FR2.L09 - PROCESSING AND ANALYSIS OF AIRBORNE SYNTHETIC APERTURE RADAR IMAGERY ACQUIRED OVER MAYA...

PROCESSING AND ANALYSIS OF AIRBORNE SYNTHETIC APERTURE RADAR IMAGERY

ACQUIRED OVER MAYA SETTLEMENTS IN CENTRAL AMERICA

Bruce Chapman1, Ronald Blom1, Thomas Garrison2, Scott Hensley1,Stephen Houston2,

Charles Golden3, Sassan Saatchi1

1Jet Propulsion Laboratory, California Institute of Technology2 Brown University3Brandeis University

The Experiment In 2004, the NASA/JPL airborne synthetic

aperture radar (AIRSAR) conducted a month -long experimental campaign in Central and South America to study:– Forest structure (La Selva, Costa Rica)– Archaeology (Central America)– Glaciology (Patagonia and Antarctica)– Hydrology (Central America)– Small scale miscellaneous studies

In 2004, the NASA/JPL airborne synthetic aperture radar (AIRSAR) conducted a month -long experimental campaign in Central and South America to study:– Forest structure (La Selva, Costa Rica)– Archaeology (Central America)– Glaciology (Patagonia and Antarctica)– Hydrology (Central America)– Small scale miscellaneous studies

The Mayan Biosphere

• The Maya Biosphere Reserve (MBR) encompasses a series of national parks in Guatemala, Mexico, and Belize.

• Thousands of archaeological sites• Heartland of ancient Maya civilization• World famous sites

– Tikal (a UNESCO world heritage site)– El Mirador– Piedras Negras– many others

• Many more undocumented sites must also exist in the region

• The Maya Biosphere Reserve (MBR) encompasses a series of national parks in Guatemala, Mexico, and Belize.

• Thousands of archaeological sites• Heartland of ancient Maya civilization• World famous sites

– Tikal (a UNESCO world heritage site)– El Mirador– Piedras Negras– many others

• Many more undocumented sites must also exist in the region

Fall 2008 AGU

The Mayan Biosphere

• Only a small portion of this landscape has been subjected to rigorous scientific study

• The greatest impediment to such research in the MBR is the difficulty of conducting ground-based reconnaissance and mapping in dense tropical forest.

• Goal of this experiment:

The development of protocols for archaeological analysis of the AIRSAR radar data

Attempt to develop a coherent and complete understanding of this ancient landscape

• Only a small portion of this landscape has been subjected to rigorous scientific study

• The greatest impediment to such research in the MBR is the difficulty of conducting ground-based reconnaissance and mapping in dense tropical forest.

• Goal of this experiment:

The development of protocols for archaeological analysis of the AIRSAR radar data

Attempt to develop a coherent and complete understanding of this ancient landscape

Fall 2008 AGU

The Mayan Biosphere

Specific tasks:• Process Interferometric SAR data from AIRSAR to obtain a very

high resolution DEM• Process the simultaneously acquired (and geographically

coincident) longer-wavelength polarimetric data to reveal the scattering mechanisms at work– Sensitive to forest structure– Man-made features (such as walls, roads, etc)– Can penetrate the forest canopy to some extent

• Compare with archaeological data obtained through field work. • Archaeological analysis of the resulting SAR data and DEM

Specific tasks:• Process Interferometric SAR data from AIRSAR to obtain a very

high resolution DEM• Process the simultaneously acquired (and geographically

coincident) longer-wavelength polarimetric data to reveal the scattering mechanisms at work– Sensitive to forest structure– Man-made features (such as walls, roads, etc)– Can penetrate the forest canopy to some extent

• Compare with archaeological data obtained through field work. • Archaeological analysis of the resulting SAR data and DEM

Single Pass InSAR to measure topography

Rosen et. al. , 2000

Two SAR images are processed from data simultaneously acquired by two antennas vertically offset from each other on the aircraft.

These two images have slightly different path lengths () to the same image pixels on the ground.

Each image pixel has an amplitude and phase. The phase is a function of the number of radar wavelengthsto that image pixel.

We calculate the product of these images to form the interferogram. From the interferogram, we may estimate the ground elevation (must be calibrated).

Two SAR images are processed from data simultaneously acquired by two antennas vertically offset from each other on the aircraft.

These two images have slightly different path lengths () to the same image pixels on the ground.

Each image pixel has an amplitude and phase. The phase is a function of the number of radar wavelengthsto that image pixel.

We calculate the product of these images to form the interferogram. From the interferogram, we may estimate the ground elevation (must be calibrated).

The accuracy of the estimated elevation is a function of the distance between the two antennas, the wavelength of the radar, aircraft motion knowledge, etc.The accuracy of the estimated elevation is a function of the distance between the two antennas, the wavelength of the radar, aircraft motion knowledge, etc.

DEM from InSAR• This technique was successfully used by the NASA Space

Shuttle SRTM mission in February 2000 to measure near-global ground elevation (between ± 60 deg. Latitude) at 30m resolution– C-band Interferometric SAR– Only 90 meter data easily available outside U.S.– 2m vertical accuracy– May be downloaded at no charge from USGS

• This resolution is insufficient for many applications, but provides an excellent reference point.– Especially as compared to C-band AIRSAR data

• This technique was successfully used by the NASA Space Shuttle SRTM mission in February 2000 to measure near-global ground elevation (between ± 60 deg. Latitude) at 30m resolution– C-band Interferometric SAR– Only 90 meter data easily available outside U.S.– 2m vertical accuracy– May be downloaded at no charge from USGS

• This resolution is insufficient for many applications, but provides an excellent reference point.– Especially as compared to C-band AIRSAR data

AIRSAR• 1988-2004 R.I.P.• Single-pass SAR

Interferometry used to measure topography

• DC-8 aircraft (operated by NASA/Dryden) flew at 30,000 ft at 450 kts

• Image ~200 hectares per minute

• Height accuracy ~2m at 5m x 5m postings

• Swath: 12km by length of flight

• 1988-2004 R.I.P.• Single-pass SAR

Interferometry used to measure topography

• DC-8 aircraft (operated by NASA/Dryden) flew at 30,000 ft at 450 kts

• Image ~200 hectares per minute

• Height accuracy ~2m at 5m x 5m postings

• Swath: 12km by length of flight

Observation Plan March 2004for archaeology

In addition,data were acquired in the Yucatan,Costa Rica, and Honduras

In addition,data were acquired in the Yucatan,Costa Rica, and Honduras

MexicoMexico

GuatamalaGuatamala

BelizeBelizeRequired about 15 hours of flight time to acquire data over all the archaeology sites

Required about 15 hours of flight time to acquire data over all the archaeology sites

Over 25,000 hectares imagedOver 25,000 hectares imaged

AIRSAR DC-8 in San Jose, Costa Rica (March 2004) DC-8 in San Jose, Costa Rica (March 2004)

Tikal from the window of the DC-8

Data Processing• The SAR data was processed using software developed

primarily by Scott Hensley of JPL• These same software tools were used in producing the

topographic maps of Angkor Wat in Cambodia (also using AIRSAR data)

• The topography was estimated from the single-pass C-band InSAR data.

• Polarimetric L-band and P-band SAR data were simultaneously acquired – These data may be directly related to forest and other

structural parameters

• The SAR data was processed using software developed primarily by Scott Hensley of JPL

• These same software tools were used in producing the topographic maps of Angkor Wat in Cambodia (also using AIRSAR data)

• The topography was estimated from the single-pass C-band InSAR data.

• Polarimetric L-band and P-band SAR data were simultaneously acquired – These data may be directly related to forest and other

structural parameters

DEM Errors and their Correction

• The initial estimates of topography suffered from residual ‘multipath’ and/or ‘switch leakage’ errors.

• These errors manifest themselves in the data as along-track height ripples in the topography. The magnitude of these ripples can be as large as 10m.

• It is caused by multiple reflections by the radar waves off the aircraft prior to reception, and/or switch leakage in the radar.

• The error is a function of the line-of-site distance from the radar to the pixel on the ground.

• The initial estimates of topography suffered from residual ‘multipath’ and/or ‘switch leakage’ errors.

• These errors manifest themselves in the data as along-track height ripples in the topography. The magnitude of these ripples can be as large as 10m.

• It is caused by multiple reflections by the radar waves off the aircraft prior to reception, and/or switch leakage in the radar.

• The error is a function of the line-of-site distance from the radar to the pixel on the ground.

Multipath error in DEM

The magnitude of these ripples (peak to trough) is about 5m

The magnitude of these ripples (peak to trough) is about 5m

20 m color wrap

Example

Multipath Correction• Calculate and average the height differences between the 90 m SRTM DEM with the AIRSAR measured heights

at the same line-of-site distance (absolute phase)• These height differences are then multiplied by the derivative of phase as a function of height to calculate a

phase offset. • The offsets are then fit as a sum of 85 Chebyshev polynomials whose coefficients are written out as the actual

phase screen file.

• The absolute phase (the line-of-site distance) is then corrected by this phase screen during reprocessing of the data

From ‘GeoSAR Calibration Documentation’ by Elaine Chapin, Scott Hensley, Delwyn K. Moller, Paul R. Siqueira, GeoSAR JPL Document 226-799, March 15, 2006

• Calculate and average the height differences between the 90 m SRTM DEM with the AIRSAR measured heights at the same line-of-site distance (absolute phase)

• These height differences are then multiplied by the derivative of phase as a function of height to calculate a phase offset.

• The offsets are then fit as a sum of 85 Chebyshev polynomials whose coefficients are written out as the actual phase screen file.

• The absolute phase (the line-of-site distance) is then corrected by this phase screen during reprocessing of the data

From ‘GeoSAR Calibration Documentation’ by Elaine Chapin, Scott Hensley, Delwyn K. Moller, Paul R. Siqueira, GeoSAR JPL Document 226-799, March 15, 2006

Before correction After correction

20 m color wrap

Mosaicking and projection

• After the image strip is processed and the DEM is obtained, the image must be mosaicked with any other image strips and re-projected to a standard projection.

• In the overlap regions, currently we are selecting the DEM from one of the image strips rather than averaging

• At the boundary of the image strips, we sometimes have discontinuities that will have to be minimized so as not to be confused with surface topography features

• The output projection is a simple equi-angular geographic projection, corresponding to 5 meters by 5 meters.

• After the image strip is processed and the DEM is obtained, the image must be mosaicked with any other image strips and re-projected to a standard projection.

• In the overlap regions, currently we are selecting the DEM from one of the image strips rather than averaging

• At the boundary of the image strips, we sometimes have discontinuities that will have to be minimized so as not to be confused with surface topography features

• The output projection is a simple equi-angular geographic projection, corresponding to 5 meters by 5 meters.

Results

• Overall features of DEM• Statistics• Topography changes between SRTM 2000 and

AIRSAR 2004

• Overall features of DEM• Statistics• Topography changes between SRTM 2000 and

AIRSAR 2004

AIRSAR DEM• 5 x 5 m horizontal

Resolution• Height accuracy

~2m• 100 m color wrap• 13 image swaths• One image strip is

still being processed

• 5 x 5 m horizontal Resolution

• Height accuracy ~2m

• 100 m color wrap• 13 image swaths• One image strip is

still being processed

Fall 2008 AGU

135

km

165 km

AIRSAR DEM – SRTM DEM

• Average height difference 0.45 m

• Some AIRSAR DEM artifacts visible

– Residual multipath

– Motion errors on two flight lines

– Discontinuities between image strips

– Some ‘voids’

• Some larger scale features may be SRTM artifacts

• Average height difference 0.45 m

• Some AIRSAR DEM artifacts visible

– Residual multipath

– Motion errors on two flight lines

– Discontinuities between image strips

– Some ‘voids’

• Some larger scale features may be SRTM artifacts

Fall 2008 AGU

25 m color wrap corresponding to height difference25 m color wrap corresponding to height difference

+5 to +10 m+5 to +10 m

Pixels south (5 m each)

Hei

ght

Diff

eren

ce (A

IRSA

R –

SRTM

) (m

eter

s)

Mean calculated for each row of image (over 100 km E/W)

Agricultural Change 2000-2004

Purple – areas cleared since 2000 (-10 m)

Yellow – areas of regrowth since 2000 (+5 m)?

- could be a resolution effect

Orange – minimal change in height since 2000

• AIRSAR C-band image is overlayed on top of height difference as brightness

confirms low vegetation where height difference < 0m

Confirms forest where height difference is > 0m

Purple – areas cleared since 2000 (-10 m)

Yellow – areas of regrowth since 2000 (+5 m)?

- could be a resolution effect

Orange – minimal change in height since 2000

• AIRSAR C-band image is overlayed on top of height difference as brightness

confirms low vegetation where height difference < 0m

Confirms forest where height difference is > 0m Fall 2008 AGU

AIRSAR – SRTM height (40 m color wrap)AIRSAR – SRTM height (40 m color wrap) Full res image

Resolution effectsAIRSAR DEM (5x5m) SRTM DEM (90x90m) AIRSAR – SRTM DEM

50 m color wrap

+19m

-13m

Likely manmade featuresLikely manmade features Full res images

Image Features• Some landscape features are visible in the SAR image, but not clearly in the DEM• Some landscape features are visible in the SAR image, but not clearly in the DEM

This line in the image extends for tens of km in an extremely straight line. It apparently marks the Mexico/Guatamala border.This line in the image extends for tens of km in an extremely straight line. It apparently marks the Mexico/Guatamala border.

20 m color wrap

El Mirador

Google Earth imageGoogle Earth image

AIRSAR image and DEMAIRSAR image and DEM

20 m color wrap

Tikal

Airsar Image and DEMAirsar Image and DEM

Hand held photo (from Panaramio website)Hand held photo (from Panaramio website)Google Earth

Landsat imageGoogle EarthLandsat image

20 m color wrap

Buena Vista ValleyP-band image/AIRSAR DEM

A : Hilltop siteB : El ZotzC: La Avispa

Garrison et al, submitted to JAS, 2010.

Buena Vista Valley AIRSAR DEM

Fall 2008 AGUGarrison et al, submitted to JAS, 2010.

La AvispaSite map overlaid on DEM

Garrison et al, submitted to JAS, 2010.

UAVSAR 2010 deployment

• UAVSAR, during a transit to Costa Rica, acquired data over several archaeological sites.– Aquateca– Copan– Buena Vista Valley

• A repeat pass insar experiment

Aquateca

Copan

Buena Vista Valley UAVSAR L-band image

Fall 2008 AGU

Buena Vista ValleyUAVSAR L-band imagery

Data Availability

• This data will be made available through the SERVIR initiative, supported by NASA and USAID to improve response to natural disasters (October, 2010)

www.servir.net

Acknowledgements

This paper was partially written at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

We thank Craig Dobson, program manager of the Space Archaeology program at NASA HQ, for funding this work.

This paper was partially written at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

We thank Craig Dobson, program manager of the Space Archaeology program at NASA HQ, for funding this work.