igarss11_takaku_dsm_report.ppt

2011 IEEE International Geoscience & Remote Sensing Symposium

July 24–29, 2011 Vancouver, Canada

High Resolution DSM Generationfrom ALOS PRISM

- Archiving and Mosaicking -

Junichi TakakuRESTEC

July 2011

Takeo TadonoJAXA


Contents

Introduction Software overview Data archiving status DSM mosaic for global data set Summary


Introduction Panchromatic Remote-sensing Instrument for Stereo Mapping

(PRISM) onboard Advanced Land Observation Satellite (ALOS)– Mission Life: Jan.2006. – Apr.2011

– Three push-broom scanners, nadir, +23.8 ﾟ & –23.8 ﾟ for stereo triplets in 35km-swath.

– Pan images: 2.5 m, 8 bit.

© JAXA© JAXA


Introduction Digital Surface Model (DSM) generation from ALOS PRISM

– with DOGS-AP (DSM and Ortho-rectified image Generation Software for ALOS PRISM) – JAXA’s exclusive triplet-image-matching software

– Grid spacing: 10m (0.3 arc-sec)– Unit: 35km x 35km / scene

PRISMTriplet stereo images DSM

(and Ortho-image)

DOGS-AP


Introduction Visual Comparison (4km x 4km)

SRTM-3 DEM ASTER-GDEM PRISM-DSM LiDAR-DSM

0m 150m

90m

N60 ﾟ ~S56 ﾟ PRISM data areas except for clouds

N/A

IfSAR from STS-99

Airborne LiDAR

Mesh Size

Data Area

Source

N83 ﾟ ~S83 ﾟ

Satellite Stereo Triangulation

ASTER GDEM is a product of METI and NASA

30m 10m 1m

Satellite Stereo Triangulation


Software overview Full support for the PRISM sensor model

– Interior parameters– Exterior parameters (including satellite orbit and attitude data)– Relative / absolute orientation

(Ba c kwa rd)

0.05

0.10

0.15

0.20

0.25

- 150 - 100 - 50 0 50 100 150

12345678

PRISM CCD unit alignments

(Nadir)

- 0.55

- 0.50

- 0.45

- 0.40

- 0.35

- 100 - 50 0 50 100 150

123456

(Fo rwa rd)

- 0.10

- 0.05

0.00

0.05

0.10

- 150 - 100 - 50 0 50 100 150

12345678

CCDA/B

CCD6CCD7

CCD8

Ground/CCD Plane

Earth rotate direction Forward

Nadir Backward

CCD3CCD4

CCD5

CCD1CCD2

CCD3CCD4

CCD5CCD6

CCD3CCD4

CCD5CCD6

CCD7CCD8

CCD2CCD1

ALOS /Focal Point

CCD1

CCD2Pixel No.1Pixel No.1

FlightDirection

CCD1CCD2

CCDA/B

CCD6CCD7

CCD8

Ground/CCD Plane

Earth rotate direction Forward

Nadir Backward

CCD3CCD4

CCD5

CCD1CCD2

CCD3CCD4

CCD5CCD6

CCD3CCD4

CCD5CCD6

CCD7CCD8

CCD2CCD1

ALOS /Focal Point

CCD1

CCD2Pixel No.1Pixel No.1

FlightDirection

CCD1CCD2

PRISM sensor alignments on optical bench (side view)

FWD

BWD

NDR

STT

Optical BenchRoll

Yaw

Pitch


Software overview (cont.) Area-based grid matching with cross-correlations Simultaneous triplet image matching (F-N & B-N)

FWD(F) NDR(N) BWD(B)

DSMs generated with different stereo pairs

F-N & B-N B-NF-N F-B F-N & B-N & F-B


Software overview (cont.) Why 10m grid spacing?

PRISM-DSMs generated in different grid spacing

100m

0m

Height ScalePRISM nadir image2.5m

20m 10m

5m 2.5m


Being used at JAXA, RESTEC, and former ALOS-Data-Node agencies (ASF/America, GSTDA/Thailand, and GA/Australia)

Accuracies from our past validations with LiDAR reference

Software overview (cont.)

-75 83 2.88 2.82 0.6043669079Kenai-W*2Flat & Mountain

-2371648.75 8.70 0.94 23465889PSLCVegetation

-73325.21 4.92 -1.70 1287801TsukubaMountainous

-178835.09 4.88 1.47 1505512Saitama*1Flat & Urban

Kenai-E*2

Fukuoka*1

Site

-102 112 5.96 5.69 -1.8030327521Steep

-2101177.39 7.05 2.20 1178376Various

Min [m]

Max [m]

RMSE [m]

SD [m]

Bias [m]

PointsTerrain

*1) provided by Geospatial Information Authority of Japan (GSI) *2) provided by Alaska Satellite Facility (ASF)


Relative-DSM

Absolute-DSM

Data archiving status Scene-DSM archives

– Absolute-DSM (GCP used) : >1300 scenes– Relative-DSM (GCP not used) : >4200 scenes (as of June 2011)

Locations of scene-PRISM-DSMs processed in JAXA/EORC as of June 2011


Stacking scene-DSMs in Japan (as of Mar. 2011)

Path ： 57~92　 (pt. +/-1.2 for each)Frame ： 2680~3115

1016 scene-frames for covering whole Japan

No. of stack scenes for each frame

34.13473~4

Color

13.71395~9

52.05281~2

0.220

%No. of frames

No. of stacks

Data archiving status (cont.)

90Path

80 70 60

2700

2800

2900

3000

3100

Fra

me


Mosaic processing

Scene-DSMin 0.3 arc-sec(path-frame) ...

Boundary smoothing among 1 ﾟ x1 ﾟ tiles and interpolating data on land-water/sea masks

Mosaic-DSM on 1 ﾟ x1 ﾟtile

Mosaicking on 1 ﾟ x1 ﾟtile

DSM mosaic for global data set

Stack-DSM

QC/QA and stacking on same path-frame


DSM data stacking

Scene-DSMs of different dates

DSM mosaic for global data set (cont.)

Cloud masksNoise in low correlation areas

Averaged only with height bias adjustmentsFilling holes of cloud masksNoise reductionStack-DSM


DSM data stacking


-2101177.39 7.05 2.20 1FukuokaVarious

-96364.374.17-1.305

-189634.394.141.444

3

1

1

Stack scenes

-73325.21 4.92 -1.70 TsukubaMountainous

-178835.09 4.88 1.47 SaitamaFlat & Urban

Site

-206896.63 6.40 1.73

Min [m]

Max [m]

RMSE [m]

SD [m]

Bias [m]

Terrain

Accuracy comparison between scene-DSM and stack-DSMDifference from LiDAR reference


Mosaicking on 1 ﾟ x1 ﾟ tiles– Offset-bias errors in relative-DSMs:

• X : 5m RMS• Y : 4m RMS• Z : 14m RMS

– Only Z offset-bias of relative-DSMs were corrected with absolute reference (e.g., SRTM) or the average of height differences on an overlap area of a neighboring absolute DSM.

Negligible for 10m grid spacing


Absolute-DSMRelative-DSMs

Z-Offset-bias correction


1 ﾟ x1 ﾟ tiles in Japan (101 tiles)


145140135130125

45

40

35

30

25


1 ﾟ x1 ﾟ tiles mosaic in Japan– 73 tiles of all 101 tiles– Offset bias of relative-DSMs were

corrected with absolute-DSMs or heights on coast lines


Gray scale image of mosaicked DSM(Preliminary result)

>2000m

0m

Cloud / Snow

Land-WaterSea

Mask:

Height scale

(Preliminary)



Number of Data Stack

1 ﾟ x1 ﾟ tiles mosaic in Japan– Total 2265 scene-DSMs

• 647 absolute-DSMs• 1618 relative-DSMs

– Max. 9 scenes stack– Max. 23 data stack

>10

0Data stack

(Preliminary)


Height accuracies on GCPs


GCP distribution(3236 GCPs)

0

200

400

600

800

1000

1200

1400

1600

0 200 400 600 800 1000 1200 1400 1600

GCP height [m]

PRIS

M/D

SM h

eigh

t [m

]

0

20

40

60

80

100

120

140

- 10 - 8 - 6 - 4 - 2 0 2 4 6 8 10

He igh t Erro r (PRISM/DSM minu s GCP) [m]

Cou

nt

-20.97 36.70 3.36 3.33 -0.47

MinMaxRMSσµ

GCP heights vs. PRISM-DSM heights(GCP height range: 11m~1567m)

Height difference statistics (in meters)

Height difference histogram


Height difference from existing global terrain data– GSI50m-DEM: 50m-DEM derived from contour lines of 1/25000 map– SRTM-3: 90m-DSM derived from C-band IfSAR


+30m

-30m

+/-0m

+30m

-30m

+/-0m

Height difference from SRTM-3for 3x3 tiles in Kanto-region

Height difference from GSI50m-DEM

for 3x3 tiles in Kanto-region


Visual comparison (6km x 6km)


PRISM-DSM(10m)

Digital Map 10m Grid (Elevation)

by GSI

ASTER-GDEM(30m)

ASTER GDEM is a product of METI and NASA

SRTM-3(90m)

Derived from contour lines of 1/25000 map

0m

200m




PRISM-DSM(10m)

Digital Map 10m Grid

(Elevation)by GSI

Kyushu Univ. / Ito campusBuilt in 2009


Contribution to the study on GLOF in Bhutan Himalayas– GLOF (Glacial Lake Outburst Flood) project supported by JICA/JST

• To extract dangerous glacial lakes which have a possibility of GLOF.• To make an inventory of expanding history for glacial lakes.• To make a flood hazard map.


WEP.M.6 “DEVELOPMENT OF GLACIAL LAKE INVENTORY IN BHUTAN USING “DAICHI” (ALOS)”By Takeo Tadono, et. al.

TH1.T04.4“TEMPORAL MONITORING OF SUPRAGLACIAL LAKES ON TSHOJO GLACIER AT BHUTAN”By Tsutomu Yamanokuchi, et. al.

Related presentations in IGARSS2011

UtilizingPRISM-DSM

Glacial lakes in Bhutan on AVNIR-2 3D-view image with PRISM-DSM



N26

N27

N28

N29

E88 E89 E90 E91 E92 E93

PRISM-DSM mosaic tiles over Bhutan area

N026E088 N026E089 N026E090 N026E091 N026E092

N027E088 N027E089 N027E090 N027E091 N027E092

N028E089 N028E090 N028E091

Contribution to a study on GLOF in Bhutan Himalayas



AVNIR-2 image mosaic over Bhutan





SRTM-3 (version 2)

Void data in SRTM-3

7000m

0m

Heightscale




7000m

0m

Heightscale

PRISM-DSM mosaic (0.3 arc-sec)

N026E088 N026E089 N026E090 N026E091 N026E092

N027E088 N027E089 N027E090 N027E091 N027E092

N028E089 N028E090 N028E091




Number of data stack (max 11 data in 160 scenes mosaic)

>5

0

DatacountOffset-bias of relative-DSMs were estimated

from the mode of difference from SRTM-3




+30m

-30m

PRISM-DSM minus SRTM-3

+/-0m

Void data in SRTM-3




PRISM-DSM(0.3 arc-sec / 10m)

SRTM-3(3.0 arc-sec / 90m)

voids

mask





PRISM-DSM(0.3 arc-sec / 10m)

SRTM-3(3.0 arc-sec / 90m)

voids



Summary

Processed thousands of PRISM-scene-DSMs in 10m-grid so far. Tested over 70 1 ﾟ x1 ﾟ tiles of their seamless mosaic in Japan

area. Processed 13 1 ﾟ x1 ﾟ tiles of their seamless mosaic in very

steep Bhutan Himalayas and provided them to the GLOF project. Planning to complete a global DSM mosaic for the whole areas

of Japan, first.

PRISM-scene-DSM is provided to general users by RESTEC as a value-added (commercial) product on order. (

http://www.alos-restec.jp/en/)

http://www.alos-restec.jp/en/

http://www.alos-restec.jp/en/

igarss11_takaku_dsm_report.ppt

Technology

s56 prism data areas

stereo triplets

matching software grid

prism ccd unit alignmentsnadir

dogsap dsm

areabased grid matching

satellite orbit

pan images