Statistical Analysis of Extreme Wind in Regional Climate Model Simulations

1. Statistical Analysis of Extreme Wind in Regional Climate Model Simulations EMS 2013Stephen Outten

2. Overview • Motivation • Statistical methods • Extreme winds in RCMs • Practical application

3. Hardanger Bridge Photo from: Norwegian Public Roads Administration

4. Current Procedure • Obtain observations for short time series at bridge and long time series at lighthouse • Relate short and long term time series to create long series at bridge and obtain distribution Hardanger bridge • Derive return events with associated uncertainties at bridge from current distribution ??? • Derive return events with associated uncertainties at bridge from current distribution Utsira

5. RCM Data • ENSEMBLES Project • Regional downscaling of IPCC models • Multiple RCMs employed • 25 km horizontal resolution • European domain • Uniform grid • Future A1B scenario • 4 downscalings • 2 GCMs x 2 RCMs • Maximum daily wind speeds

6. Statistical Methods

7. Extreme Value Theory • Theorem 1 • The maxima of multiple samples of data converge to a Generalised Extreme Value (GEV) distribution • Theorem 2 • The exceedances over a suitably chosen threshold converge to a Generalised Pareto Distribution (GPD)

8. BCM/HIRHAM5 : Bergen : 1961-1990 GEV GPD

9. BCM/HIRHAM5 : Bergen : 50 year return GEV GPD R50 : 19.58 ms-1 R50 : 19.61 ms-1 CI99% : 18.12 ms-1 26.10 ms-1 CI99% : 18.05 ms-1 30.07 ms-1

10. Parameter Space for Bergen GEV GPD Likelihood contours from inside to outside: 90%, 95%, 98%, and 99%

11. GEV : Parameter Sensitivity R50: 19.61 ms-1 R50: 28.60 ms-1

12. Generalised Extreme Value Family Generalised Extreme Value Distribution σ k μ (reversed) Weibull Gumbel Gumbel σμ Fréchet k<0 k=0 k>0

13. Likelihood Ratio Test Compares the fit of two models, one of which is a special case of the other • Procedure: • Fit both models to the data • Calculate test statistic from log-likelihoods • Use a Chi-squared to determine if fits are significantly different

14. Applying Approach to Bergen GEV GPD Confidence Interval at 99% level GEV : 18.05 ms-1 to 30.07 ms-1 GPD : 18.12 ms-1 to 26.10 ms-1 Gumbel : 18.33 ms-1 to 22.86 ms-1 Gumbel

15. Proposed Procedure • Obtain observations for short time series at bridge and long time series at lighthouse • Relate short and long term time series to create long series at bridge and obtain distribution • Use statistical tests to select the appropriate distribution to minimise the uncertainty • Derive return events with associated uncertainties at bridge from current distribution

16. Extreme Winds in Regional Climate Models

17. Model Resolution Hardanger bridge 1.3 km bridge 2-3 km wide fjord 25 km resolution Utsira

18. Proposed Procedure • Obtain observations for short time series at bridge and long time series at lighthouse • Relate short and long term time series to create long series at bridge and obtain distribution • Use statistical tests to select the appropriate distribution to minimise the uncertainty • Obtain regional climate model data at lighthouse location for reference and future periods • Relate the future distribution at the lighthouse to the bridge • Derive return events with associated uncertainties at bridge from current distribution

19. Future Change : Bergen ○ : BCM-HIRHAM5 Current ∗ : BCM-HIRHAM5 Future ○ : ECHAM5-HIRHAM5 Current ∗ : ECHAM5-HIRHAM5 Future ○ : BCM-RCA3 Current ∗ : BCM-RCA3 Future ○ : ECHAM5-RCA3 Current ∗ : ECHAM5-RCA3 Future

20. Models and Extreme Winds Source: Knutson et al. 2008

21. Models at Utsira BCM/HIRHAM5 BCM/RCA3

22. Proposed Procedure • Obtain observations for short time series at bridge and long time series at lighthouse • Relate short and long term time series to create long series at bridge and obtain distribution • Use statistical tests to select the appropriate distribution to minimise the uncertainty • Obtain regional climate model data at lighthouse location for reference and future periods • Combine projected change from models with observations from lighthouse to create future wind speed distribution at lighthouse * • Relate the future distribution at the lighthouse to the bridge • Derive return events with associated uncertainties at bridge from current and future distributions * Holland G. and Suzuki-Parker A, Journal of Climate, (submitted)

23. Practical Application

24. Proposed Approach • Obtain observations for short time series at bridge and long time series at lighthouse • Relate short and long term time series to create long series at bridge and obtain distribution • Use statistical tests to select the appropriate distribution to minimise the uncertainty • Obtain regional climate model data at lighthouse location for reference and future periods • Combine projected change from models with observations from lighthouse to create future wind speed distribution at lighthouse * • Relate the future distribution at the lighthouse to the bridge • Derive return events with associated uncertainties at bridge from current and future distributions * Holland G. and Suzuki-Parker A, Journal of Climate, (submitted)

25. Application to Utsira Lighthouse WS50 = 37.9 ms-1 WS50 = 38.2 ms-1 WS50 = 38.2 ms-1 WS50 = 38.6 ms-1 WS50 = 38.1 ms-1 Instanes A. and Outten S, Journal of Bridge Engineering, (to be submitted)

26. Summary • Developed method for including projected changes in extreme winds into the design process • Future changes in extreme winds are generally smaller than the uncertainty involved in estimating the extreme event • Inter-model spread is the largest source of uncertainty • Vital to assess uncertainties in estimates of extreme events

27. Thank You

28. Extra Slides More statistics and Winds over Europe

29. BCM/HIRHAM5 : 50 year return Reference (1961-1990) Future (2070-2099) Outten & Esau, Atmos. Chem. Phys., 2013

30. DMI/BCM : Future Change Outten & Esau, Atmos. Chem. Phys., 2013

31. BCM/HIRHAM5: Uncertainty-Future Change Outten & Esau, Atmos. Chem. Phys., 2013

32. GCM BCM ECHAM5 HIRHAM5 RCM RCA3

33. GCM BCM ECHAM5 HIRHAM5 RCM RCA3