Title: 2nd Workshop on Short Range EPS
1The INTERREG IIIB Project AMPHORE Application
des Méthodologies de Prévisions
Hydrometeorologiques Orientées aux Risques
Environnementaux
Massimo Milelli, Daniele Cane ARPA Piemonte
massimo.milelli_at_arpa.piemonte.it daniele.cane_at_tori
no2006.it
2The AMPHORE partners
- ARPA Piemonte (I)
- ARPA - SIM EMILIA ROMAGNA (I)
- CIMA (I)
- ARPACAL (ARPA Calabria) (I)
- ARPAL (ARPA Liguria) (I)
- DPC (Dipartimento Protezione Civile) (I)
- Université J. Fourier- Grenoble (F)
- Météo-France (F)
- Fundació Bosch i Gimpera (E)
- Universitat de les Illes Balears (E)
3Goals of the project
- Optimization of the alert systems for natural
hazards due to intense precipitation events - Increasing of the synergy and of the cooperation
between the scientific community and the local
administrators - Creation of new formation opportunities for
young people concerning natural hazards
prevention themes
- Improving of the precipitation forecast with the
help of probabilistic techniques
4Multimodel Theory
As suggested by the name, the Multimodel
SuperEnsemble method requires several model
outputs, which are weighted with an adequate set
of weights calculated during the so-called
training period. The simple ensemble method
with bias-corrected or biased data respectively,
is given by
(1) or
(2) The conventional
superensemble forecast (Krishnamurti et. al.,
2000) constructed with bias-corrected data is
given by (3)
5The calculation of the parameters ai is given by
the minimisation of the mean square deviation
by derivation
we obtain a set of N equations, where N is
the number of models involved
We then solve these equations using Gauss-Jordan
method.
6Model setup
- Considered variables
- Total Precipitation
- T2m
- Time interval 3h
- Models involved
- Lami (IT) (7 km)
- Aladin (FR) (10 km)
- Rams (IT) (20 km)
- MM5 (ES) (20 km)
- Bolam (IT) (20 km)
- ECMWF (UK) (40 km)
Independency of the models ! Different numerics,
physics, initialization, domain, assimilation
7Test events definition
- Poor-man Ensemble
- Piedmont, 25 November 2002 (analysis 2002112500,
forecast 36h) - Gard, 8 September 2002 (analysis 2002090800,
forecast 36h) - Montserrat, 9 June 2000 (analysis 2000060900,
forecast 36h) - Reno, 7 November 2003 (analysis 2003110700,
forecast 36h) - SuperEnsemble test cases
- SuperEnsemble
- Training August to November 2004 (skipping the
dates of the events) - Cambrils, 6 September (Spain)
- Calabria, 12 November (Italy)
- Gard, 2 November 2004 (France)
- Piedmont, 15 September (Italy)
8Work performed
- 3 models (LAMs)
- 4 models (3 LAMs ECMWF)
- 6 models (LAMs)
- 8 models (LAMs and ECMWF)
- Different training length 90/180/365 days (only
precipitation) - Different training length 60/120 days (only
temperature) - Fixed training
- Variable training
9The Method
- We evaluate the model performances with respect
to our regional high resolution network. - We applied Multimodel SuperEnsemble technique on
- 2m temperature forecasts, compared with the
measurements of 201 stations, divided in altitude
classes (lt700 m, 700-1500 m, gt1500 m). - Precipitation over warning areas
- 11 warning areas over Piedmont and Aosta
- Valley.
- For each of them mean and maximum
- precipitation values are considered
- Forecast times 12-36 h.
10The verification of the precipitation results was
made using several indices Bias score (frequency
bias) Equitable threat score (Gilbert skill
score) where ROC (Relative operating
characteristic), composed by False alarm
ratio Probability of detection (hit rate)
11Sample results precipitation
8 models, fixed training (180 days), fall 2004
forecast, average over the whole region
Mean values
12Sample results precipitation
8 models, fixed training (180 days), fall 2004
forecast, average over the whole region
Maximum values
13Sample results temperature
4 models, variable training (90 days)
May 2004 forecast, 40 selected stations in the
Olympic Area
14Conclusions
- The Multimodel technique is now implemented and
has been tested for the first time on limited
area models in the Alpine area with
high-resolution non-GTS weather station
measurements - Multimodel SuperEnsemble permits a strong
improvement of precipitation forecasts on warning
areas, both in mean and maximum values - SuperEnsemble BIAS and ETS for precipitation
events are better than those using a simple
Ensemble. The same for Hit Rate and False Alarm
Rate indices (not shown here, trust it !) - The Multimodel improves the temperature
forecasts in high mountains locations, both in
bias and RMSE and its performances are similar to
those from Kalman filter (again not shown here) - The use of different runs of the same models
improves SuperEnsemble performances
15References
- Kalman R. E., Journal of Basic Engineering, 82
(Series D) 35-45, 1960 - Krishnamurti T.N. et al., Science, 285,
1548-1550, 1999 - Krishnamurti, T. N. et al., J. Climate, 13,
4196-4216, 2000 - COSMO Newsletter, 3, 2003
- COSMO Newsletter, 5, 2005
Acknowledgements
We wish to thank the Deutscher Wetterdienst and
MeteoSwiss for providing the models outputs for
the preliminary research work and the our
partners for the realization of the project
AMPHORE.
16You are welcome to the XX Olympic Winter
Games (10-26 February) and to the IX Paralympic
Winter Games (10-19 March) of Torino 2006