Surveying and Spatial Information Systems, the

Hsing-Chung Chang

School of Surveying and Spatial Information Systems

University of New South Wales

Sydney, Australia

BIOGRAPHY

Hsing-Chung Chang is a Ph.D. student in the School of Surveying and Spatial Information Systems, the University of New South Wales. He received a Bachelor of Electrical and Electronic Engineering from the University of Auckland, New Zealand in 2002, and a Master of Engineering Science in Photonics from the University of New South Wales in 2003. He commenced his Ph.D. study in March 2003. His current research interests include applications of radar interferometry for digital elevation model generation and ground surface displacement monitoring caused by underground mining activities and earthquakes, as well as the combinations of the technologies of radar interferometry, GPS and geographic information system (GIS).

ABSTRACT

High resolution digital elevation models (DEMs) have been used in many applications such as civic planning, military mapping and navigation, natural hazard risk assessment, to name only a few. It has been a long history of using photogrammetry and dense ground surveys to draw the contour lines of the terrain and create the elevation model. Nowadays high precision DEMs with the vertical resolution from several meters to several centimeters can be derived using various remote sensing techniques, such as photogrammetry, radar interferometry, airborne laser scanning (ALS) and high resolution space imaging.

Ground control points (GCPs) that are normally obtained using traditional surveying methods and Global Positioning System (GPS) are used for DEM calibration and validation. This paper demonstrated the use of real-time kinematic GPS (RTK-GPS) to measure the elevation of terrain over an open unobstructed area, together with traditional levelling survey data for validating various high precision DEMs. One photogrammetric DEM with 1 arc-second resolution, two repeat-pass interferometric synthetic aperture radar (InSAR) DEMs derived from both C- and L-band satellite data, the C-band Shuttle Radar Topography Mission (SRTM) DEM and lastly a ALS DEM were compared against the RTK-GPS and levelling surveys. The results showed that ALS DEM has the highest height accuracy with a RMS error of less than 0.2m, followed by photogrammetric and SRTM DEMs with the RMS errors of less than 7m and 12m respectively. Finally the repeat-pass InSAR DEMs have the RMS errors between 10~30m. The sensitivity of C-band InSAR DEM to the vegetation can also be identified in the results.

INTRODUCTION

A high resolution digital elevation model (DEM) enables easy derivation of subsequent information for many applications such as urban planning, flood risk assessment, planning of electrical power tower and distribution lines, and mobility mapping in military. The majority of DEMs are generated by photogrammetry which has the longest history among the other new promising remote sensing techniques, such as high resolution stereo space imaging (Vassilopoulou et al. 2002), interferometric synthetic aperture radar (InSAR) (Zebker et al. 1994) and airborne laser scanning (ALS) (Turton and Jonas 2003). Photogrammetric DEMs are derived from the past using the airborne film photos to the present using digital imagery. As the high-resolution satellite imagery such as IKONOS and QuickBird have become available for commercial use, the stereo images from such space imaging systems have been used to generate high precision digital surface models (DSMs). Outside the visible electromagnetic radiation spectrum, ALS and InSAR systems utilize the spectrum in infrared and microwave bands to scan the earth surface hence to derive the topographic information.The DEMs derived from the various methods listed above have different horizontal (image pixel size) and vertical (height) resolution. As a rule of thumb, the DEMs in smaller pixel size would give higher accuracy in height, but they also come with greater price. Therefore, examining the accuracy of the multi-source DEMs provides users the knowledge to choose the adequate DEMs for various purposes.

Ground control points (GCPs) that are generally obtained using traditional surveying methods and static Global Positioning System (GPS) are used for DEM calibration and validation. These methods provide millimeter accuracy in height but they are also very labor intensive and time-consuming processes when surveying over a large area is required. Nowadays real-time precise positioning can be achieved even when the GPS receiver

is in motion. This is referred to as real-time kinematic systems or RTK-GPS. It is used as the solution for many time-critical applications such as engineering surveying and high precision navigation and guidance (Rizos et al. 2003).

This paper selected five DEMs derived using photogrammetry, InSAR and ALS techniques. The field surveys using levelling and RTK-GPS were conducted and used here to validate the height accuracy of these DEMs. The potential of using network RTK-GPS is also addressed in the latter section in this paper.

IMAGE DATA AND DEM GENERATION

One photogrammetric DEM, two repeat-pass interferometric synthetic aperture radar (InSAR) DEMs derived from both C- and L-band satellite data, the C-band Shuttle Radar Topography Mission (SRTM) DEM and lastly one ALS DEM are available over the test-site, Appin, which is about 75 km south-west from Sydney. This section is going to explain the generation and characteristics of these DEMs in details. Photogrammetric DEM

Photogrammetry has already proven its efficiency for a range of mapping applications including the production of orthophotos, cartographic maps, DEMs, etc. The generation of a DEM using photogrammetric principles has two operational parts: firstly the measurement phase, and secondly the derivation of the DEM. The main data source is from aerial photography, either film-based or digital. Then DEM points can be identified from the stereo-pair of aerial photos based on feature matching.

The photogrammetric DEM used in this paper was derived from film-based stereo aerial photograph. 3-D analogue plotter was used to trace the contours on the ground surface. It was then re-digitized the elevation into grid with 1 arc-second pixel size which is approximately equivalent to 30m. This photogrammetric DEM over the test-site is shown in Figure 1 (a).

InSAR DEM

Synthetic aperture radar (SAR) is a side-looking active radar-ranging system. It uses the microwave portion of the electromagnetic spectrum, encompassing frequencies ranging from 0.3~300GHz (or 1m~1mm in wavelength). Each SAR image contains information of both the amplitude and phase of the back-scattered signals from the ground objects.

Interferometric SAR (InSAR) utilizes the phase information contained in the two SAR images acquired over the identical scene at slightly different viewing locations. These two images can be acquired either at the same time by using two separate antennas mounted on the platform of aircraft or spacecraft, or acquired separately in time by re-visiting the scene with a single antenna. The latter is used typically in satellite radar systems and generally referred as repeat-pass InSAR.

Three InSAR DEMs were used for assessment in this paper. Two of them were generated using repeat-pass InSAR with the European ERS-1/2 and Japanese JERS-1 SAR satellite images respectively. The other was generated by using the single pass method in a US Shuttle Radar Topography Mission (SRTM) in 2000.

Repeat-pass satellite InSAR

Different wavelengths of electromagnetic radio waves interact with the ground objects differently, especially with the various types of vegetation. As a rule of thumb, the longer wavelength would penetrate through the canopy of vegetation further. The test site has a mixture of land-cover with farm-land, forest and built-up areas. Both European ERS-1/2 (C-band at the wavelength of 5.6cm) and Japanese JERS-1 (L-band at the wavelength of 23.5cm) SAR images were used to generate the DEMs here.

ERS-1/2 tandem mission in which the ERS-1 satellite was followed by ERS-2 in the same orbit with 24 hour time delay has been widely used for DEM generation. The advantage of the short time difference between the two acquisitions gives high conservation of coherence compared to other repeat-pass InSAR process. Comparatively, JERS-1 has the repeat cycle of 44 days but its longer wavelength make it less sensitive to the change of vegetation.

(a)

(b)

(c)

(d)

(e)

Figure 1. DEMs over the test-site derived by: (a) photogrammetry, (b) InSAR ERS-1/2 tandem, (c) repeat-pass JERS-1, (d) C-band SRTM, and (e) ALS.

The flowchart of InSAR process is shown in Figure 2. The two SAR images have to be co-registered together at sub-pixel level (1/8 ~ 1/10 of the pixel size) and the phase difference between the pixels in both images, so-called interferogram, can then be calculated.

The ambiguity height, the amount of height change when there is a 2π phase change in interferogram, can be calculated from Equation (1) with φ∆= 2π (Rosen et al. 2000). Therefore, the sensitivity to the terrain height is proportional to the perpendicular baseline and inversely to the wavelength of the microwave used by the system. It means that choosing shorter wavelength and/or longer perpendicular baseline distance will increase the sensitivity of InSAR system to the variation of terrain. However, the long baseline also leads to spatial decorrelation and hence introduces the phase noise to the system. This can be solved by carefully choosing the interferometric image pairs.

θ

λπφsin 4R H

B ∆=

∆⊥ (1)

where

∆φ: variation of phase in interferogram.

B

⊥

: perpendicular baseline between the two SAR acquisitions. ∆H : variation of terrain. λ: wavelength.

R: the range between satellite and ground object. θ: radar look angle

The phase fringe variations track the topographic contours. It is however a relative phase difference between pixels and “wrapped” between [-π, π] with many multiples of 2π. The “absolute phase” can be determined by “phase-unwrapping” the interferogram and re-inserting the constant multiple of 2π based on a height control point. Finally, the DEM can be generated by assigning the geographic coordinate to each pixel with geo-coding process.

The details of the two interferometric pairs chosen for DEM generation using ERS-1/2 tandem mission and repeat JERS-1 images are shown in Table 1. The derived InSAR DEMs are shown in Figure 1 (b) and (c) accordingly. Both DEMs have the pixel size at approximately 20m.

Satellite Date (yyyy/mm/dd) ⊥

B (m)

Time Difference

(days)

ERS-1 1995/10/29 - - ERS-2 1995/10/30 49 1 JERS-1 1995/04/21 - - JERS-1 1995/06/04 487 44

Table 1. Two interferometric pairs using ERS-1/2 tandem mission and repeat-pass JERS-1 for InSAR DEM generation.

Shuttle Radar Topography Mission

In addition to the satellite systems, an 11 day Space Shuttle Radar Topography Mission (SRTM) was successfully flown in February 2000. This mission used InSAR with signals in the C (5.6cm) and X (3cm) bands of the microwave spectrum to create the first DEM of the entire earth, in the latitude band 60°N to 57°S. Unlike SAR satellite systems, SRTM used two antennas separated 60 m apart to image the earth’s surface instantaneously (Rabus et al. 2003).

The C-band has a swath width of 225 km while the X-band was only limited to 45 km therefore the coverage of X-band is limited. As the fact that X-band wavelength cannot penetrate the vegetation and C-band wavelength will be reflected at the top of the canopies, the elevation measured by SRTM is also referred to as a digital surface model (DSM) which represents the height of the ground surface objects including vegetation.

The C-band SRTM DEM data can be downloaded from the USGS EROS Data center. By combining the data from both ascending and descending orbits, the topography data with a post spacing of 1 arc-second (approximately 30m) is released for the coverage within United States and 3 arc-second (approximately 90m) resolution data are available elsewhere. The C-band SRTM data for Australia, New Zealand and South Pacific Islands were released in October 2004. However, the test-site in this paper was not covered by the X-band SRTM data, so that only the C-band data was tested here. The C-band SRTM data for the test-site is shown in Figure 1 (d).

Airborne Laser Scanning (ALS)

ALS is a member of the so-called “Light Detection Ranging” (LiDAR) group of surveying methodologies that include airborne laser profiling and terrestrial laser scanning. Data is collected by the laser scanner mounted on the airplane as a stream of discrete reflected laser points from the ground. The system also exploits GPS, and usually an inertial measurement unit (IMU), to give precise position, attitude and acceleration information of the aircraft. At least two recordings, the first and last received signals, of each of the reflected laser points are recorded. By determining the difference between the two received signals, the height of trees or buildings can also be measured (Turton and Jonas 2003).

In general ALS derives height accuracies at grid points ranging from 0.1 ~ 0.5m, and horizontal geo-referencing accuracies ranging from 0.3 ~ 1.5m, with typical point spacing ranging from 0.2 to 4m (Turton and Jonas 2003). These accuracies are dependent upon the characteristics

Only a partial area of the test site is covered by ALS data as the swath of the system is small and the cost is higher for the same coverage by comparing to the radar system. The original of ALS data are simply the points with 3-D coordinate. The density of data is about 1 ~ 2 points per square meter and have been converted to a DEM model with pixel size 5m. The ALS DEM of this test-site overlaid with the aerial orthophoto is shown in Figure 1 (e).

GROUND SURVEY: LEVELLINGS AND RTK-GPS Levelling

Levelling is the most widely used method for obtaining the elevations of ground points relative to a reference datum and is usually carried out as a separate procedure to those used in fixing planimetric position. The height of a point can be measured at the accuracy of one millimeter. Three gridlines were surveyed in the 90’s and their locations of the gridlines are as shown by the yellow lines in Figure 3. The average separation between two pegs along the gridline is between 20~25m. The numbers of pegs along Line-A, B and C are 98, 49 and 275 respectively. RTK-GPS

RTK-GPS can deliver instantaneous point coordinates with centimeter-level accuracy. There are many applications that take advantage of RTK technology, including topographic surveying, engineering construction, geodetic control, vehicle guidance and automation, etc. (Riley et al. 2000).

RTK positioning uses a static GPS receiver as a reference station located at a known point. Another receiver is used

as the rover which can move and survey any point of interest within a limited range from the reference station. Both receivers make observations of the GPS signals at the same time and a radio data link between the two receivers permits data to be sent from the reference to rover, where the calculation of coordinates is carried out.

For the field survey at the test site, the reference receiver station was set up on top of a hill at a pillar as indicated

by a light blue dot near the centre of the image shown in Figure 3. The rover receiver was mounted on a car roof so that the survey could be easily conducted by driving along roads within the radio link coverage, which is approximately 8km from the reference station in this case study. The radio link became weak and unstable after 8km therefore a circle with the radius of 7 km from the reference station was drawn in Figure 3 as the indication

of the effective coverage of the RTK-GPS system. Only the RTK data having 3-dimensional accuracy better than

5cm were selected and used for DEM quality assessment.

Figure 3. The 3-D view of the test-site with 4 GPS-RTK routes shown in white with the location of the reference station indicated by a light blue dot and 3ground surveying gridlines for leveling shown in yellow. A coverage of 7km from the reference station was indicated by a light blue circle. The aerial photo has been draped on the photogrammetric DEM and exaggerated 3 times in the vertical direction.The assessment of the precision of the GPS-RTK survey was made by comparing the data collected along the same paths, either the round-trip path along both sides of the road or two repeated measurements collected from two independent RTK field surveys. The assessment indicates that the variation between the two trajectories is in the range of 15 ~ 26cm. This variation is the composite of the GPS-RTK measurement error at a static point, the height difference caused by the slope of the road, and the influence of surrounding obstructions such as trees or buildings. An example of the difference in height measured along the two sides of the road is given in Figure 4. The effect of tilted road surface is strongly suggested by the offset in height shown in Figure 4, with a RMS of 26cm.

Figure 4. GPS-RTK survey data collected along the two sides of the road.

DEM ACCURACY ASSESSMENT

The DEMs and the ground truth data that are levelling and RTK-GPS data were compared and analyzed with the aid of geographic information system (GIS) software. As the data may be originally projected in the different coordinate systems, it is essential to convert all the data into the same geographic coordinate system before any comparisons and analysis can be made. After geo-referencing all the data to the same coordinate system, the elevation profiles along the 3 levelling gridlines and 4 RTK-GPS routes as shown in Figure 3 were extracted from each DEMs. The profiles along the levelling gridlines and the RTK-GPR routes are shown in Figure 5 and 6 respectively.

Levelling.

GPS.

Error (m) Ref. Type Line

DEM

Min Max.RMS

Photo 0.11 20.18 6.39 Tandem 3.95 54.87 28.36 JERS-1 0.37 30.67 16.09 Levelling

A

SRTM 0.11 16.7 5.94 Photo 0.06 9.08 3.09 Tandem 0. 41.1 24.19 JERS-1 0.47 32.79 16.61 Levelling B

SRTM 0.07 27.67 10.88 Photo 0.01 13.49 3.67 Tandem 0.01 40.36 18.34 JERS-1 0.03 41.03 16.24 Levelling C

SRTM 0.01 28.55 11.02 Photo 0.01 7.35 2.96 Tandem 0.01 41.07 18.02 JERS-1 0.01 36.7 9.69 RTK-GPS

1

SRTM 0.07 3.4 1.57 Photo 0.02 7.79 3.27 Tandem 0.07 50.63 30.28 JERS-1 0.01 36.45 12.28 RTK-GPS

2

SRTM 0.01 6.9 2.03 Photo 0.01 5.51 2.17 Tandem 0.16 35.05 17.31 JERS-1 0.31 44.55 17.27 RTK-GPS

3

SRTM 0.01 8.88 3.19 Photo 0.04 5.54 2.75 Tandem 1.63 27.32 15.6 JERS-1 3.26 36.6 18.32 SRTM 0.01 8. 3.3 RTK-GPS

4

ALS 0.01 0.77 0.18

Table 2. The summary of the errors of the elevation profiles extracted from the DEMs against levelling and RTK-GPS surveying data.

The 3-D aerial photograph of the test-site in Figure 3 shows levelling gridline-A, B and approximately half of C were surveyed within forest. By contrast, the RTK-GPS would be best surveyed over the areas with open and unobstructed sky view in order to have sufficient line-of-sight to GPS satellites. The errors between each elevation profiles drawn from the DEMs against the levelling and RTK-GPS surveys are summarized in Table 2.

As mentioned earlier C-band wavelength does not penetrate through vegetation, therefore the interferometric DEMs generated with C-band would represent the height of the terrain including the height of vegetation. This can be noticed from the profiles drawn from the gridlines. Figure 5 (a) and (b) shows that in average the elevation profiles drawn from JERS-1 L-band InSAR DEM are lower than the ones drawn from ERS-1/2 tandem C-band InSAR DEM. The tandem DEM also has less RMS errors against RTK-GPS data in the open areas in contrast to the errors along the gridlines over the forest, but it is not so evident in JERS-1 results.

Figure 5 shows a sudden and unexpected increase in height of the JERS-1 DEM across one of the mine subsidence along gridline C. In order to examine the accuracy of the DEM without being biased the height values extracted from peg 140~200 were disregarded. As summarized in Table 2, the height accuracy of InSAR DEMs using ERS-1/2 and JERS-1 imagery is limited to 10~30m.

SRTM DEM which is also generated by radar interferometry with C-band wavelength however shows higher accuracy in height by using the two SAR antennas to receive the two SAR images of earth simultaneously. Even though, the sensitivity of C-band wavelength to the vegetation is still evident in reducing RMS errors in the open areas along the RTK-GPS routes (RMS < 4m) by comparing to the gridlines (RMS > 6 and 10m) as shown in Table 2. Due to the larger pixel size of 3 arc-second, SRTM DEM tends to smooth out the terrain and consequently it does not reflect the sudden change of the terrain over gorge or valley.

ALS data has the best height accuracy among all these DEMs with the RMS error of 0.18m. That is however close to the height accuracy of RTK-GPS technique. Therefore, a larger coverage of ALS DEM is needed and denser RTK-GPS data and other ground control points using static GPS or levelling are required in order to make further examination on the accuracy of ALS data in both vertical and horizontal planes.

Even though four satellite receptions are enough to resolve a point location in three dimensions, for high precision RTK-GPS 5 satellites are necessary. It can drop below 5 satellites sometimes due to the poor geometry of the satellites or the obstructed view of sky during survey. If the real-time correction for the ambiguity is not necessary for the surveying purpose, the high precision coordinates can be achieved by applying post-processing as kinematic GPS. The raw GPS data at both rover and reference station have been logged and post-processed. The results showed that the kinematic mode does not improve the accuracy in height significantly, but it resolved the ambiguity of some point samples that could not be done using RTK mode. Therefore, more data can be used for DEM validation using kinematic GPS. FURTHER WORK: NETWORK RTK-GPS

The limitation of the single base RTK is the distance between reference station and the rover due to distance-dependent biases such as orbit, ionosphere and troposphere bias (Rizos et al. 2003). But when a static reference station is available within the range of 10 ~ 15km from the survey site, then the single base RTK is still preferred.

Network RTK is a centimeter-accuracy, real-time, carrier phase-based positioning technique capable of operating over inter-receiver distances (the distance between the rover and the closest reference station receiver) up to many tens of kilometer with equivalent performance to current single-base RTK systems. The SydNet continuously operating reference stations (CORS) network has currently been established in the Sydney metropolitan area in Australia with network-based positioning capability including Network-RTK (Rizos et al. 2003).

There are 7 reference stations proposed for SydNet with 15-30km spacing between any of the two stations. The test-site being demonstrated in this paper is also covered by SydNet. The future development of this study will be to use Network-RTK to extend the survey area without being restricted to the limited range of the single base RTK systems.

CONCLUDING REMARKS

One photogrammetric DEM with 1 arc-second pixel size, two repeat-pass interferometric synthetic aperture radar (InSAR) DEMs derived from both C- and L-band satellite data, the C-band Shuttle Radar Topography Mission (SRTM) DEM and lastly a ALS DEM were compared against the RTK-GPS and levelling surveys.

Both levelling and RTK-GPS surveys are used in this paper as ground truth data for validating the DEMs. Levelling has millimeter height accuracy while the RTK-GPS samples having the 3-D accuracy better than 5cm were used for DEM quality assessment.

From the height profiles extracted from the DEMs along 3 gridlines and 4 GPS-RTK routes, the results showed that ALS DEM has the best RMS error of 0.18m against GPS-RTK data. Photogrammetric DEM has the RMS errors of 2.2 ~ 6.4m and 1.6 ~ 11m for SRTM DEM. The repeat-pass InSAR DEMs have the RMS errors of 15.6~30.3m and 9.7~18.4m for C- and L-band wavelengths respectively. The results also showed that the impact of vegetation to the InSAR DEMs derived from C-band signal is evident.

For future work, it is possible to use SydNet for Network-RTK to expend the coverage of surveying without being restricted to the limited range of the single base RTK systems after the SydNet has become operational.

ACKNOWLEDGMENTS

The author is supported by a PhD scholarship from the Cooperative Research Centre for Spatial Information (CRC-SI). The author wishes to thank all members of Satellite Navigation And Positioning Group (SNAP). I also like to thank Mr. Andrew Nesbitt of BHPBilliton for providing the GIS data and levelling surveying measurement, Prof. Makoto Omura of Kochi Women’s University, Japan, for providing the JERS-1 SAR imagery, ESA and ACRES (the Australian Centre for Remote Sensing, one of the participants of CRC-SI project) for providing the C-band SAR images.

REFERENCES

Huising, E. J. and L. M. Gomes Pereira (1998). "Errors and accuracy estimates of laser data acquired by

various laser scanning systems for topographic

applications." ISPRS Journal of Photogrammetry and

Remote Sensing 53(5): 245-261.

Rabus, B., M. Eineder, A. Roth and R. Bamler (2003).

"The shuttle radar topography mission--a new class

of digital elevation models acquired by spaceborne

radar." ISPRS Journal of Photogrammetry and

Remote Sensing 57(4): 241-262.

Riley, S., N. Talbot and G. Kirk (2000). A new system for RTK performance evaluation. Position Location and

Navigation Symposium, IEEE 2000,231-236,13-16

March 2000.

Rizos, C., D. Kinlyside, T. Yan, S. Omar and T. A. Musa (2003). Implementing network RTK: The SydNET

CORS infrastructure. 6th Int. Symp. on Satellite

Navigation Technology Including Mobile Positioning

& Location Services, Melbourne, Australia,CD-ROM

proc., paper 50,22-25 July.

Rosen, P. A., S. Hensley, I. R. Joughin, F. K. Li, S. N.

Madsen, E. Rodriguez and R. M. Goldstein (2000).

"Synthetic aperture radar interferometry."

Proceedings of the IEEE 88(3): 333-382. Turton, D. and D. Jonas (2003). Airborne Laser Scanning - Cost Effective Spatial Data. Map Asia Conference

2003, Session : Photogrammetry and LIDAR, Kuala

Lumpur, Malaysia,Page reference number: AD1,13 ~

15 October 2003.

Vassilopoulou, S., L. Hurni, V. Dietrich, E. Baltsavias, M.

Pateraki, E. Lagios and I. Parcharidis (2002).

"Orthophoto generation using IKONOS imagery and

high-resolution DEM: a case study on volcanic hazard monitoring of Nisyros Island (Greece)."

ISPRS Journal of Photogrammetry and Remote Sensing 57(1-2): 24-38.

Zebker, H. A., C. L. Werner, P. A. Rosen and S. Hensley (1994). "Accuracy of topographic maps derived from

ERS-1 interferometric radar." Geoscience and Remote Sensing, IEEE Transactions on 32(4): 823-

836.