Future perspective of
seasonal prediction system
developments at ECMWF
ECMWFにおける季節予報システム開発の展望
Roberto Buizza, Franco Molteni, Magdalena Balmaseda,
Laura Ferranti, Linus Magnusson, Kritian Mogensen,
Tim Stockdale and Frederic Vitart
European Centre for Medium-Range Weather Forecasts
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
1
Outline 概要
1. ECMWF and its forecasting systems: a brief overview
ECMWFとその予測システム：概要
2. Coupled seasonal forecasting at ECMWF: S3 and EUROSIP
ECMWFの結合季節予報：S3とEUROSIP
3. Future developments (S4) and conclusions
将来の開発（System 4）と結論
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
2
1. ECMWF: few figures (2010)

Age of ECMWF: 35 years
設立35年

Employees: 227
職員数：227

Supported by: 33 States
33の支援国

Budget: £38.8 million per annum
年予算50億4千万円
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
3
1. ECMWF objectives ECMWFの目的
Three of the key objectives of ECMWF are:
ECMWFの３つの鍵となる目的：

Operational forecasting up to 15 days ahead (including waves)
15日先までの現業予報（波浪予報も含む）

R & D activities in forecast modelling
予報モデリングにおける研究・現業活動

Operational forecasts for the coming month and season
現業１か月・季節予報
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
4
1. Baseline operational systems (2010)
基本となる現業システム（2010年）
HRES
TL1279L91 (d0-10)
Atmospheric model
EPS
TL639L62 (d0-10)
TL319L62 (d10-15/32)
SF
TL159L62 (m0-7/12)
Atmospheric model
Wave model
Wave model
Ocean model
Real Time Ocean
Analysis ~8 hours
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
Delayed Ocean Analysis
~12 days
5
1. The ECMWF systems
ECMWFのシステム
HRES/DA
# fcs
数
1
Hor resolution
水平解像度
T1279 16 km
Vert lev
鉛直解像度
91 (<0.01 hPa, 75km)
Fc length
対象期間
0-10d
高解像度
データ同化
EPS
アンサン
ブル予報
S3
季節予報
システム
51
41
T639
32 km
T319
64 km
T159
125km
62 (<5 hPa, 35km)
62 (<5 hPa, 35km)
0-10d
10-15/32d
0-13m
Wave
波浪
WAM (0.25°,
28km)
波浪モデル
WAM (0.5°, 56km)
WAM
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
Ocean
海洋
No
No
HOPE
0.3-1.4 deg
L29
HOPE
0.3-1.4 deg
L29
6
1. Numerical Weather Prediction (NWP) models
数値天気予報システム
The ECMWF model is based on
the fluid dynamics laws of
physics that describes how air
masses move, heating and
cooling processes, the water
cycle, the role of radiation, …
The interactions between the
atmosphere and the underlying
land and ocean are also very
important in determining the
weather.
The key requisite for skilful
weather prediction is a very
accurate forecast model.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
7
1. ECMWF comp in ‘78 (Cray 1A) & ‘10 (IBM p6+)
ECMWFのスーパーコンピュータ Cray 1A(1978)＆IBM p6+(2010)
Another key ingredient for skilful
weather prediction is computer
power, that should be enough to
estimate the initial state and to
integrate the model equations in a
reasonable amount of time.
Specification
CPU
1978
2010
Cray
1A
IBM
Power6+
Ratio
1
2x8700
~17000
Clock speed (ns)
12.5
0.21
~0.016
Peak perf (flops)
160 M
200 T
Sust perf (flops)
50 M
20 T
0.4·106
2.5 G
1.2 P
0.5·106
Disk space (bytes)
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
~106
8
1. HRES performance: Z500 over NH & SH
高解像度予報の精度：北半球・南半球の500hPa高度
The combination of improved data-assimilation and forecasting models, the
availability of more/better observations (especially from satellites), and higher
computer power have led to increasingly accurate weather forecasts. Today,
over NH a day-7 single forecast of the upper-air atmospheric flow has the same
accuracy as a day-5 in 1985, and a day-5 as a day-3.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
9
1. HRES performance: ECMWF, UK, US and Japan
高解像度予報の精度(地上気圧): ECMWF、UK、US、日本
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
10
1. ACC(Z500) for JJA 2008 over Europe
ヨーロッパ域の2008年6-8月の500hPa高度 偏差相関係数
Despite all the
improvements of the
past decades,
occasionally (!!)
forecasts can still fail.
Ex 1: JJA 2008. Four
models (ECMWF, UK
Met-Office, NCEP and
MSC Canada) failed
to give accurate
forecasts (ACC>0.6)
of the large scale
flow over Europe
between 5-7 June.
During two further
periods, at least two
models had similar
failures.
Time series curves
500hPa Geopotential
Anomaly correlation forecast
Europe Lat 35.0 to 75.0 Lon -12.5 to 42.5
T+120
ecmwf
met office
ncep
msc
100
80
60
40
20
0
-20
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
JUNE
JULY
AUGUST
2008
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
11
1. Why do forecasts fail?
なぜ予報が外れるのか？
Forecasts can fail because:
 The initial conditions are not accurate enough, e.g. due to poor coverage and/or
observation errors, or errors in the assimilation (initial uncertainties).初期値の誤差
 The model used to assimilate the data and to make the forecast describe only
in an approximate way the true atmospheric phenomena (model uncertainties).
モデルの誤差
 Boundary conditions (albedo, snow cover, vegetation, ..) is poorly simulated
As a further complication, the atmosphere is a chaotic system!
大気がカオス的システムだから！
t=T2
t=T1
t=0
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
12
1. Ensemble Prediction Systems (EPS)
アンサンブル予報システム (EPS)
A complete description of weather
prediction can be stated in terms
of an appropriate probability
density function (PDF).
Temperature
Temperature
fcj
Ensemble prediction based on a
finite number of deterministic
integration appears to be the only
feasible method to predict the PDF
beyond the range of linear growth.
fc0
PDF(t)
reality
Ensemble prediction can be
considered as the practical
application of chaos theory to
weather prediction.
Ensemble methods have been
used to extend the forecast
range from days to weeks,
months and seasons.
PDF(0)
Forecast time
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
13
1. EPS fc of TC Vance: T399 EPS, 18@12+96h
熱帯サイクロン Vance のT399 EPS予測1999年3月18日 96時間予報
An example of
an ensemble of
forecasts from
the ECMWF
EPS, which is
based on 51
forecasts
designed to
simulate initial
and model
uncertainties.
T399 EPS: 18@12 UTC +96h
This plot shows
the EPS t+96h
fcs of TC Vance
making landfall
in Australia on
22/03/99.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
14
1. EPS performance: T850 over NH
EPSの予測精度 北半球850hPa気温
The performance of
the EPS has been
improving
continuously for
upper level fields,
as seen by looking
at the CRPSS for
the t+72h, t+120h
and t+168h
probabilistic
prediction of T850
over NH(verified
against analyses).
Results indicate
predictability
gains of ~ 2
days/decade.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
(Thanks to Martin Janousek)
15
Outline 概要
1. ECMWF and its forecasting systems: a brief overview
ECMWFとその予測システム：概要
2. Coupled seasonal forecasting at ECMWF: S3 and EUROSIP
ECMWFの結合季節予報：S3とEUROSIP
3. Future developments (S4) and conclusions
将来の開発（System 4）と結論
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
16
2. The rationale behind seasonal prediction
季節予報の理論的根拠

Long term predictions are possible to some degree thanks to a number of
components that show variations on long time scales and, to a certain extent,
are predictable. The most important of these components is the ENSO (El
Nino Southern Oscillation) cycle which refers to the coherent, large-scale
fluctuation of ocean temperatures, rainfall, atmospheric circulation, vertical
motion and air pressure across the tropical Pacific.

ENSO‘s fluctuations are quite vast, with the changes in sea-surface
temperatures (SSTs) often affecting not just the whole width of the Pacific
but the other ocean basins too, and the changes in tropical rainfall and winds
spanning a distance of more than one-half the circumference of the earth.
The ENSO cycle is the largest known source of year-to-year climate
variability.

Changes in Pacific sea surface temperature (SST) are not the only cause of
predictable changes in the weather patterns. There are other causes of
seasonal climate variability.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
17
2. The rationale behind seasonal prediction
季節予報の理論的根拠

Unusually warm or cold sea surface temperatures in the tropical Atlantic or
Indian ocean can cause major shifts in seasonal climate in nearby continents.
Other factors that may influence seasonal climate are snow cover and soil
wetness. When snow cover is above average for a given season and region, it
has a greater cooling influence on the air than usual.

Soil wetness, which comes into play most strongly during warm seasons, also
has a cooling influence. All these factors affecting the atmospheric circulation
constitute the basis of long-term predictions.

These are some of the reasons why long-range predictions have been
developed.

Ensemble methods have been applied to build seasonal prediction
systems. At ECMWF, seasonal forecasts have been produced since
1998.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
18
2. ECMWF operational system S3 (2006-todate)
ECMWF 現業システム System 3 (2006-現在)
IFS 31R1
1.1 deg.
62 levels
OASIS-2
HOPE
~ 1.4 deg. lon
1.4/0.3 d. lat.
TESSEL
Initial Con.
4-D variational d.a.
Multivar. O.I.
Gen. of
Perturb.
Ens. Forecasts
System-3
CGCM
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
19
2. S3 Ocean multi-variate O.I. data-assimilation
System 3 多変量 最適内挿法 データ同化

Ocean ICs are the main source of predictability at seasonal time scales. The
correct initialization of the upper ocean thermal structure is considered
instrumental in the prediction of the tropical SST at seasonal timescales with
dynamical models. At the monthly time scales, the prediction of phenomena
such as the MJO requires the correct representation of the ocean-atmosphere
interactions.

A historical ocean reanalysis is required to provide initial conditions for the
calibration of the seasonal forecasts. The a-posteriori calibration of model
output requires an estimate of the model climatology, which is obtained by
performing a series of coupled hindcasts during some historical period
(typically 10-20 years, 25y 11m in the operational S3 hindcasts).

An ensemble of 5 ocean analyses is performed to estimate the uncertainty in
the ocean initial conditions. The ensemble of ocean initial conditions
contributes to the creation of the ensemble of forecasts for the probabilistic
predictions at monthly and seasonal ranges.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
20
2. S3 Ocean ICs: real-time analysis
System 3 海洋初期値: リアルタイム解析
The ocean analysis is performed every 10 days. All obs within a centered 10days window are gathered and quality controlled. In the S3 HOPE-OI system,
in addition to subsurface temperature, the scheme assimilates altimeter
derived sea-level anomalies and salinity data.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
21
2. S3 Ocean real-time analysis
System 3 海洋リアルタイム解析
Every day, a real-time ocean
analysis is produced to initialize
the monthly forecasts.
To avoid degradation of the realtime product, the analysis
always starts from the most
recent Behind Real Time (BRT)
analysis and is then brought
forward to real time every day.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
22
2. The ocean & atmosphere observation systems
海洋と大気観測システム
It is interesting to contrast the ocean and atmosphere observing systems.
Everyday, ECMWF receives ~ 5k obs of the ocean state and ~100M
atmospheric obs (~90% from sat). The atmospheric 12h 4D-Var DA system
uses ~10% of these data (~9M obs) to compute the analysis.
Synop/ship (~31000)
AMSU-A (~576000)
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
23
2. 1-year ENSO outlook with S3: 2009-2010
System 3 による 1年間のENSO見通し: 2009-2010
The tropics is the area where seasonal predictability is higher, controlled
mainly by ENSO, a coupled ocean-atmosphere phenomenon centred over the
tropical Pacific. The ECMWF S3 13m integrations (generated every quarter,
based on an 11-member ensemble) gave very good predictions of the most
recent warming conditions of 2009.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
24
2. S3 2mT climagram forecasts
System 3 2m気温 climagram 予報
One S3 product is the 2mT climagram (top-right), which shows the distribution
of the S3 forecast (purple), the model climate (grey) and the analysis. A red dot
indicates the observed anomaly.
Japan/Korea
Central tropical Pacific
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
25
2. S3 2mT climagram over central-tropical Pacific
熱帯太平洋中部のSystem3 2m気温 climagram
One S3 product is the 2mT climagram (topright), which shows the distribution of the
S3 forecast (purple), the model climate
(grey) and the analysis. A red dot indicates
the observed anomaly.
The lower plots show the ACC (left) of the
ensemble-mean forecast and the ROCA of
the PR(2mT in upper tercile).
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
26
2. S3 2mT climagram over Japan/Korea
日本/韓国域の System 3 2m気温 climagram
One S3 product is the 2mT climagram (topright), which shows the distribution of the
S3 forecast (purple), the model climate
(grey) and the analysis. A red dot indicates
the observed anomaly.
The lower plots show the ACC (left) of the
ensemble-mean forecast and the ROCA of
the PR(2mT in upper tercile).
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
27
2. S3 2mT-anomaly forecasts: 1 Nov ‘10 > DJF
System 3 2m気温偏差予報:2010年 11月1日初期値 12月～2月
Two examples of S3 forecast products: ensemble-mean (left) and probabilistic
forecasts (right) of 2mT anomalies started on 1 Nov 2010 and valid for D10JF11.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
28
2. S3 2mT-anomaly fcs from 1 Dec: ACC
System 3 2m気温偏差予報の偏差相関係数（12月1日初期値）
These plots show a measure of the accuracy of seasonal probabilistic forecasts,
the area under the Relative Operating Characteristics (ROCA) of the probabilistic
forecasts of the 2mT anomaly being below the lower tercile forecasts with 1 Dec
starting date. The ROCA has been evaluated considering 25-years (1985-2005)
11-member hindcasts.
The left panel shows the ROCA of the 2-4m average forecast, and the right
panel the ROCA of the 5-7m average forecast.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
29
2. S3 Accumulated Cyclone Energy forecast
System 3 累積サイクロンエネルギー予報
One of the operational seasonal forecasts is ACE (an index of storm activity
defined by the sum of the square of the estimated maximum sustained velocity
of every active tropical storm). The left panel shows the ACE forecast issued on
1 June 2010 for JASOND10.
The right panel shows that the accuracy of this product over WPAC (CC 52%).
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
30
2. S3 tropical storm frequency forecast over WPAC
北西太平洋におけるSystem 3 熱帯擾乱頻度予報
Over WPAC during
strong El Nino TSs
tend to avoid the
coast by curving to
the North, while
during la Nina TSs are
more likely to reach
the Philippines.
The top panels show
the observed TSs
during the 1997 El
Nino year and the
1999 la Nina year.
The bottom panels
show the probability
of tropical storm
density simulated by
the seasonal forecast
initiated in June.
1997
1999
+El Niño
La Niña
Seasonal forecasts started in June 1997 and June 1999
<-1.20
-1.20
-0.90
120°E
-0.60
-0.30
140°E
0.30
0.60
1.90
> 1.20
<-1.20
160°E
-1.20
-0.90
120°E
-0.60
-0.30
140°E
0.30
0.60
1.90
> 1.20
160°E
40°N
40°N
40°N
40°N
30°N
30°N
30°N
30°N
20°N
20°N
20°N
20°N
10°N
10°N
10°N
10°N
120°E
140°E
160°E
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
120°E
140°E
160°E
31
2. S3 tropical storm frequency forecast over WPAC
北西太平洋におけるSystem 3 熱帯擾乱頻度予報
2008 and 2009
showed a non-typical
behaviour. Despite
mild El Nino
conditions, a good
number of TSs
reached the
Philippines coast
because of the warm
SST over WPAC.
The seasonal
system predictions
of TSs match the
observed trucks
showing high
probability of TSs
over this region.
El Niño
2009
2008
Seasonal forecasts started in June 2008 and June 2009
<-1.20
-1.20
-0.90
120°E
-0.60
-0.30
140°E
0.30
0.60
1.90
> 1.20
<-1.20
160°E
-1.20
-0.90
120°E
-0.60
-0.30
140°E
0.30
0.60
1.90
> 1.20
160°E
40°N
40°N
40°N
40°N
30°N
30°N
30°N
30°N
20°N
20°N
20°N
20°N
10°N
10°N
10°N
10°N
120°E
140°E
160°E
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
120°E
140°E
160°E
32
2. S3 tropical storm frequency forecast over WPAC
北西太平洋におけるSystem 3 熱帯擾乱頻度予報
2009 showed a non-typical
behaviour. Despite mild El Nino
conditions, a good number of TSs
reached the Philippines coast
because of the warm SST over
WPAC.
ECMWF S3 ocean analysis: Anomaly
19970916 (30 days mean)
Sea Surface Temperature
respect to
Contour interval = 1 deg C
1981-2005 climatology
Interpolated in y
6
-1
4
3
O
50 N
2.5
1
2
ECMWF S3 ocean analysis: Anomaly
Latitude
1.5
20090916 (30 days mean)
Sea Surface Temperature
respect to
Contour interval = 1 deg C
1981-2005 climatology
1
1
-1
0
32
4
O
1
0.5
-0.5
-1
-1.5
-2
-2.5
-4
2
Interpolated in y
6
O
50 S
4
1
3
-6
O
50 N
2.5
-1
O
-1
2
50 E
O
100 E
Latitude
1
O
O
160 W
O
110 W
O
60 W
O
10 W
Longitude
1.5
0
O
150 E
MAGICS 6.11 bee17 - em os Wed Mar 21 14:30:25 2007
0.5
-0.5
-1
-1.5
-2
-2.5
-4
1
O
50 S
-6
O
50 E
O
100 E
O
150 E
O
160 W
O
110 W
O
60 W
O
10 W
Longitude
MAGICS 6.11 drn07 - em os Tue Oct 13 10:09:27 2009
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
33
2. EUROSIP: ECMWF, UKMO and MeteoFrance
マルチモデル季節予報システム EUROSIP: ECMWF、イギリス気象局、フランス気象局

Compared to medium-range forecasting, the predicted signals are much
smaller and the time over which model errors accumulate are longer, so the
importance of model error is much, much higher. One way to better simulate
model uncertainty is to create multi-model forecasting systems by combining
the output from several models, rather than taking just one model.

The fundamental reason for the benefit of a multi-model approach is that all
models have errors that have a different impact on a given forecast. By
averaging across a number of models, a significant part of the model error
can be reduced. Unfortunately some errors tend to be common between
models, so averaging is not a panacea nor a replacement for model
development.

One of these multi-model systems is EUROSIP, a multi-model seasonal
forecasting system consisting of three independent coupled systems: ECMWF,
Met Office and Météo-France (all integrated in a common framework).
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
34
2. 6m ENSO outlook with S3 & EUROSIP: 1 Dec 08
System 3 による 6か月ENSO見通し:2008年12月1日初期値
ECMWF (left) and EUROSIP 3-system SST anomaly forecasts for NINO3.4 area
issued on 1 Dec 2008 and valid for 6 months.
The observed SST anomaly lies at the edge of the ECMWF plum. Compared to
the single ECMWF plum, the EUROSIP plum has a larger dispersion and gives
a higher probability of cold conditions.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
35
2. 6m ENSO outlook with S3 & EUROSIP: 1 Oct 09
System 3 による 6か月ENSO見通し:2009年10月1日初期値
ECMWF (left) and EUROSIP 3-system SST anomaly forecasts for NINO3.4 area
issued on 1 Oct 2009 and valid for 6 months.
In this case the observed SST anomaly lies outside the edge of the ECMWF
plum. Compared to the single ECMWF plum, the EUROSIP plum has a larger
dispersion and includes the observed SST within the forecast range.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
36
Outline 概要
1. ECMWF and its forecasting systems: a brief overview
ECMWFとその予測システム：概要
2. Coupled seasonal forecasting at ECMWF: S3 and EUROSIP
ECMWFの結合季節予報：S3とEUROSIP
3. Future developments (S4) and conclusions
将来の開発（System 4）と結論
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
37
3. The new seasonal forecasting system S4
新しい季節予報システム（System 4）
IFS 36R4
0.7/1.1 deg.
91 levels
OASIS-3
NEMO
~ 1. deg. lon
1./0.3 d. lat.
H-TESSEL
Initial Con.
4-D variational d.a.
3-D v.d.a. (NEMOVAR)
Gen. of
Perturb.
Ens. Forecasts
System-4
CGCM
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
38
3. System 4: main features
System 4: 主な特徴

New ocean model: NEMO v. 3.0 + 3.1 coupling interface
 ORCA-1 configuration (~1-deg. resol., ~0.3 lat. near the equator)
 42 vertical levels, 20 levels with z < 300 m

Variational ocean data assimilation (NEMOVAR)





3-D var with inner and outer loop
Collaboration with CERFACS, UK Met Office, INRIA
First re-analysis (1957-2009), no assim. of sea-level anomalies
Second re-analysis and real-time system including SLA
IFS model cycle: 36r4 (currently operational)
 New physics package, including HTESSEL land-surface scheme, snow model (with
EC-Earth), new land surface initialization

Prescribed sea-ice concentration with sampling from recent years
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
39
3. Ocean Re-Analysis with NEMO at ECMWF
海洋モデルNEMOによる海洋再解析

Using NEMO/NEMOVAR

Model configuration: ORCA1, smooth coastlines, closed Caspian Sea.

Forced by ERA40 (until 1989) + ERA Interim (after 1989)

Assimilates Temperature/Salinity from EN3 and altimeter data

Strong relaxation to SST (OI_v2)

Online bias correction scheme

First ensemble reanalysis (1957-2009) completed (COMBINE project):
 5 ensemble members (perturbations to wind, initial conditions,
observation coverage)
 Corrected XBT

Second re-analysis that assimilates also altimeter data (1957-2010) just
completed (and to be continued for S4)
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
40
3. NEMOVAR re-an: verif. against altimeter data
NEMOVAR再解析: 海面高度データに対する評価
NEMOVAR improves the inter-annual variability of the ocean re-analysis, as
measured by the anomaly correlation of the sea level with the altimeter data.
Some problems remain in the Eastern North Atlantic, where assimilation
deteriorates results (under investigation).
NEMO (no obs)
NEMOVAR: T+S+Alti
(1993-2008)( 1993-2008 )
(1): faz9
RMSEcorrel
(1):sossheig
fe5x sossheig
40N
40N 40N
40N
20N
20N 20N
20N
0
0
20S
20S 20S
40S
40S 40S
60S
60N
Latitude
0
Latitude
60N 60N
Lati tude
Lati tude
correl (1): faz9 correl ( 1993-2008 )
60N
20S
40S
60S 60S
100E
160W
Longitude
(ndim): Min= -0.43, Max= 0.99, Int= 0.02
60W
0.40 0.44 0.48 0.52 0.56 0.60 0.64 0.68 0.72 0.76 0.80 0.84 0.88 0.92 0.96 1.00
0
60S
100E 100E
160W 160W
60W 60W
Longitude
Longitude
0.011.00, Int= 0.02
0.16, Int=
Max=-0.37,
Min= 0.01,Min=
(m): (ndim):
Max=
0.02 0.40 0.44 0.48 0.04
0.08 0.84 0.88 0.92 0.96
0.10 1.00
0.52 0.56 0.60 0.640.060.68 0.72 0.76 0.80
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
0.
41
3. Impact of ocean DA on SST forecasts
海面水温予報に対する海洋データ同化解析の効果
0.8
0.8
0.6
0.6
Rms error
Rms error
Rms error
0.6
0.8
0.4
0.4
0.4
Seasonal forecast experiments (20years, 11-member ensembles,
cy36r4+NEMO) indicate that ocean data assimilation with NEMOVAR improves
NSTRATL 034a rms errors
NSTRPAC 034a rms errors
SSTRATL 034a rms errors
the SST forecast skill at different lead times and different regions.
0.2
0.2
0
0
0
1
2
3
4
5
6
7
0.2
0
1
2
Fcast feik
1
Fcast ff5e
3
4
5
6
0
7
0
Persistence
Fcast feik
Ensemble sd
1
EQ2 034a anomaly correlation
Fcast ff5e
Persistence
Ensemble sd
Fcast feik
0.9
0.6
0.9
0.6
0.8
0.4
0
1
2
3
4
7
6
7
0
0.5
0
0.4
1
0
1
3
4
5
6
7
2
3
4
5
NSTRPAC 034a anomaly correlation
NSTRATL 034a anomaly correlation
6
7
1
2
5
6
7
3
4
5
6
7
0.6
0.5
0
1
2
3
4
5
6
7
0.4
0
1
2
Forecast time (months)
Forecast time (months)
MAGICS 6.12 w indmill - neh Fri Sep 3 10:59:05 2010
0
Anomaly correlation
Anomaly correlation
Anomaly correlation
0.4
4
0.4
0.7
0.5
3
7
0.8
0.6
0.5
6
0.9
0.7
0.6
5
MAGICS 6.12
1w indmill - neh Fri Sep 3 10:59:05 2010
0.8
0.7
4
wrt NCEP adjusted OIv2 1971-2000 climatology
1
0.8
3
SSTRATL 034a anomaly correlation
0.9
2
2
Forecast time (months)
MAGICS 6.12 w indmill - neh Fri Sep 3 10:59:05 2010
0.9
1
1
wrt NCEP adjusted OIv2 1971-2000 climatology
1
0
0
0.5 0
Forecast time (months)
Forecast time (months)
wrt NCEP adjusted OIv2 1971-2000 climatology
0.4
2
Forecast time (months)
MAGICS 6.12 w indmill - neh Fri Sep 3 10:59:05 2010
Ensemble sd
0.2
0.6
Forecast time (months)
5
Persistence
0.7
0.2
0.6
6
Fcast ff5e
0.4
0.7
5
7
0.8
0.4
0.2
0.6
6
Anomaly correlation
Rms error
Anomaly correlation
Rms error
0.9
0.6
Rms error
Anomaly correlation
1
0.8
0.7
5
wrt NCEP adjusted OIv2 1971-2000 climatology
1
0.8
0.8
4
EQIND 034a anomaly correlation
wrt NCEP adjusted OIv2 1971-2000 climatology
0.4
3
1
0.81
ASSIM
1
2
3
4
CONTROL
Forecast time (months)
2
64Forecast
start dates from
19930201
to 20081101
time
(months)
Ensemble sizes are 5 (feik) and 5 (ff5e)
ATL3 034a anomaly correlation
wrt NCEP adjusted OIv2 1971-2000 climatology
0
0.5
0
1
Forecast
(months)
64 start
dates fromtime
19930201
to 20081101
Ensemble sizes are 5 (feik) and 5 (ff5e)
Forecast
(months)
64 start
dates from time
19930201
to 20081101
Ensemble sizes are 5 (feik) and 5 (ff5e)
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
MAGICS 6.12 w indmill - neh Fri Sep 3 10:59:05 2010
3
4
5
6
7
Forecast time (months)
MAGICS 6.12 w indmill - neh Fri Sep 3 10:59:05 2010
42
3. Future developments and conclusions
将来の開発と結論

Ensemble-based probabilistic systems provide more complete information than
single forecasts. They can be used in weather risk management to assess the
probability of occurrence of events that can cause severe losses.

ECMWF has been producing ensemble-based forecasts since 1992. Long-range
seasonal forecasts are now based on 41 coupled integrations of the IFS-cy31r1
T159L62 atmospheric model and the HOPA 1.0-0.3 degree ocean model.

ECMWF will introduce the new seasonal forecasting system 4 (S4) in 2011. S4
is based on a more accurate, higher-resolution (T255L91) atmospheric model
and a better ocean model (NEMO 1.0-0.3 deg res). The ocean analysis will also
change from OI to a 3D-Var data assimilation system (NEMOVAR).

Further improvements of ECMWF seasonal forecasts are expected from better
model error simulation schemes, the inclusion of dynamical sea-ice and
mixed-layer (see e.g. work by Y Takaya, JMA) models, and higher resolution
ocean models. The possibility to merge the 15/32d EPS and the 7/14m SF
systems into a Seamless Probabilistic System is also been considered.
JMA WS (9 Dec 2010) - Roberto Buizza et al : Seasonal prediction at ECMWF
43