Combination of Multiple Repeat Orbits of ENVISAT for Mining ...

Combination of Multiple Repeat Orbits of ENVISAT for Mining Deformation Monitoring H.C. Chang, L. Ge, A.H. Ng, C. Rizos School of Surveying and Spatial Information Systems, The University of New South Wales, Sydney, NSW 2052, Australia H. Wang Department of Surveying Engineering, Guangdong University of Technology, Guangzhou 510006, China M. Omura Department of Environmental Science, Kochi Women's University 5-15, Eikokuji-cho, Kochi 780-8515, Japan Abstract.

In

conventional

differential

radar

differential

InSAR

(DInSAR)

and

persistent

interferometry (DInSAR), the ground surface

scatterer InSAR (PSInSAR), have demonstrated

displacement can be measured along the line-of-

their

sight. In order to measure the vertical and horizontal

information on changes to the earth’s surface

displacements

same

geometry. These techniques are considered to be

deformation derived from at least two different look

more cost-effective than, and complementary to,

directions are required. This study utilised the

conventional ground-based geodetic methods. The

DInSAR results generated from ENVISAT data

authors have been studying mine subsidence using

acquired from three different look angles to

DInSAR in the State of New South Wales, Australia

determine the 3-D displacement vectors resulting

for several years. For example, a mine subsidence

from surface deformation due to underground

profile extracted from a JERS-1 DInSAR result

mining. However, The ENVISAT results showed

illustrated a root mean square error (RMSE) of

ambiguous phases due to high phase gradients in

1.4cm against the ground levelling data (Ge, et al.

the differential interferogram. The newly acquired

2007). In addition, sub-centimetre accuracy of

ALOS/PALSAR

DInSAR

is

tandem pairs (Chang, et al. 2005, Ge, et al. 2007).

can be overcome by using interferometric signals

The results showed that a maximum subsidence of

with

1cm was developed during 24 hours, which was

finer

here

geodetic

DInSAR results was demonstrated using ERS-1/2

and

used

the

determining

to

wavelength

also

of

for

demonstrate how the high phase gradient problem longer

data

results

capabilities

imaging

resolution.

confirmed by the surveyor from the mining company.

Keywords. DInSAR, 3-D deformation vectors, ENVISAT, ALOS, mining deformation

As in most DInSAR studies for land deformation monitoring, the deformation is measured along the line-of-sight (LOS) between the radar antenna and the ground reflecting objects. Then the subsidence (vertical surface deformation) is derived under the assumption of negligible horizontal deformation.

1 Introduction Radar

interferometry

However, in-situ ground survey data shows that techniques,

including

interferometric synthetic aperture radar (InSAR),

underground mining activity may also cause horizontal surface displacement, which may be

greater than 20-30mm. In addition, this assumption

DInSAR results derived from both ascending and

of

descending images need to be combined. If the

negligible

horizontal

movement

is

more

applicable to the ERS-1/2 data as the look angle of o

deformation

is

purely

vertical,

the

vertical

ERS-1/2 is 23 . The smaller look angle makes the

displacements derived from both orbits should be

system more sensitive to vertical deformation than

the same, although the displacements measured

horizontal ones.

along the LOS may vary depending on the look

In order to study both vertical and horizontal mine

deformation,

multiple

ascending

and

angle

of

radar.

Otherwise,

the

horizontal

displacement has a signature.

descending orbits of ENVISAT/ASAR data are

The new spaceborne SAR sensors have various

analysed in this paper. The variation of the swath

scanning modes with a range of incidence angles.

mode of ENVISAT/ASAR provides the opportunity

ENVISAT/ASAR permits image acquisitions of the

to investigate the 3-D surface deformation vectors

same region in seven different swath modes. Hence

with a range of satellite look angles ranging from

the same surface deformation can be measured not

o

o

15 to 45.2 .

only in both ascending and descending orbits, but also at different viewing angles. This study

2 Methodology

examines the feasibility of combining the DInSAR

InSAR utilises two coregistered SAR images of the identical scene and measures the phase difference of the same pixels in the two images. This phase difference product, the so-called interferogram, contains topographic information, variance of water vapour in the atmosphere and ground movement that has occurred between the dates of the two SAR images acquired (if repeatpass radar interferometry is used). DInSAR extracts the interferometric phase introduced by land surface deformation by eliminating other phase contributors in the interferogram, e.g. topography. The phase difference, Δφ, caused by the height deformation, Δd, along the LOS of the radar signal is shown in equation (1), where λ is the wavelength of the radar signal. So, one complete phase cycle (2π) in a differential interferogram represents λ/2 of height displacement along the LOS:

Ä" =

Usually

the

two descending orbits of ENVISAT to reveal the 3D deformation vectors of ground subsidence. The 3-D deformation vectors can be calculated from the DInSAR solutions derived from 3 independent LOS results. This was demonstrated by Sircar, et al. (2004) using ascending and descending Radarsat-1

SAR

image

data.

It

was

also

demonstrated by Fialko and Simons (2001) for 1999 Hector mine earthquake using ascending and descending ERS-1/2 differential interferometric results together with azimuthal offsets of radar amplitude pixels. The deformation vector along the LOS signal (D LOS) is a composite of up (DU), east (DE) and north (D N) deformation components. The deformation

measured

in

the

differential

interferogram is the sum of vertical and horizontal deformation components projected onto the LOS. The contributions of D U, DE and DN to DLOS are given in equation (2):

4!!d ë

amplitude

results derived from images from one ascending and

(1)

of

vertical

height

deformation vector or subsidence is derived by assuming negligible horizontal displacement. In order to resolve the true 3-D deformation vectors,

[' cos() )

& DU # $ ! sin() ) cos(( ) sin() ) sin(( )]$ D E ! = [ D LOS ] $% D N !"

(2)

where θ is the radar incidence angle and α is the

are listed in Table 2, where pairs 1, 2 and 3 were

azimuth of the satellite heading vector (positive

acquired in the swath modes IS4, IS3 and IS2,

clockwise from North). Note that D U is defined as

respectively. The maximum temporal difference is

negative for subsiding movement.

only 3 days between pairs 1 and 3. Therefore the

Fialko and Simons (2001) utilised the azimuthal offsets and DInSAR to measure the displacements

temporal coverage of the pairs is suitable for the current purpose of 3-D deformation analysis.

along the azimuth direction and the LOS of the

The DInSAR interferograms of pairs 1~3 were

ERS-1/2 signals. The former are mostly sensitive to

derived and are shown in Figure 1. Note the high

the North-South displacements; while the latter are

phase gradient near the centre of the subsidence

sensitive to the vertical, modestly sensitive to East-

bowl. The longer baseline of pair 3 caused higher

West,

North-South

spatial decorrelation. It is indicated by a higher

displacements. Therefore, the combination provides

level of phase noise in the interferogram. As the

a good geometry in order to separate the

local topographic variation

displacement vector components DU, DE and D N.

increased noise in pair 3 was not predominantly due

However, the measurement error due to the use of

to the topographic residual phases. Consequently,

azimuthal offsets is of the order of a fraction of the

the unwrapped phase of pair 3 was noisier than the

imaging pixel. In order words, with the along-track

results

pixel size of 4m for ENVISAT, the accuracy of the

deformation vectors derived from pairs 1 and 2

azimuth offsets method is at the decimetre level. It

were used for the 3-D deformation analysis, with

is not feasible to detect the horizontal displacement

the assumption of small horizontal movement along

of mine deformation, which has a magnitude of a

the

few to tens of millimetres.

deformation maps in the slant range direction are

and

weakly

sensitive

to

of

other

North-sSuth

pairs.

is

moderate, the

Therefore

direction.

The

only

the

geocoded

In this paper, three independent look angles of

shown in Figure 2. The estimated 3-D deformation

ENVISAT data were tested for the purpose of

vectors along the East, North and Up directions are

determining the D U, DE and DN of underground

shown in Figure 3.

mining induced deformation. The geometry of the method is less than ideal as it is less sensitive to the displacement along the North-South orientation due to the characteristics of the satellite orbit. It is, however, an alternative when the azimuthal offsets method is not applicable.

Table 1. Characteristics of ENVISAT and ALOS SAR sensors.

Sensor

Band

λ

Altitude

Resolution

ENVISAT ASAR ALOS PALSAR

C

5.6cm

786km

~30m

L

23.6cm

692km

~10m

3 Input Data and Results Three ENVISAT interferometric pairs were used for the 3-D deformation vector analysis. In addition, a recently received ALOS/PALSAR interferometric pair is tested and compared against the ENVISAT results. Some characteristics of ENVISAT and ALOS/PALSAR sensors are summarised in Table 1. The ENVISAT image acquisitions over the test site

Table 2. Three ENVISAT interferometric pairs used in this study, all with 35day temporal separation between master and slave images.

Pass

Pair

Master dd/mm/yyyy

Slave dd/mm/yyyy

Bperp (m)

desc asc desc

1 2 3

8/12/2006 10/12/2006 11/12/2006

12/01/2007 14/01/2007 15/01/2007

234 238 310

A previous study showed that ground subsidence measured by DInSAR using JERS-1 data can achieve an RMS error of 1.4cm when compared to the results of ground survey (Ge, et al., 2007). In the comparison, the ground survey line was not across the centre of the subsidence bowl, therefore the maximum subsidence measured along this survey line was about 16cm. According to the (a)

ground survey data provided by the mining

(b)

company (BHPBilliton), the mining activity would cause a maximum subsidence of the order of 860 to 930mm in the area. About 65% of the total subsidence (over 50cm) becomes evident within the first 2~3 weeks after excavation. This large vertical ground movement near the centre of the subsidence bowl

causes

the

phases

in

the

differential

interferogram to be ambiguous. This has a severe (c)

impact on the ENVISAT results during the phase

(d)

Fig. 1. (a) Averaged intensity SAR image of the test site, and (b) ~ (d) differential interferograms of pair 1, 2 and 3 in slant range projection.

unwrapping process. This problem can be solved or eased by having interferometric pairs with higher imaging resolution, signals with longer wavelength and/or a shorter site revisit cycle. This is confirmed in the next section using recent ALOSPALSAR

4 Validation

data. By comparing the ground survey data collected at a similar period, the ENVISAT results do not

5 High Phase Gradient Problem

reflect the true mine subsidence because the high phase gradient at the centre of the subsidence bowl cause the phase values to be ambiguous. The ambiguous phases may be smoothed by filtering the interferogram. Therefore the discontinuous or wrapped

phase

values

in

the

differential

interferograms cannot be recovered without errors. Similar

findings

of

underestimating

mine

subsidence using ERS-1/2 DInSAR results were reported in other studies (e.g. Spreckels, et al., 2001). This problem can be eased by having a longer wavelength radar signals or a shorter repeat cycle of the radar image acquisition system.

The

maximum

phase

gradient

in

an

interferogram that avoids incoherence was defined in Massonnet and Feigl (1998). It is one fringe per pixel. The maximum phase gradient is also constrained when phase unwrapping the differential interferogram. The common phase unwrapping techniques assume the phase difference between the two adjacent pixels in the interferogram is less than π.

Otherwise

the

wrapped

phase

in

the

interferogram becomes ambiguous and cannot be unwrapped to produce continuous errorfree phase results.

Fig. 2. Deformation along the slant range direction of pair 1 (descending) and pair 2 (ascending).

Fig. 3. 3-D deformation vectors due to mining-induced subsidence over 35 days in the easting, northing and up directions.

In the application of DInSAR for monitoring

a width of 250m, the centre of the subsidence

deformation due to underground coal mining,, the

bowl should be about 125m from the edge of the

maximum subsidence usually occurs along the

panel (pillar). That is equivalent to about 4 pixels

centre of the longwall panel. At this test site the

for ENVISAT data (with a spatial resolution of

longwall panel has a width of about 200~250m.

30m). In order to avoid phase unwrapping errors,

Due to the rectangular layout of the longwall, the

the maximum height displacement over these 4

subsidence gradient across the longwall panel is

pixels has to be less than 4π. Therefore, better

much greater than along the longwall. Hence the

imaging resolution of SAR sensor improves the

phase gradient in the differential interferogram is

maximum detectable height displacement without

at its maximum in the direction across the

increasing phase unwrapping errors.

longwall panel.

Two recently received interferometric pairs of

Due to the longwall geometry, the centre of the

ALOS/PALSAR data are used in this analysis.

subsidence bowl normally lies along the centre of

ALOS/PALSAR can operate in five different

the panel. For example, for a longwall panel with

imaging modes: fine beam single polarisation

(FBS), fine beam double polarisation (FBD), direct downlink (DSN), ScanSAR and polarimetry (PLR). These two ALOS/PALSAR pairs were acquired in single and dual-polarisation modes,. For DInSAR processing the authors only used the data acquired in the FBS mode with 46 days separation (single repeat cycle of ALOS), as shown in Table 3. The differential interferogram of pair 4 shown in Figure 4 clearly indicates the fringes caused by ground subsidence at the same mine site. These are more distinguishable than the results derived from ENVISAT data. Due to the wavelength of 23.6cm and an incidence angle of 0

38.7 , the height displacement map in Figure 5

Fig. 5. Georeferenced mine subsidence map based on the ALOS differential interferogram of pair 4 in Fig. 4.

shows the maximum height displacement near the centre of the subsidence bowl as being over 40cm. The georeferenced DInSAR interferogram and subsidence map of pair 5 are shown in Figure 6.

Table 3. Interferometric characteristics of two pairs of ALOSPALSAR ascending data with an incidence angle of 38.70.

Pair

Master dd/mm/yyyy

Slave dd/mm/yyyy

Bperp (m)

Btemp (days)

4 5

27/12/2006 14/08/2007

11/02/2007 29/09/2007

530 501

46 46

(a)

(b) Fig. 4. ALOS differential interferogram of mining-induced subsidence generated from pair 4: 27/12/2006 - 11/02/2007.

Fig. 6. Georeferenced ALOS/PALSAR results of (a) DInSAR interferogram overlaid with mine plan, and (b) colour coded subsidence map overlaid on ALSO/PALSAR intensity image with mine plan, derived from pair 5: 14/08/2007 - 29/09/2007. The mine plan for the lower subsidence bowl is not yet available.

used to test their performance for mining-induced subsidence monitoring. It may be challenging to apply the SAR pixel matching technique (Tobita, et al,. 2001; Werner, et al., 2001) to measure the horizontal ground deformation with the current ENVISAT and ALOS/PALSAR data due to the large imaging resolution compared to the amount of horizontal Fig. 7. Ground survey data at the upper mine in Figure 6 for the period of 10 Apr – 28 May 2007 (48 days) and 16 Apr – 4 Jun 2007 (49 days), respectively. The maximum subsidence occurred during the time span is about 30cm.

displacement. But it may be applicable when the 1m resolution of TerraSAR-X and COSMOSKYMED become available.

7 Concluding Remarks Figure 7 shows the ground survey data (from BHPBilliton) at the upper mine in Figure 6. The

This study illustrated how 3-D deformation

maximum subsidence during a time span of 48-49

vectors due to mining-induced subsidence can be

days is

about 30cm. The ALOS/PALSAR

measured using the DInSAR technique, from a

DInSAR subsidence map in Figure 6 shows a

combination of ascending and descending image

maximum subsidence of about 35-40cm over 46

data collected with different look angles (swath

days, which is similar to the ground survey result.

modes)

More detailed validation is needed, with ground

deformation vectors with different line-of-sight

truth data with the same or similar spatial and

SAR geometry are available, it is possible to

temporal coverage.

derive the deformation components in the Up,

6 Further Work As illustrated in the previous section, using ALOS/PALSAR data (or SAR data with longer wavelength) may ease the problem of high phase gradient in interferograms caused by large and/or rapid ground deformation such as that induced by underground mining. Other options for avoiding high phase gradients are to use (if available) interferometric pairs with finer imaging resolution and/or shorter revisit times. Two new SAR satellites, TerraSAR-X and COSMO-SKYMED, were launched in June 2007. Both satellites have SAR sensors operating in the X-band frequency band. Their finest imaging resolution is less than 3m. The repeat cycles of TerraSAR-X and a single COMSMO-SKYMED are 11 and 16 days, respectively. In future studies, X-band data will be

by

ENVISAT/ASAR.

When

three

East and North directions. However,, this method is less sensitive to the displacement in the NorthSouth direction. The proposed method may be useful when a third displacement map along an independent look angle cannot be achieved, e.g. using azimuthal offsets. However, this method is limited by requiring three interferometric pairs (the triplets) with good coherence covering a similar period of time, or two ascending and descending pairs with a good estimation of the deformation in the third direction. Unfortunately, this study found that high phase gradient in the differential interferogram caused by mining-induced deformation is unavoidable using ENVISAT data. This is because of the comparatively coarse imaging resolution and long revisit orbit cycle. Therefore, the true height deformation cannot be recovered from the

ambiguous

interferometric

phases.

Consequentially errors are propagated into the final height displacement map. ALOS/PALSAR data have longer wavelength and

also

finer

ENVISAT/ASAR.

imaging The

resolution

than

derived

from

result

ALOS/PALSAR demonstrates the problem caused by high phase gradient can be mitigated. The new TerraSAR-X and COSMO-SKYMED data, with shorter site revisit cycle and finer imaging resolution, will be tested for their capability of monitoring ground deformation.

Acknowledgement This research work is supported by the Cooperative

Research

Centre

for

Spatial

Information

(CRC-SI)

Project

4.2,

whose

activities

are

funded

by

the

Australian

Commonwealth’s Cooperative Research Centres Programme. This study was also conducted in collaboration with the Strata Control Technology, supported by the Australian Coal Association Research Program. The authors wish to thank the European Space Agency for providing ENVISAT data. ALOS/PALSAR interferogram was processed under CRC-SI project 4.2, including material copyright METI and JAXA (2006, 2007) L-1.1 product: produced by ERSDAC. METI and JAXA retain ownership of the ALOS/PALSAR data. ERSDAC produced and distributed the PALSAR L-1.1 product. The PALSAR L-1.1 products were distributed to the IAG Consortium for Mining Subsidence Monitoring. The authors are grateful to ERSDAC for their support of the consortium. The authors wish to thank BHPBilliton for providing the ground survey data.

References BHPbilliton, Mine subsidence ground survey report. (private communication)

Chang, H.C., L. Ge and C. Rizos, 2005. ERS tandem DInSAR: the change of ground surface in 24 hours. IGARSS '05, Seoul, Korea, 25-29 July, vol. 7. pp. 5265-5267. Fialko, Y. and M. Simons, 2001. 'The complete (3-D) surface displacement field in the epicentral area of the 1999 Mw 7.1 Hector Mine earthquake, California, from space geodetic observation' Geophys. Res. Lett., vol. 28 no. 16: pp. 3063-3066. Ge, L., H.C. Chang and C. Rizos, 2007. 'Mine Subsidence Monitoring Using Multi-source Satellite SAR Images' Photogrammetric Engineering & Remote Sensing, vol. 73 no. 3: pp. 259-266. Massonnet, D. and K.L. Feigl, 1998. 'Radar interferometry and its application to changes in the Earth's surface' Review of Geophysics, vol. 36 no. 4: pp. 441-500. Michel, R., J.-P. Avouac and J. Taboury, 1999. ' Measuring ground displacements from SAR amplitude images: Application to the Landers earthquake' Geophys. Res. Lett., vol. 26 no. 7: pp. 875-878. Sircar, S., D. Power, C. Randell, Y. Youden and E. Gill, 2004. Lateral and subsidence movement estimation using InSAR. Geoscience and Remote Sensing Symposium, 2004. IGARSS '04. Anchorage, Alaska, 20-24 Sept. vol. 5. pp. 2991-2994. Spreckels, V., U. Wegmuller, T. Strozzi, J. Musiedlak and H.C. Wichlacz, 2001. Detection and observation of underground coal mining-induced surface deformation with differential SAR interferometry. ISPRS Workshop "High Resolution Mapping from Space 2001", Hannover. pp. 227234. Tobita, M., M. Murakami, H. Nakagawa, H. Yarai, S. Fujiwara and P.A. Rosen, 2001. '3-D surface deformation of the 2000 Usu eruption measured by matching of SAR images' Geophys. Res. Lett., vol. 28 no. 22: pp. 4291-4294. Werner, C., T. Strozzil, A. Wiesmann, U. Wegmuller, T. Murray, H. Pritchard and A. Luckman, 2001. Complimentary measurement of geophysical deformation using repeat-pass SAR. IGARSS '01. Sydney, Australia, 913 July. vol. 7. pp. 3255-3258.