qpf verification of the 4 model versions at 7 km res. (cosmo-i7, cosmo-7, cosmo-eu, cosmo-me) with...
TRANSCRIPT
QPF verification of the 4 model versions at 7 km res. (COSMO-I7, COSMO-7, COSMO-EU, COSMO-ME) with the 2 model versions at 2.8 km res. (COSMO-I2, COSMO-IT)
Specifications:
• Dataset: high resolution network of rain gauges coming from COSMO dataset and Civil Protection Department 1300 stations
• Method: 24h/6h averaged cumulated precipitation value over 90 meteo-hydrological basins
•Model selection: run 00UTC, D+1
Precipitation verification comparison in 2008/2009 among the several COSMO-Model versions (Elena Oberto, Massimo Milelli - ARPA Piemonte)
The aims
• Long period verification (seasonal trend) (from djf’04 to mam’09)
1. Statistical indices for low thres (0.2mm/24h)
2. Statistical indices for high thres (20mm/24h)
• Verification ovest last year (200806-200905)
1. Cosmo-7/Cosmo-EU comparison
2. Cosmo-I7/Cosmo-ME comparison
3. Driving model comparison: ecmwf/Cosmo-I7/Cosmo-I2
4. Driving model comparison: ecmwf/Cosmo-ME/Cosmo-IT
5. Diurnal cycle for all the model
6. Spatial error distribution (latest season: MAM’09)
% correctly forecasted not-
events (specificity)
Seasonal trend (low thres)
• Bias reduction trend
• Seasonal cycle: big peak during summertime
• Biggest overestimation peak for cosmo-I7
• Underestimation for cosmo-7 during latest seasons
Seasonal trend (low thres)
• Stable trend/slightly worsening in time
• Best performance during spring/summertime
• Worsening for Cosmo-7 during latest seasons
Seasonal trend (low thres)
• False alarm number reduction esp. in wintertime
• Worse performance during summertime, esp. for Cosmo-I7
Seasonal trend (low thres)
• Slightly improvement trend
• Seasonal cycle: better during moist seasons, worse during dry seasons
Seasonal trend (high thres)
• Bias reduction trend, at least during last year
• Seasonal cycle: big peak during spring-summertime (convective period) seems to disappear during last summer (… why?)
• General good performance during last year
• Pronounced underestimation for Cosmo-7 during last seasons
Seasonal trend (high thres)
• Slightly improvement in time
• Worse performance during summertime (except 2007)
• Worsening for Cosmo-7 during last seasons
Seasonal trend (high thres)
• False alarm number reduction
• Worse performance during summertime
Seasonal trend (high thres)
• Slightly improvement trend
• Worse performance during summertime (except 2007)
Last year verif.: Cosmo-7/Cosmo-EU vs. thres
• Good performance for Cosmo-EU
• Underestimation for Cosmo-7
• More correctly forecasted non-events for Cosmo-7
Last year verif.: Cosmo-7/Cosmo-EU fixed thres, seasonal
• Good performance for Cosmo-EU
• Underestimation for Cosmo-7
• Positive trend for both the models
Last year verif.: Cosmo-I7/Cosmo-ME vs. thres
• Similar (good) performance: slightly better for Cosmo-ME
Last year verif.: Cosmo-I7/Cosmo-ME fixed thres, seasonal
• Similar (good) performance: slightly better for Cosmo-ME
• Positive trend for both the models
Driving model comp: Ecmwf/Cosmo-I7/Cosmo-I2 vs. thres
• Big gap between Ecmwf and Cosmo-model
• Cosmo-I7 slightly overestimation; Cosmo-I2 underestimation; Ecmwf overest. for low thres/ underest. for high thres.
• I7 equivalent or slightly better then I2 (even if less false alarm for I2)
Driving model comp: ecmwf/Cosmo-I7/Cosmo-I2 fixed thres, seasonal
• Big gap between Ecmwf and Cosmo-model
• Positive trend for both I7 and I2
• I2 Underestimation tendency
• I7 is generally better
Driving model comp: Ecmwf/Cosmo-ME/Cosmo-IT vs. thres
• Big gap between Ecmwf and Cosmo-model
• IT and ME –> quite similar, it is difficult to decide the winner
• IT tendency to overestimation
Driving model comp: ecmwf/Cosmo-ME/Cosmo-IT fixed thres, seasonal
• Big gap between Ecmwf and Cosmo-model
• IT and ME –> quite similar, it is difficult to decide the winner
• IT tendency to overestimation
Last year verif.: diurnal cycle for all the versions
Low thres High thres
• Bias overestimation peak during midday
• Spin-up problem for all the models especially COSMO-I7 and COSMO-I2
• General worsening with forecast time
• The spin-up seems to disappear, underestimation during the first 6h
• Bias overestimation peak during midday
• COSMO-7 underestimates
In general: slight improvement with respect to the previous year
To sum up• Long period verification (seasonal trend) (from djf’04 to mam’09)
1. General improvement trend
2. For low thres (rain/no rain): overestimation during spring/summertime with more probability of
detection but also more false alarms
3. For high thres: the worse skills during spring/summertime (convective period)
4. General Cosmo-7 worsening during last year
• Verification over last year (200806-200905)
1. Good performance for Cosmo-EU
2. Similar performance Cosmo-I7/ Cosmo-ME (slightly better Cosmo-ME)
3. Cosmo-I7/ Cosmo-I2 comparison: similar skill, underestimation I2
4. Cosmo-ME/Cosmo-IT comparison: similar skill, overestimation IT
5. Diurnal cycle: bias overestimation peak during midday, in general slight improvement with respect
to previous year
COSMO-7 COSMO-ME
COSMO-EU COSMO-I2COSMO-IT
COSMO-I7BIAS
200806-200905
10mm/24H
COSMO-7 COSMO-ME COSMO-I7Rel Error
MAM ‘09
COSMO-EU COSMO-IT COSMO-I2
COSMO-7 COSMO-ME
COSMO-I2
COSMO-I7POD
200806-200905
10mm/24H
COSMO-ITCOSMO-EU