A comparative study of automatic drainage network extraction using ASTER GDEM, SRTM DEM and Cartosat-1 DEM in parts of Kosi basin, Bihar, India

The present study attempts to evaluate and compare the different open source 30 m resolution spaceborne digital elevation models (DEM) based on their automatic drainage network extraction efficiency. Digital elevation models from three satellite data i.e., ASTER GDEM, SRTM DEM and Cartosat-1 DEM have been employed for the extraction of stream networks. The study was carried out in the watershed situated in parts of Saharsa, Madhepura and Supaul district along the river Kosi, Bihar. Analysis was performed in ArcMap10.3 software wherein, three different thresholds of flow accumulation were used to assess the comparative evaluation. Morphometric parameters were analysed using geographical information system (GIS). The evaluation and quantification of morphometric parameters provide an effective basis for the comparison of extracted stream networks based on hydrological and morphological characteristics. Comparison and selection of appropriate DEM products for quantitative and qualitative hydrological studies are lacking in the present study area. This comparative study is helpful in selecting suitable DEM products out of the three commonly available free datasets for delineating streams in this floodplain with little variation in elevation and slope. The findings can be beneficial for using appropriate DEM data in various morphometric, prioritization, management and other GIS applications in parts of Kosi basin.


Introduction
Remote sensing and geographical information system (GIS) are efficient tools for qualitative and quantitative analysis of drainage networks in a watershed. They have been used for watershed prioritization [1, 2] watershed management [3,4] using drainage parameters. Computation of Morphometric parameters has been successfully accomplished by remote sensing and GIS [5].
Prior to the availability of DEM data, the watershed and drainage networks were drawn and extracted manually from Toposheets where all streams could not be mapped due to skill-based errors. With the advent of DEM, the inconsistency in manually drawn data has been solved. The digital elevation model (DEM) is a pixelbased matrix structure that represents the elevation of the Earth's surface excluding man-made features, vegetation or any object above ground. Each pixel (squared cell) of a DEM is linked to an elevation value [6]. Computers, application software and algorithms can model and evaluate 3D topography using DEM as an input. The automatic generation of a DEM from remotely sensed data with sub-pixel precision is promising [7]. Digital elevation models are being produced, disseminated and widely used by different organizations owing to their prospect and credence. DEMs have widespread use in various hydrologic and environmental applications such as water resource management, hydrological modelling, flow channel characterization, watershed mapping, floodplain characterization and wetland mapping. Free availability of DEM from various sources such as Cartosat-1 DEM (30 m) of ISRO, ASTER GDEM (30 m) of NASA and METI, SRTM (30 m) of NASA and NGA have steered researchers to work on them.
Automatic drainage network extraction performed on DEM is one of the many functions employed in research owing to its gamut of applications. Precise delineation of drainage networks is a necessity for many natural resource management concerns [8,9]. Manual drainage extraction is exhausting and time-consuming whereas automatic drainage network extraction is practically effortless. The automatic extraction of drainage networks is inevitable in many applications of GIS, such as hydrologic analysis, mineral deposition and land erosion [10,11]. O'Callaghan et al. [12] emphasised the importance of drainage network extraction from DEM for quantitative studies in geomorphology and hydrology. A plethora of work on automatic drainage network extraction has been accomplished by many researchers [12][13][14][15][16].
It is confounding to select an appropriate DEM in a given area from an array of freely available DEM data to perform a specific task. Studies on comparative analysis of different DEM data based on the accuracy [17,18], spot height and cell values [19], lineament extraction [20,21] and automatically extracted drainage [16,22] have been carried out.
Most of the studies using DEM in the Kosi basin deal with morphometric analysis [4,23,24], flood hazard mapping [25], Kosi dynamics and avulsion [26]. There is a lack of study to unearth the quality, usability and suitability of a better DEM from an array of freely available data for better drainage extraction in the present study area.
Determination and comparison of DEM data on the basis of morphometric parameters for automatic drainage extraction will help in quantitative and qualitative hydrological studies, prioritization of the watershed in the area based on drainage morphometry and use of suitable DEMs in other GIS applications and fields [23,24,27,28]. A comparison was performed in geographic information software ArcMap10.3 between ASTER GDEM, Cartosat-1 DEM and SRTM DEM each of 30-m spatial resolution. The efficacy was investigated based on the automatic extraction of drainage from each of the DEMs in the present study. Sentinel-2 satellite imagery was used as a reference for the analysis of the drainage network extracted.

Study area
The watershed under study lies in the Kosi mega fan covering parts of Saharsa, Supaul and Madhepura district, Bihar. River Kosi flows sinuously in the western part of the study area often with interlacing channels shifting laterally over the fan forming numerous areas inundated with water. The watershed lies between 26° 15′ 00″and 25° 30′ 00″ north latitudes and 86° 15′ 00″ and 86° 54′ 00″ east longitudes. The elevation ranges from 0 to 91 m above mean sea level (MSL). The slope of the watershed ranges from 0° to 22°. The location map of the study area is given in Fig. 1.
Geologically, the watershed comprises Quaternary sediments of Holocene age that have been classified as Gothini formation representing feebly oxidised sand, clayey sand and coarse-grained micaceous sand followed upwards by Kosi/Vaishali formation comprising silty deposits of older flood plains. Unoxidized sandy and silty deposits form the present-day Diara formation [29]. Younger alluvial plains, older floodplains, active floodplains as well as channel bar, levees, back swamps, marshes and oxbow lakes are characteristic geomorphic features of the watershed.

Data sources
Freely available satellite-acquired Digital Elevation Model datasets were processed. Three commonly employed DEM data sets-ASTER GDEM version 3, SRTM version 3 and Cartosat DEM version v3R1 with 30 m spatial resolution were utilized for the present study. The processing of these three DEMs was carried out in ArcMap 10.3 software.

ASTER (30 m) GDEM
The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) is a Japanese multispectral imaging remote sensing instrument onboard the Terra satellite launched by NASA in 1999 and has been collecting data since February 2000. ASTER GDEM V3 was generated using scenes acquired between March 1, 2000, and November 30, 2013. The GDEM collection includes 22,912 tiles (https:// aster web. jpl. nasa. gov/). Two tiles of 30 m resolution of one arc-second grid were downloaded from the website of the United State Geological Survey (USGS) Earth Explorer (http:// www. earth explo rer. usgs. gov) for the study area.

3
Center-Deutsches Zentrum für Luft-und Raumfahrt (DLR), and Agenzia Spaziale Italiana (ASI). The mission objective was to obtain single-pass interferometric SAR imagery to be used for DEM. Two SRTM DEM tiles for the present study have been downloaded from USGS Earth Explorer (http:// www. earth explo rer. usgs. gov).

Indian Space Research Organisation (ISRO) launched
Cartosat-1 on May 5, 2005, intending to deliver DEM at a higher resolution. The data is made available by ISRO and downloaded from the BHUVAN website of the National Remote Sensing Centre (NRSC) portal.

Sentinel-2 satellite imagery
Sentinel-2 image acquired in April 2020 was downloaded from Copernicus open access hub of European Space Agency (https:// scihub. coper nicus. eu). Two tiles were downloaded and False Colour Composite was prepared for better visual display of drainage on the imagery. The Sentinel-2 imagery of the study area has been used as a reference data to visually observe the overlapping of drainage network for comparative analysis. The recent pre monsoon imagery of April helps in better visibility of drainage network as the rabi crops are harvested and the study area is devoid of vegetation. The streams and the study area are properly visualised on the imagery which otherwise would have been flooded in the post monsoon imagery of this floodplain. The Table 1 shows the general DEM specifications. Figure 2 shows the DEM from the three sources used in the study.

Automatic drainage network extraction
The downloaded DEMs were reprojected to the Universal Transverse Mercator (UTM) projection zone of 45 N. The tiles were mosaiced to get the extent of the study area and the operations were performed to extract the drainage properties.
The Hydrology tool in the Spatial Analyst tool from the Arc Toolbox in Arc Map 10.3 was used for the study. The Fill tool was employed on the DEM until all sinks within the specified z-limit are filled. Then, the Flow direction tool incorporated the filled DEM as input and created a raster of flow direction from each cell to its steepest downslope neighbour. The basin tool used flow direction raster layer for automatic extraction of basins from the DEM. The desired basin was selected from each of the digital elevation models and clipped on which further operations were performed. The Flow direction raster layer was also taken as the input in the Flow Accumulation tool that calculates  The selection of flow accumulation threshold by using the raster calculator is the fundamental aspect of automating drainage extraction from DEM. The map algebra tool in the spatial analyst tools selects a specific stream and thresholds the value of stream network extraction. The threshold can also be obtained by using the conditional (con) tool. Random values were selected and input into the raster calculator equation until the pertinent stream network raster layer was attained. The chosen values fed in the raster calculator helped obtain the desired details of the stream network. The lower the value chosen for the threshold-the more detailed stream networks would be. Three-threshold values of flow accumulation greater than 5000, 10,000 and 15,000 were selected for the comparative analysis in this study. These three varying thresholds were chosen for comparison of automatically extracted drainage networks at different levels. The chosen values help in exhibiting clarity from dense to less dense drainage network. The software has been run at these varying thresholds thereby establishing accuracy as it repeatedly compares the three DEMs at three different levels of drainage extraction. Streams obtained after thresholding the flow accumulation were in the raster format (.tif ). This stream raster layer was used as the input for obtaining the stream order using the stream order tool wherein Strahler's law was followed for ordering streams. Strahler's law of stream ordering is widely accepted because the numbering of streams at hierarchy of tributaries is easy without any complex calculation [30]. The stream order network delineated in a raster format was converted into shapefile (.shp) using the Stream to Feature tool. Thus, the basin and drainage network were automatically extracted from each of the DEMs using the Spatial Analyst Tools (Figs. 2, 3). The flowchart ( Fig. 4) represents the methodology taken up for the extraction of the basin and drainage network. Stream network (.shp) layer was overlain for visual comparison of drainage networks from different DEMs on Sentinel-2 satellite imagery.

Calculation of morphometric indices
To assess the comparison of stream networks extracted at varying thresholds, morphometric indices were calculated. Mathematical formulas by Strahler [30], Horton [31], Schumm [32] have been employed in this study for the computation of linear parameters for morphometric analysis. The parameters selected under linear parameters were stream order, stream number, stream length, stream length ratio, mean stream length, bifurcation ratio and mean bifurcation ratio. Basin geometry parameters calculated were length, area and perimeter. The calculations were performed on the automatically extracted stream networks from DEM at different thresholds. The stream numbers were numbered according to the orders proposed by Strahler [30] inbuilt in ArcMap 10.3. The parameters computed using GIS technique on the extracted drainage network includes stream order, stream number, stream length and basin geometry which includes area, perimeter and basin length. The formulae used for the calculation of linear morphometric parameters are given in Table 2.

Stream order (S o) and stream number (N u )
The stream order is a measure of the relative size of streams that indicates the level of branching in a river system. Stream number is the number of stream segments for each order. Stream order and stream number are influenced by physiography, geology and geomorphology of the watershed. The order of the highest stream determines the order of the watershed and thus the present watershed is a 5th order drainage basin.
Cartosat-1 DEM extracts 5th order streams at two thresholds of 5000 and 10,000 whereas ASTER GDEM and SRTM DEM extract 5th order streams at a single threshold of 5000 only.
At all three thresholds, Cartosat-1 DEM extracts the maximum number of streams as compared to ASTER GDEM and SRTM DEM.
Thus, Cartosat-1 DEM surpasses ASTER GDEM and SRTM DEM in the extraction of stream order at two thresholds and the number of streams at all the thresholds in the study area (Table 3).

Stream length (L u )
The length of the streams was computed using Horton's [31] formula in ArcMap 10.3. The total stream length is higher (> 100 km) across all thresholds in all DEMs in the watershed. Longer stream lengths are significant to gentler slopes. Total stream length is related to the mean annual runoff [33]. The values indicate a high mean annual runoff in the present study area.
ASTER GDEM shows a comparatively lower value of total stream length across all thresholds. Cartosat-1 DEM shows consistently higher total stream length at all the thresholds.
From Table 4, it is evident that Cartosat-1 DEM surpasses ASTER GDEM and SRTM DEM in obtaining higher total stream lengths at all three thresholds.

Mean stream length (L sm )
It is calculated as the total lengths of streams of a particular order to the total number of streams in that order. Slope and topography play a vital role in controlling the L sm of a region [34]. The inconsistency of L sm across  (Table 5).

Stream length ratio (RL)
It is a dimensionless parameter calculated as the ratio of the mean stream length of a given order to the mean stream length of the next lower order. This ratio between the successive stream lengths reveals the slope and an Fig. 4 Flowchart of the methodology adopted for the study increase from lower to higher order reveals the mature geomorphic stage of development. The values vary between 0.56 to 1.55 (Table 6) indicating no drastic change in slope and geomorphic stage from one order to the next at all thresholds for all the DEM.

Bifurcation ratio (R b )
Rb is a dimensionless parameter computed as the ratio of the total number of streams of a given order to the number of streams of the next higher order. The watershed shows low R b (< 5) across all thresholds in the three DEMs ( Table 7).
The low values of Rb indicate a less distorted drainage N u = number of streams of a particular order 'u' dimensionless Strahler, 1964 [30] Bifurcation ratio (R b ) R b = (N u /N u + 1); where, N u = number of streams of a particular order 'u' , N u + 1 = number of streams of next higher order 'u + 1' Dimensionless Schumm [32] Mean bifurcation ratio (R bm ) R bm = mean of bifurcation ratios of all orders Dimensionless Schumm [32] Stream length (L u ) L u = total length of streams (km) of a particular order 'u' km Horton [31] Mean stream length (L sm ) L sm = L u /N u ; where, L u = total length of streams (km) of a particular order 'u' , N u = total number of streams of a particular order 'u' km Horton [31] Stream length ratio (R l ) R l = L sm /L sm − 1; where, L sm = mean stream length of a particular order 'u' , L um − 1 = mean stream length of next lower order 'u -1'  network and mature structural condition [35]. Although a high value from the normal low at a threshold of 5000 in SRTM DEM and Cartosat-1 DEM is observed.

Mean bifurcation ratio (R bm )
In the study area, the mean bifurcation ratio varies from 1.84 to 5.02 (Table 7) suggesting less structural disturbance and a well-developed drainage network for all three DEMs. A slight high at a threshold of 5000 in SRTM DEM is computed.

Length of the basin (L b ), basin area (A), perimeter (P)
Schumm [32] described the basin length as the maximum length of the watershed parallel to the principal drainage line.
Basins having an area of more than 100 km 2 are classified as large basins [31]. The largest area of the basin is computed for Cartosat-1 DEM (Table 8). There exists a connection between the area and the total stream lengths of a watershed [32].

Comparison of automatic drainage network extraction on sentinel-2 satellite imagery (FCC)
ASTER GDEM stream network from the extraction lies distant from the main river channel of Kosi and appears to have shifted towards the right as updated on Sentinel-2 imagery. The basin extracted also appears to cut across the main river channel of Kosi (Fig. 5a). SRTM DEM shows the maximum overlapping on the main river channel (Fig. 5b). Cartosat-1 DEM presents a fair amount of overlay on the lower side of the main river channel (Fig. 5c) but not as much as SRTM DEM as detected from the satellite imagery.

Conclusion
As revealed from the study, the morphometric parameters can be well retrieved from different sources of DEM exhibiting comparable results. Based on the morphometric analysis, the three DEMs across all thresholds indicate a high mean annual runoff, gentler slopes and mature structural conditions. It is elucidated from the comparative study that Cartosat-1 DEM exceeds ASTER GDEM and SRTM DEM in the extraction of the total number of streams at all the thresholds of 5000, 10,000 and 15,000 in the extracted watershed. Cartosat-1 extracts Vth order streams at two thresholds of 5000 and 10,000 whereas ASTER DEM and SRTM DEM extract Vth order streams at a single threshold of 5000 only. Cartosat-1 DEM shows consistently higher stream length at all three thresholds. A remarkable difference can be observed in the stream order, stream number and stream lengths of automated drainage as evident from the results. Thus, these three parameters become the deciding factors for the efficiency of DEM in this study. Cartosat-1 DEM being indigenous to India performs better than the other two DEMs in automatic extraction of the drainage network. The drainage network extracted from SRTM DEM shows the maximum overlapping on the main river channel as detected on Sentinel-2 satellite imagery. Overlaying and viewing extracted stream networks on the satellite imagery gives a visual analysis for comparison. The comparison between the automatic drainage network extraction of different DEMs (30 m) based on the morphometric parameters at varying thresholds Author contributions Both the authors contributed to the study conception, design, material preparation, data collection and analysis. The authors read and approved the final manuscript.
Funding No funding was received for conducting this study.

Conflict of interest
The authors have no relevant financial or nonfinancial interests to disclose.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.