09.11.2014 Views

Low (web) Quality - BALTEX

Low (web) Quality - BALTEX

Low (web) Quality - BALTEX

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

271<br />

Climate change impacts on extreme wind speeds<br />

S.C. Pryor 1,2 , R.J. Barthelmie 1,2 , N.E. Claussen 2 , N.M. Nielsen 2 , E. Kjellström 3 and M. Drews 4<br />

1 Indiana University, Bloomington, IN 47405 USA spryor@indiana.edu<br />

2 Risø DTU National Laboratory for Sustainable Energy, Roskilde, Denmark<br />

3<br />

Rossby Centre, SMHI, Norrköping, Sweden<br />

4 Danish Meteorological Institute, Copenhagen, Denmark<br />

1. Introduction and objectives<br />

Our objective is to quantify potential changes in extreme<br />

wind speeds across northern Europe under a variety of<br />

climate change scenarios. For wind energy applications we<br />

conform to the Wind Turbine design criteria, which define<br />

the suitable class of wind turbines based on the 50-year<br />

return period wind speed (U 50yr ). We determine extreme<br />

winds based on the Gumbel distribution so:<br />

−1<br />

⎡ ⎛ T ⎞⎤<br />

(1)<br />

U T = ln⎢ln⎜<br />

⎟⎥<br />

+ β<br />

α ⎣ ⎝ T −1⎠⎦<br />

U T wind speed for a given return period (T), and α and β are<br />

the distribution parameters.<br />

2. Methods<br />

Two downscaling tools are applied to AOGCM output to<br />

generate higher resolution realizations of surface winds from<br />

which we calculate the extreme wind speeds:<br />

1. Dynamically downscaled time series of grid-cell averaged<br />

wind speeds are used from two applications of Regional<br />

Climate Models (RCMs): (a) Simulations conducted using<br />

the Rossby Centre RCAO lateral boundary conditions from<br />

two General Circulation Models (ECHAM4/OPYC3 and<br />

HadAM3H) and two emission scenarios (SRES A2 and B2)<br />

(Pryor et al. 2005). (b) Simulations conducted using the<br />

HIRHAM RCM and lateral boundary conditions from<br />

ECHAM5/MPI-OM for the A1B SRES. Simulated time<br />

series of annual maximum wind speeds are used to compute<br />

U 50yr using the method of moments to determine α and β:<br />

ln 2<br />

(2)<br />

α =<br />

max<br />

2b1<br />

−U<br />

γ<br />

β = max<br />

(3)<br />

U −<br />

α<br />

1 n i −1<br />

(4)<br />

max<br />

b1<br />

= ∑ Ui<br />

n i=<br />

1n<br />

− 1<br />

The uncertainty on U T is given by:<br />

2<br />

(5)<br />

π 1+<br />

1.14kT<br />

+ 1.10kT<br />

σ ( UT<br />

) = α 6n<br />

Where n is the sample size, the frequency factor (k T ) is:<br />

6 ⎛ ⎡<br />

⎞<br />

⎜<br />

⎛ T ⎞⎤<br />

(6)<br />

k T = − ln − ⎟<br />

⎢ln⎜<br />

⎟⎥<br />

γ<br />

π ⎝ ⎣ ⎝ T −1⎠⎦<br />

⎠<br />

Where γ = Euler’s constant (0.577216)<br />

Assuming a Gaussian distribution of U T , then 95% of all<br />

realizations will lie with ±1.96σ of the mean, and thus σ can<br />

be used to provide 95% confidence intervals on the<br />

estimates of extreme winds with any return period.<br />

2. Empirical downscaling is used to develop site specific<br />

estimates of extreme wind speeds using output from eight<br />

AOGCMs; BCCR-BCM2.0, CGCM3.1, CNRM-CM3,<br />

ECHAM5/MPI-OM, GFDL-CM2.0, GISS-<br />

ModelE20/Russell, IPSL-CM4, and MRI-CGCM2.3.2.<br />

AOGCM output for two historical periods (1982-2000 and<br />

1961-1990) are taken from climate simulations of the<br />

twentieth century. AOGCM output for 2046-2065 and<br />

2081-2100 are from simulations conducted using the A2<br />

SRES. Wind speed observations (1982-2000) at 10-m for<br />

43 stations used to condition the transfer functions are as<br />

in Pryor et al. (2006). This empirical downscaling<br />

technique develops a probability distribution of wind<br />

speeds during a specific time window rather than a time<br />

series of wind speeds. Thus the downscaled parameters are<br />

the scale (A) and shape (k) of the Weibull distribution:<br />

⎡ k ⎤<br />

(7)<br />

⎛ U ⎞<br />

P(<br />

U ) = 1−<br />

exp⎢−<br />

⎜ ⎟ ⎥<br />

⎢<br />

⎣<br />

⎝ A ⎠ ⎥<br />

⎦<br />

This approach is advantageous in the current context<br />

because it avoids a focus on mean conditions,<br />

underestimation of variance, and difficulties associated<br />

with reproducing the time structure of wind speeds.<br />

Downscaled Weibull A and k for each time period,<br />

AOGCM and station are used to compute U 50yr using eq.<br />

(1) and:<br />

⎛<br />

( )<br />

⎟ ⎞<br />

=<br />

k ⎜<br />

− n<br />

1<br />

(8)<br />

1<br />

α ln( ) k<br />

A ⎝ ⎠<br />

1/<br />

β = A( ln( n)<br />

) k<br />

(9)<br />

Where n is the number of independent observations<br />

For the 8 AOGCMs presented and the 43 stations, the<br />

range of downscaled U 50yr for 1961-1990 lie within ±4%<br />

of the mean U 50yr and 95% lie within ±2%, so the<br />

downscaling results for extreme winds from the Weibull A<br />

and k parameters are relatively consistent for the historical<br />

period. Estimates from the future time periods are defined<br />

as being different from the historical period (1961-1990) if<br />

they lie beyond the 95% confidence intervals derived<br />

based on propagation of the uncertainty in A and k.<br />

3. Dynamically downscaled extreme wind<br />

speeds<br />

RCAO simulations for the end of the C21st conducted<br />

using boundary conditions from HadAM3 imply little<br />

change in U 50yr (Table 1). Estimated U 50yr for the future<br />

period (and both SRES) generally lie within the 95%<br />

confidence intervals derived for the control period<br />

simulations (O(4-15%) of the U 50yr in 1961-1990). This is<br />

also the case for simulations conducted using lateral<br />

boundary conditions from ECHAM4/OPYC3, although in<br />

those simulations U 50yr for the future period under either<br />

SRES show increases particularly in the southwest of the<br />

study domain (Figure 1). The choice of SRES appears to<br />

have little influence on the d(U 50yr ) in either set of runs.<br />

Table 1. Fraction of grid cells (in %) that exhibit a<br />

significant increase, decrease or no change for the 4 future<br />

simulations (2071-2100) relative to 1961-1990.<br />

Declines No change Increases<br />

ECHAM4: A2 0.1 73.2 26.7<br />

ECHAM4: B2 0.1 72.9 27.0<br />

HadAM3: A2 6.0 90.1 3.9<br />

HadAM3: B2 1.8 95.8 2.4

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!