12

Operational and Experimental Long Range Prediction

Forecasts

D. S. Pai1, Divya Surendran1, Maduri Musale1, Sreejith O. P.1 & Suryachadra Rao2.

1
India Meteorological Department, Pune
2
Indian Institute of Tropical Meteorology, Pune

This chapter discusses the various operational forecasts issued by India

Meteorological Department (IMD) for the monthly and seasonal rainfall over India and date
of monsoon onset over Kerala for the 2019 southwest monsoon season and its verification.
The chapter also discusses the experimental forecasts generated using IMD’s dynamical
global forecasting system based on coupled forecasting system (CFS) and experimental
obtained from various climate research centers within the country and abroad.

12.1. Introduction
Every year, India Meteorological Department (IMD) issues operational monthly and
seasonal forecasts for the southwest monsoon rainfall using models based on the latest
statistical techniques with useful skill (Rajeevan et al. 2007, Pai et al. 2011). In 2019, the
long range forecast for the 2019 southwest monsoon rainfall was issued in 3 stages. The
first stage long range forecast issued on 15h April consisted of only forecast for season
(June-September) rainfall over the country as a whole. In the second stage (31st May), along
with the update for the April forecast, forecast for season rainfall over the four broad
geographical regions (northwest India, central India, south Peninsula and northeast India)
and that for monthly rainfall over the country as a whole for the months of July and August
were issued. In the 3rd stage (1st August), forecast for the rainfall during the second half of
the monsoon season for the country as a whole, was issued.

184
 Since 2012, as additional forecast guidance, IMD started to use the experimental
forecasts for the monsoon rainfall generated by the dynamical model approach developed by
Indian Institute of Tropical Meteorology (IITM), Pune. The present dynamical model
forecasting system is based on the global climate forecasting system version 2 (CFSv2). The
CFSv2 is a fully coupled general circulation model (CGCM) implemented by IITM under
Monsoon Mission project launched by the Ministry of Earth Sciences (MoES) (Saha et al,
2014, Ramu et al., 2016). The global monthly and season forecasts for rainfall and
temperature prepared using CFS model is updated 15th of every month is now available
through IMD, Pune website (www.imdpune.gov.in)
A brief description of IMD‟s operational statistical and experimental dynamical
forecasting systems is discussed in this chapter along with the verification of the forecasts
generated by these forecasting systems. The experimental forecasts for the seasonal
rainfall over the country using the Monsoon Mission CFS (MM CFS) as well as that from
various national and international research institutes obtained by IMD as guidance before
issuing operational forecasts have also been discussed.

12.2 Models Used

12.2.1 Statistical Ensemble Forecasting System for the Seasonal Rainfall over the
Country as a Whole
The statistical ensemble forecasting system (SEFS) was used for the forecast for the
season rainfall over the country as a whole. For this a set of 8 predictors (Table-12.1) that
having stable and strong physical linkage with the Indian south-west monsoon rainfall is
used. The geographical domains of the predictors are shown in the Fig.12.1. For the April
SEFS, first 5 predictors listed in the Table-12.1 are used. For June SEFS, the last 6
predictors are used that include 3 predictors used for April forecast. The standard errors of
the 5–parameter and 6- parameter SEFSs were taken as ±5% and ±4% respectively. A
schematic diagram of the statistical ensemble forecasting system is shown in the Fig.12.2.
As depicted in the Fig.12.2, the forecast for the seasonal rainfall over the country as a whole
is computed as the ensemble average of best few models out of all possible models
constructed using two statistical methods; multiple regression (MR) technique and projection
pursuit regression (PPR) - a nonlinear regression technique. In each case, models were
constructed using all possible combination of predictors. Using „n‟ predictors, it is possible to
create (2n-1) combination of the predictors and therefore as many number of models. Thus
with 5 (6) predictors respectively for April (June) SEFS, it is possible to construct 31 (63)
models. Using sliding fixed training window (of optimum period of 23 years) period,
independent forecasts were prepared by all possible models for the period 1981-2018.

185
 Performance of the April & June SEFS for the independent test period of 1981-2018
is shown in the Fig.12.3a & 3b respectively. The RMSEs of the April & June SEFS for the
period 1981-2018 are 6.41% of LPA & 6.27% of LPA respectively (see Table-12.2). The
C.C. between observed and forecast rainfall of the April & June SEFS for the period 1981-
2018 are 0.70 & 0.71.

Table 12.1 : Details of the 8 predictors used for the new ensemble forecast system

Used for Correlation

No. Predictor forecasts Coefficient
in (1981-2010)
Europe Land Surface Air
1. April 0.42
Temperature Anomaly (January)
Equatorial Pacific Warm Water
2. April -0.35
Volume Anomaly (February + March)
SST Gradient Between Northwest
April and
3. Pacific and Northwest Atlantic 0.48
June
(December +January)
Equatorial SE India Ocean SST April and
4. 0.51
(FEB) June
East Asia MSLP April and
5. 0.51
(FEB + MAR) June
NINO 3.4 SST
6. June -0.45
(MAM+(MAM-DJF) Tendency)
7. North Atlantic MSLP (MAY) June -0.48
North Central Pacific Zonal Wind
8. June -0.57
Gradient 850 hPa (MAY)

Fig.12.1. Geographical domains of the predictors used in the statistical ensemble

forecasting system for the seasonal rainfall forecast over the country as a whole.

186
 In addition to the quantitative forecast, the ensemble forecasting system has also
been used to generate a five category probabilistic rainfall forecast based on the forecast
error distribution of the ensemble forecasting system. The five rainfall categories defined
based on the observed data for the period 1901-2005 are deficient (< 90% of LPA), below
normal (90-96% of LPA), normal (96-104% of LPA), above normal (104-110% of LPA) and
excess (> 110% of LPA). The climatological probabilities of these five categories are 16%,
17%, 33%, 16% & 17% respectively. The five category probability forecast is prepared using
normal probability distribution with the ensemble average of the forecast from the ensemble
forecasting system as the mean and the RMSE of the independent test period as the
standard deviation. For verification purpose, the most probable category is one that has
highest forecast probability compared to its climatological value. A forecast validating within
the same category was considered as “correct (C)”, within one category as “usable (U)” and
beyond one category as “unusable/not usable (NU)”.
The probabilistic forecast for the independent test period of 1981-2018 obtained
based on the April SEFS showed that the model forecasted correct category during 12 years
(37%), 1 category out during 19 years (50%) and 2 categories out during 5 years (13%).
The corresponding probabilistic forecast obtained based on the June SEFS showed that the
model forecasted correct category during 18 years (47%), 1 category out during 12 years
(37%) and 2 categories out during 6 years (16%). The 5 category probability forecasts
based on the April and June SEFS for the 2019 monsoon season are given in the Table-
12.3.

Fig.12.2. A Schematic
diagram of the new
ensemble forecasting
system for the
monsoon season
rainfall over the
country as a whole.
The average of the
ensemble forecasts
from best out all
possible MR (multiple
regression) and PPR
(projection pursuit
regression) models
gives the final
forecast.

187
 Fig.12.3a: Performance of the April ensemble forecasting system for the
seasonal rainfall over the country as whole for the period 1981-2018.

Fig.12.3b : Performance of the June ensemble forecasting system for the

seasonal rainfall over the country as a whole for the period 1981-2018.

12.2.2 Principal Component Regression (PCR) Models Used for the Other Operational
Forecasts
Separate PCR models were used for generating forecasts for the rainfall over the
country as a whole during the peak rainfall months (July and August) and second half of the
monsoon season (August + September) and for the season (June – September) rainfall over
the four broad geographical regions of the country. The details of the various PCR models
are given in the Table-12.4 in brief.

188
 These models used “moving training period” method for the forecasting of rainfall for
a reference year. In this method, the forecast for a reference year is prepared using fixed
number of years (constant training window period) just prior to the reference year as the
training period. As seen in the Table-12.4, models for forecasting rainfall over country as
whole (monthly, second half of the season & season) used 23 yrs and that for the season
rainfall over the four broad geographical regions used 30 yrs as the constant training window
period. In the PCR method, principal component analysis (PCA) is carried out on the
predictor data of the training period and first few PCs that explain 80% of the total variability
of the predictors during that period were retained. The retained PCs were than related
against the predictor series for the same period using the multiple linear regression method.
Scores of the selected PCs for the reference year were then calculated using the PC loading
matrix and predictor values for the reference year. These score values along with the
coefficients of the trained regression equation were used for calculating the predictor value
for the reference year. In this way, rainfall during the second half of the southwest monsoon
season over the country as a whole was predicted for the independent test period.
The skills of the models are expressed in terms of correlation coefficient (C.C) and
root mean square error (RMSE) between the model forecast and actual rainfall anomaly
during the independent test period and are shown in the last two columns of the Table-12.2.
As seen in the Table-12.2, the skills of the models for the 4 broad regions are relatively less
than that of the models for the country as a whole. Among the monthly forecast models,
model for August rainfall over country as a whole showed least skill. Among the models for
the season rainfall over the broad regions, model for Northwest India showed highest skill
and that for Central India showed lowest skill.

Table-12.2: Details of the operational models used for various monthly and seasonal
forecasts for the Indian summer monsoon rainfall
CC. Actual Root Mean
Model
Forecast Forecast Vs Forecast Square Error
(Training Window
Period Region Rainfall (RMSE) in % of
Period)
(Period) LPA (Period)
June to All India 5-P Statistical Ensemble 0.70 6.41
September Forecast System (SEFS) (1981-2018) (1981-2018)
(23 yrs)
June to All India 6-P (SEFS) 0.71 6.27
September (23 yrs) (1981-2018) (1981-2018)
July All India 6 –P Principal Component 0.62 10.79
Regression (PCR) (23 yrs) (1981-2018) (1981-2018)

August All India 5-P PCR 0.29 11.31

(23 yrs) (1998-2018) (1981-2018)
September All India 5-P PCR 0.66 15.50

189
 (23 yrs) (1981-2018) (1981-2018)
August- All India 5-P PCR 0.56 11.11
September (23 yrs) (1981-2018) (1981-2018)
June to Northwest 5-P PCR 0.54 12.89
September India (30 yrs) (1988-2018) (1988-2018)
June to Northeast 5-P PCR 0.58 10.76
September India (30 yrs) (1988-2018) (1988-2018)
June to Central 5-P PCR 0.38 12.09
September India (30 yrs) (1988-2018) (1988-2018)
June to South 6-P PCR (30 yrs) 0.46 13.34
September Peninsula (1988-2018) (1988-2018)
The PCR models were also used to generate Tercile (3 category) probability
forecasts based on the respective model forecast error distributions. For this purpose, the
tercile rainfall categories with equal climatological probabilities (33.33% each) were defined
based on the data for the period 1951-2010. The tercile probability forecast was prepared
using normal probability distribution with the forecast from the PCR model as the mean and
the model standard error during the training period as the standard deviation.
The tercile probability forecasts based on various PCR models for the 2019 monsoon
season are given in the Table-12.4a & Table-12.4b.

12.2.3 Operational Forecast Model for the Date of Monsoon Onset over Kerala
An indigenously developed statistical model (Pai and Rajeevan, 2009) was used for
preparing the operational forecast of the onset of monsoon over Kerala. The model based on
6 predictors used the principal component regression (PCR) method for its construction.
Independent forecasts were derived using the sliding fixed window period of length 22 years.
The model for 2019 was trained using data for the period 1997-2018. Fig.12.4 shows the
performance of the forecast for the period 1997-2018. The RMSE of the model is about 4
days.

Fig.12.4: Performance of the April ensemble forecasting system for the seasonal
rainfall over the country as whole for the period 1997-2018.
190
 Table-12.3: The 5 category probability forecasts for the 2018 monsoon season rainfall
over the country as a whole are given below.

Forecast Probability (%) Climatological

Rainfall Range
Category Probability
(% of LPA) April SEFS June SEFS (%)
Deficient Less than 90 17 15 16
Below Normal 90 - 96 32 32 17
Normal 96 -104 39 41 33
Above Normal 104 -110 10 10 16
Excess more than 110 2 2 17

Table-12.4a: The tercile (3) category probability forecasts for the monsoon monthly
(July and August) and that for the second half of the monsoon season (August+
September) over the country as a whole based on the PCR models.
July Model August Model
Category Rainfall Range Forecast Rainfall Range Forecast
(% of LPA) Probability (%) (% of LPA) Probability (%)
Below Normal <94 51 <94 33
Normal 94 -106 36 94 -106 39
Above Normal >106 13 >106 28

Table-12.4b: The tercile (3) category probability forecasts for the monsoon season
(June to September) rainfall over the four broad geographical regions based on the
PCR models.

NW India Central India South Peninsula Northeast India

Rainfall Foreca Foreca

Range Forecast Range Range Forecast Range
Category st st
(% of Probabilit (% of (% of Probabilit (% of
Probabi Probabi
LPA) y (%) LPA) LPA) y (%) LPA)
lity (%) lity (%)
Below
<92 43 <94 29 <93 38 <95 66
Normal
92- 93-
Normal 45 94-106 40 42 95-105 26
108 107
Above
>108 12 >106 31 >107 20 >105 08
Normal

12.3 Verification of Operational Forecasts

191
 Based on an indigenously developed statistical model, it was predicted on 15th May
2019 that monsoon will set in over Kerala on 6th June with a model error of ±4 days. The
actual monsoon onset over Kerala was on 8th June and therefore the forecast was correct.
Table-12.5 gives the summary of the various operational long range forecasts issued
for the 2019 Southwest monsoon rainfall along with the realized rainfalls
The first stage forecast for the season (June-September) rainfall over the country as
a whole issued in April was 96% of LPA with a model error of ± 5% of LPA. The update
issued in May for this forecast was (96% of LPA) with a model error of ± 4% of LPA. The
actual season rainfall for the country as a whole was 110% of LPA, which is 12% of LPA
more than the April and June forecasts respectively. Thus, the both the forecasts were not
within upper forecast limits and were underestimated the rainfall value.
Considering the four broad geographical regions of India, the forecasts issued in May
for the season rainfall over northwest India, Central India, northeast India and South
Peninsula were 94%, 100%, 91% & 97% of the LPA respectively all with model errors of ±
8%. The actual rainfalls over northwest India, Central India, Northeast India and South
Peninsula were 98%, 129%, 88% and 116% of the LPA respectively. Thus, Northwest India
and Northeast India are realized to be within the stipulated forecast error ranges. The
forecast for the seasonal rainfall over central India and south Peninsula was underestimated
to the actual value.
The forecast for the second half of the monsoon season (August –September) for the
country as a whole was 100% with a model error of 8% of LPA against the actual rainfall of
130% of LPA. Thus, the forecast for the rainfall during the second half of the monsoon
season over the country as a whole was also underestimate to the actual rainfall and was
not correct.
The forecasts for the monthly rainfall over the country as a whole for the months of
July & August issued in June were 95% & 99% of LPA respectively with a model error of ±
9%. The actual monthly rainfalls during July and August were 105% & 115% of LPA
respectively. Thus, the forecast for the monthly rainfall over the country as a whole for the
months of July & August was underestimated to the actual value.

Table 12.5 : Performance of the operational forecast issued for the 2019 southwest
monsoon rainfall.
Forecast (% of LPA) Actual
Region Period Rainfall
15thApril 31st May 1st August
(% of LPA)
All India June to September 96± 5 96± 4 96 110
Northwest India June to September 94± 8 98
Central India June to September 100± 8 129

192
Northeast India June to September 91± 8 88
South Peninsula June to September 97± 8 116
All India July 95± 9 105
All India August 99± 9 115
All India August to September 100± 8 130

12.4 Experimental Forecasting System

12.4.1 Monsoon Mission Coupled Forecast System (MMCFS)
The National Monsoon Mission (NMM) project was launched by the MoES in 2012 for
developing a state-of-the-art dynamical prediction system for monsoon rainfall on different
time scales. Climate Forecast System version 2 (CFSv2) of National Centers for
Environmental Prediction (NCEP), USA was identified as the basic modeling framework for
this purpose. The latest version of the high resolution (horizontal resolution of approximately
38km (T382)) MMCFS for the seasonal forecasting of monsoon rainfall was recently
implemented at the Climate Research and Services, IMD, Pune and it has been used to
generate the experimental forecast for 2017 southwest monsoon rainfall. This is a
worthwhile improvement over the original version which had a resolution of about 100 km.
The model climatology was prepared using retrospective forecasts generated for 27
years (1982-2008). The retrospective forecasts were prepared based on average of ten (10)
ensemble members. The skill scores of the MMCFS model for the forecasting of seasonal
rainfall over the country as a whole with different Initial conditions are given in the Table-
12.6. The performance of the model for the period 1982-2008 is given in the Fig.12.5. The
seasonal forecasts from CFS for the 2019southwest monsoon rainfall over the country as a
whole are also given in the table. For generating the 2019 JJAS forecast, 51 ensemble
members corresponding to 51 initial conditions of March month were used. Spatial pattern of
the forecasted rainfall anomaly based on March initial conditions is given in the Fig.12.6. In
month of June, updated forecast for JJAS generated with May initial conditions consisting of
40 ensemble members. Spatial pattern of the forecasted rainfall anomaly based on May
initial conditions is given in the Fig.12.7.
Based on the above modelling framework, seasonal forecasts were prepared by
using various initial conditions. The long period model average (LPMA) was computed using
28 year hindcasts for the period 1981-2008. As seen in the Table-12.6, the forecast based
on March initial conditions was 8% less than the actual rainfall 110% of LPA) whereas the
updated forecast based on May conditions was 13% of LPA less than the actual rainfall
110% of LPA). This suggests that the model forecast for the 2019 southwest monsoon
season over the country as a whole is not within forecast limits and correct.

193
Table - 12.6 : The skill scores of the Monsoon Mission CFS model for the forecasting
of seasonal rainfall over the country as a whole at different lead periods (‘3’, ‘2’ & ‘1’
months with March, April and May Initial conditions respectively). The forecast for the
2019 season rainfall over the country as a whole is given in the last column.

Initial JJAS Forecast for

conditions C.C (1981- RMSE (%of LPMA) 2019 (% of
(IC) of 2008) (1981-2008) LPMA)
March 0.45 9.28 102
April 0.35 8.70 97
May 0.23 8.66 97

Fig.12.5. Performance of the model hindcast for the southwest monsoon season
(June-September) rainfall over the country as a whole based on various initial
conditions. The model hindcasts were bias corrected using the z-score
transformation (correction for both mean and variance) method.

Fig.12.6. Rainfall anomaly forecast over Indian region for the 2019 monsoon season
computed from the MM CFS model based on March ICs.

194
Fig.12.7. Rainfall anomaly forecast over Indian region for the 2019 monsoon season
computed from the MM CFS model based on May ICs.

12.4.2 Forecasts for Seasonal Rainfall from other Indian Institutes

Apart from IMD, some other research institutions in India are also involved in the long
range forecasting research. Each year, these institutes provide experimental forecasts to
IMD prior to issuing of operational forecast. Some individual researchers also provide the
forecast. Experimental forecasts received from 6 different sources from India are given in
the Table-12.7. As seen in the Table-12.7, 2 of the 5 experimental forecasts indicated
normal (96-104% of LPA) monsoon season rainfall over the country as whole.

Table 12.7: Forecasts for 2019 southwest monsoon season rainfall over the country
as a whole received from various climate research centers/ research groups/
individuals.

Forecast (%
S.No. Institute Model
of LPA)
Space Applications Centre
Empirical model based on Genetic
1. (SAC), 91%
Algorithm
Ahmedabad
2. IMD, New Delhi Hybrid Model (CFS) 97%
South Indian Ocean convergence Weak
3. Onkari Prasad (Retired IMD)
zone based relation monsoon
4. IISc, Bangalore Stacked Auto-encoder method 99.5%
School of Earth Ocean and
5. Climate Sciences, IIT, MME of 6 dynamical models -
Bhubaneswar

195
12.4.3 Forecasts from Major International Climate Prediction Centers
Several international climate prediction centers regularly generate and provide global
seasonal forecasts based on dynamical models (Atmospheric/ coupled GCMs) through web.
Some of these centers also prepare Multi-Model Ensemble (MME) forecasts using
combinations of forecasts prepared by different centers. It may be mentioned that none of
these centers prepare forecasts specifically for the Indian region. The skill of the multi model
ensemble forecasts has been found to be better than that of the individual models.
Inferences derived from the MME forecasts for JJA/ JAS rainfall from 4 centers and
individual coupled model forecasts from 5 centers issued in April/ May are summarized in
the Table-12.8. It is seen from the Table-12.8 that the forecasts from most of the models
were indicating below normal to normal rainfall over the country as a whole during the 2019
southwest monsoon season. However, there are differences in the spatial patterns of the
various rainfall probability & anomaly forecasts.

Table 12.8: Inferences derived from seasonal forecasts from various climate centers
for the 2019 southwest monsoon season issued during April/May 2019.

S. Centre issuing Method

No the Forecast Inference for 2019

1. ECMWF, UK Coupled Model JJA & JAS (Issued: May 2019):

Normal rainfall is likely over most parts of India.
Normal to below normal rainfall is likely over
parts of eastern India and neighboring regions
near foothills of Himalaya.
EUROSIP Multi JJA and JAS (Issued: May 2019):
Model Ensemble Normal to below normal rainfall is likely over the
(MME): most parts of the country. However, there is
5 Coupled Models slight possibility to have above normal rainfall
over some of the west coastal region.
2. International MME JJA & JAS (Issued: May 2019):
Research 7 Models (AGCM Above normal rainfall is likely over some of the
Institute for & CGCM) west central and eastern parts of India. Below
Climate and normal rainfall is likely over some parts of south
Society, USA peninsular and northeastern India. For the rest
of the country climatological probabilities are
likely.
3. Japan Agency Coupled Model JJA (Issued: May 2019):
for Marine- Positive rainfall anomalies are predicted over
Earth Science most parts of south, northeast and east coast
and region of India. Negative rainfall anomalies are
Technology predicted over remaining areas like west central,
(JAMSTEC) northwest and north India.
4. APEC Climate MME June, July, August & September (Issued: May
Center 6 Models (AGCM 2019):
& CGCM) Normal to below normal rainfall is likely over
parts of northwest India. Below normal rainfall is

196
 likely over parts of south India. Above normal
rainfall likely over some parts of the eastern
India. Climatological probabilities are likely for
rest of the Country.
5. Met Office, UK Coupled Model JJA & JAS (Issued: May 2019):
Normal to below normal rainfall is likely over
most parts of the country. However, above
normal rainfall is likely over some of the
northeastern parts of the country.
6. WMO LRFMME AGCM & CGCM JJA (Issued: May 2019):
Above normal rainfall is likely over the
northernmost pars of the country. Below normal
rainfall is likely over most parts of the country.
Climatological probabilities are likely for rest of
the Country.
JJAS (Issued: May 2019):
Positive rainfall anomalies are predicted over
some of the northeastern parts of the country.
Indian region. Negative rainfall anomalies are
predicted over most parts of the country.
7. NMME MME of 8 JJA & JAS (Issued: May 2019):
Models & Normal precipitation is likely over some of the
coupled models eastern parts of the country. Above normal
precipitation is likely over parts of west and
central India. Climatological probabilities are
likely for rest of the Country.
8. NCEP CFSSv2 Coupled Model JJA & JAS (Issued: May 2019):
Above normal rainfall is predicted over parts of
west, south, northernmost and northeast India.
Below normal rainfall is predicted over parts of
eastern India and neighboring central region.
Normal rainfall is likely over remaining pats of
India.
9. JMA Coupled Model JJA (Issued: May 2019):
Normal Precipitation is likely over some of the
eastern costal parts of the country. Below
normal precipitation is likely over most parts of
the country. However, above normal
precipitation is likely over northernmost parts of
the country.
10 COPERNICUS MME of 5 JJA & JAS (Issued: May 2019):
Models & Normal to below normal precipitation is likely
coupled models over most parts of the India. Above normal
precipitation is likely over some parts of northern
India.
11. CMCC Coupled Model JJA & JAS (Issued: May 2019):
Normal to below normal precipitation is likely
over most parts of India. However, above normal
precipitation is likely over some parts of central
and eastern India.
12. DWD Coupled Model JJA & JAS (Issued: May 2019):
Normal to below normal precipitation is likely
over most parts of the India. Above normal
precipitation is likely over pockets of eastern

197
 India near coastal region.

13. ECMWF Coupled Model JJA & JAS (Issued: May 2019):
Copernicus Normal to below normal precipitation is likely
over most parts of the India.
14. Met Office Coupled Model JJA & JAS (Issued: May 2019):
Copernicus Normal to below normal precipitation is likely
over most parts of the India.
15. Meteo-France Coupled Model JJA & JAS (Issued: May 2019):
Above normal precipitation is likely over some of
the central parts of the country and its
neighboring region. Below normal precipitation is
likely over northwest and northern parts of the
country. Normal precipitation is likely over rest of
the country.

12.5 Conclusions
Based on an indigenously developed statistical model, it was predicted on 6th June
2019 that monsoon will set in over Kerala on 8th June with a model error of ±4 days.
However, the actual monsoon onset over Kerala took place within limit of model error of ±4
days and therefore the forecast was correct.
The long range forecast for the 2019 southwest monsoon rainfall was issued in 3
stages. The first stage long range forecast issued on 15th April consisted of only forecast
for season (June-September) rainfall over the country as a whole. In the second stage
(31st May), along with the update for the April forecast, forecast for season rainfall over the
four broad geographical regions (northwest India, Central India, South Peninsula and
northeast India) and that for monthly rainfall over the country as a whole for the months of
July and August were issued. In the 3rd stage (1st August), the forecast for the rainfall
during the second half of the monsoon season over the country as a whole was issued.
The first stage forecast for the season (June-September) rainfall over the country
as a whole issued in April was 96% of LPA with a model error of ± 5% of LPA. The update
issued in May for this forecast was (96% of LPA) with a model error of ± 4% of LPA. The
actual season rainfall for the country as a whole was 110% of LPA, which is 9% & 10% of
LPA more than upper forecast limits of the April and May forecasts. Thus the both the
forecasts were not within forecast limits.
Considering the four broad geographical regions of India, the forecasts issued in
May for the season rainfall over northwest India, Central India, northeast India and South
Peninsula were 94%, 100%, 91% & 97% of the LPA respectively all with model errors of ±
8%. The actual rainfalls over northwest India, Central India, Northeast India and South
Peninsula were 98%, 129%, 88% and 116% of the LPA respectively. Thus, Northwest
India and Northeast India are realized to be within the stipulated forecast error ranges.

198
The forecast for the seasonal rainfall over central India and south Peninsula was
underestimated to the actual value. The actual rainfalls of Central India and South
Peninsula were more than upper forecast limit by 21% and 11% of LPA respectively.
However, the actual season rainfall over northeast India was more than lower forecast limit
by 5% of LPA and for northwest India, the forecasts for the seasonal rainfall is realized to
be within the stipulated forecast error range. Thus, the forecasts of the seasonal rainfalls
for the two of the geographical regions (except central India and south Peninsula) were
correct.
The forecast for the second half of the monsoon season (August –September) for
the country as a whole was 100% with a model error of 8% of LPA against the actual
rainfall of 130% of LPA, which is 22% more than the upper forecast limit of (100%+8% =
108% of LPA). Thus, the forecast for the rainfall during the second half of the monsoon
season over the country as a whole was also underestimate to the actual rainfall and was
not correct.
The forecasts for the monthly rainfall over the country as a whole for the months of
July & August issued in June were 95% & 99% of LPA respectively with a model error of ±
9%. Thus the monthly forecasts for the July and August rainfalls were also underestimate
to the actual monthly rainfalls (105% & 115% of LPA respectively). But the actual rainfalls
during July is 1% more than the upper forecast limit and that of August is 7% more than
the upper forecast limit and the forecasts were therefore not correct.
While issuing the long range forecast in the month of April, El Nino conditions
where prevailed over Pacific Ocean. IMD‟s analysis of weakening of El Nino episode
prevailing since first quarter of 2018, to neutral in the second half of the monsoon season
and the emergence of positive Indian Ocean Dipole (IOD) in the middle of the monsoon
season resulting in increased rainfall activity in the second half of the season came
correct. Whereas, the June rainfall was deficient (67% of LPA) due to the negative impact
of the prevailing El Nino and delayed onset of monsoon over Kerala (8th June), the
monthly rainfalls during the latter 3 months were all above the LPA. It may be noted that
IMD forecast for monsoon onset over Kerala (6th June with model error of ±4 days) was
correct. Due to weakening of the El Nino conditions and emergence of the positive IOD
during July, rainfall activity increased significantly since last week of July and this
momentum of good rainfall activity continued through the second half of the monsoon
season. However, the rainfall during the second half was more than expected by IMD
mainly due to the longer life of a series of low-pressure systems formed over the region,
which mostly moved along the monsoon trough resulting in above normal season rainfalls
over Central India & South Peninsula and below normal season rainfall over North-East
India. Season rainfall over North-West India was normal. Thus, IMD‟s forecast for

199
geographical rainfall distribution of highest rainfall over central India and lower rainfall over
North-East India also came correct though the received rainfall over Central India and
South Peninsula was more than the upper forecast limits. Overall, the impact of synoptic
scales systems on the monsoon performance was very significant this year resulting in
increased uncertainty in the predictability of monsoon at extended and seasonal scales.
The experimental forecasts from Monsoon Mission CFS for the season rainfall over
the country as a whole based on both March and May initial conditions were underestimate
to the actual situation by about 8% and 13% of LPA. The experimental forecasts from
most of the other Indian sources as well as from the international climate centers indicated
below normal to normal rainfall over the country as a whole during the 2019 southwest
monsoon season.

References

Pai, D. S., O.P.Sreejith, S. G. Nargund, Madhuri Musale, and Ajit Tyagi, 2011,
Present Operational Long Range Forecasting System for Southwest Monsoon Rainfall
over India and its Performance During 2010, Mausam, 62, N2, pp179-196.
Pai, D. S. and M. Rajeevan, 2009, Summer monsoon onset over Kerala: New
Definition and Prediction, J. Earth Syst. Sci. 118, No. 2, pp123–135.
Rajeevan M, Pai D.S., Anil Kumar R, and Lal B, 2007, New statistical models for
long-range forecasting of southwest monsoon rainfall over India, Climate Dynamics, V28,
pp813-828.
Saha, Subodh K., Samir Pokhrel, Hemantkumar S. Chaudhari, Ashish Dhakate,
Swati Shewale, C. T. Sabeerali, Kiran Salunke, Anupam Hazra, Somnath Mahapatra, and
A. Suryachandra Rao. "Improved simulation of Indian summer monsoon in latest NCEP
climate forecast system free run." International Journal of Climatology 34, no. 5 (2012):
1628-1641.
Ramu, D. A., C. Sabeerali, R. Chattopadhyay, D. N. Rao, G. George, A. Dhakate,
K. Salunke, A. Srivastava, and S. A. Rao (2016), Indian summer monsoon rainfall
simulation and prediction skill in the CFSv2 coupled model: Impact of atmospheric
horizontal resolution, Joural of Geophysical Research: Atmospheres, pp. 1752–1775,
doi:871 10.1002/2015JD023538.

200