Presentation of M Rajeevan

R Recent t Developments D l t i in S Seasonal l
Prediction of Indian monsoon rainfall
using Coupled Climate Models
M Rajeevan
National Atmospheric Atmospheric Research Research Laboratory
Laboratory
Department of Space

Introduction
IMD’s seasonal forecasts of Indian monsoon
rainfall are mainly based on statistical models.
Limited skill
Limited scope
Constraints due to epochal variation of predictor predictor- predictor
rainfall relationships
Limited understanding g of observed monsoon
variability
Users demand forecasts on higher g
resolution

Introduction
However, during the past 10 years, dynamical
models (AGCMs) ( ) have been extensively y used
for experimental seasonal forecasts.
IMD Pune
IITM Pune
Pune
IISc Bangalore
CMMACS Bangalore
SAC Ahmedabad
CDAC Pune
NAL Bangalore
IIT Delhi
NCEP coupled climate forecast system (CFS)
IITM Pune running on operational basis
CDAC Pune on experimental basis
NCEP coupled climate forecast system (CFS)

Dynamical Models
Evaluation of dynamical models is being
done from the AMIP AMIP-1 1 era
Gadgil and Sajini 1998
Kang et et al. al. 2002
2002
Wang et al. 2004
Seasonal prediction of Indian Monsoon
(SPIM) project project- sponsored by Department of
Science Science and Technology
Gadgil and Srinivasan (2010) Current Science
5 AGCMs AGCMs, period: period: 1985 1985-2004 1985 2004

Dynamical Models
Evaluation of coupled dynamical models
Preethi ee e et a al. 2009 009 o on DEMETER results esu s
Pattanaik and Kumar 2010, Yang et al.2008,
Drbohlav et al 2010 on NCEP CFS
Rajeevan and Nanjundiah 2010 on IPCC models
Very modest results - AGCMs vs Coupled
Models
Not Not better than than statistical statistical models
Stronger ENSO linkage, poor Indian Ocean
li linkage
k

EU EU- ENSEMBLES project p j
The Development of a European Multi Multi-model model Ensemble
System y for seasonal to inter inter-annual annual prediction p ( (DEMETER) )
project (Palmer et al. 2004) provided an ideal opportunity to
examine the simulation characteristics of the Indian summer
monsoon.
ENSEMBLE
ENSEMBLE-based based predictions of climate changes and their
impacts (ENSEMBLES) is an EU EU-funded funded integrated project
th that t intends i t d t to develop d l an ensemble bl prediction di ti system t f for
climate change based on the principal state state-of of-the the-art, art, high
resolution global models developed in Europe.
The project has been designated to produce for the first
time, an objective probabilistic estimate of uncertainty in
future climate climate at the seasonal to decadal and longer
timescales

Dealing with forecast uncertainty
● Model error is a major source of forecast uncertainty. Three
approaches to deal with model error are being investigated in
ENSEMBLES: multi multi-model, model model, stochastic physics and perturbed
parameters.
● The multi multi-model model is built from ECMWF, , Met Office, , Météo Météo-
France operational activities and the DEMETER experience at
IFM IFM-Kiel, Kiel, CERFACS and INGV.
● Perturbed parameter system stems from the decadal
prediction system (DePreSys) created at the Met Office.
● Stochastic Stochastic physics ph sics s system stem uses ses the the ECMWF ECMWF stochastic
stochastic
backscatter system developed for medium medium-range range forecasts.
● ~20 20,000 000 years years of of integrations.
integrations

ENSEMBLE Models
No models except the
Depresys used the flux
corrections.
Depresys ensemble
members are generated
by perturbing
parameters in the
physical
parameterization
schemes.

ENSEMEBLES models
Common hindcast period: 46 years (1960 (1960-2005) 2005)
For For each each year year, 77-month
7 month long long seasonal seasonal forecasts
starting from 1 st February, May May, , August and
November.
9E 9 Ensemble bl M Members b
Since DEMETER, the contributing seasonal
prediction p systems y have improved p in all aspects p
increase in resolution
better representation of sub sub-grid grid physical processes
land surface process process,
sea sea-ice ice
greenhouse gas boundary forcing
more widespread use of assimilation for ocean
initialization

Monsoon Seasonal Rainfall (June to September) in mm/day
Observed ENSEMBLES Multimodel

Both DEMETER and
ENSEMBLES predict more
rainfall over the equatorial
Indian Ocean compared to
observations.

Biases in the Multi model monsoon rainfall

Mean rainfall
(cm)
Coefficient
of Variation
CV (%) ( )
Observed 82.9 11.9 1.00
ECMWF 93 93.2 2 45 4.5 037 0.37
IFM-
GEOMAR
79.8 5.7 0.34
METEO
FRANCE
45.0 8.2 0.34
UKMO 80.0 12.2 0.39
CMCC-
INGV
91.0 5.4 0.39
DePreSys 65.6 8.9 0.27
ENSEMBL
ES MME
DEMETER
MME
(1960-2001)
75.8 5.7 0.45
76.0 4.3 0.28
Correlation
Coefficient
1960-2005

In comparison with DEMETER, ENSEMBLES results
improved in drought g yyears
like 1972, 1974 and 1979 and
excess monsoon year like 1961.
Problem remains for years like 1983 and 1997

In comparison with DEMETER, ENSEMBLES
MME has better prediction skill. Each of
ENSEMBLES models except Depresys
showed better skill than the DEMETER MME.

Probabilityy
1.2
1
0.8
0.6
04 0.4
0.2
Probability forecasts of Indian Monsoon Rainfall
Drought Excess
Normal
0
1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005
Year
Positive skill for probabilistic forecasts of droughts
BBrier i Skill SScore
Drought: 00.201 201
Excess : 0.060
Normal= -0.002

ENSEMBLES predict
an extended t d d warming i
into central Pacific
Ocean during g the
1997 El Nino.
Indian Ocean Dipole
(IOD) signal is
missing i i ffrom th the
model predictions

ENSEMBLES predict
an extended warming
into central Pacific
Ocean during 1983
monsoon season.

Correlation between observed and model predicted rainfall

Monsoon Teleconnections

Monsoon Tele-
connections
ENSO-Indian
monsoon
teleconnection has
weakened during
the recent decades
North Atlantic
North Atlantic
teleconnection has
strengthened
during the recent
decades

Role of north Atlantic Ocean
Rajeevan and Latha Sridhar, 2008,
Geophys. Res. Letters

ENSEMBLES could
not capture the
recent weakening of
the ENSO-Indian
ENSO Indian
monsoon
teleconnection.
Correctly captures
y p
the north Atlantic
teleconnection

ENSEMBLES could not capture the
ENSEMBLES could not capture the
recent weakening of the El Nino- Indian
monsoon relationship

Prediction skill:
1960-1979
PPrediction di ti skill: kill
1980-2005

Best of best IPCC models also showed excessive rainfall over
the equatorial Indian Ocean.
Rajee an and Nanj ndiah (2009) in “ C rrent Trends in
Rajeevan and Nanjundiah (2009), in “ Current Trends in
Science” Indian Academy of Sciences.

IPCC models showed colder SST bias
over the Tropical Indian Ocean

Too stronger coupling of SST-rainfall in the
IPCC coupled climate models
Rajeevan and Nanjundiah (2009) in “ Current Trends
Rajeevan and Nanjundiah (2009), in Current Trends
in Science” Indian Academy of Sciences.

Observations ENSEMBLES
Too stronger local air-sea coupling
over the Indian ocean in the models

Observations ENSEMBLES

Observations ENSEMBLES

Scope for further improvement
improvement
Initialization of land surface conditions
GHG forcing
Stratospheric Processes

GHG effect in seasonal forecasts
Constant GHG
Correlation = 0.29
Variable GHG
Correlation = 0.68
Francisco J Doblas-Reyes
ECMWF
WCRP workshop on seasonal
Prediction Prediction, Barcelona Barcelona, 2007

Th Thank k you