PI : Prof. Sutapa Chaudhuri

1

Bias estimation and effort for removal of UM/CFS
coupled model output with adaptive techniques for
improving forecast skill of ISM.
Research Project under National Monsoon Mission

• PI Prof. Sutapa Chaudhuri
• Co-PIs Prof. M.Majumdar Dr.D.Lohar
• JRFs Debanjana Das, Sayantika Goswami Arumita
  Roy Chowdhury
• Department of Atmospheric Sciences
• University of Calcutta

2
The objectives of the Proposed research

1. Study and analysis of various components of
   monsoon and their seasonal and monthly
   variability (spatial and temporal) during June,
   July, August and September (JJAS). Seasonal
   variability of monsoon will be the major concern.
2. Correlation analysis between monsoon rainfall and
   various components like SST anomaly, ENSO, NAO
   etc. Predictability of the phases (active or
   break) and intra-seasonal variability and
   identification of relevant predictors
3. Error analyses for both CFS (v2) and Unified
   model (UM) of UKMO generated products and their
   comparison.
4. Performance analysis of model outputs using
   various skill scores.
5. Identification of the model bias due to various
   mesoscale weather components
6. Error minimization of the model products.
7. Scrutinizing the validation of model after bias
   correction and error minimization

3

• The preliminary objectives (i) and (ii) of the
  project have been taken up and completed , which
  are Published in referred journals
• Chaudhuri Sutapa, Sayantika Goswami, Debanjana
  Das Anirban Middey (2014) Meta-heuristic ant
  colony optimization technique to forecast the
  amount of summer monsoon rainfall skill
  comparison with Markov chain model, Theor Appl
  Climatol, 116, 3-4, 585-595 (IF 1.742)
• Chaudhuri Sutapa Jayanti Pal (2014) The
  influence of El Niño on the Indian summer
  monsoon rainfall anomaly a diagnostic study of
  the 1982/83 and 1997/98 events, Meteorol Atmos
  Phys, 124, 3-4, 183-194 (IF 1.245)
• Chaudhuri Sutapa Jayanti Pal (2014)
  Cloudaerosol coupled index in estimating the
  break phase of Indian summer monsoon, Theor Appl
  Climatol, 118, 3, 447- 464(IF 1.742)
• Chaudhuri Sutapa, Jayanti Pal, Suchandra
  Guhathakurta (2015) The influence of galactic
  cosmic ray on all India annual rainfall and
  temperature, Advances in Space Research, 55, 4,
  11581167 (IF 1.238)

4
Research undertaken to attain the prime
objectives of the project (iii) Bias
estimation for CFS (v2) model generated products
Motivation Scientists have shown that the summer
monsoon of India encounters with episodes of
abundant precipitation during active phase and
scarce precipitation during break phase. It has
also been observed that the onset of Indian
summer monsoon (ISM) is triggered by northward
propagation of the wet phase of boreal summer
intra-seasonal oscillation (BSISO) with
considerable inter-annual variability.
Approach Considering the crucial role of BSISO
on ISM, proper simulation of this variability
through the state of the - art coupled model
is felt to be important for the short-term and
long-term prediction of the ISM. In view of this,
the efficiency of (CFSv2) over the Indian monsoon
region is investigated with 50 years long run.
The skill of the model in simulating the spatial
distribution of Precipitation, SST, northward and
eastward propagation of the system is estimated.
5
Mean JJAS Precipitation Comparison between
CFSv2 OBS
Seasonal mean rainfall is characterized by three
strong convective zones over Indian landmass
during monsoon season Western Ghat Central
India and North-east India CFSv2 model
simulates the mean JJAS rainfall but
overestimates rainfall over Western Ghat and
north - east India. However, it underestimates
the rainfall over central India. The model has a
tendency to show more rain over regions having
elevated orography. In central India as there is
no remarkable elevation in orography, the model
underestimates the actual rainfall pattern.
6
Mean JJAS Precipitation Comparison between
CFSv2 OBS
The oceanic rainfall is a major challenge for
CFSv2 model simulation as it is clear from the
rainfall plots. The significant wet rainfall
bias is evident over western equatorial Indian
Ocean (WIO) and Arabian Sea (AS). The presence
of dry rainfall bias of similar magnitude over
eastern equatorial Indian Ocean (EIO) and Bay of
Bengal (BoB) is also observed.
7
Mean JJAS Precipitation
Mean JJAS SST - Comparison
The wet rainfall biases over western equatorial
Indian Ocean (WIO) and Arabian Sea (AS).
coincides with the cold SST biases. Dry
rainfall bias over eastern equatorial Indian
Ocean (EIO) and Bay of Bengal (BoB) also
coincides with cold SST bias Thus, CFSv2 model
shows cold SST bias over ocean.
8
Intra-seasonal precipitation variance
The variance of CFSv2 model output with
observation of rainfall and SST during JJAS
period shows the regions with maximum bias
(variance). The plots show that Somali upwelling,
which is a common feature during SW monsoon is
over simulated by the model.
Intra-seasonal SST variance
9
Lead Lag SST-precipitation regression
Both the northward and eastward propagations are
simulated by CFSv2. However, it is not very
prominent.
Northward Propagation
Eastward Propagation
10
Findings The analyses show that the CFSv2 model
successfully simulates many major features like
precipitation, SST variation and propagation
during ISM. However, the model overestimates
precipitation over Western Ghats and North East
India having elevated orography. The significant
overestimate of precipitation is observed over
the eastern equatorial Indian Ocean (EIO) and
Arabian Sea (AS). The oceanic rainfall is a major
impediment in the CFSv2 simulation as is evident
from rainfall variation. The presence of
significant dry bias (3 - 7 mm/day) over central
India implies that CFSv2 underestimates the
rainfall over the monsoon trough zone. The model
has a tendency to underestimate the precipitation
over central India having comparatively flat
orography. The study further shows that CFSv2
model depicts cold SST bias over ocean.
11
The presence of bias in the CFSv2 model products
in simulating ISM motivated to Investigate on the
ambiguity in predictability of Summer Monsoon in
various climate models along with
CFSv2 Preamble Characterizing and quantifying
ambiguity in simulation of Summer Monsoon is of
primary importance not only for the purposes of
detection and attribution, but also for tactical
approaches for adaptation and mitigation.
Uncertainty in prediction derives from three main
sources forcing, model response, and internal
variability (Hawkins and Sutton 2009 Tebaldi and
Knutti 2007). Internal variability is the natural
variability of the climate system that occurs in
the absence of external forcing, and includes
processes inherent to the atmosphere, the ocean,
and the coupled ocean-atmosphere system. Internal
atmospheric variability, also termed as climate
noise (Madden 1976 Schneider and Kinter
1994Wunsch 1999 Feldstein 2000), arises from
chaotic dynamical processes inherent to the
atmosphere. Internal variability, on the other
hand, depends on dynamics. Approach Predictability
of CFSv2 model assessment is done along with
few other existing global model with the help of
signal to noise ratio. Attempt is made to
calculate and analyze the predictability limit.
The Internal variability of model is evaluated.
The Predictability is computed at the
significance level of 99.9, 99, 95.
12
Model names Ensemble member (with May initial condition)
NCEP CFSv2 24
GFDL CM2p1 10
NASA GMAO 6
COLA RSMAS CCSM3 29
Parameter taken
Precipitation (mm/day)
Sea Surface temperature (degree centigrade)
Observation Used
GPCP for precipitation
HadISST for Sea Surface Temperature
13
(No Transcript)
14
Inter comparison of Model bias during JJAS for
(b) sea surface temperature
Inter comparison of Model bias during JJAS for
Total precipitation
CFSv2 model products show cold bias almost over
the entire tropical ocean except some parts of
west pacific. Different models show different
kinds of biases in some region. Each model
suffers from cold as well as warm biases
throughout the globe.
15
Inter comparison of Model signal during JJAS for
SST
Inter comparison of Model signal during JJAS for
Total precipitation
SST signal is reasonably captured by the four
global coupled models. NASA GMAO model has
serious problem in the polar region.
16
Inter comparison of Model noise during JJAS for
(b) sea surface temperature
Inter comparison of Model noise during JJAS for
Total precipitation
Among the four models CFSv2 shows maximum
internal variability that is the differences in
the realization. This is also observed to be high
over the tropics in each model. In CFSv2 the
noise is highest over the tropical Indian ocean
and over tropical Pacific.
17
Inter comparison of Model SNR during JJAS for
Total precipitation
Inter comparison of Model SNR during JJAS for sea
surface temperature
SNR is highest in the tropics as signal is
highest in the tropics and noise is also high for
precipitation. However, the signal is higher than
the noise thus, SNR is high.
18
Inter comparison of Model Signal to total square
root during JJAS for (a) Total precipitation
Inter comparison of Model Signal to total square
root during JJAS for (b) sea surface temperature.
19
Inter comparison of predictability at different
significance level during JJAS for (a) Total
precipitation
Inter comparison of predictability at different
significance level during JJAS for (b) sea
surface temperature
the theoretical predictability of seasonal mean
monsoon is quite high
20
Findings

• SNR is highest in the tropics as signal is
  highest in the tropics and noise is also high for
  precipitation. However, the signal is higher than
  the noise thus, SNR is high.
• Location of Maximum signal to Noise is different
  in different models. The main reason behind this
  is that the signal is changing from model to
  model. Signal to Noise is high in tropics. This
  means that boundary condition of SST are the main
  forcing behind producing large signal.
• SST is high in the tropics mostly because noise
  is low in the tropics. Signal is higher than the
  noise here. But in the mid-latitudes Noise is
  higher than signal
• Region of cool and warm SST bias vary in model to
  model. Bias is minimum in ENSO region.CFSv2 is
  having cool SST bias
• It is the measure of internal variability.
  Internal variability depends on dynamics. It is
  very surprising that noise is very different in
  different models.
• Noise in the CFS is highest among all the models.
  This noise is highest in the Indian Ocean and
  Pacific Ocean region.
• It is not known that which one is correct because
  the noise is not known in observation as the
  observation has only one realization and
  therefore in observation, noise as well as
  signal is mixed. In nature there is no ensemble
  member.
• SST is highly predictable during JJAS almost over
  the whole globe.
• From the analysis it is clear that
  predictability is higher in tropics.
• Signal is well captured in CFSv2 but at the same
  time internal variability(noise) is also maximum
  in case of CFSv2.
• Orographay and moist convection seem to be most
  important factors in simulating monsoon rainfall.

21
The most important finding of this study is that
the theoretical predictability of seasonal mean
monsoon is quite high however it is known from
practical experience that actual skill of
seasonal operation prediction by coupled models
is not very high. Thus, the question is why the
predictability is high but the prediction skill
is low?
The answer to this question is beyond the scope
of this study but it has been suggested by many
authors that it is partly because of the large
intra-seasonal variability within the monsoon
season and partly because the models have very
large biases. The speciality of CFSv2 model is
that the internal dynamics - generated noise is
much higher than other models at the same time
signal is also highest among the other models
or closest to observation.
22
(iii) Bias estimation for UKMO Unified model (UM)
generated products Motivation The Indian summer
monsoon and associated rainfall is a matter of
concern for decades. Modeling aspects and
improved plotting platforms are coming up besides
statistical methods for better understanding of
the model products in simulating rainfall. The
main purpose of this study is to estimate the
skill of United Kingdom Met Office - Unified
Model (UKMO - UM) in estimating the Indian summer
monsoon rainfall (ISMR) from June to September
(JJAS), 2011 within the confine of 5º to 36º
latitude and 60º to 100º longitude over
India. Approach Firstly Orography (m) dataset
which is an adapted data in the Unified Model is
collected. For the verification Terrain-Base
(T-base) etopo2 orography (m) dataset from
NOAA/NGDC site (http//dss.ucar.edu/datasets/ds759
.3) is also collected. The resolution of the
Orography dataset for both T-Base and Unified
Model is converted into 0.25x0.25 grid
resolution. Secondly, three important surface
parameters produced by the model have been taken
into account for analyzing the Orography induced
effect. Those are total precipitation amount for
Indian Summer Monsoon Rainfall (ISMR), Surface
temperature (ST) and Surface vector wind (SVW).
Total precipitation amount (kg/m2/TS, equivalent
to mm/TS), 6-hourly data set is prepared for JJAS
period as accumulated precipitation in mm/day
(ISMR), Surface Temperature (ST) is computed in
degree celcius (ºC) from its absolute value and
surface vector wind (SVW) is in m/s.
23
Data Arrangements
Parameters Format of the observational data Format of UKMO-UM data Format converted for the study
Elevation NOAA-NGDC Terrain-Base 2-minute Bathymetry/Topography Data, 0.033x0.033 grid in meters 0.83x0.55 in meters 0.25x0.25 grid in meters
Precipitation IMD observation of 1x1 grid of daily file in mm 6-hourly file of 0.35x 0.23 grid in kg/m2/TS (TS stands for time step of 6-hour) Monthly (JJAS) 1x1 grid in mm/day
Surface Temperature NCEP/NCAR observation of 2.5x2.5 grid in degree celcius 6-hourly file of 0.35x 0.23 grid in abosulte temperature (K) Monthly (JJAS) of 1x1 in degree celcius
Surface Vector Wind NCEP/NCAR observation of 2.5x2.5 in meter/sec 6-hourly file of 0.35x 0.23 grid in Monthly (JJAS) of 1x1 in meter/sec
24
The performance of UKMO-UM to produce
precipitation w.r.t. IMD observation
UKMO-Unified Model UKMO-Unified Model
Standard Deviation 0.444
Correlation Coefficient 0.584
RMS Difference 0.616
Estimated Bias -3.025
Precipitation (ISMR) observed by TRMM-merged IMD
data (left), UKMO-UM model output (middle),
estimated bias (right)
25
Zonal divisions of India
Zonal Divisions of India Coordinates of the enclosed area
Western Ghat (WG) 8ºN 21.5ºN, 73ºE - 78ºE
Eastern Ghat (EG) 10ºN - 22ºN, 78ºE - 87ºE
Gangetic West Bengal (GWB) 21ºN - 26ºN, 86ºE - 91ºE
Central Part of India 17ºN - 27ºN, 73ºE - 83ºE
Himalayan Ranges 27ºN - 36ºN, 74ºE - 96ºE
WG EG GWB CPI HR
26
ISMR from IMD observation (left), UKMO-UM output
(mid) and estimated model bias (right), over WG
T-base elevation from NOAA/NGDC (left), UKMO-UM
output (middle), and estimated bias in Elevation
(right) over WG
ST from NCEP/NCAR Reanalysis (left), UKMO-UM
output (mid) and estimated bias in ST (right)
over WG
27
SVW from NCEP/NCAR Reanalysis (left) and UKMO-UM
output (right) over WG
28
ISMR from IMD observation (left), UKMO-UM output
(mid) and estimated bias in ISMR (right) over EG
T-base elevation from NOAA/NGDC (left), UKMO-UM
output (middle), and estimated bias in Elevation
(right) over EG
ST from NCEP/NCAR Reanalysis (left), UKMO-UM
output (mid) and estimated bias in ST (right)
over EG
29
SVW from NCEP/NCAR Reanalysis (left) and UKMO-UM
output (right) over EG
30
ISMR from IMD observation (left), UKMO-UM output
(mid) and estimated bias in ISMR (right) over GWB
T-base elevation from NOAA/NGDC (left), UKMO-UM
output middle), and estimated bias in Elevation
(right) over GWB
ST from NCEP/NCAR Reanalysis (left), UKMO-UM
output (mid) and estimated bias in ST (right)
over GWB
31
SVW from NCEP/NCAR Reanalysis (left) and UKMO-UM
output (right) over GWB
32
ISMR from MD observation (left), UKMO-UM output
(mid) and estimated bias in ISMR (right) over CPI
T-base elevation from NOAA/NGDC (left), UKMO-UM
output (middle), and estimated bias in Elevation
(right) over CPI
ST from NCEP/NCAR Reanalysis (left), UKMO-UM
output (mid) and estimated bias in ST (right)
over CPI
33
SVW from NCEP/NCAR Reanalysis (left) and UKMO-UM
output (right) over CPI
34
ISMR from IMD observation (left), UKMO-UM output
(mid) and estimated bias in ISMR (right) over HR
T-base elevation from NOAA/NGDC (left), UKMO-UM
output (middle), and estimated bias in Elevation
(right) over HR
ST from NCEP/NCAR Reanalysis (left), UKMO-UM
output (mid) and estimated bias in ST (right)
over HR
35
SVW from NCEP/NCAR Reanalysis (left) and UKMO-UM
output (right) over HR
36
Planned activities for the remaining period to
meet the remaining objectives
On going Research

• Study of internal variability with high
  resolution data.
• Estimation of orography induced bias on ISMR with
  UM model products
• Future Plan
• Long-range forecast of Indian Summer Monsoon
  Rainfall using adaptive neuro-fuzzy inference
  system and validation with CFSv2 and UKMO model
  products
• The model bias arises due to various mesoscale
  weather components will be taken care off

37
Acknowledgement
I thank the Ministry of Earth Science (MoES), GOI
for giving me the opportunity to work for
National Monsoon Mission I thank the team of
Monsoon Mission Scientists of IITM , Pune for
their kind cooperation and help provided to the
research fellows of the project, whenever
needed. I thank the University of Calcutta for
giving me the facilities for the success of the
project. I thank my research fellows of the
project for hard work and taking keen interest
for the success of the mission. I thank
IITM-ICTP TTA activity for enhancing the
technical efficiency .
38
Thank You
Those who hate rain hate life  ? Dejan
Stojanovic