Production of a multi-model, convective-

scale superensemble over western Europe

as part of the SESAR project

PHY-EPS Workshop, June 19th, 2013

Jeffrey Beck, F. Bouttier,

O. Nuissier, and L. Raynaud*

CNRM-GAME

*GMAP/RECYF

Météo-France/CNRS

European Convective-Scale EPS

Transition toward convection-resolving ensembles (e.g.): France: PEArome (2.5 km, 12 members, 24-hour forecasts) – Pre-Op UK: MOGREPS-UK (2.2 km, 12 members, 24-hour forecasts) – Pre-

Op Germany: COSMO-DE (2.8 km, 20 members, 21-hour forecasts) – Op Others?

Computational resources focused toward high-resolution representation of small-scale features (e.g., extreme events, fog), but creates limitations: Number of members and therefore ensemble sampling/performance is

restricted Size of domain and forecast duration also constraints

Potential solution is to combine multiple national models in a “super”-ensemble

Single European Sky ATM Research (SESAR)

Collaborative project to overhaul European airspace and Air Traffic Management (ATM)

Goal is to unify ATM over EU states, similar to NextGen ATM program in the USA

Key necessity: Continent-wide convective-scale modeling for aviation hazards with ensemble (probabilistic) forecasts

Within the context of the SESAR project, an experimental version of a superensemble is being created (operational in several years)

http://www.sesarju.eu

Regional Model Domains

MOGREPS + AROME =

24 members

COSMO +

AROME = 32

members

Uniform resolution, grid, and forecasts required in order to merge individual models from Met Office, Météo-France, and DWD: 0.022° lat x 0.027° lon grid, ~2.2 km resolution Slightly adjusted (interpolated) domains allowing for collocated grid points Hourly forecasts out to 21 hours (00Z or 03Z initialization)

Parameters collected: 10-m variables, pressure level temperature, wind, and hydrometeor content,

plus total surface accumulated precip since initialization Derived variables: simulated reflectivity, echotop, and VIL for hazardous

weather forecasting

Preliminary dataset collected during convective events between July and August 2012 (42 days)

Model Specifics for Superensemble

Calculate simulated reflectivity at each grid point using rain, snow, and hail hydrometeor mixing ratios

Find upper-most pressure level with 18 dBZ = Echotop

Integrate reflectivity factor for column above grid point to derive vertically integrated liquid (VIL) for hail detection (Z D∝ 6)

Initial Superensemble Derived Variables

z

x

Echotop ~ 18 dBZ

VIL = kg m-2

850 mb Simulated Reflectivity

Sim. Ref. Example (AROME and COSMO)

MOGREPS + AROME =

24 members

Model overlap region

15/8/2012 – 18 UTC

850 mb simulated reflectivity ensemble mean (dBZ)

Sim. Ref. Example (AROME and COSMO)

MOGREPS + AROME =

24 members

15/8/2012 – 18 UTC

850 mb simulated reflectivity ensemble spread is qualitatively similar in single domain and overlap regions

Sim. Ref. Animation (AROME and COSMO)

850 mb simulated reflectivity

21-hr simulation from 03Z 5/8/2012 to 00Z 5/9/2012

Echotop Example (AROME and COSMO)

MOGREPS + AROME =

24 members

Echotop (in mb) defined using 18 dBZ

Warm colors indicate lower cloud tops

Echotop Example (AROME and COSMO)

MOGREPS + AROME =

24 members

Echotop ensemble spread (mb)

Warmer colors indicate more spread

Similar spread in overlap regions

Echotop Animation (AROME and COSMO)

Echotop (mb) 21-hr simulation from

03Z 5/8/2012 to 00Z 5/9/2012

Superensemble Challenges

How to interpret output: Initial focus is to meet SESAR deliverables with regard to aviation hazards:

Strong convection, echotop, hail threat (VIL), turbulence, upper-level variables

Use of quantiles, ensemble spread, and probability in both overlap and single model regions

Identify potential inconsistencies and biases when merging ensembles Currently employ a linear decrease in member weight < 50 km from boundary

Point data versus different types of objective analysis smoothing

weight

x or y

Model 2 (red)Model 1 (black)w=1

w=0

PDF = { wi xi } for all members “I”

Impact of spatialization method on reflectivity quantiles

Methods to derive a PDF at a given point x:● m1 'point': use member values at point x

m2 'square': equiprobable values in square around xm3 'circle': like m2, in circle centered around xm4 'cone': use values in disk, with decreasing weight as

●

●

●

distance to x increases.

Notes:● size of square or circle is uniform (here: circle radius = 12 km,square with same area)● 12-member Arome ensemble

Impact of Spatialization on Mean

point circle

square cone

Future Work

Incorporation of other data: Add MOGREPS-UK (“the data are in the mail”) Produce and derive other variables of interest, both for SESAR and for

ensemble modeling research purposes Other countries interested in participation?

Analyse and verify data: Inter-comparison between models in overlap zones (score analyses) Rain gauge and surface based observations for precipitation total and 10-m

forecast variables Validation of probabilistic products using 3D archived radar data from the

French ARAMIS radar network Verification to show impact/benefit of increased ensemble sampling

Thank You

Questions, comments, or suggestions welcome!