Poster_IGARSS_4913 [Mode de compatibilité] - i-sea

Hyperspectral field database in support to coastal wetland mapping
A. Dehouck1,2, V. Lafon1,2, B. Lubac2,3, S. Kervella1,2, D. Bru2,3, M. Schmeltz2,3, A. Roubache1,2
1
GEO-Transfert, ADERA, avenue du Dr Schweitzer, 33600 Pessac, France
de Bordeaux, EPOC, UMR 5805, avenue des facultés, 33405 Talence, France
3 CNRS, EPOC, UMR 5805, avenue des facultés, 33405 Talence, France
2 Université
Introduction and objectives
Coastal areas are very attractive places worldwide that undergo significant changes due to natural variability, climate change and anthropogenic
activities. The macrotidal lagoon of Arcachon (SW of France, Figure 1) is characterized by a wide range of natural habitats classified in the Natura
2000 network. The upper tidal levels are occupied by coastal salt-marshes (Figure 2) which constitute valuable habitats with regards to biodiversity
and ecosystem functioning. Intertidal mud and sand flats are partially covered by a thin vegetation cover of Zostera noltii aquatic plants (Figure 3).
Figure 2. Salt marsh
Figure 3. Zostera noltii over intertidal flats
Spaceborne remote sensing is an efficient and low-cost tool to ensure long-term monitoring over wide coastal areas and assess their changes over
time [1,2,3]. This contribution forms a part of an ongoing research work that aims to derive accurate spatial maps of the Arcachon lagoon from highresolution multispectral optical imagery (Spot, Formosat). This paper investigates the hyperspectral signatures of the main intertidal facies over
space and time in order to determine if the vegetation species can be dissociated one from each other using multispectral bands or band
associations (e.g. NDVI), and if not, what is the potential of hyperspectral imagery to do so.
Figure 1. Arcachon lagoon. Red boxes
locate field sites.
Data and methods
The hyperspectral database consists of about three hundred and twenty-five reflectance
spectra. Various sediment grain size, vegetation species, vegetation cover have been
sampled (Fig. 4) at a seasonal timescale. In particular, we collected spectra over Zostera
noltii seabeds with various sediment backgrounds (sand and muddy sand) and typical
Water Framework Directive densities (0-25%, 25-75%, 75-100%), endemic Spartina
maritima and invasive Spartina anglica (lower salt-marsh vegetation), green algal deposits
as well as bare sediments which were occasionally associated with a microphytobenthos
biofilm. Sampled areas are reported on Figure 1. Data presented here are remote sensing
reflectance (Rrs in sr-1) measurements performed at nadir. A series of relevant
accompanying measurements was also realized consisting in sediment grain size, sediment
moisture, microphytobenthos biomass and characterization of vegetation species. Each
measurement was geo-positioned using a Trimble GEO XT GPS (0.5 m accuracy).
Figure 4. Content of the hyperspectral field data base
Results
Zostera noltii vs Spartina
Effect of sediment background
In both summer and winter time, Spartina
maritima presents low reflectances from
the blue to the red domain in comparison
to Z. noltii and Spartina anglica covers
(Figure 8). Also, both Spartina sp. winter
spectra are clearly discriminable from
Zostera spectra. This indicates that
coupling winter and summer bands in a
classification scheme will increase the
discrimination of the lower salt marsh
from intertidal seagrass meadows and the
discrimination between endemic and
invasive Spartina species.
Reflectance data indicate a strong
dependence of Zostera noltii seabeds to
sediment
background
(Figure
6),
particularly for lowly- to moderatelyvegetated covers. As demonstrated in
laboratory [1], reflectance is higher over
the whole spectra for brighter sediments
than for darker ones. Variations in the
NDVI are also background-dependent. This
suggests that the NDVI cannot be used to
retrieve Zostera noltii density, except if it is
associated with a detection tool of the
underlying substrate.
Biofilm detection
A set of reflectance spectra acquired over
sands and muddy sands shows strong light
absorption at 670 nm due to the presence
of chlorophyll (Figure 7). Here, tiny
reflectance variations in the blue to red
range suppose discrimination of the
microphytobenthos from bare sediment
and Zostera seabed may be possible using
multispectral imagery. But in many other
cases, the spectra are much more
confused. Thus, more data are still needed
to assess the capability of either
multispectral and hyperspectral sensors to
detect microphytobenthic biofilms. Again,
the use of NDVI may also bring confusion
with similar NDVI values being both
associated with biofilms and Zostera noltii
covers.
Figure 6. Reflectance spectra of low-density
Zostera noltii seabeds (0-25 %) over sands (red)
and a muddy sand (green) background
Figure 7. Reflectance spectra of bare sediment,
microphytobenthos biofilms and low Zostera noltii
seabeds
Zostera noltii vs green algae
Several green algae species (Ulva,
Enteromorpha, Monostroma, Cladophora)
proliferate in the Arcachon lagoon at
various seasons. Figure 9 compares algae
and
Zostera
reflectance
spectra.
Reflectance behaves differently from the
blue to red wavelengths depending on the
considered algae species: some algae
(Enteromorpha, Cladophora) experiment a
slightly weakened reflectance while
filamentous deposits are quite similar to
Zostera seabed. Thus, it appears necessary
to consider small differences in narrow
bands. More investigations are needed
here to clarify the discrimination potential
of all algae species from Zostera noltii from
the use of both commercial space sensors
and airborne hyperspectral imagery.
Figure 5. Deployment of TriOS sensors over a patch
of Spartina anglica at wintertime
Figure 8. Typical reflectance spectra acquired in summer
(continuous line) and winter time (dashed) of Zostera
noltii seabeds (green), Spartina maritima (red) and
Spartina anglica (purple)
Figure 9. Comparison between Zostera noltii seabeds
(green) and green algae reflectance spectra (blue)
Conclusions
The hyperspectral field dataset recorded along the Arcachon lagoon is crucial to adapt a mapping strategy based on optical imagery, in particular when space imagery is the only accessible tool. The
dataset demonstrates the potential of e.g. SPOT and Formosat imagery to derive biosedimentary maps of the intertidal lagoon. These results demonstrate the potential of selecting carefully
multispectral bands or by combining them using a multi-temporal strategy to achieve discrimination of the habitats of interest. In particular, we established that several vegetation species are
discriminable using multispectral imagery (intertidal Zostera noltii, endemic Spartina martima and invasive Spartina anglica). However, algae deposits must be treated carefully, since only some
species seem discriminable only when using narrow spectral bands. Finally, we must highlight the counterproductive use of associating the NDVI to qualify the vegetation density over intertidal flats
in a multispectral classification scheme, since it may confuse biofilms with seagrass covers.
References:
[1] L. Barillé, J.-L. Mouget, V. Méléder, P. Rosa, and B. Jesus, “Spectral response of benthic diatoms with different sediment backgrounds”, Remote Sensing of Environment, 115, pp. 1034–1042, 2011.
[2] L. Barillé, M. Robin, N. Harin, A. Bargain, and P. Launeau, “Increase in seagrass distribution at Bourgneuf Bay (France) detected by spatial remote sensing”, Aquatic Botany, 92, pp. 185–194, 2010.
[3] V. Lafon, V. Marieu, R. Butel, A. Dehouck, J.M. Froidefond, and G. Trut “Cartographie des faciès bio-sédimentaires du Bassin d’Arcachon par imagerie Formosat-2”, Proceedings of the 10èmes Journées Nationales Génie Côtier Génie Civil, Sophia-Antipolis, pp. 563-572, in French,
2008.
This work has recently been funded by the CNES (French Space Center) in the frame of the ORFEO program (Pleiades preparatory program) in 2010 and TOSCA SYNIHAL in 2011.