Home > Application > Natural Hazard Management > Flood & Cyclones






Using remotely sensed data to detect changes of riverbank in Mekong River, Vietnam


Method
  1. Data used
    There were 10 digital images of two types - optical and radar, consisting of 7 dates during the period 1989 - 1999. Optical images include Messr (MOS-1b, Japan), Landsat TM, both with high quality (low cloud). Messr images with 4 bands cover only 80 km x 80 km, thus the two adjoining scenes of 1989 and 1990 were used as one date. Radar image of ERS-2 in September and October 1999 were also analyzed as one date because its swath width is just 100 km x 100 km. Landsat TM (180km) and Radarsat (170 km in extended mode) images cover almost the whole Mekong River (of Vietnam territory). (Table 1)

    Topographic maps constructed in period 1966 - 1968 from aerial photos and geometric measurement, in scale of 1: 50,000 and projection UTM, were used as a baseline data source in this study because of their precision. Shorelines of these maps were taken to be an initial datum for change detection analysis.


  2. Table 1 List of data used
    No Date Type Format Images Bands Resolution (m) Notes
    1 04.12.1989 Optic Digital Messr (MOS-1b) 4 50 x 50 1 date
    2 24.01.1990 Optic -- Messr (MOS-1b) 4 50 x 50
    3 13.03.1995 Radar -- Radarsat 1 25 x 25 C - 5.6cm
    4 21.02.1996 Optic -- Landsat TM 3 30 x 30  
    5 01.10.1999 Radar -- ERS-2 1 30 x 30 C - 5.7cm1 date
    6 12.09.1999 Radar -- ERS-2   30 x 30  
    8 12.04.2000 Radar -- Radarsat 1 25 x 25 1 date
    9 24.11.2000 Radar -- Radarsat 1 25 x 25
    10 06.09.2001 Optic -- Landsat 7-ETM Pan 15 x 15  
    11 1966-1968 Vector -- Topographic maps 1/50,000 UTM

  3. Data processing
    All images were geo-rectified to topographic maps of UTM before processing, interpreting and analyzing. After Landsat image had been registered to UTM projection, Messr images were resized accordingly for geo-rectification.

    In order to extract shorelines from images, each type of imageries was processed in different method based on the level of distinction between water - land and soil - vegetation that the image can reveal. In order to distinguish shoreline or riverbank objects on optical images, NDVI (normalized difference vegetation index) was computed, taken from the formula of [Infra Red - Red] / [Infra Red + Red]. For Landsat images, spectral bands 4 and 3 are used, and for Messr they are band 4 and band 2. Radar images were analyzed on their grayscale texture to create a set of descriptor images. Shoreline is the border between two objects land - water, difference of those make a slight change of backscatter reflectance of radar signals. This also was the process of discriminating land from water, wet land with or without vegetation cover.


  4. Change identification
    Changes of riverbank were directly traced out by comparing the two or three images in pairs. Results of interpretation were transformed into GIS layers by years, in vector format. Shorelines were re-corrected at segments, as they have been mis-interpreted because wetland areas are located next to the bank or optical images cloudy, speckle noise of radar images.

    Change of river segments was detected by superimposing data layers together by the order of raster - vector or vector - vector. Erosion and accretion on river were located and an estimation was made with the aid of GIS.
Page 2 of 4
| Previous | Next |