Using Satellite LiDAR Data Using Satellite LiDAR Data

Sampling

Sampling & & Mapping Forest Resources Mapping Forest Resources

Ross Nelson Ross Nelson NASA

NASA –– Goddard SpaceGoddard Space

Using Satellite LiDAR Data Using Satellite LiDAR Data

NASA

NASA –– Goddard Space Goddard Space Flight Center Flight Center

Biospheric Sciences Branch Biospheric Sciences Branch Code 614.4

Code 614.4

With guest appearances by:

Dan Kimes, Jon Ranson, Guoqing Sun, Paul Montesano, Slava Kharuk,

Erik Næsset, Terje Gobakken,

Jonathan Boudreau, Hank Margolis, André Beaudoin, Chhun-Huor Ung, Luc Guindon, Philippe Villemaire, Tim Gregoire, Göran Ståhl

But first, a little background….

Outline:

Outline:

- Siberia

Q b

- Quebec

WILL LAND ECOSYSTEMS

EXACERBATE OR MITIGATE WARMING ?

Largest remaining uncertainties about the Earth’s carbon budget are in its terrestrial components.

7.6 ± 0.4

NEED TO REDUCE THESE UNCERTAINTIES

Global Carbon

Budget (Canadell et al., 2007)

To Atmosphere

Unidentified

4.1 ± 0.04

1.5 ± 0.5

UNCERTAINTIES

Fossil Fuels Land Use

Ocean Uptake

Unidentified (“missing”)

Terrestrial Sink Atmospheric

To L d/O

Fossil Fuels

Change Peta (10¹⁵) grams of carbon/year

Carbonp

2.2 ± 0.4

Land/Ocean 2.2 ± 0.4

2.8 ± 0.7 TROPICAL BIOMASS

LARGEST CONTRIBUTOR

The map shows areas with a canopy cover of at least 40%

least 40%

by woody plants taller than 5

meters

From 1990 to 2000, the global area of temperate forest increased by almost 3 million hectares per year, while deforestation in the tropics occurred at an

Millennium Ecosystem Assessment Synthesis Report (2005)

average rate exceeding 12 million hectares per year over the past two decades.

Uncertainty in biomass change is greatest in tropics. Biomass density more uncertain than changes in forested area.

So how can we use lasers to measure the amount of wood on the ground? g

Airborne lasers (LiDARs) measure distance:

ground dist – tree dist. → tree height

tree height α biomass α carbon

photo: Kaiguang Zhao, TAMU

ICESat Land Applications ICESat Land Applications

Courtesy of Dave Harding, NASA GSFC

GLAS Waveform Range Offsets & Elevations

1064 nm Laser Pulse

Travel TTime

Return Amplitude

Outline:

Outline:

- Siberia

- Quebec

Study Area #1: South Central Siberia

(Ranson/Sun/Kimes/Kharuk/Montesano)

- just NW of Lake Baikal

(N of Irkutsk to Krasnoyarsk, - 10° x 12° area,

811 414 k ² - 811,414 km²,

- 55 GLAS orbits, L2a, - 101,831 GLAS pulses, - ~17.6 km between fls.

- MODIS for land cover, - GLAS for biomass

Study Area in Central Siberia

•

80˚ to 110˚ Longitude and 50˚ to 75˚

Latitude Latitude

• The Russian forest contains 22% of the Earth’s forest area.

• Biomass/carbon estimates for these vast forest have only recently been characterized.

F i d i hi i i

• Forest inventory data in this region is lacking and decades old.

90N

MODIS MOD09 8-day composite, date: July 12 2003, Band combination: True Color

GODDARD SPACE FLIGHT CENTER BIOSPHERIC SCIENCES BRANCH

Kimes, Ranson, Sun, Nelson, Kharuk, Montesano.

Goals:

Goals:

Goals:

Goals:

•Produce the most accurate maps possible of timber volume and above ground forest phytomass/carbon stocks

for the Siberian study area.

• Develop test and integrate new remote sensing Develop, test, and integrate new remote sensing

methods for extracting forest canopy structure measures using MODIS and GLAS data.

• Compare these new remote sensing methods with existing ground estimates.

• Produce a realistic error bound on the remotely sensed carbon and timber volume estimates.

Carbon1.ppt

carbon and timber volume estimates.

Predicted Timber Volume - NN - 6 GLAS Variables

(MedH, Ht2, Fslope, Mjp2loc, Ga3, Npk); R²=0.78, RMSE=81 m³/ha

picture: J. Boudreau

GODDARD SPACE FLIGHT CENTER BIOSPHERIC SCIENCES BRANCH

Kimes et al.: Landcover Attributes from ICESat GLAS Data

Methods Methods

• Produce MODIS classification (500 m res.)

• Process waveform data for each GLAS shotProcess waveform data for each GLAS shot

• Develop models to extract timber volume, biomass, and height using field data.

• Apply models to all GLAS shots in the study area.

• Produce timber volume and biomass maps and a statistical error bound on the estimates for each class and for the total study area

Timber Volume Map Timber Volume Map

10

10 ° ° x 12 x 12 ° ° Study Area Study Area 10

10 x 12 x 12 Study Area Study Area GLAS and MODIS data:

GLAS and MODIS data:

•• MODIS forest classes MODIS forest classes

•• MODIS forest classes MODIS forest classes

•

• MODIS % tree cover MODIS % tree cover

•

• GLAS timber volume GLAS timber volume

Total Timber Volume (m

Total Timber Volume (m

³³

/ha): /ha):

Mean Forested Area:

Mean Forested Area:

Mean Forested Area:

Mean Forested Area:

MODIS/GLAS (10

MODIS/GLAS (10°° slope)slope)

163.4

163.4

±±

11.8 m 11.8 m

³³

/ha /ha

MODIS/GLAS (90

MODIS/GLAS (90°° slope)slope) MODIS/GLAS (90

MODIS/GLAS (90 slope)slope)

171.9

171.9

±±

12.4 m 12.4 m

³³

/ha /ha Shepashenko et al. (1998) Shepashenko et al. (1998)

162.1 m

162.1 m

³³

/ha /ha

Above Ground Phytomass Above Ground Phytomass

10

10 ° ° x 12 x 12 ° ° Study Area Study Area

Using GLAS and MODIS data:

Using GLAS and MODIS data:

•• MODIS forest class, MODIS forest class,,,

•• MODIS % tree cover, MODIS % tree cover,

•• GLAS timber volume GLAS timber volume

Total Phytomass:

Total Phytomass:

forested area (10

forested area (10

°°

slope): slope):

86 7

86 7

±±

6 3 6 3 /h /h 86.7

86.7

±±

6.3 ton/ha 6.3 ton/ha forested area (90

forested area (90

°°

slope): slope):

90.3

90.3

±±

6.7 ton/ha 6.7 ton/ha

Above Ground Carbon Estimates (Pg)

Central Siberia (10X12 degree study area)

Forest:

( g y )

4

2 3

etagrams)

95% CI

1 2

Weight (pe

0

Russian Forest Map (1990)

Land Resources of

SPOT-based 1km Gobal

MODIS MOD12- IGBP 1km Land

MODIS-based 500m Land

GLAS/MODIS 500m (2003) Map (1990) Resources of

Russia

1km Gobal Land Cover

(2000)

IGBP 1km Land Cover (2001)

500m Land Cover (2003)

500m (2003)

Dataset

[1.95±0.14 Pg]

Nelson et al.

(2008) Stolbovoi

et al. (2002)

Siberia – results:

1. Siberia: GLAS estimates of merchantable volume agree with ground estimates on a 811,414 km

²

study area in

south-central Siberia.

MODIS/GLAS (10

MODIS/GLAS (10 ° ° slope) slope) 163.4 163.4 ± ± 11.8 m 11.8 m

³³

/ha /ha MODIS/GLAS (90

MODIS/GLAS (90 ° ° slope) slope) 171.9 171.9 ± ± 12.4 m 12.4 m

³³

/ha /ha Shepashenko et al. (1998), ground

Shepashenko et al. (1998), ground 162.1 m 162.1 m

³³

/ha /ha

10°forest estimates: + 0.8% difference, CV: 7.2%

90°forest estimates: + 6.0% difference, CV: 7.2%

Outline:

Outline:

- Siberia

Q b

- Quebec

“A Realistic Analysis of the Variability of

QCLP – Quebec Carbon LiDAR Project

y y

Carbon Estimates Using Airborne and Space LiDAR”

The QCLP Team:

- Ross Nelson – NASA/GSFC Dan Kimes NASA/GSFC - Dan Kimes – NASA/GSFC - Hank Margolis – Laval Univ.

- Jonathan Boudreau – Laval Univ.

- André Beaudoin – CFS - Chhun-Huor Ung – CFSg - Luc Guindon – CFS

- Philippe Villemaire - CFS - Tim Gregoire – Yale Univ.

- Erik Næsset – Nor. Univ. Life Sci.

- Terje Gobakken – Nor Univ Life Sci - Terje Gobakken – Nor. Univ. Life Sci.

- Göran Ståhl – Swed. Univ of Ag. Sci.

+ Ryan Collins, Capitol Air

The Quebec sampling plan:

Satellite LiDAR:

GLAS ^(9-11/03)

b_PALS= f(ht, cc)_GLAS

Airborne LIDAR:

PALS ^(8/05)

b_ground= f(ht, cc)_PALS

Ground Reference:

MNRQ and CFS ground plots (400 m²).

(y2000 – 2004)

photo:

J. Boudreau

l t t 59 75N

~62N

last tree ~59.75N 1710

km 1056

km

62N

1061 km

~55N

~50N

775 km

~49N

The Quebec sampling plan:

Satellite LiDAR:

) ( 753 . 0 ) ( 587 . 6 ) ( 846 . 3 ) 515 . 4 ˆ (

SRTM GLAS

GLAS

PALS w f r

b = − + − −

GLAS ^(9-11/03)

N 1325 Rsq 0.6012 AdjRsq 0.6003 RMSE 31.986

75 100 125 150 175

) ( )

( )

( )

( _{GLAS GLAS SRTM}

PALS f

ass (t/ha)

b_PALS= f(ht, cc)_GLAS

-50 -25 0 25 50 75

GLAS bioma

Airborne LIDAR:

PALS ^(8/05)

535 . 2 ) ( 154 . ˆground=10 hqa − b

-25 0 25 50 75 100 125 150 175 200 225 250

PALS biomass (t/ha)

b_ground= f(ht, cc)_PALS

biomass (t/ha)ry biomass (t/ha) drydr

) (m h_qa

Ground Reference:

MNRQ and CFS ground plots (400 m²).

(y2000 – 2004)

photo:

J. Boudreau

MNRQ Ground Reference GLAS estimates

Accuracy Assessment: Quebec Southern Ecozones - Regional Estimates:

biomass stan.err. no. Biomass stan.err. percent

(t/ha) (t/ha) plots (t/ha) (t/ha)‡ difference

Northern Temperate Ecozone: (109,769 km²) Ecozone: (109,769 km )

Conifer 76.60 5.82 49 64.69 1.86 - 15.5

Deciduous 77.85 4.95 176 89.91 4.62 + 15.5

Mixedwood 65.91 2.79 313 82.22 1.21 + 24.7

Mi d d E

Mixedwood Ecozone:

(98,101 km²)

Conifer 85.90 1.57 583 74.61 3.14 - 13.1

Deciduous 75.00 2.98 290 82.98 3.33 + 10.6

Mixedwood 87.15 1.43 1177 81.80 1.53 - 6.1

Southern Boreal Ecozone:

(374,665 km²)

Conifer 86.36 0.37 10,007 63.80 5.07 - 26.1

Deciduous 56 71 1 77 617 60 68 3 42 + 7 0

Deciduous 56.71 1.77 617 60.68 3.42 + 7.0

Mixedwood 82.16 0.73 3602 68.54 1.93 - 16.6

Provincial Comm. For. 81.90 0.50 16814 76.14 2.52 - 7.0

‡ SRS standard errors, stratified linear model, w/prediction error, w/covariance.

photo:

J. Boudreau

…that would be

5.04 ^± 0.42 Gt dry biomass,

or 2.52 ^{± 0.21 Gt C}

in Quebec

Thank you.

Summary:

1. Quebec:

- GLAS estimates of dry biomass within 1-26% of MNRQ ground estimates for commercial forest cover types - 582,536 km

²

.

- assessed 1.27 million km

²

using a space LiDAR.

ETM+/PALS/GLAS:

76.1 ± 2.5 t/ha (97 orbits)

MNRQ ( d)

81 9 ± 0 5 t/h (16 814 l t )

MNRQ (ground):

81.9 ± 0.5 t/ha (16,814 plots)

% difference: -7.0%

CV, GLAS: 3.3%

So can you do large area forest inventories with ICESat/GLAS?

Yes.

Any problems associated with using GLAS for large area forest inventory and monitoring?

Yes, significant problems.

Problems include:

1. An apparent inability of GLAS to accurately measure short-stature, open forest.

PALS Mean Height GLAS Mean Height

PALS & GLAS Heights x Latitude

a convolved

2. Slope + large footprint waveform LiDAR = topography-forest canopy waveform Problems include:

Slope effects can be mitigated with DTMs.

Problems include:

3 C S /G S 2008 201

3. An observational hole in ICESat/GLAS data acquisitions, ca. 2008 – 2015.

4. A possibility that ICESat II (launch ~2015) will have degraded forest 4. A possibility that ICESat II (launch 2015) will have degraded forest

measurement capability due to - ∆ footprint size (50m to 70m), - ∆ pulse width (6 ns to 10 ns).

5. Unable to assess/measure forest degradation, i.e., the intermittent, scattered removal of individual high-value trees from a forest.

Satellite optical data, e.g., ETM, SPOT, MODIS → forest location and type.

Satellite LiDAR data to measure structure, estimate biomass and carbon.

Current Outlook on U.S. Satellite LiDARs:

ICESat I / GLAS:

(ice mission)( ) - 65m footprint, 172 m post spacing, waveform, 6 ns pulse width,

- single beam system, 15 km between orbits at equator (91 day repeat), - launched Jan. 2003,

died October 19 2008 (will it rise again ?) - died October 19, 2008 (will it rise again ?)

ICESat II/GLAS:

- launch due sometime in 2015 (ice mission) - launch due sometime in 2015, (ice mission)

- 50-70m footprint, 140 m post spacing, waveform, 6-10 ns pulse width, - single beam system, 15 km between orbits, possible off-nadir pointing

to pick up across-track slope, - 3-5 year mission.

DESDynI:

with L-band radar (?) (solid earth mission) l h 2015 2017 (??)

- launch ~2015-2017 (??),

- 25 m footprint, contiguous, 3-5 beams across-track, 3-5 year mission.

LIST:

a swath mapper launch 2017++??? (hydrology mission)

LIST:

a swath mapper, launch 2017++??? (hydrology mission) - 5 m footprint, contiguous footprints, waveform

- complete global coverage over 5 years.

GLAS-related Publications:

1 Boudreau J R F Nelson H A Margolis A Beaudoin L Guindon D S Kimes 2008 1. Boudreau, J., R.F. Nelson, H.A. Margolis, A. Beaudoin, L. Guindon, D.S. Kimes. 2008.

Regional aboveground forest biomass using airborne and spaceborne LiDAR in Québec. Remote Sensing of Environment 112(10): 3876 – 3890.

2 Gregoire T G Q F Lin J Boudreau and R Nelson 2008 Regression estimation following 2. Gregoire, T.G., Q.F. Lin, J. Boudreau, and R. Nelson. 2008. Regression estimation following

the square-root transformation of the response. Forest Science 54(6): 597-606.

3. Nelson, R.F., E. Næsset, T. Gobakken, G. Ståhl, and T. Gregoire. 2008. Regional Forest Inventory Using an Airborne Profiling LiDAR Journal of Forest Planning 13: 287 294 Inventory Using an Airborne Profiling LiDAR. Journal of Forest Planning 13: 287 - 294.

4. Nelson, R., J. Boudreau, T. Gregoire, H. Margolis, E. Næsset, T. Gobakken, and G. Ståhl.

2009. Estimating Québec Provincial forest resources using ICESat/GLAS. Canadian Jo rnal of Forest Research in press

Journal of Forest Research, in press.

5. Nelson, R., K.J. Ranson, G. Sun, D. Kimes, V. Kharuk, and P. Montesano. 2009.

Estimating Siberian timber volume using MODIS and ICESat/GLAS. Remote Sensing of Environment 113(3): 691 701 doi:10 1016/j rse 2008 11 010

of Environment 113(3): 691-701. doi:10.1016/j.rse.2008.11.010.

6. Nelson, R. 2009. Model effects on GLAS-based regional estimates of biomass and carbon.

International Journal of Remote Sensing, accepted for publication.

Statistical Framework:

n

_ijk

∑ ^ˆ

p

ijkp k

b b

∑ =

= ¹ ˆ

ijk

ijk n

b

The orbit becomes my unit of observation

of observation.

i vegetation zone i - vegetation zone, j - cover type,

k - orbit, p – pulse.

Cover Type Biomass Estimate within Vegetation Zone:

ijk n

ijk

ij w b

b

ij

ˆ

ˆ ⁼ ∑ ^{ijk where w}

^ijk

⁼ _∑

^nij

ⁿ

^ijk

k

ijk ij

1

∑ = ∑

= k

n

ijk 1

∑

n

= nij

k

wijk 1

= 1.0

n_ij

1 ˆ ) ( ˆ

ˆ ) r(

aˆ

v

¹

−

2

= ∑

k=

ij ijk

ijk ij

b b

w b

j

i vegetation zone

SRS:

) 1

(

^ij

n

_ij

−

ⁱ - vegetation zone,

j - cover type, k - orbit,

p – pulse.

Vegetation Zone Estimates of Biomass:

∑

=

ⁿⁱ _{ij ij}

i

w b

b ˆ ˆ

^ij

ij

w = a

^and

∑

ⁿⁱ

^w

^{ij = 1.0}

∑

j=

j j

1 i

j

a ∑

= j

ij 1

1

ˆ ) ˆ , v(

oˆ c 2

ˆ ) r(

aˆ v ˆ )

r(

aˆ v

1

1 1

1 2

im ij

n

j

n

j m

im ij

n

j

ij ij

i

w b w w b b

b

i i

i

∑ ∑

∑

⁻

= = +

=

+

=

ˆ ) ( ˆ

ˆ ) ( ˆ

) ,

( ( , )

n

∑ ∑

n

b b

b b

m j

i i j m

1

) (

) (

ˆ ) ˆ , v(

oˆ

c

₂

) , (

1 1

−

−

−

= ∑ ∑

= =

m j i

im iml

k

ij ijk

l im

ij

n

b b

b b

b b

) , ( j

Provincial Estimates – four models:

(n=104,044 , 97 orbits, SRS)

dry biomass estimates Prov. biom. totals

model mean(t/ha) SE (t/ha) CV(%) total(Gt) SE (Gt)

model mean(t/ha) SE (t/ha) CV(%) total(Gt) SE (Gt)

believe GLAS

Sqrt, nostr, noreg 41.72 2.82 6.8 5.29 0.35

Lin, nostr, reg 40.65 5.13 12.6 5.15 0.65

believe ETM+

Lin, str, reg 39.78 3.17 8.0 5.04 0.40

Sqrt, str, noreg 38.94 2.17 5.6 4.94 0.28

ΔC = 0.18 Gt

(~7% diff.)

- cannot add regression error to sqrt model (positive bias)

- stratified models lower due to nonforest biomass = 0 regardless of GLAS