Chapter – 3
REDEFINING THE NEKA RIVER WATERSHED
BOUNDARY LINE OF IRAN COMPARING ASTER,
SRTM, DIGITAL TOPOGRAPHY DEM AND TOPO
SHEET USING GIS AND REMOTE SENSING
TECHNIQUES
40
CHAPTER - 3
REDEFINING THE NEKA RIVER WATERSHED
BOUNDARY LINE OF IRAN COMPARING ASTER, SRTM,
DIGITAL TOPOGRAPHY DEM AND TOPO SHEET USING
GIS AND REMOTE SENSING TECHNIQUES
3.1 Abstract
Accurate area calculation and for land use evaluation, land use change,
Geomorphologic classification, Hydrological analysis delineation of watershed
boundaries are the base. With the advent of advanced spatial tools redefining the
manual drawn topographical boundaries is necessary. An integrated approach of data
analysis and modeling can accomplish the task of delineation. The main objectives of
this paper is to redraw the already drawn manual boundary by comparing four
different data source, such as Aster DEM, SRTM DEM, Digital Topography DEM
and Topo sheet data. Secondly, it is also required to ascertain the most accurate
method and the appropriate remote sensing data and GIS for delineation purpose. A
true comparison of all the four methods the mean distance existing between the
ground GPS data and the Aster Dem was 43 mts, between Ground data and SRTM
was 307 mts, the mean distance between the Digital Topographic map and the ground
GPS points were 269. It is also tested between Topo sheet and the GPS ground points
was 304. The regression analyses comparing 230 x-y points along the complete
boundaries yielded an R2 of 0.082 between the ASTER and SRTM boundaries; the R2
for the comparison between the ASTER and the digital topography boundaries was
0.0157, between ASTER and Topo sheet Map was 0.171. After, One-way Analysis of
variance (ANOVA) in SPSS, the means obtained from the four methods of SRTM,
ASTER, Digital Topography and Top sheet were compared. The initial results showed
41
that there was a statistically significant difference among the four means. Apart from
this method the comparison of stream distance with the ground point proved that aster
stream path is closer than the other three stream path difference. Thus it is proved that
the Aster Dem is the most suitable application for the delineation of the rugged terrain
when compared with the other three methods.
3.2 Introduction
Land use evaluation and forest management in the mountain regions is gaining
importance at the government and at community level. The proper demarcation of
boundary is the real challenge ahead. Hydrological research in mountain watersheds of
developing countries is a relatively new field, therefore, hydrologic and erosion data are
necessary for the development of mathematical watershed models that can simulate and
evaluate existing and proposed management scenarios (de Jong et al., 2005).
Across the world only few research works have been carried at the small scale
to delineate the boundary. The demarcation boundary study done by the American
Society of Agriculture and Biological Research team (Julia K. Pryde 2007 et.al) has
published in its report the demarcation done for Illamanga Sub watershed in North
America.
3.3 Digital Elevation Data
Digital elevation data are of limited availability in many less developed areas or in
a rugged terrain with steep slopes. The space borne earth observation sensors and google
earth images have reduced the complexness of the authenticity of the elevation data. The
Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER)
42
onboard and the NASA Terra satellite has the capability of taking along-track stereo
images allowing the generation of high-resolution DEMs (Kääb, 2002; Stevens
etoral., 2004; Kääb, 2005).
Digital elevation models (DEMs) are topographic models of the Earth's terrain
(bare ground) that have had the heights of vegetation, buildings, and other cultural
features digitally removed. DEMs are commonly referred in the remote sensing world
as digital terrain models (DTMs) typically offered as a continuous elevation surface
as a grid (Podobnikar, 2009). Different techniques for the generation of DTMs have
been developed since their inception more than fIfty years ago (Miller and Laflamme,
1958; Gesch et al., 2002; Hirano et al., 2003; Maune, 2007; Intermap, 2009).
Significant advances in remote sensing technologies have led to a new era of higher
quality global topographic observations, where reliable topographic measurements are
becoming a possibility (Homer et al., 2007). At small scales, space borne systems
(coarse ground sampling distance (GSD)) such as shuttle radar topographic mission
(SRTM) collected 80% of the earth's landmass with 30 m or 90 m resolution (Rabus
et al., 2003). At medium scales radar interferometric techniques (medium to high
resolution) had been applied to generate global DTMs (Madsen et aI., 1993; Farr and
Kobrik, 2000; Maune, 2001: Walker et al., 2007; Intermap, 2009). For larger scales
and more local usage, airborne laser scanning (LiDAR) and aerial photogrammetric
techniques (high spatial resolutions) have been applied to create DTMs (e.g. Lefsky et
al., 2002; Nresset, 2002; Andersen et al., 2006). Remote sensing and GIS applications
of DTMs have become widespread. Forest and water resource management
applications, including watershed management, flood hazard mapping, timber harvest,
and fire management are dominant users of DTMs. Terrain attributes often provide
direct inputs for environmental, forestry, topographic and hydrological models and
43
thus accuracy of the elevation models is critical to environmental modeling
(Kellndorfer et al., 2004; Thirion et al., 2006; Balzter et al., 2007a; 200Th; Anderson
et al., 2008). Mapping standards have tended to accept the data. If it is within
mapping standards such as the National Standard for Spatial Data Accuracy (FGDC,
1998) and the National Digital Elevation Program (NDEP, 2004).
3.4 SRTM DTM Accuracy Assessment
The SRTM radar signal measurement result in a reflec tive surface elevation
which depends on terrain cover and is a complicated function of the electromagnetic
and structural properties of the scattering medium (Bhang et al., 2007). In snow, the
penetration depth of the radar signal depends on wetness, temperature, and porosity
(Braun et al., 2007). Vegetation presents an even more complex scattering
environment. It has been estimated that C-band only penetrates a quarter or a third of
the canopy height (Carabajal, 2005). Performance evaluations by NIMA, the USGS,
and the SRTM project team have shown the absolute vertical error to be much
smaller, with the most reliable estimates being approximately 5 m (Rosen et al., 2001;
Sun et al., 2003). Brown et al. (2005) used GPS and NED data to evaluate the
accuracy of the SRTM data for southeastern Michigan. They reported that the SRTM
mission specifications for absolute and relative height errors for the GPS ground
control point targets were exceeded. A more extensive analysis of the SRTM DGPS
data indicates that it meets the absolute and relative accuracy requirements even for
bare surface areas. Previous research efforts indicated that accuracy for an IFSAR
derived DTM could be terrain dependent. According to the mission objectives,
SRTM data were expected to have an absolute horizontal circular accuracy of less
than 20 m. Absolute and relative vertical accuracy was anticipated to be less than 16
and 10m, respectively (Kellendorfer et al., 2004).
44
3.5 ASTER DTM Accuracy Assessment
As part of ASTER digital elevation model (DEM) accuracy evaluation efforts
by the US/Japan ASTER Science Team, stereo image data for four study sites around
the world have been employed to validate prelaunch estimates of height accuracy
(Hirano et aI., 2003). Automated stereo correlation procedures were implemented
using the Desktop Mapping System (DMS) software on a personal computer to derive
DEMs with 30 to 150 m postings. Results indicate that a root-mean-square error
(RMSE) in elevation between ±7 and ±15 m can be achieved with ASTER stereo
image data of good quality. An evaluation of an ASTER DEM data product produced
at the US Geological Survey (USGS) EROS Data Center (EDC) yielded an RMSE of
± 8.6 m. Overall; the ability to extract elevations from ASTER stereo pairs using
stereo correlation techniques meets expectations. Studies were conducted by a large
group of international investigators, working under the joint leadership of U.S and
Japan ASTER Project participants, to validate the estimated accuracy of the new
ASTER Global DEM product and to identify and describe artifacts and anomalies
found in the ASTER GDEM (ASTER, 2009). They reported an overall vertical
RMSE for the 934 lOX 1° GDEM tiles of 10.87 meters, as compared to NED data;
which would equate to a an accuracy at 95% confidence of 21.31 meters, or a little
more than the 20 m accuracy at 95% confidence estimated for the ASTER GDEM
prior to its production. Vertical accuracy of NED data is approximately 2-3 m RMSE.
When compared with more than 13,000 GCPs the RMSE dropped to 9.35 meters.
These values convert, respectively, to vertical errors of just over and just under the
estimated ASTER GDEM vertical error of 20 meters at 95% confidence. The ASTER
(2009) found the ASTER DTM to contain significant anomalies and artifacts, due to
clouds and the algorithm used to generate the final GDEM, which will affect its
45
usefulness for certain user applications. Another shortcoming of the current ASTER
GDEM Version 1 is the fact that no inland water mask has been applied.
Consequently, the elevations of the vast majority of inland lakes are not accurate, and
the existence of most water bodies is not indicated in the ASTER GDEM. The
vertical accuracy of this ASTER DEM was checked against 40 DGPS survey points
and 12 points digitized from USGS 1 :24,000-scale topographic quadrangles, yielding
an RMSEz of ±8.6 m. This generally corresponds with other validation results
reported by EDC (EDC DAAC, 2001).
Since the development of Geographic Information Systems (GISs), digital
elevation models (DEMs) have been generated throughout the world. DEMs provide
good terrain representations and are applied routinely in watershed modeling. DEMs
can be used to derive flow networks and then automatically generate watershed
boundaries for given outlet points using GIS technology. Therefore, an essential
component to watershed delineation is a hydrologically sound DEM of the area of
interest.
The Advanced Spaceborne Thermal Emission and Reflection Radiometer
(ASTER) is an advanced multispectral imager that was launched on board NASA’s
Terra spacecraft in December, 1999. ASTER covers a wide spectral region with 14
bands from the visible to the thermal infrared with high spatial, spectral, and
radiometric resolution. The spatial resolution varies with wavelength: 15 m in the
visible and near- infrared (VNIR), 30 m in the short wave infrared (SWIR), and 90 m in
the thermal infrared (TIR).
46
The ASTER Digital Elevation Model (DEM) product is generated using bands
3N (nadir-viewing) and 3B (backward-viewing) of an ASTER Level-1A image
acquired by the Visible Near Infrared (VNIR) sensor. The VNIR subsystem includes
two independent telescope assemblies that facilitate the generation of stereoscopic data.
The Band-3 stereo pair is acquired in the spectral range of 0.78 and 0.86 microns with a
base-to-height ratio of 0.6 and an intersection angle of about 27.7°. There is a time lag
of approximately one minute between the acquisition of the nadir and backward images
(M. Lorraine Tighe and Drew Chamberlain 2009).
3.6 Accuracy Assessment of Google Image
Google Earth data has been used by millions of people and its potential is not
well harnessed. The google earth has potential application that extends beyond
visualization. The contribution of google for the study of land-cover and land use change
science, (Himiyama 2009), (Arun das et al. 2010) are significant. The recent high
resolution images with less than 2.5 meters resolution covers nearly 20 percent of the
earth’s surface. The high resolution images allow to extract Natural features and also
human built environment. To characterize the horizontal positional accuracy of the
high-resolution Google Earth archive, the locations of 436 control points in the GE
imagery to their equivalent positions in the Landsat GeoCover data set was used,
which has positional accuracy of 50 meters root-mean-squared error (RMSE). In
an ideal assessment of spatial accuracy, it would determine the position of these
Sensors 2008, 8 7976 control points through a global ground-based campaign using
global positioning satellites (GPS). Done for below cities. Sao Paolo, Brazil, San
Salvador, El Salvador, Chonan, South Korea, and Anqing, China (David Potere
2008).
47
3.7 Data Base
Delineation of watershed boundary and comparing between the different data
sources has demanded various sources of data. All useful forms of scientific data
sources have been explored to derive the result Table3.1 Show the all data. t. The
topographical map of 1:50000 scale published in 1965 by the Iranian geographical
organization was scanned and geo referenced at WGS 1984 datum and projected with
UTM zone 39 having RMS error of less than 0.2 pixels. To cover the study area 11
topographical maps were mosaiced in ERDAS and digitized in Arc GIS to determine
the manual boundary.
The ASTER DEM data of 2009 was downloaded from G-ASTER and
resample to 29mtr, to cross verify SRTM data (DEM) of 90mtr resolution (2002) was
downloaded from USGS. The digital topographical map published in 2004 by Iranian
geographical organization, at the scale of 1:25,000, was employed to obtained in the
watershed boundary shape file of Neka River. Two types of field GPS points were
used, one for demarcating boundary and another type for calculating the distance
difference between the stream and the GPS point. The Google image downloaded
through Google map catcher was stitched using python programmer, more than 50
ground control points were used to geo-reference and the RMS error was less than 0.4
pixels. The resample was done using 1.60 meters SPOT of 2010. WGS 1984 datum
was assigned and was projected on UTM 39 zone.
The IRS panchromatic data of 1st June 2004, was georeferenced with 50 GPS
Points and the RMS error was less than 0.5 pixels, image Geo referenced to the UTM
zone 39, projection based on the WGS84 datum. The resampled was done at 5.5 mtr,
Similarly, Four resample scenes were mosaiced.
48
Table 3.1 : Data Base
Data Date Spatial
Resolution Source
SRTM (DEM) 2002 90 M USGS
ASTER (DEM) 2009 29 M ASTER G DEM
Digital Topographical Map 2004 1:25000 Iranian Geographical Organization
Topographical Map 1965 1:50000 Iranian Geographical Organization
PAN IRS 2004 5.5 M Indian Remote Sensing
Google Image 2010 1.5 M Google Earth (spot)
3.8 Methodology
Four different sources of data were used to delineate watershed boundaries for
the analysis. The Iranian Geographical Organization topo sheet was used to digitize the
Neka river watershed boundary through ArcGIS 9.3. The total area was 188.62 sq kms
(figure 3.1). Digital topography DEM, Aster DEM and SRTM DEM was analyzed
using spatial tools of ArcGIS hydrology and obtained the watershed boundary. To
enable spatial analysis tools ArcGIS Extension extension was installed.
Figure 3.1 Digitized watershed boundary line based on the toposheet mosaic
49
The GIS technique for watershed delineation consists of the following steps.
First, the “Fill” tool was used to fill sinks in the elevation grid; this removed small
imperfections in the data and enabled the “Flow Direction” tool (the second step) to run
properly and create a grid of flow direction from each cell in the elevation grid to its
steepest down slope neighbor. Then, the “Flow Accumulation” tool was used to create a
grid of accumulated flow to each cell from all other cells in the flow direction gr id. The
next step was to identify the watershed outlet grid, ensuring that was located directly
over a grid cell from the drainage network. Finally, the “Watershed” tool was used to
delineate the watershed for the specified outlet. Boundaries (in grid format) were
defined. Using Spatial Analyst, the watershed boundary and the stream grids were then
vectorized to produce polygon and polyline themes, respectively, for further analysis
and comparison (stream).
3.9 Analysis
The four watershed boundaries were compared visually. Regression analyses
were employed to compare each of the DEM-based watershed boundaries with the 230
GPS points. For the regression analyses, a Cartesian coordinate system was used to
compare the values of x at the same y location on the three boundaries to determine
how similar they were. A total of 230 points, at constant intervals of 120 m, were
utilized in each regression analysis for the complete watershed boundary. Then, one
way Anova was conducted using ssps to determine the difference between the GPS
point and the water shed boundary line.
Visually it is evident from the figure 3.2, there are many differences compared
between the ground GPS points and the delineated ASTER, SRTM, Topo sheet
(manual) and Digital topographical map boundaries. The area of the watershed
50
delineated from topo sheet (manually) is 1887.62 sq km, while the digital topography
DEM is 1906.72the SRTM-based watershed area is 1934.31 sq km and the ASTER-
based watershed area is 1901sq km. Through this method the area difference is lesser
than between one another. For land evaluation research instead of checking the error in
the total area in the watershed much importantly the exact water divide point is required.
Eminently, finding error in area was discarded and instead the distance error between the
GPS point and other boundary line has been calculated.
Figure 3.2 : Comparison of Neka (Iran) river watershed boundaries with ASTER,
SRTM, Digital Topographical DEM and TopoSheet.
The ArcGIS - tool that measures the straight-line distance from each GPS point
cell to the closest boundary line source were used to obtain the statistical descriptions of
the differences in distance and compare between four DEM-based boundary to find out
which boundary is closer to the exact ground data as shown in Table3. 2 Calculation of
error were made between ground GPS points and the boundary line derived from
ASTER DEM, Top sheet hand boundary, Digital Topography DEM and SRTM DEM
using Analysis tools, Proximity and Generate Near Table in ArcGIS. As per the
analysis the ASTER DEM boundary line is having mean variation of 43 mts distance
from the GPS point which is lesser than the other three boundary line while SRTM is
304, Topo Sheet is 307 mts and Digital topography DEM is 269 mts.
51
Table 3.2 : Descriptive statistics of the difference in distance between limits
SRTM Aster
digital topography
DEM
topo
sheet
Mean 304.29 43.41 269.87 307.33
Standard Error 42.08 13.22 32.47 32.12
Median 111.67 23.44 129.33 171.75
Mode 68.93 8.60 129.36 73.23
Standard Deviation 639.56 200.89 493.52 488.13
Sample Variance 409037.6
0 40355.9
0 243562.81 238267.1
3
Kurtosis 10.37 217.29 14.98 15.73
Skewness 3.33 14.54 3.82 3.78
Range 3155.17 3037.78 3190.11 3403.47
Minimum 1.54 0.01 1.77 1.35
Maximum 3156.71 3037.79 3191.88 3404.82
Sum 70291.21 10027.0
5 62339.95 70992.57
Count 231.00 231.00 231.00 231.00
Confidence Level (95.0%) 82.91 26.04 63.98 63.28
The regression analyses comparing 230 x-y points along the complete
boundaries yielded an R2 of 0.082 between the aster and srtm boundaries; the R2 for
the comparison between the ASTER and the digital topography boundaries was 0.0157,
between ASTER and Topo sheet Map was 0.171 as shown in figure 3.3, 3.4 and 3.5
respectively. Therefore the regression analysis performed to test further any relation
between the other three boundaries and the Aster boundary indicates Aster is having
less relationship between the three boundary.
52
Figure 3.3, 3.4 and 3.5 : Regression analysis compared between ASTER – with
SRTM, Digital Topographical Map and Topo sheet
y = 0.9135x + 264.64 R² = 0.0823
0.00
500.00
1000.00
1500.00
2000.00
2500.00
3000.00
3500.00
0.00 500.00 1000.00 1500.00 2000.00 2500.00 3000.00 3500.00
(3.3) Aster and srtm
53
To compare the four means, the researcher subjected them to the One-way
Analysis of variance (ANOVA) in SPSS. The means obtained from the four methods of
SRTM DEM, ASTER DEM, Digital Topography DEM and Topo sheet manual
boundary were compared. The initial results showed that there was a statistically
significant difference among the four mean as shown in Table 3.3 for SPSS output. As
it is observed in the last column the P-value is less than 0.001. (p< 0.001).
Table 3.3 : One way Analysis of variance (ANOVA)
To see where the differences are exactly lied, the post hoc multiple comparisons
(Scheffe) was used. The level of significance was set at 0.001 (P=0.001). The results of
the post hoc multiple comparisons revealed a statistically significant difference between
the distances as measured through ASTER and all the other methods of distance
measurement. The distances measured by the other three groups were equally inexact
and much less exact than that of ASTER as shown in the table 3.4 below for the
summary of ad hoc multiple comparisons.
Sum of Squares df Mean Square F Sig.
Between Groups 11064704.063 3 3688234.688 15.843 .000
Within Groups 214181392.113 920 232805.861
Total 225246096.176 923
54
Table 3.4 : Post HOC Multiple Comparisons (SCHEFFE )
Data
Mean
Difference
(I-J)
Std.
Error Sig.
99.9% Confidence
Interval
Lower
Bound
Upper
Bound
SRTM ASTER 260.88 (*) 44.89 .000 79.05 442.70
Digital TOPO 34.42 44.89 .899 -147.40 216.24
Toposheet -3.03 44.89 1.000 -184.86 178.78
ASTER SRTM -260.88(*) 44.89 .000 -442.70 -79.05
Digital TOPO -226.46(*) 44.89 .000 -408.28 -44.63
Toposheet -263.91
(*) 44.89 .000 -445.74 -82.09
Digital
TOPO
SRTM -34.42 44.89 .899 -216.24 147.40
ASTER 226.46 (*) 44.89 .000 44.63 408.28
Toposheet -37.455 44.89 .874 -219.28 144.36
Toposheet SRTM 3.03 44.89 1.000 -178.78 184.86
ASTER 263.91 (*) 44.89 .000 82.095 445.74
Digital TOPO 37.45 44.89 .874 -144.36 219.28
3.10 Visual Cross Examination of four boundary line with Google Image and
PAN IRS.
Visually the four boundary line overlaid on Google image, are over lapping on
each other on steep slopes and on gentle slopes the boundary line are deviating. Among
the four boundary lines, the Aster DEM boundary line is exactly cutting across the
water divide as shown in figure 3.6 Similarly when the boundary lines were overlaid on
PAN IRS it is confirmed that the boundary lines merge on steep slopes but on gentle
slopes they deviate each other as one can see from the figures 3.7 In this case also Aster
Dem boundary line is cutting exactly on the water divide point, either on steep or gentle
slope.
55
Figure 3.6 : Visual Cross Examination of four boundary line with Google Image
Figure 3.7 : Visual Cross Examination of four boundary line with PAN IRS
56
3.11 Stream Network Analysis
Based on the GPS points picked up on the steam path were compared between
each stream line. The mean distance between GPS point and the Aster DEM is 58.70
mts, Digital Topo graphical stream is 129.34, topo sheet is 118.79 mts and SRTM is
98.76 mts. Only ASTER DEM is closer to the ground data compared between the other
three as shown in the figure 3.8 and 3.9
Figure 3.8 : Stream Network comparison between ASTER, SRTM, Digital
Topography DEM and Toposheet of Neka River Watershed
Figure 3.9 : Stream Network and watershed boundary comparison between
ASTER, SRTM, Digital Topography DEM and Toposheet of Neka River
(Iran)
57
3.12 Conclusions
The methodology described in this paper allows evaluate watershed delineation
on DEMs of different source. The accuracy of the watershed delineation it is highly
dependant on the accuracy and good quality of the Digital Elevation Model available
(DEM).
Secondly it is also proved that, the ASTER is good for demarcation on rugged
and steep slopes.
The Stream network analysis also proved that the ASTER DEM possess less
error compared between the SRTM, Digital Topography and Topo sheet.
Lastly, the Google visual comparison also has proved that, the ASTER data has
less error compared with the other three boundary lines.
Similarly, the comparison made using IRS PAN data also proved that, the
ASTER data is the best for delineation and to good to extract the watershed boundary.
ASTER data have several advantages, including low cost, high spatial resolution, good
correlation over vegetated areas. Its disadvantages include mainly the potential masking
by clouds. On the other hand, elevation models produced from SRTM data will be the
highest resolution topographic dataset ever produced for the Earth’s land surface.
Therefore, an obvious advantage of SRTM is the significant increase in spatial
resolution and vertical accuracy over existing global elevation data. Although, the
accuracy is clearly dependent upon the terrain vegetation as a radar cannot penetrate it.
Finally, ASTER DEMs appear to be highly complementary to other types of satellite-
derived data, such as Shuttle Radar Topography Mission (SRTM). It had been shown
that a fusion of DEM from different sources (optics and radar) leads to improved results
in comparison to the reference DEM.
58
The overall methodology adopted in this paper has evaluated the delineation of
the watershed boundary comparing with each other and proved that, ASTER is the best
source of data for the delineation.
Based on the above testing and comparisons made, it is also strongly felt that,
the future researcher can straight away use the ASTER data for any Hydrological and
Land use and land evaluation studies.