Transcript 2. Dynamical downscaling with LAMs

Diversas técnicas de regionalización del clima para el estudio del
cambio climático en Europa
Different methodologies for the study of climate change in
Europe at regional scale
E. SanchezGomez, S. Somot, M. Déqué, J. Najac, J. Beauvier, P. Quintana-Seguí
OUTLINE
Introduction : the need of high resolution data for climate studies over
European-Mediterranean region
1.
Dynamical downscaling with ARPEGE-Climat
Application as a high resolution atmospheric forcing on NEMOMED8
2. Dynamical downscaling with limited area models (LAMs).
Study of the Mediterranean water and heat budgets with an ensemble of LAMs
3. Hybrid (statistical+dynamical) Methodology to study the impact of climate
change on winds in France
To study the Mediterranean climate we need high spatial resolution data
Mediterranean orography
Mistral
Bora
Etesian
Wind over the Gulf of Lyon
IPCC, 350km
High spatial resolution is required
150 km
(ERA40)
50 km
(RCM)
Arpege Climat (strectched grid)
Ruti et al. 2007, JMS
1. Dynamical downscaling of ERA40 with ARPEGE-Climat
•
•
•
•
ARPEGE-Climate
Global Spectral AGCM
Zoom facility available
Mediterranean resolution: 50 km
Can be spectrally driven by
ERA40
ARPERA
•
•
•
•
Spectral nudging
Large spatial scales (> 125
km) driven by ERA40
Small spatial scales are free
Vorticity (6h), Divergence
(24h), Temperature (24h),
Mean Sea Level Pressure
(48h) are relaxed (Humidity
free)
ARPERA run: 1961-2000
dT
 dynamics  physics  T
dt
Nudging term
Large scale
ERA40
ARPEGE
Nudging
coef.
Nudging term
159
T  
n
C
n 1 m   n
m
n
m
n
F
m
m

k
(
A

B
m
n
n )
Cn  
0

if n  63
if n  63
1. Dynamical downscaling of ERA40 with ARPEGE-Climat
Using ARPERA to perform an « hindcast » for the Mediterranean Sea


NEMO-MED8 driven by ARPERA
Simulation hindcast 1960-2001
Détroit de
Gibraltar
Golfe
du
Lion
Mer
Adriatique
Seuil
d’Otrante
Détroit de
Sicile
Mer
Egée
Mer
Ionienne
Buffer
zone
Bassin
Levantin
NEMOMED8 (Madec, 2008)
- New version of OPAMED8 with
some changes in the
physics (partial step, free
surface etc)
- 1/8° horizontal resolution, 43
vertical levels
Density at 1550 m (Med Est)
Mai
1991
Mai
1994
montly volume of
dense water (m3)
Good results: Correct simulation of the EMT (1992-1993)
1. Dynamical downscaling of ERA40 with ARPEGE-Climat
Deep water formation in the Gulf of Lyon
Heat Flux (W/ m2)
1986-1987
Daily maximum mixed layer depth (m)
ARPERA
ERA
Dec
Jan
Feb
Mar
ARPERA heat flux
Quantile-Quantile plot
Apr
MED-ERA
MED-ARPERA
LS91
Dec
ERA heat flux
Jan
Feb
Mar
Apr
Herrmann and Somot, GRL, 2008
2. Dynamical downscaling with LAMs
RCMs: FP6 ENSEMBLES project
Domaine spatial RCM
High resolution database to study the impact
of global warming on Europe and its
uncertainties
12 LAMs
Spatial resolution : 25 km
Two numerical experiments: ERA40 et GCM
Some pararel studies by using LAMs models
 Internal variability of LAMs (Laprise et al., 2008, Luchas-Picher et al.,
2008a,2008b, SanchezGomez et al. 2008) associated to large-scale circulation.
 Ability of LAMs to simulate the water and heat budgets on the
Mediterranean Sea.
2. Dynamical downscaling with LAMs
Internal variability of LAMs
LAMs may provide different solutions within the domain, even forced by the same LBCs.
This variability is often called the internal variability of RCMs and can be determined by the
spread among the members in an ensemble of simulations driven by identical LBCs with
the same LAM.
Spread among 10 members normalized by the transient
variabillity for Z500, for ALADIN-Climat (50km)
The spatial structure de RIV
depends on the large scale
situation (weather regimes)
Composites of RIV according to North
Atlantic Weather Regimes in winter
Blocking
At. Ridge
NAO+
NAO+
2. Dynamical downscaling with LAMs
Ability of LAMs to simulate the Mediterranean water budget
From Mariotti et al. 2002
Mediterranean Sea water content:
atmo
dM/dt = G + BLK + R –D
D
P
GW = net water transport from Gibraltar
E
sea
R
GW
BLK = water input from Black Sea
R = river discharge
M
BLK
Long term mean : dM/dt  0
D  E – P (on annual mean basis)
E – (P + R +BLK)  GW
Water Budget Estimates
WB: 520 – 950 mm/yr (observations)
WB: 391 – 524 mm/yr (NCEP, ERA)
GW: 630 –1135 mm/yr
Castellari et al. 1994, Gilman and Garret 1994,
Bethoux and Gentili 1999, Boukthir and Barnier
2000, Mariotti et al. 2002, Bryden and Kinder
(1991), Bryden (1994) et Garret (1996),
Tsimplis et Bryden (2000), Candela (2001),
Baschek et al. (2001), Garcia de La Fuente et
al. (2007 )
2. Dynamical downscaling with LAMs
Motivation
The water budget influence the water mass characteristics (density, salinity),
and hence the Mediterranean thermohaline circulation.
 Changes in the water budget can impact the Atlantic thermohaline circulation
through the Mediterranean Outflow Water transport at Gibraltar.
 Changes in the water budget may influence the amount of moisture that
flows into Europe, northeast Africa and the Middle East.
 Validation of LAMs. High resolution atmospheric forcing for a ocean model of
the Mediterranean Sea.
MOW
LIW
AW
WMDW
EMDW
2. Dynamical downscaling with LAMs
Area-averaged Water budget on the Mediterranean Sea
DATA
E-P-R
CNRM
1615
DMI
2072
ETHZ
1392
KNMI
1316
METNO
1823
MPI
1352
OURANOS
1128
UCLM
1347
WB1
710
WB2
1593
ERA40
1119
km3/yr
WB1 -> Combination GPCP + OAFlux
WB2 -> HOAPS
Net mass transport estimates of G : (1577 – 2838) km3/yr
The LAMs and WB2 provide water budgest estimates consistent
with G estimates.
SanchezGomez et al., in preparation
2. Dynamical downscaling with LAMs
Water budget changes over the EBRO river catchment
Precipitation
Evaporation
Quintana-Segui et al. 2008
2. Dynamical downscaling with LAMs
Ability of LAMs to simulate the Mediterranean Heat budget
Short wave
absorbed
by water
(SW)
Net surface
emission,
Long wave
(LW)
dHC/dt= GH + HF
Latent
Heat (LH)
Sensible
Heat (SH)
GH:Net heat transport at
Gibraltar
HF:SW+LW+LH+SH
GH
Long terme: dHC/dT=0
SW+LW+LH+SH  GH
Mediterranean Sea
DATA
ERA40
OAFLUX
ISCCP
NOC
HOAPS
ISCCP
LH
84
84
86
89
Heat budget estimates
SH
11
13
7
13
LW
79
74
84
74
SW
162
180
184
180
HB
- 12
+9
+6
+3
GH: (-3 ,– 8.5) W/m2 (mooring based)
HB : +6 W/m2 (S. Josey, observations)
HB : –5 W/m2 (E. Tragou, observations)
HB : -3. 9 W/m2 (S. Somot, coupled model)
2. Dynamical downscaling with LAMs
ERA40 forced runs :
RCM
CNRM
C4I
DMI
ETHZ
ICTP
KNMI
METNO
HC
MPI
SMHI
UCLM
OURA
HB
+10
+4
-21
-36
-36
-10
-15
+19
+3
+14
+11
+12
GCM forced runs :
ARPEGE ECHAM5
RCM
CNRM
HB
-5
C4I
+14
BCM
HadCM3 ECHAM5 ECHAM5
DMI
ETHZ
ICTP
KNMI
METNO
-49 -17
-0.6
-49
Heat gain
Heat loss
Heat budget estimates
GH: (-3 ,– 8.5) W/m2 (mooring based)
HB : +6 W/m2 (S. Josey, observations)
HB : –5 W/m2 (E. Tragou, observations)
HB : -3. 9 W/m2 (S. Somot, coupled model)
HadCM3 ECHAM5
HC
-11
MPI
+9
HadCM3
SMHI
CGCM3
UCLM
OURA
-11
-14
2. Dynamical downscaling with LAMs: Climatic Change
17%,
+208mm/yr
20 years mean
+21%
94mm/yr
+1.94std
90%
-33%
-1.94std
-148mm/yr
-7%, -94mm/yr
Ensemble mean: +4.5%, +57mm/yr
Ensemble mean: -6%, -27mm/yr
+61%
60mm/yr
+22%,
+22 mm/yr
-85%
-82mm/yr
-42%, -40 mm/yr
Ensemble mean: -10%, -9 mm/yr
Ensemble mean: -11%, -11mm/yr
2. Dynamical downscaling with LAMs: Climatic Change
+30%, +376 mm/yr
Ensemble : +13%, +164mm/yrmean
-4%, -47 mm/yr
More statistically significant changes in
hydrological variables from 2050…
+3%, +15 mm/yr
Ensemble mean: -16%, -73 mm/yr
-35%,-160 mm/yr
2. Dynamical downscaling with LAMs: Climatic Change
+445 mm/yr, +55 %
+28 mm/yr, +3%
Ensemble mean: + 236 mm/yr, +29%
Mean (1979-2000) = 793 mm/yr
SanchezGomez and Somot , in preparation
Increase of +55% of fresh water deficit in the Mediterranean Sea (excluiding the
runoff terms), due to precipitation reduction and warming-enchanced evaporation.
Progressive drying of the Mediterranean region (Mariotti et al. 2008)
3. Statistical + dynamical downscaling : 10m wind in France
Statistical downscaling scheme (SDS):
Large-scale circulation
(prédicteurs)
Daily Wind 850hPa
UV850
Statistical Model
Local state
Daily wind at 10m
UV10
Advantages:
 Low computational cost.
 We can obtain a big number of future climate projections from different statistical models,
predictors etc.
 Links between LSC circulation and the local climate offers attractive physical interpretation.
3. Statistical + dynamical downscaling : 10m wind in France
Statistical downscaling scheme (SDS):
Large-scale circulation
(prédicteurs)
Statistical Model
Daily Wind 850hPa
UV850
Local state
Daily wind at 10m
UV10
Disadvantages:
 Stationarity hypothesis : links between the LSC and local variables are conserved in the future
climate (Frias et al. 2008, Najac et al. 2009)
 We can reconstruct distributions only where we have station data.
 In general SDS underestimates the observed trends (Boe et al. 2006, Najac et al. 2008)
3. Statistical + dynamical downscaling : 10m wind in France
UV850 (daily means)
(ERA40 reanalyses – 1974-2002)
Type1
Weather types
Weather types are divided into wind classes,
taking into account a criteria
for inter and intra group variance
Najac el al. 2008
One day is ramdonly chosen
inside each wind class
Mesoscale simulations for selected days
{ U10m, i / i  wind class}
Winter and summer separately
Type6
3. Statistical + dynamical downscaling : 10m wind in France
UV850 (daily means)
(ERA40 reanalyses – 1974-2002)
Impact of climate change
Climate Models
UV850 (Daily mean)
Weather types
Weather types are divided into wind classes,
taking into account a criteria
for inter and intra group variance
Najac el al. 2008
Frequency of occurrence
of the wind classes f i
One day is ramdonly chosen
inside each wind class
Distributions reconstructed by weighting
each simulation by the corresponding
wind class frequency
Mesoscale simulations for selected days
{ U10m, i / i  wind class}
u   f i  u10m ,i
i
Distributions of wind at 10 m
3. Statistical + dynamical downscaling : 10m wind in France
Méso-NH:
- non-hydrostatic mesoscale atmospheric model
- Developped in the Laboratoire d'Aérologie (UPS/CNRS) and the CNRM (CNRS/Météo-France)
- 3 embedded domains:
D1: 36km, 4300 x 4300 km2, 72s
D2: 9 km , 1300 x 1300 km2, 18s
D3: 3 km , 480 x 290 km2,
6s
- 40 vertical levels
- Simulations of 24 hours
(+ initialisation of 6h)
- Boundary and initial conditions
(6h): Réanalyses ERA40
- Simulation of 200 days
Validation : the grid point the closest to
the station point
0
10
100
500
1000
2000
3000
Relief
(m)
3. Statistical + dynamical downscaling : 10m wind in France
Origin of errors (%)
o Errors due to simulation are
large in regions of low relief.
o Errors due to simulation are
large in the high orography
regions, Pyrenees, Massif Central.
%
%
100 90 80 70
60 50
Simulation
50 60 70 80 90 100
Sampling
3. Statistical + dynamical downscaling : 10m wind in France
Multi-model mean changes for the
Wind mean speed at 10 m and anomaly vectors
2046-2065 (Ref 1971-2000)
UV10 DS Hyb
Winter
Summer
UV10 DS Stat
UV10 Simu
3. Conclusions

We need high resolution datasets (models, observations) to study and
interpret Mediterranean climate (over land and sea).

The dynamical dowscaling of ERA40 with ARPEGE-Climat as a high spatial
resolution atmospheric forcing to NEMOMED8 show improvements on the
simulation of circulation in the Mediterranean Sea.

The Mediterranean Sea water budget estimates obtained from an ensemble
of LAMs simulations are correct and coherent with net mass transport at
Gibraltar.

Future climate projections simulated by LAMs show an increase of freshwater
deficit in the Mediterranean Sea (+55%). The low resolution CMIP3 models
predicts +24% (Mariotti et al. 2008).

The statistical+dynamical dowscaling scheme show results consistent with
SDS methodology for changes in winds at 10 m in France. Though it has a high
computational costs, this method provides an homogenous information that
can be used for policymakers.