Tree Height Estimation Methodology With Xband and P-band InSAR Data. Lijun Lu Guoman Huang Qiwei Li CASM

Transcription

1 Tree Height Estimation Methodology With Xband and P-band InSAR Data Lijun Lu Guoman Huang Qiwei Li CASM

2 Outline CASMSAR dataset and test area Height Estimation with RVoG method Height Estimation with dual-band InSAR method Results validation and analysis Conclusions

3 CASMSAR Cessna Citation aircraft CASMSAR System X-band dual-antenna SAR P-band repeat pass SAR CASM-SAR System Parameters Airborne Platform Airborne PolInSAR System Band X-band P-band Interferometry Dual-antenna single pass single-antenna repeat-pass Baseline Length 2.2m -- Mode Biastics Mode Ping-Pong Mode Monostatic Polarization HH HV VH VV HH HV VH VV Resolution(m) Incident Angle( ) Bandwidth(MHz) Flight Altitude m Velocity 450km/h Endurance 3.5hours

4 Test area Sichuan Danling Forest Eucalyptus Broadleaf

5 CASMSAR dataset P-band Pauli synthesized image X-band Pauli synthesized image

6 Danling forest area Samples distribution (Google Earth)

7 Tree Height Measurement Measured Data in Danling Plot Species Tree Number Mean(m) Std.(m) 1 Eucalyptus ±1.7 2 Eucalyptus ±2.1 3 Eucalyptus ±4.4 4 Eucalyptus ±1.7 5 Eucalyptus ±1.3 6 Eucalyptus ±1.5 7 Eucalyptus ±5.4 8 Eucalyptus ±0.7 9 Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Arborvitae ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ± Eucalyptus ±4.4

8 Height Estimation with RVoG method P band data preprocessing steps Track Compensation Registration Interferometry

9 Height Estimation with RVoG method γ ( w) β z mn = mn iβ ( ) z z γ 0 C + µ w = e 1 + µ ( w) 4π B λr sinθ Tree Height h Volume Coherence γ C Ground-to-volume Ratio Topographic Phase z 0 Wavenumber β z v Wave Extinction γ = h v 2σ z ( ww 1 2) iβz z cosθ e e dz 0 C hv 2σ z cos σ 0 e θ dz µω ( )

10 Height Estimation with RVoG method 1 Least squares line fit Complex coherence can be rewritten using multiple polarization channels as the equation of a straight line in the complex plane. The points of multiple polarization channels can be fit a line. γ ( w, w ) 1 2 = e γ + µ ww ( ww 1 2) iβz z0 C 1 2 ( ) 1 + µ ( ww ) 1 2

11 Height Estimation with RVoG method 2 Vegetation bias removal Decide the point on unit cycle as underlying ground topographic phase and remove the phase related to the underlying ground topography.

12 Height Estimation with RVoG method 3 Height and extinction estimation γ C = p1 e p e 2 ( ph 2 v 1) ( ph 1 v 1) p 1 2 cos σ θ ww 1 2 = p 2 2σ = + i cos θ β ( ) z Parameter Range Step Tree Height 0 h v 30m 1m Wave Extinction 0 σ 0.4db/m 0.01db/m

13 Height Estimation with RVoG method P-band Interferometric phase generated from 9 scattering mechanisms Polarization channel HH HV VV Singular Value Decomposition SVD1 SVD2 SVD3 Numerical Radius NR1 NR2 NR3

14 Height Estimation with RVoG method Height estimation result with RVoG method

15 Height Estimation with dual-band InSAR method Dual-band interferometry calculate the X-band and P-band DEM height difference X band radar reflects on the top of the vegetation cover produces a Digital Surface Model (DSM) P band signal penetrates towards the forest canopy generates a ground level Digital Elevation Model(DEM)

16 Height Estimation with dual-band InSAR method Flowchart X-band SLC Registration InSAR Processing Elevation Correction Difference P-band SLC Tree Height

17 Height Estimation with dual-band InSAR method X-band Interferometric phase generated from 9 scattering mechanisms Polarization channel HH HV VV Singular Value Decomposition SVD1 SVD2 SVD3 Numerical Radius NR1 NR2 NR3

18 Height Estimation with dual-band InSAR method X-band and P-band SAR imagery coregistration Tie-points on X-band SAR image Tie-points on P-band SAR image

19 Height Estimation with dual-band InSAR method X-band and P-band SAR imagery coregistration X-band coregistered SAR image RMSE(root meam square error): 1.9 pixel P-band coregistered SAR image

20 Height Estimation with dual-band InSAR method Height difference correction Canopy height derived from X-band SAR Ground height derived from P-band SAR

21 Height Estimation with dual-band InSAR method Height difference correction Ground points with high coherence are selected to correct height difference of two elevation results by using a polynomial model Num X-band height P-band height Difference (m) (m) (m) A B A B

22 Height Estimation with dual-band InSAR method The elevation results Canopy height derived from X-band SAR Ground height derived from P-band SAR

23 Height Estimation with dual-band InSAR method Height estimation result with dual-band InSAR method

24 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Error mean:2.21m RMSE=2.85m Results validation and analysis RVoG height estimation validation Plot Mean (m) Std (m) RVoG(m) Error (m)

25 Results validation and analysis 14 samples are in the range of measured tree height, more than half of samples are not in the range of measured tree height that are overestimated.

26 Results validation and analysis Dual-band InSAR validation Plot Mean(m) Std(m) P-band RVoG (m) X-P band InSAR (m) Error (m) ± exception null ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Error mean:1.92m RMSE=2.47m

27 Results validation and analysis 15 samples are in the range of measured tree height, 5 samples are not in the range of measured tree height which of one part are overestimated and other part are underestimated

28 Results validation and analysis Comparison for two methods P-band RVoG X-P band InSAR Robustness Precision Low (dependent on model, algorithm) Mean:2.21m RMSE:2.85m High Mean:1.92m RMSE:2.47m Reliability 50% 75% Efficiency Low High

29 Conclusion Forest tree height can be estmated by RVoG model Dual-band InSAR Dual-band InSAR method has more advantages than RVoG method Robustness Precision Reliability Efficiency

30