Potential uses of TerraSAR-X for mapping herbaceous halophytes over salt marsh and tidal flats

Estuarine, Coastal and Shelf Science 115 (2012) 366e376

Contents lists available at SciVerse ScienceDirect

Estuarine, Coastal and Shelf Science journal homepage: www.elsevier.com/locate/ecss

Potential uses of TerraSAR-X for mapping herbaceous halophytes over salt marsh and tidal flats Yoon-Kyung Lee a, b, Jeong-Won Park a, Jong-Kuk Choi b, Yisok Oh c, Joong-Sun Won a, * a

Department of Earth System Sciences, Yonsei University, 134 Shinchon-dong, Seodaemun-gu, Seoul 120-749, Republic of Korea Korea Ocean Satellite Center, Korea Institute of Ocean Science & Technology, 787 Haean-ro, Sangrok-gu, Ansan, Gyeonggi-do 426-744, Republic of Korea c Department of Electronic Information and Communication Engineering, Hongik University, 94 Wausan-ro, Seoul 121-791, Republic of Korea b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 19 June 2012 Accepted 8 October 2012 Available online 17 October 2012

This study presents a method and application results of mapping different halophytes over tidal flats and salt marshes using high resolution space-borne X-band synthetic aperture radar (SAR) that has been rarely used for salt marsh mapping. Halophytes in a salt marshes are sensitive to sea-level changes, sedimentation, and anthropogenic modifications. The alteration of the demarcations among halophyte species is an indicator of sea level and environmental changes within a salt marsh. The boundary of an herbaceous halophyte patch is, however, difficult to determine using remotely sensed data because of its sparseness. We examined the ecological status of the halophytes and their distribution changes using TerraSAR-X and optical data. We also determined the optimum season for halophyte mapping. An annual plant, Suaeda japonica (S. japonica), and a typical perennial salt marsh grass, Phragmites australis (P. australis), were selected for halophyte analysis. S. japonica is particularly sensitive to sea level fluctuation. Seasonal variation for the annual plant was more significant (1.47 dB standard deviation) than that for the perennial grass, with a pattern of lower backscattering in winter and a peak in the summer. The border between S. japonica and P. australis was successfully determined based on the distinctive X-band radar backscattering features. Winter is the best season to distinguish between the two different species, while summer is ideal for analyzing the distribution changes of annual plants in salt marshes. For a single polarization, we recommend using HH polarization, because it produces maximum backscattering on tidal flats and salt marshes. Our results show that high resolution SAR, such as TerraSAR-X and Cosmo-SkyMed, is an effective tool for mapping halophyte species in tidal flats and monitoring their seasonal variations. Ó 2012 Elsevier Ltd. All rights reserved.

Keywords: salt marshes TerraSAR-X halophyte tidal flats backscattering coefficients

1. Introduction Salt marshes develop in the upper intertidal zone when dominated by halophytic herbaceous plants under favorable conditions and influenced by tides and salinity at the transition point between tidal flats and inland (Adam, 1990). A salt marsh plays several important roles: 1) it acts as a buffer zone from storms and contamination (or heavy metals), 2) it exchanges materials between tidal flats and open water, and 3) it removes a large amount of carbon from the atmosphere (Kirwan and Murray, 2007; Mitsch and Gosselink, 2007). Communities of salt marsh vegetation called halophytes play a fundamental role in the stability and topography of coastal wetlands. Halophyte communities in salt marshes have vertical zonation, a spatial segregation based on * Corresponding author. E-mail address: [email protected] (J.-S. Won). 0272-7714/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ecss.2012.10.003

plant competition and physical gradient characteristics of the habitat such as salinity, water level, and exposure time (Bertness et al., 2002). Halophytes are particularly sensitive to sea level changes, the rate of marsh accretion, sediment supply, and anthropogenic modifications (Gardner and Porter, 2001). Halophyte zonation is closely related to salinity and exposure time (Lee et al., 2006). The imbalance between sea level rise and sediment accretion is one of the major factors driving the loss of salt marshes (Warren and Niering, 1993). Rizzetto and Tosi (2011, 2012) demonstrated the aptitude of modern salt marshes to counteract relative sea-level rise, as well as the rapid response of tidal channel networks to sea-level variations. Halophyte species alteration and demarcation is one of the environmental indicators of sea level changes in a salt marsh (Gardner and Porter, 2001). Therefore, accurate mapping of salt marshes is useful in understanding wetland functions and monitoring their responses to natural and anthropogenic actions (Baker et al., 2006).

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

Spaceborne synthetic aperture radar (SAR) has been used to monitor coastal regions and wetlands. Recently, high-resolution space-borne X-band SAR systems such as TerraSAR-X and CosmoSkyMed have been made available to the public, but they have rarely been used for salt marsh mapping. X-band SAR applications have gradually increased since the launch of the TerraSAR-X in 2007 (Strozzi et al., 2009). This study is to examine the potential of the high resolution X-band SAR system as a tool for salt marsh mapping. While optical satellite images are very useful for land vegetation studies, it is often difficult to acquire data from tidal flats due to the lack of optimal water and cloud conditions. In addition, optical data are insufficient for identifying salt marsh vegetation because of limitations in the spectral resolution and the lack of spatial resolution for detecting vegetation types (Adam et al., 2010). Radar reflection provides different information than optical sensors (Ozemi and Bauer, 2002). While free from weather condition limitations, radar backscatter is sensitive to dielectric properties (inundation level, soil moisture, soil salinity) and surface roughness (Henderson and Lewis, 2008). Various vegetation communities can be distinguished based on canopy structure, soil moisture, and the presence or absence of flooding (Kasischke and Bourgeau-Chavez, 1997). Hong et al. (2010) reported a strong dependence of scattering mechanisms on vegetation type over wetlands and demonstrated the effectiveness of using TerraSAR-X radar interferometry for monitoring water-level change. A strong correlation of double bounce between the water surface and vertical pole structures in coastal areas was also reported (Lee et al., 2006). Ramsey (1998) recommended shorter wavelength imagery at low incidence angles for herbaceous areas. X-band (3.1 cm) SAR exhibits more sensitivity to leaves and small branches because it has a shorter wavelength and light penetration depth (Huang et al., 2010). Wet troposphere (atmospheric water vapor) can, however, cause artifacts in SAR interferograms, and these artifacts are larger for shorter wavelengths (Zebker et al., 1997). L-band and C-band SAR data have also been used for coastal wetlands mapping (Townsend, 2002; Lang et al., 2008; Slatton et al., 2008; Kwoun and Lu, 2009). Lucas et al. (2007) effectively mapped mangroves using ALOS PALSAR data. Crevier et al. (1996) reported that late autumn is the best season for detecting wetlands when using ERS-1 C-band. Polarimetric analysis is generally considered effective for vegetation monitoring. Like-polarized radars are well suited for detecting flooded vegetation, and L-HH is preferred for wooded vegetation and C-HH for herbaceous wetlands (Kasischke et al., 1997). Crosspolarization is necessary for differentiating herbaceous vegetation from forested areas (Bourgeau-Chavez et al., 2001). Quad-polarization is another benefit of estimating biomass using TerraSAR-X (Englhart et al., 2011). The potential for using high resolution space-borne X-band SAR in salt marsh mapping has not been thoroughly explored. The primary objective of this study was to investigate the potential of space-borne X-band SAR as a tool for salt marsh mapping. Multitemporal TerraSAR-X data were acquired under flood/ebb conditions to differentiate halophyte species over tidal flats and to determine their temporal variations. The characteristics of X-band radar backscattering from different halophyte species and their seasonal variations were examined. We also studied the relationship between radar backscattering and a vegetation index derived from Landsat ETMþ to better understand how radar signal is related to vegetation information for different seasons. We aimed to determine the optimal season and tidal flat conditions for satellite observation. Major halophyte species around the west coast of the Korean peninsula include Suaeda japonica (S. japonica) and Phragmites australis (P. australis), among others. We determined the distribution of different species and the border between S. japonica and P. australis. We focused on S. japonica, which is an

367

annual, because it changes dynamically according to sea level fluctuation. To differentiate halophyte species and determine the optimal season for salt marsh mapping, a multiple regression analysis and an independent t-test were performed for different polarizations. A final salt marsh map was constructed using a decision tree based on a statistical analysis of the backscattering coefficient to validate our approach. 2. Halophytes in the study area The Ganghwa tidal flat is located on the mid-west of the Korean Peninsula near the estuaries of the Hang-gang (or Han River), as shown in Fig. 1. Tides are semi-diurnal with a mean tidal range of 6.5 m (spring tide ¼ approximately 8 m, neap tide ¼ approximately 4 m) (Choi et al., 2011). The surface sedimentary facies are primarily mud flats in the eastern part of the tidal flat, sand flats in the western part, and mixed flats in-between, as shown in Fig. 1 (Woo and Je, 2002). The dominant area of Phragmites australis is the already stabilized salt marsh, which is affected by seawater only a few times each month. Conversely, the seaward or landward migration of Suaeda japonica is an indicator of sea level change, because of the species’ sensitivity to sea level. S. japonica is an annual plant belonging to the Chenopodiaceae family. It has a stem that grows up to 50 cm high, and it emerges primarily in the spring (March to May). The rapid growth of the underground portion of S. japonica during its beginning stage allows it to adapt to the ebb and flood tidal environment. After the roots are firmly fixed in the ground (typically by May), the plant’s above-ground portion grows quickly (Lee and Ihm, 2004). Abrupt changes in S. japonica distribution represent environmental changes in the local area, as S. japonica is particularly sensitive to local sea level changes (Min, 2005). As the frequency of seawater inundation decreases, S. japonica appears and migrates seaward. Due to the sparseness of the population at the seaward front, it is, however, difficult to determine the precise boundary for annual observation. Migration of the rear boundary between S. japonica and P. australis is also an indirect indicator of sea level change. Phragmites australis, otherwise known as the common reed, has been found at the edges of high marshes and is a perennial grass with annual shoots emerging from underground rhizomes (Orson et al., 1997). This plant grows to 3e4 m in height in optimal conditions (Poulin et al., 2010). The stem dies in the early winter but remains standing as a rigid cane until the plant is reborn the next spring (Burgess and Evans, 1989). P. australis spreads primarily through the growth of surface runners and underground rhizomes at the edges of salt marshes (Philipp and Field, 2005). 3. Data and processing 3.1. TerraSAR-X data processing In this study, a total of 16 TerraSAR-X data sets acquired between 2008 and 2010 were used for analyses of backscattering, polarization, incidence angle, and other characteristics, as summarized in Table 1. Due to the restrictions on quad-polarization data in the study area, only single- and dual-polarization data were used in this study. Halophyte characteristics were initially examined using TerraSAR-X HH single polarization data, which is generally better for discriminating between water-free and partially water-covered surfaces in tidal flats. Single HH-polarization data were the most abundant in this study and were thus utilized as the primary source in the analysis. Dual-polarization of VV and VH TerraSAR-X data was used to examine their effectiveness in distinguishing different species. One data set of 13 single-look slant range complex (SSC)

368

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

Fig. 1. Location of the study area, acquired on 27 July 2008 by TerraSAR-X HH. The boundary refers to the surface sediment distribution (green: mud flats, blue: mixed flats, red: sand flats). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

images was acquired in the standard stripmap mode with a 3 m pixel resolution, 39 incidence angle, and HH polarization. The other data set of three SSC images was acquired in the dualpolarization (VH/VV combination) stripmap mode with a 6 m resolution and 23 and 31 incidence angles. Initially, intensity images from the SSC data were converted into multi-look intensity images (MLI) and geographically rectified using orbital data. Absolute radiometric calibration functioned to minimize differences in the image radiometry using Equation (1) (Infoterra, 2008), which transforms the intensity of each pixel digital number (DN) into a backscattering coefficient, sigma naught (s ), in decibels.





s0 ¼ 10  log10 ks  DN2  NEBN þ 10  log10 ðsinwloc Þ ðdBÞ (1) ks: Calibration and processor scaling factor provided by the ground station DN: Pixel image values or digital numbers NEBN: Noise equivalent beta naught, represents the influence of different noise contributions to the signal

qloc: Local incidence angle derived from the geocoded incidence angle mask It is necessary to normalize radar signals according to incidence angle for a comparison between signals obtained at different incidence angles (Ulaby et al., 1982). The angular dependence of the backscattering coefficients can be reduced if the measured backscattering coefficients are normalized by the function of s0 ¼ acosbq, where the coefficient b is related to the target backscattering properties and a, depending on the dominant scattering mechanism and sensor conditions (Ulaby et al., 1982; Baghdadi et al., 2008). For the VH/VV dual polarization data, the beta value was computed for each polarization. Sigma naught values over the dense forest were used for correction. Conversion values of 1.135 and 1.696 were used for VH and VV, respectively. Although the TerraSAR-X images were georectified, there were position discrepancies of approximately 3 or 4 pixels compared with the field survey data. To refine the geolocation of the images, seven ground-control points (GCPs) obtained from the field survey were used. Speckle noise in the images was suppressed by applying an enhanced Lee filter (Lopes et al., 1990) with a 3 by 3

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376 Table 1 Summary of satellite data used in this study. Satellite and data characteristics

Image acquisition date mm/dd (tidal condition, tide level, precipitation if not zero)

TerraSAR-X (HH polarization) Orbit pass: Ascending Incidence angle at scene center: 39.2 Product type: SSC

2008: 06/13 07/27 10/12 10/23 11/25 12/06 2009: 04/17 07/03 12/15 2010: 03/24 07/01 08/03 09/05

TerraSAR-X (VV/VH polarization) Orbit pass: Descending Incidence angle at scene center: 31.6 Product type: SSC

2008: 09/24 (Flood, 365 cm) 10/05 (Flood, 719 cm)

TerraSAR-X (VV/VH polarization) Orbit pass: Ascending Incidence angle at scene center: 23.3 Product type: SSC

2008: 07/11 (Flood, 328 cm)

Landsat ETM+

2008: 01/30 (Ebb, 611 cm), 04/19 (Ebb, 108 cm), 05/05 (Ebb, 839 cm), 08/09 (Ebb, 608 cm), 08/25 (Ebb, 630 cm), 09/10 (Flood, 540 cm), 10/12 (Flood, 291 cm), 12/15 (Flood, 106 cm) 2009: 02/01 (Ebb, 501 cm), 03/21 (Flood, 510 cm), 4/06 (Flood, 392 cm), 05/08 (Ebb, 134 cm), 05/24 (Flood, 210 cm), 09/13 (Ebb, 603 cm), 10/15 (Flood, 493 cm), 12/18 (Ebb, 78 cm) 2010: 01/03 (Ebb, 114 cm), 02/04 (Ebb, 513 cm), 04/09 (Flood, 507 cm), 09/16 (Ebb, 504 cm), 10/18 (Flood, 460 cm)

(Ebb, 225 cm) (Flood, 286 cm, 0.2 mm) (Ebb, 438 cm) (Flood, 252 cm, 6.0 mm) (Ebb, 374 cm, 0.4 mm) (Flood, 391cm) (Flood, 414 cm) (Ebb, 284 cm) (Ebb, 636 cm) (Flood, 401 cm) (Flood, 677cm) (Flood, 465 cm) (Ebb, 269 cm, 2.5mm)

369

reported by Svoray and Shoshany (2003). Twenty-two Landsat ETMþ images were acquired between January 2008 and October 2010 and are summarized in Table 1. Although these images were collected with the scan line corrector (SLC)-off due to an instrument defect, most areas of interest were not affected. All of the images were geometrically rectified with image-to-image coregistration using an Indian Remote Sensing Satellite (IRS) 1C panchromatic image that was already rectified by 1:5000 topographic maps. Atmospheric correction was applied using the Flaash model in ENVI. To minimize the effect of the soil background conditions, the soil-adjusted vegetation index (SAVI), developed by Huete (1988), was calculated from the images because halophytes are sparsely populated in salt marshes compared to inland vegetation. The SAVI accounts for the spectral contribution of the soil and is recommended for predicting biomass when soil exposure is high relative to vegetation cover (Zhang et al., 1997).

SAVI ¼

1:5  ðb4  b3Þ ðb4 þ b3 þ 0:5Þ

(2)

where b3 and b4 are pixel values of the Landsat ETMþ band 3 and band 4, respectively. 3.3. Decision tree A decision tree classifier is an efficient tool for classifying patterns in data (Parmuchi et al., 2002). Advantages of a decision tree include: 1) no assumption is required for data distribution or feature independence (Ke et al., 2010), 2) it can utilize both nonlinear and hierarchical associations (Pal and Mather, 2003), and 3) it is easy to interpret with all measurement scales (Wright and Gallant, 2007). We designed a rule-based salt marsh mapping method using TerraSAR-X and the boundaries of surface sedimentary facies based on the rule that if a condition is met, then an inference can be made. To determine thresholds for each step, the mean and standard deviation were used for each class and condition in data acquisition. Inland areas were eliminated from the decision tree classification process because they are a potentially large source of misclassification error. The boundaries determined by the decision tree were verified using GPS fieldmeasured data. 4. Analysis and results

window, which preserved spatial variations such as edges and texture. The tidal levels recorded at the Inchon station were corrected to the local tidal conditions at the moment of the SAR data acquisition, accounting for the time differences, tide ratios, and mean sea levels. Precipitation records were obtained from the Ganghwa weather station, which is the station closest to the study area. This was done because precipitation alters the moisture content of vegetation and soil, thus impacting the SAR backscattering return (Lu and Meyer, 2002). A field survey was conducted in August 2009 to identify the boundaries of Phragmites australis. To determine the boundaries between the non-vegetated tidal flats and Suaeda japonica, field surveys were conducted in May, July, and October 2010 using GPS (eTrex Visata HCx, GARMIN). 3.2. Landsat ETMþ data A close relationship between radar backscattering and spectrally unmixed Landsat TM data for herbaceous vegetation was

4.1. General characteristics of X-band backscattering Typical classes were chosen to represent the salt marsh types that exist in this region and to clarify the range of backscatter observed in the X-band images. The vegetated marsh in the study area was primarily covered by Phragmites australis and Suaeda japonica, and non-vegetated marsh areas were identified as tidal flats. P. australis, S. japonica, tidal flats, and water were then characterized as typical classes. While the bare surface of tidal flats is generally characterized as having relatively low backscattering, tidal channels produce higher backscattering. Thus, the effect of tidal channels also needs to be considered in halophyte mapping by SAR. Approximately 350 pixels from each class were selected in areas where this type of land cover was confirmed via field survey in order to examine the seasonal variation of sigma naught in each class. Other halophyte species were excluded because their distributions were too small to trace. Fig. 2 displays the temporal variations of sigma naught in each class. The stability of sigma naught throughout the different seasons was initially evaluated using an average sigma naught for an industrial

370

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

4.2. Characteristics of soil-adjusted vegetation index (SAVI) for different classes

Fig. 2. Temporal variations of sigma naught of HH polarization in each class over a three-year period.

area because backscattering from a large number of artificial structures does not change over time (Ferretti et al., 2001). The sigma naught of the selected industrial area remained stable for three years at an average of 1.75 dB and a standard deviation of 0.23 dB, indicating that further compensation was not needed for the analysis. The average sigma naught for seawater among the classes was at its lowest at 26.94 dB with a standard deviation of 1.12 dB. Backscattering from seawater is generally affected by surface conditions linked to tides, currents, and wind. The average sigma naught for the tidal flats demonstrated significant variation with a standard deviation of 4.27 dB, according to the surface conditions at the moment of data acquisition. The average sigma naught for Phragmites australis was higher and slightly more stable than that for Suaeda japonica. Standard deviations for P. australis and S. japonica were 1.06 dB and 1.47 dB, respectively. These findings are discussed further in the following sections.

The SAVI was calculated from Landsat ETMþ visible and nearinfrared (VNIR) bands to examine the spectral characteristics of Phragmites australis, Suaeda japonica, and the tidal flats according to seasonal change, as in Fig. 3. Due to a strong chlorophyll presence in the near-infrared bands, the seasonal variation pattern of SAVI matched well with the phenological cycle of the halophytes, increasing from March until August/September and then rapidly decreasing. All three years of data demonstrated the same cyclic pattern. The seasonal variation of SAVI is useful in understanding characteristics of the growing season. The annual SAVI variation patterns of S. japonica (Fig. 3(a)) and P. australis (Fig. 3(b)) were generally similar, while that of tidal flats (Fig. 3(c)) differed considerably. The lengths of the growing seasons for the species were similar, but the annual variation of S. japonica was slightly larger than that of P. australis. The maximum SAVI value of S. japonica was slightly higher than that of P. australis, but the lowest SAVI value of S. japonica was slightly lower than that of P. australis. The chlorophyll content of S. japonica was high in February, declined gradually through April, increased during the summer, and decreased sharply in late autumn. Throughout all of these changes, the betacyanin content demonstrated an inverse relationship with chlorophyll (Hayakawa and Agarie, 2010). Although S. japonica contains betacyanin, it is difficult to identify this plant using only the vegetation index. It is important to clearly identify the area of S. japonica habitation because it changes dynamically according to sea level, while the habitat of P. australis remains relatively stable. S. japonica dies in the autumn and decomposes over the winter months. The annual SAVI variation of the tidal flats was less significant than those from the halophytes, as seen in Fig. 3(c). The SAVI value of the tidal flats was low and relatively stable throughout the year. The different tidal conditions (ebb/ flood) and the different contents of the surface remnant water and algae blooms are thought to contribute to the limited variation in the tidal flats. In summary, Phragmites australis and Suaeda japonica are distinguishable in the tidal flats during the summer, but it is difficult to distinguish these plants in the tidal flats during the winter because they exhibit similar reflectance values. While it is possible to discriminate the halophyte-developed areas from bare tidal flat surfaces, it is difficult to classify different halophyte species using optical remote sensing alone. There is a trade-off between spatial

Fig. 3. Seasonal changes of SAVI in (a) P. australis, (b) S. japonica, and (c) tidal flats.

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

and spectral resolution for optical remote sensing sensors; consequently, it is often difficult for a space-borne optical remote sensing system to meet all the requirements necessary to map different halophytes. The spatial resolution and spectral sensitivity of the Landsat ETMþ make it difficult to differentiate these halophytes, even during the summer. 4.3. Characteristics of radar backscattering and seasonal variation HH polarization produces maximum backscattering on tidal flats and salt marshes. As previously discussed in Section 4.1, the seasonal variation for perennial grass was insignificant, with slightly higher values in the winter. The seasonal variation for the annual plant was more significant similar to the plant life cycle, with a pattern of lower backscattering in winter, which increased during the spring and then reached a peak during the summer, as seen in Fig. 2. While Suaeda japonica is dead in November, the presence of its plant structure contributes to a high s value. After the structure of the annual plant collapses due to the tides in December, its backscatter value decreases dramatically. The sigma naught of Phragmites australis was larger than that of S. japonica, by 0.84 dB on average, and increased by 2.34 dB in the winter, as in Fig. 2. The sigma naught behavior of the perennial plant was not synchronized with the SAVI or plant life cycle. While its leaves are dry during the winter, the P. australis structure, with a thin leaf cover and the presence of surface water, results in a higher s value than in the growing season. As a result, winter is the best season for distinguishing between P. australis and S. japonica. Additionally, the variation patterns of S. japonica correlate with that of the tidal flats to a certain degree. The correlation coefficient between the two classes was 0.66. A certain amount of radar energy is thought to penetrate through S. japonica and hit the ground surface due to the sparseness and the small, short succulent leaves. Because the ground surface beneath S. japonica is composed of mud flats and remnant surface water, the backscattering signal is enhanced due to the double-bounce between the remnant water surface and the plant structure. While P. australis grows vigorously when the salt marsh is relatively stable, the growth of S. japonica begins at the border of dynamic sea level change and is indicative of the seaward extension of the salt marsh. The mixed nature of S. japonica in terms of sigma naught from vegetation and the tidal flats is typical. However, this particular area of S. japonica exhibited a 2.21e6.25 dB higher sigma naught than the bare surface of the tidal flats. Halophyte signatures of the TerraSAR-X VV/VH dual polarization data were also examined for a limited number of data sets. Radar returns from tidal flats via VV-polarization were slightly lower than those acquired from the salt marsh (average range: e10.5 to 13.3 dB). The seasonal variation of Suaeda japonica was greater than that of Phragmites australis with regard to VV/VH-polarization. The average difference between VV and VH polarization was 6.2 dB and 5.8 dB for P. australis and S. japonica, respectively, while that for the non-vegetated tidal flats was 13.2 dB, which was much larger than values obtained from the salt marsh. The average VV/VH difference of 5.8 dB for S. japonica was smaller than the ground radar measurement of 10.2 dB (Park et al., 2009). The VV/VH ratio demonstrated that volume scattering for S. japonica and the nonvegetated tidal flats was dominant during the summer. TerraSAR-X VV/VH dual polarization has the potential to be used in the identification of salt marsh and tidal flat boundaries, according to larger VV/ VH ratios. However, VV/VH ratio-based discrimination may not be effective in discriminating between S. japonica and P. australis within a salt marsh. Better results are attained when focusing on the seasonal variation of S. japonica versus the year-round stability of P. australis. While it is possible to discriminate both halophytes from nonvegetated tidal flats through the year, it is necessary to determine

371

the optimum season for this. We examined the differences in radar backscattering between Phragmites australis and Suaeda japonica, P. australis and the non-vegetated tidal flats, and S. japonica and the non-vegetated tidal flats. The difference between P. australis and S. japonica was the greatest in February, followed by a decrease through August and then an increase until December. The difference between P. australis and the non-vegetated tidal flats exhibited large variation throughout the year due to different tidal conditions. Although tidal conditions yielded large variation, the difference between S. japonica and the non-vegetated tidal flats was large in the summer. Thus, summer is the best season to discriminate between S. japonica and the tidal flats, but it is difficult to determine a specific season to differentiate between P. australis and the non-vegetated tidal flats. To distinguish between P. australis and S. japonica, the start of the growing season in April until the end of November was chosen as the “on-season” based on the temporal variation of S. japonica, with the rest of the year considered the “off-season”. We subsequently examined the relationship between TerraSARX HH polarization sigma naught and Landsat EMTþ SAVI, as shown in Fig. 4. A negative correlation with a correlation coefficient of 0.648, with a p-value of 0.024, was determined between X-band sigma naught and SAVI for Phragmites australis, as shown in Fig. 4(a). This implies that the vegetation vitality of the perennial grass increased the chlorophyll reflectance, while the X-band radar backscattering remained stable or slightly decreased independent of the chlorophyll content. It should be noted that the height and density of P. australis do not change dramatically throughout the year. For Suaeda japonica, conversely, a positive linear relationship with a correlation coefficient of 0.68, with a p-value of 0.036, was acquired between the X-band sigma naught and SAVI, as shown in Fig. 4(b). This positive correlation implies that radar backscattering is enhanced with an increase of SAVI. This suggests that, while the seasonal variation of the annual plant as recorded by X-band HHpolarization does not completely align with the phenological cycle, increases in SAVI can be correlated with an increase of surface or volume scattering of the X-band. This may be due to an increasing density of population and leaf moisture content during the growing season. 4.4. Contribution to radar backscattering A multiple regression analysis was performed to examine the relationship between the sigma naught-dependent and independent variables, as well as to identify significant independent variables using SPSS. Polarization (HH, VV, and VH), land cover class, season (on-season or off-season), tidal height (below 100, 200, 300, 400, 500, 600, and above 700 m), and tidal condition (ebb or flood) were selected as independent variables. Stepwise selection was used to achieve an optimum regression model. The results are listed in Table 2. The relative contribution of each parameter to sigma naught can be understood by the standardized coefficient. Polarization among the five parameters showed the largest contribution to the sigma naught, followed by land cover class and season. Sigma naught changes were approximately 4.70 dB in potential cases of polarization change when the other variables remained constant. The tidal height and tidal condition contributed the least to the radar return according to the standardized coefficients in Table 2. These results support the use of SAR for land cover classification in tidal flats with a careful consideration of seasonal effects. Based on the results, the sigma naught for each class was averaged according to season and tidal condition. Although tidal height had a greater relative contribution than tidal condition, it was not included in the sub-division due to the limited number of tidal heights in the sub-tidal height class. The average sigma naught

372

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

Fig. 4. Relationship between SAVI and sigma naught for (a) P. australis and (b) S. japonica.

for each sub-class is plotted in Fig. 5. The mean Phragmites australis HH polarization during the on-season at flood tide was approximately 8.09 dB, while VV polarization was 12.83 dB. The mean value for Suaeda japonica was approximately 8.34 dB during the on-season, which decreased to approximately 12.07 dB during the off-season regardless of the tidal conditions. The sigma naught in VV polarization during the on-season at flood conditions was comparable to HH polarization during the off-season. Sigma naught values for P. australis and S. japonica HH polarization were very similar during the on-season for all tidal conditions. As a result, it is difficult to use HH to discriminate between these two halophyte species during the on-season (summer). However, it is effective to discriminate between the two species during the off-season (autumn to winter). These results indicate that ebb tide during the off-season created the best conditions for distinguishing the difference between the two species by 4.6e8.6 dB. Additionally, VV single polarization or VV/VH dual polarization was determined to be more effective for discrimination. However, the behavior of the average sigma naught between the different polarizations was similar to that of P. australis. The tidal flats demonstrated the largest standard deviation, which was dependant on surface sediment as well as tidal condition. One control factor of radar backscattering over a tidal flat surface is the effective exposure area associated with tidal conditions, which is defined as the ratio of water-free surface area to total area (Ryu et al., 2002). The other major control factor is the effect of ripple, as it plays an important role in sand-dominant surfaces and sand flats (Gade et al., 2008). However, ripples are not well developed in mud flats, in which tidal energy is low and halophytes are beginning to develop. Flood tides resulted in a higher mean sigma naught than ebb tide conditions, probably due to the presence of

remnant surface water during ebb tides. The lowest value for tidal flats marked by the open square was as low as that of seawater and was affected by the spring tide. The average sigma naught of seawater was the lowest mean value for all polarizations, seasons, and tidal conditions. We calculated the mean difference and performance of an independent t-test between classes to quantify the difference in sigma naught with respect to polarization, tidal condition, and season, as shown in Table 3. The p value from the t-test was noted if the difference was smaller than the significance level (a ¼ 0.05). If the p value was larger than the significance threshold (p > a), the two compared classes were difficult to distinguish. Significant p values are marked in bold. Our primary concerns included distinguishing halophytes from tidal flats and identifying Suaeda japonica versus Phragmites australis. As seen in Table 3, the result of the independent t-test between S. japonica and P. australis was large, with a t-score value of 37.71 under ebb tide conditions during the off-season. This was encouraging, as it is difficult to distinguish

Table 2 Coefficients of the multiple linear regression model (n ¼ 26,662). Independent variable

Un-standardized coefficients B

Std. error

Polarization Land cover class Season Tidal height Tidal condition (Constant)

4.695 0.729 2.722 0.220 0.702 44.040

0.053 0.019 0.098 0.000 0.087 0.572

Standardized coefficient

t

0.558 0.196 0.160 0.046 0.046

89.232 38.482 27.730 8.240 8.109 77.011 Fig. 5. Average sigma naught and standard deviation for each sub-class.

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376 Table 3 Results of t-test with HH and VV/VH polarization. Class

Polarization

Tidal condition

Season*

P. australis

S. japonica

S. japonica

HH

All

On Off On Off On Off On On On Off On Off On Off On On

8.49 41.72 1.31 (p ¼ 0.30) 37.71 10.63 31.60 15.43 12.48 70.88 56.02 51.05 53.00 50.00 39.56 L5.05 (p ¼ 0.62) 69.15

e

Ebb Flood

Tidal flats

VV VH HH

All Ebb Flood

VV VH

83.48 43.46 64.35 35.98 56.25 33.63 13.51 69.15

* The “On” season represents the growing season of S. japonica from April until the end of November, with the rest of the year considered as the “Off” season.

between these species using optical remote sensing. The difference between the halophyte species in the off-season yielded a t-score of 41.72 at a 95 percent confidence level. The independent t-score between S. japonica and non-vegetated tidal flats was large at ebb tide during the on-season. With a p value of 0.30, this test discouraged the use of HH polarization during the on-season under flood tide conditions to distinguish the different species. A p value of 0.62 between P. australis and the tidal flats also suggests that VV polarization is not suitable for use to distinguish P. australis from tidal flats during the on-season under flood tide conditions. For long-term monitoring of S. japonica distribution linked to sea level fluctuation, it is necessary to identify an annual date for ideal data acquisition. In summary, an optimum data acquisition plan utilizing high resolution space-borne X-band SAR HH-polarization should focus on the off-season, particularly during ebb tide. VV/VH dual or full polarization data would have a great potential for monitoring salt marshes; however, the results for VV/VH dual or full polarization in this study were inconclusive due to a small number of data sets.

4.5. Effect of tidal channel on radar backscattering within tidal flats The standard deviation of the tidal flats was the largest among all classes, as shown in Fig. 5. Grain size, surface roughness and ripple, remnant water, and tidal level governed the backscattering yield for the tidal flats. Another important tidal flat feature impacting surface backscattering is the tidal channels and creeks, which are developed according to grain size and topography. The stream velocity of the tidal currents is known to be higher near tidal creeks than in open flats (Gade et al., 2008). As a result, roughness in the vicinity of channels typically produces a higher backscattering coefficient. The high backscattering from tidal channels is often problematic when mapping halophytes within low backscattered tidal flats. Table 4 Average sigma naught and standard deviation of open flats and channel vicinities according to tidal flat type. 20080727 (flood) (clear sky condition) Open flats

20081023 (flood) (raining)

Channel vicinity Open flats

Mud flats 14.65 (2.74) 4.69 (1.74) Mixed flats 16.58 (4.47) 11.99 (4.07)

Channel vicinity

15.77 (3.27) 9.89 (2.85) 24.63 (3.13) 12.96 (3.20)

373

To distinguish areas affected by channels from open flats, we reviewed the mean difference in sigma naught. Two images were acquired on 27 July 2008 (tidal height ¼ 286 cm) and 23 October 2008 (tidal height ¼ 252 cm; precipitation ¼ 6.0 mm), as shown in Table 4. The area in the vicinity of the channels produced strong backscattering compared with those of open mud and mixed flats. The backscattering was particularly significant on the mud flats because the steep meandering slope of channels in mud flats acted to cause scatter. The average sigma naught of the channels in the mud flats (4.69 dB) was much higher than that in the mixed flats (11.99 dB). The area in the vicinity of the channels in the mud flats had stronger backscattering than those of halophytes by about 3.3e4.4 dB. Caution is required if data are obtained under rainy condition. On rainy days, the difference in the mixed flats dropped to less than 5 dB, and the average sigma naught of the channels in the mud flats was similar to those of Phragmites australis and Suaeda japonica. In clear weather, the average sigma naught of the mixed flats during flood tide was approximately 16 dB; however, rain dramatically decreased it to approximately 24 dB. Conversely, the effect of rain on mud flats and in the vicinity of channels was limited. The difference in the backscattering coefficient between the open flats and the vicinity around the channels in mixed flats nearly doubled when it rained. The images taken on rainy days were also useful in distinguishing channels from open mixed flats. 4.6. Classification and validation Decision tree analyses were performed for halophyte mapping using a statistical analysis of the backscattering coefficients. Decision rules incorporating knowledge about halophyte habitats during the off-season facilitated a distinction of Phragmites australis from Suaeda japonica that could not be made during the on-season. Classification began with the HH image from the on-season, when Suaeda japonica demonstrated the highest backscattering. In VV polarization, the water exhibited the highest difference with the tidal flat in terms of mean value (14.0 dB). The average sigma naught of HH polarization for the seawater in flood conditions was 26.3  2.19 dB, and a threshold of 23.6 dB for HH polarization in flood conditions was used to identify seawater. The sigma naught values of the two different halophytes overlapped in large areas during the on-season. While Phragmites australis had more backscattering features (7.4  2.0 dB) even during the off-season, S. japonica yielded a lower backscattering coefficient (14.1  2.0 dB). As a result, a threshold should be determined during the off-season to aid in distinguishing P. australis from the tidal flats. Although channels near halophyte patches have higher backscattering coefficients (4.69  1.7 dB), the channels were not identified in this step because some channels had similar backscattering coefficients to P. australis. In addition, some channels in the mixed flats had low sigma naught, similar to that of open flats. Based on this information, the boundary between P. australis and S. japonica could be identified. The threshold of >11.3 dB for HH polarization was identified as the boundary between S. japonica and the tidal flats. Mixed flat channels were easily detected because of their low backscattering coefficients. A threshold of >16.1 dB was used to discriminate channels and bare surfaces in the mixed flats, while a threshold of >11.9 dB was used to differentiate channels from bare surfaces in mud flats. However, it is sometimes difficult to discriminate channels from the tidal flat surface using backscattering intensity only. Thus, channels were also independently obtained from the Kompsat-2 images for classification. The classification results from the decision tree are shown Fig. 6. Because of the high backscattering of tidal channels at high elevation, some parts of the channels were identified as Suaeda

374

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

Fig. 6. Classification maps generated according to decision tree (a) using only TerraSAR-X images, and (b) using TerraSAR-X images and a tidal channel vector map.

Fig. 7. (a) Extracted boundary of P. australis from 2009 (pink) and 2010 (green) classification results, as well as the tracked line from 2009 field survey results (blue), and (b) enlargement of the square area in (a). (c) The field survey line in 2009 (green) with a buffer of 3 m, equal to the TerraSAR-X ground resolution and the matched line of P. australis (red) extracted from TerraSAR-X data acquired in 2009, (d) extracted boundary of S. japonica via classification of TerraSAR-X data acquired in 2009 (pink) and 2010 (green), as well as the tracked line from 2010 field surveys (blue), and (e) enlargement of the square area in (c). (f) The field survey line in 2010 (green), with a 3 m buffer, and the matched boundary of S. japonica (red) derived from 2010 TerraSAR-X data.

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

japonica. Tidal channel identification errors at the boundary of S. japonica and non-vegetated tidal flats were due to the relatively high backscattering when only TerraSAR-X images were involved in classification, as shown in Fig. 6(a). The errors caused by tidal channels were, however, removed when utilizing tidal channels that had been independently mapped from optical images, as shown in Fig. 6(b). We then compared the boundary of Phragmites australis and Suaeda japonica extracted from the classification results with the boundary tracked from the field surveys conducted in 2009 and 2010, as shown in Fig. 7. The extracted boundary of P. australis is displayed in Fig. 7(a) and (b) and is located at the edge of the upper tidal flats where fresh water is supplied. The boundary between P. australis and S. japonica was relatively stable in the eastern region of the island between 2009 and 2010, as shown in Fig. 7(b), and the delineated line from the field survey agreed with the extracted line from the classification. The reference boundaries obtained from the field survey (green lines in Fig. 7(c)), with a buffer of 3 m, were also in agreement with the P. australis boundary derived from the classification of the TerraSAR-X image (red lines in Fig. 7(c)). The agreement between the two lines was 85.4%. The migration of the seaward boundary of S. japonica is noticeable in the western region of the island, as shown in Fig. 7(d) and (e). The southwestern boundary between S. japonica and the non-vegetated tidal flats was not continuously field tracked because of the low density of S. japonica and poor accessibility. The boundary of TerraSAR-X derived S. japonica was compared at the northwestern island, and 70.6% of the two boundaries were in agreement, as in Fig. 7(f). The results demonstrate that a radar backscattering analysis of TerraSAR-X images is a suitable approach for mapping the distribution of different halophyte species over tidal flats as well as their seasonal variation. 5. Conclusions The accurate mapping and monitoring of salt marshes is important in understanding their responses to sea-level changes, as well as marsh accretion and anthropogenic modification. In this study, high-resolution TerraSAR-X was used to map different halophytes. The most significant result is the possibility of halophyte species mapping over tidal flats and determination of their seasonal variation using radar backscattering analysis. Two typical salt marsh vegetation types in the study area showed distinctive Xband radar backscattering features, fitting for either an annual plant or a perennial grass. The typical perennial salt marsh grass Phragmites australis demonstrated consistent X-band radar backscattering throughout the year, with some slightly higher values in the off-season. The annual plant Suaeda japonica begins to grow in March, consequently increasing radar backscattering throughout the summer. In September, the chlorophyll content of S. japonica begins to decrease, but X-band radar backscattering remains high until November, when the plant structure is collapsed by tidal currents. These results indicate relatively high X-band radar backscattering from May through October, as well as a cyclic phenological pattern in terms of backscattering. While P. australis is relatively stable, S. japonica represents the frontier of the seaward expansion of salt marshes in response to sea level changes. Although areas of S. japonica concentration are naturally mixed with tidal flats, the border between S. japonica and bare tidal flats can be distinguished based on seasonal backscattering coefficient behavior. According to X-band backscattering analysis, the seasonal variation for Phragmites australis was not significant (1.06 dB standard deviation). Seasonal variation for Suaeda japonica was, however, more significant (1.47 dB standard deviation), with a pattern of

375

lower backscattering in winter that increased in the spring to reach a peak in the summer. The border between S. japonica and P. australis was determined based on the higher sigma naught of P. australis in winter during ebb tide, at approximately 2.34 dB. Winter is the best season to distinguish between P. australis and S. japonica, while S. japonica exhibited the highest backscattering in summer at HH polarization. The ebb tide during the off-season provided the best conditions for distinguishing the two species. Dual-polarized data was also used to assess the impact of shorter wavelengths on the backscattering properties of halophytes. Although this research has been focused on Suaeda japonica as an annual salt marsh plant which is popular in the study area, Suaeda is a globally popular genus thriving in salt marsh (Ungar and Boucaud, 1974). There are about 110 species in the Suaeda genus which thrive in salty habitats. For instance, the most common one in northwestern Europe is Suaeda maritima and is also known in English as seablite. It grows along the coasts, especially in salt marsh area, and is dominant annual plants with 10e20 cm height in northwestern Europe such as England and Portugal. Suaeda australis, with 10e90 cm height, is a common salt marsh plant in Australia (Neumeier, 2005). Therefore, the approach demonstrated in this study, identifying and differentiating annual salt marsh vegetation from perennial grass using X-band SAR, can be further extended to all salt marshes with short annual plants in the world. However, the radar backscattering characteristics and their seasonal variation for each species should be examined first at each coastal wetland. This study demonstrated the potential of space-borne high resolution X-band SAR systems, such as TerraSAR-X and CosmoSkyMed, for monitoring salt marshes. Unfortunately, high resolution X-band SAR data has only been available to the public since 2007, and a five-year data span is too short to conclude halophyte migration and succession. Consequently, this research is limited in terms of future long-term monitoring. Follow-up studies with more sophisticated approaches are necessary to improve classification accuracy. Future investigation for wetland studies should be devoted to radar interferometry and polarimetric analysis utilizing full polarization data sets. As Hong et al. (2010) reported previously, interferometric pairs obtained by high resolution SAR systems with short temporal baselines would have the advantage of monitoring the growth and seasonal variation of wetland vegetation. Although HH polarization produces maximum backscattering on wetlands and effectively discriminates the boundaries of different halophyte species, a single polarization may not be sufficient for salt marsh study. The X-band VV/VH dual polarization is generally very useful in discriminating salt marshes from tidal flats. VH polarization sometimes, however, produces strong radar returns on intertidal channels, which can lead to the misclassification of salt marshes. Polarimetric analysis using quad-polarization systems would be more effective than single- or dual-polarization analysis. When the microwave scattering mechanisms associated with herbaceous halophytes and the surrounding areas are fully understood via a polarimetric scattering decomposition approach, it will greatly contribute to tidal flat studies. Acknowledgments This research was funded by the NSL (National Space Lab) program through the Korea Science and Engineering Foundation, funded by the Ministry of Education, Science and Technology (M105DA000004-08D0100-0011A) and by the public application research of a satellite data project funded by the Korea Aerospace Research Institute. The TerraSAR-X data used in this study were provided by DLR as a part of TerraSAR-X AO project (COA0047).

376

Y.-K. Lee et al. / Estuarine, Coastal and Shelf Science 115 (2012) 366e376

References Adam, E., Mutanga, O., Rugege, D., 2010. Multispectral and hyperspectral remote sensing for identification and mapping of wetland vegetation: a review. Wetlands Ecology Management 18, 281e296. Adam, P., 1990. Saltmarsh Ecology. Cambridge University Press, London, p. 461. Baghdadi, M., ZeibiI, M., Delorme, A., 2008. Assessment of the ASAR sensor radiometric quality in comparison to ERS-2 and RADARSAT-1 SAR data. International Journal of Remote Sensing 29 (16), 4653e4665. Baker, C., Lawrence, R., Montagne, C., Patten, D., 2006. Mapping wetlands and riparian area using Landsat ETMþ imagery and decision-tree-based models. Wetlands 26, 465e474. Bertness, M.D., Ewanchuk, P.J., Silliman, B.R., 2002. Anthropogenic modification of New England salt marsh landscapes. Processing of the National Academy of Society 99 (3), 1395e1398. Bourgeau-Chavez, L.L., Kasischke, E.S., Brunzell, S.M., Dudd, J.P., Smith, K.B., Frick, A.L., 2001. Analysis of space-borne SAR data for wetland mapping in Virginia riparian ecosystem. International Journal of Remote Sensing 22 (18), 3665e3687. Burgess, N.D., Evans, C.E., 1989. The Management of Reedbeds for Birds. The Royal Society for the Protection of Birds, p. 78. Choi, J.K., Eom, J., Ryu, J.H., 2011. Spatial relationships between surface sedimentary facies distribution and topography using remotely sensed data: example from the Ganghwa tidal flat, Korea. Marine Geology 280, 205e211. Crevier, Y., Pultz, T., Lukowske, T., Toutin, T., 1996. Temporal analysis of ERS-1 SAR backscatter for hydrology application. Canadian Journal of Remote Sensing 22, 65e77. Englhart, S., Keuck, V., Siegert, F., 2011. Aboveground biomass retrieval in tropical forests - the potential of combined X- and L-band SAR data use. Remote Sensing of Environment 115, 1260e1271. Ferretti, A.C., Prati, C., Rocca, F., 2001. Permanent scatterers in SAR interferometry. IEEE Transactions on Geosciences and Remote Sensing 39 (1), 8e20. Gade, M., Alpers, W., Melsheimer, C., Tanck, G., 2008. Classification of sediments on exposed tidal flats in the German Bight using multi-frequency radar data. Remote Sensing of Environment 112, 1603e1613. Gardner, L.R., Porter, D.E., 2001. Stratigraphy and geologic history of a southeastern salt marsh basin, North Inlet, South Carolina, USA. Wetlands Ecology and Management 9, 371e382. German Aerospace Center (DLR), 2009. TerraSAR-X ground segment basic product specification document, Doc. TX-GS-DD-3302 (1.6), 1e108. Hayakawa, K., Agarie, S., 2010. Physiological roles of betacyanin in halophyte, Suaeda japonica Makino. Plant Product Science 13 (4), 351e359. Henderson, F.M., Lewis, A.J., 2008. Radar detection of wetland ecosystems: a review. International Journal of Remote Sensing 29 (20), 5809e5835. Hong, S.H., Wdowinski, S., Kim, S.W., 2010. Elevation of TerraSAR-X observation for wetland InSAR application. IEEE Transactions on Geoscience and Remote Sensing 48 (2), 864e873. Huang, S., Potter, C., Crabtree, R.L., Hager, S., Gross, P., 2010. Fusing optical and radar data to estimate sagebrush, herbaceous, and bare ground cover in Yellowstone. Remote Sensing of Environment 114, 251e264. Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment 25, 53e70. Infoterra, 2008. Radiometric Calibration of TerraSAR-x Data (Beta Naught and Sigma Naught Coefficient Calculation). Infoterra, Doc. TSXX-ITD-TN-0049radiometric_calculations _l1.00, 1e16. Kasischke, E.S., Bourgeau-Cahvez, L.L., 1997. Monitoring South Florida wetlands using ERS-1 SAR imagery. Photogrammetric Engineering and Remote Sensing 33 (3), 281e291. Kasischke, E., Melack, J., Dobson, M., 1997. The use of radars for ecological application: a review. Remote Sensing of Environment 59, 141e156. Ke, Y., Quackenbush, L.J., Im, J., 2010. Synergistic use of QuickBird multispectral imagery and LIDAR data for object-based forest species classification. Remote Sensing of Environment 114, 1141e1154. Kirwan, M.L., Murray, A.B., 2007. A coupled geomorphic and ecological model of tidal marsh evolution. Proceedings of the National Academy of Sciences of the United States of America 105 (15), 6118e6122. Kwoun, O.I., Lu, Z., 2009. Multi-temporal RADARSAT-1 and ERS backscattering signatures of coastal wetlands in southeastern Luisiana. Photogrammetric Engineering & Remote Sensing 75 (5), 607e617. Lang, M.W., Kasischke, E.S., Prince, S.D., Pittman, K.W., 2008. Assessment of C-band synthetic aperture radar data for mapping and monitoring coastal plain forested wetlands in the Mid-Atlantic Region, U.S.A. Remote Sensing of Environment 112, 4120e4130. Lee, J.S., Ihm, B.S., 2004. Growth strategies of four salt marsh plants on Mankyung river estuary in Korea. Ecological Research 19, 37e42. Lee, S.K., Hong, S.H., Kim, S.W., Yamaguchi, Y., Won, J.S., 2006. Polarimetric features of oyster farm observed by AIRSAR and JERS-1. IEEE Transactions on Geoscience and Remote Sensing 44 (10), 2728e2735.

Lopes, A., Touzi, R., Nezry, E., 1990. Adaptive speckle filters and scene heterogeneity. IEEE Transactions on Geoscience and Remote Sensing 28 (6), 992e1000. Lu, Z., Meyer, D.J., 2002. Study of high SAR backscattering caused by an increase of soil moisture over a sparsely vegetated area: implications for characteristics of backscattering. International Journal of Remote Sensing 23 (6), 1063e1074. Lucas, R.M., Mitchell, A.L., Rosenqvist, A., Proisy, C., Melius, A., Ticehurst, C., 2007. The potential of L-band SAR for quantifying mangrove characteristics and change: case studies from the tropics. Aquatic Conservation: Marine and Freshwater Ecosystems 17 (3), 245e264. Min, B.M., 2005. Ecological characteristics of a Suaeda japonica population and the effects of early-season air temperatures on population formation. Journal of Plant Biology 48 (4), 411e421. Mitsch, W.J., Gosselink, J.G., 2007. Wetlands, fourth ed. John Wiley & Suns, p. 582. Neumeier, U., 2005. Quantification of vertical density variations of salt-marsh vegetation. Estuarine, Coastal and Shelf Science 63, 486e496. Orson, R.A., Brown, L., Sharp, P.C., 1997. Ecological context and vegetation restoration. In: Casagrande, D.G. (Ed.), Restoration of an Urban Salt Marsh. Yale School of Forestry and Environmental Studies Bulletin Series, vol. 100, pp. 136e150. New Haven, Connecticut. Ozemi, S.L., Bauer, M.E., 2002. Satellite remote sensing of wetlands. Wetlands Ecology and Management 10, 381e402. Pal, M., Mather, P., 2003. An assessment of the effectiveness of decision tree methods for land cover classification. Remote Sensing of Environment 86 (4), 554e565. Park, J.W., Lee, Y.K., Won, J.S., 2009. Investigation of intertidal zone using TerraSARX. Korean Journal of Remote Sensing 25 (4), 383e389. Parmuchi, M.G., Karszenbaum, H., Kandus, P., 2002. Mapping wetlands using multitemporal RADARSAT-1 data and a decision-based classifier. Canadian Journal of Remote Sensing 28, 175e186. Philipp, K.R., Field, R.T., 2005. Phragmites australis expansion in Delaware bay salt marshes. Ecological Engineering 25, 275e291. Poulin, B., Davranche, A., Lafebvre, G., 2010. Ecological assessment of Phragmites australis wetlands using multi-season SPOT-5 scenes. Remote Sensing of Environment 114 (7), 1602e1609. Ramsey III, E., 1998. Radar remote Sensing of wetlands. In: Lunetta, R., Elvidge, C. (Eds.), Remote Sensing Change Detection: Environmental Monitoring Methods and Applications, pp. 211e243. Rizzetto, F., Tosi, L., 2011. Aptitude of modern salt marshes to counteract relative sea-level rise, Venice Lagoon (Italy). Geology 39 (8), 755e758. Rizzetto, F., Tosi, L., 2012. Rapid response of tidal channel networks to sea-level variations (Venice Lagoon (Italy). Global and Planetary Change 92-93, 191e197. Ryu, J.H., Won, J.S., Min, K.D., 2002. Waterline extraction from Landsat TM data in a tidal flat: a case study in Gomso bay, Korea. Remote Sensing of Environment 83 (3), 442e456. Slatton, K.C., Crawford, M.M., Chang, L.D., 2008. Modeling temporal variations in multipolarized radar scattering from intertidal coastal wetlands. ISPRS Journal of Photogrammetry & Remote Sensing 63, 559e577. Strozzi, T., Teatini, P., Tosi, L., 2009. TerraSAR-X reveals the impact of the mobile barrier works on the Venice coastal stability. Remote Sensing of Environment 113, 2682e2688. Svoray, T., Shoshany, M., 2003. Herbaceous biomass retrieval in habitat of complex composition: a model merging SAR images with unmixed Landsat TM data. IEEE Transactions on Geoscience and Remote Sensing 41 (7), 1592e1601. Townsend, P.A., 2002. Relationships between forest structure and the detection of flood inundation in forested wetlands using C-band SAR. International Journal of Remote Sensing 23 (3), 443e460. Ulaby, F.T., Moore, R.K., Fung, A.K., 1982. Microwave Remote Sensing: Active and Passive. In: Radar Remote Sensing and Surface Scattering and Emission Theory, vol. 2. Wesley, Advanced Book Program, Reading, Massachusetts, p. 609. Ungar, I.A., Boucaud, J., 1974. Comparison of seed proteins in the genus Suaeda (Chenopodiaceae) by means of disc gel electrophoresis. American Journal of Botany 61 (4), 325e330. Warren, R.S., Niering, W.A., 1993. Vegetation change on a northeast tidal marsh: interaction of sea-level rise and marsh accretion. Ecology 74 (1), 96e103. Wright, C., Gallant, A., 2007. Improved wetland remote sensing in Yellowstone National Park using classification trees to combine TM imagery and ancillary environment data. Remote Sensing of Environment 107, 582e605. Woo, H.J., Je, J.G., 2002. Changes of sedimentary environments in the southern tidal flat of Kanghwa island. Ocean and Polar Research 24 (4), 331e343. Zebker, H.A., Rosen, P.A., Hensly, S., 1997. Atmospheric effects in interferometric synthetic aperture radar surface deformation and topographic maps. Journal of Geophysical Research 102 (B4), 7547e7563. Zhang, M., Ustin, S., Rejmankova, E., Sanderson, E., 1997. Monitoring pacific coast salt marshes using remote sensing. Ecological Applications 7 (3), 1039e1053.