Transcript Daily OI Analysis for Sea Surface Temperature

Daily OI Analysis for Sea
Surface Temperature
Richard W. Reynolds (NOAA, NCDC)
Thomas M. Smith (NOAA, STAR)
Chunying Liu (NOAA, NCDC)
Dudley B. Chelton (Oregon State University)
Kenneth S. Casey (NOAA, NODC)
Michael G. Schlax (Oregon State University)
NOAA’s National Climatic Data
Center
Asheville, NC
1
The Good, the Bad
and the Ugly
Richard W. Reynolds (NOAA, NCDC)
Thomas M. Smith (NOAA, NCDC)
Chunying Liu (NOAA, NCDC)
Dudley B. Chelton (Oregon State University)
Kenneth S. Casey (NOAA, NODC)
Michael G. Schlax (Oregon State University)
NOAA’s National Climatic Data
Center
Asheville, NC
2
Outline
1.
2.
3.
4.
Introduction
Justification for a daily analysis
Version 1
Problems with Version 1 &
improvements to Version 2
5. Intercomparisons
6. Final comments
3
Jan '03: Number of Days
with Nighttime Obs
Top: AVHRR Pathfinder
Bottom: AMSR-E
For AVHRR:
• Absolute latitudes > 40° have
roughly only 5 days of data
• Number of days increases
toward the tropics
• Drop offs due to cloud cover
For AMSR:
• Absolute latitudes > 40° have
more than 20 days of data
• Drop offs due to precipitation
in ITCZ and SPCZ
4
Magnitude of Gradient:
Gulf Stream
• Daily OI using
AMSR-E for
January - March
2003
• Gradients have a
stationary part
due to topography
– Thus, limited
AVHRR data are
useful
5
Magnitude of Gradient:
Tropical Eastern Pacific
• Daily OI using
AMSR-E for August October 2003
• Gradients propagate
westward
– Limited coverage
not as useful here
– Monthly averaging
smoothes out most
of gradient signal
6
Outline
1.
2.
3.
4.
Introduction
Justification for a daily analysis
Version 1
Problems with Version 1 &
improvements to Version 2
5. Intercomparisons
6. Final comments
7
OI SST Analysis
Weekly (OI.v2)
• Grid: 1o
• Data: Satellite plus in
situ (ship and buoy)
data
– Satellite data
• Infrared AVHRR
– 7-Day large-scale satellite
bias correction for each
satellite
Daily
• Grid: 0.25o
• Data: Satellite plus in
situ (ship and buoy)
data
– Satellite data to include
• Infrared AVHRR
• Microwave AMSR-E
– 7-Day large-scale satellite
bias correction for each
satellite
• Spatial error correlation: • Spatial error correlation:
~ 700 km
~ 100 km
8
Nighttime Zonal Bias: AVHRR - In Situ
90N
AVHRR
Pathfinder
- In Situ
90S
AVHRR
Operational
- In Situ
90N
'82
'05
90S
Pathfinder Advantages: Lower bias variability; data delayed
Operational Advantages: Real-time data; 1981-84 data
9
Ocean Coverage (%)
• Day and Night Coverage for
– AVHRR Pathfinder ~12%
– AVHRR Operations ~ 8%
– AMSR
~40%
• Note Pathfinder coverage
better then Operation
– Pathfinder cloud
contamination?
• AMSR coverage
considerably better than
AVHRR
– Note data drop outs
10
OI Zonal Spatial
Error Correlation
Scales
From the figure note:
• Daily spatial scales
strongly reduced from
weekly scales
• Daily scales reduced in
high gradient areas
– Scales <100 km in Gulf
Stream
• Daily scales between 100
& 200 km in most regions
11
3-Day Gulf Stream
Average Gradients
• Data
– AVHRR
– AMSR
• Analyses
–
–
–
–
Weekly OI.v2: AVHRR
Daily RTG-SST: AVHRR
Daily OI: AVHRR only
Daily OI: AMSR+AVHRR
• Missing Data: AMSR &
AVHRR
• Daily OI resolution
improved over OI.v2
• AMSR+AVHRR best
analysis resolution
12
3-Day Eastern Tropical Pacific Gradients
• Daily OI
resolution
improved over
OI.v2
• AMSR+AVHRR
best analysis
resolution
• High Gradients
in AVHRR lost
in analyses
• Features are
progress waves
not partly fixed
by topography
13
Gulf Stream Gradient Index
• Index:
– 70°W-40°W
– 35°N-50°N
• Analyses
–
–
–
–
Weekly OI.v2
RTG-SST
AVHRR-only
AMSR+AVHRR
• Seasonal cycle
range
– Similar minimum
for daily analyses
– Maximum roughly
40% for
AVHRR+AMSR
14
Jan 2003:
Standard Deviation
Top: Pathfinder
Bottom: AMSR-E
• AMSR-E has stronger
standard deviations than
Pathfinder
– Especially in mid-latitude
winter
– Clouds reduce Pathfinder
sampling
• These differences plus
gradient differences
suggest that separate
Pathfinder OI and
Pathfinder + AMSR OI
needed
15
Empirical Orthogonal
Teleconnections (EOT)
• 130 modes from Extended
Reconstruction SST
• Fit modes to 7-day in situ and
satellite anomalies on 2o grid
• Only use modes if supported
by both types anomalies
– In figure mode 4 would probably
not be used due to lack of in situ
data
• Reconstruct anomalies from
modes
• Bias is difference between
the two reconstructions
16
Random +
Sampling Error
• OI Derived
• Sampling error
dominant term
• Sampling and
random error
reduced by
observations
– Lower for
AMSR+AVHRR
17
Bias Error
• Bias error derived
from variance of EOT
modes used in
correction
– From modes supported
by satellite data but not
by in situ data
– Bias error consider
independent if from
different satellites
– Thus error lower for
AMSR+AVHRR than
AVHRR-only
• Total error: Σ (Bias +
Random + Sampling)
Variances
18
Daily OI: Products
All products on 0.25° spatial grid
• AVHRR only
– AVHRR Pathfinder Daily OI: January 1985-end of
Pathfinder data (currently December 2006)
– AVHRR Operational Daily OI: January 2007-present
• AMSR + AVHRR
– AMSR + AVHRR Pathfinder Daily
OI: June 2002 to end of Pathfinder data
(currently December 2006)
– AMSR + AVHRR Operational Daily OI: January
2007-present
19
Daily OI
SST Anomalies
EOT satellite bias correction
Left: Pathfinder AVHRR OI
Right: Pathfinder AVHRR +
AMSR-E OI
• 27 August: Katrina
– Weak anomaly in Gulf in
Path + AMSR-E
• 28 August: Katrina
– Strong anomaly in Path +
AMSR-E
• 29 August: Katrina
– Strong anomaly in Path +
AMSR-E
– Modest anomaly in Path
20
Operational Daily OI SST Data Flow
21
Sample
Warning
& Error
Daily OI
Emails
22
Outline
1.
2.
3.
4.
Introduction
Justification for a daily analysis
Version 1
Problems with Version 1 &
improvements to Version 2
5. Intercomparisons
6. Final comments
23
Data Anomalies: 11 Jan 2003
Differences lead to day-to-day noise in OI
• AVHRR day &
night
– Note data
scarcity
– Pathfinder:
local time
• AMSR day &
night
– Note swath
width &
precipitation
– Day night
differences
not always
diurnal
warming
24
4 Days of Buoy SST Data
0Z 22 March 22 - 0Z 26 March 26, 2006
• Resolution
– 1 minute
• Averaged to
– 1 hour
– 1 day
• Consider
random
sampling
– AMSR could
approach 2
obs/day
– AVHRR would
be lower
especially in
winter
25
Buoys with Good Temporal Coverage
• Use moored
buoys to
determine if
temporal
smoothing is
needed for the
daily OI
• 43 buoys
selected with
at least 99%
daily coverage
from '03-'05
26
Average Daily Spectra for 3 Years
• Buoy spectra
closer to daily OI
spectra when 3days of data are
used
– Note: by day,
AVHRR-only
lower than
AMSR+AVHRR
spectra
• 3-days of data
used in Version
2
27
SSTs & Gradients:
3-Day Average
• Data
– AMSR
• Analyses
– Daily OI: AVHRR only
– Daily OI: AMSR+AVHRR
• Precipitation data void
near 35°N & 130°W
– Noise near boundary
leads to OI interpolation
errors
– Note gradients
28
AMSR Data
Anomalies
9 Feb 2003
• Data in color passed the
extra QC which includes
– SST changes
– Proximity to rain
• Data in black failed the
extra QC but was used in
the original OI
– Note the relatively small
number
• Typical AMSR obs which
fail the extra QC
– Daytime: 1.0-1.5%
– Nighttime: 0.5-1.0%
29
AMSR+AVHRR OI
Anomalies: 9 Feb 2003
Top: OI with
AMSR+AVHRR data
Bottom: OI with
AMSR+AVHRR data:
AMSR has extra QC
• SST anomalies reduced in
oval region when extra QC
obs not used
– This procedure reduces
some day-to-day noise
– Of course, the extra QC is
not be prefect
• AMSR with extra QC used
in Version 2
30
Analysis
Difference
July 2006
• NOAA 17 AVHRR OI:
Pathfinder - Navy
– Top Panel: No zonal
bias first guess
correction
– Bottom Panel: With
zonal bias first
guess correction
– Note: Differences
smaller between
40oS- 60oS with
zonal bias correction
• Bias correction with
zonal first guess
used in Version 2
31
Average Bias Correction Spectra
2000-2005
• Night AVHRR
bias
correction
– 1 Field: no
smoothing
– Binomial filter:
3, 5 or 7 days
• 5 day binomial
filter selected
for Version 2
32
Daily OI: Ship Biases
• Satellite biases are corrected with respect to
in situ data (ships & buoys)
• Ships have larger random and bias
errors than buoys
– Ship observations using buckets tend to
be biased cold due to evaporation
– Ship observations using engine intake tend
to be biased warm due to engine room heating
– Ship metadata needs improvement
• Try to correct ship bias using buoy data
– Would EOT procedures over a 30 day time period
work?
33
• EOT monthly
ship & buoy
analyses
• Buoy SST obs
increase with
time
– Significant buoy
obs 1994 to
present
• First correct ship
biases than
correct satellite
biases
34
'94
Time
Ship-Buoy SST
Zonal Biases
'05
Monthly Averaged Collocated
Buoy and Ship Anomaly SSTs
• Buoy vs. Ship:
data and fit
– Red: 19982006
– Black: 19891997
– 2 Periods
similar
• Clear need to
adjust ships:
0.14oC
subtracted
from all ship
obs in Version
2
35
Daily OI Revisions
Version 2: Next Daily OI Version (Oct 2007)
1.
2.
3.
4.
5.
Use 3-days of data in the Daily OI analysis
Smooth 7-day satellite bias corrections
Use the extra QC from AMSR data
Improve the land-sea table
Correct ship SSTs with respect to in buoy SSTs
Version 3: Following Daily OI Version 2
1.
2.
3.
4.
Use additional satellite data: TMI, ATSR, MODIS, etc.
Process NOAA-7: November 1981 - December 1984
Harden operational procedures
Improve bias correction
36
Outline
1.
2.
3.
4.
Introduction
Justification for a daily analysis
Version 1
Problems with Version 1 &
improvements to Version 2
5. Intercomparisons
6. Final comments
37
Difference:
V2 - V1
Upper Panel: AVHRR-only
• Ship bias adjustment Cool differences: Lat > 40oS
• Zonal bias first guess Warm differences: Lat < 40oS
Lower Panel: AMSR-only
• Additional changes due to
improved AMSR QC
– Removal of tropical rain
contamination reduces
positive bias
– Removal of mid-latitude
rain contamination reduces
negative bias
38
Difference:
V2 - OI.v2
Upper: (AVHRR-only) - OI.v2
• Tropical Negative Bias
• Negative Bias near 60oS
• Positive Bias above 70oS
Lower: (AMSR-only) - OI.v2
• Biases generally lower than
Pathfinder AVHRR
Residual bias error
• ~ 0.1oC except 0.3oC north
of 70oN and south of 60oS
39
Thanks to
Sudhir Nadiga
Difference:
V2 - OI.v2
Upper: (AVHRR-only) - OI.v2
• Tropical Negative Bias
• Negative Bias near 60oS
• Positive Bias > 70oS
Lower: (AVHRR-only
No bias correction) - OI.v2
• Biases generally much
larger without correction
• Note TAO moorings
How should this bias
error be corrected?
40
Convert Sea Ice
to SST
• Least squares fit
between SST and
sea ice, x, (0 ≤ x ≤ 1)
– SST2 = A x2 + B x + C
– SST1 = b x + c
– where SST = freezing
temp for x = 1
– For each month &
region find A, B, C, &
b, c
41
Sea Ice to SST
• Least squares fit: 19851994
• Verification: 1995-2004
– Bias & RMS
differences for N. Hem.
• RMS Difference
– RMS increases with
decreasing ice conc.
• Bias: Data - Fit SST
– Data > 0.3oC warmer
than fit when ice < 0.7
Linear Fit selected
Range ice conc. >0.5
42
Weekly & Daily OI
December 1991
• Both OIs have bias correction
for AVHRR data
• Daily OI uses Pathfinder
• Weekly OI uses Operational
From the figure note:
• Operational data problem along
35oS
– Delayed mode processing allows
data correction
• Daily scales reduced in high
gradient areas
43
OI SSTs for December 2006
• Gulf Stream flows
off Carolina coast
• Cold shelf water
should be present
between Gulf
Stream and coast
– Not present in
weekly OI.v2
– Weak in
Pathfinder and
AMSR daily OI
– Strongest in
Navy daily OI
• Both Pathfinder
and Navy use
NOAA-17 data!
44
SST Data for 28 December 2006
• Data for 28
December, 2006,
typical when
AVHRR
available
• Cold shelf water
off Carolinas is:
– Not present in
Pathfinder and
AMSR data
– Weak in buoy
data
– Stronger in
Navy data
45
SST Analyses for 28 December 2006
• UK OSTIA
– (1/20) oC
• NCEP RTG-HD
– (1/12) oC
• NCDC Daily OIs
– (1/4) oC
• Shelf water
weak in OSTIA
• RTG-HD
smoother than
daily OI
46
Suggestions for High Resolution SST
•
First Stage: low
resolution using
(1/4)o ship, buoy,
AVHRR, & AMSR
data
–
AMSR
dominates
47
Suggestions for High Resolution SST
•
First Stage: low
resolution using
(1/4)o ship, buoy,
AVHRR, & AMSR
data
–
•
AMSR
dominates
Second Stage:
high resolution
using 4-5 km
AVHRR
–
•
AVHRR
improves
resolution
where available
High Resolution
First Guess
–
Low resolution
+ damped high
resolution
48
Outline
1.
2.
3.
4.
Introduction
Justification for a daily analysis
Version 1
Problems with Version 1 &
improvements to Version 2
5. Intercomparisons
6. Final comments
49
Time
AMSR + AVHRR Daily OI SST Anomaly : 2oS - 2oN
50
(oC)
SST Daily OI Anomaly
(oC)
AMSR + AVHRR
51
To Maintain Credibility
• Stress both the good and the bad
–E.g., grid and spatial resolution differ
• Keep the descriptions short
–Users will not read documentation
• Keep the number of products small
–Users will not read documentation
• Keep data formats simple
–Users will not read documentation
52
The End
53
OI Anomalies with Bias Correction
January 1988: Along equator
From the figure note:
• Operational AVHRR did
not have strong biases
in this region
• Negative biases in daily
OI except at TAO
mooring sites near:
155oW, 140oW & 125oW
• OI bias correction fails
to correct AVHRR
Pathfinder satellite
biases
54
OI Anomalies with Bias Correction
January 1988: Along equator
•
•
•
•
From the figure note:
Operational AVHRR did
not have strong biases in
this region
Negative biases in daily
OI except at TAO mooring
sites near: 155oW, 140oW
& 125oW
OI bias correction fails to
correct AVHRR Pathfinder
satellite biases
Empirical Orthogonal
Teleconnections (EOT)
bias correction better at
correcting biases
55
Daily OI computed with EOT bias correction
SST Analyses for 28 December 2006
Frontal features
• Common to both
Navy and AMSR
– Probably
realistic as data
independent
– Weak in
Pathfinder
• Pathfinder +
AMSR keeps
features from
AMSR
– AMSR
overwhelms
AVHRR in open
ocean
56
SST Analyses for 28 December 2006
Frontal features
• Common to
both Navy and
AMSR
• Weak in OISTA
• Strong in RTGHD
57