Postseismic Deformation Postseismic Deformation from the 1991 Racha, from the 1991 Racha, Georgia EarthquakeGeorgia Earthquake

May 16, 2006May 16, 2006

Joel PodgorskiJoel Podgorski

Earth and Ocean SciencesEarth and Ocean Sciences

University of British ColumbiaUniversity of British Columbia

Purpose of studyPurpose of study

To determine if and how postseismic To determine if and how postseismic deformation has occurred following deformation has occurred following 1991 earthquake in Racha, Georgia1991 earthquake in Racha, Georgia

This information will provide This information will provide constraints on rheological constraints on rheological properties of lithosphere in Arabia-properties of lithosphere in Arabia-Eurasia continental collision zoneEurasia continental collision zone

OutlineOutline Racha earthquake and similar earthquakesRacha earthquake and similar earthquakes Tectonic settingTectonic setting GPS dataGPS data

• Racha rupture area and GPS sitesRacha rupture area and GPS sites

• data acquisition and analysisdata acquisition and analysis

• velocity decay in datavelocity decay in data

• correcting data for secular displacementcorrecting data for secular displacement Earthquake afterslipEarthquake afterslip

• afterslip processafterslip process

• inverse modelinginverse modeling

• modeling resultsmodeling results Viscoelastic relaxationViscoelastic relaxation

• viscoelasticity and modelingviscoelasticity and modeling

• modeling resultsmodeling results SummarySummary

Racha earthquake and similar Racha earthquake and similar earthquakesearthquakes

Racha earthquake, Georgia (April 29, 1991)Racha earthquake, Georgia (April 29, 1991)

• MMww=6.9 (same as Loma Prieta)=6.9 (same as Loma Prieta)

• Dip slip/thrust mechanismDip slip/thrust mechanism

• 3030˚ dip, hypocenter 6 km depth˚ dip, hypocenter 6 km depth

Similar thrust events:Similar thrust events:

• Chi-Chi, Taiwan (1999), MChi-Chi, Taiwan (1999), Mww=7.6, hypocenter =7.6, hypocenter

depth 8 km, dip 30depth 8 km, dip 30˚̊

• Northridge, California (1994), MNorthridge, California (1994), Mww=6.7, =6.7,

hypocenter depth 20 km, dip 45hypocenter depth 20 km, dip 45˚̊

• Loma Prieta, California (1989), MLoma Prieta, California (1989), Mww=6.9, =6.9,

hypocenter depth 16 km, dip 70hypocenter depth 16 km, dip 70˚̊

UP DIP

DOWN DIP

Setting

Greater Caucasus:Greater Caucasus:• 1000 km long, ~5000 1000 km long, ~5000

m highm high• middle of Alpine middle of Alpine

Himalayan fold beltHimalayan fold belt• uplift began 3.5 Ma uplift began 3.5 Ma

after collision of after collision of ArabiaArabia

Greater Caucasus

25 MM/YR

ARABIA

Racha rupture area and GPS sitesRacha rupture area and GPS sites

Racha Epicenter

Afterslip Plane

GPS Sites

GPS data acquisition and analysisGPS data acquisition and analysis

Data are from 8 sites for some or all of time Data are from 8 sites for some or all of time points: 1991.56 (~3 months after earthquake), points: 1991.56 (~3 months after earthquake), 1994.77, 1996.74, 1998.71, 2000.761994.77, 1996.74, 1998.71, 2000.76

Field work by collaborators at MITField work by collaborators at MIT Data analyzed at MIT with GAMIT/GLOBK software:Data analyzed at MIT with GAMIT/GLOBK software:

• Estimate site coordinates, satellite orbital Estimate site coordinates, satellite orbital parameters, atmospheric delay corrections, parameters, atmospheric delay corrections, and earth orientation parametersand earth orientation parameters

• Combine parameter estimates and covariances Combine parameter estimates and covariances and apply position and velocity constraints and apply position and velocity constraints from global core sites from global core sites

50% precision improvement 1991-1994 due to 50% precision improvement 1991-1994 due to increased satellite coverageincreased satellite coverage

Logarithmic versus linear fit to dataLogarithmic versus linear fit to data

Logarithmic fit indicative of afterslipLogarithmic fit indicative of afterslip Residuals to linear fit and log fit Residuals to linear fit and log fit

comparedcompared Sites showing better fit with Sites showing better fit with

logarithmic curve (misfit halved or logarithmic curve (misfit halved or better):better):

• LESO (N,E,Up)LESO (N,E,Up)

• KHUR (N)KHUR (N)

• SACC (E, Up)SACC (E, Up)

Data correction for secular Data correction for secular displacement: two optionsdisplacement: two options

Subtract 1996-2000 Subtract 1996-2000 velocitiesvelocities

• Assumption: Assumption: postseismic postseismic deformation deformation finished by 1996 – finished by 1996 – likely true if no likely true if no viscoelastic viscoelastic deformationdeformation

• Advantage: values Advantage: values based on actual based on actual measurementsmeasurements

Subtract predicted Subtract predicted MIT block model MIT block model velocitiesvelocities

• Based on 1988-2005 Based on 1988-2005 GPS measurements to GPS measurements to fit large-scale fit large-scale block model for block model for eastern eastern Mediterranean Mediterranean regionregion

• Advantage: data Advantage: data from entire from entire earthquake cycleearthquake cycle

• Disadvantage: rough Disadvantage: rough fit, not meant for fit, not meant for smaller scalesmaller scale

Earthquake afterslipEarthquake afterslip Motion on a fault after an earthquake due to Motion on a fault after an earthquake due to

stresses induced by the earthquakestresses induced by the earthquake Happens on coseismic fault; sometimes also on Happens on coseismic fault; sometimes also on

adjacent faultsadjacent faults Usually occurs where coseismic slip was at a Usually occurs where coseismic slip was at a

minimumminimum Occurs at shallow depths in zone of velocity Occurs at shallow depths in zone of velocity

strengtheningstrengthening Begins immediately after earthquake and can last Begins immediately after earthquake and can last

for several years – modeled as logarithmic decay for several years – modeled as logarithmic decay Examples of afterslipExamples of afterslip

• Chi-Chi: Chi-Chi: 16% of coseismic moment in 15 16% of coseismic moment in 15 months months

• Northridge: Northridge: 22% of coseismic moment in 2 22% of coseismic moment in 2 yearsyears

• Loma Prieta: Loma Prieta: 10% of coseismic moment in 5 10% of coseismic moment in 5 yearsyears

Afterslip inverse modelingAfterslip inverse modeling

Kinematic model inverting for fault Kinematic model inverting for fault displacement in an elastic half-space using displacement in an elastic half-space using displacements on Earth's surfacedisplacements on Earth's surface

Green‘s functions, Green‘s functions, GG, are calculated on each , are calculated on each fault tile to relate slip to each GPS fault tile to relate slip to each GPS displacementdisplacement

Smoothing parameter, Smoothing parameter, ßß, facilitates trade-off , facilitates trade-off between best fit to data and a smoothly between best fit to data and a smoothly varying solution by minimizing:varying solution by minimizing:

||||WW((GGss--dd)||)||22 + + ßß2 2 ||||LLss||||22

(misfit)(misfit) (roughness) (roughness)ssii is slip on each fault tileis slip on each fault tileddjj is displacement at each GPS site is displacement at each GPS site

L is the Laplacian operatorL is the Laplacian operatorWWTTWW = ∑ = ∑-1-1 (data covariance matrix) (data covariance matrix)

Fault plane resolution testsFault plane resolution tests Forward modeled checkerboard slip distribution to Forward modeled checkerboard slip distribution to

GPS sitesGPS sites Inverted for slip using forward modeled Inverted for slip using forward modeled

displacement vectorsdisplacement vectors

Modeled Afterslip Plane

SLIP INPUT

INVERSION OUTPUT

Racha Hypocenter

SURFACE

DEPTH

40 K

M

120 KM

Afterslip 1991-1994Afterslip 1991-1994 Subtracting 1996-2000Subtracting 1996-2000

• 90% error reduction90% error reduction

• 35% coseismic moment35% coseismic moment

max slip: 35 cmmax slip: 35 cm

Subtracting block Subtracting block modelmodel• 66% error reduction66% error reduction

• 28% coseismic moment28% coseismic moment

max slip: 45 cmmax slip: 45 cm

SURFACE

DOWN DIP

1991-1994 afterslip with coseimic slip 1991-1994 afterslip with coseimic slip distributionsdistributions

Using data subtracting 1996-2000 velocities

Using data subtracting block model

Coseismic model

DOWN DIP

SURFACE

Forward-modeled afterslip versus dataForward-modeled afterslip versus dataRed:Red: data with 1- data with 1-σσ errors (67% confidence interval) errors (67% confidence interval)Blue: modelBlue: model

Subtracting 1996-2000 Subtracting 1996-2000 velocitiesvelocities

Subtracting block modelSubtracting block model

Afterslip with seismicityAfterslip with seismicity• Afterslip plotted with first 2 months of Afterslip plotted with first 2 months of

aftershocksaftershocks Afterslip is from 3 months to 3 years after Afterslip is from 3 months to 3 years after

earthquakeearthquake

““shallow” aftershocks shallow” aftershocks beneath 1500m high Racha beneath 1500m high Racha RidgeRidge

QuickTimeª and aTIFF (LZW) decompressor

are needed to see this picture.

Afterslip 1994-1996Afterslip 1994-1996

Subtract 1996-2000Subtract 1996-2000

22% error reduction22% error reduction

max slip: 18 cmmax slip: 18 cm

Afterslip was likely not occurring after Afterslip was likely not occurring after 19941994

Subtract block Subtract block modelmodel

27% error reduction27% error reduction

ViscoelasticityViscoelasticity Viscoelasticity: elastic on short time Viscoelasticity: elastic on short time scale, viscous on long time scale (e.g. scale, viscous on long time scale (e.g. Earth's mantle)Earth's mantle)

Indication in lower lithosphere: seismic Indication in lower lithosphere: seismic attenuation and high heat flowattenuation and high heat flow

Maxwell viscoelastic material explained by:Maxwell viscoelastic material explained by:

ddε/dt = σ/2η + dσ/Edtε/dt = σ/2η + dσ/Edt

from which:from which:

σ = σσ = σ00exp(-Et/2η) exp(-Et/2η)

(stress decay at 0 strain rate)(stress decay at 0 strain rate)

where:where:

σ-stress, ε-strain, η-viscosity, E-Young's σ-stress, ε-strain, η-viscosity, E-Young's modulusmodulus

Viscoelastic modelingViscoelastic modeling Basis: P and S attenuation in lower crust Basis: P and S attenuation in lower crust of Caucasus of Caucasus

Used code to forward model response of Used code to forward model response of coseismic slip using different viscosities coseismic slip using different viscosities and layer thicknessesand layer thicknesses

Best model:Best model:

• 4e+17 Pa s for bottom 4e+17 Pa s for bottom

20 km of crust20 km of crust

• Only 21% error reduction Only 21% error reduction

Viscoelastic relaxation not responsible for Viscoelastic relaxation not responsible for early postseismic deformationearly postseismic deformation

SummarySummary Eight GPS sites produced sparse time series of nine-year Eight GPS sites produced sparse time series of nine-year

postseismic period of 1991 Racha earthquakepostseismic period of 1991 Racha earthquake Logarithmic decay in position measurements from three Logarithmic decay in position measurements from three

sites indicate postseismic deformationsites indicate postseismic deformation Two methods of correcting data for secular motions Two methods of correcting data for secular motions

produced similar afterslip inversions over 1991-1994:produced similar afterslip inversions over 1991-1994:

• shallow aseismic afterslipshallow aseismic afterslip

• 65-90% error reduction65-90% error reduction

• ~30% of coseismic moment~30% of coseismic moment No evidence for afterslip post-1994No evidence for afterslip post-1994 No evidence for viscoelastic relaxation in 1991-1994No evidence for viscoelastic relaxation in 1991-1994 Afterslip dominated Racha deformation as anticipated by Afterslip dominated Racha deformation as anticipated by

studies of similar earthquakesstudies of similar earthquakes A viscoelastic layer may be in the lower lithosphere, but A viscoelastic layer may be in the lower lithosphere, but

Racha event not strong enough to activate itRacha event not strong enough to activate it

Thank YouThank You

Velocity calculation from raw dataVelocity calculation from raw data

Displacements found by Displacements found by differencing measurements differencing measurements relative to station SACC relative to station SACC located near fault on footwalllocated near fault on footwall

Secular displacements corrected Secular displacements corrected by subtracting velocities from by subtracting velocities from the 1996-2000 time periodthe 1996-2000 time period

Errors found by summing squares Errors found by summing squares of errors used in differencing of errors used in differencing measurementsmeasurements

Fault afterslipFault afterslip Inverse modeling in an elastic half-Inverse modeling in an elastic half-space with Poisson‘s ratio of 0.25space with Poisson‘s ratio of 0.25

Green‘s functions, Green‘s functions, GG, are calculated , are calculated on each fault slip tileon each fault slip tile

Roughness, Roughness, LL, of inversion result is , of inversion result is minimized by applying smoothing minimized by applying smoothing factor, factor, ßß, and minimizing weighted , and minimizing weighted residual sum of squares (WRSS):residual sum of squares (WRSS):

||||WW((GsGs--dd)||)||22 + + ßß2 2 ||||LsLs||||22

where:where:

ssii is slip on each fault tileis slip on each fault tile

ddjj is displacement at each GPS is displacement at each GPS sitesite

WWTTWW = ∑ = ∑-1-1 (data covariance matrix) (data covariance matrix)

Determine smoothingDetermine smoothing

Choose Choose smoothing smoothing where where curvature of curvature of misfit vs. misfit vs. roughness roughness graph is graph is greatestgreatest

Afterslip fault planes & Afterslip fault planes & aftershocksaftershocks

Afterslip inversion resultAfterslip inversion result

Smoothing = 5.8Smoothing = 5.8

EQEQ

slipslip

Smoothing = 10.0Smoothing = 10.0

Comparison of inversion and dataComparison of inversion and data(BLUE: model, RED: data)(BLUE: model, RED: data)

All stationsAll stations Close up near Close up near faultfault

Viscoelastic relaxation codeViscoelastic relaxation code

Code forward models response of Code forward models response of earth to earthquake stressesearth to earthquake stresses

Uses spherical layers with Uses spherical layers with variable density, bulk modulus, variable density, bulk modulus, shear modulus, and viscosityshear modulus, and viscosity

Calculates spherical harmonic Calculates spherical harmonic expansion of spheroidal and expansion of spheroidal and toroidal motion components toroidal motion components

Viscoelastic relaxation vs. fault Viscoelastic relaxation vs. fault afterslip afterslip

Viscosity = 10Viscosity = 101717 Pa s (below Pa s (below 16km)16km)

Things to do…Things to do…

Confirm that investigating only Confirm that investigating only 1991-1994 is adequate1991-1994 is adequate

Do more modeling of Do more modeling of viscoelastic relaxation to gain viscoelastic relaxation to gain clearer picture of what clearer picture of what viscosity is needed to fit dataviscosity is needed to fit data

Perhaps try viscoelastic Perhaps try viscoelastic modeling of 1994-1996 modeling of 1994-1996 deformationdeformation