1. Cosmo General Meeting Moscow 6-9 Sept 2010 Two (maybe three) convective case studies with COSMO 1 km simulations Antonella Morgillo Arpa-Simc Bologna

2. Cosmo General Meeting Moscow 6-9 Sept 2010 Outline motivations of the work namelist and set up of the run case studies results and preliminary conclusions future works

3. Cosmo General Meeting Moscow 6-9 Sept 2010 27/05/2009 – 14:00 Colorno (PR) 21.2 27/05/2009 – 14:45 Parma T. P. (PR) 29.4 27/05/2009 – 14:45 Parma (PR) 31.4 27/05/2009 – 15:15 Casatico – Langhirano (PR) 36.8 27/05/2009 – 15:45 Collagna (RE) 20.2 27/05/2009 – 16:00 Ligonchio (RE) 30 27/05/2009 – 16:00 Collagna (RE) 25 Observed precipitation in 15 minutes Courtesy of Federico Grazzini Operational CosmoI2 fc +18 27 may 2009 Parma’s flood

4. Cosmo General Meeting Moscow 6-9 Sept 2010 Most of (present) high resolution model are not able to resolve explicity the deep convection; this is (maybe) because for mesh size between 3 and 7 km a convective system is partly resolved and partly still a subgrid process; Several studies demostrate that an increasing spatial resolution towards the 1 km spatial scale improved the rappresentation of convective phenomena (like thunderstorm, squall line and so on ); in the 1km model the surface terrain are much better rappresented; in future: 1km scale are able to make use of higher resolution input data (data assimilation from radar or high resolution gauge rain network). Motivation 2

5. Cosmo General Meeting Moscow 6-9 Sept 2010 Model set up and integration area 1.COSMO I7 (v 4.9; int2lm 1.10) with 0.625° grid length, dt=40s; 2.297*313 grid points and 40 vertical levels; 3.IC from 12h previously assimilation cycle, BC from IFS forecasts; 4.fc +36 hours; 5.Tiedke convection; 6. five prognostic species (vapour, rainwater, ice, snow, graupel); 1.COSMO 1km: One way nesting on COSMO I7 from 0.0625° to 0.01°; dt=5s 2.551*400 grid points and 45 vertical levels; 3. no assimilation; 4. fc +24 hours; 5. Shallow convection; 6.five prognostic species (vapour, rainwater, ice, snow, graupel); 7. Runge Kutta scheme;

6. Cosmo General Meeting Moscow 6-9 Sept 2010 Model errors and problems Mainly a CFL violation error in the Alps area when the model encounters a steep orography like that; CFL point lat =46.675 lon =11.002 (0.05 x 0.05) rotated coord lat= -10.9 a -10.7, lon=0.7 Photo by D. Cesari

7. Cosmo General Meeting Moscow 6-9 Sept 2010 The vertical section in the CFL point _l The CFL point from the Initial orography 0.1° of spatial resolution The problem: the model isn’t able to handle the steepness of the valley by a simple application of an orographic filter: The solution: the application of the orographic filter at least three times with an extra filter that acts when the altitude difference between two near points is more than 250 m. m The same point after the application of the filter &INPUT lfilter_oro =.true., eps_filter=0.1, ilow_pass_oro=6, lxso_first=.true., ilow_pass_xso=8, num_filt_xso=3, rxso_mask=250.0, m 1300m 1800m 1500m 1300m

8. Cosmo General Meeting Moscow 6-9 Sept 2010 The nesting Test the sensitivity of the 1km model to the “mother” (several tests nesting the 1km over 2.8 and 7km); Nesting a storm model over a model with an assimilation cycle (nudging) and a parametrized convection; Perform a true comparison between the 2.8km and 1km (nested over the same “mother”). COSMO 7 KM COSMO 1KM COSMO 2.8KM

9. Cosmo General Meeting Moscow 6-9 Sept 2010 Disclaimer secondly because it’s simple to compare our model forecast with the radar’s network (they have more or less the same resolution). We don’t take into account the radar error (at this stage), and we don’t make any verification agaist the rain gauge. At first we glance at precipitation; We stress our attention on convective activity and choose some (few) case study of convective precipitation affected the Emilia Romagna region and North Italy; first of all because we are interested to the possibility of a very high resolution model to create by himself the deep convection; We want to compare (at a glance) the localization and the growth of convective cells comparing the model precipitaton maps with the radar precipitation maps. It’s unlikely that the model predicted exactly in the space and in the time every showers and this has to be borne in mind when we’ll look to the maps. Neverthless the structure predicted by the model give us some usefull information about the situation. Next we stress our attention to the model capacity to reproduce the right “initiation of convection”. We define the “initiation of convection” as the time when the radar accumulated precipitation is over 0.1 mm/h

10. Cosmo General Meeting Moscow 6-9 Sept 2010 Summary of Case studies 27 may 2009 – flood over Parma city – strong convective activity over North Italy. 07 July 2009 – organized thunderstorm over North Italy with growth of a lot of intense but very localized thunderstorms over the Emilia Romagna. 29 may 2010 – extended thunderstorms in the North Italy and a strong organized thunderstorm along the Po river (about 70mm/2h recorded by the rain gauge of Copparo station (Ravenna) )

11. Cosmo General Meeting Moscow 6-9 Sept 2010 The observation: Reflectivity from the Emilia Romagna meteorological radar network for the case study 27 may 2009 At 12 UTC a convective cell moves from Lago di Garda towards Sud. A large orographyc convective systems spread along the Tosco Emiliano’s Appennine. In the afternoon a strong thunderstorm hit Parma city (55mm/h recorded by the rain gauge). In the following hours an organized convective line spring up and moving towards est. 12.15 14.15 17.15 19.15

12. Cosmo General Meeting Moscow 6-9 Sept 2010 1h precipitation from italian DPCN radar network 1h precipitation from Cosmo 1km forecast 1 hour accumulated precipitation maps 27 may 2009

13. Cosmo General Meeting Moscow 6-9 Sept 2010 1 hour accumulated precipitation averaged in the area — cosmo1km — cosmo 2.8km — cosmo 7km — radar Integration area Lat = 43.5°N to 45°N Lon = 10°E to 12.5°E Cosmo 7km has the same 24h accumulated averaged precipitation of the radar but the initiation of convection is about 4 hours earlier; Cosmo 2.8 km – the initiation of convection is 1 hour earlier of the radar’s. It produce too much rain in a too short time, about13 mm in 5 hours (as we can see from the steepness of the curve); Cosmo 1km begin the convection 1hour later the radar’ beginning, with a high rain rate (about 7mm in 2 hours), but the same total 24 accumulated averaged precipitation. It reproduce (more or less) both in time and in space the convective cell that hits Parma.

14. Cosmo General Meeting Moscow 6-9 Sept 2010 — cosmo1km — cosmo 2.8km — cosmo 7km Instability index cape&cin averaged in the area The CAPE averaged map shows that the 7km and 1 km have about the same energy; The 2.8km has a very high CAPE (averaged in the region) with respect to the other model (and the same CIN of 1km). This large energy is released in a high rain rate.

15. Cosmo General Meeting Moscow 6-9 Sept 2010 The observation: Reflectivity from the Emilia Romagna meteorological radar network for the case study of 07 July 2009 In the night between the 6 and 7 of July a lot of intense thunderstorms develop in the North Italy. In the morning of the next days at 9 a large thunderstorm system moves from north-west to north-est of North Italy, and there is the growth of a lot of convective cells over Emilia Romagna region that hits, in particular, Modena and Bologna city. At the end the sistem moves towards the Adriatic sea. 3.15 4.15 9.15 10.15 11.15 12.15 14.15

16. Cosmo General Meeting Moscow 6-9 Sept 2010 1hour accumulated precipitation maps 07 July 2009 1h precipitation from italian DPCN radar network 1h precipitation from Cosmo 1km forecast

17. Cosmo General Meeting Moscow 6-9 Sept 2010 — cosmo1km — cosmo 2.8km — cosmo 7km — radar Integration area Lat = 43.5°N to 45°N Lon = 10°E to 12.5°E 1 hour accumulated precipitation averaged in the area Cosmo 7km performs better than the other models. The “initiation of convection” has no particular delay. The accumulated averaged precipitation is in close to the radar estimation; Cosmo 2.8km definitely underestimates the precipitation; Cosmo 1km: the “initiation of convection” has more than 11 hours of delay. The accumulated averaged precipitation is underestimated;

18. Cosmo General Meeting Moscow 6-9 Sept 2010 — cosmo1km — cosmo 2.8km — cosmo 7km Instability index cape&cin averaged in the area cin cape J/kg In the first hours of the run both 1km and 2.8km have a large CAPE, but not enough CIN for triggering the convection.

19. Cosmo General Meeting Moscow 6-9 Sept 2010 10.30 11.30 The observation: Reflectivity from the Emilia Romagna meteorological radar network for the case study 29 may 2010 In the late morning a large convective cell located over Lago Maggiore merge with some localized convective cells over the Est part of the Emilia Romagna to form an organized thunderstorm line along the Po river. Some isolated thunderstorms spring up in the west part of the region. In the following hours the system moves towards sud-est. The rain gauge of Copparo (located near Ravenna in the est part of the ER region) records about 70mm/2hours. 12.30 13.30 15.00 16.00

20. Cosmo General Meeting Moscow 6-9 Sept 2010 29 may 2010 1h accumulated precipitation maps 1h accumulated precipitation from radar 1h precipitation forecast from Cosmo 1km

21. Cosmo General Meeting Moscow 6-9 Sept 2010 Integration area Lat = 43.5°N to 45°N Lon = 10°E to 12.5°E 1hour accumulated precipitation averaged in the area — cosmo1km — cosmo 2.8km — cosmo 7km — radar Cosmo 7km and the radar curve are similar. but it has come early the “initiation of convection” of about 6 hours. The behaviour and the accumulated total precipitation looks like; Cosmo 2.8km; the “initiation of convection” is of about 1 hour earlier and it underestimates the total averaged precipitation; Cosmo 1km has no delay in the “initiation of convection”, but it underestimates the prec.

22. Cosmo General Meeting Moscow 6-9 Sept 2010 — cosmo1km — cosmo 2.8km — cosmo 7km Instability index cape&cin averaged in the area cin cape Both the 1km and 2.8km, in the first hours of the run, don’t have enough CAPE, (and have negative CIN), so the convection isn’t trigged on. The 7km has a large CAPE (and energy available)

23. Cosmo General Meeting Moscow 6-9 Sept 2010 summarizing …. Radar Cosmo 7km Cosmo 2.8km Cosmo 1km 27/05/0912 UTC- 4 h-2 h+1 h 07/07/0901 UTC+ 1 h (no initiation +11 h 29/05/0910 UTC- 6 h- 0.5 h0 h Delay in the “Initiation convection” relative to the radar expressed in hours (the minus sign means the model begin before the radar Radar Cosmo 7km Cosmo 2.8km Cosmo 1km 27/05/0966 13.5 7 07/07/092.82.50.10.5 29/05/094.3522.6 Total precipitation in 24/hours expressed in mm/24h. In blue the underestimation, in red the overestimation, in green the right estimate  Cosmo 7km generally has come early the “initiation of convection”, but the averaged total precipitation is almost right forecasted;  Cosmo 2.8km has come early too the “initiation of convection” and generally underestimates the averaged total precipitation. Only in one case it overestimates the forecasted precipitation (when it began to rain It rains too much!!)  Cosmo 1 km has a short delay (of about 1h or less) in the “initiation of convection” and has a general, but not crucial, underestimate of the averaged total precipitation. This underestimate is probably due because the explicit convection takes some time to begin, and when the convective activity is too close to the beginning of run (like in the 7 July case study when the convection begin very early in the morning) the model generally insn’t able to forecast the convective activity.

24. Cosmo General Meeting Moscow 6-9 Sept 2010 Some final and philosofical questions The first question is if it’s reasonable to push up the model towards this resolution. The benefits would be great as regards the forecast of extreme localized events, or fog forecast or for the air quality. Besides we pawn a lot of expensive extra computer resources and a lot of work for updating the state of art of our high resolution models. We would ask if running a 1 km model give us a real improvements to the understanding of weather risk situations and early warning The nswear is yes! At least if our aim is not to miss an extreme event. The second question is the verification problem. At this stage we know that the details at the grid-scale is not believable. We do not expect a 1km forecast to get the locations of showers accurate to 1 km. It’s not wise. Besides one of the problem would be what we use to verify the high-resolution forecasts against. In this work we use the radar networks that have more or less the same spatial scale. However radar data may not have the accurancy we would like. Rain gauge is more accurate, but which type of verification?

25. Cosmo General Meeting Moscow 6-9 Sept 2010 Future work Sensitivity to the convection scheme of the “mother” model Sensitivity to the increased vertical level Study regarding the horizontal diffusion assimilation of high resolution input data (like radar data or regional gauge network)? A new tools for the verification? Thanks to all for the attention and many thanks to Davide Cesari and Oliver Furher for usefull discussions

26. Cosmo General Meeting Moscow 6-9 Sept 2010 Slide di supporto

27. Cosmo General Meeting Moscow 6-9 Sept 2010 Sensitivity al nesting: 24 hours total precipitation COSMO I7COSMO I2 COSMO I1 su I7 COSMO I1 su I2

28. Cosmo General Meeting Moscow 6-9 Sept 2010 Model set up Cosmo 7 kmCosmo I 1 km Time step40 s.5s Vertical levels4045 Domain (ie x je)297*313551*400 Grid length0.0625°0.01° Prognostic rainyesYes Convection schemeOn (Tiedke convection)Off (shallow convection) Assimilation12h previous cycle nudgingNone Microphysics 4 species (water vapour, cloud water, rain, precipitation ice (graupel, hail, snow) DynamicsLeap frogRunge Kutta

29. Cosmo General Meeting Moscow 6-9 Sept 2010 Carattestiche dell’esperimento. La suite dell’esperimento:  Versione: Int2lm 1.8 - Cosmo 4.10; risoluzione: cosmo I7, cosmo I2, cosmo I1  Inizio run 27 maggio 2009 00 FC +24  IC da ciclo di analisi di 12 h di COSMO I7. BC da I7 e da I2 BC da I7 BC da IFS BC & IC Graupel, Runge Kutta, SSO spenta, convezione shallow, diffusione orizzontale attiva Risoluzione 0.01 (circa 1 km)‏COSMO I1 Graupel, Ruinge Kutta, sso spenta, convezione shallowRisoluzione 0.025 (circa 2.8km)‏COSMO I2 Graupel, Tiedke, sso attivo, Risoluzione 0.625 ( circa 7km di risoluzione spaziale)‏ COSMO I7 Principali caratteristicheDESCRIZIONERISOLUZIONE

30. Cosmo General Meeting Moscow 6-9 Sept 2010 27 maggio 2009

31. Cosmo General Meeting Moscow 6-9 Sept 2010 Hourly total precipitation from the model 27 maggio 2009 I1 su I7 0102030405 06 07080910 1112131415 1617181920 24232221

32. Cosmo General Meeting Moscow 6-9 Sept 2010 Hourly total precipitation from radar 27 maggio 2009 0102030405 06 07080910 1112131415 1617181920 24232221

33. Cosmo General Meeting Moscow 6-9 Sept 2010 COSMO I1 - exp8 (I1 su 2.8 km) 27 maggio 2009

34. Cosmo General Meeting Moscow 6-9 Sept 2010 COSMO I1 - exp9 (I1 su I 7km ) 27 maggio 2009

35. Cosmo General Meeting Moscow 6-9 Sept 2010 Comments on 27/05/2009 Cosmo 7km has the same 24h accumulated averaged precipitation of the radar but the initiation of convection is about 4 hours earlier; Cosmo 2.8 km – the initiation of convection is 1 hour earlier of the radar’s. It produce too much rain in a too short time, about13 mm in 5 hours (as we can see from the steepness of the curve); Cosmo 1km begin the convection 1hour later the radar’ beginning, with a high rain rate (about 7mm in 2 hours), but the same total 24 accumulated averaged precipitation. It reproduce (more or less) both in time and in space the convective cell that hits Parma. The CAPE averaged map shows that the 7km and 1 km have about the same energy; The 2.8km has a very high CAPE (averaged in the region) with respect to the other model (and the same CIN of 1km). This large energy is released in a high rain rate.

36. Cosmo General Meeting Moscow 6-9 Sept 2010 7 luglio 2009

37. Cosmo General Meeting Moscow 6-9 Sept 2010 Integration area 3 Lat = 43°N to 47°N Lon = 8°E to 13°E — cosmo 7km — cosmo1km — cosmo 2.8km — radar 07/07/09

38. Cosmo General Meeting Moscow 6-9 Sept 2010 CAPE accumulated over 1 hour averaged in the area — cosmo1km — cosmo 2.8km — cosmo 7km Integration area 3 Lat = 43°N to 47°N Lon = 8°E to 13°E 07/07/09

39. Cosmo General Meeting Moscow 6-9 Sept 2010 7 km 2.8 km accumulated precipitation maps at 12 utc for 7km and 2.8km cosmo model

40. Cosmo General Meeting Moscow 6-9 Sept 2010 Hourly total precipitation from the model 07 07 2009 0102030405 06 07080910 1112131415 1617181920 24232221

41. Cosmo General Meeting Moscow 6-9 Sept 2010 Hourly total precipitation from radar 07 07 2009 0102030405 06 07080910 1112131415 1617181920 24232221

42. Cosmo General Meeting Moscow 6-9 Sept 2010 Comments Cosmo 7km performs better than the other models. The initiation of convection has no particular delay. The main goal of the parametrization scheme; The accumulated averaged precipitation is in close to the radar estimation; Cosmo 2.8km definitely underestimates the precipitation; Cosmo 1km: in the first hours in the morning the model forecast the large convective sistem (more or less) over the North west Italy, but it doesn’t move towards Est. The model lacks almost completely the thunderstorms in the Emilia Romagna (but cosmo 2.8km too). The accumulated averaged precipitation is underestimated; Both 1km and 2.8km in the first hours have a large CAPE, but not enough CIN for trigging the convection.

43. Cosmo General Meeting Moscow 6-9 Sept 2010 29 maggio 2010

44. Cosmo General Meeting Moscow 6-9 Sept 2010 Comments on 29 maggio 2010 Cosmo 7km and the radar curve are similar. CosmoI7 anticipate the “initiation of convection” of about 6 hours, but the behaviour and the accumulated total precipitation looks like; Cosmo 2.8km anticipate the “initiation of convection” of about 1 hour and underestimate the total averaged precipitation; Cosmo 1km has no delay in the “initiation of convection”, but underestimate the tp. Both 1km and 2.8km, in the first hours, have a not enough CAPE, (and negative CIN) and the convection isn’t trigged on (but later they still don’t produce enough precipitation).