MULTI-STAGE SEGMENTATION OF RADAR INTERFEROMETRY IMAGERY FOR LANDFORM MAPPING IN TROPICAL REGION

摘要

Dense vegetation that covers most of the landscape becomes a common limitation in mapping landform in tropical region. This paper examines the use of radar interferometry, geometrically corrected by land-cover height classification, in landform mapping; the use of segmentation method to develop multi-stage method in landform mapping; and the possible appropriate scale obtained from SRTM data. Using Idrisi® and "eCognition ®" softwares, segmentation and multi-spectral classification were applied to identify landform elements which are the types of land-cover from Landsat 8, and elevation, slope, relief intensity and curvatures from SRTM (DEM and slope). Visual interpretation on DEM and land-cover fusion imagery was conducted to derive basic control landform and land-cover maps. The result shows that multi-stage scale parameter, shape and compactness level applied in obtaining land-cover, elevation and slope are prominent in determining the most appropriate each elements classification borders. Despite a complex procedure has to be applied in determining landform classification, the combination of landform elements segmentation outcomes present the border of each landform class distinctly. The comparison between landform maps derived from segmentation process and visual interpretation method demonstrates slight dissimilarities, meaning that multi-stage segmentation approach can improve the digital landform mapping in tropical region. The vertical geometric correction on elevation using land-cover height approach is also effective in revealing the original landform that related to the elements of slope, relief and curvature. This finding is also significant in reducing effort in mapping landform using visual interpretation method for a very large coverage but in detail scale level.