Modelling and Inversion of EMI data collected over ... · Modelling and Inversion of EMI data collected over magnetic soils Len Pasion 1, Kevin Kingdon 1, Stephen Billings 1, Douglas

UBCGeophysical Inversion FacilityModelling and Inversion of EMI data

collected over magnetic soils

UXO/Countermine/Range ForumAugust 26, 2009Len Pasion1, Kevin Kingdon1, Stephen Billings1, Douglas W. Oldenburg1

1. Sky Research, 2. University of British Columbia

Waikaloa, Hawaii

Waimea, Hawaii

Examples of EMI data acquired at sites with magnetic soils

• Geophysical Proveouts• Geonics EM63 Data• First time channel

Former Camp Sibert, Alabama

60.0 mV

51.0

41.9

32.9

23.9

14.8

5.8

-3.2

-10.0

50.0 mV

42.3

34.5

26.8

19.0

11.3

3.6

-4.2

-10.0

Map Data

A Typical Processing Flow1 Filter to Remove

Geologic Features2

Target Picking via Threshold on Filtered data

3Data Inversion/Discrimination Estimate dipole model parameters: • Location• Orientation• Polarizabilities

4

UBCGeophysical InversionFacilityCell 185Base Plate

Example of inverting for dipole polarizabilities

m=F -1 [d ]

Cell 64Scrap

L1 > L2 = L3

L1 ≠ L2 ≠ L3

InvertData

m=F -1 [d ]

InvertData

UBCGeophysical InversionFacility� At sites with a strong geologic signal, small spatial scale signals can result:

1. Position of SensorAbove Ground

2. Topography

Un-modeled correlated signal will negatively affect estimated dipole parameters

UBCGeophysical InversionFacilityOutline� Introduction1. Incorporating ground clearance when

processing EMI data2. Investigating the effects of magnetic

topography� Conclusion

UBCGeophysical InversionFacility

[ ] [ ] [ ]bgt pp,,, bgt FFF += θφrm

• When the target response and background response are additive we can rewrite the forward operator asBackgroundTarget

• We will use a dipole model

• The EM parameters of the background can vary laterally

χ(ω) χ(ω)= +

Target in a Background Host

Modeling EMI sensor data for a target in a magnetic host

UBCGeophysical InversionFacility� We have developed a fast method of modeling the sensor response due to changes in sensor position above magnetic soil� Assume response has form: V(x,y,t)=A(height,pitch,roll,yaw)G(χ)f(t)

Man Portable Vector (MPV) TEM Sensor Data

Sky Research UXO test plot in Ashland, OR

R2

R3R4

VerticalComponents

RadialComponents

AzimuthalComponents

Rx2,Rx3,Rx4

Modeling EMI sensor data for a magnetic host

UBCGeophysical InversionFacilityHeight Test

Front Center - coaxial

Height Above Ground (m)

Orientation Test

Front Center - coaxial

Sensor Tilt (degrees)

tilt

Modeling EMI sensor data for a magnetic host

UBCGeophysical InversionFacilityTwo approaches developed:1. Subtract soil response prior to inversion. Solve for a

smooth background susceptibility model, then subtracting the predicted from the smooth susceptibility model.� Approach successfully demonstrated using both frequency domain (GEM-3 at Fort Lowry Bombing and Gunnery Range) and time domain (Geonics EM63 at Camp Sibert) data2. Simultaneously solve for host parameters and target parameters� Approach successfully demonstrated using Geonics EM63 TEM data acquired at Camp Sibert, Alabama

Incorporating ground clearance when processing EMI data

UBCGeophysical InversionFacilityExample: Inversion of Geonics EM63

TEM data at Camp Sibert, AL

Channel 1 Channel 20

• ESTCP Discrimination Pilot Project• Data collected in cued mode• 4.2 inch mortarsMethod 1: Subtract soil response prior to inversion

UBCGeophysical InversionFacilityMethod 1: Subtract soil response prior to inversion • Use elevation to estimate height above ground• Solve for the property distribution G(x,y) with regularized inversion

( ) ( ) 2 2

soil soilminimize ( )bgd mFϕ β= − + obsm W d m W m

d =F [m]m=F-1 [d ]

• Small spatial wavelength features in the data can be modeled with a background host that has slowly spatially varying EM properties

UBCGeophysical InversionFacility- =

Corrected data can then be inverted for dipole model parameters

Background

χ(ω) χ(ω)

Target in a Background Host Target in Freespace

- =

UBCGeophysical InversionFacilityMethod 2. Simultaneous inversion for background and target parameters

Observed data Predicted data ResidualChannel 1

Channel 20

( ) ( )( ) 2

target soil( ) ( )t bgd F Fϕ = − +obsm W d m m

• Simultaneous inversion for dipole parameters plus a background susceptibility that varies spatially as a plane: G(x,y) = A + Bx + Cy

UBCGeophysical InversionFacilityChannel 20

Time (ms)

EM

63 R

esp

on

se (

mV

)

S1

S2

S2

S1

Method 2. Simultaneous inversion for background and target parameters

UBCGeophysical InversionFacilityPolarizations obtained from TEMTADS in-air measurements

Estimated polarizations when inverting data for 3 unique polarizations

Estimated polarizations when inverting data for 2 unique polarizations

Method 2. Simultaneous inversion for background and target parameters

UBCGeophysical InversionFacilityExample:� 37 mm projectiles buried at 20 cm depth� 60 mm mortars buried at 40 cm depth� Data synthetically generated by using IMU and GPS records from Camp Sibert Geonics EM63 survey data� Position error: σ = 1 cm, IMU error: σ = 1 degrees � Magnetic soil signal is approximately 15 mV in the first time channel

Using simulations to determine the importance of including ground clearance when processing data

Does modeling the soil response due to sensor movement improve the ability to discriminate?

UBCGeophysical InversionFacilityTraditional MethodMedian filtered data

- Poor separation of features for the 60 and 37 mm

FAR = 0.68AUC = 0.84

60 mm

37 mm

Modeling the response due to sensor movement can improve our ability to discriminate between different target types

60 mm

37 mm

FAR = 0.00AUC = 1.00

Including soil response from sensor movement

- 60 mm and 37 feature clusters are well separated

UBCGeophysical InversionFacilityOutline� Introduction1. Incorporating ground clearance when

processing EMI data2. Investigating the effects of magnetic

topography� Conclusion

UBCGeophysical InversionFacility� EH3D Code developed at the University of British Columbia Geophysical Inversion Facility� Maxwell’s Equation solved numerical using a finite volume approach� Complex magnetic susceptibility used to represent soils with viscous remnant magnetization

Modeling the EMI response of topography using numerical modeling

UBCGeophysical InversionFacilityModelling the MPV response to a bump

Modeling the EMI response of topography using numerical modelling

UBCGeophysical InversionFacilityModelling the MPV response to a Trench

Modeling the EMI response of topography using numerical modelling

UBCGeophysical InversionFacilityInvestigating the effects of magnetic topography via simulationsExample:� 40 mm projectiles buried at 15 cm depth� Assume a bump running N-S� Position σ = 0.5 cm� Orientation σ= 0.5 degrees� 100 realizations

X comp Y comp Z comp

40 mm

Bump Response

X comp Y comp Z comp

UBCGeophysical InversionFacilityPolarizabilities Polarizability Sum

Time (ms) Time (ms)

L(t) Σ Li(t)

Investigating the effects of magnetic topography via simulationsCase 1: No magnetic soil topography� Accurate recovery of polarizabilities� Shape and Size information accurately recovered

UBCGeophysical InversionFacilityPolarizabilities Polarizability Sum

Time (ms) Time (ms)

L(t) Σ Li(t)

Case 2: Magnetic Bump� Magnetic properties typical of that found at the Ashland, Oregon Airport� Accuracy deteriorates at early and later times � Start to lose shape information� Size info accurate over most the time windowInvestigating the effects of magnetic topography via simulations

UBCGeophysical InversionFacilityPolarizabilities Polarizability Sum

Time (ms) Time (ms)

L(t) Σ Li(t)

Investigating the effects of magnetic topography via simulationsCase 3: A More Magnetic Bump� Magnetic properties 2 x that found at the Ashland Airport� Complete loss of shape information� Size information accurate over most the time window Z comp

UBCGeophysical InversionFacilityConclusions� Methodologies have been developed to estimate geologic and target parameters. � We have shown that a background with slowly spatially varying geologic properties can model the observed small spatial wavelength features in the data.� We have developed a method for modeling the EMI sensor response to topography� Recovery of the polarization tensor is less affected by topography than sensor position above the ground

This research is funded by the Strategic Environmental Research and Development Program (SERDP MM-1573)

UBCGeophysical InversionFacilityThis research is funded by the Strategic Environmental Research and Development Program (SERDP MM-1573)

Acknowledgements

UBCGeophysical InversionFacility

-100 -80 -60 -40 -20 0 20 40 60 80 1000

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5x 10

-3

Offset (cm)

Hz

6cm bump, 40cm loop10 cm bump, 40cm loop12 cm bump, 40cm loop14 cm bump, 40cm loop18cm bump, 40cm loop6cm bump, 100cm loop10 cm bump, 100cm loop12 cm bump, 100cm loop14 cm bump, 100cm loop18cm bump, 100cm loop

UBCGeophysical InversionFacilityModel based features:� Inversion of data for parameters of a physics-based model� These parameters reflect the size, shape, and material

properties of the target

Data based features:� Amplitude� Spatial extent

Sensor data: d

Model Parameters: m

d =F [m]

m=F-1 [d ]

Forward Operator

Inverse Operator

UBCGeophysical InversionFacilityMedian filtered data

Pre-filter data by subtracting modelled soil response

Simultaneous inversion for dipole parameters and soil spatial variability

Principlepolarizability Secondary polarizabilities

UBCGeophysical InversionFacilityExample: MTADS EM61 TEM Array - Camp Sibert, Alabama

MTADS first time channel - Detrended

EM61 mV Channel 1

Cell 644• An approximately 40 mV anomaly was

detected in the NS lines• The detrended elevation suggests that

there is a variation of approximately 13 cm in the ground clearance

Ground Clearance Estimated from Elevation Data

meters

Modeling the sensor response due to magnetic soils (Task 1.2)

UBCGeophysical InversionFacility

Line 76

Predicted Geologic ResponseMTADS first time channel - Detrended

EM61 mV Channel 1

Line 15 Line 518

Modeling the sensor response due to magnetic soils (Task 1.2)

Example: MTADS EM61 TEM Array - Camp Sibert, Alabama

UBCGeophysical InversionFacilityHx

0 1 2

0

0.5

1

1.5

2

-5

0

5

x 10-4

Hx

0 1 2

0

0.5

1

1.5

2-1

0

1

2x 10

-5

Hx

0 1 2

0

0.5

1

1.5

2-1

0

1

2

3

x 10-5

Hy

0 1 2

0

0.5

1

1.5

2

-5

0

5

x 10-4

Hy

0 1 2

0

0.5

1

1.5

2 -5

0

5

10

x 10-6

Hy

0 1 2

0

0.5

1

1.5

2-1

-0.5

0

0.5

1

x 10-5

Hz

0 1 2

0

0.5

1

1.5

2

-15

-10

-5

0

x 10-4

Hz

0 1 2

0

0.5

1

1.5

2 -3

-2

-1

0

x 10-5

Hz

0 1 2

0

0.5

1

1.5

2

-3

-2

-1

0

1

x 10-5

Example: Modelling the response of the TEMTADS to a lump of magnetic soil

Backgroundσ = 10-1 S/mχ(ω) 18cm18cm30cm

UBCGeophysical InversionFacilityTask 2.2: Include appropriate interaction components into modeling and inversion

• Different geologic scenarios have been modelled using a finite volume numerical modelling code for Maxwell’s Equations (EH3D)

• We have confirmed the ability of EH3D to correctly model the viscous remnant magnetization (VRM) response

χχχχ(ωωωω)

Magnetic susceptibility model • Based on lab measurements of

Kaho’olawe soil (MM1414)

Real

Imag

Frequency (Hz)

Real

Imag

Frequency (Hz)

Example: Modelling the VRM Response

UBCGeophysical InversionFacilityProcessing method FAR AUCMedian Filter data directly 0.68 0.84Pre-filter assuming plane 0.04 1.00Simultaneous inversion 0.00 1.00

Modeling the soil response due to sensor movement improves our ability to discriminate between different target types

FAR = 0.68AUC = 0.84

UBCGeophysical InversionFacilityBackgroundσ = 10-1 S/mχ(ω)

1m1mBump: 10x10x50cm(also ran 6,12,14,18cm)10cm10cm30cm

Backgroundσ = 10-1 S/mχ(ω) Trench: σ = 10-9 S/m

10cm10cm30cmHx

Hx Hy Hz

HzHy

Modelling different topographic features

UBCGeophysical InversionFacilityMedian filtered data

60 mm

37 mm

Modeling the response due to sensor movement can improve our ability to discriminate between different target types

Pre-filter data by subtracting modelled soil response

Simultaneous inversion for dipole parameters and soil spatial variability

FAR = 0.68AUC = 0.84

FAR = 0.04AUC = 1.00

FAR = 0.00AUC = 1.00

UBCGeophysical InversionFacilityTransmitter

( )( )

( )( )

0 030 020 0 1

L t

t L t

L t

=

ML1: Axial Polarization

L2 =L3 : Transverse Polarization

The Point Dipole Model

L1

L2 =L3

L1 ≠ L2 ≠ L3

Axi-symmetric (UXO-like) No Axial Symmetry (e.g. scrap)

L1 > L2 = L3

L3

L2L1

• The size and shape of a target is reflected in the elements of the diagonalized polarization tensor:

UBCGeophysical InversionFacility

Northing (m)

Example: Camp Sibert, Alabama• Data collected for the ESTCP Discrimination Pilot project • 4.2 inch mortars

UBCGeophysical InversionFacilityIncorporating Ground Clearance when Inverting EMI Data�Methodologies have been developed to estimate

geologic and target parameters. �We have developed a fast method of modeling the sensor response due to changes in sensor position above a magnetic earth�We have shown that a background with slowly spatially varying geologic properties can model the observed small spatial wavelength features in the data. �Techniques have been applied to synthetic and field data