IUKWC Workshop November 2016: Developing Hydro-climatic Services for Water Security Session 2.2 AK Sahai

1. Extended Range Prediction
Activities of IITM
Dr. A. K. Sahai
Extended Range Prediction Group, Monsoon Mission
Indian Institute of Tropical Meteorology,
Pune – 411 008, INDIA
E-mail: sahai@tropmet.res.in

2. Why we need ERP
Seasonal rainfall anomalies are nearly homogeneous over Indian region during extreme
monsoon years (droughts/floods).
But, mostly (~70%) monsoon years are normal and during normal years the rainfall
anomalies are inhomogeneous over the country, contributing to large degree of spatial
variability !!!
Adding to this is the variability of rainfall on temporal scales.............
Drought (2002) Flood (1961) Normal (1998)
Seasonal JJAS
rainfall anomaly
during drought,
flood and normal
years
Drought (2002) Flood (1961) Normal (1998)
Although the prediction of a ‘normal’ all India
rainfall may have a comfort factor, it may not be
useful for agricultural planning.
Therefore, in addition to the seasonal mean All
India rainfall, we need to predict some aspects of
monsoon 3-4 weeks in advance on a relatively
smaller spatial scale that will be useful for
farmers.
Spatial Variability Temporal Variability

3. Phase1:
Peninsular India
Phase2:
Central India
Phase3:
Central India
Phase4:
North India
Phase5:
Foothills
Phase6:
South IO
Phase7:
Indian Ocean
Phase8:
Southern tip
Composite Rainfall anomalies in different MISO phases
Predictability of MISO provides hope for extended range prediction:

4. Examples of evolution of MISO indices

5. Statistical Extended Range Prediction: The start of a
New ERA
• As seasonal prediction has limited applicability for short
term planning, the extended range prediction beyond
weather scale started growing prominence.
• The temporal evolution of atmospheric variables may be
considered as a continuous process which can have a
mixing of processes arising from multiple scales of motion
and some inherent chaotic component.
• The goal is to separate the scale motions of our interest
(signal) from other scales and the chaotic part (noise).
• The larger the signal to noise ratio captured by a particular
statistical method for a particular variable, the better the
method of prediction.

6. Different methods of Statistical Extended Range
Prediction
Prediction of ISO :
• Based on Principal Oscillation Pattern (POP): Von Storch and Xu
(1990), Clim. Dyn
• based on Singular Vector Decomposition: Waliser et al., 1999. J.
Clim.
• Based on EOF only : Lo and Hendon, 2000, MWR
• Based on EOF and Multiple-regression: Jones et al., 2004
• Based on EOF and Analog techniques: Xavier and Goswami.,
2007, MWR
• Based on Self-Organizing Map and Analogues: Chattopadhyay,
Sahai and Goswami., 2008, JAS.

7. Early Developments FOR Extended Range
Prediction over Indian Subcontinents
• Initial Gain was realized for flood forecasting .
A study by Webster et al., 2004, BAMS
(Limitation: End Point Problem)
• Goswami et al., 2007 used a EOF based
analogue method for prediction of OLR
(Limitation: at times OLR and Rainfall mismatch; linear
method)

8. Initial Attempt:
A SOM based Non-Linear Analogue Technique
Sahai et al., 2006
START OF OUR JOURNEY TOWARDS DEVELOPMENT
OF EXTENDED RANGE PREDICTION SYSTEM

9. JAS 2008

10. Towards Real-Time Probabilistic ISO prediction
Presented in IMSP 2009

11. Well-predictedNot so well-predicted P4
JGR, 2014

12. ..Though Statistical Extended Range
Forecast gives a reliable forecasts..
Unaccounted error may arise due to
variability in predictability skill of
predictor themselves.
Moving towards Dynamical Prediction….

13. CFS1 CFS2
SOM
2005, P4

14. Time Line of development of IITM ERPS using CFSv2
2011: Ensemble Prediction System developed, [Abhilash etal., 2014, IJOC]
2012: Bias Correction of CFS forecasted SST implemented
[Abhilash etal., 2014, ASL; Sahai etal., 2013, Cur. Sci.]
2013: High Resolution CFST382 implemented
[Sahai etal., 2014, CD;Borah etal, 2014, IJOC]
2014: CFS based Grand EPS Implemented
[Abhilash etal., 2015, JAMC; Sahai etal., 2015, Cur. Sci]
2015: Forecast for winter and other seasons started
2016: Forecast for Heat Waves started
[Applications: Onset Prediction: Joseph etal, 2014, JC; Uttrakhand Heavy Rainfall: Joseph etal,
2014, CD; Skill of CFST126: Abhilash etal., 2014, CD; Comparison 2013 and 2014 June extremes:
Joseph etal., QJRMS, 2015; Prediction skill of MJO: Sahai et al., 2016, IITM-RR]

15. Zonal wind at 1000 hPa from (a) Analysis,
(b) Perturbed initial condition and (c)
Perturbation.
Development of Perturbation Technique
Each ensemble member is generated by slightly
perturbing the initial atmospheric conditions
with a random matrix (random number at each
grid point) generated from a random seed.
Fraction of the 24 hour tendency of different
model variables are added to or subtracted from
the unperturbed analysis with random
perturbation between -1 and +1 times the 24
hour tendency so that the perturbation follow
Gaussian distribution.
The perturbed IC,
X´x,y,z,t = Xx,y,z,t – n [r ΔXx,y,z,t]
where, ΔX = Xx,y,z,t – Xx,y,z,t-1 ; r -> taken from a
random matrix and lies between -1 and +1; n ->
tuning factor such that 0≤n≤1
We perturb the wind, temperature and moisture
fields and the amplitude of perturbation for all
variables are scaled according to the magnitude
of each variable at a given vertical level.
Abhilash et al. (2012, IITM Res Rep); Abhilash et al. (2014, Int . J. Climatol.)

16. SST Bias from Long Simulation
Development of Bias-correction Technique

17. Spatial correlation between forecasted
rainfall and observed rainfall
Implementation of High Resolution Version
Pentad correlation skill and
RMSE w.r.t. observations for
CFST126 and CFST382
Sahai et al. 2015, Clim Dyn
CFS126-IMD CFS382-IMD
OBS
IMD
Climatological Rainfall bias

18. OBS
IMD
CFS126-IMD GFS126-IMD CFS382-IMD MME-IMD
Deterministic prediction Skill
Probabilistic Prediction Skill
Abhilash et al. 2015, JAMC; BAMS
Seasonal mean and difference from OBS
CFS126
SME
CFS382
SME
GFSbc
SME
MME
P1 Lead 0.87 0.84 0.87 0.89
P2 Lead 0.85 0.85 0.81 0.88
P3 Lead 0.83 0.82 0.78 0.87
P4 Lead 0.79 0.80 0.76 0.86
Spatial pattern correlation over 40°E-120°E
and 40°S-40°N
MME has been formulated using 21
ensembles of GFSbc, 11 ensembles of CFS126
and 11 ensembles of CFS382.
Hence, total 43 ensemble members were
produced independently from 3 variants of
CFS model to generate the CGEPS and
forecast consensus is done by making simple
average among the members.
Development of MME

19. RMSE and spread of MISO indices
Considerable improvement in MME is contributed from the increased spread,
which overcomes the under-dispersive nature of the individual models in EPS.
Abhilash et al. 2015, JAMC; BAMS
Bivariate RMSE: RMSE w.r.t.
observation
Bivariate Spread: Std. Dev of
iindividual models w.r.t.
Ensemble mean

20. Relook: Why MME?
Comparison of IITM-ERPS
with ECMWF

21. Skill of CFST126/T382 is much less
than ECMWF in longer leads
Comparison of IITM-ERPS with ECMWF
Skill improved due to
bias correction
How to improve the skill and make it
comparable to that of ECMWF?

23. 0531: Low Pressure System (LPS) over southern
tip of peninsula is likely to intensify and move
towards Oman coast. This system may dissipate
around 11th June and till then the monsoon
activity will be weaker than normal over India.
0725: It was forecasted that Monsoon activity
will be normal and there is a possibility that it
may enter in the break phase around 10th Aug.
0620: There will be a large scale reduction
of rainfall during 1st half of July.
0710: Large scale monsoon activity is
expected to increase by the end of July
resulting in revival of monsoon.
0903: A fresh spell of good rainfall will propagate
from Indian ocean to southern peninsula around 20th
September and may reach central India around 25th
September.
0605: It is likely that by 17th June the offshore
trough along the west coast will be established
and within one week after that, monsoon may
reach central India as a feeble current.
Prediction of 2015 SW monsoon season

24. OBS
Verification based on 30th June IC
Monitoring and prediction of MISO during 2016

25. Joseph et al. 2015, J Climate

26. MOK forecast of 2014 based on IC: 0516
Key points from the forecast:
 A low pressure system might form over Bay of
Bengal around 25 May 2014 and move
northwards.
 South-west monsoon of 2014 would make its
onset over Kerala on 04 June.
 However, the strengthening and progression of
the monsoon might be slackened till 15 June due
to the presence of a low-level anticyclonic
circulation over central India. Afterwards,
monsoon might strengthen and progress.
Monsoon Onset over Kerala

28. Progression of ISM 2016
IC: 0605
Peculiar progression of ISM from
eastern side (via Bay of Bengal
branch) around 17 June was
predicted by MME.

30. The revival of monsoon due to the formation of a LPS around 13-14
September and subsequent westward movement was forecasted well
from 08 September IC.
Withdrawal of ISM 2016 (IC: 0908)

32. Year June Rainfall Departure from Mean
2001 219.0 35.6
2002 180.1 9.4
2003 179.9 9.8
2004 158.7 -0.8
2005 143.2 -9.5
2006 141.8 -12.7
2007 192.5 18.5
2008 202.0 24.3
2009 85.7 -47.2
2010 138.1 -15.6
2011 183.5 12.2
2012 117.8 -28.0
2013 219.8 34.4
2014 92.4 -43.5
Observed June Rainfall during 2001-14
Joseph et al. 2016, QJRMS

33. The model has remarkable skill in predicting the June extremes !!!
ERP of June extremes by CGEPS MME
Joseph et al. 2016, QJRMS

35. IC: 05 JunePrediction of Heavy Rainfall Events
Uttarakhand event in
June 2013
Evolution of Potential Vorticity (PV; x10-7 s-1) anomalies at 700 hPa and mean sea level pressure
Forecast
OBS
Joseph et al 2016, Clim. Dyn.

36. Low Pressure System (LPS) over southern tip of
peninsula is likely to intensify and move towards
Oman coast. This system may dissipate around 11th
June and till then the monsoon activity will be
weaker than normal over India.
Cyclone “Ashobaa” during Onset
phase of 2015 monsoon
IC: 0531
Prediction of Cyclogenesis
MME
OBS

37. MME
OBS
IC: 11 May
Cyclone Roanu in May 2016
Prediction of Cyclogenesis

39. OBS model
MJO Forecast during March-April 2009 period

40. Prediction of Heat Waves

41. Heat Wave in May 2016
Observed
Heat Index
Predicted Heat
Index
IC: 0516
Prediction of Heat Waves

42. Prediction of North-East Monsoon (NEM)

43. Hindcast Skill for Post Monsoon/NEM
NEM
SLK
NEM and SLK region Climatological rainfall
NEM (SLK) region exhibits
useful skill up to 3(2) Pentad
Hindcast Skill

44. Area averaged rainfall over NEM region during 2015 predicted by MME
The CGMME system
predicted the above normal
rainfall activity over Chennai
and NEM region well in
advance. The CGMME
system able to capture this
high impact continuous
rainfall activity during the
last week of November and
first week of December
around Chennai region from
4th pentad lead

45. Application to Model Output
• Bias Correction and
Downscaling
45

46. SOM based Bias correction and Downscaling: Application to ERP of Mahanadi flood in September 2011
Sahai et al 2016, Clim. Dyn.
0.30
0.47
0.17
0.31
0.22
0.60
0.49
0.47

47. Critical Success Index of Predicting
Anomalies
• CSI of predicting dry
anomalies in precipitation
forecast from CFSv2 and
IITM_ensemble is higher in
northwestern region whereas
lower in Himalayan range and
southern peninsula
• Runoff and Soil moisture
shows higher CSI than
precipitation and
temperature due to
persistence in Initial
conditions.
Precipitation
Temperature
Runoff
Root Zone
Soil moisture

48. Prediction of Drought : 15th July 2009
Better predictability of precipitation and temperature from bias-corrected IITM
ensemble along with persistence in Initial hydrologic condition lead to better forecast
of anomalies in runoff and root-zone soil moisture
Precipitation Temperature Runoff
Root zone
Soil Moisture

49. ERP has:
• Application in agricultural
• Application on health sector
• Application on energy sector
• Application in urban/rural planning
• Application in extremes prediction
• Application in dam/watershed management
• Application in insurance/reinsurance
The current version of ERP System has been transferred to IMD.

50. picture from Nasa
Thank You
http://www.tropmet.res.in/erpas/index.php