Title: Some developments at ECMWF during 2005
1Some developments at ECMWFduring 2005
2Overview
- New PBL scheme
- Wavelet Jb in 4D-Var
- TL799L91 in preoperational mode
- TL2047
3Is there any skill in cloud forecasts?
4ECMWF PBL parameterization (- 2004)(similar at
NCAR, NCEP, Arpege, DWD,)
moist mass-flux convection
dry K diffusion PBL
5PBL cloud unification an MK approachMartin
Köhler, ECMWF
- key ingredients
- moist conserved variables
- combined Mass-flux/K-diffusion solver
- cloud variability
- transition between stratocumulus and shallow
convection
6A statistical mass flux framework for organized
updrafts
7improved marine stratocumulus (MK PBL-old)
T511 time10d n140
old CY28R4
new MK PBL
8Peruvian stratocumulus model column vs EPIC
observations
9European Stratus in December 2004
10Wavelet Jb
- Wavelet Jb defines the analysis
change-of-variable and the background cost
function as - Index j denotes scale (wavenumber band).
- The functions ?j are localized spatially, and in
wavenumber. - The Vj are block-diagonal, with vertical
covariance matrix blocks, and one block per
gridpoint. - Convolution with ?j limits the influence of Vj to
a band of wavenumbers and to nearby gridpoints. - The result is a covariance model that allows both
spatial and spectral variation of covariances.
11Wavelet Jb 2D Example
- As an example, we illustrate a 2D covariance
model on the sphere approximately-Gaussian
structure functions with spatially-varying length
scale (300km 1300km). - The rows of the covariance matrix can be shown by
plotting Bx, with x a set of delta functions. - For the wavelet Jb covariance model, we have
12Wavelet Jb 2D Example
- The prescribed length scales are
13Wavelet Jb
Where x is a set of delta functions
14Wavelet Jb
Scale with sj2 at each gridpoint, and for each
scale.
15Wavelet Jb
Convolve again with
16Wavelet Jb
17Wavelet Jb
- Example Horizontal and vertical Vorticity
Correlations.
North America
Equatorial Pacific
18Against analysis
Against analysis
Against observations
Against observations
19L91
20Higher horizontal and vertical resolution
User Meeting June 2005
Anomaly correlation of 500hPa height for northern
hemisphere
T799 L91
T511 L60
Mean over 148 cases from 1 August 2004
Day
21Higher horizontal and vertical resolution
User Meeting June 2005
Anomaly correlation of 500hPa height for Europe
T799 L91
T511 L60
Mean over 148 cases from 1 August 2004
Day
22Higher horizontal and vertical resolution
Anomaly correlation of 500hPa height for southern
hemisphere
T799 L91
T511 L60
Mean over 197 00UTC cases from 1 January 2005
Day
23Statistical significancet test, 500hPa
root-mean-square error
Day 3 Day 5 Day 7
N Hem 0.2 - -
S Hem 0.1 0.1 0.1
Europe 5 5 -
Green denotes T799L91 better
24RMS tropical temperature errors against
radiosondes
300hPa
500hPa
T511 L60
T799 L91
100hPa
50hPa
Day
Day
Mean over 197 00UTC cases from 1 January 2005
25Fit to TEMP observations
26Forecasts of KatrinaD108h
T511 L60
T799 L91
27Hurricane Katrina at T511 and T799
Observed low of 902hPa
T799
959
921
T511
28IFS Model Run-times on IBM p690
Resolution time-step CPUs Wall time for 1-day forecast (minutes) Gflops on IBM p690 of Peak
T511 L60 900 sec 128 7 81 8.4
T799 L91 720 sec 256 16 177 9.1
T1279 L60 450 sec 512 23 354 9.1
T2047 L60 300 sec 768 68 537 9.2
Peak is 7.6 Gflops per PE for IBM p690
29Profile for T2047on IBM p690 (768 CPUs)
Legendre Transforms 17 of total cost of model
Physics 36 of total cost
30Comparison of profilesat different resolutions
31Power spectra of total kinetic energy at T511,
T799 T2047
32Case-study of an event of heavy rainfall
(orographically induced) with severe flooding in
the South of France in September 2002(Observed
rainfall in excess of 500mm/24h)Comparison of
precipitation forecast (24-48h range) at T511,
T799, T1279 and T2047
33 10m-Wind for the South of France on 9 September
2002 at 0UTC
10m-wind from T2047 36h-forecast
34Total precipitation accumulated over 24-48h
FC-range
T799
T511
max 150mm
max 78mm
T2047
T1279
max 321mm
max 197mm
35Rainfall Verification over South of Francefor
T2047 24-48h forecast
36Precipitation skill as a function of model
resolution
2000/07/18 2005/02/15
ACC Z500
ACC PT2
Station location TP
Resolution Europe France
T2047 0.56 0.61
T1279 0.52 0.60
T799 0.45 0.55
T511 (op) 0.38 0.41
90 confidence intervals indicated
37Thank you very much for your attention