Prediction of Seasonal Mean Monsoon Rainfall

1
Prediction of Seasonal Mean Monsoon Rainfall

India Meteorological Department

• M.Rajeevan
• National Climate Centre
• India Meteorological Department
• Pune

2
Inter-annual variation of South-west Monsoon
rainfall
A Drought over the country is declared when the
all-India monsoon (June to September) rainfall is
less than 90 of Long period average and at least
20 of the country experiences rainfall
deficiency of 25 or more.
3
GDP and Indian Monsoon Rainfall

• Impact of a severe drought on GDP remains 2
  to 5 throughout, despite the substantial
  decrease in the contribution of agriculture to
  GDP over the five decades.

Gadgil and Gadgil (2006), Economic and Political
Weekly, Vol XLI no 47, p 4887-4895
4
Walkers Contributions

• Sir Gilbert Walker made significant contribution
  to long range forecasting research.
• He introduced the correlation and regression
  techniques and objective models.
• His research for global predictors led to the
  discovery of Southern Oscillation and North
  Atlantic Oscillation.
• His regression methods have been more or less
  followed by IMD for the operational work.

5
HISTORY OF IMD OPERATIONAL FORECASTS ALL
STATISTICAL MODELS
1924--1987 Forecast NW INDIA / PENINSULA
1988 16 PARA MODEL All India Deterministic
FORECASTS
1999 Forecasts for 3 REGIONS introduced
2003 8/10 Parameter Model for (Jun-Sept) July
F/C for India
2004 All India Forecast along with 4 Homogeneous
Regions of India
2007 New Statistical Models Introduced
16 P Model Continued till 2002
8/10 Para Models
6
Forecast Performance 1932-1987
Gadgil, Rajeevan and Nanjundiah (2005, Current
Science) Monsoon Prediction Why yet another
failure?
7
Verification of forecasts 1988-2006
8
Present Long Range Forecast Schedule
Seasonal (June to September) Rainfall
April June
July Rainfall
June
Seasonal (June to September) Rainfall
June
Monsoon Onset Date over Kerala
Mid-May
9
This year, IMD introduced new statistical models
as operational modelsRajeevan et al. (2007)
Climate Dynamics
10
IMDs new statistical models

• Specifically three major issues were addressed
  to improve the models.
• New Predictor data set
• A smaller but more physically linked predictor
  data set was used. Search for new predictors.
• New Model Development method
• Ensemble Method was used instead on relying on a
  single model.
• New Statistical Tools
• A New non-linear technique was adopted.

11
New list of predictors
12
S.No New Predictors Used for forecasts in
1 North Atlantic Sea Surface Temperature (December January) April and June
2 Equatorial SE Indian Ocean Sea Surface Temperature (February March) April and June
3 East Asia Mean Sea Level Pressure (February March) April and June
4 NW Europe Land Surface Air Temperatures (January) April
5 Equatorial Pacific Warm Water Volume (February March) April
6 Central Pacific (Nino 3.4) Sea Surface Temperature Tendency (MarAprMay) (DecJanFeb) June
7 North Atlantic Mean Sea Level Pressure (May) June
8 North Central Pacific wind at 1.5 Km above sea level (May) June
13
Role of North Atlantic Circulation and Indian
monsoon rainfall

• Chang et al (2001, J.Climate)
• Possible role of North Atlantic Circulation,
  mid-latitude jet stream, Eurasian Temperature etc
  on ENSO-Monsoon weakening
• Srivastava and Rajeevan (2004, Geophys.Res.Letters
  )
• Statistical relationship between SST and OLR over
  North Atlantic and Indian monson
• Goswami et al (2006, Geophys.Res.Letters)
• Relationship between North Atlantic Circulation
  and Indian summer monsoon on multi-decadal time
  scale
• Ding and Wang (2005, 2007, J.Climate)
• Coupled pattern of intra-seasonal variability
  between mid-latitude circulation and Indian
  summer monsoon

14
North Atlantic MSLP (May) and ISMR
15
Relationship with Tropical Pacific Warm Water
Volume (WWV)
Rajeevan and McPhaden, 2004, Geophysical Research
Letters
16
PPR is similar to Artificial Neural Network, but
training algorithm is much superior. We used R
Software for training the model.
Multiple Regression Method
Pursuit Projection Regression Method
Where, fm is the ridge function derived from the
data itself.
17
ENSEMBLE METHOD
Rajeevan et al. ( 2007) Climate Dynamics
18
(No Transcript)
19
Ensemble Multiple Regression Model
20
Rajeevan et al. ( 2007), Climate Dynamics
21
(No Transcript)
22
COMPARISON OF FORECASTS COMPARISON OF FORECASTS COMPARISON OF FORECASTS COMPARISON OF FORECASTS COMPARISON OF FORECASTS

YEAR RAINFALL IN OF NORMAL RAINFALL IN OF NORMAL RAINFALL IN OF NORMAL RAINFALL IN OF NORMAL
YEAR ACTUAL OPERATIONAL FORECAST NEW Model Forecast
2000 92 99 96
2001 91 98 92
2002 81 101 91
2003 102 98 106
2004 87 100 85
2005 99 98 102
2006 100 92 95
23
Relative role of predictors
24
Tropical vs Extra-Tropical Predictors
25
Tropical vs Extra-Tropical
26
Tropical Vs Extra-Tropical parameters
27
Forecasts for 2007
28
2007 monsoon performance
29
2007 monsoon performance
30
ECMWF Forecast for 31 July
A very weak monsoon
31
DYNAMICAL PREDICTION OF INDIAN MONSOON

• Many model validation programs undertaken (MONEG,
  AMIP I II, PROVOST, DEMETER, CLIVAR Monsoon
  Panel, SPIM).
• Dynamical models have shown serious problems in
  predicting / simulating inter-annual variability
  of Indian monsoon.
• Statistical models in comparison perform better,
  at least for all-India summer monsoon rainfall.

32
SFM (ECPC) results

• Wang et al (2004, J.Climate) using the results
  of 11 AGCMs reported serious problems in
  simulations over the Indian region during the
  1997-1998 El Nino episode (CLIVAR International
  monsoon Panel).

SFM showed serious problems in two years, 1994
and 1997.
33
EU-Ensembles project

• 9 member ensembles
• May initial conditions, ERA-40 atmosphere and
  soil initial conditions
• Realistic boundary forcings GHGs, aerosols,
  solar forcing etc
• Hindcast production period for 1991-2001
• 4 coupled models considered
• ARPEGE /OPA- CNRM
• ECHAM/OM- MPI
• GloSea- UKMO
• IFS/HOPE - ECMWF

34
Anomalous year 1994
All India Monsoon Rainfall 110
35
Scatter plots
METEOFRANCE
ECHAM
1994
1994
1997
1997
UK Met Office
ECMWF
1994
1994
1997
1997
36
(No Transcript)
37
NE Monsoon (Oct-Dec) rainfall

• NE monsoon rainfall during the period
  October-December is also important for South
  Peninsula, especially for Tamil Nadu.
• NE Monsoon is not getting attention that it
  deserves.
• Some work was done under IMD/IITM collaboration

38
NE Monsoon Rainfall Time Series
39
NE Monsoon and ENSO
Pankaj Kumar et al. 2006. Climate Dynamics
40
NE Monsoon Hindcasts DEMETER Results
41
DEMETER RESULTS
ECMWF MODEL
UK MET OFFICE MODEL
42
NE Monsoon Statistical forecasts
43
NE Monsoon Statistical forecasts
Based on the research done at IMD/IITM, a
statistical model was developed and an
operational forecast was issued for the 2006 NE
Monsoon Season.
44
Conclusions

• IMD has been issuing operational long range
  forecasts using statistical models.
• IMD is also using a dynamical model on
  experimental basis.
• IMD is working on developing Extended range
  forecast models to meet the requirement of users.
• Dynamical models have shown better skill for
  forecasts of NE monsoon rainfall.
• IMD also exploring possibility of preparing
  seasonal forecasts for seasonal mean
  temperatures.
• I believe better understanding of monsoon
  variability and predictability is very crucial to
  improve the seasonal forecasts.

45
Thank you