Automated Near-Realtime Flood Detection and
Mapping Using Terra Modis
PREPROCESSING
In an earlier paper, Low et al (2003) described the processing of MODIS Direct Broadcast at
CRISP using NASA algorithms from Level 0 to Level 2. One of the Level 2 products is the
surface reflectance product produced from NASA MOD09 algorithm (Vermote et al, 1999).
The seven shortwave bands are extracted from this product and reprojected onto Latitude-Longitude
grid over the area of interest using another NASA software called MODIS
Reprojection Tool Swath. Subsequently, the image is screened for clouds and shadow using a
CRISP-developed cloudmask software. This software adapts selected 250m cloud and shadow
detection tests from the NASA MOD35 cloudmask algorithm (Ackerman et al, 2002) by
optimising the thresholds to the region. However, some amount of cloud and shadow still
evade detection and the subsequent compositing algorithm attempts to further reduce such pixels
prior to classification.
COMPOSITING ALGORITHM
During a flood event, the region of interest is normally cloud covered and therefore mapping the
water area using one single image will not produce desirable results. In fact, during the rainy
season, some degree of cloud contamination is inevitable even with one month of images for
compositing. Over water pixels, the surface reflectance of shortwave infrared (Band 6) tends to
be low. By selecting minimum surface reflectance of this band over the month, the algorithm
will select days of flooding. However, shadow pixels also tend to have low surface reflectance
in this band, so another criteria was added to reduce selecting the shadow pixels. The
compositing algorithm ranks the pixels in order of increasing values of Band 6. Instead of
choosing the day with the smallest value in Band 6 i.e. first candidate, it takes the first three
candidates and examines the shortwave infrared Band 5 and selects the day with the highest
Band 5 of the three days. In this way, both selection of shadows and cloud contamination is
further reduced. Of course, this could inadvertently remove some genuine water pixels.
