The observational dataset most RT

1
The observational dataset most RTs are waiting
for the WP5.1 daily high-resolution gridded
datasets
2
Why?

• developing daily high-resolution gridded
  observational datasets for Europe?
• Evaluation of the ENSEMBLES simulation/prediction
  system
• Scenario construction
• Impact assessment
• Analysis of climate extremes

3
Project partners

• KNMI, Lisette Klok Albert Klein Tank
• MeteoSwiss, Evelyn Zenklusen Michael Begert
• University of East Anglia, Malcolm Haylock Phil
  Jones
• University of Oxford, Nynke Hofstra Mark New

4
Overview

• Variables and grid
• Stations and series
• Homogeneity
• Interpolation
• Data availability

5
Variables and grid

• daily observations
• Tmax, Tmin, Tmean, RR, mslp, snow depth
• regular 0.25 degree grid and/or an equal area
  grid
• 1960-2004 or present

Domain
6
Station locations

• Data sources
• ECAD
• EMULATE
• STARDEX
• GSN
• GHCN - daily
• MAP project
• MARS

2033 stations
7
Series

• 1831 precipitation
• 1384 Tmax
• 1388 Tmin
• 1244 Tmean
• 317 mslp
• 180 snow depth

• quality controlled
• updated with SYNOP data

8
Homogeneity results of the absolute test
following the method of Wijngaard et al., 2004
Homogeneous for periods gt 10 years absolute test Homogeneous for periods gt 40 years absolute test
Precipitation 78 35
Temperature 64 20
Air pressure 80 41
Snow depth 89 77
9
Homogeneity results of the relative test

• Vera-QC (Begert et al., in preparation)
• Only complete series
• Period 1960-2000
• Number of break-points detected
• 0(?)
• 1(?)
• 2(?)
• 3(?)
• gt4(?)
• undefined (?)

10
Homogeneity results of the relative test
Homogeneous for periods gt 10 years absolute test Homogeneous for periods gt 40 years absolute test Homogeneous over 1960 2000 relative test
Precipitation 78 35 32
Temperature 64 20 Tmean 23 Tmax 22 Tmin 12
Air pressure 80 41 9
Snow depth 89 77 -
11
Homogeneity results of the relative test
C
C
C
Frequency distribution of shift dimensions for
temperature
12
Interpolation methods

• Natural Neighbour Interpolation
• Angular Distance Weighting
• Thin Plate Splines
• Kriging
• Conditional Interpolation (only rainfall)

13
Selection of best method and validation

• Cross validation
• Remove one station and interpolate to location of
  that station
• Compare results with observed values and
  calculate skill scores (e.g. RMSE, LEPS)
• Comparison with grids from high resolution
  station series
• E.g. UK 55 km rainfall, Switzerland rainfall and
  Norway
• Compare results with gridded datasets and
  calculate skill scores

14
Interpolation results
LEPS skill scores averaged across all methods
15
Uncertainties in the interpolation results

• Still looking for appropriate method to determine
  uncertainties
• Will be in the form of uncertainty bands around
  the interpolated value
• Different uncertainty bands for every grid for
  every day

16
Data availability

• Gridded datasets (September 2007)
• http//www.ensembles-eu.org/ gtgt RT5 site
• Daily station series (if public!)
• http//eca.knmi.nl

17
(No Transcript)