forest height and ground topography at l-band from an experimental

Forest Height and Ground Topography at L-Band from an Experimental Single-Pass Airborne Pol-InSAR System

Bryan Mercer1, Qiaoping Zhang1, Marcus Schwaebisch2, Michael Denbina1, Shane Cloude3

(1) Intermap Technologies Corp., Calgary, Canada(2) Intermap Technologies GmbH, Munich, Germany

(3) AEL Consultants, Edinburgh, Scotland

PolInSAR 2009: Frascati, Italy Jan 26-31, 2009

© 2008 Intermap Technologies. All rights reserved.2

Contents

Objectives

The System

PolInSAR Approach

The Test Site

ResultsCanopy HeightGround Elevation

Conclusions


The L-Band Program Objective

To determine whether we can …Extract bare earth elevation (DTMs) beneath forest canopy and also extract the forest height,With adequate accuracy and consistency to justify an operationalairborne system?

Some Considerations:L-Band vs P-Band

» Forest Penetration» Bandwidth restrictions/Resolution (6 MHz vs 85 MHz)

Single-Pass vs Repeat-Pass» Eliminate temporal decorrelation» Optimum baseline

PolInSAR Algorithms


What is the Expected L-Band Attenuation?

2-Way Attenuation (db) vs Incidence Angle (80% ) for 20 meter Tree Height

0

5

10

15

20

25

30 40 50 60

Incidence Angle (deg

2-W

ay A

tten

uatio

n (d

b)

L-Band (80%)P-Band (80%)

Forest attenuation coefficients: Bessette and Ayasli (2001)

L-BandHH and VVMeasurements(Burnstick)


Under what conditions do we see the ground?

Forest attenuation coefficients: Bessette and Ayasli (2001)

Penetration depth definition: Kugler, et. al. (2006)

Forest Penetration Depth (HH) vs Incidence Angle Assuming NESZ = -30db ; (80% of

forest samples)

020

4060

80100

30 40 50 60

Incidence Angle

Pene

tratio

n D

epth

(m

) P-Band (80%)L-Band (80%)

Desirable

NESZ: - 40 db


Single-Pass L-Band PolInSAR System: design philosophy

Objectives:Demonstrate DEM extraction performance in several sets of forest / topography conditionsLearn enough to make sound recommendation for a potential operational follow-on system

Constraints:InexpensiveRapid turn-around: one-year design / build / test window

Approach:Modify existing TopoSAR (X/P system)

» Flexible digital architectureMaintain S/N performance (NESZ = -40 db)

» Low power, low gain antennas» Fly low (1km), sacrifice swath (1km)


System Design

Gulfstream Commander platform

Radar hardware based on TopoSARsystem

originally X- and P-band

Antennas: log periodicH and V planesMounted on rigid beam passing through the un-pressurized part of the fuselage

Single Tx/Rx chainPulse sequential switching between polarizations and antennasMonostatic and bistatic modes possible

» Full and half baseline» Only full baseline discussed here


System Design Parameters

Wavelength 0.226 mMax Bandwidth 135 MHzPeak Power 0.4 kwPRF/channel 2120 hz

No. of Channels 8,12,16Polarization QuadBaseline 3.5 m

Test Altitude 1000 mNESZ (max) -40 db

Ground Swath 1100 mAz. Res. 1.0 m

Slant Rg. Res 1.1 mIncidence Angles 30 - 60 deg

Kz @ 45° 0.14 rad/m


The PolInSAR Approach

assume RVOG* model applicable,

extract ground phase, Φ , from the complex coherence

* Treuhaft and Siqueira (2000), Papathanassiou and Cloude (2001)

Real

Imaginary

Φ

Ground Phase

0~

vγ

⎥⎥⎦

⎤

⎢⎢⎣

⎡

+

+Φ=)(1

)(~exp)(~ 0

w

ww

m

mvj γ

γ

hv

Canopy Polarization Independent

Ground Polarization Dependant

F(z) =structure function


Extraction of Ground Phase and Canopy Height

+* X

(1) Cloude, S. and K. Papathanassiou, “Three-Stage Inversion Process for PolarimetricSAR Interferometry”, IEEE Proc.-Radar Sonar Navig. (2003)

(2) Tabb, M., et. al., “Phase Diversity: A Decomposition for Vegetation Parameter Estimation using Polarimetric SAR Interferometry”, EUSAR2002

(3) Cloude, S., “Polarization Coherence Tomography”, Radio Science, Vol 41 (2006)

Coherence Region

Φ Real

Imaginary

To obtain ground phase, use:• Coherence Mag. Optimization(1)

• for low vegetation• Phase Optimization(2)

• for high vegetation

To obtain hv:• use the inversion expression(3)

z

w

z

wv k

c

kh vv

)~(sin2ˆ)~arg( 1 γε

φγ −

+−

=


kz across swath

Near Range Far Range

kz changes by x 6 across full swath

This presentation – will examine

performance in far part of the swath

kz = 0.10 to 0.05 rad/m

Ha ~ 60 to 120 m

~650 m

Optimum criterion for parameter inversion (1)

(kzhv / 2) = 1.3

kz = (4π/λ) (Δθ/sinθ) = (4πBn/λHtanθ)

(1) Cloude, S., K. P. Papathanassiou, ‘Forest Vertical Structure Estimation using Coherence Tomography’, IGARSS 2008, Boston


kz vs ‘Design’ Canopy Height

Near Range Far Range

~650 m

‘Design’ hv = 2.6/kz

26 m

13 m

46 m


Calibration

Geometric and polarimetric calibration components

Polarimetric calibration based on Queganmethod(1)

Range-dependant cross-talk and imbalance correction terms calculated using:

» Trihedrals (10) placed across swath» Forest (flat, relatively homogeneous,

extensive)

Cross-talk (observed) less than -25 dbHH/VV imbalance corrected across swath to

» Amplitude 1.0 +/- 0.05» Phase 0.0 +/- 5 degrees

~ 1km

VV Image Front Antenna: Trihedrals

30 35 40 45 50 55 60 650

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

CR Incidence Angle (Degree)

Mag

nitu

de

Estimated Imbalance (K2) from CRs: Magnitude

Mag: 0.951±0.1111

30 35 40 45 50 55 60 65-50

-40

-30

-20

-10

0

10

20

30

40

50

CR Incidence Angle (Degree)

Pha

se (D

egre

es)

Estimated Imbalance (K2) from CRs: Phase

Phase: 1.36±3.225

(1) (1) S. Quegan, “A unified algorithm for phase and crosstalk calibration of polarimetric data - theory and (2) Observations”, IEEE Trans. on Geoscience and Remote Sensing, vol. 32, no. 1, pp. 89–99, 1994.


Test Program; Three sites in Western Canada

Two areas in Alberta, CanadaEdson, BurnstickLodgepole pine species5 – 30 m heights

» Various ages» Clearcut areas (bare or with recent regrowth

Winter , late spring acquisition

One area in British Columbia, CanadaFraser DeltaSeveral deciduous species15 – 40 mLate spring acquisition

Varying amounts of ground truth available


The Edson test area

Test area near Edson: a forested region of Alberta, Canada

Patchwork of lodgepole pine forest and clearcut areasClearcuts may have been re-planted and in a regrowth phaseTypically 15-30 m highL-Band data acquired in February and again in June 2008

Ancillary dataX-Band DSM (from 2006) Lidar ground elevations and point cloud (courtesy Terrapoint 2007)Color air photo (Valtus 2007)

~20 m


Canopy Height Definition from Lidar Feature File

Lidar ‘feature file’ or point cloud approx 1 sample/m²

The lidar canopy height ‘h100’ is defined by the highest lidar sample value within a 5.6 meter search radius of each horizontal grid point (~100 m²)

A surface is then fitted to these points to represent the canopy smoothly

Lidar data provided by Terrapoint

5.89465.89475.89485.89495.8955.89515.89525.89535.8954

x 106

0

5

10

15

20

25

30

UTM Y

Hei

ght A

bove

Lid

ar B

are

Ear

th (m

) --- Lidar Feature Grid (5m Search Radius)+ Lidar Point Cloud


Edson Test Site: Ancillary Data

Air Photo Lidar DSM Lidar DTM Lidar (DSM-DTM)

Forest: 15 – 30 m

Bare or regrowth (<5m)

Forest: 15 – 30 m


Forest: 15 – 30 m


~ 1.3km


Radar Amplitude Images: X (2006), L (2008)

XHH(2006) LHH (near ant) LVV (near ant) LHV (near ant)


Multipath Issues and Mitigation

During first set of tests (winter), severe multipath in one of the antennas limited the usable swath to the far half of the swath

Subsequent absorber placement reduced the problem

Full swath available in summmer testsStill some multipath effects seen in the magnitude imagesInterferometric phase effects handled by calibration (CORVEC)Some coherence magnitude impact

Near Far Coherence

VV VVVV


Results

Canopy height hv

With respect to lidar canopy surface height h100

Ground elevationWith respect to lidar DTM


Profiles showing: Lidar DTM, Lidar Canopy Height, L-Band hv

1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross Section

Met

ers

Meters

1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross Section

Met

ers

Meters

--- Lidar Bald--- Lidar Feature--- L-Band Tree Height

LeftProfile

RightProfile

M9054 (summer data)

30 m

30 m


L-Band hv vs Lidar Canopy Heights (Kz ~ 0.1)

15 20 25 30 3515

20

25

30

35

Lidar Tree Height (m)

L-B

and

Tree

Hei

ght (

m)

Statistics:

Mean: -0.73m

Std. Dev.: 1.34m

RMSE: 1.53m

15 20 25 30 3515

20

25

30

35

Lidar Tree Height (m)

L-B

and

Tree

Hei

ght (

m)

Statistics:

Mean: -0.73m

Std. Dev.: 1.34m

RMSE: 1.53m

15 20 25 30 3515

20

25

30

35

Lidar Tree Height (m)L-

Ban

d Tr

ee H

eigh

t (m

)

Statistics:

Mean: 0.51m

Std. Dev.: 2.12m

RMSE: 2.18m

15 20 25 30 3515

20

25

30

35

Lidar Tree Height (m)L-

Ban

d Tr

ee H

eigh

t (m

)

Statistics:

Mean: 0.51m

Std. Dev.: 2.12m

RMSE: 2.18m

15 30 m20 25

20

25

30

15

20

25

30

15

Mean(Δ)=0.5mStdvn(Δ)=2.1m

Mean(Δ)=-0.7mStdvn(Δ)= 1.3m

Δ=hv–h(lidar)


1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross Section

Met

ers

Meters

1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross SectionM

eter

s

Meters

--- Lidar Bald--- Lidar Feature--- L-Band Ground--- L-Band Tree Height

1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross SectionM

eter

s

Meters


Lidar DTM, L-Band Ground (summer), Lidar Canopy Height,, L-Band hv (summer)

LeftProfile

RightProfile

M9054 (summer data)

30 m

30 m


1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross Section

Met

ers

Meters

1070

1080

1090

1100

1110

1120

1130

1140

1150

1160

0 500 1000 1500 2000 2500

Cross SectionM

eter

s

Meters

Lidar DTM, L-Band Ground (winter), Lidar Canopy Height,, L-Band hv (winter)

LeftProfile

RightProfile

M9041 (winter data)

30 m

30 m



Error Statistics: Edson Forest Samples (H>20m)

DTM Errors CanopyHeight


Conclusions

Results presented from tests of the first single-pass L–Band PolInSAR system

Edson test area – winter and early summer15-30 m tree heights

Derived forest heights in center/far swath mostly consistent with lidar-derived canopy as shown in profiles (<1 meter bias)

Sampled areas show hv(L-Band) ~ h(lidar) with standard deviation ~ 2 metersUncertain yet how robust this is over a larger range of tree heights, kz , other forest types, etc

Ground elevations noisier and exhibit local biases compared to lidar DTMMean offsets in local areas typ. 2-3 meters (winter data set), larger (summer data set) with standard deviation ~ 2.5 – 3.5 meters

There is a summer/winter difference in the ground elevation accuraciesNot yet clear whether this is due to forest target changes or to system differences

forest height and ground topography at l-band from an experimental

Documents