Richard Moffet(2) ,Normand Gagnon(1)and Juan Fontecilla(1) Canadian Meteorological Centre, development (1) and operations(2) divisions, Meteorological Service of Canada, Environment Canada Dorval, Québec

Monthly and seasonal weather - Is it

predictable? (Ensemble forecasting at the

Canadian Meteorological Centre)

Weather forecasting and uncertainty

Ensemble prediction system

Seasonal (up to 120 days) MAIN TOPIC

Canadian EPS and NAEFS

Products

Summary: web links

Outline

Principles of Numerical Weather Prediction

OBSERVATIONS INITIALIZATION NUMERICAL

MODEL

Sampling of current state

of the atmosphere.

Processing data to

initialize models.

Projection forward

in time.

• Upper air soundings.

• Surface land stations.

• Ships / buoys.

• Aircraft observations.

• Satellite data.

• Quality control.

• Brings observed

atmospheric variables

onto model grid.

• Provides forecast

atmospheric variables on

model grid.

Trial field

Models are improving!

T. Robinson, CMC

~11 years ~11 years

48-h gain in

predictability in 22 years

But still one big problem: “chaos”

120-h integration – mean sea level pressure

Two integrations done with identical NWP models but on different computers

M. Lajoie, CMC

Bit Flip

• Test of model sensitivity via minor changes

• The last bit of dynamical fields (HU,ES,VT,TT,UU et VV) switched

from 1 to 0 and vice-versa. The flip is done on all levels and at all 10

points in latitude and longitude.

2 runs on NAOS

(RUN PAR) - ( RUN with BIT FLIP)

240h, GZ 500 mb Up to 24 mb at surface and et 28 dm GZ at 500mb

Principles of NWP – modeling Primitive equations

spp

xs

Vd Fx

pTR

xfv

x

uu

t

u

ln

ys

Vd Fy

pTR

yfu

y

vv

t

v

ln

0

VdTRHydrostatic

Momentum

Ts

p

VdV Fdt

pd

dt

d

C

TR

dt

dT

ln1

qs F

dt

dqV

dt

Pd ; 0

ln

RTp

Continuity

Thermodynamics

State 0 at T and 1

u

tu

u

x

Sources of error create uncertainties in

initial conditions and then in forecasts

OBSERVATIONS INITIALIZATION

NUMERICAL

MODEL

Uncertainty Uncertainty Uncertainty

Sampling of current state

of the atmosphere

Processing data to

initialize models

Projection forward

in time

Trial field

Then comes… Ensemble forecasting

Initial states Final states

True initial state True final state

Climatology

Ensemble

mean

Analysis

Deterministic

forecast

Uncertainty on

initial state

R. Verret, N. Gagnon, CMC

Atmospheric Ensemble forecasting

basics

• An Ensemble Prediction System is a set

of integrations of one or several NWP

models that differ in their initial states

(and sometimes in their configurations

and boundary conditions).

• Ensemble prediction is an attempt to

estimate the non-linear time evolution of

the forecast error probability distribution

function.

• With Ensemble forecast, it is possible

to evaluate, express and forecast

uncertainty.

Context

• Common usages of Ensemble forecasts:

– Ensemble mean as a substitute for a single deterministic

forecast.

– Clustering to produce a small set of forecast states

characterized with the cluster mean.

– A priori prediction of forecast skill.

– Ensemble probability distribution function.

– Measure of uncertainty.

– Extension of forecast range.

Monthly and seasonal forecast system

Introduction

• The Canadian Meteorological Centre (CMC) is producing seasonal forecasts in operational mode since September 1995.

• Seasonal forecasts are purely objective and automated.

• Seasonal forecasts provide three-month (90-day) temperature and precipitation anomaly outlooks in a priori three equally probable categories “above normal”, “near normal” and “below normal”.

• Twelve three-month seasons:

• Dynamical models are used for season 1 with zero and one month lead time.

• A statistical model is used for seasons 2, 3 and 4 with 3, 6 and 9 months lead time respectively.

• The set-up for season 1 is also used for monthly temperature anomaly forecasts issued on the 1st and 15th days of each month.

DJF JFM FMA MAM AMJ MJJ

JJA JAS ASO SON OND NDJ

Current system:

• Dynamical models used for season 1.

– Forecasts are issued on the first day of each month (12

seasons).

– GEMCLIM: 1.875 50 levels ptop 5mb

– GCM2: T32 10 levels ptop 10 mb

– GCM3: T63 32 levels ptop 5 mb

– SEF: T95 27 levels ptop 10 mb

• Historical forecasts (HFP2):

– 41 years (1969-2010).

Current seasonal forecasting set-

up: overview

Four models

GEM + GCM2

+ GCM3 + SEF

ICE

GEM

SEF

GCM3

GCM2

CMC analysis relaxed to

climatology during first 15 days

(GCM2 is a step function)

SNOW

SEF

CMC analysis relaxed to

climatology during first 15

days

GEM

GCM2

GCM3

prognostic variable

SST

last 30-day anomaly

persisted throughout

J0-9 J0-8

J0-7 J0-6

J0-5 J0-4

J0-3 J0-2

J0-1 J0

J0-9 J0-8

J0-7 J0-6

J0-5 J0-4

J0-3 J0-2

J0-1 J0

J0-9 J0-8

J0-7 J0-6

J0-5 J0-4

J0-3 J0-2

J0-1 J0

GEM

GCM2

J0-9 J0-8

J0-7 J0-6

J0-5 J0-4

J0-3 J0-2

J0-1 J0

GCM3

SEF

Month 1 Month 2 Month 3 Month 4 2 runs per day (00 and 12 UTC)

Current seasonal forecasting set-

up: overview

• Four-month integrations:

– Zero lead time forecasts.

– One-month lead time forecasts.

10 lagged runs of

GEM + GCM2 + GCM3 + SEF 40 member

ensemble

Mth 0 Mth 3 Mth 1 Mth 2 Mth 4

Monthly forecast

Monthly forecast at mid-month

Seasonal forecast 0

month lead time

Seasonal forecast one month lead

time

Forecast periods

Anomalies classified into three equally probable categories using 0.43

for temperature anomaly and 0.43 for precipitation anomaly.

20001971

T

20001971

P

Current seasonal forecasting set-up:

deterministic forecasts

2000-1971

SEF ,

20001971

SEF

20001971

SEF

20001971

GCM3 ,

20001971

GCM3

runs 10

GCM3

2000-1971

GCM2 ,

20001971

GCM2

runs 10

GCM2

2000-1971

GEM ,

20001971

GEM

runs 10

GEM

forecastanomaly 4

1

TT

TT

TTTT

TTTT

T

2000-1971

SEF ,

20001971

SEF

20001971

SEF

20001971

GCM3 ,

20001971

GCM3

runs 10

GCM3

2000-1971

GCM2 ,

20001971

GCM2

runs 10

GCM2

2000-1971

GEM ,

20001971

GEM

runs 10

GEM

forecastanomaly 4

1

PP

PP

PPPP

PPPP

P

Class thresholds

normalabove0.43

normal0.430.43

normalbelow0.43

TT

TTT

TT

APCC (Asia Pacific Climate Center) outlook

Pcpn Thresholds

Current seasonal forecasting set-up:

probabilistic forecasts

• To generate probabilistic forecasts:

– Calculate forecast anomaly for each member of the ensemble

using each model’s own climatology

– Classify the forecast anomalies into three equally probable

categories using 0.43 as threshold (above, normal, below)

– Count the number of members in each category

– Divide by the total number of members (40) and multiply by 100

Probabilistic forecasts for:

Above

Normal

Below

Probabilic forecasts

Reliability

Historical Percent Correct

Reliability Diagram

Temperature

Historical Percent Correct

Reliability Diagram

Precipitation

For seasons 2-3-4: Canonical Correlation

Analysis (CCA)

• Purely statistical

• Based on sea surface temperature anomalies and ortho-

normal empirical functions.

• Under revision and improvement

Verification

• Deterministic scores

– Correlation

– Percent correct

• Probabilistic scores

– Reliability

– Sharpness

– ROC

A look back at summer 2010

Summer 2010

Since the summer of 1992, every summer season has seen above normal

Summer 2010

Explaining the present forecast

• La Nina winter…and spring…

500 hPa circulation: La Nina winters

La Nina years: winter,trend removed

La Nina: winter with trend

Strong La Nina: winter T anomalies

La Ninayaers: winter:pcpn

Spring La Nina : trend removed

Spring La Nina: with trend

Spring La Nina years: pcpn

What else?

CEPS/NAEFS

MSC Ensemble Prediction System ( planned to be

used for monthly forecast in the next year)

• Members:

– 20+1 members:

– GEM 0.9° (~100 km resolution) L28 (forecast), L58(analyses).

– 16-day integration.

– Twice a day (00 and 12 UTC).

• Simulation of initial condition uncertainties:

– ensemble Kalman filter data assimilation with perturbed observations.

• Simulation of model uncertainties:

– A multi-model approach, each member having its own physics

parameterization.

– Stochastic perturbations added to tendencies in the parameterized

physical processes.

– Stochastic kinetic energy back-scattering scheme to re-introduce

dissipated energy.

current – As of May 2010

Current parameterizations used in

global MSC EPS

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

No

Back-

scattering

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

Strong

Weak

std

GWD

0.85

1.0

0.85

0.85

1.0

1.0

0.85

0.85

1.0

1.0

0.85

0.85

1.0

0.85

1.0

1.0

0.85

0.85

1.0

1.0

1.0

Vertical

mixing

parameter

No Bougeault ISBA Kain & Fritsch 0

Stochastic

Physics

Mixing length Land surface scheme Deep convection #

Yes

Yes

Yes

Yes

Yes

Bougeault

Bougeault

Blackadar

Bougeault

Blackadar

ISBA

Force-restore

ISBA

ISBA

Force-restore

Relaxed Arakawa Schubert

Kuo Symmetric

Kain & Fritsch

Oldkuo

Relaxed Arakawa Schubert

16

17

18

19

20

Yes

Yes

Yes

Yes

Yes

Blackadar

Bougeault

Blackadar

Bougeault

Blackadar

Force-restore

Force-restore

Force-restore

Force-restore

ISBA

Relaxed Arakawa Schubert

Kuo Symmetric

Oldkuo

Kain & Fritsch

Kuo Symmetric

11

12

13

14

15

Yes

Yes

Yes

Yes

Yes

Blackadar

Bougeault

Blackadar

Blackadar

Bougeault

Force-restore

ISBA

ISBA

ISBA

ISBA

Kain & Fritsch

Kuo Symmetric

Relaxed Arakawa Schubert

Kain & Fritsch

Oldkuo

6

7

8

9

10

Yes

Yes

Yes

Yes

Yes

Bougeault

Blackadar

Bougeault

Blackadar

Bougeault

ISBA

ISBA

Force-restore

Force-restore

Force-restore

Kain & Fritsch

Oldkuo

Relaxed Arakawa Schubert

Kuo Symmetric

Oldkuo

1

2

3

4

5

From P. Houtekamer, ARMA

NAEFS

• Global ensembles: – NOAA, MSC, NMS of Mexico: official agreement signed in

November 2004.

– FNMOC (US NAVY) may join NAEFS in 2010.

• Advantages: – Larger ensemble allowing better PDF definitions (super-ensemble).

– Improved probabilistic forecast performance.

– Seamless suite of forecast products across international boundaries and across different time ranges (1-14 days).

– Minimal additional costs – levering computational resources.

– Synergy with NCEP on R&D work.

• Problems: – Combination of multi-model ensembles into a super-ensemble.

– Real time exchange (operational considerations).

NAEFS

• Raw data exchange (00 and 12 UTC runs).

– ~ 50 selected variables.

– 6-hourly output frequency over 16 days.

– GRIB format.

• Basic products:

– Using same algorithms/codes.

– Bias correction algorithm.

– Forecast products in terms of climatological anomalies.

– Week 2 (days 8 to 14) forecasts based on the combined MSC/NCEP ensembles.

Produced with

EPS forecasts!

MSC official Forecasts for days 6 and 7

From Canadian EPS system only

More for my mother …

Forecast confidence

• The spread-skill relationship can be used to assess

forecast confidence.

• Forecast is incomplete without information on

expected flow dependent skill.

Probabilistic forecasts • Evaluation of possible scenarios.

• EPS outputs downscaling application models

• Construction of pdf from a finite ensemble.

• Can be used to extend forecast range beyond day 5.

For you, decision makers…

Risk management • Probabilities of event occurrences.

• Risk calculation.

Decision making • Decision making based on probabilities and

cost/loss ratio.

Mean and standard deviation maps

Meteograms

Probability maps

% of getting > 10 mm of rain over 6 days

Done by member counting

Access to digital data

• Currently available global data on grid :

http://www.weatheroffice.gc.ca/grib/Low-resolution_GRIB_e.html

• And on our datamart:

– Raw model outputs on a higher resolution grids in GRIB2 format,

– Values at points (cities) in XML format: meteograms (soon)

http://dd.meteo.ec.gc.ca/ensemble/

http://www.ec.gc.ca/meteo-weather/default.asp?lang=En&n=136568DB-1

http://www.weatheroffice.gc.ca/saisons/index_e.html

http://www.weatheroffice.gc.ca/ensemble/naefs/index_e.html

http://www.meteo.gc.ca/ensemble/index_e.html

http://www.ec.gc.ca/adsc-cmda/default.asp?lang=En&n=30EDCA67-1

http://www.ec.gc.ca/adsc-cmda/default.asp?lang=En&n=98231106-0

Thank you! Merci!