Geoscience Reference
In-Depth Information
accuracy of *1 pixel. Pixel sizes range from *3to*100 m in space-borne imagery (e.g.,
Horritt
2000
; Di Baldassarre et al.
2009
), depending on the orbit revisit time, and can
potentially be excellent for flood extent determination. Misclassification errors do occur
however, with flattened and wet vegetation behaving, in certain situations, in the same way
as open water, and emergent vegetation disrupting the specular reflections in shallow open
water to appear more like dry land. Moreover, orbit repeat times may be low (3 days for
ASAR wide swath mode, 7-10 days for RADARSAT and 35 days for ERS-1 and ERS-2)
compared to the flood dynamics in many basins. From records of flood events all around
the world since 1985 collected by the Dartmouth Flood Observatory (
http://www.
dartmouth.edu/*floods/archiveatlas/index.htm
), it appears that the mean duration of floods
is around 9.5 days and the median duration is 5 days. Accordingly, there may only be a
low probability of a SAR overpass occurring simultaneously with a flood in all but the
largest river systems. Moreover, SAR sensors are designed to be all-purpose instruments
and may not be optimal for flood mapping (see, for example, Bates et al.
2004
). Con-
stellations of satellites are likely to be the only way to achieve a suitable combination of
resolution and revisit frequency (Garc´a-Pintado
2013
). For example, the COSMO-Sky-
Med constellation can offer a revisit time as short as 12 h. The few studies to have obtained
simultaneous aerial photograph and satellite SAR data have shown that the accuracy of
satellite radars in classifying flood extent to be only of the order 80-85 % (Biggin and
Blyth
1996
). As a consequence, significant research effort has been expended in devel-
oping sophisticated techniques to classify SAR imagery into wet and dry areas (see, for
example, Matgen et al.
2007
; Mason et al.
2007
; Giustarini et al.
2013
).
Combining the observations of inundation extent available from optical, passive
microwave and active microwave systems, a number of researchers have developed global
floodplain and wetland inundation extent data sets. For example, Prigent et al. (
2007
) used
passive microwave land surface emissivities calculated from SSM/I and ISCCP observa-
tions, ERS scatterometer responses, and AVHRR visible and near-infrared reflectances
from 1993 to 2000 to calculate average monthly inundated fractions of equal-area grid cells
(0.25 9 0.25 at the equator). Similarly, the Dartmouth Flood Observatory (
http://
floodobservatory.colorado.edu/
, see Adhikari et al.
2010
) uses 250 m resolution MODIS
and other data, such as the SRTM Water Body Data set (SWBD), to map flooding in near
real time and from this compile an archive of large floods. Such data sets provide a first
comprehensive global view of surface water dynamics and flooding.
2.3 Remote Measurements of Water Elevation
Remote measurements of water surface elevation can be obtained from (a) profiling
altimeters such as the JASON and Topex-Poseidon radar altimeters or the Geoscience
Laser Altimeter System (GLAS) on board the ICESat satellite; (b) interferometric mea-
surements of water surface elevation change using pairs of synthetic aperture radar images;
and (c) the intersection of shorelines derived from inundation extent data (e.g., a satellite
SAR scene) with a suitable digital elevation model.
Satellite radar altimeters were primarily designed from oceanic studies and have a
footprint of *2 km and vertical accuracy of decimetres to metres (Birkett et al.
2011
).
Such instruments also have wide (hundreds of kilometres) spacing between tracks which
miss many of the world's rivers and most of the world's lakes. Over the continental land
surface, such instruments therefore only record elevations over the very largest rivers;
however, sophisticated retracking algorithms have recently been developed (e.g., Berry
et al.
2005
), which allow separation of water and other signals in mixed pixels. In this way,
