GISdevelopment.net ---> AARS ---> ACRS 1989 ---> Digital Image Processing 1

Land conditional analysis for flood disaster combining raster / vector data

Masahiro Setojima, Yukio Akamatsu and Yoko Hirose
Kokusai Kogyo Co. Ltd. 3-6-1, Asahigaoka,
Hino-city, Tokyo 191 Nippon

Tokumaru Kimiaki
Nissho Iwai Corp., 4-5, Akasaka 2-chome,
Minato-ku, Tokyo 107 Nippon


Abstract
As Japan has steep topographic features and rivers are so short, there exist high risks for flood disaster when heavy rain attacks. Thanks to river works, great disasters scarcely happen, however big flood disaster does happen every several years.

The study area is Kokai River where flood covered its basin in August 1986. Combining Lands at TM data and various Geographic Information on the system which enable us to process both RASTER AND VECTOR information, we analyzed Land Conditions for larger area.
  1. Analyze land conditions for flood covered area in Kokai River basin, using Lands at data and various geographic information regarding land conditions, adjusting them into raster information, and overlaying them each other pixel by pixel.

  2. Based on the above analysis, extract highly worried area for future flood and collect necessary geographic information.

  3. Land-conditional analysis operating Raster/Vector Geographic Information System.
We expect that the above methodology would give us a holistic approach for worried areas utilizing limited variety of geographic information.

Introduction
Taiphoon # 10 of 1986 passed through Kanto district (Metropolitan Tokyo included) in the night of 4-5 August 1986, and caused Kokai River flood disaster with enormous damages. (Fig. 1)


Fig. 1

Landsat data can effectively be used in these cases, which require macroscopic point of view and urgent processing/analysis. While Landsats bring us broad, simultaneous and periodical information, it has a restriction as it depends on earth surface reflection and emission.

If one want to grasp the realities of the flood and use them for prevention of future disaster, correlation among flooded area, land conditions and land use conditions should be considered.

This study tried (1) to grasp realities of flooded area and extract worried areas in flooded zone, utilizing land sat data and several geographic information, then,(2) to collect more detailed information for the above worried areas for the analysis of land conditions for analysis of entire basin for future floods.

Realities of Kokai River Flood Disaster
Heavy rain, caused by a low pressure derived from Taiphoon .# 10, in the night of 4-5 August 1986 made many rivers in Kanto/Tohoku district overflow. Banks of Kokai River collapsed, and broad areas are covered by water, which was confirmed by Landsat TM in the morning of 6 August1986.

Kokai River has another experience of flood in August 1981, construction of river facilities had been imminent. As, at Kokao River, half of its works have not yet completed, flood disaster may occur again in future heavy rain.

We can point out, as the features of this flood, that (1) bank collapsed near artificial structures (i.e. water induction gate) and that (2)flood water flowed into former river channel and rice field.

Method used for analysis
Multi dimensional and holistic approaches are needed in this study. We chose the following 2 methods for the analysis.
  1. Over laying various images representing land conditions
    - to prepare various geographic information for flooded area and
    - to overlay pixel by pixel, in order to extract worried area.

  2. Combination of Raster/Vector information
    -analyze the features of highly worried areas, chosen by the analysis (1), combining raster and vector information, mainly that of social information.
Extracting worried areas through overlaying image(raster)information
  1. General
    We chose approx. 30,000 ha of Kokai River and Kinu River basin. We used Landsat TM data, path/Row 107/35 of 23-Jan-1985 (CCT) and of 06-Auyg-1986 (False color photo). Procedures are followings.

    1. Land over classification: Using TM data before the flood, extracting 30 x 30 km area, and resample them into pixel size of 30 meter. Classification is made using geocoded 6 bands (excluding band-6). Classes are Rice field, Field, grassland, village, water zone, etc.

    2. Extracting flooded zone: After digitizing TM false color photo after the flood, Gooding it, then conducting level slicing with its blue data (TM Band-2)to extract flooded zone. (photo-1)


      Photo-1 Extraction of flooded zone

    3. Input of land Conditional Information: Based on Topographical Map and Land Condition Map, we input land form classification, slope classification, relative height, in order to establish Image Database. Each pixel is formed as 30x 30 meter, in order to correspond 1to 1with TM data.

    4. Analysis of flooded zone: Correlation was analyzed between flood realities and land conditions, overlaying pixel based land coverage information and land conditional information, which have been extracted from Landsite TM data, for the flooded zone.

    5. Extracting highly worried areas: Taking the results of (4) and already existing data into account, the importance of each land conditional categories was defined. Flood worriness classification for the entire basin was estimated based on this analysis.

  2. Analysis of flood disaster area
    Correlation among land use, land form classification, slope and relative height were analyzed for the flooded zone.

    1. Analysis through Land use: Land use proportion is shown in table1 (a). Rice field occupies 60 percent of flooded zone, and field, open space and grass land occupy 10 percent respectively.

      Table-1(a) Land Use Proportion
      Category Cover %
      Rice field 710.46 61.39
      Field 110.88 9.58
      Open space 102.06 8.82
      Glass Land 146.52 12.66
      Forest 5.40 0.47
      Villages 56.25 4.86
      Water Zone 25.74 2.22
      Total 1,157.31 100.00

      Rice field can be easily flooded, we could say with this proportion. Forests are seldom flooded, and they could be relatively safe area against flood.

    2. Analysis through land form classification: The classification is shown in table 1 (b). Talley and flood plain occupy 70 %, and no other category exceed 10 %. 3 % for table land indicates that flooded/non-flooded boundaries exist along with table land/plain boundaries.

      0 % for former river channel and back swamp seem to be occasional just in this case, normally they would show high correlation with flooded zone.

      Not a few number of natural banks flooded means that this flood was relatively huge and with deep water coverage.

      Table-1(b) Landform Classification
      Category Cover (ha) %
      Valley/Flood Plain 848.07 73.28
      Natural Banks 100.53 8.69
      Former River Channel 0.63 0.05
      Back Swamp 0.00 0.00
      Table Land 37.44 3.24
      River 43.56 3.76
      High Water Channel 57.60 4.98
      Low Water Channel 69.48 6.00
      Total 1,157.31 100.00

      Table-1(c) Slope Classification & Relative Height
      Category Cover %
      0-1/1500
      1/1500-1/500
      1/500- more
      945.81
      202.50
      0.00
      82.50
      17.50
      0.00
      Total 1157.31 100.00
      --- 1.5m
      -1.5 - - 0.5
      -0.5 - + 0.5
      + 0.5 -
      0.36
      466.38
      677.61
      12.96
      0.03
      40.30
      58.55
      1.12
      Total 1157.31 100.00

    3. Analysis through slope classification and relative height: slope class, more than 80 % of flooded zone is the area of less than 1/1500 slope, and area with slope of more than 1/500 was not at all flooded. We can estimate that subtle difference of slope would result in being flooded.

      Table 1 (c) indicates slope classes and relative height coverage. As for In regard to relative height, -0.5-+0.5 m zone was found in every part of flooded zone, which indicate that relatively lower zones have positive correlation with flood.

  3. Macroscopic extraction of worried area in entire basin
    Taking the results of the above 4-2 analysis into account, we extracted worried areas, through pixel based overlay of each land conditions. Photo-2 (a) thru (d).

    Photo-2 (e) is its result. It classified most of the flooded area as highly worried, which testify this analysis as an appropriate one.

Photo 2

Land conditional analysis for worried areas combining Raster/Vector information
  1. General
    In 4., we grasped worried areas against flood through microscopic point of view, using various image data and overlaying them each other. Here, we combine Raster data showing the extracted flood disaster area and Vector data of social and geographical data.

    Object areas are squares of 5 x 5 km located in the center of worried area. One from Kokai River Basin (A), and another between Kokai River and Kinu river (B). Area A was actually flooded in 1986.

    Selected vector data are H\Ground height, micro-topography, Equipment (refuge, water level observation house, etc), river structures (banks, bridges, etc), railroads, roads, villages, former river channels, etc. These vector data are registered to worried area images in raster form, and overlaid onto it in vector form.

    Land conditional analysis for worried areas for flood is made with the above results.

  2. Land conditional analysis of worried area
    Photo 4 shows the result of raster/vector analysis. Then land conditional tendencies of worried area analyzed from various information are stated.

    1. Ground height tendency: Photo 4-(a) (b) are the overlay images of worriness classification images and ground height.

      In area (A), most risky areas have ground height lower than 19meters, and in area(B) lower than 16 meters. Both figures are lower than natural banks more or less.

    2. Micro Topography: Most worried areas correspond mainly to rice field along with rivers. It is because rice field is favorably located in flood plain. The further from rice field and the closer to natural banks, the less worried.

    3. Villages distribution: Villages distribution in worried areas is shown in Photo 4-(c) and (d).While most of houses are located on natural banks where risk for flood is low, some are located near rice field, which is in worried area.

    4. Railroad/Road distribution: As railroads run straight North/South in both (A) and (B), we cannot tell tendencies. Roads are mainly located on natural banks and on table land. In some cases, roads are constructed cutting the edges of natural banks.

    5. Conditional analysis for worried areas
      We would conclude that:

      • The most worried areas correspond to rice field located lower than natural banks.
      • No villages/houses are located in the most worried areas.
      • Some of transportation networks (i.e. railway, roads, etc) go across most worried areas.

Photo-3 Object areas A & B for Raster/Vector overlay


Photo-4(a) Site A


Photo-4(b) Site B


Photo-4(c) Site A


Photo-4(d) Site B


Photo 4 (a)- (d) Raster/Vector Overlay Analysis
4-(a) & (b) are overlay Images of Worriness Classification (raster)&Ground Height (vector).
4-(c) & (d) are overlay images of worriness Classification (raster) & Villages Distribution (vector).

Conclusion
  1. This study aimed to grasp land conditional features of flooded zone and to extract worried areas for flood disaster. While accuracy and definitions of land conditional information used in this analysis are insufficient, we did confirm that land conditional data base for river base in was necessary.

  2. For river basins where sufficient information have not yet equipped, land conditional information should be imminently collected and flood risky areas should be extracted in urgent manner, in order to establish disaster counter measures. For this purpose, this study should be applicable.

  3. For the extraction of worried flood areas from macroscopic view, raster overlay analysis is effective, and for land conditional analysis of zoned worried areas and disaster counter measure planning, vector analysis is effective.

Flow chart

Reference
  • M. Setojima, Y. Akamatsu, Y.Oyama : Analysis of flooded zone using LandsatTM data (JSPRS, 6th symposium proceedings, Dec. 1986)
  • M. Setojima, Y. Akamatsu: Collapse risk classification using image overlay processing )Symposium of Japan Society of Civil Engineering, Oct. 1985)
  • M. Setojima, Y. Akamatsu: Land collapse risk classification using image database 9Symposium of JSCE, Oct.1986)
  • This study was made using MICROIMAGE and TERRAPAK software of Terra-Mar Resource Information Services, Inc.