SAR tomography for the retrieval of forest biomass and height : Results from two tropical forests in French Guiana

Ho Tong Minh Dinh1, Thuy Le Toan1, Fabio Rocca2, Stefano Tebaldini2,

Ludovic Villard1, Maxime Réjou-Méchain3, and Jerôme Chave3

1 Centre d’Etudes Spatiale de la Biosphere (CESBIO), France 2 Politecnico di Milano, Italy

3 Evolution and Diversité Biologique (EDB), France

Email : htmdinhrs@gmail.com

Background By exploiting SAR Tomography (TomoSAR), we are able to transform the multi-baseline stack of SAR images into a multi-layer stack of SAR images, where each image represents the complex reflectivity associated with a layer at a certain height above the ground [1].

Track 1

Reference Track (Master)

Track n

azimuth

elevation

slant range θ

π/2

cross range

Duration: 3 months (option : 1 year) and Repeat pass time : ≤ 4 day The main goal is to improve forest biomass and height retrieval methods by addressing 2 questions:

- main scattering mechanisms (SMs) at forest and ground level - SMs changes across the global forest biomes

BIOMASS Tomographic Phase [2]

Hei

ght

[m]

Capon spectrum - HH channel

200 400 600 800 1000 1200 1400 1600 1800 2000

0

20

40

60

LiDAR height

Hei

ght

[m]

Capon spectrum - HV channel

200 400 600 800 1000 1200 1400 1600 1800 2000

0

20

40

60

LiDAR height

Hei

ght

[m]

Capon spectrum - VV channel

Slant range [m]

200 400 600 800 1000 1200 1400 1600 1800 2000

0

20

40

60

LiDAR height

Background

The retrieval of forest biomass has been demonstrated at the Paracou forest in French Guiana [1], [3].

heig

ht [m

]

2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000

0

20

40

60

LiDAR height

range [m]

0 100 200 300 400 500 600 0

100

200

300

400

500

600

Above-ground biomass (t/ha)

Retr

ieve

d b

iom

ass (

t/h

a)

RMSE = 34.5 (t/ha)

= 9.41 (%)

r P = 0.84

1. The backscatter from the layer at 30 m above the ground level is significantly correlated to tropical forest Above Ground Biomass (AGB).

2. Tropical forest height can be estimated as 4m accurate or better.

Background

Goal : To assess the robustness of the method, the analysis is performed at Nouragues forest with strong topography and high AGB (up to 600 t/ha).

Paracou

Nouragues

French Guiana

Problem : Phase calibration, to be carried out in order to recover the interferometric ground phase, is difficult for the harsh hilly terrain.

Tomography processing

Phase calibration solution:

Retrieval of an initial guess for ground phases by Capon spectra for HH channel; Correct the original data using this ground phase; Re-retrieval of ground phases by Sum of Kronecker Products (SKP) model.

Initialization 0

sin

4gn

initial

n zBr

);(maxarg0 HHzSz Caponz

g where:

Tomography processing

vvgg

K

k

kkk RCRCRCW 1

Assuming 2 main Scattering Mechanisms - ground and volume scattering : The best LS approximation of W is obtained by retaining the first 2 terms of the SKPD The matrices Rg, Rv, Cg, Cv can be determined from the terms of the SKPD via a linear transformation

SKP Decomposition [4]

Multi-polarimetric multi-baseline covariance matrix can be expressed :

Tomography processing

Retrieval of the ground phases

By this way, we get two advantages : The removal of the propagation disturbances, which allows a correct focusing along the vertical direction by exploiting the Fourier Transform. The removal of terrain topography, resulting in the contributions from the terrain to be automatically focused at 0m, independent of the actual topography.

ground = [1ground , 2

ground , … Nground ], can be obtained by :

ngn

ground

n zBr

sin

4

initial

g

ground RPL )(

Phase calibration is carried out by removing the ground phase value of the original SAR images.

TomoSAR to understand Scattering Mechanisms

After phase calibration based on the ground profile, most of phase center are located on the ground.

Even if the topography is very harsh, the SKPD makes it possible to retrieve the ground phases for phase calibration, which allows a correct focusing along the vertical direction.

Ground spectrum

Heig

ht

[m]

Azimuth [m]

1000 1500 2000 2500 3000 3500 4000 4500 5000-20

0

20

40

60

Volume spectrumH

eig

ht

[m]

Azimuth [m]

1000 1500 2000 2500 3000 3500 4000 4500 5000-20

0

20

40

60

0 20 40 60

LiDAR top height

0 0.5 1

0 0.5 1

0 20 40 60

LiDAR top height

HH spectrum

Heig

ht

[m]

Azimuth [m]

1000 1500 2000 2500 3000 3500 4000 4500 5000-20

0

20

40

60

HV spectrumH

eig

ht

[m]

Azimuth [m]

1000 1500 2000 2500 3000 3500 4000 4500 5000-20

0

20

40

60

Contributions from the ground level beneath the forest are observed. However, significant scattering contributions are observed at the canopy level in HH polarisation, whereas this volume scattering contribution is dominating in HV polarisation.

The scattering mechanisms in tropical forest shown by these results are quite different from those in boreal forests where for all polarisations the dominating contribution was observed to be associated with the ground level.

TomoSAR to understand Scattering Mechanisms

TomoSAR to retrieve forest biomass and height

Forest biomass retrieval

0 2 4 6 -25

-20

-15

-10

-5

0 rP = 0.8, Slope = 2.5

AGB (100 t/ha)

P H

V

[dB

]

Best fit line

in-situ Paracou

in-situ Nouragues

LiDAR Nouragues

•Training on

stratified subset of

Nouragues data.

•Performance

assessed on data

from Paracou.

•Training on

stratified subset

of Paracou data.

•Performance

assessed on data

from Nouragues. 0 100 200 300 400 500 600

0

100

200

300

400

500

600

Reference biomass (t/ha)

Retr

ieved b

iom

ass (

t/ha)

RMSE = 50.86 (t/ha)

= 13.88 (%)

rP = 0.88

The backscatter from the layer at 30 m above the ground level is significantly correlated to tropical AGB in both tropical forests in French Guiana.

0 100 200 300 400 500 6000

100

200

300

400

500

600

Reference biomass (t/ha)

Retr

ieved b

iom

ass (

t/ha)

RMSE = 57.17 (t/ha)

= 15.6 (%)

rP = 0.77

Forest height retrieval The retrieval of forest height has been assessed through a direct investigation of the shape of the 3D backscattered power distributions at each location.

LiDAR

500 1000 1500 2000 2500 3000 3500 4000 4500

500

1000

1500

10

20

30

40

50

Tomography

500 1000 1500 2000 2500 3000 3500 4000 4500

500

1000

1500

10

20

30

40

50

Relative Error

500 1000 1500 2000 2500 3000 3500 4000 4500

500

1000

1500

0

0.5

1

LiDAR (m)

To

mo

gra

phy (

m)

Normalized joint distribution

10 20 30 40 50 10

20

30

40

50

The estimation appears to be reliable for vegetation layers ranging from 20m to 40m. For this range height, standard deviation is about 4m.

The relative error has been evaluated as : |Htomogaphy – HLiDAR|/HLiDAR

TomoSAR to retrieve forest biomass and height

Conclusions

TomoSAR can be used to improve our knowledge on scattering

mechanisms of different forests.

TomoSAR can be used to retrieve forest biomass and forest height: the robustness and transferability of the tomographic method have been demonstrated for 2 tropical forests.

Hence, TomoSAR results could be used as references to fine tune PolSAR and PolInSAR methods (model approximation, model parameterisation…)

The feasibility under the 6 MHz bandwidth constraint by frequency allocation at P band has been proved in Paracou [3]. Future work will extend the BIOMASS simulation to the Nouragues site.

REFERENCES

[1] Ho Tong Minh Dinh, T. Le Toan, F. Rocca, S. Tebaldini, M. Mariotti d’Alessandro and L. Villard, “Relating P-band SAR tomography to tropical forest biomass”, TGRS, 2013. [2] T. Le Toan, S. Quegan, M. Davidson, H. Balzter, P. Paillou, K. Papathanassiou, S. Plummer, F. Rocca, S. Saatchi, H. Shugart, and L. Ulander, “The BIOMASS Mission : Mapping global forest biomass to better understand the terrestrial carbon cycle”, Remote Sensing of Environment, pp. 2850–2860, Jun. 2011. [3] Ho Tong Minh Dinh, S. Tebaldini, F. Rocca, T. Le Toan, L. Villard, and P. Dubois-Fernandez, “Capabilities of BIOMASS tomography for investigating tropical forests,” TGRS, under revision. [4] Tebaldini, S. "Algebraic Synthesis of Forest Scenarios From Multibaseline PolInSAR Data", TGRS, 2009;

ACKNOWLEDGMENTS

The authors would like to thank Dr. Pascale Dubois-Fernandez, along with the whole TropiSAR 2009 team. We also acknowledge the CIRAD for in situ data availability in Paracou.