2. Latest Forecasts
a. Forecasts of NINO area SST anomalies
Dynamical
seasonal forecasts from the POAMA-1 system
contributed by Oscar Alves, Guomin Wang and Harry
Hendon
Bureau of Meteorology
Research Centre, Melbourne, Australia.
POAMA (Predictive Ocean Atmosphere Model for
Australia) is an intra-seasonal to inter-annual climate prediction system based
on coupled ocean and atmosphere general circulation models (Alves et al, 2003).
The first version (POAAM-1) was developed in a joint project involving the
Bureau of Meteorology Research Centre (BMRC) and CSIRO Marine Research (CMR).
The POAMA model uses the Australian Community Ocean Model 2 (ACOM2) and
the latest version of the Bureau’s Atmosphere Model (BAM). Real time oceanic
and atmospheric initial states are used to initialise the coupled model. These
are provided by an ocean data assimilation system that is run in real time as
part of the POAMA system and by the Bureau of Meteorology operational weather
analysis and prediction system (called GASP – Global Assimilation and
Prediction System).
A seasonal forecast is produced
each day using the latest ocean and atmospheric initial conditions. Products
are based on the latest 30 daily forecasts, forming an effective 30-member
ensemble. A basic set of forecasts is presented in section 2. Many other plots
from both real-time forecasts and hind-casts are available on the POAMA web
site (http://www.bom.gov.au/bmrc/ocean/JAFOOS/POAMA).
These include horizontal plots of SST anomalies for each lead time, as well as
equatorial/time plots with daily resolution of SST, 20C isotherm depth,
outgoing long-wave radiation and surface zonal wind anomalies. These are
produced for means of monthly ensembles, mean of last 30 forecasts and also
individual forecasts.
One feature of the POAMA system is the ability of the atmosphere model
to represent the Madden-Julian Oscillation (MJO). This, together with the use
of real-time ocean and atmospheric data, means that POAMA can also produce
forecasts of intra-seasonal variability out to a few weeks lead-time.
Experimental intra-seasonal forecasts are also available on the POAMA web site.
The POAMA-1 system uses the
latest version of the Bureau of Meteorology unified atmosphere model (BAM
version 3.0d). It uses a modified convection closure that allows the model to
have a good representation of the MJO. It has a horizontal spectral resolution
of T47 and has 17 vertical levels. The performance of this model forced with
observed SST is described in Colman et al (2003).
The ocean model component is ACOM2. It was developed
by CMR, and was based on the Geophysical Fluid Dynamics Laboratory Modular
Ocean Model (MOM version 2). The grid spacing is 2 degrees in the zonal
direction. The meridional spacing is 0.5° within 8° of the equator, increasing
gradually to 1.5° near the poles. There are 25 levels in the vertical, with 12
in the top 185 metres. Technical details of ACOM2 are given in Schiller et al.,
1997 and Schiller et al. 2002.
The ocean and atmosphere models
were coupled using the Ocean Atmosphere Sea Ice Soil (OASIS) coupling software
(developed by CERFACS, France; Valcke et al, 2000).
The ocean data assimilation scheme is
based on the optimum interpolation (OI)
technique described by Smith et al (1991). Only temperature observations are
assimilated and only measurements in the top 500m are used. There are several
improvements over the scheme described by Smith et al (1991). The OI scheme is
used to correct the model background field every 3 days using a 3 day
observation window, one and a half days either side of the assimilation time.
Current corrections are calculated by applying the geostrophic relation to the
temperature corrections, similar to the method described by Burgers et al. (2002).
For the real time forecasts the atmospheric component
is initialised with weather analysis from the Bureau of Meteorology’s operational
Numerical Weather Prediction (NWP) system (GASP). This means that the seasonal
forecast model knows about the latest intra-seasonal state of the tropical
atmosphere.
The POAMA-1 system has been run daily in real-time by
the Bureau of Meteorology operations branch since 1st October 2002.
This real-time system consists of two suites: main ocean analysis cycle and
forecast cycle. These are described below.
The main analysis cycle aims to use as many ocean sub-surface
observations as possible to provide an estimate of the ocean state in near-real
time. To allow as many observations as possible to be used the system is run
approximately 10 days behind real-time. Each day the ocean state is integrated
forward one day using the ocean model. The ocean model is forced with
six-hourly fields from the Bureau of Meteorology GASP NWP system.
Every third day observations are assimilated into the ocean model. All
available sub-surface temperature observations from the Global
Telecommunications System (GTS) are used. Sea surface temperature observations
are not assimilated. Instead, the ocean model surface temperature is relaxed to
the SST analysis field used in the GASP system with an e-folding time scale of
3 days.
The second suite is the forecast
cycle, which is run every day. This consists of an ocean only catch-up analysis
to bring the ocean state up to the last day, followed by a coupled model
forecast. A significant number of ocean observations are received over the GTS
within a day of real-time, for example, observations from the TRITON/TAO array.
This means that more is know about the latest state of the ocean than the
information that went to produce the main analysis because the main analysis is
produced 10 days behind real-time. For this reason a catch up analysis is
produced each day as part of the forecast cycle. The ocean model is integrated
forward from the main analyses to the present and observations are assimilated
every three days as in the main analyses. The catch-up analysis does not have
any impact on the main analyses. Then a 9-month coupled model forecast is
produced in real-time using the very latest ocean state from the catch-up
analysis and the latest atmospheric state from the GASP NWP analysis. The daily forecasts are combined to form a
30-member monthly ensemble and an ensemble of the last 30 forecasts (updated
daily on the POAMA web site).
Figure 1. SST anomalies in the NINO3 (left) and
NINO4 (right) regions. The latest 30 daily forecasts are shown (as available on
day of submission of report, the start dates of the forecasts are shown in the
plot heading). The red forecasts show the latest 15 forecasts and blue the
previous 15. The green line is an estimate of the observed SST anomaly from the
BMRC ocean analysis system. All anomalies are relative to the 1987-2001 period.
The focus of the POAMA-1
system is the prediction of tropical SST anomalies. Figure 1 shows the forecast
SST anomalies in the NINO3 and NINO4 regions. The last 30 forecasts are shown
(those available on the day of submission of this report). All forecast
anomalies are calculated relative to the model hind-cast climatology over the
period 1987-2001, using all the hind-casts starting at the same time of the
year.
Figure 2 shows
distribution plots of the NINO3 SST anomalies. It shows the percentage of
ensemble members whose anomaly lie between within each 0.4°C bin interval, for various ranges of forecast
lead-time. These provide forecast distributions for NINO3 SST anomalies, based
on the ensemble spread.
Figure 2. NINO3 SST anomaly distribution for lead times 4-6 months (left) and
6-9 months (right). These show the percentage of ensemble members lying within
the specified 0.4°C temperature anomaly bins.
Figure 3. Ensemble mean SST anomalies at 3 months lead (left) and six months
lead (right). These are based on the latest 30 daily forecasts (as available on
day of submission of report). All anomalies are relative to the forecast
climatology over the 1987-2001 period.
The daily forecasts are used to create an ensemble
mean based on the latest 30 forecasts. Horizontal distributions of ensemble
mean SST anomalies are shown in figure 3, for different lead times.
The spread about the ensemble mean can been seen in
figure 4, which shows “spaghetti plots” containing SST anomalies for each
individual forecast. These plots show the spatial consistency of the ensemble
members. For example, many contours of the same value in the same region
indicate consistence, however, contours of different values in the same region
indicate that there is ensemble spread.
Figure 4. “Spaghetti” plots of SST anomalies at 3 month (top) and six month
(bottom) lead times. The last 30 forecasts are plotted. The SST anomaly
contours from each individual forecast are plotted to show the ensemble spread.
Contours are as follows: Black: +3°C, Red: +2°C, Orange: +1°C, Green: -1°C, light Blue: –2°C, Blue: –3°C.
Testing of the skill in the
coupled model used the so-called “hind-cast test”. This method initializes a
prediction on, for example, 1 March 1987 using only information available
before that date. The test of the model is then based on comparing the
prediction to information collected during the remainder of 1987. Hind-cast
tests initialized at many past dates can be combined into statistics to
evaluate the accuracy of the forecasts, and to provide a measure of “skill” of the model.
A set of 180 forecasts, one per month (started on the
1st of each month) for the years 1987 to 2001, have been used to
assess the performance of the model. Forecast anomalies are calculated relative
to the hind-cast climatology using all the hind-casts starting at the same time
of the year. Ocean initial conditions were taken from an ocean assimilation
that was carried out from 1982 to 2001 using the same assimilation system as
used for the operational version. GASP atmospheric initial conditions were not
available for all of this period. Instead the atmospheric state was taken from
the appropriate date during an integration of the atmosphere model forced with
observed weekly Reynolds SST.
The skill measures presented here are likely to be an
under-estimate of the skill of the real-time system for several reasons:
(a)
Because
in the hind-casts the model was not initialized with the true atmospheric state
from GASP, we do not expect it to perform as well as in the real time forecasts.
(b)
The
ocean observing system is being continually improved. For example, it was only
in the early 1990’s that the TOGA-TAO array was implemented in the tropical
Pacific. This array consists of moored temperature sensors and is the main
source of observations in the equatorial Pacific. Another major revolution in
the ocean observing system is happening right now. Around 300 autonomous
floats, which drift around in the ocean measuring temperature and for the first
time salinity, are being deployed in a project called Argo (Argo 1998). As new
Argo floats become available over the GTS they are automatically used by the
POAMA system.
(c)
Only
a single forecast member was used per month for the hind-casts, whereas the
real-time system uses an ensemble of 30 forecasts per month.
Figure 5. Left - NINO3 anomaly correlation as a function of lead time. Right – rms error (solid) and standard deviation (dashed) for NINO3 SST anomalies. Based on 180 forecasts, one per month during the period 1987-2001. Red – persistence of initial SST anomalies, Green – POAMA-1 coupled model.
One measure of forecast skill is the anomaly correlation coefficient for
NINO3 SST anomalies. This is shown in figure 5 for both model and persistence,
as a function of lead-time. The plot shows that the model beats persistence at
all lead times, even during the first month of the forecast. At eight months
lead-time the skill is relatively high at 0.7 and significantly better than
persistence at 0.2. Also shown in figure 5 is the rms error of the model and
persistence. Again the model beats persistence at all lead times. The standard
deviation of NINO3 SST anomalies (also in figure 5) shows that the model
maintains a realistic level of variability at all lead times.
Horizontal patterns of anomaly correlation skill are
shown in figure 6 for lead times of three and six months. At all lead times the peak in skill is
concentrated in the central and eastern Pacific, associated with the El Nino/La
Nina phenomena. At three months lead time the anomaly correlation reaches up to
0.9 in the central Pacific. At six months lead-time it reaches up to 0.8 in the
central Pacific south of the equator. These skill measures are competitive with
the best international models, especially when taking into account that only
one hind-cast per month was used.
Figure6. Spacial distribution
of anomaly correlation of SST anomaly
(x100). Top- 3 month lead time and bottom – 6 month lead time.
Alves, O., G. Wang, A. Zhong, N. Smith, F. Tseitkin, G.
Warren, A. Schiller, S. Godfrey and G. Meyers, 2003. POAMA: Bureau of
Meteorology operational coupled model seasonal forecast system. Proc. National
Drought Forum, Brisbane, Apr 2003, pp. 49-56. Available from DPI Publications,
Department of Primary Industries, GPO Box 46, Brisbane, Qld 4001, Australia.
ARGO, 1998. On the design and implementations of Argo-
A global array of profiling floats. The ARGO science team. 26 pp.
Burgers, G., M. A.
Balmaseda, F. C. Vossepoel, G. van Oldenborgh, P. van Leeuwen, 2002: Balanced
Ocean-Data Assimilation near the Equator. Journal
of Physical Oceanography: Vol. 32,
No. 9, pp. 2509–2519.
Colman, Deschamps, Naughton,
Rikus, Sulaiman, Puri, Roff, Sun and Embery, 2003. BMRC Atmospheric Model (BAM)
version 3.0: comparison with mean climatology'. BMRC research report in press. Bur. Met., Melbourne, Australia
Schiller, A.,
J. S. Godfrey, P. C. McIntosh, G. Meyers, N. R. Smith, O.Alves, G. Wang and R.
Fiedler, 2002: A New Version of the Australian Community Ocean Model for Seasonal
Climate Prediction. CSIRO Marine Research Report No. 240.
Schiller, A., J.
S. Godfrey, P. McIntosh and G. Meyers, 1997: A global ocean general circulation
model climate variability studies. CSIRO Marine Research Report No. 227. 60 pp.
Smith, N. R., J. E. Blomley and G. Meyers, 1991. A
univariate statistical interpolation scheme for subsurface thermal analyses in
the tropical oceans. Prog. Oceanog., 28, 219-256.
Valcke, S., L. Terray and A. Piacentini, 2000. OASIS
2.4 Ocean Atmospheric Sea Ice Soil users guide, Version 2.4. CERFACS Tech. Rep,
CERFACS TR/CMGC/00-10,85pp.
Wang, G., R. Kleeman, N. Smith, and F. Tseitkin, 2001:
The BMRC coupled general circulation model ENSO forecast system. Mon. Wea. Rev. 130, 975-991.