Effects of spectral response function on surface reflectance and NDVI measured with moderate resolution satellite sensors


Trishchenko, A.P.; Cihlar, J.; L.Z.anQing

Remote Sensing of Environment 81(1): 1-18

2002


We report the results of a modelling study on the sensitivity of normalized difference vegetation index (NDVI) and surface reflectance to differences in instrument spectral response functions (SRF) for various Advanced Very High Resolution Radiometers (AVHRR) onboard the National Oceanic and Atmospheric Administration's (NOAA) satellites NOAA-6-16 as well as the Moderate Resolution Imaging Spectroradiometer (MODIS), the Vegetation sensor (VGT), and the Global Imager (GLI). The surface types assumed in this study were coniferous forest, deciduous broadleaf forest, closed and open shrubland, drygrass/savannah, grassland, cropland, crop/natural vegetation mosaic, barren/desert, water bodies, fresh snow, and coarse granular snow. Modelling results were validated against real satellite observations employing AVHRR/NOAA-14 and -15 and MODIS, with a very good agreement. It is shown that for identical atmospheric state and similar surface spectral reflectance, the NDVI and spectral reflectances are sensitive to the sensor's SRF. Relative to a reference SRF for AVHRR/NOAA-9, the differences in reflectance among the AVHRRs range from -25 to 12% for visible channel (red) and from -2 to 4% for near-infrared (NIR) channel. Absolute change in NDVI among various AVHRRs ranged from -0.02 to 0.06. The most significant difference was observed for the AVHRR/3. Consistent results were obtained with the AVHRR sensors aboard the following afternoon satellites: NOAA-9, -11, and -12, whereas important discrepancies were found for other AVHRRs aboard NOAA-6 and -10 and especially those launched more recently (NOAA-15 and -16). Reflectance and NDVI measured by MODIS channels 1 and 2 also exhibit significant differences (up to 30-40%) relative to AVHRR. GLI and VGT have some specific features that should be taken into account when intercomparing surface or top of the atmosphere (TOA) reflectance as well as NDVI. Sensitivity of the SRF effect to variable atmospheric state (water vapour, aerosol, and ozone) was also investigated. Polynomial approximations are provided for bulk spectral correction with respect to AVHRR/NOAA-9.

Remote
Sensing
of
Environment
ELSEVIER
Remote
Sensing
of
Environment
81
(2002)
1-18
www.elsevier.com/locateirse
Effects
of
spectral
response
function
on
surface
reflectance
and
NDVI
measured
with
moderate
resolution
satellite
sensors
Alexander
P.
Trishchenko",
Josef
Cihlar
a
,
Zhanqing
Li
a
'
b
'Canada
Centre
for
Remote
Sensing,
Natural
Resources
Canada,
588
Booth
Street
Ottawa,
Ontario,
Canada
Kl
A
0Y7
b
University
of
Maryland,
ESSIC,
College
Park,
MD,
20742,
USA
Received
14
June
2001;
received
in
revised
form
15
October
2001;
accepted
15
October
2001
Abstract
We
report
the
results
of
a
modeling
study
on
the
sensitivity
of
normalized
difference
vegetation
index
(NDVI)
and
surface
reflectance
to
differences
in
instrument
spectral
response
functions
(SRF)
for
various
Advanced
Very
High
Resolution
Radiometers
(AVHRR)
onboard
the
National
Oceanic
and
Atmospheric
Administration's
(NOAA)
satellites
NOAA-6-16
as
well
as
the
Moderate
Resolution
Imaging
Spectroradiometer
(MODIS),
the
Vegetation
sensor
(VGT),
and
the
Global
Imager
(GLI).
Modeling
results
were
validated
against
real
satellite
observations
employing
AVHRR/NOAA-14
and
-15
and
MODIS,
with
a
very
good
agreement.
It
is
shown
that
for
identical
atmospheric
state
and
similar
surface
spectral
reflectance,
the
NDVI
and
spectral
reflectances
are
sensitive
to
the
sensor's
SRF.
Relative
to
a
reference
SRF
for
AVHRR/NOAA-9,
the
differences
in
reflectance
among
the
AVHRRs
range
from
25%
to
12%
for
visible
channel
(red)
and
from
2%
to
4%
for
near-infrared
(NIR)
channel.
Absolute
change
in
NDVI
among
various
AVHRRs
ranged
from
0.02
to
0.06.
The
most
significant
difference
was
observed
for
the
AVHRR/3.
Consistent
results
were
obtained
with
the
AVHRR
sensors
aboard
the
following
afternoon
satellites:
NOAA-9,
-11,
and
-12,
whereas
important
discrepancies
were
found
for
other
AVHRRs
aboard
NOAA-6
and
-10
and
especially
those
launched
more
recently
(NOAA-15
and
-16).
Reflectance
and
NDVI
measured
by
MODIS
channels
1
and
2
also
exhibit
significant
differences
(up
to
30-40%)
relative
to
AVHRR.
GLI
and
VGT
have
some
specific
features
that
should
be
taken
into
account
when
intercomparing
surface
or
top
of
the
atmosphere
(TOA)
reflectance
as
well
as
NDVI.
Sensitivity
of
the
SRF
effect
to
variable
atmospheric
state
(water
vapor,
aerosol,
and
ozone)
was
also
investigated.
Polynomial
approximations
are
provided
for
bulk
spectral
correction
with
respect
to
AVHRR/NOAA-9.
CO
2002
Elsevier
Science
Inc.
All
rights
reserved.
Keywords:
Satellite;
Surface
reflectance;
AVHRR;
MODIS;
GLI;
VGT;
NDVI;
Spectral
response
function;
Spectral
correction;
Global
change
detection
1.
Introduction
Satellite
observation
is
a
convenient
and
feasible
tool
for
global
monitoring
of
atmospheric
and
terrestrial
envir-
onment
due
to
frequent
and
global
coverage.
Among
various
satellite
sensors,
the
Advanced
Very
High
Resolu-
tion
Radiometer
(AVHRR)
onboard
the
National
Oceanic
and
Atmospheric
Administration's
(NOAA)
polar
orbiting
satellites
has
the
longest
record
for
research
and
application
(Cracknell,
1997).
There
are
three
series
of
the
AVHRR
instruments.
The
four-channel
radiometers
AVHRR/1
were
launched
onboard
the
Tiros-N
and
NOAA-6,
-8,
and
-10.
*
Corresponding
author.
Tel.:
+1-613-995-5787;
fax:
+1-613-947-
1406.
E-mail
address:
(A.P.
Trishchenko).
The
five-channel
radiometers
AVHRR/2
were
deployed
on
the
platforms
NOAA-7,
-9,
-11,
-12,
and
-14
followed
by
a
six-channel
radiometer
AVHRR/3
onboard
the
NOAA-15
and
-16.
The
range
of
AVHRR
data
applications
is
very
broad.
To
name
a
few,
e.g.,
the
visible
and
near-infrared
(NIR)
channels
of
AVHRR
are
used
for
retrieving
cloud
parame-
ters
(Rossow,
1989),
solar
radiation
budget
(Hucek
&
Jacobowitz,
1995),
determination
of
absorbed
photosyn-
thetically
active
radiation
(APAR;
Li,
Moreau,
&
Cihlar,
1997),
and
retrievals
of
aerosol
optical
depth
(AOD;
Stowe,
Ignatov,
&
Singh,
1997)
and
other
parameters
(Gutman,
Csiszar,
&
Romanov,
2000;
Nakajima,
Higurashi,
Kawa-
moto,
&
Penner,
2001).
One
of
the
most
important
appli-
cations
of
the
AVHRR
thermal
channels
lies
in
estimation
of
global
sea
surface
temperature
(SST;
Reynolds
&
Smith,
1993).
The
thermal
channels
are
also
used
for
determining
0034-4257/01/$
see
front
matter
2002
Elsevier
Science
Inc.
All
rights
reserved.
PIE
S0034-4257(01)00328-5
2
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
land
surface
temperature
and
emissivity
(Qin
&
Karnieli,
1999).
Thermal
AVHRR
channels
in
combination
with
channels
1
and
2
are
employed
for
forest
fire
detection
and
monitoring
(Li,
Nadon,
&
Cihlar,
2000).
An
important
application
of
AVHRR
solar
channels
is
the
retrieval
of
surface
reflectance
to
determine
different
land
surface
parameters
such
as
surface
cover
type
(Cihlar
et
al.,
in
press),
normalized
difference
vegetation
index
(NDVI;
Kidwell,
1994),
leaf
area
index
(LAI;
Chen,
Rich,
Gower,
Norman,
&
Plummer,
1997),
and
other
surface
characteristics.
New
opportunities
for
global
mon-
itoring
of
terrestrial
ecosystem
are
unfolding
with
the
availability
of
Moderate
Resolution
Imaging
Spectroradi-
ometer
(MODIS)
data.
The
processing
of
satellite
data
involves
many
steps.
The
final
purpose
of
satellite
data
processing
in
land
surface
studies
is
to
obtain
the
systematic
maps
of
various
quant-
itative
physical
parameters
corrected
for
the
intervening
effect
of
atmosphere,
effect
of
varying
observational
geo-
metry,
and
specific
sensor
properties.
Some
of
these
correc-
tions
can
be
done
quite
accurately,
like
correction
for
Raleigh
molecular
scattering.
Nonetheless,
most
of
them
may
be
implemented
with
some
uncertainty
due
to
limited
knowledge
of
input
information.
The
important
processing
step
is
the
data
calibration.
Despite
numerous
efforts,
the
results
often
vary
among
different
investigators
(Brest,
Rossow,
&
Roiter,
1997;
Gut-
man,
1999;
Masonis
&
Warren,
2001;
Rao
&
Chen,
1999;
Tahnk
&
Coakley,
2001).
Accurate
calibration
requires
continuos
monitoring
of
the
gain
and
offset
due
to
degrada-
tion
of
sensor
sensitivity
with
time.
The
degradation
may
not
necessarily
be
a
linear
function
of
time
(Tahnk
&
Coakley,
2001).
It
is
commonly
agreed
that
for
satellite
sensors
lacking
onboard
calibration
in
solar
spectrum,
the
total
relative
uncertainties
of
calibration
are
within
5%
(Rossow
&
Schiffer,
1999).
An
essential
part
of
this
uncertainty
is
related
to
the
effect
of
spectral
response
function
(SRF),
when
it
is
not
accounted
for
properly
during
vicarious
calibration
or
sensor
intercalibration
(Teillet
et
al.,
2001).
Variable
sun
and
observational
geometry
induces
another
source
of
systematic
noise
(Gutman,
Gruber,
Tarpley,
&
Taylor,
1989;
Li,
Cihlar,
Zheng,
Moreau,
&
Ly,
1996).
This
angular
effect
is
a
combination
of
anisotropic
reflective
properties
of
the
atmosphere
and
land
surface.
The
effect
must
be
accounted
for
in
long-term
studies
of
satellite
data
to
obtain
unbiased
results
(Cihlar
et
al.,
1998;
Gutman,
1999).
This
is
achieved
by
normalizing
satellite
image
to
common
geometry
using
empirical
anisotropic
factors.
They
are
derived
either
from
sequence
of
satellite
scenes
collected
over
long
period
of
time
(Cihlar
et
al.,
1998;
Trishchenko,
Li,
Park,
&
Cihlar,
in
press)
or
from
special
directional
obser-
vations,
like
those
ones
from
POLDER
or
MISR
instruments
(Csiszar,
Gutman,
Romanov,
Leroy,
&
Hautecoeur,
2001).
Neglecting
angular
correction
in
the
AVHRR
data
pro-
cessing,
for
example,
may
introduce
biases
in
composite
coarse
resolution
long-term
reflectance
datasets
of
the
order
of
1-2%
depending
on
spectral
band
and
surface
type
(Gutman,
1999).
The
effect
becomes
more
significant
(5-10%)
for
solar
zenith
angles
(SZA)
greater
than
55
°
.
Numerous
vegetation
indices
have
been
developed
to
monitor
the
state
of
vegetation
from
spaceborne
instruments
(Bannari,
Morin,
&
Bonn,
1995).
They
were
constructed
to
diminish
atmospheric
contamination,
mitigate
the
influence
of
soil
spectral
reflectance
signatures,
or
emphasize
certain
features
of
vegetation
conditions.
The
set
of
advanced
vegetation
indices
optimised
for
upcoming
sensors
is
dis-
cussed
by
Gobron,
Pinty,
Verstraete,
and
Widlowski
(2000).
Nevertheless,
NDVI
remains
the
basic
vegetation
index
most
widely
employed
for
global
monitoring
of
vegetation.
It
is
defined
as
the
following
ratio:
NDVI
=
p
N1R
p
red
7
where
p
and
p
red
are
reflectances
for
visible
(red)
and
NIR
spectral
bands.
Attempts
have
been
made
to
use
the
AVHRR
data
for
long-term
monitoring
of
land
reflectances
and
vegetation
indices
(Cihlar
et
al.,
in
press;
Gutman,
1999;
Kaufman
et
al.,
2000).
These
and
other
studies
on
long-term
monitor-
ing
are
motivated
by
the
availability
of
quality
AVHRR
time
series
for
the
period
of
nearly
20
years.
Although
the
construction
and
characteristics
of
all
AVHRR
instruments
are
quite
similar,
they
are
not
identical
among
all
missions.
Consequently,
the
effect
of
varying spectral
response
may
create
an
artificial
noise
imposed
upon
a
subtle
natural
variability.
This
artifact
should
be
examined
thoroughly
before
comparing
data
between
different
missions
to
deter-
mine
possible
changes
in
satellite
climatic
records.
So
far,
the
effects
of
SRFs
have
not
been
considered
carefully
in
such
studies.
Some
influence
of
the
spectral
characteristic
of
the
satellite
sensors
on
remote
sensing
of
vegetation
indices
has
been
studied
for
forested
regions
(Teillet,
Staenz,
&
Williams,
1997)
and
during
vicarious
calibrations
proce-
dures
(Teillet
et
al.,
2001).
Nevertheless,
systematic
charac-
terisation
of
these
effects
for
various
representative
surface
spectral
signatures
on
a
global
scale
and
for
all
AVHRR
sensors
has
not
been
addressed
properly.
Analysis
of
long-
term
satellite
products
from
various
missions
may
require
corrections
to
account
for
differences
in
SRF
that
have
not
been
investigated.
Our
study
is
aimed
to
fill
this
gap
and
to
provide
quantitative
estimates
for
the
effect
of
SRF
among
all
AVHRR
missions.
Differences
between
AVHRR
and
MODIS, Global
Imager
(GLI),
and
Vegetation
sensors
(VGT)
are
also
considered.
To
achieve
this
goal,
some
representative
surface
spectral
reflectance
curves
were
selected
from
two
observation
sources.
The
first
is
a
database
of
spectral
observation
made
by
the
PROBE-1
instrument
(Seeker,
Staenz,
Budkewitsch,
&
Neville,
1999)
at
the
Canada
Centre
for
Remote
Sensing
(CCRS).
The
second
one
is
the
Advanced
Spaceborne
Thermal
Emission
NIR
red
P
P
Ch.1
C
h
.2
I
N6
N10
green
vegetation
N
7
N-9
-
N
11
N-12
N-14
\
••
•••1
0
.
1.01.1.
N15
- -
-N16
MODIS
VGT/SPOT
GLI/ADEOS-II
:;
i;
.
71;t
i
t
4Th
ji
100
80
60
40
20
100
80
60
40
20
100
80
60
40
20
0
0
0
0
(r)
V
0
a)
Q
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
3
and
Reflection
Radiometer
(ASTER)
spectral
library
(avail-
able
from
http://speclibbl.nasa.gov
).
The
details
of
the
databases
are
given
below.
The
6S
radiative
transfer
code
(Vermote,
Tame,
Deuze,
Herman,
&
Morcette,
1997)
was
employed
for
model
simulation
of
the
signal
at
the
top
of
the
atmosphere
(TOA)
level
under
various
atmospheric
condi-
tions
and
observational
geometries.
The
paper
is
organized
as
follows.
Section
2
describes
the
special
features
of
instrument
SRFs.
Section
3
discusses
the
surface
spectral
library
and
modeling
of
satellite
signal
at
the
TOA
level.
Section
4
presents
results
and
an
analysis
of
comparisons
between
various
sensors.
Validation
results
of
the
model
simulation
using
real
satellite
observations
are
shown
in
Section
5.
Section
6
summarizes
the
research.
2.
Sensor
SRFs
The
SRFs
for
AVHRR
NOAA-6-16,
MODIS,
GLI,
and
VGT
compatible
channels
in
visible
and
NIR
are
shown
in
Fig.
la—c.
The
three
panels
of
Fig.
1
present
SRFs
for
different
types
of
AVHRR:
AVHRR/1
(a),
installed
on
morning
satellites,
AVHRR/2
(b),
which
was
operational
on
afternoon
satellites,
and
a
morning
satellite
NOAA-12.
The
bottom
panel
(c)
shows
SRFs
for
AVHRR/3
(NOAA-15
and
-16),
MODIS,
GLI,
and
VGT.
A
typical
spectrum
of
green
vegetation
is
also
plotted
for
reference.
Though
similar,
these
curves
differ
in
shape,
the
central
wavelength
location,
the
bandwidth,
and
the
degree
of
overlap
between
channels,
especially
with
respect
to
the
transition
from
chlorophyll
absorption
band
to
foliage
reflection
band
(0.68-0.72
µm).
Noticeable
differences
are
seen
among
various
AVHRRs.
The
most
notable
differences
exist
between
AVHRR/3
on
NOAA-15
and
-16
and
other
AVHRRs
(Fig.
lb
and
c).
The
channels
of
the
new
AVHRR/3
have
narrower
bandwidths
and
a
much
smaller
overlap
over
the
vegetation
transition
band.
All
these
factors
affect
the
magnitude
of
spectral
reflectance
observed
by
the
sensors
and
lead
to
higher
NDVI
values
over
vegetated
surfaces.
Spectral
response
functions
0.6
0.7
0.8
0
.
9
1.0
Wavelenghth
[1.1m]
Fig.
1.
SRFs
of
visible
(red)
and
NIR
channels
for
AVHRR
NOAA-6-16
and
MODIS,
VGT,
and
GLI.
(a)
AVHRR/1.
Morning
satellites
NOAA-6,
-8
and
-10.
(b)
AVHRR/2.
Afternoon
satellites
NOAA-7,
-9,
-11,
and
14
and
morning
satellite
NOAA-12.
(c)
AVHRR/3
(NOAA-15
and
-16),
VGT/SPOT,
and
GLI/ADEOS-Il.
Typical
spectral
reflectance
curve
for
green
vegetation
is
shown
on
each
panel.
40
30
20
10
0
50
40
30
20
1
2
I
...
..
I
...
..
.
I
-
..3--
a
80
-
1
Barren/desert
- - -
2
Water
70
-
3
Fresh
snow
4
Coarse
granular
snow
60
-
5
Drygrass/savanna
50
-
6
Coniferous
1
-
7
Coniferous
2
8
Deciduous
1
-
9
Deciduous
2
10
Closed
Shrubs
-
11
Open
Shrubs
12
Grass
1
-
13
Grass
2
-14
Grass
3
-
15
Grass
4
16
Cropland
-----17
Cropmosaic
....
11
-12
16'
.
_
_
:.
13
-
7
b)
-13
-
-
-
----------
100
-
90
Re
flecta
nce
[
%]
Re
flec
ta
nce
(
%)
10
-15
'"
,I
I I I I I
I
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
Wavelength
[pin]
i
)
a
Visible
(red)
channel
V
A
surface
TOA
N.
-
b)
Near-IR
channel
r////d
surface
TOA
-
c
NDVI
5
;
/'
surfac
N
TOA
s
c1)
E
8
0
a)
-o
2
_c
T4
-0
2
0
_c
O
co
0)
0)
-cs
0,
O
0
4
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
The MODIS
channels
are
also
quite
distinct
from
the
AVHRR
ones
(Fig.
1c).
They
are
much
narrower
and
have
no
overlap
with
each
other
over
the
vegetation
transition
band.
The
MODIS
channel
1
is
also
shifted
further
to
NIR
region
relative
to
the
AVHRR
channel
1.
The
GLI
red
channel
is
close
to
MODIS,
although
it
is
shifted
more
toward
the
MR.
The
width
of
the
GLI
MR
channel
is
between
those
for
AVHRR
and
MODIS.
The
VGT
red
channel
extends
beyond
0.7-pm
limit,
thus
causing
signifi-
cant
impact
on
the
red
reflectances
and
NDVI.
Fig.
1
indicates
that
a
direct
comparison
of
spectral
reflectance
or
vegetation
indices
produced
by
various
sensors
should
be
made
with
caution.
3.
Surface
spectral
data
and
modeling
To
encompass
a
potential
range
of
variability
in
surface
reflectance
and
NDVI,
a
set
of
representative
spectra
for
various
surface
targets
were
compiled,
following
the
clas-
sification
scheme
used
in
the
NASA
Surface
and
Atmo-
spheric
Radiation
Budget
(SARB)
Project
(Rutan
&
Charlock,
1997).
The
complete
scheme
for
the
SARB
Fig.
2.
Spectra
of
the
surface
targets
used
in
simulations.
They
were
normalized according
to
Eq.
(1).
Top
panel
(a)
shows
spectra
of
nonvegetated
surfaces
and
bottom
panel
(b)
shows
spectra
for
vegetated
surfaces.
Table
1
Surface
types
assumed
in
this
study
Surface
types
1.
Coniferous
forest
(1
'
2)
2.
Deciduous
broadleaf
forest
(1
.
2)
3.
Closed
shrubland
4.
Open
shrubland
5.
Drygrass/savannah
6.
Grassland°
-4)
7.
Cropland
8.
Crop/natural
vegetation
mosaic
9.
Barren/desert
10.
Water
bodies
11.
Fresh
snow
12.
Coarse
granular
snow
Superscripts
in
denote
the
number
of
spectral
curves
used
in
simulations
(Fig.
2).
Project
included
20
different
surface
classes.
Since
we
had
no
measurements
for
some
of
the
surface
types
and
yet
the
particular
focus
of
this
study
is
on
the
boreal
ecosystem,
12
classes
were
adopted
in
this
investigation
(Fig.
2
and
Table
1).
Surface
classes
of
lower
class
number
nevertheless
covered
the
bulk
of
variability
in
spectral
reflectance
Surface
and
TOA
reflectances
and
NDVI
for
AVHRR/NOAA-9
Fig.
3.
Surface
and
TOA
reflectances
and
NDVI
for
the
AVHRR/NOAA-9
for
selected
spectra.
AVHRR
onboard
the
NOAA-9
satellite
is
considered
as
the
reference
sensor
following
to
ISCCP
approach
(Rossow
&
Schiffer,
1999).
Visible
(red),
NIR
reflectances,
and
NDVI
are
presented
in
sequence
from
top
to
bottom.
Note
the
break
in
vertical
scale
between
0.4
and 0.8
for
the
visible
channel.
1.0
0.3
0.2
0.1
0.0
0.8
0.6
0.4
0.2
0.0
0.8
0.6
0.4
0.2
0.0
-0.2
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
5
occurred
in
nature.
Classes
1
and
2
are
depicted
by
two
spectral
curves
each,
and
class
6
(grassland)
is
described
by
four
curves
with
different
values
of
NDVI
to
reflect
various
vegetation
density
levels.
Spectral
curves
were
derived
from
two
sources:
aircraft
observations
from
the
PROBE-1
instrument
made
by
the
Canada
Centre
for
Remote
Sensing
(Secker
et
al.,
1999).
The
PROBE-1
is
airborne
hyperspectral
sensor
covering
visible
and
NIR
spectral
regions
with
128
spectral
bands.
The
second
source
was
acquired
from
the
ASTER
spectral
library.
We
normalized
each
spectrum
to
reproduce
SARB
broadband
albedo,
i.e.,
Xmax
Xmin
7
Xmax
f
So(X)p(X)dX
Xmin
where
S
o
(X)
and
p(X)
are
the
solar
spectral
constant
and
observed
spectral
reflectance,
and
u
sARB
is
the
broadband
albedo
from
Rutan
and
Charlock
(1997).
The
normalization
(Eq.
(1))
provides
a
link
of
individual
spectral
curves
to
a
specific
SARB
surface
type.
The
derived
spectra
(Fig.
2)
are
shown
for
the
wavelength
interval
0.35-1.25
µm,
which
essentially
covers
visible
and
NIR
portion
of
the
solar
P(x)
=
otsARB
(1)
Difference
in
surface
reflectance
for
visible
(red)
channel
relative
to
AVHRR/NOAA-9
15
10
5
0
-5
-10
-15
relative
difference
N
6
absolute
difference
a'
I
I
_
N7_
N8=
0.015
0.010
0.005
0.000
-0.005
-0.010
At
'Er
=
=
0
I
0.015
0.015
N10
E
N11=
N12
=
e
15
0
10
(7)
0.010
5
0.005
<
0
A.
0.000
-AA-S-A
fa
A
0
-5
z
-0.005
-10
ak
=
-0.010
0
-15
CC
0.015
z
0)
°
20
z
1
-
N14
N15
N16
0.015
>
cc
0.010
co
10
0.005
CC
CC
0
0.000
-10
5
••
7
-0.005
-0.010
a
CC
-20
1
8,
=
-0.015
I
I
I
30
MODIS
VGT/SPOT
GLI/ADEOS]
0.02
20
0.01
10
A
0
A
0.00
A
A
7
=
-10
-0.01
-20
-30
-0.02
°
-0.2
0.0
0.2
0.4
0.6 0.8
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6 0.8
NDVI
surf
Fig.
4.
Absolute
(solid
triangles)
and
relative
(open
circles)
differences
in
visible
(red)
channel
reflectances
at
the
surface
level
with
respect
to
AVHRR/NOAA-
9.
The
left
scale
is
for
relative
difference
and
the
right
scale
is
for
absolute
difference.
All
data
points
are
plotted
versus
NDVI
of
particular
sensor.
Quadratic
best
fits
for
absolute
(solid)
and
relative
(dashed)
differences
are
also
shown.
Parameters
of
fitting
curves
are
given
in
Table
2.
6
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
Difference
in
surface
reflectance
for
NIR
channel
relative
to
AVHRR/NOAA-9
4
2
0
-2
-4
0.03
0
02
0.01
0.00
-0.01
-0.02
o
relative
difference
N6
-
A
-
absolute
difference
_
N7
-
N8
0.03
0.03
I
'
I
N10
N11
-
N12
-
4
0
0.02
4c
2
0.
0.01
0
A
<
0
z
-2
cc
z
c-4
0.00
-0.01
-0.02
z
-AAA
0.03
cc
0.03
-A.--
9.,
-
A&
D
.
A
6,
-
AA-
4
0
z
2
cc
z
0
cc
-2
0.02
0.01
0.00
-0.01
cc
CC
A
A
A
-
cc
2
-4
N14
_
N15
7
N16
02
.
CC
I I
0.03
0.06
A
-
15
0.04
10
5
k
0
.
0.02
0
o
0.00
-5
-0.02
-10
-0.04
-15
MODIS:
VGT/SPOT
GLI/ADEOS:
-0.06
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6 0.8
NDVI
surf
Fig.
5.
Similar
to
Fig.
4,
but
for
the
NIR
channel
at
the
surface
level.
Parameters
of
fitting
curves
are
given
in
Table
3.
spectrum
under
study.
The
top
panel
(Fig.
2a)
depicts
spectra
of
nonvegetated
surfaces
and
a
dry
grass/savannah
surface,
and
the
bottom
panel
(Fig.
2b)
shows
spectra
for
various
vegetated
surfaces.
The
steep
increase
in
reflectivity
above
0.7
µm
is
a
characteristic
feature
of
these
curves.
The
data
selected
encompass
spectral
differences
between
various
natural
surface
classes.
4.
Modeling
results
4.1.
Radiative
transfer
modeling
The
6S
radiative
transfer
model
was
employed
to
simu-
late
the
TOA
signal.
The
wavelength
increment
of
the
model
is
2.5
nm,
which
allows
us
to
accurately
resolve
all
spectral
features
of
the
targets
and
instrument
SRF.
Baseline
simulations
were
conducted
for
US62
atmospheric
profile
with
total
water
vapor
(TWV)
columnar
amount
scaled
to
1.5
cm,
representative
for
the
boreal
region
in
summer
time
(Cihlar,
Tcherednichenko,
Latifovic,
Li,
&
Chen,
2000).
The
ozone
content
was
set
to
350
Dobson
units
(DU).
Since
satellite
studies
of
surface
properties
usually
employ
clear-sky
composites
selected
for
the
high-
est
atmospheric
transparency,
the
AOD
was
set
to
0.06,
as
recommended
by
Fedosejevs
et
al.
(2000)
for
the
Canadian
boreal
zone.
The
surface
reflectance
was
assumed
to
be
Lambertian,
i.e.,
independent
of
sun-sensor
geometry.
The
TOA
reflectance
has
a
certain
dependence
on
observational
geometry
due
to
atmospheric
effects.
We
conducted
com-
putations
for
various
geometrical
conditions:
the
SZA
0
0
varying
from
0
°
to
75
°
with
15
°
steps,
the
viewing
zenith
angle
(VZA)
0
from
0
°
to
45
°
with
15
°
steps,
and
the
relative
azimuth
angle
(RAA)
it,
was
set
to
0
°
,
90
°
,
and
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
7
Difference
in
NDVI
computed
at
the
surface
level
relative
to
AVHRR/NOAA-9
6
N6
-
N7
-
I
I
N8
-
0.09
4
5er"
'
Ei
-a
0
0.06
2
0
-
o
00
.
--------
Tre--
0
-
°
6
61" -7
0.03
0
0.00
-2
00
7
-
0
-0.03
0
-4
.6
o
A
relative
difference
-
absolute
difference
I
,
I
I
-
0
I
I
-0.06
-0.09
0
6
0.09
d
-
o
0
N10
-
N
j
1
N12
=
4
0.06
9'
0
2
0.03
0
0
AA,
a
-
4
0.00
>
-2
-0.03
-4
0
-0.06
0
-6
I
I
-0.09
z
=C
15
0.09
I
I
0
tr,
12
9
6
00
-
-
0
o
0
sP
:-"-
0.06
0.03
0
z
3
0
0
0.00
-3
-0.03
-6
-9
-0.06
0
Z
-12
-15
N14
=
N15
-_
N16
I
-0.09
30
25
0.09
A
20
4
0.06
15
10
5
0
A
s
0.03
0
0.00
-5
-10
-0.03
-15
-20
-25
MODIS
=
VGT/SPOT
°
GLI/ADEOS
-0.06
0.09
-30
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
NDVI
surf
Fig.
6.
Similar
to
Figs.
4
and
5,
but
for
NDVI
at
the
surface
level.
Parameters
of
fitting
curves
are
given
in
Table
4.
Table
2
Parameters
of
quadratic
best
fit
to
absolute
spectral
correction
Ap
=
p
-
pN0AA-9
and
relative
spectral
correction
Ap
=
(p
-
p
N0AA
_
9
)/(p
N0AA
-
9
)
(%)
for
visible
(red)
channel
Instrument
Absolute
correction
r
2
a
Relative
correction
(%)
r
2
v
(%)
AVHRR/N-6
0.00035
-
0.0189X+
0.0141X
2
.80
0.0013
-
0.160
-
0.445X-
19.525X
2
.97
0.97
AVHRR/N-7
0.00026
-
0.0153X+
0.0127X
2
.77
0.0010
0.108
-
2.230X-11.050X
2
.96
0.76
AVHRR/N-8
-
0.00056
+
0.0014X+
0.0033X
2
.77
0.0007
0.087
-
9.037X+
21.721X
2
.81
1.37
AVHRR/N-10
0.00037
-
0.0195X+
0.0153X
2
.79
0.0014
-
0.159
-
1.411X-
16.949x
2
.96
1.0
AVHRR/N-11
0.00001
-
0.0012X+
0.0005X
2
.76
0.0002
0.006
+0.0744X-
2.335X
2
.96
0.12
AVHRR/N-12
-
0.00022
+
0.0027X-
0.0035X
2
.21
0.0005
0.073
+0.1604X-
0.947x
2
.14
0.47
AVHRR/N-14
-
0.00046+
0.0112X-
0.0077X
2
.82
0.0008
0.116
-
2.951X+
18.076X
2
.94
0.87
AVHRR/N-15
0.00029
-
0.0222X+
0.0117X
2
.81
0.0021
-
0.105
+3.115X-
36.306x
2
.98
1.31
AVHRR/N-16
0.00028
-
0.0217X+
0.0123X
2
.80
0.0020
-
0.096
+2.044X-
32.746X
2
.98
1.26
MODIS
-
0.00037
-
0.0118X-
0.0051X
2
.73
0.0035
0.046
+12.136X-
56.504x
2
.98
1.87
VGT/SPOT
-
0.00086+
0.0305X-
0.0404X
2
.42
0.0034
0.570
+16.234X-
27.183X
2
.58
2.29
GLI/ADEOS
-
0.00063
+
0.0103X-
0.0297X
2
.62
0.0045
0.419
+21.432X-
65.063x
2
.96
2.51
Surface
level.
X
denotes
NDVI
for
particular
sensor
computed
at
the
surface
level.
8
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
Table
3
Similar
to
Table
2,
but
for
NIR
channel
Instrument
Absolute
correction
r
2
Relative
correction
(%)
r
2
v
(%)
AVHRR/N-6
-
0.00069+0.00443X-
0.0021X
2
.79
0.0005
-
0.0777+0.8707X-
0.27881
2
.95
0.05
AVHRR/N-7
-
0.00049+0.00142X-
0.0045X
2
.77
0.0004
-
0.0682
-
0.2536X-
0.5465X
2
.90
0.08
AVHRR/N-8
0.00005
-
0.00385X-
0.002X
2
.88
0.0006
-
0.0205
-
1.5927X+
0.1487X
2
.96
0.10
AVHRR/N-10
-
0.00073
+0.00745X+
0.0031X
2
.89
0.0012
-
0.0539
+
2.3498X+
0.3563X
2
.98
0.13
AVHRR/N-11
-
0.00008
-
0.0001X+
0.0002X
2
.11
0.0001
-
0.0140
-
0.1087X+
0.1505X
2
.19
0.02
AVHRR/N-12
-
0.00025
+
0.00264X+
0.0014X
2
.90
0.0004
-
0.0184
+
0.8199X+
0.2289X
2
.98
0.04
AVHRR/N-14
-
0.00335
+0.02615X-
0.0168X
2
.74
0.0026
-
0.3457+5.4112X-
3.1057X
2
.89
0.33
AVHRR/N-15
-
0.00082
+0.01153X+
0.005
1X
2
.88
0.0020
-
0.03087
+
4.0655X+
0.210X
2
.96
0.29
AVHRR/N-16
-
0.00164
+0.01696X+
0.0002X
2
.86
0.0023
-
0.12313
+5.0171X-
0.669X
2
.96
0.32
MODIS
0.00101
+0.017881+0.028X
2
.82
0.0069
0.4773
+9.8974X+
1.9483X
2
.89
1.46
VGT/SPOT
0.00349
-
0.00826X+
0.0532X
2
.81
0.0059
0.4960
+
4.1976X+
7.6205X
2
.88
1.25
GLI/ADEOS
0.0056
-
0.021X+
0.0605X
2
.73
0.0070
0.6173
+2.1544X+
8.6714X
2
.83
1.52
Surface
level.
X
denotes
NDVI
for
particular
sensor
computed
at
the
surface
level.
180
°
.
For
sensitivity
tests, the
following
values
were
chosen:
0,
0.06,
and
0.6
for
AOD;
0.5,
1.5,
and
5.0
cm
for
TWV;
270,
350,
and
430
DU
for
ozone.
The
simulations
were
done
for
a
viewing
geometry
at
0
0
=45
°
,
0=0
°
,
and
=
0
°
,
a
typical
geometry
for
normalizing
satellite
obser-
vations
(Trishchenko
et
al.,
in
press).
The
range
of
variability
in
reflectance
and
NDVI
values
simulated
for
the
AVHRR
on
board
NOAA-9
over
all
surface
targets
is
illustrated
in
Fig.
3.
This
radiometer
is
often
considered
as
a
reference
instrument
(e.g.,
Rossow
&
Schiffer,
1999).
Note
the
break
in
vertical
scale
for
visible
channel
in
Fig.
3a,
which
was
introduced
to
show
enough
details
for
low
reflective
surfaces
as
well
as
the
highly
reflective
snow
surface.
The
reflectance
in
visible
(red)
band
for
vegetated
surfaces
ranges
from
0.05
to
0.15.
At
the
surface,
the
values
are
typically
smaller
because
scat-
tering
by
atmospheric
molecules
adds
to
the
signal
reflected
from
darker
surfaces.
The
opposite
is
true
for
the
NIR
band
where
surface
reflectance
is
higher
than
at
the
TOA,
because
atmospheric
attenuation
outweighs
scat-
tering
back
to
the
sensor.
These
relationships
also
explain
why
NDVI
at
the
TOA
level
is
typically
smaller
than
at
the
surface
level.
The
NDVI
is
either
very
small
or
even
negative
for
nonvegetated
surfaces,
such
as
water,
snow,
and
barren/desert
classes.
4.2.
Surface
level
Figs.
4-6
show
the
surface
level
results
for
visible
(red),
NIR
reflectances,
and
NDVI.
The
absolute
and
relative
differ-
ences
with
respect
to
AVHRR/NOAA-9
values
are
plotted
against
NDVI
for
each
sensor.
The
difference
is
also
referred
to
as
the
spectral
correction
factor.
The
least
differences
are
found
for
AVHRR/NOAA-11
followed
by
NOAA-12.
Other
AVHRR/1,2
are
reasonably
close
to
AVHRR/NOAA-9,
although
the
differences
could
reach
0.01
(10-15%
relative)
for
red
channel
(Fig.
4),
0.01
(2-3%)
forNIR
channel
(Fig.
5),
and
0.03
(4-6%)
for
NDVI
of
vegetated
surfaces
(Fig.
6).
Since
NDVI
for
sparse
vegetation
and
nonvegetated
targets
are
small,
the
relative
differences
in
NDVI
for
these
surface
types
are
larger.
The
sensor
spectral
reflectances
and
NDVI
differ
sys-
tematically
for
AVHRR/3
onboard
the
NOAA-15
and
-16.
The
visible
(red)
reflectance
for
AVHRR/3
is
smaller
by
0.01-0.015
(20-25%;
Fig.
4),
while
the
NIR
channel
reflectance
is
larger
by
0.01-0.015
(3-4%;
Fig.
5).
As
a
Table
4
Similar
to
Tables
2
and
3,
but
for
NDVI
Instrument
Absolute
correction
r
2
Relative
correction
(%)
r
2
v
(%)
AVHRR/N-6
0.00005
+
0.052X-
0.022781
2
.97
0.0021
3.84
+3.7437X-
5.227X
2
.43
0.48
AVHRR/N-7
0.00001
+0.03632X-
0.0196X
2
.93
0.0018
3.315
-
0.3486X-1.459X
2
.50
0.50
AVHRR/N-8
0.00006
-
0.0018%-
0.0205X
2
.88
0.0021
-
3.0562
+
9.4069X-
10.072X
2
.50
0.84
AVHRR/N-10
0.0002+0.0648%-
0.0372X
2
.97
0.0020
5.301
+2.526X-
5.774X
2
.77
0.45
AVHRR/N-11
-
0.0001
+
0.0031X+
0.00072X
2
.87
0.0005
-
0.1952+
1.601X-1.137X
2
.47
0.25
AVHRR/N-12
-
0.00032
+
0.0031X+
0.0007X
2
.39
0.0015
-
1.316+5.1X-
3.795X
2
.44
0.82
AVHRR/N-14
-
0.00201
+0.0099X-
0.0304X
2
.93
0.0013
6.986
-
34.3909X+
30.8637X
2
.83
1.81
AVHRR/N-15
-
0.00026+0.0877X-
0.0307X
2
.97
0.0038
5.4
+10.478X-
10.894X
2
.57
0.82
AVHRR/N-16
-
0.00061+0.091X-
0.0391X
2
.97
0.0034
8.139
-
0.4926X-
2.1904
X
2
.29
1.25
MODIS
0.00068+0.1199X-
0.03383X
2
.94
0.0084
-
3.993
+61.4265X-
53.129X
2
.59
5.70
VGT/SPOT
-
0.0006
-
0.0153X+
0.05836X
2
.54
0.0104
-
15.758+61.013X-
47.087X
2
.82
3.77
GLI/ADEOS
-
0.00086
+
0.0295X+
0.0667X
2
.81
0.0149
-
20.982
+97.84X-
74.127X
2
.89
4.62
Surface
level.
0
difference
relative
N81
0.010
0.005
0.000
-0.005
-0.010
6
A
absolute
difference
N10
N1
1
11
.
_
N12
0.010
0.005
0.000
%
42
"
1611
0
6
0*
26
iRa
P-
___
rn
-0.005
0
-0.010
z
0
rn
CC
N14
N15:_
N16
0.010
0
0.005
CC
0.000
'ov
e
,
2„
-0.005
-0.010
1'11'1'
MODIS
-
VGT/SPOT
G
LI/AD
EOS-
0.03
0.02
0.01
0.00
-0.01
-0.02
-0.03
I
15
10
5
0
-5
-10
-15
0
15
T-
10
rn
0
0
z
-5
0
(/)
-10
5
CC
-15
nI
rn
0
z
C s
-
0
15
10
5
ci)
>
CC
0
o
-
5
co
5
-10
-15
20
10
0
-10
-20
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
9
Difference
in
visible
(red)
channel
TOA
reflectance
relative
to
AVHRR/NOAA-9
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
NDVI
toa
-0.2
0.0
0.2
0.4
0.6
0.8
Fig.
7.
Absolute
(solid
triangles)
and
relative
(open
circles)
differences
in
visible
(red)
channel
reflectances
at
the
TOA
levels
with
respect
to
AVHRR/NOAA-9.
Solid
lines
correspond
to
quadratic
fit
to
relative
difference.
Parameters
of
fitting
curves
are
given
in
Table
5.
result,
NDVI
derived
from
AVHRR/NOAA-15
or
-16
is
higher
by
0.03-0.06
(5-10%;
Fig.
6).
These
differences
are
due
to
(1)
a
significantly
narrower
spectral
band
of
the
visible
(red)
channel
that
is
much
less
contaminated
by
the
elevated
reflection
in
MR
and
(2)
the
MR
channel
that
is
less
influenced
by
the
transition
band
(Fig.
1c).
The
SRFs
of
MODIS
and
GLI
are
so
different
from
that
of
AVHRR/NOAA-9
that
the
surface
reflectance
differences
reaches
0.02
(20-30%)
in
visible
channel
(Fig.
4),
0.04-
0.05
(10-15%)
in
NIR
channel
(Fig.
5),
and
0.06-0.09
(20-
25%)
in
NDVI
(Fig.
6).
The
SRF
effect
for
VGT
is
smaller
than
for
MODIS
and
GLI
and
comparable
to
AVHRR/3.
The
10
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
Difference
in
NIR
channel
TOA
reflectance
relative
to
AVHRR/NOAA-9
4
2
0
-2
-4
0.03
0.02
0.01
0.00
-0.01
-0.02
0
relative
*•••••••4.g./...
4,
difference
difference
N
6
•••.4-....44,:ite
N7
N8
absolute
0.03
0.03
N10
-
t4i;
41
N11
N12
O
O
4
0.02
rn
0.01
<
0
4,
0.00
rn
0
z
6
0
-2
-0.01
0
z
cc
-0.02
z
-4
cc
CC
0.03
CC
(3)
4
0.03
Co
I
T
I
0
z
2
6
,
>•.
*•
11
"
0.02
0.01
CC
CC
z
0
0.00
CC
0
cc
z
-2
-4
N14
0
N15
N16
-0.01
-0.02
CC
1
0.03
40
I
-
I
I
H
30
J
f
0.15
20
I
:t IT
!
tu,2111;
0.10
10
0
47:
4
44
4
;Z,
ele
,
0.05
0
0.00
-10
-0.05
-20
-0.10
-30
MODIS
VGT/SPOT
GLI/ADEOS::
-0.15
-40
I
I
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
-0.2
0.0
0.2
0.4
0.6
0.8
NDVI
toa
Fig.
8.
Similar
to
Fig.
7,
but
for
NIR
channel
reflectances
at
the
TOA
level.
Parameters
of
fitting
curves
are
given
in
Table
6.
difference
could
be
as
large
as
0.01
(
10%)
in
the
visible
(red)
channel,
0.02-0.04
(5-10%)
in
the
NIR
channel,
and
0.03
(
5-10%)
in
NDVI
for
vegetated
surfaces.
The
discrepancies
caused
by
different
SRFs
may
be
corrected
using
the
second
degree
polynomial
functions,
as
shown
in
Figs.
4-6.
The
curves
were
produced
by
fitting
I
'
I ' I
'
.
;
VGT/SPOT-
k
I
,
I ,
1,
I
I
-0.2
0.0
0.2
0.4
0.6
0.8
NDVI
toa
9
6
3
0
-3
-6
o
-9
0
0
9
6
<
3
0
0
0
-3
z
-6
-9
cr)
<
15
O
12
c6
9
9
6
>
3
0
0
z
-3
-6
0
-9
-12
-15
120
100
80
60
40
20
0
-20
-40
N6-
relative
difference
!
absolute
difference
-
-
I
1'1'1'1'
N
0
-
-1
,`.•
N7-
_
-
N11-
I I
N8
N12
-
I.
I
,
I I
_
I I
t
I
N14
I I
I '
MODIS
,I I
-0.2
0.0
0.2
0.4
0.6
0.8
N15
-
I
.-
_
-
i
N16
1
-
GLI/ADEOS
I
r
I
-0.2
0.0
0.2
0.4
0.6
0.8
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
11
Difference
in
NDVI
computed
at
the
TOA
level
relative
to
AVHRR/NOAA-9
0.04
0.02
0.00
-0.02
-0.04
0.04
a)
0.02
0
0.00
0
-0.02
CI
z
-0.04
ca
0
0.09
5
0
0.06
Z
0.03
0.00
-0.03
-0.06
-0.09
0.30
0.25
0.20
0.15
0.10
0.05
0.00
-0.05
-0.10
Fig.
9.
Similar
to
Figs.
7
and
8,
but
for
NDVI
at
the
TOA
level.
Quadratic
fit
is
plotted
for
absolute
difference
in
NDVI
due
to
SRF
effect.
Parameters
of
fitting
curves
are
given
in
Table
7.
the
data
points.
Tables
2-4
give
the
coefficients
of
the
quadratic
functions
that
best
fit
the
data,
correlation
coef-
ficient,
and
standard
deviation
of
the
fit
for
each
sensor
for
visible,
NIR,
and
NDVI,
respectively.
The
quality
of
the
fits
is
quite
good
for
AVHRRs
and
MODIS,
while
data
for
VGT
and
GLI
sensors
are
more
scattered.
12
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
Table
5
Parameters
of
polynomial
fit
to
the
relative
spectral
correction
Ap
=
(p
-
p
NoAA
_
9
)/(p
N0AA
-
9
)
(%)
for
visible
(red)
channel
Instrument
Relative
correction
(%)
r
2
v
(%)
AVHRR/N-6
-
0.01588
-
4.62556X-
7.96852X
2
.93
0.77
AVHRR/N-7
0.04201
-
4.12357X-
3.67224X
2
.88
0.68
AVHRR/N-8
-
0.41911
-
2.87354X+
8.89133X
2
.65
0.60
AVHRR/N-10
0.01277
-
4.84277X-
6.23082X
2
.90
0.84
AVHRR/N-11
-
0.01258
-
0.48667X-
1.27999X
2
.94
0.09
AVHRR/N-12
-
0.09189
-
0.2534X-
1.66716X
2
.53
0.35
AVHRR/N-14
-
0.15411
+
1.30802X+
7.27126X
2
.92
0.47
AVHRR/N-15
0.10932
-
6.31443X-
18.2578
1X
2
.97
0.98
AVHRR/N-16
0.04244
-
6.30091X-16.02976X
2
.96
0.96
MODIS
1.37765
+3.11606X-
40.78357X
2
.96
1.71
VGT/SPOT
1.01934
+
9.33224X-
28.91508X
2
.70
2.38
GLI/ADEOS
1.5794+6.874681-
49.41134X
2
.93
2.60
TOA
level.
X
denotes
NDVI
for
particular
sensor
computed
at
the
TOA
level.
4.3.
TOA
level
The
general
trends
of
the
effect
of
SRF
at
the
TOA
are
similar
to
those
at
the
surface
but
differ
in
detailed
features
as
a
result
of
the
distortion
by
the
atmosphere.
Discrepancies
in
TOA
visible,
NIR
reflectance,
and
NDVI
with
respect
to
AVHRR/NOAA-9
are
shown
in
Figs.
7-9.
Quadratic
fits
to
the
relative
differences
in
reflectances
and
absolute
differ-
ence
in
NDVI
are
also
plotted.
Other
data
points
are
more
scattered
and
no
fits
are
presented.
For
example,
the
relative
difference
for
NDVI
may
be
extremely
large
for
some
combinations,
simply
because
NDVI
values
computed
at
the
TOA
level
are
very
close
to
zero.
Figs.
7-9
contain
more
points
than
figures
for
the
surface
level
because
of
the
variable
effects
of
the
atmosphere
and
observation
geo-
metry.
Therefore,
fitting
all
data
points
with
one
curve
is
just
a
bulk
approximation
of
the
SRF
effect
to
account
for
the
large
SRF
effect
(e.g.,
MODIS,
VGT,
and
GLI).
In
the
case
of
AVHRR
sensors,
the
approach
still
provides
a
good
approximation
of
the
effect
of
SRF.
The
parameters
of
fitting
curves
are
given
in
Tables
5-7.
Table
6
Similar
to
Table
4,
but
for
NIR
channel
reflectance
at
the
TOA
level
Instrument
Relative
correction
(%)
r
2
Q
(%)
AVHRR/N-6
1.13467+1.64781X-1.26708X
2
.72
0.17
AVHRR/N-7
-
0.05851
-
0.53685X-
0.44445X
2
.89
0.08
AVHRR/N-8
1.00893
-
1.28713X-
0.49973X
2
.90
0.14
AVHRR/N-10
0.67763
+3.41197X-
0.56705X
2
.95
0.21
AVHRR/N-11
0.41666+0.15057X-
0.16803X
2
.14
0.04
AVHRR/N-12
0.65109+1.4419X-
0.42437X
2
.92
0.10
AVHRR/N-14
-
0.06817+5.81785X-
3.93554X
2
.89
0.36
AVHRR/N-15
-
0.07092+4.90503X-
0.23422X
2
.96
0.30
AVHRR/N-16
-
0.62499+5.51994X-
0.84905X
2
.96
0.30
MODIS
16.69042+23.8168X-
9.82829X
2
.72
3.10
VGT/SPOT
10.65846
+12.30469X+
0.975721
2
.80
1.88
GLI/ADEOS
11.94219
+
9.91694X+
2.78224X
2
.75
2.14
X
denotes
NDVI
for
particular
sensor
computed
at
the
TOA
level.
Table
7
Parameters
of
polynomial
fit
to
the
absolute
spectral
correction
for
NDVI
Instrument
Absolute
correction
(%)
r
2
a
(%)
AVHRR/N-6
0.00659+
0.0435X-
0.02586X
2
.93
0.0023
AVHRR/N-7
-
0+0.02435X-
0.0125X
2
.82
0.0023
AVHRR/N-8
0.00668
-
0.00023X-
0.02523X
2
.84
0.0019
AVHRR/N-10
0.00431+
0.05377X-
0.03415X
2
.95
0.0022
AVHRR/N-11
0.00224+
0.00428X-
0.00276X
2
.77
0.0004
AVHRR/N-12
0.00383
+
0.00911X-
0.00633X
2
.46
0.0017
AVHRR/N-14
0.00003
+
0.01558X-
0.03521X
2
.66
0.0018
AVHRR/N-15
0.00112+0.08104X-
0.02105X
2
.98
0.0032
AVHRR/N-16
-
0.00138+0.08156X-
0.02569X
2
.98
0.0028
MODIS
0.06948+
0.16993X-
0.1358
1X
2
.82
0.0105
VGT/SPOT
0.04608+
0.04565X-
0.01774X
2
.35
0.0134
GLI/ADEOS
0.04879+
0.08439X-
0.0035X
2
.71
0.0160
TOA
level.
X
denotes
NDVI
for
particular
sensor
computed
at
the
TOA
level.
Similar
to
the
surface
case,
the
best
agreement
with
AVHRR/NOAA-9
was
found
for
AVHRR/NOAA-11.
For
all
remaining
AVHRRs,
the
atmospheric
effect
generally
diminishes
the
spectral
difference
for
the
visible
(red)
chan-
nel
and
slightly
increases
it
in
the
NIR.
The
absolute
discrepancies
in
NDVI
remain
essentially
the
same
as
at
the
surface,
0.03-0.06
(Fig.
9).
The
effect
on
NDVI
for
NOAA-7,
-8,
-11,
-12,
and
-14
is
typically
within
±0.01.
For
NOAA-6
and
-10,
the
differences
in
NDVI
relative
to
NOAA-9
were
as
much
as
0.02-0.03
(3-5%
for
vege-
tated
surfaces).
The
largest
discrepancy
was
observed
for
NOAA-15
and
-16.
The
absolute
difference
in
NDVI
could
be
as
high
as
0.03-0.06,
which
is
larger
than
10%
for
vegetated
targets.
The
corrections
for
visible
channels
of
MODIS,
VGT,
and
GLI
sensors
at
the
TOA
level
shown
in
Fig.
7
are
similar
in
magnitude
to
those
at
the
surface,
but
the
magnitude
of
spectral
correction
for
the
NIR
channels
is
much
higher
(Fig.
8).
Apart
from
the
same
reasons
as
for
surface,
the
narrow
NIR
spectral
channels
are
not
affected
by
atmospheric
absorption
in
the
0.94-µm
water
vapor
absorption
band.
Consequently,
the
relative
differences
between
these
sensors
and
AVHRR/NOAA-9
were
as
much
as
20-40%,
with
the
largest
differences
occurred
between
AVHRR
and
MODIS
(up
to
40%).
The
results
for
VGT
and
GLI
are
similar
but
smaller
than
for
MODIS.
The
difference
(p
NIR
-
pNIR
,
N0AA_9)
is
positive
in
all
cases
and
increases
with
NDVI.
4.4.
Sensitivity
of
spectral
correction
to
atmospheric
state
Although
the
spectral
effects
discussed
above
are
rep-
resentative
for
"nominal"
conditions
described
above,
it
is
also
likely
that
in
any
particular
comparison
the
magnitude
of
the
differences
is
affected
by
the
variable
atmospheric
state.
Major
variables
influencing
satellite
measurements
are
AOD,
TWV,
and
ozone
columnar
amounts.
To
invest-
igate
how
variations
in
the
atmospheric
state
affect
the
spectral
correction
factors,
a
sensitivity
factor
of
spectral
I
T
=0.6
A
CO
A
s 0
0
A
®
;
-
18
I.
'
`•
a
e
at
6,
r=0.0
=
N6
0
relative
difference
A
absolute
difference
-
T=0.6
as
A
Q
A
.
••
A
0
A
-
84k
AG
-
A
AI
et,
T=0.0
N7
0.003
0.002
0.001
0.000
-0.001
-0.002
0.003
0.003
0.002
0.001
0.000
-0.001
-0.002
0.003
0.003
0.002
0.001
0.000
-0.001
-0.002
-0.003
0.006
0.004
0.002
0.000
-0.002
-0.004
-0.006
co
O
O
ci
0
0
O
0_
0
00
t
0
0
O
0
0
GLI/ADEOS
N14
_
••
r=0.0
AA
9-6_
A
A
°
4
22
®
as
o
ti
0
••
r=0.6
0
-0.2
0.0
0.2
0.4
0.6 0.8
-0.2
0.0
0.2
0.4
0.6
0.8
NDVI
toa
T=0.6
Ia
-401-a
r=0.0
N11
0
T=0.6
0
5
0
2:
41
16
r=0.0
N15
r=0.0-
1
00
F4,ci)
ask
sea
0
0 0
T=0.6
-
0
0
O
0
'3
0
VGT/SPOT
N12
T=0.6
.
e
0
A
A
O
5
T=0.0
N16
_
3
2
1
0
0
T=0.6
°
A
A
00
.222
o
0
6:2,
r=0.0
0
-1
MODIS
-2
3
-0.2
0.0
0.2
0.4
0.6
0.8
3
2
O
O
0
Cu
0
Q.
"
3
3
CD
O
O
2
-1
o
rs
0_
-2
-a
-
.•
T
=0.0
N10
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
13
Sensitivity
of
spectral
correction
for
visible
(red)
channel
TOA
reflectance
to
aerosol
optical
depth
T.
Basic
state
T=0.06
_
A
=-
44
)
A
44-1
104
-
414
-
040:
N8
3
2
1
0
-1
-2
-3
Fig.
10.
Sensitivity
of
SRF
effect
to
AOD
T.
Visible
(red)
channel
reflectance
at
the
TOA
level.
AOD
for
the
basic
atmospheric
state
is
0.06,
TWV
is
1.5
cm,
and
ozone
content
is
350
DU.
Computations
were
done
for
no
aerosol
(T=0)
and
hazy
(T=0.6)
cases.
correction
(Op)'
to
the
atmospheric
constituent
A
is
defined
as
(AP);
=
APi(A)
APi(Ao),
(2)
where
APi
(A)
=
N
(A)
-
PAVHRIVNOAA-9
(A)
,
(
3
)
A
is
the
amount
of
TWV,
ozone,
or
AOD;
A
0
is
the
corresponding
amount
for
the
basic
(or
reference)
atmospheric
state;
index
i
refers
to
a
specific
satellite
sensor;
and
p
denotes
the
reflectance
or
NDVI.
Normalized
(relative)
sensitivity
is
computed
as
the
ratio
[(Aa/p
AvHRR/
NOAA-91
x
100%
(Eqs.
(2)
and
(3)).
0.15
0.10
-
0.05
z
<
0.00
-0.05
-0.10
-0.15
02
0.3
0.4
0.5
N
DV
I
AVHRR/N-14
0.6
I
I
I
O
OZ
,
arnItal
7,
D
°
I I
Nu
m
ber
o
f
cases
2000
-
1500
-
1000
-
500
-
14
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
4.4.1.
Water
vapor
amount
The
sensitivity
tests
of
spectral
effect
to
the
amount
of
TWV in
the
atmospheric
column
are
computed
for
the
NIR
channels
and
NDVI
(figures
not
included),
since
water
vapor
effect
in
the
visible
band
is
not
significant.
The
tests
were
conducted
for
the
basic
atmospheric
state
(1.5
cm)
and
two
extreme
values,
0.5
cm
(dry
case)
and
5
cm
(humid
case).
The
ozone
amount
was
fixed
at
350
DU
and
the
AOD
at
0.06.
The
sensitivity
of
spectral
correction
to
precipitable
water
vapor
is
rather
small
for
the
AVHRR
NIR
channels
relative
to
the
magnitude
of
the
spectral
correction
itself.
It
is
well
within
1%
of
AVHRR/NOAA-9
reflectance
and
typically
three
to
five
times
smaller
than
spectral
correction
computed
for
the
basic
atmospheric
state.
Since
water
vapor
has
essentially
no
effect
on
visible
reflectance,
corrections
to
NDVI
are
also
small.
For
other
sensors
(MODIS,
VGT,
and
GLI),
the
sensitivity
to
water
vapor
is
somewhat
larger.
The
spectral
adjustment
for
MODIS
NIR
channel
is
about
-
6%
and
7-8%
for
TWV=
0.5
and
5
cm,
respectively,
relative
to
the
standard
case
of
TWV
=1.5
cm.
Corrections
for
NIR
channel
reflectance
of
VGT
and
GLI
sensors
are
within
±
3%.
For
NDVI,
absolute
corrections
range
from
-
0.03
to
0.04
for
MODIS
and
from
-
0.015
to
0.02
for
NDVI
of
VGT
and
GLI.
Based
on
these
results,
we
may
conclude
that
the
effect
of
atmospheric
water
vapor
is
negligible
when
making
spectral
adjustments
among
various
AVHRRs.
This
is
because
the
various
sensors
have
similar
spectral
cov-
erage,
in
particular
with
respect
to
water
vapor
absorption
bands.
Note
that
the
spectral
correction
effect
under
study
should
not
be
confused
with
the
absolute
effect
of
TWV
on
NIR
reflectance
itself.
The
latter
is
a
lot more
significant
(Cihlar
et
al.,
2000).
The
situation
is
more
complicated
for
MODIS,
VGT,
and
GLI
sensors.
The
TOA
reflectances
measured
by
these
sensors
are
less
sensitive
to
water
vapor,
since
SRF
for
these
instruments
do
not
include
the
strong
water
absorption
band
around
0.94
pm.
Nevertheless,
since
we
estimate
the
effect
of
spectral
correction
relative
to
AVHRR,
which
is
quite
sensitive
to
water
vapor
amount
(Cihlar
et
al.,
2000),
the
sensitivity
of
spectral
correction
to
water
vapor
emerges
for
these
sensors.
Comparison
NDVI
at
the
TOA
level
from
AVHRR/NOAA-14
and
NOAA-15
NDVIAvHRR8,1-14,
as,er0.428
NDVI
AVHRR/N-
15
aver
=0.445
AN
DVI
aver
=0.
617
DVI
mod
=0.025
1
-0.10
-0.05
0.00
0.05
0.10
0.15
ANDVI=NDVI
AVHRR/N-15
-NDVI
AVHRR/N
-
14
Fig.
11.
Top
panel
shows
comparison
of
NDVI
computed
at
the
TOA
level
from
AVHRR/NOAA-14
and
-15.
Small
dots
denote
satellite
observations.
Open
circles
denote
model
simulations.
Bottom
panel
shows
distribution
of
difference
in
ANVI.
0
-0.15
7
I ' I
I
-
4:b;i:
%s
o
F.X.°
itty
sVccns?
'%07,
"
We,
*kr
°
N•Yrilit
***
0
48
$
,
igoe
Po
e
e~~e88°
a
n
d
°
9
I I
0.4
0.3
5
0.2
<I
0.1
0.0
-0.1
-0.2
0.4
0.3
0.2
0.1
0.0
-0.1
0.2
NDVI
AvHRRIN
_
14
=0.414
NDVI
moDis
=0.568
AN
DVI
aver
=0.154
AN
DVI
d
=0.140
I
-I-1
-
Fr
I I
I
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
ANDVI=NDVI
MODIS
-NDVIA
VHRRIN-14
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
15
4.4.2.
AOD
The
sensitivity
of
the
spectral
effects
to
AOD
was
examined,
assuming
a
continental
aerosol
with
AOD
=
0,
0.06,
and
0.6,
TWV=
1.5
cm,
and
ozone
amount
=
350
DU.
Since
the
effect
is
very
small
(<0.2%)
for
low
aerosol
loading,
we
focus
on
the
effect
caused
by
moderately
large
aerosol
amounts.
The
sensitivity
of
the
spectral
correction
to
aerosol
for
the
red
channel
of
AVHRR
(relative
to
AVHRR/
NOAA-9)
is
negligible
for
AVHRR/NOAA-8,
-11,
and
-12
(Fig.
10).
It
may
reach
2-3%
for
AVHRR/NOAA-6,
-7,
and
-10
and
exceed
3%
for
NOAA-15
and
-16.
The
correction
is
negative
for
AVHRR/NOAA-14
(from
-
1%
to
-
2%).
The
absolute
magnitude
of
the
correction
ranges
from
-
0.001
to
0.002.
The
sensitivity is
similar
for
other
sensors
(Fig.
10).
The
sensitivity
for
NIR
channels
of
AVHRR
is
even
smaller
in
general
(figures
omitted;
from
-
0.3%
to
-
1%)
and
somewhat
larger
for
MODIS,
VGT,
and
GLI
(from
-
2%
to
-
5%)
NIR
channels.
The
magnitude
of
sensitiv-
ity
of
the
NDVI
spectral
correction
to
AOD
falls
within
-
3%
and
3%.
While
such
an
effect
is
comparable
with
the
instrument
uncertainties,
the
effect
is
persistent
for
aerosol
laden
atmosphere
and
is
thus
recommended
to
be
taken
into
consideration
in
intercomparison
studies,
if
possible.
4.4.3.
Ozone
amount
The
sensitivity
of
the
spectral
correction
to
ozone
amount
is
studied
for
270,
350,
and
430
DU.
Since
ozone
has
a
weak
absorption
in
the
visible
region
centered
around
0.6
µm
(Chappius
band;
Liou,
1992),
the
sensiti-
Comparison
NDVI
at
the
TOA
level
from
MODIS
and
NOAA-14
0
1
0.2
0.3
0.4
0.5
0.6
0.7
08
NDVI
AVHRR/N-14
u)
1500
C.)
'
45
1000
E
Z
500
Fig.
12.
Similar
to
Fig.
11,
but
for
comparison
of
NDVI
at
the
TOA
level
derived
from
MODIS
and
AVHRR/NOAA-14.
Like
in
previous
comparison,
good
agreement
is
found
on
average
between
modeling
and
observations
both
in
the
sign
and
magnitude
of
SRF
effect.
The
scattering
of
observed
points
is
due
to
possible
residual
cloud
contamination
and
resampling
of
MODIS
image
from
Integerized
Sinusoidal
Projection
(ISP)
to
Lambert
conformal
conic
projection,
which
alters
the
image
resolution.
16
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
vity
is
overly
small.
For
all
AVHRRs
and
MODIS,
the
correction
is
typically
less
than
0.5%
or
0.001
in
absolute
units.
For
VGT
and
GLI,
the
effect
is
slightly
larger,
0.5-1%
relative
or
0.002
absolute.
The magnitude
of
the
ozone
effect
on
the
NDVI
spectral
correction
is
equally
small,
as
it
essentially
affected
by
changes
in
visible
reflectances
only.
5.
Application
to
real
satellite
data
To
test
the
modeling
results,
we
compared
two
pairs
of
images
over
identical
areas.
One
pair
is
for
AVHRR
images
acquired
by
NOAA-14
and
-15
and
another
pair
is
for
AVHRR/NOAA-14
and
MODIS.
The
images
in
each
pair
were
taken
very
close
in
time
so
that
temporal
changes
do
not
affect
the
comparisons.
Two
AVHRR
images
were
taken
over
an
area
of
Northern
Ontario
(Canada)
observed
on
July
15,
2000
in
the
morning
(NOAA-15)
and
afternoon
(NOAA-14).
The
area
is
approximately
250
x
250
km
centered
around
(53
°
N,
87.5
°
W).
The
second
comparison
is
between
clear-sky
compo-
sites
from
MODIS
and
AVHRR/NOAA-14
over
an
area
of
Southern
Great
Plains
(USA).
MODIS
covers
the
period
July
19-26,
2000
and
the
AVHRR
for
the
period
July
21-31,
2000.
All
images
were
resampled
to
Lambert
conformal
conic
projection
with
1-km
resolution.
The
area
of
comparison
is
10
x
8
°
centered
around
(36
°
N,
97
°
W).
Joint
statistical
distributions
of
reflectances
and
brightness
temper-
atures
were
analysed
and
additional
thresholds
were
applied
to
all
images
to
detect
and
to
remove
cloud-contaminated
pixels
in
addition
to
clear-sky
compositing
procedures.
Water
pixels
were
excluded
from
the
comparison
due
to
strong
directional
effects
that
would
complicate
the
compar-
ison
(Cihlar
et
al.,
in
press).
The
directional
effects
resulting
from
different
local
observation
times
and
geometry
do
exist
over
land
as
well
(Li
et
al.,
1996)
but
less
pronounced
than
water
bodied.
The
effect
is
further
reduced
for
NDVI
due
to
considerable
cancellation
of
the
effects
in
the
visible
and
MR
channels
(Gutman,
1999;
Li
et
al.,
1996).
The
comparisons
of
NDVI
between
AVHRR/NOAA-14
and
-15
and
between
AVHRR/NOAA-14
and
MODIS
are
presented
in
Figs.
11
and
12.
Statistical
analyses
using
the
t
test
showed
that
the
two
comparisons
have
statistically
significant
nonzero
mean
difference
at
a
significance
level
.01
or
lower.
The
modal
value
of
the
distribution
of
NDVI
differences
between
AVHRR/NOAA-15
and
AVHRR/
NOAA-14
shown
in
Fig.
11
is
0.025
or
5.7%.
The
difference
between
average
values
is
slightly
smaller
(0.017
or
3.9%)
because
of
skewness
of
the
distribution.
This
difference
is
in
conformity
with
modeling
results,
which
also
indicate
a
greater
magnitude
of
NDVI
for
AVHRR/NOAA-15
than
for
AVHRR/NOAA-14.
The
magnitude
of
differences
between
modeled
NDVIs
is
slightly
larger
(1.4%)
due
to
contributions
by
various
factors,
in
particular
due
to
aerosol
and
surface
directional
effects.
The
directional
effect
is
lessened
for
the
comparison
between
MODIS
and
AVHRR/NOAA-14
due
to
closer
observation
times
(Fig.
12).
The
observed
modal
value
of
the
NDVI
difference
is
0.14
(29%)
and
the
mean
difference
of
0.15
(31%),
in
comparison
with
the
modeled
NDVI
difference
of
about
0.125
(25%).
Slightly
smaller
values
of
NDVI
difference
for
the
modeling
case
may
reflect
the
contribution
of
water
vapor
effect,
as
discussed
in
Section
4.
Good
overall
agreement
between
modeling
esti-
mates
and
satellite
observations
bolsters
our
confidence
in
the
estimates
of
spectral
correction
effects
derived
in
this
paper.
6.
Conclusions
Long-term
monitoring
of
the
Earth's
environment
by
satellite
sensors
require
consistent
and
comparable
measure-
ments.
In
this
paper,
we
evaluated
the
effect
of
a
major
sensor
parameter,
namely,
the
SRF,
on
the
consistency
of
observations
made
by
moderate
resolution
sensors
com-
monly
used
for
surface
and
atmospheric
studies.
Starting
with
TIROS-N
in
1978,
these
sensors
have
provided
a
long
time
series
of
satellite
data,
which
contain
rich
information
pertaining
to
the
state
and
changes
of
many
important
environmental
and
meteorological
variables.
However,
use
of
such
diverse
data
sets
requires
a
careful
evaluation
of
their
compatibility
and
consistency
to
avoid
any
artifact.
This
study
elaborates
the
influence
of
different
SRF
on
reflectance
measurements
in
the
visible
and
MR
channels
and
on
their
combination
in
the
form
of
NDVI.
The
sensors
under
study
include
AVHRRs
from
NOAA-6
to
the
latest
NOAA-16
as
well
as
MODIS,
VGT,
and
GLI.
All
the
sensors
are
compared
to
the
AVHRR/NOAA-9,
which
was
chosen
as
a
reference.
The
study
illustrated
that
the
differences
in
SRF
are
significant
enough
to
be
taken
into
account,
in
particular
for
studies
concerning
interannual
variations.
It
is
comparable
in
magnitude
to
the
uncertainties
caused
by
sensor
calibration
and
the
angular
correction
procedure.
Even
among
"the
same
type"
instruments
such
as
AVHRR,
the
effect
of
the
varying
SRF
on
surface
and
TOA
spectral
reflectances
and
NDVI
vegetation
index
is
sufficiently
large
to
require
correction.
Relative
to
the
AVHRR/NOAA-9,
differences
range
from
25%
to
12%
for
visible
reflectance
(red)
and
from
2%
to
4%
for
MR
reflectance.
The
absolute
differences
in
NDVI
among
various
AVHRRs
range
from
0.02
to
0.06.
The
most
consistent
with
AVHRR/NOAA-9
results
were
obtained
for
AVHRR/NOAA-11
and
-12
where
the
corrections
are
small
and
optional.
The
corrections
must
be
implemented
for
other
AVHRRs
and
especially
for
the
AVHRR/3
on
NOAA-15
and
-16.
Reflectances
and
NDVI
from
MODIS
differ
from
AVHRR/NOAA-9
by
as
much
as
30-40%.
Likewise,
VGT
and
GLI
also
exhibit
considerable
differences
relative
to
AVHRR
observations.
Given
the
significant
effect
of
SRF,
simple
polynomial
approximations
were
derived
that
may
be
used
for
correc-
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
17
tion.
They
provide
a
good
accuracy
of
approximation
for
the
AVHRR
sensors.
Other
sensors
(MODIS,
VGT,
and
GLI)
require
more
significant
correction
to
adjust
for
spectral
differences
in
comparison
with
AVHRR.
Polynomial
approximations,
we
propose
for
these
sensors,
may
be
used
as
first-order
corrections.
Higher
accuracy
may
be
achieved
by
taking
into
consideration
the
atmospheric
variables,
observational
angles,
and
information
from
additional
spec-
tral
channels
available
from
these
instruments
(Gitelson
&
Kaufman,
1998).
Sensitivity
tests
of
the
SRF
effect
to
various
atmospheric
variables
(water
vapor,
aerosol,
and
ozone)
were
conducted.
In
general,
their
influences
are
rather
small.
The
largest
effect
is
caused
by
aerosol,
which
may
reach
a
few
percent.
Water
vapor
affects
the
spectral
correction
between
AVHRR
and
other
sensors
but not
within
AVHRR
modifications.
Ozone
variation
generally
exerts
a
small
effect
for
all
sensors
and
may
be
neglected.
The
effects
of
SRF
are
further
reinforced
by
analyses
of
two
pairs
of
real
satellite
imagery
data
for
AVHRR
from
NOAA-14
and
-15
and
from
MODIS.
The
observational
results
are
generally
in
good
agreement
with
model
simulations
both
in
sign
and
mag-
nitude
of
the
SRF
effect.
Acknowledgments
Authors
are
grateful
to
J.-C.
Deguise
and
R.
Hitch-
cock
of
CCRS
for
making
PROBE-1
data
available
for
this
study.
We
acknowledge
the
use
of
spectral
data
from
JPL
ASTER
spectral
library
(http://speclibbl.nasa.gov
).
The
authors
also
thank
Gunar
Fedosejevs
for
his
valuable
comments
and
discussion.
This
research
was
partially
supported
by
the
Biological
and
Environmental
Research
Program
(BER),
U.S.
Department
of
Energy,
Grant
No.
DE-FG02-02ER63351.
References
Bannari,
A.,
Morin,
D.,
&
Bonn,
F.
(1995).
A
review
of
vegetation
indices.
Remote
Sensing
Reviews,
13,
95
-120.
Brest,
C.
L.,
Rossow,
W.
B.,
&
Roiter,
M.
D.
(1997).
Update
of
radiance
calibrations
for
ISCCP.
Journal
of
Atmospheric
and
Oceanic
Technol-
ogy,
14,
1091-1109.
Chen,
J.
M.,
Rich,
P.
M.,
Gower,
T.
S.,
Norman,
J.
M.,
&
Plummer,
S.
(1997).
Leaf
area
index
of
boreal
forests:
theory,
techniques
and
meas-
urements.
Journal
of
Geophysical
Research,
102,
29429-29444.
Cihlar,
J.,
Chen,
J.
M.,
Li,
Z.,
Huang,
F.,
Latifovic,
R.,
&
Dixon,
R.
(1998).
Can
interannual
land
surface
signal
be
discerned
in
composite
AVHRR
data.
Journal
of
Geophysical
Research,
103,
23163
-
23172.
Cihlar,
J.,
Latifovic,
R.,
Chen,
J.,
Trishchenko,
A.,
Du,
Y.,
Fedosejevs,
G.,
&
Guindon,
B.
(2001).
Systematic
corrections
of
AVHRR
image
com-
posites
for
temporal
studies.
Remote
Sensing
of
Environment
(Accep-
ted,
August
2001).
Cihlar,
J.,
Tcherednichenko,
I.,
Latifovic,
R.,
Li,
Z.,
&
Chen,
J.
(2000).
Impact
of
variable
atmospheric
water
vapor
content
on
AVHRR
data
corrections
over
land.
IEEE
Transactions
on
Geoscience
and
Remote
Sensing,
39
(1),
173-180.
Cracknell,
A.
P.
(1997).
The
advanced
very
high
resolution
radiometer
(AVHRR)
(534
pp.).
London:
Taylor
&
Francis.
Csiszar,
I.,
Gutman,
G.,
Romanov,
P.,
Leroy,
M.,
&
Hautecoeur,
0.
(2001).
Using
ADEOS/POLDER
data
to
reduce
angular
variability
of
NOAA/AVHRR
reflectances.
Remote
Sensing
of
Environment,
76,
399-409.
Fedosejevs,
G.,
O'Neill,
N.,
Royer,
A.,
Teillet,
P.
M.,
Bokoye,
A.
I.,
&
McArthur,
B.
(2000).
Aerosol
optical
depth
for
atmospheric
correction
of
AVHRR
composite
data.
Canadian
Journal
of
Remote
Sensing,
26,
273
-
284.
Gitelson,
A.,
&
Kaufman,
Y.
J.
(1998).
MODIS
NDVI
optimization
to
fit
the
AVHRR
data
series-spectral
consideration.
Remote
Sensing
of
Environment,
66,
343
-350.
Gobron,
N.,
Pinty,
B.,
Verstraete,
M.
M.,
&
Widlowski,
J.-L.
(2000).
Advanced
vegetation
indices
optimized
for
up-coming
sensors:
design,
performance,
and
applications.
IEEE
Transactions
on
Geoscience
and
Remote
Sensing,
38
(6),
2489-2505.
Gutman,
G.
(1999).
On
the
use
of
long-term
global
data
of
land
reflectances
and
vegetation
indices
derived
from
the
advanced
very
high
resolution
radiometer.
Journal
of
Geophysical
Research,
104
(6),
6241-6255.
Gutman,
G.,
Csiszar,
I.,
&
Romanov,
P.
(2000).
Using
NOAA/AVHRR
products
to
monitor
El
Niflo
impacts:
focus
on
Indonesia
in
1997-98.
Bulletin
of
the
American
Meteorological
Society,
81,
1189-1205.
Gutman,
G.,
Gruber,
A.,
Tarpley,
D.,
&
Taylor,
R.
(1989).
Application
of
angular
models
to
AVHRR
data
for
determination
of
the
clear-sky
plan-
etary
albedo
over
land
surfaces.
Journal
of
Geophysical
Research,
94,
9959-9970.
Hucek,
R.,
&
Jacobowitz,
R.
(1995).
Impact
of
scene
dependence
on
AVHRR
albedo
models.
Journal
of
Atmospheric
and
Oceanic
Technology,
12
(4),
697-711.
Kaufman,
R.
K.,
Zhou,
L.,
Knyazikhin,
Y.,
Shabanov,
N.
V.,
Myneni,
R.
B.,
&
Tucker,
C.
J.
(2000).
Effect
of
orbital
drift
and
sensor
changes
on
the
time
series
of
AVHRR
vegetation
index
data.
IEEE
Transactions
on
Geoscience
and
Remote
Sensing,
38
(6),
2584-2597.
Kidwell,
K.
(1994).
Global
vegetation
index
user's
guide.
U.S.
Dept.
of
Commerce,
NOAA/National
Environmental
Satellite
Data
and
Information
Service,
National
Climatic
Data
Center,
Satellite
Data
Services
Division,
Suitland,
MD
20746.
Li,
Z.,
Cihlar,
J.,
Zheng,
X.,
Moreau,
L.,
&
Ly,
H.
(1996).
The
bidirectional
effects
of
AVHRR
measurements
over
boreal
regions.
IEEE
Transac-
tions
on
Geoscience
and
Remote
Sensing,
34,
1308-1322.
Li,
Z.,
Moreau,
L.,
&
Cihlar,
J.
(1997).
Estimation
of
photosynthetically
active
radiation
absorbed
at
the
surface.
Journal
of
Geophysical
Re-
search,
102,
29717-29727.
Li,
Z.,
Nadon,
S.,
&
Cihlar,
J.
(2000).
Satellite
detection
of
Canadian
boreal
forest
fires:
development
and
application
of
an
algorithm.
International
Journal
of
Remote
Sensing,
21,
3057-3069.
Liou,
K.
N.
(1992).
Radiation
and
cloud
processes
in
the
atmosphere:
theory,
observation,
and
modeling.
New
York:
Oxford
University
Press.
Masonis,
S.
J.,
&
Warren,
S.
G.
(2001).
Gain
of
the
AVHRR
visible
channel
as
tracked
using
bidirectional
reflectance
of
Antarctic
and
Greenland
snow.
International
Journal
of
Remote
Sensing,
22,
1495
-1520.
Nakajima,
T.,
Higurashi,
A.,
Kawamoto,
K.,
&
Penner,
J.
E.
(2001).
A
possible
correlation
between
satellite-derived
cloud
and
aerosol
micro-
physical
parameters.
Geophysical
Research
Letters,
28,
1171-1174.
Qin,
Z.,
&
Karnieli,
A.
(1999).
Progress
in
the
remote
sensing
of
land
surface
temperature
and
ground
emissivity
using
NOAA-AVHRR
data.
International
Journal
of
Remote
Sensing,
20,
2367-2393.
Rao,
C.
R.
N.,
&
Chen,
J.
(1999).
Revised
post-launch
calibration
of
the
visible
and
near-infrared
channels
of
the
Advanced
Very
High
Resolu-
tion
Radiometer
(AVHRR)
on
the
NOAA-14
spacecraft.
International
Journal
of
Remote
Sensing,
20,
3485-3491.
Reynolds,
R.
W.,
&
Smith,
T.
M.
(1993).
Improved
global
sea
surface
temperature
analysis
using
optimum
interpolation.
Journal
of
Climate,
7,
929-948.
Rossow,
W.
B.
(1989).
Measuring
cloud
properties
from
space:
a
review.
Journal
of
Climate,
2,
201-213.
18
A.P.
Trishchenko
et
al.
/
Remote
Sensing
of
Environment
81
(2002)
1-18
Rossow,
W.
B.,
&
Schiffer,
R.
A.
(1999).
Advances
in
understanding
clouds
from
ISCCP.
Bulletin
of
the
American
Meteorological
Society,
80,
2261
2287.
Rutan,
D.,
&
Charlock,
T.
P.
(1997).
Spectral
reflectance,
directional
reflectance,
and
broad-band
albedo
of
the
Earth's
surface.
In:
Proc.
9th
conference
on
atmospheric
radiation,
Long
Beach,
California,
(pp.
466-470).
Boston,
MA:
American
Meterological
Society.
Secker,
J.,
Staenz,
K.,
Budkewitsch,
P.,
&
Neville,
R.
A.
(1999).
A
vicarious
calibration
of
the
Probe-1
hyperspectral
sensor.
In:
4th
International
airborne
remote
sensing
conference
and
exhibition/
21st
Canadian
symposium
on
remote
sensing
(pp.
21-24,
75-82).
Ottawa,
Ontario,
Canada,
Ann
Arbor,
MI,
USA:
ERIM
International
Inc.
(June).
Stowe,
L.
L.,
Ignatov,
A.
M.,
&
Singh,
R. R.
(1997).
Development,
validation,
and
potential
enhancements
to
the
second-generation
op-
erational
aerosol
product
at
the
National
Environment
Satellite,
Da-
ta,
and
Information
Service
of
the
National
Oceanic
and
Atmospheric
Administration.
Journal
of
Geophysical
Research,
102,
16923-16934.
Tahnk,
W.
R.,
&
Coakley,
J.
A.
(2001).
Updated
calibration
coefficients
for
NOAA-14
AVHRR
channels
1
and
2.
International
Journal
of
Remote
Sensing,
22
(15),
3053
—3057.
Teillet,
P.
M.,
Barker,
J.,
Markham,
B.
L.,
Irish,
R.
R.,
Fedosejevs,
G.,
&
Storey,
J.
C.
(2001).
Radiometric
cross-calibration
of
the
Landsat-7
ETM+
and
Landsat-5
TM
sensors
based
on
tandem
data
sets.
Remote
Sensing
of
Environment,
78,
39-54.
Teillet,
P.
M.,
Staenz,
K.,
&
Williams,
D.
J.
(1997).
Effects
of
spectral,
spatial,
and
radiometric
characteristics
on
remote
sensing
vegetation
indices
of
forested
regions.
Remote
Sensing
of
Environment,
61,
139-149.
Trishchenko,
A.
P.,
Li,
Z.,
Park,
W.,
&
Cihlar,
J.
(2001).
Corrections
for
the
BRDF
and
topographic
effects
in
satellite
retrieval
of
surface
spectral
reflectance
in
solar
spectral
region.
In:
W.
L.
Smith,
&
Y.
Tymofeyev
(Eds.),
Proc.
of
the
Int.
Radiation
Symp.
IRS
2000.
St.
Petersburg,
Russia,
August
2000,
vol.
147
(pp.
44-47).
Hampton,
VA,
USA:
A.
Deepak
Publishing.
Vermote,
E.,
Tanre,
D.,
Deuze,
J.
L.,
Herman,
M.,
&
Morcette,
J. J.
(1997).
Second
simulation
of
the
satellite
signal
in
the
solar
spectrum:
an
over-
view.
IEEE
Transactions
on
Geoscience
and
Remote
Sensing,
35,
675-686.