Transcript Diapositive 1

Reference title: Empirical downscaling
of wind speed probability distribution
Présenté par Tamara Salameh, pour le groupe de lecture
SAMA
2ième année de thèse au Laboratoire de Météorologie
Dynamique/ Ecole Polytechnique
Empirical downscaling of wind
speed probability distribution
S. C. Pryor and J.T.Schoof
Atmospheric ScienceProgram, Department of Geography, Indiana University,
Bloomington, Indiana, USA
R. J. Barthelmie
Departement of wind energy and Atmospheric Physiscs, Riso National laboratory,
Roskilde, Denmark
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110, 2005
• My work: Downscaling of
the wind in the western
Mediterranean basin
• Pryor’s article summary:
A probabilistic approach
to empirically downscale
the wind speed and
energy density
to multiple stations in
northern Europe
21/07/2015
3
why Pryor’s article?
1. Develop novel techniques to empirically
downscaling wind speed and energy density
from GCM’s outputs
2. Empirical downscaling for climate projection
(2071-2100)
3. Evaluation of climate change on downscaled
winds
4. Empirical downscaling results compared
with dynamical downscaling
21/07/2015
4
• Introduction
Plan
• Data
Introduction
Data
• Methods
Methods
• Results
Results
Discussion
• Discussion
21/07/2015
5
•
Near surface wind for: wind energy,
coastal erosion, storm surges, air sea
exchanges
•
Dynamical downscaling:
Plan
Introduction
1. Theoretically preferable
2. No need for data
Data
Methods
•
Empirical downscaling:
Results
1. Computationally more efficient
2. No need for detailed surface morphology
maps.
Discussion
21/07/2015
6
Plan
Introduction
Data
Methods
Results
1. ECHAM4/OPYC3, coupled with
AOGCM (2.8° × 2.8°) (1982-2002 and
1961-1990 with A2 scenario  20712100) 
1. Predictors: mean and standard deviation of
relative vorticity at 500hPa and mean sea level
pressure
2. Boundary conditions for RCM
Discussion
21/07/2015
7
Plan
2. Rossby centre coupled RCM (RCAO)
(0.44° × 0.44°) for 1961-1991 and with
the A2 scenario for 2071-2100
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
+
Introduction
Data
Methods
Results
Discussion
Large scale observations:
NCEP/NCAR Reanalysis (NNR) (2.5° ×
2.5°)
&
Fine scale observations: near surface
wind speeds from National Climatic
Data Center (46 stations)
21/07/2015
8
•
Relative vorticity ζ at 500hPa (mean
and standard deviation)
•
Sea level pressure gradients
Plan
Introduction
Data
Methods:
Predictors
Predictands
Methodology
Evaluated using:
Er '      2 m o r
2
2
m
2
o
Results
Discussion
21/07/2015
9
Plan
Introduction
Data
1. Comparison of the ECHAM4/OPYC3 to
NNR data:
•
•
Methods
Results
Evaluation
of the
AOGCM
and Of the
Correlation of the derived mean sea
level pressure is >0.88
Very good agreement during winter, for
the potential vorticity and overestimation
of the variability during the summer.
(a)
(b)
Emp. Down
(c)
Taylor diagram for
(a) mean sea level
pressure, (b) mean
relative vorticity and
(c) standard
deviation of relative
vorticity.
Discussion
21/07/2015
10
Plan
The two Weibull probability density
function A the scale parameter, and K the
shape parameter
Introduction
Data
Methods:
Predictors
Predictands
Methodology
Results
Discussion
(U is the time series of wind speed observations)
---------------------------------------------------------------------------------------In general:
K is between 1 and 4
The more K is weak, the more the
wind speed distribution is wide.
21/07/2015
11
Cumulative distribution function
Plan
  U k 
P(U )  1  exp    
  A  
Expected energy density
1 3  3
E  A 1  
2
 k
Introduction
Г is the gamma function and ρ is the air density
Data
Methods:
1

U  A1  
k

Predictors
Predictands
Methodology
U 50%  A(ln 2)1/ k
Results
Discussion
U 90%  A( 1. * ln( 0.1))1/ k
For the mean wind speed, the median wind speed and the 90th percentile
respectively.
21/07/2015
12
Plan
Introduction
Example of a regression equation for a Weibull
parameter:
Ai  c1  PG j  c2   j  c3  ( j )
Data
Methods:
Predictors
Predictands
Methodology
i is the station and j is the value of the circulation from
ECHAM4/OPYC3. Ai , PG j ,  j and  ( j ) are
vectors of 12 values (one for each month).
Results
Discussion
21/07/2015
13
Plan
Introduction
2. Wind speed probability distribution
function over Copenhagen, from
observations and from dwnscaling
technique.
Data
Methods
Results
Evaluation
of the
AOGCM
and Of the
Emp. Down
Discussion
21/07/2015
14
1982-2002
Plan
Introduction
Data
Winter
1982-2002
Methods
Results
Evaluation
of the
AOGCM
and Of the
Emp. Down
Discussion
Observed and downscaled energy
density, 90th percentile and mean
wind speed for 1982-2002 at each of
the sites.
21/07/2015
15
1961-1990
and 2071-2100
Plan
Introduction
Data
Winter 1961-1990
and 2071-2100
Methods
Results
Evaluation
of the
Emp. Down
Discussion
Downscaled energy density, 90th
percentile and mean wind speed for
1961-1990 and 2071-2100 at each of
the sites.
21/07/2015
16
Plan
Introduction
Data
Methods
Results
Evaluation
of the
Emp. Down
Discussion
Normalized change of the wind energy density, 90th
percentile and mean wind speed for 1961-1990 and
2071-2100 [(2071-2100 – 1961-1990)/(1961-1990)]
First row is for the entire period, second row is for the
winter time
21/07/2015
17
Plan
Introduction
Data
Methods
Results
Evaluation
of the
Emp. Down
Discussion
Comparison between outputs from the RCAO and from
the empirical downscaling. (a) mean wind speed for
1961-1990 and 2071-2100 (on the all the stations and on
the grid cells containing the station location). (b) mean
wind speed at grid cells containing the station for 19611990 and 2071-2100. Also shown the ED. (c) grid cell
average change in mean wind speed between 19611990 and 2071-2100, projected from ARCAO and ED.
21/07/2015
18
Plan
1. Discrepancies between changes in
mean wind speed from ED and
from DD.
Introduction
Data
Methods
Results
Discussion
The problem is not resolved. Sites that
exhibit largest decrease in mean wind
speeds for 2071-2100 show the smallest
increase in the DD. Sites with the largest
discrepancies are not characterized by lower
quality Weibull fit, but they are located in
general in regions of relatively complex
terrain and land cover heterogeneity.
21/07/2015
19
Plan
Introduction
Data
Methods
Results
Discussion
2. This method can be applied to
other geophysical variables. It could
be applicable too, to any temporal
window for which stable probability
distribution can be derived.
Even for long time series of “stable”
probability distributions of the wind, our
domain present a very complex and
heterogenic terrain  probability distribution
of the wind on some stations canot be
explained by the Weibull variables
21/07/2015
20
Plan
Introduction
Data
3. The study is conducted with an
apriori knowledge of the predictors
in Northern Europe. Other large
scale predictors can be
appropriate in other domains.
Methods
Results
In process on our domain.
Discussion
21/07/2015
21
Plan
4. The study can be sensitive to the
selection of the AOGCM from
which the predictors are derived.
Introduction
Data
Methods
Results
 Analyses of multiple AOGCM
simulations. We started by working
on the ERA40 re-analysis
Discussion
21/07/2015
22
Fin!!
Merci pour votre attention
21/07/2015
23