Assessment of Cartosat-1 DEM for Modeling Floods in Data Scarce Regions

Abstract

Digital Elevation Model (DEM) plays an important role in modeling floods. In data scarce developing countries, unavailability of high resolution topography and river cross-sections data are the prime limitations for simulating hydrodynamic models for modeling floods. In the present study, we assess the quality of Cartosat-1 DEMs in providing accurate river cross-sections and floodplain elevations; and hence their suitability in modeling floods. Cartosat-1 DEMs are prepared using Ground Control Points (GCP) of surveyed elevation and bias corrected Shuttle Radar Topography Mission (SRTM) elevation. Surveyed elevation based Cartosat-1 DEM is found to be of best quality while bias corrected SRTM elevation based Cartosat-1 DEM is found to be of reasonable quality on the basis of cross-section representation as well as elevation statistics. Cross-sections derived from the Cartosat-1 DEMs as well as surveyed cross-sections are later used independently in MIKE11 model for 1-dimensional flow modeling. Simulated water levels from models based on Cartosat-1 DEMs are compared graphically with the observed water levels. Modeling performance is also evaluated using different statistical performance criteria. Results show that the models based on Cartosat-1 DEM derived cross-sections perform similar to the model based on surveyed cross-sections. Study concludes that a reasonably accurate DEM, prepared from moderate survey in data scarce region, can be used for deriving requisite river cross-sections for hydrodynamic modeling.

Full text

Translate

Turn on search term navigation

Water Resour Manage (2016) 30:12931309 DOI 10.1007/s11269-016-1226-9

Prachi Pratyasha Jena1 & Banamali Panigrahi2 & Chandranath Chatterjee3

Received: 4 July 2015 /Accepted: 4 January 2016 / Published online: 10 January 2016# Springer Science+Business Media Dordrecht 2016

Abstract Digital Elevation Model (DEM) plays an important role in modeling floods. In data scarce developing countries, unavailability of high resolution topography and river cross-sections data are the prime limitations for simulating hydrodynamic models for modeling floods. In the present study, we assess the quality of Cartosat-1 DEMs in providing accurate river cross-sections and floodplain elevations; and hence their suitability in modeling floods. Cartosat-1 DEMs are prepared using Ground Control Points (GCP) of surveyed elevation and bias corrected Shuttle Radar Topography Mission (SRTM) elevation. Surveyed elevation based Cartosat-1 DEM is found to be of best quality while bias corrected SRTM elevation based Cartosat-1 DEM is found to be of reasonable quality on the basis of cross-section representation as well as elevation statistics. Cross-sections derived from the Cartosat-1 DEMs as well as surveyed cross-sections are later used independently in MIKE11 model for 1-dimensional flow modeling. Simulated water levels from models based on Cartosat-1 DEMs are compared graphically with the observed water levels. Modeling performance is also evaluated using different statistical performance criteria. Results show that the models based on Cartosat-1 DEM derived cross-sections perform similar to the model based on surveyed cross-sections. Study concludes that a reasonably accurate DEM, prepared from moderate survey in data scarce region, can be used for deriving requisite river cross-sections for hydrodynamic modeling.

* Prachi Pratyasha Jena [email protected]

1 Department of Agricultural and Food Engineering, Indian Institute of Technology Kharagpur, West

Bengal 721 302, India

2 GVK informatics Pvt. Ltd, Hyderabad 500039, India

3 Department of Agricultural and Food Engineering, Indian Institute of Technology, Kharagpur, India 721302

http://crossmark.crossref.org/dialog/?doi=10.1007/s11269-016-1226-9&domain=pdf

Web End = http://crossmark.crossref.org/dialog/?doi=10.1007/s11269-016-1226-9&domain=pdf

Web End = Assessment of Cartosat-1 DEM for Modeling Floods in Data Scarce Regions

1294 P.P. Jena et al.

Keywords Digitalelevationmodel.Verticalbias.Cartosat-1DEM.Floodmodeling.MIKE-11 model . SRTM DEM

1 Introduction

Accurate representation of river channels and floodplain are essential for river flood modeling. In developing countries, unavailability of high resolution topography and river cross-section data is the prime limitation for simulating hydrodynamic models. Digital Elevation Model (DEM) is a great source to acquire topographical information. It has been widely used in hydrological modeling for drainage network extraction and watershed delineation for decades (Fairfield and Leymarie 1991; Vogt et al. 2003; Sahoo et al. 2006; Metz et al. 2011; Mani et al. 2014; Sharma et al. 2014). In recent years, globally available DEMs like Shuttle Radar Topography Mission (SRTM) and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) with suitable modifications (Patro et al. 2009a, b; Gichamo et al. 2012; Yan et al. 2015) and advanced LIght Detection And Ranging (LiDAR) based DEMs (Chatterjee et al. 2008; Forster et al. 2008; Merwade et al. 2008a; Teng et al. 2015) have been used to extract terrain data for flood modeling. Uncertainties in terrain data bring about uncertainties in hydrodynamic model output in many different ways, such as discharge from hydrological model, water surface elevation and inundation extent (Merwade et al. 2008b; Chatterjee et al. 2008). Therefore, terrain analysis needs meticulous attention and effort to get accurate topographic information. Resolution of DEM has a significant effect on understanding different earth features (Tarolli 2014), especially, extracting cross-section of narrow-width rivers where coarse resolution DEMs fail to represent the accurate geometry.

Different topographic sources have significant implications on the outputs of hydraulic and hydrodynamic models (Ali et al. 2014). Hydrodynamic model produces highly accurate outputs when input topographic data are obtained using ground survey techniques, such as using Global Positioning System (GPS), Differential Global Positioning System (DGPS) and total station method (Casas et al. 2006). Due to time and budget constraints, surveying is not always a feasible task for large scale applications. The expensive LiDAR based DEMs are the best sources of terrain data for flood modeling, channel network extraction and channel bed morphology analysis (Liu 2008; Cook and Merwade 2009; Tarolli 2014). However, use of globally available DEMs is also a great advantage to hydrodynamic modeling (Wilson et al. 2007; Tarekegn et al. 2010; Yamazaki et al. 2012; Gichamo et al. 2012). In particular, large scale studies in data scarce regions are very much dependent on these freely available DEMs (Paiva et al. 2011; Yan et al. 2015). Many studies have integrated Geographical Information System (GIS) with globally available DEMs or satellite DEMs to get better topographic features such as river channel, cross section and floodplain geometry, and floodplain extent (Merwade et al. 2005; Sanders 2007; Paiva et al. 2011). Synthetic aperture radar based interferometer technology does not capture the river bed elevation below the water surface (Farr et al. 2007). As a result, SRTM data overestimates the channel bed elevation, which contributes to uncertainties in hydrodynamic model output. Researchers have dealt with this issue using different approaches of bias correction (Patro et al. 2009a, b; Pramanik et al. 2010; Gichamo et al. 2012; Yan et al. 2015). Yan et al. (2015) corrected SRTM elevation of river bed using a free model parameter which is obtained from calibration of the best fit model. Gichamo et al. (2012) have extracted triangular cross-sections from ASTER DEM and corrected the vertical bias to improve the cross section geometry.

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1295

Flood is a very frequent phenomenon in many parts of India. For example, two recent devastating floods have occurred in Uttarakhand during June 2013 (Times of India, retrieved on 18th June 2013) and in Jammu and Kashmir during September 2014 (Hindustan Times, retrieved on 8th September 2014). Huge loss of economy and life is confronted due to the floods. An accurate flood hazard map is one of the non-structural measures for minimising loss of life and property. In India, cross-section data are generally available at the gauging sites only (Central Water Commission 1996). Sometimes these gauging sites are more than 50 km apart with widely varying cross-sections in between. Use of such widely spaced cross-sections in flood inundation model is expected to produce inaccurate results. In addition, existing topographic maps in India have sparse coverage of data and poor vertical resolution in low lying areas. Researchers have used vertical bias corrected SRTM DEM data to deal with the lack of topographic information in Mahanadi river delta (Patro et al. 2009a, b; Samantaray et al. 2015) and Brahmani river basin (Pramanik et al. 2010) in India. However, coarser resolution SRTM DEM (90 m 90 m) may not produce accurate elevations in small width rivers. So, a high or medium resolution DEM may be a substitute for filling the cross-section gaps in the river channel. A good quality DEM and accurate river cross-sections are very essential for reliable flood inundation modeling and thus development of flood hazard maps. In this study, we have addressed the issues of lack of cross-sections and good quality DEM for flood inundation modeling by preparing a DEM from high resolution Cartosat-1 stereo images for a part of the low lying deltaic region of Mahanadi river basin in India.

Cartosat-1 stereo image is the first Indian Remote sensing Satellite (IRS) image mainly developed for topographic mapping and digital terrain model generation. Two stereo images with resolution of 2.5 m (for aft) and 3 m (for fore) make it possible to acquire high resolution topographic information. Applications of Cartosat-1 DEM have not been explored widely, and are not well documented in literature. Few researchers have evaluated the Cartosat-1 data products for terrain modelling and large scale mapping applications (Pieczonka et al. 2011; Bothale and Pandey 2013; Giribabu et al. 2013). Mostly, studies on DEM from Cartosat-1 stereo images have been performed for hilly areas, where quite good agreement in terrain data was observed with low RMSE (Ahmed et al. 2007; Evans et al. 2008; Gianinetto 2009). Comparisons of SRTM DEM and Cartosat-1 DEM have been documented by researchers such as Rawat et al. (2012) and Yarrakula et al. (2013). Yarrakula et al. (2013) reveal that Cartosat-1 DEM produces more accurate river profile elevation as compared to SRTM DEM.

A pair of high resolution stereo images would generate better quality DEM when good ground control points (GCPs) are used. Generally, GCPs are selected at those locations which are well identified in the stereo images. Information on GCPs can be collected through ground survey or benchmarks in topomaps. Cost prohibitive ground survey and lack of benchmarks in topomaps is a real challenge for preparing DEM from Cartosat-1 stereo image. Here, we have attempted to develop Cartosat-1 DEMs for data scarce conditions and assess its suitability in 1-D flood modeling. The present study is carried out in two phases. In the first phase of the study, we developed Cartosat-1 DEMs using GCPs from (i) surveyed ground truth elevation, and (ii) bias corrected global DEM elevation. For applying bias correction to a global source of elevation, we assessed two global DEMs, i.e., SRTM and ASTER with respect to ground surveyed elevation. The elevations as well as river cross-sections obtained from Cartosat-1 DEMs were then compared with ground survey data. In the second phase of the study, one dimensional river flow modeling was carried out using MIKE11 model setup of the surveyed cross-sections and the derived cross-sections from the generated Cartosat-1 DEM. The performance of the model setups were studied for two cross-section sets.

1296 P.P. Jena et al.

2 Study Area and Data Used

The study area is located in the lower delta region of Mahanadi river basin in India (Fig.1a). The Mahanadi river basin is one of the major interstate river basins in eastern India. Delta region of the Mahanadi river lies between longitudes of 85o 30 to 86o 52E and latitudes of 19o 40 to 20o 45N. More than 80 % of total cropped area in the delta region is affected by floods during the monsoon period. The area is covered by plain alluvial track and has average surface elevation of 1030 m. The climate is tropical with an average annual rainfall of 1400 mm. In the last decade, five high floods of magnitude more than 35,000 m3/s (a 10 year return period flood) at the head of delta region have occurred, i.e., in years 2001 (39,887 m3/s), 2003 (38,223 m3/s), 2006 (36,344 m3/s), 2008 (43,795 m3/s) and 2011 (38,832 m3/s). These floods have caused significant damages to life, crops, and property. The floods of 2008 and 2011 caused a financial loss of more than US $ 400 million (Flood Workshop 2011). The increasing trend of extreme rainfall in the upstream catchment of the delta region is supposed to be responsible for these frequent high floods in the last decade (Jena et al. 2014). In addition, heavy downpour from mid-July to mid-September in the delta region and flow release from the Hirakud Dam located in the upper catchment also contributes to flooding. Because of complicated river distributaries in the region, it is very cumbersome and expensive to carry out survey throughout the rivers in entire delta region. Therefore, the study area has been selected as a small part of the delta region, which includes rivers like Kushabhadra, Kuakhai, Bhargavi and Daya (Fig. 1a) covering an area of 960 km2. Approximate width of rivers in the study area ranges between 300 m-700 m. In these rivers, escapes (weir structure at embankment of the river) are provided at different locations to prevent failure of embankments during high floods. These are the locations where usually river flow overtops the embankments and causes floods in adjacent areas. The escapes are named as Achyutpur in river Bhargavi, Jogisahi in river Kushabhadra, and Kanti and Madhipur in river Daya, as shown later in Fig. 5.

Two types of data are used in this study. First is topographic data for DEM generation and second is hydrometry data for calibration and validation of the 1-D hydrodynamic model. Topographic data include (i) SRTM DEM (90 m 90 m) (Source: http://srtm.csi.cgiar.org

Web End =http://srtm.csi.cgiar.org .) and ASTER DEM (30 m 30 m) (Source: https://earthexplorer.usgs.gov

Web End =https://earthexplorer.usgs.gov .), (ii) surveyed cross-section elevations in different river reaches, (iii) surveyed DGPS elevations in floodplains, and (iv) three IRS Cartosat-1 stereo pair images (Fig. 1b) which have been procured from National Remote Sensing Center (NRSC), Hyderabad, India. Details of images are: (1) Path-579, Row- 303, date of pass-16 March 2008, (2) Path- 580, Row-303, date of pass- 26 December 2008, (3) Path-581, Row-304, date of pass- 23 October 2009. A total of 90 cross-sections are measured through ground survey, i.e., 39 in river Bhargavi, 45 in river Kushabhadra, and 6 in river Daya (Fig. 1b). These 90 cross-sections comprise 1213 elevation points in all which were measured using the total station method and GPS with very negligible error in vertical and horizontal traverse. Similarly, elevations of 122 locations were surveyed in floodplain (Fig. 1b) using Trimble-R3 DGPS system. Hydrometry data include monsoon (June to September) daily discharges at Balianta, which is located at the upstream of the river reach, and water level of four gauging stations located on the three river reaches (Fig. 1a). Daily discharge data at Balianta station was collected from the State Water Resources Department for the years 2001 to 2006. Daily water levels at the Nimapara gauging station was collected from the Central Water Commission, Bhubaneswar for the years 2001 to 2006. Available water levels of the gauging sites at escapes,i.e., Jogisahi, Achyutpur, Madhipur, and Kanti were collected from the sub-divisional irrigation offices in the region. Water level data at Achyutpur was available for the flood years 2001, 2003 and 2006; at Kanti for the years 2001 and 2003; and at Jogisahi and Madhipur for the year 2003.

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1297

Fig. 1 Location of study area a in Mahanadi Delta with the river network and gauge stations, b Cartosat-1 Image with surveyed cross sections (Black cross marks) and DGPS survey points (Red circles)

3 DEM Analysis

3.1 Comparison of Freely Available Global DEMs

SRTM and ASTER DEMs are widely used freely available online sources of global elevations.

These are primary substitutes for topographic data in data scarce regions. Prior to DEM generation from Cartosat-1 image, we compared the accuracy of these freely available elevation data with respect to ground truth surveyed elevations. Elevation errors of SRTM DEM and ASTER DEM at 122 locations (DGPS survey locations) in the floodplain are shown

1298 P.P. Jena et al.

in Fig. 2. SRTM DEM shows consistent positive error; whereas ASTER shows an inconsistent error pattern, i.e., both positive and negative errors occur at different locations, with respect to surveyed points. Accuracy of SRTM and ASTER DEMs were also analysed based on the statistical parameters, such as minimum, maximum, mean, standard deviation (SD), standard error of mean (SEM), and root mean squared error (RMSE). Table 1 presents the statistical parameter values of three sources of topographic information. Mean values for both SRTM and ASTER DEMs were found nearly same; but SD, SEM and RMSE are lower for SRTM DEM as compared to ASTER DEM. Both the DEMs were found to be inferior products for the region. However, with certain modifications SRTM DEM has been used for flood modeling in literatures such as Patro et al. (2009a, b), Pramanik et al. (2010) and Samantaray et al. (2015). So, we have also used the bias corrected SRTM DEM elevation for generating Cartosat-1 DEM in the next section. Moreover, preference of SRTM DEM over ASTER DEM in hydrological and hydraulic modeling has also been reported in literatures (Forkuor and Maathuis 2012; Ali et al. 2014; Sharma et al. 2014).

3.2 DEM Generation from Cartosat-1 Stereo Image3.2.1 Ground Control Points

Selection of ground control points (GCPs) is an important prerequisite for extracting digital terrain model from stereo image pairs. We have selected 42 points (from 122 surveyed DGPS elevations) in the floodplain which are well identified locations such as intersection of roads, fields and small water bodies. In addition, we have also selected few points from 30 cross-sections (out of 90 surveyed cross-section elevations) which are at identifiable locations such as bends and bifurcations in the river reaches. Three points are selected from each of the 30 locations (one each from left and right embankment and one from river bed), which sum-up to 90 points. In all, 132 (42 + 90) locations are selected from surveyed data for GCPs. Two sets of GCPs were prepared to depict two types of data scarce conditions: (i) availability of moderate survey data where few elevation points are required in floodplain as well as river, and (ii) availability of limited survey data where few elevation points are required in floodplain only. Locations of the GCPs are same for the two cases. In the first case, surveyed elevations were taken as elevations of the selected 132 GCPs. In the second case, bias corrected elevations of

Fig. 2 Comparison of elevation error of SRTM and ASTER with reference to surveyed elevation (Elevation error = SRTM/ASTER elevation-Surveyed elevation)

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1299

Table 1 Statistics of elevations of

DEMs and DGPS data

a: Standard Error of Mean

b: Root Mean Square Error

SRTM DEM were taken as elevations at the same 132 locations. Bias correction was done by a simple approach where original elevations were reduced by the RMSE value. A similar bias correction method was applied by Patro et al. (2009a), where they compared SRTM elevations with spot heights from topomaps. In this study, the elevation of 42 points in floodplain and 90 points in river were reduced by 4.53 m. This is the RMSE value found with respect to surveyed elevation as mentioned in Table 1. In this way, bias corrected SRTM DEM elevations were used as elevations of 132 GCPs. The idea of using this set of GCPs is to generate Cartosat-1 DEM with limited survey data. Subsequently, two Cartosat-1 DEMs were prepared using the above-mentioned two different sets of GCPs. Two different cases for GCP selection were applied to tackle issues of data scarcity such as lack of extensive ground survey data and inadequate elevation information in topomaps. For example, in the first case, few surveyed elevations from floodplain and river were used. In the second case, limited surveyed data were used to determine the RMSE of SRTM DEM; and then bias corrected SRTM elevations were taken as elevation of ground control points.

3.2.2 Terrain Extraction

In order to obtain fine resolution DEM, we used Indian Remote sensing Satellite (IRS)

Cartosat-1 stereo pair image data. Ground control points were taken in two different ways (as stated in section 3.2.1) for extracting terrain data from the stereo image. Terrain data were extracted from Cartosat-1 stereo pair images using Leica Photogrammetry Suite (LPS) software version 9.1. A total of 132 GCPs were used to refine the block file in LPS which was initially refined using rational polynomial coefficients (RPC). After the block adjustment in LPS, tie points were generated by automatic areal triangulation and verified by visual inspection. The aerial triangulation method was used for processing stereo image pair to reconstruct a complete surface in terms of geometry and radiometry (Shenk 1996). Using tie points and GCPs, DEM was extracted from the stereo pair image. DEM generated from the GCP of surveyed elevation and bias corrected SRTM elevations are named as Survey based Cartosat-1 DEM (hereafter referred as Survey-Carto) and Reduced SRTM based Cartosat-1 (hereafter referred as Reduced-SRTM-Carto), respectively. Cartosat-1 DEMs were developed with resolution of 30 m 30 m. These DEMs were later validated against 1123 available river profile elevation points (excluding the 90 GCPs used for DEM generation from the total number of 1231 surveyed cross-section elevation points) and 80 floodplain elevation points (excluding 42 GCPs used for DEM generation from the total number of 132 surveyed floodplain elevation points). Subsequently, quality of these DEMs was assessed.

Statistical Parameters DGPS ASTER SRTM

Locations (numbers) 122 122 122

Minimum (m) 03.63 05.00 07.00

Maximum (m) 16.60 37.00 20.00

Mean (m) 09.17 13.64 13.68

Standard Deviation (m) 03.13 05.74 02.94

SEM (m)a 00.28 00.52 00.26

RMSE (m)b - 05.58 04.53

1300 P.P. Jena et al.

3.3 Assessment of DEM Quality

The DEMs were compared against the reference elevation using two approaches. Firstly, the elevation differences and accuracies of elevations at different locations in the DEM are assessed with respect to surveyed elevations; and secondly, the cross-section profiles of different DEMs are compared with the surveyed cross-section profiles. Elevations of two generated Cartosat-1 DEMs are compared with the surveyed elevations as shown in scatter plot diagrams for both floodplain (Fig. 3a) and river (Fig. 3b). Fig. 3a shows the comparison of elevations at 80 points (12242) in floodplain which are not used for DEM generation. While Fig. 3b shows the comparison of elevations at 1123 points (123190) which are not used for

Fig. 3 Scatter plots of elevations derived from two Cartosat-1 DEMs in a floodplain and b river

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1301

DEM generation. Both Cartosat-1 DEM points show similar pattern of scatter on the 1:1 line (Fig. 3a, b). Statistics of elevation difference of the DEM points are compared for both floodplain and river (Table 2). Statistical parameters such as maximum, mean, and standard deviation of the elevation differences for Survey-Carto DEM are lower than that of Reduced-SRTM-Carto DEM in both floodplain and river. In addition, comparative statistics, i.e., RMSE and MAE (Mean Absolute Error) for the Survey-Carto are lower than that of Reduced-SRTMCarto DEM in floodplain and river (Table 2). Higher error of Reduced-SRTM-Carto DEM may be due to use of RMSE based bias corrected SRTM elevations as GCPs. So, the Survey-Carto DEM prepared from moderate survey data appears to be best, but Reduced-SRTM-DEM prepared from limited survey is also found to be reasonably accurate and hence, may be adapted for data scarce condition.

In the second approach of assessment, cross-section profiles from generated DEMs were compared with reference to surveyed cross-sections. The river cross-section profiles at upstream and downstream sections of the two river reaches, viz., Bhargavi and Kushabhadra, are presented in Fig. 4. Cross-sections of Bhargavi river are shown in Fig. 4a and Kushabhadra river in Fig. 4b. Cross-sections derived from Survey-Carto DEM closely adhere to the actual cross-section profiles. Cross-sections derived from Reduced-SRTM-Carto DEM have deviations from actual cross-sections. However, at few locations such as upstream and downstream of Kushabhadra, pattern of cross-section profile and bed level agrees with the actual cross-sections (Fig. 4). In case of availability of limited survey data, the Reduced-SRTM-Carto may be used for extracting river cross-sections.

Both the statistical and graphical comparison revealed that Survey-Carto DEM is the best one for the data scarce region where a moderate survey is required. To assess the suitability of Cartosat-1 DEM in flood modeling, we used the cross-sections from the Survey-Carto DEM in MIKE11 HD model.

Table 2 Statistics of elevation differences and measures of accuracy for two Cartosat-1 DEMs with respect to surveyed elevation in floodplain and river

Location Statistical Parameters Survey-Carto Reduced-SRTM-Carto

Floodplain elevations DGPS points 80 80

Minimum (m) 00.01 00.01

Maximum (m) 03.41 06.35

SD (m) 01.02 01.58

Mean (m) 01.31 02.33

Range (m) 03.40 06.34

RMSE (m) 1.65 2.94

MAE (m) 1.04 2.05

River elevations Surveyed points 1123 1123

Minimum (m) 00.00 00.02

Maximum (m) 11.96 14.60

SD (m) 02.05 02.36

Mean (m) 02.41 02.97

Range (m) 11.91 14.57

RMSE (m) 3.17 3.80

MAE (m) 2.41 2.97

1302 P.P. Jena et al.

Fig. 4 Cross-section profiles in Bhargavi river a and Kushabhadra river b at upstream and downstream locations

4 Hydrodynamic Modeling

4.1 Description of 1-D Model Setup

MIKE-11 HD model was used to simulate water level in the river system. MIKE11 model solves fully dynamic flow equation (Saint-Venant equation) using finite difference mathematical technique (DHI 2014). Numerical solution of the flow equations are determined using implicit finite difference scheme known as Six point Abbot scheme developed by Abbott and Ionescu (1967).

Model inputs are river network, cross-section geometry, hydrodynamic parameter and hydrodynamic boundary condition. Channel slope is taken by the model through the horizontal chainage of the river network and minimum elevation of the cross-section profile for each kinematic routing element (DHI 2014). The network file was prepared from a topographic base map. The river network of three river branches, i.e., Bhargavi, Kushabhadra, and Daya with the five gauging sites, i.e., Nimapara, Achyutpur, Jogisahi, Madhipur and Kanti are shown in Fig. 5. Ninety cross-sections from Survey-Carto DEM (developed for the condition of moderate data availability) were used in one model setup. Similarly, another model setup was prepared using 90 cross-sections from detailed survey data. The model was set up with five open hydrodynamic boundaries in the river branches. One inflow boundary was given at the upstream gauging site, Balianta (shown in Fig. 5) and four constant water level boundaries were given at the downstream end of the rivers.

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1303

Fig. 5 MIKE11 river network with locations of gauging sites and escapes used in calibration and validation of the model

4.2 Calibration and Validation of MIKE-11 Model

Two MIKE11 model setups of two cross-section sets were calibrated independently.

Mannings roughness coefficient, n was taken as the only calibrating parameter in the model. With recommended n values in literature (Chow 1959) for rivers, the initial value of n was taken as 0.02. One global n value was taken for entire three river reaches. For each model setup, n value was adjusted and fixed such that model outputs would closely agree with the observed values at Nimapara and four escape locations. Global n values were calibrated and found to be 0.028 for the model based on surveyed cross-sections and 0.027 for the model based on cross-sections of Cartosat-1 DEM. Daily water level data from 1st Aug 2003 to 7th Oct 2003 was used for calibrating the two model setups. Simulated water levels from each model setup were compared with the observed water levels at five gauging sites, which are discussed in section 4.3. After calibration, models were validated with the available observed water level data for the years 2001, 2002, 2004, 2005 and 2006. Water level data at Nimapara are available for all the 5 years; whereas water level data at Achyutpur are available for the years 2001 and 2006, at Kanti for the year 2001, and at Jogisahi for the year 2006. The calibration and validation performances of the model were measured through different performance indices, such as Nash-Sutcliffe Efficiency (NSE), coefficient of determination (R2), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). These criteria of

1304 P.P. Jena et al.

goodness of fit have been invariably used by many researchers in hydraulic and hydrodynamic modeling (Chatterjee et al. 2008; Chavarri et al. 2013; Domeneghetti et al. 2014).

4.3 Assessment of Cartosat-1 DEM from MIKE11 Model PerformanceModel performances were compared on the basis of graphical and statistical performance measures. The graphical evaluation includes comparison of model simulated water levels with the observed water levels at different gauging sites. Simulated and observed water levels were compared at five gauging sites, Nimapara, Achyutpur, Kanti, Madhipur and Jogisahi during calibration. A comparison of water levels at two of these gauging sites are shown in Fig 6a. Model based on Cartosat-1 DEM derived cross-sections simulate marginally higher water level than the observed water level at Nimapara and Achyutpur (Fig. 6a). Over-estimation of water level in Cartosat-1 DEM based model may be explained through the higher elevation of the cross-section bed level extracted from the Cartosat-1 DEM. Besides the graphical comparison, performance indices of the two model setups during calibration were also evaluated (Table 3). The model setup based on surveyed cross-section has performed best among the two in terms of higher NSE (ranging from 81.79 % to 91.73 %) and R2 (ranging from 0.88 to 0.96) and

Fig. 6 Comparison of observed and simulated water levels at a Nimapara and Achyutpur during calibration for the year 2003 and b Nimapara for 2004 and Achyutpur for 2001during validation

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1305

Table 3 Performance indices of model setups with cross-sections from survey measurement and Cartosat-1 DEM during calibration at five gauging sites

Model setups based on cross sections obtained from

lower RMSE (ranging from 0.35 m to 0.62 m) and MAE (ranging from 0.24 m to 0.52 m) for all five gauging sites. The performance of the model setup based on cross-sections from Survey-Carto DEM is also found comparable with the performance of the first model setup, reporting NSEs of more than 79 % at all the locations (ranging from 79.01 % to 87.94 %). Other indices such as R2, RMSE and MAE are also found to be closer to that of the first model setup.

During validation, simulated water levels were compared with the observed water levels at three gauging sites for different years. A comparison for two of these gauging sites, i.e., Nimapara and Achyutpur are shown in Fig. 6b. Peaks of the water level hydrographs are captured well by the model based on cross-sections from the Survey-Carto DEM as well as surveyed cross-sections (Fig. 6b). The performance indices of the two setups were evaluated during validation (Table 4). For the first model setup, performance indices such as NSE, R2,

RMSE and MAE were found within the range of 71.10 % to 90.19 %, 0.72 to 0.94, 0.37 m to0.87 m and 0.28 m to 0.74, respectively. We observed that the performance of each model during validation (Table 4) is close to that of the model during calibration (Table 3), thus highlighting robustness of the model. In the model setup based on Survey-Carto DEM, NSE, R2, RMSE and MAE were found within the range of 60.5 % to 79.54 %, 0.68 to 0.90, 0.40 m

Table 4 Performance indices of model setups with cross-sections from survey measurement and Cartosat-1 DEM during validation at three gauging sites

Cross sections from Index Nimapara Achyutpur Kanti

2001 2002 2004 2005 2006 2001 2006 2001

Survey NSE (%) 84.56 86.60 90.19 86.9 83.02 89.75 71.11 74.12

R2 0.91 0.88 0.91 0.89 0.94 0.92 0.72 0.81

RMSE (m) 0.87 0.49 0.56 0.62 0.87 0.37 0.44 0.49

MAE (m) 0.74 0.34 0.42 0.48 0.74 0.28 0.34 0.36

Survey-Carto DEM NSE (%) 79.54 73.25 72.56 71.03 72.81 87.79 60.53 69.50

R2 0.81 0.75 0.78 0.72 0.89 0.90 0.68 0.76

RMSE (m) 1.00 0.70 0.93 0.93 1.10 0.40 0.51 0.53 MAE (m) 0.88 0.46 0.78 0.69 0.97 0.33 0.39 0.40

Index Nimapara Achyutpur Madhipur Kanti Jogisahi

2003 2003 2003 2003 2003

Survey NSE (%) 91.73 88.65 81.79 88.04 89.02

R2 0.96 0.92 0.88 0.90 0.93

RMSE (m) 0.62 0.46 0.48 0.35 0.44

MAE (m) 0.52 0.35 0.37 0.24 0.34

Survey-Carto DEM NSE (%) 83.81 85.44 79.01 81.77 87.94

R2 0.86 0.90 0.83 0.87 0.98

RMSE (m) 0.86 0.53 0.51 0.43 0.46

MAE (m) 0.75 0.42 0.39 0.31 0.41

1306 P.P. Jena et al.

to 1.10 m and 0.33 m to 0.97 m, respectively for the three gauging sites (Table 4). Although the performance of model based on cross-sections from Survey-Carto DEM is lower than that of the model based on surveyed cross-sections; NSE > 60 % indicates fair performance of the model during validation (according to Moriasi et al. (2007), NSE > 50 % indicates satisfactory performance of a hydrological model). Thus, in the present study on the assessment of Cartosat-1 DEMs, Survey-Carto DEM was found to have a good potential for deriving river cross-sections in data scarce condition for use in one-dimensional flood modeling.

We evaluated the performance of a model setup based on only limited surveyed cross-sections. As stated in section 3.2.2, 90 elevation points from the 30 cross-sections were used to generate the Survey-Carto DEM. These 30 cross-sections of the entire river reach constitute what is termed as limited cross-sections. These cross-sections were used to develop a model setup. This analysis has been carried out to understand how the model would perform if one wants to carry out flow modeling based on only these 30 cross-sections without opting for other alternatives like Cartosat-1 DEM derived cross-sections. The model setup was calibrated and validated at Nimapara gauging site. NSEs were obtained as 56.23 % and 44.48 % for calibration (for the year 2003) and validation (for the year 2001), respectively. The NSE values are remarkably lower than the NSEs of the two earlier model setups, i.e., Survey and Survey-Carto (presented in Tables 3 and 4), during calibration and validation at Nimapara gauging station. This indicates unsatisfactory performance of the model based on 30 cross-sections in comparison to models based on 90 cross-sections. Better performance of these two model setups are explained through the use of more number of cross-sections which is required for accurate representation of the river geometry. It may be noted that higher number of cross-sections were derived from the Cartosat-1 DEM which was generated using GCPs from the 30 cross-section locations and few floodplain locations. Thus, comparison of model performances provides an understanding that limited ground survey is advantageous in generating the Cartosat-1 DEM which can be further used for extracting required number of cross-sections. Although it is anticipated that using more number of cross-sections (i.e., less spacing between cross-sections) increases model performance in simulating 1-dimensional river flow, increasing number of cross-sections beyond an optimal cross-section set does not enhance the model performance (Castellarin et al. 2009). However, hydraulic models are very sensitive to the geometric description of the channel such as number of cross-sections and spacing between cross-sections (Merwade et al. 2008b). Rivers with bifurcations and bends require more number of cross-sections to retain its geometry. In this context, instead of opting for the extensive ground survey, a reasonably accurate DEM prepared using Cartosat-1 stereo pairs, limited surveyed cross-section and limited floodplain elevation data can be used for deriving the requisite number of cross-sections.

5 Conclusion

Cartosat-1 DEMs were developed for two conditions, i.e., availability of moderate survey data in floodplain and river; and limited survey data in floodplain only. These two cases were identified for addressing the problem of data scarcity in the region. A comparison of elevation error of SRTM and ASTER DEM revealed that SRTM DEM is the better global DEM with consistent vertical bias. Few surveyed elevations and modified SRTM elevations were used as elevation of GCP in preparing Cartosat-1 DEMs. Cross-sections of Cartosat-1 DEM developed for a data scarce condition (where moderate survey is

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1307

required) were found to be in close agreement with the surveyed cross sections. Use of these cross-sections in MIKE11 model resulted in satisfactory simulation of water level in the river. In addition, poor performance of model based on limited cross-sections revealed that Cartosat-1 DEM developed from moderate surveyed data is advantageous for developing a robust model in data scarce conditions. However, future studies may be taken up to analyse the minimum number of GCPs required for preparing a good quality Cartosat-1 DEM. Also, as flood inundation modeling highly depends on the topography of flood-plain, further studies on two-dimensional flood inundation modeling are necessary to validate the DEM prepared from Cartosat-1 data.

Acknowledgments This study is a part of the project titled BFlood inundation zoning for different return periods in Mahanadi River basin^ sponsored by Indian National Committee on Surface Water (INCSW), Ministry of Water Resources, Govt. of India.

Abbott MB, Ionescu (1967) On the numerical computation of nearly horizontal flows. J Hydraul Res 5:

97117

Ahmed N, Mahtab A, Agrawal R, Jayaprasad P, Pathan SK, Ajai SDK, Singh AK (2007) Extraction and validation of cartosat-1 DEM. Photonirvachak-J In Soc Remote Sens 35(2):121127Ali MA, Solomatine DP, Di Baldassarre G (2014) Assessing the impact of different sources of topographic data on 1-D hydraulic modeling of flood. Hydrol Earth Syst Sci Discuss 11:73757408Bothale RV, Pandey B (2013) Evaluation and comparison of multi resolution DEM derived through cartosat-1 stereo pair - a case study of damanganga basin. J Indian Soc Re Sens 41(3):497507Casas A, Benito G, Thorndycraft VR, Rico M (2006) The topographic data source of digital terrain models as a key element in the accuracy of hydraulic flood modelling. Earth Surf Proc Land 31(4):444456 Castellarin A, Di Baldassarre G, Bates PD, Brath A (2009) Optimal cross-sectional spacing in preissmann scheme 1D hydrodynamic models. J Hydraul Eng 135(2):96105Central Water Commission (CWC) (1996) Report of the working group on flood management for the ninth five year plan (19972002). CWC, New DelhiChatterjee C, Foerster S, Bronstert A (2008) Comparison of hydrodynamic models of different complexities to model floods with emergency storage areas. Hydrol Process 22(24):46954709Chavarri E, Crave A, Bonnet MP, Mejia A, Da Silva JS, Guyot JL (2013) Hydrodynamic modelling of the amazon river: factors of uncertainty. J S Am Earth Sci 44:94103Chow VT (1959) Development of uniform flow and its formulas. Open-channel hydraulics. (pp. 101113).

McGraw-Hill Book Company, New York

Cook A, Merwade V (2009) Effect of topographic data, geometric configuration and modeling approach on flood inundation mapping. J Hydrol 377(12):131142. doi:http://dx.doi.org/10.1016/j.jhydrol.2009.08.015

Web End =10.1016/j.jhydrol.2009.08.015 DHI (2014) A modelling system for rivers and channels, MIKE 11 reference manual. Danish Hydraulic Institute,

Horsholm

Domeneghetti A, Tarpanelli A, Brocca L, Barbetta S, Moramarco T, Castellarin A, Brath A (2014) The use of remote sensing-derived water surface data for hydraulic model calibration. Remote Sens Environ 149:130 141. doi:http://dx.doi.org/10.1016/j.rse.2014.04.007

Web End =10.1016/j.rse.2014.04.007 Evans AG, Ramachandran B, Zhang Z, Baily GB, Cheng P (2008). An accuracy assessment of Cartosat-1 stereo image data-derived Digital Elevation Models: A case study of the Drum Mountains, Utah. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing, 11611164Fairfield J, Leymarie P (1991) Drainage networks from grid digital elevation models. Water Resour Res 27(5):

709717. doi:http://dx.doi.org/10.1029/90wr02658

Web End =10.1029/90wr02658

Farr TG, Rosen PA, Caro E, Crippen R, Duren R, Hensley S, Kobrick M, Paller M, Rodriguez E, Roth L, Seal D,

Shaffer S, Shimada J, Umland J, Werner M, Oskin M, Burbank D, Alsdorf D (2007) The shuttle radar topography mission. Reviews of Geophysics 45(2). http://dx.doi.org/10.1029/2005rg000183

Web End =10.1029/2005rg000183 Flood workshop (2011) Workshop on floods in Odisha: civil society perspectives. 16th October 2011, Conference

Hall, Odisha Environmental Society. http://www.scribd.com/doc/69550673/Brief-Minutes-of-Flood-Workshop-revised%23scribd

Web End =http://www.scribd.com/doc/69550673/Brief-Minutes-of-Flood- http://www.scribd.com/doc/69550673/Brief-Minutes-of-Flood-Workshop-revised%23scribd

Web End =Workshop-revised#scribd

References

1308 P.P. Jena et al.

Forkuor G, Maathuis B (2012) Comparison of SRTM and ASTER derived Digital Elevation Models over two regions in Ghana - Implications for hydrological and environmental modeling. Studies on Environmental and Applied Geomorphology, Dr. Tommaso Piacentini (Ed.), ISBN: 978-953-51-0361-5. http://cdn.intechopen.com/pdfs-wm/32991.pdf

Web End =http://cdn. http://cdn.intechopen.com/pdfs-wm/32991.pdf

Web End =intechopen.com/pdfs-wm/32991.pdf

Forster S, Chatterjee C, Bronstert A (2008) Hydrodynamic simulation of the operational management of a proposed flood emergency storage area at the middle Elbe river. River Res Appl 24(7):900913

Gianinetto M (2009) Evaluation of cartosat-1 multi-scale digital surface modelling over France. Sensors 9(5):32693288

Gichamo TZ, Popescu I, Jonoski A, Solomatine D (2012) River cross-section extraction from the ASTER globalDEM for flood modeling. Environ Model Softw 31:3746

Giribabu D, Kumar P, Mathew J, Sharma KP, Murthy YVNK (2013) DEM generation using cartosat-1 stereo data: issues and complexities in Himalayan terrain. Euro J Remote Sens 46:431443Jena PP, Chatterjee C, Pradhan G, Mishra A (2014) Are recent frequent high floods in Mahanadi basin in eastern

India due to increase in extreme rainfalls? J Hydrol 517:847862

Liu X (2008) Airborne LiDAR for DEM generation: some critical issues. Prog Phys Geogr 32(1):3149Mani P, Chatterjee C, Kumar R (2014) Flood hazard assessment with multiparameter approach derived from coupled 1D and 2D hydrodynamic flow model. Nat Hazards 70(2):15531574Merwade VM, Maidment DR, Hodges BR (2005) Geospatial representation of river channels. J Hydrol Eng

10(3):243251

Merwade V, Cook A, Coonrod J (2008a) GIS techniques for creating river terrain models for hydrodynamic modeling and flood inundation mapping. Environ Model Softw 23(1011):13001311Merwade V, Olivera F, Arabi M, Edleman S (2008b) Uncertainty in flood inundation mapping: current issues and future directions. J Hydrol Eng 13(7):608620Metz M, Mitasova H, Harmon RS (2011) Efficient extraction of drainage networks from massive, radar-based elevation models with least cost path search. Hydrol Earth Syst Sci 15(2):667678Moriasi DN, Arnold JG, Van Liew MW, Bingner RL, Harmel RD, Veith TL (2007) Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans ASABE 50(3): 885900Paiva RCD, Collischonn W, Tucci CEM (2011) Large scale hydrologic and hydrodynamic modeling using limited data and a GIS based approach. J Hydrol 406(34):170181Patro S, Chatterjee C, Mohanty S, Singh R, Raghuwanshi NS (2009a) Flood inundation modeling using MIKE

FLOOD and remote sensing data. J Indian Soc Remote Sens 37(1):107118

Patro S, Chatterjee C, Singh R, Raghuwanshi NS (2009b) Hydrodynamic modelling of a large flood-prone river system in India with limited data. Hydrol Process 23(19):27742791Pieczonka T, Bolch T, Buchroithner M (2011) Generation and evaluation of multitemporal digital terrain models of the Mt. Everest area from different optical sensors. ISPRS J Photogramm Remote Sens 66(6):927940 Pramanik N, Panda RK, Sen D (2010) One dimensional hydrodynamic modeling of river flow using DEM extracted river cross-sections. Water Resour Manag 24(5):835852Rawat KS, Mishra AK, Sehgal VK, Ahemd N, Tripathi VK (2012) Comparative evaluation of horizontal accuracy of elevations of selected ground control points from ASTER and SRTM DEM with respect to CARTOSAT-1 DEM: a case study of Sahjahanpurdistrict, Uttar Pradesh, India. http://dx.doi.org/10.1080/10106049.2012

Web End =10.1080/10106049.2012 . 724453Sahoo B, Chatterjee C, Raghuwanshi NS, Singh R, Kumar R (2006) Flood estimation by GIUH-based Clark and

Nash models. J Hydrol Eng 11(6):515525

Samantaray D, Chatterjee C, Singh R, Gupta PK, Panigrahy S (2015) Flood risk modeling for optimal rice planning for delta region of Mahanadi river basin in India. Nat Hazards 76:347372Sanders BF (2007) Evaluation of on-line DEMs for flood inundation modeling. Adv Water Resour 30(8):1831

1843

Sharma CS, Mishra A, Panda SN (2014) Assessing impact of flood on river dynamics and susceptible regions: geomorphometric analysis. Water Resour Manag 28(9):26152638Shenk T (1996) Digital areal triangulation. Int Arch of Photogramm Remote sens Vol. XXXI, PartB3, Vienna Tarekegn TH, Haile AT, Rientjes T, Reggiani P, Alkema D (2010) Assessment of an ASTER-generated DEM for

2D hydrodynamic flood modeling. Int J Appl Earth Obs 12(6):457465

Tarolli P (2014) High-resolution topography for understanding earth surface processes: opportunities and challenges. Geomorphology 216:295312Teng J, Vaze J, Dutta D, Marvanek S (2015) Rapid inundation modeling in large floodplains using LiDAR DEM.

Water Resour Manag 29(8):26192636

Vogt EV, Colombo R, Bertolo F (2003) Deriving drainage networks and catchment boundaries: a new methodology combining digital elevation data and environmental characteristics. Geomorphology 53(34):281298

Assessment of Tartosat-1 DEM for Modeling Floods in Data scarce 1309

Wilson M, Bates P, Alsdorf D, Forsberg B, Horritt M, Melack J, Frappart F, Famiglietti J (2007) Modeling large-scale inundation of Amazonian seasonally flooded wetlands. Geophys Res Lett 34(15)

Yamazaki D, Baugh CA, Bates PD, Kanae S, Alsdorf DE, Oki T (2012) Adjustment of a spaceborne DEM for use in floodplain hydrodynamic modeling. J Hydrol 436:8191

Yan K, Tarpanelli A, Balint G, Moramarco T, Baldassarre G (2015) Exploring the potential of SRTM topography and radar altimetry to support flood propagation modeling: Danube case study. J Hydrol Eng 20(2): 04014048

Yarrakula K, Deb D, Samant B (2013) Comparative evaluation of cartosat-1 and SRTM imageries for digital elevation modeling. Geo Spat Inf Sci 16(2):7582

Word count: 7215

Show less

Springer Science+Business Media Dordrecht 2016

Assessment of Cartosat-1 DEM for Modeling Floods in Data Scarce Regions

Content area

Abstract

Full text