Waveform simulations and analysis of Lidar data for … - International Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016 1 Waveform simulations

JAXA - International Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016 1

Waveform simulations

and analysis of Lidar data

for the development

of a space-based

vegetation lidar system

International Workshop on Vegetation Lidar

and Application from Space

January 6-7, 2016, Kyoto, Japan

S. Durrieu1, E. Grau1, 2, J-P Gastellu-Etchegorry3, G. Vincent4,

C. Antin5, J. Costeraste6, J. Lochard7, F. Tinto6, H. Debise1,

M. Bouvier1, C. Lavalley1,8, F. Fabre7

1 Irstea-UMR TETIS, 2 CNES , 3 CESBIO, 4 IRD-UMR AMAP, 5Cirad-UMR AMAP

7 AIRBUS- Defence & Space, 8 CNRS-UMR AMAP

General context

Vegetation structure is the driver and the result of forest ecosystem

ecological functions

Structure measurements provide information on processes and on

biophysical parameters and are needed

● to advance our understanding on forest ecosystems functioning

● and to sustainably manage forests

Spaceborne Lidar missions are currently the best solution to expand

forest structure measurements in space and time

Several missions are planned or under study by space agencies:

GEDI/NASA , MOLI/JAXA, LEAF/CNES


Outline

LEAF, the vegetation lidar mission under study in France

A modeling framework developed in the frame of

experimentation/validation study

an ongoing

Results on vegetation representations for RT modeling

Instrumental studies

Research avenues to improve forest parameters assessment from Lidar

Perspectives


JAXA - International Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016

LEAF: Lidar for EArth and Forests

A vegetation mission under study

in France

2014 : Recommended as a possible mid- term mission following the scientific prospective seminar of CNES

4

LEAF a satellite mission coupling a Lidar and an imager

Primary objective is to provide spatialized information on forest 3D structure and on its dynamics

Secondary objectives :

● Ground altimetry data for Geodesy, Geohazards and Cryosphere studies

● Continental water level data for Hydrologists

Design by Airbus Defense and Space (former EADS Astrium) with the support of CNES

LEAF mission

A standard satellite platform

Square modular structure ± 3° depointing capability

870 Kg launch mass

nadir

1500 mm

Measured track

V

A small launcher

(Vega- ESA)

5 years life

Launch in 2

Gro

C

Ground Station

Measu track

5 years life

Launch in 2

Gro

C

LEAF Satellite

Ground Station


LEAF instruments

Enhance geolocation

Improve accuracy of forest parameters

Provide data for extrapolation of lidar results

After 3 years

Spatial and temporal sampling characteristics 500 m

● 15 Hz repetitive rate

●

●

3 year revisit period

5/6 years lifetime

LEAF: current design

Fullwaveform lidar system

• NIR wavelength : 1064 nm (eye safety)

• Pulse duration < 2ns

• Footprint diameter: 20 m with an array detector (4) slope

• Energy : 86mJ laser output => 60mJ instrument output

• Signal sampling = 1 ns Vertical resolution = 15 cm z res.

on the shelves

VHR XS Imager

• Panchromatic + B,G,R,NIR

• 2m resolution, 14 km swath width

• Co-registered with the Lidar

2.5 km


LEAF concept is not set in stone

Models used to size the system are considered to be insufficiently

reliable still pending issues to optimize the mission and enhance its

capacity to monitor forest

This led CNES to implement an experimentation/validation stage

Objective for LEAF: size a system providing complete vegetation

profiles even for dense forests without any risk of calling into question

the concept or the system architecture in the latter mission project

phases

Establishment of a working plan to address pending issues


u i r e m e n t s q e u s e r r g m e e t i n

c i t y o s i z e a s p a c e s y m e s t t

u l a t e d & m e a s i g n a l

A i r b o r n e m e a n t s

+ A n c i a a a

( f u l l y c h a r a

nadir

(full

d e l i n g

i

i a l

p r o c e s s g

C o c y

e t

n d c h a r a c t e r i s t i c s

S t a n d c h a r a c t e r i s t i c s

d e r

i i

v e o m s g a l

A l l i o m e t r c r s

C a a p

e u l

n d f r

a

n

f o s t p m e e r s

S t

t a a r e r

w e e b

n s t e n s i

n

n S g

o e l s m T R

i i ti

d o

ll t

l

c t e r i z & c a i d s y s t e m

“ S u f f i c e n t l y ” r e a l s c

s t a n m d

b r a e t e d

d y r

e u r m e s

s u r d e

y b e t w e e n

S m i

c n s t e n s i o C

Experiment / Validation stage

F e d m e a s

u l a t e d s i

a l n g

i l u r e m e n t s

S m i

Field measurements

Simulated signal

Consistency between

Simulated & measured signal

Airborne measurements

+ Ancillary data

y characterized & calibrated system

“Sufficiently” realistic

stand models RT modeling

Signal

processing

Consistency

between

forest parameters

Stand characteristics

Stand characteristics

derived from signal

Allometric rules

V

Capacity to size a space system

meeting user requirements



A modeling framework developed

in the frame of the experiment /

validation study

Accurate modeling requires to develop

An efficient and reliable radiative transfer model

“Sufficiently” realistic stand models

9

[Dauzat et al., 2009]

al Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016 10

1- Radiative transfer model

DART - Discrete Anisotropic Radiative Transfer (CESBIO) :

● “Ray-Carlo” method (combination of ray tracing and Monte-Carlo methods)

● DART can model

Any lidar system (position, angular divergence, and pulse characteristics)

Multiple diffusions

Interactions between lidar signal and atmosphere

Solar noise

● Compatible with complex and realistic vegetation representations

[Gastellu-Etchegorry et al., 2013]

JAXA - Internation

CESBIO

2- Realistic forest scenes

« AMAP studio»

stand models based on

architectural representation

& stochastic tree simulations

(calibrated in the field)

« AMAP » scenes - 3D mesh trees

Voxelization of TLS data

(multiple scans, merged) Collaboration with the University of

Sherbrooke (Pr. R. Fournier

+ DTM, created from TLS measures

+ low vegetation layer (turbid) is added

+ standard optical properties (soil & vegetation)

« Lollipop » scenes - simple geometrical trees

field measurements

& allometries

« TLS » - 3D LAI distribution


altM

ext

- Stand height related metrics: ext, altM - Energy related metrics: N, iVeg, pVeg, pGrd - Vegetation profile related metrics: altP, rh15, rh50, rh85

N

iVeg

Ground/Vegetation limit

pVeg

pGrd

altP rh15

rh50 rh85

3- Analysis & validation of simulated waveforms

Definition of waveform metrics


METRICi_simulation – METRICi_reference

METRICi_reference

APEi = ABS * 100

Simulated WF

Absolute percent error computed for each metric

Bravais-Pearson Coefficient of Correlation (CC)

Waveform comparison

Reference WF


3- Analysis & validation of simulated waveforms

Simulation plateform

Integration of field measurements

and preprocessing of

ALS and TLS data

Building forest scenes (3 types)

Simulation of ALS and large footprint

Lidar data with DART

Analysis and validation

of simulated Lidar data

2


1

3


Preliminary studies on vegetation

representations

Questions:

What is the best way to represent a

forest plot in RT model for lidar

signal simulations?

What are the impacts of

uncertainties of the inputs used to

create forest scenes on the

simulated waveforms ?

15

Data sites

© T. Allouis

Island of REUNION

Tropical forests

© E. Grau

The LANDES forest

Maritime Pine

plantations


JAXA - Inter

7 plots on the Landes site

TLS

Lollipop

AMAP

Focus on 1 plot of Landes site : LAI distribution

national Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016 17

4 plots on Island of Reunion (tropical) site


3D view LAI distribution TLS Lollipop

No architectural models for tropical species no AMAP representations

op on Vegetation Lidar and n. 2016

Analysis

Evaluation of the quality of the 3 forest representations (Landes site)

● Reference for waveform comparison = ALS data aggregated at plot level

● Analyzed waveforms = simulated ALS data aggregated at plot level

TLS

JAXA - International Worksh Application from Space, Kyoto, Japan – 6-7 Ja 19

Lollipop AMAP

Analysis

Sensitivity analysis to evaluate the impact of uncertainties of the inputs on

the simulations (2 sites)

● Reference = large footprint simulated waveform

● Compared to waveforms obtained by changing initial parameters

Simplification of AMAP 3D mesh

Changes in : veg, ground,total LAI, LAI distribution, LAIunderlayer, slope

Example : Impact of change in Vegetation reflectance

Evolution of the waveform for 1 plot and 1 forest representation


Mean evolution for all the plots

Results

Which representation is compatible with RT modelling?

● Lollipop representation: inaccurate crown shapes poor results

● AMAP & TLS: shape of the vegetation peak well reproduced but differences in energy metrics were observed

● Different calibration needs

TLS : only reflectance Lollipop: LAI calibration (+allometry) AMAP (3D mesh) : very complex calibration

TLS based approach appears to be a good candidate to represent forest scenes

Sensitivity analysis :

● Shape metrics are stables (except when slope changes)

● Energy related metrics are very sensitive to changes in veg, LAI, ground

No absolute validation at this stage future experiments will include

● Characterization of the airborne lidar system

● Additional field measurements (spectroradiometer, atmospheric condition)

Simulations with errors in WF metrics < 10% can be expected

Increasing needs

for calibration



Instrumental studies

Impact of systems characteristics for high

energy fullwaveform systems

Evaluation of a low energy photon-counting

Lidar solution as an alternative to high

energy FW lidar

22

Sensitivity analysis a FW system

Sensitivity analysis to evaluate the impact on waveforms of:

● viewing angle, footprint size

● atmospheric condition

● solar noise

Main conclusions

● Intensity of ground and vegetation peaks decreases with viewing angle

and footprint size due to the signal spread risk to miss the ground

peak in dense forest

● Minor impact of sun and atmospheric noises

● Instrumental noise is likely to be the most significant source of noise and

a determining characteristic to size the system


Simulation of LEAF waveforms on

tropical forests

ICESAT data on tropical forest – emitted energy : 75 to 87 mJ

1st order diffusions

75 m

All diffusions


LEAF V2

1064 nm, 20 m footprint , 60 mJ,

Pulse width : 2ns Ø 60 cm 500 km

Comparing LEAF, GEDI & MOLI

LEAF V2

1064 nm, 20 m footprint , 60 mJ (= 15 mJ / detector) Pulse width : 2ns Ø 60 cm

500 km

GEDI central footprint

1064 nm, 25 m footprint , 17 mJ

Pulse width : 11-13 ns Ø 60 cm (?) 400 km

[Coyle et al., SPIE 2015]

GEDI lateral footprints

7,7 mJ

MOLI

1064 nm, 25 m footprint , 15 mJ (60/4)

Pulse width : 7 ns Ø 60 cm (?) 400 km

[Asai et al., IGARSS 2014]

With instrumental noise,


ground peak hardly detectable on this WF

Low energy Lidar: a viable alternative solution?

Operating high energy lasers in space (e.g. full waveform systems with

few 10 mJ at few 10 Hz) is associated with risks: laser induced damage,

contamination, eye safety…

The design of ATLAS on ICESat-2 shows that alternative approach is

possible (few 10-100 µJ/shot at 10 KHz)

ATLAS design optimized for hard ground targets, vegetation (tree

height) comes as secondary goal.

Study conducted by Airbus Defense and Space to verify if the low

energy approach is suited for tree height determination


[Lochard, J et al., SilviLaser 2015]

Technical point on emission & detection parts

Two main options considered

Two candidates identified

Crystal amplified lasers Optical fiber amplified lasers

Advantages + High TRL for space mission

-Reduced alignment constraints

-Lower sensitivity to vibrations

and to contamination

- Better thermal dissipation…

Drawbacks - Complex production

-Limitation in peak power

-Concern on the resistance to

radiations

Photo-multiplier tubes (PMT) Avalanche Photodiode (APD)

Advantages +

-Higher TRL (Caliop, soon

ICESat-2

- Low dark current.

- No blind time

-Large accessible spectral domain

(UV to IR, with various substrates)

- Higher quantum efficiency

Drawbacks - Reduced spectral domain

[0.3µm-0.9µm]

Need to operate at lower

temperature for IR

JAXA - International Workshop on Vegetation Lidar and Application from Space, Kyoto, Japa[nL–o6c-h7 aJradn., 2J01e6t al., SilviLaser 2015] 27

Performance and design sizing approach

Performance requirement:

● Measurement noise < 2 m rms

● Bias < 4 m

(= random error on tree height)

Overall approach toward instrument design:

● Definition of reference scenes and waveform simulations

● Performance estimation through performance model and comparison to the

requirement

● Setting of a threshold value for the lidar product (laser energy x telescope

collecting surface)

● Instrument design based on the lidar product.


X nber of simulation runs

Mediterranean maquis Tropical forest

Savanna Northen forest

Tree height

Ve

g.

De

ns

ity

Mediterranean Tropical forest maquis Mangrove

Savanna Temperate forest

Pine plantation

Accumulation of 14 successive shots

(i.e. over 10 m for 10 KHz PRF)

Basic detection criteria motivated

by robustness, simple

implementation and interpretation

DART

Design sizing approach

4 λ : 355 , 532 , 1064 & 1550 nm

20 m diameter spot

System altitude 500 km.

3 % of integrated energy

97 % of integrated energy

Max stand height


First performance results and operating point

Low performances

● For UV and visible linked to low values of both ground and leaves reflectance

● For the lower stands due to a long dead time of the detector (10 ns i.e. 1.5 m

height)

First instrument design sizing based on to 2 scenes (temperate and

tropical forests) and two IR wavelengths.

Measurement noise set for an accumulation over 14 shots at 0,5 m for

the temperate forest

derived lidar product threshold are 0.10mJ.m² at 1.064µm and 0.07mJ.m²

at 1.55µm.


Budget and conclusions

Operating point: a 0,8 m Ø telescope and

● 200 µJ / shot at 1.064µm

● 140 µJ / shot at 1.55µm

Resulting instrument mass: 220 kg (incl. 25 % margin)

Power budget : 240 Watt (incl. 20 % margin)

Simulations show the viability of such an approach for tree height

determination with metric accuracy for scenes of major interest

including the tropical forests

Limitations of the study

● A full error budget would require to consider additional contributors (e.g.

uncertainty on the atmosphere characteristics and effects linked to the

platform (e.g. pointing, altitude uncertainties))

● Results highly dependent on the method used to estimate height

Same satellite class

than for LEAF

with current FW specifications



Research avenues that could be

explored to improve forest parameters

assessment from Lidar

Promising recent studies using ALS

data

To what extend these approaches can

be transferred to process spaceborne

data?

32

User requirements not always met

with existing ALS based ABA

ABA approaches are widely used to assess forest parameters (Volume, AGB, basal area) from ALS data

Metrics are derived from height and density distributions of ALS echoes

Most explanative metrics are used to build a predictive model

But low robustness level for such approaches

● Metrics are highly correlated

● One model / site

● Insufficient accuracy for complex stands (e.g. multilayered broadleaved

stands), e.g. RMSE > 20% for biomass assessments

Development of new methods to overcome these issues


Explored approaches

Seeking for new metrics

● Using normalized vegetation profiles [Bouvier et al., 2015], [Vincent et al., 2015]

● Penetration distribution metrics instead of height metrics [Vega et al., 2016]

Using few meaningful metrics to build more general models (one shape)

“Combining LIDAR and growth and yield models”

[Bouvier et al., 2015]

Heterogeneity of the

canopy surface

RSD < 20%

Vertical

heterogeneity

R² > 0,85 -

even for broadleaved multilayered stands

Aillons

Beaufort

Vercors (O, NE)


Landes

OPE HayeLanguimberg

Bure Vosges

● Strengths of each approaches (growth and yield, LIDAR) integrated in a consistent and comprehensive way – not only additive or multiplicative combination of metrics

● No bias among 10 different sites with a single model for BA estimation

[St-André et al., SilviLaser 2015]

Remaining challenges

Transfer these approaches to tropical forests and to large footprint FW

lidar signal

Relief One of the limitation

when processing ICESat 1

● Improved height prediction expected with future systems

Smaller footprint

Local slope assessment (better than SRTM DTM)

● Vegetation profile?

Photon-counting data: vegetation profile? Smithsonian Environmental Research Center, MD


Crédit: R. Nelson, FLORESTA project- 2013

Mix between ground and vegetation signal


Perspectives

36

n t s

l t s i g n l

C o e t w e e n

S i l

o r n e m e a s u r e m

a t a

a r a c e r i z e a l b r a e d

u f f i n t l

s i c d & h

e n t

+ A n c a r d y l

b r

s i n

s i s t e n

e t n

f m e e r s

S n d c h a r a c t e r s t c s a t

t a p a r e s t o r

e w e b

c y n o

g

C

s e

a l

c

n g

c t e r s t c

e r v e d f r m s a l n g i o d

s h a a r c d n S a

A e m u r e s a F i d m e e l

m o d e s s t a d n d o m T R a l s t c i i e y ” r e c i

a

“ S

d S i m u a e

m u l a t e d & m e a d s g n a s u e r

n c b y e n s i s t

p a c i t y t p a c e m s y s e t s o s z a e a C nadir

(full

l i s

Experiment / Validation step

F e d m e a s u r e m

u l a t e

n t s

+ A n c

ll t

l a a

( f u l l d s y s t e m

“ S u f f i c e n t l y ” r e a l s c

e l d e i n

i

i a l

p r o c e g

C o n s s t

e r s

S n d c h a r a c t e r i s t i c s

S t a n d c h a r a c t e r i s t i c s

d e r

i i

v e d f o m a l

A l o m e t r c r u e l

s g n r

a t

m e t

e t w e n

f s t p a a r e o r

e

c y

b

n e i

s s n

l g

S n g

o m T R

i i ti

s d o m s t a n d

e y c h a r a c t e r i z e & c a i b r a t d

d y i a r

e m

s i g n a l

A i r b o r n e m e u r e a s

d

a l

n s y b e t w

u l a t e d & m e a s u r e m

e e n

S i

c s t n e i o C

i n d s g m

i l e n t s

S i

i

Ai s

il

y c

llometric rules

t t ystem

French Guyana experiment

Sept 2016


l eling

Si

pro

i i

t i i

i

V

Capacity to size a space system

What can be expected from ICESat2 & GEDI ?

+ Comparison between FW and PC lidars (crossing orbits)

+ Complementarity between green and NIR wavelength

+ Complementarity between BIOMASS and lidar missions to improve biomass

assessments

+ Evaluation of several sampling strategies

- Boreal forests not covered (~ 30% of the world's total forest area)

- The interest of an imager or stereo-imager co-registered with a lidar will not be

evaluated

2000 2010 2020 2030 2040

ICESat (6 years, non continous)

ICESat-2

(5-7 years)

BIOMASS

(5 years)

GEDI (? years)

CESBIO


MOLI LEAF….

Technology is not everything…

Research is needed to refine and integrate user requirements in link

with emerging research topics; e.g. assimilation of lidar data in

● Biodiversity models

● Ecosystem models for carbon stocks and fluxes assessment

New global models developed to

include inter-grid variability &

management practices

The capacity to meet user requirements is highly dependent on field

measurement efforts

● Allometry

Tree climbing for

height and crown

architecture studies

© AMAP

Résultats Marc

Landes Forest – France - Pine plantations


Allometric equation

Deleuze et al.

(2013)

O. Shaiek et

al. (2011)

Cacot, (2007) JY Fraysse et

L Cotten

Predictive model for

biomass (T/ha)

R² 0.95 0.93 0.79 0.50

RSD (%) 12.7 13.0 19.8 32.6

RMSE 10.8 10.8 18.8 31.3


Thank you for your attention

© I.Piney

Documents

Waveform simulations and analysis of Lidar data for … - International Workshop on Vegetation Lidar and Application from Space, Kyoto, Japan – 6-7 Jan. 2016 1 Waveform simulations