Subseasonal Prediction with the NCEP-CFS: Forecast Skill and Prediction Barriers for Tropical Intraseasonal Oscillations

1
Subseasonal Prediction with the NCEP-CFS
Forecast Skill and Prediction Barriers for
Tropical Intraseasonal Oscillations
Experiments conducted under NOAAs Climate Test
Bed
Augustin Vintzileos and Hua-Lu
Pan EMC/NCEP/NWS/NOAA
2
Messages to take back home

• The CFS is a useful tool in forecasting Tropical
  Intraseasonal Oscillations which are the basis
  for subseasonal prediction its skill is similar
  to other centers
• The reason for the drop in skill is found in the
  Maritime Continent which presents a Barrier to
  the eastward propagation of the active convective
  phase of the TIO
• Increasing the horizontal resolution of the
  atmospheric model has not improved the skill of
  TIO forecast
• A better set of initial conditions is shown to be
  crucial for improving skill by 3-5 days. Both
  better quality and better compatibility with the
  forecast model appears to be a factor
• Intraseasonal variations of oceanic initial
  states are not well represented by GODAS.
  However, ocean atmosphere coupling is quite
  important for the TIO. It follows that
  inconsistencies between the ocean and atmospheric
  initial state at this portion of the spectrum may
  damage the forecast

3
What is Subseasonal Forecasting?
4
The seamless forecasting suite from Weather to
Climate
Atmospheric initial conditions
Forecast lead times
0-14 days
15-60 days
60 days and beyond
Land initial conditions
Synoptic
Seasonal-to-Interannual
Subseasonal
Oceanic initial conditions
Mainly affected by Atmospheric I.C.
Mainly affected by Oceanic I.C. but also by land
I.C. (e.g., snow cover, soil moisture)
Affected by all I.C.
TIO affecting weather statistics
ENSO affecting weather statistics
Weather
5
Issues concerning subseasonal forecasting

• How critical are Initial Conditions?
• How critical is model resolution?
• How critical are model drifts and biases?
• What are the most adequate ensemble generation
  techniques?

Answers to such questions will allow to
prioritize development efforts and thus optimize
the operational tool
6
CPC Global Tropics Benefits/Hazards Assessment
Description Week 1-2 outlooks for
enhanced/suppressed rainfall and
favorable/unfavorable conditions for TC
activity Purpose Provides regional planners
with global interests advanced notice on
potential hazards/impacts Physical Basis MJO,
ENSO, other coherent and/or persistent anomalies,
interaction with the extratropics Outside
Collaboration ESRL, TPC, NWS WR/CR, and
others Tools Detailed monitoring, ENSO/MJO
composites, MJO objective forecasts
(statistical/dynamical), GFS/CFS
forecasts Plans Product more objective in
nature, evaluate and apply input associated with
subseasonal variability from additional dynamical
models
See poster by Jon Gottschalck
7
The Tropical Intraseasonal Oscillation (TIO)
8
Tropical Intraseasonal Oscillations some points
to remember

• TIO consists of large-scale coupled patterns in
  atmospheric circulation and deep convection all
  propagating eastward slowly through the portion
  of the Indian and Pacific oceans where the sea
  surface is warm. It constantly interacts with the
  underlying ocean and influences many weather and
  climate systems (from Zhang, 2005)
• TIO are the scientific basis for subseasonal
  forecasting i.e., they are what ENSO is to
  seasonal forecasting
• No theoretical context yet
• Comprehensive dynamical models do not represent
  them perfectly though there is consensus that
  coupling with the ocean improves their simulation
• Observations show that sometimes the MJO
  collapses to higher modes as it crosses the
  Maritime Continent

9
Defining a metric for the TIO

• A CLIVAR-MJO panel recently made recommendations
  on a number of metrics to use. One of these
  metrics combine winds at 200 hPa and 850 hPa and
  precipitation i.e., represents the coupling
  between the large scale circulation and diabatic
  forcing.
• We have hindcasts from 2002 to 2006 i.e., a
  mostly quiet period in regard to ENSO events.
  Nevertheless, in order to avoid possible
  sampling issues for defining mean annual cycles
  and drifts we only use the smoothest possible
  variable for defining an index.
• We use zonal wind at 200 hPa averaged from
  20S-20N and we next show that for our purpose
  this is an adequate measure

10
Defining a metric for the TIO
The Recipe

• Our verifying fields will be from Reanalysis-2
• Consider the zonal wind at 200 hPa from 2002 to
  2006 averaged between 20S-20N
• Compute and remove the mean annual cycle and the
  zonal mean
• Perform and EOF analysis of the resulting field
  (no time filtering)

11
First and second EOFs of the zonal wind at 200 hPa
Indian
Atlantic
Pacific
10 days
EOF1
EOF2
-EOF1
-EOF2
r0.6
A full oscillation in 40 days
12
Reconstructed U200 vs. GPCP Precipitation, May
July, 2002
Upper level divergence
5S-5N averaged, total unfiltered precipitation
field
20S-20N averaged, filtered U200 anomaly field
13
Defining a metric for the TIO
The Recipe

• Projection of the observed and forecast U200
  anomalies on the two first EOFs isolates the TIO
  signal (no filtering in the time domain)
• Pattern correlation between the observed and
  forecast projections

14
Some initial experimentation

• Used the T126 version of the operational T62 CFS
• Hindcasts were run up to 65 days and were
  initialized four times per day from CDAS2 and
  GODAS from May 7th to July 15th and from November
  7th to January 15th from 2000 to 2004 (run by
  Saha, Vintzileos, Thiaw and Johanson )

15
The Maritime Continent Barrier
16
Pattern Correlation for initialization dates from
May to June 2002
The Maritime Continent Predictability Barrier
June 6th-9th
June 6th-9th
June 6th-9th
17
Reconstructed U200 vs. GPCP Precipitation, May
July, 2002
June 8th
Upper level divergence
5S-5N averaged, total unfiltered precipitation
field
20S-20N averaged, filtered U200 anomaly field
18
We designed a series of subseasonal retrospective
forecasts with the CFS for the systematic study
of the Maritime Continent Barrier (Proposed to
and endorsed by the Climate Test Bed FY2007)
19
Retrospective forecast design
May 23rd to August 11th from 2002 to 2006 1
forecast every 5 days, with additional
re-forecasts at the beginning of each
month Forecast lead 60 days
Model resolution Atmosphere T62
200Km x 200Km T126
100Km x 100Km T254
50Km x 50Km Ocean the standard CFS
resolution
Initial conditions Atmosphere, Land from
Reanalysis 2 (CDAS2) and from GDAS Ocean from
GODAS
20
Forecast skill for TIO as a function of
Resolution and Initial Conditions
21
Skill for the TIO mode (verification CDAS2)
GDAS
Persistence forecast
Skill up to 14 18 days
CDAS2
Persistence forecast
GDAS
22
MJO forecast skill at ECMWF
0.4
Skill up to 14-18 days
From Vitart et al. 2007
23
Reasons for the drop in forecast skillThe
Maritime Continent Barrier
24
Reconstructed U200 vs. GPCP Precipitation, May
July, 2002
June 8th
Upper level divergence
5S-5N averaged, total unfiltered precipitation
field
20S-20N averaged, filtered U200 anomaly field
25
Pattern correlation as a function of
initialization day and lead time
When initialized by GDAS the CFS shows a somehow
better behavior during the first few days of the
forecast near the barrier.
June 8th
June 8th
26
and the Ocean?

• There is consensus that the ocean plays an
  important role for the evolution of the TIO
• CFS is initialized by GODAS which in turn is
  optimized for Seasonal-to-Interannual forecast
• GODAS
• Comes in pentads
• Its SST is damped to the weekly Reynolds SST
• Contains information from 2 weeks before and two
  weeks after

27
Standard Deviation of the 20-90 day filtered SST
2002 - 2006
2002 - 2006
With MOM3 we use climatological SST for the
majority of the Maritime continent. MOM4
alleviates this issue
28

• As suspected, energy in the subseasonal portion
  of the spectrum is low in the GODAS product
• What about the normalized variability modes
  (more physical meaning for the Indian Ocean than
  simple EOFs) ?

29
(No Transcript)
30
(No Transcript)
31
Is there any relevance between the daily OI SST
EOF modes and the TIO?
32
The TIO EOFs
Compare to the correlation between Principal
Component 1 and Principal Component 2 of the
daily OI SST and the anomalies of Zonal Wind at
200 hPa at each grid point
33
The TIO EOFs
There is remarkable resemblance between the U200
EOFs and the correlation of U200 anomalies and
the SST Principal components
34
There is an empirical relationship between the
SST and the TIO suggesting that initial states
for the ocean and the atmosphere should be
coherent
35
Messages

• The CFS is a useful tool for forecasting
  Tropical Intraseasonal Oscillations its skill
  is similar or better to other centers
• The reason for the drop in skill is found in the
  Maritime Continent which presents a Barrier to
  the eastward propagation of the active convective
  phase of the TIO
• Increasing the horizontal resolution of the
  atmospheric model has not improved the skill of
  TIO forecast
• A better set of initial conditions is shown to be
  crucial for improving skill by 3-5 days. Both
  better quality and better compatibility with the
  forecast model appear to be a factor
• Intraseasonal variations of oceanic initial
  states are not well represented by GODAS.
  However, ocean atmosphere coupling is quite
  important for the TIO. It follows that
  inconsistencies between the ocean and atmospheric
  initial state at this portion of the spectrum may
  damage the forecast

36
Conclusions

• We have shown that a set of atmospheric initial
  conditions which is more realistic and which is
  more compatible with the forecast model is
  crucial for TIO forecast. This underlines the
  importance of the new reanalysis project carried
  out at NCEP.
• We have shown here that horizontal resolution is
  not critical for forecast of the TIO. However
  there are areas (Sahel) were resolution higher
  than T126 is beneficial. The next version of the
  CFS will be at T126. Could downscaling from T126
  provide results as good as the ones obtained with
  a CFS at T254 in these areas?
• The role of oceanic initial conditions has not
  yet been explored. How to improve the
  intraseasonal part of the ocean initial state?

37
Questions?
38
Number of strong summer TIO events during the
period of hindcasts 6
This is equivalent to 24-30 years of hindcasts
for assessing seasonal prediction skill
39
Initialization shocks

• The GDAS initial conditions are more compatible
  to the CFS atmosphere than CDAS2
• This difference could result to a stronger
  initialization shock when CFS is initialized by
  CDAS2
• We quantify the initialization shock by
  investigating forecast skill for the mean annual
  cycle

40
Forecast skill for the mean annual cycle of
U200 (verifying against CDAS2)
As expected there is a stronger initialization
shock when CFS is initialized by CDAS2 than by
GDAS. In fact bias correction improves the
forecast skill of the TIO when the CFS is
initialized by CDAS2. However bias correction is
not affecting the skill of the CFS when
initialized by GDAS (we obtain same results by
removing the observed mean state).
41
(No Transcript)
42
Tropical Atlantic
43
August 2002 weak tropical activity
OBS
44
August 2005 very strong tropical activity
OBS
45
Zonal mean of U 200 hPa averaged from 20S to 20N
Associated with the tropical Easterly Jet the
mean Boreal Summer tropical flow at 200 hPa is
non-divergent
There is strong intraseasonal variability of this
quantity during all seasons
46
Effects from removing the mean zonal signal
Standard Deviation of the total signal
Maritime Continent and Western Pacific
Standard Deviation of the total minus the zonal
mean signal
47
Skill for the zonal mean u200 (verification CDAS2)
Persistence Forecast
48
(No Transcript)
49
CDAS2 vs. GDAS

• Older version of GFS at T62L28
• This is a multi-year long estimation of the
  Atmospheric state obtained with the same, albeit
  older, model and same assimilation methodologies
• Quality is time-invariant

• Newer version of GFS at T254L64 and T382L64
• This is the best available estimation of the
  Atmospheric state obtained by the best model and
  assimilation techniques available each day
• Quality improves with time

50
GDAS vs. GPCP vs. Reanalysis-2 for June 2002
GDAS Precipitable Water
Reanalysis 2 Precipitable Water
GPCP Precipitation
drift
Time evolution of mean energy at wave numbers
10-40 when CFS is initialized by CDAS2 (red) or
by GDAS (blue).
51
(No Transcript)
52
(No Transcript)