Document Sample
					                         OCEAN MODEL ANALYSIS AND PREDICTION
                         FORECASTING BASED ON NEAR REAL-TIME
                            SATELLITE ALTIMETRY AND ARGO
                                                Gary B. Brassington

                                     Bureau of Meteorology Research Centre,
                                   700 Collins Street, Melbourne, 3000, Australia

ABSTRACT/RESUME                                              management of real-time observations; quality control;
                                                             archive management; and data servicing.
BLUElink> is a joint Australian government initiative
to develop Australia’s first operational ocean               The principle user of the operational products and joint-
forecasting system called OceanMAPS. The project has         partner in the project is the RAN. However the
transitioned to the implementation and trial phase using     operational system being developed at the Bureau is
the infrastructure of the Bureau of Meteorology.             aimed at a more general public service for the benefit of
OceanMAPS has a global grid with 1/10° by 1/10°              the wider Australian maritime community. Australia is
resolution in the Australian region (90E-180E, 70S-          an island continent surrounded by three of the worlds
16N) and uses the Modular Ocean Model version 4              major oceans. Australia’s 200nm Exclusive Economic
optimised for the NEC SX6. The analysis uses an              Zone ranks as the third largest in the world behind the
ensemble based multi-variate optimal interpolation           United States and France. Guiding the sustainable
scheme where model error covariances are derived from        management of this vast resource is a critical activity for
a 72-member ensemble of intra-seasonal anomalies             the Australian government. Important users will include:
based on a 12-year ocean only model integration. The         the Australian Maritime Safety Authority for ocean
scheme has been formulated to assimilate near real-time      surface tracking; marine management such as the Great
sea level height anomalies processed from Jason-1,           Barrier Reef Marine Park Authority; recreational ocean
ENVISAT and Geosat Follow-On and profile                     racing such as the Sydney to Hobart Yacht race and
observations including Argo, XBT and the TAO array.          other related activities including tourism, angling diving
The operational configuration including the data             and boating.
management of the near real-time observations is
reviewed.                                                    2. OCEAN OBSERVING SYSTEM

1. INTRODUCTION                                              Advances in developing the global ocean observing
                                                             system have been critical to Australia realising an
In 2003, the Australian Government initiated a $15M          operational ocean prediction capability. Australia
project called “BLUElink – Ocean Forecasting                 continues to actively engage in international ocean
Australia” to deliver operational short-range ocean          observing programmes that contribute to the real-time
forecasts for the Asian-Australian region. The partners      system through SOOP and Argo. Investment by the
assigned to accomplish the task included the Bureau of       Australian government in satellite ocean observing
Meteorology (the Bureau), the Royal Australian Navy          systems has largely been restricted to scientific
(RAN), and the Commonwealth Scientific and Industrial        contributions. The BLUElink project is providing a
Research Organisation (CSIRO).                               focal point for evaluating the quality of this system in
                                                             Australia’s regional sea.
CSIRO has lead the collaborative development of both
the Ocean Forecast Australia Model (OFAM) [1] and            2.1 Argo
the BLUElink Ocean Data Assimilation System
(BODAS) [2]. This development culminated in                  Argo was initiated in 1999 with a target of maintaining a
BLUElink’s first major achievement, a 15 year ocean          network of 3000 autonomous profiling floats. At Feb
reanalysis experiment referred to as the BLUElink            2006 the deployment had reached ~2400 floats.
ReANalysis (BRAN). The Bureau has lead the                   Australia has contributed 117 deployments to the
development of the ocean prediction system referred to       network with an on-going commitment to manage the
here as the Ocean Model Analysis and Prediction              real-time and delayed mode retrievals. Argo science
System (OceanMAPS) [3]. Developments include                 team provide images of the float distribution as well as
extending the model and assimilation system,                 maps of the density updated monthly as shown in Fig.1.
developing the system infrastructure including data          Focusing on the density found in Australia’s marginal
seas reveals three regions that have consistently shown a   Sea with part of the northern domain showing a counter
low density of floats: the North West shelf; Coral Sea;     current. This suggests that on average there is a low
and the Great Australian Bight. Forecasts of heat           probability for drifters moving into the region during
content in the Coral Sea are of particular interest for     these months. However, in August (and September not
applications of tropical cyclone forecasting and GBR        shown) the streamlines demonstrate a broad region of
marine park management for coral bleaching.                 penetrating streamlines along both sections indicating a
                                                            high probability for westward propagation of drifters
                                                            into the Coral Sea. All streamlines that penetrate into
                                                            the Coral Sea, across all seasons, show a dominant
                                                            westward propagation of ~6°/month, many of which
                                                            reach the GBR. This indicates a high attrition rate for
                                                            Argo drifters and a relatively short residence time.
                                                            (Trajectories from the global surface drifter program
                                                            show similar behaviour, not shown).
Figure 1: Argo worldwide density at the 31 January
2006. (Courtesy of SIO/JCOMM)
The bathymetry of the Coral Sea shown in Fig, 2 shows
that its eastward boundary is defined by several island
groups formed along the edge of the continental plate
including the Solomon Islands and Bank Islands.

Figure 2: Ocean bathymetry for the Coral Sea. Land is
shown in white and the 1000m depth contour line is
shown in black.                                             Figure 3: Streamlines of the horizontal currents and
                                                            temperature averaged over the top 200m. The
This chain of islands act as a significant barrier to the   streamlines are initiated from the 172.5E meridion and
South Equatorial Current (SEC) that is a feature of the     the latitude line 7S at 0.25 degree intervals. The data is
general circulation of this region. The Australian          based on a monthly average of BRAN from 1993-
continent represents the western boundary to the SEC        2002.(a) February and (b) August.
however the equatorial return flow is yet to be clearly     The BRAN analysis and the use of streamlines are not
observed and defined. BRAN has provided the best            definitive evidence in of themselves to evaluate the
available model analysis for the Australian region for      Argo observing system for the Coral Sea. The findings
the period 1991-2004. Streamlines derived from the          indicate that a more comprehensive study based on
monthly means for February and August averaged for          tracing historical records of surface drifters and Argo
the years 1993-2002 are shown in Fig.3a and 3b              floats is required. These results are suggestive only that
respectively Sreamlines have been traced from the west      the low density shown in Fig.1 for this region may
Pacific along the 172.5E meridion and from the              continue to persist when Argo reaches full deployment
equatorial Pacific along the latitude line of 7S. The       and optimal distribution. It also indicates that the GBR
velocity field represents the average over the top 200m.    is likely to be a net sink for both the Argo and global
Streamlines are not equivalent to Lagrangian paths in       surface drifter program.
unsteady flow, however the persistence of westward
flow throughout the climatological model year suggests      2.2 Satellite altimetry
a quasi-steady mean circulation. February is typical of a
signifcant portion of the climatological model year (Oct-   At present there are three instruments available for
May) where few streamlines penetrate into the Coral         operational oceanography, Jason, ENVISAT and GFO.
Each platform has unique orbit characteristics as shown
in Tab. 1 that impact the resolution of sampling and the
coverage of the total observing system.

Platform      Period       Along-track          Latitude
              (days)      Resolution (km)        range
  Jason        ~9.92            ~7.6           66S – 66N
ENVISAT         ~35             ~5.8            79S - 79N
  GFO          ~16.9            ~7.0            72S - 72N

Table 1: Characteristics of orbit tracks for satellite
altimeter platforms
The global coverage of satellite platforms make them a
crucial component to the observing system for countries
with modest in-situ programs like Australia. However
                                                            Figure 5: Satellite altimetry passes for Jason, ENVISAT
the coverage of sub-orbits is not uniformly distributed
                                                             and GFO over the Australian domain from 1st, 2nd and
over the domain of interest. An example of one days
                                                                                  3rd Aug. 2003
coverage for all three platforms is shown in Fig. 4 for
the Australian region.                                      The quality of the distribution of altimeter observations
                                                            is a function of the scale of interest and gaps remain at
                                                            scales on the order of 4°!4°. The spatial scale for
                                                            BLUElink is determined by the model error covariances
                                                            combined together with localisation (L=8° for BRAN
                                                            [2]) used in the data assimilation scheme.

Figure 4: Satellite altimetry passes for Jason, ENVISAT
and GFO over the Australian domain for 3rd Aug, 2003.
The distribution of satellite observations for the 3rd
August 2003 is an example where the entire East             Figure 6: Average number of sea level height
Australia Current (a key area for maritime activity) was    observations from Jason, ENVISAT and GFO per day
not observed. However, in other regions such as (130E,      per 2°!2 ° bin. Black squares indicate zero observations.
55S) there is a convergence of multiple tracks from
multiple platforms. This pattern changes from day to        The average number of observation per day per 2°! 2°
day however during March 2003 the large data gap in         area for sea surface height observations from Jason,
the EAC region occurred as frequently as every three        ENVISAT and GFO combined is shown in Fig. 6. The
days. If we round up the orbit periods to 10 and 17 days    higher density of observations at high latitudes is
for Jason and GFO respectively the combined repeat          consistent with standard orbit tracks. The noticeable
period of the observing system is approximately 3.25        checkerboard indicates the gaps between orbit crossings.
years and the pattern in Fig. 4 should occur again later    The relatively low density in the tropics and equatorial
this year (Oct. 2006). The strategy that was adopted for    region is a result of larger separation of passes and the
BRAN to improve this distribution was to widen the          interference from precipitation. The large volume and
window to 3 days. An example of which is shown in           coverage of quality observations from these three
Fig. 5 for the 1st – 3rd Aug. 2003.                         platforms makes this the dominant observing system for
                                                            given for the 2nd March 2006. Argo peaks at ~2 days
2.3 Other observations                                      behind real-time. Additional profile observations from
                                                            Coriolis peak 3-4 days behind real-time.
Sea surface temperature from satellite platforms has a
longer history than altimetry and provides a much larger
volume of quality observations. BLUElink is supporting
research for the development of high resolution sea
surface temperature analyses as a contribution to the
GHRSST project. However these quality controlled
observations are not directly assimilated into
OceanMAPS. The SST analyses are instead used to
apply surface restoring. Plans for a follow-on project to
BLUElink include assimilating SST as part of the
prediction system.

The global drifter program is another significant global
observing program that is targeted toward observing
mesoscale oceanography. BLUElink has not included
these into the assimilation system and are used only as
an independent validation.
                                                            Figure 7: The total number of unique observations
2.4 Real-time data management                               based on retrievals from USGODAE, Coriolis and GTS
                                                            for 2nd Mar. 2006 and the timeliness of those profile
The GTS is the primary operational communication for        data.
in-situ ocean observations. The Bureau have established
                                                            None of the satellite altimeter platforms are rated as
infrastructure to receive and manage this stream. There
                                                            operational systems and delivery of products is under a
are essentially two formats for ocean observations,
                                                            best efforts basis. The first operational platform to be
BATHY (bathythermal observations) and TESAC
                                                            managed by NOAA will be Jason-2, scheduled for
(TEmperature, SAlinity and Currents). These formats
                                                            launch in mid-2008. The dependence of BLUElink on
have the advantage of compact size but prove less
                                                            the existing altimetry makes this a clear vulnerability to
convenient for data handling compared with NetCDF
                                                            the prediction system. This status will likely prevent
the standard adopted by the Argo program. The Bureau
                                                            OceanMAPS being certified as a fully operational
have developed an internal program to convert the GTS
                                                            system at the Bureau in 2007.
message into the Argo format. A priority for BLUElink
has been to ensure the maximum number of Argo
                                                            The BLUElink project does not process the raw satellite
retrievals available in real-time. The Argo program have
                                                            altimetry observations and relies on the managing
devised a distributed data management system with each
                                                            centres JPL, ESA and NOAA providing a processed sea
partner forming a Data Assembly Centre to manage the
                                                            surface height anomaly product. The quality of these
retrieval, real-time and delayed mode quality control
                                                            products is impacted by the algorithm used to compute
and distribution through both the GTS and the two
                                                            the Geophysical Data Record. The faster algorithms
GDAC’s (USGODAE and Coriolis). Analysis of the
                                                            required for the near real-time delivery impact the
three sources of real-time Argo files have revealed that
                                                            quality of the products. Sensitivity of analyses to real-
none of these sources contain a complete superset of all
                                                            time ssha products will be assessed during system trials.
the real-time observations. In order to maximise the
number of Argo observations available to BLUElink the
                                                            Each of the three sources of sea level anomaly products
Bureau have developed a system based on all three
                                                            are obtained in a unique way. Jason is obtained via
sources. A duplicate checker was developed to locate all
                                                            OCEANIDS and ftp push service provided by
duplicates and form a single file containing the best
                                                            PO.DAAC. ENVISAT is obtained via an ftp push from
copy (the highest level of quality control) amongst the
                                                            ESA. GFO is retrieved by ftp from NOAA. Both Jason
duplicates for each day. The sorting is performed in the
                                                            and GFO have alternative retrieval pathways through
order of USGODAE, Coriolis then GTS. Initial
                                                            USGODAE providing useful backup. Each data stream
processing amongst the USGODAE and Coriolis
                                                            has its own unique file format which is processed at the
highlighted the differences in the updating cycles and
                                                            Bureau and passed through a series of quality control
the high number of duplicates that could be found
                                                            checks and reformatted to BUFR for storage and
within each source. Current behaviour is shown in Fig.7
                                                            netCDF for use in the analysis.
where the relative contribution of unique profiles and
timeliness in days behind real-time of each source is
                                                            3. SURFACE FLUXES
                                                               The broad features of shallow mixed layers over much
The Bureau maintain a set of operational NWP forecast          of the central domain; deeper mixed layers in the
systems including: GASP [4] a global assimilation and          equatorial and high latitude region show good
prediction system based on a spectral model and LAPS           correspondence. OFAM shows evidence of deeper
[5] a limited area prediction system for the Australian        mixed layers over the tropics indicating stronger
region. GASP has a horizontal resolution of 0.75°!0.75°        westerlies and La Nina like conditions. There is also
resolution in the tropics and uses a gaussian grid in high     evidence of significant penetration of SEC into the
latitudes and has 33 sigma levels. The analysis uses a         Coral Sea compared with CARS. The high latitudes in
generalised multi-variate statistical interpolation scheme     OFAM are characterised by very distinct narrow
(GenSI) which has been recently upgraded to include            filaments that retain this characteristic after averaging.
scatterometer observations from QuikSCAT. The                  This could be a realistic feature determined by
analysis for GASP is performed on a 6 hourly cycle with        bathymetric steering. The same region in CARS is
10day forecasts issued every 12 hours as 06Z and 18Z.          smooth consistent with the 1° resolution. BLUElink has
Analyses have demonstrated modest improvements in              developed a high resolution climatology for the
bias and skill in forecast winds [6, 7]. LAPS has a            Australian region that is due to be published in 2006.
horizontal resolution of 0.375°!0.375° over the
Australian domain 65.00E-184.6250E and 65.000S-
17.1250N. LASP uses a GASP background state to
perform an analysis using the GenSI scheme. LAPS
performs a 3 day forecast every 12 hours at 0Z and 12Z.
GASP fluxes are retrieved from the operational disk
storage and interpolated onto the OFAM grid using
Delaunay triangulation.


BLUElink have implemented the GFDL Modular Ocean
Model version 4 (MOM4) [1, 8] with specific
enhancements for mixed layer physics [9] and the
algorithm for calculating penetrative solar radiation. In      Figure 8: Depth the of temperature that is 1.0° C cooler
addition, the software was vectorised for use on the           than the surface. (a) mean January from CARS, (b)
NEC SX6 computing architecture a joint facility of the         mean January from OFAM based on 1998-2004
Bureau and CSIRO. OFAM uses a global grid with a
horizontal resolution of 0.1°!0.1° in the Australian
                                                               5. BLUELINK OCEAN DATA ASSIMILATION
region defined by 90E-180E and 75S-16N which is
designed to resolve mesoscale ocean dynamics. Outside
this region the grid is stretched up to a resolution of 0.9°
                                                               BLUElink has developed a new software for ocean data
in the Indian Ocean and South Pacific Ocean and
                                                               assimilation which uses an ensemble based multi-variate
tropical North Pacific. Beyond this domain the
                                                               optimal interpolation scheme [2]. A specific feature of
resolution in the North Atlantic is 2°!2°. This                the scheme is the use of an ensemble of mesoscale
configuration has 71% of the total grid points within the      anomalies from OFAM to form the statistics for the
Australian region and 72% of all model grid points are         model error covariance. BRAN was implemented based
water cells. The OFAM grid provides dynamically                on a 72 member ensemble which was constrained by the
consistent and numerically well behaved boundary               length of the free model run available and the
conditions to the Australian region.                           computational cost of the scheme. In general, the
                                                               undersampling of the statistics gave rise to large far
The OFAM model has been integrated over the period             field covariances. These were assumed to be unrealistic
1991-2004 using ERA40 and ECMWF forecast fluxes.               features for mesoscale variability and were controlled
These spinup runs have been used to diagnose model             through localisation. It is critical to this scheme that the
performance such as mixed layer depth against the              free ocean model driven by surface fluxes maintains
CSIRO Atlas of Regional Seas climatology [10] as               both a realistic mean ocean state and a realistic
shown in Fig. 8. The CARS dataset is derived from              distribution of ocean variability. A complete analysis of
historical observations from the NODC world ocean              OFAM is a significant on-going task. Analyses to date
atlas 1998 and CSIRO data holdings. The results for            have discovered some biases in surface fluxes and in the
OFAM are based on the period 1998-2004 which is                mixed layer parameterisation leading to improvements
beyond the CARS climatology so that a portion of               that are seen in Fig. 8 (and other diagnostics not shown).
variation will be attributable to interannual variability.
BRAN was the first test of the OFAM and BODAS                8. REFERENCES
system. It was conducted for a period determined by the
availability of altimeter observations, July 1992– March     1. A Schiller, G B Brassington, R Fiedler, D A Griffin, J
2005. This included ERS1 and Topex-Poseidon in               Mansbridge, P R Oke, K R Ridgway, and N R Smith,
addition to those in Tab. 1 and all available quality        ‘Eddy-resolving ocean circulation in the Asian-
controlled in-situ profile data. The primary independent     Australian region inferred from an ocean reanalysis
observation used to compare with BRAN has been               effort’. manuscript in preparation.
surface drifters. In regions where both tidal induced and
wave induced surface motion are weak, BRAN shows             2. P R Oke, A Schiller, D A Griffin and G B
remarkably good agreement with drifter tracks [2].           Brassington, ‘Ensemble data assimilation for an eddy-
BRAN was performed on 6 nodes of an NEC SX6 using            resolving ocean model of the Australian region’, Q. J.
priority queues and was completed in 5 months                Roy. Meteorol. Soc., in press.
generating ~8Tbytes for daily averaged prognostic
variables. Analyses of BRAN are on-going and follow-         3. G. B. Brassington, G. Warren, N. Smith, A. Schiller,
on analysis “BRANII” is scheduled for 2007.                  P. R. Oke, BLUElink> Progress on operational ocean
                                                             prediction for Australia, Bulletin of the Australian
6. OCEANMAPS/CONCLUSION                                      Meteorological and Oceanographic Society, 18, pp104-
                                                             109, 2005.
BRAN was an important milestone in the BLUElink
project that has provided a level of confidence that the     4. Seaman, R., Bourke, W., Steinle, P., Hart, T.,
ocean prediction system being developed will have            Embery, G., Naughton, M and Rikus, L., Evolution of
some level of forecast skill. The on-going analysis of       the Bureau of Meteorology's Global Assimilation and
BRAN will continue to lead to refinements of parameter       Prediction System, Part 1: Analyses and Initialization,
choices and further improvements in skill. However,          Aust. Met. Mag., 44, 1-18, 1995.
there are also several distinctions between BRAN and
OceanMAPS that are likely to deteriorate forecast skill.     5. Puri, K., Dietachmayer, G., Mills, G. A., Davidson,
These include: (1) a larger forecast period of 7 days; (2)   N. E., Bowen, R. A. and Logan, L. W., The new BMRC
the use of NWP forecasts; (3) real-time observations;        Limited Area Prediction System, LAPS, Aust. Met.
and (4) operational scheduling for delivery of forecasts.    Mag., 47, 203-233, 1998.

The Bureau’s experience in supporting operational            6. J D Kepert, D J M Greenslade and G D Hess.
systems was the logical choice for developing and            Assessing and improving the marine surface winds in
supporting OceanMAPS. A large array of operationally         the Bureau of Meteorology numerical weather
supported infrastructure is in place that have been          prediction systems. BMRC Research Report No. 10,
adapted into the ocean prediction system. These include      2005.
the high performance computing facility, existing
communication networks for retrieval of GTS and other        7. Eric W. Schulz, Jeffrey D. Kepert, and Diana J. M.
real-time observations, real-time databases and archive      Greenslade , An Assessment of Marine Surface Winds
devices. Key issues addressed by the BLUElink team           in Numerical Weather Prediction Systems at the
include handling non-standard file formats that occur in     Australian Bureau of Meteorology (in press)
the ocean community and servicing of the large data
                                                             8. S M Griffies, M J Harrison, R C Pacanowski, and A
volumes. In particular the WMO standards of GRIB and
                                                             Rosati, ‘A technical guide to MOM4’. GFDL ocean
BUFR are not used in the ocean community. BLUElink
                                                             group technical Report No. 5, 339pp, 2004.
is developing internet server solutions based on
OPeNDAP intended to insulate the maritime user
                                                             9. D Chen, L M Rothstein and A J Busalacchi. ‘A
community from the details of the data management
                                                             hybrid vertical mixing scheme and its application to
system. OceanMAPS is in trial mode with a focus on
                                                             tropical ocean models’. J. Phys. Oceanogr., 24, pp2156-
testing the robustness of infrastructure components. The
                                                             2179, 1994.
first system trial is focusing on selected target periods
including Jan-Mar 2005 which overlaps with BRAN
                                                             10. Ridgway K.R., J.R. Dunn, and J.L. Wilkin, Ocean
and includes dynamical features such as tropical
                                                             interpolation by four-dimensional least squares -
cyclone Ingrid and southern extension of the EAC.
                                                             Application to the waters around Australia, J. Atmos.
                                                             Ocean. Tech., Vol 19, No 9, 1357-1375, 2002

BLUElink science team and Dr. Spillman for Fig. 7.