Presentation is loading. Please wait.

Presentation is loading. Please wait.

Subsurface scattering of SAR signals assessed using GPR Matthew B. Charlton School of Geography The University of Nottingham.

Similar presentations


Presentation on theme: "Subsurface scattering of SAR signals assessed using GPR Matthew B. Charlton School of Geography The University of Nottingham."— Presentation transcript:

1 Subsurface scattering of SAR signals assessed using GPR Matthew B. Charlton School of Geography The University of Nottingham

2 Identification of duricrusts using remote sensing in the Libyan Sahara to facilitate location of archaeological sites and palaeo-environmental reconstruction. AIM Assess why SAR detection of such sites is variable. In particular, determine the role of subsurface scattering using GPR.

3 Duricrusts 3 types found in interdunes Calcretes Silcretes Gypcretes Protect underlying sediments, forming inverted relief Preserve sedimentary contexts otherwise removed by deflation processes Duricrust Lacustrine sediments Dune sand

4 Archaeology Hand axes Burials Rock Art

5 Optical: Enhanced Thematic Mapper Calcretes can be discriminated from multispectral data 3 km

6 L-band Radar Landsat ETMJERS-1 JERS detects some (not all) silcretes invisible in TM imagery...

7 C-band Radar Landsat TMJERS-1Radarsat Other silica-rich duricrusts evident in Radarsat fine beam mode image, but not in JERS or TM

8 Factors influencing backscatter from duricrusts SAR system: incidence angle, frequency, orbit, etc. Surface roughness variation. Dielectric constant. Subsurface scattering.

9 Methods Acquire SAR data (ERS, JERS, RADARSAT, SIR-C). Acquire ETM imagery. Identify sites (NSGP1, UBR1, UL1).

10 12.80 12.95 13.10 26.80 26.65 26.50 12.80 12.95 13.10 26.80 26.65 26.50 Longitude (decimal degrees) Latitude (decimal degrees) JERS-1 RADARSAT Site Locations (Lat / Lon) (decimal degrees): NSGP1: 26.59865N 13.06504E UBR1: 26.68181N 12.84430E UL1: 26.71133N 12.90280E Libyan Fezzan NSGP1: thin gravel crust UBR1: thinner calcrete / silcrete crust UL1: thick calcrete / gypsum crust

11 Methods Acquire SAR data (ERS, JERS, RADARSAT, SIR-C). Acquire ETM imagery. Identify sites (NSGP1, UBR1, UL1). Field survey of surface and subsurface: Surface roughness (10m profiles at 0.0084m resolution) Dielectric constant (Thetaprobe) Local topographic surveys Surface samples (texture, etc.) GPR profiles (30-75m, 450 and 900 MHz, CMP) Pits (structure and samples)

12 Analysis Process SAR data (georegister, convert to backscatter). Use surface properties to run simplified IEM. Compare estimated and observed backscatter. Basic interpretation and processing of GPR data, plus determination of penetration depth, attenuation, and scattering.

13 IEM Inputs Dielectric Constant = 2.5 - 4.1 Average value: 2.88 No significant variation Roughness Parameters NSGP1  = 0.53 L = 47.68 UBR1  = 0.66 L = 35.87 UL1  = 0.24 L = 20.53 Exponential and gaussian functions Sites smooth at L- & C-band High variability at shorter profile lengths Drift and periodicity in correlograms Frequency 1.275 GHz 5.3 GHz Incidence Angle 10-50° JERS: 35.21° RADARSAT: 36.9°

14 Estimated and Observed Backscatter Backscatter (dB): JERS Predicted: -34.12 Observed: -0.52 Predicted: -30.95 Observed: 2.60 Predicted: -37.70 Observed: 4.33 Backscatter (dB): RADARSAT Predicted: -27.44 Observed: -22.39 Predicted: -23.87 Observed: -21.64 Predicted: -31.67 Observed: -23.86

15 Backscatter (dB): JERS Predicted: -34.12 Observed: -0.52 Predicted: -30.95 Observed: 2.60 Predicted: -37.70 Observed: 4.33 Backscatter (dB): RADARSAT Predicted: -27.44 Observed: -22.39 Predicted: -23.87 Observed: -21.64 Predicted: -31.67 Observed: -23.86 Backscatter coefficient is underestimated. Patterns are wrong for JERS: Predicted and RADARSAT: UBR1 > NSGP1 > UL1 JERS UL1 > UBR1 > NSGP1 Estimated and Observed Backscatter

16 IEM Prediction Error Assumption of constant incidence angle. Assumption of constant range direction. Dielectric constant. Drift and periodicity in correlograms results in high L. Trade-off between  and L. Variability in roughness parameters. Subsurface scattering not accounted for in simplified IEM.

17 Ground-penetrating radar surveys  PulseEKKO 1000A GPR  450 & 900 MHz  0.02 - 0.05 m Step Size  30 - 75 m Profiles  Ground-coupled  80ns Time Window  10 ps Sampling Interval  CMP surveys Basic Processing trace edit dewow time zero topographic

18 Two Way Travel Time (ns) Horizontal Position (m) 6.0 22.0 37.5 Horizontal Position (m) 0.0 15.0 30.0 NSGP1 (Constant Gain = 50) 0.0 20.0 40.0 Two Way Travel Time (ns) 0.0 20.0 40.0 900 MHz GPR Profiles UL1 (Constant Gain = 50)  NSGP1 dominated by horizontal layering, ringing and limited penetration.  UL1 dominated by horizontal layering (more layers), greater layer roughness, greater penetration and significant numbers of diffractions of varying size.

19 GPR signal scattering: NSGP1 Layer reflection at sand- crust interface of variable depth. Reflection depends on depth of burial, layer roughness, and nature of materials. Complicated by laminations and proximity to direct arrivals. 0.45 m 0.10 m

20 GPR signal scattering: UL1 Layer reflection at sand- crust interface of variable depth. Layer reflection at different crust interfaces (powder, massive, inhomogeneous). Multiple scattering from inhomogeneities (cracks, voids, etc.).

21 Two Way Travel Time (ns) Horizontal Position (m) 6.0 22.0 37.5 Horizontal Position (m) 0.0 15.0 30.0 NSGP1 (Constant Gain = 50) 0.0 20.0 40.0 Two Way Travel Time (ns) 0.0 20.0 40.0 900 MHz GPR Profiles UL1 (Constant Gain = 50) GPR response demonstrates that signal scattering varies between sites in a way that depends on the nature of the duricrust. UL1 would appear to have higher subsurface scattering potential than NSGP1.

22 Backscatter (dB): JERS Predicted: -34.12 Observed: -0.52 Predicted: -30.95 Observed: 2.60 Predicted: -37.70 Observed: 4.33 Backscatter (dB): RADARSAT Predicted: -27.44 Observed: -22.39 Predicted: -23.87 Observed: -21.64 Predicted: -31.67 Observed: -23.86 C-band has limited penetration and greater sensitivity to variability in surface roughness. Conforms to modelled pattern. L-band is less sensitive to limited surface roughness. With greater signal penetration, subsurface complexity becomes more important. NSGP1 has a less complex subsurface and therefore a lower observed backscatter compared to UL1. Re-interpreting Backscatter

23 Quantifying Subsurface Scattering Maximum penetration depth (from ungained images, assuming velocity of 0.177mns -1 ) : NSGP1: 0.8m UL1: 1.2m Attenuation calculation (from ungained images, assuming velocity of 0.177mns -1 ) : Convert to Instantaneous Amplitude. Calculate Mean Trace.

24 Quantifying Subsurface Scattering Attenuation calculation (from ungained images, assuming velocity of 0.177mns -1 ) : Convert to instantaneous amplitude. Calculate mean trace to give attenuation curve (Grandjean et al., 2001). Convert to dB: (where CF = -93.98) Calculate dB difference between two points of known depth using (Farr et al., 1986) : To avoid direct arrivals attenuation was calculated between 0.5- 1.5m. Maximum penetration depth (from ungained images, assuming velocity of 0.177mns -1 ) : NSGP1: 0.8m UL1: 1.2m

25 Quantifying Subsurface Scattering Attenuation results: Very similar between sites: NSGP1: 1.04 dB/m UL1: 1.15 dB/m Relatively high for dry sand (salt content?) Subject to averaging error and local fluctuations in instantaneous amplitude at depth. Confirm that simplicity of NSGP1 GPR data is due to lack of reflections, not excessive attenuation.

26 Quantifying Subsurface Scattering Preliminary assessment of subsurface scattering in upper metre Descriptive statistics (using pit traces) : Standard deviation of amplitudes does not work. Nor does coefficient of variation (NSGP1 = 1.38; UL1 = 0.99). Due to greater contrast between high (direct arrivals, reflections, scattering) and low amplitude (no reflections) zones in NSGP1 (more of UL1 has similar higher magnitude values). RMSE: Subtract mean pit trace from individual pit traces and calculate RMSE. Where there is greater scattering there should be greater error. Analysis confirms greater scattering at UL1 and shows intra-site variability. Such a value could be scaled to produce an equivalent to RMS surface height. PitUL1NSGP1 136202439 237471277 344441538

27 Quantifying Subsurface Scattering Preliminary assessment of subsurface scattering in upper metre Layer roughness: Previous analysis considered all sources of scattering. For the most simple inclusion in IEM layer roughness can be calculated from GPR data: Two Way Travel Time (ns) Horizontal Position (m) 0.0 15.0 30.0 0.0 20.0 40.0 UL1 (Constant Gain = 50) Has yet to be done. Ignores other sources of scattering. Complex task in multi-layered environments.

28 Variations in surface roughness do not explain observed backscatter differences. C-band conforms to model results; L-band does not. GPR confirms considerable subsurface layer and volume scattering potential at two sites. Scattering is related to the nature of the duricrust (layering, nodules, voids, cracks, depth of sand overlay). In arid environments, at low frequencies, subsurface scattering may become more important in determining backscatter. Further work to (1) assess SAR penetration, (2) assess scattering dependency on subsurface complexity, and (3) develop GPR techniques to derive subsurface scattering parameters. Conclusions

29 CREDITS Field Work Dr. Nick Brooks (University of East Anglia) Dr. Kevin White (University of Reading) Pit Photographs Toby Savage


Download ppt "Subsurface scattering of SAR signals assessed using GPR Matthew B. Charlton School of Geography The University of Nottingham."

Similar presentations


Ads by Google