ABSTRACT: Flood is one of the most severe nature disaster. It is taken place very year in the
Northeastern part of Thailand. It is impractical to explore the flooded area through field
investigating. The methods to extract flood extent from passive remotely sensed data such as
LANDSAT TM , SPOT data are hardly carried out because flooded area is usually covered by
cloud. Microwave remotes sensing is very useful for monitoring flood because it can obtain a
good image especially in bad weather. Flood extent is mostly extracted from SAR images by
visual interpretation. It is accurate but very labor intensive. The purpose of this study is to
extract flood extent using Neural Network method in order to enhance flood analysis. The
RADARSAT SAR data acquired in 2001, 2002 and 2003 were used to monitor flood extent in
the lower Songkhram river basin, Northeast Thailand. The back propagation of Neural Network
Classifier was used to identify flooded area. The Kappa analysis was used to measure an
accuracy of the flooded area from the classifier. The result shows that the overall accuracy of
flood extent in 2001, 2002 and 2003 were 86.01 %, 91.68 and 87.74 % respectively when the
reference data were the flooded area obtained from visual interpretation. Therefore, the Neural
Network technique plays effectively role to extract the flood extent from SAR data.