Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss
WG4activities
Pierre EckertMeteoSwiss, Geneva
2 COSMO General meeting ¦ Rome, September 2011Pierre.Eckert[at]meteoswiss.ch
Topics
• Guidelines for forecasters, incl. stratified verification (↔ WG5)
• Postprocessing
• Sochi Olympic games PP CORSO
• FIELDEXTRA presentation by Jean-Marie Bettems
3 Automatic weather classifications| COSMO GM 2011
Tanja Weusthoff / Pierre Eckert
New (automatic) weather classifications (MeteoSwiss) The old manual weather classifications are replaced with new
automated weather classifications.
OLDNEW
Alpenwetterstatistik AWS
Perret
Zala-KlassifikationMan
ual,
until
31.
12.2
010 GWT & CAP/PCACA
auto
mat
ed
Sin
ce J
anua
ry 2
011,
Cal
cula
ted
back
until
01.
09.1
957
4 Automatic weather classifications| COSMO GM 2011
Tanja Weusthoff / Pierre Eckert
1. CAP = Cluster Analysis of Principal Component
1. Neue (automatisierte) Wetterlagenklassifikationen
2. GWT = GrossWetterTypes
3. GWTWS = adapted GWT
GWT10, GWT18 and GWT26 based on (1) MSLP and (2) Z500
GWTWS with 11 classes based on GWT8 for Z500, mean wind at 500 hPa and mean MSLP
CAP9, CAP18 and CAP27 based on MSLP
10 classifications are computed every day, based on two different kind of methods
Methods
5 Automatic weather classifications| COSMO GM 2011
Tanja Weusthoff / Pierre Eckert
• For daily computation (since 01.01.2011), use of the operational IFS 12z run from ECMWF; Analysis and forecasts out to 10 days are classified
• Classifications computed back using ECMWF reanalyses 01.09.1957-31.08.2002 ERA40
01.09.2002-31.12.2010 ERA interim
• Domain: alpine region41N - 52N (12pts)
3E - 20E (18pts)
1. Neue (automatisierte) Wetterlagenklassifikationen
Database
6 Verification results at MeteoSwiss in 2011
COSMO GM / WG5 Parallel Session, 05.09.2011
Results for 20103h accumulated precipitation sumsover the domain of the Swiss radar composite
models: COSMO-2 and COSMO-7for all 8 daily forecast runs, precipitation sums from +3 to +6h
observation precipitation estimates of the swiss radar composit
in case of a missing value, the full date will not be evaluated
Neighbourhood verification for precipitation(MeteoSwiss, T. Weusthoff)
7 Verification results at MeteoSwiss in 2011
COSMO GM / WG5 Parallel Session, 05.09.2011
NE
(11x)
N
(18x)
NW
(38x)
SE
(4x)
S
(10x)
SW
(49x)
E
(4x)
W
(56x)
F
(78x)
H
(73x)
L
(25x)
COSMO-7 better COSMO-2 better
differences in Fractions Skill Score for weather-type dependant verif
COSMO-2 minus COSMO-7
YEAR 2010
8 Verification results at MeteoSwiss in 2011
COSMO GM / WG5 Parallel Session, 05.09.2011
Summary neighbourhood verification precipitation in 2010
• The skill of the models varies for different weather types and the differences between COSMO-2 and COSMO-7 varies also:- best skill: Autumn and Spring, south to northwest weather types- greatest difference COSMO-2 minus COSMO-7: Summer and Winter, north- and east types, convective cases
Tanja Weusthoff
9
Conditional verificationConditional verification
Flora Gofa
11
Percentage of weather regimes
0
5
10
15
20
25
30
Z C
Z AC
N-NWC
N-NWAC
N-NE C
N-NEAC
S-SW C
S-SWAC
S-SE C
S-SEAC
Cut-off STNAC
Per
centa
ge
%
1 2 3 4 5 6 7 8 9 10 11 12
For southerly weather situations the cloud
cover is more overestimated….
12
Weather type Dependent Verification
w.r.t. high density rainguage network
Maria Stefania Tesini
13
6-Northerly cyclonic
14
10-Central Mediterranean Low
16
Some considerations on models performances
• At low threshold (e.g. 1 mm/24h) – Cosmo Models perform well in cyclonic situations (CLM,CMT,MC) –
high TS and BIAS ≈1– ECMWF is strongly biased– In anticyclonic situation COSMO-MED and ECMWF are better in terms
of POD but they tend to overestimate the number of events• At higher thresholds (e.g. 5 m/24h and 10 mm/24h)
– COSMO-I7 and I2 miss the anticyclonic situation– still good performance for all models for the cyclonic
situations
17 COSMO General meeting ¦ Rome, September 2011Pierre.Eckert[at]meteoswiss.ch
Postprocessing
• COSMO-MOS
• Diagnostics of turbulence for aviation
• Exchange of postprocessing methods
18
Turbulence index = 1 (light) Turbulence index = 4 (moderate)
Turbulence index = 5 (severe)Colours for measurement height in [m]
Matthias Raschendorfer COSMO Rome 2011DWD
Diagnostics of turbulence for aviation, M. Raschendorfer DWD
19
Matthias Raschendorfer
Distribution between Model- and ARCAS-EDR:
- Prediction-pedictor correlation: 0.44
COSMO Rome 2011DWD
20
Matthias Raschendorfer
Final distribution after successive regression:
- 21 predictors- most effective besides edr: p, dt_tke_(con, sso, hsh)- Successive cubic regression of residuals- Prediction-pedictor correlation: 0.627- Variance reduction: 39.9 %
COSMO Rome 2011DWD
Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss
Accounting for Change:Local wind forecasts from the high-
resolution model COSMO
Vanessa Stauch (MeteoSwiss)
ECAC & EMS, September, 14th 2010COSMO-GM, September 2011, Roma
22 Local wind forecasts | ECAC/EMS 2011, BerlinVanessa Stauch, [email protected]
Spatial verification of wind speed
Model topographyfairly complex
Model performancepretty good
23 Local wind forecasts | ECAC/EMS 2011, BerlinVanessa Stauch, [email protected]
Spatial verification of wind speed
Model topographyfairly complex
Model performancepretty good
Model performanceat some stationsrather poor
24 Local wind forecasts | ECAC/EMS 2011, BerlinVanessa Stauch, [email protected]
Accounting for change
Length of database ~
complexity of statistical correction
temporal flexibility (e.g. when model error changes)
“global MOS”
“KF”
“UMOS” “COSMO-MOS”
„global MOS “: e.g. MOSMIX at DWD, multiple linear regression based on global NWP models (GME and IFS)
“UMOS”: ‘updateable’ MOS of Canadians, weighting when model chsnges
“KF”: Kalman Filter based estimation, online update
+ Sampling for many cases, good discrimination
- A bit inert when model changes
+ insensitive to model changes
- simple error model, poor discrimination of weather condition
Need for models with few parameters
“MOS with reforecasts”
25 Local wind forecasts | ECAC/EMS 2011, BerlinVanessa Stauch, [email protected]
Extended logistic regression
Wilks 2009
Sam
ple
clim
atol
ogy
Wind speed
threshold
ObsFcst
lnp q
1 p q
b0 bix i
i1
n
g q
Add thresholds as predictor, estimate one additional parameter
26 Local wind forecasts | ECAC/EMS 2011, BerlinVanessa Stauch, [email protected]
Results: bias correction for vmax