Extraction of Subsurface Features from InSAR-derived...

Extraction of Subsurface Features from InSAR-derived Digital Elevation Models

Si-Ting Xiong a, Jan-Peter Muller a,b * aImaging Group, Mullard Space Science Laboratory (MSSL), University College London bChair, Terrain Mapping Sub-Group, CEOS WG on Calibration & Validation

Fringe2015, 25/03/2015, Frascati (Rome), Italy

Outline

•  Context – Radar penetration & Subsurface mapping – Global DEMs (Digital Elevation Models)

•  Materials and Methods •  Results

– Extraction of subsurface features – Comparisons of different DEMs

•  Conclusions

•  radar penetra-on: –  roughness of surface rela-ve to wavelength:

homogeneous, fine-‐grained thin (1-‐2 m thick) cover materials.

–  dielectric property: moisture content is less than about 1.0% .

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. GE-22, NO. 4, JULY 1984

RH

0'1

(,(a)

(b)

Fig. 3. Geometry of the wave ray. (a) Exposed rough surface of dielec-tric constant e 1; (b) a layer of sand of thickness L and dielectric con-stant C2 covers the rough surface.

(a)

10.0

5.0

e = 50"

RH E2 = 2.5

\ :=8.0

SRHV\

2.0

1.0o

0.5

0.3

0.230 40 50 60 70 0

(b)Fig. 5. Behavior of R as a function of 0 for different values of Lid. (a)For Lld = 0.2, RH, RHV are larger than 1 for all values of 0. (b) ForLld = 0.8 the enhancement effect occurs only at large incidence angles.

I I I I ,

0 0.2 0.4 0.6 0.8 1.0 L/d

Fig. 4. Ratio R of the backscatter cross section of the covered surfaceass over the backscatter cross section of the uncovered surface as.For small value of Lld (d = skin depth), R is larger than 1 because ofthe enhancement due to surface refraction. As Lld increases, theabsorption losses in the sand become dominant, cancelling the favor-able refraction effect.

illustrated in Fig. 6. As the dielectric constant e1 increasesabove 4, which would be the case for almost all solid rocks,RH is always larger than 1 for Lld = 0.2 and 0 = 500. In thecase of Lid = 0.8, this enhancement occurs only for el largerthan 12 which is not a common case. For the vertical polariza-tion, the enhancement is less favorable.

III. FORMULATION FOR CROSS POLARIZEDBACKSCATTER

In the case of the cross polarized cross section, c4' is given by[7] [8]

where

g = 2irk8 cos4 0(e -I)' (e - sin2 0)

(cos 0 + -e-sin2 0)2 ( cos 0 + E - sin2 0)2

and I is the surface roughness autocorrelation. The ratio RHVfor the layer-covered interface is then

S5 -T 2 Lid cosORHV = , = H(O) TV(0) e-2/ O

Us

g(e,2V 0) (8)

9(El I XI a)

The behavior ofRHV is also shown in Figs. 5 and 6. In general,it can be said that the cross polarized signal enhancement isweaker than in the case of the direct HH polarized signal butstronger than in the case VV polarized signal.

5-

4-

3-

2-

386

(6)Igst= g(e, X, O)I

1676 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 41, NO. 7, JULY 2003

(a)

(b)

(c)

Fig. 5. Location of GPR profile for site 2. (a) Geological map from Klitzschet al. [9]. (b) Landsat-7 ETM scene (panchro). (c) SIR-C scene (L band, HHpolarization).

45 with a pixel size of 12.5 m. We performed enhanced Leeadaptive filtering on image data in order to reduce high-fre-quency noise (speckle) while preserving high-frequency fea-tures (edges). For each of our three test sites, we defined “homogeneous” areas according to the geological context (eacharea containing at least 60 pixels) in order to derive statisticaldistributions for the scattering coefficients. Results of the po-larimetric analysis are presented in Figs. 8–10. Top parts of thefigures present the extract of SIR-C L-band image containing

(a)

(b)

(c)

Fig. 6. Location of GPR profile for site 3. (a) Geological map from Klitzschet al. [9]. (b) Landsat-7 ETM scene (panchro). (c) SIR-C scene (L band, HHpolarization).

the GPR profile, with the selected “ homogeneous” areas. Westudied the distribution of , and , theand ratios, and the correlation between andfor each of the C and L bands. We present normalized his-tograms (i.e., axis ranges between zero and one), and axisis in decibels for scattering coefficient distribution. Copolarizedscattering coefficients contain both surface and nonsurface scat-tering components, whereas the nonsurface scattering compo-nent, i.e., subsurface and volume scattering components, is the

Landsat-7 ETM+

SIR-C (L-band HH)

•  subsurface mapping: –  a low-‐loss thin sand layer enhances the

capability to image the subsurface sand-‐bedrock interface.

–  the skin depth ranges from 1.5 to 6 m at L-‐band, the SIR-‐A frequency shown by laboratory experiment.

–  field observa-ons showed that the sand layer thickness ranges from 0.8 to 2 m.

Courtesy P. Paillou et al.

Courtesy C. Elaichi et al.

Introduc-on – Radar penetra-on & subsurface mapping

•  Global DEMs up to 30 m are available for public use

–  SRTM-‐C and –X DEM

–  ASTER GDEM

–  ICESat Eleva-on dataset •  InSAR-‐derived DEM for specific requirement

–  ALOS-‐PALSAR –  TanDEM-‐X

•  Problems of applica-on to hyperarid areas

–  DSM (Digital Surface Model) or DSSM (Digital SubSurface Model) in hyperarid area?

–  Detect subsurface features? Bedrock height Mapping? Or surface lowered due to deep channels

Introduc-on – Global DEMs

•  Eastern Sahara – Kufrah River

•  southeast Libya, 23°N-24°N, 23°E-24°E

•  a tributary of a paleodrainage system passing from south to north till Sarir Dalmah in Libya.

–  Gilf Kebir Plateau •  southwest Egypt, 23°N-24°N,

26°E-28°E •  1,300 crater-like features over 4,000

km2 (Paillou, P., Schuster, M., Tooth, S., et al., 2009)

•  Ténéré Desert •  northeast Mali and

northwest Chad •  15°N-20°N, 10°E-20°E

Study Sites

•  Several global DEM datasets are used to inves-gate their penetra-on depth, which is a key characteris-c in terms of subsurface features detec-on and bedrock mapping.

ASTER GDEM SRTM-C DEM SRTM-X DEM ICESat /GLA14

Tile size 1°× 1° 15′ ×15′ －

Posting interval 1 arc-second (~ 30 m) 70 m footprint / 150 m

Geographic coordinates Geographic lat/lon －

Datum WGS84/EGM96 WGS84/ WGS84 ellipsoid

TOPEX/Poseidon

Nodata Value -9999 -32767 －

Horizontal accuracy ±20 m ±20 m(abs.)

±15 m(rel.) ±20 m(abs.) ±15 m(rel.) －

Vertical accuracy ±30 m ±16 m(abs.) ±10 m(rel.)

±16 m(abs.) ±6 m(rel.) cm scale

Materials and Methods

Part A

Part B

•  ALOS-PALSAR 25 m Forest and non-Forest (FNF) map was used to delineate paleo-channels.

–  Japan Aerospace Exploration

Agency（JAXA) , has produced the 4

year, 25m spacing global PALSAR

mosaics from 2007 to 2010 using the

accurate SAR processing.

–  PALSAR HH, HV backscatter is slope

corrected and ortho-rectified using the

SRTM, and the radiometrically

calibrated.

Courtesy M. Schimada et al., JAXA

Materials and Methods

paleo-channels

Eastern Sahara A paleoriver is delineated on PALSAR HH backscatter image and overlaid on Gaussian- filtered and hill-shaded SRTM-C DEM.

Results -‐ Extrac-on of subsurface features

Eastern Sahara A paleoriver is delineated on PALSAR HH backscatter image and overlaid on Gaussian- filtered and hill-shaded SRTM-C DEM.


an alluvial fan

A paleoriver is delineated on PALSAR HH backscatter image and overlaid on Gaussian- filtered and hill-shaded SRTM-C DEM.

circular features

Eastern Sahara Results -‐ Extrac-on of subsurface features

Ténéré Desert – Part A

Part A


Part A

Ténéré Desert – Part A Results -‐ Extrac-on of subsurface features

PALSAR doesn’t show this linear features due to affluence by low backscatter from background.

Delineated from PALSAR HH backscatter

Part A


Part A

fractures or faults?


PALSAR doesn’t show this linear features due to affluence by low backscatter from background.

Delineated from PALSAR HH backscatter

Part A

Part A

None of the linear and circular features are shown in Landsat image, whereas most recognisable in PALSAR HH backscatter image.

Most circular features are shown in SRTM-C DEM, but less apparent in PALSAR images.


Optical image : only shows outcrop rock.

SRTM-C DEM: rock with elevation different from background,

also show different forms of sand cover.

ALOS-PALSAR image: bedrock under thin sand layer.

bedrock under thin sand layer

different forms of sand cover

only outcrop rock can be seen

Part B

Ténéré Desert – Part B Results -‐ Extrac-on of subsurface features

Landsat image SRTM-C DEM ALOS/PALSAR

ASTER DEM

SRTM-C DEM

ASTER – SRTM-C

v Kufrah River

ASTER – SRTM-C

SRTM-C DEM ASTER DEM

DEMs appear to show surface expression of buried paleorivers.

v Gilf Kebir Plateau Results -‐ Comparisons of different DEMs

q  In this study area, ASTER GDEM is very noisy.

q  SRTM-C DEM agrees closely with ICESat/GLA14.

q  However, SRTM-X DEM has a large bias with ICESat/GLA14 and SRTM-C DEM, which appears to be caused by phase adjustment difficulties (Marschalk, U and Roth, A et al., 2004).

KT1 KT2

KT3

Results -‐ Comparisons of different DEMs v Kufrah River

GT1 GT2

GT3 GT4

Results -‐ Comparisons of different DEMs v Gilf Kebir Plateau

q  ASTER GDEM is also very noisy.

q  SRTM-C DEM agrees closely with ICESat/GLA14, but 2-3 m bias within steep slope area.

q  SRTM-X DEM has a negative bias of ≈2~5 m from ICESat/GLA14.

GT5 GT6

GT7

Results -‐ Comparisons of different DEMs v Gilf Kebir Plateau

Detail inves6ga6on of eastern Sahara

R： SRTM-C DEM, G:PALSAR-HH, B:ASTER GDEM

q  ASTER GDEM acts the worst to show the profile of paleoriver . q  SRTM-C DEM shows lower elevations across paleoriver bed,

while PALSAR-HH image displays lower backscatter.

Topographic features analysing

Results -‐ Comparisons of different DEMs

R： SRTM-C DEM, G:PALSAR-HH, B:ASTER GDEM

Topographic features analysing

q  More information shown by combining both SRTM-C DEM and ALOS-PALSAR backscatter.

q  Both SRTM-C DEM and PALSAR backscatter image show the crater-like feature.

Results -‐ Comparisons of different DEMs Detail inves6ga6on of eastern Sahara

1.  SRTM-‐C DEM agrees closely with ICESat eleva-on and are superior to ASTER GDEM and SRTM-‐X DEM, but is of insufficient accuracy in terms of heigh-ng expressions of subsurface features ability over hyperarid regions.

2.  Comparing with SRTM DEM, ALOS/PALSAR image shows more detailed subsurface features, but is easily affected by similar backsca`ers from background.

3.  Combining subsurface features extrac-on results of SRTM-‐C DEM and PALSAR image, the study shows a great poten-al to map the bedrock eleva-on by using InSAR of lower frequency and longer baseline, such as ALOS-‐2/PALSAR-‐2 datasets.

Conclusions

Thank you for your attention! Questions & Suggestions please!

Extraction of Subsurface Features from InSAR-derived...

Documents

Transcript of Extraction of Subsurface Features from InSAR-derived...