Automated Near-Realtime Flood Detection and
Mapping Using Terra Modis
John LOW, Soo Chin LIEW, Leong Keong KWOH
Centre for Remote Imaging, Sensing and Processing (CRISP)
National University Singapore
Blk. SOC1 Level 2, Lower Kent Ridge Road, Singapore 119260
Fax: (+65) 7757717
E-mail: crslkkj@nus.edu.sg,
crslsc@nus.edu.sg,
crsklk@nus.edu.sg
ABSTRACT
A prototype of a near-realtime flood mapping system using TERRA MODIS is being described.
This system uses the seven shortwave bands of MODIS with Bands 1 and 2 at 250m and Bands
3 to 7 resampled from 500m to 250m resolution. Having all the bands at the same resolution
will help facilitate compositing of the data, visualisation and classification of the data. The
seven shortwave bands undergo atmospheric correction, cloud and shadow filtering. The 'good'
pixels are composited from Day 1 of each month upto the end of the month over the area of
interest. The compositing algorithm preferentially chooses water pixels over the composite
period. A supervised Maximum Likelihood scheme classifies the resulting composite to three
classes, namely, water, forest and cropland(other non-water, non-forest class). The classified
image is compared to another such image of the normal season. The pixels that have changed
from a non-water class to the water class indicate high confidence that flooding has occurred.
The system was tested with MODIS data over Cambodia and selected SPOT quicklook images
were used as ground truth.
BACKGROUND
Floods are an annual occurrence over many countries in the Southeast Asia region. In the event
of floods, relief agencies and governments would appreciate timely data of the flood extent for
mitigation and assessment. Satellite imagery would be the best resource due to its temporal
frequency and wide spatial coverage. Flood occurrence frequently coincides with intense cloud
activity and optical sensors like MODIS may not be suitable to provide immediate images of the
affected areas. Whilst active sensors like SAR provide cloud-penetrating capabilities to image
at the ground surface, the costs and temporal frequency may be a setback and obstacle to lesser
developed countries. Notwithstanding, SAR imagery has proven to be very useful in flood
detection (Ping Chen, et al 1999).
If delayed access to satellite data is acceptable, then MODIS data would be a low-cost but
effective option to flood-mapping. MODIS is a moderate resolution sensor. There are seven
optical bands optimised for land imaging, with 2 bands at 250m resolution and 5 bands at 500m.
MODIS is availabe on two satellite platforms, namely AQUA and TERRA thereby providing
ample opportunities for monitoring of flood activities. Zhan et al (2002) used MODIS 250m
bands to detect land cover changes such as floods, deforestation and burnt scars.
