Short-Range Ensemble Prediction

System at INM

José A. García-MoyaSMNT – INM

27th EWGLAM & 12th SRNWP MeetingsLjubljana, October 2005

October 2005 27th EWGLAM & 12th SRNWP Meetings

2

Introduction

Surface parameters are the most important ones for weather forecast.

Forecast of extreme events (convective precip, gales,…) is probabilistic even for the short-range.

Short Range Ensemble prediction can help to forecast these events.

Forecast risk (Palmer, ECMWF Seminar 2002) is the goal for both Medium- and, also, “Short-Range Prediction” (quotation is mine).

October 2005 27th EWGLAM & 12th SRNWP Meetings

3

October 2005 27th EWGLAM & 12th SRNWP Meetings

4

Errors in LAMs

Due to the model formulation Multimodel techniques

Due to uncertainties in the initial state Singular vectors, breeding

Due to uncertainties at boundaries From different deterministic global models From a global ensemble

Due to the parameterization schemes Multiphysic Stochastic physic techniques

October 2005 27th EWGLAM & 12th SRNWP Meetings

5

Multi-model

Hirlam. HRM

from DWD. MM5 UM

Unified Model from UKMO.

October 2005 27th EWGLAM & 12th SRNWP Meetings

6

Multi-Boundaries

From different global deterministic models: ECMWF UM

UKMO AVN

NCEP GME

DWD.

October 2005 27th EWGLAM & 12th SRNWP Meetings

7

Ensemble

72 hours forecast four times a day (00, 06, 12 y 18 UTC).

Characteristics: 4 models. 4 boundary conditions. 4 last ensembles (HH, HH-6, HH-12, HH-18).

16 member ensemble every 6 hours Time-lagged Super-Ensemble of 64

members every 6 hours.

October 2005 27th EWGLAM & 12th SRNWP Meetings

8

Actual Ensemble

72 hours forecast once a day (00 UTC). Characteristics:

4 models. 4 boundary conditions.

13 (of 16 expected) member ensemble every 24 hours

October 2005 27th EWGLAM & 12th SRNWP Meetings

9

Actual Ensemble II

BCs / Model

AVN ECMWF

GME UM

Hirlam X X X X

Hrm X X X X

MM5 X X X X

UM O O O X

October 2005 27th EWGLAM & 12th SRNWP Meetings

10

Road Map2003-2004

Research to find best ensemble for the Short Range

Jun 04 – Jun 05

Building Multimodel System

Jun 05-Dec 05

Mummubn/16 members

Daily run non-operational

Mar 06 Mummub 16/16

members

Full operations

Jun 06 Mummub+4lag64 members

First try

October 2005 27th EWGLAM & 12th SRNWP Meetings

11

Post-processing

Integration areas 0.25 latxlon, 40 levels Interpolation to a common area

~ North Atlantic + Europe Grid 380x184, 0.25º

Software Enhanced PC + Linux ECMWF Metview + Local developments

Outputs Deterministic Ensemble probabilistic

October 2005 27th EWGLAM & 12th SRNWP Meetings

12

Post-processing II

October 2005 27th EWGLAM & 12th SRNWP Meetings

13

Monitoring in real time

Intranet web server Deterministic outputs

Models X BCs tables Maps for each couple (model,BCs)

Ensemble probabilistic outputs Probability maps: 6h accumulated

precipitation, 10m wind speed, 24h 2m temperature trend

Ensemble mean & Spread maps EPSgrams (not fully-operational)

Verification

October 2005 27th EWGLAM & 12th SRNWP Meetings

14

Monit 1: home

October 2005 27th EWGLAM & 12th SRNWP Meetings

15

Monit 2: all models X bcs

October 2005 27th EWGLAM & 12th SRNWP Meetings

16

Monit 5: All Prob 24h 2m T trend

October 2005 27th EWGLAM & 12th SRNWP Meetings

17

Monit 7: Spread - Emean maps

October 2005 27th EWGLAM & 12th SRNWP Meetings

18

Validation ECMWF operational analysis as reference. Verification software

~ ECMWF Metview + Local developments Deterministic scores

Bias & Rms for each member Probabilistic ensemble scores

Rank histograms ROC Spread skill

15 days of comparison (Aug, 17 to 31, 2005).

October 2005 27th EWGLAM & 12th SRNWP Meetings

19

October 2005 27th EWGLAM & 12th SRNWP Meetings

20

Rank histograms

Ensemble members ranked from smallest to greatest value.

Percent of cases which verifying analysis falls in an interval.

First interval, below smallest member.

Last one, above greatest member. Z500, T500, Msl Pressure

H+24, H+48

October 2005 27th EWGLAM & 12th SRNWP Meetings

21

October 2005 27th EWGLAM & 12th SRNWP Meetings

22

Spread Skill

Spread vs Ensemble Mean Error Z500

H+00 to H+72 T500

H+00 to H+72 Msl Pressure

H+00 to H+72

October 2005 27th EWGLAM & 12th SRNWP Meetings

23

October 2005 27th EWGLAM & 12th SRNWP Meetings

24

ROC Curves

10m Wind Speed Thresholds: 10m/s, 15m/s H+24, H+48

24h Accumulated Precipitation Thresholds: 1mm, 5mm, 10mm, 20mm H+24, H+48

October 2005 27th EWGLAM & 12th SRNWP Meetings

25

October 2005 27th EWGLAM & 12th SRNWP Meetings

26

October 2005 27th EWGLAM & 12th SRNWP Meetings

27

Advantages: Better representation of model errors (SAMEX and

DEMETER). Consistent set of perturbations of initial state and

boundaries. Better results (SAMEX, DEMETER, Arribas et al., MWR

2005). Disadvantages:

Difficult to implement operationally (four different models should be maintained operationally).

Expensive in terms of human resources. No control experiment in the ensemble.

Conclusions for Multimodel

October 2005 27th EWGLAM & 12th SRNWP Meetings

28

Future 16 members full-operational Bias removal Calibration: Bayessian Model

Averaging Verification against observations Time-lagged 64 members 4runs/day More Post processing software

(targeting clustering)

October 2005 27th EWGLAM & 12th SRNWP Meetings

29

Thanks to…

MetOffice Ken Mylne, Jorge Bornemann

DWD Detlev Majewski, Michael Gertz

ECMWF Metview Team