Potential Predictability and Extended Range Prediction of Monsoon ISO

1
International Asian Monsoon Symposium, Honolulu,
Hawaii
Potential Predictability and Extended Range
Prediction of Monsoon ISOs
B.N. Goswami Centre for Atmospheric and Oceanic
Sciences Indian Institute of Science, Bangalore.
B.N. Goswami and Prince K. Xavier, 2003, Geophys.
Res. Lett., 30(18), 1966
2
Basis for Potential Predictability of monsoon ISOs
Convectively coupled large-scale spatial
structure
Rainfall (shaded, mm/day) and 850 hPa wind
anomalies associated with a strong (right) and
weak (left) phases of monsoon ISO.
3
Basis for Potential Predictability of monsoon ISOs
Existence of low frequency quasi-periodicity
Time series of daily rainfall anomaly (mm/day)
over central India (blue) during 1 June 30
Sept. for three years and 10-90 day filtered
(red) rainfall.
4
What could possibly be predicted with lead time
of more than 10 days? Large-scale low frequency
component of intraseasonal variability (10-90
day) of rainfall

• What is the usefulness of such predictions?
• Would predict the dry and wet spells. Planning
  for sowing, harvesting, water management.
• As ISOs also cluster the synoptic variability,
  it would also give probability of occurrence of
  high or low rainfall.

5
Goswami, et al. 2003, GRL
Frequency distribution of genesis of low pressure
systems (LPS) as a function of normalized monsoon
intraseasonal index (MISI) based on 40 years of
data
6
Spatial clustering of tracks of LPS during active
ISO phase, MISI gt 1, during 1954-1983.
Few LPS and their tracks during weak (break)
ISO phase, MISI lt-1, during 1954-1983.
7
Long term seasonal mean (JJAS) winds at 850 hPa
(m/s) and associated relative vorticity (10-6 s-1)
Active - Weak composite wind anomalies at 850 hPa
(m/s) and associated relative vorticity (10-6
s-1) based on 40 year (1954-1993)
8

• How can we make an estimate of potential
  predictability for active and break conditions
  from observations?
• A simple procedure is described to make such an
  estimate from observations

9
Data Used

• Daily rainfall over Indian continent from rain
  gauge stations (1971-1989)
• CMAP pentad data (linearly interpolated to daily
  values) , 1979-2001
• NCEP/NCAR Reanalysis daily winds 1979 2001
• NOAA daily OLR 1979-2001
• 10-90 day band-pass Lanczos filter is used to
  isolate ISO

10
Regions over which potential predictability of
precipitation is examined
11
10-90 day filtered precipitation (CMAP) averaged
over Box I normalized by its own standard
deviation shown here for 10 summers (1 June- 30
Sept.). Blue circles ? peak wet spells (active
conditions) red squares ? peak dry spells
(break in monsoon).
12
CMAP Box-I
2
2
2
13
Same as the previous figure (a), but for the
precipitation averaged over the eastern
equatorial Indian ocean Box II
Same as in (a) but for rainfall data from rain
gauge stations averaged over monsoon trough
region.
14
Same as in (a) but for relative vorticity
averaged over the monsoon trough (70E-90E,
15N-25N). Active and break dates were taken from
precipitation.
Same as (a) but for Zonal winds at 850 hPa
averaged over 80E-95E, 12N-18N. Active and break
dates were taken from precipitation.
Thus, the differences in divergence between
transitions from active to break and break to
active is similar in circulation and rainfall.
15
Conclusion Predictability The transition from
break to active conditions is intrinsically more
chaotic than transitions from active to break
conditions. A fundamental property of monsoon
ISOs. Why? Break ?Active convective
instabilityfast error growth Active?Break
dying convection -- slow error growth due to
slow oscillation Consequence, The potential
predictability limit for monsoon breaks is about
20 days while that for monsoon active conditions
is only about 10 days
16
Empirical Extended Range Prediction of Monsoon
ISOs
17
Lo and Hendon (2000) Mon. Wea. Rev., 128,
25282543. The prediction of MJO in the OLR and
200 hPa streamfunction was attempted by using a
simple multiple linear regression model. The
predictors were OLR and 200 hPa streamfunction
themselves. The predictants were two leading
Principal Components (PC) of OLR and three
leading PCs of 200 hPa streamfunction. Skillful
forecasts of the MJO in OLR and 200 hPa
streamfunction were achieved out to about 15
days. The model performed well when the MJO is
active at the initial condition but not so well
when it is inactive. They also found that the
empirical forecasts were better than the DERFs
for lead times longer than one week.
18
First two EOFs of CMAP pentad (interpolated to
daily) rainfall for JJAS (1979-2001)
19
Predictors PC1-4 of 10-90 day filtered CMAP
PC1-2 of Surface Pressure
Predictants PC1-4 of filtered rainfall
Predicted rainfall at lead time ?
20
Model developed on 1 June-30 Sept. data for
1979-1995 Model is tested on independent data for
1996-2001
15-day predictions and verifications of rain
anoms ave(70E-90E)
21
(No Transcript)
22
Mean of 18-day predictions of breaks (mm/day)
Mean of 57 verifications of the 18-day
predictions from CMAP (mm/day)
(Mean of an ensemble of 57 such predictions
starting from initial conditions around active
conditions)
23
Mean of 18-day predictions of Active conditions
(mm/day) Mean of an ensemble of 54 such
predictions starting from initial conditions
around active conditions
Mean of 54 verifications of the 18-day
predictions from CMAP (mm/day)
24
Correlation between 18-day predictions of active
conditions starting from break conditions with
corresponding verifications (n54)
Correlation between 18-day predictions of breaks
starting from active conditions with
corresponding verifications (n57).
25
--------------------------------------------------
------------------- Lead Time Prediction
Prediction of Breaks of Active ------------
--------------------------------------------------
------- 15 days 0.65 0.38 18
days 0.56 0.43 -----------------------------
----------------------------------------- Correlat
ions between predictions and observations of
rainfall averaged over the monsoon trough region
(70E-85E, 10N-22N)
26
18-day predictions of rainfall over the monsoon
trough (red) together with actual observations
(blue) for the period June 1-Sept. 30, of 2000
and 2001.
27
Conclusion The model demonstrates useful skill
of prediction of breaks up to 18 days in
advance. However, the useful skill for
prediction of active conditions is limited to
about 10 days. Scope for improvement preparatio
n for real-time prediction employ different
method of empirical prediction
28
(No Transcript)
29
(No Transcript)
30
Table 1. Errors and Skill of model forecasts and
persistence of anomalies averaged over 70-85E,
10-22N.
31
15-day predictions and verifications of rain
anoms ave(70-90E)
32
(No Transcript)
33
(No Transcript)
34
Empirical prediction of dry and wet spells of rain
The multiple linear regression model
?i - Regression coefficients
? - Lead time
N - Number of Predictors
35
Predictors PC1-4 of 10-90 day filtered CMAP
PC1-2 of Surface Pressure
Predictants PC1-4 of filtered rainfall
Predictors are added one by one and the model
gives optimum performance with the above 6
predictors
The model is developed on 17 monsoon seasons
(1979-1995) and they are tested on the next 5
years (1996-2001).
36
Having generated the predicted values of the
first four PCs of rainfall, the predicted
rainfall anomalies (P) are constructed using
Where, PC is the Principal components E is the
EOFs