Relationships between PALSAR

backscattering data and forest above

ground biomass in Japan

○ Takeshi Motohka (Japan Aerospace Exploration Agency)

2011/07/28: IGARSS 2011, Vancouver, Canada.

Masanobu Shimada (Japan Aerospace Exploration Agency)

Osamu Isoguchi (Remote Sensing Technology Center of Japan)

Masae I. Ishihara (Japan Wildlife Research Center)

Satoshi N. Suzuki (Japan Wildlife Research Center)

Outline

- Background

- Data

- In situ biomass data

- PALSAR yearly mosaic data

- Results

- Relationship between biomass and PALSAR data

- Mapping forest biomass

- Summary and future works

BackgroundForest biomass is a key parameter to assess

Emission of greenhouse gasses (CO2, CH4, etc.)

Accumulated carbon in forests Biodiversity

A large-scale, time-series, globally consistent biomass monitoring is important for various projects such as REDD+.

Biomass monitoring by PALSAR

- Microwave backscatter shows high correlation with forest tree biomass especially for longer wavelength (i.e. L-band, P-band, …).

- Spatial (10 – 100 m) and temporal (46 days) resolution meet the REDD+ or FCT methodologies.

- Well calibrated global datasets for 5 years (2006 ～ )

- ALOS mission was ended in 2011, but next ALOS-2 (PALSAR-2) will be launched in 2013.

Phased Array type L-band Synthetic Aperture Radar

PALSARPALSAR-2

ALOSALOS-2

Lucas et al., 2011

Mitchard et al., 2009

Englhart et al., 2011

Many studies have revealed the relationships between forest biomass and PALSAR data at various regions and forest types.

AfricaAustralia

Indonesia

Purpose of the study

• Target ＝ Japan

• Investigating the relationships between PALSAR backscattering data and above ground biomass of Japanese forests

• Testing the retrieval of forest biomass using the obtained empirical relationships and PALSAR yearly mosaic data

In situ Biomass Data

Website: 　 http://www.biodic.go.jp/moni1000/

“Monitoring Site 1000” project 　by the Ministry of Environment of Japan (since 2003)

• Network of long-term research sites for biodiversity assessment

• 49 tree census sites

• Located at various forest types

• Only natural forests were selected in the study (not including artificial forests)

Deciduous broadleaf forest(Tomakomai, Hokkaido)

Evergreen coniferous forest(Otanomousu-daira, Nagano)

Deciduous broadleaf forest(Chichibu, Saitama)

Evergreen broadleaf forest(Yona, Okinawa)

Tree diameter of breast height [cm] (DBH)

Tree diameter of breast height [cm] (DBH)

Dry weight [kg]Dry weight [kg]

measured all trees in about 1 ha plotexcept for DBH < about 5 cm.

Allometric equationsfor each specie

Ʃ (dry weight) / stand-size

Biomass [t/ha]Biomass [t/ha]

Processing of tree census data

Statistics of the forest stands (n=44)

Range Mean Median

Observation　year 2005　-　2009 - 2007

Elevation　(m) 40　-　1880 583 460

Annual　mean　temperature　(degree) 2.5　-　21.9 10.5 9.4

Stand　size　(ha) 0.1　-　1.2 0.9 1.0

Tree　density　(number/ha) 493　-　3975 1304 1164

Mean　DBH　(cm) 8.7　-　26.8 16.9 17.3

Basal　area　(m2/ha) 13.0　-　78.2 44.8 44.0

Above　ground　biomass　(t/ha) 46.7　-　467.9 270.7 270.7

DBH:　Diameter　at　Breast　Height

PALSAR yearly mosaic

R: HH

G: HV

B: HH/HV

Year: 2007, 2009

Mode ： Fine beam dual

(HH, HV)

Mosaicking period ：

Jun. - Sep.

Pixel sampling ： 10 m

Orbit: Ascending

- Long-strip processing- Ortho-rectification- Slope-correction- Mosaicking

γ0　[dB]　=　10　log 〈 DN2 〉 -　83

15 x 15 pixels averaging

Shimada & Otaki (2011); Shimada (2011) in “IEEE JSTAR special issue on Kyoto and Carbon Initiative”

Generation of PALSAR yearly mosaics

Converting DN to gamma naught

xbay ln

Saturation level:

- dy/dx = 0.01… 91 [t/ha]

- dy/dx = 0.005… 182 [t/ha]

HV

RMSE: 0.703 [dB]

PALSAR γ0 vs. biomass

Saturation level:

- dy/dx = 0.01… 68 [t/ha]

- dy/dx = 0.005… 136 [t/ha]

PALSAR γ0 vs. biomass xbay ln

RMSE: 1.053 [dB]

HH

Red ●: After correction 　

　

Green +: Before correction

RMSE: 1.053 [dB] RMSE 0.703 [dB]

RMSE: 2.312 [dB] RMSE 2.073 [dB]

Effect of slope-correction

HV

HH

Precipitation vs. PALSAR gamma naughtrainfall (over 10mm during 3 days before obs.)

RMSE:HH: 0.670 dB HV: 0.402 dB

Mean bias between rainy and non-rainy data:

HH: +0.177 dB HV: +0.044 dB

Inversion of forest biomass

　 RMSE: 　 106.23 t/ha

　 %RMSE: 　 39.3 %　　 (=RMSE/mean)

PALSAR HV γ0

(15 x 15 pix average)

Water, Lay-over, Shadowing mask

Urban area mask

Biomass map

b

aAGB

0

exp

400 ～0 100 200 300

Above ground biomass (t/ha)

400 ～0 100 200 300

Above ground biomass (t/ha)

Cropland

Cropland

Wetland

1 km

(c) Google Earth

400 ～0 100 200 300

Above ground biomass (t/ha)

Summary

• We examined the relationships between PALSAR backscattering data and forest biomass in Japan.

HV polarization was better to use (low RMSE and high saturation level).

Slope correction was very important to reduce the error especially in mountainous regions.

More data points were needed to investigate the difference among vegetation types. Airborne LiDAR can be good solution of this.

Summary

• Simple inversion of forest biomass

Spatial pattern seems to be good.

Problems: accuracy and saturation

Possible solution:

• Additional SAR analysis (multi-temporal data, full-polarimetric data, etc…)

• Data fusion with ALOS/PRISM DSM and ALOS/AVNIR-2 (10-m res. VIS&NIR) data

• Data fusion with ICESAT/GLAS data

• More in-situ data points and more evaluation