GHRSST XVIII
Qingdao, June 2017
#32
Long
-
term Impact of Sampling Bias in NASA
MODIS and AVHRR
-
Pathfinder Level 3
SSTs
Yang Liu,
Kay
Kilpatrick,
Sue
Walsh,
and Peter J.
Minnett
Clouds
and
inter
-
swath
gaps
are
the
primary
reasons
for
incomplete
coverage
of
satellite
measurements
causing
sampling
errors
in
averaged
satellite
SST
fields
.
Previously
(Liu
&
Minnett,
2016
)
we
found
that
the
MODIS
monthly
sampling
error
referenced
to
MUR
SSTs
is
up
to
O
(
1
K),
which
far
exceeds
the
error
threshold
needed
for
climate
research
.
The
largest
sampling
error
(>
5
K)
is
found
in
the
Arctic
.
Later,
Liu
et
al
.
(
2017
)
concluded
that
the
sampling
errors
due
to
spatial
averaging
can
be
reduced
by
removing
the
seasonal
cycle,
while
those
due
to
temporal
averaging
still
exist
in
the
seasonal
anomaly
fields
(Figure
1
)
.
The
next
question
we
are
going
to
seek
answers
to
is,
whether
the
temporal
sampling
error
can
be
negligible
in
a
climate
time
-
scale
mean
or
its
impact
can
decrease
the
quality
of
long
-
term
applications
of
the
satellite
IR
SSTs
.
We sample 9km gap free Level 4
SST fields (reference) using a daily
satellite infrared
Level
3
SST cloud
mask. The sampling error
generated in averaging will be the
difference between the averaged
sampled
(pseudo
L
3)
and
reference fields. The global error
characteristics depend on the
averaging size and dimension
(spatial or temporal).
Figure
1
.
Global
seasonal
sampling
errors
quantified
using
MUR
SSTs
(red)
and
MUR
SST
anomalies
(blue)
.
1. Sampling
Errors & Background
2. SST Climatology
•
High
latitude
warm
sampling
errors
still
exist
in
the
monthly
MODIS
SST
climatology,
derived
using
13
-
year
MODIS
SST
mask
and
MUR
fields
.
Errors
in
some
regions
of
the
Arctic
are
>
1
K
.
•
Sampling
errors
of
the
MODIS
night
-
time
SST
(NSST)
mask
shows
slightly
smaller
magnitudes
especially
around
Antarctica
.
The
larger
MODIS
SST
sampling
error
is
caused
by
the
daytime
cloud
screening
imperfection
in
the
current
version
.
Figure
3
.
Global
sampling
errors
(in
K)
in
the
AVHRR
pathfinder
5
.
2
SST
climatology,
quantified
as
the
difference
between
the
climatologies
generated
by
10
-
year
AVHRR
SST
cloud
mask
sampled
MUR
and
MUR
SSTs
.
•
The negative sampling
errors along the tropical
instability waves still
exist in the MODIS and
AVHRR
climatologies
.
•
Sampling errors in the
AVHRR climatology
shows larger magnitudes
in the low latitudes.
Daytime (SST) and
nighttime (NSST)
patterns are similar.
3
. SST trends
4. Gap fraction and cloud persistence trends
5. Discussion
Figure
4
.
200207
-
201506
SST
trend
derived
from
13
-
year
daily
Aqua
MODIS
SST
(upper
left)
and
NSST
(upper
right)
masks
sampled
MUR
(pseudo
level
3
)
and
MUR
(center)
;
The
difference
between
the
pseudo
level
3
trends
and
the
reference
trend
are
shown
in
the
third
row
.
Figure
5
.
As
Figure
4
third
row
.
The
SST
(top)
and
NSST
(bottom)
masks
are
from
10
-
year
daily
AVHRR
pathfinder
5
.
2
SST
and
NSST
fields
.
•
The
MODIS
monthly temporal sampling
errors also lead to warmer trends
sampled in the high latitudes, except
north of the Bering Strait.
Both MODIS
and AVHRR sampled trends show similar
results in the high latitudes.
•
For
MODIS,
t
he center of the basins and
low latitudes show negligible difference
in the trends.
However,
AVHRR sampled
trends (Figure 5) are slightly warmer in
the low latitudes, with some cooler
exceptions in the nighttime sampled
Indonesian islands, ITCZ and west Africa
equatorial upwelling regions.
•
Trends
in
the
gap
fraction
and
cloud
persistence
can
be
attributed
for
the
sampling
errors
in
trends,
as
increasing
trends
are
also
found
in
the
high
latitudes
.
Monthly
climatology
of
the
MODIS
and
AVHRR
sampled
MUR
fields
shows
warm
sampling
biases
at
the
high
latitudes
are
intrinsic
and
are
not
reduced
at
climate
change
scales
of
at
least
a
decade
.
The
SST,
gap
fraction,
and
cloud
persistence
trends
of
the
13
years
indicate
that
a
biased
trend
signal
for
the
last
decade
is
likely
to
be
found
especially
in
the
high
latitudes,
and
this
is
very
likely
because
of
the
increasing
gap
fraction
and
cloud
persistence
trends
found
in
the
similar
locations
.
References
Liu
, Y.
, Chin, T. M. and Minnett, P.J
. (2017).
Sampling Errors in Satellite
-
Derived Infrared Sea
-
Surface Temperatures. Part II: Sensitivity and Parameterization.
Remote Sensing of
Environment
(in revision).
Liu, Y.
, & Minnett, P.J. (2016).
Sampling Errors in Satellite
-
Derived Infrared Sea
-
Surface Temperatures. Part I: Global and Regional MODIS fields.
Remote Sensing of Environment, 177,
48
-
64
Figure
2
.
Global
sampling
errors
(in
K)
in
the
Aqua
MODIS
SST
climatology,
quantified
as
the
difference
between
the
climatologies
generated
using
13
-
year
MODIS
SST
cloud
mask
sampled
MUR
and
MUR
SSTs
.
Figure
6
.
Trends
in
the
monthly
Aqua
MODIS
SST
(upper
row)
and
NSST
(lower
row)
gap
fraction
(left)
and
cloud
persistence
(right)
.
Acknowledgements
NASA Graduate fellowship to Yang Liu. Access to super
-
computers at the
Cooperative Institute for
Climate and Satellites
-
North
Carolina; a
NOAA and North Carolina State University cooperative
institute
.