An example of hyperspectral imaging and LIDAR integration for the detection of landslide parameters (presented at Geological Remote Sensing Group meeting in London, 2011).
Integrated horizontal-scan remote sensing for landslide imaging and evaluation
1. Integrated horizontal-scan remote
sensing for landslide imaging and
evaluation
Amer Smailbegovic1, Michael Mendenhall2,
Jeffrey Clark2, Kyle Gray3, and Richard Wooten4
1TeraElementLtd., 2US Air Force Institute of Technology, 3College
of Charleston, and 4NC Geological Survey
2. The Landslide: Franklin, NC
Located in the
Appalachian
Mountains of the
Eastern United States
Image/map courtesy of NC Geological Survey
3. Landslide: Geology requirements
Unstable slope:
35 – 65 degrees
Expansive Clays
Deteriorating host
Water
Host rock: Amphibolites/schist
withsignificant concentrations of pyrite
aswell as deteriorating igneous rocks
Alluvium, sands, clays all present as
process of host rock decomposition
continues
4. Landslide: How it works
Deforestation: decreasing vegetative cover can
increase risk by as much as 50%, change in water
absorption-holding factor-climate.
Soil mineralogy: water expansive
clays, poorly draining soils, mixed soil
types, lateritization and deterioration
of the host rock by
weathering/leaching
Increased loading: Loading unstable, poorly
rooted layers may result in collapse (e.g.
water load, structure load)
Slope angle: determination of risky slopes particular for an area
5. Landslide: Observables
Fracture
Sulfates
SPECTRAL SIDE: LIDAR side:
-Identify zones of weathered rock and -Identify zones of weakness (fractures,
byproducts of acid-digestion: sulfates faults, sliding blocks)
-Identify zones of clay minerals that -Identify zones of high / risky slope
swell by as much as 200% when
hydrated (e.g. montmorillonite)
6. The People (is it a landslide if no one is around to notice)
Currently the area is contentious and attracting
different attention for variety of purposes: study,
safety, litigation, development, transportation and
infrastructure.
One must exercise caution in approach not to trigger
existing and other non-geological landslides
8. Equipment
Velodyne High Def LIDAR (HDL-64E S2)
360 Deg Horizontal FOV
HyperSpecTIR (HST) -3 26.8 Deg Vertical FOV
Third model of SpecTIR’s airborne scanner, which 1.3M Points per second
saw airborne action until 2006 when it was 2 Co-mounted Hi Res framing cameras
relegated to ground use.
227 bands, 1mRad FOV, 450-2450 spectral range.
9. Technical Challenge
• Low signal-to-noise ratio (hyperspectral)
• Cause: scene geometry & horizontal config
– Specular component is the largest contributor of RTR
– Specular component hits the ground plane
– Only weak diffuse component reaches the imager
Result low SNR…
RDiff RTR=RDiff+RSpec
RSpec RDiff
RSpec
Hill Side Imager
Ground Plane
10. Mitigating SNR Issues
Transform-based denoising
may occur prior to atmospheric
compensation – work in progress
• Processing Chain
Image Acquisition
I1
Reflectance
I2 Super Resolution: ISR Atmospheric C ˆC
I Signal
ISR I1, I 2 , , I n SR ISR
Compensation
…
In Denoising
• Super Resolution:
– 10’s of images with stationary imager – average yields
good recovery of true radiance image
• Atmospheric Compensation: Currently Flat Field
• Denoising Efforts: Savitzy-Golay
– Savitzy-Golay filtering used in this study
– Ongoing: Discrete Wavelet Transform & Discrete Fourier
Transform denoising method
11. Some Intermediate Results
• Super Resolution • (μOrig- μSR) white panel
– (σOrig-σSR) white panel – Orig overly bright w/large
• Significant reduction in σ in VIS deviation from estimate of μ
• Notable reduction in σ in SWIR
80
120 60
100
Aprrox crossover point 40
80 between VNIR & SWIR
Detectors 20
60
0
40
-20
20
0 -40
600 800 1000 1200 1400 1600 1800 2000 2200 2400
Wavelength in nm
-20
600 800 1000 1200 1400 1600 1800 2000 2200 2400
Wavelength in nm
• Savitzy-Golay Smoothing
1
0.4
0.35
0.95
0.3
0.9 0.25
FFSG , FF
2
FFSG , FF
0.2
0.85
2
0.15
0.8 White Panel Mean: 0.1
White Panel Variance:
0.75
Savitzy-Golay (red) 0.05
Savitzy-Golay (red)
0.7
Non-Noised (black) 0
Non-Noised (black)
600 800 1000 1200 1400 1600 1800 2000 2200 2400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
12. Hyperspectral Data
Scene 3 Scene 1
Cal Panels for ELM The King of Pop has reemerged
In-scene spectral diversity
13. Target Minerals
Reflectance Spectra De-noised
using Savitzky-Golay regression
14. On Scene
Example of hyperspectral detection of
minerals – classification results overlain
on natural-color RGB composite:
RED – water expanding mineral suite with
evidence of sulfate minerals indicating
rock-decomposition processes
GREEN – water expanding clay minerals
only
15. LIDAR
Color-shaded, horizontal-
looking LIDAR data showing
some of the apparent structural
elements of significance to
landslide formation and/or
propagation.
16. LIDAR
Water-expanding mineral suite on
LIDAR scan
17. LIDAR
Isolated block Some of the possible
targets for another
landslide: note
potentially
compromised block
in the center
Structure Large area of
compromised rock
Zone of Weakness and clay minerals on
the right is an
evidence of an
exposed face, but
lack of structural
elements does not
Example of integrated LIDAR and HSI-produced maps make it as risky as
the location on the
left
18. Directions
• Technical
– Atmospheric Correction
• Tradeoffs between MODTRAN, on-site irradiance spectra,
ELM, and Flat Field for the horizontal scan configuration
– Digital Signal Processing
• Incorporate Sensor Noise Model
• Detailed study on denoising affects (S-G, DWT, DFT)
– Create a workable geotechnical model from new
input data
19. Conclusions
– It is possible to delineate zones of “problematic”
mineralogy on a wider scale using horizontal-
looking hyperspectral imagery
– It is possible to delineate zones of structural
weakness and/or landslide formation elements
using horizontal-looking LIDAR data
– Integrating LIDAR and hyperspectral data shows
promising results as a tool in observing and
tracking existing landslide areas
– Fills important elements in geotechnical
evaluation: mineralogy and structure