Flood Inundation Mapping from Satellite Images: a Case Study of Some Flood Prone Districts of Bangladesh
Total Page:16
File Type:pdf, Size:1020Kb
4thInternationalConference on Advances in Civil Engineering 2018 (ICACE 2018) 19 –21 December 2018 CUET, Chittagong, Bangladesh www.cuet.ac.bd FLOOD INUNDATION MAPPING FROM SATELLITE IMAGES: A CASE STUDY OF SOME FLOOD PRONE DISTRICTS OF BANGLADESH M. M. Rahman*, R. M. Tumon& R. Chakma Department of Urban & Regional Planning, Rajshahi University of Engineering & Technology, Rajshahi-6204, Bangladesh. E-mail: [email protected] *Corresponding Author ABSTRACT Bangladesh is one of the most flood prone countries in the world due to its geographical settings. Every year a major portion of land is inundated by flood. Efficient and timely monitoring of flood events is prerequisite for flood risk management as well as disaster response during and after flood. But near real time flood monitoring is quasi impossible without the use of Earth Observation (EO) data from space. Synthetic aperture radar (SAR) image from space offer opportunity for near real time flood monitoring regardless of time and weather. In this paper Sentinel-1 satellite images have been used in order to delineate flood inundated area. Firstly, the images have been calibrated through radiometric correction the image filtering has been done in 7X7 window using Speckle Filtering method. Finally, Binarization has been applied on filtered images in order to separate water from non-water area selecting a threshold value through iteration process and histogram analysis. The total work has been done in SNAP and ArcGIS software. This study suggests that real time automatic delineation of flood extent from satellite images can be the scope for further research in this field. Keywords: Flood Inundation, Earth Observation, Satellite Images, Synthetic aperture radar (SAR), Calibration, Sentinel Image. INTRODUCTION Bangladesh is one of the most flood prone countries in the world with about 21 percent of the country is subject to annual flooding and an additional 42 percent is at risk of flood with varied intensity (Ahmed and Mirza, 2000). Approximately 20% to 25% of its territory is inundated during the monsoon season (Siddiqui & Hossain, 2006). Two-thirds of Bangladesh is less than 5 meters above sea level, making it one of the most flood prone countries in the world. Severe flooding during a monsoon causes significant damage to crops and property, with severe adverse impacts on rural livelihoods (IFAD, 2012). A number of research works have identified that the monsoon flood scenario will be aggravated with future climate change context (WMO & GWP, 2003). Earth Observation (EO) data facilitate the mapping of flood extent over large areas. In this context, remotely sense data has the potential during the various phases of flood management process of providing an overview of the situation on the ground without direct contact with the flooded area that allows the decision makers to follow the water extent during the disaster (Sghaier et al., 2018).Numerous remote sensing techniques have been used to detect and delineate flooded area based on synthetic aperture radar (SAR) imagery (Liu et al., 2004). SAR delivers an all-weather, all-day tool for flood mapping at near real-time. In the study by Smith (1997), the capability to operate during daytime and night time and in almost all-weather conditions, Synthetic Aperture Radar (SAR) sensors 990 have emerged as one of the most important tools for providing reliable and near-real time information on flood disasters. The objective of this research is to determine the flood extents using multi-temporal satellite SAR data sets. In this paper Sentinel-1 satellite images have been used in order to delineate flood inundated area. Firstly, the images have been calibrated through radiometric correction the image filtering has been done in 7X7 window using Speckle Filtering method. Finally, Banalization has been applied on filtered images in order to separate water from non-water area selecting a threshold value through iteration process and histogram analysis. The methods are selected which are fast and easy to apply over large areas and images are inexpensive and easily available. Several image are analyzed from the year 2017. The change detection and thresholding was conducted to separate from flood area to non-flood. METHODOLOGY Data Used Sentinel-1A images of Sirajganj, Bogra and Tangail have been used in this study to delineate flood inundation area. The 10 Sentinel-1A images of different dates have been downloaded as a zip format from Copernicus website (https://scihub.copernicus.eu/dhus/#/home). The dates of images have been presented in the Table 1. Table 1: Downloaded Sentinel-1A images Sentinel-1 image Date S1A_IW_GRDH_1SDV 24/04/2017 S1A_IW_GRDH_1SDV 5/7/2017 S1A_IW_GRDH_1SDV 17/07/2017 S1A_IW_GRDH_1SDV 29/07/2017 S1A_IW_GRDH_1SDV 31/07/2017 S1A_IW_GRDH_1SDV 12/8/2017 S1A_IW_GRDH_1SDV 22/08/2017 S1A_IW_GRDH_1SDV 5/9/2017 S1A_IW_GRDH_1SDV 17/09/2017 S1A_IW_GRDH_1SDV 29/09/2017 Data preparation Sentinel-1 images are preserved in manifest.safe format. First, manifest.safe was selected from the image folder to view in the SNAP tool. Then folder was viewed on the left corner. There were two bands for each polarization recorded: Amplitude and Intensity. (The Intensity band is a virtual one. Which is the square of the amplitude). A subset helped creating a smaller portion of image which necessary. Pre-processing Firstly, image calibration has been done. The Radiometric Calibration provides imagery in which the pixel values can be directly related to the radar backscatter of the scene (ESA).This created a new product with calibrated values of the backscatter coefficient. The radiometric calibration is applied by the following equation which are acquired from ESA: value (i) = , where, depending on the selected Look Up Tables (LUT), value (i) = one of or original . = One of the betaNought (i), sigmaNought(i), gamma(i) or dn(i) Noise (i) = . The de-noise LUT must be calibrated matching the radiometric calibration LUT applied to the DN: Noise (i) = calibrated noise profile for one of one of or original . = One of the betaNought (i), sigmaNought (i), gamma (i) or dn(i) 991 The calibrated noise profile can then be applied to the data to remove the noise by subtraction. Application of the radiometric calibration LUT and the calibrated de-noise LUT can be applied in one step as follows: Value (i)= Secondly, Speckle filtering has been applied to remove the image noise. Speckle noise was considered as a common phenomenon in all coherent imaging systems like laser, acoustic and SAR imagery (Gan et al., 2012). Here Lee filter was used to reduce the speckle noise with window size 7 by 7. The equation is , L= , is the estimated pixel value (Horritt et al., 2001). Where x is an image pixel corrupted by a stationary multiplicative noise n such that y = nx (Gagnon and Jouan, n.d.). The after image from figure shows that the noise is reduced. Third, optimal thresholding has been applied to detect flooded area based in the histogram of the image and separate water areas and non-water areas (Ezzini et al, 2018). Histogram thresholding encompasses separating the image into several gray scale ranges based on peaks in the histogram (Townse & Foster 2002). The expression 255*(image name< threshold value) has been applied to separate water from non-water. Fourth, Radiometric Terrain Correction was applied using digital elevation model SRTM 3sec. for geometric correction. Finally, unsupervised classification was conducted to separate the whole area into two classes: Non-Water and Water. A map was prepared to visualize the flood condition of that particular date. RESULTS AND DISCUSSIONS Flood inundation areas for the month of July and August in the year 2017 have been illustrated by the Figure 2. A reference image during non-flood season is presented in Figure 1 to compare the flooded area. In both figures light yellow colour represent non-flooded area and blue for flooded area. In [Fig.2] it is clearly acknowledged that the blue portion is larger than the Reference image from [Fig.1]. Figure 2 shows the spatial temporal variation of flooded area during July, August and September of 2017 where Figure 3 and Figure 4 show the temporal variation. From the Figure 4 it is seen that highest flood is observed on 17 July 2017 whereas minimum inundated area is observed on 29 September 2017. Actually in the context of Bangladesh July month is considered as Monsoon season. In July average rainfall is too high in comparing with other months. Fig. 1: Water and Non-Water Area in non-flooded season. 992 Fig. 2: Flood inundated area in different dates 993 Fig. 3: Percentage of flooded and non-flooded area in the month of July and August 2017. Source: calculated by Author from satellite Images, 2018 Fig. 4: Temporal variation of inundated area in the month of July and August 2017 Source: calculated by Author from satellite Images, 2018 Table 2: Percentages of Flooded area in three districts Bogra Sirajganj Tangail Date Flooded Area % Flooded Area % Flooded Area % 24/4/2017 191.60 5.95 337.31 12.57 160.99 4.23 5/7/2017 482.31 14.98 726.28 27.06 377.90 9.93 17/7/2017 484.32 15.04 898.25 33.47 615.47 16.18 29/7/2017 521.75 16.20 584.22 21.76 381.57 10.03 31/7/2017 522.18 16.22 609.24 22.70 358.68 9.43 12/8/2017 243.83 7.57 581.32 21.66 284.06 7.47 22/8/2017 295.75 9.19 593.27 22.10 365.71 9.61 5/9/2017 353.36 10.97 736.10 27.43 402.37 10.58 17/9/2017 295.75 9.19 593.27 22.10 365.71 9.61 29/09/2017 243.83 7.57 581.32 21.66 284.06 7.47 Source: calculated by Author from satellite Images, 2018 994 Table 2 shows the percentage of flooded area by Districts and Dates.