-
Flores de Santiago and Neil Latour for their
sible to discern dead white mangrove stands from
assistance in the field data sampling.
healthy ones due to a significant relationship be-
tween LAI and corresponding RADARSAT-1 fine
beam backscatter coefficients. The results also
References
indicate that a steeper incident angle and the
acquisition of SAR in the dry season may improve
Dwivedi R.S., Rao B.R.M. and Bhattacharya S. 1999. Mapping
this relationship. Although the coefficients of
wetlands of the Sundaban Delta and it’s environs using ERS-
determination were not as high as those deter-
1 SAR data. Int. J. Remote Sens. 20: 2235–2247.
mined by others using airborne SAR, the results of ´
Flores-Verdugo F.J., Gonzalez-Farias F., Ramirez-Flores O.,
this study would suggest that RADARSAT-1 Amezcua-Linares F., Yanez-Arancibia A., Alvarez-Rubio M.
could be used as an alternative method for moni- and Day J.W. 1990. Mangrove ecology, aquatic primary
´
productivity, and fish community dynamics in Teacapan–
toring mangrove forests, particularly for regular
Agua Brava lagoon–estuarine system (Mexican Pacific).
surveys of mangrove damage incurred by major
Estuaries 13: 219–230.
disturbances (e.g. hurricanes, pollution). The use Flores-Verdugo F.J., Gonzalez-Farias F., Blanco-Correa M.
of spaceborne SAR imagery can provide resource ´
and Nunez-Pasten A. 1997. The Teacapan-Agua Brava
managers with quick synoptic assessments of large Marismas Nacionales mangrove ecosystem on the Pacific
coast of Mexico. In: Kjerfve B., Drude L. and Diop E.H.S.
areas of mangrove at a known schedule and for
(eds), Mangrove Ecosystem Studies in Latin America and
regions where persistent cloud cover is the norm.
Africa. UNESCO, pp. 35–46.
Based on the results of this investigation, it is Green E.P., Mumby P.J., Alaisdair E.J., Clark C.D. and
recommended that further studies with RADAR- Ellis A.C. 1997. Estimating leaf area index of mangroves
SAT-1 data be conducted for other white man- from satellite data. Aquat. Bot. 58: 11–19.
Green E.P., Clark C.D., Mumby P.J., Edwards A.J. and
grove areas, particularly those with structural
Ellis A.C. 1998. Remote sensing techniques for mangrove
parameters that vary considerable from this study
mapping. Int. J. Remote Sens. 19: 935–956.
area. Moreover, it is suggested that investigations Hess L.L., Melack J.M. and Simonett D.S. 1990. Radar
of these data for other mangrove species be con- detection of flooding beneath the forest canopy: A review.
sidered. Given the outcome of this study, it is Int. J. Remote Sens. 11: 1313–1325.
408
Jensen J.R., Lin H., Yang Y., Ramsey E., Davis B.A. and Leckie D.G. and Ranson K.J. 1998. Forestry applications using
Thoemke C.W. 1991. The measurement of mangrove char- imaging radar. In: Henderson F.M. and Lewis A.J. (eds),
acteristics in Southwest Florida using SPOT multispectral Manual of Remote Sensing. 3rd ed. Vol. 2. Principles and
data. Geocarto Int. 2: 13–21. Applications of Imaging Radar. John Wiley & Sons, Tor-
Kovacs J.M. 1999. Assessing mangrove use at the local scale. onto, Canada, pp. 435–509.
Landscape Urban Plan. 43: 201–208. Mougin E., Proisy C., Marty G., Fromard F., Puig H.,
Kovacs J.M. 2000. Perceptions of environmental change in a Betoulle J.L. and Rudant J.P. 1999. Multifrequency and
tropical coastal wetland. Land Degrad. Dev. 11: 209–220. multipolarization radar backscattering from mangrove
Kovacs J.M., Wang J. and Blanco-Correa M. 2001a. Mapping forests. IEEE Trans. Geosci. Remote Sens. 37: 94–102.
mangrove disturbances using multi-date Landsat TM imag- Proisy C., Mougin E., Fromard F. and Karam M.A. 2000.
ery. Environ. Manage. 27: 763–776. Interpretation of polarimetric radar signatures of mangrove
Kovacs J.M., Blanco-Correa M. and Flores-Verdugo F. 2001b. forests. Remote Sens. Environ. 71: 56–66.
A logistic regression model of hurricane impacts in a man- Proisy C., Mougin E., Fromard F., Trichon V. and Karam
grove forest of the Mexican Pacific. J. Coastal Res. 17: 30–37. M.A. 2002. On the influence of canopy structure on the
Kovacs J.M., Flores-Verdugo F., Wang J. and Aspden L.P. backscatter of mangrove forests. Int. J. Remote Sens. 23:
2004a. Estimating leaf area index of a degraded mangrove 4197–4210.
forest using high spatial resolution satellite data. Aquat. Bot. Ramsey E. and Jensen J.R. 1996. Remote sensing of mangrove
80: 13–22. wetlands: relating canopy spectra to site-specific data. Pho-
Kovacs J.M., Malczewski J. and Flores-Verdugo F. 2004b. togramm. Eng. Remote Sens. 62: 939–948.
Examining local ecological knowledge of hurricane impacts Townsend P.A. 2002. Estimating forest structure in wetlands
in a mangrove forest using an Analytical Hierarchy Process using multitemporal SAR. Remote Sens. Environ. 79: 288–
(AHP) approach. J. Coastal Res. 20: 792–800. 304.
Kovacs J.M., Wang J. and Flores-Verdugo F. 2005. Mapping Ulaby F.T., Moore R.K. and Fung A.K. 1986. Microwave
mangrove leaf area index at the species level using IKO- Remote Sensing: Radar Remote Sensing and Surface Scat-
NOS and LAI-2000 sensors. Estuarine, Coastal Shelf Sci. tering and Emission Theory, Vol. 2. Artech House Inc.,
62: 377–384. Norwood, MA, USA.
Wetlands Ecology and Management (2006) 14:401–408
DOI 10.1007/s11273-005-6237-x
-1
Assessing fine beam RADARSAT-1 backscatter from a white mangrove
(Laguncularia racemosa (Gaertner)) canopy
John M. Kovacs1,*, Casey V. Vandenberg2 and Francisco Flores-Verdugo3
1
Department of Geography, Nipissing University, North Bay, Ontario, P1B 8L7, Canada; 2Department of
Geography, University of Lethbridge, Lethbridge, Alberta, T1K 3M4, Canada; 3Instituto del Ciencias del Mar
y Limnologı´a, UNAM, 82000 Mazatla´n, Sinaloa, Mexico; *Author for correspondence (e-mail: johnmk@
nipissingu.ca; phone: +705-474-3461; fax: +705-474-1947)
Received 31 October 2005; accepted in revised form 21 December 2005
Key words: Backscatter coefficients, Biophysical parameters, LAI, Mangroves, Mexico, RADARSAT-1,
SAR
Abstract
To determine whether spaceborne Synthetic Aperture Radar (SAR), specifically fine beam RADARSAT-1
C-band, could be used to provide quantitative data on white mangrove (Laguncularia racemosa (Gaertner))
forests, backscatter coefficients (r°) were examined in relation to structural parameter data collected from
plots located in a mangrove forest of the Mexican Pacific. Significant coefficients of determination were
recorded between the backscatter coefficients and the logarithms of both Leaf Area Index (LAI) and mean
stem height at two incident angles for both the dry and wet seasons. The highest coefficients of determi-
nation for LAI (r2 = 0.60) and mean stem height (r2 = 0.72) were observed using the shallower ($40°) and
steeper ($47°) incident angles, respectively. No significant relationships were recorded between the back-
scatter coefficients and either stem density, basal area or mean DBH. Given the results of this investigation,
it is recommended that for cloud covered regions, fine beam RADARSAT-1 data could be used by resource
managers when they require a quick method for surveying structural damage to mangroves resulting from
both natural and anthropogenic causes.
parameters such as Leaf Area Index (LAI). The
Introduction
results of these investigations do suggest that, in
Numerous studies have been conducted in the use conjunction with field data, these satellite sensors
of remotely sensed data from satellites for moni- can provide sufficiently accurate estimations of
toring mangrove forests but few have examined many of these structural parameters. For example,
the use of these data for estimating quantitative Kovacs et al. (2005) recently mapped estimated
measures of these forested wetlands (Green et al. mangrove LAI at the species level using a combi-
1998). Consequently, there has been a recent nation of very high resolution IKONOS data and
interest (Jensen et al. 1991; Ramsey and Jensen in situ LAI collected with a LAI-2000 Plant Can-
1996; Green et al. 1997; Kovacs et al. 2004a) in opy Analyzer.
determining whether these sensors can be used as a Although the results of the aforementioned
method for extracting mangrove forest biophysical remote sensing studies on mangroves are quite
402
promising, the sensors they have examined are Materials and methods
limited to optical (passive) satellite platforms. In
many tropical and subtropical regions persistent Study area
cloud cover is the norm and thus the availability
The Agua Brava Lagoon is located along the
of useful optical data of many mangrove regions
Pacific Coast of Mexico within the State of Nayarit
is quite limited. One alternative to optical data is
(Figure 1). This mangrove forest belongs to the
Synthetic Aperture Radar (SAR). Unhampered
´
much greater Teacapan–Agua Brava Lagoon–
by cloud cover, SAR imagery can also provide
Estuarine System, which is considered one of the
unique information on the surface targets based
largest mangrove wetlands of the Pacific coast of
on the interaction with the active energy and the
the Americas (Flores-Verdugo et al. 1990). This
geometry and dielectric constants of the ground
system experiences sharp seasonal variation in
features imaged (Ulaby et al. 1986). Moreover,
precipitation, typical of the tropical sub-humid
due to characteristic interactions of SAR with
climatic zone, with distinct wet and dry seasons
persistently flooded forests and relatively flat
occurring in the months of June–October and
terrain, mangroves are considered ideal forest
November–May, respectively. Although consid-
canopies for SAR investigations (Hess et al.
ered a very important ecological reserve for
1990). As with other forested wetlands, the
numerous organisms (Flores-Verdugo et al. 1997)
backscatter from SAR is enhanced due to the
and an important local source of renewable prod-
smooth water below which enhances scattering
ucts (Kovacs 1999), recent investigations (Flores-
mechanisms within the forest components (e.g.
Verdugo et al. 1997; Kovacs 2000; Kovacs et al.
stems, branches). Although limited in number,
2001a, 2001b, 2004a, 2004b) have indicated that
recent studies (Mougin et al. 1999; Proisy et al.
the system is experiencing considerable degrada-
2000; Proisy et al. 2002) of a mangrove forest in
tion, resulting primarily from anthropogenic cau-
French Guiana have shown the utility of air-
ses. Consequently, a recent study of estimated
borne SAR for estimating structural parameters
mangrove LAI (Kovacs et al. 2005) has identified,
of mangrove forest. Their results, based on 12
based on state of degradation and species compo-
sample stands, revealed significant correlations
sition, four types of mangroves located within the
between radar backscatter coefficients and
Agua Brava Lagoon. These mangrove classes in-
numerous stand parameters (e.g. basal area, stem
clude dead white mangrove (Laguncularia race-
density) at various SAR frequencies and polar-
mosa (Gaertner)), poor condition white mangrove,
izations, including C-band HH polarization.
healthy condition white mangrove and healthy red
The purpose of this investigation is to determine
mangrove (Rhizophora mangle (L.)). The latter
whether C-band HH polarization data, captured
class, red mangrove, does not appear to be affected
from a satellite platform (RADARSAT-1 fine
but is limited in distribution to the main edge of the
beam), can provide similar success in estimating
Agua Brava Lagoon as well as along the numerous
structural parameter data of mangrove forest. In
tidal channels that extend further inland.
comparison to airborne SAR, spaceborne SAR
data may be more applicable for mangrove mon-
itoring systems as the data are available on a
Field data collection
repetitive basis and cover much larger areas. Un-
like the previous SAR studies, this investigation
Fourteen circular plots (0.03 ha) were laid out in
will also focus on potential differences associated
the Agua Brava Lagoon system during the month
with multiple incident angles and multitemporal
of May 2004. Within each plot, all trees of a DBH
data (i.e. seasonal acquisitions). Although using
greater than or equal to 2.5 cm were measured and
standard beam RADARSAT-1 and focusing on
the central location recorded at a sub-meter
non-mangrove forested wetlands, the results of
accuracy using a differential post-processing GPS
Townsend (2002) suggest that the accuracy of the
unit. The selection of the stands was chosen at
estimation of the mangrove biophysical parame-
random from homogeneous white mangrove areas
ters are likely to vary according to these two
that represented the three conditions that persist in
variables.
403
Figure 1. The Agua Brava Lagoon of the Mexican Pacific.
this basin mangrove forest (Kovacs et al. 2005). Remote sensing data acquisition & processing
Specifically, four of the plots were laid out in poor
Two fine beam mode RADARSAT-1 (C-band)
condition white mangrove stands and seven of the
scenes of the study area, representing two incident
plots in healthy condition sites. The remaining
angles, were acquired for both the dry and wet
three plots were laid out in dead tree stands. All
seasons. Specifically, the two band positions cho-
fourteen plots were located in areas where the
substrate is completely saturated or flooded. An sen, SAR F2Near and SAR F5Far, represent
incident angles of approximately 40° and 47°,
average stem height, an average DBH and both
respectively. At present RADARSAT-1 fine beam
stem density and basal area was determined for
has the highest spatial resolution for a SAR based
each location (Table 1). In addition, to confirm the
satellite platform with a spatial resolution of
homogeneity of the locations, average tree heights
approximately 8 m and a pixel spacing of 6.25 m.
and quick estimates of basal areas, calculated
Meteorological records from an adjacent town
using a forest prism (BAF2), were determined at a
were used to determine if any precipitation events
20 m distance from the centre of the plot in all four
cardinal directions. Using the technique described occurred prior to or during the SAR acquisitions.
For each scene, radar backscatter coefficients (r°)
by Kovacs et al. (2004a), estimated LAI maps for
were then recorded, in decibels, by using the ori-
the two seasons, dry and wet, were then derived
ginal brightness values, the georeference segments,
from an April 2004 and an October 2004 IKONOS
the orbital segments and arrays provided. All four
scene, respectively. As a result, using the estimated
of the scenes were then co-registered to a geo-
LAI maps, an average LAI value was then deter-
metrically corrected 2004 IKONOS scene. Using
mined for each plot, for both the dry and wet
the sub-meter GPS data collected, a series of
seasons.
404
Table 1. Biophysical parameter data for the Laguncularia racemosa dominated plots.
Basal area (m2/ha)
Plot # Vegetated Stem density (stems/ha) Mean DBH Mean stem LAI dry LAI wet
condition total (L.r/R.m.) total (L.r/R.m.) (cm) height (m) season season
1 dead 3633 9.2 5.5 0.5 0.20 0.01
2 4367 13.5 6.1 0.5 0.10 0.31
3 4533 17.5 6.7 1.2 0.01 0.01
4 poor 4633 15.6 6.1 3.7 1.10 1.41
5 5133 24.6 7.1 3.5 1.16 1.28
6 7067 19.9 5.3 2.6 0.87 1.81
7 2902 4.7 4.2 3.0 1.10 1.63
8 healthy 6833 24.3 6.1 2.5 2.00 3.36
9 5133 19.5 6.2 3.0 2.57 3.95
10 9767 29.5 5.7 5.5 2.02 3.44
11 3712 (3666/46) 16.0 (15.93/0.02) 6.5 4.5 2.27 3.66
12 3237 (3116/121) 11.2 (10.93/0.29) 6.0 7.0 2.78 4.12
13 4282 (4068/214) 11.1 (10.88/0.22) 5.3 3.5 2.57 4.10
14 4988 (4926/62) 8.6 (8.54/0.05) 4.2 4.5 2.38 4.06
L.r. = Laguncularia racemosa and R.m. = Rhizophora mangle.
masks, to approximate the area of sampling, were the mean RADARSAT-1 backscattering coeffi-
then created for each plot. Consequently, a mean cient (Table 2), only two parameters, LAI and
backscatter coefficient, in decibels, was then de- mean stem height, were identified as having a sig-
rived from the pixels located under the masks. The nificant relationship (p < 0.01) for both incident
masks were slightly larger than the actual plot angles and for both seasons. Specifically, lower
areas but given the homogeneity of the areas, mean backscatter coefficients were recorded from
determined from the IKONOS maps and prism white mangrove plots having lower LAI and lower
sweeps, this was deemed appropriate. The size of mean stem heights (Figure 2). It must be noted
the masks was selected to reduce radiometric res- that although both of these parameters were
olution errors originating from speckle for the identified separately as statistically significant they
homogeneous targets (Ulaby et al. 1986). Finally, are interrelated with one another and thus should
simple linear regression techniques were employed not be expected to respond separately or uniquely
to examine whether the radar backscatter coeffi- in relation to the total backscatter observed from
cients could be used to predict any of the bio- each plot. Specifically, according to Table 1, LAI
physical parameters. and mean stem height are highly correlated.
However, given the characteristic properties of the
radar wavelength selected (C-band), it is postu-
lated that LAI plays the dominant role in the
Results and discussion
contribution of the observed backscatter. Specifi-
Based on the linear coefficients of determination cally, it is commonly agreed (Leckie and Ranson
between the logarithm of the stand parameters and 1998) that the primary contributor to backscatter
Table 2. Linear coefficients of determination (r2) between the logarithm of structural parameters of white mangrove plots (n = 14)
and RADARSAT-1 fine beam backscattering coefficients (dB) at two incident angles (F2N=40°; F5F=47°). Boldface numbers
represent significant relationships at p < 0.01.
Parameter Dry season Wet season
F2N F5F F2N F5F
LAI 0.60 0.56 0.48 0.45
Mean stem height (m) 0.68 0.55 0.61 0.72
Stem density (stem/ha) 0.04 0.16 0.10 0.05
Basal area (m2/ha) 0.01 0.06 0.14 0.00
Mean DBH (cm) 0.06 0.02 0.01 0.13
405
Figure 2. Fine beam RADARSAT-1 backscattering coefficients, at two incident angles (F2N=40°; F5F=47°), vs. Leaf Area Index
(LAI) and mean stem height for 14 white mangrove plots located in a mangrove forest of the Mexican Pacific.
from forest canopies in the shorter wavelengths tributions. In contrast, the interactions of SAR
(e.g. K, X, C) is from the interactions with the with trunks and main branches are considered the
leaves and smaller canopy elements (i.e. volume main contributors of forest stand backscatter for
scattering). Consequently, little or no backscatter the longer SAR wavelengths (e.g. L and P). As a
should be expected from dead mangrove areas result, for such SAR wavelengths, the backscatter
(Figure 3) which contain few, if any, leaves (i.e. coefficients from the dead areas, where trunks
LAI $ 0), especially with the flooded regime of the remain, should be significant. This would be
mangroves which limits ground reflectance con- especially true in these flooded forests where the
406
Figure 3. Comparison of RADARSAT-1 F5F backscatter from a white mangrove forest with an enhanced optical IKONOS data false
colour composite (4, 3, 2). Note the large expanse of dead mangroves in the top portions of the images. Refer to Figure 1 for
geographic position.
double-bounce effect would result from the inter- SAR backscatter coefficients and other man-
actions between the standing water and the trunks grove structural parameters. For example, in these
which act as dihedral corner reflectors. studies, in addition to biomass, coefficients as high
Although LAI and the corresponding back- as 0.90 and 0.88 were recorded for basal area and
scatter coefficients were deemed significant, the mean DBH in relation to AIRSAR C-HH back-
data plotted in Figure 2 indicate that the discrep- scatter. The much higher spatial resolution of their
ancy between no leaf content (i.e. LAI $ 0) and SAR platform and, possibly, the greater range of
some leaf content (i.e. low LAI) is more apparent the parameters recorded in their study plots may
than between low LAI and high LAI for the RA- explain the discrepancy. For example, in this
DARSAT-1 data. This circumstance could explain study, although the LAI varied considerably
the lower coefficients of determination recorded among the sample plots, this was not the case for
for the wet season data, when the range of the LAI the basal areas and stem densities (Table 1).
increased dramatically. The limited ability to dis- Enhanced backscatter from flooded areas is
cern LAI values for the healthier mangroves may angle dependent and thus an important consider-
be due to saturation of the SAR signal occurring ation for mangrove areas. However, with regards
much earlier than that which might be expected to the examination of two incident angles, the
from longer SAR wavelengths (Dwivedi et al. results of this study indicate that the use of the
1999). Specifically, shorter wavelengths are known shallower incident angle (40°) only slightly im-
to be constrained in their ability to penetrate proves the relationship between LAI and the cor-
closed forest canopies. responding backscatter coefficients. At the fine
Unlike the previous work of others using air- beam mode, RADARSAT-1 data selection is
borne SAR (Mougin et al. 1999; Proisy et al. 2000; restricted to an incident angle range of approxi-
Proisy et al. 2002), the results of this study do mately 38°–47°. At the standard beam mode, the
not indicate significant relationships between range of incident angles available does improve
407
(24°–47°) but at the expense of a lower spatial anticipated that with the planned launch of
resolution (25 m). RADARSAT-2 the extraction of mangrove bio-
With regard to seasonality of the data, the physical parameter data from spaceborne SAR
coefficients of determination for LAI and SAR will improve significantly. The launch of RA-
backscatter coefficients were greater for the dry DARSAT-2 will ensure the continuity of all
season. LAI values for the plots did, however, vary existing RADARSAT-1 beam modes, which
quite dramatically as a result of the phenological would allow for long term change detection stud-
changes associated with the white mangroves of ies. Moreover, RADARSAT-2 will provide new
this study area. In this investigation no precipita- capabilities for improving mangrove forest struc-
tion events were recorded three days prior to or tural parameter extraction including the addition
during the acquisition of the wet season imagery. of an ultra fine beam mode (3 m resolution) and
Consequently, it was not possible to determine quad-polarization.
whether surface moisture, which increases the
dielectric constant of the leaves, would have
altered the relationship between LAI and the
Acknowledgements
backscatter coefficients.
J.M. Kovacs wishes to acknowledge financial sup-
port (grant # 249496-02) of the Natural Sciences
Conclusion
and Engineering Research Council of Canada. The
Canada Space Agency provided the RADARSAT-
The results of this investigation suggest that
1 data to J.M. Kovacs and F. Flores-Verdugo as
spaceborne SAR, specifically fine beam RADAR-
part of the Data for Research Use (DRU) program
SAT-1, can be used to some extent for extracting
(project # 02-04). The authors would also like to
biophysical parameter data from mangrove for-
extend their thanks to Lance P. Aspden, Francisco
ests. Specifically, in this investigation it was pos-
Flores de Santiago and Neil Latour for their
sible to discern dead white mangrove stands from
assistance in the field data sampling.
healthy ones due to a significant relationship be-
tween LAI and corresponding RADARSAT-1 fine
beam backscatter coefficients. The results also
References
indicate that a steeper incident angle and the
acquisition of SAR in the dry season may improve
Dwivedi R.S., Rao B.R.M. and Bhattacharya S. 1999. Mapping
this relationship. Although the coefficients of
wetlands of the Sundaban Delta and it’s environs using ERS-
determination were not as high as those deter-
1 SAR data. Int. J. Remote Sens. 20: 2235–2247.
mined by others using airborne SAR, the results of ´
Flores-Verdugo F.J., Gonzalez-Farias F., Ramirez-Flores O.,
this study would suggest that RADARSAT-1 Amezcua-Linares F., Yanez-Arancibia A., Alvarez-Rubio M.
could be used as an alternative method for moni- and Day J.W. 1990. Mangrove ecology, aquatic primary
´
productivity, and fish community dynamics in Teacapan–
toring mangrove forests, particularly for regular
Agua Brava lagoon–estuarine system (Mexican Pacific).
surveys of mangrove damage incurred by major
Estuaries 13: 219–230.
disturbances (e.g. hurricanes, pollution). The use Flores-Verdugo F.J., Gonzalez-Farias F., Blanco-Correa M.
of spaceborne SAR imagery can provide resource ´
and Nunez-Pasten A. 1997. The Teacapan-Agua Brava
managers with quick synoptic assessments of large Marismas Nacionales mangrove ecosystem on the Pacific
coast of Mexico. In: Kjerfve B., Drude L. and Diop E.H.S.
areas of mangrove at a known schedule and for
(eds), Mangrove Ecosystem Studies in Latin America and
regions where persistent cloud cover is the norm.
Africa. UNESCO, pp. 35–46.
Based on the results of this investigation, it is Green E.P., Mumby P.J., Alaisdair E.J., Clark C.D. and
recommended that further studies with RADAR- Ellis A.C. 1997. Estimating leaf area index of mangroves
SAT-1 data be conducted for other white man- from satellite data. Aquat. Bot. 58: 11–19.
Green E.P., Clark C.D., Mumby P.J., Edwards A.J. and
grove areas, particularly those with structural
Ellis A.C. 1998. Remote sensing techniques for mangrove
parameters that vary considerable from this study
mapping. Int. J. Remote Sens. 19: 935–956.
area. Moreover, it is suggested that investigations Hess L.L., Melack J.M. and Simonett D.S. 1990. Radar
of these data for other mangrove species be con- detection of flooding beneath the forest canopy: A review.
sidered. Given the outcome of this study, it is Int. J. Remote Sens. 11: 1313–1325.
408
Jensen J.R., Lin H., Yang Y., Ramsey E., Davis B.A. and Leckie D.G. and Ranson K.J. 1998. Forestry applications using
Thoemke C.W. 1991. The measurement of mangrove char- imaging radar. In: Henderson F.M. and Lewis A.J. (eds),
acteristics in Southwest Florida using SPOT multispectral Manual of Remote Sensing. 3rd ed. Vol. 2. Principles and
data. Geocarto Int. 2: 13–21. Applications of Imaging Radar. John Wiley & Sons, Tor-
Kovacs J.M. 1999. Assessing mangrove use at the local scale. onto, Canada, pp. 435–509.
Landscape Urban Plan. 43: 201–208. Mougin E., Proisy C., Marty G., Fromard F., Puig H.,
Kovacs J.M. 2000. Perceptions of environmental change in a Betoulle J.L. and Rudant J.P. 1999. Multifrequency and
tropical coastal wetland. Land Degrad. Dev. 11: 209–220. multipolarization radar backscattering from mangrove
Kovacs J.M., Wang J. and Blanco-Correa M. 2001a. Mapping forests. IEEE Trans. Geosci. Remote Sens. 37: 94–102.
mangrove disturbances using multi-date Landsat TM imag- Proisy C., Mougin E., Fromard F. and Karam M.A. 2000.
ery. Environ. Manage. 27: 763–776. Interpretation of polarimetric radar signatures of mangrove
Kovacs J.M., Blanco-Correa M. and Flores-Verdugo F. 2001b. forests. Remote Sens. Environ. 71: 56–66.
A logistic regression model of hurricane impacts in a man- Proisy C., Mougin E., Fromard F., Trichon V. and Karam
grove forest of the Mexican Pacific. J. Coastal Res. 17: 30–37. M.A. 2002. On the influence of canopy structure on the
Kovacs J.M., Flores-Verdugo F., Wang J. and Aspden L.P. backscatter of mangrove forests. Int. J. Remote Sens. 23:
2004a. Estimating leaf area index of a degraded mangrove 4197–4210.
forest using high spatial resolution satellite data. Aquat. Bot. Ramsey E. and Jensen J.R. 1996. Remote sensing of mangrove
80: 13–22. wetlands: relating canopy spectra to site-specific data. Pho-
Kovacs J.M., Malczewski J. and Flores-Verdugo F. 2004b. togramm. Eng. Remote Sens. 62: 939–948.
Examining local ecological knowledge of hurricane impacts Townsend P.A. 2002. Estimating forest structure in wetlands
in a mangrove forest using an Analytical Hierarchy Process using multitemporal SAR. Remote Sens. Environ. 79: 288–
(AHP) approach. J. Coastal Res. 20: 792–800. 304.
Kovacs J.M., Wang J. and Flores-Verdugo F. 2005. Mapping Ulaby F.T., Moore R.K. and Fung A.K. 1986. Microwave
mangrove leaf area index at the species level using IKO- Remote Sensing: Radar Remote Sensing and Surface Scat-
NOS and LAI-2000 sensors. Estuarine, Coastal Shelf Sci. tering and Emission Theory, Vol. 2. Artech House Inc.,
62: 377–384. Norwood, MA, USA.