end of June. The inflorescences or panicles start developing in July and
turn purplish-brown with a fluffy aspect at maturation in autumn.
Seeds are wind-dispersed in early winter with the panicles becoming
thinner and switching to a beige colour. Leaves remain green until
October and turn yellow before drying and falling down in winter. The
stems then dry but stand for a few years before breaking down. In
order to provide sustainable conditions for reed harvesting in winter,
reedbeds are flooded from March to June, dried in summer, flooded in
autumn and drained in winter for mechanical harvest (December–
March). Flooding can be extended through spring or summer if
waterfowl hunting occurs. The total area of reedbeds in the Camargue
is estimated at about 8000 ha, of which 2000 ha is harvested every
year (Mathevet & Sandoz, 1999).
Beds of submerged macrophytes develop in unmanaged marshes
that dry in summer as well as in marshes managed for waterfowl
hunting, which are either permanently flooded with freshwater inputs
or drained shortly in spring (Tamisier & Grillas, 1994). These open
marshes, which vary in size from 0.02 ha to 250 ha, can develop dense
mono- or multispecific stands of pondweeds (Potamogeton pectinatus,
Potamogeton pusillus), Eurasian water milfoils (Myriophyllum spicatum)
or widgeon grasses (Ruppia maritima). Some Chara spp. characteristic of
unmanaged marshes can develop in spring but are generally quickly
replaced by the species mentioned above that are more competitive in
quasi-permanently flooded marshes. Thus, depending upon the water
management and the species, submerged macrophytes develop from
mid-February to late March, reaching their maximum growth in May
through July. A progressive senescence starts from early winter but
some plants can remain until the next spring. Water inputs generating
new emergence can also be observed in autumn. Water turbidity is
generally low because submerged plants limit sediment movement. A
continuous surficial water flow is sometimes favoured by managers to
increase marsh attractiveness to ducks. The total area of submerged
macrophytes in the Camargue has not been estimated.
2.3. Field sampling
2.3.1. Reed and submerged macrophyte beds
Physical access to many wetlands was hindered by water too
shallow for boat access and by roads too bad for vehicle access. The
difficulties associated with the sampling of remote flooded marshes
were further hampered by land privacy. Selection of study plots was a
compromise between admittance, accessibility, and getting a repre-
sentative sample of the Camargue marshes based on aerial photo-
graphs and videos collected during flights by plane or ultra-light
motorized engines (ULM). The number of plots surveyed was further
limited by the relatively short sampling period of optimal plant
growth. The training sample, collected in 2005, consisted of 46 plots of
common reed and 25 plots of submerged macrophytes spread
throughout the Camargue (Fig. 1). The independent validation sample
of 2006 consisted of 21 sites of common reed, and 83 sites of aquatic
beds. All study plots were located in seasonal or permanent shallow
marshes either covered with reed-dominated helophytes or with
submerged macrophytes during some time of the year.
For each habitat type, water and vegetation measures were taken
within 20×20 m squares (i.e., four pixels of a SPOT-5 scene) of
homogeneous vegetation placed within a larger homogeneous zone
and located at least 70 m from the border to reduce edge effects in
spectral response. Plot size was defined in order to contain at least one
purepixel (10×10 m) of a SPOT-5 scene. Each sampling plot was placed
in a different hydrological unit to increase structural diversity and avoid
autocorrelation. They were geolocated with a GPS (Holux GR-230XX)
situated in the centre of the plot at 3m above ground to avoid
interferences caused by high reed stems, using the average position
obtained during the whole process of field data gathering (1–2h).
Water level, plant cover and composition were estimated along two
diagonals crossing the entire plot between May and July, depending
upon the development of the vegetation. Common reed density was
measured by counting the green and dry stems inside four quadrats of
50×50 cm per plot in June or July located at 7m from the center of the
plot in each cardinal direction. Homogeneity throughout the plot was
visually estimated and coded from 1 to 4. Vegetation cover was
evaluated with four digital pictures taken vertically from the ground
level upwards in the centre of the 50-cm quadrats and processed with
CANEYE (Baret & Weiss, 2004), a software that derive canopy
characteristics such as LAI, fAPAR and the cover fraction with several
photographs. The estimation of the canopy characteristics are based on
the transmittance of light through the canopy consider ing the
vegetation elements as opaque (Baret & Weiss, 2004).
Water levels were measured at a permanent rule during veg-
etation sampling, as well as monthly or twice monthly at each
hydrological unit sampled.
2.3.2. Other land covers
Tamarisk (Tamarix gallica), riparian forest, rush, grassland, sand
(dune or beach), salt pan, saline marsh (more or less covered by
perennial halophytes such as Arthrocnemum spp.), other forests
(including pine forest), agricultural and urban areas were extracted
from a vector layer created by the Réserve Nationale de Camargue from
aerial photographs in 2001 provided by the Parc Naturel Régional de
Camargue. Additional categories were digitized based on aerial
photographs and ground or aerial (airplane and ultra-light aircrafts)
surveys: sea, rice, sawgrass (Cladium mariscus), club rush (B. maritimus,
S. littoralis, S. lacustris), cattails (Typha spp.), and groundsel bush
(Baccharis halimifolia). In 2006, an updated land cover was available,
pro viding details for agricultural crops. Homogenous stands of
groundsel bush and cattails were unfortunately too few to be included
in the validation sample. We therefore obtained a total of 640 polygons
for the training sample and 587 polygons for the validation sample.
2.4. Image processing
The Camargue can be covered with a si ngle SPOT-5 scene
(60×60 km). Two seasonal time series of SPOT-5 images (SPOT/
Programme ISIS. Copyright CNES) and field datasets were acquired at
one year intervals for model building (2004–2005) and validation
(2005–2006). Thanks to the Spot satellite programming service,
scenes were acquired in late December, March, May, June, July and
September (October in 2006) of both years. These dates had been
selected based on vegetation phenology and seasonal water manage-
ment of the targeted habitats. The programming service provided
several possible images within a two-week period when meteoro-
logical conditions were not optimal. SPOT-5 has 10- m spatial
resolution and four bands: B1 (green: 0.50 to 0.59 μ m), B2 (red:
0.61 to 0.68 μ m), B3 (near-infrared NIR: 0.79 to 0.89 μ m) and B4
(shortwave-infrared SWIR: 1.58 to 1.75 μ m). Spot scenes came with
radiometric correction of distortions due to differences in sensitivity
of the elementary detectors of the viewing instrument that is the
preprocessing level called 1A (Spot image, 2008).
Scenes were radiometrically normalized using the 6S atmospheric
code (Davranche et al., 2009) developed by Vermote et al. (1997), and
projected to Lambert conformal conic projection datum NTF (Nou-
velle Triangulation Française) using a second-order transformation
and nearest-neighbour resampling. The scenes were georeferenced to
a topographic map at 1:25,000 scale.
We extracted the mean refl
ectance value for each plot of reed and
aquatic beds and each polygon of other land covers from each band of
each scene using the ‘Spatial Analyst’ of ArcGis ve rsion 9.2
(Environmental Systems Research Institute, Meudon, France). Using
these data, we further calculated for each plot and polygon the most
common multispectral indices (Table 1), and multitemporal indices
corresponding to subtractions between each pair combination of
dates. In the data file, these variables were labelled as follow:
554 A. Davranche et al. / Remote Sensing of Environment 114 (2010) 552–562