© Crown copyright 2011 07/0XXX Met Office and the Met Office logo are registered trademarks Met...
1
© Crown copyright 2011 07/0XXX Met Office and the Met Office logo are registered trademarks Met Office FitzRoy Road, Exeter, Devon, EX1 3PB United Kingdom Tel: 01392 885680 Fax: 01392 885681 Email: [email protected] Ensemble Forecasts for an Extreme Rainfall Event Southern France 15-16 June 2010 Ken Mylne 1 , Piers Buchanan 1 , Rob Neal 1 , Naomi Keet 2 , Nicole Girardot 3 , Massimo Milelli 4 , Andrea Montani 5 1 Met Office, Exeter 2 Met Office Work Experience Student 3 Meteo-France, Toulouse 4 ARPA Piemonte, Torino 5 ARPA-SIMC, Bologna Introduction On the 15th and 16th of June 2010, a severe rainfall event caused 20 deaths and a large amount of destruction in SE France, and localised flooding and landslides in NW Italy. The rainfall resulted from a low pressure system which generated a persistent moist unstable SE’ly flow from the Mediterranean. The event was forecast well in advance by forecast models and ensembles, and because of the extreme nature of the forecast, ensemble products were exchanged in real-time between forecast offices in the UK, France and Italy. It thus provides a useful case study for the performance of ensemble prediction systems in heavy convective rain. Rainfall analyses (below) from Meteo-France (left) and ARPA- Piemonte (right) show that up to ~400mm in 24h was observed in France and up to 150mm in 24h was observed in Italy where the village of Coazze recorded 350mm over the event. Total rainfall (mm) in 24h between 12UTC on 15 June to 12UTC on 16 June, 2010 in the North West of Italy. Cars damaged by floods in Draguignan (Image courtesy of bbc.co.uk) Total rainfall (mm) in 24h between 06UTC on 15 June and 06UTC on 16 June analysed by Meteo-France. ECMWF Ensemble Prediction System (EPS) Medium-Range Forecasts Early indications of heavy rainfall were provided in the ECMWF EPS from around 9 June and the risk of a severe event became apparent in the Meteogram for Nice from the forecast from 00UTC on 11 June (below, left) which shows a high probability of 20-40mm/24h and a maximum of over 100mm/24h. Later forecasts progressively enhanced the signal so the forecast initiated at 00UTC on 14 June (below, right) indicated a high probability of 80-120mm with a risk of over 160mm/24h. Considering that this is from a global model ensemble with parameterised convection, these are extreme predictions and provide a strong signal for an exceptional event. Short-range EPS Forecasts Several European Met Services now run EPS using higher resolution regional models with the aim of understanding uncertainty in short-range forecasts of extreme events. We compare here forecasts for the probability of exceeding 50 (left) and 100mm (right) in 24h from the ECMWF EPS, the MOGREPS ensemble of the UK Met Office, the PEARP ensemble of Meteo- France and the COSMO-LEPS ensemble run by ARPA-SIMC (Bologna) and used by ARPA-Piemonte. Not all forecasts are available for exactly the same times but are shown for c. 2-day forecasts which would be available to forecasters at around the same time, with validity between 15/0600 and 16/0600UTC PEARP 06UTC 14/06/10 T+48 50mm/24h MOGREPS 06UTC 14/06/10 T+48 50mm/24h ECMWF 00UTC 14/06/10 T+54 50mm/24h MOGREPS 06UTC 14/06/10 T+48 100mm/24h PEARP 06UTC 14/06/10 T+48 100mm/24h COSMO-LEPS 12UTC 13/06/10 T+72 100mm/24h (or 6h later in the case of COSMO-LEPS). All four systems successfully highlighted a risk of heavy precipitation in the two correct regions, with the highest probabilities for over 100mm/24h in the Torino region of Italy. (MOGREPS gave a higher risk here in the T+54h image, not shown.) MOGREPS gave the highest probabilities in SE France where the greatest damage occurred, although slightly too far SW. ECMWF also gave a good alert in this area, as did the following run of COSMO-LEPS (not shown). Systems gave a remarkably consistent message as the event got closer, with the regions at greatest risk highlighted well from an early stage. There was some fluctuation in forecast probabilities, particularly for 100mm/24h from MOGREPS and PEARP which significantly reduced high probabilities in some runs. (PEARP actually gave a slightly stronger signal in the previous run started at 18UTC on 13 June.) Forecasters in Italy reported that this did not cause problems as they had already issued warnings and had continuing alerts from other systems, but it does suggest some potential benefit from combining probabilities from consecutive forecast cycles to reduce run-to-run variability in automated forecasts – such “lagged” ensemble forecasts are an ongoing research topic. Forecasts from other global ensembles at NCEP (USA) and CMC (Canada) also gave significant probabilities for over 50mm/24h in the correct general area, but lacked detail. Higher resolution global (ECMWF) or regional models gave greater localised detail, although none consistently indicated the location of the heaviest rain in France, so such detail must be interpreted with care. MOGREPS Meteograms for Nice and Torino (below) illustrate a common feature of ensemble forecasts for extreme events. Two days ahead they indicate potential for extreme rainfall up to 160mm/12h, but the most likely amounts are much less: 20-40mm/12h. To ensure that the observed extremes are included in forecasts it is important to take account of the ensemble maxima as well as the most likely amounts. The observed maxima, of course, occurred at slightly different locations from where the point forecasts are available. The complete forecast information is best provided by a combination of charts showing the horizontal distribution and point forecasts showing the probability distribution of rainfall. Conclusions - This case illustrates that ensemble forecasts can provide valuable guidance on some very extreme rainfall events and pinpoint the areas at risk with several days notice. High model resolution is important for regional detail but cannot be relied on at the grid-scale. All these ensembles used parameterised convection, and this illustrates that for dynamically-forced convection this does not prevent the prediction of extreme rainfall amounts. However it should not be expected that these ensembles will perform so well for most convective flash-flood events, for which km-scale convection-permitting models are required.
-
Upload
suzanna-bryant -
Category
Documents
-
view
217 -
download
0
Transcript of © Crown copyright 2011 07/0XXX Met Office and the Met Office logo are registered trademarks Met...
© Crown copyright 2011 07/0XXX Met Office and the Met Office logo are registered trademarks
Met Office FitzRoy Road, Exeter, Devon, EX1 3PB United KingdomTel: 01392 885680 Fax: 01392 885681Email: [email protected]
Ensemble Forecasts for an Extreme Rainfall EventSouthern France 15-16 June 2010 Ken Mylne1, Piers Buchanan1, Rob Neal1, Naomi Keet2, Nicole Girardot3, Massimo Milelli4, Andrea Montani51 Met Office, Exeter 2 Met Office Work Experience Student 3 Meteo-France, Toulouse 4 ARPA Piemonte, Torino 5 ARPA-SIMC, Bologna
Introduction
On the 15th and 16th of June 2010, a severe rainfall event caused 20 deaths and a large amount of destruction in SE France, and localised flooding and landslides in NW Italy.
The rainfall resulted from a low pressure system which generated a persistent moist unstable SE’ly flow from the Mediterranean. The event was forecast well in advance by forecast models and ensembles, and because of the extreme nature of the forecast, ensemble products were exchanged in real-time between forecast offices in the UK, France and Italy. It thus provides a useful case study for the performance of ensemble prediction systems in heavy convective rain.
Rainfall analyses (below) from Meteo-France (left) and ARPA-Piemonte (right) show that up to ~400mm in 24h was observed in France and up to 150mm in 24h was observed in Italy where the village of Coazze recorded 350mm over the event.
Total rainfall (mm) in 24h between 12UTC on 15 June to
12UTC on 16 June, 2010 in the North West of Italy.
Cars damaged by floods in Draguignan (Image courtesy of bbc.co.uk)
Total rainfall (mm) in 24h between 06UTC on 15 June and
06UTC on 16 June analysed by Meteo-France.
ECMWF Ensemble Prediction System (EPS) Medium-Range Forecasts
Early indications of heavy rainfall were provided in the ECMWF EPS from around 9 June and the risk of a severe event became apparent in the Meteogram for Nice from the forecast from 00UTC on 11 June (below, left) which shows a high probability of 20-40mm/24h and a maximum of over 100mm/24h. Later forecasts progressively enhanced the signal so the forecast initiated at 00UTC on 14 June (below, right) indicated a high probability of 80-120mm with a risk of over 160mm/24h. Considering that this is from a global model ensemble with parameterised convection, these are extreme predictions and provide a strong signal for an exceptional event.
Short-range EPS Forecasts
Several European Met Services now run EPS using higher resolution regional models with the aim of understanding uncertainty in short-range forecasts of extreme events. We compare here forecasts for the probability of exceeding 50 (left) and 100mm (right) in 24h from the ECMWF EPS, the MOGREPS ensemble of the UK Met Office, the PEARP ensemble of Meteo-France and the COSMO-LEPS ensemble run by ARPA-SIMC (Bologna) and used by ARPA-Piemonte. Not all forecasts are available for exactly the same times but are shown for c. 2-day forecasts which would be available to forecasters at around the same time, with validity between 15/0600 and 16/0600UTC
PEARP 06UTC 14/06/10 T+48 50mm/24h
MOGREPS 06UTC 14/06/10 T+48 50mm/24h
ECMWF 00UTC 14/06/10 T+54 50mm/24h
MOGREPS 06UTC 14/06/10 T+48 100mm/24h
PEARP 06UTC 14/06/10 T+48 100mm/24h
COSMO-LEPS 12UTC 13/06/10 T+72 100mm/24h
(or 6h later in the case of COSMO-LEPS).
All four systems successfully highlighted a risk of heavy precipitation in the two correct regions, with the highest probabilities for over 100mm/24h in the Torino region of Italy. (MOGREPS gave a higher risk here in the T+54h image, not shown.) MOGREPS gave the highest probabilities in SE France where the greatest damage occurred, although slightly too far SW. ECMWF also gave a good alert in this area, as did the following run of COSMO-LEPS (not shown). Systems gave a remarkably consistent message as the event got closer, with the regions at greatest risk highlighted well from an early stage.
There was some fluctuation in forecast probabilities, particularly for 100mm/24h from MOGREPS and PEARP which significantly reduced high probabilities in some runs. (PEARP actually gave a slightly stronger signal in the previous run started at 18UTC on 13 June.) Forecasters in Italy reported that this did not cause problems as they had already issued warnings and had continuing alerts from other systems, but it does suggest some potential benefit from combining probabilities from consecutive forecast cycles to reduce run-to-run variability in automated forecasts – such “lagged” ensemble forecasts are an ongoing research topic.
Forecasts from other global ensembles at NCEP (USA) and CMC (Canada) also gave significant probabilities for over 50mm/24h in the correct general area, but lacked detail. Higher resolution global (ECMWF) or regional models gave greater localised detail, although none consistently indicated the location of the heaviest rain in France, so such detail must be interpreted with care.
MOGREPS Meteograms for Nice and Torino (below) illustrate a common feature of ensemble forecasts for extreme events. Two days ahead they indicate potential for extreme rainfall up to 160mm/12h, but the most likely amounts are much less: 20-40mm/12h. To ensure that the observed extremes are included in forecasts it is important to take account of the ensemble maxima as well as the most likely amounts. The observed maxima, of course, occurred at slightly different locations from where the point forecasts are available. The complete forecast information is best provided by a combination of charts showing the horizontal distribution and point forecasts showing the probability distribution of rainfall.
Conclusions - This case illustrates that ensemble forecasts can provide valuable guidance on some very extreme rainfall events and pinpoint the areas at risk with several days notice. High model resolution is important for regional detail but cannot be relied on at the grid-scale. All these ensembles used parameterised convection, and this illustrates that for dynamically-forced convection this does not prevent the prediction of extreme rainfall amounts. However it should not be expected that these ensembles will perform so well for most convective flash-flood events, for which km-scale convection-permitting models are required.
