Spectral Characterization of Mangrove Leaves in the Brazilian Amazonian Coast: Turiaçu Bay, Maranhão State

Mangrove communities are tropical systems which have fewer species than tropical forests, especially in Latin Amer-ica and display a single architecture, usually lacking the various strata commonly found in other forest ecosystems. The identification of mangrove communities by orbital data is not a difficult task but the most interesting challenge is to identify themselves by the dominant species. The first step toward that floristic identification is the spectral characterization of detached leaves. Leaves from four species of mangrove trees were spectrally characterized considering the Directional Hemispherical Reflectance Factor (DHRF) determined through radiometric measurements using an integrating sphere LICOR 1800 attached to a spectroradiometer SPECTRON SE-590. In the visible bands (0.45-0.69 µm) the button-shaped mangrove Conocarpus erectus was brighter and the red mangrove Rhizophora mangle was darker than the other two species which shows very close DHRF values. Otherwise the black mangrove Avicennia germinans and the white mangrove Laguncularia racemosa can be distinguished from one another in the Near Infra Red (NIR) region (0.76-0.90 µm and in this region of the spectrum the DHRF of C. erectus and R. mangle become very close.


INTRODUCTION
Mangroves are valuable coastal systems (Cintrón and Schaeffer-Novelli, 1992, Robertson, 1992. They are known to protect the environment against erosion and flooding, to process pollutants and waste discharge contaminated with heavy metals (Silva et al., 1990), to export organic matter to estuarine areas and thereby increasing fisheries yield Schaeffer-Novelli, 1983, Twilley, 1985), as to shelter threatened species (Rebelo-Mochel et al., 1991, Rebelo-Mochel, 1993.
The Amazonian coast, comprised by the Brazilian states of Maranhão, Pará and Amapá, contains the largest mangrove system in the world, measuring 8,900 km2 (Kjerfve et al. (2002). The state of Maranhão is located in the north coast of Brazil and it contains about 750,000 ha of mangroves (Sant'Anna andWhately, 1981, Kjerfve and. The macro-tidal coast of Maranhão contains 500,000 ha of these mangroves, more than 30% of the total for all Brazil. Biomass may reach 280 ton/ha or more, with trees taller than 40 m in some places (Cintrón and Schaeffer-Novelli, 1983, Lacerda and Schaeffer-Novelli, 1992, Kjerfve and Lacerda, 1993, Rebelo-Mochel, 1995. Mangrove species in this region are three species of red mangroves Rhizophora mangle, Rhizophora racemosa, Rhizophora harrisonii, two species of black mangroves Avicennia germinans, Avicennia schaueriana, the white mangrove Laguncularia racemosa and the buttom-shaped mangrove Conocarpus erectus (Santos, 1986, Rebelo-Mochel et al 1991.
The west coast of Maranhão is sharply cut by rivers, streams and tidal channels filled with clay and silt which foster the development of mangroves. The mouth of Turiaçu River is one of the largest entrances of the region with a surface area of 900 km 2 . Satellite remote sensing might be the most cost-effective technique for evaluating mangrove distribution and biomass.
Long and Skewes (1996) described a technique using Thematic Mapper (TM) satellite images to inventory Australian mangroves. The use of Multispectral Scanner (MSS) and TM satellite images to identify mangrove areas along the Brazil coast has previously been accomplished by Herz et al. (1984), Herz and Jaskow (1985), , Pires (1986), Pires andHerz (1987, 1988) and Herz (1991). Indeed, delimitation of mangrove occurrences using multispectral data is not a hard task considering the spectral contrast between this morphological unit and other land cover types. The problem however is set when it is necessary to identify different mangrove types, especially species composition, due to the existence of mixed stands and the characteristics of coastal land forms. But the potential advantage in that identification procedure through remote sensing technology is the fact that there are few species on mangrove environments compared with other tropical rain forests. Indeed, the Atlantic coast of Americas has 3 species of Rhizophora, 2 of Avicennia, 1 of Conocarpus and 1 of

Laguncularia.
Some attempts have been made to describe mangrove communities by their spectral reflectance in the field. These techniques are generally expensive and laborious, but there are some aspects that become promising in mangrove studies. As mentioned above, mangrove communities have fewer species than tropical forests, especially in Latin America and display a single architecture, usually lacking the various strata commonly found in other forest ecosystems. Besides, mangroves grow in coastal plains with low relief and estuarine waters, forming extensive zones dominated by 1 or 2 species, with relative homogeneous physiography.
Considering leaves as the most important part of a plant in terms of its interaction with the electromagnetic radiation, the spectral reflectance of leaves has great influence on the canopy bidirectional reflectance. Thus, the spectral characterization of leaves can give information about the opportunities and constrains of the canopy identification. These facts reinforce our belief that in mangrove environment it would be possible to find a strong correlation between spectral reflectance of detached leaves and the spectral response of mangroves canopies INPE ePrint: sid.inpe.br/ePrint@80/2006/07.31.19.21 v1 2006 measured by an orbital located sensor. One most important aspect to be considered here is related to the real possibility to spectrally characterize leaves of each mangrove species, in other words, if we can segregate mangrove species from their leaves reflectance spectra. The complete understanding of that aspect is the first step to be carried out toward the use of orbital data for discriminating mangrove species.
The main goal of this work is to characterize the reflectance of leaves of four mangrove species that occur in the State of Maranhão, Amazonian coast of Brazil, in order to distinguish them one from another. This spectral characterization can later help the inventory and description of mangroves communities in that region utilizing orbital data.

Study Area
Turiaçu Bay is located between latitudes S 1º 20' -1º 45' and longitudes W 45º 10' -45º 20' in the western coast of Maranhão State, northern Brazil ( Figure 1). The extension of the ecosystem, the presence of pure and mixed stands of mangrove species and the low human density and high pristine areas, make this region profitable for regional studies. The field work was performed in Bacuri municipality, right side of Turiaçu Bay, during July 1997. There were selected 26 sample plots over 4 mangrove environments: fringe, basin, riverine and hypersaline habitats.

Leaf Sampling and Structural Data
The data collection was carried out inside plots of about 30m x 30m in each of the 26 selected sample plots. Leaves on the top of the canopy were detached from arbitrarily selected trees by climbing and cutting the upper branch exposed to the sun. There were collected about 42 leaves from each mangrove species that were find in each sample plot. The leaves were stocked in plastic bags and carried in ice boxes according to Sousa et. al (1996). All the reflectance measurements were performed 8 hours after leaves were detached since  didn't find any significant changes in the leaf reflectance measurements up to 10 hours after leaf detaching.
Structural data comprised tree height measured with a Ranging TLR 75 Telemeter and the diameter at the breast height with a calibrated ribbon, for all the trees over 2.5 cm in diameter as stated in Schaeffer-Novelli et al. (1986). All the 26 sites were referenced with a Scout GPS Trimble Navigation with an error of less than 90m in the field.
Green leaves were collected and taken to the Laboratory of Botany of the Universidade de Taubaté where their internal structures were examined taking into account the expected differences between species.

Radiometric Measurements
Radiometric measurements were performed with an integrating sphere LICOR 1800 attached to a spectroradiometer SPECTRON SE-590 that runs from 0.4 µm to 1.1 µm, with 0.03 µm of spectral resolution. As the minimal number of leaves necessary to guarantee confidence to the statistical tests was not known in the beginning of the field work, Directional Hemispherical Radiance was measured for each of 42 detached leaves from each mangrove species in all the 26 sample plots by placing the addaxial (upper) surface of the leaf in the integrating sphere. After each group of 10 leaves, the reference panel (BaSO 4 ) inside the sphere was measured to calculate the Directional Hemispherical Reflectance Factors (DHRF). DHRFs were calculated considering the group of leaves closest (in time terms) to the spectra radiance of the reference panel.
The DHRF is given by the equation: The data were processed with the software "ESPECTRO" developed by the Radiometry fine ones in DHRF of Eucalyptus saligna detached leaves. The authors found no differences using both spectral resolutions on an ANOVA statistical procedure. Thematic Mapper bands were adopted in this study with the aim of checking out possible modeling integrating radiometric data of leaves and Landsat/TM ones, which is the object of another study.

Statistical Analysis
To determine the minimum number of leaves necessary to set confidence to the spectral characterization, it was considered (for each spectral band) the equation: where: n = minimum number of leaves; t = 1.83 (for 9 degrees of freedom and 10% of significance); d = minimum difference to be ratified (in this case, 10%); CV = coefficient of variation between DHRFs.
Ten spectra from all available ones for each species were randomly selected in order to determine the n values for each band. The minimum number was identified as the higher n value reached in this procedure, independent on the species.
Variance analysis (ANOVA) was applied for the averages of each spectral band to ratify if there were significant differences between the DHRFs for each species in each spectral band.
The F value was considered at 10% of significance. After finding significant differences we tested the differences between the averages with Duncan's test also at 10% of significance.

RESULTS AND DISCUSSION
Considering the 26 sample plots, C.erectus was found in only one of them, L.racemosa in nine, A.germinans in nineteen and R.mangle in eighteen.
Results for the minimum number of leaves needed to set confidence to the statistical analysis were: 6 leaves for R. mangle; 8 for L. racemosa; and 33 for A germinans. It was acquainted the spectra of 1,848 leaves (total) r considering the 4 mangrove species in the 26 sample plots. 9 The complete DHRF spectra and the DHRF for each spectral band of mangrove leaves are shown in Figure 2. The DHRFs presented on Figure 2 are average values calculated from all available spectra acquired from the 26 sample plots in the field. Some data were lost in data processing task, thus we decided to consider 39 leaf spectra in each of these sites to perform the statistical analysis for all species.
[Insert Figure  As expected, in bands 1 and 3 the DHFRs of all species were low due to photosynthetic pigments activity (Ponzoni, 2001) with lower standard deviation values comparing with band 2 results. In band 4 both DHFR and standard deviation were higher than those observed in the visible bands.

A scatter plot composed by DHFR values from band 4 (y axe) and band 3 (x axe) is presented
in Figure 4 in order to allow a complementary evaluation of the spectral differences between the leaves of the four species.
[Insert Figure 4 about here] Considering the places of the DHFR values (four species) in the bi-dimensional scatter plot presented in Figure 4 one can observe that both L.germinans and C.erectus (even considering only one point for this last one) could be individualized from the others two species, besides being individualized themselves. The DHFR's dispersion for R.mangle was lower than those observed for the others, indicating higher homogeneity of DHFR values for this species.
The performance and results of the Analysis of Variance (ANOVA) are shown in was higher and L. racemosa was lower than the other 2 species which spectra became closer.
The spectral bands used in this study showed satisfactory results, finer bands, with more wavelengths seem to have little advantage for characterizing leaf spectral reflectance (Ramsey and Jensen, 1996).
[Insert Table 1 about here] [Insert Table 2 about here] Radiant energy interacts with the leaf structure by absorption and by scattering ( Gates et al.,1965). In the visible bands (1,2 and 3) Cardoso and  found similar spectral reflectance between 5 species of terrestrial shrubs, differences appearing in the NIR band 4 probably due to leaf mesophyll influence. Gausman (1974) demonstrated that compacted leaves had lower reflectance than leaves with porous mesophyll or with many cell wall-air space interfaces. The anatomical analysis of the epidermis of mangrove leaves showed all the species are dorsiventral, or in other words with a marked dissimilarity between the upper and the lower halves, and a colorless tissue represented by a thick addaxial hypodermis (Tomlinson, 1986). The white mangrove L.racemosa has the most compacted epidermis below the cuticular membrane of all the species examined. It has cuticular membrane and flanges strongly developed besides having a concentration of palisade layers below a very thin lacunal parenchyma. This is coherent with the low brightness explained by the lowest values of DHFR in the NIR bands (Figures 3 and 4). The red mangrove R.mangle and the black mangrove A.schaueriana have a similar arrangement in the structure of the epidermis but Avicennia has far well developed lacunae in the parenchyma between the cuticular membrane and the palisade layers. This may explain the higher reflectance of this species in the NIR region. In this study we were not able to make anatomical analysis of C.erectus but Stace (1965) made a review of Combretaceae and reports that C. erectus has narrow-walled epidermal cells arranged in a very regular way. The weak arrangement of cells of C. erectus could be related to the high brightness of this species in the NIR region.
In the visible bands the spectral response should be explained due to differences on the chemical components (differences), including photosynthetic pigments amount (Ponzoni, 2001). As no chemical analysis were performed, it was impossible to compare chemical differences with radiometric ones. Gitelson and Merzlyak (1996) verified that even a small increase in the chlorophyll content of green leaves lead to a significant decrease in the reflectance, so it can be possible that in the visible bands the significant differences between DHFR values could be associated to differences on the chlorophyll content (differences).
Anyway, the significant differences in DHFR were quite important even at laboratory investigation since they are at least (because at least they are) indicatives that orbital data from visible and NIR spectral regions could also be explored in mangrove species identification.
The optical analysis of leaf epidermis showed that Rhizophora has far more chloroplasts than the other species although the total amount of chlorophyll was not measured in this study.
Photosynthetic and chemical components probably are responsible for the lowest brightness of Rhizophora mangle in the visible bands. The reflectance of the leaf water content is measured by the wavelengths of 1. 45 -1.93 µm (Thomas et al., 1971) and it was not discussed in this paper.
In spite of Ponzoni and Gonçalves (1999) have recommended the usage of larger spectral bands, it should be interesting to consider the possibility to obtain orbital hyperspectral (Hyperion data, for instance) from mangrove areas. Considering the greater amount of mangroves spectra that could be extracted from the hyperspectral images, statistical procedures like those utilized by Galvão et al. (2004), for example, could be applied in the entire spectral range data in order to identify differences or similarities between mangroves canopies.

CONCLUSIONS
The spectral identification of the 4 mangrove species was possible. All mangrove species were distinguished in band 1. In the visible bands 2 and 3 the spectra of A germinans and L.
racemosa were confused and otherwise in the NIR band 4 the spectra of C. erectus and R.
mangle would be confused.
Considering the mangrove species in this study Conocarpus erectus had the highest reflectance and Rhizophora mangle had the lowest in the visible bands, while in the NIR region the leaf reflectance of Avicennia germinans was higher and Laguncularia racemosa was lower than both of the other 2 species.
Disregarding the aspects related to the architecture of canopies dominated by these species one should expect when using airborne or orbital data high reflectance values in canopies of     List of Tables   Table 1 -Results of ANOVA application on all species and spectral bands.    Band 1 * Significative at 10% level of significance. Table 1