Trends in extremes in the ENSEMBLES daily gridded observational datasets for Europe

1. Trends in extremes in the ENSEMBLES daily gridded observational datasets for Europe Nynke Hofstra and Mark New Oxford University Centre for the Environment

2. ENSEMBLES dataset • Daily dataset • Europe • 1950-2006 • Precipitation and mean, minimum and maximum temperature • Four different RCM grids • Kriging interpolation method for anomalies, Thin Plate Splines for monthly totals/means • 95% confidence intervals Haylock et al. Submitted to JGR

3. Introduction • How can this dataset be used for comparison with extremes of RCM output • Required: ‘true’ areal averages

4. Introduction • Several ways to calculate ‘true’ areal averages: • Interpolation of stations within grid (e.g. Huntingford et al. 2003) • Osborn / McSweeney (1997, 2007) method using inter-station correlation • More focused on extremes: • Method of Booij (2002) • Areal Reduction Factors, like Fowler et al. (2005) • But not enough station data available

5. Introduction • Variance of the areal average influenced by amount of stations used • Density of station network differs in time and space

6. Introduction Haylock et al. (submitted JGR) Klok and Klein Tank (submitted Int. J. Climatol.)

7. Objective • Understand what the influence of station density is on the distribution and trends in extremes of gridded data • Focus: • Precipitation • Gamma distribution • Extreme precipitation trends

8. Contents • Experiment • Gamma distribution results • Trends in extremes results • Conclusions so far • Further questions and applications

9. Experiment • Similar setup to interpolation done for ENSEMBLES dataset • One grid with 7 stations in or nearby • 252 stations with 70% or more data available within a 450 km search radius

10. Experiment

11. Experiment

12. Experiment • Calculate ‘true’ areal average of 7 stations • Use Angular Distance Weighting (ADW) interpolation of • 100 random combinations of 4 – 50 stations • all stations • First interpolate to 0.1 degree grid, then average over 0.22 degree grid • ADW uses 10 stations with highest standardised weights and needs minimum 4 stations for the interpolation

13. Experiment • Calculate the parameters of the gamma distribution • Using Thom (1958) maximum likelihood method • Calculate linear trends in extreme indices • Using fclimdex programme

14. α = 0.5 α = 1 α = 2 α = 3 α = 4 β = 0.5 β = 1 β =2 β = 5 β = 10 Gamma distribution McSweeney 2007

15. N=9051 Gamma distribution • How well does the gamma distribution fit the data?

16. Gamma distribution • Dry day distribution and gamma parameters

17. Gamma distribution 95th percentile α=0.6, β=4 α=0.8, β=7

18. Gamma distribution

19. Trends in extremes

20. Trends in extremes

21. Conclusions so far • Gamma scale parameter smaller for interpolated values • Smoothing • Small differences between interpolated and ‘true’ • Small differences using 4 or 50 stations for the interpolation

22. Conclusions so far • Trend in interpolated values larger than in station values • Small differences using 4 or 50 stations for the interpolation • It seems that local trend is picked up even if the amount of stations used for the interpolation is small

23. Further questions and applications • Is the smoothing that we have observed over-smoothing? • What is the distance to the closest station for all combinations of stations? • What happens to the trend of the grid value if only stations with a negative trend are used? • Split the study into two parts: interpolation to 0.1 degree grid and averaging to 0.22 degree grid • Do a similar experiment for minimum and maximum temperature

24. Thank you! Nynke Hofstra Oxford University Centre for the Environment nynke.hofstra@ouce.ox.ac.uk Questions, ideas and remarks very welcome!