the future of scientific computing for insar geodetic … change due to subsidence, ... insar...

InSAR Scientific Computing

EarthScope/EarthCube Workshop, Arizona State University, Tempe, AZ, October 2012

- 1 - Copyright 2012 California Institute of Technology. Government sponsorship acknowledged.

The Future of Scientific Computing for InSAR Geodetic Imaging

Paul A. Rosen

Jet Propulsion Laboratory

California Institute of Technology

October 29, 2012




Outline

• Background – geodetic imaging measurements

• Geodetic imaging science, present and future

• Anticipated missions and data sets

• Implications for computational infrastructure and

capabilities




Differential Interferometry for Change

Two observations are made from different locations in space and at different times,

so the interfero-metric phase is proportional to topography and topographic change.

Df =4p

l(r(t1)- r(t2 )) =

4p

l(dLOS -Drtopo )+ noise

t1

t2

dLOS(t1 )

(t2 )

t1 t2

top o

B

Radar in space

First Pass, time t1

Second Pass, time t2

Assumes surface element remains geometrically and electrically similar

Df - differential phase

r - range

l - wavelength

B- baseline

t - time




Basic Geodetic & Imaging Products

Interferogram (Taiwan) Correlation (California) Simons & Rosen (2007)

Polarimetry (Africa)




Dense Sampling in Space and Time to Understand Earthquake Mechanisms

1 month Millions of “GPS-like” points in each image frame.

Frequent temporal snapshots will reveal new processes and improve models

~40 k

m

PBO Western US

permanent stations

1 month

Gomberg et al. (2010) doi:10.1130/B30287.1

Parkfield CARH GPS Station




Spatial Synoptic coverage

Lin et al. (2012) Submitted to JGR

2010 Mw 8.8 Maule (Chile)




Long Valley Deformation Time Series

Agram, Jolivet, Simons et al (2012 in prep)

• New methods for InSAR time series analysis are becoming mature, and are giving a new view of Earth processes

• These geodetic measurements can potentially support more complex geophysical models, provided: - Data are available - Data have suitable quality

MInTS analysis, ECMWF-corrected




Etna Deformation Time Series

Agram, Jolivet, Simons et al (2012 in prep)

With appropriate data, methods can be applied to understanding eruption dynamics on global scale

NSBAS analysis, ECMWF-corrected




Hydrology with InSAR Time Series

Credits: InSAR analysis – Zhen Liu, Tom Farr (JPL); Visualization – Vince Realmuto (JPL) http://photojournal.jpl.nasa.gov/catalog/PIA16293

http://photojournal.jpl.nasa.gov/catalog/PIA

http://photojournal.jpl.nasa.gov/catalog/PIA




Assessing Devastation with Remote Sensing: Christchurch, NZ

In Hours to Days – Radar Remote Sensing In 4 Months – Engineering Assessment

The Damage Proxy Map indicates significant centimeter

scale change due to subsidence, inundation, or structure

collapse allowing early assessment of the scale of

damage. [1]

Re-zoning map indicating where repair and rebuilding will

be allowed. This map is the product of many man-hours

of effort to assess structures and the land that they were

built on. [2]

Christchurch Cathedral

Canterbury TV Building




Landslide occurrences correlated to earthquake localities

Comprehensive landslide mapping through deformation can lead to better understanding of landslide mechanisms worldwide

Scheingross et al. (2012) Accepted to JGR




International SAR Missions “The Golden Age of SAR”

1980 1985 1990 1995 2000 2005 2010 2015

SeaSAT

SIR-A SIR-B SIR-C/

X-SAR

JERS-1

ERS-1 ERS-2

RADARSAT-1

Envisat-1

TerraSAR-X

SRTM

TanDEM-X

RADARSAT-2/RCM

Challenger

Cassini

Launch Orbit Launch Orbit

Magellan

Earth

Planetary

ALOS-1/2

InSAR?

Cosmo-Skymed

LRO

Chandryan

See export compliance restrictions on cover Mission Concept Review

Decadal Survey:

US L-band Mission Science

Ice sheets and sea level

Will there be catastrophic collapse of the major ice

sheets, including Greenland and West Antarctic and, if

so, how rapidly will this occur?

What will be the time patterns of sea level rise as a

result?

Changes in ecosystem structure and biomass

How does climate change affect the carbon cycle?

How does land use affect the carbon cycle and biodiversity?

What are the effects of disturbance on productivity, carbon, and other ecosystem functions and services?

What are the management opportunities for minimizing disruption in the carbon cycle?

Extreme events, including earthquakes and volcanic eruptions

Are major fault systems nearing release of stress via strong earthquakes?

Can we predict the future eruptions of volcanoes?

Recommended by the NRC Decadal Survey for near-term launch to address important scientific questions of high societal impact:

What drives the changes in ice masses and how does it relate to the climate?

How are Earth’s carbon cycle and ecosystems changing, and what are the consequences?

How do we manage the changing landscape caused by the massive release of energy of earthquakes and volcanoes?

Planned by NASA as one of the following 4 Decadal Survey TIER 1 Missions

SMAP

ICESat-II

DESDynI

CLARREO

Biomass Ice Dynamics Deformation 3 - 13




Coverage Area and Frequency

41,637,823 km2; Every Cycle, A/D; Single-Pol 87,061,332 km2; 2 per yr, A/D; Single-Pol

87,776,900 km2; 1 per seas., Quad-Pol

34,839,285 km2 ;5 per cycle; Single-Pol

Solid Earth

Solid Earth

Eco-systems

Ice




Average access latency to reimage any point on Earth with a single 8 day orbiting radar

The key is free, open and low latency access to raw data and the infrastructure to exploit it fully for societal and scientific benefit




User Communities for SAR Data

Wide-range of needs and capabilities

• Individual Scientists and Modelers

- Small areas, specific problems

• Science “Power Users”

- Large scale problems (interseismic, global carbon, pan-icesheet)

• Operational Agencies

- Ship tracking, water resource monitoring, frac-ing, oil-spills, levees

• Disaster/hazard response

- Earthquakes, volcanoes, fires, floods, etc.

Implies flexible and extensible algorithms and processing




Computational Complexity

PrepIgramStack.py

ProcessStack.py

SBASInvert.py(or)

NSBASInvert.py(or)

TimefnInvert.py

DatatoWavelet.py

MetadataCommon

MaskCoherence

Unwrapped IFGs

Weather models

GPS data

InvertWaveletCoeffs.py(or)

InvertWaveletCoeffs_folds.py

WavelettoData.py

SBAS chain

MInTS chain

Visualization

GIAnT overview

HDF5 file from PrepIgramStack

Inputs

data.xmlsbas.xml

(or) mints.xml

Empirical multiscale approach

Weather model approach

Network inversion of parameters

IFG corrections

PyAPS - every SAR acquisition

Lat+Lon+DEM from

PrepIgramStack

Automatic download of

weather model

DEM fromPrepIgramStack

Deramp each IFG

Network inversion of parameters

IFG corrections

Compare GPS to each IFG

Network or direct inversion of parameters

Output: HDF5 file

GPS data from SOPAC

or file

Atmospheric corrections

Orbital error correction

Flowchart for ProcessStack

Inputs

MetadataCommon

MaskCoherenceUnwrapped

IFGs

userfn.py data.xml

Mask + Crop

List of IFGs + Bperp

Filenames Parameters

Multilook

Common reference

Output: HDF5 file Output: Lat, Lon, DEM

Metadata includes DEM, Lat, Lon files in radar

coordinates

Flowchart for PrepIgramStack

Condition Data Condition Data

Form SLC 1 Form SLC 2

Resample Image #2&

Form Interferogram&

Estimate Correlation

Remove Topography

Filter & Look Down

Unwrap Phase

Geocode

Post-Process&

Model

RemoveModel

DEM

(Re)EstimateBaseline

GPS

IndependentData

EstimateTie Points

Orbits

ReturnModel




Automated Rapid Analysis Data System

! "#$%&' "() $*$(+, -*' . (

! / 0! 123 (4/5 6) 7(89: *: *, ; ' (+, -*' . (

/ 28! 0/ (4) <-$-*' 9-7((((((() 8= (4! >07(

((((((

?' #' @1A((/$B () $*$(

! @: %C1*9$&D() <-; @$&' . ' %*(

E 9F<*() $*$( 6<. ' (+' 9<' -(89: &(

G: %HCI : I -(8$<9-(

?<%' 1: J1-<CK*() <-; @$&' . ' %*(

) 8= (

/$"$9((+$*' @@<*' (

+' %-: 9-( 89: #<"' 9-L! 9&K<#' -(

2+! (42/+1MLNO(2%#<-$*O(+' %H%' @1M$LF7(

P! Q! (4! ?E +1MLN7((! +0(4GE += E 1+D, = ' "7(R! +! (4) 2+) , %07(

! /0! 123 (4P8?(/5 6) 7(

! /0! 1= S (4R! +! (! 0+67(= ! 0(4T: 9' $%(UI %"7(

) 8= (4R! +! (! > 07(

GI -*: . ' 9(89: "I &*-(

) $<@, (8: <%*(8: -V(

>8+(+*$H: %(

+I F1"$<@, (8: <%*(8: -V(

+E 8! G(WR! XGE (>+0(Y(E *K' 9-(

>8+((G: %-*' @@$H: %(

R! +! (1(P8?(

) $. $C' (89: Z, (= $; (

+' <-. : . ' *' 9(

= $--<#' (89: "I &H: %(: J(2+) / -(4= 2$+W/2-7(

G: K' 9' %&' (GK$%C' (

6<. ' (+' 9<' -(

! @: %C1*9$&D() <-; @$&' . ' %*(6<. ' (+' 9<' -(

?<%' 1: J1-<CK*() <-; @$&' . ' %*(6<. ' (+' 9<' -(

+! / (0. $C' -(

2#' %*-(

G: K' 9' %&' (

G: - ' <- . <&(0%*' 9J' 9: C9$. (

R20G(8! >2/ (

0%+! / (Y(>8+(0%*' C9$H: %(

0+G2(4! 0+67(

0+G2(4R! +! (! 0+67(

E +G! / (4R! +! (! 0+67(

U$I @*(= : " ' @(

+2+2+((

R+U(> ' : <%J: 9. $H&-(

2$9*K(+&<' %&' (X' %*I 9' (

! <9F: 9%' (= <--<: %-(

>8+(2Z; @: 9' 9(4R! +! (! 0+67(6+(2+/ ) -(4R! +! (= 2$+W/2-7(

= 0%6+(4G$. ; I -7(

89: [' &*-(: %(G$. ; I - ( / 28! 0/ (4R! +! () <-$-*' 9-7(

+2+2+(4R! +! (= 2$+W/2-7(

) ' ; ' %"' %&, (U@: B (

9G= 6(

2$9*K\I $D' (

X: @&$%: (

?$%"-@<"' (

U@: : "(

+I F-<"' %&' (

! / 0! 1= S (0%JI -<: %(

S XE ((X! ?X2(

69$%-<' %*() ' *' &H: %(

] ) (6<. ' (+' 9<' -(

! / 0! 1= S (4! 0+67((G@: I "1F$-' "(G: . ; I H%CO(+*: 9$C' O($%"() $*$(= $%$C' . ' %*(

! /0! 1= S (

! /0! 1= S (

! /0! 1= S (

! /0! (89: [' &*-(NAMNVA VA (

! /0! 123 (4/5 6) 7(<-(" ' #' @: ; <%C($@@(&: %%' &H: %-(<%(*K' (F@I ' (9' C<: %-O(B K<@' (! /0! 1= S (4! 0+67(B <@@("' #' @: ; ($"#$%&' "(JI %&H: %-(J: 9(. $*I 9<%C(*K' ("$*$(-, -*' . (&$; $F<@<H' -($%"(' %$F@<%C(&@: I "1F$-' "(K$_$9"(. : %<*: 9<%CO("$*$(; 9: &' --<%CO(. $%$C' . ' %*O($%"("<-*9<FI H: %V(

W+>+(R20G(>@: F$@(G= 6(0/0+(4 $#' J: 9. 7(

U$I @*(= : " ' @(

+2+2+((

E *K' 9(E ; ; : 9*I %<H' -(

?' #' @1A(/0R2Q() $*$(

E 9F<*() $*$(

) $<@, (

S : I 9@, (

M(S _(4> ) >8+7(

Y(E *K' 9-(

?' C' %"(

U<%<*' (U$I @*(0%#(4G$. ; I -7(

G: - ' <-. <&(>8+(X' &*: 9(= $; (

> 8+(6+(

0%+! / (6+(

>2(4! 0+67(((((((

69: ; : - ; K' 9<&(G: 99' &H: %(

` ' $*K' 9(

>8+(

E +G! / (4! 0+67(

((((((

2G= ` U(

= E ) 0+(` <@"a9' (




The Data Explosion

Gigabytes of International Civil SAR Data (cumulative)

-

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2025

Cu

mu

lati

ve

Da

ta V

olu

me

(G

B)

Year

Speculative




Sample Sizing for 350,000 ALOS scenes

Processing Steps Processing hours per scene

Processing hours for 350K scenes

Number of Nodes1

Common Processing Steps 0.20 70,000 9 LOS Surface Displacement Maps

2 1.33 466,667 62

AT Surface Displacement Map3 1.50 131,250 17

Atmospheric Correction 0.08 29,167 4

Surface Change Map2 0.08 17,500 2

LOS Surface Deformation Time Series 0.25 87,500 12 AT Surface Deformation Time series

2 0.25 21,875 3

Surface Change Time Series 0.02 5,833 1

Total 3.68 829,792 110

Product Temp Storage (per product)

Archived Volume

LOS Surface Deformation Maps 20 Gb 735 Tb

AT Surface Deformation Maps 20 Gb 735 Tb

Surface Change Maps 20 Gb 735 Tb

LOS Deformation Time Series ~35 – 480 Gb 110 Gb AT Deformation Time Series ~35 – 480 Gb 110 Gb

Surface Change Time Series ~35 – 480 Gb 110 Gb

350,000 scenes =250 Tb




US L-band SAR?

Added computational complexity: data isn’t always optimal for our purposes

2000 2010 2020

P

L

C

X

ERS-1/2 (ESA) ENVISAT (ESA)

Radarsat-1 (CSA) Radarsat-2 (CSA)

PalSAR (JAXA)

Tandem-X (DLR) TerraSAR-X (DLR)

Biomass (ESA)

S Chinese HJ-1-C (China)

Sentinel-1 (ESA)

Radarsat-C (CSA)

Indian RISAT (ISRO)

Cosmo/SKYMED (ASI)

PalSAR-2 (JAXA) SAOCOM (CONAE)

Cosmo/SKYMED-2 (ASI)

TerraSAR-X2 (ASI)

Defunct or Unavailable

Commercial, or Unavailable

Wav

ele

ngt

h B

and

Wrong Frequency and/or Commercial and/or unknown

UAVSAR/Airborne (difficult processing)




Software and Methods for Diverse Users

• Processing this sea of data will require modern methods

- Computational architecture that can address complex data analysis/processing and interoperability of products, processing and models in a heterogeneous web-based environment

- Seamless communication between users, data, and models in the field and elsewhere

- Software scalable from single users to large computational systems

Individual users

Cloud systems DAACs

• The next generation tools being built have the following attributes:

- Python-based, with consequent rich ready-made tool set

- Object-oriented - Interoperable - Provenance - Recipes for incorporating legacy code

• Likely basis of US SAR mission production

processor • Being tested by other large scale projects




The future

• A lot of data – many petabytes

• New homogeneous data sets suitable for global time series

- Historical and new heterogenous datasets from legacy systems, small-sats, and other international sensors.

• Mature Algorithms for time-varying geodetic imaging and geophysical exploitation thereof

• Desirable future architecture features:

• Loosely-coupled architecture to enable rapid assimilation of new technology advancements in different components as they develop

• Elastic compute resources for processing on demand

• Distributed data storage to facilitate low-latency data access across geographically dispersed compute nodes.

• Interoperable metadata models and encoding formats that are infused into tools, data systems, and used across different communities.




The future – cont’d

• More architectural features

• Federated data discovery and access enabling scalable handling of “big data”. InSAR data alone cannot be effectively handled by one data center.

• Data product preservation and stewardship enabling product provenance, transparency, and reproducibility.

• Visualization environments with sufficient network and processing bandwidth to enable innovation and discoveries not otherwise possible.

• Robustness and ease-of-use of geodetic imaging software given variable quality data




Synthetic Aperture Radar

6-26

Small antenna -

big beam

Big antenna -

small beam

Synthesized long antenna




Some Examples of Data Access Issues for US Investigators

• ALOS-2

• L-band, 14-day repeat: good for US science objectives

• Multi-mode, campaign mapping like ALOS-1 (as far as we know)

• Some commercial pressure

• Radarsat-2

• C-band, 24-day repeat: Limited science potential

• Fully commercial - $3K per image – 1M images = $3B

• Multi-mode

• Pre-existing orders limit data-taking flexibility

• Cosmo-Skymed

• X-band, 1- to 11-day repeat: often good for US science objectives

• Multi-mode, dual-use limits data-taking flexibility

• Undersized downlink capacity and distribution system (to date)




Polarization - A Measure of Surface Orientations and Properties

Vertical polarization

passes through

horizontally arranged

absorbers.

Polarization Filters Wave Polarization

Horizontal polarization

does not pass through

horizontally arranged

absorbers.

Mostly horizontal polarization

is reflected from a flat surface.

Colo

r figure

s fro

m w

ww

.colo

rado.e

du

/ph

ysic

s/2

000




Phase and Radar Interferometry

1l 2 l

2p 4p

0 1

2

Cycle number Collection of random path lengths jumbles the phase of the echo

The total phase is two-way range measured in wave cycles + random component from the surface

The phase of the radar signal is the number of cycles of oscillation that the wave executes between the radar and the surface and back again.




Radar Interferometry for Topography

• The two radar (SAR) antennas act as coherent sources

• When imaging a surface, the phase fronts from the two sources interfere

• The surface topography slices the interference pattern

• The measured phase differences record the topographic information

Df =4p

l(r(s1)- r(s2 )) =

4p

lDrtopo + noise




Radar Design to Meet Critical Requirements

Repeat Period requirement for Deformation science drives the Radar Swath

12M-day Repeat Period => 240/M-km Swath Width

Sensitivity requirement for Biomass (cross-pol) measurement drives Antenna

Size and Radar Power

Accuracy requirements for Deformation and Biomass drive Electronics &

Mechanical Stability and Calibration

A new SweepSAR technique was adopted as a means to achieve much wider

swath than conventional SAR strip-mapping, without the performance

sacrifices associated with the traditional ScanSAR technique

Conventional StripMap:

<~70km Swath

Conventional ScanSAR:

non-uniform along-track

sampling

Radar

antenna

Radar

antenna

Resulting ~40 day repeat

does NOT meet

proposed Deformation

and Ice Science

Requirements

Resulting degradation in

effective azimuth looks

does NOT meet proposed

Ecosystem Science

Requirements




New SweepSAR Technique to Meet Science Needs

• On Transmit, all Feed Array elements are

illuminated (maximum Transmit Power),

creating the wide elevation beam

• On Receive, the Feed Array element echo

signals are processed individually, taking

advantage of the full Reflector area

(maximum Antenna Gain)

Uses digital beamforming to provide wide

measurement swath

DBF allows multiple simultaneous echoes in

the swath to be resolved by angle of arrival

Uses large reflector to provide high aperture

gain

Full-size azimuth aperture for both transmit and

receive

Full-sized elevation aperture on receive

Only need data from feed array elements

being illuminated by an echoes

These elements can be predicted a priori




Jakobshavn Isbrae Highly Variable in Time and Space

Position of Ice Front

Despite spare spatial and sporadic temporal sampling, existing SAR data reveal large variations in glacier flow.

Science requires fine temporal sampling of all rapidly evolving outlet glaciers and ice streams.

~5%/yr velocity increase

Jakobshavn Isbrae is one of the few glaciers where frequent InSAR observations are available.

Per

iod

wh

ere

no

dat

a ac

qu

ired

3 -

33




Global Carbon Monitoring

3 -

34

• Global mapping of carbon and biomass dynamics at high spatial resolution and high accuracy

- Determines carbon stored in vegetation and the net effect of changes from fires and other disturbances and subsequent regrowth on concentrations of atmospheric CO2

- Supports climate treaty implementation by providing spatially explicit estimates of carbon stocks, accumulation in growing forests and losses to existing stocks

- Essential to reliable Carbon Monitoring System

• A US mission would then enable global carbon modeling at spatial scales commensurate with disturbances and environmental gradients

- Critical for initializing models that evaluate policy actions by predicting future land/atmosphere carbon dynamics

Snapshot of California carbon from available foreign data




How Much Data is Enough Data?

It depends on perspective:

• Science

- Problem of interest (Time and spatial scales, required accuracy, etc.)

- Region of Interest (Vegetation, Access, Cloud cover, etc.)

- Resources (available grants, computers, students)

• Policy/Program

• National priorities (science or society?)

• Science priorities (e.g. climate or hazards?, continuity or discovery?)

• Cost (of buying data vs. building a mission)

the future of scientific computing for insar geodetic … change due to subsidence, ... insar...

Documents