Embed Size (px)
Citation preview
Global Wetlands Mapping Using ALOS PALSAR data
Bruce Chapman1, Kyle McDonald1, Laura Hess2
1Jet Propulsion Laboratory, California Institute of Technology2University of California, Santa Barbara
An Inundated Wetlands Earth System Data Record: Global Monitoring of Wetland Extent and Dynamics
Principal Investigator: Kyle McDonald (JPL/Caltech)
Project Members: Mahta Moghaddam (The University of Michigan)Bruce Chapman (JPL/Caltech)Laura Hess (University of California, Santa Barbara)John Kimball (University of Montana)Elaine Matthews (NASA/GISS)Collaborators:Wenjun Chen (Canadian Centre for Remote Sensing)Catherine Prigent (LERMA - France) Ake Rosenqvist (JRC- Italy)Masanobu Shimada (EORC-JAXA – Japan)Nick Davidson (Ramsar)Max Finlayson
(International Water Management Institute)Martti Hallikainen
(Helsinki University of Technology)
I. Regional inundated wetlands data sets from Synthetic Aperture Radar (SAR)
- Spatial coverage: Major global wetland regions, 100m resolution
- Temporal coverage: 1-2 year time series at 17-to-46 day intervals during 2006-2009 † ‡
- Retrospective 1990’s-era from archived JERS data covering Alaska, Canada, Amazon
Wetland extent (maximum inundatable area, including water bodies).
2. Wetland vegetation type (Non-vegetated, Herbaceous, Shrub, Woodland, Forest).
3. Inundation state (Flooded, Non-flooded; 17-46 day intervals) ‡
4. Annual inundation duration
II. Global monthly inundation data sets derived from multiple satellite data sources
- Spatial coverage: Global, 25 km resolution
- Temporal coverage: Monthly monitoring with annual summaries, 1992-2009 †
1. Globally gridded (0.25º) monthly inundated area fraction
2. Globally gridded (0.25º) annual inundation duration
Components of the Inundated Wetlands Earth System Data Record
† The domain of the 25-km and 100-m data sets excludes permanently frozen regions and seasonally frozen landscapes during the frozen season, although data from frozen seasons is used to improve classification accuracy.
‡ PALSAR ScanSAR mode has 46-day exact repeat orbit with 17-day sub-cycles.
Alaska & CanadaBOREAS, Quebec, SE U.S.Northern S AmericaPantanal & La PlataCongo Basin, Sudd Niger Delta & Senegal Okavango, ZambeziCentral Siberia & Finland Pripet-Biebrza marshesLena delta; Tibetan plateauEast India & Sri LankaSoutheast AsiaTigris marshesMalyasia, Indonesia, PNGAlligator RiverMurray-Darling River
PALSAR cycle Year Month
ALOS Wetlands ScanSAR Acquisitions (Descending swaths)7 8 9 10 11 12 13 14 15 16 22 23 2417 18 19 20 29 30
2007 2008 200925 26 27 2821
2006O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A S OJM J A
PALSAR Regional Coverage: ScanSAR regions
•Inundated wetland area (swath-by-swath)•Principal wetland vegetation classes (non-vegetated, herbaceous, shrub, woodland, forest), •Seasonally based summary products describing timing and extent of wetland inundation•Production is phased according to K&C acquisitions
Decision-tree rules generation
Polygon classification
Training DatasetsSegmentation
ALOS 50mHH-HV mosaic
1-2 dates
SRTM DEM & Water Body
Urban & Roads
Hydro SHEDS
ALOS 100mScanSAR swaths
8+ dates
Segmentation
Decision-tree rules generation
Polygon classification
Multi-date inundation
Monthly-mean inundation extent
Non-wetland Woody
Wetland extent Wetland type
Segmentation
Decision-Tree Rules Generation
Classification
Multi-date inundation
Wetland extent
Wetland type
Backscatter Image
Regional training data
Ancillary layers
Algorithm for classification of L-band SAR imagery into 100-meter resolution inundated wetlands data sets. The SAR classification algorithm employs object-oriented image segmentation and statistically-based decision tree classification. This construct follows directly from our previous work on mapping of tropical and boreal wetlands with JERS SAR imagery.
Regional inundated wetlands data sets from Synthetic Aperture Radar (SAR): Algorithms
Baseline Algorithm
Enhanced Algorithm
Example Legacy Data Products from SAR
Flood duration for a sub-region of the Amazon Basin based on JERS–1 SAR.The colors indicate flooding duration over a 1 year period, or 8 cycles of coverage.
Rosenqvist et al, 2002
Example Legacy Data Products from SAR
Image mosaics were assembled under the Global Boreal Forest and Rain Forest Mapping Projects. The JERS SAR constitutes a pathfinder mission for application of satellite-borne SAR for mapping global wetlands. A wetlands area map is shown for the Amazon Basin, developed directly from the JERS SAR data. We will employ algorithms and techniques established utilizing the JERS SAR data sets in constructing our global ESDR.
Richey et al, 2002
Example Legacy Data Products from SAR
A 100-meter resolution wetlands map of Alaska has been developed using JERS-1 SAR imagery. A statistically based decision tree classification scheme was applied using multiple data layers to develop the regional-scale map. Data layers employed include maps of wetlands ground reference data, DEM, JERS SAR-based open water maps, SAR acquisition date, and image texture. Whitcomb, Moghaddam, McDonald, Kellndorfer, and Podest, 2008 (in review)
Whitcomb et al, submitted 2008
ALOS Image Mosaics ‐ North/South America• Dual Polarization mosaics
– HH, HV, Ratio
– Topography, Date, Incidence Angle
– Inundation state (derived product)
• ScanSAR mosaics– HH
– Topography, Date, Incidence Angle
– Inundation state (derived product)
• Output projection– Variable by choice (EQA, UTM, Albers equal area, etc.)
– ~1 arcsecond (30 m) pixel spacing
• Format– Raw (unsigned 16 bit), Geotiff, …
Image Mosaics• Method for ortho‐rectification
– Using software from Gamma Remote Sensing
– Using SRTM DEM data interpolated to 1 arcsec (~30x30 meters)
• Using CGIAR SRTM gap‐filled data
– Careful – shifted 0.5 3‐arcsec pixels relative to Standard SRTM DEM
– Have to ‘pad’ DEM for Gamma software
• http://srtm.csi.cgiar.org/
• Stitch SRTM‐like tiles together to make continental scale image mosaics
• But – we will keep all the data, so that the image mosaics may be custom built like a jigsaw puzzle based on science objectives
– This is especially important for the SCANSAR coverage areas, and for filling in gaps
– Have to accommodate overlap and zero fill
Image Strip Calibration• There are radiometric calibration errors in the KC image strip products that need to be corrected– Images fall off in brightness in the near and far range
• Some trends are due to target‐dependent incidence angle variations in brightness
• If there was more overlap between strips, might not be an issue.
Cross track calibration
RSP249HH
Average over entire 1500 km long track
All data± 1 σ of mean value at each range bin
Range bin Range bin
Step 1: for selected segment of image strip, find mean and standard deviation for each range binStep 2: recalculate mean excluding data outside 1 standard deviation of meanStep 3: smoothed version of range dependence applied to entire image strip (0 dB overall gain)
‐5
‐10
‐15
‐20
Cross track calibration
RSP249HV
Average over entire 1500 km long track
All data± 1 σ of mean value at each range bin
Range bin Range bin
‐5
‐10
‐15
‐20
The spatial extent of the ‘fall off’ in the near and far range appears to slowly vary along track.
Applying a single correction for an entire image strip does not entirely eliminate the problem
DEM ~1 arcsecDays since Launch (583 in this case)
Incidence Angle Pixel Area
RSP79 RSP78 RSP77 RSP76
Brightness : HH Backscatter
SRTM‐like tiles
Scansar and Fine Beam data can be easily interchanged
Must still accommodate zero fill and overlap areas
Facilitates multi‐temporal analysis (tile –by‐tile)
S04W062 S04W062 S04W061 S04W061 S04W061 S04W060
A ‘replacement’ tile can be easily inserted, if desired
For the casual user and for data discovery:Make zoomable image mosaics available for browsing on web
With links to other available image products
How do we make these image mosaics available for use?
(64°,‐163°) to (60°, ‐159°)HH – red, HV – green, HH/HV ‐ blue
http://uavsar.jpl.nasa.gov/zoomify/KC/N61W162_HHHVRT.html
For an intermediate user, perhaps unfamiliar with SAR imagery:
Enable geo‐referenced Google‐Earth compatible viewing, with links to image mosaic products
How do we make these image mosaics available for use?
Alaska
For the ALOS KC science team:
SRTM‐like tiles of HH, HV, HH/HV, Date, Incidence Angle, DEM, etc., in addition to mosaic products.
How do we make these image mosaics available for use?
These image products can be used to make thematic products (wetlands/non‐wetlands, forest/non‐forest, etc.)
These image products can be used to make thematic products (wetlands/non‐wetlands, forest/non‐forest, etc.)
Status• DEM ‐ assembled
• Ortho‐rectification – Gamma software
• Calibration – in progress
• Processing to mosaics – in progress
• Distribution – being developed
• South America dual‐pol data processed with old version of EORC processor and must be reprocessed.
Status• Global DEM (SRTM plus best available) at 1 arc sec resolution
• Processing flow developed to project each image strip to a series of SRTM‐like tiles
• Calibration software under refinement
• 30 HH/HV strips processed (each ~1,500km in length) over North America, over 1,000 SRTM‐like tiles of image data.
Cabaliana, Brazil
Open WaterOpen Water
Flooded VegetationFlooded Vegetation
Low vegetationLow vegetation
ForestForest
Red HHGreen HVBlue HH/HV
Yukon River (Holy Cross), Alaska
ALOS SAR (HH‐red, HV‐green, HH/VV‐blue) Aerial photo
Lat: 62.2, Lon: ‐159.64, Summer 2007
scrubscrub
grasses grasses
BareBare
Current Coverage (May 2008)(dual‐pol)
Processed by JAXA EORC
Almost 1000 image strips
Conclusions
• ALOS global acquisition strategy will enable continental and regional scale mapping of inundated wetlands
• ALOS image quality exceeds JERS, both geometrically and radiometrically.
• Addition of HV channel will increase classification accuracy.
Acknowledgements
• This research was undertaken within the framework of the ALOS Kyoto & Carbon Initiative. The ALOS data were provided by JAXA EORC.
• This paper was partially written at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
