CHARACTERIZING LANDSLIDE MOVEMENT AT THE BOULDER CREEK
EARTHFLOW, NORTHERN CALIFORNIA, USING L-BAND INSAR
by
LAURA LYN STIMELY
A THESIS
Presented to the Department of Geological Sciences
and the Graduate School of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Master of Science
September 2009
11
"Characterizing Landslide Movement at the Boulder Creek Earthflow, Northern
California, Using L-band InSAR," a thesis prepared by Laura Lyn Stimely in partial
fulfillment of the requirements for the Master of Science degree in the Department of
Geological Sciences. This thesis has been approved and accepted by:
Committee
3 (- A~-I_-_2._o_0_9,----- _
Date
Committee in Charge:
Accepted by:
Dr. Joshua J. Roering, Chair
Dr. David A. Schmidt
Dr. John M. Logan
Dean of the Graduate School
111
An Abstract of the Thesis of
Laura Lyn Stimely for the degree of Master of Science
in the Department of Geological Sciences to be taken September 2009
Title: CHARACTERIZING LANDSLIDE MOVEMENT AT THE BOULDER CREEK
EARTHFLOW, NORTHERN CALIFORJ~IA,USING L-BAND INSAR
Approved: --/
Spatial and temporal patterns of movement of the Boulder Creek earthflow were
investigated using 26 interferograms derived from ALOS satellite radar images acquired
between February 2007 and February 2008. Persistently unstable hillslopes in Northern
California are ideally suited to the study of the dynamics and morphological signature of
earthflows, as the deeply sheared melange lithology, high seasonal rainfall, and fast uplift
rates promote widespread deep-seated landsliding. In addition to identifying multiple
active landslides in the region, L-band InSAR reveals varying defonnation rates in the
accumulation, transport, and toe regions ofthe Boulder Creek earthflow. Downslope
displacement rates up to 1.8 m/yr are observed on the earthflow over a I-year period. The
pattern of defonnation is similar to that observed from 1944-2006 inferred from aerial
photography. Interferograms highlight spatially variable rates controlled by lithology and
gullies, and movement correlates with seasonal rainfall with a phase lag of~2 months.
IV
CURRICULUM VITAE
NAME OF AUTHOR: Laura Lyn Stimely
PLACE OF BIRTH: State College, Pennsylvania, USA
DATE OF BIRTH: June 4, 1982
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR
The Pennsylvania State University, University Park, PA
DEGREES AWARDED:
Master of Science, Geology, 2009, University of Oregon
Bachelor of Science, Engineering Science & Mechanics, 2005, The Pennsylvania
State University
AREAS OF SPECIAL INTEREST:
Geomorphology
Satellite Interferometry
PROFESSIONAL EXPERIENCE:
Graduate Teaching Fellow, Department of Geological Sciences, University of
Oregon, Eugene, OR, 2007-2009
Engineer, Bettis Atomic Power Laboratory, West Mifflin, PA, 2005-2007
Engineering Intern, Data Storage Institute, Singapore, 1/2005-7/2005
Technical Documentation Writer & Assistant, Minitab, Inc., State College, PA,
2000-2005
Geochemistry Laboratory Assistant, Materials Characterization Laboratory, Penn
State University, 2000-2001
PUBLICATIONS:
Roering J, Stimely L, Schmidt D, Mackey B (2009). Landslide movement and sediment
production quantified with InSAR, airborne LiDAR, and archival air photos.
Submitted to Geophysical Research Letters.
v
VI
ACKNOWLEDGMENTS
I give my most sincere thanks to Professors Josh Roering and David Schmidt for
their valuable support in the prepration of this manuscript and for their generosity and
insight throughout the research process. Josh and David have provided me a positive
environment to mature as a scientist, I am forever indebted to them for the skills I have
learned and the patience they have shown. I especially want to thank my third thesis
committee member John Logan for his honesty and kindness. I also would like to thank my
family and friends for all of their support both academically and emotionally.
Vll
TABLE OF CONTENTS
Chapter Page
I. INTRODUCTION 1
II. STUDY AREA AND GEOLOGIC SETTING 6
Geologic Setting......................... 6
Details of the Boulder Creek Earthflow......................................................... 7
III. DATA AND METHODS 12
Overview 12
InSAR Data Specifics.................................................................................... 13
Interferogram Formation................................................................................ 15
Unwrapping Errors and Decorrelation........................................................... 16
Stacking Interferograms....................................................................... 20
Atmospheric Artifacts..................................... 20
Projecting Line-of-Sight Deformation onto Landscape Surface 21
Time Series of Deformation.................................................. 23
IV. RESULTS 24
Overview. 24
Regional Landslide Evaluation & Individual Interferogram Observations... 24
Detailed Examination ofBoulder Creek Stacks 27
Downslope Velocity Projections.................................................................... 33
Seasonal Signal.............................................................................................. 36
V. DISCUSSION OF RESULTS 38
Overview.................. 38
Investigating the Spatial and Temporal Variations in Velocity 38
Investigating InSAR Line-of-sight Sensitivity 43
VI. SUMMARY AND CONCLUSIONS 50
V111
Chapter Page
APPENDICES 53
A. LIST OF INTERFEROGRAMS USED 53
B. GLOSSARY OF TERMS 55
REFERENCES 56
IX
LIST OF FIGURES
Figure Page
1. Location of the Study Area 5
2. The Boulder Creek Earthflow 9
3. Tree Vectors 13
4. Location of ALOS Scenes 14
5. Example of Correction for Unwrapping Error 19
6. Azimuth and Dip of Downslope Earthflow Movement 22
7. Regional View Highlighting Five Earthflows Found in Path 223 25
8. Select Individual Interferograms 26
9. Stack ofInterferograms for Satellite Paths 223 and 224 29
10. Downslope Velocity Projections.................................................................... 35
11. February 2007 to February 2008 Time Series of Average Daily Velocities 37
CHAPTER I
INTRODUCTION
Landslides frequently occur in hilly and mountainous areas, both presenting a
significant natural hazard and having profound effects on the landscape. Hillslopes adjust
their shape via diffusion creep and mass wasting in response to external forces such as
uplift, weathering, and climate change (e.g. Cmden and Varnes, 1996). Over geologic
timescales slope angles adjust to incision and uplift rates. Uplift usually results in
increased fluvial incision, which lowers base level and increases slope angles (Burbank
and Anderson, 2001). Hillslopes adjust to oversteepening via mass wasting processes,
which quickly removes material and reduces slope angles. Hillslopes have been observed
to maintain slope angles up to a particular threshold. Burbank et al. (1996) argued for
such a balance between uplift and erosion in the Himalayas where landsliding appears to
prevent slopes from continuing to steepen as uplift and incision increase. The idea can be
likened to a pile of sand, which can maintain slopes no steeper than its angle of repose.
Oversteepening, by adding material near the top of the pile or removing material near the
bottom, inevitably results in a landslide of material and relatively constant slopes.
Landslide activity also varies on short timescales. Seasonal precipitation has been shown
to playa large role in the activity of landslides (e.g. Keefer and Johnson, 1983; Iverson
2and Major, 1987). Increased displacements have been observed during El Nino years and
winter storms have resulted in earthflow surges of up to tens of meters in a day (Calabro,
2008; Iverson & Major, 1987). By studying landslides over broad areas and timescales,
we seek to decipher the signature of past and present landslides on the landscape and
what forces drive these features.
Detecting and monitoring landslide behavior tends to be highly site-specific and
costly (Iverson & Major, 1987; Kelsey, 1978). Techniques such as leveling and GPS
reveal valuable information about earthflow movement but may not be suitable to reveal
detailed spatial heterogeneities and can only be maintained for a relatively limited period
of time. Regional studies tend to be coarse, using data such as aerial photographs or
digital elevation models to identify locations and timing of landslides (Hovius et aI.,
1997; Hovius et aI., 2000; Fuller et aI., 2004). Quantifying the spatial and temporal
patterns of movement at high spatial resolution and over a broad area will provide insight
into what a landslide-dominated landscape can tell us about the interplay between
hillslope evolution and tectonic and climate processes at work.
Earthflows are among the most common mass-movement phenomena in nature.
They are identified by round scarps, inflated toes, and by long narrow teardrop-shaped
forms. Earthflows commonly display evidence of both fluid-like flow and rigid body
translation. They occur in weak, fine-grained sediments. Earthflows often extend from
ridge to channel and are typically found on relatively uniform slopes (Keefer and
Johnson, 1983). The continuous and slow-moving nature of earthflows offers an excellent
laboratory in which to study their short and long-term behavior and triggers.
3The primary data for this study is L-band Interferometric Synthetic Aperture
Radar (InSAR). InSAR is a powerful tool for measuring surface deformation over broad
regions with sub-centimeter scale precision. In recent decades, InSAR has proven its
capability to detect surface displacements caused by a variety of events such as
earthquakes, ice sheet motion, volcanic activity, and subsidence (Massonnet et ai., 1993;
Zebker et ai., 1994; Goldstein et ai., 1993; Massonnet et ai., 1995; Carnec et ai., 1996;
Fruneau et ai., 1996). Despite some challenging environmental conditions (high
vegetation density, steep topography, and high deformation rates) the capability of
InSAR to detect landslide movements has also been demonstrated (Carnec et ai., 1996;
Fruneau et ai., 1996; Rott et ai., 1999; Vietmeier et ai., 1999; Squarzoni et ai., 2003;
Colesanti et ai., 2003; Farina et ai., 2004; Catani et ai., 2005). Early InSAR studies on
landslides primarily used short repeat-cycle data, collected during the ERS
commissioning phase in 1991 (3-day repeat cycles), or on TANDEM missions in 1995,
1996, 1997 and 1999 (I-day repeat cycles). More recent studies have utilized new
processing techniques (e.g. Hilley et aI, 2004) or data from new satellite systems (e.g.
Kimura et ai., 2000).
Little InSAR data was available for this study due to the recent launch of the
ALOS satellite. To date, there are no InSAR studies published about earthflows in the Eel
River basin of Northern California. The Japanese Aerospace Exploration Agency (JAXA)
launched their L-band SAR system in 2006. Prior to this the primary InSAR data
available was from C-band systems that perform poorly in heavily vegetated regions.
There have been other L-band satellites (such as JERS), but they had limited data
4availability. Hilley et al. (2004) was able to quantify movement of earthflows in the
Berkeley vicinity of the eastern San Francisco Bay area with C-band data using an
alternative processing method called PS (persistent scatterers) InSAR, which tracks
persistently bright radar scatterers through time and can thus calculate displacement rates.
Hilley's study took advantage of an urban setting containing many buildings with comers
acting as persistent scatterers.
This study uses L-band InSAR data to identify localized landslide movement over
broad regions (~l 00 km) while also resolving fine details in the movement of individual
slides. InSAR is used to survey landslide activity in the Eel River basin ofNorthern
California (Figure I). Historical aerial photography data and high-resolution LiDAR
topography are used to assist in the analyses and interpretation of movements at Boulder
Creek. Among the various signals detected, I specifically constrain surface deformation
at the Boulder Creek earthflow between February 2007 and February 2008. Boulder
Creek was chosen for the study because it is optimally oriented with respect to the look
direction of the satellite and exhibits measurable amounts of movement throughout the
year. Displacement rates are shown to be associated with seasonal variations in rainfall. I
will use these results to examine relationships between slope movement and topographic,
fluvial, and lithologic properties. This in turn will help address broader questions
regarding the mechanics of earthflows and the timescales on which they are active.
5A
20 km
Figure 1. Location of the study area - nOlihern California. The Eel and Van Duzen Rivers
are drawn in blue. The Boulder Creek eatihflow is labeled with the red star.
6CHAPTER II
STUDY AREA AND GEOLOGIC SETTING
Geologic Setting
Persistently unstable hillslopes of the California Coast Ranges contain many
active and ancient earthflows of varying sizes. The regional bedrock of the coast range is
Franciscan melange, a late Jurassic and Cretaceous age clay-rich rock unit consisting of
highly sheared sandstones and siltstones in which are dispersed blocks and boulders of
greenstone, chert, schist, and serpentine (Kelsey, 1978). The rocks are interpreted to
represent a Mesozoic to Cenozoic aged accretionary wedge that has been uplifted and
exhumed. Recent uplift (~1 mm/yr) over the past 5 Ma is driven by the migration of the
Mendocino Triple Junction (Furlong and Schwartz, 2004).
The coast range of Northwestern California has a Mediterranean climate and an
average annual rainfall of 0.8 m, 80% of which falls between October and April.
Vegetation cover is locally variable as most slopes feature large grassland prairies
separating patchy oak, madrone, and conifer forests. The highly sheared bedrock, rapid
uplift, and bimodal precipitation pattern favor large landslide initiation. Both active and
ancient earthflows comprise roughly 10% of this melange area (Kelsey, 1978). The
7Boulder Creek earthflow is located along the main stem of the Eel River, approximately
60 miles southeast of Eureka, California (Figure 1).
Details ofthe Boulder Creek Earthflow
It is important to understand the surface morphology and expected movement
styles when interpreting deformation measured from space. The Boulder Creek earthflow
is a large landslide complex that appears to be deforming by multiple processes,
including block gliding, slumping, and viscous flow (Kelsey, 1978). No geotechnical or
hydrologic data has been published for this site. Therefore, information such as depth to
the water table and depth to a basal shear plane are not known. Mackey (personal
communication) has used orthorectified aerial photographs to examine decadal
movement rates at Boulder Creek. This data will be used to complement the InSAR
observations. To date this appears to be the only published work that has been done at the
Boulder Creek earthflow.
Anomalously large for earthflows in the region, the ~3 km2 Boulder Creek
earthflow is as large as the drainages of neighboring tributaries (Figure 2). It exhibits
classic earthflow morphology as described in Keefer and Johnson (1983), extending from
channel to ridge with a bowl-shaped accumulation zone, narrow transport zone, and
bulbous toe. It is 5 km in length from the top of the accumulation zone to the bottom of
the toe along the Eel River. At its widest, the accumulation zone is 3 km across. The
transport zone is almost 2 km long and consistently 0.5 km wide. The toe is 1 km long
and 0.75 km wide. The downslope gradient in the transport zone is oriented 2400 (east-
of-north). The accumulation zone exhibits a wide range of slopes and downslope
orientations due to its bowl-shaped structure.
8
9Figure 2. The Boulder Creek earthflow. Fig. 2a) Aerial photograph of the Boulder
Creek earthflow. Fig. 2b) LiDAR shaded relief image of the Boulder Creek earthflow.
10
A prominent lithologic boundary within the earthflow appears to playa dominant
role in the spatial distribution of movement. The boundary cuts obliquely across the slide
(NW-SE) near the top of the transport zone (Figure 2). Above this contour there is a
relatively flat bench-like region which marks the top of the transport zone. Below this
line the slope becomes oversteepened (23°) for a short distance before adjusting to the
slope angle of the rest of the transport zone. This strip of more resistant bedrock can be
seen extending beyond the extent of Boulder Creek when looking at the bare shaded
relief map created by LiDAR data in Figure 2.
Gullies have been interpreted to be a key component of sediment removal on the
surface of earthflows (Kelsey, 1978; Harvey, 2001, see references therein). Boulder
Creek follows the classic earthflow model in this regard with one prominent longitudinal
gully extending the entire length of the transport zone as well as several smaller gullies
throughout the transport and accumulation zones. LiDAR records the "central" gully to
be as deep as 10 meters in places. Several smaller gullies have developed in the southern
half of the transport zone, but very few in the northern half. There is also a significant
gully that appears to mark the northern shear margin of the transport zone. All of these
gullies deflect around the toe rather than cut through it before reaching the Eel River
(Figure 2).
Earthflows extending from channel to ridge are often strongly coupled to the
channel. Sometimes earthflows move when the river erodes material at the toe,
essentially removing the earthflows natural buttressing (Keefer and Johnson, 1983). At
other times the earthflow encroaches on, dams, or permanently pushes the river to a new
location. From observations of the shape of the river, the Boulder Creek earthflow
appears to have pushed the Eel River a few hundred meters to the west. Boulder Creek
resides on an outside bend of the river. This is a common location for slope instabilities
due to high fluvial shear stresses on the channel margins, which can instigate bank
instabilities.
11
12
CHAPTER III
DATA AND METHODS
Overview
To constrain the spatial patterns and magnitudes of movement at the Boulder
Creek earthflow, I use synthetic aperture radar data from the ALOS satellite
supplemented with decadal-scale historical rates derived from orthorectified aerial
photographs and LiDAR data. InSAR data provides detailed information on the spatial
distribution of movement over one year. Decadal-scale Boulder Creek movement is
tracked using an orthorectified aerial photograph taken in 1944 and LiDAR flown in
October 2006 (Mackey, personal cornrnuncation). The positions of 259 trees growing on
the earthflow surface were tracked over this 62 year period to create vectors of horizontal
movement. These vectors (referred to as "tree vectors") represent a historical record of
magnitude and direction of Boulder Creek earthflow movement (Mackey, personal
communcation). High-resolution (~1 m) topography from LiDAR data also contributes a
map of terrain slope and allows for a detailed investigation of the relationships between
velocity and surficial features (Figure 3).
13
Figure 3. Tree vectors representing average yearly rates of individual trees on the
surface of the Boulder Creek eatihflow, tracked from 1944 through 2006. Active lobes
in the accumulation zone are labeled.
InSAR Data Specifics
The ALaS satellite was launched by the Japanese Space Agency in 2006. Boulder
Creek lies within the overlapping region covered by ascending satellite paths 223 and
224. The satellite line-of-sight (LOS) heading is 075° (east-of-north) with a look angle of
34.3° downward. For both paths, Boulder Creek lies within Frame 790 (Figure 4). Data
are collected over the region approximately every 1.5 months. For path 223, 5 synthetic
aperture radar scenes were used to make 9 interferograms spanning from March 14, 2007
to September 14,2007. For path 224,8 scenes were used to make 17 interferograms
14
spanning from February 13,2007 through February 16,2008 (Table 1). The availability
of data for this study was limited to a short window of time due to the recent launch of
the satellite. ERS data was also examined, but proved incoherent over the field area.
A
c
20 km
u
,
o
Figure 4. Location of ALOS paths 223 and 224. The Boulder Creek earthflow is located
in the overlapping region of the two tracks and labeled with a red star.
15
Interferogram Formation
To create an interferogram, a satellite reflects radio waves off of the earth's
surface at two separate times, recording the amplitude and phase of the returning signal
each time. The reflected chirps are then projected back to the individual scatters on the
ground, using the Doppler shift and two-way travel time to produce a single-look
complex image for each satellite pass. The two complex synthetic aperture radar (SAR)
images are interfered by multiplying the first by the complex conjugate of the second
(Massonnett et aI., 1995). The resulting phase shift indicates the distance the surface
scatterers have moved toward or away from the satellite as well as the surface height
from a reference ellipsoid. By removing the topographic signal in the phase using a
DEM, the surface deformation is isolated. The final step in creating an interferogram is
unwrapping, which involves transforming the repeated 2:rt cycles of deformation into a
cumulative amount of deformation. Every 2:rt cycle of deformation represents 11.8 cm of
line-of-sight (LOS) deformation for the ALaS satellite (Massonnet and Feigl, 1998).
For this study, I processed the SAR data using the ROI_PAC (Repeat Orbit
Interferometry Package) software suite developed at JPLlCaltech (Rosen et aI., 2004).
Raw ALaS SAR data is first converted to single look complex (SLC) images. A I-arc
second (30 m resolution) digital elevation model (DEM) collected during the Shuttle
Remote Topography Mission (SRTM) is used to remove the topographic component
from the phase and then the SLCs are interfered. Interferograms were processed at the
standard 4-100ks resolution with standard processing parameters. The displacement field
obtained from my InSAR analysis continuously covers an area approximately 70x70 km2
16
with a pixel resolution of~30 meters. Line-of-sight displacement values are precise to the
sub-centimeter scale.
Unwrapping Errors and Decorrelation
Phase decorrelation occurs in areas where the phase in neighboring pixels is not
spatially correlated. Decorrelation is caused by low coherence, which can be quantified
using the following expression:
< CIC * 2>Y= ----;=======
-J<C£*2><C£*I> (1)
where Cj is the complex phase measurement for scene i, c* is the complex conjugate, and
the brackets < > represent ensemble averaging. Generally, pixels of low yare disregarded
in the analysis because they do not show quantifiable evidence of deformation. Here, the
decorrelated patches are used to infer locations of rapid ground movement.
Decorrelation can be caused by a number of phenomena. Anything that causes the
phase to appear random or speckled, such as a change in the properties of the surface
scatterers will create an incoherent patch. If the phase change across a single pixel is 2Jt
or more, the interferometric signal also decorrelates. This can be due to steep topography,
differential surface motion, or large antenna baseline (Blirgmann et aI., 2000). Baseline
decorrelation occurs when the satellite path for the two passes is sufficiently far apart that
the orientation of the LOS look vector differs, rendering the technique's ability to detect
surface deformation infeasible. All but two interferograms with a perpendicular baseline
of more than 1 kilometer were not used in this study because of baseline decorrelation.
17
The exceptions to this baseline threshold exhibited complete coherence and fully
captured the landslide signal over Boulder Creek.
Two problems were discovered during data processing. The first issue regarded
the power spectral filter, which is used to remove high frequency noise from the
interferograms. It was found to be dampening the landslide signal. Reducing this filter
strength from the default value of 0.75 to 0.30 for most interferograms proved to preserve
more of the signal during unwrapping while also minimizing the random phase speckle.
The second problem was an unwrapping error I eventually attributed to very steep
deformation gradients at the margins of the transport zone. This processing error reduced
and in some cases entirely removed the signal due to landslide motion at Boulder Creek
but did not result in any decorrelation. By manually adjusting parameters and utilizing
additional ROI PAC tools, I was able to correct the error. The error consistently occurred
in a fast moving region of the transport zone. The ROI PAC scripts that unwrap the 2n
cycling of the phase failed to properly integrate the phase in the areas with high
displacement gradients. I discovered the problem by seeing unexpected jumps in the
range change (Figure 5a). Where deformation was expected to increase (in the heart of
the transport zone), it instead decreased dramatically and then resumed the expected
pattern of deformation (as determined by "good" interferograms). This unwrapping error
occurred in four interferograms from path 223 and six interferograms from path 224
(Table 1).
The solution for this error was to manually set the phase jump across this shear
boundary. The problematic region was isolated by constructing a mask and the number of
18
2n: phase cycles was manually assigned. The BRIDGE function within ROlPAC was
used to unwrap the region with the added cycles of deformation.
The spatial extent of the region needing repair and the number of cycles to add
was determined by examining pairs of short-term interferograms (with no unwrapping
errors) that together span the same duration as the interferogram exhibiting unwrapping
errors. The sum of the deformation from the two short-term interferograms should equal
the deformation from the long interferogram. By comparing where the deformation does
not correspond, the spatial extent and magnitude of unwrapping errors can be determined.
(Figure 5b).
19
a) (1) May 16 ~ Jull + (2) Jull - Aug 16
A Unexpecl d
!
b) = 05/16 - 08/16 bridged
A'
Le end
Gl!ner.J1 d ·lerm.Hien
trE'tld f.,J IlhJ~tr.l1 !J
(1)'12)
pt-rt
del mUll nldr
I "yIG" ",If,
3.5 A
1.5
-0.5
-2.5
-4.5
-65
Amr)u t f
l'ima!th
-.~, f.J..d·ln~
A'
Figure 5. Example of correction for unwrapping error due to large deformation gradients in
the transport zone. Fig. 5a) Two short-term interferograms, May 16 - July 1 and July 1 ­
August 16 (both of which do not contain the error). The total deformation over these two
intervals should be equivalent to the amount of deformation in the long interferogram
spanning the total duration, May 16 - August 16. Grey areas indicate where phase is
decorre1ated. Profiles from A to A' of displacement (in radians) for each interferogram are
shown below each interferogram. The dashed black line indicates the expected trend of
deformation (smooth acceleration (more negative) towards the center of the transport zone
and slowing down as the edges are approached). Interferograms (1) and (2) show an
expected profile of deformation. Note the unexpected phase jump in the long interferogram,
altering the profile and highlighting the unwrapping error. Fig. 5b) The corrected
interferogram for May 16 - August 16 is plotted after manual bridging is performed. The
corrected profile is shown (in red) along with the uncorrected (green) and the expected (sum
of (1) and (2)) (in blue).
20
Stacking Interferograms
By stacking multiple interferograms, I determine the average LOS rate of
deformation over a given period of time. This method reduces noise and emphasizes
features that are most consistent though time. Stacking also highlights areas with small
amounts of deformation that may not be resolved using single interferograms. Stacking
consists of removing the mean phase value from each interferogram. Then, pixel-by-
pixel, the range change is summed and divided by the total duration of all the
interferograms. For the ith pixel:
n
~ cpn,i
<Pi = ---,-1__
n
~I!!.tn,i
1
(2)
where <Pi is the average phase change, CPn,i is the phase of the nth interferogram, and Mn,i
is the time span of the interferogram. If a pixel is incoherent for anyone interferogram,
the algorithm removes that value from the calculation.
Interferograms from different paths must be stacked separately because of the
difference in look angle (33° for path 223 and 35° of path 224) (Table 1). Interferograms
with a temporal baseline greater than one year are completely decorrelated and not used.
Atmospheric Artifacts
Water vapor, generally in the form of clouds or fog, can contribute significant
phase artifacts (Zebker et aI., 1997). Artifacts following the valleys and ridges can be
seen in many interferograms, which are likely of atmospheric origin. The dimensions of
21
Boulder Creek, however, are much smaller than the typical dimensions of clouds (tens of
kilometers) or the other long wavelength artifacts unique to single interferograms.
Therefore these artifacts in any interferogram should not mask the signal of interest in a
stack. However, persistent fog in valley bottoms may cause a systematic signal ambiguity
near the toe of Boulder Creek unrelated to landslide deformation. All SAR data for this
study was flown around 6:30am when fog often fills the Eel River valley up to tens of
meters above river level, potentially altering signal strength on the toe of Boulder Creek.
Since atmospheric artifacts are associated with individual SAR scenes, stacking helps to
minimize the potential impact of an artifact in a given interferogram.
Projecting Line-oi-sight Deformation Onto Landscape Surface
Quantifying downslope rates of the Boulder Creek earthflow are necessary in
order to compare InSAR rates to tree vector rates as well as investigate earthflow
dynamics, such as sediment flux rates and the role of surficial features. Because InSAR
reveals only one component of the three dimensional displacement, additional constraints
on the complete deformation field are needed to extract the downslope component of
deformation. For this analysis, I assume that the horizontal component of deformation
occurs parallel to the direction of the tree vectors. The gradient of the unit displacement
vector was derived from the LiDAR data; ArcGIS was used to calculate the slope at a
resolution of 60 meters. Tree vector orientations were interpolated across the earthflow at
30 meter resolution to provide the horizontal component of the unit displacement vector
22
(Figure 6). This calculation was performed across the slide pixel-by-pixel to allow for
variations in heading and dip.
Azimuth (x- & y- components) Surface dip (Z-COl1lpOnent)
<,
"
o· (nat}
Figure 6. Azimuth and dip of downslope eatthflow movement. On the left is the heading of
eatthflow movement derived from tree vector data. Tree vector headings were interpolated
across the slide at 30m pixel resolution. The image on the right displays surface dip derived
from LiDAR DEM.
The magnitude of downslope velocity was calculated using the following equation
that relates LOS deformation to the complete 3D deformation field.
II ~ II L1¢iDi = A AD"l( (3)
where Dj is the line-of-sight range change (from the stack), IIDill is the magnitude of the
deformation vector at pixel i, and I is a unit vector in the direction of the satellite LOS.
For each pixel, the LOS signal is projected onto the surface using the local surface dip for
23
the vertical component and the tree vectors for the horizontal component, creating a map
of downslope velocity values averaged over the stack duration.
Time Series ofDeformation
A time series of Boulder Creek deformation from February 13,2007 to February
16, 2008 was generated from 17 interferograms on path 224 (Table 1). The algorithm
(Schmidt & Blirgmann, 2003) calculates the relative change in phase between epochs for
each pixel (30m x 30m) in the scene. The algorithm used to perform the time series
inversion uses single value decomposition to linearly invert for the average phase change
per epoch. Pixels that are decorrelated in any single interferogram are removed from the
inversion. Because of significant decorrelation, the bulk of the transport zone is not
included in the time series. A location in the transport zone just below the sharp break in
slope was recoverable, as was the small tributary earthflow and the toe. To generate the
location-specific time series plots, the phase values for a small region (90x90 meters) are
averaged together for each epoch and plotted relative to another stable location in the
scene. The stable location was determined by experimentation and examination of the
LiDAR.
24
CHAPTER IV
RESULTS
Overview
In this section I present the results from 26 ALOS interferograms in the form of
stacks to examine spatial patterns of movement and a time series to explore temporal
patterns ofmovement as well as a look at some of the exemplary individual
interferograms. A time series inversion was performed on the path 224 interferograms to
investigate the time dependency ofmovement. Large deformation rates and deformation
orientation were the primary challenges to this study.
Regional Landslide Evaluation & Individual Interferogram Observations
The Boulder Creek earthflow is one of five distinct and large earthflows seen in
nearly all ofthe individual interferograms from path 223. The Taliaferro landslide
complex was also identified. Boulder Creek was the only apparent earthflow captured by
path 224. While this study will only investigate the Boulder Creek earthflow in detail, it
is important to emphasize the power ofL-band InSAR to capture a regional view of real­
time earthflow activity in vegetated areas such as northern California. Figure 7 shows the
location and extent of the five large landslides seen in the data from satellite path 223.
25
Rel.Hi l! In"l~ u! \i~hl
,jIS~14HI'lr -(Il at
(rnjyr)
leIJ2 IIebl008
0045
o
Figure 7. Regional view highlighting five earthflows found in path 223. Images are
stacked using the 9 interferograms on path 223 spanning a duration of six months from
March 14,2007 to September 14,2007.
All of the individual interferograms for this study record some amount of
localized deformation at the Boulder Creek emthflow. Figure 8 presents the Boulder
Creek earthflow in several individual interferograms from paths 223 and 224 highlighting
various features of interest including localized signals in the transport zone and toe
region, winter decOlTelation due to large deformation gradients, and ideal interferograms.
The nearly identical interferograms shown in Figures 8a and 8b highlight the consistent
movement pattern throughout the year. Figures 8c and 8d span winter months and
26
highlight decorrelation (grey areas) because ofthe increased amounts of deformation.
Figures 8e and 8f display the localized toe signal along the Eel River and the overall
inactivity of the toe region.
2
LOS displacement in radians
,---------------,r------------..,
1.9 rad
68 r"d
L.4 rJ
1\ 3 rdd
Figure 8. Select individual interferograms of the Boulder Creek earthflow. Fig. 8a&b)
Exemplary interferograms with complete data coverage. These interferograms
highlight increased activity in the transpo11 zone. Fig. 8c&d) These interferograms
show decorrelation in winter interferograms due to large amounts of deformation. Fig.
8e&f) close-up of the slump signal at the base of the earthflow, along the Eel River.
27
As discussed in Chapter 3, the satellite look direction is 75°east-of-north, which is
nearly sub-parallel to the overall direction of flow (240°). The earthflow is traveling
almost directly towards the satellite, allowing for the optimal imaging of the spatial
patterns of movement at the Boulder Creek earthflow. To interpret the interferograms,
one must first understand the relationship between range change measured by the satellite
and deformation on the surface. Westward movement of the earthflow translates to a
decrease in phase because the surface is moving closer to the satellite, reducing the range
between the two. Downward movement, such as subsidence or deflation, translates to an
increase in phase because the surface is moving further away from the satellite. Based on
average values for the heading and dip of the transport zone, 240° and 14° respectively, 1
mm of westward movement would cancel out almost 3 mm of subsidence. The lack of
additional constraints on the complete deformation field requires an assumption that all of
the motion captured by the satellite is parallel to the surface. In other words, any
synchronous subsidence or inflation will be interpreted as horizontal deformation. This is
a small source of error and will not significantly affect interpretations of the LOS or
projection results.
Detailed Examination ofBoulder Creek Stacks
Interferograms are stacked over the summer months because of data availability.
Nine interferograms from path 223 are stacked to generate an average summer LOS
deformation rate spanning March 14,2007 to September 14,2007. Six interferograms
from path 224 are stacked to represent a comparable rate spanning May 16, 2007 to
28
October 1,2007. An average yearly rate is presented by stacking all 17 interferograms
from path 224 (February 13,2007 to February 16,2008). For L-band InSAR with a pixel
width of~30 meters and an epoch length of 46 days, the maximum theoretical observable
deformation rate is 1.3 m/yr along the satellite LOS (Calabro, 2008). The maximum LOS
summer rate (~0.5 m/yr) is well below this value. The interferograms used to make
summer stacks have complete data coverage with little or no decorrelation. Accelerated
landslide movement occurs during wet, winter months, enough to overcome the 1.3 m/yr
threshold and cause patches of decorrelation in the transport zone (Hilley et aI., 2004;
Iverson, 2000). Boulder Creek deformation in the dry season is observed to be well below
InSAR's limit of resolving surface deformation. Covering similar periods of time and
displaying nearly identical spatial patterns and rates of movement, summer stacks on
paths 223 and 224 are discussed together.
Summer stacks on paths 223 and 224 image localized deformation at the Boulder
Creek earthflow (Figure 9). Stacks are draped over a high-resolution hillshade image
from LiDAR data to highlight surface features. Tree vectors representing historical
average horizontal rates are also displayed for comparison. The light green fringe
representing an average summer LOS rate of ~0.1 m/yr generally follows the
morphologic boundaries of the earthflow. Beyond this fringe and off the earthflow, the
blue regions represent areas of negligible movement. The maximum average LOS
summer rate observed for Boulder Creek is ~0.5 m/yr, located in the transport zone.
Other areas of interest include slumping at the toe margin along the Eel River, a small
tributary earthflow along the south margin, and regions of continuous movement in the
29
accumulation zone. The spatial variation of LOS rates is in close agreement with tree
vector relative magnitudes, indicating that the relative spatial pattern of deformation is
the same over the short and longer time frame. Most notably, the regions of greatest
activity remain the same.
Track 223
March 14,2007 - September 14, 2007
9 inter 'erogram~
Track 224
Febru ry 13, 2007 - February 16, 2008
17 inlerl~r gr"m~
LOS displacement rate (mm/yr)
Tnbul tv n
80
500
awaylrom t
sawllil~
Figure 9. Stack of interferograms for satellite paths 223 and 224, representing average
LOS rates (mrn/yr) for the duration of the data. Fig. 9a) Stack 223 spans March 14,2007­
September 14,2007. Fig. 9b)Stack 224 spans February 13, 2007-Febmary 16,2008. Fig.
9c) Stack 224 with tree vector overlay.
InSAR and the tree vectors indicate that the majority of movement in the summer
on Boulder Creek is occurring within the transport zone. The transport zone is defined as
the elongate midportion of an earthflow that connects the region of collection
30
(accumulation zone) with the region of deposition (toe). Flow in the transport zone is
characterized as "plug" flow (Kelsey, 1978) where the material undergoes block-glide
style deformation and moves downslope as a coherent unit with little internal
deformation. LOS summer deformation rates are two orders of magnitude higher here
than elsewhere on the slide, reaching a maximum LOS rate of ~0.5 m/yr.
A deeply incised gully (~10 m deep) bisects the transport zone. Nearly all
movement in the transport zone observed by InSAR occurs within the southern "half' of
the transport zone ("half' defined as the region south of the bisecting gully). Movement
rates drop off steeply beyond this region. Within this region, LOS deformation rates are
the highest in the center at 0.45 m/yr and decelerate towards the boundaries with rates
along the gullies of ~0.3 m/yr. Additional movement beyond the borders just described
occurs at rates of 0.1-0.2 m/yr.
A steep velocity gradient occurs at the top end of the transport zone. This
coincides with a sharp break in slope, also called a knickpoint, where the slide surface
changes from a relatively flat bench with a dip of nearly 0° to a very steep drop-off with
an average slope of 23° degrees. This knickpoint is correlated to a lithologic boundary
that can be seen extending well beyond Boulder Creek in the LiDAR data. Here I
presume that this narrow band of more-resistant rock is responsible for the knickpoint,
below which the earthflow gains speed. While there is not such a steep gradient of
deformation at the bottom end of the transport zone, the green fringe representing a LOS
rate of ~0.15 m/yr noticeably stops at the beginning of the toe, highlighting the
corresponding change in movement with changes in surface features.
31
Because the tree vectors represent horizontal motion and the InSAR results reflect
the component of displacement along the satellite's LOS, which is predominantly
vertical, tree vector rates cannot be directly compared to LOS rates. They can, however,
corroborate the spatial distribution of movement captured by InSAR. The tree vectors
agree with significantly higher rates in the southern half of the transport zone. They also
agree with a distinct deceleration when approaching the toe. The tree vectors highlight
one region of discrepancy between the two datasets in the transport zone. Along the
northern margin of the transport zone tree vectors indicate a region of westward
movement that is slightly oblique to the overall direction of motion (240°). InSAR
records the LOS rates in this area to be negligible. This could be due to little or no
deformation during 2007 or because the motion was close to orthogonal to the satellite's
look direction, as discussed more fully in Chapter 5.
The stacks highlight a small tributary flow entering the transport zone near the toe
on the south side at a LOS rate of ~0.15 m/yr. There are no tree vectors to corroborate
this InSAR result, however the aerial photographs used to derive the tree vectors indicate
historical movement of this feature from the breakup and transport of a large rock outcrop
(Mackey, personal communication).
The toe of the Boulder Creek earthflow is large, bulbous, and appears to be
moving very little according to both InSAR and tree vectors observations. This classic
shape for earthflow toes is representative of the region where material from the transport
zone is deposited. Despite the general inactivity of the toe, a distinct InSAR signal
32
persists at the steep earthflow front along the Eel River. The average summer LOS
deformation rate for this region is ~O.12 m1yr.
There are no tree vectors at the margin of the toe to corroborate the InSAR signal
and the region may be within the zone of persistent morning fog, both of which lead to
uncertainty in the interpretation of this signal. However, the signal is isolated to one
portion along the toe margin and is situated directly over a fresh-looking arcuate scarp
seen in the LiDAR data (Figure 8e & f). Were the signal due to fog alone, it would appear
along the entire margin. This supports the interpretation that the signal is due to a
localized slump. The magnitude of the signal may have been corrupted by contributions
from water vapor; however, the relative change in magnitude over time can still provide
insight into landslide dynamics. Another possible source of the signal is an error in the
DEM, but that cannot be argued without further evidence.
The accumulation zone is a large bowl-shaped hollow where hillslope material is
collected and funneled into the transport zone (Kelsey, 1978). Collection zones
characteristically exhibit less continuous deformations styles such as slumping and fault
block crumbling (Nolan and Janda, 1995). Little deformation is observed in the
accumulation zone from InSAR data. What is observed is a relatively linear path directly
in-line with the transport zone moving at LOS rates of ~O.1-0.15 m1yr. Tree-vector data
for this region denote three lobes of continuous movement. InSAR activity corresponds
to the central lobe of tree vector movement, however does not detect movement on the
other two lobes. These styles of internal deformation are expected to cause decorrelation
in the interferograms. Instead there is a spatially extensive coherent signal with negligible
33
magnitudes. While these non-continuous deformation styles are most likely occurring in
part, the tree vectors prove that continuous deformation is also occurring within the
region. The other possible reasons for the lack of InSAR movement are the wide range of
slope orientations and local terrain roughness. Of the three active lobes captured by the
tree vectors, only the central lobe is in-line with the satellite look direction. The north and
south lobes are much closer to being orthogonal to the satellite LOS.
Downslope Velocity Projections
As described in Chapter 3, the LOS deformation rates averaged over 2007
summer months (stacks) were projected onto the slope using tree vector orientations for
the horizontal component and slope derived from the DEM for the vertical component of
the complete deformation field. The results of the projection can be seen in Figure 10.
Using the average heading and dip for the transport zone (2400 and 140 , respectively), the
projection results in a downslope velocity ~3 times larger than the satellite LOS rate.
Deviations from this average occur where orientations of the surface are not optimal for
imaging by the satellite. This is notable along the northern margin of the transport zone
where the tree vectors indicate oblique motion of ~0.5 rn/yr directly towards the gully
rather than straight down the transport zone. The InSAR projection, however, reports a
negligible velocity along the margin.
At some locations on the slide, aspect and dip combinations approach angles that
are orthogonal to the satellite LOS. This results in a singularity in the computation of the
downslope velocity for a pixel. Pixels with the scaling factor iJ·[ (from equation 3)
34
between -0.06 and +0.06 were removed from the calculation and do not appear in the
final projection images. Recall that the satellite look direction is nearly sUb-parallel to the
direction of flow. This is why there are few singularities in the transport zone. Deeply
incised gullies in the transport zone do, however, result in a small number of singularities
because of the local terrain roughness they create. There are many singularities in the
accumulation zone due to complex surface morhpology and the wide range of slope
orientations. These singularities result in low data coverage for the region. Because
InSAR captured so little movement in the accumulation zone it was already a poor
candidate for the projection. Therefore the downslope projection is most appropriate for
the transport zone and toe region of Boulder Creek.
35
Track 223
March 14, 2007 - September 14, 2007
9 ir tt:rferogr.)m~
Downhill displacement rate (m/yr)
1,5
o
2 m/vr
Track 224
February 3,2007 - February 16, 2008
17 int~rlero81<1m,
Figure 10. Downslope velocity projections for each stack. Colorscale saturated at 0 and 3
m/yr. Values below 0 m/yr appear a deep blue. Values above 3 m/yr appear red. These
extreme values are interpreted to be artifacts of the projection due to complex surface
morphology and are not physically representative. This projection is most relevant for the
intermediate values found in the transport zone. Fig. lOa) Downslope velocity for path
224. Fig. lOb) Downslope velocity for path 224. Fig 1Oc) Downslope velocity for path
224 with tree vector overlay.
The downslope velocity values from InSAR range from 0.5 to 1.5 m/yr in the
transport zone and ~0.3 m/yr elsewhere. Tree vectors for this region measure 2.0 to 2.5
m/yr in the transport zone and ~0.5 m/yr elsewhere. As the InSAR-derived rates are for
summer only on path 223 and weighted towards the summer months on path 224 (due too
loss of coherence in the transport zone in winter interferograms), it is expected that they
will be lower than the tree vector rates, which record yearly averages including
accelerated movement in the winter. Tree vector rates are consistently around 2 times
36
larger than the InSAR-derived summer rates, thus corroborating the InSAR projection
results.
Seasonal Signal
A strong seasonal signal can be seen in all regions of the earthflow included in the
inversion (Figure 11). Rates of movement are plotted in millimeters per day and represent
an average rate for the given epoch. All epochs are 46 days except for the first, which is
92 days. Movement decelerates in the spring of 2007 in concert with the rain. The
earthflow is still slowing down in October 2007 when the winter rains begin again. It is
not until the beginning of January 2008 that the earthflow begins to accelerate. This ~2
month lag can be seen in all three locations plotted in Figure 11, however the magnitude
of movement lowers by an order of magnitude as you progress from the transport zone to
the small tributary flow to the slumping signal at the toe margin. The transport zone
(Figure 11, red diamonds) reaches a peak rate of 1.2 mm/day (0.44 m/yr) in June 2007
and then steadily slows down to 0.5 mm/day (0.18 m/yr) by the end of December. At this
point the transport zone begins accelerating again to a final rate of 0.75 mm/day (0.27
m/yr). This same pattern of deceleration is seen in the small tributary flow entering the
transport zone near the toe (green diamonds), however its fastest rate is 0.3 mm/day and
slows to nearly zero by the end of the year. The toe slump (yellow diamonds) also shows
a dampened but consistent pattern of maximum speed in June and slowing to nearly zero
before beginning to accelerate again in 2008.
37
Possible sease ality signal
February 2007-Febru ry 2008
>:
bO 1'0
-0
-}I'; E
Ll)
c:
0
;0 'S:;
19
.:iL! 'Q.
U
J'. Q.J
a.
N 2='
'ro
Ie 0
"
'-- T_Jl_IJ_1_'V_no_\\' 010< IC',J1ge'Jled 10.) -l)J I '"1IJI. I
c:
Q.JE (In
Q.J
U
~ Il
a.
VI
"0 01
~
<! u
Nov
fL!\f\l> ~hd,.:
b~Cln '" ~hllsifH It
211"Jlllhl !>: I'1'0
-0
-
1 IE
E
-
IIIQ.J
....
1'0
~
Figure 11. February 2007 to February 2008 time series of average daily velocities for
three regions of the Boulder Creek earthflow. Daily rainfall collected at the Zenia
meteorological station, approximately 15km north of Boulder Creek, is plotted in blue.
The red diamonds represent movement in the transport zone. Green diamonds represent a
location in the tributary flow and yellow diamonds represent movement at the margin of
the toe.
38
CHAPTER V
DISCUSSION OF RESULTS
Overview
In the previous chapter, I analyzed ALOS satellite data in conjunction with
historical movement rates and LiDAR data in an effort to constrain spatial and temporal
patterns of movement at the Boulder Creek earthflow. In this chapter, I discuss the
implications of interpreting landslide movement measured in the satellite LOS as well as
how patterns of surface movement through space and time relate to earthflow
morphology and dynamics.
Investigating the Spatial and Temporal Variations in Velocity
The downslope velocity projections represent earthflow movement averaged over
one year, from February 13,2007 to February 16,2008. While the projections are
estimates of downslope velocities and do not consider synchronous vertical signals (such
as uplift or subsidence of the surface), they provide a first order map of where and how
much the earthflow is moving over one year. The following sections first discuss the
spatial variability of velocities averaged over one year and then the temporal variability
within that year. The spatial variability highlights connections between lithology and
39
gullies and surface movement. Temporal variability highlights seasonal patterns of
movement that correlate with rainfall.
The spatial variability of downslope velocities determined from the InSAR data
highlight distinct zones of activity within the earthflow. Movement is predominantly
occurring in the transport zone with minimal activity seen in the accumulation zone and
toe regions. Between this "hot spot" of activity and the Eel River is a relatively inactive
toe, prompting the question of what is driving movement. It is often speculated that
earthflow movement is driven by erosion at the toe by the channel below. The inactive
toe of Boulder Creek indicates that stresses are not being transmitted upslope from the
Eel River. Further, the slump at the toe margin does not appear to have any effect
upslope.
From examining the LiDAR, it appears possible that the earthflow is driven from
above by material collected in the accumulation zone and funneled into the transport
zone. A lobe of material can be seen approaching the sharp break in slope near the top of
the transport zone. I interpret this to suggest a more episodic supply of material to the
transport zone; when enough material collects it forms lobes with enough force from
above to advance over the flat bench top and enter the transport zone conveyor belt.
With an inactive toe, gullies appear to be the primary mechanism connecting the
channel to the transport zone. Several deep gullies «15 m) dissect the fastest regions of
the transport zone. Downslope, the gullies coalesce and wrap around both sides of the
toe before merging with the Eel River. It is also notable that the gullies are absent from
the slow regions of the transport zone (the northern half). This suggests that the locations
40
of gullies can be associated with regions of activity and that gullies playa large role in
the removal of sediment from the earthflow. Erosion from gullies has been found to be a
significant contributor for sediment flux in channels over short and long term (Harvey,
200 I; Betts et aI., 2003).
In addition to the accumulation zone, a smaller amount of material enters the
transport zone from the tributary earthflow. This region, too, abuts the locked toe region
and again gullies appear to be responsible for transporting that material off the earthflow.
A deep gully forms immediately downslope from where the tributary flow enters the
transport zone, again implying a direct relationship between gullies and regions of
activity and sediment removal.
The rapid increase in velocity after the sharp break in slope near the top of the
transport zone provides insight into the depth and shape of the Boulder Creek earthflow.
This velocity increase is most likely due to the lithologic boundary discussed in Chapter
4. Above this boundary the surface is relatively flat and immediately below it the surface
dips steeply, up to 23 0 , for ~100 meters before leveling out to the average transport zone
dip of~14o. Because of the broad extent and sharp linearity of the feature, this lithologic
boundary is likely to have persisted through much of Boulder Creek's history and appears
to largely control the location of the top of the transport zone. Cae et al. (2008)
determined that when landslides have moved distances larger than the dimensions of the
largest basal topographic irregularities, landslide surface morphology can be used as a
guide to the morphology of the basal slip surface. Due to its large size and significant
amounts of deformation over historic timescales (~150 meters in 64 years according to
41
tree vectors), I assume that such an argument about surface morphology mimicking a
basal slip surface is relevant for Boulder Creek, at least locally in the upper reaches of the
transport zone. The prominent knickpoint suggests that bedrock is close to the surface at
that location, which would mean that the earthflow is quite thin there. Transport zone
thickness cannot be determined, however, fast-moving flows are often less thick than
slower ones of the same surface area in order to balance the amount of material entering
and leaving the region. This suggests a gradual thickening just below the knickpoint to
some thickness which is most likely maintained along the transport zone.
Individual interferograms of Boulder Creek provide information on seasonal
variations in rates and styles of movement. Interferograms spanning summer months
show continuous data coverage and lower rates than winter months (Figure 8a & b).
Small patches of decorrelation in interferograms spanning winter months are attributed to
large amounts of deformation or internal deformation (Figure 8c & d) (Calabro, 2008).
Because of the high precipitation in the winter, this is when the earthflow is expected to
move most rapidly. Decorrelation patches occur in regions where the most activity is
expected: over gullies in the transport zone.
The time series of range change explores the relationship between precipitation
and slide movement. Despite the drastic spatial variability in velocities, the time series
highlights a similar temporal pattern across the slide which correlates strongly with
seasonal rainfall patterns. Figure 11 plots average LOS velocities (mm/day) at different
locations on the earthflow along with daily rainfall data collected by the California
Department of Water Resources at the Zenia meteorological station, approximately 15
42
kIn north of the study site (Figure 1). While the magnitudes of the rates are not directly
applicable because they are in the LOS and not downslope, the relative change provides
insight into seasonal patterns. To the first order, it can be seen that as the rains begin in
October 2007, the earthflow is still decelerating. However, about 2 months later, at the
beginning of January, the earthflow begins to accelerate. I interpret this to be a delayed
response to rainfall, which would suggest that the Boulder Creek earthflow is
hydrologically driven. As the water table rises, pore pressures and sliding velocities
increase. When the water table lowers, pore pressures and sliding velocities decrease
(Ehlig, 1992). Iverson (2000) describes the relationship between sliding rate and time
since rainfall onset. The diffusion of pore pressure through the slide decreases the normal
force and frictional resistance, thereby allowing the slide to accelerate. As pore pressure
diffuses through the slide there is a lag between the onset of rainfall and the acceleration
of the slide. The relatively long lag time of ~2 months could indicate low diffusivity for
the earthflow, perhaps caused by the clay-rich material or a deep basal shear zone (Keefer
and Johnson, 1983.
Despite being much larger than most documented earthflows in the region,
Boulder Creek appears to be similar in its rates and styles of movement. Velocities in the
transport zone of Boulder Creek approach a yearly average of 2 m/yr. Elsewhere on the
slide velocities range from 0 - 1 m/yr. Earthflow velocities at other field sites have been
observed to range from 0.5 m/yr to 6 m/yr with occasional surges of tens of meters in a
matter of days. The Minor Creek earthflow in northwestern California was monitored
from 1982 to 1985 (Iverson and Major, 1987). This 800 meter-long eartht10w has a
43
summer creep rate of 1-4 mm/yr and accelerates up to 0.5 m/yr in the winter.
Acceleration into winter rates occurred between November and March depending on the
year. Rapid movement persisted into Mayor June. The inconsistent delays between onset
of seasonal rainfall and onset of rapid movement at different field sites highlight the
episodic nature of earthflow movement. Hilley et at. (2004) observed a lag of ~3 weeks
for earthflows in the East Bay Hills region near San Francisco, California. These
earthflows are ~1 km long and move 27-38 mm/yr downslope, as inferred from InSAR
range change. The Portuguese Bend landslide (~1 km2) on the Palos Verdes peninsula
near Los Angeles, California, shows a lag of approximately 1 month. Due to extensive
monitoring, the depth of this earthflow is known to be 18 meters on average (Calabro,
2008). The fact that rates and styles of deformation observed at Boulder Creek are
consistent with other earthflows in the area suggests that L-band InSAR can successfully
characterize and quantify landslide-related deformation.
Investigating InSAR Line-aI-sight sensitivity
The previous sections and Figure 7 highlight the power ofL-band InSAR to
identify and quantify earthflows in the vegetated terrain ofNorthern California. Prior to
the introduction of this broad remote sensing tool, earthflows were found and studied on
a case-by-case basis. Five large earthflows were identified using InSAR in this study,
thus providing a "snapshot" of real-time regional earthflow activity in the Eel River
basin. An understanding of the limitations that come with measuring deformation using a
single look direction is important, but even one component of deformation is enough to
44
highlight areas of interest and infer relationships between surface deformation and
earthflow morphology and dynamics.
There are a number of limitations that must be considered when interpreting
InSAR data. The largest error source comes from atmospheric artifacts. Other than the
possible fog signal at the base of the earthflow, no significant atmospheric artifacts
appear to be present in any interferograms. Limited temporal sampling, the inability to
see a signal in long-duration interferograms because of high rates, and no descending data
are additional limitations of this study. Temporal sampling is, on average, every 46 days
for this dataset, which is the best that can be attained for the ALaS satellite orbit.
InSAR's sensitivity to the orientation of deformation is highlighted in this study.
Deformation orthogonal to the LOS will go undetected because the interferometric
analysis is insensitive to motion in this direction. The transport zone of Boulder Creek is
optimally oriented for imaging by an ascending InSAR satellite because surface motion is
heading nearly sub-parallel to the satellite look direction. The accumulation zone, on the
other hand, exhibits greater variability in the orientation of surface movement, and the
regions where the expected movement is orthogonal to the LOS show no signal in the
interferograms. The toe region is extremely smooth and the direction of motion is
generally in-line with the transport zone, making it another optimal region for InSAR
imaging. The following sections compare ascending InSAR data (Figure 9) to horizontal
rates derived from aerial photography to investigate what parameters (i.e. deformation
orientation, surface roughness) are most important or limiting when interpreting
earthflow movement using InSAR.
45
While the average LOS rates represent only one component of the total movement
of the Boulder Creek earthflow, they do provide a sufficient level of detail with which
InSAR can resolve landslide-related deformation. Tree vectors could only be created
where trees persist on the earthflow through time. As seen in figure 3, tree vector
coverage is relatively dense across the earthflow, highlighting specific and interesting
regions of activity such as viscous-flow characteristics in the southern half of the
transport zone and oblique movement near the northern margin of the transport zone.
Tree vectors are absent from the tributary flow, where no trees were able to be tracked.
Despite a lack of trees, boulders and shrubs could be seen moving downslope through
time indicating activity of this feature (Mackey, personal communication). In the
following sections I compare spatial patterns of movement detected by InSAR and the
tree vectors.
The extraordinary detail that can be seen in the transport zone highlights why
Boulder Creek was chosen to be the subject of this study. The movement in this region is
heading nearly sub-parallel to the satellite look direction, resulting in the satellite
detecting as much movement as possible. While the other earthflows identified in this
study also demonstrate internal displacement gradients, the large size of the transport
zone of Boulder Creek combined with its optimal orientation provide the best resolution
with which to study earthflow movement. Prominent displacement gradients appear
consistently in all of the individual interferograms highlighting the steady pattern of
movement. The margins of movement seen by the satellite coincide well with the
interpreted geomorphic boundaries of the slide. Similar to what is seen at Boulder Creek,
46
the transport zones at the Kekawaka and Halloween earthflows (see Figure 1; Halloween
Earthflow along the Van Duzen River) also show a similar style of deformation where
material moves as a coherent unit downslope (Mackey et aI., 2009; Kelsey 1978). This
styIe of continuous deformation is ideal for detection by InSAR because of the small
amounts of internal rotation and shear that could otherwise result in too much change in
the surface scatterers and cause decorrelation.
Limitations in resolving deformation that is oblique to the LOS is more apparent
in the accumulation zone. This bowl-shaped collection zone takes material from near the
ridges where gradients are higher, and funnels the material downslope and inward
towards the top ofthe transport zone. Tree vectors indicate three discrete active lobes of
material moving downslope from prominent headscarps towards the transport zone: a
central lobe moving relatively parallel to the transport zone and a lobe on either side
(referred to as the "north" lobe and the "south" lobe), each approximately 30° off the
central lobe heading (Figure 3). The InSAR data captures distinct movement of ~0.13
m1yr (LOS) in the central lobe that nicely agrees with the location of tree vectors, which
record horizontal rates of 1-2 m/yr. The north and south lobes, however, appear to be
inactive according to the InSAR data, but very active according to the tree vectors. Tree
vectors indicate that the north lobe is the most active of the three with horizontal rates
ranging from 0.5 to 1.5 m/yr. I attribute the difference between agreement in the central
lobe and disagreement in the north and south lobes to InSAR's dependence on the
orientation of deformation relative to the LOS look direction. Deformation in the central
47
lobe is optimally oriented with respect to the satellite while the north and south lobes are
approaching orthogonality to the LOS.
I assume discontinuous deformation to be another challenging parameter when
landslide-related movement in the accumulation zone. Deformation styles commonly
observed in collection zones of earthflows, slump blocks, rotation, tension cracks, and
crumbling, result in small-scale surface roughness. Rapid movement and small-scale
rotations from crumbling and rolling material is expected to cause decorrelation in the
interferograms. While the magnitude of LOS deformation is negligible for the north and
south lobes, there remains a spatially extensive coherent signal across the region. The fact
that there is always a coherent signal across the accumulation zone indicates that
deformation is either not occurring or, in the central lobe at least, occurring in a
continuous fashion. This is corroborated by the tree vectors, which highlight continuous
movement in all three lobes. Thus, it appears that the complex surface morphology is not
an impediment to using InSAR in this locality. Spatial averaging during processing and
continuous flow-like motion in the area provides coherent data cross the zone.
As discussed in Chapter 4, an interesting phase signal resides at the steep margin
of the toe along the Eel River. This localized signal (~200xI00 meters) must be
interpreted with caution because of the possible water vapor contribution to the phase. A
signal attributed to fog does not appear in every interferogram but it does appear in both
stacks, indicating that it is a persistent signal. While the magnitude of the signal may be
questionable, the location of the signal proves that it can be attributed to landslide-related
motion. Not only is the signal unique to a small section of the margin, but the LiDAR
48
data highlights a fresh arcuate shape in the same location as the InSAR signal. Were the
signal to be due solely to fog, it would extend across the entire margin (Figure 8e & f).
While LOS displacement rates in the transport and accumulation zones are
assumed to be primarily horizontal, I assume the slump signal to be predominantly
vertical. A rotational slump would likely result in the surface moving downward. Because
the majority of the toe is not moving at all, horizontally or vertically, as indicated by
InSAR and the tree vectors, it is unlikely that the signal of interest would be moving
horizontally. This assumption holds when considering the expected direction of range
change for downward vertical motion. Positive range change indicates that the surface is
moving away from the satellite. This can be interpreted as downslope surface-parallel
movement or subsidence/deflation of the toe margin. The range change observed for the
toe slump, however, is negative. This indicates some form of uplift. Folding and multiple
shear surfaces due to compression from upslope could result in uplift of the toe margin.
Further investigation of this feature is necessary, however, to validate the signal. It is also
possible that fog has affected the signal.
The magnitude of the signal may have been corrupted by contributions from water
vapor and is therefore not directly applicable. I assume the fog to behave as a volumetric
scatterer of the satellite signal. Properties of the fog that could affect the phase signal,
such as thickness, density, and mere presence, can vary from day to day. The calculated
phase change along the river could be reflecting changes in the fog parameters rather than
changes in the surface. The time series discussion later in this chapter, however,
highlights a seasonal variation in displacement for this feature that is identical to patterns
elsewhere on the slide. While this also is an argument for the validity of the signal, it
does not necessarily defend the magnitudes. Further investigation would be required to
gain confidence in this result.
49
50
CHAPTER VI
SUMMARY AND CONCLUSIONS
This study of the Boulder Creek earthflow has incorporated InSAR, aerial photo
analysis, and LiDAR in an attempt to monitor the slide and better understand its
dynamics. Two-pass L-band satellite interferometry is used to determine deformation
rates during the summer months of 2007. The downslope velocity values range from 0.5
to 1.5 m/yr in the transport zone and ~0.3 m/yr elsewhere. A comparison is made
between these results and horizontal rates derived from orthorectified aerial photographs
and LiDAR spanning 1944-2006. Tree vectors for this region measure 2.0 to 2.5 m/yr in
the transport zone and ~0.5 m/yr elsewhere. Tree vector rates are consistently ~2 times
larger than the InSAR-derived summer rates. Because tree vector rates represent yearly
averages, which include accelerated winter movement, and InSAR rates are for the
summer only, these results are in agreement.
My results highlight the ability of L-band InSAR to identify and quantify
earthflow activity in higWy-vegetated terrain. Five large earthflows (including Boulder
Creek) are identified in one frame, providing a snapshot of earthflow activity for a 70km
x 70 km region. Within the borders of the Boulder Creek earthflow, InSAR images
spatial heterogeneities of movement at high spatial resolution (30m x 30m). The most
51
displacement occurs in the transport zone with less activity seen in the accumulation and
toe regions. Within the transport zone, the regions of highest displacements are seen to
correlate with the regions of highest gully density, suggesting that gullies either form in
response to or are the cause of increased surface movement. If they are the cause of
movement, is movement only happening as deep as the gully thalwegs? Perhaps this
creates a positive feedback loop where the gullies remove enough material to lessen the
normal load on the basal shear plane and accelerate movement of the whole unit. If gully
incision is a response to movement, then perhaps the activity breaks up the surface and
encourages channel formation, which would then continue to assist in moving material. A
sharp break in slope attributed to a lithologic boundary appears to dictate the start of the
high displacements of the transport zone.
The time series inversion for path 224 interferograms displays variations in LOS
rates between February, 2007 and February, 2008. All regions of the slide show a
significant deceleration from March 2007 through December 2007 before beginning to
accelerate in January 2008, approximately 2 months after the winter rains began. This
relatively long phase lag between the onset of rain and the onset of increased earthflow
activity could help to constrain physical parameters of the slide such as diffusivity and
depth to a basal shear zone.
This study examined limitations to the interpretation ofInSAR results derived
from a single look vector. InSAR is shown to be sensitive to the orientation of the
deformation such that significant amounts of deformation go unseen if moving
orthogonal to the satellite line-of-sight. Such is the case in the accumulation zone of the
Boulder Creek earthflow, where only one of three lobes of activity observed by the tree
vectors is observed by InSAR.
The study of the Boulder Creek earthflow could be furthered by traditional site­
specific methods such as surveying, boreholes, and stake lines as well as GPS and
additional InSAR data. This would expand temporal data coverage, constrain vertical
signals, verify the spatial extent of movement in the accumulation zone, constrain the
large winter rates in the transport zone which InSAR could not, and investigate the toe
margin "uplift" signal.
52
APPENDIX A
LIST OF INTERFEROGRAMS USED
53
Path Starting Scene Ending Scene Baseline (m) Bridge required?
223 2007 March 14 2007 April 29 -571
223 2007 March 14 2007 June 14 -139 Yes
223 2007 March 14 2007 September 14 -878 Yes
223 2007 April 29 2007 June 14 431
223 2007 April 29 2007 July 30 62 Yes
223 2007 April 29 2007 September 14 -333 Yes
223 2007 June 14 2007 July 30 -369
223 2007 June 14 2007 September 14 -765 Yes
223 2007 July 30 2007 September 14 -395
224 2007 February 13 2007 May 16 -670 Yes
224 2007 February 13 2007 July 1 -775
224 2007 May 16 2007 July 1 3140
224 2007 May 16 2007 August 16 -290 Yes
224 2007 May 16 2007 October 1 -747 Yes
224 2007 July 1 2007 August 16 -277
224 2007 July 1 2007 October 1 -640
224 2007 July 1 2007 November 16 -810 Yes
224 2007 July 1 2008 January 1 -916
224 2007 August 16 2007 October 1 -455
224 2007 August 16 2007 November 16 -620
224 2007 August 16 2008 January 1 -730 Yes
224 2007 October 1 2007 November 16 -170
224 2007 October 1 2008 January 1 -275
224 2007 November 16 2008 January 1 -105
224 2007 November 16 2008 February 16 -1065 Yes
224 2008 January 1 2008 February 16 -960
54
APPENDIXB
GLOSSARY OF TERMS
55
ALaS: Advanced land observation satellite
ERS: European Remote Sensing satellites
GPS: Global Positioning System
InSAR: Interferometric Synthetic Aperture Radar
JAXA: Japan Aerospace Exploration Agency
JERS: Japanese Earth Resources Satellite
LiDAR: Light Detection And Ranging
LOS: Line of Sight
SAR: Synthetic Aperture Radar
56
57
REFERENCES
Betts, H.D., Trustrum, N.A., and De Rose, R.C. (2003). Geomorphic changes in a
complex gully system measured from sequential digital elevation models and
implications for management. Earth Surface Processes and Landforms, 28,
1043-1058.
Burbank D.W., Leland, J., Fielding, E., et al. (1996). Bedrock incision, rock uplift and
threshold hillslopes in the northwestern Himalayas. Nature, 379,505-510.
Burbank, D.W. and Anderson, R.S. (2001). Tectonic Geomorphology. Blackwell Science
Ltd.
Bfugmann, R. et al. (2000). Synthetic aperture radar interferometry to measure Earth's
surface topography and its deformation. Annual Reviews ofEarth and Planetary
Sciences, 28, 169-209.
Calabro, M.D., 2008. An examinatino ofsurface displacement at the Portuguese Bend
landslide, southern California, using radar interferometry. Unpublisehd Master of
Science thesis, University of Oregon, Eugene, Oregon.
Camec, C., Massonnet, D., and King, C. (1996). Two examples of the use ofSAR
interferometry on displacement fields of small spatial extent. Geophysical Research
Letters, 23, 3579- 3582.
Catani, F., Farina, P., Moretti, S., et al. (2005). On the application of SAR interferometry
to geomorphological studies: Estimation of landform attributes and mass
movements. Geomorphology, 66, 119-131.
Colesanti, c., Ferretti, A., Novali, F., et al. (2003). SAR Monitoring of Progressive and
Seasonal Ground Deformation Using the Permanent Scatterers Technique. IEEE
Transactions on Geoscience and Remote Sensing, 41 (7), 1685-1701.
Cruden, D.M. and Varnes, D.J. (1996). Landslide types andprocesses. In Turner, A.K.
and Schuster, R.L. (eds.), Landslides--Investigation and Mitigation. Washington
58
D.C., National Academy Press, Transportation Research Board Special Report 247,
36-75.
Farina, P., Colombo, D., Fumagalli, A., et al. (2004). On the integration of remote
sensing data within landslide risk analysis. International Geological Congress,
Abstracts, 32(2), 183.
Fruneau, B., J. Achache, and C. Delacourt (1996). Observation and modeling ofthe
SaintEtienne-de-Tinee landslide using SAR interferometry. Tectonophysics, 265,
181-190.
Fuller, C. W., Willett, S. D., Hovius, N. and Slingerland, R (2003). Erosion rates for
Taiwan mountain basins: New determinations from suspended sediment records and
a stochastic model of their temporal variation. Geology, 111, 71-87
Furlong, K.P., and Schwartz, S.Y. (2004). Influence ofthe Mendocino triple junction on
the tectonics of coastal California. Annual Review ofEarth and Planetary Sciences,
32,403-433.
Goldstein, RM., Engelhardt, H., Kamb, B., Frolich, RM. (1993). Satellite radar
interferometry for monitoring ice sheet motion; application to an Antarctic ice
stream. Science, 262, 1525-1530.
Harvey, AM. (2001). Coupling between hillslopes and channels in upland fluvial
systems: implications for landscape sensitivity, illustrated from the Howgill Fells,
northwest England. Catena, 42, 225-250.
Hilley, G. E., BOrgmann, R, Ferretti, A, et al. (2004). Dynamics ofslow-moving
landslides from permanent scatterer analysis. Science, 304, 1952-1955.
Hovius, N. Stark, C.P., and Allen, P.A (1997). Sediment flux from a mountain belt
derived by landslide mapping. Geology (Boulder), 25, 231-234.
Hovius, N., Stark, C.P., Chu, H., et al. (2000). Supply and removal of sediment in a
landslide-dominated mountain belt; Central Range, Taiwan. Geology, 108, 73-89.
Iverson, R M. and Major, 1.J. (1986). Rainfall, ground-water flow, and seasonal
movement at Minor Creek landslide, northwestern California: Physical interpretation
of empirical relations. Geological Society ofAmerica Bulletin, 99, 579-594.
59
Iverson, R.M. (2000). Landslide triggering by rain infiltration. Water Resources
Research,36,1897-1910.
Keefer, D.K., and Johnson, A.M. (1983). Earth flows-Morphology, mobilization and
movement. us. Geological Survey Professional Paper, 1264.
Kelsey, H. M. (1978). Earthflows in Franciscan melange, Van Duzen River basin,
California. Geology, 6, 361-364.
Kimura, H. and Yamaguchi, Y. (2000). Detection oflandslide areas using satellite radar
interferometry. Photogrammetric Engineering and Remote Sensing, 66,337- 344.
Mackey, B.H., Roering, ll, and McKean, lA. (2009). Long-term kinetics and sediment
flux of an active earthflow, Eel River, California. Geology, 37, 803-806.
Massonnet D, Rossi M, Carmona C, Adragna F, Peltzer G, et al. (1993). The
displacement field of the Landers earthquake mapped by radar interferometry.
Nature, 364, 138-142.
Massonnet D, Briole P, Arnaud A. (1995). Deflation of Mount Etna monitored by
spaceborne radar interferometry. Nature, 375,567-570.
Massonnet, D., Feigl, K.L. (1998). Radar interferometry and its application to changes in
the Earth's surface. Reviews ofGeophysics, 36,441-500.
Nolan, K.M. and Kelsey, H.M. (1995). Movement and sediment yield of two earthflows,
northwestern California. Us. Geological Survey Professional Paper, 1454.
Rosen, P.A., et al. (2004). Updated Repeat Orbit Interferometry Package Release, Eos,
Trans. AGU, 85,47.
Rott, H., Scheuchl, B., Siegel, A., Grasemann, B. (1999). Monitoring very slow slope
movements by means of SAR interferometry: a case study from a mass waste above
a reservoir in the Otztal Alps, Austria. Geophysical Research Letters, 26,
1629- 1632.
Squarzoni, C., Delacourt, c., and Allemand, P. (2003). Nine years of spatial and temporal
evolution ofthe La Valette landslide observed by SAR interferometry. Engineering
Geology, 68(1-2), 53-66.
60
Vietmeier, J., Wagner, W., and Dikau, R. (1999). Monitoring Moderate Slope
Movements (Landslides) In The Southern French Alps Using Differential SAR
Interferometry. Procedings of FRINGE'99, Web site: http://esrin.esa.it/-fringe99
Zebker, H.A., and Rosen, P. (1994). On the derivation of coseismic displacement fields
using differential radar interferometry; the Landers earthquake. International
Geoscience and Remote Sensing Symposium, 1,286-288.