Environ Monit Assess (****) ***:*** ***
Remote sensing of aquatic vegetation:
theory and applications
Thiago S. F. Silva Maycira P. F. Costa John M. Melack Evlyn M. L. M. Novo
Received: 9 February 2007 / Accepted: 24 May 2007 / Published online: 26 June 2007
Springer Science + Business Media B.V. 2007
Abstract Aquatic vegetation is an important com- successful results. The present paper reviews the
ponent of wetland and coastal ecosystems, playing theoretical background and possible applications
a key role in the ecological functions of these of remote sensing techniques to the study of
environments. Surveys of macrophyte communi- aquatic vegetation.
ties are commonly hindered by logistic problems,
Keywords Remote sensing Macrophytes
and remote sensing represents a powerful alter-
native, allowing comprehensive assessment and Aquatic vegetation
monitoring. Also, many vegetation characteristics
can be estimated from re ectance measurements,
such as species composition, vegetation struc- Introduction
ture, biomass, and plant physiological parameters.
However, proper use of these methods requires Aquatic plants are an important component of
an understanding of the physical processes behind wetland and coastal ecosystems, playing a key role
the interaction between electromagnetic radiation in ecological function (Marion and Paillison 2003;
and vegetation, and remote sensing of aquatic Junk 1997). Many macrophyte communities are
plants have some particular dif culties that have characterized by high growth rates, rapid biomass
to be properly addressed in order to obtain accumulation and, in seasonal ecosystems such as
wetlands and oodplains, by a tight connection
with the ooding pattern of the landscape (Junk
T. S. F. Silva (B) M. P. F. Costa 1997). These plants have a large capacity to absorb
Department of Geography, University of Victoria,
harmful substances and pollutants, and can be
P.O. Box 3050 STN CSC,
indicators of the eutrophic status of a water body
Victoria, BC V8W 3P5, Canada
(Onaindia et al. 1996).
e-mail: abpyyu@r.postjobfree.com
Surveys of macrophyte communities are com-
J. M. Melack
monly hindered by limited accessibility (Vis et al.
Bren School of Environmental Science
2003). Hence, remote sensing is a valuable tool for
and Management, University of California,
assessment of macrophyte stands and associated
Santa Barbara, CA 93106, USA
biophysical and ecological parameters. The use
E. M. L. M. Novo
of remotely sensed images allows multitemporal
Instituto Nacional de Pesquisas Espaciais,
studies and provides comprehensive information
Avenida dos Astronautas, 1758, CEP 12227-10,
from surrounding areas. With the advance of
S o Jos dos Campos, SP, Brazil
132 Environ Monit Assess (2008) 140:131 145
sensor technology and processing techniques, veg- distinguish between submerged and oating or
etation characteristics such as species composi- emergent plants, as these factors act differently in
tion, leaf area index, biomass, photosynthetically each case.
active radiation absorbed and even chemical com-
position can be determined by analysis of radio- Spectral behavior of submerged vegetation
metric data (Tilley et al. 2003; Pe uelas et al.
1993). The green region of the spectrum is considered as
Aquatic plants and their properties, however, the most suitable for sensing submerged macro-
are not as easily detectable as terrestrial vegeta- phytes, followed by the red and red edge regions.
tion. Proper understanding of the physical interac- Several studies highlight the same narrow spec-
tion between electromagnetic energy and both the tral regions as optimal for submerged macrophyte
vegetation and its environment, as well as careful discrimination (Table 1). This convergence indi-
application of pre-processing steps prior to the cates that common underlying conditions such as
analysis of remotely sensed data are requirements pigment concentration and cellular structure are
for obtaining successful results. Most remote sens- responsible for the main differences among
ing techniques have been employed to assess macrophyte species. Also, the green region pro-
macrophyte properties: eld spectrometry, aerial vides greater light penetration in waters with
photography, aerial/orbital multispectral systems, higher concentrations of suspended and dissolved
hyperspectral systems, microwave sensors, digital material (Kirk 1994).
airborne videography and sonar systems. In the Water strongly absorbs the electromagnetic
present paper, the theoretical background and radiation in the optical spectral region, resulting
applications of remote sensing to aquatic plants in signi cant dampening of the radiometric sig-
are examined, in order to provide a comprehen- nal. Because of this, re ectance measurements for
sive perspective on its present and future capabil- submerged species are usually very low, on the
order of 10 10 2 (Pinnel et al. 2004; Dierssen and
ities and needs.
Zimmerman 2003; Fyfe 2003; Han and Rundquist
2003; Heege et al. 2003; Paringit et al. 2003;
Optical remote sensing
Everitt et al. 1999). In the absence of water (i.e.
The principles behind aquatic vegetation spectral laboratory conditions), higher re ectance values
characteristics are the same as behind its ter- can be obtained (Paringit et al. 2003; Armstrong
restrial counterparts. At the leaf level, presence 1993). The main challenge of remote sensing of
and concentration of leaf pigments determine the submerged aquatic plants is thus to isolate plant
response in the visible region of the spectrum, and signal from the overall water column interference.
leaf morphology and water content are the main Due to the reduced magnitude of the signal, a
factors acting on the infrared wavelengths (Fig. 1). careful and adequate correction of atmospheric
At the individual level, biophysical factors such effects is necessary prior to the analysis of sub-
as leaf distribution, leaf density and orientation, merged vegetation radiometry derived from air-
and overall canopy structure are important. Ver- borne and orbital data. This correction is usually
tically oriented plants or reduced leaf area offer obtained through (1) image-based procedures,
less available surface to interact with the down- which employ pixels with known spectral char-
welling radiation, while highly branched canopies acteristics to correct for atmospheric noise; and
and broadleaved plants have a more effective (2) model based procedures, which use radiative
re ective area (Williams et al. 2003). At the transfer equations to model atmospheric condi-
community level, plant biomass and density are tions and the radiation pathway, and then pre-
also important variables. Although the spectral dict the expected surface re ectance for these
response of aquatic vegetation resembles that of conditions.
terrestrial vegetation, the submerged or ooded Image-based methods usually consist of Dark
conditions introduce factors that alter its overall Object Subtraction (DOS) (Chavez 1988), which
spectral characteristics. It is therefore useful to uses objects with near zero re ectance to
Environ Monit Assess (2008) 140:131 145 133
Fig. 1 Leaf-radiation
interactions at
microscopic level. Arrow
thickness is proportional
to the magnitude of
radiation uxes
estimate and correct for atmospheric haze, or Radiative transfer models are based on param-
some of its improved versions, which also correct eterization of atmospheric conditions. However,
atmospheric transmittance (Chavez 1996). The these parameters are seldom available for speci c
major drawback associated with this approach is locations, and the model applications thus rely on
that often the existing dark objects within a scene estimated or average parameters, which can result
are in fact the water bodies, and a poor choice in erroneous corrections (Song et al. 2001). Com-
of haze values can actually cancel out the water- mon methods include the MODTRAN (Berk et al.
leaving radiance. 1999), and 6S algorithms (Vermote et al. 1997).
Table 1 Appropriate spectral regions for discriminating submerged macrophyte species, as suggested by different authors
Wavelength (nm) Plant species
Williams et al. (2003) 574 / 681 Vallisneria americana, Myriophyllum spicatum
Fyfe (2003) 530 580 / 520 530 / 580 600 Zostera capricorni, Posidonia australis, Halophila ovalis
Pinnel et al. (2004) 550 / 656 Chara spp., Naja marina, Nitellopsis obtusa, Potamogeton spp.
Han and Rundquist (2003) 538 / 706 Ceratophyllum demersum
134 Environ Monit Assess (2008) 140:131 145
Ultimately, the choice of methods should be were estimated with the use of IKONOS satellite
based on the amount and reliability of available imagery, and compared to eld measurements.
atmospheric data. If available, then model-based Finally, another important source of varia-
procedures are the logical choice. If not, image- tion for submerged vegetation re ectance is the
based procedures are expected to yield more accu- presence of epibiont organisms, especially epi-
rate results and minimize error introduction; refer phytes, which can cover the plant surface. Fyfe
to Zilioli and Brivio (1997) and Song et al. (2001) (2003) showed signi cant differences between the
for further information. re ectance of cleaned and fouled leaves, in all
Apart from water, the presence of optically wavelengths, for different macrophyte species.
active material (i.e. plankton, sediment, organic These effects were more signi cant between 570
molecules) affects the scattering and absorption of and 590 nm (Fyfe 2003; Williams et al. 2003). The
radiation (Han and Rundquist 2003; Kirk 1994). presence of epiphytes can also smooth the spec-
In addition, bottom re ectance is a factor to be tral curve, reducing the difference in re ectance
considered when interpreting the radiometric sig- between wavelengths and masking subtle spectral
nal of macrophyte beds in shallow waters. features (Armstrong 1993).
Ackleson and Klemas (1987) used a single-
scattering volume re ectance model to represent Spectral behavior of emergent species
the interaction between the three main compo-
nents of the signal from submerged vegetation With the absence of water attenuation, the av-
(water, bottom, plants). Using this physical repre- erage re ectance of emergent macrophytes is
sentation and a set of pre-determined, represen- usually higher than the observed for submerged
tative parameter values, the authors showed that, plants. Values can range between 0.02 and 0.1
in shallow depths, the overall re ectance signal in the visible spectrum (with an usual peak in
is determined mainly by the vegetation density, the green region) and 0.06 0.65 in the near in-
assuming that bottom re ectance is constant and frared (Tilley et al. 2003; Jakubauskas et al. 2000;
differs signi cantly from the vegetation. As depth Malthus and George 1997; LaCapra et al. 1996;
increased, dominance of re ectance shifted to the Pe uelas et al. 1993; Best et al. 1981).
water column components. Hence, Ackleson and The presence of ooding, however, introduces
Klemas (1987) suggested that incorporating depth variability in re ectance values due to the mixing
information into the classi cation method can of plant and water signals (Malthus and George
reduce the in uence of water column variation. 1997). This mixing usually results in a decrease
Armstrong (1993) accomplished this for Landsat in total re ected radiation, especially in the near
TM visible bands through a linearization proce- to mid infrared regions where water absorption is
dure developed by Lyzenga (1978), which yields stronger. The intensity of such effect will be deter-
depth invariant bands. mined by vegetation density and canopy structure
Water column optical models also include (Jakubauskas et al. 2000) (Fig. 2), as well as by
bathymetric information as one of the vari- the nature of the water signal. As noted before,
ables used to correct water and bottom effects the latter is a function of the amount and nature
(Dierssen and Zimmerman 2003; Heege et al. of suspended materials and depth of the water
2003). Paringit et al. (2003) attempted to develop column, plus substratum composition for shallow
a seagrass canopy model to predict the spectral depths.
response of submerged macrophytes in shallow Physiological status of vegetation can be
areas. The model considered not only the effects another source of variation in plant spectral signa-
of the water column through radiative transfer tures. Best et al. (1981) demonstrated that a single
modeling, but also viewing and illuminating con- species, in different phenological stages, exhibited
ditions, leaf and bottom re ectance, leaf area in- signi cant variation in its re ectance. In addition,
dex and the vertical distribution of biomass. With physiological stress can lead to spectral variability
model inversion, plant coverage and abundance (Tilley et al. 2003; Pe uelas et al. 1997; Bajjouk
Environ Monit Assess (2008) 140:131 145 135
Benton and Newman 1976; Edwards and Brown
1960). Although most airphoto analyses rely on
visual interpretation, plant species can be often
discriminated due to its high spatial resolution
(Moore et al. 2003; Schulz et al. 2003). Digitiza-
tion of aerial photography may allow the appli-
cation of computer aided classi cation algorithms
(Valta-Hullkonen et al. 2003; Marshall and Lee
1994). Aerial photography, however, often lacks
the capacity to record in multiple spectral bands,
a hindrance especially signi cant for submerged
vegetation.
Fig. 2 Effect of plant density in the spectral pro le
Digital multispectral airborne systems can pro-
of Nuphar polysepalum. On higher densities, the spec-
vide high spatial resolution coupled with an
tral curve is similar to what is expected from a veg-
etated surface. As it decreases, re ectance values are increased number of spectral bands. Its spectral
reduced, especially in the 700 1,000 nm region, and the
re nement can support more accurate quantita-
overall response approaches the one for a water sur-
tive analysis and classi cation of data (Malthus
face (Adapted from Jakubauskas et al. (2000): Interna-
and George 1997). Nevertheless, as data acquisi-
tional Journal of Remote Sensing, Taylor & Francis Ltd,
http://www.tandf.co.uk/journals) tion from these sensors can be more expensive
than the acquisition of aerial photography, the
latter remain as a common data source when
et al. 1996; Pe uelas et al. 1993; Best et al. 1981). information at high spatial resolution (meter to
Stress usually implies alterations on biochemical sub-meter range) is required (Maheu-Giroux and
status and morphological characteristics, which in de Blois 2005).
turn determine the spectral response in the differ- Another alternative to airphoto surveying is
ent regions of spectra. Factors such as chlorosis, the use of videographic systems, which employ
desiccation or disease can be detected in the spec- a digital video camera instead of photographic
tral signature of plants. sensors. These devices can attain ne spatial res-
Because of the wide range of re ectance val- olutions (sub-meter), and by using lters and
ues, the spectral signature of emergent aquatic multiple cameras, acquire simultaneous images in
vegetation often overlaps the signals from terres- different spectral bands. Videography has been
trial vegetation, water and occasionally soil. This employed with success to map both emergent and
variability can lead to poor results from simple submerged vegetation (Sprenkle et al. 2004; Hess
automated classi cation procedures and hinder et al. 2002; Everitt et al. 1999).
visual interpretation (Silva 2004; Ozesmi and In recent years, numerous studies employ-
Bauer 2002; Best et al. 1981). In such cases, the ing hyperspectral imaging sensors have been
use of alternative image classi cation algorithms performed (Pinnel et al. 2004; Dierssen and
such as decision tree (Baker et al. 2006) or neural Zimmerman 2003; Thomson et al. 2003; Williams
network classi ers (Filippi and Jensen 2006) may et al. 2003; Anstee 2001; Alberotanza 1999;
help. Thomson et al. 1998; Bajjouk et al. 1996;
Lacapra et al. 1996; Zacharias et al. 1992).
These sensors offer good spatial resolution (about
Optical remote sensing applications to aquatic
1 4 m) and the capacity of recording full spectra
vegetation studies
for each pixel. Such richness of data is of special
interest to the study of submerged vegetation,
Aerial photography was the rst remote sensing
since the overall signal is low and an acceptable
method to be employed for studying and map-
degree of discrimination can be only obtained by
ping plant stands, and early studies date back to
the examination of subtle spectral characteristics.
the 1960s and 1970s (Austin and Adams 1978;
136 Environ Monit Assess (2008) 140:131 145
Speci c features in the re ectance curves can 70 and 96% can be achieved (Pasqualini et al.
often be related to physiological and biophysical 2005; Sawaya et al. 2003; Valta-Hullkonen et al.
parameters, allowing the indirect estimation of 2003; Anstee 2001; Everitt et al. 1999; Malthus and
these. The application of hyperspectral imagery is George 1997; Bajjouk et al. 1996). High overall
one of the most promising uses of remote sensing accuracy can be obtained with the correct choice
to the study of aquatic vegetation. Speci cations and application of mapping techniques. Another
of some of the more widely used hyperspectral inherent advantage of satellite imagery is the reg-
sensors are listed in Table 2. ular temporal acquisition, allowing utilization of
Satellite systems have also been successfully time series to analyze seasonal patterns (Silva
applied to the study of aquatic vegetation. 2004; Jensen et al. 1993) or landscape changes
Although the spatial resolution of these systems (Moore et al. 2003; Jensen et al. 1995). Instru-
is, in most cases, incapable of discriminating ments such as the Landsat series provide almost
aquatic vegetation at the species level (Jensen 30 years of imagery, which is a valuable and
et al. 1993), satellite imagery is useful for map- unparalleled source of temporal data.
ping macrophytes communities. Landsat MSS Remote sensing can be employed as a tool for
and TM images have been employed for map- estimating biophysical measures. Plant biomass
ping submerged (Zhang 1998; Armstrong 1993; can be estimated by means of spectral data, mainly
Ackleson and Klemas 1987) and emergent vege- through the use of regression analysis, with bands
tation. Images with spatial resolutions higher than or band combinations (ratios, indexes) as predic-
Landsat have also been applied to both vege- tor variables. It is important to note that with
tation types, e.g., SPOT data (Pasqualini et al. increases in biomass, the relationship between the
2005; Jensen et al. 1995, 1993, 1986) and IKONOS spectral signal and the actual biomass approaches
images, with 1m resolution (Sawaya et al. 2003). an asymptote (Pe uelas et al. 1993).
Coarser resolution data have been proved useful Zhang (1998), using the rst and second princi-
as well, e.g. the Indian IRS-LISS I, with ground pal components of a PCA transformed TM image,
resolution of 72.5 m (Pal and Mohanty 2002; estimated the biomass of submerged stands in the
Chopra et al. 2001), and MODIS images (250 and Honghu Lake (China), obtaining a coef cient of
determination of R2 = 0.85. Also, submerged veg-
500 m resolution) have been shown to be able
to map macrophyte occurrence after the use of etation biomass has been estimated by Armstrong
spatial resolution enhancement techniques (Silva (1993), using depth normalized TM images and
obtaining an overall R2 = 0.79. This high degree
2004).
The usual application of remote sensing of agreement, considering the radiometric (8 bit),
imagery is to produce cover maps for aquatic spectral (few, broad bands) and spatial (30 m) lim-
plants, in general or for different populations or itations of such images, suggests that even better
communities. Considering both airborne and or- results could be acquired with the use of systems
bital imaging sensors, accuracies ranging between with more resolving power. Other biophysical
Table 2 Some of the most widely used hyperspectral sensors currently in operation
Sensor Number Spectral interval Bandwidth Spatial Platform Manufacturer
of bands (excl. thermal) resolution
MIVIS 102 430 2,500 nm 8 20 nm Variable Airborne Daedalus Enterprises
CASI-2 19 288 400 1,050 nm 1.9 nm Variable Airborne ITRES research
PHILLS 128 380 1,000 nm 0.5 3 nm Variable Airborne Naval Research Laboratory, US
HyMap 100 200 450 2,500 nm 10 20 nm Variable Airborne Integrated Spectronics
AVIRIS 224 400 2,500 nm 10 nm Variable Airborne Jet Propulsion Lab
Hyperion 220 400 2,500 nm 10 nm 30 m NASA EO-1 TRW Inc.
Environ Monit Assess (2008) 140:131 145 137
indexes can also be estimated through the use of wavelengths tend to have deeper canopy pene-
remote sensing, such as percentage cover (Heege tration and less sensitivity to smaller biophysical
et al. 2003; Pinnel et al. 2004) and Leaf Area In- variations. In addition to wavelength, every radar
dex (Dierssen and Zimmerman 2003). These mea- system has de ned polarizations for sending and
sures represent important ecological variables, receiving the radiation pulse, either vertically (V)
and are often employed as inputs to ecosystem or horizontally (H). Same-polarization systems
models. Physiological characteristics can also be are usually referred as HH and VV systems, and
inferred from the spectral response of macro- cross-polarization systems as HV or VH. Differ-
phytes, due to alterations in optically active sub- ent polarizations, as well as ratios or differences in
stances. Examples are chlorophyll concentration polarizations can highlight speci c characteristics
(Pe uelas et al. 1993), photosynthetic ef ciency for some types of targets (Lewis and Henderson
(Pe uelas et al. 1997, 1993), chemical composition 1998).
(LaCapra et al. 1996) and environmental pres- Many of the current applications of SAR sys-
sures (Tilley et al. 2003). tems are derived from satellite-borne sensors,
such as the Japanese Earth Resources Satellite 1
(JERS-1), the Canadian Radarsat 1, and the Euro-
pean systems Earth Resources 1 and 2 (ERS-1 and
Synthetic aperture radar (SAR) systems
ERS-2) and Envisat ASAR. Important research
The use of SAR data have been long acknowl- has been also generated by data collected from the
edged as a valuable tool for studying wetlands. SIR-C/X-SAR instrument own on a space shuttle
In the microwave range, differences in the signal in 1994, and applications of airborne SAR systems
recorded from dry and ooded vegetation allow are also signi cant (Table 3).
the mapping of ooding extent (Costa 2004; Hess To understand the radiometric responses in
et al. 1995). In addition, numerous studies have SAR data, it is necessary to realize that radar
shown that SAR images can be utilized to study sensors are side-looking systems, meaning that the
aquatic vegetation (Costa 2005; Kasischke et al. electromagnetic pulse hits the surface in a sub-
2003; Moreau and Le Toan 2003; Costa et al. nadir angle. For this reason, it is expected that,
2002; Novo et al. 2002; Noernberg et al. 1999; for smooth plain surfaces, most of the radiation
Le Toan et al. 1997; Pope et al. 1997; Kasischke is re ected specularly and does not return to the
and Borgeau-Chavez 1997; Hess et al. 1995). sensor. With increasing surface roughness and
Synthetic Aperture Radar data offers infor- addition of volume components, such as vegeta-
mation about canopy biophysical characteristics tion, the backscattered radiation increases (Lewis
and dielectric properties (a proxy for water con- and Henderson 1998).
tent), instead of biochemical and morphologi- The overall radar signal from aquatic vege-
cal features. The longer microwave wavelengths tation is composed primarily of the volumetric
penetrate into the canopy, resulting in a volu- backscatter from the canopy elements, the sur-
metric signal. Coupled with its active source of face backscatter from the ground surface, and the
energy, image acquisition can be performed re- double-bounce interaction from radiation that is
gardless of weather conditions or time of day. forward scattered from the surface but bounces
Such capability is valuable as wetland environ- off the canopy elements and returns to the sensor
ments frequently occur in cloudy locations. How- (Fig. 3).
ever, as radar wavelengths do not penetrate into The geometry of the canopy, moisture content
water, these systems can only be applied to emer- and the presence of strongly vertically or hori-
gent macrophytes. zontally oriented features may affect the result-
Radar systems operate in speci c regions of the ing signal at some wavelength and polarization
electromagnetic spectrum, and radar bands are combinations. For instance, dense, tall (1.5 m
usually coded by a single letter. The most common or more), vertically-oriented wetland herbaceous
bands used are X (3 cm wavelength), C (5.6 cm), plants show double-bounce in L band (HH and
S (10 cm), L (23 cm) and P (75 cm). Longer VV), and even C-HH at low incidence angles
138
Table 3 Past, current and planned spaceborne radar systems (expected launch date in parenthesis)
Sensor Bands Polarization Incid. angle Spat. ses. (m) Swath width (km) Orbit cycle (days) Agency
ERS-1a C VV 23 28 100 35 European Space Agency (ESA)
JERS-1a L HH 38 18 74 44 National Space Development
Agency of Japan (NASDA) -
currently JAXA
SIR-C/X-SARa L, C, X Full Pol. **-**-**-**-**-** Jet Propulsion Lab, USA
(L,C)
VV (X)
ERS-2 C VV 23 28 100 35 European Space Agency (ESA)
Radarsat-1 C HH 10-59-10-100-** 500 24 Radarsat International
Envisat ASAR C HH, VV, 15 45 30 1,000-**-***-** European Space Agency (ESA)
HV, VH
Radarsat-2 (2007) C Full Pol. 10-60-3-100-**-*** 24 Radarsat International
Alos PALSAR L Full Pol. 8-60-10-100-** 350 46 Japan Aerospace Exploration
Agency (JAXA)
MAPSAR (2010) L HH, VV, **-**-*-**-**-** 7 Brazilian National Institute
HV, VH for Space Research (INPE)/
German Aerospace Center
(DLR)
a Currently deactivated
Environ Monit Assess (2008) 140:131 145
Environ Monit Assess (2008) 140:131 145 139
Fig. 3 Schematic
representation of the
scattering mechanisms at
C and L bands for aquatic
vegetation (Adapted
from Costa 2004)
(Costa et al. 2002; Pope et al. 1997). Double- higher incidence angles (Hess et al. 1990; Ford
bounces are caused by the interaction of the and Casey 1988). At lower angles, the pathway of
radiation with the stem/trunk, followed by a the incident wave through the canopy is minimal;
change in direction towards the surface (water) therefore, the radiation is less attenuated by the
and a strong bounce back towards the radar an- canopy.
tenna (dihedral corner re ector behavior). The For less dense herbaceous plants in ooded
inverse is also possible. wetlands, backscattering values are not as high
The characteristics from both the plants and as those observed for high density stands, due
the sensor are needed to explain double-bounce to the increase in the forward scattering of
interaction. The combination of a long wave- water patches (Pope et al. 1997). For herba-
length (L band), horizontal polarization (HH), ceous plants (varying densities), at either C-VV or
and steep incidence angle allows higher penetra- cross-polarized and low incidence angles, double-
tion of the radiation through the canopy. At L bounces were not observed, but signals related
band plant leaves are quasi-transparent; hence to canopy volume-scattering (Pope et al. 1997;
the radiation interacts mostly with the stem and Kasischke and Borgeau-Chavez 1997; Hess et al.
the underlying water. For the same con gura- 1995) were observed.
tion of radiation/target, but with VV polarization, Volume-scattering mechanisms are character-
the interaction is mostly with the upper canopy ized by the interaction of the radiation within the
(Ulaby et al. 1986). Double-bounce mechanisms vegetation canopy, i.e. leaves and stems. The radi-
are enhanced for radiation at lower incidence ation is scattered by the elements in all directions
angles (i.e. closer to nadir) when compared with within the volume, and the resulting backscatter-
140 Environ Monit Assess (2008) 140:131 145
ing towards the antenna is not as strong as it in overall discrimination (Proisy et al. 2000; Hess
is for double-bounce mechanisms (Ulaby et al. et al. 1995). The new generation of full polarimet-
1982). ric sensors, for example, does not suffer from this
The backscatter from aquatic vegetation stands limitation (Table 3). Multiple bands and/or po-
usually has low values, in all wavelengths and larization are also more commonly found among
polarizations. The low return is caused mainly airborne SAR systems.
by forward scattering from the water surface For macrophyte mapping, accuracies ranging
and the attenuation of the signal from the from 65 to 97% can be achieved (Costa 2004; Hess
canopy. Some controversy exists about the factors et al. 2003, 1995; Novo et al. 2002), and species can
affecting aquatic vegetation signal in different be differentiated in some degree, such as grasslike
con gurations and about which is the most versus broadleaved (Noernberg et al. 1999).
appropriate con guration to remotely sense Another promising application for radar remote
macrophytes. Costa et al. (2002) showed that sensing is biomass estimation. Studies show re-
a combination of L and C band signal was lationships between stand biomass and radar
backscatter ranging from R2 = 0.59 to 0.78
sensitive to stand height and biomass, while
C band alone responded only to the latter, (Moreau and Le Toan 2003; Costa et al. 2002;
and presented a lower signal saturation value. Novo et al. 2002). Saturation values range from
470 g m 2 to 2000 g m 2 of above water biomass,
Rosenthal et al. (1985), however, suggested that
C band should be more sensitive to plant height depending on community characteristics (Moreau
than biomass. For a more comprehensive dis- and Le Toan 2003; Costa et al. 2002).
cussion of the microwave radiometric behavior To overcome the issue of signal saturation,
of aquatic vegetation, the reader is suggested radar interferometry has also been used as an
to refer to Kasischke et al. (2003); Costa et al. alternative to backscatter signal analysis. In-
(2002); Noernberg et al. (1999); Pope et al. (1997); terferometry is a technique where two radar
Kasishcke and Borgeau-Chavez (1997). images taken at different locations are used to
Overall, the total backscattering from wetland map ground elevation (topography). Each pixel
herbaceous plants is dependent on the interaction at the radar scenes contains not only amplitude
of the microwave energy with both the canopy and but also the phase information, corresponding
the canopy-ground. Not only plant characteris- to the distance between platform and a given
tics, such as density, distribution, orientation, leaf place at the Earth s surface. The phase differences
shape, dielectric constant, height and components between these two images are used to derive
of the canopy, but also the sensor parameters precise information on surface height (Lu et al.
(polarization, incidence angle and wavelength) 2007). Vegetation height can be also determined
play an important role in determining the amount by interferometry, and later on be used as a proxy
of radiation backscattered toward the radar an- for biomass determination (Simard et al. 2006;
tenna. Due to this multitude of factors, visual Dutra et al. 2007; Santos et al. 2004).
interpretation usually requires more training and A somewhat less developed application of
familiarity with radar imagery than optical data. radar is the merging of both optical and radar
One of the main hindrances of spaceborne data by image fusion techniques. Since the infor-
radar systems is that most have a single band/ mation content in each one differs, there is low
polarization con guration, reducing the data redundancy when joining these sources, permit-
available for an accurate identi cation of macro- tin better mapping and discriminative results.
phyte stands (Hess et al. 2003). This limitation can At usual land cover mapping, accuracies can be
be overcome with the combination of different increased by as much as 10% by optical-radar
satellite imagery (Costa et al. 2002) or the use of fusion (Haack and Bechdol 2000). For aquatic
textural and contextual measures (Simard et al. macrophytes, the fusion between Radarsat-1
2000, 2002; Noernberg et al. 1999). Also, the use and Landsat TM allowed species-level discrimi-
of multitemporal, multi-incidence angle or multi- nation of macrophyte stands (Graciani and Novo
polarization data could offer some improvement 2003).
Environ Monit Assess (2008) 140:131 145 141
Other remote sensing systems tide levels both during and between ights must
be acknowledged, as it introduces signi cant mea-
Optical and radar remote sensing together com- surement errors (Brennan and Webster 2006).
prise the vast majority of systems and applica- The overall precision of the system in use is
tions. However, other methods can also provide a factor that must be considered, as vegeta-
valuable information about macrophyte commu- tion heights shorter than the minimum mea-
nities. For submerged vegetation mapping, suc- surable return difference cannot be properly
cessful results have been obtained by the use determined (Hopkinson et al. 2006). For the
of side-scan (Pasqualini et al. 2005) and multi- same reason, proper calibration and validation of
beam sonar systems (Komatsu et al. 2003). Multi- LiDAR heights from ground truth is paramount.
beam systems offer the advantage of generating Nonetheless, differences in the signal from veg-
three-dimensional information, including vertical etated and non-vegetated areas can still be use-
height distribution, and allowing visualization of ful for thematic classi cation, as shown by Rosso
the community structure. et al. (2006) for Spartina spp.
Recently, airborne LiDAR(Light Detection Wang and Philpot (2007) applied bathymetric
and Ranging) systems have been applied with LiDAR to map submerged vegetation. Again, the
success to the study of aquatic vegetation. These effect of the water column on the LiDAR signal is
sensors employ a high-frequency laser pulse, us- the main source of interference to be dealt with.
ing differences between the return time of each For bathymetric systems, green wavelengths are
beam to derive height and terrain information and used instead of infrared, as they offer the best
produce 3-D datasets. The accuracy of LiDAR trade off between water absorption and scattering
systems is usually very high, attaining meter to due to suspended material (Wang and Philpot
sub-meter spatial resolution and less than 0.5m 2007).
vertical accuracy (Brennan and Webster 2006;
Rosso et al. 2006; Hopkinson et al. 2005). These
systems were initially utilized for the generation Conclusions
of digital terrain models, but vegetation mapping
and estimation of biophysical parameters have The use of remote sensing for studying aquatic
been successful (Kotchenova et al. 2004; Maltamo vegetation is well established. From the ear-
et al. 2004; Patenaude et al. 2004; Popescu et al. lier mapping applications, employing analog aer-
2002). ial photography and visual interpretation, to the
As with other optical systems, a major factor use of modern digital high resolution systems
affecting the LiDAR response is water absorp- and complex automated classi cation algorithms,
tion. As most LiDAR systems operate in the there are many opportunities and advantages in
infrared region, saturated soils and free water sur- applying remote sensing techniques to obtain a
faces will dampen the returning signal. The result- synoptic view of macrophyte communities and its
ing reduction on the number of returns from the properties. Plant cover and distribution, biomass
substratum then affects the proper determination and other biophysical and physiological parame-
of canopy heights (Hopkinson et al. 2005). On ters can be estimated from eld spectral data, or
the other hand, signal penetration in the canopy by images. This information can then be used for
may generate the same problem; if canopy is environmental assessment and modeling, and for
sparse or too vertically oriented, lesser returns are better understanding of the ecological dynamics of
expected from the top elements, thus underesti- aquatic plant communities.
mating height. Hopkinson et al. (2005) studied Among the new developments of remote sens-
both aquatic and terrestrial grasses, and found ing science, the use of hyperspectral imagery
that these factors combined resulted in a mean appears to be a very promising tool for studying
difference of 53% between LiDAR estimations aquatic vegetation in the present and near future.
and ground measured height, against only 33% for Many airborne hyperspectral systems are avail-
terrestrial plants. In tidal systems, differences in able nowadays (AVIRIS, CASI, HyMap), and
142 Environ Monit Assess (2008) 140:131 145
orbital hyperspectral sensors are becoming avail- Baker, C., Lawrence, R., Montagne, C., & Patten, D.
(2006). Mapping wetlands and riparian areas us-
able, such as NASA Hyperion. The coupling of
ing Landsat ETM+ imagery and decision-tree-based
spatial data with rich spectral information allow
models. Wetlands, 26(2), 465 474.
better treatment of the main problems associated Benton, A. R., & Newman, R. M. (1976). Color aerial
with the usual multispectral systems, and provide photography for aquatic plant monitoring. Journal of
Aquatic Plant Management, 14, 14 16.
more detailed and sensitive information. Medium
Berk, A., Anderson, G., Bernstein, L., Acharya, P.,
resolution sensors, such as MODIS, MERIS and
Dothe, H., Matthew, M., et al. (1999). MODTRAN4
SPOT-Vegetation can provide useful information radiative transfer modeling for atmospheric correc-
for regional and global scale studies. Although tion. In Proceedings of SPIE The International So-
ciety for Optical Engineering (Vol. 3756, pp. 348 353).
SAR orbital sensors were restrained to only a
Best, R. G., Wehde, M. E., & Linder, R. L. (1981). Spectral
few bands and polarizations, new multi-polarized
re ectance of hydrophytes. Remote Sensing of Envi-
and full polarimetric systems are currently avail- ronment, 11, 27 35.
able (Envisat ASAR, ALOS Palsar) or expected Brennan, R., & Webster, T. L. (2006). Object-oriented land
cover classi cation of lidar-derived surfaces. Canadian
to become operational in the upcoming years
Journal of Remote Sensing, 32(2), 162 172.
(Radarsat-2, MAPSAR). This type of informa-
Chavez Jr., P. S. (1988). An improved dark-object subtrac-
tion, especially if combined with optical data, can tion technique for atmospheric scattering correction of
supply a good set of data for the study of emergent multi-spectral data. Remote Sensing of Environment,
24, 459 479.
macrophytes. Remote sensing is a powerful tool
Chavez Jr., P. S. (1996). Image-based atmospheric cor-
to be considered when studying large scale phe-
rections revisited and improved. Photogrammetric
nomena in aquatic vegetation communities, and is Engineering and Remote Sensing, 62(9), 1025 1036.
capable of delivering information unmatched by Chopra, R., Verma, V. K., & Sharma, P. K. (2001).
Mapping, monitoring and conservation of Haruke
any other surveying techniques.
wetland ecosystem, Punjab, India, through remote
sensing. International Journal of Remote Sensing,
22(1), 89 98.
Costa, M. (2005). Estimate of net primary productivity of
aquatic vegetation of the Amazon oodplain using
References
Radarsat and JERS-1. International Journal of Remote
Sensing, 26(20), 4527 4536.
Ackleson, S. G., & Klemas, V. (1987). Remote sensing of Costa, M. P. F. (2004). Use of SAR satellites for mapping
submerged aquatic vegetation in lower Chesapeake zonation of vegetation communities in the Amazon
bay: A comparison of Landsat MSS to TM imagery. oodplain. International Journal of Remote Sensing,
Remote Sensing of Environment, 22, 235 248. 25(10), 1817 1835.
Alberotanza, L., Brando, V. E., Ravagnan, G., & Costa, M. P. F., Niemann, O., Novo, E., & Ahern, F. (2002).
Zandonella, A. (1999). Hyperspectral aerial images. A Biophysical properties and mapping of aquatic veg-
valuable tool for submerged vegetation recognition in etation during the hydrological cycle of the Amazon
the Ortobello lagoons, Italy. International Journal of oodplain using JERS-1 and Radarsat. International
Remote Sensing, 20(3), 235 248. Journal of Remote Sensing, 23(7), 1401 1426.
Anstee, J., Dekker, A., Brando, N., Pinnel, N., Byrne, G., Dierssen, H. M., & Zimmerman, R. (2003). Ocean color
Danieal, P., et al. (2001). Hyperspectral imaging for remote sensing of seagrass and bathymetry in the
benthic species recognition in shallow coastal waters. Bahamas Banks by high-resolution airborne imagery.
In Proceedings of the International Geoscience and Limnology and Oceanography, 48(1), 444 455.
Remote Sensing Symposium 01 (Vol. 6. pp. 2513 Dutra, L. V., Treuhaft, R., Mura, J. C., Santos, J. R. D.,
1515). & Freitas, C. D. C. (2007). Estimating 3-dimensional
Armstrong, R. A. (1993). Remote sensing of submerged structure of tropical forests from radar multi-
vegetation canopies for biomass estimation. Interna- baseline interferometry: The Tapaj s FLONA case. In
tional Journal of Remote Sensing, 14(3), 621 627. Anais do XIII Simp sio Brasileiro De Sensoriamento
Austin, A., & Adams, R. (1978). Aerial color and color Remoto. Florian polis, Brasil (pp. 1657 1662).
infrared survey of marine plant resources. Pho- Edwards, R. W., & Brown, M. W. (1960). An aerial photo-
togrammetric Engineering and Remote Sensing, 44(4), graphic method for studying the distribution of aquatic
469 480. macrophytes in shallow waters. Journal of Ecology, 48,
Bajjouk, T., Guillaumont, B., & Populus, J. (1996). 161 163.
Application of airborne imaging spectrometry system Everitt, J. H., Yang, C., Escobar, D. E., Webster, C. F.,
data to intertidal seaweed classi cation and mapping. Lonard, R. I., & Davis, M. R. (1999). Using remote
Hydrobiologia, 326/327, 463 471. sensing and spatial information technologies to detect
Environ Monit Assess (2008) 140:131 145 143
and map two aquatic macrophytes. Journal of Aquatic Jensen, J. R., H. M. E., & Christensen, E. (1986). Remote
Plant Management, 37, 71 80. sensing inland wetlands: A multispectral approach.
Filippi, A. M., & Jensen, J. R. (2006). Fuzzy learning vec- Photogrammetric Engineering and Remote Sensing,
tor quantization for hyperspectral coastal vegetation 52(1), 87 100.
classi cation. Remote Sensing of Environment, 100(4), Jensen, J. R., Narumalani, S., Weatherbee, O., & Mackey,
512 530. J. H. E. (1993). Measurement of seasonal and yearly
Ford, J., & Casey, D. (1988). Shuttle radar mapping with cattail and waterlily changes using multidate SPOT
diverse incidence angles in the rainforest of Borneo. panchromatic data. Photogrammetric Engineering and
International Journal of Remote Sensing, 9(5), 927 Remote Sensing, 59(4), 519 525.
943. Jensen, J. R., Rutchey, K., Koch, M., & Narumalani, S.
Fyfe, S. K. (2003). Spatial and temporal variation in spec- (1995). Inland wetland change detection in the
tral re ectance: Are seagrasses spectrally distinct?. Everglades water conservation area 2A using a time
Limnology and Oceanography, 48(1), 464 479. series of normalized remotely sensed data. Pho-
Graciani, S. D., & Novo, E. M. L. M. (2003). Determina o togrammetric Engineering and Remote Sensing, 61(2),
da cobertura de macr tas aqu ticas em reservat rios 199 209.
tropicais. In Anais do XI Simp sio Brasileiro de Sen- Junk, W. (Ed.) (1997). The Central Amazon Floodplain:
soriamento Remoto. (pp. 2509 2516). Ecology of a Pulsing System, Vol. 126 of Ecological
Haack, B., & Bechdo, M. (2000). Integrating multisensor Studies. Springer.
data and RADAR texture measures for land cover Kasischke, E. S., & Borgeau-Chavez, L. L. (1997). Mon-
mapping. Computers & Geosciences, 26, 411 421. itoring south Florida wetlands using ERS-1 SAR
Han, L., & Rundquist, D. (2003). The spectral responses imagery. Photogrammetric Engineering and Remote
of Ceratophyllum demersum at varying depths in an Sensing, 63(3), 281 291.
experimental tank. International Journal of Remote Kasischke, E. S., Smith, K. B., Borgeau-Chavez, L. L.,
Sensing, 24(4), 859 864. Romanowicz, E. A., Brunzell, S., & Richardson, C. J.
Heege, T., Bogner, A., & Pinnel, N. (2003). Mapping of (2003). Effects of seasonal hydrologic patterns in
submerged aquatic vegetation with a physically based south Florida wetlands on radar backscatter measured
process chain. In SPIE Proceedings on Remote Sensing from ERS-2 SAR imagery. Remote Sensing of Envi-
(Vol. 5233). CD-ROM. ronment, 88, 423 441.
Hess, L., Melack, J., Filoso, S., & Wang, Y. (1995). Kirk, J. T. O. (1994) Light and Photosynthesis in Aquatic
Delineation of inundated area and vegetation along Ecosystems, 2nd edn. Cambridge University Press.
the Amazon oodplain with the SIR-C synthetic aper- Komatsu, T., Igarashi, C., Tatsukawa, K., Sultana, S.,
ture radar. IEEE Transactions on Geoscience and Re- Matsuoka, Y., & Harada, S. (2003). Use of multi-beam
mote Sensing, 33(4), 896 904. sonar to map seagrass beds in Otsuchi Bay on the
Hess, L., Melack, J., Novo, E. M. L. M., Barbosa, C. C. F., Sanriku coast of Japan. Aquatic Living Resources, 16,
& Gastil, M. (2003). Dual-season mapping of wetland 23 230.
inundation and vegetation for the Central Amazon Kotchenova, S. Y., Song, X., Shabanov, N. V., Potter, C. S.,
Basin. Remote Sensing of Environment, 87, 404 428. Knyazikhin, Y., & Myeni, R. B. (2004). Lidar remote
Hess, L., Melack, J., & Simonett, D. S. (1990). Radar detec- sensing for modeling gross primary production of
tion of ooding beneath the forest canopy: A review. deciduous forests. Remote Sensing of Environment, 92,
International Journal of Remote Sensing, 11(7), 131*-***-***.
1325. LaCapra, V. C., Melack, J. M., Gastil, M., & Valeriano, D.
Hess, L. L., Novo, E. M. L. M., Slaymaker, D. M., Holt, J., (1996). Remote sensing of foliar chemistry of inun-
Steffen, C., Valeriano, D. M., et al. (2002). Geocoded dated rice with imaging spectrometry. Remote Sensing
digital videography for validation of land cover map- of Environment, 55(1), 50 58.
ping in the Amazon basin. International Journal of Le Toan, T., Ribbes, F., Wang, L., Floury, N., Ding, K.,
Remote Sensing, 23(7), 1527 1555. King, J. A., et al. (1997). Rice crop mapping and mon-
Hopkinson, C., Chasmer, L., Lim, K., Treitz, P., & itoring using ERS-1 data based on experiment and
Creed, I. (2006). Towards a universal lidar canopy modeling results. IEEE Transactions on Geoscience
heig