SAR Guidebook

sarmap, Sept ember 2008
Sy nt het i c Aper t ur e Radar

Sy nt het i c Aper t ur e Radar
and
and
The The SAR SAR- - Gui debook Gui debook
I nt r oduct i on I nt r oduct i on
The aim of t his t ut orial is t o int roduce beginners t o land applicat ions based on spaceborne
Synt het ic Apert ure Radar (SAR). I t is int ended t o give users a basic underst anding of SAR
t echnology, t he main st eps involved in t he processing of SAR dat a, and t he t ype of informat ion
t hat may be obt ained from SAR images.
Not e t hat t his t ut orial is based on an int roduct ory course developed by sarmap in collaborat ion
wit h t he UNESCO BI LKO group and financed by t he European Space Agency. Wit h respect t o
t he original one, it has been ext ended (t o Polarimet ry and Polarimet ric SAR I nt erferomet ry)
and adapt ed t o int roduce t he use of SARscape
.
Usi ng t he modul e nav i gat i on t ool s Usi ng t he modul e nav i gat i on t ool s
The navigat ion t ools at t he bot t om left of t he pages and t o t he t op right are int ended t o help
you move around bet ween slides as easily as possible.
Bot t om l ef t navi gat i on Bot t om l ef t navi gat i on
Takes you t o t he main cont ent slide; part icularly useful for navigat ing around t he
t heory sect ion of t he module
Takes you back t o t he slide you viewed previously, a useful way of ret urning t o
t he cont ent list s when you are t rying t o find a specific slide
Takes you t o t he previous slide in t he series
Takes you t o t he next slide in t he series
Top r i ght nav i gat i on Top r i ght nav i gat i on
Takes you t o t he slide indicat ed by t he adj acent ent ry in t he cont ent s list
Cr edi t s Cr edi t s
For t he preparat ion of t his Tut orial, document s of t he following inst it ut ions/ companies have been used:
Alaska SAR Facilit y
At lant is Scient ific I nc.
European Space Agency
I nfoSAR Limit ed
Japan Aerospace Explorat ion Agency
Radarsat I nt ernat ional
TeleRilevament o Europa
Universit y of I nnsbruck, I nst it ut e for Met eorology & Geophysics
Universit y of Not t ingham
Universit y of Pavia
Universit y of Trent o, Remot e Sensing Laborat ory
Universit y of Zurich, Remot e Sensing Laborat ory
US Geological Survey
Acr ony ms Acr ony ms
ASAR Advanced SAR
ASI Agenzia Spaziale I t aliana
DEM Digit al Elevat ion Model
DESCW Display Eart h remot e sensing Swat h Coverage for Windows
DI nSAR Different ial I nt erf eromet ric SAR
DLR Deut sche Luft und Raumfahrt
DORI S Doppler Orbit ography and Radioposit ioning I nt egrat ed by Sat ellit e
ENL Equivalent Number of Looks
EOLI Eart hnet On-Line I nt eract ive
ESA European Space Agency
GCP Ground Cont rol Point
I nSAR I nt erf eromet ric SAR
JAXA Japan Aerospace Explorat ion Agency
JPL Jet Propulsion Laborat ory
NASA Nat ional Aeronaut ics and Space Administ rat ion
PAF Processing and Archiving Facilit y
PDF Probabilit y Densit y Funct ion
Acr ony ms Acr ony ms
PolSAR Polarimet ric SAR
PolI nSAR Polarimet ric I nt erf eromet ric SAR
PRF Pulse Repet it ion Frequency
RADAR Radio Det ect ion And Ranging
RAR Real Apert ure Radar
SAR Synt het ic Apert ure Radar
SI R Shut t le I maging Radar
SLC Single Look Complex
SRTM Shut t le Radar Terrain Mission
UTM Universal Transf er Mercat or
WGS World Geodet ic Syst em
Sy mbol s Sy mbol s
A Amplit ude
|
o
Bet a Nought
c Speed of Light
|

Phase Difference
f
D
Doppler Frequency
I I nt ensit y
P
d
Received Power for Dist ribut ed Target s
P
t
Transmit t ed Power
P Power
L Number of Looks

Wavelengt h
o
o
Backscat t ering Coef ficient or Sigma Nought
u

I ncidence Angle
t

Pulse Durat ion
Tabl e of Cont ent s Tabl e of Cont ent s
1. What is Synt het ic Apert ure Radar (SAR)? 1. What is Synt het ic Apert ure Radar (SAR)?
2. How SAR product s are generat ed 2. How SAR product s are generat ed
3. Appropriat e land applicat ions 3. Appropriat e land applicat ions
4. Operat ional and fut ure 4. Operat ional and fut ure spaceborne spaceborne SAR sensors SAR sensors
5. Glossary 5. Glossary
6. References 6. References
1. What i s Sy nt het i c Aper t ur e Radar ( SAR) ? 1. What i s Sy nt het i c Aper t ur e Radar ( SAR) ?
1.1 The Syst em 1.1 The Syst em
1.2 Specific Paramet ers 1.2 Specific Paramet ers
1.3 Acquisit ion Modes 1.3 Acquisit ion Modes
1.4 Scat t ering Mechanisms 1.4 Scat t ering Mechanisms
1.5 Speckle 1.5 Speckle
1.6 Dat a St at ist ics 1.6 Dat a St at ist ics
1.7 Geomet ry 1.7 Geomet ry
1.1 The Sy st em 1.1 The Sy st em
Radar I maging
I maging radar is an act ive illuminat ion syst em. An ant enna, mount ed on a plat form, t ransmit s
a radar signal in a side-looking direct ion t owards t he Eart h's surface. The reflect ed signal,
known as t he echo, is backscat t ered from t he surface and received a fract ion of a second lat er
at t he same ant enna (monost at ic radar).
For coherent radar syst ems such as Synt het ic Apert ure Radar (SAR), t he amplit ude and t he
phase of t he received echo - which are used during t he focusing process t o const ruct t he
image - are recorded.
SAR versus ot her Eart h Observat ion I nst rument s
Li dar Opt i cal Mul t i -Spect r al SAR
airborne airborne/ spaceborne airborne/ spaceborne
own radiat ion reflect ed sunlight own radiat ion
infrared visible/ infrared microwave
single frequency mult i-frequency mult i-frequency
N.A. N.A. polarimet ric phase
N.A. N.A. int erferomet ric phase
day/ night day t ime day/ night
blocked by clouds blocked by clouds see t hrough clouds
Pl at f or m
Radi at i on
Spect r um
Fr equency
Pol ar i met r y
I nt er f er omet r y
Acqui si t i on t i me
Weat her
Real Apert ure Radar (RAR) - Principle
Apert ure means t he opening used t o
collect t he reflect ed energy t hat is used t o
form an image. I n t he case of radar
imaging t his is t he ant enna.
For RAR syst ems, only t he amplit ude of
each echo ret urn is measured and
processed.
Real Apert ure Radar - Resolut ion
The spat ial resolut ion of RAR is primarily det ermined by t he size of t he ant enna used: t he
larger t he ant enna, t he bet t er t he spat ial resolut ion. Ot her det ermining fact ors include t he
pulse durat ion (t) and t he ant enna beamwidt h.
Range resolut ion is defined as
where c is t he speed of light .
Azimut h resolut ion is defined as
where L is t he ant enna lengt h, R t he dist ance ant enna-obj ect , and

t he wavelengt h. For
syst ems where t he ant enna beamwidt h is cont rolled by t he physical lengt h of t he ant enna,
t ypical resolut ions are in t he order of several kilomet res.
L
R
res
azimuth
=
t
2
c
res
range
=
Synt het ic Apert ure Radar - Principle
SAR t akes advant age of t he Doppler hist ory of t he radar echoes generat ed by t he forward
mot ion of t he spacecraft t o synt hesise a large ant enna (see Figure). This allows high azimut h
resolut ion in t he result ing image despit e a physically small ant enna. As t he radar moves, a
pulse is t ransmit t ed at each posit ion. The ret urn echoes pass t hrough t he receiver and are
recorded in an echo st ore.
SAR requires a complex int egrat ed array of
onboard navigat ional and cont rol syst ems,
wit h locat ion accuracy provided by bot h
Doppler and inert ial navigat ion equipment .
For sensors such as ERS-1/ 2 SAR and
ENVI SAT ASAR, orbit ing about 900km from
t he Eart h, t he area on t he ground covered by
a single t ransmit t ed pulse (foot print ) is about
5 km long in t he along-t rack (azimut h)
direct ion.
synt het ic apert ure
Synt het ic Apert ure Radar - Range Resolut ion
The range resolut ion of a pulsed radar syst em is limit ed fundament ally by t he bandwidt h of
t he t ransmit t ed pulse. A wide bandwidt h can be achieved by a short durat ion pulse. However,
t he short er t he pulse, t he lower t he t ransmit t ed energy and t he poorer t he radiomet ric
resolut ion. To preserve t he radiomet ric resolut ion, SAR syst ems generat e a long pulse wit h a
linear frequency modulat ion (or chirp).
Aft er t he received signal has been compressed, t he range resolut ion is opt imised wit hout loss
of radiomet ric resolut ion.
Chirp, real part
Synt het ic Apert ure Radar - Azimut h Resolut ion
Compared t o RAR, SAR synt het ically increases t he ant enna's size t o increase t he azimut h
resolut ion t hough t he same pulse compression t echnique as adopt ed for range direct ion.
Synt het ic apert ure processing is a complicat ed dat a processing of received signals and phases
from moving t arget s wit h a small ant enna, t he effect of which is t o should be t heoret ically
convert t o t he effect of a large ant enna, t hat is a synt het ic apert ure lengt h, i.e. t he beam
widt h by range which a RAR of t he same lengt h, can proj ect in t he azimut h direct ion. The
result ing azimut h resolut ion is given by half of real apert ure radar as shown as follows:
- Real beam widt h |

=

/ D
- Real resolut ion AL = |

.
R = Ls (synt het ic apert ure lengt h)
- Synt het ic beam widt h |s =

/ 2
.
Ls = D / (2
.
R)
- Synt het ic resolut ion ALs = |s
.
R = D / 2
where

is t he wavelengt h, D t he radar apert ure, and R t he dist ance ant enna-obj ect (refer t o
t he Figure on t he next page).
This is t he reason why SAR has a high azimut h resolut ion wit h a small size of ant enna
regardless of t he slant range, or very high alt it ude of a sat ellit e.
Synt het ic Apert ure Radar - Azimut h Resolut ion
1.2 Speci f i c Par amet er s 1.2 Speci f i c Par amet er s
Wavelengt h
Radio waves are t hat part of t he elect romagnet ic spect rum t hat have wavelengt hs
considerably longer t han visible light , i.e. in t he cent imet re domain. Penet rat ion is t he key
fact or for t he select ion of t he wavelengt h: t he longer t he wavelengt h (short er t he frequency)
t he st ronger t he penet rat ion int o veget at ion and soil. Following wavelengt hs are in general
used:
P-band = ~ 65 cm AI RSAR
L-band = ~ 23 cm JERS-1 SAR, ALOS PALSAR
S-band = ~ 10 cm Almaz-1
C-band = ~ 5 cm ERS-1/ 2 SAR, RADARSAT-1/ 2, ENVI SAT ASAR, RI SAT-1
X-band = ~ 3 cm TerraSAR-X-1 , COSMO-SkyMed
K-band = ~ 1.2 cm Milit ary domain
Polarizat ion
I rrespect ive of wavelengt h, radar signals can t ransmit horizont al (H) or vert ical (V) elect ric-
field vect ors, and receive eit her horizont al (H) or vert ical (V) ret urn signals, or bot h. The basic
physical processes responsible for t he like-polarised (HH or VV) ret urn are quasi-specular
surface reflect ion. For inst ance, calm wat er (i.e. wit hout waves) appears black. The cross-
polarised (HV or VH) ret urn is usually weaker, and oft en associat ed wit h different reflect ions
due t o, for inst ance, surface roughness.
HV polarizat ion
The plot shows t he radar reflect ivit y variat ion for
different land cover classes (colours), while t he
dashed lines highlight t he swat h range for ENVI SAT
ASAR dat a.
Not e t hat t his angular dependence of t he radar
backscat t er can be exploit ed, by choosing an
opt imum configurat ions for different applicat ions.
I ncidence Angle
The incidence angle (u) is defined as t he angle formed by t he radar beam and a line
perpendicular t o t he surface. Microwave int eract ions wit h t he surface are complex, and
different reflect ions may occur in different angular regions. Ret urns are normally st rong at low
incidence angles and decrease wit h increasing incidence angle (see Figure).
incidence angle
r
a
d
a
r

r
e
f
l
e
c
t
i
v
i
t
y
1.3 Acqui si t i on Modes 1.3 Acqui si t i on Modes
St ripmap Mode - Principle
When operat ing as a St ripmap SAR, t he ant enna
usually gives t he syst em t he flexibilit y t o select an
imaging swat h by changing t he incidence angle.
Not e t hat t he St ripmap Mode is t he most commonly
used mode. I n t he case of ERS-1/ 2 SAR and JERS-1
SAR t he ant enna was fixed, hence disabling
select ion of an imaging swat h. The lat est generat ion
of SAR syst ems - like RADARSAT-1/ 2, ENVI SAT
ASAR, ALOS PALSAR, TerraSAR-X-1, COSMO-
SkyMed, and RI SAT-1 - provides for t he select ion of
different swat h modes.
S
w
a
t
h

w
i
d
t
h
ScanSAR Mode - Principle
While operat ing as a St ripmap SAR, t he syst em is limit ed t o
a narrow swat h. This const raint can be overcome by
ut ilising t he ScanSAR principle, which achieves swat h
widening by t he use of an ant enna beam which is
elect ronically st eerable in elevat ion.
Radar images can t hen be synt hesised by scanning t he
incidence angle and sequent ially synt hesising images for
t he different beam posit ions. The area imaged from each
part icular beam is said t o form a sub-swat h. The principle
of t he ScanSAR is t o share t he radar operat ion t ime
bet ween t wo or more separat e sub-swat hs in such a way as
t o obt ain full image coverage of each. S
w
a
t
h

w
i
d
t
h
Spot light Mode - Principle
During a Spot light mode dat a collect ion, t he sensor
st eers it s ant enna beam t o cont inuously illuminat e t he
t errain pat ch being imaged.
Three at t ribut es dist inguish Spot light and St ripmap
mode:
S
w
a
t
h

w
i
d
t
h
Spot light mode offers finer azimut h resolut ion t han
achievable in St ripmap mode using t he same physical
ant enna.
Spot light imagery provides t he possibilit y of imaging a
scene at mult iple viewing angles during a single pass.
Spot light mode allows efficient imaging of mult iple
smaller scenes whereas St ripmap mode nat urally
images a long st rip of t errain.
1.4 Scat t er i ng Mechani sms 1.4 Scat t er i ng Mechani sms
General
SAR images represent an est imat e of t he radar backscat t er for t hat area on t he ground. Darker
areas in t he image represent low backscat t er, while bright er areas represent high backscat t er.
Bright feat ures mean t hat a large fract ion of t he radar energy was reflect ed back t o t he radar,
while dark feat ures imply t hat very lit t le energy was reflect ed.
Backscat t er for a t arget area at a part icular wavelengt h will vary for a variet y of condit ions,
such as t he physical size of t he scat t erers in t he t arget area, t he t arget 's elect rical propert ies
and t he moist ure cont ent , wit h wet t er obj ect s appearing bright , and drier t arget s appearing
dark. (The except ion t o t his is a smoot h body of wat er, which will act as a flat surface and
reflect incoming pulses away from t he sensor. These bodies will appear dark). The wavelengt h
and polarisat ion of t he SAR pulses, and t he observat ion angles will also affect backscat t er.
Surface and Volume Scat t ering
A useful rule-of-t humb in analysing radar images is t hat t he higher or bright er t he backscat t er
on t he image, t he rougher t he surface being imaged. Flat surfaces t hat reflect lit t le or no radio
or microwave energy back t owards t he radar will always appear dark in radar images.
Veget at ion is usually moderat ely rough on t he scale of most radar wavelengt hs and appears as
grey or light grey in a radar image.
Surface Scat t ering
Volume Scat t ering
Double Bounce
Double Bounce
Surfaces inclined t owards t he radar will have a st ronger backscat t er t han surfaces which slope
away from t he radar and will t end t o appear bright er in a radar image. Some areas not
illuminat ed by t he radar, like t he back slope of mount ains, are in shadow, and will appear dark.
When cit y st reet s or buildings are lined up in such a way t hat t he incoming radar pulses are
able t o bounce off t he st reet s and t hen bounce again off t he buildings (called a double-bounce)
and direct ly back t owards t he radar t hey appear very bright (whit e) in radar images. Roads and
freeways are flat surfaces and so appear dark. Buildings which do not line up so t hat t he radar
pulses are reflect ed st raight back will appear light grey, like very rough surfaces.
Combinat ion of Scat t ering Mechanisms
I t is wort h ment ioning - in part icular for low frequency (like L- or P-band) SAR syst ems - t hat
t he observed radar reflect ivit y is t he int egrat ion of single scat t ering mechanisms - such as
surface (o
s
), volume (o
v
), and double bounce (o
t
) scat t ering - as shown, as example for
forest ry, in t he Figure. Not e t hat , a t heoret ical modelling (usually based on t he radiat ive
t ransfer t heory) of t he radar backscat t er is very complex and, t hereby, simplificat ions of t he
t arget and assumpt ions on t he basic scat t ering processes must be done.
Crown layer
Trunk layer
Ground layer
An Example
Sur f ace Scat t er i ng
Lake
Vol ume Scat t er i ng
For est r y
Doubl e Bounce
House
ERS-1
ERS-1 SAR (C-band) sample (ca. 17 km x 10 km)
Penet rat ion
Depending on t he frequency
and polarizat ion, waves can
penet rat e int o t he veget at ion
and, on dry condit ions, t o some
ext ent , int o t he soil (for
inst ance dry snow or sand).
Generally, t he longer t he wave-
lengt h, t he st ronger t he pene-
t rat ion int o t he t arget is. Wit h
respect t o t he polarizat ion,
cross-polarized (VH/ HV) acquisi-
t ions have a significant less
penet rat ion effect t han co-
polarized (HH/ VV) one.
Dielect ric Propert ies
Radar backscat t er also depends on t he dielect ric propert ies of t he t arget : for met al and wat er
t he dielect ric const ant is high (80), while for most ot her mat erials it is relat ively low: in dry
condit ions, t he dielect ric const ant ranges from 3 t o 8. This means t hat wet ness of soils or
veget at ed surfaces can produce a not able increase in radar signal reflect ivit y.
Based on t his phenomenon, SAR syst ems are also used t o ret rieve t he soil moist ure cont ent -
primarily - of bare soils. The measurement is based on t he large cont rast bet ween t he dielect ric
propert ies of dry and wet soils. As t he soil is moist ened, it s dielect ric const ant varies from
approximat ely 2.5 when dry t o about 25 t o 30 under sat urat ed condit ions. This t ranslat es t o an
increase in t he reflect ed energy. I t is wort h ment ioning t hat t he inference of soil moist ure from
t he backscat t ering coefficient is feasible but limit ed t o t he use of polarimet ric and dual
frequency (C-, L-band) SAR sensors, in order t o separat e t he effect of soil roughness and
moist ure.
1.5 Speck l e 1.5 Speck l e
General
Speckle refers t o a noise-like charact erist ic produced by coherent syst ems such as SAR and
Laser syst ems (not e: Suns radiat ion is not coherent ). I t is evident as a random st ruct ure of
pict ure element s (pixels) caused by t he int erference of elect romagnet ic waves scat t ered from
surfaces or obj ect s. When illuminat ed by t he SAR, each t arget cont ribut es backscat t er energy
which, along wit h phase and power changes, is t hen coherent ly summed for all scat t erers, so
called random-walk (see Figure). This summat ion can be eit her high or low, depending on
const ruct ive or dest ruct ive int erference. This st at ist ical fluct uat ion (variance), or uncert aint y, is
associat ed wit h t he bright ness of each pixel in SAR imagery.
When t ransforming SAR signal dat a int o act ual imagery -
aft er t he focusing process - mult i-look processing is
usually applied (so called non-coherent averaging). The
speckle st ill inherent in t he act ual SAR image dat a can be
reduced furt her t hrough adapt ive image rest orat ion
t echniques (speckle filt ering). Not e t hat unlike syst em
noise, speckle is a real elect romagnet ic measurement ,
which is exploit ed in part icular in SAR int erferomet ry
(I nSAR).
Speckle Model and Speckle Filt ering Principle
A well accept ed appropriat e model for fully developed speckle is t he mult iplicat ive fading
random process F,
I = R . F
where I is t he observed int ensit y (speckle observed reflect ivit y), R is t he random radar
reflect ivit y process (unspeckle reflect ivit y).
The first st ep in speckle filt ering is t o check if speckle is fully developed in t he neighbourhood
of t he pixel considered. I f t his is t he case, an est imat ion of t he radar reflect ivit y is made as a
funct ion of t he observed int ensit y, based on some local st at ist ics and of some a priori
knowledge about t he scene. Good speckle removal requires t he use of large processing
windows. On t he cont rary, good preservat ion of t he spat ial resolut ion is needed so as not t o
blur t hin image det ails like t ext ural or st ruct ural feat ures.
I n high spat ial resolut ion images, speckle can be part ially developed in some areas (e.g.
urban), when a few st rong scat t ers are present in t he resolut ion cell. I n t he ext reme case of an
isolat ed point t arget , int ensit y fluct uat ions are dominat ed by a det erminist ic funct ion which
should not be affect ed by t he speckle filt ering process. I n t hese cases, small window sizes are
preferable.
Speckle Model and Speckle Filt ering Principle (cont .)
A speckle filt ering is t herefore a compromise bet ween speckle removal (radiomet ric resolut ion)
and t hin det ails preservat ion (spat ial resolut ion).
Adapt ive filt ers based on appropriat e scene and speckle models are t he most appropriat e ones
for high spat ial resolut ion SAR images, when speckle is not always fully developed. Generally,
such filt ers are all adapt ive as a funct ion of t he local coefficient of variat ion and can be
enhanced by fixing a minimum value for bet t er speckle smoot hing and an upper limit for
t ext ure or point t arget preservat ion. The coefficient of variat ion (e.g. mean/ st andard deviat ion)
is a good indicat or of t he presence of het erogeneit y wit hin a window. I t is well adapt ed when
only isot ropic (homogeneous) t ext ure is present and can be assist ed by rat io operat ors for
anisot ropic orient ed t ext ural feat ures.
Enhanced speckle filt ers also include t he possibilit y t hat t he coefficient of variat ion is assist ed
by geomet rical det ect ors, and t hat t he rat io det ect or is ext ended t o t he det ect ion of linear
feat ures and isolat ed scat t erers.
1.6 Dat a St at i st i cs 1.6 Dat a St at i st i cs
Single Look Complex, Amplit ude, I nt ensit y (Power) Dat a
SAR dat a are composed by a real and imaginary part (complex dat a), so-called in-phase and
quadrat ure channels (see Figure).
Not e t hat t he phase of a single-channel SAR syst em (for example C-band, VV polarizat ion) is
uniformly dist ribut ed over t he range -t

t o + t. I n cont rast , t he amplit ude A has a Rayleigh
dist ribut ion, while t he I nt ensit y I (or Power P) = A
2
has a negat ive exponent ial dist ribut ion.
I n essence: I n single-channel SAR syst ems (not t o be confused wit h t he I nSAR, DI nSAR,
PolSAR, and PolI nSAR case) phase provides no informat ion, while Amplit ude (or I nt ensit y /
Power) is t he only useful informat ion.
1.6 Dat a St at i st i cs 1.6 Dat a St at i st i cs
I nt ensit y (or Power) Dat a
SAR dat a - aft er t he focusing process - are usually mult i-looked by averaging over range
and/ or azimut h resolut ion cells - t he so-called incoherent averaging. Fort unat ely even mult i-
looked I nt ensit y dat a have a well-known analyt ic Probabilit y Densit y Funct ion: I n fact , a L-look-
image (L is t he number of looks) is essent ially t he convolut ion of L-look exponent ial
dist ribut ions (see Figure).
An import ant charact erist ic are t he moment s
(expect ed mean and variance) of t he Gamma
funct ion (a st at ist ical funct ion closely relat ed t o
fact orials) for an homogeneous area, i.e.
mean = 2
.
(st andard deviat ion)
2
variance = 4
.
(st andard deviat ion)
4
/ L
This mot ivat es t he definit ion of t he Equivalent
Number of Looks (ENL) as
ENL = mean
2
/ variance
The ENL is equivalent t o t he number of
independent I nt ensit y I values averaged per pixel.
1.7 Geomet r y 1.7 Geomet r y
General
Due t o t he complet ely different geomet ric propert ies of SAR dat a in range and azimut h
direct ion, it is wort h considering t hem separat ely t o underst and t he SAR imaging geomet ry.
According t o it s definit ion, dist ort ions in range direct ion are large. They are mainly caused by
t opographic variat ions. The dist ort ions in azimut h are much smaller but more complex.
Geomet ry in Range
The posit ion of a t arget is a funct ion of t he pulse t ransit t ime bet ween t he sensor and t he
t arget on t he Eart hs surface. Therefore it is proport ional t o t he dist ance bet ween sensor and
t arget . The radar image plane (see figure included in t he next slide) can be t hought of as any
plane t hat cont ains t he sensors flight t rack. The proj ect ion of individual obj ect point s ont o t his
plane, t he so-called slant range plane, is proport ional t o t he sensor dist ance, and causes a
non-linear compression of t he imaged surface informat ion.
Geomet ry in Range (cont .)
The point s a,b, and c are imaged as a,b, and c in t he slant range plane (see figure). This
shows how minor differences in elevat ion can cause considerable range dist ort ions. These
relief induced effect s are called foreshort ening, layover and shadow.
Layover is an ext reme case of foreshort ening, where t he slope o

is bigger t han t he incidence
angle (u). Wit h an increasing (horizont al) dist ance, t he slant range bet ween sensor and t arget
decreases.
Shadow is caused by obj ect s, which cover part of t he t errain behind t hem.
Geomet ry in Range (cont .) - An Example
I n mount ainous areas, SAR images are i) generally st rongly geomet rically dist ort ed, and, ii)
from a radiomet ric point of view, SAR facing slopes appear very bright (see Figure). St eeper
t opography or smaller incidence angles - as in t he case of ERS-1/ 2 SAR - can worsen
foreshort ening effect s.
Not e t hat foreshort ening effect s can be
correct ed during t he geomet ric and radiomet ric
calibrat ion assuming t he availabilit y of high
resolut ion Digit al Elevat ion Model (DEM) dat a.
Layover and Shadow areas can be exact ly
calculat ed, but not correct ed. These areas have
no t hemat ic informat ion.
Slant versus Ground Range Geomet ry
Oft en SAR dat a are convert ed from t he slant range proj ect ion (i.e. t he original SAR geomet ry)
int o t he ground range one (see Figure).
Not e t hat SAR dat a in ground range proj ect ion are neit her in a cart ographic reference syst em
(for inst ance UTM Zone 32) nor geomet rically correct ed. The only way t o correct ly geocode
SAR dat a (i.e. t o convert t he SAR dat a int o a map proj ect ion) is by applying a rigorous range-
Doppler approach st art ing from SAR dat a in t he original slant range geomet ry.
u
u
i
Ground range (G
R
)
Slant range
(S
R
)
u sin
S
G
R
R
=
Geomet ry in Azimut h
The frequency of t he backscat t ered signal depends on
t he relat ive velocit y bet ween sensor and t he t arget .
Part s of t he signal, which are reflect ed from t arget s in
front of t he sensor, are regist ered wit h a higher t han
t he emit t ed frequency, since t he ant enna is moving
t owards t he t arget . Similarly, t he regist ered frequency
of obj ect s t hat are behind t he sensor is lower t han t he
emit t ed frequency.
All t arget s wit h a const ant Doppler frequency shift
describe a cone. The t ip of t his cone is t he phase
cent re of t he SAR ant enna and it s axis is defined by
t he velocit y vect or of t he plat form. The cut t ing
bet ween t his Doppler cone and t he surface of t he
Eart h is a hyperbola, which is called t he I so-Doppler
line (see Figure).
I so-Doppler line
2. How SAR Pr oduct s ar e Gener at ed? 2. How SAR Pr oduct s ar e Gener at ed?
SAR syst ems can acquire dat a in different ways, such as i) single or dual channel mode (for
inst ance HH or HH / HV or VV / VH), ii) int erferomet ric (single- or repeat -pass) mode, iii)
polarimet ric mode (HH,HV,VH,VV), and finally, iv) by combining int erferomet ric and
polarimet ric acquisit ion modes. Obviously, different acquisit ion modes are subj ect t o different
processing t echniques. They are:
Pr ocessi ng of SAR I nt ensi t y
The product generat ion is limit ed t o t he int ensit y processing.
I nt er f er omet r i c SAR ( I nSAR/ DI nSAR) Pr ocessi ng
The product generat ion includes int ensit y, and int erferomet ric phase processing.
Pol ar i met r i c SAR ( Pol SAR) Pr ocessi ng
The product generat ion includes int ensit y, and polarimet ric phase processing.
Pol ar i met r i c-I nt er f er omet r i c SAR ( Pol I nSAR) Pr ocessi ng
The product generat ion includes int ensit y, polarimet ric, and int erferomet ric phase
processing.
SARscape SARscape

- - Modul es Modul es
Basi c
I t includes a set of processing st eps for t he generat ion of SAR product s based on
int ensit y. This module is complement ed by a mult i-purpose t ool.
This module is complement ed by:
Focusi ng
I t support s t he focusing of RADARSAT-1, ENVI SAT ASAR, and ALOS PALSAR dat a.
Gamma & Gaussi an Fi l t er
I t includes a whole family of SAR specific filt ers. They are part icularly efficient t o
reduce speckle, while preserving t he radar reflect ivit y, t he t ext ural propert ies and t he
spat ial resolut ion, especially in st rongly t ext ured SAR images.
SARscape SARscape

- - Modul es Modul es
I nt er f er omet r y
I t support s t he processing of I nt erferomet ric SAR (2-pass int erferomet ry, I nSAR) and
Different ial I nt erferomet ric SAR (n-pass int erfereomet ry, DI nSAR) dat a for t he generat ion
of Digit al Elevat ion Model, Coherence, and Land Displacement / Deformat ion maps.
This module is complement ed by:
ScanSAR I nt er f er omet r y
I t offers t he capabilit ies t o process I nSAR and DI nSAR dat a over large areas (400 by
400 km).
SAR Pol ar i met r y / Pol ar i met r i c I nt er f er omet r y
The PolSAR/ PolI nSAR module support s t he processing of polarimet ric and polarimet ric
int erferomet ric SAR dat a.
Per si st ent Scat t er er s
I t enables t o det ermine mm-scale displacement s of individual feat ures on t he ground.
SARscape SARscape

Modul es Modul es i n ENVI i n ENVI

Envi r onment Envi r onment
SARscape SARscape

Tool s Tool s i n ENVI i n ENVI

Envi r onment Envi r onment
2.1 SAR I nt ensit y 2.1 SAR I nt ensit y Processing Processing
2.2 2.2 I nt erferomet ric SAR (I nSAR/ DI nSAR) Processing * I nt erferomet ric SAR (I nSAR/ DI nSAR) Processing *
2.3 2.3 Polarimet ric SAR (PolSAR) Processing Polarimet ric SAR (PolSAR) Processing
2.4 2.4 Polarimet ric Polarimet ric- - I nt erferomet ric SAR (PolI nSAR) Processing I nt erferomet ric SAR (PolI nSAR) Processing
Pr ocessi ng Techni ques Pr ocessi ng Techni ques
* St ripMap and Spot Light modes are support ed in t he I nt erferomet * St ripMap and Spot Light modes are support ed in t he I nt erferomet ry Module, ry Module,
ScanSAR mode in t he ScanSAR I nt erferomet ry Module ScanSAR mode in t he ScanSAR I nt erferomet ry Module
2.1.1 Focusing 2.1.1 Focusing
2.1.2 Mult i 2.1.2 Mult i - - looking looking
2.1.3 Co 2.1.3 Co- - regist rat ion regist rat ion
2.1.4 Speckle Filt ering 2.1.4 Speckle Filt ering
2.1.5 2.1.5 Geocoding Geocoding
2.1.6 Radiomet ric Calibrat ion 2.1.6 Radiomet ric Calibrat ion
2.1.7 Radiomet ric Normalizat ion 2.1.7 Radiomet ric Normalizat ion
2.1.8 2.1.8 Mosaicing Mosaicing
2.1.9 Segment at ion 2.1.9 Segment at ion
2.1.10 Classificat ion 2.1.10 Classificat ion
2.1 SAR I nt ensi t y Pr ocessi ng 2.1 SAR I nt ensi t y Pr ocessi ng
2.1 SAR I nt ensi t y Pr ocessi ng 2.1 SAR I nt ensi t y Pr ocessi ng
2.1.1 Focusi ng 2.1.1 Focusi ng
Purpose
SAR processing is a t wo-dimensional problem. I n t he raw dat a, t he signal energy from a point
t arget is spread in range and azimut h, and t he purpose of SAR focussing is t o collect t his
dispersed energy int o a single pixel in t he out put image.
Focusing for SAR image format ion involves sampled and quant ized SAR echoes dat a and
represent s a numerical evaluat ion of t he synt het ic apert ure beam format ion process. A large
number of arit hmet ic comput at ions are involved. The numerical nat ure of t he digit al
correlat ion calls for t he formulat ion of an accurat e mat hemat ical procedure, oft en referred t o
as a SAR correlat ion or focusing algorit hm, in order t o manipulat e t he sample echo signals t o
accomplish t he SAR correlat ion process.
Met hod
SAR processing is performed in range and azimut h direct ions. I n r ange, t he signal is spread
out by t he dur at i on of t he linear Frequency Modulat ion (FM) t r ansmi t t ed pul se. I n
azi mut h, t he signal is spread out by t he l engt h of t he per i od it is illuminat ed by t he
ant enna beam, or t he synt het ic apert ure. As a point t arget passes t hrough t he azimut h ant enna
beam, t he range t o t he t arget changes. On t he scale of t he wavelengt h, t his range variat ion
causes a phase variat ion in t he received signal as a funct ion of azimut h. This phase variat ion
over t he synt het ic apert ure corresponds t o t he Doppler bandwidt h of t he azimut h signal, and
allows t he signal t o be compressed in t he azimut h direct ion.
The range variat ion t o a point t arget can result in a variat ion in t he range delay (dist ance
sensor-t arget ) t o t he t arget t hat is larger t han t he range sample spacing, result ing in what is
called range migrat ion. This range migrat ion of t he signal energy over several range bins must
be correct ed before azimut h compression can occur. The range-Doppler algorit hm performs
t his correct ion very efficient ly in t he range-t ime, azimut h-frequency domain.
Met hod (cont .)
I n order t o process t he correct part of t he azimut h frequency spect rum, t he range-Doppler
algorit hm requires as input t he Doppl er cent r oi d. I t also requires knowledge of t he
t r ansmi t t ed pul se for range compression, and of t he i magi ng geomet r y such as t he range
and sat el l i t e vel oci t y for const ruct ion of t he azimut h mat ched filt er. The main st eps are
shown in t he block diagram below, and are described in t he following sect ions.
SAR Raw Dat a
Range Compression
- Range Migrat ion
- Aut ofocus
- DC ambiguit y est imat ion
Single Look Compex Dat a
Azimut h Compression
Doppler Cent roid Est imat ion
Doppler Cent roid Est imat ion
The Doppler Cent roid (DC) frequency of SAR signal is relat ed
t o locat ion of t he azimut h beam cent re, and is an import ant
input paramet er when processing SAR imagery. DC locat es
t he azimut h signal energy in t he azimut h (Doppler) frequency
domain, and is required so t hat all of t he signal energy in t he
Doppler spect rum can be correct ly capt ured by t he azimut h
compression filt er, providing t he best signal-t o-noise rat io
and azimut h resolut ion. Wrong DC est imat ion result s in areas
of low signal-t o-noise rat io, st rong discret e t arget s, and
radiomet ric discont inuit ies. Even wit h an accurat e knowledge
of t he sat ellit e posit ion and velocit y, t he point ing angle must
be dynamically est imat ed from t he SAR dat a in order t o
ensure t hat radiomet ric requirement s are met .
A number of algorit hms have been developed t o est imat e t he
Doppler cent roid frequency. Oft en, t he key challenge is t o
define t echniques t hat will yield sufficient ly accurat e
est imat es for all processing modes.
I f t he ant enna is squint ed
(i.e. not perpendicular t o
t he flight direct ion), t he
Doppler spect rum is not
symmet ric.
Range Compression
I n collect ing t he SAR dat a, a long-durat ion linear Frequency Modulat ion (FM) pulse is
t ransmit t ed. This allows t he pulse energy t o be t ransmit t ed wit h a lower peak power. The linear
FM pulse has t he propert y t hat , when filt ered wit h a mat ched filt er (e.g. t he reference
funct ion), t he result is a narrow pulse in which all t he pulse energy has been collect ed t o t he
peak value. Thus, when a mat ched filt er is applied t o t he received echo, it is as if a narrow
pulse were t ransmit t ed, wit h it s corresponding range resolut ion and signal-t o-noise rat io.
This mat ched filt ering of t he received echo is called range compression. Range compression is
performed on each range line of SAR dat a, and can be done efficient ly by t he use of t he Fast
Fourier Transform (FFT). The frequency domain range mat ched filt er needs t o be generat ed
only once, and it is applied t o each range line. The range mat ched filt er may be comput ed or
generat ed from a replica of t he t ransmit t ed pulse. I n addit ion, t he range mat ched filt er
frequency response t ypically includes an amplit ude weight ing t o cont rol sidelobes in t he range
impulse response.
Range FFT
X
Raw Dat a ( Radar Echos)
The energy of one single
point is spread along bot h
direct ions.
Range Compressed Dat a
The t ransmit t ed chirp
(pulse) is compressed in one
range bin, while t he echoes
are spread along t he
azimut h direct ion.
Range Compression (cont .) - An Example
I n t he following pict ures an example is shown of how SAR dat a are processed for a single
synt het ic point . I n horizont al is shown t he azimut h direct ion, in vert ical t he range one.
Compr essed Chi r p
(Reference Funct ion)
Range FFT
-1
Azimut h Compression
Azimut h compression is a mat ched filt ering of t he azimut h signal, performed efficient ly using
FFT's. The frequency response of t he azimut h compression filt er is precomput ed using t he
orbit al geomet ry. The azimut h filt er also depends on range. Thus t he dat a is divided int o range
invariance regions, and t he same basic mat ched filt er is used over a range int erval called t he
Frequency Modulat ion (FM) rat e invariance region. The size of t his invariance region must not
be large enough t o cause severe broadening in t he compressed image. Also included is an
amplit ude weight ing t o cont rol sidelobes in t he azimut h impulse response. Not e t hat t he posit ion
of t he amplit ude weight ing in t he azimut h frequency array depends on t he Doppler Cent roid,
which also depends on range.
The ext ract ed frequency array for each look is mult iplied by t he mat ched filt er frequency
response and t he inverse FFT is performed t o form t he complex look image. The mat ched filt er
frequency response is adj ust ed by a small linear phase ramp for each look. I n addit ion, azimut h
int erpolat ion may also performed aft er look compression t o achieve a desire azimut h pixel
spacing, and it is done on each look separat ely.
Range Compressed Dat a Azimut h FFT
Range Cell Migrat ion Correct ion
Azimut h FFT
-1
X Reference funct ion
Azi mut h compr essed dat a
All t he energy backscat t ered
by one single resolut ion cell on
ground is compressed in one
pixel.
Azimut h Compression (cont .) - An Example
Range compressed dat a are, aft er t he Range Cell Migrat ion Correct ion, azimut h compressed.
Not e t hat during focusing, t hese st eps are execut ed on t he whole image, t o obt ain a complex
image (Single Look Complex) where amplit ude is relat ed t o t he radar reflect ivit y, and phase t o
t he acquisit ion geomet ry and on t he ground t opography.
From Single Look Complex t o Det ect ed (1-Look) Dat a
Aft er look compression, each of t he look images is det ect ed, i.e. t he dat a is convert ed from
complex t o real numbers (r
2
+ i
2
= P
2
). That is, t he Power (or I nt ensit y) of each complex
sample is calculat ed. Not e t hat t he pixel of t he Single Look Complex (SLC) and Power dat a
does not have t he same dimensions as t he resolut ion cell during t he dat a acquisit ion, due t o
t he variat ion of range resolut ion wit h incidence angle.
The pict ure below shows - as example - an ENVI SAT ASAR AP (HH polarizat ion) dat a of
Licht enburg (Sout h Africa) wit h 1-look.
azimut h direct ion
r
a
n
g
e

d
i
r
e
c
t
i
o
n

SARscape SARscape

- - Focusing Module Focusing Module
2.1.2 Mul t i 2.1.2 Mul t i - - l ooki ng l ooki ng
Purpose
The SAR signal processor can use t he full synt het ic
apert ure and t he complet e signal dat a hist ory in
order t o produce t he highest possible resolut ion,
albeit very speckled, Single Look Complex (SLC) SAR
image product . Mult iple looks may be generat ed -
during mult i-looking - by averaging over range
and/ or azimut h resolut ion cells. For an improvement
in radiomet ric resolut ion using mult iple looks t here is
an associat ed degradat ion in spat ial resolut ion. Not e
t hat t here is a difference bet ween t he number of
looks physically implement ed in a processor, and t he
effect ive number of looks as det ermined by t he
st at ist ics of t he image dat a.
ENVI SAT ASAR AP
(HH polarizat ion) dat a
of Licht enburg (Sout h
Africa) wit h 1 look
(left ) and mult i-looked
wit h a fact or 4 in
azimut h (right ).
How t o select an appropriat e number of looks - An Example
The number of looks is a funct ion of
- pixel spacing in azimut h
- pixel spacing in slant range
- incidence at scene cent re
The goal is t o obt ain in t he mult i-looked image approximat ely squared pixels considering t he
ground range resolut ion (and not t he pixel spacing in slant range) and t he pixel spacing in
azimut h. I n part icular, in order t o avoid over- or under-sampling effect s in t he geocoded image,
it is recommended t o generat e a mult i-looked image corresponding t o approximat ely t he same
spat ial resolut ion foreseen for t he geocoded image product .
Not e t hat ground resolut ion in range is defined as
ground range resolut ion = pixel spacing range
sin(incidence angle)
How t o select an appropriat e number of looks (cont .) - An Example
- pixel spacing azimut h = 3.99 m
- pixel spacing range = 7.90 m
- incidence angle = 23
-> ground resolut ion = 7.90 / sin(23) = 20.21 m
-> result ing pixel spacing azimut h = 3.99
.
5 = 19.95 m
-> recommended pixel size of geocoded image 20 m
I t is import ant t o not e t hat t his example refers t o ERS-1/ 2 SAR dat a, which is charact erized by
a fixed acquist ion geomet ry. Advanced current SAR syst ems - such as RADARSAT-1, ENVI SAT
ASAR, and fut ure SAR missions - can acquire images wit h different incidence angle (i.e.
different Beam Modes). I n t his case, t o each different acquisit ion mode a different mult i-
looking fact or in range and azimut h must be applied.
SARscape SARscape

- - Basic Module Basic Module
2.1.3 Co 2.1.3 Co - - r egi st r at i on r egi st r at i on
Purpose
When mult iple images cover t he same region
and, in part icular, a speckle filt ering based
on t ime-series will be performed, or image
rat ioing (or similar operat ions) are required
in slant (alt ernat ively ground) range
geomet ry, SAR images must be co-
regist ered. This requires spat ial regist rat ion
and pot ent ially resampling (in cases where
pixel sizes differ) t o correct for relat ive
t ranslat ional shift , rot at ional and scale
differences. Not e t hat co-regist rat ion is
simply t he process of superimposing, in t he
slant range geomet ry, t wo or more SAR
images t hat have t he same orbit and
acquisit ion mode. This process must not t o
be confused wit h geocoding ( or
georeferencing), which is t he process of
convert ing each pixel from t he slant range
geomet ry t o a cart ographic reference
syst em.
ENVI SAT ASAR AP (HH polarizat ion) dat a of
Licht enburg (Sout h Africa) acquired at t hree
different dat es have been focused, mult i-
looked and co-regist ered. The color composit e
have been generat ed by combining t he t hree
co-regist ered images assigned t o t he red,
green and blue channel.
A local non-paramet ric shift est imat e is comput ed on t he basis of t he orbit al dat a and t he
Digit al Elevat ion Model (if provided in input ). I n case of inaccurat e orbit s a large cent ral
window (Cross-correlat ion Cent ral Window) is used inst ead.
A set of windows (Cross-correlat ion Grid) is found on t he reference image.
The input dat a cross-correlat ion funct ion is comput ed for each window.
The maximum of t he cross-correlat ion funct ion indicat es t he proper shift for t he select ed
locat ion.
The residual paramet ric shift , which is summed t o t he original local non-paramet ric
est imat e, is calculat ed by a polynomial depending on t he azimut h and range pixel posit ion.
I n case t he input SAR dat a are represent ed by SLC product s, t he residual paramet ric shift
is furt her refined by comput ing "mini-int erferograms" on small windows (Fine Shift
Paramet ers) dist ribut ed t hroughout t he image.
Met hod
This st ep is performed in an aut omat ic way, according t o t he following procedure:
SARscape SARscape

2.1.4 Speck l e Fi l t er i ng 2.1.4 Speck l e Fi l t er i ng
Purpose
I mages obt ained from coherent sensors such as SAR (or Laser) syst em are charact erized by
speckle. This is a spat ially random mult iplicat ive noise due t o coherent superposit ion of
mult iple backscat t er sources wit hin a SAR resolut ion element . I n ot her words, speckle is a
st at ist ical fluct uat ion associat ed wit h t he radar reflect ivit y (bright ness) of each pixel in t he
image of a scene. A first st ep t o reduce speckle - at t he expense of spat ial resolut ion - is
usually performed during t he mult i-looking, where range and/ or azimut h resolut ion cells are
averaged. The more looks t hat are used t o process an image, t he less speckle t here is.
I n t he following sect ions select ed algorit hms are short ly present ed.
Filt er
Minimum Mean
Square Error
- Gamma MAP
- Annealing
- Mult i- Temporal
Merge Using
Moment s
- Annealed
Segment at ion
- Mult i- Temporal
st ruct ure
reconst ruct ion
speed
qualit y
speed qualit y
Speckle Filt ering
or Segment at ion?
Speed or
Qualit y?
Segment at ion
Speed or
Qualit y?
Speckle Filt ering or Segment at ion?
I n essence, t here is no golden rule solut ion: speckle filt ering or segment at ion (Sect ion 2.1.7)
should be applied wit h respect t o t he specific needs.
Overview
Minimum Mean
Square Er ror
Speck l e Fi l t er i ng
Non- SAR
Speci f i c
SAR Speci f i c
Gamma- Gamma MAP
Gamma- Gaussian MAP
Gaussian- Gaussian MAP
Annealed Correlat ed
MAP
I
n
c
r
e
a
s
e

i
n

r
e
l
i
a
n
c
e

u
p
o
n

a

m
o
d
e
l
I
n
c
r
e
a
s
e
d

n
u
m
b
e
r

o
f

a
s
s
u
m
p
t
i
o
n
s
Mult i- Tempor al
Polarimet ric
Non- Edge Preserving
( Mean, Median, et c. )
Edge Preser ving
( Mat suyama)
Anisot ropic
Non- Linear Diffusion
Mult i- Tempor al
Anisot ropic
Non- Linear Diffusion
Minimum Mean Square Error Filt ers
The most well known adapt ive filt ers are based on a mult iplicat ive model and t hey consider
local st at ist ic.
The Fr ost filt er is an adapt ive filt er, and convolves t he pixel values wit hin a fixed size window
wit h an adapt ive exponent ial impulse response.
The Lee and Kuan filt ers perform a linear combinat ion of t he observed int ensit y and of t he
local average int ensit y value wit hin t he fixed window. They are all adapt ive as a funct ion of t he
local coefficient of variat ion (which is a good indicat or of t he presence of some het erogeneit y
wit hin t he window) and can be enhanced by fixing a minimum value for bet t er speckle
smoot hing and an upper limit t ext ure or point t arget preservat ion.
Gamma-Gamma and Gaussian-Gaussian Maximum A Post eriori (MAP)
I n t he presence of scene t ext ure, t o preserve t he useful spat ial resolut ion, e.g. t o rest ore t he
spat ial fluct uat ions of t he radar reflect ivit y (t ext ure), an A Priori first order st at ist ical model is
needed. Wit h respect t o SAR clut t er, it is well accept ed t hat t he Gamma-dist ribut ed scene
model is t he most appropriat e. The Gamma-dist ribut ed scene model, modulat ed by, eit her an
independent complex-Gaussian speckle model (for SAR SLC images), or by a Gamma speckle
model (for mult i-look det ect ed SAR images), gives rise t o a K-dist ribut ed clut t er. Nevert heless,
t he Gaussian-dist ribut ed scene model remains st ill popular, mainly for mat hemat ical t ract abilit y
of t he inverse problem in case of mult i-channel SAR images (mult ivariat e A Priori scene
dist ribut ions). I n t his cont ext , t he following filt er families has been developed:
- Single channel det ect ed SAR dat a
- Gamma-Gamma MAP filt er
- Gamma-Dist ribut ion-Ent ropy MAP filt er
- Gamma-A Post eriori Mean filt er
- Mult i-channel det ect ed SAR dat a
- Gamma-Gaussian MAP filt er for uncorrelat ed speckle
- Gaussian-Gaussian MAP filt er for correlat ed speckle
- Gaussian-Dist ribut ion-Ent ropy MAP filt er for correlat ed speckle
Mult i-Temporal Filt ers
Wit hin t he Mult i-Temporal filt ering - besides t he considerat ion of a speckle specific filt er - an
opt imum weight ing filt er is int roduced t o balance differences in reflect ivit y bet ween images at
different t imes. I t has t o be point ed out t hat mult i-t emporal filt ering is based on t he
assumpt ion t hat t he same resolut ion element on t he ground is illuminat ed by t he radar beam in
t he same way, and corresponds t o t he same coordinat es in t he image plane (sampled signal) in
all images of t he t ime series. The reflect ivit y can of course change from one t ime t o t he next
due t o a change in t he dielect ric and geomet rical propert ies of t he element ary scat t ers, but
should not change due t o a different posit ion of t he resolut ion element wit h respect t o t he
radar. Therefore proper spat ial co-regist rat ion of t he SAR images in t he t ime series is of
paramount import ance.
Single-I mage and Mult i-t emporal Anisot ropic Non-Linear Diffusion Filt er
First int roduced for single opt ical images, t his part icular t ype of single-image filt ering allows a
high level of regularizat ion in homogenous areas while preserving t he relevant feat ures
ult imat ely used for segment at ion (edges or more generally discont inuit ies). This is realized by
int roducing an edge-direct ion sensit ive diffusion cont rolled by a t ensor of values specifying t he
diffusion import ance in t he feat ures direct ion
The drawback of exist ing mult i-t emporal speckle filt ers is t hat t hey are st rongly sensor and
acquisit ion mode dependant , because based on st at ist ical scene descript ors. Moreover, if
feat ures masks are used, an accuracy loss can be int roduced when regarding part icular shape
preservat ion, mainly due t o t he lack of a priori informat ion about size and t ype of t he feat ures
exist ent in t he image. Therefore, in order t o t ake advant age of t he redundant informat ion
available when using mult i-t emporal series, while being fully independent regarding t he dat a
source, a hybrid mult i-t emporal anisot ropic diffusion scheme is proposed.
ENVI SAT ASAR AP (HH polarizat ion) mult i-looked unfilt ered (left ) and Gamma-Gamma MAP filt ered
image (right ). Not e t he speckle reduct ion while preserving t he st ruct ural feat ures of t he Gamma-
Gamma MAP one.
Example I
Mean (left ) and Mult i-Temporal (De Grandi) filt ered (right ) ENVI SAT ASAR AP (HH polarizat ion)
image. Not e t he blurring effect s of t he mean filt er, and t he st rong speckle reduct ion while
preserving t he st ruct ural feat ures of t he mult i-t emporal (De Grandi filt er) one.
Example I I
The pict ure shows a sample of t hree
speckle filt ered ENVI SAT ASAR AP
(HH polarizat ion) images from t he
area of Licht enburg (Sout h Africa)
acquired at different dat es. Not e t hat
t he images have been focused,
mult i-looked, co-regist ered and
speckle filt ered using a Mult i-
Temporal (De Grandi) filt er. Compare
t his image wit h t he mult i-t emporal
unfilt ered one (2.1.3 Co-
regist rat ion).
Example I I I
SARscape SARscape

SARscape SARscape

- - Gamma and Gaussian Filt er Module Gamma and Gaussian Filt er Module
2.1.5 2.1.5 Geocodi ng Geocodi ng
Purpose
Geocoding, Georeferencing, Geomet ric Calibrat ion, and Ort ho-rect ificat ion are synonyms. All of
t hese definit ions mean t he conversion of SAR images - eit her slant range (preferably) or ground
range geomet ry - int o a map coordinat e syst em (e.g. cart ographic reference syst em). A
dist inct ion is usually made bet ween
El l i psoi dal Geocodi ng, when t his process is performed wit hout t he use of Digit al Elevat ion
Model (DEM) dat a
Ter r ai n Geocodi ng, when t his process is performed wit h t he use of DEM dat a
Not e t hat t he only appropriat e way t o geocode SAR dat a is by applying a range-Doppler
approach (refer t o SAR geomet ry sect ion for t he j ust ificat ion). I n fact , SAR syst ems cause non-
linear compressions (in part icular in t he presence of t opography), and t hus t hey can not be
correct ed using polynomials as in t he case of opt ical images, where (in t he case of flat Eart h) an
affine t ransformat ion is sufficient t o convert it int o a cart ographic reference syst em.
Range-Doppler Approach
The removal of geomet ric dist ort ions requires a high precision geocoding of t he image
informat ion. The geomet ric correct ion has t o consider t he sensor and processor charact erist ics
and t hus must be based on a rigorous range-Doppler approach. For each pixel t he following
t wo relat ions must be fulfilled:
where R
s
= Slant range
S, P = Spacecraft and backscat t er element posit ion
v
s
, v
p
= Spacecraft and backscat t er element velocit y
f
0
= Carrier frequency
c = Speed of light
f
D
= Processed Doppler frequency
Range equat ion
Doppler equat ion
P S R =
s
s s p
D
R c
R v v f
f
) ( 2
0

=
Range-Doppler Approach (cont .)
Using t hese equat ions, t he relat ionship bet ween t he sensor, each single backscat t er element
and t heir relat ed velocit ies is calculat ed and t herefore not only t he illuminat ing geomet ry but
also t he processors charact erist ics are considered. This complet e reconst ruct ion of t he imaging
and processing geomet ry also t akes int o account t he t opographic effect s (foreshort ening,
layover) as well as t he influence of Eart h rot at ion and t errain height on t he Doppler frequency
shift and azimut h geomet ry.
The geocoding is usually implement ed using a backward solut ion (see Figure), i.e. t he Out put
(DEM or ellipsoidal height ) is t he st art ing point .
Out put
Digit al Elevat ion Model (or
ellipsoidal height ) in a car-
t ographic reference syst em
(for inst ance UTM zone 32
in WGS-84 syst em)
I nput
Slant range
geomet ry
1. Range-Doppler equat ion
2. Resampling
Nominal versus Precise Geocoding
I n t he geocoding procedure following nomenclat ure is used:
Nomi nal Geocodi ng, if
- No Ground Cont rol Point (GCP)
- Orbit al paramet ers (posit ions and velocit ies)
- Processed paramet ers (Doppler, range delay, pixel spacing, et c.)
- Digit al Elevat ion Model or Ellipsoidal height
are used during t he geocoding process.
Pr eci se Geocodi ng, if
- Ground Cont rol Point s (1 GCP is sufficient )
- Orbit al paramet ers (posit ions and velocit ies)
- Processed paramet ers (Doppler, range delay, pixel spacing, et c.)
- Digit al Elevat ion Model
are used during t he geocoding process.
Not e t hat geocoded images achieve a pixel accuracy even wit hout t he use of GCPs, if proper
processing is performed and orbit al paramet ers (so-called precise orbit s) are available.
Geodet ic and Cart ographic Transforms
During t he geocoding procedure
geodet ic and cart ographic t ransforms
must be considered in order t o
convert t he geocoded image from t he
Global Cart esian coordinat e syst em
(WGS-84) int o t he local Cart ographic
Reference Syst em (for inst ance UTM-
32, Gauss-Krueger, Oblique Mercat or,
et c.). The Figure shows t h e
conversion st eps included in t his
t ransform.
Some Basic Geodet ic and Cart ographic Nomenclat ure
Pr oj ect i on represent s t he 3-dimensional Eart hs surface in a 2-dimensional plane.
El l i psoi d is t he mat hemat ical descript ion of t he Eart hs shape.
El l i psoi dal Hei ght is t he vert ical dist ance above t he reference ellipsoid and is measured
along t he ellipsoidal normal from t he point t o t he ellipsoid.
Geoi d is t he Eart hs level surface. The geoid would have t he shape of an oblat e ellipsoid
cent red on t he Eart hs cent re of mass, if t he Eart h was of uniform densit y and t he Eart hs
t opography did not exist .
Or t homet r i c Hei ght is t he vert ical dist ance above t he geoid and is measured along t he
plumb line from t he point t o t he geoid.
Topogr aphy is t he Eart hs physical surface.
Dat um Shi f t Par amet er s convert t he ellipsoids origin int o t he Eart hs cent re.
Ellipsoidal Geocoding - ENVI SAT ASAR AP (HH) Mode
The pict ure shows t hree ENVI SAT ASAR AP
(HH polarizat ion) images from t he area of
Licht enburg (Sout h Africa) acquired at
different dat es. The images, which have a
pixel spacing of 15 m, have been focused,
mult i-looked, co-regist ered, speckle filt ered
using a mult i-t emporal (De Grandi) filt er,
and finally ellipsoidal geocoded in t he WGS-
84, UTM zone 35 reference syst em. The
ellipsoidal geocoding using a reference
height has been performed in a nominal way
based on precise orbit s from t he DORI S
(Doppler Orbit ography and Radioposit ioning
I nt egrat ed by Sat ellit e) syst em.
Terrain Geocoding - ERS-2 SAR
The pict ure shows an ERS-2 SAR
image of t he Bern area (Swit zerland).
This image has been focused, mult i-
looked, speckle filt ered using a
Gamma-Gamma MAP filt er, and finally
t errain geocoded in t he Oblique
Mercat or (Bessel 1814 ellipsoid)
reference syst em. The t errain
geocoding using a high resolut ion
Digit al Elevat ion Model (DEM) has
been performed in a nominal way
using precise orbit s of t he DORI S
syst em.
Ellipsoidal Geocoding - ERS-2 SAR
The pict ure shows an ERS-2 SAR image
of t he Bern area (Swit zerland). The
image has been focused, mult i-looked,
speckle filt ered using a Gamma-Gamma
MAP filt er, and ellipsoidal geocoded in
t he Oblique Mercat or (Bessel 1814
ellipsoid) reference syst em. The
ellipsoidal geocoding using a (const ant )
reference height has been performed in
a nominal way using precise orbit s of
t he DORI S syst em. Not e t he locat ion
inaccuracies - due t o t he lack of t he
DEM informat ion - wit h respect t o t he
t errain geocoded image. Compare t his
image wit h t he corresponding t errain
geocoded one.
SARscape SARscape

2.1.6 Radi omet r i c Cal i br at i on 2.1.6 Radi omet r i c Cal i br at i on
Purpose
Radars measure t he rat io bet ween t he power of t he pulse t ransmit t ed and t hat of t he echo
received. This rat io is called t he backscat t er. Calibrat ion of t he backscat t er values is necessary
for int ercomparison of radar images acquired wit h different sensors, or even of images obt ained
by t he same sensor if acquired in different modes or processed wit h different processors.
I n order t o avoid misunderst anding, not e t he following nomenclat ure:
Bet a Nought ( ) is t he radar bright ness (or reflect ivit y) coefficient . The reflect ivit y per
unit area in slant range is dimensionless. This normalizat ion has t he virt ue t hat it does not
require knowledge of t he local incidence angle (e.g. scat t ering area A).
Si gma Nought ( o
o
) , t he backscat t ering coefficient , is t he convent ional measure of t he
st rengt h of radar signals reflect ed by a dist ribut ed scat t erer, usually expressed in dB. I t is a
normalized dimensionless number, which compares t he st rengt h observed t o t hat expect ed
from an area of one square met re. Sigma nought is defined wit h respect t o t he nominally
horizont al plane, and in general has a significant variat ion wit h incidence angle, wavelengt h,
and polarizat ion, as well as wit h propert ies of t he scat t ering surface it self.
Gamma ( ) is t he backscat t ering coefficient normalized by t he cosine of t he incidence
angle.
The Radar Equat ion
The radiomet ric calibrat ion of t he SAR images involves, by considering t he radar equat ion law,
correct ions for:
The scat t ering area (A): each out put pixel is normalised for t he act ual illuminat ed area of
each resolut ion cell, which may be different due t o varying t opography and incidence
angle.
The ant enna gain pat t ern (G
2
): t he effect s of t he variat ion of t he ant enna gain (t he rat io
of t he signal, expressed in dB, received or t ransmit t ed by a given ant enna as compared t o
an isot ropic ant enna) in range are correct ed, t aking int o account t opography (DEM) or a
reference height .
The range spread loss (R
3
): t he received power must be correct ed for t he range dist ance
changes from near t o far range.
The Radar Equat ion (cont .)
The base of t he radiomet ric calibrat ion is t he radar equat ion. The formulat ion of t he received
power for dist ribut ed scat t ers P
d
for a scat t ering area A can be writ t en as:
Scat t ering Area A
Ant enna Gain Pat t ern
Range Spread Loss
Backscat t ering Coefficient - ERS-2 SAR
The pict ure shows t he backscat t ering
coefficient (o
o
) est imat ed from an ERS-2 SAR
image of t he Bern area (Swit zerland). The
image has been focused, mult i-looked,
speckle filt ered using a Gamma-Gamma MAP
filt er, and finally t errain geocoded in t he
Oblique Mercat or (Bessel 1814 ellipsoid)
reference syst em. The radiomet ric calibrat ion
has been performed during t he t errain geo-
coding procedure. Not e t hat a rigorous
radiomet ric calibrat ion can be exclusively
carried out using DEM dat a, which allows t he
correct calculat ion of t he scat t ering area.
Compare t his image wit h t he corresponding
t errain geocoded (but not radiomet rically
calibrat ed) one in t he previous slide.
Local I ncidence Angle Map
The pict ure shows t he local incidence
angle map (e.g. t he angle bet ween t he
normal of t he backscat t ering element
and t he incoming radiat ion). This
direct ion result s from t he sat ellit es
posit ion vect or and t he posit ion vect or
of t he backscat t ering element . The gray
t ones correspond t o t he angle and
achieve t he bright est t ones in areas
close t o shadow. The darkest t ones
corresponds t o areas close t o layover.
Not e t hat t he local incidence angle map
is used t o calculat e t he effect ive
scat t ering area A.
Layover and Shadow Map
The pict ure shows, in red, t he areas of
layover for t he ERS-2 SAR image from t he
area of Bern (Swit zerland). Not e t hat
from a t hemat ic point of view t hese areas
cannot be used t o ext ract informat ion,
since t he radar energy is compressed t o a
few resolut ion cells only. Furt hermore, in
t hese areas t he local spat ial resolut ion
t ends t o t he infinit e. I t has t o be point ed
out t hat t he calculat ion of a layover and
shadow map can only be carried out by
using a high resolut ion Digit al Elevat ion
Model (DEM). I n t his case no shadow
areas exist due t o t he st eep incidence
angle.
SARscape SARscape

2.1.7 Radi omet r i c Nor mal i zat i on 2.1.7 Radi omet r i c Nor mal i zat i on
Purpose
Even aft er a rigorous radiomet ric calibrat ion, backscat t ering coefficient variat ions are clearly
ident ifiable in t he range direct ion. This is because t he backscat t ered energy of t he illuminat ed
obj ect s is dependent on t he incidence angle. I n essence, t he smaller t he incidence angle and
t he wider t he swat h used t o acquire an image, t he st ronger t he variat ion of t he backscat t ering
coefficient in t he range direct ion. Not e t hat t his variat ion represent s t he int rinsic propert y of
each obj ect , and t hus may not be correct ed.
This plot shows t he backscat t ering variat ion
for different land cover classes (colours),
while t he dashed lines highlight t he swat h
range for ENVI SAT ASAR dat a.
I n order t o equalize t hese variat ions usually
a modified cosine correct ion is applied.
incidence angle
r
a
d
a
r

r
e
f
l
e
c
t
i
v
i
t
y
Backscat t ering Coefficient - ENVI SAT ASAR Wide Swat h
coefficient (o
o
) est imat ed from an
ENVI SAT ASAR Wide Swat h mode
image (405 km swat h) from t he area
of Malawi (Africa). The image, which
has a pixel spacing of 150 m, has
been focused, mult i-looked, speckle
filt ered using a Gamma-Gamma MAP
filt er, t errain geocoded in t he WGS-84
(UTM zone 36 reference syst em), and
radiomet rically calibrat ed. Not e t he
st rong bright ness variat ions from near
(left ) t o far range (right ).
Normalized Backscat t ering Coefficient - ENVI SAT ASAR Wide Swat h
The pict ure shows t he normalized
backscat t ering coefficient (o
o
) est i-
mat ed from an ENVI SAT ASAR Wide
Swat h mode image (405 km swat h)
from t he area of Malawi (Africa). The
image, which has a pixel spacing of
150m, has been focused, mult i-
looked, speckle filt ered using a
Gamma-Gamma MAP filt er, t errain
geocoded in t he WGS-84 (UTM zone
36 reference syst em), and radio-
met rically calibrat ed. Not e t he
bright ness homogeneit y aft er t he
radiomet ric normalizat ion procedure.
Backscat t ering Coefficient - ENVI SAT ASAR Global Mode
coefficient (o
o
) est imat ed from an
ENVI SAT ASAR Global Mode image
(405km swat h) from an area covering t he
I vory Coast , Mali, Burkina Faso, and
Maurit ania (Africa). The image, which has
a pixel spacing of 1 km, has been
focused, mult i-looked, speckle filt ered
using a Gamma-Gamma MAP filt er,
ellipsoidal geocoded in t he WGS-84
(geographic reference syst em), and finally
radiomet rically calibrat ed. Not e t he st rong
bright ness variat ions from near (right ) t o
far range (left ).
Normalized Backscat t ering Coefficient - ENVI SAT ASAR Global Mode
The pict ure shows t he normalized
o
) est imat ed
from an ENVI SAT ASAR Global Mode
image (405 km swat h) of an area
covering t he I vory Coast , Mali, Burkina
Faso, and Maurit ania (Africa). The image,
which has a pixel spacing of 1km, has
been focused, mult i-looked, speckle
filt ered using a Gamma-Gamma MAP
filt er, ellipsoidal geocoded in t he WGS-84
(geographic reference syst em), and finally
radiomet rically calibrat ed. Not e t he
bright ness homogeneit y aft er t he radio-
met ric normalizat ion procedure.
2.1.7 Radi omet r i c 2.1.7 Radi omet r i c Nor mal i zat i on Nor mal i zat i on
SARscape SARscape

2.1.8 2.1.8 Mosai ci ng Mosai ci ng
Purpose
Terrain geocoded, radiomet rically calibrat ed, and radiomet rically normalized backscat t ering
coefficient (o
o
) dat a acquired over different sat ellit e pat hs/ t racks are usually mosaiced, making
it possible t o cover large areas (count ry t o cont inent al level) - usually wit h dat a of high spat ial
resolut ion.
I t is wort h ment ioning t hat t he monost at ic nat ure of t he syst em and t he penet rat ion
capabilit ies t hrough t he clouds means t hat it is not necessary t o correct for sun inclinat ion or
at mospheric variat ions. This is t he great advant age of SAR over opt ical sensors, primarily
because large areas wit h t he same illuminat ion geomet ry can be imaged exact ly in t he same
way regardless of weat her condit ions or seasonal variat ions in sun angle. This great ly
facilit at es t he det ect ion of land cover changes.
JAXA
Mosaic of 1250 JERS-1 SAR
ellipsoidal geo-coded images
of Cent ral Africa. Not e t hat
t he mosaicing procedure has
been complet ely aut omat ic.
The mosaicing followed
ellipsoidal geocoding of each
single high resolut ion scene,
and was obt ained using t he
following met hodology:
Cent ral Africa imaged by JERS-1 SAR dat a
The amplit ude rat io is
est imat ed bet ween nei-
ghbouring images in t he
overlapping area.
The correct ion fact ors are
obt ained by means of a
global opt imizat ion.
Malawi imaged by mult i-t emporal ENVI SAT ASAR and ALOS PALSAR dat a
The color composit e
on t he left illust rat es
a mult i-t emporal dat a
set based on
ENVI SAT ASAR AP
(120 images) and
ALOS PALSAR FBS
(70 scenes) dat a
covering t he whole
Malawi (100,000 km
2
,
15m resolut ion). The
image on t he right
shows an int erfero-
met ric color compo-
sit e based on ALOS
PALSAR FBS dat a
(70 image pairs). All
processing has been
performed st art ing
from raw dat a.
SARscape SARscape

2.1.9 Segment at i on 2.1.9 Segment at i on
Purpose
Segment at ion assumes t hat images are made up of regions, separat ed by edges, in which
t he radar reflect ivit y is const ant . The number and posit ion of t hese segment s and t heir
mean values are unknown and must be det ermined from t he dat a. I nst ead of at t empt ing t o
reconst ruct an est imat e of t he radar reflect ivit y for each pixel, segment at ion seeks t o
overcome speckle by ident ifying regions of const ant radar reflect ivit y.
Edge Det ect ion
The principle select ed is derived from t he Canny edge det ect or. The advant age of t his
algorit hm is t hat it does not rely on st at ist ical a-priori on regions dist ribut ions and it is purely
based on an analyt ical scheme for det ect ing cont ours. Moreover, such a det ect or is valid for
all possible edge orient at ions and subj ect t o a limit ed number paramet er est imat ion.
For single image, t he Canny edge det ect or consist s in t he sequent ial execut ion of five st eps.
First t he image is smoot hed t o eliminat e and noise. I t t hen finds t he image gradient t o
highlight regions wit h high spat ial derivat ives. The algorit hm t hen t racks along t hese regions
and suppresses any pixel t hat is not at t he maximum (non-maximum suppression) in t he
gradient direct ion. The gradient array is furt her reduced by hyst eresis t hresholding.
I n order t o process vect or-valued dat a, t he algorit hm present ed for single image is ext ended
using t he similar approach used for vect or-valued diffusion filt ering. I n t he vect or-valued dat a
case, t he main difference wit h original Canny lies in t he comput at ion of t he direct ion and
magnit ude of gradient .
Edge Det ect ion - Example
Edge det ect ion is applied in
conj unct ion wit h t he anisot ropic non-
linear diffusion filt er. The t wo images
on t he t op show int ensit y dat a aft er
t he mult i-t emporal anisot ropic non-
linear diffusion filt ering st ep. I n t he
t wo images on t he bot t om, t he mult i-
t emporal edge map obt ained wit h
t he Canny ext ension has been super-
posed t o t he original unfilt ered
int ensit ies images.
SARscape SARscape

Region Merging
The approach is based on an hypot het ical image coding scheme t hat decomposes a given
image int o homogeneous segment s and encodes t hese segment s independent ly of each ot her.
Such an approach also support s sit uat ions in which t he encoder and t he decoder have some
common informat ion about t he image, i.e. in our case an edge map ext ract ed from t he filt ered
images, t o improve segment at ion. Thus, t he ent ire coding procedure can be t hen described as
a t wo-part source channel coding wit h side informat ion. The final region delineat ion is achieved
by a region growing st ep sat isfying t he minimum code lengt h const raint described.
The final st ep for achieving t he required segment at ion is t o find t he segment at ion
corresponding t o a minimal cost wit h t he previous equat ion. To t his end, a region growing
st age is t aken. First , images are divided in one-pixels wide dist inct regions. Then, all t he
merging cost s bet ween 4-connect ed regions are comput ed and st ored in a list . Hence, t he
merging operat ion corresponding t o t he lower cost can be select ed. Finally, t he merge list is
updat ed. The only paramet er int ervening here is t he number of merge operat ions t o perform.
This is a crucial mat t er as it defines t he final scale for segment at ion.
Region Merging - Example
The Figure below shows t he segment at ion images mult i-t emporal segment at ion achieved at
t wo scales, namely wit h 150 and 250 segment s respect ively.
SARscape SARscape

Time Varying Segment at ion
Aft er t he mult i-t emporal region growing st ep a common edge and segment at ion map is
available. However, such global maps are not appropriat e, if single-dat e specific changes (i.e.
anomalies) should be ident ified. I n order t o fully exploit t he mult i-image segment at ion, global
region maps are considered as a st art ing point and process each segment obt ained separat ely
using t he single-image informat ion.
Short y summarized single-dat e specific changes are obt ained by:
1. Consider each region R at each dat e.
2. Est imat e t he number of classes by means of t he Expect at ion Maximizat ion algorit hm.
3. Segment R using an agglomerat ive segment at ion process using t he opt imal number of
classes founded.
4. Time Varying Segment at ion result s from t he union of ident ified segment s in R.
Time Varying Segment at ion - Example
The Figure below illust rat es t he result of t he t ime varying segment at ion algorit hm for a
part icular area. The pict ures on t he t op are t he original unfilt ered images, whereas t hose on
t he bot t om are t he t ime varying segment ed ones.
SARscape SARscape

2.1.10 Cl assi f i cat i on 2.1.10 Cl assi f i cat i on
Purpose
The select ion of a classificat ion t echnique depends very much on t he level of det ail required in
t he result ing map. Whilst relat ively simple classificat ion algorit hms and a choice of dat a sources
may be used for maps cont aining a limit ed number of basic classes, more sophist icat ed
algorit hms must be used when more numerous and / or specific classes are select ed, and not
every dat a source may be suit able. This fact arises from t he concept of classificat ion it self: a
class is oft en clear and underst andable t o a human user, but may be very difficult t o define wit h
obj ect ive paramet ers.
Supervised or Unsupervised?
Looking at t he exist ing classificat ion algorit hms, a first dist inct ion may be made based on t he
involvement of an operat or wit hin t he classificat ion process, i.e.
Super vi sed Al gor i t hms require a human user t hat defines t he classes expect ed as
result of t he classificat ion and reference areas t raining set s t hat can be considered as
represent at ive of each class. I n a similar way, an operat or may define a set of rules t hat ,
if fulfilled, will result in a pixel assigned t o one of t he possible classes.
Unsuper vi sed Al gor i t hms aut onomously analyse t he input dat a and aut omat ically
ident ify groups of pixels wit h similar st at ist ical propert ies.
Supervised Classifiers - Overview
A variet y of aut omat ed met hods have been developed over t he past t wo decades in order t o
classify Eart h Observat ion dat a int o dist inct cat egories. Decision Trees, Maximum Likelihood
(ML), which considers t he appropriat e Probabilit y Densit y Funct ions, and Neural Net works
(NN) have been proven t o be t he most useful (see Figure). Recent ly algorit hms based on t he
principle of Support Vect or Machines (non-paramet ric) have also shown a significant pot ent ial
in t his field.
I t is wort h ment ioning t hat over t ime ML and NN met hods have been ext ended from a pixel
based t o a cont ext ual (i.e. pixel neighbourhood informat ion) approach by considering, for
inst ance, Markov Random Fields (MRF), which provide a flexible mechanism for modelling
spat ial dependence.
Decision Trees
Maximum Likelhood
Neural Net works
Supervised
paramet ric
non-paramet ric (dist ribut ion free)
Supervised Classifiers - Maximum Likelihood
Maximum Likelihood (ML) est imat ion is a mat hemat ical expression known as t he Likelihood
Funct ion of t he sample dat a. Loosely speaking, t he likelihood of a set of dat a is t he probabilit y
of obt aining t hat part icular set of dat a, given t he chosen probabilit y dist ribut ion model. This
expression cont ains t he unknown model paramet ers. The values of t hese paramet ers t hat
maximize t he sample likelihood are known as t he Maximum Likelihood Est imat es (MLE).
The advant ages of t his met hod are:
ML provides a consist ent approach t o paramet er est imat ion problems. This means t hat MLE
can be developed for a large variet y of est imat ion sit uat ions.
ML met hods have desirable mat hemat ical and opt imalit y propert ies. Specifically,
i) t hey become minimum variance unbiased est imat ors as t he sample size increases; and
ii) t hey have approximat e normal dist ribut ions and approximat e sample variances t hat can
be used t o generat e confidence bounds and hypot hesis t est s for t he paramet ers.
Supervised Classifiers - Maximum Likelihood (cont .)
The disadvant ages of t his met hod are:
The likelihood equat ions need t o be specifically worked out for a given dist ribut ion and
est imat ion problem. The mat hemat ics is oft en non-t rivial, part icularly if confidence int ervals
for t he paramet ers are desired.
The numerical est imat ion is usually non-t rivial.
ML est imat es can be heavily biased for small samples. The opt imalit y propert ies may not
apply for small samples.
ML can be sensit ive t o t he choice of st art ing values.
Supervised Classifiers - Decision Tree
Decision t ree classificat ion t echniques have been successfully used for a wide range of
classificat ion problems. These t echniques have subst ant ial advant ages for Eart h Observat ion
classificat ion problems because of t heir flexibilit y, int uit iveness, simplicit y, and comput at ional
efficiency. As a consequence, decision t ree classificat ion algorit hms are gaining increased
accept ance for land cover problems. For classificat ion problems t hat ut ilise dat a set s t hat are
bot h well underst ood and well behaved, classificat ion t ree may be defined solely on analyst
expert ise.
Commonly, t he classificat ion st ruct ure defined by a decision t ree is est imat ed from t raining
dat a using st at ist ical procedures. Recent ly, a variet y of works has demonst rat ed t hat decision
t ree est imat es from t raining dat a using a st at ist ical procedure provide an accurat e and efficient
met hodology for land cover classificat ion problems in Eart h Observat ion. Furt her, t hey require
no assumpt ion regarding t he dist ribut ion of input dat a and also provide an int uit ive
classificat ion st ruct ure.
Supervised Classifiers - Neural Net works
NN are supervised classificat ion t echniques, which can model t he complex st at ist ical
dist ribut ion - as in t he SAR case - of t he considered dat a in a proper way. The main
charact erist ics of NN are
A non-linear non-paramet ric approach capable of aut omat ically fit t ing t he complexit y of t he
different classes
Fast in t he classificat ion phase
Different models of NNs have been proposed. Among t hem, t he most commonly used are t he
Mult i-Layer Percept ron (MLP) and t he Radial Basis Funct ion (RBF) neural net works. Despit e t he
MLPs neural net works t rained wit h t he error back-propagat ion (EBP) learning algorit hm are t he
most widely used, t hey exhibit import ant drawbacks and limit at ions, such as:
The slow convergence of t he EBP learning algorit hm
The pot ent ial convergence t o a local minimum
The inabilit y t o det ect t hat an input pat t ern has fallen in a region of t he input space wit hout
t raining dat a
Supervised Classifiers - Neural Net works (cont .)
RBF Neural Net works overcome some of t he above problems by relying on a rapid t raining
phase and by present ing syst emat ic low responses t o pat t erns t hat fall in regions of t he input
space wit h no t raining samples. This propert y is very import ant because it is st rongly relat ed t o
t he capabilit y of rej ect ing crit ical input samples. Moreover, t he simple and rapid t raining phase,
makes t he RBF neural net works an efficient and suit able t ool also from t he user point of view.
As previously ment ioned, assuming t hat t he geomet rical resolut ion of t he sensors is sufficient ly
high, t he MRF t echnique can be addit ionally applied, giving a class label for each pixel t hat also
considers t he correlat ion among neighboring pixels. I n t his way, it is possible t o obt ain an
improvement in t he performances in t erms of robust ness, accuracy and reliabilit y.
Unsupervised Classifiers - K-Means
K-Means unsupervised classificat ion calculat es init ial class means evenly dist ribut ed in t he dat a
space t hen it erat ively clust ers t he pixels int o t he nearest class using a minimum dist ance
t echnique. Each it erat ion recalculat es class means and reclassifies pixels wit h respect t o t he
new means. All pixels are classified t o t he nearest class unless a st andard deviat ion or dist ance
t hreshold is specified, in which case some pixels may be unclassified if t hey do not meet t he
select ed crit eria. This process cont inues unt il t he number of pixels in each class changes by
less t han t he select ed pixel change t hreshold or t he maximum number of it erat ions is reached.
Example
The pict ure on t he left shows a colour composit e of int erferomet ric t errain geocoded ERS-1/ 2
(so-called ERS-Tandem) SAR int erferomet ric images from t he area of Morondava
(Madagascar). The colours correspond t o t he int erferomet ric coherence (red channel), mean
amplit ude (green channel), and amplit ude changes (blue channel). On t he right side, t he
obt ained land cover classificat ion based on NN met hod is illust rat ed.
Mult i-Temporal Analysis - Overview
The use of mult i-t emporal SAR dat a for t he generat ion of land cover maps - and changes in
part icular - has increasingly become of int erest for t wo reasons:
Wit h mult i-t emporal SAR dat a t he radiomet ric qualit y can be improved using an appropriat e
mult i-t emporal speckle filt er (or segment at ion) wit hout degrading t he spat ial resolut ion;
Classes and changes can be ident ified by analysing t he t emporal behaviour of t he
backscat t ering coefficient and/ or int erferomet ric coherence.
I n essence, change det ect ion t echniques can be dist inguished int o:
Changes t hat can be det ect ed by using det erminist ic approaches.
Changes t hat can be det ect ed by using supervised classifiers (ML, NN, et c.).
Condit io sine qua non for t he exploit at ion of mult i-t emporal analysis is t hat SAR dat a are
rigorously geomet rically and radiomet rically calibrat ed and normalized, in order t o enable a
correct comparison.
Mult i-t emporal Analysis - Det erminist ic Approach
The basic idea of a det erminist ic approach is t he analysis of reflect ivit y changes in t he acquired
SAR dat a over t ime. Measurement s of t emporal changes in reflect ivit y relat es t o t ransit ions in
roughness (for inst ance changes in t he surface roughness due t o emergence of crop plant s) or
dielect ric changes (for inst ance drying of t he plant , or frost event ).
I t is wort h ment ioning t hat t his t ype of classifier can be used in a pixel-based as well as on an
area-based way. Typical applicat ions of t he met hod are in t he domain of agricult ure, but it can
be ext ended t o ot her applicat ions relat ed t o land dynamics mapping (flooding, forest ry, et c.).
A simple but efficient way t o det ermine t he magnit ude of changes in a mult i-t emporal dat a set
is t o perform a rat io bet ween consecut ive images.
Example
The pict ure on t he left shows a colour composit e of ellipsoidal geocoded ERS-2 SAR mult i-
t emporal images of t he Mekong River Delt a (Viet nam). The images have been focused, mult i-
looked, co-regist ered, speckle filt ered using a mult i-t emporal speckle filt ering, geomet rically
and radiomet rically calibrat ed and normalized. The mult it ude of colours in t he mult i-t emporal
image corresponds t o t he variet y of rice cropping syst ems pract ised in t his area. Using a
det erminist ic classifier t he different plant ing moment s and corresponding crop growt h st ages
have been det ermined (on t he right ).
SARscape SARscape

ENVI ENVI

Funct ion Funct ion
2.2.1 Baseline Est imat ion 2.2.1 Baseline Est imat ion
2.2.2 2.2.2 I nt erferogram I nt erferogram Generat ion Generat ion
2.2.3 Coherence and Adapt ive Filt ering 2.2.3 Coherence and Adapt ive Filt ering
2.2.4 Phase Unwrapping 2.2.4 Phase Unwrapping
2.2.5 Orbit al Refinement 2.2.5 Orbit al Refinement
2.2.6 Phase t o Height Conversion and 2.2.6 Phase t o Height Conversion and Geocoding Geocoding
2.2.7 Phase t o Displacement Conversion 2.2.7 Phase t o Displacement Conversion
2.2 2.2 I nt er f er omet r i c I nt er f er omet r i c SAR ( SAR ( I nSAR I nSAR) Pr ocessi ng ) Pr ocessi ng
St r i pMap St r i pMap and and Spot Li ght Spot Li ght Mode Mode
ScanSAR ScanSAR Mode Mode
2.2.1 Basel i ne Est i mat i on 2.2.1 Basel i ne Est i mat i on
General
The difference r
1
and r
2
(Ar) can be measured by t he
phase difference (|) bet ween t wo complex SAR images.
This is performed by mult iplying one image by (t he
complex conj ugat e of) t he ot her one, where an
int erferogram is formed. The phase of t he int erferogram
cont ains fringes t hat t race t he t opography like cont our
lines.
Crit ical Baseline
The generat ion of an int erferogram is only possible when t he ground reflect ivit y acquired wit h
at least t wo ant ennae overlap. When t he perpendicular component of t he baseline (B
n
)
increases beyond a limit known as t he crit ical baseline, no phase informat ion is preserved,
coherence is lost , and int erferomet ry is not possible.
The crit ical baseline Bn,cr, can be calculat ed as
Bn,cr = R t an(u)
2 Rr
where Rr is t he range resolut ion, and u

is t he incidence angle. I n case of ERS sat ellit es t he
crit ical baseline is approximat ely 1.1 km.
The crit ical baseline can be significant ly reduced by surface slopes t hat influence t he local
incidence angle.
SARscape SARscape

- - I nt erferomet ric Module I nt erferomet ric Module
2.2.2 2.2.2 I nt er f er ogr am I nt er f er ogr am Gener at i on Gener at i on
General
Aft er image co-regist rat ion an int erferomet ric phase (|) is generat ed by mult iplying one image
by t he complex conj ugat e of t he second one. A complex int erferogram (I nt ) is formed as
schemat ically shown in t he illust rat ion below.
P
1
: R
1
= R
2
P
0
: R
1
< R
2
P
2
: R
1
> R
2
I nt = S
1
. S
2
*
| = 4t (R
1
- R
2
) /
Click left mouse-but t on t o visualise t he illust rat ion
|
R
S
1
R
1
R
2
P
1
P
2
P
0
S
2
B
u
Spect ral Shift - Principle
The t wo ant enna locat ions used for image acquisit ion induce not only t he int erferomet ric
phase but also shift t he port ions of t he range (ground reflect ivit y) spect ra t hat are acquired
from each point . The ext ent of t he spect ral shift can be approximat ed by
where o

is t he local ground scope, and u

t he incidence angle of t he mast er image. This
frequency (Af) shift must be t hereby compensat ed.
) tan( ) tan(
0 0
o u o u
u
A
= A
R
B
f f f
n
S
1
S
2
Common Doppler Filt ering - Principle
Just as t he range spect ra become shift ed due t o variable viewing angles (S
1
and S
2
in t he
illust rat ion) of t he t errain, different Doppler can cause shift ed azimut h spect ra. As a
consequence an azimut h filt er applied during t he int erferogram generat ion is required t o fully
capt ure t he scenes pot ent ial coherence at t he cost of poorer azimut h resolut ion.
Orbit direct ion
Look direct ion
I nt erferogram - Example
The Figure illust rat es t wo ENVI SAT ASAR images of t he Las Vegas area (USA). Not e t hat t he
t wo images have been acquired wit h a t ime int erval of 70 days. The t wo scenes, which in
I nSAR j argon are defined as Mast er and Slave image, are in t he slant range geomet ry and are
used t o generat e t he int erferomet ric phase, corresponding int erferogram, and int erferomet ric
coherence.
Mast er image Slave image
I nt erferogram - Example
The Figure illust rat es t he int erferogram
generat ed from t he t wo ENVI SAT ASAR
images (previous page). I n essence, t he
complex int erferogram is a pat t ern of
fringes cont aining all of t he informat ion on
t he relat ive geomet ry. The colours (cyan t o
yellow t o magent a) represent t he cycles
(modulo 2t) of t he int erferomet ric phase.
Due t o t he slight ly different ant ennae
posit ions, a syst emat ic phase difference
over t he whole scene can be observed. I n
order t o facilit at e t he phase unwrapping,
such low frequency phase differences are
subsequent ly removed (refer t o t he next
slides).
I nt erferogram Flat t ening
I n preparat ion for t he phase unwrapping st ep t o come, t he expect ed phase, which is
calculat ed using a syst em model, is removed, producing a flat t ened int erferogram, t hat is
easier t o unwrap.
Neglect ing t errain influences and Eart h curvat ure, t he frequency t o be removed can be
est imat ed by t he int erferogram it self. However, t he most accurat e models for removal of t he
fringes are t hose t hat est imat e t he expect ed Eart h phase by assuming t he shape of t he Eart h
is an ellipsoid or, more accurat ely, by using a Digit al Elevat ion Model.
I nt erferogram Generat ion Flat t ening wit h Ellipsoid - An Example
flat t ened assuming t he Eart hs surface an
ellipsoid. I f compared t o t he init ial
int erferogram, it is evident t hat t he
number of fringes has been st rongly
reduced, hence making it possible t o
facilit at e t he phase unwrapping process
and t he generat ion of t he Digit al Elevat ion
Model or, in case of different ial
int erferomet ry, t he measurement s of
ground mot ions.
2.2.2 2.2.2 I nt er f er ogr am I nt er f er ogr am Fl at t eni ng Fl at t eni ng
I nt erferogram Flat t ening wit h Digit al Elevat ion Model - An Example
flat t ened by considering a low resolut ion
Digit al Elevat ion Model, i.e. t he t opography.
I f t his is compared t o t he init ia l
int erferogram, or t o t he ellipsoidal flat t ened
one, it is evident t hat t he number of fringes
has been st rongly reduced, hence facilit at ing
t he phase unwrapping process and t he
generat ion of t he Digit al Elevat ion Model or,
in case of different ial int erferomet ry, t he
measurement s of ground mot ions.
SARscape SARscape

2.2.3 Coher ence and Adapt i v e Fi l t er i ng 2.2.3 Coher ence and Adapt i v e Fi l t er i ng
Purpose
Given t wo co-regist ered complex SAR images (S
1
and S
2
), one calculat es t he int erferomet ric
coherence () as a rat io bet ween coherent and incoherent summat ions:
Not e t hat t he observed coherence - which ranges bet ween 0 and 1 - is, in primis, a funct ion of
syst emic spat ial decorrelat ion, t he addit ive noise, and t he scene decorrelat ion t hat t akes place
bet ween t he t wo acquisit ions.
I n essence coherence has, in primis, a t wofold purpose:
2
2
2
1
*
2 1
) ( ) (
) ( ) (

=
x s x s
x s x s
To det ermine t he qualit y of t he measurement (i.e. int erferomet ric phase). Usually, phases
having coherence values lower t han 0.2 should not be considered for t he furt her
processing.
To ext ract t hemat ic informat ion about t he obj ect on t he ground in combinat ion wit h t he
o
).
Coherence - An Example
The Figure illust rat es t he est imat ed coherence.
Bright values correspond t o values approaching
t o 1, while dark values (black = 0) are t hose
areas where changes (or no radar ret urn, radar
facing slope, et c.) occurred during t he t ime
int erval, 70 days in t his case. Not e t hat
coherence is sensit ive t o microscopic obj ect
propert ies and t o short -t erm scat t er changes.
I n most cases, t he t hemat ic informat ion
cont ent decreases wit h increasing acquisit ion
int erval, mainly due t o phenological or man-
made changes of t he obj ect or weat her
condit ions. Since t he select ed sit es are locat ed
in dry areas, high coherence informat ion is
observed even over long t imescales.
Filt ered DEM Flat t ened I nt erferogram - An Example
The Figure illust rat es t he filt ered
int erferogram aft er an adapt ive filt ering.
Compare t his pict ure wit h t he unfilt ered one
(I nt erferogram Flat t ening wit h Digit al
Elevat ion Model). The basic idea of t his
adapt ive filt ering is t o use t he coherence
values in order t o obt ain irregular windows
and t hus specifically filt er t he different
feat ures.
SARscape SARscape

2.2.4 Phase Unw r appi ng 2.2.4 Phase Unw r appi ng
Purpose
The phase of t he int erferogram can only be modulo 2t. Phase Unwrapping is t he process t hat
resolves t his 2t

ambiguit y. Several algorit hms (such as t he branch-cut s, region growing,
minimum cost flow, minimum least squares, mult i-baseline, et c.) have been developed. I n
essence, none of t hese are perfect , and depending on t he applied t echnique some phase
edit ing should be carried out in order t o correct t he wrong unwrapped phases. The most
reliable t echniques are t hose in which different algorit hms are combined.
I nt erferomet ric Phase
The int erferomet ric phase (|) is expressed as |

= t an(I maginary(I nt ) / Real(I nt )), modulo 2t.
Absolut e Phase
t
t
Wrapped Phase
t
t
I n order t o resolve t his inherent ambiguit y, phase unwrapping must be performed.
Unwrapped Phase - An Example
The Figure illust rat es t he unwrapped phase.
At t his st age t he grey levels represent ing
t he phase informat ion are relat ive and must
be absolut ely calibrat ed in order t o convert
it t o t errain height . Not e t hat no grey level
discont inuit ies can be observed. This
indicat es t hat t he phase unwrapping
process has been correct ly performed.
SARscape SARscape

2.2.5 Or bi t al Ref i nement 2.2.5 Or bi t al Ref i nement
Purpose
The orbit al refinement is crucial for a correct t ransformat ion of t he phase informat ion int o
height values. I n essence, t his procedure, which requires t he use of some accurat e Ground
Cont rol Point s, makes it possible t o
i) calculat e t he phase offset (hence allowing t o calculat e t he absolut e phase), and
ii) refine t he orbit s and t hus obt ain a more accurat e est imat e of t he orbit s and t he
corresponding baseline. Generally, t his st ep is performed by t aking int o account
Shift in azimut h direct ion
Shift in range direct ion
Convergence of t he orbit s in azimut h direct ion
Convergence of t he orbit s in range direct ion
Absolut e phase
Filt ered DEM Flat t ened I nt erferogram aft er Orbit al Refinement - Example
The Figure illust rat es t he filt ered DEM
flat t ened int erferogram aft er t he orbit al
correct ion. Compare it wit h t he non
correct ed one. From a visual comparison it
is clear how fringes t hat have been induced
by inaccurat e orbit s may be removed in
order t o obt ain a proper int erferogram.
SARscape SARscape

2. 2. 6 Phase t o Hei ght Conv er si on and 2. 2. 6 Phase t o Hei ght Conv er si on and Geocodi ng Geocodi ng
Purpose
The absolut e calibrat ed and unwrapped phase values are convert ed t o height and direct ly
geocoded int o a map proj ect ion. This st ep is performed in a similar way as in t he geocoding
procedure, by considering t he Range-Doppler approach and t he relat ed geodet ic and
cart ographic t ransforms. The fundament al difference wit h t he geocoding st ep is t hat t he
Range-Doppler equat ions are applied simult aneously t o t he t wo ant ennae, making it possible
t o obt ain not only t he height of each pixel, but also it s locat ion (x,y,h) in a given cart ographic
and geodet ic reference syst em. Formally, t he syst em is:
=
= +
|
.
|
\
|

|
.
|
\
|

=
= +
|
.
|
\
|

|
.
|
\
|

=

|
t
4
0
2
0
0
2
0
2 1
2
2
2 2
1
1
1 1
2
1
R R
f
S P
V V S P
R S P
f
S P
V V S P
R S P
D
S P
D
S P
Digit al Elevat ion Model, Las Vegas (USA) - Example
The Figure illust rat es t he derived Digit al
Elevat ion Model (DEM) in t he UTM
coordinat e syst em. Dark values correspond
t o t he low height values, while bright areas
correspond t o higher elevat ions.
The height accuracy of t he obt ained DEM,
which has a spat ial resolut ion of 20m, is of
+ / - 5 met er.
3D DEM View, Las Vegas (USA) - Example
The Figure illust rat es t he Digit al Elevat ion Model in a 3D view: The colours of t he image are
provided by t he combinat ion of int erferomet ric coherence and backscat t ering coefficient dat a.
SARscape SARscape

2.2.7 Phase t o Di spl acement Conv er si on 2.2.7 Phase t o Di spl acement Conv er si on
Purpose
The t emporal separat ion in repeat -pass int erferomet ry of days, mont hs, or even years, can be
used t o advant age for long t erm monit oring of geodynamic phenomena, in which t he t arget
changes posit ion at a relat ively slow pace, as in t he case of glacial or lava-flow movement s.
However, it is also useful for analysing t he result s of single event s, such as eart hquakes.
I n essence, t he observed phase (|
int
) is t he sum of several cont ribut ions. The obj ect ive of
Different ial int erferomet ry (DI nSAR) is t o ext ract from t he different component s, t he
displacement one (|
Movement
).
Noise Atmosphere Movement Change Topography Int
R R
| | | | |
t | + + + + =
=
2 1
4
Principle
Different ial int erferogram can be generat ed in different ways, as shown below. I n t he first case,
passes 1 (pre event ) and 2 (aft er event ) in combinat ion wit h a DEM (used for subt ract ing t he
t opography induced fringes) are considered. I n t he second case, in order t o avoid t he DEM
generat ion and isolat e movement s associat ed wit h t he event , passes 1 and 2 combined wit h
passes 1 and 3 are t aken int o account .
SAR dat a 1
SAR dat a 2
SAR dat a 3
I nt erferogram 1
I nt erferogram 2
Diff erent ial I nt erf erogram Coherent change
SAR dat a 1
SAR dat a 2
DEM
I nt erferogram 1
I nt erferogram 2
Diff erent ial I nt erf erogram Coherent change
Different ial I nt erferogram, Eart hquake Bam (I ran) - Example
The Figure shows a different ial
int erferogram generat ed from t he ENVI SAT
ASAR pair acquired on 3 December 2003
(pre-eart hquake) and 11 February 2004
(post -eart hquake). An I nSAR DE M
(generat ed from an ERS-Tandem pair) was
used for subt ract ing t he t opography
induced fringes.
I nt erferomet ric Coherence , Eart hquake Bam (I ran) - Example
The Figure illust rat es t he int erferomet ric
coherence relevant t o t he 3 December 2003 /
11 February 2004 ENVI SAT ASAR pair. I t is
import ant t o not e t hat lowest coherence
values (darkest values) correspond bot h t o
st eep slopes or veget at ed areas (especially
visible in t he lower part of t he image) and t o
t he Bam built zone (image cent er), which was
complet ely dest royed by t he eart hquake.
Deformat ion Map, Eart hquake Bam (I ran) - Example
The Figure illust rat es t he deformat ion map
obt ained considering a st rike slip fault
(horizont al movement ) wit h a Nort h -Sout h
orient ed fault plane. Red and green t ones
represent t he areas of largest deformat ion
(in opposit e direct ion), while yellowish areas
correspond t o smaller deformat ion zones.
SARscape SARscape

2.3.1 St at us 2.3.1 St at us
2.3.2 Principle 2.3.2 Principle
2.3.3 2.3.3 Polarimet ric Polarimet ric Calibrat ion Calibrat ion
2.3.4 2.3.4 Polarimet ric Polarimet ric Speckle Filt ering Speckle Filt ering
2.3.5 Polarizat ion Synt hesis 2.3.5 Polarizat ion Synt hesis
2.3.6 2.3.6 Polarimet ric Polarimet ric Signat ure Signat ure
2.3.7 2.3.7 Polarimet ric Polarimet ric Feat ures Feat ures
2.3.8 2.3.8 Polarimet ric Polarimet ric Decomposit ion Decomposit ion
2.3.9 2.3.9 Polarimet ric Polarimet ric Classificat ion Classificat ion
2.3 2.3 Pol ari met ri c Pol ari met ri c SAR ( SAR ( Pol SAR Pol SAR) Pr ocessi ng ) Pr ocessi ng
2.3 2.3 Pol ar i met r i c Pol ar i met r i c SAR ( SAR ( Pol SAR Pol SAR) Pr ocessi ng ) Pr ocessi ng
2.3.1 St at us 2.3.1 St at us
Over t he past 20 years t here has been considerable progress in t he applicat ion of polarimet ric
SAR dat a for land observat ion. The following point s cont ribut ed t o est ablish t he PolSAR
t echnique:
I ncreased availabilit y of polarimet ric dat a.
Development in PolSAR dat a processing (i.e. calibrat ion, despeckling t echniques).
Development in PolSAR decomposit ion t echniques (analysis of fundament al scat t ering
propert ies of land surfaces/ volumes).
Development in PolSAR dat a classificat ion.
Nowadays, new PolSAR t echniques make use of model-derived dat a int erpret at ion (forest ry,
agricult ure, ice/ snow surfaces, ).
2.3.2 Pr i nci pl e 2.3.2 Pr i nci pl e
Convent ional SAR syst ems operat e wit hin a single, fixed-polarizat ion ant enna for bot h
t ransmission and recept ion of radio frequency signals. I n t his way a single radar reflect ivit y is
measured, for a specific t ransmit and receive polarizat ion combinat ion, for every resolut ion
element of t he image. A result of t his implement at ion is t hat t he reflect ed wave, a vect or
quant it y, is measured as a scalar quant it y and any addit ional informat ion about t he scat t ering
process cont ained in t he polarizat ion propert ies of t he scat t ered signal is lost . To ensure t hat
all t he informat ion of t he scat t ered wave is ret ained, t he polarizat ion of t he scat t ered wave
must be measured t hrough a vect or measurement process, enabling t o use bot h t he
amplit ude and phase informat ion t o dist inguish bet ween different scat t ering mechanisms. This
differs from ot her approaches, not only because it is more complet e in t he sense t hat more
informat ion is used during t he informat ion ext ract ion process, but in part icular because a
bet t er underst anding of t he radar wave-medium int eract ion can be gained.
More informat ion is available at ht t p: / / envisat .esa.int / polsarpro/ t ut orial.ht ml
2.3.3 2.3.3 Pol ar i met r i c Pol ar i met r i c Cal i br at i on Cal i br at i on
Polarimet ric calibrat ion allows, in general, t o minimize t he impact of non ideal behaviours of a
full-polarimet ric SAR acquisit ion syst em, t o obt ain an est imat e of t he scat t ering mat rix of t he
imaged obj ect s as accurat e as possible from t heir available measurement . Calibrat ion is
applied according t o t he following model:
(
+
(
O O
O O
(
O O
O O
(
=
(
VV VH
HV HH
VV VH
HV HH j
VV VH
HV HH
N N
N N
f S S
S S
f
Ae
Z Z
Z Z
1 2
1
2 4
3
1
cos sin
sin cos
cos sin
sin cos 1
o
o
o
o
|
Scat t ering Mat rix
Faraday Rot at ion Absolut e Calibrat ion
Thermal Noise
Channel Amplit ude I mbalance Cross Talk
2.3.3 2.3.3 Pol ar i met r i c Pol ar i met r i c Cal i br at i on Cal i br at i on
SARscape SARscape

- - Polarimet ry and Polarimet ric Polarimet ry and Polarimet ric- - lI nt erferomet ric SAR Module lI nt erferomet ric SAR Module
2.3.4 2.3.4 Pol ar i met r i c Pol ar i met r i c Speck l e Fi l t er i ng Speck l e Fi l t er i ng
The mult iplicat ive speckle model for t he single polarizat ion case (refer t o Sect ion 1.5) may be
ext ended t o t he polarimet ric case by considering t hat polarimet ric channels are affect ed by
independent mult iplicat ive speckle component s.
Two fully polarimet ric st at ist ically adapt ive speckle filt ers for single or mult i-look polarimet ric
SAR are proposed: t he Wishart -Gamma Maximum A Post eriori, and t he Dist ribut ion-Ent ropy
Maximun A Post eriori filt er. Bot h filt ers use t he same speckle model. Moreover, for low look
correlat ion, t he condit ional probabilit y densit y funct ion of t he covariance mat rix of t he speckle
measurement is modelled by a complex Wishart dist ribut ion.
The complex Wishart -Gamma Maximum A Post eriori uses an a priori knowledge about t he
scene, which is modelled by a Gamma dist ribut ed scalar paramet er equal t o t he normalized
number of scat t erers by resolut ion cell. This model is suit ed t o ensure a sat isfact ory adapt ivit y
of t he filt er t o rest ore most t ext ured scenes at t he usual spaceborne and airborneSAR
resolut ions, when t he det ect ed int ensit y channels can be assumed as being K-dist ribut ed.
2.3.4 2.3.4 Pol ar i met r i c Pol ar i met r i c Speck l e Fi l t er i ng Speck l e Fi l t er i ng
SARscape SARscape

- - Gamma and Gaussian Filt er Module Gamma and Gaussian Filt er Module
2.3.5 Pol ar i zat i on Sy nt hesi s 2.3.5 Pol ar i zat i on Sy nt hesi s
Knowledge of t he scat t ering mat rix (i.e. t he 2x2 complex
element s, where t he diagonal element s are t he co-polar (HH,
VV) t erms, while t he off-diagonal are known as cross-polar
(HV, VH) t erms) permit s est imat ion of t he received power for
any possible combinat ion of t ransmit t ing and receiving
ant ennas (i.e. polarizat ion synt hesis or format ion of t he
scat t ering mat rix in any arbit rary polarizat ion basis). This
t echnique is what gives polarimet ry it s great advant age over
convent ional fixed-polarizat ion radars - more informat ion may
be inferred about t he scat t ering mechanisms on surfaces or
wit hin volumes if t he scat t ering mat rix can be measured.
The pict ure shows a polarizat ion synt hesis based on Shut t le
I maging Radar dat a acquired during t he t hird radar shut t le
mission (SI R-C) over t he area of Flevoland (The Net herlands).
The image has been generat ed by combining t he t hree
polarizat ions HH, VV, and HV in t he red, green, and blue
channels respect ively.
2.3.5 2.3.5 Pol ar i zat i on Sy nt hesi s Pol ar i zat i on Synt hesi s
SARscape SARscape

2.3.6 Pol ar i zat i on Si gnat ur e 2.3.6 Pol ar i zat i on Si gnat ur e
A part icular graphical represent at ion of t he variat ion of backscat t ering as a funct ion of
polarizat ion, known as polarizat ion signat ure (see Figure), is quit e useful for describing
polarizat ion propert ies of a t arget . The response consist s of a plot of synt het ized (and
normalized) scat t ering cross sect ions as a funct ion of t he ellipt icit y and orient at ion angles of
t he received wave.
e
l
l
i
p
t
i
c
i
t
y
o
r
ie
n
t
a
t
i
o
n
p
o
w
e
r
2.3.6 Pol ar i zat i on Si gnat ur e 2.3.6 Pol ar i zat i on Si gnat ur e
SARscape SARscape

2.3.7 2.3.7 Pol ar i met r i c Pol ar i met r i c Feat ur es Feat ur es
The simplest way t o represent t he polarimet ric informat ion cont ained in t he scat t ering or
St okes mat rix is given by calculat ing some co- and cross polarized power, as well as select ed
rat ios. Furt hermore, t hese polarimet ric feat ures enable, t o some ext ent , t o ident ify t he most
suit able combinat ions for classificat ion purposes. They are:
Span = HH+ HV+ VH+ VV
HH HH*
VV VV*
HV HV*
Re { HH VV* }
I m { HH VV* }
Re { HV VV* }
I m { HV VV* }
Re { HH HV* }
I m { HH HV* }
Polarizat ion Rat io = HH HH* / VV VV*
Linear Depolarizat ion Rat io = HV HV* / VV VV*
2.3.7 2.3.7 Pol ar i met r i c Pol ar i met r i c Feat ur es Feat ur es
SARscape SARscape

2.3.8 2.3.8 Pol ar i met r i c Pol ar i met r i c Decomposi t i on Decomposi t i on
The obj ect ive of t he coher ent decomposi t i on is t o express t he measured scat t ering mat rix
S as t he combinat ion of t he scat t ering responses of simpler obj ect s.
The symbol S
i
st ands for t he response of every one simpler obj ect s, whereas c
i
indicat es t he
weight of S
i
in t he combinat ion leading t o t he measured S.
I t has t o be point ed out t hat t he scat t ering mat rix S can charact erise t he scat t ering processes
produced by a given obj ect , and t herefore t he obj ect it self. This is possible only in t hose cases
in which bot h, t he incident and scat t ered waves are complet ely polarized waves.
Consequent ly, coherent t arget decomposit ions can be only employed t o st udy t he coherent
t arget s. These scat t ers are known as point or pure t arget s.
I n a real sit uat ion, t he measured scat t ering mat rix S corresponds t o a complex coherent
t arget . Therefore, in a general sit uat ion, a direct analysis of t he scat t ering mat rix, wit h t he
obj ect ive t o infer t he physical propert ies of t he scat t erer under st udy, is shown very difficult .
Thus, t he physical propert ies of t he scat t er are ext ract ed and int erpret ed t hrough t he analysis
of simpler responses S
i
and corresponding coefficient s c
i
.
The decomposit ion exposed in pervious page in not unique in t he sense t hat it is possible t o
find number of infinit e set s S
i
in which t he scat t ering mat rix S can be decomposed.
Nevert heless, only some of t he set s S
i
are convenient t o int erpret t he informat ion cont ent of
S. Three met hods are employed t o charact erize coher ent scat t er er s based on t he
scat t ering mat rix S :
The Pauli Decomposit ion
The Krogager Decomposit ion
The Cameron Decomposit ion
The scat t ering mat rix S is only able t o charact erize, from a polarimet ric point of view,
coherent scat t erers. On t he cont rary, t his mat rix can not be employed t o charact erize
dist ribut ed t arget s. This t ype of scat t erers can be only charact erized, st at ist ically, due t o t he
presence of speckle noise. Since speckle noise must be reduced, only second order
polarimet ric represent at ions can be used t o analysed dist ribut ed scat t erers. These second
order descript ors are t he 3 by 3, Hermit ian average covariance (C) and t he coherency (T)
mat rices. These t wo represent at ions are equivalent .
The complexit y of t he scat t ering process makes ext remely difficult t he physical st udy of a
given scat t er t hrough t he analysis of C or T. Hence, t he obj ect ive of t he i ncoher ent
decomposi t i ons is t o separat e t he C or T mat rices as t he combinat ion of t he second order
descript ors corresponding t o t o simpler obj ect s, present ing an easier physical int erpret at ion.
These decomposit ion t heorems can be expressed as:
where p
i
and q
i
denot e t he coefficient s of t he component s in C and T .
As in t he case of t he coherent decomposit ion, it is desirable t hat t hese component s present
some propert ies. First at all it is desirable t hat t he component s C
i
and T
i
correspond t o pure
t arget s in order t o simplify t he st udy. I n addit ion t he component s should be independent , i.e.
ort hogonal. The bases in which C orT are not unique. Consequent ly different i ncoher ent
decomposi t i ons can be expressed. They are:
The Freeman Decomposit ion
The Huynen Decomposit ion
The Eigenvect or-Eigenvalue Decomposit ion
I n t he next page an overview of t he different coher ent and i ncoher ent decomposi t i ons is
given.
ESA
SARscape SARscape

2.3.9 2.3.9 Pol ar i met r i c Pol ar i met r i c Cl assi f i cat i on Cl assi f i cat i on
Cloude and Pot t ier proposed an algorit hm t o ident ify in an unsupervised way polarimet ric
scat t ering mechanisms in t he H-o

(Ent ropy-Mean alpha angle) plane. The basic idea is t hat
ent ropy arises as a nat ural measure of t he inherent reversibilit y of t he scat t ering dat a and t hat
t he mean alpha angle can be used t o ident ify t he underlying average scat t ering mechanism.
The H-o

plane is divided in 9 basic zones charact erist ic of different scat t ering behaviours (see
Figure in t he next page). The basic scat t ering mechanism of each pixel can be ident ified by
comparing it s ent ropy and mean alpha angle paramet ers t o fixed t hresholds. The different
class boundaries, in t he H-o

plane, have been det ermined so as t o discriminat e surface
reflect ion, volume diffusion, and double bounce reflect ion along t he o

axis and low, medium,
and high degree of randomness along t he H axis.
ESA
The present ed procedure may be furt her improved by explicit ly including t he anisot ropy
informat ion (see Figure in t he next page). This polarimet ric indicat or is part icularly useful t o
discriminat e scat t ering mechanisms wit h different eigenvalue dist ribut ions but wit h similar
int ermediat e ent ropy values. I n such cases, a high anisot ropy value indicat es t wo dominant
scat t ering mechanisms wit h equal probabilit y and a less significant t hird mechanism, while a
low anisot ropy value corresponds t o a dominant first scat t ering mechanism and t wo non-
negligible secondary mechanisms wit h equal import ance.
ESA
SARscape SARscape

2.4.1 Principle 2.4.1 Principle
2.4.2 Co 2.4.2 Co- - regist rat ion regist rat ion
2.4.3 2.4.3 I nt erferogram I nt erferogram Generat ion Generat ion
2.4.4 2.4.4 Polarimet ric Polarimet ric Coherence Coherence
2.4.5 Coherence 2.4.5 Coherence Opt imizat ion Opt imizat ion
2.4 2.4 Pol ari met ri c Pol ari met ri c- - I nt er f er omet r i c I nt er f er omet r i c SAR ( SAR ( Pol I nSAR Pol I nSAR) Pr ocessi ng ) Pr ocessi ng
2.4 2.4 Pol I nSAR Pol I nSAR Pr ocessi ng Pr ocessi ng
Polarimet ric int erferomet ry and coherence opt imizat ion are an ext ension of convent ional
int erferomet ry t o t he case where t he int erfered pairs are not const it ut ed by single images, but
by full-polarimet ric dat aset s, originat ing t he case of vect or int erferomet ry as opposed t o scalar
int erferomet ry (see Figure next page).
Basically, int erferomet ric SAR measurement s are sensit ive t o t he spat ial dist ribut ion of t he
scat t erers, polarimet ric SAR measurement s are relat ed t o t he orient at ion and/ or dielect ric
propert ies of t he scat t erers. While using lower syst em frequencies (i.e. L- or P-band), t he
combinat ion of t he t wo t echniques allows t he analysis of t he spat ial dist ribut ion of t he
polarimet ric scat t ering mechanisms wit hin scat t ering volumes (veget at ion, snow/ ice, ).
More informat ion is available at ht t p: / / eart h.esa.int / polinsar/
ESA
1. Coarse est imat e based on t he orbit informat ion.
2. Mast er t o slave amplit ude image cross-correlat ion.
3. Fine shift calculat ion based on t he phase informat ion. During t his st ep spect ral shift and
common Doppler bandwidt h filt ering are performed.
I nt erferomet ric and polarimet ric-int erferomet ric processing ext ract s informat ion from pairs of
complex SAR dat a (SLCs). For t his reason SAR image pair must be co-regist ered. The
coregist rat ion process is divided in t wo st eps: i) est imat ion of t he relat ive shift s bet ween a
reference and a slave image, and, ii) int erpolat ion of t he slave image t o accomplish for t he
est imat ed shift s. The shift is est imat ed using t he first image of t he t wo list s as mast er and
respect ively slave image, while t he same est imat ed shift is applied t o t he whole list of slave
images, t o avoid relat ive misalignment s. I t is t o be recommended, t herefore, t hat t he
polarizat ion for which may be expect ed t he higher coherence (usually HH) is provided as first
of t he input file list s.
The shift is calculat ed in 3 st eps:
2.4.2 2.4.2 Co Co - - r egi st r at i on r egi st r at i on
SARscape SARscape

For each resolut ion element t wo co-regist ered scat t ering mat rices are available. The complet e
informat ion measured by t he SAR syst em can be represent ed in form of t hree 3 by 3 complex
mat rices T
11
, T
22
, and O
12
formed formed using t he out er product s of t he scat t ering vect ors k
1
and k
2
as
T
11
= k
T
1
. k
1
T
22
= k
T
2
. k
2
O
12
= k
T
1
. k
2
T
11
and T
22
are t he convent ional polarimet ric coherency mat rices which describe t he
polarimet ric propert ies for each individual image separat ely, and O
12
is a complex mat rix
cont aining polarimet ric and int erferomet ric informat ion. The t wo complex scalar images i
1
and
i
2
for forming t he int erferogram are obt ained by proj ect ing t he scat t ering vect ors k
1
and k
2
ont o t wo unit ary complex vect ors w
1
and w
2
, which define t he polarizat ion of t he t wo images
respect ively as:
i
1
= w
T
1
.
k
1
and i
2
= w
T
2
.
k
2
The int erferogram relat ed t o t he polarizat ions given by w
1
and w
2
is t hen
i
1
i
*
2
= (w
T
1
.
k
1
) (w
T
2
.
k
2
)
T
Two cases should be dist inguished:
w
1
is equal t o w
2
, i.e. images wit h t he same polarizat ion are used t o form an
int erferogram. I n t his case t he int erferomet ric phase cont ains only t he int erferomet ric
const ribut ion dut o t o t he t opography and range variat ion, while t he int erfeormet ric
coherence expresses t he int erferomet ric correlat ion behaviour.
w
1
is not equal w
2
, i.e. images wit h different poalrizat ion are used t o form t he
int erferogram. I n t his case t he int erferomet ric phase cont ains besides t he int erferomet ric
also t he phase difference bet ween t he t wo polarizat ions. The int erferomet ric coherence
expresses apart from t he int erferomet ric correlat ion behaviour also t he polarimet ric
correlat ion bet ween t he t wo polarizat ions

(w
1
, w
2
) =
I nt
.
2
SARscape SARscape

2.4.4 2.4.4 Pol ar i met r i c Pol ar i met r i c Coher ence Coher ence
Based on t he formulat ions obt ained in Sect ions 2.4.3 (I nt erferogram Generat ion) and 2.3.4
(Coherence) t he general formulat ion of t he int erferomet ric coherence is derived as:
where T
11
and T
22
are t he convent ional polarimet ric coherency mat rices which describe t he
polarimet ric propert ies for each individual image separat ely, O
12
is a complex mat rix
cont aining polarimet ric and int erferomet ric informat ion, w
1
and w
2
t he t wo unit ary complex
vect ors.
2.4.4 2.4.4 Pol ar i met r i c Pol ar i met r i c Coher ence Coher ence
SARscape SARscape

2.4.5 Coher ence Opt i mi zat i on 2.4.5 Coher ence Opt i mi zat i on
The dependency of t he int erferomet ric coherence on t he polarizat ion of t he images used t o
form t he int erferogram leads t o consider t he quest ion of which polarizat ion yields t he highest
coherence. I n essence, t he problem is t o opt imize t he general formulat ion of t he
int erferomet ric coherence, i.e.
Aft er t edious algebra, it can be demonst rat ed t hat t he maximum possible coherence value
opt 1
t hat can be obt ained by varying t he polarizat ion is given by t he square root of t he maximum
eigenvalue. Each eigenvalue is relat ed t o a pair of eigenvect ors (w
1i
,w
2i
) one for each image.
The first vect or pair (w
11
,w
21
) represent s t he opt imum polarizat ions. The second and t hird
pairs (w
12
,w
22
) and (w
13
,w
23
), belonging t o t he second and t hird highest singular values,
represent opt imum solut ions in different polarimet ric subspaces.
The t hree opt imum complex coherences can be obt ained direct ly by using t he est imat ed
eigenvalues:
2.4.5 Coher ence Opt i mi zat i on 2.4.5 Coher ence Opt i mi zat i on
SARscape SARscape

3. What ar e Appr opr i at e Land Appl i cat i ons? 3. What ar e Appr opr i at e Land Appl i cat i ons?
Some Ex ampl es Some Ex ampl es
3.1 Agricult ure monit oring 3.1 Agricult ure monit oring
3.2 Aquacult ure mapping 3.2 Aquacult ure mapping
3.3 Digit al Elevat ion Model 3.3 Digit al Elevat ion Model
3.4 Flood mapping 3.4 Flood mapping
3.5 Forest mapping 3.5 Forest mapping
3.6 Geomorphology 3.6 Geomorphology
3.7 Monit oring of Land Subsidence 3.7 Monit oring of Land Subsidence
3.8 Subsidence 3.8 Subsidence Monit oring of Building Sinking Monit oring of Building Sinking
3.9 Rice mapping 3.9 Rice mapping
3.10 Snow mapping 3.10 Snow mapping
3.11 Urban mapping 3.11 Urban mapping
3.12 Wet lands mapping 3.12 Wet lands mapping
3.1 Agr i cul t ur e Moni t or i ng 3.1 Agr i cul t ur e Moni t or i ng
3.1.1 Purpose
Reliable and obj ect ive informat ion on cropped area is import ant t o farmers, local and nat ional
agencies responsible for crop subsidies and food securit y, as well as for t raders and re-
insurers. I n t he past years it has been shown t hat SAR based product s can provide informat ion
on field processing condit ions like ploughing, field preparat ion et c. and on crop growt h st at us
such as plant ing, emerging, flowering, plant mat urit y, harvest t ime, and frost condit ions. These
product s - complement ed wit h opt ical based product s (such as chlorophyll, leaf area index,
salinit y, veget at ion indexes, et c.) - offer a suit e of informat ion which will allow bet t er
management of bot h t he cropped areas but also of t he environment .
3.1.2 Met hod
The basic idea behind t he generat ion of useful informat ion for agricult ure using SAR t echniques
is t he analysis of changes in t he acquired dat a over t ime. Measurement s of t emporal changes
in reflect ivit y relat e in primis
i) t o t ransit ions in surface roughness, which are indicat ive of soil t illage, and, while crop-
specific, useful in furt her specificat ion of crop classes, and lat er
ii) t o t he phenological crops st at us.
3.1 Agr i cul t ur e Moni t or i ng 3.1 Agr i cul t ur e Moni t or i ng
3.1.3 Example
The SAR based product s illust rat ed below show t he different crop growt h moment s during a
crop season in Sout h Africa. Different colours represent different growing st ages at different
dat es at field level.
No data
Before ploughing
After ploughing
Weeds emergence
Weeds removal/frost
Plants just emerged
Plant growing 1
Flowering
Plant drying
Plant growing 2
Full plant development
Full maturiry
sarmap
3. 2 Aquacul t ur e Mappi ng 3. 2 Aquacul t ur e Mappi ng
3.2.1 Purpose
I nvent ory and monit oring of shrimp farms are essent ial t ools for decision-making on
aquacult ure development , including regulat ory laws, environment al prot ect ion and revenue
collect ion. I n t he cont ext of government aquacult ure development policy, much at t ent ion is
focused on t he ident ificat ion and monit oring of t he expansion of shrimp farms. Therefore, t he
availabilit y of an accurat e, fast and mainly obj ect ive met hodology t hat also allows t he
observat ion of remot e areas assumes a great value.
3.2.2 Met hod
The ident ificat ion of shrimp farms on SAR images is based on several element s:
i) t he radar backscat t er received from t he wat er surface of t he ponds and from t heir
surrounding dykes,
ii) t he shape of t he individual ponds, and
iii) t he pat t ern of groups of ponds and t he relat ive direct ion of t he dykes vis--vis t he incoming
radar beam.
The locat ion of shrimp farms is also t ypical, t hus t he analysis of t heir posit ion and of t he former
land use of t he area is necessary t o verify t he ident ificat ion.
3.2.2 Met hod
Based on t his principle, FAO, has developed a met hodology, which considers t he following
processing st eps:
- Pre-processing : Speckle filt ering and geocoding
- Wat er classificat ion : Hist ogram t hresholding
- Boundary det ect ion : Edge det ect ion
- Proximit y analysis : The occurrence of highly reflect ive surfaces around wat er surfaces is an
indicat ion of t he presence of shrimp farms. The proximit y analysis examines t he boundaries
of wat er bodies obt ained from t he classificat ion, up t o a user-specified dist ance, t o locat e
bot h highly reflect ive surfaces in t he classified image and edges in t he Sobel filt ered
images. The proximit y analysis produces t wo "summary images" t hat synt hesise t he shrimp
ponds-relat ed informat ion cont ained in an ERS SAR image. The summary images allow t he
operat or t o locat e t he areas where t here is a great er evidence of t he occurrence of shrimp
farms, and help in t racing t he farms boundaries.
3.2.3 Example
The met hodology is applied t o ERS-2 SAR dat a acquired in 1996, 1998 and 1999 in an area of
Sri Lanka. The map, illust rat ed below, shows t hree classes: Wat er bodies (blue), Shrimp farms
occurring up t o 18 April 1996 (red), and Expansion of shrimp farms up t o 16 Oct ober 1998
(pink).
FAO
3. 3 Di gi t al El ev at i on Model Gener at i on 3. 3 Di gi t al El ev at i on Model Gener at i on
3.3.1 Purpose
The Digit al Elevat ion Model product represent s a reconst ruct ion of t he height of t he surface of
a select ed region. Such surfaces may also be of urban and forest ed areas, where t he t ops of
buildings or t rees are represent ed, respect ively. The product is generat ed over a regular grid of
equidist ant point s, where t he corresponding height is a measure of t he average height wit hin
t he cell. Taking advant age of int erferomet ric t echniques it is possible t o generat e high qualit y
DEM in a semi-aut omat ed fashion.
3.3.2 Met hod
The basic idea for t he generat ion of DEM product s is t he conversion of int erferomet ric phase
informat ion, derived from t wo SAR acquisit ions at different dat es from slight ly different orbit al
posit ions, int o height s. I n order t o obt ain accurat e product s, t he SAR dat a must go t hrough a
dedicat ed processing chain, which carries out dat a focussing, int erferomet ric processing, geo-
referencing and finally mosaicing. For a complet e descript ion of t he met hodology refer t o t he
I nSAR Sect ion.
3.3.3 Product s Accuracy
The achievable resolut ion wit h ERS-Tandem dat a is up t o 25 m on t he horizont al plane (x-y),
while t he achievable height accuracy can be summarized as follows:
5-8 met res in flat -moderat e rolling areas where high t emporal correlat ion is available (dry
areas, sparse veget at ion or wint er condit ions)
10-15 met res in st eep t opography areas where high t emporal correlat ion is available
Worst accuracy in areas where low t emporal correlat ion is available (large forest ed or wat er
areas), where t he DEM may be reconst ruct ed by int erpolat ion.
3.3.4 Example
This DEM of Swit zerland covers an area of approximat ely 41,000 km
2
wit h height s ranging
from 200 m t o 4650 m. This product has been generat ed using 70 ERS scenes, namely 22
descending pairs and 13 ascending pairs. The DEM, which is proj ect ed in t he Swiss
cart ographic syst em (Oblique Mercat or), has a grid size of 25 met res.
sarmap
3.4 Fl ood Mappi ng 3.4 Fl ood Mappi ng
3.4.1 Purpose
Flooding is a maj or hydrological hazard which occurs relat ively frequent ly. During t he last
decade floods have affect ed approximat ely 1.5 billion people - more t han 75% of t he t ot al
number of people report ed as affect ed by nat ural disast ers world-wide. Around t he world t here
are on average about 150 serious floods per year, wit h significant rises in wat er level ranging
from severely overflowing st reams, lakes or reservoirs t o maj or ocean-driven disast ers in
exposed coast al regions. Like drought s, floods are cat ast rophic event s t hat t o some ext ent
follow a nat ural cycle t hat is oft en predict able. Timely informat ion about flood phenomena t hat
may provide st rong indicat ors of a fort hcoming disast er can oft en help t o t rack and ident ify t he
areas t hat will be affect ed most severely.
3.4.2 Met hod
The SAR can easily det ect wat er-covered areas, charact erised by a much lower int ensit y t han
any ot her feat ure in t he surroundings. The main limit at ions are induced by t he presence of
nearby veget at ion cover or t he presence of wind, but change det ect ion t echniques using SAR
acquisit ions from different dat es (in t he normal condit ions and during t he flood), prove t o be a
robust way t o overcome t hese difficult ies.
3.4.3 Example
The map shown on t he right is based on t he
rat ioing of t wo pairs of ENVI SAT ASAR Wide
Swat h scenes (e.g. 25 July 2004 and 23 March
2003) covering t he whole of Bangladesh, and
combining t his wit h informat ion relevant t o t he
t errain height (e.g. a Digit al Elevat ion Model)
during t he classificat ion st ep. The map shows
flooded areas (blue and cyan), permanent wat er
(black), and urban areas (red).
3.4.3 Example
Based on a single ENVI SAT ASAR
Alt ernat ing Polarizat ion (HH/ HV) scene
acquired on August 2003, a map
indicat ing t he river (blue) and ot her
land cover t ypes (such as Agricult ure,
Forest ry, and Urban) has been
produced for t he area of Dresden,
Germany, during a flood event . I t is
wort h ment ioning t hat in t his case t he
act ual ext ent of t he flooded area can
not be est imat ed. However, t his
product can be int egrat ed wit h already
exist ing GI S informat ion, t hus making
it possible t o derive t he ext ent of wat er
covered areas.
Forest
Agricult ure
Wat er
Urban
Ot hers
sarmap
3.5 For est Mappi ng 3.5 For est Mappi ng
3.5.1 Purpose
Eart h Observat ion has demonst rat ed t hat it can offer useful informat ion for forest ry
applicat ions, especially in cases of maj or forest change such as t hat caused by severe st orms
or fires. Using low spat ial resolut ion opt ical imagery from sensors such as NOAA AVHRR, ERS
ATSR-1/ 2, ENVI SAT AATSR and SPOT-VEGETATI ON, act ive fire maps are generat ed for large
regions and even t he whole globe. Given t heir low spat ial resolut ion (approximat ely 1 km),
t hese sensors are not suit able for deriving burn scars. The most popular sensors t o dat e for
responding t o such needs are t he Enhanced Themat ic Mapper (ETM+ ), t he Themat ic Mapper
(TM) bot h on board t he Landsat sat ellit es and SPOT-4/ 5 High Resolut ion inst rument s. Wit h
spat ial resolut ions ranging from 30 t o 5 met res, such imagery can provide accurat e est imat es
of t he burnt area, as it is relat ively easy t o discriminat e burned from non-burned areas, at least
in cert ain veget at ion t ypes. However, t hese inst rument s have a maj or drawback in t hat it can
be difficult t o get imagery in areas wit h long periods of significant cloud cover. This applies in
part icular t o t ropical and boreal regions. I n t hese sit uat ions, t he possibilit y of exploit ing high
spat ial resolut ion (10 t o 25 met res) SAR dat a provides great advant ages over opt ical sensors,
because dat a acquired from act ive microwave syst ems are not affect ed by wat er vapour,
smoke or clouds.
3.5.2 Met hod
The basic met hodology for using SAR dat a t o det ect forest burn scars relies on change
det ect ion, comparing dat a acquired aft er a fire wit h reference dat a obt ained beforehand. Wit h
t his obj ect ive, applicat ions relying on SAR dat a t radit ionally based on amplit ude images may be
fruit fully ext ended by exploit ing int erferomet ric t echniques. A benefit of repeat -pass SAR
int erferomet ry is t he feasibilit y of exploit ing coherence informat ion as well as t he usual
backscat t ering coefficient informat ion and backscat t ering coefficient variat ion bet ween t he t wo
acquisit ion t imes.
Change det ect ion based on amplit ude images may be implement ed by using rat ioing. This
mult i-t emporal change approach, which t racks changes in SAR backscat t er over t ime using
mult iple images, makes it possible t o dist inguish different land-cover t ypes and changes in
t hese, based on t heir unique t emporal signat ures. One drawback of t his met hod is t hat mult iple
acquisit ions are required, and t hat t hey must be available at key point s in t he phenological
cycle of t he different land-cover t ypes. More import ant ly, for mapping changes aft er a disast er
event , images must be available soon aft erwards (maximum 1-2 mont hs), in order t o maximise
t he informat ion ret rieved from t he imagery. Coherence informat ion can add support ing
evidence t o t his approach, and may also give addit ional informat ion suit able t o produce a
reference forest -non forest map and t o help t he subsequent classificat ion.
3.5.3 Example
The figure below shows a colour composit e image from part of Saskat chewan, Canada. Wat er
bodies are shown in dark green and dark blue. To t he left is an area t hat burnt during t he
summer of 1995, clearly visible in cyan. To t he right t he blue-green area is an area t hat burnt
prior t o 1995. I n t he graphic on t he right , normalised SAR values are shown for a forest area
and an area t hat burned in summer 1995. There is an increase of over 6 dB in t he backscat t er
values of t he burned area wit h respect t o forest in t he Oct ober image. Alt hough t his falls again
by January, it remains almost 3 dB above t he backscat t er of unburned forest det ect ed.
3.5.3 Example
Result s of t he semi-aut omat ic burnt area det ect ion are shown in t he figure below. Burnt areas
have been mapped as t hose which burned in 1995 and t hose which burned in previous years.
Burnt area polygons provided by t he Canadian Forest Service are overlaid.
sarmap
3.6 Geomor phol ogy 3.6 Geomor phol ogy
3.6.1 Purpose
Eart h Observat ion dat a are part icularly appropriat e for mapping morphological feat ures such as
morpholineament s, which are import ant fact ors det ermining geomorphological st ruct ures and
geologically act ive areas. Due t o t he monost at ic acquisit ion mode of radar syst ems, SAR
images in part icular are well suit ed for morpholineament mapping.
3.6.2 Met hod
The combinat ion of ort horect ified opt ical (mult i-spect ral) and t errain geocoded SAR images is a
simple and suit able met hodology t o ident ify morpholineament s, uplift / subsidence evidence,
drainage pat t erns, act ive fault syst ems, duct ile st ruct ures, et c. A common way t o generat e
such a product is based on t he t ransformat ion of a mult ispect ral image (e.g. Landsat TM),
which is in t he Red Green Blue (RGB) syst em, int o an I nt ensit y, Hue, and Sat urat ion (I HS)
syst em. Subsequent ly t he SAR image, whose informat ion is primarily represent ed by it s
t ext ure, replaces t he I nt ensit y channel before re-t ransforming int o t he RGB colour syst em.
3.6 Geomor phol ogy 3.6 Geomor phol ogy
3.6.3 Example
The figure below shows an ERS-1 (left ) and corresponding Themat ic Mapper (cent re) image of
an arid area in Sudan. On t he right part t he SAR and mult i-spect ral dat a have been merged by
means of a colour t ransform (RGB t o I HS). This example highlight s how opt ical-radar dat a
fusion can significant ly enhance t he informat ion cont ent for morphological and geological
mapping.
sarmap
3. 7 Moni t or i ng of Land Subsi dence 3. 7 Moni t or i ng of Land Subsi dence
3.7.1 Purpose
Ground subsidence is a phenomenon caused by nat ural or man-induced compact ion of
unconsolidat ed sediment s. I t s effect is a sinking of t he ground surface. I n many cases t his is a
consequence of t he ext ract ion of ground wat er, geot hermal fluids, coal, gold and ot her general
mining act ivit y. Due t o t he rapidly increasing use of underground nat ural resources such as
wat er, oil and gas, most of t he maj or subsidence areas around t he world have developed at
accelerat ed rat es in t he past years. Two t radit ional met hods for det ect ing t his are based on
direct measurement s, such as levelling and GPS t echniques. However, t here are t wo import ant
problems in t he use of t hese approaches:
i) t he cost of t he inst rument at ion, and
ii) t he difficult y t o ext rapolat e point measures over wide areas.
3.7.2 Met hod
The basic idea for t he generat ion of land subsidence product s is t he conversion of different ial
int erferomet ric phase informat ion, derived from t hree or more SAR acquisit ions at different
dat es from slight ly different orbit al posit ions, int o displacement s (so-called convent ional
DI nSAR t echnique) as present ed in t he DI nSAR Sect ion.
3. 7 Moni t or i ng of Land Subsi dence 3. 7 Moni t or i ng of Land Subsi dence
3.7.3 Example
The figure shows a land subsidence map from an area of Algeria, which has been produced
using ERS-1 and 2 SAR dat a acquired in t he period 1992-2000. I n t his case t he land subsidence
is due t o significant wat er ext ract ion act ivit ies. What is visible in t he figure below, is a general
subsidence t rend t hat crosses t he cent er of t he image in t he NW-SE direct ion. I t is wort h
not ing t hat t he most of t he wells (blue crosses) are dist ribut ed over t his subsiding area.
s
a
r
m
a
p
3. 8 Moni t or i ng of Bui l di ng Si nk i ng 3. 8 Moni t or i ng of Bui l di ng Si nk i ng
3.8.1 Purpose
As for land subsidence.
3.8.2 Met hod
The Permanent Scat t erer (PS) met hod (PSI nSAR
TM
), developed by Polit ecnico di Milano
(POLI MI ) and TRE (a POLI MI spin-off), is a new approach int roduced t o improve t he abilit y t o
det ermine mm-scale displacement s of individual feat ures on t he ground. I t uses all dat a
collect ed by a SAR syst em over t he t arget area. As long as a significant number and densit y of
independent radar-bright and radar-phase st able point s (i.e. permanent scat t erers) exist wit hin
a radar scene and enough radar acquisit ions have been collect ed, displacement t ime series and
range-change rat es can be calculat ed.
Using t he PS met hod, surface mot ions can be resolved at a level of ~ 0.5 mm/ yr. This resolves
very small-scale feat ures, including mot ions of individual t arget s/ st ruct ures (e.g. a bridge or a
dam), not previously recognised in convent ional DI nSAR. PS usually correspond t o buildings,
met allic obj ect s, out crops, exposed rocks, et c. exhibit ing a radar signat ure t hat is const ant
wit h t ime. Once t hese radar benchmarks have been ident ified from a t ime series of dat a, very
accurat e displacement hist ories can be obt ained for t he period 1992 t o t he present . The effect
is akin t o suddenly having a dense GPS net work ret rospect ively available for t he last t en years
in any moderat ely urbanised area (at least areas where ERS SAR dat a have been collect ed).
3. 8 Moni t or i ng of Bui l di ng Si nk i ng 3. 8 Moni t or i ng of Bui l di ng Si nk i ng
3.8.3 Example
The figure on t he right
shows t he average
deformat ion velocit y of
radar benchmarks (PS)
ident ified in an area of Paris
(France) by using ERS-1/ 2
SAR dat a from 1992 t o
2003. Displacement s are
measured along t he
direct ion of t he sensor
t arget conj unct ion.
TRE
3. 9 Ri ce Mappi ng 3. 9 Ri ce Mappi ng
3.9.1 Purpose
Rice is t he most import ant food crop in developing count ries, which st ill produce 1.6 t imes as
much rice as wheat , t he second most import ant st aple. Recent proj ect ions made by t he
I nt ernat ional Food Policy Research I nst it ut e show t hat t he demand for rice will increase by
about 1.8% per year over t he 1990-2020 period. This means t hat over t he period of t he next
30 years, rice consumpt ion will increase by nearly 70%, and Asian rice product ion must
increase t o about 840 million t ons by t he year 2025, from t he present level of about 490 million
t ons, if rice prices are t o be maint ained at current levels.
3.9.2 Met hod
The basic idea behind t he generat ion of rice acreage st at ist ics using SAR t echniques is t he
analysis of changes in t he acquired dat a over t ime. Measurement of t emporal changes of SAR
response leads t o t he ident ificat ion of t he areas subj ect t o t ransplant ing / emergence moment ,
since an increase in t he SAR backscat t er corresponds t o a growt h in t he rice plant s.
3. 9 Ri ce Mappi ng 3. 9 Ri ce Mappi ng
3.9.3 Example
Based on mult i-t emporal ENVI SAT ASAR
and RADARSAT-1 dat a, a rice map,
indicat ing t he rice emergence moment s
has been generat ed for an area in t he
Philippines. The rice acreage st at ist ics are
st ored in map format showing t he rice
ext ent (right ) and, in form of numerical
t ables (left ), quant ifying t he dimension of
t he area cult ivat ed by rice at t he smallest
administ rat ive level - t ypically village unit .
sarmap
0
1000
2000
3000
4000
5000
6000
7000
8000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
H
e
c
t
a
r
Municipalit ies
3.10 Snow Mappi ng 3.10 Snow Mappi ng
3.10.1 Purpose
The ext ent of snow covered area is a key paramet er for predict ing snowmelt runoff. Because
SAR sensors provide repeat pass observat ions irrespect ive of cloud cover, t hey are of int erest
for operat ional snowmelt runoff modelling and forecast ing.
3.10.2 Met hod
The met hodology for mapping melt ing snow, developed by t he Universit y of I nnsbruck, is
based on repeat pass images in t he C-band SAR and applies change det ect ion t o eliminat e t he
t opographic effect s of backscat t ering. At C-band dry snow is t ransparent and backscat t ering
from t he rough surfaces below t he packed snow dominat es. This is t he reason why t he ret urn
signals from dry snow and snow-free areas are very similar. When t he snow becomes wet ,
backscat t ering decreases significant ly. Wet snow can t herefore be dist inguished from dry snow
or snow- free condit ions using analysis of t emporal backscat t er changes.
3.10 Snow Mappi ng 3.10 Snow Mappi ng
3.10.3 Example
The figure on t he right shows a snow
map on 12 May 1997 (blue and green)
and 16 June 1997 (green only), based
on ERS-2 SAR images of ascending
and descending passes. I n areas of
residual layover (magent a) no
informat ion can be ext ract ed. The
boundaries of t he Schlegeis basin
(Aust ria) are shown in yellow. On t he
lower right t he fract ion of residual
layover is high because it is covered
only by t he descending image. The
snow area decreased from 97 km
2
on
12 May t o 61 km
2
on 16 June.
Universit y of I nnsbruck
3. 11 Ur ban Mappi ng 3. 11 Ur ban Mappi ng
3.11.1 Purpose
Geo-informat ion is by definit ion spat ially relat ed and t herefore reliant on mapping. Some sort
of base map is t he foundat ion upon which every Geographic I nformat ion Syst em (GI S) or geo-
spat ial service is built . Hence, cart ography can be seen as a horizont al element of t he geo-
informat ion service indust ry, supplying an input t o every processing and applicat ion-based
chain and for a broad range of t hemat ic applicat ions. Cart ographic product ion chains t hat
exploit bot h SAR and opt ical dat a exist and operat e in bot h t he civilian and securit y sect ors.
3.11.2 Met hod
Urban areas are difficult t o analyse, primarily due t o t he many different land cover t ypes (e.g.
st reet s, buildings, parks, et c.), each of which have t heir own shape, geomet ry and dimension
charact erist ics. The met hodology developed by t he Universit y of Pavia, which makes it possible
t o ident ify different classes depending upon building densit y, uses a t ext ure analysis approach
(i.e. a t echnique t hat t ake int o account of t he spat ial relat ionship bet ween neighbouring
pixels). The ext ract ed feat ures are finally classified using supervised non-paramet ric classifiers,
such as t he Fuzzy Art map.
3. 11 Ur ban Mappi ng 3. 11 Ur ban Mappi ng
3.11.3 Example
The figure below illust rat es an ERS-1 SAR scene of t he Pavia (I t aly) acquired on August 13t h
1992 and t he result ing map.
U
n
i
v
e
r
s
i
t
y

o
f

P
a
v
i
a
Building high densit y
Building medium densit y
Building low densit y
Veget at ion
Wat er
3. 12 Wet l ands Mappi ng 3. 12 Wet l ands Mappi ng
3.12.1 Purpose
Wet lands are areas where wat er is t he primary fact or cont rolling t he environment and t he
associat ed plant and animal life. They occur where t he wat er t able is at or near t he surface of
t he land, or where t he land is covered by shallow wat er. Half of t he worlds wet lands are
est imat ed t o have been lost during t he 20t h cent ury, as land was convert ed t o agricult ural and
urban areas, or filled t o combat diseases such as malaria. The Ramsar Convent ion provides t he
framework for int ernat ional co-operat ion for t he conservat ion of wet lands. I t obliges it s part ies
t o designat e wet lands of int ernat ional import ance for inclusion in a list of so-called Ramsar
sit es and t o wisely manage t he wet lands in t heir t errit ories.
3.12.2 Met hod
The primary ut ilisat ion of SAR dat a is in t he ident ificat ion and mapping of open wat er and
flooded veget at ion. This informat ion is oft en used t o complement land use/ cover analysis
(based on opt ical images), but may also be ut ilized as a st and alone product . This so-called
Wat er Cycle Regime Product indicat es t he ext ent of wat er at dry and wet periods of t he year,
and is generat ed by classifying mult i-t emporal SAR dat a.
3. 12 Wet l ands Mappi ng 3. 12 Wet l ands Mappi ng
3.12.3 Example
Mult i-t emporal RADARSAT-1 dat a
acquired on August , Sept ember and
November 2004 over t he Lit t oral Audois
(France) have been used t o generat e
t he Wat er Cycle Regime Product (figure
right bot t om). This product has been
obt ained by combining t he t hree wat er
/ flooded veget at ion maps (figures t op
right , t op left , bot t om left ) , which were
produced by t hresholding each image.
The result ing product gives a clear
indicat ion of t he wat er cycle during t his
period.
At lant is Scient ific
25 August 2004 11 Sept ember 2004
12 November 2004
4. What ar e Oper at i onal and 4. What ar e Oper at i onal and Fut ur e Fut ur e Spacebor ne Spacebor ne SAR Sensor s? SAR Sensor s?
4.1 ERS 4.1 ERS- - 1 and 2 SAR 1 and 2 SAR
4.2 JERS 4.2 JERS- - 1 SAR 1 SAR
4.3 RADARSAT 4.3 RADARSAT- - 1 1
4.4 SRTM 4.4 SRTM
4.5 ENVI SAT ASAR 4.5 ENVI SAT ASAR
4.6 ALOS PALSAR 4.6 ALOS PALSAR- - 1 1
4.7 TerraSAR 4.7 TerraSAR- - X X- - 1 1
4.9 COSMO 4.9 COSMO- - SkyMed SkyMed
4.10 RI SAT 4.10 RI SAT- - 1 1
4.11 SENTI NEL 4.11 SENTI NEL- - 1 1
4.1 ERS 4.1 ERS- - 1 and 2 SAR 1 and 2 SAR
Agency European Space Agency
Frequency C-band
Polarizat ion VV
Ground Resolut ion 25 m
Acquisit ion Mode St ripmap (I mage)
Swat h 100 km
Repeat cycle 35 days
Launched 1991-2000 / 1995
Furt her I nformat ion ht t p: / / www.esa.int
Not e t hat t he j oint use of ERS
(Eart h Remot e Sensing Sat ellit e) -1
and ERS-2 SAR is called ERS-
Tandem mode. I n t his part icular
case, ERS-1 and ERS-2 SAR dat a
have been acquired t ime-shift ed by
24 hours. For almost 5 years t his
at ypical acquisit ion mode made it
possible t o collect ed repeat -pass
int erferomet ric (I nSAR) dat a used
mainly for t he generat ion of Digit al
Elevat ion Model dat a.
4.2 JERS 4.2 JERS- - 1 SAR 1 SAR
Agency Japan Aerospace Explorat ion Agency
Frequency L-band
Polarizat ion HH
Ground Resolut ion 20 m
Acquisit ion Mode St ripmap (I mage)
Swat h 70 km
Launched 1993-1998
Furt her I nformat ion ht t p: / / www.eorc.j axa.j p
Agency Canadian Space Agency
Frequency C-band
Polarizat ion HH
Ground Resolut ion 10 t o 100 m
Acquisit ion Modes St ripmap (Fine, St andard, Wide) and ScanSAR
Swat h 50 t o 500 km
Launched 1995
Furt her I nformat ion ht t p: / / www.rsi.ca
As indicat ed in t he Table below, images acquired in different modes (Fine, St andard,
Wide, et c.) can be delivered in different format s (Signal Dat a, Single Look Complex, Pat h
I mage, et c.). I t is wort h ment ioning t hat t he most appropriat e format is (if available)
Single Look Complex dat a. This is primarily because i) t he dat a are in t he original SAR
geomet ry and ii) t he dat a are unt ouched (in radiomet rical t erms), t hus making it possible
t o perform t he most suit able processing for t he generat ion of t he envisaged product .
Furt her inf ormat ion at ht t p: / / www.rsi.ca/ product s/ sensor/ radarsat / cl_ra_bm.asp
4.4 SRTM 4.4 SRTM
Agency NASA/ JPL & DARA/ ASI
Frequency X- and C-band
Polarizat ion VV
Acquisit ion Modes St ripmap
Mission lenght 11 days
Launch 2000
Furt her I nformat ion ht t p: / / srt m.usgs.gov
4.4 SRTM 4.4 SRTM
The Shut t le Radar Terrain Mission (SRTM) is a
j oint proj ect bet ween NASA and Nat ional
Geospat ial-I nt elligence Agency t o map t he world
in t hree dimensions. SRTM ut ilized dual
Spaceborne I maging Radar (SI R-C) and dual X-
band Synt het ic Apert ure Radar (X-SAR) configured
as a baseline int erferomet er. Flown aboard t he
NASA Space Shut t le Endeavour February 11-22,
2000, SRTM successfully collect ed dat a over 80%
of t he Eart h's land surface, for most of t he area
bet ween 60 degrees N and 56 degrees S lat it ude.
SRTM dat a are being processed at t he Jet
Propulsion Laborat ory int o research-qualit y digit al
elevat ion models (DEMs). The dat a are 3 X 3
averaged t o 3-arc second spacing (90 met re) from
t he original 1-arc second dat a. The absolut e
horizont al and vert ical accuracy is 20 met res
(circular error at 90% confidence) and 16 met res
(linear error at 90% confidence), respect ively.
Mount St . Hel ens, USA
USGS
Agency European Space Agency
Frequency C-band
Polarizat ion HH or VV or HH/ HV or VV/ VH
Acquisit ion Modes St ripmap (I mage), AP, ScanSAR (Wide Swat h, Globe)
Launch 2001
Furt her I nformat ion ht t p: / / www.esa.int
As indicat ed in t he Table below, images acquired in different modes (Fine, St andard,
Wide, et c.) can be delivered in different format s (Raw, Single Look Complex, Precision,
et c.). I t is wort h ment ioning t hat t he most appropriat e format is (if available) t he Single
Look Complex one, primarily because i) t he dat a are in t he original SAR geomet ry and ii)
t he dat a are unt ouched (in radiomet rical t erms), t hus making it possible t o perform t he
most suit able processing for t he generat ion of t he envisaged product .
Raw SLC Pr eci si on El l i psoi dal Medi um Br owse
Geocoded Resol ut i on
I mage Yes Yes Yes Yes Yes Yes
Al t er nat i ng
Pol ari zat i on Yes Yes Yes Yes Yes Yes
Wi de Swat h No No Yes No Yes Yes
Gl obal No No Yes No No Yes
Furt her inf ormat ion ht t p: / / envisat .esa.int / dat aproduct s/ asar/ CNTR2-1.ht m
4.6 ALOS PALSAR 4.6 ALOS PALSAR- - 1 1
Agency Japan Aerospace Explorat ion Agency
Frequency L-band
Polarizat ion Single Pol, Dual Pol, Full Pol
Acquisit ion Modes St ripmap (Fine) and ScanSAR
Repeat Cycle 46 days
Launch 2006
Furt her I nformat ion ht t p: / / www.eorc.j axa.j p
4.7 Ter r aSAR 4.7 Ter r aSAR- - X X- - 1 1
Agency I nfot erra, Germany
Frequency X-band
Acquisit ion Modes St ripmap, ScanSAR and Spot light
Swat h 15 t o 60 km
Launch 2007
Furt her I nformat ion ht t p: / / www.t errasar.de
Agency Canadian Space Agency and MacDonald Det t wiler (MDA)
Frequency C-band
Polarizat ion Single Pol, Dual Pol and Full Pol
Acquisit ion Modes St ripmap and ScanSAR
Launch 2007
Furt her I nformat ion ht t p: / / www.mda.ca
4.9 COSMO 4.9 COSMO- - Sk y Med Sk y Med
Agency Agenzia Spaziale I t aliana (ASI )
Frequency X-band
Acquisit ion Modes St ripmap, ScanSAR, and Spot light
Launch 2007 t o 2009 - Const ellat ion of 4 sat ellit es
4.10 RI SAT 4.10 RI SAT- - 1 1
Agency I ndian Space Agency
Frequency C-band
Acquisit ion Modes St ripmap, ScanSAR and Spot light
Repeat cycle ? days
Launch 2009
Furt her I nformat ion ht t p: / / www.isro.org
4.11 SENTI NEL 4.11 SENTI NEL- - 1 1
Nearly all European SAR sat ellit e syst ems current ly in orbit have t heir nominal lifet ime
t erminat ing in 2008. Cont inuit y of ESA SAR C-band dat a is vit al t o ensure effect ive
exploit at ion of user invest ment and gaps in dat a availabilit y will affect on-going monit oring
programs.
The following 3 modes - relevant for land applicat ions - are planned:
St ripmap I nt erferomet ric Ext ra-ScanSAR
ScanSAR
Azimut h Resolut ion (m) 5 < 20 < 80
Ground Range resolut ion (m) 4 < 5 < 25
Swat h (km) > 80 > 240 > 400
Polarizat ion HH-HV, VV-HV HH-HV, VV-HV HH-HV, VV-HV
Repeat Cycle (days) 14 14 14
5. Gl ossar y 5. Gl ossar y
Range (pixel) spacing Pixel spacing across t rack
Azimut h (pixel) spacing Pixel spacing along t rack
I ncidence angle Angle from nadir at which t arget is viewed
Swat h Widt h of t he imaged scene in t he range
Slant Range
I mage direct ion as measured along t he
sequence of line-of-sight rays from t he
radar t o each and every reflect ing point in
t he illuminat ed scene.
Ground Range
Range direct ion of a side-looking radar
image as proj ect ed ont o t he nominally
horizont al reference plane, similar t o t he
spat ial display of convent ional maps.
Some Basic Terminology
Some Basic Terminology
Ascending Orbit
Descending Orbit
orbit direct ion
orbit direct ion
Look direct ions
Right
Left
Left
Right Look direct ions
Across-t rack - The across-t rack dimension is t he imaging direct ion of t he sensor t hat is ort hogonal t o t he direct ion in which t he
plat form is moving.
Act i ve Remot e Sensi ng Syst em - A syst em t hat provides it s own source of energy and illuminat ion (i.e. radar syst em). A
remot e sensing syst em t hat t ransmit s it s own elect romagnet ic emanat ions at an obj ect (s) and t hen records t he energy reflect ed
or refract ed back t o t he sensor.
Al ong-t rack - The along-t rack dimension is t he imaging direct ion of t he sensor t hat is parallel t o t he direct ion in which t he
plat form is moving.
Ampl i t ude - Measure of t he st rengt h of a signal, and in part icular t he st rengt h or height of an elect romagnet ic wave (unit s of
volt age). The amplit ude may imply a complex signal, including bot h magnit ude and t he phase.
Ant enna - Part of t he radar syst em, which t ransmit s and/ or receives elect romagnet ic energy.
Ant enna Ar r ay - An arrangement of several individual ant ennas so spaced and phased t hat t heir individual cont ribut ions add in
t he preferred direct ion and cancel in ot her direct ions. SAR syst ems, employ a short physical ant enna, but t hrough modified dat a
recording and processing t echniques, t hey synt hesise t he effect of a very long ant enna. The result of t his mode of operat ion is a
very narrow effect ive ant enna beamwidt h, even at far ranges, wit hout requiring a physically long ant enna or a short operat ing
wavelengt h. For example, in a SAR syst em, a 2m ant enna can be made effect ively 600 m long.
At t enuat i on - Decrease in t he st rengt h of a signal. The decrease in t he st rengt h of a signal, is usually described by a
mult iplicat ive fact or in t he mat hemat ical descript ion of signal level. A signal is at t enuat ed by applicat ion of a gain less t han unit y.
Common causes of at t enuat ion of an elect romagnet ic wave include losses t hrough absorpt ion and by volume scat t ering in a
medium as a wave passes t hrough.
Azi mut h - The relat ive posit ion of an obj ect wit hin t he field of view of an ant enna in t he plane int ersect ing t he moving radar's
line of flight . The t erm commonly is used t o indicat e linear dist ance or image scale in t he along-t rack direct ion.
Azi mut h Ambi gui t y - A form of ghost ing t hat occurs when t he sampling of ret urned signals is t oo slow.
Azi mut h Bandpass Fi l t eri ng - Bandpass filt ering select s a cert ain band of frequency component s in t he signal. Azimut h
bandpass filt ering refers t o filt ering in t he azimut h direct ion of t he t wo-dimensional SAR signal. The locat ion of signal energy in
t he azimut h frequency domain depends on t he ant enna point ing angle, so bandpass filt ering is necessary t o maximize t he signal
energy in t he processed image.
Azi mut h Compressi on - I n t he SAR signal domain, t he raw dat a is spread out in t he range and azimut h direct ions and must
be coherent ly compressed t o realise t he full-resolut ion pot ent ial of t he inst rument . Azimut h compression consist s of coherent ly
correlat ing t he received signal wit h t he azimut h replica funct ion. The appropriat e Hamming weight ing is applied also t o t he
reference funct ion. Subsequent correlat ion has t he effect of modulat ing bot h signal and noise by similar amount s and hence t he
signal-t o-noise rat io is unchanged by t his process.
Azi mut h Resol ut i on - Resolut ion charact erist ic of t he azimut h dimension, usually applied t o t he image domain. Azimut h
resolut ion is fundament ally limit ed by t he Doppler bandwidt h of t he syst em. Excess Doppler bandwidt h is usually used t o allow
ext ra looks, at t he expense of azimut h resolut ion.
Azi mut h Ti me - The t ime along t he flight pat h.
Backscat t er - I t is t he port ion of t he out going radar signal t hat t he t arget redirect s direct ly back t owards t he radar ant enna.
Backscat t eri ng is t he process by which backscat t er is formed. The scat t ering cross sect ion in t he direct ion t oward t he radar is
called t he backscat t ering cross sect ion; t he usual not at ion is t he symbol sigma . I t is a measure of t he reflect ive st rengt h of a
radar t arget . The normalised measure of t he radar ret urn from a dist ribut ed t arget is called t he backscat t er coefficient , or sigma
nought , and is defined as per unit area on t he ground. I f t he signal formed by backscat t er is undesired, it is called clut t er.
Ot her port ions of t he incident radar energy may be reflect ed and scat t ered away from t he radar or absorbed.
Band - A select ion of wavelengt hs or range of radar frequencies.
Bandwi dt h -A measure, according t o a st andard definit ion (see widt h), of t he span of frequencies available in a signal or ot her
dist ribut ion, or of t he frequency limit ing st ages in t he syst em. Typical bandwidt hs in t he range channel of a SAR are on t he order of 20
Megahert z, and in t he azimut h channel are on t he order of 1 Kilohert z. Bandwidt h is a fundament al paramet er of any imaging syst em,
and det ermines t he ult imat e resolut ion available. For any pulse, t he basic paramet er t hat describes it s st ruct ure is t he t ime bandwidt h
product .
Beam - A focused pulse of energy. The ant enna beam of a side-looking radar (SLAR) is direct ed perpendicular t o t he flight pat h and
illuminat es a swat h parallel t o t he plat form ground t rack. Due t o t he mot ion of t he sat ellit e, each t arget element is illuminat ed by t he
beam for a period of t ime, known as t he int egrat ion t ime.
Beam Mode - The SAR operat ing configurat ion defined by t he swat h widt h and resolut ion.
Bet a Nought ( ) - A radar bright ness coefficient . The reflect ivit y per unit area in slant range is dimensionless.
Br i ght ness - Propert y of an image in which t he st rengt h of t he radar reflect ivit y is expressed as being proport ional t o a digit al
number (digit al image file) or t o a grey scale (phot ographic image), which for a phot ographic posit ive shows bright as whit e. The
at t ribut e of visual percept ion in accordance wit h which an area appears t o emit more or less light . Bright ness may be a result of
variat ions in t one, t ext ure, or in t he case of radar imagery, radar art efact s. The t opography and surface roughness of t he t errain will
affect t he image bright ness. Where t he local incidence angle is large, t he image will be dark. Conversely, t he image will be bright er
where t he local incidence angle is small.
C-Band - A nominal frequency range, from 8 t o 4 Ghz (3.75 t o 7.5 cm wavelengt h) wit hin t he microwave port ion of t he
elect romagnet ic spect rum. C-band has been t he frequency of choice for several experiment al aircraft SAR syst ems as well as a series
of single-band sat ellit e SAR syst ems, including t he ERS-1/ 2 and Envisat SAR syst ems and RADARSAT-1/ 2 SAR. The corresponding
wavelengt h for t hese syst ems is on t he order of 5.6 cm, which has been found useful in sea ice surveillance as well as in ot her
applicat ions. I t s penet rat ion capabilit y wit h regard t o veget at ion canopies or soils is limit ed and is rest rict ed t o t he t op layers.
Cal i brat i on - Process whereby one may relat e t he digit al numbers describing an image t o physical quant it ies such as reflect ivit y,
geomet ry (posit ion or size), or phase.
Chi rp - Typical phase coding or modulat ion applied t o t he range pulse of an imaging radar designed t o achieve a large t ime-
bandwidt h product . The result ing phase is quadrat ic in t ime, which has a linear derivat ive. Such coding is oft en called linear
frequency modulat ion (FM) .
Chi r p Compr essi on - The echo signal is correlat ed wit h a suit able reference funct ion. This correlat ion is performed in t he
frequency domain aft er suit able Fast Fourrier Transform from t he t ime domain. The reference funct ion of int erest should represent
t he chirp signal which illuminat es t he t arget .
Coher ence - Coherence is t he magnit ude of an int erferogram's pixels, divided by t he product of t he magnit udes of t he original
image's pixels. I t is usually calculat ed on a small window of pixels at a t ime, from t he complex int erferogram and images. I t ranges
from 0.0, where t here is no useful informat ion in t he int erferogram; t o 1.0, where t here is no noise in t he int erferogram. Coherence
can serve as a measure of t he qualit y of an int erferogram; t ell you more about t he surface t ype (veget at ed vs. rock); or t ell you
when a t iny, ot herwise invisible change has occured in t he image, and it is only visible in t he phase image of an int erferogram.
Compl ex Number - For radar syst ems, a complex number implies t hat t he represent at ion of a signal, or dat a file, needs bot h
magnit ude and phase measures. I n t he digit al SAR cont ext , a complex number is oft en represent ed by an equivalent pair of
numbers, t he real in-phase component (I ) and t he imaginary quadrat ure component (Q). For coherent syst ems such as SAR, t he
role of complex numbers is an essent ial part of t he signal, since signal phase is used in t he processor t o obt ain high-resolut ion.
Co-pol ar i sat i on Si gnat ur e - The received signat ure when t he t ransmit and receive ant ennas have t he same polarisat ion
propert ies.
Cor ner Ref l ect or - A combinat ion of t wo or more int ersect ing specular surfaces t hat combine t o enhance t he signal reflect ed in
t he direct ion of t he radar. The st rongest reflect ion is obt ained for mat erials having a high conduct ivit y (i.e. ships, bridges).
Cross Pol ari sat i on Si gnat ure -The received signat ure when t he t ransmit and receive ant ennas have ort hogonal polarisat ions.
Deci bel ( dB) - Measurement of signal st rengt h, properly applied t o a rat io of powers. Decibels oft en are used in radar, such as in
measures of reflect ivit y, for which t he dynamic range may span several fact ors of t en.
Depol ari sat i on - The polarisat ion st at e of an elect romagnet ic wave can change when t he wave scat t ers from a t arget .
Depolarisat ion is a measure of t he change in t he degree of polarisat ion of a part ially polarised wave upon scat t ering. For example, a
t arget may scat t er a wave wit h a great er degree of polarisat ion t han t he incident wave, in which case t he depolarisat ion is negat ive.
Depolarisat ion is also used t o indicat e spat ial or t emporal variat ion of t he degree of polarisat ion for a complet ely polarised wave
Depr essi on Angl e - Usually refers t o t he line of sight from t he radar t o an illuminat ed obj ect as measured from t he horizont al plane
at t he radar. For image int erpret at ion, use of t he t erm is not recommended because it does not account for t he effect s of Eart h
curvat ure, and it does not convenient ly include effect s of local slope in t he scene. I t is more appropriat e for engineering descript ion
of t he vert ical ant enna pat t ern at t he radar it self.
Det ect i on - Processing st age at which t he st rengt h of t he signal is det ermined for each pixel value. Det ect ion removes phase
informat ion from t he dat a file. The preferred det ect ion scheme uses a magnit ude squared met hod, which is energy conserving, and
has unit s of volt age squared per pixel.
Doppl er Fr equency - The Doppler frequency depends on t he component of sat ellit e velocit y in t he line-of-sight direct ion t o t he
t arget . This direct ion changes wit h each sat ellit e posit ion along t he flight pat h, so t he Doppler frequency varies wit h azimut h t ime.
For t his reason, azimut h frequency is oft en referred t o as Doppler frequency.
Di el ect r i c - Mat erial which has neit her "perfect " conduct ivit y nor is perfect ly "t ransparent " t o elect romagnet ic radiat ion. The
elect rical propert ies of all int ermediat e mat erials, such as ice, nat ural foliage, or rocks, may be described by t wo quant it ies relat ive
dielect ric const ant ; and loss t angent . Reflect ivit y of a smoot h surface and t he penet rat ion of microwaves int o t he mat erial are
det ermined by t hese t wo quant it ies.
Di el ect ri c Const ant - Fundament al (complex) paramet er, also known as t he complex permit t ivit y, t hat describes t he elect rical
propert ies of a lossy medium. (See permeabilit y.) By convent ion, t he relat ive dielect ric const ant of a given mat erial is used, defined
as t he (absolut e) dielect ric const ant divided by t he dielect ric const ant of "free space".
Di hedr al - Corner reflect or formed by t wo surfaces ort hogonally int ersect ing. For enhanced backscat t er, t he dihedral must be open t o
t he radar, and have t he axis of int ersect ion at right angles t o t he direct ion of illuminat ion.
Di st ri but ed Scat t ers -Element s of a scene consist ing of many small scat t erers of random locat ion, phase, and reflect ivit y in each
resolut ion cell.
El l i pt i cal Pol ari sat i on - A polarisat ion st at e in which t he t wo perpendicular component s of t he elect ric field have unequal magnit udes
and a non-zero phase difference. I n t his case, t he t ip of t he elect ric field vect or t races an ellipse on a plane t hat is t ransverse t o t he
wave propagat ion direct ion.
Foreshort eni ng - Spat ial dist ort ion whereby t errain slopes facing a side-looking radar's illuminat ion are mapped as having a
compressed range scale relat ive t o it s appearance if t he same t errain were level. Foreshort ening is a special case of elevat ion
displacement . The effect is more pronounced for st eeper slopes, and for radars t hat use st eeper incidence angles. Range scale
expansion, t he complement ary effect , occurs for slopes t hat face away from t he radar illuminat ion.
Fr equency - Number of oscillat ions per unit t ime or number of wavelengt hs t hat pass a point per unit t ime. Rat e of oscillat ion of a
wave. I n remot e sensing, t his t erm is most oft en used wit h radar. The frequency bands used by radar (radar frequency bands) were
first designat ed by let t ers for milit ary secrecy. I n t he microwave region, frequencies are on t he order of 1 GHz (Gigahert z) t o 100 GHz.
("Giga" implies mult iplicat ion by a fact or of a billion). For elect romagnet ic waves, t he product of wavelengt h and frequency is equal to
t he speed of propagat ion, which, in free space, is t he speed of light .
Fr equency Modul at i on - A t echnique in which t he frequency of a signal is changed about a fundament al or carrier frequency.
Geocodi ng ( or Georef er enci ng or Ort ho-rect i f i cat i on) -The process t o t ransform an image from slant range proj ect ion t o a
cart ographic reference syst em considering ellipsoidal height or a Digit al Elevat ion Model.
Ground Range - Range direct ion of a side-looking radar image as proj ect ed ont o t he nominally horizont al reference plane, similar t o
t he spat ial display of convent ional maps. For spacecraft dat a, an Eart h geoid model is used, whereas for airborne radar dat a, a planar
approximat ion is sufficient . Ground range proj ect ion requires a geomet ric t ransformat ion from slant range t o ground range, leading t o
relief or elevat ion displacement , foreshort ening, and layover unless t errain elevat ion informat ion is used.
Ground Range - Range direct ion of a side-looking radar image as proj ect ed ont o t he nominally horizont al reference plane, similar to
t he spat ial display of convent ional maps. For spacecraft dat a, an Eart h geoid model is used, whereas for airborne radar dat a, a planar
approximat ion is sufficient . Ground range proj ect ion requires a geomet ric t ransformat ion from slant range t o ground range, leading t o
relief or elevat ion displacement , foreshort ening, and layover unless t errain elevat ion informat ion is used.
Hori zont al Pol ari sat i on - Linear polarisat ion wit h t he lone elect ric vect or orient ed in t he horizont al direct ion in ant enna co-ordinat es.
I nci dence Angl e - Angle bet ween t he line of sight from t he radar t o an element of an imaged scene, and a vert ical direct ion
charact erist ic of t he scene. The definit ion of "vert ical" for t his purpose is import ant . One must dist inguish bet ween t he ( nominal)
"incidence angle" det ermined by t he large scale geomet ry of t he radar and t he Eart h's geoidal surface, and t he local incidence angle
which t akes int o account t he mean slope wit h each pixel of t he image. Smaller incidence angle refers t o viewing line of sight being
closer t o t he (local) vert ical, hence "st eeper". I n general, reflect ivit y from dist ribut ed scat t erers decreases wit h increasing incidence
angle.
I nt ensi t y - St rengt h of a field or of a dist ribut ion, such as an image file, proport ional t o magnit ude, squared (see Power).
I nt er f eromet ri c Synt het i c Apert ur e Radar ( I nSAR) - SAR int erferomet ry is a t echnique involving phase measurement s from
successive sat ellit e SAR images t o infer different ial range and range changes for t he purpose of det ect ing very subt le changes on, or
of, t he Eart h surface wit h unprecedent ed scale, accuracy and reliabilit y. SAR int erferomet ry has been demonst rat ed successfully in a
number of applicat ions, including t opographic mapping, measurement of t errain displacement as a result of eart hquakes, and
measurement of flow rat es of glaciers or large ice sheet s. The t erm I nSAR, is most commonly associat ed wit h repeat -pass
int erferomet ry.
I nt er f er omet r y - A t echnique t hat uses t he measured differences in t he phase of t he ret urn signal bet ween t wo sat ellit e passes t o
det ect slight changes on t he Eart h's surface. The combinat ion of t wo radar measurement s of t he same point on t he ground, t aken at
t he same t ime, but from slight ly different angles, t o produce st ereo images. Using t he cosine rule from t rigonomet ry t o calculat e t he
dist ance bet ween t he radar and t he Eart h's surface, t hese measurement s can produce very accurat e height maps, or maps of height
changes. Mapping height changes provides informat ion on eart hquake damage, volcanic act ivit y, landslides, and glacier movement .
L-Band - A nominal frequency range, from 1 t o 3 Ghz (30 t o 10 cm wavelengt h) wit hin t he microwave port ion of t he
elect romagnet ic spect rum. L-band has been t he frequency of choice for several experiment al aircraft SAR syst ems as well as a
series of single-band sat ellit e SAR syst ems, including t he SEASAT SAR and JERS-1 SAR syst ems. The corresponding wavelengt h for
t hese syst ems is on t he order of 23.5 cm, which has been found useful in sea ice surveillance as well as in ot her applicat ions. I t s
penet rat ion capabilit y wit h regard t o veget at ion canopies is significant .
Layover - Ext reme form of elevat ion displacement or foreshort ening in which t he t op of a reflect ing obj ect (such as mount ain) is
closer t o t he radar (in slant range) t han are t he lower part s of t he obj ect . The image of such a feat ure appears t o have fallen over
t owards t he radar. The effect is more pronounced for radars having smaller incidence angle.
Li near Pol ar i sat i on - A polarisat ion st at e in which one of t he perpendicular component s of t he elect ric field has zero magnit ude.
I n t his case, t he polarisat ion ellipse collapses t o a st raight line; t he t ip of t he elect ric field vect or t races a st raight line on a plane
t hat is t ransverse t o t he wave propagat ion direct ion.
Looks - I t refers t o individual looks as groups of signal samples in a SAR processor t hat split s t he full synt het ic apert ure int o several
sub-apert ures, each represent ing an independent look of t he ident ical scene. The result ing image formed by incoherent summing of
t hese looks is charact erised by reduced speckle and degraded spat ial resolut ion. The SAR signal processor can use t he full synt het ic
apert ure and t he complet e signal dat a hist ory in order t o produce t he highest possible resolut ion, albeit very speckled, single-look
complex (SLC) SAR image product . Mult iple looks may be generat ed by averaging over range and/ or azimut h resolut ion cells. For an
improvement in radiomet ric resolut ion using mult iple looks t here is an associat ed degradat ion in spat ial resolut ion. Not e t hat t here is
a difference bet ween t he number of looks physically implement ed in a processor, and t he effect ive number of looks as det ermined
by t he st at ist ics of t he image dat a.
Magni t ude -One of t hree paramet ers required t o describe a wave. Magnit ude is t he amplit ude of t he wave irrespect ive of t he
phase. For a complex signal described by in-phase (I ) and quadrat ure ( Q) component s, t he magnit ude is given by sqrt (I
2
+ Q
2
). For
complex amplit ude A, magnit ude is, by definit ion, lAl.
Mi crowave - A very short elect romagnet ic wave. The port ion of t he elect romagnet ic spect rum lying bet ween t he far infrared (I R)
and t he convent ional radio frequency port ion. While not bounded by definit ion, it is commonly regarded as ext ending from 1 mm to
1 m in wavelengt h (300 GHz t o 0.3 GHz frequency). Passive syst ems operat ing at t hese wavelengt hs somet imes are called
microwave syst ems. Act ive syst ems are called radar, alt hough t he lit eral definit ion of radar requires a dist ance measuring capabilit y
not always included in act ive syst ems.
Monost at i c Radar - A monost at ic radar syst em t ransmit s and receives it s energy t hrough t he same ant enna syst em or t hrough
collocat ed ant ennas.
Mot i on Compensat i on - Adj ust ment of a sensing syst em and/ or t he recorded dat a t o remove effect s of plat form mot ion, including
rot at ion and t ranslat ion, and variat ions in along-t rack velocit y. Mot ion compensat ion is essent ial for aircraft SARs, but usually is not
needed for spacecraft SARs.
Mul t i -l ook - See Looks
Mul t i f requency Radar - Broadband syst ems t hat t ransmit pulses in a range of frequencies and wavelengt hs.
Mul t i pol ari sat i on Radar -A radar capable of simult aneously and coherent ly acquiring several independent complex polarisat ion
measurement s for every pixel in t he image.
P-Band - A frequency range from 0.999 t o 0.2998 GHz (30 t o 100 cm wavelengt h) wit hin t he microwave (radar) port ion of t he
elect romagnet ic spect rum. P-band is an experiment al SAR frequency t hat has only been used t o-dat e for research and development
purposes. I t is part of t he NASA JPL AI RSAR mult i-frequency (C-, L- & P-band) SAR syst em designed for Eart h observat ion
experiment s. P-band is not hindered by at mospheric effect s and is capable of seeing t hrough heavy rain showers. P-band SAR
penet rat ion capabilit ies are very significant wit h regard t o veget at ion canopies, glacier or sea ice, and soil.
Phased Ar r ay Radar - A phased array radar uses an ant enna t hat consist s of an array of ant enna element s along wit h signal
processing t hat allows t he ant enna t o be st eered elect ronically.
Phase Pr eser vi ng - When t he phase at t he peak is correct , t he processing algorit hm is referred t o as phase preserving, regardless
of t he phase variat ion across t he impulse response.
Phase Unwr ap - I n SAR int erferomet ry, t he phase delay of t he carrier signal at a cert ain point in t he int erferogram is a funct ion
of t he t errain height at t hat point . However, t he phase of t he carrier signal can only be measured t o wit hin one cycle, or 360
degrees. Phase unwrapping refers t o convert ing t he measured phase t o t he absolut e phase, by adding t he appropriat e number of
cycles, or mult iple of 360 degrees, t o t he measured phase.
Pol ari met ri c Act i ve Radar Cal i brat or ( PARC) - Device used t o receive and ret ransmit radar pulses. These devices usually
consist of a polarisat ion sensit ive receive and t ransmit ant enna and a st able amplifier which boost s t he signal level so t hat t he
device being calibrat ed receives a high signal of a given polarisat ion.
Pol ari met ri c Radar - A radar which permit s measurement of t he full polarisat ion signat ure of every resolut ion element .
Pol ari sat i on - Orient at ion of t he elect ric (E) vect or in an elect romagnet ic wave, frequent ly "horizont al" (H) or "vert ical" (V) in
convent ional imaging radar syst ems. Polarizat ion is est ablished by t he ant enna, which may be adj ust ed t o be different on
t ransmit and on receive. Reflect ivit y of microwaves from an obj ect depends on t he relat ionship bet ween t he polarizat ion st at e
and t he geomet ric st ruct ure of t he obj ect . Common short hand not at ion for band and polarizat ion propert ies of an image file is to
st at e t he band, wit h a subscript for t he receive and t he t ransmit st at e of polarizat ion, in t hat order.
Pol ari sat i on El l i pse - For an ellipt ically polarised wave, t he t ip of t he elect ric field vect or t races an ellipse on a plane t hat is
t ransverse t o t he wave propagat ion direct ion. This polarisat ion ellipse describes t he polarisat ion propert ies of t he elect romagnet ic
waves, including t he rat io of t he perpendicular elect ric field component s and t heir relat ive phases.
Power - St rengt h of a field or of a dist ribut ion, such as an image file, proport ional t o magnit ude, squared (see I nt ensit y).
Pul se - A short burst of elect romagnet ic radiat ion t ransmit t ed by t he radar. Also described as a group of waves wit h a
dist ribut ion confined t o a short int erval of t ime. Such a dist ribut ion is described in t he t ime domain, or in spat ial dimensions, by
it s widt h and it s amplit ude or magnit ude, from which it s energy may be found. I n radar, use is made of modulat ed or coded
pulses which must be processed t o decode or compress t he original pulse t o achieve t he impulse response observed in t he
image.
Pul se Repet i t i on Frequency ( PRF) - Rat e of recurrence of t he pulses t ransmit t ed by a radar.
Radar Ant enna -The radar ant enna is a st ruct ure for t ransmit t ing and receiving radiat ed energy; it is an import ant subsyst em t hat
defines, t o a great ext ent , a radar's operat ional capabilit ies and cost . I n radar remot e sensing t he main funct ion of t he ant enna is t o
concent rat e a radiat ed microwave energy int o a beam of required shape, referred t o as t he ant enna pat t ern, t o t ransmit it int o t he
desired direct ion (look direct ion), and t o receive t he ret urned energy from surfaces or obj ect s. Radar remot e sensing ant ennas
provide scene illuminat ion.
Radar Cross Sect i on ( RCS) - Measure of radar reflect ivit y. The Radar Cross Sect ion (RCS) is expressed in t erms of t he physical
size of an hypot het ical uniformly scat t ering sphere t hat would give rise t o t he same level of reflect ion as t hat observed from t he
sample t arget .
Radar Equat i on - Mat hemat ical expression t hat describes t he average received signal level (or, somet imes, t he image signal
level), compared t o t he addit ive noise level, in t erms of syst em paramet ers. Principal paramet ers include t ransmit t ed power,
ant enna gain, noise power, and radar range. The range effect is somet imes called t he spreading fact or, since effect ive power
decreases significant ly wit h a small increase in range. All else equal, t he power received by a SAR per image pixel is proport ional t o
R
3
.
Radi o Echo - The signal reflect ed by a radar t arget , or t he t race produced by t his signal on t he screen of t he cat hode-ray t ube in a
radar receiver.
Radi omet ri c Resol ut i on - The expect ed spread of variat ion in each est imat e of scene reflect ivit y as observed in an image.
Smaller radiomet ric resolut ion is "bet t er". Radiomet ric resolut ion for a given radar may be improved by averaging, but at t he cost of
spat ial resolut ion.
Radi omet er - An inst rument for quant it at ively measuring t he int ensit y of elect romagnet ic radiat ion in some band of wavelengt hs in
any part of t he elect romagnet ic spect rum. Usually used wit h a modifier, such as an infrared radiomet er or a microwave radiomet er.
Radi omet ri c Cal i brat i on - The process t o radiomet rically calibrat e SAR dat a considering t he Ant enna Gain Pat t ern, Range Spread Loss
and Scat t ering area according t o t he Radar Equat ion.
Range Resol ut i on - Resolut ion charact erist ic of t he range dimension, usually applied t o t he image domain, eit her in t he slant range
plane or in t he ground range plane. Range resolut ion is fundament ally det ermined by t he syst em bandwidt h in t he range channel.
Range Ti me - The fast t ime wit hin a received pulse, relat ive t o t he pulse t ransmission t ime.
Raw Dat a - Raw dat a are dat a as received from t he SAR syst em.
Ref l ect i vi t y - Propert y of illuminat ed obj ect s t o reradiat e a port ion of t he incident energy. Reflect ivit y, in general, is larger in t he
specular direct ion for smaller surface roughness. For side looking radars, backscat t er is t he observable port ion of t he energy reflect ed.
Backscat t er, in general, is increased by great er surface roughness. I n general, reflect ivit y is increased for higher conduct ivit y of t he
scat t ering surface. The relat ive st rengt h of radar reflect ivit y is t abulat ed by sigma, for discret e obj ect s, and by sigma nought for nat ural
t errain surfaces.
Repeat Pass I nt er f eromet r y - Met hod based on t wo image acquisit ions of t he same scene from slight ly displaced orbit s of a sat ellit e.
Phase informat ion of t he t wo image dat a files are superimposed. The t wo phase values at each pixel are t hen subt ract ed, leading t o an
int erferogram t hat records only t he differences in phase bet ween t he t wo original images. Phase differences can be relat ed t o t he alt it ude
variat ion at each posit ion in t he swat h and enable t he product ion of a Digit al Elevat ion Model (DEM). For opt imum result s, t here should
be no change in t he backscat t er t o maint ain coherence; veget at ed sit es are t herefore a problem. For det ect ion of feat ure movement (e.g.
t racking glaciers) orbit s should be as close as possible. And knowledge of t he sensor locat ion is crit ical. Wit h a good baseline and
coherence, t his t echnique can be bet t er t han st ereo (~ 10 m vert ical accuracy).
Resol ut i on - Generally (but loosely) defined as t he widt h of t he "point spread funct ion", t he "Green's funct ion", or t he " impulse
response funct ion", depending on whet her one has an opt ics, a physics, or an elect ronic syst ems background. More properly, "resolut ion"
refers t o t he abilit y of a syst em t o different iat e t wo image feat ures corresponding t o t wo closely spaced small obj ect s in t he illuminat ed
scene when t he bright ness of t he t wo obj ect s in quest ion are comparable and fall wit hin t he dynamic range of t he radar in question.
(Definit ion adapt ed from Lord Rayleigh [ 1879] ). "Higher resolut ion" refers t o a syst em having a smaller impulse response widt h.
Resol ut i on Cel l - A t hree-dimensional cylindrical volume surrounding each point in t he scene. The cell range dept h is slant range
resolut ion, it s widt h is azimut h resolut ion, and it s height , which is conformal t o t he illuminat ion wavefront , is limit ed only by t he vert ical
beam widt h of t he ant enna pat t ern. Resolut ion cell oft en is defined wit h respect t o t he local horizont al.
SAR Focusi ng - I n a long synt het ic apert ure (array), SAR focusing involves t he removal and compensat ion of pat h lengt h differences
from t he ant enna t o t he t arget on t he ground. The main advant age of a focused synt het ic apert ure is t hat it increases it s array lengt h
over t hose radar signals t hat can be processed, and t hus increases pot ent ial SAR resolut ion at any range. SAR focusing is a necessary
process when t he lengt h of a synt het ic array is a significant fract ion of t he range t o ground being imaged, as t he lines-of-sight (range)
from a part icular point on t he ground t o each individual element of t he array differ in dist ance. These range differences, or pat h lengt h
differences, of t he radar signals can affect image qualit y. I n a focused SAR image t hese phase errors can be compensat ed for by
applying a phase correct ion t o t he ret urn signal at each synt het ic apert ure element . Focusing errors may be int roduced by unknown or
uncorrect ed plat form mot ion. I n an unfocused SAR image, t he usable synt het ic apert ure lengt h is quit e limit ed.
S-Band - A nominal frequency range from 4 t o 2 GHz ( 7.5 t o 15 cm wavelengt h) wit hin t he microwave (radar) port ion of t he
elect romagnet ic spect rum. S-band radars are used for medium-range met eorological applicat ions, for example rainfall measurement s,
as well as airport surveillance and specialised t racking t asks.
Scanni ng Synt het i c Aper t ur e Radar ( ScanSAR) - A having t he capabilit y t o illuminat e several subswat hs by scanning it s ant enna
off-nadir int o different posit ions.
Sensi t i vi t y Ti me Cont rol ( STC) - Pre-programmemed change in radar amplit ude due t o weaker backscat t er from great er ranges and
varying incidence angles across t he imaged swat h.
Shadow - From an opt ical point of view as seen from t he posit ion of a radar, a region hidden behind an elevat ed feat ure in t he scene
would be out of sight . This region corresponds t o t hat which does not get illuminat ed by t he radar energy, and t hus is also not visible
in t he result ing radar image. The region is filled wit h "no reflect ivit y", which appears as small digit al numbers, or a dark region in hard
copy.
Si del obes - Non-zero levels in a dist ribut ion t hat are separat ed from t he desired cent ral response. Sidelobes arise nat urally in
ant enna pat t erns, for example, alt hough in general t hey are a nuisance, and must be suppressed as much as possible. Large side-
lobes may lead t o unwant ed mult iple images of a single feat ure.
si gma ( o) - The convent ional measure of t he st rengt h of a radar signal reflect ed from a geomet ric obj ect (nat ural or
manufact ured) such as a corner reflect or. Sigma specifies t he st rengt h of reflect ion in t erms of t he geomet ric cross sect ion of a
conduct ing sphere t hat would give rise t o t he same level of reflect ivit y. (Unit s of area, such as met res squared). (See radar cross
sect ion.)
si gma nought ( o
o
) - Scat t ering coefficient , t he convent ional measure of t he st rengt h of radar signals reflect ed by a dist ribut ed
scat t erer, usually expressed in dB. I t is a normalized dimensionless number, comparing t he st rengt h observed t o t hat expect ed
from an area of one square met re. Sigma nought is defined wit h respect t o t he nominally horizont al plane, and in general has a
significant variat ion wit h incidence angle, wavelengt h, and polarizat ion, as well as wit h propert ies of t he scat t ering surface it self.
Sl ant Range - I mage direct ion as measured along t he sequence of line-of-sight rays from t he radar t o each and every reflect ing
point in t he illuminat ed scene. Since a SAR is looks down and t o t he side, t he slant range t o ground range t ransformat ion has an
inherent geomet ric scale which changes across t he image swat h.
Speckl e - St at ist ical fluct uat ion or uncert aint y associat ed wit h t he bright ness of each pixel in t he image of a scene. A single look
SAR syst em achieves one est imat e of t he reflect ivit y of each resolut ion cell in t he image. Speckle may be reduced, at t he expense
of resolut ion, in t he SAR processor by using several looks. Speckle appears as a mult iplicat ive random process whose variance and
spat ial correlat ion are det ermined primarily by t he SAR syst em.
Synt het i c Apert ur e - A synt het ic apert ure, or virt ual ant enna, consist s of a long array of successive and coherent radar signals
t hat are t ransmit t ed and received by a physically short (real) ant enna as it moves along a predet ermined flight or orbit al pat h. The
synt het ic apert ure is formed by point ing t he real radar ant enna of relat ively small dimensions, which are rest rict ed in size by t he
sat ellit e plat form, broadside t o t he direct ion of forward mot ion of t hat plat form. The point s at which successive pulses are
t ransmit t ed can be t hought of as t he element s of a long synt het ic array, which a signal processor will t hen use and process to
generat e a SAR image. This det ailed array of radar signal dat a is t he key t o achieving high azimut h resolut ion. This long virt ual
ant enna concept is t he basis for synt het ic apert ure radar, or SAR.
Synt het i c Apert ur e Radar ( SAR) - A synt het ic apert ure radar, or SAR, is a coherent radar syst em t hat generat es high-
resolut ion remot e sensing imagery. Signal processing uses magnit ude and phase of t he received signals over successive pulses
from element s of a synt het ic apert ure t o creat e an image. As t he line of sight direct ion changes along t he radar plat form
t raj ect ory, a synt het ic apert ure is produced by signal processing t hat has t he effect of lengt hening t he ant enna. The achievable
azimut h resolut ion of a SAR is approximat ely equal t o one-half t he lengt h of t he act ual (real) ant enna and does not depend on
plat form alt it ude (dist ance). High range resolut ion is achieved t hrough pulse compression t echniques. I n order t o map t he ground
surface t he radar beam is direct ed t o t he side of t he plat form t raj ect ory; wit h a sufficient ly wide ant enna beam widt h in t he along-
t rack direct ion, an ident ical t arget or area may be illuminat ed a number of t imes wit hout a change in t he ant enna look angle.
St okes Mat ri x - 4x4 array of real numbers t hat describes t he t ransformat ion of t he St okes paramet ers of t he incident wave int o
t he St okes paramet ers of t he elect romagnet ic wave reflect ed by each element of a scene illuminat ed by a radar. The St okes mat rix
describes t he complet e polarizat ion signat ure of t he reflect ive medium.
St okes Paramet ers - Set of four real numbers t hat t oget her describe t he st at e of polarizat ion of an elect romagnet ic wave.
Swat h - Widt h of t he imaged scene in t he range dimension, measured eit her in ground range or in slant range.
Text ur e - Second order spat ial average of bright ness. Scene t ext ure is t he spat ial variat ion of t he average reflect ivit y. For areas of
nominally const ant average reflect ivit y, image t ext ure consist s of scene t ext ure mult iplied by speckle.
Tone - First order spat ial average of image bright ness, oft en defined for a region of nominally const ant average reflect ivit y.
Tr ansmi ssi on - Energy sent by t he radar, normally in t he form of a sequence of pulses, t o illuminat e a scene of int erest .
Tri hedral - Corner reflect or formed from t hree mut ually ort hogonal surfaces.
Vol ume Scat t er i ng - Mult iple scat t ering event s occurring inside a medium, generally neit her dense nor having a large loss
t angent , such as t he canopy of a forest . The relat ive import ance of volume scat t ering is governed by t he dielect ric propert ies of
t he mat erial.
Vert i cal Pol ari sat i on - Linear polarisat ion wit h t he lone elect ric vect or orient ed in t he vert ical direct ion in ant enna co-ordinat es.
Wavel engt h - I n a periodic wave, t he dist ance bet ween t wo point s of corresponding phase in consecut ive cycles
X-Band - A nominal frequency range from 12.5 t o 8 GHz (2.4 t o 3.75 cm wavelengt h) wit hin t he microwave (radar) port ion of t he
elect romagnet ic spect rum. X-band is a suit able frequency for several high-resolut ion radar applicat ions and has oft en been used
for bot h experiment al and operat ional airborne SAR syst ems, designed for milit ary as well as civilian remot e sensing applicat ions.
The corresponding wavelengt h for t hese syst ems is on t he order of 3 cm, which has been found useful for mapping and
surveillance t asks.
Zero Doppl er Ti me - I t is t he along-t rack (azimut h) t ime at which a t arget on t he ground would have a Doppler shift of zero wit h
respect t o t he sat ellit e ( i.e. when t he t arget was perpendicular t o t he flight pat h). Also called t he closest approach azimut h t ime.
6. Ref er ences 6. Ref er ences
Carrara W., R. Goodman, and R. Maj ewsky, Spot light Synt het ic Apert ure Radar: Signal Processing Algorit hms,
Art ech House, 1985. - Advanced
Curlander J.C. and R.N. McDonough, Synt het ic Apert ure Radar: Syst ems and Signal Processing, Wiley-
I nt erscience, November, 1991. - Advanced
Dixon T. (Edit or), SAR I nt erf eromet ry and Surface Change Det ect ion, Report of a Workshop held in Boulder,
1995, ht t p: / / sout hport .j pl.nasa.gov/ science/ dixon. - Basic
Elachi C.,T. Bicknell, R. Jordan, and C. Wu, Spaceborne Synt het ic Apert ure I maging Radars: Applicat ion,
Techniques, and Technology, I EEE Vol. 70, Oct ober 1982. - Basic
ESA, ASAR product handbook, ht t p: / / www.envisat .esa.int / dat aproduct s. - Basic
ESA, Polarimet ric SAR I nt erf eromet ry, ht t p: / / eart h.esa.int / polinsar/ - Advanced.
Henderson F. and Lewis A. (Edit ors), Manual of Remot e Sensing, Volume 2, Principles and Applicat ions of
I maging Radar, I SBN: 0-471-33046-9, 1998. - Basic
Hovanessian S., I nt roduct ion t o synt het ic array and imaging radar, Art ech House, 1980. - Basic
Massonnet D. and K. Feigl, Radar int erferomet ry and it s applicat ions t o changes in t he Eart hs surface, Review
of Geophisics, 36/ 4, 1998. (ht t p: / / www.ingv.it / barba/ igl/ 2003/ Massonnet _98.pdf) - Basic t o Advanced
6. Ref er ences 6. Ref er ences
Oliver C. and S. Quegan, Underst anding Synt het ic Apert ure Radar I mages, Art echHouse, 1998. - Basic t o
Advanced
Olmert C., Alaska SAR Facilit y Scient ific SAR Users Guide, ht t p: / / www.asf.alaska.edu/ SciSARuserGuide.pdf -
Basic
Schreier G. (Edit or), SAR Geocoding: Dat a and Syst ems, Wichmann, 1993. - Basic t o Advanced
Ulaby F.T., R. Moore, and A.K. Fung, Microwave Remot e Sensing (Volume 1,2,3), Addison Wesley, Reading
(MA), 1981, 1982, 1986. - Advanced
Ulaby F.T. and C. Elachi (Edit ors), Radar Polarimet ry for Geoscience Applicat ions, Art ech House, Nordwood,
1989. - Advanced
For a complet e reference list refer t o ht t p: / / sout hport .j pl.nasa.gov/ science/ SAR_REFS.ht ml

SAR Guidebook

Uploaded by

Copyright:

Available Formats

SAR Guidebook

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SAR Guidebook

Uploaded by

Copyright:

Available Formats

sarmap, Sept ember 2008

Sy nt het i c Aper t ur e Radar

You might also like