BLUElink High Resolution Sea Surface Temperature Analysis System Helen Beggs, Anthony Rea and Nevill

1
BLUElink High Resolution Sea Surface Temperature
Analysis SystemHelen Beggs, Anthony Rea and
Neville SmithAustralian Bureau of
MeteorologyPresented at the GHRSST-PP Science
Team Workshop, Exeter, 16-20 May, 2005

2
Overview

• BLUElink High Resolution SST Analysis System at
  the Australian Bureau of Meteorology
• Regional AVHRR SSTs produced by Bureau
• Proposed method for blending SST data
• Conversion of AATSR skin temperatures to sub-skin
• Comparison between AVHRR and AATSR SSTs
• BLUElink contribution to GHRSST-PP
• Future developments for BLUElink (2007 - 2010)

3
BLUElink HR SST Analysis System at Australian
Bureau of Meteorology

• High resolution SST analysis combining infra-red
  and microwave satellite measurements with in situ
  data to produce improved estimates of SST
• Australian region 20?N to 65?S, 50?E to 160?W
• Blend all data sources into SST(1m) and, later,
  SST(skin) estimates
• Output 6-hourly real-time SST analyses at 5 km
  resolution regionally
• Analyses will be generated within 6 hours of data
  reception
• Aim to be operational end of 2006
• See web page at http//www.bom.gov.au/bmrc/ocean/B
  LUElink/SST_external.html

4
SST Data Streams to be Blended

• Regional real-time AVHRR SST (Bureau) 1 km,
  bulk, skin?

• Global real-time AVHRR SST (NESDIS) 9 km, bulk

• Geostationary MTSAT-1R Imager SST (JMA) 5 km,
  bulk, skin?

• Geostationary FY-2C Radiometer SST (CMA?) 5
  km, bulk?, skin?

• Envisat AATSR Meteo SST product (ESA) 17 km,
  skin

• Aqua AMSR-E SST (Remote Sensing Systems) 38 km,
  sub-skin

• In Situ SST from ARGO profiling floats, drifting
  buoys, ships and XBTs (GTS) 15 km, bulk

5
Regional AVHRR SSTs currently produced by Bureau

• Currently using NOAA-15 and NOAA-17
• Coverage 20?N to 70?S, 50?E to 160?W depending
  on receiving station
• Resolution 1.1 km at nadir
• Digitization 10-bit
• Updated in real time
• Data Product Single swath files of L2A SST(0.5m)
• Mean bias of bulk SST compared with in situ
  data 0.04 to 0.26 K
• Mean RMS error (bulk SST) 0.83 to 0.87 K

6
Current Bureau AVHRR Processing

• Data received at 4 reception stations and sent to
  head office (Melbourne) as raw data
• Converted to McIDAS format using in-house
  software
• Zenith and sun angle checks
• Land/sea checks (but no sea-ice checks)
• Cloud clearing (pixels with cloud currently not
  included in products)
• SSTs computed for 0.5 m depth using NESDIS NLSST
  algorithms and regression coefficients based on
  historical buoy data
• Navigation information updated using the CAPS
  package developed by CSIRO Marine Research
• approx. 1-pixel accuracy

7
Proposed improvements to Bureau AVHHR SST data

• Migration of system from FORTRAN/McIDAS to
  Tcl/Tk/CAPS allowing
• netCDF and HDF output formats in addition to
  McIDAS
• Addition of a quality/confidence flag along the
  lines of the GHRSST-PP L2P flags
• Improved land/sea mask
• Reduce the digitization error in the full
  resolution (1 km) AVHRR SST data by using a 3x3
  or 5x5 array about an individual pixel to
  calculate the difference in brightness
  temperatures between channels 4 and 5, then apply
  this value to 1 km resolution data (see Ian
  Barton, Remote Sens. Environ., 2987-89, 1989)
• Calibrate the 1 km AVHRR brightness temperatures
  to skin temperatures by regressing against
  co-located 17 km AATSR skin temperatures measured
  within /- 1 hour of AVHRR data
• Produce a cleanly stitched single orbit AVHRR
  SST product using time at start of each scan line
  and pixel number in scan to remove overlapping
  pixels.

8
Proposed Initial Blending Method for BLUElink
SST(1m) Analysis
9
Converting AATSR skin SST to sub-skin SST

• 17km Meteo AATSR skin temperatures (10?m)
  converted to sub-skin temperatures (1 mm) using
  Fairall et al. (1996) cool skin correction
  algorithms
• Input 3-hourly averaged surface wind stresses
  and net heat fluxes from 0.75 GASP
  analyses/forecasts
• All available AATSR Meteo product data used for 8
  days over the period 1 Oct 2004 1 Mar 2005 over
  50E-180 E, 65S- 20N
• Compared with Donlon et al. (2002) equations
  using only surface wind speeds from GASP.
• Mean(Fairall sub-skin SST skin SST) 0.19
  0.02C (Day), 0.19 0.02C (Night)
• Mean(Donlon sub-skin SST skin SST) 0.17
  0.0C (Day), 0.20 0.01C (Night)

C
C
C
10
Converting AATSR skin SST to sub-skin SST-
Future Work

• Use Ian Bartons DDS shipboard skin and sub-skin
  temperatures to determine proportionality
  constant (?0) and systematic offset in Fairall
  algorithms
• Use 1 hourly averaged LAPS wind stresses and net
  heat fluxes

11
Comparing 17 km AATSR SST with 1 km AVHRR bulk
SST

• 10 AATSR Meteo product SST 1 km AVHRR bulk
  SST
• Period 1 31 Oct 2004
• Area 50E-180 E, 65S- 20N
• Match-up Colocated data in 10 pixel, 1 hour

C
C
C
C
12
Comparing 17 km AATSR SST with 1 km AVHRR bulk
SST

• ?T 10 AATSR Meteo product SST 1 km AVHRR
  bulk SST
• N Number of AATSR/AVHRR matchups over a single
  day or night
• Standard atmospheric conditions Dual Nadir
  SST
• Non-Standard atmospheric conditions Dual
  Nadir SST ? 0.0 C (Day), 0.1 C (Night)

13
Comparing 17 km AATSR SST with 1 km AVHRR bulk
SST

• Compare with UK Met Office results (OCarroll et
  al. (2004))
• ?T 10 AATSR Meteo product SST buoy SST

14
Future HR SST developments for BLUElink Phase II
(2007 2010)

• Geostationary SSTs from MTSAT-1R and possibly
  Fengyun-2C will provide hourly data, increased
  area of coverage and the ability to resolve the
  diurnal cycle
• 6-hourly skin SST analyses (representing the
  diurnal cycle) at 5 km resolution
• Expand the Bureaus X-Band network, allowing more
  accurate 1 km resolution SSTs to be generated
  from MODIS and VIIRS 12-bit brightness
  temperatures
• Develop local SST coefficients using AATSR for
  other satellite radiometers (MTSAT-1R, FY-2C,
  MODIS?, VIIRS?) in addition to AVHRR
• Attempt a daily, skin and bulk SST product at 1
  km resolution
• combine AVHRR, AATSR, MODIS?, VIIRS?
• Incorporate GHRSST-PP L2P SST data sets into the
  Bureaus blended SST analysis system
• Will AATSR and MODIS L2P be available?

15
Proposed Eventual Blending Method for SST(1m) and
SST(skin) Analyses
16
BLUElink contribution to GHRSST-PP

• Regional Real Time AVHRR SST L2P files
• Bureau can provide stitched, single swath L2
  SST(skin) and/or SST(bulk) data at 1 km
  resolution
• SSES Bias and Standard Deviation error based on
  confidence flags
• Rejection Flags SST out of range, IR_cloudy,
  land, sun glint, high zenith angle
• Confidence Flags Would like low wind speed flag
  as well as high wind speed retrieval flag
• Possibly add cloud coverage and cloud type to SST
  data if wanted by GHRSST-PP
• No source of regional insolation, wind speed or
  sea ice data at 1 km resolution, only real-time
  GASP Model 10m wind speeds at 0.75 resolution
• Hope to have operational early 2006
• Diagnostic Data Set (Ian Barton CSIRO)
• Regional Real Time 5 km resolution L4 SST(1m) and
  SST(skin) data every 6 hours
• Operational end of 2006

17
(No Transcript)
18
Variation over 10 pixels of 1 km AVHRR bulk SST

• Mean(standard deviation 1km AVHRR 'bulk' SST over
  10 pixel)
• 0.28 0.04C (Day), 0.31 0.04C
  (Night)
• Probably mainly due to 10-bit digitization error

19
Converting AATSR skin SST to sub-skin SST

• 17km Meteo AATSR skin temperatures (10?m)
  converted to sub-skin temperatures (1 mm) using
  Fairall et al. (1996) cool skin correction
  algorithms and 3-hourly averaged surface wind
  stresses and net heat fluxes from 0.75 GASP
  analyses/forecasts.
• Net Heat Flux Sensible Heat Flux Latent Heat
  Flux - Net Long-wave Radiation
• All available AATSR Meteo product data used for 8
  days over the period 1 Oct 2004 1 Mar 2005 over
  50E-180 E, 65S- 20N
• Compared with Donlon et al. (2002) equations
  using only surface wind speeds from GASP.
• Mean(Fairall sub-skin SST skin SST) 0.19
  0.02C (Day), 0.19 0.02C (Night)
• Mean(Donlon sub-skin SST skin SST) 0.17
  0.0C (Day), 0.20 0.01C (Night)

20
Comparing 17 km AATSR skin SST with 1 km AVHRR
bulk SST

• All available AATSR Meteo product data used for 1
  31 Oct 2004 over 50E-180 E, 65S- 20N
• Match-up Colocated data in 10 pixel, 1 hour
• Mean(AATSR skin SST - AVHRR 'bulk' SST) -0.31
  0.19C (Day), 0.02 0.18C (Night)
• Mean number of match-ups 2004 ? 1281 (Day),
  2859 ? 1840 (Night)
• Mean ? in ?T over all 10 pixels 0.21 0.04C
  (Day), 0.24 0.08C (Night)
• Average of the 24-hourly standard deviations of
  all temperature differences over the study region
  0.87 ? 0.28?C

C
C
C
21
Comparing 17 km AATSR skin SST with 1 km AVHRR
bulk SST

• All available AATSR Meteo product data used for 1
  31 Oct 2004 over 50E-180 E, 65S- 20N
• Match-up Colocated data in 10 pixel, 1 hour
• Mean(AATSR skin SST - AVHRR 'bulk' SST) -0.31
  0.19C (Day), 0.02 0.18C (Night)
• UK Met Office Mean(AATSR skin SST buoy 'bulk'
  SST) 0.03 0.46C (Day), 0.07 0.35C
  (Night) (OCarroll et al., 2004)
• Mean number of match-ups per day 2004 ? 1281
  (Day), 2859 ? 1840 (Night)
• Mean ? in ?T over all 10 pixels 0.21 0.04C
  (Day), 0.24 0.08C (Night)
• Average of the 24-hourly standard deviations of
  all temperature differences over the study region
  0.87 ? 0.28?C

22
Comparing 17 km AATSR sub-skin SST with 1 km
AVHRR bulk SST

• All available AATSR Meteo product data used for 1
  31 Oct 2004 over 50E-180 E, 65S- 20N
• Match-up Colocated data in 10 pixel, 1 hour
• Mean(AATSR subskin SST - AVHRR 'bulk' SST)
  -0.10 0.24C (Day), 0.18 0.18C (Night)
  0.16C c.f. skin
• UK Met Office Mean(AATSR subskin SST buoy
  'bulk' SST) 0.19 0.35C (Day), 0.23 0.34C
  (Night) (OCarroll et al., 2004)
• Mean number of match-ups per day 1100 ? 1048
  (Day), 2859 ? 1840 (Night)
• Average of the 24-hourly standard deviations of
  all temperature differences over the study region
  0.91 ? 0.33?C 0.04C c.f. skin

23
Comparing AATSR sub-skin SST with AVHRR bulk SST
for Standard Atmospheric Conditions

• AATSR Meteo product data for Standard
  atmospheric conditions used for 1 31 Oct 2004
  over 50E-180 E, 65S- 20N
• Only used AATSR data where Dual Nadir SST C (Day) 0.1 C (Night)
• Match-up Colocated data in 10 pixel, 1 hour
• Mean(AATSR subskin SST - AVHRR 'bulk' SST)
  -0.13 0.22C (Day), 0.10 0.17C (Night)
  Decrease of 0.08C
• Mean number of match-ups per day 1100 ? 1069
  (Day), 2157 ? 1368 (Night)
• Average of the 24-hourly standard deviations of
  all temperature differences over the study region
  0.86 ? 0.30?C Decrease of 0.05C

All Atmospheric Conditions
Standard
24
Comparing AATSR sub-skin SST with AVHRR bulk SST
for Non-Standard Atmospheric Conditions

• AATSR Meteo product data for Non-Standard
  atmospheric conditions used for 1 31 Oct 2004
  over 50E-180 E, 65S- 20N
• Only used AATSR data where Dual Nadir SST ? 0.0
  C (Day) 0.1 C (Night)
• Match-up Colocated data in 10 pixel, 1 hour
• Mean(AATSR subskin SST - AVHRR 'bulk' SST) 0.89
  0.44C (Day), 0.45 0.29C (Night)
• Mean number of match-ups per day 43 ? 70 (Day),
  704 ? 516 (Night)
• Average of the 24-hourly standard deviations of
  all temperature differences over the study region
  1.03 ? 0.51?C

25
SST Definitions
26
Limitations to current Bureau AVHHR data

• No SST value under cloudy conditions
• Coverage limited to Bureau of Meteorology LAC
  coverage area
• Infrequent coverage towards extremes of coverage
  area
• Degraded performance in areas of extreme humidity
• SSTs are produced at a nominal depth of 0.5 m
  using historical buoy data but infra-red
  radiometers actually measure skin temperature
• NOAA AVHRR instruments transmit 10-bit data
  resulting in a digitizing accuracy in brightness
  temperatures of 0.1C near 7 C. The NLSST
  algorithms use difference in BTs between
  channels 4 and 5 resulting in an error in SST for
  a single 1 km pixel of as high as 0.5C (Ian
  Barton, Remote Sens. Environ., 2987-89, 1989)

27
Real Time Quality Assurance

• Online quality control system
• Matches with ships/buoys computed 3 hours after
  each orbit
• RMS and bias computed on a daily basis
• Email warnings automatically sent to key staff
• Flagged observation platforms where spiking or
  high bias evident

28
Areas for Improvement

• Image Navigation
• Critical if a high-resolution product is to be
  generated
• Must be fully automatic for operational use in
  real-time processing (i.e. no human interaction)
• Must be accurate (1 pixel or better), consistent
  and reliable
• Ephemeris data is not sufficient

29
Improved Navigation using CAPS

• Common AVHRR Processing System http//www.dar.csi
  ro.au/rs/Capshome.html
• CAPS is an extension of Tcl/Tk for processing
  AVHRR, and ATSR-2 GBT satellite data. CAPS can
  also display and reproject MODIS, MISR and ASTER
  data.
• Uses the Clift navigation model for AVHRR
  delivering approximately 1-pixel accuracy
  (consistently)
• Incorporates Numeric Array Processor (NAP) which
  allows simple manipulation of array data
• Orbital elements, satellite attitude and clock
  correction files downloaded from CSIRO Marine
  Research, Hobart via FTP every 30 minutes
• elements computed by pattern matching,
  least-squares fit to image chips after each pass
  (Hobart data only at present)
• Ephemeris works well in other parts of the world
  and also has good accuracy over the short term
  meaning it can be used on passes where there is
  little or no land

30
Envisat AATSR Meteo SST product produced by ESA

• ATS_MET_2P is a level 2 product, containing SSTs
  for all cloud-free pixels
• ESAs Envisat-1 polar-orbiting satellite
• Skin SST at ? 10 ?m
• Resolution 10 arc minute x 10 arc minute (17 km)
• Digitization 12-bit
• Data up to 3 hours old (3 day repeat)
• Accuracy 0.3 K
• Biases may occur due to cloud contamination,
  atmospheric aerosols and water vapour
• Less affected by aerosols compared with other
  infra-red sensors
• Obscured by clouds

31
Aqua AMSR-E SSTs

• Passive-microwave radiometer
• Sub-skin SST at ? 1 mm
• Resolution 0.25? ? 0.25? grid
• Data 1 day old (daily repeat)
• Source Remote Sensing Systems
• Accuracy ?1 K
• Relatively insensitive to atmospheric effects
  compared with most infra-red sensors
• Largely unaffected by clouds
• Sensitive to ocean surface roughness and
  precipitation

32
MTSAT-1R Imager SSTs

• JMAs MTSAT-1R satellite was launched Feb 2005
• Regional coverage around a geostationary orbit at
  140 E. Useful information between 70N and
  70S.
• Resolution Bureau Space Based Observations
  Section will grid the 4 km raw data at 5 km grid
  resolution
• Hourly data
• Affected by aerosols and atmospheric water vapour
• Initially intend to collaborate with JMA to
  compute bulk SSTs from MTSAT-1R using similar
  equations as currently used for AVHRR
• Bureau will later attempt to calibrate the raw
  brightness temperatures to skin temperatures
  using AATSR Meteo Product skin SSTs

33
In Situ SSTs

• ARGO floats 4 5 m depth, ? 0.1? lat/lon,
  accuracy at best 100 m from ARGOS System, data
  accuracy
• Drifting buoys 0.2 m and 1 m depth, ? 0.1?
  lat/lon, accuracy at best 100 m from ARGOS
  System, data accuracy ? 0.25 K
• XBTs 3.7 m depth, ? 0.1? lat/lon, accuracy at
  best 100 m from ARGOS System, data accuracy ?
  0.15 K depending on air temperature
• Ships 0.3 m to 10 m depth, ? 0.1? lat/lon,
  accuracy ? 0.1? lat/lon, data accuracy 0.1 K

34
BoM Operational SST Analysis System

• Running operationally at BoM since 1993
• Global analyses are done once per week on a 1
  grid
• Regional analyses are done daily on a 0.25 grid
• http//www.bom.gov.au/bmrc/ocean/results/SST20ana
  ls

35
BoM SST Analysis System Inputs

• In situ SST data - buoys, volunteer observing
  ships and XBTs
• AVHRR SST data processed by NESDIS at 200 km
  global resolution
• Sea-ice edge data - NOAA/NCEP
• Local AVHRR retrievals used for regional (daily)
  analysis
• Climatology from Reynolds and Smith (1994)
• Statistical parameterisations for
  forecast/first-guess error variances and
  covariances, and for observational errors
• Statistical forecasts based on a combination of
  climatology (low weight) and the previous
  analysis.

36
BoM SST Analysis System Processing

• Univariate statistical (optimal) interpolation
• In situ SST data used to provide an initial
  (unbiased) estimate of SST
• Broad-scale (750 km) in situ analysis is compared
  with a similar broad-scale analysis of the AVHRR
  data, and the difference is used to correct all
  the AVHRR data
• Fine-scale (100-200 km) analysis is performed
  using all available SST data
• For the regional analysis, the scales are 30-50
  km and the 1 degree analysis is used to correct
  possible bias in the local retrievals.

37
BLUElink High Resolution SST Analysis System

• Output real-time sea-surface temperature (SST)
  analyses at 5 km resolution every 6 hours
• Australian region 20?N to 65?S, 50?E to 160?W
• Data sources
• AVHRR based on retrievals for bulk SST (Local
  1km Global 9 km)
• ENVISAT AATSR Meteo Product similar to AVHRR
  but skin, higher quality, 17 km
• Regionally geostationary satellite data from
  MTSAT-1R and possibly FY-2C to resolve the
  diurnal cycle and fill in some cloud gaps 5 km
• Microwave SST data for persistent cloud gaps 38
  km
• In situ bulk SST for validation
• CSIRO (Ian Barton) - diagnostic data set
  comparing ship-mounted radiometer data with
  satellite temperatures
• An Australian component of GODAE High-Resolution
  SST (GHRSST) Pilot Project
• See web page at http//www.bom.gov.au/bmrc/ocean/
  BLUElink/SST_external.html