Hyperspectral imaging for mapping ...

5 downloads 94 Views 9MB Size Report
strongly correlated to plant biomass and physiology (Curran et al., 1990; ...... (Tsai & Philpot, 1998), but this adds variability to derivative indices, linked to the ...... Geology, BS Siegal and AR Gillespie eds, John Wiley, New York, 1980, 5-45.
Hyperspectral imaging for mapping microphytobenthos in coastal areas Méléder V., Launeau P., Barillé L., Combe J-P., Carrère V., Jésus B., Verpoorter Ch. ABSTRACT Estuaries and coastal areas are highly important natural habitats, amongst the most productive marine ecosystems on earth. They provide vital ecosystem services to mankind and are particularly valuable as habitats and feeding grounds for a variety of organisms, such as birds, shellfish, demersal fish and invertebrates, which often support the local economy. However, these ecosystems are under strong anthropogenic pressure caused by intensive human exploitation (e.g. industry, fishery, tourism, domestic sewers). One of the major primary producing groups of these areas are microscopic algae that inhabit intertidal sediments (microphytobenthos, MPB), even though their photosynthetic activity is restricted to the narrow illuminated layer of surface sediment, where they form dense aggregates called biofilms. Microphytobenthic biomass often exceeds that of the phytoplankton in the overlying tidal waters. MPB provides an important energy source for the estuarine food web, supplying up to 45% of the organic resources of an estuary, and has a central role in moderating carbon flow in coastal sediments. Some relationships have also been established between the surface concentration of MPB and critical erosion shear-stresses, therefore providing important information about sediment erodability and dynamics in coastal areas. These vital ecosystem functions emphasize the need to detect and monitor microphytobenthic biofilms accurately at the ecosystem scale. However, sampling intertidal flats is a complex, and frequently difficult, task. Intertidal sediments are often dominated by small sediment particles (< 63 µm) forming large mudflats that are very hard to navigate or walk. It is therefore almost impossible to monitor MPB spatial dynamics using regular field sampling. This has led to a growing interest in the use of remote sensing techniques to detect microphytobenthic assemblages. Microphytobenthos can exhibit considerable taxonomic and pigment variability, which is lost when chlorophyll a is used as a single descriptor of the community. These different microalgae groups display different pigment compositions with specific absorption features, which can be captured by hyperspectral sensors. In this chapter, we will first introduce the principles of imaging spectrometry (also called hyperspectral remote sensing). Because of its high spectral resolution (from 1 to 35 nm) and continuous sampling of the solar spectrum (more than 100 images acquired contiguously and simultaneously), imaging spectrometry can identify pigments based on their specific absorption features. We then present some algorithms developed to map microphytobenthos in coastal areas. Analyzing the strength and shape of the specific absorption features leads to quantitative estimations of pigment concentration that would be impossible to obtain from standard multispectral images. Finally, imaging spectrometry also provides access to information at the subpixel level, dealing with the problem of intimate (non-linear) versus areal (linear) mixtures.

Key words: imaging spectrometry, spectral signature, pigments, microphytobenthos, concentration, mixtures, intertidal zone.

1 INTRODUCTION Microphytobenthos are an artificial grouping of algae and photosynthetic bacteria (e.g. cyanobacteria, euglenids, crysophyceans, dinoflagelates and diatoms) (MacIntyre et al., 1996) that colonize benthic substrata. These microorganisms frequently accumulate at the sediment surface at low tide in high densities forming photosynthetic biofilms (Figure 1) and they perform many important ecosystem functions in estuarine and coastal environments (Boogert et al., 2006). They are important in the mediation of nutrient fluxes between sediment and the water column (e.g. Cabrita and Brotas, 2000) and exude extracellular polymeric substances (EPS) which play a role in sediment stabilization (e.g. Paterson, 1989).They can be responsible for more than 50% of estuarine total primary productivity (Underwood and Kromkamp, 1999), frequently being the major primary producers in estuarine mudflat systems (MacIntyre et al., 1996; Underwood and Kromkamp, 1999), where they represent a significant food source for grazers (e.g. Middelburg et al., 2000).

1

b

c

a Figure 1. Microphytobenthos from the mudflat scale to the cell scale. (a) Bourgneuf Bay DAIS image (RGB colors associated with channels at 496, 675 and 798 nm, respectively). Green: microphytobenthos; light red: terrestrial vegetation; dark blue: macroalgae; white: sand; gray: mud; blue: water. From Combe et al. (2005). (b) Field biofilm covering mudflat at low tide. (c) Biofilm seen by Low Temperature Scanning Microscopy: sediment surface fully covered with diatom cells. Above a densely packed biofilm of smaller diatoms some larger cells of Pleurosigma angulatum, another diatom species, are visible. Scale bar is 100 µm. The taxonomic composition of microphytobenthos assemblages (hereafter called MPB) can show significant variability and is the result of a complex interaction of biotic and abiotic factors. The most important of these are light, salinity, nutrients, grazing and sediment type (e.g. Oppenheim, 1988; Pinckney and Sandulli, 1990; Underwood, 1994; Delgado et al., 1991; Van der Grinten et al., 2004; Van der Grinten et al., 2005; Jesus et al., 2005). Microphytobenthic biomass is partially controlled by tidal exposure, dynamic and flow and by the specific characteristics of the sediment they colonize (e.g. Brotas et al., 1995; Jesus et al., 2009). Sediment type also appears to be the most important variable in the structuring of MPB taxonomical composition, with muddy sediments generally dominated by diatoms while sandy sediments frequently exhibit more diverse assemblages including cyanobacteria and euglenids (e.g. Underwood and Barnett, 2006; Jesus et al. 2009). MPB has been described as a ‘secret garden’ (MacIntyre et al., 1996) for its roles in the functioning of intertidal mudflat systems, yet its structure and dynamics at the ecosystem level have seldom been quantified. MPB biofilms can indeed cover several hectares at low tide, which are often difficult to reach and sample, particularly on mudflats. Only a few studies have assessed the structure (diversity and biomass) and 2 2 dynamics (spatio-temporal changes) of MPB at mesoscale (several m ) and macroscale (several km ) (Brotas and Plante-Cuny, 1998; Lucas and Holligan, 1999; Méléder et al., 2005, 2007; Jesus et al., 2009) as 2 compared to microscale (few m ) (e.g. Cariou-Le Gall and Blanchard, 1995; Kelly et al., 2001; Riaux-Gobin et al., 1987; Wiltshire, 2000). The work of Guarini et al. (1998) is the only study that has mapped microphytobenthos structure at macroscale, by means of a geostatistical approach based on an extensive field study, using a systematic sampling grid over the area to be mapped. Remote sensing, using airborne and space borne sensors, would appear to provide a simpler alternative for studying and mapping the communities formed by unicellular photosynthetic organisms. . 2

The first study using remote sensing to map MPB assemblages was performed by Uno and Gotoh (1992). They showed the area of greatest abundance of MPB using the infrared band of two space-borne sensors: LANDSAT-TM and MOS-MESSR. These sensors, described as multispectral because they have few spectral bands (less than ten) which are broad, of variable width and are non-contiguously. Moreover, spatial 2 resolution is coarse (a few hundred km ). Thus, these attributes are limited to measuring and mapping vegetation and MPB biomass. Méléder et al. (2003b) were confronted with these limitations in their mapping of MPB using SPOT imagery. The use of a hyperspectral sensor by Combe et al. (2005) overcame these multispectral limitations to produce the first MPB biomass map at the ecosystem level. This chapter presents the use of hyperspectral remote sensing to study and map MPB assemblages at various scales, from field spectroradiometry to airborne imagery. Hyperspectral technology and capabilities are described in section 2, with the main processes involved in remote sensing. The link between the optical properties of MPB, its diversity, biomass, habitat type (mud/sand) and eco-physiology, is detailed in section 3, with an emphasis on the absorption processes due to photosynthetic and accessory pigments. Quantification of MPB at macroscale is developed in section 4 using a case study along the French Atlantic coast. Comprehensive maps of MPB biomass are presented and critically evaluated, based on non-linear inversion of visible-infrared hyperspectral images.

2 PRINCIPLE OF IMAGING SPECTROMETRY 2.1 What is imaging spectrometry? Imaging spectrometry was originally developed to obtain compositional information from inaccessible planetary surfaces within the solar system (Hunt, 1977; Clark et al., 1990; Goetz, 1992; Pieters and Englert, 1993). In the past twenty-five years, this technique has also shifted towards the observation of the Earth with airborne hyperspectral sensors such as the NASA-JPL Airborne Visible InfraRed Imaging Spectrometer (AVIRIS), Integrated Spectronics HyMap™ (http://www.intspec.com) or the Earth Orbiter 1 (EO-1) hyperspectral sensor Hyperion, the first NASA operational test for a spaceborne Earth orbiting imaging spectrometer (Pearlman et al., 1999). The principle of imaging spectrometry and its main differences with broad-band sensors that are commonly used by the Earth observation community are highlighted in Figures 2 to 4. Landsat Thematic Mapper (TM, 7 bands), SPOT HRV (5 bands), MODIS (36 bands) or ASTER (14 bands) generally undersample the solar electromagnetic spectrum, with discontinuous spectral bands that can be up to several hundred nanometers wide (Figure 2). The wavelength location and width of spectral bands in broad-band sensors are designed with specific applications in mind. For example, the location and width of bands 3 and 4 of the Landsat TM were specifically designed for vegetation mapping. In contrast, imaging spectrometers sample the whole spectrum within a certain wavelength range.

(a)

(b)

Figure 2: Spectra of the mineral alunite shown as measured in the laboratory and for (a) broadband remote sensing instruments and (b) some imaging spectrometers. The FWHM is the full width at half maximum (see Figure 11). The alunite is sample HS295.3B from the USGS spectral library (Clark et al., 1993). Each spectrum is offset upward by (a) 0.6 units and (b) 0.3 units from the one below for clarity. From Clark (1999).

3

Figure 3: 224 band AVIRIS image cube of Moffett Field, San Francisco Bay, California. x and y axes represent spatial data (1024 x 614) as a false-color composite image made to accentuate the structure of the water and evaporation ponds on the right of the image. The z axis represents spectral data as 224 contiguous bands from 400 (“top”) to 2500 nm (“bottom”) in pseudocolor (rainbow). The two horizontal black bands correspond to the saturated atmospheric water vapor bands where all the photons are absorbed, preventing any surface observation (source: http://aviris.jpl.nasa.gov/html/aviris.cube.html). Imaging spectrometers acquire contiguous bands of high spectral resolution (between 10 and 25 nm wide), resulting in what is commonly called an image cube (Figure 3). Imaging spectrometry is defined as “the simultaneous acquisition of images in many narrow, contiguous spectral bands” (Goetz et al., 1985). Depending on the author, it can also be called hyperspectral remote sensing or imaging spectroscopy. A detailed spectrum, comparable to those that can be measured by laboratory instruments, is obtained for each pixel of the image. High spectral resolution reflectance spectra collected by imaging spectrometers enable the direct identification (and in some instances, abundance determinations) of individual materials based upon their reflectance characteristics. Most natural materials of the Earth’s surface have diagnostic features in the 400-2500 nm range of the reflected solar spectrum, most commonly observed by imaging spectrometers. If features observed in the VNIR (from 0.4 to 1 µm) related to pigments or electronic processes due to transition metals such as iron are broad, the diagnostic absorption features for most material in the SWIR (from 1.0 to 2.5 µm) appear as very narrow spectral bands, typically 20-40 nm in width at the halfband depth for vibrational features (Hunt, 1980). Therefore, differences between materials can only be identified if the spectrum is sampled at a sufficiently high resolution. Only spectral imaging systems that acquire data in contiguous 10 nm-wide spectral bands can produce data with sufficient resolution for the direct identification of those materials with diagnostic features. Multispectral sensors cannot resolve these features because their spectral bandwidths are 100-200 nm and are not contiguous. As a result, multispectral broad band images only allow to separate different surface types when imaging spectrometer can identify them.

Figure 4: The imaging spectroscopy concept (source: NEMO). 4

An analysis of imaging spectrometer data provides a detailed spectrum for each picture element (pixel) of the image. Reflectance characteristics of individual materials can be used to map, and sometimes estimate concentrations of minerals (Goetz et al., 1985; Lang et al., 1987; Pieters and Mustard, 1988; Kruse, 1988; Kruse et al., 1993; Crowley, 1993; Boardman and Kruse, 1994; Clark et al., 1996; Boardman and Huntington, 1996; Crowley and Zimbelman, 1996), atmospheric constituents and volcanic gases (Gao and Goetz, 1990; Carrère and Conel, 1993; Borel and Schläpfer, 1996; de Jong and Chrien, 1996; Spinetti et al., 2008), vegetation (Peterson et al., 1988; Gamon et al., 1993; Elvidge et al., 1993; Chen et al., 1996; Green, 1996; Sabol et al., 1996), snow and ice (Nolin and Dozier, 1993; Green and Dozier, 1996), dissolved and suspended constituents and water quality in lakes and other water bodies and the near-shore environment (Hamilton et al., 1993, Carder et al., 1993; Richardson and Ambrosia, 1996; Kruse et al., 1997). Currently, spectrometers are in use in the laboratory, in the field, on aircraft and on satellites (most of them looking at other planets from space). A list of existing imaging spectrometers can be found in annex 1. Several Earth observation missions are also planned for the near future and these are briefly presented in annex 2. Because spectroscopy is sensitive to so many processes, spectra can be very complex. However, it is because of this sensitivity that spectroscopy has great potential as a diagnostic and mapping tool. Reflectance spectroscopy can be used without sample preparation, is non-destructive, and can be carried out remotely from airborne and satellite sensors. The next section explains these general processes and how they influence the spectral signatures of surface materials. 2.2 Factors influencing the spectral signature Natural targets are usually illuminated by the whole hemisphere of the sky and therefore receive both direct solar flux and scattered sky light, as well as a fraction of light scattered from the surface itself due to its structure (Clark, 1999). A proportion of the incident radiation is reflected, either directly from the surface, or after multiple interactions within the surface of the material that is translucent to the incoming radiation. Natural objects are generally not perfectly diffuse (Lambertian) reflectors (i.e. they do not scatter light equally in all directions), and therefore the intensity of the reflection varies with the angle that it leaves the surface. Consequently, the radiation environment comprises two hemispherical distributions of electromagnetic radiation, one incoming and one outgoing (Clark, 1999). These interactions of absorption and reflection form the basis of spectroscopy and hyperspectral analysis. Scattering is also another factor influencing the spectral signature.

Figure 5: Interactions between electromagnetic http://rst.gsfc.nasa.gov/Intro/Part2_3html.html)

radiation

and

a

medium

(source:

2.2.1 Absorption Two main general processes cause absorption bands in the spectra of materials: electronic and vibrational. Burns (1993) examines the details of electronic processes and Farmer (1974) covers vibrational processes. Shorter introductions to the causes of absorption bands in minerals are given by Hunt (1977, 1982) and 5

Gaffey et al. (1993) for the visible and near-infrared. The following sections present some examples for typical Earth surfaces. 2.2.2 Rocks and soils (Figure 6) The reflectance characteristics shorter than 1 µm for minerals that constitute rocks and soils are influenced by the presence of transition metals. Almost all minerals contain some iron, and that metal dominates the shape of most mineral reflectance curves in this spectral region. Charge transfer bands, which are a result of the exchange of electrons between neighboring metal ions, create absorptions in the UV region shorter than 0.4 µm. The wings of these bands extend into the visible portion of the spectrum and are responsible for the general increase in reflectance between 0.4 and 0.8 µm. Electronic transitions in the transition elements, which are due to energy level changes in the d-shell electrons within the crystal field of the mineral, result in absorption features near 0.9 µm (Clark, 1999). As a consequence, some minerals show color due to absorption by what are called color centers. Natural crystals often have lattice defects caused generally by impurities that disturb the periodicity of the crystal. These defects can produce discrete energy levels. Irradiation (e.g. solar UV radiation in this case) provides photon energy to allow the electrons to move into the defect. A good example of color centers is the yellow, purple or blue color of fluorite. See Hunt (1977), Gaffey et al. (1993) and a book by Burns (1993) and references therein for more details. At wavelengths greater than 1.0 µm, vibrational features associated with bound and unbound water become important in determining the reflectance spectrum of the material. Major absorption features are seen at 1.4 and 1.9 µm. There are also strong atmospheric water bands at these wavelengths, making them unusable for used in passive remote sensing, because nearly all of the incoming sunlight is absorbed by the atmosphere. Combination bending-stretching overtones of the fundamental OH vibration at 2.74 µm are seen in the region of 2.1-2.4 µm. Overtones for H2O and CO3 are also found in this region. Minerals such as alunite, calcite, gypsum and many others have uniquely diagnostic reflectance spectra in this region. Some minerals, however, (e.g. the clay minerals halloysite, kaolinite and dickite) have absorption features which are very similar and sometimes difficult to distinguish.

(a)

(b) Figure 6: Examples of characteristic spectra of rock and soil-forming minerals; (a) Fe-bearing minerals; (b) vibrational features. From Clark (1999).

6

2.2.3 Vegetation (Figure 7) The spectral reflectance for vegetation in the 0.4-2.5 µm region does not vary as much as that for minerals. The chlorophyll absorption features are centered at 0.48 and 0.68 µm. The absorption at 0.48 µm is also the result of electronic transitions in the carotenoid pigments, accessory pigments to the chlorophyll pigments in the photosynthetic process. The 0.68 µm absorption is due to electronic transitions in the chlorophyll molecule, centered around the magnesium component of the photoactive site. The area between these two absorption features (i.e. the green wavelengths) is only slightly absorbed by pigments (chlorophylls and carotenoids), and this results in the green color of plants. The steep rise in reflectance of vegetation at wavelengths longer than 0.8 µm is called the red edge of the chlorophyll band. Its position and slope is strongly correlated to plant biomass and physiology (Curran et al., 1990; Gitelson et al., 1996, for example). The spectral region between 0.8 and 1.3 µm is called the near-infrared plateau. The high reflectance in this spectral region is characteristic of leaf tissue, depending strongly on the scattering of photons by leaf cells (see below for the definition of scattering). The infrared plateau also contains potentially diagnostic features that may be related to both cellular arrangement within the leaf and the hydration state (Gates, 1970; Gausman et al., 1977; 1978). The cellular arrangement within the leaf is genetically controlled, resulting in differences in amounts of scattering for different vegetation types. This can therefore, potentially, be used for separating or identifying vegetation types or communities of vegetation (Esau, 1977). At longer wavelengths, the reflectance spectrum of healthy vegetation is mainly influenced by water content of the plant cells. Dry vegetation presents absorption features mostly attributable to lignin and cellulose that were masked by water in healthy green vegetation.

Figure 7: Examples of spectra of various vegetation types. From Clark (1999).

2.2.4 Scattering Scattering is the process that makes reflectance spectroscopy possible. Photons enter a surface, are scattered one or more times and, while some are absorbed, others are scattered from the surface so they may be seen and detected. As mentioned above, the scattering of photons by leaf cells can give information about vegetation type. Scattering is a non-linear process so it makes the recovery of quantitative information more difficult. The reflectance of a particulate surface, such as soils, or of vegetation (see above), however, is much more complex and the optical path of photons is a random walk. At each grain (or leaf cell) that the photons encounter, a certain percentage is absorbed. If the grain is bright, like a quartz grain at visible wavelengths, most photons are scattered and the random walk process can go on for hundreds of encounters. If the grains are dark, like magnetite, the majority of photons will be absorbed in only a few encounters. The random walk process of photons scattering in a particulate surface also enhances weak features not normally seen in transmittance. Scattering is also influenced by the size of the grains or objects (Figure 8).

7

favored favored

Figure 8: Effect of grain size on absorption and scattering. Modified from Combe (2005).

Finally, the intensity of spectral features in reflectance is a function of the intrinsic absorption strength, the scattering properties and the abundance of an absorbing material (Hapke, 1981, and references therein). In nature and at the size of a pixel, most natural surfaces are mixtures of different constituents. There are two main kinds of mixtures (Figure 9): surficial or areal mixtures where the resulting spectrum is a linear combination of spectra of each elementary constituent. This corresponds to the case where photons only interact with one surface type (Figure 9a). Often, this type of mixing is dominant when materials are distributed as discrete patches within a pixel – so called checkerboard mixing (Figure 9a’). This is because light can only interact with more than one cover type at the boundaries of the patches. Generally, however, the observed spectrum corresponds to a non-linear, intimate mixture of different elements (Figure 9b). In this case, photons interact with more than one medium before reaching the instrument. Non-linear mixing becomes dominant when materials that are intimately distributed within a pixel – so-called ‘salt and pepper’ mixtures (Figure 9b’). In reality, however, both linear and non-linear mixing occur within a single pixel, but the dominant type of mixing is determined by the patchiness of materials and how they are distributed within a pixel. For a number of reasons, the absorptions of one material may dominate an entire wavelength region in a mixture (Clark, 1983, 1999). For example, the spectrum of an intimate mixture of equal amounts of alunite and jarosite is dominated in the vibrational region by alunite, with its intense 2.17 µm absorption, compared 3+ to jarosite with its weaker absorption at 2.27 µm (Figure 10). Jarosite, with a 0.95 µm Fe absorption, dominates the electronic region because alunite lacks any significant absorptions at these wavelengths. This distinction, identified while testing single components, is applicable to spectra of intimate (non-linear) and areal (linear) mixtures that are spectrally dominated by one component in a given wavelength region, a situation that occurs in many surface environments (Clark et al., 1991, 1992; Swayze et al., 2000).

8

(a’)

(b’)

Figure 9: Definition of (a) areal (linear) mixture, so called checkerboard mixing (a’) versus (b) intimate (nonlinear) mixture so called mixing ‘salt and pepper’ (b’). Modified from Combe et al. (2005). IO: incident radiation; IRO: incident light IO partially reflected at the top of the biofilm; ITRT: measured emergent radiance; IT: incident light partially transmitted through the total thickness of the biofilm; ITR: second transmission through the biofilm after reflectance by the substrate.

Figure 10: Spectra of a linear (areal) and an intimate mixture of alunite and jarosite. From Swayze et al. (2003).

2.3 Some terminology This section defines some terminology frequently encountered in imaging spectrometry. Some terms describe the characteristics of the instrument, which will influence the quality of the data and the possibility of extracting relevant information. Others define radiation quantities. 2.3.1 Instrument-related terms There are four general parameters that describe the capability of a spectrometer: (1) spectral range, (2) 9

spectral bandwidth, (3) spectral sampling, and (4) signal-to-noise ratio. (1) Spectral range is the range of wavelengths measured by a spectrometer. It is important that this range encompasses enough diagnostic spectral absorptions to solve the desired problem. There are general spectral ranges that are in common use, each controlled to first order by the detector technology: ultraviolet (UV): 0.001 to 0.4 µm, visible: 0.4 to 0.7 µm, near-infrared (NIR): 0.7 to 3.0 µm, mid-infrared (MIR): 3.0 to 30 µm, and far-infrared (FIR): 30 µm to 1 mm. The approximate wavelength range 0.4 to 1.0 µm is sometimes referred to, in the remote sensing literature, as the VNIR (for visible-near-infrared) and the range 1.0 to 2.5 µm is sometimes referred to as the SWIR (short wave infrared). The mid-infrared covers thermally emitted energy, which for the Earth starts at about 2.5 to 3 µm, peaks near 10 µm then decreases beyond the peak, with a shape controlled by gray body emission. (2) Spectral bandwidth is the width of an individual spectral channel in the spectrometer. The narrower the spectral bandwidth, the narrower the absorption feature the spectrometer will measure accurately, if enough adjacent spectral samples are obtained. As mentioned above, some systems have a few broad channels, not contiguously spaced, and thus are not considered spectrometers (Figure 2). Others have many narrow bandwidths, contiguously spaced. Figure 2 shows spectra for the mineral alunite that could be obtained by some broadband and spectrometer systems. Note the loss of subtle spectral detail in the lower-resolution systems compared to the laboratory spectrum. Bandwidths and sampling greater than 25 nm rapidly lose the ability to resolve significant mineral absorption features. The shape of the bandpass profile is also important. The most common bandpass in spectrometers is described by a Gaussian profile, where the middle of the bandpass is more sensitive than its extremities. The width of the bandpass is usually defined as the width in wavelength at the 50% response level of the function, as shown in Figure 11, called the full width at half maximum (FWHM).

Figure 11: Gaussian profile with a full width at half maximum (FWHM) of 10 nm. This profile is typical of spectrometers such as AVIRIS, which has 224 such profiles spaced at about 10 nm. From Clark (1999). (3) Spectral sampling is the distance in wavelength between the spectral bandpass profiles for each channel in the spectrometer as a function of wavelength. Spectral sampling is often confused with bandpass, with the two lumped together and called resolution. The Nyquist theorem states that the maximum amount of information is obtained by sampling at one-half the FWHM. Spectrometer design, however, sometimes dictates a different sampling, and many modern spectrometers currently in use sample at half Nyquist, a sampling interval approximately equal to the FWHM. (4) Finally, a spectrometer must measure the spectrum with enough precision to record details in the spectrum. The signal-to-noise ratio (SNR) is dependent on the detector sensitivity, spectral bandwidth (the narrower the band the less light there is to detect), and intensity of the light reflected or emitted from the surface being measured. A few spectral features are quite strong, and an SNR value of only about 10 will be adequate to identify them, whereas others are weak, and an SNR value of several hundred (and higher) is often needed. 2.3.2 Physical terms Radiance Radiance is the most important quantity measured in spectrometry because much of remote sensing is concerned with radiant flux leaving extended areas in a direction toward the sensor. Consequently, the concept of radiance is very frequently used. Spectral radiance is defined as the radiant flux in a beam per unit wavelength, per unit area and solid angle of that beam i.e. it is the amount of light reaching the surface. -2 -1 -1 It is usually expressed in the SI units [W.m .sr .nm ]. Radiance includes both atmospheric and surface effects. So, for example, radiance can change significantly as a result of cloud cover or atmospheric aerosols. An example of an AVIRIS radiance spectrum is shown in Figure 12a.

10

Reflectance, reflectance factors The amount of light reaching the earth’s surface can change as a function of wavelength. For example, absorption and scattering of light by atmosphere can dominate the shape of the spectral curve. The sun also emits light in different amounts at different wavelengths (the so-called solar irradiance curve). It is therefore often desirable in remote sensing to standardize the amount of reflected light to the amount of light reaching the surface. This process removes absorption features due to atmospheric absorption, leaving only those features which are associated with the target under consideration. Dividing the radiance leaving the surface by the incident radiation onto the surface gives the so-called reflectance. Following the concept of energy conservation, the values of reflectance lie between 0 and 1 (or 0 and 100%) inclusive. The reflectance factor is the ratio of the radiant flux reflected by the surface to that reflected into the same reflected-beam geometry by an ideal (lossless) and diffuse (Lambertian) standard surface, irradiated under the same conditions. For measurement purposes, a Spectralon® panel commonly approximates the ideal diffuse standard surface. Most natural surfaces are not Lambertian so reflectance factors, adapted to the remote sensing problem and respecting particular directional issues, can generally be defined as follows, using the notations in Table 1: F(θi,φi,ωi;θr,φr,ωr;λ) where the direction and solid angle of the circular cone of the incoming and reflected radiance are indicated. An extensive description and definition of directional reflectance factors can be found in Schaepman-Strub et al. (2006). Table 1: Notations used for the definition of at-surface reflectance quantities. Symbol ρ R θ φ

Explanation reflectance (dimensionless) reflectance factor (dimensionless) zenith angle, in a spherical coordinate system (rad) azimuth angle, in a spherical coordinate system (rad)

ω

solid angle =

λ f()

wavelength of the radiation (nm) function

∫ dω = ∫∫ sin θdθdφ

(sr)

sub- and superscripts i r

incident reflected

Converting data from radiance to reflectance removes the effects of the atmosphere. The surface reflectance spectrum corresponding to the radiance spectrum is shown in Figure 12b. The next section briefly describes the impact of the atmosphere on imaging spectrometry data.

11

(a)

(b)

Figure 12: (a) Simulated AVIRIS radiance spectrum of kaolinite; the simulation was performed using the MODTRAN radiative transfer code (Berk et al., 1989) with a “standard” atmosphere; (b) reflectance spectrum of kaolinite from USGS reference spectral data base. 2.4 Atmospheric effects Remote sensing measurements of the Earth’s surface are strongly influenced by the atmosphere (Goetz et al., 1985). Both scattering and absorption by gases and particulates affect the amount and wavelengths of light reaching the sensors. Absorption by atmospheric gases is dominated by water vapor with smaller contributions from carbon dioxide, ozone, and other gases (Gao and Goetz, 1990). Strong atmospheric water absorption bands make the atmosphere opaque in many regions (for example the 1.4 and 1.9 µm regions) and only small atmospheric windows are available for terrestrial remote sensing (Figure 13). The drop toward the ultraviolet is due to scattering and strong ozone absorption at wavelengths shorter than 0.35 µm. Ozone also displays an absorption at 9.6 µm. Oxygen absorbs at 0.76 µm in a narrow feature. CO2 absorbs at 2.01 and 2.06, with a weak doublet near 1.6 µm. Water causes most of the rest of the absorption throughout the spectrum and hides additional absorptions from other gases. Finally, more haze translates to more diffuse radiation than direct beam solar radiation reaching the surface.

Figure 13: MODTRAN (Berk et al., 1989) modeled atmospheric transmittance, visible to near-infrared. Most of the absorptions are due to water. Oxygen occurs at 0.76 µm, carbon dioxide at 2.0 and 2.06 µm. From Clark (1999). Consequently, the most critical step in most imaging spectrometry data analysis strategies is to convert the 12

data from radiance to reflectance so that individual spectra can be compared directly with laboratory or field data for identification. This can be accomplished by using either radiative transfer based algorithms, such as ATCOR4 (Richter and Schläpfer, 2002) or FLAASH (Adler-Golden et al., 1999; Matthew et al., 2000), which are commercially available, or empirical approaches, such as the “empirical line” method (Roberts et al. 1985; Moran et al., 2001). Empirical approaches require field spectra from reference targets within the field of view of the image. More details about atmospheric corrections and existing algorithms can be found, for example, in Staenz et al. (2002), Goetz et al. (2003), and Ben-Dor et al. (2004).

3 ASSESSING MICROPHYTOBENTHOS SPECTRORADIOMETRY

BY

LABORATORY

AND

FIELD

3.1 Parameters of interest and how they transcribe into spectral signatures 3.1.1 Spectroradiometry vs. pigments Organisms belonging to microphytobenthos (MPB) assemblages are characterized by the presence of photosynthetic pigments: chlorophylls, carotenoids and phycobilins. These pigments, associated to pigmentprotein complexes, constitute light-harvesting complexes and reaction centers within thylakoid membranes, which absorb light energy to transfer it into chemical energy (Porra et al., 1997). This absorption of light energy involves wavelengths ranging from 300 to 800 nm. Downwelling light at these wavelengths is referred to as photosynthetically active radiation (PAR). While the absorbed wavelengths are converted into chemical energy and reducing power forms carbohydrate from CO2, the non-absorbed wavelengths are transmitted and reflected, giving pigments their characteristic colors (Porra et al., 1997): green for chlorophylls, orange for carotenoids, and blue or red for phycobilins (Figure 14). Thus, all microphytobenthic photosynthetic eukaryotes and cyanobacteria, containing chlorophyll a, the only pigment found in all plants, algae and cyanobacteria (Table 2), show an absorption peak around 440 nm, in the blue region of electromagnetic radiation, and another one around 675 nm in the red region (Figure 14). The addition of accessory pigments, such as chlorophyll b or c, carotenoids and other phycobilins, widens the spectral range for light absorption, either because the absorption peaks are slightly different from chlorophyll a (the case of other chlorophylls), or because they are located in a spectral region which does not overlap with chlorophyll absorption (the case of carotenoids and phycobilins) (Figure 14). Algae employ pigments other than chlorophyll a to capture additional light for photosynthesis, for example, algae that live underwater can employ pigments that harvest those wavelengths of light that can easily pass thorough the water-column. The pigment composition of these pigment-protein complexes varies widely throughout vegetal phyla, and can thus be used as a taxonomic fingerprint (Gieskes, 1991). Indeed, MPB biofilm stains the intertidal sediment surface with colors depending on its taxonomic diversity (Table 2) and the relative abundance of its constituents. Hence, intertidal surfaces are detectable by spectroradiometry using their spectral characteristics (Méléder et al., 2003a; Combe et al., 2005). In fact, whereas the incident light reaching biofilm is partially absorbed for photosynthesis, the remaining energy can be transmitted and/or reflected. The fraction of absorbed light, and thus of reflected light, depends on the pigment composition and the relative and quantitative abundance (biomass) of microphytobenthic biofilm.

13

Absorbance

1

6

2 5

3

400

4

450

500

550

600

650

700

750

800

Wavelength (nm)

Figure 14: Absorption spectra of pigments from varying algae sources measured by spectrophotometry: 1 chlorophyll a, 2 chlorophyll c, 3 chlorophyll b, 4 diadinoxanthin, 5 fucoxanthin, 6 R-phycoerythrin (a phycobilin). 1 2 4 5 were isolated from a diatom species (Entomoneis paludosa) (Méléder et al., 2003a), 3 from a Chlorophyceae species (Dunaliella sp.) (Méléder, unpublished data) and 6 from a macroalgae Rhodophyceae (Grateloupia turuturu) (Denis et al., 2009). The ordinate represents a relative scale of absorbance.

Chlorophyceae

Cyanobacteria

Diatoms

Dinophyta

Euglenophyceae

Table 2: Distribution of characteristic pigments within algae groups belonging to MPB, used as a taxonomic fingerprint. ++: major pigment (> 10%) ; +: minor pigment (1 – 10%) ; -: trace (< 1%). From Jeffrey and Vesk (1997). Encircled numbers refer to Figure 14.

Chlorophyll a 1

++

++

++

++

++

Chlorophyll b 3

++

++

Chlorophyll c 2

β, ε carotene

-

β, β carotene

+

Antheraxanthin

-

+

++

++

-

-

-

Diadinoxanthin 4 Diatoxanthin

++

++

++

-

-

-

Dinoxanthin

+

Fucoxanthin 5

++

Lutein

++

Neoxanthin

++

-

Peridinin

++

Pyrrhoxanthin Zeaxanthin Phycobilins 6

+

+

++ ++

14

To assess the contribution of photosynthetic and accessory pigments to the reflectance spectrum of organisms constituting MPB in order to use hyperspectral imaging to map MPB biomass on the cohesive sediment surface of tidal areas, analysis of MPB reflectance spectra is firstly carried out in the laboratory under controlled conditions of incident light, water content and biomass. Biofilm complexity is reduced as much as possible, using only one characteristic species of an MPB taxonomic group, the diatoms (Table 2). Secondly, laboratory results are used to validate field measurements with the aim of calibrating hyperspectral images (see § 3.1.3 and 4). Species are cultivated in the laboratory and monospecific cultures are diluted at varying concentrations. Suspended cells from each daughter culture are deposited on filters (glass fiber or polycarbonate), producing a uniform distribution of cells at the filter surface, simulating a biofilm (Figure 15). Spectral measurements are performed using a spectroradiometer to determine radiance -2 -1 -1 (mW.cm .sr .nm ) for a wavelength range covering at least the PAR. Radiance spectra for the ambient incident light are recorded by the radiance of a near-perfect diffuser (a piece of white Spectralon®). The ratio of the biofilm and incident spectra radiances gives the reflectance (without units). After measuring reflectance, the pigment content of each biofilm on the filter is analyzed using High Performance Liquid Chromatography (HPLC). This technique, widely used in oceanography to assess phytoplankton, MPB, seaweed and seagrass composition (Jeffrey, 1997), enables the identification and quantification of more than 30 liposoluble pigments: chlorophylls, carotenoids and their breakdown products. These pigments are extracted into solvent and eluted following varying solvent mixtures during gradient and/or isocratic phases (Jeffrey, 1997). Then, each pigment is detected and characterized by diode array measurements of their absorption spectra expressed between 400 and 800 nm (Figure 14).

Monospecific culture Daughter cultures (varying concentrations)

Reconstituted bioflm

Figure 15: Biofilm reconstitution using a monospecific culture, diluted at varying concentrations. Spectrum reflectance is measured for each biofilm, before pigment analyses by HPLC. Color intensity illustrates the biomass variation. It is then possible to link absorption bands from reflectance spectra, obtained by spectroradiometry, and absorption spectra of major pigments, obtained by HPLC, to interpret the spectral characteristics of MPB. For example, a reconstituted biofilm using a diatom species (Entomoneis paludosa) absorbs light energy at 440 nm due to all pigments, and shows particular shoulders around 500 and 550 nm due to diadinoxanthin and fucoxanthin respectively (Figures 14 and 16). This biofilm also absorbs around 630 nm, due to both chlorophyll c and a, and around 670 nm, due to chlorophyll a. For comparison, a reconstituted biofilm using a Chlorophyceae (Dunaliella sp.) does not show a shoulder around 550 nm because fucoxanthin is not a constituent of its pigment-protein complexes, whereas the absorption around 500 nm is due to lutein (Figure 16). Chlorophyll b is responsible for the shoulder around 650 nm in the reflectance spectrum of the Chlorophyceae, before the absorption band of chlorophyll a. This shoulder is absent from the reflectance spectrum of the diatom species, which does not have chlorophyll b within its pigment-protein complexes (Figure 16). This interpretation of spectral absorption bands by pigment maxima absorbance could be easier nd using a 2 derivative process on reflectance, showing peaks for each absorption band or shoulder (see § 3.2.2).

15

Reflectance

3

1 Lutein

All pigments 400

1

and

5

2

4 450

1

500

550

600

650

700

750

800

Wavelength (nm)

Figure 16: Laboratory reflectance spectrum measured by spectroradiometry on reconstituted biofilm from two MPB species: a diatom (Entomoneis paludosa, fine black line), a Chlorophyceae (Dunaliella sp., gray line). Each absorption band and shoulder is linked to the absorbance spectrum of major pigments measured by HPLC. Encircled numbers refer to Figure 14 and Table 2. The ordinate represents a relative scale of reflectance.

3.1.2 Spectroradiometry vs. biomass Pigment studies provide useful information about oceanic plant communities. Whereas carotenoid pigments are commonly used as taxonomic biomarkers to characterize planktonic and benthic communities (Table 2, Gieskes, 1991; Jeffrey, 1997), measurements of chlorophyll a in superficial sediments provide estimates of plant biomass and production. Chlorophyll a is usually used as a biomass estimator in phytoplanktonic and microphytobenthic studies (e.g. De Jonge and Colijn, 1994; MacIntyre et al., 1996; Brotas and Plante-Cuny, 1998; Guarini et al., 1998; Kelly et al., 2001; Perkins et al., 2003; Méléder et al., 2005). MPB biomass can be -2 expressed in concentration (mass vs. unit area: mg Chl a.m ) or content (mass vs. sediment mass: mg Chl -1 a.g dry weight sediment) (Perkins et al., 2003). However, Murphy et al. (2005a) have shown that chlorophyll measured as mass per unit mass of sediment is inappropriate for ground-truthing remotely sensed observations of chlorophyll in intertidal benthic sediments. The amplitude of the absorption bands of reflectance spectra measured on reconstituted biofilm on filters increases with increasing biomass (Figure 17). The signal and its absorption bands, however, tend to flatten towards the strongest biomass. A -2 reflectance saturation phenomenon appears beyond the threshold biomass at around 100 mg Chl a.m (Figure 17, gray ellipses). The fucoxanthin- and diadinoxanthin-related shoulders between 500 and 550 nm disappear while reflectance values reach a minimum measured between 400 and 550 and around 670 nm. This saturation phenomenon does not concern the absorption band around 630 nm due to both chlorophyll c -2 and a because absorbance in this wavelength keeps increasing above 100 mg Chl a.m (Figure 17).

1.2

1.0

Reflectance

0.8

0.6

mg Chl a.m-2

0.4

2.9 13.5 34.7 68.3 105.5 146.1 182.3

0.2

0.0 400

450

500

550

600

650

700

750

800

Wavelength (nm)

Figure 17: Reflectance spectra measured in the laboratory on reconstituted biofilm on filters for varying biomasses from Entomoneis paludosa (diatom) cultures. Gray ellipses indicate the saturation phenomenon -2 for the threshold biomass (> 100 mg Chl a.m ). The gray arrow indicates the increase in the chlorophyll c -2 absorption band beyond 100 mg Chl a.m .

16

3.1.3 Field spectroradiometry and sampling depth Whereas the use of field spectrometry as a stand-alone, low-cost method for rapidly mapping the distribution of major habitat types of an intertidal system has been demonstrated by Foster and Jesus (2006), its use to estimate field biomass still has a limitation: the sampling depth (Barillé et al., 2007). Indeed, the detected energy in spectroradiometry is a proportion of the energy returned toward the instrument, after a variable path through the sediment. The length of this path depends not only on the MPB biomass but also on the sediment texture, varying from mud to muddy-sand on intertidal mudflats, organic and water contents/concentrations, and the wavelengths of incident light. Thus, after reflectance measurements on field biofilm, it is essential to sample this biofilm at a sediment depth in accordance with the length of the path of the reflected light. This depth could be linked to the photic zone, corresponding to the depth receiving more than 1% of the incident light. Throughout this zone, biomass is photosynthetically active whereas below this 1% threshold light energy is not available for photosynthesis (Kelly et al., 2001). On intertidal mudflats, the photic zone does not exceed 2 mm (Jorgensen and Des Marais, 1986; Kühl et al., 1994; MacIntyre et al., 1996; Paterson et al., 1998). Comparing reflectance data using the NDVI (see § 3.2.1 for details) for varying biomass measured either on reconstituted biofilm on filters in the laboratory, or in the first 2 mm of sediment from the field, a shift is -2 observed (Figure 18). For example, an NDVI value around 0.2 corresponds to a biomass of 2.9 mg Chl a.m -2 on a filter or 45 mg Chl a.m in the first 2 mm of sediment. Sampling biomass in the first 2 mm of sediment is not suitable for reflectance measurement by spectroradiometry (Barillé et al., 2007, Jesus et al., 2006). This depth is too great compared to the length of the path of the reflected light in sediment. Biomass sampled using a technique based on the phototactism capability of MPB organisms, particularly diatoms (Eaton and Moss 1966), shows results more similar to those obtained on filters (Figure 18). Nevertheless, trapping these cells in lens-cleaning tissue using the natural migration of organisms has limitations: a possible saturation of the tissue by a maximum number of cells, which prevents all the biomass from being sampled as some cells remain below the tissue. This saturation could explain why low biomass values are obtained. A second limitation is the potential selectivity: only cells with a migration ability could be trapped but MPB is not solely constituted by moving cells (Méléder et al., 2007; Jesus et al., 2006, 2009). To overcome these limitations, sampling biomass in sediment remains the best technique, but at a depth less than 2 mm. A depth of 500 µm seems to be a good compromise (Jesus et al., 2006; Méléder, unpublished data). 0.8

0.7 0.6

NDVI

0.5 0.4

0.3 0.2

0.1 0 0

20

40

60

80

100

120

140

160

180

200

Biomass (mg Chl a.m-2)

Figure 18: Reflectance data vs. biomass from the field and the laboratory.  : biomass sampled in the first 2 mm of sediment;  : biomass sampled by trapping cells in lens tissue; : biomass measured on reconstituted biofilm on filters in the laboratory. Reflectance are expressed using the NDVI (for details, see § 3.2.1). 3.2 Processing reflectance spectra Compared to terrestrial applications, field and laboratory spectroscopy has been applied only recently to aquatic botany. MPB spectral reflectance analysis is thus partly inspired not only by studies performed on higher plants, but also by apparently unconnected research fields such as planetology, which has developed many hyperspectral remote sensing methods, transferable to MPB studies (Combe, 2005; Combe et al., 2005). The information about MPB that can be extracted from reflectance data depends firstly on the spectral resolution of the sensor (see § 2.3.1). The spectral resolution represents the number of spectral bands (spectral sampling or sampling interval) and the wavelength range of each band, estimated by the full width half maximum (FWHM, Figure 11). Spectra acquired with a spectroradiometer at a high spectral resolution are sometimes degraded after spectral resampling, when spectroscopy is used to calibrate images obtained at a lower spectral resolution, typically broadband satellite multispectral images (e.g. LANDSAT and SPOT). 17

Multispectral sensors are distinguished from hyperspectral sensors when the number of bands is reduced (arbitrarily, below 10) and the width of each band is broad (ca. 100 nm). A major distinction can therefore be proposed between multi- and hyperspectral processing of reflectance spectra. In the first case, indices are calculated as a combination of a few broad spectral bands, while in the second case the full reflectance spectrum is exploited to retrieve relevant information, and subtle absorption features are investigated. However, high resolution reflectance spectra obtained with spectroradiometers are also used to compute vegetation indices based on a few spectral bands.

Figure 19: Comparison of different types of spectral processing applied to a microphytobenthic biofilm to estimate the chlorophyll a absorption at 673 nm. The chlorophyll c absorption band, characteristic of a microalgal community dominated by diatoms, is clearly visible around 630 nm. (a) Single-band ratio Rb/Rc, based on visible-near-infrared bands. (b,c,d) Indices based on continuum removal. (b) Continuum is fitted between selected end-points C1 and C2; band depth (Db) computed using reflectance (Rb) and corresponding continuum reflectance (Rc) at the band center. Dashed lines represent mean surface reflectance (10 measurements) plus or minus standard deviation. (c) Normalized ratio or scale-band depth after continuum removal. (d) Scaled-band area (in gray) after continuum removal. From Carrère et al. (2004).

3.2.1 Single-band vegetation index (VI) VIs are spectral transformations of two or more bands designed to enhance the contribution of vegetation properties (Huete, 1988). These spectral transformations are then used to establish relationships between bioptical and biological quantities (e.g. biomass, pigment composition). These indices are calculated as single-band ratios or normalized differences from Near-Infrared (NIR) and visible bands. The commonly-used VI index is the Normalized Difference Vegetation Index (NDVI) involving red and NIR wavelengths: NDVI = (NIR-red)/(NIR+red) (Rouse et al., 1973). The red band, selected around 675 nm, corresponds to chlorophyll a absorption, a salient spectral feature associated to this ubiquitous pigment (Figure 19). A wide array of VIs has been tested by Murphy et al. (2005b) to quantify benthic chlorophyll on an emersed mudflat in Australia. Many of these indices have been developed to study vegetated canopies (Huete, 1988), with a physical structure very different from that of an intertidal biofilm of microalgae, which may explain the variable 18

success of VIs designed for terrestrial vegetation (Table 3). Murphy et al. (2005b) selected a visible-band ratio, R562/R647, to estimate the chlorophyll a concentration used as a proxy for biomass estimation. However, these authors studied intertidal sediments near Sydney colonized by filamentous macroalgae distributed at the surface and within the sediment. European mudflats are dominated by benthic diatoms (Underwood & Kronkamp, 1999) for which the NDVI appears as a robust index to quantify MPB biomass in muddy sediments (Kromkamp et al., 2006). A specific adaptation of the NDVI, using the in vivo absorption band of chlorophyll c around 632 nm, a diagnostic band for the class of diatoms (Méléder et al., 2003a), has been successfully used under the name of Phytobenthos Index (PI, Foster & Jesus, 2006; Jesus et al., 2006; Serôdio et al., 2009) where PI = R750-R635/ R750-R635. This index is less sensitive to the biomass saturation phenomenon (Méléder et al., 2003a) and to chlorophyll a breakdown products, whose overlapping absorption with the chlorophyll a absorption band at 675 nm may result in equivocal results (Barillé et al., 2007). In spite of its robustness due to the ratioing concept which reduces many forms of multiplicative noise (Huete et al., 2002), the NDVI has some disadvantages. It tends to saturate at high biomass, exhibiting a typical -2 asymptotic relationship beyond a concentration of ca. 100 mg Chl a.m (Figure 20, Méléder et al., 2003a). It can also be influenced by chlorophyll fluorescence in the 670-700 nm wavelength range (Figure 21d) with an underestimation of microalgal biomass to reach over 25% (see Fig. 3d, Serôdio et al., 2009). It is also supposed to be influenced by sediment brightness variations, by analogy with the effect of soil background brightness on canopy spectral responses (Huete, 1988). In intertidal environments, such albedo variations can be related to sediment water content, grain size or the presence of a mucilaginous matrix secreted by microalgal communities (Decho et al., 2003; Carrère et al., 2005; Verpoorter et al., 2009). Residual water forming a layer above MPB biofilms is frequently found in mudflats at low tide. Since water is more absorptive at NIR compared to visible wavelengths (Kirk, 1996), this may affect the band ratios constructed at these wavelengths (Murphy et al., 2008). The spectral response of the substrata may also influence band ratio VIs, with sandy sediment characterized by a steeper visible-NIR slope than muddy sediments (Murphy et al., 2005b). In terrestrial science applications, a family of VI was designed to remove the influence of soil background brightness for a better estimation of canopy or grassland biophysical parameters (Vescovo and Gianelle, 2008). The soil-adjusted vegetation index (SAVI) proposed by Huete (1988), was the first of these VI integrating a soil correction. It was tested by Murphy et al. (2005b) to predict chlorophyll a on emersed intertidal sediments, but provided results very similar to those obtained with the NDVI.

1 0.9 0.8

NDVI

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0

50

100

150

-2

200

250

Biomass (mg chl a.m )

Figure 20: Relationship between NDVI (Normalized Difference Vegetation Index) and increasing biomass (estimated by the concentration of chlorophyll a, mg Chl a. m-2) of benthic diatoms. NDVI is calculated with the chlorophyll a absorption band around 675 nm. The non-linear and asymptotic relationship illustrates the saturation of the chlorophyll a absorption band at high biomass. Data from experimental biofilms obtained with monospecific cultures of Navicula ramosissima and Entomoneis paludosa. Modified from Méléder et al. (2003a). 3.2.2 Derivative analysis The derivation of reflectance spectra is a technique recently applied to reflectance spectroscopy to remove background signals caused, for example, by differences in sediment grain-size and to resolve overlapping or weak spectral features (Figure 21, Demetriades-Shah et al., 1990). Derivation also places all absorption features on the same reference plan (i.e. centred on zero), thus removing, at least to a large extent, 19

brightness variations of the substratum. Derivatives can be calculated using simple geometric methods that quantify the band-to-band change in slope or by differentiation. This hyperspectral processing uses spectrally continuous data, and provides different types of information according to the level of derivation (Tsai & Philpot, 1998). This technique has been applied to MPB up to the fourth derivative order (Murphy et al., 2008). However, differentiation generates high frequency noise, and the signal-to-noise ratio tends to increase with increasing derivative order (Tsalky, 1994). Smoothing algorithms must therefore be applied (Tsai & Philpot, 1998), but this adds variability to derivative indices, linked to the smoothing intensity. A common method used to generate derivatives is the combined differentiation and smoothing method of Savitsky and Golay (1964) but using the corrected coefficients presented in Steiner et al. (1972). Demetriades-Shah et al. (1990) mentioned that derivative indices should be less sensitive than classical VI to soil background variations. This was confirmed for the intertidal environment by Murphy et al. (2005b), who obtained the strongest statistical relations between second order derivative measurements and chlorophyll a concentrations. The first order derivative, showing peaks above the zero baseline at the inflection points of absorption features, is mainly used to identify the Red Edge Inflection Point (REIP). In a zero-order spectrum, the REIP is the wavelength of maximal slope between red and NIR wavelengths, while in the first order it corresponds to the peak between 670 and 750 nm (Horler et al., 1983) (Figure 21). This index, which tends to shift to longer wavelengths for increasing biomass, has been used to retrieve information about higher plant physiology (Filella & Peñuelas, 1994; Jago et al., 1999). Applied to intertidal mudflats, it was however a poor estimator of benthic chlorophyll, compared to derivative indices at other wavelengths (Murphy et al., 2005b). The second order derivative, showing peaks at the band-centers, has been applied to MPB reflectance spectra by several authors (Murphy et al., 2005b, 2008; Jesus et al., 2006; Serôdio et al., 2009). It has more potential than the previous order, because wavelength positions of positive peaks in the second order derivative could be associated to the absorption of the main pigments (Jesus et al., 2006, Figure 21e), chlorophyll fluorescence (Serôdio et al., 2009, Figure 21d), sediment-stability variables (Murphy et al., 2008). It is also less sensitive to noise than the fourth order. With monospecific cultures of benthic diatoms, the peak positions are consistent with the absorption bands of the main pigments: chlorophyll a, chlorophyll c, fucoxanthin, diadinoxanthin (Figure 21e, Jesus & Barillé, original data). With natural communities, the value of second derivative indices is less convincing (Murphy et al., 2005b; Jesus et al., 2006). A coupling between derivative analysis and the Modified Gaussian Method (MGM), presented in the next paragraph, has been proposed by Verpoorter (2009) as an Automated MGM approach (AMGM) for hyperspectral mapping of sediment pigment, water content, and grain size.

20

Table 3. Multispectral (Single-band Vegetation Index) and hyperspectral (Derivative analysis and continuum removal) indices used to estimate MPB biomass and pigment composition. Some indices were tested to study filamentous macroalgae intimately mixed with the sediment (Murphy et al., 2005b). For the SAVI, L represents a soil adjustment factor ranging from 0 to 1 (Huete, 1988). Single-band Index R700 / R675 R673 / R720 R562 / R647 R400 / R500

Vegetation

Abbreviation

Calculation

Reference

– – – –

R700 / R675 R673 / R720 R562 / R647 R400 / R500

Near-Infrared-Blue Index

IR-B

(R750 – R435) / (R750 + R435)

Blue-Green Index

BG

(R590 – R435) / (R590 + R435)

Green-Red Index

GR

(R590 – R675) / (R590 + R675)

Log Red-Infrared index

R-IR

log(R750) / log(R673)

Ratio Vegetation Index Normalized Difference Vegetation Index Green NDVI Normalized Pigment Chlorophyll a Index

RVI

R750 / R672

Gitelson et al., 1993 Carrère et al., 2004 Murphy et al., 2005b Murphy et al., 2005b Kromkamp et al.,2006 Kromkamp et al., 2006 Kromkamp et al., 2006 Hakvoort et al., 1998 Jordan, 1969

NDVI

(R750 – R672) / (R750 + R672)

Rouse et al.,1973



(R750 – R550 ) / (R750 + R550)

NPCI

(R680 – R430) / (R680 + R430)

Phytobenthos Index

PI

(R750 – R635) / (R750 + R635)

Gitelson et al., 1996 Peñuelas et al., 1993 Forster and Jesus, 2006

With soil correction Soil Adjusted Vegetation Index Derivative analysis

SAVI

[(R750 – R672) / (R750 + R672+L)](1+L)

Huete, 1988

First derivative measure (δ) λREIP Wavelength of the REIP Second derivative measure (δδ) δδR661, Derivative reflectance δδR693, δδR721 (661, 693, 721 nm) δδR422, δδR444, Derivative reflectance δδR468, (422, 444, 468, 492, 538, δδR492, 643, 678 nm) δδR538, δδR643, δδR678 Continuum removal Normalized ratio (or scaledDb band depth)

(Rc– Rb) / Rc

Carrère et al., 2004

Scaled-band area

Ab

∫ (Rc- Rb) / Rc

Carrère et al., 2004

MGM

λ500, λ550, λ623, λ675

AMGM



Modified Gaussian Model (band depth or strength) Automated Modified Gaussian Model



Horler et al., 1983



Murphy et al., 2005b



Jesus et al., 2006

λ

Combe et al., 2005 Verpoorter 2009

et

al.,

21

(d)

(e)

Figure 21: Derivative analysis of a sugar beet canopy (a-c, from Demetriades-Shah et al., 1990) and two benthic diatoms (d-e, Jesus & Barillé, original data). Spectral reflectance of a sugar beet canopy, a bare dry soil and a bare wet soil (a), and their first order (b) and second order (c) derivatives. The main peak in first order differentiation identifies the red edge inflection point around 720 nm. With increasing derivative order, the soil contribution is gradually eliminated (b-c). Reflectance spectrum and second derivative (red line) of Cylindrotheca sp. (d). The double peak at 668 and 686 nm is due to the interference of chlorophyll fluorescence (see Serôdio et al., 2009). Reflectance spectrum and second derivative (red line) of Entomoneis paludosa sp. (e). Wavelength positions of positive peaks in the second order derivative can be associated to the absorption of the main pigments: chlorophyll a (676 nm), chlorophyll c (630 nm), fucoxanthin (540 nm), diadinoxanthin (499 nm).

3.2.3 Continuum removal The continuum represents the overall shape of a reflectance spectrum (Figures 19b and 22). Removing the continuum enables the absorption features of interest (pigment absorption bands for vegetation studies, mineral absorption bands for geological applications) to be isolated from other contributions changing albedo and slopes due to other absorbing materials (Clarke & Roush, 1984). For intertidal sediments, the continuum is mainly influenced by water content and grain size (Carrère et al., 2004), but also by unknown processes that contribute to the overall shape of reflectance spectra. Continuum removal techniques are therefore important for MPB studies to retrieve true absorptions due to photosynthetic and accessory pigments in the visible wavelength range. Carrère et al. (2004) applied a fast and easy continuum removal technique (Figure 19b,c,d), for which the continuum is defined by a straight line within a wavelength range of the reflectance spectrum. To remove the continuum, the reflectance spectrum is then divided by the straight line and transformed into a scaled reflectance, from which scale-band depth or scale-band area can be estimated. A more elaborate continuum-removal technique, the Modified Gaussian Model (MGM), developed by Sunshine et al. (1990) for mineral analysis, was adapted by Combe (2005) to study MPB. The MGM provides a set of 22

Natural logarithm ic reflectance

mathematical functions that enable all the absorption bands and a continuum to be modeled simultaneously. Thus, absorption features that are specific to MPB pigments can be retrieved (Combe et al., 2005; Barillé et al., 2007). The continuum is modeled by a straight line in wave numbers (1/wavelength) and the absorption bands (band depth or strength, in logarithmic reflectance) are fitted with Gaussian curves (Figure 22). Hoepffner and Sathyendranath (1991) had already successfully applied Gaussian curve deconvolution to determine the contribution of individual pigments to the in vivo absorption spectrum of phytoplanktonic algal cultures. MGM was applied by Barillé et al. (2007) to compare reflectance spectra of field and laboratory benthic diatoms. In spite of very different substrates (sediment vs. fiberglass filters), the absorption features could be compared after continuum removal (Figure 23). Significant relationships were obtained between the specific band depth of carotenoid pigments and their concentrations estimated by high performance liquid chromatography. The MGM approach is therefore of double interest for MPB studies: 1) to remove the sediment background influence and provide spectral indices that should be better than single-band VI, 2) to retrieve specific pigment absorption features that only appear as subtle slope changes in reflectance spectra. The application of MGM to hyperspectral image processing is further developed in a later section.

Reflectance (%)

0.16 0.12 0.08 0.04 0 400

A

500

600

700

800

900

0 -0.2 -0.4 -0.6 -0.8 -1

B

-1.2 400

500

Wavelength (nm)

600

700

800

Wavelength (nm)

Natural logarithmic reflectance

Natural logarithmic reflectance

Figure 22: Example of a benthic diatom reflectance spectrum fitted by a Modified Gaussian Model (MGM) with several Gaussian curves and a continuum. (A) Reflectance spectrum and its continuum (dashed line) in the wavelength space. (B) Removed-continuum spectrum in the logarithmic reflectance vs. wavelength space and its corresponding Gaussian curves. The vertical dashed arrow indicates the band depth (or strength) at the 675 nm absorption band. The horizontal dashed arrow indicates the Full Width Half Maximum at the 632 nm absorption band. Three other Gaussian curves are represented: fucoxanthin (550 nm), diadinoxanthin (500 nm), and a broad band at 435 nm accounting for the absorption of many pigments. From Barillé et al. (2007).

0 A -0.5 -1 -1.5 -2 400

500

600

700

Wavelength (nm)

800

0 B -0.4 -0.8 -1.2 -1.6 400

500

600

700

800

Wavelength (nm)

Figure 23: Benthic diatom reflectance spectra for a range of biomass concentrations and two different types of substrate. Spectra are shown in the logarithmic reflectance vs. wavelength space after continuum removal by a Modified Gaussian Model. (A) Field spectra of natural assemblages from Bourgneuf Bay mudflat -2 (France), with a chlorophyll a biomass ranging from 7.5 to 141.8 mg m . The limits correspond to the lower and upper curves respectively. (B) Monospecific culture spectra, on fiberglass filters (Méléder et al., 2003a), -2 with a chlorophyll a biomass ranging from 3 to 238 mg m . The limits correspond, to the lower and upper curves respectively. From Barillé et al. (2007). 23

900

3.3 Two examples of data processing 3.3.1 Assessing MPB physiology by spectroradiometry Although the major use for spectral reflectance has been the estimation of microphytobenthic biomass, the non-destructive nature of this tool can be expanded to investigate further aspects of MPB taxonomy and ecophysiology. Estuarine intertidal flats can exhibit high primary productivity rates as a result of the extensive cover of MPB biofilms (Underwood and Kromkamp, 1999). Frequently, the presence of diatom-dominated biofilms is the origin of this high productivity. This microphytobenthos group is particularly important because it exhibits an exceptional capacity to cope with variations in exposure to ambient light. This feature allows it to maintain high productivity rates over a wide range of light levels and under rather unpredictable light environments (Lavaud et al., 2007). The success of this group in sustaining these high productivity rates is partially attributed to the efficiency of diatoms in the dissipation of excess energy via a thermal dissipation mechanism (non-photochemical quenching, i.e. NPQ). Photosynthetic electron transport leads to the buildup of a transthylakoid ∆pH which promotes the enzymatic epoxidation of diatoxanthin (hereafter called DT) into diadinoxanthin (hereafter called DD). Thus, when light increases, the conversion of DD into DT is promoted, and when light decreases, the conversion of DT into DD is facilitated (Figure 24). This process is known as the xanthophyll cycle and, in diatoms, it can account for up to 90% of energy dissipation (Lavaud et al., 2002). In terrestrial plants, a comparable mechanism exists involving different pigments but working in a very similar way, converting vioxanthin into zeaxanthin (Bilger and Björkman, 1990).

Diadinoxanthin (DD)

+ Light - Light

Diatoxanthin (DT)

Figure 24: Exposure to increasing light levels leads to the epoxidation of diadinoxanthin (DD) into diatoxanthin (DT). Conversely, a decrease in light levels leads to the reverse conversion. The importance of this cycle in the photoregulation of photosynthetic organisms has led to an obvious interest in developing remote sensing indices that could be used to follow the xanthophyll cycle. The Photochemical Reflectance Index (PRI, Gamon et al., 1992) is the most widely used to follow zeaxanthin production in terrestrial vegetation. Recently, Jesus et al. (2008) investigated the hypothesis that a similar index could be developed to follow the diatom xanthophyll cycle in microphytobenthic biofilms. They used a combination of PAM fluorescence, NPQ inhibitors and spectral reflectance and concluded that raw reflectance spectra were insufficient to detect the changes caused by the xanthophyll cycle. However, the analysis of second derivative spectra showed that the ratio δδ508/ δδ630 was strongly related to the ratio DT/DD and could successfully be used to follow the diatom xanthophyll cycle non-destructively (Figure 25).

24

Figure 25: Relationship between DT/DD and δδ508/ δδ630 in two diatom species (Amphora sp. and Cylindrotheca sp., circles and squares respectively), cultivated at two different light regimes (LL - 25 µmol -2 -1 -2 -1 photons.m .s and HL - 400 µmol photons.m .s , black and white symbols respectively). From Jesus et al. (2008). Although this study was based on artificial biofilms constructed from diatom cell suspensions, it showed a very high sensitivity to low DT concentrations and very good results with different diatom species. With further validation, this index could prove to be very useful as a PRI replacement in aquatic and intertidal sediment communities, which are frequently dominated by diatoms.

3.3.2 Moisture content effect on sediment reflectance spectra Remote sensing provides a tool for mapping MPB and their sedimentological environment. The mapping of types of substrata and their MPB biomass has been successfully done by Combe et al. (2005). Few studies have, however, taken into account the physical properties that affect the reflectance properties of microalgal communities. The ability to map MPB abundance from hyperspectral images is dependent on the substrate background (sand vs. mud), the amount of moisture, the particle size and, to some extent, amount of organic matter. Indeed, sediment reflectance is strongly affected by slight differences in moisture content, reducing the spectral contrast needed to identify the absorption due to specific photosynthetic pigments or minerals, and consequently determining the ability to map sediment through remote sensing techniques. Additonally, by lowering the reflectance, moisture has an impact on SNR, decreasing it and making identification of weaker features more difficult. In a sediment matrix, water is present in macropores and micropores and is adsorbed by surface of particles. Depending upon the sediment texture, three hydration states can be defined; (1) ‘Saturated water’ is where water fills both types of pore spaces and is also sufficient to cover the surface of most particles that constitute the sediment, (2) ‘Free water’ is where capillary water or bulk water is contained within pore spaces but does not cover all particle surfaces, (3) ‘Surface adsorbed water’, also called ‘hygroscopic water’, is a few layers of water molecules loosely bound to the external surface of sediment particles. Such surface adsorbed water may be linked to minerals over a wide range of wavelengths. Free water and adsorbed water are contained together in the sediment interstices; this water is also called ‘interstitial water’. There is an additional situation that may occur for hydrated minerals where the water forms part of the structure of the mineral. This water is known as ‘structural water’. Surface water can mask the true spectral characteristics of the underlying sediment, probably causing an underestimation of MPB abundance. The degree of underestimation may also be a function of the moisture content. It is possible to identify all surface water directly from the imagery. This may allow a suitable correction to be carried out on the affected areas of the abundance map derived from hyperspectral imagery. To retrieve the

25

diagnostic absorption features of the diatom species, it would be preferable to use a method which is based on the overall spectral shape rather than single band ratio indices.

®

Figure 26. Laboratory spectra of sand sediment measured with an ASD Fieldspec3 FR spectroradiometer from 0.3 to 2.5 µm. (a) Effects of particle size variation on reflectance spectra of 8 granulometric mean size fractions for a dry sediment (blue-red). After total dehydration, major moisture wavebands (see arrows) remaining within beach sand spectra are explained by hygroscopic water and structural water (hydroxyl vibration). (b) Effects of moisture content variation on reflectance spectra of beach sand sediment. Major absorption bands are assigned to vibrations of water molecule processes (stretching and bending). Water content increases from 0 (red spectrum) to 25% (blue spectrum).

In Figure 26a, it can be seen that, for sediments with a similar water content (dry sandy sediment), there is a decrease in the overall shape of the spectra with an increase in particle size. This can be attributed to a difference in granulometric properties: each sediment sample has a different sediment structure and porosity, and the particle size affects the reflectance, as predicted by the scattering theory. Although sediment grain size has a significant effect on reflectance values, our analysis is focused on the moisture content effect (Figure 26b). Rainey et al. (2000) demonstrated that the spectral contrast between coarse and fine intertidal sediments is primarily a product of their very different spectral responses to moisture loss. Moreover, it is important to take moisture content into account as it is the main source of temporal variability in sediment reflectance and can be considered as a proxy for one of the major stress elements in an intertidal environment (salinity and temperature and desiccation stress, etc.). Laboratory reflectance measurements Sediment samples were collected in various study sites and reflectance was measured in the laboratory. ® Spectral reflectance was measured over the 0.35 – 2.50 µm wavelength region with an ASD Fieldspec3 FR portable spectroradiometer. The main objective was to investigate absorption parameters; the spectral reflectance changes with sediment surface moisture varying from very high moisture content (i.e. saturated water) to interstitial moisture content (i.e. free water), down to very low moisture content (i.e. adsorbed/hygroscopic water). The links between sediment moisture and absorption parameters described hereafter are discussed in relation to sediment type (from mud (< 63 µm) to sand (< 2 mm)). Sediment samples were first oven-dried at 150°C. After measu ring the reflectance of the oven-dried sample, deionized water was homogeneously distributed in the sediments until the sediment was saturated. The reflectance was measured frequently during the drying process until the sediment mass returned to its dry state. For each spectral measurement, the corresponding moisture content was concurrently measured by immediately weighing the samples. Moisture content was determined from the relative mass loss, defined as the difference in mass before and after heating divided by the sample mass before heating. The mass loss of the different sample types when heated to 150°C was mea sured in situ to determine the gravimetric moisture measurement. Water absorption band analysis Moisture content has a significant effect on reflectance values (Figure 27). Essentially, moisture decreases the reflectance of sediments and affects the shape of the spectra due to the occurrence of well-defined water absorption bands. Moisture content influences not only the general shape of the spectrum (or continuum), mainly in the infrared region, but also the strength of the specific water absorption features (i.e. 0.97; 1.20; 1.40; 1.90; 2.80 µm). Hydration bands are assigned to molecular vibrations (e.g. stretching and bending) and their comparative energy absorption strengths (Hunt and Salisbury, 1970). In most cases, previous 26

investigations have used moisture content based on hydration absorption bands (Ben-Dor et al., 1999; Lobell and Asner, 2002; Liu et al., 2002; Whiting et al., 2004). Results show a quantitative relationship between moisture content and the way in which the shape of the molecular water band changes. However, some of these hydration absorption overtones, specifically those centred on 1.4 and 1.9 um, cannot be used with images because they are corrupted by atmospheric water vapor. As mentioned above, moisture content also influences the general shape of the spectrum or continuum. The MGM continuum can be considered as a possible proxy for moisture content, excluding the hydration bands (Verpoorter et al., 2009).

Figure 27: Sand spectral reflectance measured during dehydration (blue to red) between 0.3 and 2.5 µm with an ASD (Fieldspec3 Pro FR) spectroradiometer. Effects of moisture variation on reflectance spectra; major absorption region within sand spectra; bands with numbers 1, 2, 3 4, 5, 6 assigned to combinations between fundamental and overtone vibrations in water stretching and bending (γ1 symmetric stretching at 2.73 µm; γ2 bending mode at 6.27 µm; γ3 asymmetric stretching at 2.66 µm), and their comparative energy absorption strengths. Modified from Verpoorter (2009). • Spectral Derivative Analysis (SDA) Spectral Derivative Analysis (SDA) is a useful technique that potentially enables the elimination of background signals and the resolution of overlapping spectral features (Demetriades-Shah et al., 1990). The derivative is calculated by dividing the difference between successive spectral values by the wavelength interval separating them. Goodin et al. (1993) suggested that the first and second order derivative transformations reduce interference from background pure water and suspended sediment effects, respectively. The effect of sediment background on the derivative spectra will be small and it should be possible to analyze the specific absorption of the liquid water observable in the VIS-NIR region and also reduce the moisture effect on the general shape of the spectra. The use of derivative spectroscopy for estimating moisture content is not frequently reported in the literature. The change in derivative spectra for sand dehydration as the experiment progressed is shown in Figure 28. Typical derivative spectra are measured over the 0.3 – 2.5 µm wavelength range. Regions of the spectrum at water band wavelengths show particularly large changes in derivative reflectance with moisture content and these are, therefore, candidate spectral regions for the estimation of moisture content with derivative spectroscopy. The derivative reflectance spectra vary in a regular way with moisture content. As the degree of hydration increases, there is an increase in the magnitude of the peaks and a progression of the shoulder position. Magnitudes are more pronounced near the fundamental absorption of water at 2.8 µm bands than at longer wavelengths (Figures 28b, c, arrows showing the increasing magnitudes of the peaks).

27

Figure 28: (a) Spectral reflectance of sand sediment measured during dehydration (blue-red) between 0.3 and 2.5 µm with an ASD (Fieldspec3 Pro FR) spectroradiometer; (b) change in first derivatives of reflectance sand plots (dR/dλ); (c) change in second derivatives. Modified from Verpoorter (2009). Spectra derived from sediment with high moisture content show high magnitude peaks, arrows showing the increasing magnitude of peaks. Note that the peak around 0.5 µm in the first derivative disappears in the second derivative and is not related to water but to background sediment slope.

Here, first and second order spectral derivatives are performed on sediment spectra primarily to locate the position and inflection point of the water absorption bands (Figure 28a, b, c). First derivative spectra are presented in Figure 28b. Peaks in the first derivative reveal the rate of change in slope of the original spectrum; spectral regions with increasing reflectance with wavelength have positive first derivative values while areas with decreasing reflectance have negative values. Zero values in first derivatives pinpoint accurately the minima position of broad absorptions or peaks. The sign of ‘zero crossing’ indicates local 28

maxima (negative sign) or local minima (positive sign) in the original spectrum. The second derivative (Figure 28c) gives access to asymmetry (high magnitude), minima and maxima (derivative sign) and spectral slope (when equal to zero). Consequently, the position of hydration bands can be accurately determined by the use of the first and second derivative orders. In the second derivative order, the peak magnitude is also proportional to the water abundance. •

Water abundance vs. hydration states of sediment

The 0.97 µm band is less corrupted by atmospheric water vapor (Figure 29). The effect of sediment moisture on reflectance was estimated from laboratory spectral measurements by determining the strength of the specific liquid water absorption at 0.97 µm (Figure 29b). This effect was summarized from saturation to ovendried.

Figure 29: (a) Evolution of observed sandy sediment reflectance spectrum in the 0.35 – 1.10 µm spectral range during dehydration (blue-red) and scaled to the minima of their reflectance. As the hydration degree increases, the spectral slope is more pronounced. (b) Evolution of the observed water absorption band at 0.97 µm after continuum removal. Modified from Verpoorter (2009). After continuum removal (see § 3.2 and § 4.2 for details), the band depth parameter was regressed against the gravimetric moisture content (Figure 30a). The same approach can be applied to all reflectance spectra that include hydration absorption bands at 1.20, 1.40, 1.90, and 2.80 µm. Strong variations in the band strength are observed (Figures 29b, 30). Clearly, reflectance responses to moisture variations are non-linear and can be described by an exponential model according to the modified Beer-Lambert law (Figure 30a). The relationship between moisture content and absorption parameters enables the various types of water present in sediment (saturated, free, or adsorbed) to be identified and separated. Three different sections can be detected in the regression curve that can be correlated to different hydration states: saturated water, free water filling macro- and micropores in the sediment, and adsorbed water molecules loosely bound to particles. Three different models, corresponding to the three hydration states, were adjusted to the data. The wavelength limits between these three water forms are not strictly defined, depending on sediment type (i.e. sand, sandy-mud, muddy-sand, mud) and the fixation/removal kinetics. Validation performed on test data sets also showed high coefficients of determination and a slope close to one (Figure 30b).

29

Figure 30: (a) Exponential models to estimate moisture content. X-axis is the measured moisture content in the sample; Y-axis is the estimation of the strength of the specific liquid water at 0.97 µm for a sand sediment sample. It is possible to separate 3 water forms according to the Beer-Lambert law; (1) saturated water, (2) free water, (3) adsorbed water. (b) Comparison between moisture content measured in the sample and water derived from inverting the equations derived from the test data set. From Verpoorter et al. (2009). The strong exponential correlation between moisture content and strength at 0.97 µm appears the most promising spectral criterion to separate water forms. Eliminating the sediment spectral water effect is fundamental to estimate biomass, organic matter, minerals, particle size, and sediment roughness etc. more accurately. According to these relations, it is possible to distinguish spectrally between different sediments in an intertidal area that has been exposed to drying conditions. To classify the sediments accurately within the intertidal zone, it is necessary to minimize the spectral complexity of the image first by identifying and isolating the saturated surface water. •

Soil Moisture Gaussian Model (SMGM) analysis

Most measurements presented in the previous section cover only the 0.4 – 1.1 µm wavelength range, which includes the 0.97 µm water band. If data available also cover the SWIR wavelength range (1.1 – 2.5 µm), other spectral properties than strength of a specific water absorption can be used to estimate moisture content. In the SWIR, the general shape of the spectrum is controlled by the fundamental water absorption at 2.8 µm. A method, known as the ‘Soil Moisture Gaussian Model (SMGM)’ is more practical for this spectral range (Whiting et al., 2004). The same approach was performed on mudflat sediments by Verpoorter (2009). Sediment spectra can be fitted by an inverted Gaussian function in the 1.2 – 2.5 µm spectral range. The inverted Gaussian function is centered on the 2.8 µm fundamental absorption of water. The Gaussian curve defines a convex hull also called the ‘water continuum’. The baseline is extrapolated from the maximum reflectance wavelength to the center of the Gaussian curve. Contrary to the traditional SMGM, the spectra are not normalized to natural log reflectance (Figure 31a). The distance to the inflection point (σ) or Full Width at Half Maximum (FWHM) is the best predictor for water estimates. It was found that the shape of the specific absorption band overtone at 0.97 µm in the VIS-NIR depends, to a large extent, on the fundamental water absorption at 2.8 µm of the sediment (Figure 31b). To evaluate the effectiveness of the Gaussian function at 0.97 µm in order to describe the response of the sediment moisture levels, the overtone band parameter (i.e. the strength of absorption at 0.97 µm) was plotted and compared to the inverted Gaussian function parameter (FWHM at 2.8 µm). In Figure 31b, a region of dispersed points, where the inverted Gaussian model does not converge, is observed in the saturated water region. A global and positive correlation between the fundamental water absorption band at 2.8 µm and the VIS-NIR water absorption band overtone at 0.97 µm was found.

30

Figure 31: (a) The inverted Gaussian function of sand sediment is fitted (solid line) to the convex hull points; original spectrum (dashed line) from the point of the maximum reflectance in the 1.2-2.5 µm spectral range. (b) Evolution of observed water absorption band parameters (FWHM at 2.8 µm and strength at 0.97 µm). The inverted model does not converge for saturated water sediments. Modified from Verpoorter (2009).

4 MAPPING MICROPHYTOBENTHOS FROM IMAGING SPECTROMETER DATA Application to Bourgneuf Bay, France. The bay was imaged by the Digital Airborne Imaging Spectrometer (DAIS) in 2002 during the European Hyperspectral Sensor (HySens) airborne campaign. This instrument has 80 channels, covering the 50012600 nm spectral range: 72 from 496 to 2412 nm, 1 in the Mid-Infrared (MIR) and 6 in the Thermal Infrared (TIR). The spectral region used here, between 496 and 1035 nm, including PAR (photosynthetically active radiation, see § 3.1.1) corresponds to 32 channels with a spectral resolution between 15 and 20 nm. Flight conditions (flight altitude 3400 m, field of view (FOV) 52°, instantaneous field of view (IFOV) 0.189° ) were chosen for a spatial resolution of around 5 m and an image width of 511 pixels, 3.3 km. It occurred under suitable atmospheric conditions: high visibility, a stable atmosphere and low humidity. In situ reflectance spectra (Figure 32) were acquired with a GER 3700 field spectroradiometer (spectral resolution about 1.5 nm) concurrently with the overpass to help validate image calibration to surface reflectance. Field reflectance was determined by first measuring the light reflected from a calibrated white reference panel (Spectralon) and then the light reflected from the surface. Atmospheric corrections were performed by DLR (Deutsche Zentrum für Luft-und Raumfahrt), using the ATCOR 4 algorithm (Richter, 1996). Field spectra of dry sand from surface area 3 times larger than a pixel have been use as the reference homogeneous bright target for this calibration (Figure 32a). With no topographic or shadow effects over the area of interest, the calibration provided good results (see Combe et al., 2005 for details).

31

a

40

c

Radiance

35

Reflectance

60%

g

50%

30 40%

25 20

30%

15

20%

10

b

10%

5 0 0,25

0,5

0,75

1

1,25

1,5

1,75

2

0% 2,5

2,25

e

d

Figure 32: Field acquisition during the DAIS 2002 airborne campaign; (a) surface area 3 times larger than a pixel used for calibration; (b) GER3700 spectroradiometer and Spectralon; (c) radiance spectra of the Spectralon in red, of the sand in black and reflectance spectrum in blue given by the radiance ratio sand/Spectralon ; (d) water pond used to double check reflectance; (e) population of tourists changing the color of the beach in the DAIS image in (f). This shows that a population can be estimated by looking at its impact on spectra (color) even if individuals cannot be distinguished from each other. Reflectance spectra of the main components that can be present within a pixel were measured in the field with a GER 3700 field spectroradiometer: MPB, macroalgae, bare muddy sediments and water layers can be mixed together. Dry sand and terrestrial vegetation are not expected to be in close contact with MPB but are present in the scene. In Figure 33a, MPB, macroalgae and terrestrial herbaceous vegetation spectra show low reflectances in the visible where plant pigments absorb light for photosynthesis. As a consequence, spectra have a similar red edge between 680 nm and 750 nm, characteristic of vegetation, but very different levels for the Infrared (IR) plateau, due to differences in light scattering. In the visible, MPB presents a typical spectral signature because of a lower absorption near 586 nm and the presence of the chlorophyll c absorption band at 630 nm (see § 3.1.1).

Reflectance

0.8

sea water

0.8

Reflectance

0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

grass THV fucus Fucus

sp.

microphytobenthos MPB dry sand

nm

0 400

500

600

700

800

Wavelength (nm)

900

1000

0 400

500

600

700

800

900

1000

Wavelength (nm)

Figure 33: Field spectra of the main components of the image at two resolutions: (a) GER spectroradiometer resolution; (b) DAIS resolution. (THV: terrestrial herbaceous vegetation; MPB: microphytobenthos). 32

For comparison with DAIS pixels, field data were resampled at DAIS resolution (Figure 33b). Such a process flattens the narrow chlorophyll a absorption band at 675 nm and erases the chlorophyll c one. Despite this loss of information, due to the low spectral resolution of DAIS data, there are still enough channels to distinguish clearly MPB from macroalgae and the herbaceous vegetation of the shoreline. The fastest detection of MPB is achieved by a simple display of the three channels producing its characteristic peak, formed by a maximum at 586 nm (ρ586) flanked by minimum reflectances at 496 (ρ496) and 675 (ρ675) nm respectively in the blue, green and red channels of a color composite image as shown in Figure 34a. Areas covered by various amounts of MPB appear green whereas all other vegetation is dark gray. The MPB index (MPBI) defined by Combe et al. (2006) is therefore a simple quantification of this green hue. Similarly, a herbaceous vegetation index (THVI) can be defined with the green peak of reflectance at 511 nm. The presence of MPB is also characterized by a low value of the Normalized Difference Vegetation Index (NDVI), rewritten here at DAIS resolution, whereas macroalgae and terrestrial vegetation show high NDVI values.

MPBI =

2 ⋅ ρ 586 ρ 496 + ρ 675

THVI =

2 ⋅ ρ551 ρ 496 + ρ 605

NDVI =

ρ 798 − ρ 675 ρ 798 + ρ 675

By combining THVI and MPBI with NDVI in a color composite image, as shown in Figure 34b, the spatial distribution can be simply obtained for brown algae (green in Figure 34b), herbaceous vegetation (cyan), sediments (mud and wet sand are dark green to brown while dry sand is black), water (purple) and MPB (light pink).

.496 .586 .675

THVI NDVI MPBI

f(NDVI)

f(MPBI)

mg Ch a. m-2

a b c d Figure 34: Varying color composite of the DAIS image. (a) RGB colors associated with channels at 496, 675 and 798 nm, respectively, to detect the presence of MPB in green; (b) combining THVI and MPBI with NDVI, coloring the MPB in light pink whereas other vegetation (macroalgae, herbaceous vegetation, meadow and marsh) is green, light blue or cyan. Microphytobenthos biomass as a function of NDVI in (c) and as a -2 function of MPBI in (d) expressed in mg Chl a.m (see § 3.1 and § 4.1). Scale bar is in meter. Both NDVI and MPBI are quite efficient for detecting MPB. However, the full quantification requires some field and laboratory experiments to set up a function of transformation between the index of reflectance and -2 biomass in mg Chl a.m .

33

0.6

Reflectance

mg Chla.m-2

200

y = 20835,93x2 - 24212,93x + 7102,06 R2 = 1,00

0.5 150 0.4 Glass fiber 2.9 13.5 34.7 68.3 105.5 146.1 182.3

0.3 0.2 0.1 0.0 400

0.2

500

600

700

800

900

100 y = 323,06x2 - 83,70x + 5,95 R2 = 1,00

50

nm 1000

Reflectance

NDVI 0 0,0

0,1 mg

200

0,2

0,3

0,4

0,5

0,6

0,7

Chl a.m-2

y = 15259,60x2 - 14908,64x + 3709,07 R2 = 1,00 Mud 2.9 13.5 34.7 68.3 105.5 146.1 182.3 0.60 6.37 14.75 24.53

0.1

0.0 400

0.2

150 y = 150,82x2 - 21,68x + 0,19 R2 = 1,00

100

y = 275,92x2 - 2,35x + 1,74 R2 = 1,00

50

nm

NDVI 0

500

600

700

800

900

1000

Reflectance

0,0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

mg Chl a.m-2

200

y = 7485,21x2 - 7059,61x + 1726,91 R2 = 1,00 Mud 2.9 13.5 34.7 68.3 105.5 146.1 182.3 0.60 6.37 14.75 24.53 DAIS

0.1

0.0 400

500

600

700

800

900

150 y = 265,15x2 - 32,46x + 0,17 R2 = 1,00 100 y = 391,45x2 - 63,23x + 2,28 R2 = 1,00 50

nm

NDVI 0

1000

0,0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

Figure 35: Analysis of reflectance spectra of biofilms reconstituted on filters (glass fiber) in the laboratory (colored spectra) and field biofilms trapped in lens tissues on mudflats (gray spectra). For each spectra, -2 biomass, measured by HPLC, is expressed in mg Chl a.m . For details, see § 3.1. (a) initial laboratory data; (b) relation between biomass and NDVI at GER spectroradiometer spectral resolution; (c) laboratory spectra divided by filter spectrum and multiplied by mudflat spectrum and field biofilm spectra; (d) relation between biomass and NDVI to compare field and laboratory spectra. (e) and (f) same data at DAIS imaging spectrometer resolution. 4.1 First considering that biofilms are homogeneous Figure 35 displays GER spectroradiometer spectra of the increasing biomass of reconstituted biofilms measured by HPLC and the best fitting curve between this biomass and its corresponding NDVI. A careful observation of the spectra (Figure 35a) shows that all absorption bands (carotenoids and chlorophyll a) quickly reach a minimum for blue wavelength as indicated by blue arrow, Figure 35a, around 100 mg Chl -2 a.m . Then, the peak at 586 nm characterizing the presence of MPB collapses quickly (orange arrow, Figure 35a) with increasing biomass, illustrating the saturation phenomenon of reflectance. Consequently, the best fitting curves have to be calculated in two parts, below and above the last biomass value measured before -2 saturation (i.e. 68.3 mg Chl a.m , Figure 35a and b). It follows that (Figure 35b) the NDVI increases quickly -2 with low biomass from 0 (reference spectrum of fiberglass filter) to 68.3 mg Chl a.m and very slowly from -2 -2 68.3 mg Chl a.m up to 200 mg Chl a.m where spectra reach saturation. To simulate the same evolution of biofilm spectra on a mudflat, each of them is divided by the laboratory 34

fiberglass filter spectrum and multiplied by the field mudflat spectrum. The shorter range of NDVI values produced by the darker mud of the mudflat induces new fitting curves and therefore shows that an NDVI from the MPB spectrum can be influenced by background variations. The match between fitting curves of biomass measured on fiberglass in the laboratory or after trapping in lens tissue in the field shows a small shift (Figure 35d). This could be due both to the lens tissue selectivity of the biomass (see § 3.1.3) and to light scattering differences that tilt field spectra toward 400 nm. It could also result from the use of a halogen light source in the laboratory compared to the sun in the field. However, at DAIS resolution (Figure 35e and f), laboratory and field data match each other better. This is confirmed by the addition of new data from the laboratory and the field, as shown in Figure 36. 250

mg Chl a. m-2 diat > 70 mg/m² diat < 70 mg/m²

200

a

y = 9610.28x 2 - 9304.32x + 2311.06 R2 = 0.94

mg Chl a. m-2

250

b

200

field GER field ASD

150

150

100

100

y = 694.17x 2 - 2699.01x + 2672.24 R2 = 0.85

50

y = 51.99x 2 - 50.55x – 17.55 R2 = 0.87

y = 360.23x 2 - 63.23x + 2.22 R2 = 0.93 50

DAIS NDVI

0 0

0.1

0.2

0.3

0.4

0.5

0.6

DAIS MPBI

0

0.7

1

1.2

1.4

1.6

1.8

2

Figure 36: (a) Determination of the function of transformation of NDVI, calculated at DAIS resolution, and biomass obtained by HPLC; (b) the same analysis versus MPBI showing a clear backward evolution over 70 -2 mg Chl a.m . Although MPBI is quite efficient for detecting MPB (Figure 34), it is sensitive to saturation at high biomass. This saturation of spectra produces a backward evolution of the MPBI for biomass values up to ~70 mg Chl -2 -2 a.m (Figure 36b). Fortunately, field data mostly display biomass values much lower than 70 mg Chl a.m , -2 with a maximum at 35 mg Chl a.m . To avoid confusion with areas of macroalgae, the use of MPBI to predict biomass on a DAIS image is limited to values greater than 1.16 and ranging from 0.1 to 0.7 NDVI (Figure -2 34c and d). An NDVI value below 0.5 corresponds to a biomass less than 70 mg Chl a.m while an NDVI -2 above 0.5 corresponds to a biomass greater than 70 mg Chl a.m (blue and orange respectively in Figure 36a). The resulting biomass map from NDVI is restricted to the area dominated by microphytobenthos (compare Figure 34a and b with c) without confusion with macroalgae. This is particularly efficient in the south of the image where a relatively high MPB area is in contact with brown macroalgae belonging to the Fucal order. This would not be possible without the contribution of the MPBI, which masks all macroalgae pixels. Classical satellite multispectral data would not be able to produce such a map of using NDVI. As stated in the introduction, the band width is fundamental even at the relatively low spectral resolution of DAIS.

4.2 Now considering the patchy spatial distribution of biofilms As can be seen on the following picture (Figure 37), the MPB is scattered in patches over the mudflat. Each patch, from 1 cm² to 6 m², is therefore a mixture of various biofilm thicknesses. It behaves spectrally as a linear combination of homogenous subpixels, made of intimate mixtures of translucent MPB and mudflats.

35

a c d b

Figure 37: (a) MPB biofilms observed on a typical mudflat with moderate ripple marks; (b) picture of a detail taken at nadir minimizing the specular reflection of the sky. Scale bars are 10 cm. (c) lens tissue used to trap biofilms by phototactism. They appear whiter when they are still dry (c) and become almost identical to their background when diatoms migrate through the wet lens tissue (d). The complex relationship between MPB biomass and its reflectance is illustrated by the following simulation of pixel spectra covered from 100% to 25% of their pixels by a biofilm. The shape of each spectrum characterizes low and high biomasses from 100% to 50% cover well, but below 25% they are almost identical to each other. Small MPB patches are then difficult to detect. Confusion will occur more frequently for biomass widely scattered over a mudflat. Better results are expected when the coverage is greater than 50%. 0,18

0,18

100% cover

0,16

Scattered patches of high biomass

0,14

0,12

100% 2,9 = 2,9 100% 13,5 = 13,5

0,12

0,10

100% 34,7 = 34,7 100% 68,3 = 68,3 100% 105,5 = 105,5 100% 146,1 = 146,1

0,10

0,08 0,06

100% 182,3 = 182,3

0,06

50% 2,9 = 1,5 50% 13,5 = 6,8 50% 34,7 = 17,4 50% 68,3 = 34,2 50% 105,5 = 52,8 50% 146,1 = 73,1 50% 182,3 = 91,1

0,08

0,04

Reflectance

50% cover

0,16

0,14

0,04

0,02

0,02

a

0,00

c

0,00 400

500

600

700

800

900

1000

0,18

400

500

600

700

800

900

1000

0,18

75% cover

0,16

25% cover

0,16

0,14

0,14

0,12

75% 2,9 = 2,2 75% 13,5 = 10,2 75% 34,7 = 26,1 75% 68,3 = 51,2 75% 105,5 = 79,1 75% 146,1 = 109,6 75% 182,3 = 136,7

0,10 0,08 0,06

0,12

25% 2,9 = 0,7 25% 13,5 = 3,4 25% 34,7 = 8,7 25% 68,3 = 17,1 25% 105,5 = 26,4 25% 146,1 = 36,5 25% 182,3 = 45,6

0,10 0,08 0,06

0,04

Homogeneous low biomass

0,04

0,02

0,02

b

0,00

d

0,00 400

500

600

700

800

900

1000

400

500

600

700

800

900

1000

Wavelength in nm

Figure 38: Variations in MPB spectral response according to four fractional covers. (a) 100% MPB spread on a mudflat corresponding to laboratory spectra divided by filter spectrum and multiplied by mudflat spectrum (Figure 35e); (b) to (d) linear combinations between 75% to 25% of MPB spread on a mudflat and 25% to 75% of mudflat spectra; (e) comparison of various combinations. -2

The reflectance of a pixel mudflat 100% covered by an MPB biofilm of 34.7 mg Chl a.m biomass displays a -2 stronger absorption band than the reflectance of a mudflat 50% covered by a biofilm of 68.3 mg Chl a.m biomass (Figure 38e), whereas the total biomass values covering the pixel are similar: 34.7 and 34.2 mg Chl -2 a.m respectively. Thus, the NDVI is not able to rank both biomasses; neither can they be modeled by a linear combination of extreme endmembers. Consequently, classical NDVI analysis and unmixing are not 36

applicable. Another example of a non-linear mixture is illustrated in Figure 38e: the spectrum corresponding -2 -2 to 75% fractional cover by 34.2 mg Chl a.m biomass best fits the 50% cover by 68.3 mg Chl a.m biomass although the biomass values are quite different. Despite the fact that the use of linear mixing to model MPB biomass must be avoided, a subtle discrepancy between the shapes of both spectra seems to be sufficient to distinguish them. Since each biomass, associated to a sediment type at a given coverage, is characterized by a specific spectrum shape, it appears necessary to build a synthetic spectral database (spectral library) of all possible combinations of biomass, substrate and coverage. Then, the biomass on the image would be given by the match between its spectrum and the spectral library. However, each substrate may display small spectral shape variations due to superficial roughness and the tilt of the surface, as shown in Figure 37, which are incompatible with the accuracy required for the detection of differences between the spectra of Figure 38e. Combe et al. (2005) used the Modified Gaussian Model (MGM) technique to identify a continuum carrying such information without altering the detection of pigments. The overall shape of a reflectance spectrum is related to the physical properties of the surface (e.g. grain size, roughness, moisture content, local slope) which may change from place to place despite a homogeneous composition. This continuum has to be removed to focus on absorptions. Following the definition of Clark et al. (1987), the continuum needs anchorage points at local maximum reflectances, but there is no stable anchorage toward the blue with MPB (see Figure 38). This is overcome by the MGM developed by Sunshine et al. (1990) in which the continuum is a straight line, in a wavenumber scale, insensitive to a set of Gaussian absorption features as illustrated in Figure 39b and c. The MGM source code is available on-line (Sunshine et al., 1998) in IDL (Interactive Data Language) and Fortran. The spectral deconvolution by MGM uses a non-linear least-squares analysis. As described in Figure 39a, the measured spectrum reflectance R at a given wavelength λk is fitted by a superposition of n Gaussian distributions (each is defined by three parameters: central wavelength µ, half-width σ and negative strength s) and a continuum curve (slope a and intercept b for a straight line in wavenumber). Continuum + n gaussian functions

a

centre

 (λ k − µ i )2  n −1 ln (R k )= a λ k + b + ∑ s i ⋅exp  −  2 2σ i i =1  

(

)

b nm

c

Figure 39: Modified Gaussian Model. (a) equation; (b) synthetic spectra in black made up of 5 absorption bands and a continuum in gray; (c) continuum removal by MGM in black and standard spectrum envelope (as given by ENVI software) in magenta showing the better results of the MGM method which fits the absorption band well.

37

Dry sand Microphytobenthos Macroalgae (Fucus species) Terrestrial vegetation (Spartina species)

Spectra of specific materials from individual DAIS pixels (25 m2 area) 40

b

MGM continuum

100

35

Reflectance (%)

Reflectance (%)

30 25 20 15 10 5

80 60 DAIS continuum removed

40 20

b

a

0

0 500

600

700

800

900

Wavelength (nm)

1000

500

600

700

800

900

1000

Wavelength (nm)

Figure 40: (a) Determination of the MGM continuum and (b) normalization to that continuum. The normalization to the MGM continuum also facilitates the distinction between MPB and macroalgae (Figure 40). The normalized spectral distance used by Combe et al. (2005) was therefore able to enhance the detection of MPB as described in Figure 41. The quantification of the match between DAIS spectra and the spectral library mixture is tested by direct spectral comparisons based on either angle or distance (Bakker et al., 2002). The angular separation between two spectra w and v, defined as a spectral angle (SA) by Kruse et al. (1993), can be calculated with a normalized vector product:

SA ( v , w ) = cos

−1

  v⋅w   v ⋅ w 

 .   

(1)

The spectral angle is independent of the length of the two vectors. In Figure 41b, spectra from dark and bright materials are indexed with subscripts a (va, wa) and b (vb, wb) respectively. If only spectral angle is used, intensity information is lost as dark and bright spectra lie along the same vector.

(

)

(

SA v a , w b = SA v b , w b

)

. (2)

Since such variations contain information that is needed for quantifying the amount of any component present within a pixel, intensity distance is more pertinent than spectral angle. The difference between two vectors v and w becomes the ratio of their respective reflectances. The intensity distance (ID) is calculated as the absolute difference in length of v and w:

ID ( v , w ) = v − w

.

(3)

The intensity distance (Eq. 3) is very sensitive to the reference spectrum. That is why the spectrum to be matched (from the image) is considered as a reference, while the matching spectrum to be found (from the library) is the unknown. The library spectrum that minimizes ID is searched for. When the relative brightness increases (from A to B, Figure 41b), the distance also increases. Consequently, comparison of an unknown vector v to a reference vector w is more relevant if the intensity distance is divided by the length of the reference vector (Eq. 4):

38

1 n

SD ( v , w ) =

∑ λ

 ρ v (λ ) − ρ w (λ )    ρ w (λ )  

2

=

v− w . (4).

w

Launeau et al. (2002) used such a distance SD with normalized reflectances to map mineral components. In this study, Eq. 4 was applied with reflectance after continuum removal. This calculation is more discriminating than the spectral angle: va and vb display identical angles while SD is different (Eq. 4 and Figure 41b):

(

)

(

SD v a , w b > SD v b , w b

)

. (5)

For an accurate evaluation of the biomass, spectra have to be compared to the library in the spectral range where MPB pigments dominate reflectance, namely between 496 and 727 nm (14 channels).

Figure 41. Description of metrics that can be used for comparison between spectra from a library and image spectra. Spectral Angle (SA) (Kruse et al., 1993) and Intensity Distance (ID) are commonly used. (a) Definition of SA and ID with two-band spectra (v and w) represented as vectors in a 2-D space (Bakker, 2002). With many channels, SA and ID are calculated in the same way. (b) Consequences of SA and ID calculations. Given a set of airborne spectra (vA, vB) and a set of reference spectra (wA, wB), pairs of vectors (vA, wA), (vB, wB), (vA, wB) and (vB, wA) define the same angle α: SA is insensitive to the mean value of spectra. On the other hand, ID (vA, wA) < ID (vB, wB) and ID (vB, wA) < ID (vA, wB): the mean value of the spectra is preserved when using distance. ID is more sensitive to the difference between high level spectra (bright surface) (circle centered on B) than between low level spectra (dark surface) (A). In order to have equivalent distances, one must normalize with respect to a reference spectrum.

The comparison between spectra from DAIS pixels and the spectral library of all possible mixtures of MPB patches on the sediments of Bourgneuf Bay considerably enhances the detection of MPB. In the area where the fraction cover is greater than 50% (Figure 42b), the map of biomass (Figure 42a) is similar to the map constructed from NDVI with the hypothesis of homogeneous biofilm in Figure 34c. The higher biomass of 35 -2 mg Chl a.m is located at the same spot but lower biomass values are increased from 10 to 15 or 20 mg Chl -2 a.m with the library technique. This means that the heterogeneity of the MPB spatial distribution has no influence in areas of high biomass and is only sensitive in lower ones. As expected, the main differences between both methods are located in areas of fraction cover lower than 50%. Indeed, the sensitivity of this -2 new approach to mapping MPB can detect biofilm at very low biomass (~ 5 mg Chl a.m ) either covering sediment well (point 4, Figure 42) or not (point 1, Figure 42). MPB is also detected in an area rich in macroalgae (point 3, Figure 42) or in the water column, due to the resuspension of photosynthetic organisms by tidal currents (point 2, Figure 42). These detections were not previously possible using MPBI and NDVI (Figure 34c and d).

39

mg Chl a.m²

Figure 42 Biomass and fraction cover maps determined by comparison of the DAIS image with the synthetic library of all possible combinations between intimate mixtures of MPB and various sediments from Combe et al. (2005). See text for explanation of points 1 to 4.

5 PERSPECTIVES The best characterization of MPB biofilms is based on the removal of the continuum from hyperspectral data and the comparison of image spectra with a spectral library in order to map the spatial distribution and the biomass of MPB. Because of the advances in hyperspectral technology, in terms of spectral and spatial resolution, the next stage is to build a new spectral library with different mixtures (sand, macroalgae, euglena, diatoms) and biomass (diatoms) using a better spectral resolution field spectrometer, with the aim of discriminating more surface components through their specific signatures on images of higher hyperspectral resolution. Confusion between the micro- and macro- algae should be reduced, as well as the biomass saturation phenomenon. This objective will rely on new sets of microalgal monospecific cultures at various biomasses, in particular euglenophyceae. The isolation and cultivation of MPB species from natural assemblages thus remains a key step for this remote sensing research project. This could be overcome by a full radiative transfer model of the biofilm currently in progress. Nevertheless, current MPB maps are particularly useful to study the functioning of intertidal flats. For example, these maps could be integrated into a functioning model, associated to primary production, hydrodynamic and resuspension models. Primary production maps could also be obtained. Moreover, with advances in hyperspectral technology, information will not refer to MPB as only an entire entity, but as an assemblage, where ecological or taxonomic groups are distinguishable, such as diatoms and euglenophyceae, or even within diatoms. Thus, mapping the biomass of each group constituting MPB, characterized by varying primary production capabilities, will result in a more realistic primary production map. 40

ACKNOWLEDGEMENTS Authors thank Dr. Richard Murphy from the University of Sydney, for the critical comments on the manuscript. REFERENCES Adler-Golden, SM, Matthew, MW, Bernstein, LS, Levine, RY, Berk, A, Richtsmeier, SC, Acharya, PK, Anderson, GP, Felde, G, Gardner, J, Hoke, M, Jeong, LS, Pukall, B, Mello, J, Ratkowski, A, Burke, HH. Atmospheric correction for short-wave spectral imagery based on MODTRAN4, in Proc. SPIE, Imaging Spectrometry V, vol. 3753, 1999. Ball, DW. Defining terms, Spectroscopy, 10, 1995, 16-18. Barillé, L, Méléder, V, Combe, J-P, Launeau, P, Rincé, Y, Carrère, V, Morançais, M. Comparative analysis of field and laboratory spectral reflectance of benthic diatoms with a Modified Gaussian Model approach. J. Exp. Mar. Biol. Ecol., 343, 2007, 197-209. Bakker, WH, Schmidt, KS. Hyperspectral edge filtering for measuring homogeneity of surface cover types. ISPRS J. Photogramm. Remote Sens., 56, 2002, 246-256. Ben-Dor, E, Irons, JR, Epema, GF. Soil reflectance. In AN Rencz (ed.) Remote sensing for earth sciences: Manual of remote sensing. Wiley and Sons, New York, 1999, pp. 11-188. Ben-Dor, E, Kindel, B, Goetz, AFH. Quality assessment of several methods to recover surface reflectance using synthetic imaging spectroscopy data, Remote Sens. Environ., 90, 2004, 389-404. Berk, A, Bernsten, LS, Robertson, DC. MODTRAN, a moderate resolution model for LOWTRAN7, Air Force Geophysics Laboratory, Hanscom AFB, MA, GL-TR-89-0122, 1989, pp. 38. Bilger, W, and Björkman, O. Role of the xanthophyll cycle in photoprotection elucidated by measurements of light-induced absorbance changes, fluorescence and photosynthesis in leaves of Hedera canariensis. Photosynth. Res., 25, 1990, 173-185. Boardman, JW, Kruse, FA. Automated spectral analysis: A geologic example using AVIRIS data, North Grapevine Mountains, Nevada, Proc. Tenth Thematic Conference on Geologic Remote Sensing, Environmental Research Institute of Michigan, Ann Arbor, MI. 1994, I-407-I-418. th

Boardman, JW, Huntington, JH. Mineral mapping with 1995 AVIRIS data, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 9-11. Boogert, NJ, Paterson, DM, Laland, KN. The implications of niche construction and ecosystem engineering for conservation biology. Biosciences 57(7), 2006, 570-578. Borel, CC, Schläpfer, D. Atmospheric pre-corrected differential absorption techniques to retrieve columnar th water vapor: Theory and simulations, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 13-22. Brotas, V, Cabrita, T, Portugal, A, Serôdio, J, Catarino, F. Spatio-temporal distribution of the microphytobenthic biomass in intertidal flats of Tagus Estuary (Portugal). Hydrobiologia 300/301 (1), 1995, 93-104.

Brotas, V and Plante-Cuny, M-R. Spatial and temporal patterns of microphytobenthic taxa of estuarine tidal flats in the Tagus Estuary (Portugal) using pigment analysis by HPLC. Mar. Ecol. Prog. Ser., 171, 1998, 4357. Burns, R. Mineralogical applications of crystal field theory, 2 Cambridge, 1993, 551 p.

nd

edition, Cambridge University Press,

41

Cabrita, MT, Brotas, V. Seasonal variation in denitrification and dissolved nitrogen fluxes in intertidal sediments of the Tagus estuary, Portugal. Mar. Ecol. Prog. Ser., 202, 2000, 51-65. Carder, KL, Reinersman, P, Chen, RF, Muller-Karger, F, Davies, CO, Hamilton, MK. AVIRIS calibration and application in coastal oceanic environments, Remote Sens. Environ., 44, 2-3, 1993, 205-216. Carrère, V, Conel, JE. Recovery of atmospheric water vapor total column abundance from imaging spectrometer data around 940 nm – Sensitivity analysis and application to Airborne Visible/InfraRed Imaging Spectrometer (AVIRIS) data, Remote Sens. Environ., 44, 2-3, 1993, 179-204. Carrère, V, Spilmont, N, Davoult, D. Comparison of simple techniques for estimating chlorophyll a concentration in the intertidal zone using high spectral-resolution field-spectrometer data. Mar. Ecol. Prog. Ser., 274, 2004, 31-40. Chen, Z, Elvidge, CD, Groeneveld, DP. Monitoring seasonal dynamics of arid land vegetation using AVIRIS th data, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 23-38. Clark, RN. Spectral properties of mixtures of montmorillonite and dark carbon grains: implications for remote sensing minerals containing chemically and physically adsorbed water, J. Geophys. Res., 88, 1983, 1063510644. Clark, RN. Spectroscopy of rocks and minerals and principles of spectroscopy. Remote sensing of the Earth rd sciences: Manual of Remote Sensing, 3 edition, vol. 3, AN Rencz Ed., John Wiley & Sons Inc, 1999, 3-58. Clark, RN, Roush, TL. Reflectance spectroscopy: quantitative analysis techniques for remote sensing applications. J. Geophys. Res., 89 (B7), 1984, 6329-6340. Clark, RN, King, TVV, and Gorelick, NS. Automatic continuum analysis of reflectance spectra. Proceedings, Third AIS workshop, 2-4 June, 1987, JPL Publication 87-30, Jet Propulsion Laboratory, Pasadena, California, 1987, 138-142. Clark, RN, King TVV, Klejwa, M, Swayze, G, Vergo, N. High spectral resolution reflectance spectroscopy of minerals, J. Geophys. Res., 95, 1990, 12653-12680. Clark, RN, Swayze, GA, Gallagher, AJ, Gorelick, N, Kruse, F. Mapping with imaging spectrometer data using the complete band shape least-squares algorithm simultaneously fit to multiple spectral features form rd multiple minerals, in Proc. of the 3 JPL Airborne Visible/Infrared Spectrometer (AVIRIS) Workshop, JPL Pub. 91-28, AVIRIS Pasadena, Calif., 1991, 2-3. Clark, RN, Swayze, GA, Gallagher, AJ. Mapping the mineralogy and lithology of Canyonlands, Utah, with rd imaging spectrometer data and the multiple spectral features mapping algorithm, in Proc. of the 3 JPL Airborne Visible/Infrared Spectrometer (AVIRIS) Workshop, JPL Pub. 91-28, AVIRIS Pasadena, Calif., 1992, 11-13. Clark, RN, Swayze, GW, Gallagher, A. Mapping minerals with imaging spectroscopy, USGS Bull. 2039, U.S. Geological Survey, Office of Mineral Resources, Washington, D.C., 1993, 141-150. Clark, RN, Swayze, GA, Rowan, LC, Livo, KE, Watson, K. Mapping surficial geology, vegetation communities and environmental materials in our national parks: The USGS imaging spectroscopy integrated th geology, ecosystems, and environmental mapping project, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 55-56. Combe, JP. Etude des surfaces planétaires par télédétection visible infrarouge hyperspectrale, Thèse de Doctorat de l’Université de Nantes, 2005, pp. 284. Combe, JP, Launeau, P, Carrère, V, Despan, D, Méléder, V, Barillé, L, Sotin, C. Mapping microphytobenthos biomass by non-linear inversion of visible infrared hyperspectral images, Remote Sens. Environ., 98, 2005, 371-387. Crowley, JK. Mapping playa evaporite mineral with AVIRIS data: a first report from Death Valley, California, 42

Remote Sens. Environ., 44, 2-3, 1993, 337-356. Crowley, JK, Zimbelman, DR. Mapping hydrothermally altered rocks on Mount Rainier, Washington: th Application of AVIRIS data to volcanic hazard assessments, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 63-66. Curran, PJ, Dungan, JL, Gholz, HL. Exploring the relationship between reflectance red edge and chlorophyll content in slash pine, Tree Physiol., 7, 1990, 33-48. Decho, AW, Kawaguchi, T, Allison, MA, Louchard, EM, Reid, RP, Stephens, FC, Voss, KJ, Wheatcroft, RA, Taylor, BB. Sediment properties influencing upwelling spectral reflectance signature: the biofilm gel effect. Limnol. Oceanogr., 48 (1, part 2), 2003, 431-443. de Jong, SM, Chrien, TG. Mapping volcanic gas emissions in the Mammoth Mountain area using AVIRIS. th Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 75-82. de Jonge, VN, and Colijn, F. Dynamics of microphytobenthos in the Ems estuary. Mar. Ecol. Prog. Ser., 104, 1994, 185-18. Delgado, M, de Jonge, VND, Peletier, H. Effect of sand movement on the growth of benthic diatoms. J. Exp. Mar. Biol. Ecol., 145 (2), 1991, 221-231. Demetriades-Shah, T, Steven, M, Clark, J. High resolution derivative spectra in remote sensing. Remote Sens. Environ., 33, 1990, 55-64. Eaton, JW, and Moss, B. The estimation of numbers and pigment content in epipelic algal populations. Limnol. Oceanogr., 11, 1966, 584-595 Elvidge, CD, Chen, Z, Groeneveld, DP. Detection of trace quantities of green vegetation in 1990 AVIRIS data, Remote Sens. Environ., 44, 2-3, 1993, 271-280. Esau, K. Anatomy of seed plants, 2

nd

ed., Wiley, New York, 1977, 550 p.

Farmer, VC. The infrared spectra of minerals, Mineralogical Society, London, 1974, 539 p. Filella, I, Peñuelas, J. The red edge position and shape as indicators of plant chlorophyll content, biomass and hydric status. Int. J. Remote Sens., 15, 1994, 1459-1470. Forster, RM, Jesus, B. Field spectroscopy of estuarine habitats. Int. J. Remote Sens., 27 (17), 2006, 36573669 Gaffey, SJ, McFadden, LA, Nash, D, Pieters, CM. Ultraviolet, visible, and near-infrared reflectance spectroscopy: laboratory spectra of geologic materials, Remote Geochemical Analysis: Elemental and mineralogical composition, CM Pieters and PAJ Englert Eds, Cambridge University Press, Cambridge, 1993, 43-78. Gates, DM. Physical and physiological properties of plants, Remote Sensing with special reference to agriculture and forestry, National Academy of Sciences, Washington, DC, 1970, 224-252. Gamon, JA, Peñuelas, J, Field, CB. A narrow-waveband spectral index that tracks diurnal changes in photosynthetic efficiency. Remote Sens. Environ., 41, 1992, 35-44. Gamon, JA, Field, CB, Roberts, DA, Ustin, SL, Valentini, R. Functional patterns in an annual grassland during and AVIRIS overflight, Remote Sens. Environ., 44, 2-3, 1993, 239-254. Gao, BC and Goetz, AFH. Column atmospheric water vapor and vegetation liquid water retrievals from airborne imaging spectrometer data, J. Geophys., Res., 95, D4, 1990, 3549-3564. Gausman, HW, Escobar, DE, Knipling, EB. Relation of Peperomia obtusifolia’s anomalous leaf reflectance to its leaf anatomy, Photogramm. Eng. Remote Sens., 43, 1977, 1183-1185. 43

Gausman, HW, Escobar, DE, Everitt, JH, Richardson, AJ, Rodriguez, RR. Distinguishing succulent plants from crop and woody plants, Photogramm. Eng. Remote Sens., 44, 1978, 487-491. Gieskes, WWC. Algal pigment fingerprints: clue to taxon-specific abundance, productivity and degradation of phytoplankton in seas and oceans. In: Demers, S (ed) Particle Analysis in Oceanography. Heidelberg: Springer Verlag, Berlin, 1991, pp 61-99. Gitelson, A, Garbuzov, G, Szilagyi, F, Mittenzwey, KH, Karnieli, A, Kaiser, A. Quantitative remote sensing methods for real-time monitoring of inland waters quality. Int. J. Remote Sens., 14, 1993, 1269–1295. Gitelson, A, Kaufman, YJ, Merzlyak, MN. Use of a green channel in remote sensing of global vegetation from EOS-MODIS. Remote Sens. Environ., 58, 1996, 289–298. Gitelson, AA, Merzlyak, MN, Lichtenthaler, HK. Detection of red edge position and chlorophyll content by reflectance measurements near 700 nm, J. Plant Physiol., 148, 1996, 501-508. Goetz, AFH. Imaging spectrometry for Earth observation, Episodes, 15, 1992, 1-14. Goetz, AFH, Vane, G, Solomon, JE, Rock, BN. Imaging spectrometry for Earth remote sensing, Science, 228, 1985, 1147-1153. Goetz, AFH, Kindel, BC, Ferri, M, Qu, Z. HATCH: results from simulated radiances, AVIRIS and Hyperion, IEEE Transactions on Geoscience and Remote Sensing, vol. 41, 2003, 1215-1222. Goodin, DG, Han, L, Fraser, RN, Rundquist, DC, Stebbins, WA, Schalles, JF. Analysis of suspended solids in water using remotely sensed high resolution derivative spectra. Photogramm. Eng. Remote. Sens., 59, 1993, 505-510. Green, RO. Estimation of biomass fire temperature and areal extent from calibrated AVIRIS spectra, th Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 105-114. Green, RO, Dozier, J. Retrieval of surface snow grain size and areal extent from calibrated AVIRIS spectra, th Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 127-134. Guarini, J-M, Blanchard, G, Bacher, C, Gros, P, Riera, P, Richard, P, Gouleau, D, Galois, R, Prou, J, Sauriau, P-G. Dynamics of spatial patterns of microphytobenthic biomass: inferences from a geostatistical analysis of two comprehensive surveys in Marennes-Oléron Bay (France). Mar. Ecol. Prog. Ser., 166, 1998, 131-141. Hakvoort, JHM, Heineke, M, Heymann, K, Kühl, H, Riethmüller, R, Witte, G. Optical remote sensing of microphytobenthic biomass: a method to monitor tidal flat erodibility. Senckenbergiana maritima, 29, 1998, 77–85. Hamilton, MK, Davies, CO, Rhea, WJ, Pilorz, SH, Carder, KL. Estimating chlorophyll content and bathymetry of Lake Tahoe using AVIRIS data. Remote Sens. Environ., 44, 2-3, 1993, 217-230. Hapke, B. Bidirectional reflectance spectroscopy, I, Theory, J. Geophys. Res., 86, 1981, 3039-3054. Hapke, B. Introduction to the theory of reflectance and emittance spectroscopy, Cambridge University Press, New York, 1993. Herzberd, G. Molecular spectra and molecular structure, vol. 2, Infrared and Raman spectra of polyatomic molecules, Van Nostrand Reinhold, New York, 1945, 632 p. Hoepffner, N and Sathyendranath, S. Effect of pigments composition on absorption properties of phytoplankton. Mar. Ecol. Prog. Ser., 73, 1991, 11-23. Horler, DN, Dockray, M, Barber, J. The red edge of plant leaf reflectance. Int. J. Remote Sens., 4, 1983, 273288.

44

Huete, AR. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ., 25, 1988, 295-309. Hunt, GR. Spectral signatures of particulate minerals, in the visible and near-infrared, Geophysics, 42, 1977, 501-513. Hunt, GR. Spectroscopic properties of rocks and minerals, in Handbook of Physical Properties of Rocks, vol. 1, RS Carmichael Ed., CRC Press, Boca Raton, Fla., 1982, 295-385. Hunt, GR. Electromagnetic radiation: the communication link in remote sensing, in Remote Sensing in Geology, BS Siegal and AR Gillespie eds, John Wiley, New York, 1980, 5-45. Jago, RA, Cutler, ME, Curran, PJ. Estimating canopy chlorophyll concentration from field and airborne spectra. Remote Sens. Environ., 68, 1999, 217-224. Jeffrey, SW. Application of pigment method to oceanography. In: Jeffrey, SW, Mantoura, RFC and Wright, SW (eds) Phytoplankton pigments in oceanography. Monographs on oceanographic methodology. UNESCO publishing, Paris, 1997, pp 127-166. Jeffrey, SW, and Vesk, M. Introduction to marine phytoplankton and their pigment signatures. In: Jeffrey, SW, Mantoura, RFC and Wright, SW (eds) Phytoplankton pigments in oceanography. Monographs on oceanographic methodology. UNESCO publishing, Paris, 1997, pp 37-84. Jesus, B, Brotas, V, Marani, M, Paterson, D. Spatial dynamics of microphytobenthos determined by PAM fluorescence. Estuar. Coast. Shelf S., 65 (1-2), 2005, 30-42. Jesus, B, Mendes, CR, Brotas, V, Paterson, DM. Effect of sediment type on microphytobenthos vertical distribution: modelling the productive biomass and improving ground truth measurements. J. Exp. Mar. Biol. Ecol., 332, 2006, 60-74. Jesus, B, Mouget, JL, Perkins, RG. Detection of diatom xanthophyll cycle using spectral reflectance. J. Phycol., 44, 2008, 1349-1359. Jesus, B, Brotas, V, Ribeiro, L, Mendes, CR, Cartaxana, P, Paterson, DM. Adaptations of microphytobenthos assemblages to sediment type and tidal position. Cont. Shelf Res., 29 (13), 2009, 1624-1634. Jordan, CF. Determination of leaf area index from quality of light on the forest floor. Ecology, 50, 1969, 663– 666. Jorgensen, BB, and Des Marais, DJ. A simple fiber-optic microprobe for high resolution light measurements: Application in marine sediment. Limnol. Oceanogr., 31, 1986, 1376-1383. Kelly, JA, Honeywill, C, and Paterson, DM. Microscale analysis of chlorophyll-a in cohesive, intertidal sediments: implication of microphytobenthos distribution. J. Mar. Biol. Assoc. UK, 81, 2001, 151-162. Kirk, JTO. Light and photosynthesis in aquatic ecosystems. Cambridge: Cambridge University Press. 1996. Kromkamp, JC, Morris, EP, Forster, RM, Honeywill C, Hagerthey, S, Paterson, DM. Relationship of intertidal surface sediment chlorophyll concentration to hyperspectral reflectance and chlorophyll fluorescence. Estuaries and coasts 29 (2), 2006, 183-196. Kruse, FA. Use of Airborne Imaging Spectrometer data to map minerals associated with hydrothermally altered rocks in the northern Grapevine Mountains, Nevada and California, Remote Sens. Environ., 24, 1, 1988, 31-51. Kruse, FA, Lefkoff, AB, Boardman, JW, Heidebrecht, KB, Shapiro, AT, Barloon, PJ and Goetz, AFH. The spectral image processing system (SIPS)--interactive visualization and analysis of imaging spectrometer data. Remote Sensing of Environment 44, 1993, 145-163. Kruse, FA, Richardson, LL, Ambrosia, VG. Techniques developed for geologic analysis of hyperspectral data th applied to near-shore hyperspectral ocean data, Proc. ERIM 4 International Conference, Remote Sensing for Marine and Coastal Environments, Environmental Research Institute of Michigan (ERIM), Ann Arbor, 1, 1997, I-233-I-246.

45

Kühl, M, Lassen, C Jorgensen, BB. Light penetration and light intensity in sandy marine sediments measured with irradiance and scalar irradiance fiber-optic microprobes. Mar. Ecol. Prog. Ser., 105, 1994, 139-148. Lang, HR, Adams, SL, Conel, JE, McGuffie, BA, Paylor, ED, Walker, RE. Multispectral remote sensing as stratigraphic tool, Wind River Basin and Big Horn Basin areas, Wyoming, American Association of Petroleum Geologists Bull., 71, 4, 1987, 389-402. Launeau, P, Sotin, Ch, Girardeau, J. Cartography of the Ronda peridotite (Spain) by hyperspectral remote sensing. Bulletin de la Société Géologique de France, vol. 173, no. 6, 2002, 491-508. Lavaud, J. Fast regulation of photosynthesis in diatoms: Mechanisms, evolution and ecophysiology. Funct. Plant Sci. Biotech., 1, 2007, 267-287. Lavaud, J, Rousseau, B, Van Gorkom, H, Etienne, A-L. Influence of the diadinoxanthin pool size on photoprotection in the marine planktonic diatom Phaeodactylum tricornutum. Plant Physiol., 129, 2002, 1398-1406. Liu, W, Baret F, Gu, X, Tong, Q, Zheng, L. and Zhang, B. Relating soil surface moisture to reflectance. Remote Sens. Environ., 81, 2002, 238-246. Lobell, DB and Asner, GP. Moisture effects on soil reflectance. Soil Sci. Soc. Am. J., 66, 2002, 722-727. MacIntyre, H, Geider, R, Miller, D. Microphytobenthos: the ecological role of the "secret garden" of unvegetated, shallow water marine habitats. I. Distribution, abundance and primary production. Estuaries, 19, 1996, 186-201. Méléder, V, Barillé, L, Launeau, P, Carrère, V, Rincé, Y. Spectrometric constraint in analysis of benthic diatom biomass using monospecific cultures. Remote Sens. Environ., 88, 2003a, 386-400. Méléder, V, Launeau, P, Barillé, L, Rincé, Y. Cartographie des peuplements du microphytobenthos par télédétection spatiale visible-infrarouge dans un écosystème conchylicole. CR Acad. Sci. Paris-Biologie, 326, 2003b, 377–389. Méléder, V, Barillé, L, Rincé, Y, Morençais, M, Rosa, P, Gaudin, P. Spatio-temporal changes in microphytobenthos structure analysed by pigment composition in a macrotidal flat (Bourgneuf bay, France). Mar. Ecol. Prog. Ser., 297, 2005, 83-99. Méléder, V, Rincé, Y, Barillé, L, Gaudin, P, Rosa, P. Spatio-temporal patterns of microphytobenthos assemblages in a macrotidal flat (Bourgneuf bay, France). J. Phycol., 43, 2007, 1177-1190. Middelburg, JJ, Barranguet, C, Boschker, HTS, Herman, PMJ, Moens, T, Heip, CHR. The fate of intertidal 13 microphytobenthos carbon: An in situ C-labelling study. Limnol. Oceanogr., 45 (6), 2000, 1224-1234. Moran, MS, Bryant, R, Thome, K, Ni, W, Nouvellon, Y, Gonzalez-Dugo, MP, Qi, J, Clarke, TR. A refined empirical line approach for reflectance factor retrieval from Landsat-5 TM and Landsat-7 ETM+, Remote Sens. Environ., 78, 2001, 71-82. Murphy, RJ, Tolhurst, MG, Chapman, MG, Underwood, AJ. Remote-sensing of benthic chlorophyll: should ground-truth data be expressed in units of area or mass? J. Exp. Mar. Biol. Ecol., 316, 2005a, 69–77. Murphy, RJ, Tolhurst, MG, Chapman, MG, Underwood, AJ. Estimation of surface chlorophyll-a on an emersed mudflat using field spectrometry: accuracy of ratios and derivative-based approaches. Int. J. Remote Sens., 26 (9), 2005b, 1835-1859. Murphy, RJ, Underwood, AJ, Tolhurst, MG, Chapman, MG. Field-based remote-sensing for experimental intertidal ecology: case studies using hyperspatial and hyperspectral data for New South Wales (Australia). Remote Sens. Environ,. 112, 2008, 3353-3365. Nolin, AW, Dozier, J. Estimating snow grain size using AVIRIS data, Remote Sens. Environ., 44, 2-3, 1993, 231-238.

46

Oppenheim, DR. The distribution of epipelic diatoms along an intertidal shore in relation to principal physical gradients. Botanica Marina, 31 (1), 1988, 65-72. Paterson, DM. Short-term changes in the erodibility of intertidal cohesive sediments related to the migratory behavior of epipelic diatoms. Limnol. Oceanogr., 34 (1), 1989, 223-234. Paterson, D, Wiltshire, K, Miles, A, Blackburn, J, Davidson, I, Yates, M, McGrorty, S, Eastwood, J. Microbiological mediation of spectral reflectance from intertidal cohesive sediments. Limnol. Oceanogr., 43 (6), 1998, 1207-1221. Pearlman, J, Carman, S, Lee, P, Liao, L, Segal, C. Hyperion imaging spectrometer on the new millennium program Earth Orbiter-1 system, in Proc. Int. Symp. Spectral Sensing Research (ISSSR), Systems and Sensors for the New Millennium, 1999. Peñuelas, J, Gamon, JA, Griffin, KL, Field, CB, 1993. Assessing community type, plant biomass, pigment composition, and photosynthetic efficiency of aquatic vegetation from spectral reflectance. Remote Sens. Environ., 46, 110–118. Peterson, DL, Aber, JD, Matson, PA, Card, DH, Swanberg, N, Wessman, C, Spanner, M. Remote sensing of forest canopy and leaf biochemical contents, Remote Sens. Environ., 24, 1988, 85-109. Pieters, CM, Mustard, JM. Exploration of crustal/mantle material for the Earth and Moon using reflectance spectroscopy, Remote Sens. Environ., 24, 1988, 151-178. Pieters, CM, Englert, PAJ. Remote geochemical analysis: Elemental and mineralogical composition, Cambridge University Press, Cambridge, 1993, Topics in remote sensing 4, 594 pp. Pinckney, J, Sandulli, R. Spatial autocorrelation analysis of meiofaunal and microalgal populations on an intertidal sandflat: scale linkage between consumers and resources. Estuar. Coast. Shelf S., 30 (4), 1990, 341-353. Porra, RJ, Pfündel, EE, Engel, N. Metabolism and function of photosynthetic pigments. In: Jeffrey, SW, Mantoura, RFC and Wright, SW. (eds) Phytoplankton pigments in oceanography. Monographs on oceanographic methodology. UNESCO publishing, Paris, 1997, pp 85-126. Rainey, MP, Tyler, AN, Bryant, RG, Gilvear, DJ. The influence of surface and interstitial moisture on the spectral characteristics of intertidal sediments: implications for airborne image acquisition and processing. Int. J. Remote Sens., 21, 2000, 3025-3038. Richardson, LL, Ambrosia, VG. Algal accessory pigment detection using AVIRIS image-derived spectral th radiance data, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 189-196. Richter, R, Schläpfer, D. Geo-atmospheric processing of airborne imaging spectrometry data. Part 2: Atmospheric/Topographic Correction. Int. J. Remote Sens., 23(13), 2002, 2631-2649. Richter, R. Atmospheric correction of DAIS hyperspectral image data. Computers & Geosciences, 22(7), 1996, 785– 793. Roberts, DA, Yamaguchi, Y, Lyon RJP. Calibration of airborne imaging' spectrometer data to per cent reflectance using field spectral measurements. Nineteenth International Symposium on Remote Sensing of Environment, Ann Arbor, Michigan. October 1985 (Ann Arbor, Michigan: Environmental Research Institute, Michigan), 1985, pp. 679-688. Rouse, JW, Haas, RH, Schell, JA, Deering, DW. Monitoring vegetation systems in the Great Plains with ERTS. Proceedings of the Third ERTS Symposium, NASA, SP-351, vol.1 (Washington, DC: NASA), 1973, 309–317. Sabol, DE, Smith, MO, Adams, JB, Gillespie, AR, Tucker, CJ. Monitoring forest regrowth using a multith platform time series, Summaries of the 6 annual JPL Airborne Earth Science Workshop, JPL Pub. 96-4, vol. 1, AVIRIS Workshop, Jet Propulsion Laboratory, California Institute of Technology, Pasadena, Calif., 1996, 203-208.

47

Savitzky, A. and Golay, MJE. Smoothing and differentiation of data by simplified least squares procedures. Analytical Chemistry 36, 1964, 1627-1639. Schaepman-Strub, G, Schaepman, MF, Painter, TH, Dangel, S, Martonchik, JV. Reflectance quantities in optical remote sensing – definitions and case studies, Remote Sens. Environ., 103, 2006, 27-42. Serôdio, J, Cartaxana, P, Coelho, H, Viera, S. Effects of chlorophyll fluorescence on the estimation of microphytobenthos biomass using spectral reflectance indices. Remote Sens. Environ., 113, 2009, 17601768. Spinetti, C, Carrère, V, Buongirono, F, Jefferson Sutton, A, Tamar, E. Carbon dioxide of Pu ù’O o volcanic plume at Kilauea retrieved by AVIRIS hyperspectral data, Remote Sens. Environ., 112, 2008, 3192-3199. Steiner, J, Termonia, Y. and Deltour, J. Comments on smoothing and differentiation of data by simplified least square procedure. Analytical Chemistry 44, 1972, 1906-1909. Sunshine, J, Pieters, CM, Pratt, SF. Deconvolution of mineral absorption bands: an improved approach. J. Geophys. Res., 93 (B5), 1990, 6955-6966. Sunshine, J, Pieters, CM, Pratt, SF, McNaron-Brown, KS. Absorption band modeling in reflectance spectra: th Availability of the Modified Gaussian Model (at www.planetary.brown.edu). Proceedings of the 29 Lunar and Planetary Space Conference, 1998. Swayze, GA, Smith, KS, Clark, RN, Sutley, RJ, Pearson, RM, Vance, JS, Hageman, PL, Briggs, PH, Meier, AL, Singleton, MJ, Roth, S. Using imaging spectroscopy to map acidic mine waste, Environ. Sci. Technol., 34, 2000, 47-54. Swayze, GA, Clark, RN, Goetz, AFH, Chrien, TG, Gorelick, NS. Effects of spectrometer band pass, sampling and signal-to-noise ratio on spectral identification using the Tetracorder algorithm, J. Geophys. Res., 108, E9, 2003, doi:10.1029/2002JE001975. Talsky, G. Derivative spectrophotometry. Low and high order. VCH publishers, Inc. New York, 1994. Tsai, F, Philpot, W. Derivative analysis of hyperspectral data. Remote Sens. Environ., 66, 1998, 41-51. Underwood, GJC. Seasonal and spatial variation in epipelic diatom assemblages in the Severn estuary. Diatom Research, 9, 1994, 451-472. Underwood, G, and Kromkamp, J. Primary production by phytoplankton and microphytobenthos in estuaries. Adv. Ecol. Res., 29, 1999, 93-153. Underwood, GJC, Barnett, M. What determines species composition in microphytobenthic biofilms? Functioning of microphytobenthos in estuaries. Proceedings of the Microphytobenthos symposium, Amsterdam, The Netherlands, 2006, pp 121-138. Van der Grinten, E, Simis, S, Barranguet, C, Admiraal, W. Dominance of diatoms over cyanobacterial species in nitrogen-limited biofilms. Archiv für Hydrobiologie, 161 (1), 2004, 98-111. Van der Grinten, E, Janssen, APHM, De Mutsert, K, Barranguet, C, Admiraal, W. Temperature- and LightDependent Performance of the Cyanobacterium Leptolyngbya Foveolarum and the Diatom Nitzschia Perminuta in Mixed Biofilms. Hydrobiologia, 548 (1), 2005, 267-278. Vane, G, Duval, JE, Wellman, JB. Imaging spectroscopy of the Earth and other solar system bodies, in Remote Geochemical Analysis: Elemental and Mineralogical Composition, CM Pieters and PA Englert eds. Cambridge University Press, Cambridge, 1993, 121-143. Vescovo, L, Gianelle, D. Using MIR bands in vegetation indices for the estimation of grassland biophysical parameters from satellite remote sensing in the Alps region of Trentino (Italy). Adv. in Space Res., 41, 2008, 1764-1772. Verpoorter, C, Carrère, V, Robin, M. Retrieval of physical properties of mudflat sediments from hyperspectral th data using the Modified Gaussian Model and spectral curve fitting. Proceedings of the 5 EARSeL Workshop on Imaging Spectroscopy. Bruges, Belgium, 2007, p.10. 48

Verpoorter, C, Carrère, V, Combe, JP. Sediment characterization in the intertidal zone of the Bourgneuf Bay th (France) using the Automated Modified Gaussian Model (AMGM). Proceedings of the 6 EARSeL Workshop on Imaging Spectroscopy. Tel-Aviv, Israel, 2009, p. 6. Verpoorter, C. Télédétection hyperspectrale et cartographie des faciès sédimentaires en zone intertidale. Application à la Baie de Bourgneuf. Ph.D. thesis, Nantes University, 2009, 423p. http://tel.archivesouvertes.fr/tel-00391184/fr. Whiting, ML, Li, L, Ustin, SL. Predicting water content using Gaussian model on soil spectra. Remote Sens. Environ., 89, 2004, 535-552.

ANNEXES 1.

EXISTING IMAGING SPECTROMETERS

Table “Sources of Imaging Spectrometry Data” (source: http://hydrolab.arsusda.gov/rsbasics/sources.php)

Acronym

Full Name

Manufacturer

AAHIS

Advanced Airborne Hyperspectral Imaging System Airborne Hyperspectral Scanner

SETS Technology

AHS

AIP AIS-1

AIS-2 AISA

AMS

AMSS ASAS

ASTER Simulator AVIRIS

CAESAR

CASI

Daedalus Enterprises, Inc. Airborne Instrument Lockheed Program Martin Airborne Imaging NASA, JPL Spectrometer

Geoscan Ltd. NASA

Number Bands 288

48

NASA, JSC

Airborne Imaging NASA, JPL Spectrometer Airborne Imaging Specim Ltd. Spectrometer for Applications Airborne MODIS NASA Simulator (based on AADS-1268)

Airborne Multispectral Scanner MK-II Advanced Solid State Array Spectroradiometer GER Corp.

Operator

433-12700

2000-6400

NASA, JPL

128

900-2100, 1200-2400

NASA, JPL

128

800-1600, 1200-2400 450-1000

Specim Ltd., 286 3Di, Inc., Galileo Corp. 50

Pty Geoscan Pty 46 Ltd. NASA, GSFC 62

JAPEX Geosciences Institute, Tokyo NASA, JPL NASA Ames

Airborne Visible/Infrared Imaging Spectrometer CCD Airborne NLR Experimental Scanner for Applicators in Remote Sensing Compact Airborne Itres Research Spectrographic Imager

of Spectral Range (nm) 432-832

530-15500

500-12000 400-1060

24

760-12000

224

400-2450

12

520-780

1-288

430-870

49

CASI-2

Compact Airborne Itres Research Spectrographic Imager

1-288

400-1000

CASI-3

Compact Airborne Spectrographic Imager Compact High Resolution Imaging Spectrograph Sensor

1-288

400-1050

40

430-860

91

400-12500

211

400-12000

37

400-12000

79

400-12000

32

400-12000

CHRISS

CIS

DAIS 21115

DAIS 3715 DAIS 7915

EPS-A FLI/PMI

FTVFHSI

GERIS

HYDICE

HyMAP

IISRB IMSS

IRIS ISM LIVTIRS 1

LIVTIRS 2

MAIS

MAS

Itres Research

Science SETS Applications Technology, Int. Corp. Inc. (SAIC) Chinese Imaging Shanghai Inst. Spectrometer Of Technical Physics Digital Airborne GER Corp. Imaging Spectrometer Digital Airborne GER Corp. Imaging Spectrometer Digital Airborne GER Corp. Imaging Spectrometer

Environmental Probe System Fluorescence Line Imagery/Programmable Multispectral Imager Fourier Transform Visible Hyperspectral Imager Geophysical and Environmental Research Imaging Spectrometer Hyperspectral Digital Imagery Collection Experiment HyMAP Imaging Spectrometer

DLR, Germany

GER Corp. Moniteq Ltd.

Kestrel FIT

Dept. Fisheries Oceans

of 228 &

Corp.,

GER Corp.

Naval ERIM Research Laboratory Integrated Spectronic Pty Ltd. Imaging Bomem

Infrared Spectrometer Image Multispectral Pacific Sensing Advanced Technology Infrared Imaging ERIM Spectroradiometer Imaging Spectroscopic DESPA Mapper Livermore Imaging Lawrence Fourier Transform Livermore Imaging Spectrometer Labs Livermore Imaging Lawrence Fourier Transform Livermore Imaging Spectrometer Labs Modular Airborne Shanghai Inst. Imaging Spectrometer Of Technical Physics MODIS Airborne Daedalus NASA Ames Simulator Enterprise Inc. & GSFC

430-805

256

440-1150

63

400-2500

210

413-2504

128

400-2504

1720

3500-5000

320

2000-5000

256

2000-15000

128

800-3200 3000-5000

8000-12000

71

440-11800

50

530-14500

50

MERIS MIDIS

MIVIS

PROBE-1 ROSIS

SASI

SFSI SMIFTS

SSTI HSI

TRWIS III VIFIS

VIMS-V WIS

2.

Medium Resolution Imaging Spectrometer Multiband Identification and Discrimination Imaging Spectroradiometer Multispectral Infrared and Visible Imaging Spectrometer PROBE-1 Reflective Optics System Imaging Spectrometer Shortwave (Infrared) Airborne Spectrographic Imager SWIR Full Spectrographic Imager Spatially Modulated Imaging Fourier Transform Spectrometer Small Satellite Technology Initiative Hyperspectral Imager TRW Imaging Spectrometer Variable Interference Filter Imaging Spectrometer Visible Infrared Mapping Spectrometer Wedge Imaging Spectrometer

ESA

15

400-1050

Surface Optics JPL Corp.

256

400-30000

Daedalus CNR, Rome Enterprise, Inc.

102

433-12700

ESSI

100-200

400-2400

128

450-850

160

850-2450

22-115

1220-2420

75

1000-5000

384

400-2500

384

400-2500

60

440-890

DLR, MBB

ESSI, Australia GKSS, DLR

Itres Research

CCRS

CCRS

Hawaii Inst. Of Geophysics

TRW Inc.

NASA

TRW Inc. Univ. Dundee ASI

of

NASA Cassini 512 Mission Hughes Santa 170 Barbara Research Center

300-1050 400-2500

FUTURE SPACEBORNE MISSIONS

EnMAP (Germany) (source: http://www.enmap.org/) EnMAP (Environment Mapping and Analysis Program) is a German hyperspectral satellite mission that will provide high quality hyperspectral images on a timely and frequent basis. Its main objective is to investigate a wide range of ecosystem parameters encompassing agriculture, forestry, soil and geological environments, coastal zones and inland waters. This should increase our understanding of coupled biospheric and geospheric processes and thus enable the management and ensure the sustainability of our vital resources. The envisaged launch of the EnMAP satellite is 2012 and the program should last 5 years. Mission objective The primary goal of EnMAP is to measure and analyze quantitative diagnostic parameters describing key processes of the Earth’s surface. Derived geochemical, biochemical and biophysical parameters are assimilated in physically based ecosystem models and ultimately provide information reflecting the status and evolution of various terrestrial ecosystems. In particular, the scientific mission objectives are as follows: • To provide high-spectral resolution observations of biophysical, biochemical and geochemical variables over the wavelength range from 420 to 2450 nm in contiguous 5-10 nm wide bands sampled at 10 nm intervals. The spatial resolution is 30 m x 30 m. • To observe and develop a wide range of ecosystem parameters encompassing agriculture, forestry, soil/geological environments, and coastal zones and inland waters. • To acquire high resolution spatial and spectral data from space that will enable/improve the retrieval of quantitative parameters needed by the users, but that are not provided by multispectral sensors. • To provide high quality calibrated data and data products to be used as inputs for improved modeling 51

and understanding of biospheric/geospheric processes. This will further contribute to the assimilation of data/information into such process models. • To develop and market high-level information products meeting the demands of stakeholders in natural resource management. Based on these iterative measurements, remote sensing standard products can be substantially improved and new user-driven information products will be established that could so far only be produced in the frame of scientific airborne hyperspectral campaigns. Sensor characteristics The EnMAP instrument provides information based on 218 contiguous spectral bands in the wavelength range from 420 to 2450 nm. EnMAP is a hyperspectral imager of pushbroom type with two separate spectral channels: one for the VNIR range from 420 to 1000 nm and one for the SWIR range from 900 to 2450 nm. Both channels share a common telescope (TMA) with a field splitter placed in the telescope focal plane. Both spectrometers are designed as prism spectrometers thus providing the highest possible optical transmission with low polarization sensitivity. It is characterized by an SNR of > 500:1 in the VNIR and > 150:1 in the SWIR range at a ground resolution of 30 m x 30 m.

Figure A1: The hyperspectral sensor on EnMAP (source: http://www.enmap.org/) With a nominal swath width of 30 km at nadir and an across track pointing capability of ± 30°, the accessible target range is ± 390 km. This results in a target revisit frequency of roughly 4 days. The sun-synchronous orbit enables the satellite to pass over any given point of the Earth’s surface at the same local solar time, which results in a consistent illumination.

Figure A2: (source: http://www.enmap.org/)

52

The sensors are able to collect up to 5000 km data length per day, which means up to 1000 km within an orbit. Expected products The EnMAP processing chain will deliver the following products to the users: • Level1: systematic and radiometric correction The Level 1 processor will correct the hyperspectral image for known effects, e.g. radiometric nonuniformities, and will convert the system corrected data to physical at-sensor radiance values based on the currently valid radiometric calibration values and dark current measurements. • Level 2a: geometric correction The Level 2a processor will create ortho-images by direct geo-referencing utilizing an adequate digital elevation model. The extraction of Ground-Control-Points from existing reference images by image matching techniques – if suitable reference images are available – serve to improve the Line-of-Sight vector and therefore to increase the geometric accuracy of the ortho-images. • Level 2b: atmospheric correction without geometric correction The Level 2b processor will convert the physical at-sensor radiance values to surface reflectance values separately for land and water applications. This includes the estimation of the aerosol optical thickness and the columnar water vapor. • Level 2: atmospheric correction with geometric correction

Figure A3: The EnMAP processing chain and products (source: http://www.enmap.org/) Planning phases The EnMAP Mission is divided into five phases with respect to the mission progress. Currently, we are in the Mission Preparation Phase where the development of the preliminary design has been successfully finished (Phase B). Based on the requirements of the scientific user community the technical specifications for the EnMAP mission are defined within the Phase A study. In Phase B, the preliminary design has to be developed that delivers the baseline and leads to a preliminary definition of interfaces between the mission segments. Phase C comprises the development of detailed design of the components of the Space and Ground Segment. In Phase D, all mission components are to be verified and validated, full readiness of the ground segment for in-orbit operations has been ensured and utilization for spacecraft in-orbit operations has been authorized. The mission preparation phase will end officially with the countdowns of the EnMAP launch.

PRISMA (Italy) (source: http://www.asi.it/en/activity/earth_observation/prisma_) PRISMA (Precursore IperSpectrale of the application mission) is an earth observation system with innovative electro-optical instrumentation which combines a hyperspectral sensor with a panchromatic medium53

resolution camera. The advantage of this combination is that, in addition to a classical capability of observation based on the recognition of the geometrical characteristics of the scene, there is the one offered by hyperspectral sensors which can determine the chemical-physical composition of objects present on the scene. This offers the scientific community and users many applications in the field of environmental monitoring, resource management, crop classification, pollution control, etc. Further applications are possible even in the field of national security. Operational life will be 5 years, launch is foreseen for 2010. Objectives PRISMA is a small Italian mission of demonstrative/technological and pre-operational nature. The fundamental objectives are the following: • to develop a small mission entirely in Italy for monitoring natural resources and atmosphere characteristics taking advantage of the prior development carried out by ASI; • to make available in a short period of time the data necessary to the scientific community for developing new applications for environmental risk management and observation of the territory, based on high-resolution spectral images; • to test the hyperspectral payload in orbit.

Figure A4: PRISMA (source: http://www.asi.it/en/activity/earth_observation/prisma_) Sensor characteristics Spatial resolution: 20-30 m (hyperspectral) / 2.5-5 m (Panchro) Swath width: 30-60 km Spectral range: 0.4-2.5 µm (hyperspectral) / 0.4-0.7 µm (Panchro) Continuous coverage of spectral range with 10 nm resolution Application domains Mapping of land cover and agricultural landscape Pollution monitoring Quality of inland waters Coastal zones and Mediterranean sea Soil moisture Carbon cycle monitoring HyspIRI (USA) (Source: http://www.nap.edu/catalog/11820.html) Science objectives The Hyperspectral Infrared Imager (HyspIRI) mission aims to detect responses of ecosystems to human land management and climate change and variability. For example, drought initially affects the magnitude and timing of water and carbon fluxes, causing plant water stress and death and possibly wildfires and changes in species composition. Disturbances and changes in the chemical climate, such as O3 and acid deposition, cause changes in leaf chemistry and the possibility of vulnerability to invasive species. The HyspIRI mission can detect early signs of ecosystem change through altered physiology, including agricultural systems. Observations can also detect changes in the health and extent of coral reefs, a bellwether of climate change. Those capabilities have been demonstrated in spaceborne imaging spectrometer observations but have not been possible with existing multispectral sensors. Variations in mineralogical composition result in variations in the optical reflectance spectrum of the surface that indicate the distribution of geologic materials and the condition and types of vegetation on the surface. Gases from within the Earth, such as CO2 and SO2, are sensitive indicators of volcanic hazards. They also 54

have distinctive spectra in both the optical and near-IR regions. The HyspIRI mission would yield maps of surface rock and soil composition that in many cases provide equivalent information to what can be derived from laboratory x-ray diffraction analysis. The hyperspectral images would be a valuable aid in detecting the surface expression of buried mineral and petroleum deposits. In addition, environmental disturbances accompanying past and current resource exploitation would be mapped mineralogically to provide direction for economical remediation. Detection of surface alterations and changes in surface temperature are important precursors of volcanic eruptions and will provide information on volcanic hazards over areas of Earth that are not yet instrumented with seismometers. Variations in soil properties are also linked to landslide susceptibility. Mission and Payload The HyspIRI mission uses imaging spectroscopy (optical hyperspectral imaging at 400-2500 nm and multispectral IR at 8-12 µm) of the global land and coastal surface. The mission would obtain global coverage from LEO with a repeat frequency of 30 days at 45 m spatial resolution. A pointing capability is required for frequent and high resolution imaging of critical events, such as volcanoes, wildfires, and droughts. The payload consists of a hyperspectral imager with a thermal multispectral scanner, both on the same platform and both pointable. Given recent advances in detectors, optics, and electronics, it is now feasible to acquire pushbroom images with 620 pixels cross-track and 210 spectral bands in the 400 to 2500 nm region. If three spectrometers are used with the same telescope, a 90 km swath results when Earth’s curvature is taken into account. A multispectral imager similar to ASTER is required in the thermal IR region. For the thermal channels (five bands in the 8 to 12 µm region), the requirements for volcano-eruption prediction are high thermal sensitivity of about 0.1 K and a pixel size of less than 90 m. An optomechanical scanner, as opposed to a pushbroom scanner, would provide a wide swath of as much as 400 km at the required sensitivity and pixel size. The HyspIRI mission has its heritage in the imaging spectrometer Hyperion on EO-1 launched in 2000 and in ASTER, the Japanese multispectral SWIR and thermal IR instrument flown on Terra. The hyperspectral imager’s design is the same as the design used by JPL for the Moon Mineralogy Mapper (M3) instrument on the Indian Moon-orbiting mission, Chandrayaan-1, and so will be a proven technology. Launch is foreseen mid-2015. Japan (source: http://www.spacemart.com/reports/) Japan’s Ministry of Economy, Trade and Industry (METI) recently announced the initiation of a 5-year program for the research and development of a next generation Earth observation satellite payload with hyperspectral capabilities. This future payload will combine traditional multispectral imaging at 5-meters spatial resolution with hyperspectral imaging at lower spatial resolution but at over 185 different spectral bands.

55