Topic 3.3: Targeted observations and data assimilation in track prediction

1
Sixth International Workshop on Tropical Cyclones
Topic 3.3 Targeted observations and data
assimilation in track prediction
Rapporteur Chun-Chieh Wu PSA Working
Group Sim Aberson, Brian Etherton, Sharanya
J. Majumdar, Seon Park, Melinda S. Peng, Zhaoxia
Pu, Michael Morgan, Steve Tracton, Samuel
Westrelin, and Munehiko Yamaguchi
2
Outline

• Introduction
• Surveillance programs using the
  dropwindsondes
• Targeted observations for tropical cyclones
• Comparison of targeted observing methods
• Other data to be targeted and assimilated
• Issues and concerns
• Recommendations

3
Improving the understanding and forecasting of TCs
Dynamics of the typhoon system
Cost effective?
Dynamics of the model
Initial condition
Multi-scale interaction Air-sea
interaction Terrain/PBL effect
Data assimilation and/or Initialization
4
Introduction

• Accurate forecast of tropical cyclones
• Realistic numerical models
• Accurate representation of meteorological fields
• Observation data
• Surface observations, soundings, and ships
• Dropwindsonde data
• Satellite data
• Radar data
• Data assimilation

5
Surveillance programs using the Dropwindsondes

• The impact of dropwindsonde data
• Between 1982 and 1996, the HRD conducted 20
  synoptic flow experiments.
• Burpee et al. (1996)
• The average error reductions in the consensus
  forecasts from three dynamical models varied from
  16 to 30.

6
Dropwindsondes

• In 1997, the HRD began operational synoptic
  surveillance mission with the G-IV jet aircraft.

• Aberson and Franklin (1999)
• The dropwindsonde observations improved the mean
  track forecasts of the GFDL model by as much as
  32.

7
Dropwindsondes

• Aberson (2002)
• The additional dropwindsonde data from the
  synoptic surveillance missions provided
  statistically significant improvements in the
  GFDL forecast only at 12 h.
• TC vortex initialization schemes
• The amount of data coverage
• Aberson (2003) and ongoing
• The improvement through targeted observations

8
DOTSTAR (Dropwindsonde Observations of Typhoon
Surveillance near the Taiwan Region)
6
Astra jet of AIDC
9
(Wu et al. 2005a, BAMS)
9
DOTSTAR observations
Up to 2006, 24 missions have been conducted in
DOTSTAR for 20 typhoons, with 386 dropsondes
deployed during the 129 flight hours. 18 typhoons
affecting Taiwan 8 typhoons affecting mainland
China 4 typhoons affecting Japan 2 typhoons
affecting Korea 5 typhoons affecting Philippines
21. Kaemi
20. Bilis
23. Saomai
22. Bopha
24. Shanshan
10
The impact of DOTSTAR data on global models in
2004
Melinda Peng
Sim Aberson
NCEP GFS 14
NOGAPS 14
Tetsuo Nakazawa
JMA GSM 19
ENSEMBLE 22
(Wu et al. 2006a, WF)
11
Background on targeted observations

• Adaptive observations observations targeted in
  sensitive regions can reduce the initial
  conditions uncertainties, and thus decrease
  forecast error.
• Targeted observation is an active research topic
  in NWP, with plans for field programs, tests of
  new observing systems, and application of new
  concepts in predictability and data assimilation.
  (Langland 2005)
• Factors associated with adaptive observations
• - Observation density, variables and errors
• - Magnitude of uncertainty
• - Data assimilation system
• - Growth of uncertainty

12
Adaptive observation strategies

• Dynamics-based strategy
• SV, adjoint sensitivity, and PV.
• Uncertainty-based strategy.
• Ensemble variance
• Joint dynamics-uncertainty based strategy.
• The ideal one would be the strategy
  that use both of dynamics and uncertainty
  information (e.g., ETKF, VARSV).
• (Since 1997, developed for mid-lat, FASTEX)

13

• Since 2003, several objective methods, have been
  proposed and tested for operational surveillance
  missions in the environment of Atlantic
  hurricanes conducted by HRD/NOAA (Aberson 2003)
  and NW Pacific typhoons by DOTSTAR (Wu et al.
  2005).
• NCEP/GFS ensemble variance
• (collaborating with Aberson)
• ETKF
• (collaborating with Majumdar)
• NOGAPS Singular Vector
• (collaborating with Reynolds and Peng)
• Adjoint-Derived Sensitivity Steering Vector
  (ADSSV)
• JMA moist Singular Vector
• (collaborating with Yamaguchi)

(Aberson 2003, MWR)
(Majumdar et al. 2006, MWR)
(Peng and Reynolds 2006, JAS)
(Wu et al. 2006b, JAS)
14
Comparison of targeted observations in DOTSTAR
Ensemble Variances, Toth and Kalnay (1993)
ETKF, Bishop and Majumdar (2001)
FNMOC SV, Palmer et al. (1998)
ADSSV, Wu et al. (2006)

• DOTSTAR
• (Wu et al. 2006b)

• G-IV surveillance
• Comparison of
• targeted techniques
• (Etherton et al. 2006)
• Maumdar et al. 2006
• Reynolds et al. 2006

More comprehensive comparisons are ongoing.
15
How the dropsonde data improve the forecast?
Typhoon Conson (2004) as an example
(Nakazawa 2004, THORPEX meting)
Typhoon Conson (2004) 8 June 1200UTC
JMA-GSM
16
Evaluate a SV method as a strategy for Targeting
Observation
JMA has executed Observing System Experiments
(OSEs) to investigate the usefulness of the
singular vector method as a strategy for
sensitive analysis.

• For the initial time of 12UTC 08 June 2004 when
  totally 16 dropsondes were dropped into typhoon
  CONSON by the DOTSTAR (Dropsonde Observation for
  Typhoon Surveillance near the Taiwan Region)
  project, 4 predictions with JMA Global Spectral
  Model (TL319L40) about the use of the dropsondes
  in the global 4D-Var analysis are executed.
• all dropsonde observations are used for making
  the initial condition
• dropsondes are not used at all
• only 3 data within a sensitive region are used
  (4, 9, 12)
• only data outside of a sensitive region are used
  (6, 8, 10, 13, 15, 16)

(From Yamaguchi)
Sensitive analysis result

• The distribution means vertically accumulated
  total energy by the 1st moist singular vector.
• Targeted area for the SV calculation is N25-N30,
  E120-E130.
• Optimization time interval is 24 hours.

x
CONSONs center position
17
OSEs result on CONSONs track forecast
Red (I) all dropsonde observations are used for
making the initial condition Blue (II)
dropsondes are not used at all Green (III) only
3 data within a sensitive region are used (4, 9,
12) Water (IV) only data outside of a
sensitive region are used (6, 8, 10, 13, 15, 16)
(From Yamaguchi)
(III)
(I)
(IV)
similar
(II) is almost same with (IV)
18
Other data to be targeted and assimilated

• Observations for data assimilation
• To date, Targeted observations for TCs are
  mainly dropwindsondes deployed from the aircraft.
• There is considerable scope for extending
  targeted observing strategies to include other
  types of data, most prominently from satellites.
• GOES (Zou et al. 2001) and TRMM (Pu et al. 2004)
• Microwave radiances (Bauer et al. 2006a, b).
• The collection of satellite and in-situ data from
  field programs (e.g. CAMEX-4, Kamineni et al.
  2006) with different spatial and temporal
  resolutions and error characteristics (Fisher
  2003 Berre et al 2006, Westrelin et al 2006)
  will continue to play a very important role in
  improving tropical cyclone track forecasts.

19
Other data o be targeted and assimialted

• Questions more specific to targeted observations
  can be addressed over the next decade
• Given the abundance of satellite data that will
  be available for assimilation, what subsets of
  the data are the most necessary for assimilation
  to improve the tropical cyclone forecast?
  (satellite data thinning)
• What are the optimal variables, three-dimensional
  structures, and spatial and temporal density that
  are necessary for observation?

20
Issues of concerns (Langland 2005 and THORPEX)

• Although the impact of observations is greater
  when selected in a sensitive area, the few
  observations deployed may not make a substantial
  impact on the forecasts.
• The statistical evaluation of the significance of
  the measured impact requires a large number of
  cases.
• Current diagnostics used to evaluate forecasts
  provides a good assessment of the validity of
  forecasts (skill), but it may not be sufficient
  to reveal whether these improvements are relevant
  to applications (value).
• The use of climatological sensitivities may lead
  to improvements on average and be more cost
  effective than targeted observations on demand.
• Overall, there was a considerable question as to
  the value of targeting, especially when isolated
  from the more general issues of observing system
  sensitivities in design of an optimal mix of
  available observing platforms.

21
Recommendation

• Need to assess the influence of the data
  assimilation scheme on the effectiveness of
  targeted observations.
• More studies of varying definitions,
  interpretations, and significance of sensitive
  regions (e.g., different methods, metrics)
• More work on sampling strategies in sensitive
  areas, e.g., immediate storm environment for
  shorter range prediction versus remote areas
  relevant to longer range forecasts including
  the impact of large scales in meso-scales models.
• More work on metrics to assess the impact of
  targeting or more generally on any changes in
  the observation network.
• Emphasis of the potential value of OSEs and OSSEs
  in assessing potential observing system impacts
  prior to actual field programs.
• Stronger efforts to develop alternative observing
  platforms (other than the dropwindsondes) for
  targeting, especially adaptively selecting
  satellite observations by revising the data
  thinning algorithms currently used.
• Improvement and continuous refinement of targeted
  observing strategies.

22
THORPEX-PARC Experiments and Collaborating
Efforts (from Dave Parsons)
Upgraded Russian Radiosonde Network for IPY
Winter storms reconnaissance and driftsonde
NRL P-3 and HIAPER with the DLR Wind Lidar
JAMSTEC/IORGG