Cliquez pour modifier le style du titre Cliquez pour modifier le style des sous-titres du masque 1 Nowcasting strategies : Rapid analysis refresh and high resolution modelling Ludovic Auger, Pierre Brousseau, Olivier Dupont METEO-FRANCE WMO/WWRP Workshop on Use of NWP for Nowcasting Boulder, Colorado, USA October, 2011

2 Outlines 1.Introduction 2.Spin-up and cycling issues 3.Arome nowcasting configuration 4.Arome airport configuration

3 Introduction (1/3) : Nowcasting and NWP In the past there was a clear distinction between nowcasting and numerical weather forecast –Nowcasting was more relying on diagnostic tools or extrapolation of current observations (and still is) –Forecast models were meant to predict large scale phenomena, but only had a weak potential for mesoscale events forecast. CAPE diagnostic, issued from observations spatialisation Radar reflectivities extrapolation Those kind of approaches are less interesting beyond 2 hours. Examples of nowcasting outputs :

4 Introduction (2/3) : Today Mesoscale models With the onset of convective scale models, now there is a possibility for NWP systems to outperform more traditional and empirical Nowcasting tools : Eg : The French operational model AROME running at 2.5km resolution succeeds in anticipating, quite accurately sometimes, risk inducing events such as heavy rainfall Arome operational configuration : –2.5 km model with its own assimilation system, 720x750 grid points, 60 levels on the vertical, running 4 times a day with a 36 hours term –dedicated assimilation system : adapted background statistics, assimilation of all the data available in large scale model in a higher density mode, assimilation of radar reflectivities. enables AROME to perform a forecast with relevant small scale information (though not always at the right position )

5 Introduction (3/3) : AROME nowcasting There is an obvious potential for nowcasting with such refined models. A way to transform AROME forecast model into a nowcasting model is to perform more frequent forecasts with the use of more recent observations. Arome nowcasting project Requirements of such a system : –Short delivery time (~ one hour) due to the short time-range of nowcasting (up to 6 hours) –Must cover a sufficiently large domain, since one of our client is aeronautic, we must cover one entire flight region –High density observations assimilation capacities Consequences : –Very short cut-off : some observation might not have the time to reach our observation database system. –Asynchronous coupling technique must be used. –Configuration files come from older forecasts.

6 Spin-up issues Spin-up is a general term to describe spurious behaviour of the model during the first time-steps of the integration. General cause is the imbalance of meteolorogical fields that produce gravity waves Particularly relevant for nowcasting because we are interested in short range forecasts. Also an issue if we want to use a one-hour cycled system. Many different filtering techniques provide solutions to spin-up issues Sometimes, a higher coherence between fields helps a lot Although we can control quite efficiently spurious oscillations, filtering alters the initial fields and can degrade some forecast parameters, we did not apply any filtering so far. dPs/dt Spinup as a function of time for different model configurations More details will be given for this topics in Thibauts presentation tomorrow Difference between filtered And non-filtered analysis, Temperature for the last level

7 Cycling or not cycling ? Rapid refresh means more frequent analysis, ideal strategy would be to use the previous most recent forecast to feed the new analysis. But we have some difficulties to improve our current 3 hours cycle AROME model, depending on the parameter or forecast initial time we can improve or worsen 3 hour cycle. Even with filtering techniques we hardly improved scores Problem might be somewhere else (observation error correlation ?) We do not cycle our AROME-nowcasting configuration 2m temperature score (RMSE) as a function of forecast range red : 1 h cycle, black 3h cycle Rain Heidke Skill Score as a function of forecast range : 1 h cycle, black 3h cycle More details will be given for this topics in Thibauts presentation tomorrow

8 AROME-Nowcasting Our current 2.5 km resolution AROME model is running 4 times a day up to 36 hours term. Due to observation cut-off and runing time, model outputs are available more than 3 hours after the forecast initial time. We designed our new system with the main objective to have a model available every hour with very recent observations. Arome-nowcasting configuration : –Analysis every hour –- 45 min/+15 min assimilation window –Lateral boundary conditions coming from operational mesoscale AROME model –Provides a 7 hour forecast –No cycling, we use the most recent intial condition from mesoscale model (ie a 2 hour up to 6 hour forecast) –Asynchronous coupling –Specific background error covariances matrix.

9 Arome-Nowcasting AROME nowcasting will take its initial and boundary conditions from the 8-times-per- day AROME operational. General diagram for AROME-nowcasting configuration. Initial and coupling altitude field Initial surface field

10 Arome-Nowcasting Results shown here come from 3 testing periods (27 days total) over a 600x600 gridpoints domain. Forecast range Forecast initial time Comparison between arome nowcasting and arome operational as a function of forecast initial time and forecast range, the thicker the circle is the better Arome nowcasting is (the outer circle diameter corresponds to RMSE from Arome nowcasting, and the inner one corresponds to Arome operational).

11 Arome-Nowcasting The scores are better for most of the parameters (rain scores are under study) Surface pressure is worse in the 3 first hours of run, then is better, this illustrates the spin-up impact. After 3 hours the gain is weak.

12 AROME nowcasting : Loss of observations due to short cut-off Due to the short cut-off time of 15 min some observations are missing. Since we start from a fresh guess from another model every hour, we somehow get the missing information from the initial file

13 AROME airport configuration Part of a R&D project on Wake-Vortex prediction systems. The goal is to provide relevant parameters for Wake-Vortex prediction : temperature, humidity, wind but also Kinetic energy related parameters : Turbulent Kinetic Energy and Eddy Dissipation Rate. Typical nowcasting issue, we have to provide relevant parameters for a specific goal, people from air traffic management want the best data available. We plan to design an hectometric-scale model (0.5 km horizontal resolution), with a refresh analysis every hour Wake Vortex after take off Oragraphy at 0.5 km Oragraphy at 2.5 km

14 AROME airport configuration Justification of the horizontal fine resolution : we hope with such a resolution to resolve explicitly a bigger part of the boundary layer kinetic energy, but we are limited by the computational cost. We still use a shallow convection scheme for parameterization of subgrid boundary layer movements. Subgrid/Total kinetic energy ratio as a function of grid size divided by boundary layer height, according to Honnert and Masson, 2011,. Example with a boundary layer height of 2km, Dx/H=0.25 : more than half of the boundary layer energy is explicit Subgrid/Total kinetic energy Dx/H Ratio for AROME airport for a typical Day.

15 AROME-airport configuration Deep convection representation can also be improved at 0.5 km resolution. Not only is the representation of convection closer to reality, but the timing of the event can also be much more realistic. Simulated reflectivites from 6 hour forecast at 2.5km (left), 0.5km (right). Radar reflectivities for the 05 june 2011, 18H00 UTC (center). 2.5 km model Radar reflectivites0.5 km model

16 Conclusion Today mesoscale models are developed enough to provide useful forecast to nowcasting systems. We developed in that spirit a mesoscale nowcasting model, with forecast every hour. Due to spin-up issues and maybe something else, we are not satisfied with cycling one hour forecast, as a consequence we prefer starting from a new guess every hour. It was shown that due to the use of more recent observation in this system, we significantly increase forecast performance compared to our operational forecast system. Another product derived from our AROME-nowcasting configuration is a nowcasting tool dedicated to an airport coupled with a 0.5 km gridsize model, the goal of this tool will be to feed a Wake-Vortex forecast model.

17 THANK YOU FOR YOUR ATTENTION ! For more information contact