Bambang Trisakti and Ita Carolita
Research Officer of Indonesian National Institute of Aeronautic and Space (LAPAN)
Email: [email protected]
This paper explains a method to generate DEM from ASTER (Advance Space borne Thermal Emission and Reflection Radiometer) stereo data and evaluates the generation of ASTER DEM and SRTM DEM (Space Shuttle Radar Topography Mission) with 90 m spatial resolution. ASTER DEM is generated from 3n (nadir) and 3b (backward) level 1b, with 10 ground control points (XYZ coordinate) derived from ASTER RGB 321 geometric-corrected and SRTM DEM. Almost all tie points are collected automatically and several tie points are added manually. The triangulation and DEM extraction process are done automatically by using ERDAS Imagine Software. DEM Evaluation is carried out by comparing between ASTER DEM and SRTM DEM for the height distribution of vertical and horizontal transect lines and the height value of the whole DEM images. The process is continued by analyzing the height differences between ASTER DEM and SRTM DEM. The results show that ASTER DEM has 15 m spatial resolution with height differences less than 30 m for about 67% of total area, and absolute mean error is 27 m (compared with SRTM DEM). This absolute mean error is large enough, because the GCP’s used in this study are only in a small amount and most of study area is in the high terrain area (mountainous area) with dense vegetation coverage.
ASTER (Advance Space borne Thermal Emission and Reflection Radiometer) on board of Terra spacecraft is multi spectral optical sensor that is launched on December 1999. ASTER sensor has 14 spectral bands that range from visible to thermal infrared band. All spectral bands of ASTER are divided into three radiometers: VNIR (Visible Near Infrared Radiometer), SWIR (Short Wave Infrared Radiometer) and TIR (Thermal Infrared Radiometer) (Ersdac, 2003). VNIR has a high performance, high resolution optical instrument (15 m spatial resolution) which is able to detect reflection from the ground surface ranging from visible to near infra red (0.52 – 0.86 m). VNIR has 2 (two) near infra red bands which have similar wavelengths, those are 3n (nadir looking) and 3b (backward looking). The 3b band is used to achieve the backward looking, with setting angle between the backward looking and the nadir looking is design to be 27,60° (Ersdac, 2002). The objection of band 3b addition is to obtain stereoscopic image that will be processed to generate the height of land surface or DEM (Digital Elevation Model).
DEM of land surface provides significant information for many research activities and important data as the input of image processing and image analysis, such as image correction due to height of land surface (Ortho rectification), contour mapping, 3D images generation, disaster management (determination of vulnerable area due to landslide, flood and tsunami disaster), monitoring land subsidence phenomenon and many others. Therefore, the capabilities of ASTER stereo image that provides DEM with high spatial resolution (15 m) is very important for remote sensing and GIS (Geographic Information System) users to enhance the accuracy of desired height information. The method of DEM generation from satellite stereo image (Such as ASTER, SPOT, etc) and the accuracy of generated DEM have been published in many research reports, for examples: (Goncalves and Oliveira, 2004; Tsakiri-Strati et al, 2004; Pantelis et al, 2004; Ulrich et al, 2003). Several study results show that the vertical accuracy of ASTER DEM approaches to 25 m, but in area with less vegetation coverage, the accuracy can rise approximately to 9-11 m (Goncalves and Oliveira, 2004; Richard Selby, PCI Geo). Furthermore, Ulrich et al (2003) have reported that ASTER DEM has better accuracy for medium scale mapping (1:100.000 and 1:50.000).
This paper describes DEM generation method from ASTER stereo image, and evaluates the accuracy of generated ASTER DEM by comparing it to SRTM DEM (Shuttle Radar Topography Mission) with 90 m spatial resolution. DEM is generated from 3n (nadir looking) and 3b (backward looking) band of ASTER data level 1b by using Ground Control Points (XYZ coordinate) from ASTER RGB 321 geometric-corrected image and SRTM DEM. Furthermore, evaluation of DEM accuracy is done by analyzing the height different between ASTER DEM and SRTM DEM.
This study uses level 1b ASTER data (Source: ERSDAC and Indonesia RS-GIS Forum) with 15 m spatial resolution, SRTM DEM (Source: obtaining from USGS website) with 90 m spatial resolution and ASTER RGB 321 geometric-corrected image based on IKONOS image. The Study area is located in Lhok Nga (Nanggroe Aceh Darussalam Province, Indonesia) as shown in Picture 1. The coastal region of this area is affected by tsunami that happened on 26th December 2004. Data pre-processing is started by cropping the interest area for 3n (nadir looking) and 3n (backward looking) band. Then both images are rotated 270 degrees clockwise. Destriping process should be done for level 1a of ASTER data, but it is no need to be done for level 1b data, since this level has been corrected from striping distortion. The next processes are done by using Erdas Imagine Software (Orthobase-Pro module).
The processes are started by pyramid layer making, and then insert appropriate sensor model and orientation parameters (such as: side incidence, sensor column, pixel size, etc) which some of them can be obtained from ancillary data. Ground Control Points (GCPs) collecting (XY coordinate point and height (Z) point) is carried out by using ASTER RGB 321 geometric-corrected image based on IKONOS for XY references and SRTM DEM for Z references. This study uses 10 GCPs, and those GCPs will be used as the reference for tie point making. Almost all tie points are collected automatically and several tie points are added manually. The next step is triangulation process which aims to relate the XY points on image, GCPs and sensor specification information, so that the formulation of these 3 (three) related parameters can be developed. The last step is generating DEM from the overlapping area between 3n and 3b bands.
Picture 1 Study area and image of band 3n and 3b of ASTER data
The Accuracy of the generated ASTER DEM is evaluated by comparing height value of ASTER DEM and SRTM DEM. The vertical and horizontal transect lines are drawn along the both DEM images, and then the height distribution of each transect lines are compared. Finally, the height differences between ASTER DEM and SRTM DEM are extracted and classified to investigate mean absolute error between the both DEM images. Classification is done in land area only (water area is not included).
3. RESULTS and DISCUSSION
Pictures 2a and 2b perform ASTER DEM images are produced from ASTER stereo data, as well as 3D images of ASTER RGB 321 composites. ASTER DEM is generated from 3n and 3b stereo bands using 10 GCPs. ASTER DEM has 15 m spatial resolution, and the contour of DEM looks softer compare to contour of SRTM DEM that only has 90 m spatial resolution. Blue color shows the area with lower height (lower terrain). The gradation from blue to red color shows that the heights of land surface increase. Relative height obtained from ASTER DEM in the study area achieves to 752 meter. The 3D images in picture 2b show union result from DEM data and ASTER RGB 321 composite images, whereas the most of study area consists of mountain areas with dense vegetation coverage area, in the contrary the land that has lower height and less vegetation coverage area is located in certain small coastal area.
Picture 2 DEM ASTER and 3D image of ASTER RGB 321 composite
Picture 3 Comparison of height distribution of ASTER DEM and SRTM DEM along vertical and horizontal transect lines
Evaluation to the accuracy of ASTER DEM is done by comparing height distribution along transect lines vertically and horizontally. Picture 3 shows vertical and horizontal transect lines in ASTER images, and comparison of height distribution between ASTER DEM and SRTM DEM along the transect line’s track. Height distributions of the both DEM data are almost same along vertical and horizontal lines. On the other hand absolute values of height are not same, especially in the high topography areas such as mountain with dense vegetation coverage areas.
Table 1 shows measurement results of ASTER DEM and SRTM DEM for whole images (full image). Minimum, maximum and mean values from ASTER DEM and SRTM DEM show relative same values, even though ASTER DEM has wider range values comparing to SRTM DEM. This result shows that ASTER DEM is more sensitive than SRTM DEM (It is related to spatial resolution of ASTER DEM higher than SRTM DEM). Negative values at ASTER DEM minimum values are pixel values located in the coastal water area, where some certain areas (such as: water and cloud area) will cause distortion to the generated DEM values.
Table 1 Measurement values of ASTER DEM and SRTM DEM
|Full image||Min Value (m)||Max Value (m)||Mean Value (m)|
Table 2 Measurement values of height difference of ASTER DEM and SRTM DEM
|ASTER-SRTM||Min Value (m)
Max Value (m)
|Mean Error (m)||Mean Absolute Error (m)|
Table 3 Classification results of height difference between ASTER DEM and SRTM DEM (classification is done in land area only)
|Height difference (m)||Percentage of coverage area|
|0 – 10 / -10 m||29 %|
|11 – 20 / -20 m||23%|
|21 – 30 / -30 m||15%|
|31 – 40 / -40 m||10%|
|41 – 50 / -50 m||7%|
|51 – 60 / -60 m||5%|
|> 60 / -60 m||11%|
The accuracy of generated ASTER DEM is analyzed by extracting the height difference between the both DEM images. Minimum and maximum values, mean error and mean absolute error values are calculated. Classification process of height differences between those both image is done to analyze percentage of coverage area for each height difference interval. Classification is done in land area only (water area is not included). Table 2 and 3 show measurement results of height difference between the both DEM images and coverage area percentage for each height difference interval. Measurement result shows that minimum and maximum height difference values are more than 100 m, meanwhile absolute mean error is up to 27 m (ASTER DEM value to SRTM DEM value). These results perform quite wide for error level, but it is not significant different compare to previous research’s results. The previous research has reported that ASTER accuracy is up to 25 m for mountain areas with dense vegetation coverage (Richard Selby, PCI Geo).
Classification result for 0-10 m of height difference is covering at 29% of land area, 11-20 m is at 23% and 21-30 meter is at 15%, therefore 0-30 meter for this field study is covering 67% of land area in whole DEM image. Meanwhile, 31-60 meter of height difference is at 12% and more than 60 meter is at 11% of land area. Obtaining results shows that the height difference between ASTER DEM and SRTM DEM data are quite big (the height difference more than 31 meter is up to 23% from the study area), but this difference can be less in certain ways, (to increase the accuracy of generated DEM) such are :
- More Ground Control Points is added to use them as reference. The correlation between XY point in image, GCPs and specification sensor information will be higher if more numbers of GCPs are used. Geosystem (2002) in Rob (2004) has recommended to use 54 GCPs or more to generate DEM from one ASTER scene data with high accuracy.
- Increasing GCPs accuracy. Small numbers of GCPs will be helped by increasing GCPs accuracy, such as using ground control point from high resolution map/image reference or from field measurement results.
This paper describes DEM generation method from ASTER stereo image and evaluates the accuracy of generated ASTER Dem by analyzing height different between ASTER DEM and SRTM DEM. Several results are obtained below:
- DEM with spatial resolution 15 meter can be generated from ASTER stereo image (3n (nadir looking) and 3b (backward looking)) level 1b using Erdas Imagine software (Othobase Pro).
- ASTER DEM that has been generated has mean absolute error up to 27 m (compared to SRTM DEM) with 0-30 m of height difference is covering 67% of land area in the study area. The mean absolute error and percentage of coverage area have lower accuracy than we expected. It is caused by the less of GCPs that be used in the DEM generation process and the most of area study are mountain side with high topography and dense vegetation coverage. Anyway, the accuracy level is not too different compared to the results in previous ASTER research reports (Richard Selby, PCI Geo).
- ERSDAC, 2002, ASTER User’s Guide Part-III (Ver. 1.0), Earth Remote Sensing Data Analysis Center, JAPAN.
- ERSDAC, 2003, ASTER Reference Guide Version 1.0, Earth Remote Sensing Data Analysis Center, JAPAN.
- Goncalves J.A. and Oliveira A.M., 2004, Accuracy analysis of DEMS derived from ASTER imagery, ISPRS XX, Istambul,Turkey.
- Pantelis M. and Ian D., 2004, A rigorous model and DEM generation for SPOT 5-HRS, ISPRS XX, Istambul,Turkey.
- Richard Selby, PCI Geo, Creating digital elevation models and orthoimages from ASTER Imagery, PCI Geomatics, United Kingdom.
- Rob Van Ede, 2004, Destriping and Geometric Correction of an ASTER level 1A Image, Faculty of GeoSciences, Dept. of Physical Geography, Utrecht University
- Tsakiri-Strati M., Georgoula O. and Patias P., 2004, DEM evaluation generated from HRS SPOT 5 data, ISPRS XX, Istambul,Turkey.
- Ulrich K., Tobias B. and Jeffrey O, 2003, DEM generation from ASTER satellite data for geomorphometric analysis of Cerro Sillajhuay Chile/Bolivia, ASPRS 2003 annual conference proceeding, Anchorage, Alaska