Printer Friendly Format

Page 2 of 3
| Previous | Next |

BASIC DATA
The three NOAA AVHRR images covering the whole area of Bangladesh and parts of neighboring countries were collected from NOAA. The three images of September 18, 1988; October 31, 1995; and September 18, 1998 were used for the flood of 1988, 1995 and 1998, respectively. These three images from three flood events of 1988, 1995 and 1998 have been found to be of suitable quality, with less cloud amount, and fairly close to the combined peak flow of the three mighty rivers among the available less-cloud-covered images.

Correction is needed to avoid geometric distortion from a distorted image. To establish the relationship between the image coordinate system and the geographic coordinate system positions of several ground control points (GCPs) were chosen on the satellite images. GCPs were selected from points easily identified on the satellite images and on the topographic map of Bangladesh. Geometric corrections were carried out until the root-man square errors were less than one pixel. Both the lengths of one pixel and of one line for NOAA AVHRR images represent 1.1 km on the ground surface. This is the ground resolution of NOAA AVHRR data. After the geometric correction the whole area of Bangladesh was extracted by using a vector layer which was prepared from the geographic map of Bangladesh.

Digital elevation data, physiographic divisions, geological divisions, administrative districts, drainage network and population density data were prepared in a common coordinate system for Bangladesh. Each of GIS and RS image yields 118736 pixels on a display monitor, and each pixel covers 1.1km ´ 1.1km on ground surface after geo-coding. Finally, all digital GIS data were incorporated with geometrically corrected NOAA AVHRR data within a GIS approach.

ESTIMATION OF FLOODED AREA

Estimation of Flooded Area
Although each image for a particular date during the flood permitted mapping of all water bodies present at the time of data acquisition, only three NOAA AVHRR image data of September 18, 1988; October 31, 1995; and September 18, 1998 were used to estimate the flood effects of 1988, 1995 and 1998, respectively.

In our previous studies, we used three NOAA AVHRR image data for the event of 1988 (Islam & Sado 2000a,b,c). To differentiate between water and non-water, Iterative Self-Organizing Data Analysis Technique (ISODATA) clustering of unsupervised classification was performed. ISODATA clustering is an iterative non-hierarchical clustering that use minimum spectral distance to assign a cluster for each candidate pixel. Initially all pixels were classified into several categories, and then these categories were divided into three classes of water, non-water and cloud. After the interpretation of cloud cover pixels, the three categories were divided into two categories of water and non-water. Cloudy skies are expected during the flood, and the NOAA AVHRR cannot receive the radiance from a cloud-covered ground surface. The presence of cloud over the damaged areas after an event limits the usefulness of this data, and difficulty arises with the interpretation of whether a given area beneath the clouds is dry or covered by water.

Therefore, low cloud covered images (cloud coverage for the images taken September 18, 1988; October 31, 1995; and September 18, 1998 were 15.00, 9.08 and 13.07%, respectively) were employed to estimate flooded area, flood affected frequency and floodwater depth category. The cloud-covered pixels were interpreted by using the algorithm that was developed for the recovery of cloud-covered pixels as water or non-water. The details of the algorithm have been described by Islam & Sado (2000b).

Estimated flooded areas for September 18, 1988; October 31, 1995; and September 18, 1998 are 34.74, 27.71 and 36.08%, respectively, when the drainage network map was superimposed onto the flood season images. The flooded area was estimated after subtracting the normal water area (river, lake, pond, etc.) from total inundated area; it was then converted to percentage of land area (non-water area in dry season) of the whole country. Our previous study showed that the flooded area for September 18, 1988 was 40.00% when the dry season image of January 20, 1988 was superimposed onto the September 18, 1988 flood season image. Moreover, the total flooded area estimated by different methods ranged from 47% to 50% where three different images of September 18 and 24 and October 8, 1988 were considered within one event of 1988 (Islam & Sado 2000b,c). The flooded area decreased due to the consideration of drainage network map instead of dry season’s classification image (two classes of water and non-water). Because the dry season image of 1988 indicated the normal water area was only 5.32% whereas drainage map showed the normal water area as 15.15%. Our previous study included error pixels (water area in dry season and non-water area in flood season, 4.57%). Those did not appear in this study, due to superimposing of drainage network map instead of dry season’s classified image. Some studies indicated that flood extent surpassed the previous record and the flood of 1988 covered nearly 60% of the country (Bangladesh Government & UNDP 1989; Rasid & Pramanik 1990). According to the local organization, the flood of 1998 also affected more than 50% of the country. The flood effect of 1998 is still under investigation.


Fig. 2. Concept for flood-affected frequency

Fig. 3. Map of flood-affected frequency using the three events of 1988, 1995 and 1998 floods
CONSIDERED HYDRALIC FACTORS

Flood-affected Frequency
The flood-affected frequency was estimated by using the images of September 18, 1988; October 31, 1995; and September 18, 1998. The flood-affected frequency is determined for each pixel as the ratio of the number of NOAA images within the three flood events of 1988, 1995 and 1998, showing inundation to the total number of cloud free NOAA images available near the peak flood time. The concept of the different degree of flood-affected frequency is shown in Fig. 2. The inundated area that did not appear in any of the above-mentioned three images was considered to be a non-hazard area, and that which appeared in a single image was considered to be a low hazard area. The common inundated area that appeared in two images (MH in Fig. 2) and that which appeared in all three images (HH in Fig. 2) were considered to be medium and high hazard areas, respectively. The flood-affected frequency for the area flooded for the three events was categorized as non-hazard, low, medium and high, according to their hit number of the floods, as shown in Fig. 3. Fig. 3 shows that high flood affected areas are Sylhet trough, flood plain areas of the Meghna and Brahmaputra Rivers and Faridpur trough.


Fig. 4. Map of flood depth for 18 September 1988
Flood Depth
Although flood depth determining from remote sensing imagery is very difficult, maximum likelihood method of supervised classification was used to determine flood depth categories for the individual pixel. Flood depths were classified as shallow, medium and deep by using the images of September 18, 1988; October 31, 1995; and September 18, 1998. Training areas of shallow, medium and deep floods were selected on each image, according to the visual interpretation of difference in color and gray scales for different categories of depth for supervised classification. These were interpreted after superimposing the NOAA images onto a digital elevation image of Bangladesh. The results of different categories of flood depth were examined by the estimated average albedo of the pixels with the same flood depth category.

Generally, deeper waters show lower albedo; the results of albedo estimation show good agreement with these results (Islam & Sado 2000a,c). The ranking for flood depth was categorized as no-flooding, shallow, medium and deep flooding. Three flood depth maps were constructed for three flood events of 1988, 1995 and 1998 by using the above-mentioned images, respectively. Different categories of flood depth for 1988 image are shown in Fig. 4. Fig. 4 shows that Sylhet trough is affected by very deep flooding, with some scattered areas across the Brahmaputra and Meghna Rivers. Medium flooding appear in Sylhet trough; floodplain areas of the Meghna, Ganges and Brahmaputra Rivers; and the lower part of the Ganges River, including Faridpur trough. The flood-affected frequency map (Fig. 3) shows the maximum areas affected by high flooding.

FLOOD HAZARD ASSESSMENT

Best Combination of Thematic Data for a Flood Hazard Map
The concept of two and three dimensional ranking matrix in multiplication mode for different combination of thematic maps is shown in Fig. 5. The best combination of thematic data for flood hazard map was examined among the ten combinations (4C2=6 and 4C3=4) of the thematic data; physiography, geology and land cover classification, for flood-affected frequency and flood depth using a ranking matrix for three dimensional multiplication mode. The selection technique for the best combination of GIS data for flood hazard maps created for flood depth and flood-affected frequency was described in detail in authors previous paper (Islam and Sado, 2000a). Summation of diagonal elements for the possible combinations, 4C2 and 4C3, are shown in Table 1. Possible combinations show that the summations of diagonal elements ranges from 37% to 66%. The combination which shows the maximum total pixels for diagonal elements is considered as the best –in this case, the combination of physiography, geology and land cover. The combination of physiography, and geology is the second best, whilst that of land cover and elevation is the worst. Therefore, a flood hazard map may be derived from land cover classification, physiographic and geological divisions using either flood-affected frequency or flood depth.

In this study, flood hazard assessments were undertaken using land cover, physiographic and geologic features and drainage network data. Land cover classification was carried out using dry season’s NOAA AVHRR image of January 20, 1988. Flood depth and floodaffected frequency were used as hydraulic parameters of the floods.

Hazard Rank Assessment through Land Cover, Physiograpy and Geology Flood hazard ranks were estimated based on a weighted score for land cover, physiographical and geological data for each pixel of the land area of Bangladesh. A weighted score was estimated by


where A, B, C and D represent the occupied area percentage by nonhazardous area, and low, medium, respectively, when flood-affected frequency was considered to be a hydraulic factor, for each category of GIS component. Included are land cover categories (9 categories), physiographic divisions (31 divisions), and geologic divisions (28 divisions) (Islam & Sado 2000a,b). Similarly, A, B, C and D represent the occupied area percentage by nonflooded area, shallow, medium and deep flooding, respectively, for each category of above-mentioned GIS components, when flood depth was considered to be a hydraulic factor.

The coefficients of 0.0, 1.0, 3.0 and 5.0 for A-D in Eq. (1) were used to describe the weight for the flood damage. The top upper half of Table 2 shows the acquired area percentage by nonhazardous, low, medium and high damage areas only for the land cover categories for flood affected-frequency while the lower half of the Table 2 shows the acquired area percentage of nonflooded, shallow, medium and deep flooding only for land cover classification for flood depth of 18 September 1988, with calculated weighted score and hazard rank (HR). Points for the categories of land cover were estimated on the basis of linear interpolation between 0 and 100, where 0 corresponds to the lowest (0), and 100 to the highest (230.21 in the top half of Table 2) weighted score. To quantify the flood hazard, the three rankings for flood damage (HR 1~3) were obtained from the allocated point. Hazard ranks were fixed according to the corresponding value of the points. Points 0-33 corresponded to hazard rank 1, 33-66 correspond to rank 2, and 66-100 correspond to rank 3, as shown in Table 2. Hazard ranks were determined using the same algorithm for geologic divisions and physiographic divisions, using flood-affected frequency.


Fig. 5. Concept of the ranking matrix
Table 1 Summation of diagonal elements of square matrixes for determining the best combination of thematic data for flood hazard map.

Four different categories of flood depth (no-flooding, shallow, medium and deep depth) were estimated independently for the images of September 18, 1988; October 31, 1995; and September 18, 1998. Hazard ranks were determined for each event using the above-mentioned algorithm for physiographic divisions, geologic divisions, and land cover categories by using the flood depth.

Page 2 of 3
| Previous | Next |