Page 1© Crown copyright 2004 Seasonal forecasting activities at the Met Office Long-range Forecasting Group, Hadley Centre Presenter: Richard Graham ECMWF.

© Crown copyright 2004 Page 1

Seasonal forecasting activities at the Met Office

Long-range Forecasting Group, Hadley Centre

Presenter: Richard Graham

ECMWF Forecast Products Users Meeting, 15-17th June 2005


Outline

Brief summary 2004/5

Recent Met Office (GloSea) seasonal forecasts

New website products

Multi-model forecast combination/calibration


Overview, 2004/5

Met Office GloSea model (version of HadCM3) main tool for seasonal prediction – used in conjunction with statistical methods

now runs in parallel configuration with system2 in ECMWF Op. suite (multi-model)

15-member hindcasts 1987-present have been completed (12 starts/year)

Met Office seasonal forecast website update - planned for July multi-model products (temperature and precipitation)

probability for outer-quintile categories

extensive upgrade of verification/validation information

(research on multi-model combination/calibration)

Use of ECMWF monthly system Met Office post-processed products supplied to various end users

Talks: Bernd Becker and Stefan Meulemans - later today

EU ENACT project completed 43-year global ocean re-analyses, different models and ODA methods

analysis quality assessed through seasonal forecast impacts


Recent forecasts: Nino3.4 SST anomaly, Dec04 – May05

Dec04 Jan05 Feb05

Mar05 Apr05 May05


Recent forecasts:Outer-quintile temperature forecasts and verif DJF04/05

P(outer quintile)

>40%

IRI verification


Skill assessment for outer quintiles – new website

products


Hindcast datasets used

BACKRUNS (‘operational’ set) period: 1987-present (16 yrs, 1987-2002, used here)

15-member ensemble allows skill assessment for higher order (than tercile) categories

DEMETER (‘research’ set) period: 1959-2001 (43 yrs)

9-member ensemble


BACKRUNS verification: temperature, ROC: selectable by region/initial date/lead time:

terciles outer-quintiles

Example for Central America, JJA, 1-month lead (1987-2002)


BACKRUNS verification: precipitation, ROC: selectable by region/initial date/lead time:




BACKRUNS: ROC skill mapsExample for 2m temperature, JJA, 1m lead

from GloSea hindcasts 1987-2002 (16 yrs)

local gridpoint ROC scores (area under the curve)

Skill for outer-quintile ‘extremes’ similar to that for outer terciles

NB. But sample of events is smaller

Event = temperature in upper tercile Event = temperature in upper quintile


DEMETER: ROC skill mapsExample for 2m temperature, JJA, 1m lead

from GloSea DEMETER hindcasts 1959-2001 (43 yrs)

local gridpoint ROC scores (area under the curve)

Event = temperature in upper tercile Event = temperature in upper quintile


BACKRUNS verification: precipitation, reliability: selectable by region/initial date/lead time:




European multi-model

GloSea GloSea+ECMWF multi-modelprobability of well-above/below temperatures, Jun-Jul-Aug

well above

well below


multi-model combination, calibration

prediction of decile ‘extremes’


Discriminant Analysis for multi-model combining/calibration

Generalised squared distance;

where,t category of interest (eg. upper quintile)x vector of predictor variables (elements = eg. t2m from different models) mt corresponding vector of mean predicted values (from hindcasts) when

category ‘t’ was observed (cross validated)S covariance matrix for different predictors (models)

Predicted probability of category ‘t’ given by…

Take average of pt(x) over all ensemble members

)()()(2 t1

t mxSmxx tD

))(5.0exp()( 2 xx tt Dp


Area with BS > 95% significant for 2m tempouter quintiles DEMETER (1959-1999)

Black: Raw Ensemble GloSea. Red: Raw Ensemble Multi-model Blue: Discriminant GloSea. Green: Discriminant Multi-model.

All seasons

0,1,2,3 months lead


Area with ROC skill > 95% significant 2m tempDEMETER (1959-1999)

Black: Raw Ensemble GloSea. Red: Raw Ensemble Multi-model Blue: Discriminant GloSea. Green: Discriminant Multi-model.

All seasons

0,1,2,3 months lead


European multi-model

GloSea GloSea calibrated

probability of well-above/below temperatures, Jun-Jul-Aug

well above

well below


ROC significance for calibrated forecasts of outer deciles: temperature, Europe, JAS (1959-1999)

>95% significance

GloSea

GloSea+ system2+ Meteo-France

Multi-model


ROC significance for calibrated forecasts of outer deciles: temperature, Europe, AMJ (1959-1999)

>95% significance

GloSea

GloSea+ system2+ Meteo-France

Multi-model


Summary + key plans

New website products – July 05 Multi-model

Outer-quintiles probabilities

Expanded verification information

Next year Calibrated products

Higher order extremes (deciles) Need to resolve trade off between reliability and resolution

EU ENSEMBLES project Investigate multi-annual/decadal skill

Compare ensemble initialisation strategies

Next generation GloSea Investigate/prepare new Hadley Centre climate model (HadGEM1) for use in seasonal

prediction


ENSEMBLES project: Met Office seasonal/multi-annual/decadal runs

ModelDePreSys(HadCM3)

Currentoper. range

decadal

assimilationmethod

GloSea(HadCM3)

Seasonal(6months)

Conventional (OI type)

calibrated anomalies

9-ensembleexperiments1991-2001

pert. ODA

pert. phys.

lagged avge

pert. phys.

lagged avge

Hindcasts

period: 1991 - 2001

GloSea:->7m: 1st/15th May/Nov 1st June/Dec ->14m: 1st May/June/Nov/Dec-> 10y: 1st May 1964, 1994

DePreSys:

-> 10y: 1st May/Nov (all years)


Questions & Answers


Seasonal forecasting activities at the Met Office

Long-range Forecasting Group, Hadley Centre

Presenter: Richard Graham

ECMWF Forecast Products Users Meeting, 15-17th June 2005


Outer quintile temperature forecasts and verif FMA 05

P(outer quintile)

>40%

IRI verification


Seasonal forecasting system : GloSea

• Enhanced version of the Hadley Centre Climate model HadCM3

• Six month real-time 41 member ocean atmosphere global forecast ensemble

• 5 ocean analyses from perturbed wind stresses

• Ocean analyses further perturbed with instantaneous SST perturbations

• hindcast (‘BACKRUN’) period, 1987-present (1987-2002 used here)

Ocean Analysis - 5 member ensemble

Real - TimeForecast

41 memberensemble

Retrospective Forecasts - 15 member ensemble

1987 20041988

Atmosphere NWP/re- analyses

15 member


BACKRUNS verification: temperature, reliability: selectable by region/initial date/lead time:




Discriminant Analysis

-used to calibrate dynamical forecasts and combine ensembles from multi-model-Multi-variate: different predictors (models) considered together, relative skill taken into account

Pdfs for above, middle and below terciles, given value of a predictor (illustration)

Calculated using historical data and hindcasts

Page 1© Crown copyright 2004 Seasonal forecasting activities at the Met Office Long-range Forecasting Group, Hadley Centre Presenter: Richard Graham ECMWF.

Documents

crown copyright

discriminant multimodel

suite multimodel

upper quintile slide

significance glosea

member ensemble slide

discriminant glosea

raw ensemble multimodel