1

ANALYSIS OF REPEATED ICESAT FULL WAVEFORM DATA: METHODOLOGY AND

LEAF-ON / LEAF-OFF COMPARISON

Hieu Duong1

Norbert Pfeifer2

Roderik Lindenbergh1

1: DEOS, MGP-FRS, 2: University of Innsbruck, Institute of Geography

2

Outline

• Introduction• ICESat/GLAS.• Waveform groundtracks in study area

• Waveform processings• Results and Discussion.• Conclusion

3

ICESat Facts

• Main goals: measuring ice sheet mass balance, cloud and aerosol heights, as well as land topography and vegetation characteristics.

• Altitude: 600km• NASA mission• Launched 2003• Life time 5 years• Carries GLAS instrument

4

GLAS instrument

• Footprint: 70m diameter• Spot space: 175m• 3 sensors: L1 (failed), L2 and L3• Geolocation: star tracker and

GPS

Full waveform laser altimetry!

5

ICESat repeated tracks

February:1840 waveforms

September:2942 waveforms

Forest pairs:358

6

ICESat repeated tracks

• Green: February 2003• Red: September 2003

40 60 80 1000

10

20

30

40

50Histogram of footprint shift

← mean =73.86[m]

• Select forest footprints manually by comparing to Landsat images.

7

Target: detect waveform changes

Fitting algorithm

Waveform Normalization

Conversion-Bin-ASCII-Counts-Voltage

-Smoothing-Initial parameters estimation

Shifting Computation

Waveform Distances-> DI, Intensity Dist-> RP, Peak Ratio

25% smallest RP & DI

-> C_DI threshold-> C_RP threshold

GLA01

GLA14

Extraction of ELevation

Points by IDL

8

Target: detect waveform changes

Class: sameDI & RP small

Class: not comparable

DI & RP not small

DI smallRP not small

DI not smallRP small

Class: slight change (pine)

Class: strong change (deciduous)

9

Waveform units

Y axis: Original waveform in counts (0-255) is converted to voltage units

X axis: 1ns ~ 15cm

0 100 200 300 400 50020

40

60

80

100Original waveform: 235714742:1

Relative Time (ns)0 100 200 300 400 500

-0.2

0

0.2

0.4

0.6Volt-waveform: 235714742:1

Volts

(V)

Relative Time (ns)

10

Normalization

Total intensity: ∑=

544

1iiV

=> Normalization step:

∑=

= 544

1ii

inori

V

VV

Area under waveform equals to 1.

After that we smooth the waveform.

Mean total intensity February: 116.48September: 39.90

11

Fitting: Gaussian decomposition

• Transmitted pulse is in Gaussian model• Return waveform is a sum of Gaussian components• Fitting algorithm: least square estimation

(Vienna Univ. of Tech.)

12

Results of Gaussian decompositionRed: raw waveformGreen: Gaussian components

First mode: left most Gaussian componentLast mode: right most

0 200 400 600-0.01

0

0.01

0.02

0.03

0.04Myfit-date:27-02-2003-49338080:19

Relative Time (ns)0 200 400 600

-0.01

0

0.01

0.02

0.03

0.04Myfit-date:30-09-2003-235714742:1

Relative Time (ns)

13

Shift

Blue: winter waveformRed: summer waveformGreen: shifted winter waveform

0 100 200 300 400 500 600-0.01

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07Waveform:27-02-2003<->30-09-2003

Nor

Volts

(V)

Relative Time (ns)

27-02-2003: in winter30-09-2003: in summer27-02-2003:shifted in winter

The average shift of 4.26[m] maybe caused by changes in the GLASconfiguration

-400 -200 0 200 4000

10

20

30

40

50X: -28.5Y: 50

14

Waveform distances

Intensity distance:

Along vertical axis

Along horizontal axis

Peak ratio:

2544

1

( ( ) ( ))( , )

544F S

F Si

V i V iDI W W

=

−= ∑

1,( , )

1,

F FF F S S

S SF S

S SF F S S

F F

LM FMLM FM LM FM

LM FMRP W W

LM FMLM FM LM FM

LM FM

−⎧− − > −⎪ −⎪= ⎨ −⎪ − − ≤ −⎪ −⎩

15

Waveform parameters

Peak locations• last mode ~ ground surface• first mode ~ canopy top

Widths of modes:~ feature roughness or slope surface, or both, canopy thickness

Amplitudes of modes• first mode ~ canopy density

Distance first-last mode~ tree height

0 200 400 600-0.005

0

0.005

0.01

0.015

0.02

0.025y

Relative Time (ns)

16

Mean parameter valuesFirst mode width• February: 4.28m• September: 3.74m

First mode amplitude• February: 0.00418• September: 0.0056

Last mode amplitude• February: 0.0083• September: 0.0088

First-Last peak distance• February: 20.41m• September: 19.18m

0 0.01 0.02 0.03 0.040

20

40

60

80

100Histogram of amplitude of first mode in February

← mean =0.0041892

0 0.01 0.02 0.03 0.040

20

40

60

80Histogram of amplitude of f irst mode in September

← mean =0.005615

17

Results: Waveform classes

0 200 400 6000

0.01

0.02

0.03

0.04same: 49338080-19<>235714742-1

wintersummer

0 200 400 600

0

5

10

15

20x 10-3

pine forest: 49337720-20<>235714382-2

wintersummer

0 200 400 600

0

0.01

0.02

0.03deciduous f orest49338120-36<>235714782-18

wintersummer

0 200 400 600

0

5

10

15

x 10-3 not comparable:49337780-20<>235714442-2

wintersummer

20 pairs

small-small

70 pairs

small-not small: deciduous?

69 pairs

not small-small

199 pairs

not small-not small

18

Conclusion

• Possible to compare repeated track ICESat waveform data, but, standardization steps are needed.

• Standardization steps, waveform parametrizations and waveform distances can be applied to non-ICESat waveform data as well.

• Waveform processing procedure should be extended, and tested more extensively on a more controlled data set.

19

Upcoming

• Article on land cover classification via analyzing ICESat’s full waveform data

(ISPRS Symposium, Enschede, May 2006)