Abstract: In this study, the development of a flood hazard map and assessment of flood hazard are described using RADARSAT SAR and GIS data for the historical flood event of 1998. A flood hazard map was developed on the basis of ranking matrix in two dimensional multiplication mode which was calculated using the digital elevation and land-cover data. Flood-affected frequency estimated from multi-temporal SAR imageries was considered as a hydraulic component for the evaluation of flood hazard. Assessment of flood hazard was performed by overlaying thematic data onto derived hazard map. It is demonstrated that the evaluation of flood risk can be done efficiently using GIS and RS data. It is expected that the developed flood hazard map will be useful to mitigate losses of lives and property from future flood disasters in third world cities, particularly in Greater Dhaka, Bangladesh.
INTRODUCTION
Flood delineation and hazard evaluation has been important subject of environmental research for decades. Identifying the area having higher hazard potential is considered vital within the latest strategy for flood management[1]. Recent emphasis on flood management largely corroborate with flood disaster preparedness, prevention and mitigation[2], particularly developing potential flood hazard and risk map of a given area[3], which can contribute to the development of an efficient flood management system.
Floods are an increasing problem in Dhaka, the capital of Bangladesh. In the summer of 1998, Greater Dhaka experienced a devastating flood which was generally described as the flood of the century. The flood duration, rainfall and damages surpassed all previous records[4]. Although the Bangladesh government in association with international aid agencies, has invested large amount of resources in protecting Dhaka and its adjoining areas from recurrent floods by constructing embankment, flood damage and flood affected areas have increased significantly in recent years[5]. The flood vulnerability of Greater Dhaka is aggravated by unplanned urbanization and infilling of lowlands. Besides, flood problem is a neglected priority for urban planning. Furthermore, it is also clear that the increasing flood hazards of Dhaka have not been analyzed for purposes of disaster mitigation[6]. Therefore, following the 1998 catastrophe urban planners and water experts have called for the precise delineation of flood prone areas[7] and to develop flood hazards maps for historical events[8] that can help save life and to mitigate losses from future flood disasters.
During the last two decades, Remote Sensing (RS) has played an increasing role in the fields of hydrology and water resources management[9]. Geographic Information System (GIS) is also used extensively to model surface water and flood damage assessment[10-12]. Recently, the integration capabilities of satellite data with GIS have opened opportunities for quantitative analysis of hydrological events, such as flood, at all geographic and spatial scales. The potential of this facility for improving hazard evaluation and risk reduction is continually being explored[13].
Even though flood management has shifted to its assessment aspect, the big challenge is the lack of generally accepted methods for producing hazard maps or even on the scope of producing such maps[14]. Moreover, many well developed methods usually applied in developed nations may not be accessible to developing country, particularly in Bangladesh due to the lack of digital data or restriction to access data sources. For example, DEM-based detail flood mapping may not possible in Bangladesh since higher resolution digital elevation data for the whole country or part of it is not available. Although satellite remote sensing data has shown great potential for hazard assessment[15], the studies that have used satellite data with GIS for flood hazard assessment remain fairly limited in Bangladesh[16]. This technique is yet to be utilized in Greater Dhaka for the development of hazard maps and assessment of flood hazard.
This study assessed flood hazard in Greater Dhaka by integrating Synthetic Aperture Radar (SAR) data with GIS for the historical greatest flood of 1998.
MATERIALS AND METHODS
The study area was the Greater Dhaka City, Bangladesh. The latitudes and longitudes of the study area are 23°68’ N, 90°33’ E and 23°90’ N, 90°50’ E, respectively. The area is located mainly on an alluvial terrace, known as the Modhupur terrace of the Pleistocene period. Topographically, Dhaka is a flat land. The surface elevation of the area ranges between 1 and 14 m[17]. It lies in the sub-tropical monsoon zone and experience humid climatic conditions. The City experiences about 2000 mm annual rainfall, of which more than 80% occurs during the monsoon season.
In order to analyze the 1998 flood event, seven RADARSAT SAR (Narrow and Wide Beam mode) data were acquired. Six imageries were for the flood season (July-September) and one for the dry season of 1998 (15 Dec). The SAR data was despeckled using the GAMA-MAP filter with 5x5 window size. After suppressing the speckle inherent to SAR images, geometric correction was carried out using a referenced 1999 Landsat TM image of the same area until the root mean square errors resulted in less than one pixel. A second order polynomial fit was applied and the pixel values were resampled to 50 m.
The major impediment to studying flood hazard in Greater Dhaka was the unavailability of GIS data. Therefore, in this study, a number of GIS layers were generated. A Digital Elevation Model (DEM) with 50 m resolution of Greater Dhaka area was obtained from Institute of Water Modeling, Bangladesh. Using the DEM, an elevation map was created. The elevation data were sliced into fourteen categories, at one meter intervals. The land-cover map was generated in a GIS platform, using one IRS-1D Panchromatic data of 29 February 2000 (path 110 and row 055). By displaying the IRS-1D image on a computer, on-screen digitization was performed using ARCVIEW software to generate a digital land-cover map. Accurate locations of land cover categories were guided by the recent topographic map, visual interpretation and field data. Using PC ARC/INFO software, the land-cover map was edited and leveled. Nine land-cover categories were identified for the whole study area. After finalizing the land-cover map, it was converted from vector to raster for flood hazard assessment. An administrative division map of Greater Dhaka was also digitized using the recent topographic map (scale 1:10000). A total of thirty-four administrative boundaries were obtained in the study area. In addition, the drainage network of Greater Dhaka was extracted from IRS-1D PAN data. Finally, GIS and SAR dataset were transformed to the Bangladesh Transverse Mercator Projection (BTM)[18]. This produced a total of 166560 pixels (480x347) on a computer screen for the individual image and GIS layers.
In this study, a simple procedure[19] is adopted for the development of flood hazard map and subsequent hazard assessment. Land-cover, elevation heights, drainage network and administrative data are employed to assess flood hazard potential. The notion of flood-affected frequency is used as a hydraulic component.
In order to assess flood hazard in Greater Dhaka for the 1998 historical event, the estimation of flooded areas from multi-temporal SAR images were performed using a threshold algorithm[20] . Individual flood time image was classified according to flooded and non-flooded categories. After classifying the individual images into two discrete categories (water and non-water), every image was then superimposed on a classified dry seasonal SAR image which has also two classes (water and non-water). This was carried out in order to estimate the net inundated areas. The classification accuracy[21] of derived flood maps was conducted using ground reference data collected from the field by a GPS.
The concept of flood-affected frequency[16] was used to assess flood hazard. In order to develop a flood hazard map of Greater Dhaka, a flood frequency map based on flood duration was developed using the multi-temporal flood time SAR imagery. The classified imageries obtained by threshold algorithm were used to develop a flood frequency map. Total six images classified into water and non-water areas, which were acquired on July 7 and 31 1998; August 10 and 25 1998 and September 10 and 17 1998, were used to develop flood frequency image. Initially, the classified images of July 7 and 31 were combined to construct a single classified image of July; similarly two images of August and two images of September were combined to construct the single images for August and September, respectively. These combinations provided an opportunity to get a common boundary needed to develop a flood-affected frequency map, using the technique suggested by Islam and Sado[16]. The inundated area that appeared in all the images considered the highly damaged area. The common inundated areas that appeared in two and one of the three images were deemed as medium damaged and low damaged areas, respectively. An inundated area that did not appear in any of the images was considered as non-damaged areas. Thus, four flood frequency areas were obtained, corresponding to damage rankings of class 1, class 2, class 3 and class 4 as high, medium, low and non-hazard areas, respectively.
To assess flood hazard for each category of land-cover and elevation, a weighted score was estimated and hazard ranks were determined. This was done by overlaying the GIS database onto the derived flood frequency map. The weighted score was obtained from:
Weighted score= Class 1x1.0+Class 2x3.0+Class
3x5.0+Class 4x7.0 |
(1) |
where, classes 1 to class 4 represent the area percentages occupied by the categories of flood-affected frequency for land-cover and elevation. To describe the degree of flood damage, the coefficients 1.0, 3.0, 5.0 and 7.0 for classes 1 to 4 in equation 1 were used. Furthermore, the points for each land-cover and elevation category/ID (here denoted by Identification Number or ID) was estimated by linearly interpolating between 0 and 10, which corresponded to 0 as lowest and 10 as highest score. From the score, four ranks were decided according to the scale of flood severity. Hazard ranks were fixed by considering the corresponding value of the point, 1-2.5 for hazard rank 1, 2.5-5 for hazard rank 2, 5-7.50 for hazard rank 3 and 7.50-10 for hazard rank 4.
The final flood hazard map was further superimposed on land-cover, elevation and administrative divisions’ maps in order to evaluate the flood risk in Greater Dhaka. An overlaying function was applied and areas within each hazard zone were computed for land-cover, elevation and administrative units to assess flood risk.
RESULTS
Earlier studies on flooded area estimation during the 1998 flood event in Greater Dhaka, revealed that the extent of flooded areas were 34 to 53%[20,22]. The estimation of flooded areas using the threshold algorithm showed that the flood extent ranged from 34.65 (7 July 1998) to 53% (25 August 1998) (Table 1). Three hypotheses can be made concerning the presence of water in each SAR image. These are that: it rained again; more waters came down from the upstream and spread over the already flooded zones, consequently increasing the affected area and both phenomena occurred simultaneously. Analysis of daily rainfall and water level records of the study areas supported the assumptions[4,20].
| Table 1: | Flooded area percentage (%) derived from SAR imageries |
| Table 2: | Flood hazard ranks for land-cover category by flood-affected frequency. (ID: land-cover identification number; HR: hazard rank) |
The derived flood maps from multi-temporal SAR imagery were validated using a ground truth map, On the average, the derived flood maps resulted in 70% spatial agreement[20].
The acquired area percentages of each class number of land-cover (9 categories) and elevation unit (14 divisions) for flood-affected frequency were estimated to develop flood hazard map. The estimated results of land cover classification categories for the flood frequency are shown in Table 2. Two flood hazard maps were developed on the basis of land-cover and elevation intervals, which were used to evaluate flood risk. Using the ranking matrix on a two dimensional multiplication mode[19] a new hazard map was created, considering the interactive effect of land-cover, elevation and flood-affected frequency, which shows the higher hazard ranks. The resulting flood hazard map (with 10 classes) was further re-classified into distinct categories according to the intensity of hazards. This breakdown provided five hazard categories i.e. very high, high, moderate, less and least hazard zones (Fig. 1).
The least hazard zone corresponded to the non-flooded areas, usually do not inundate during monsoon season. This means that the areas not exposed to potential flooding are within the category of least hazard zone. Highly elevated lands (>13 m) of the study area represented the least hazard zone, which is only 1.90% of total land (Table 3). Moderately high lands are within the group of less hazard zone. These are the most extensively developed areas in Greater Dhaka. Most of the fringe areas of metropolitan Dhaka were characterized by moderate flood hazard zone. High hazard and very high hazards were assigned to cells where higher exposure to flooding has been observed. Lowland areas and cultivated land with scattered settlements in and around Dhaka were designated as high hazard zone. This is due to high hazard rank for their low elevation. Extreme lowlands and areas adjacent to river are designated as very high hazard zone. It is found that 19.20% of the study area constitutes very high hazard zone (Table 3).The final flood hazard map derived by using ranking matrix does not contain river networks. Therefore, river networks is overlaid in the final hazard map and shown in Fig. 1.
The relationships between flood hazard and elevation are shown in Table 4. It is observed that certain low lying areas are not highly susceptible to flooding, whereas areas with moderate elevation show higher susceptibility to flooding. It is also shown that a major portion of the study area corresponded to moderate to very high hazard zone. The elevation categories of 3, 4 and 5 with the elevation from 3 to 5 m corresponded to moderate hazard potential. This means that agricultural lowlands are the first to be inundated. In contrast, categories 6, 7 and 8 with the elevations of 6 to 8 m are at high risk areas, indicating that newly developed fringe areas around Dhaka will be inundated as soon as floodwater submerges the lowlands and agricultural lands. As the recent development of Dhaka is mainly concentrated on medium elevated lands, it is imperative to take initiatives to save life and property from imminent flood disasters.
| Fig. 1: | Flood hazard map evaluated from the interactive effect of
land cover and elevation by flood-affected frequency |
| Table 3: | Areas in Greater Dhaka with various degree of flood hazard |
| Table 4: | Flood hazard areas according to elevation heights |
| Table 5: | Flood hazard areas according to administrative divisions |
Overlaying land-cover on the derived flood hazard map provided invaluable information for managing potential flood risk in Greater Dhaka. It is demonstrated that 42.34% of the area of the intensely developed land-cover category is under moderate to high hazard zone. Since this category are extensively urbanized and highly populated, huge flood damage for infrastructure and buildings may occur. In contrast, 52% of the less intensely developed area is within the less to high hazard zone. This implies middle and low income groups of the City are to be affected by the potential flood hazard. More than 72% of the areas under floodplain category are at very high hazard zone. It is projected that, by the year 2010, most of the floodplains and lowlands of Greater Dhaka will be urbanized if the present pace of urbanization continues[17]. Therefore, it is essential to commence precautionary measures.
The hazard areas, according to administrative divisions, are presented in Table 5. The acquired area percentage within the different hazard zones under the different administrative units showed that in the protected part of Dhaka, i.e. Uttara, Pallabi, Cantonment, Mirpur and Mohammadpur, are exposed to flood hazard in varying degrees. In the non-protected part, Demra, Gulshan and Sabujbagh are highly susceptible to flood hazard.
CONCLUSIONS
In this study, the development of a flood hazard map for Greater Dhaka of Bangladesh was carried out by using the integrated techniques of remote sensing and GIS. Hazard assessment, on the basis of different thematic groups, was also performed.
RADARSAT SAR imageries with digital elevation data and land-cover categories are utilized to model flood hazard. Using a ranking matrix on a two dimensional multiplication mode, flood hazard map was developed and categorized into five distinct hazard zones. This shows the potential flood hazard areas in Greater Dhaka. It is imperative to note that only 1.90% of the study area found to be the least vulnerable to potential flood hazard while 19.20% area constitutes very high hazard potential. The derived hazard map was superimposed onto different GIS layers to assess flood hazard. The result shows that major part of Greater Dhaka is highly vulnerable to impending flood disasters.
The results described in this study gave an insight into potential flood hazard in Greater Dhaka and can be used to manage future flood events by the relevant agencies. It can be used in assigning priority for the development of high hazard areas. This study is useful for flood emergency management, including relief and rehabilitation programs. The most important advantage of the digital hazard map is that the developed map can be updated from time to time for effective management of flood disasters.