Adaptive targeting in OSSE

1
Adaptive targeting in OSSE
Yucheng Song and Zoltan Toth

• Outline
• Adaptive observing / data processing
  techniques in OSSE
• Addition to OSSE
• Link with THORPEX
• Link with T-PARC

2
(1)Adaptive data observing/processing techniques
in OSSE

• Test methods/platforms/application in OSSE
  framework
• Develop software into OSSE
• Ensemble (T126 or T170) product generation in
  OSSE
• ETKF targeting strategy (certain instruments)
• Evaluate data impact by certain instruments like
  UAS, Doppler Wind Lidar

3
NCEP Operational GEFS

• NAEFS (NCEP/GEFS)
• 80 perturbations in cycling (see next slide)
• Replaced previous 56 perturbations in ensemble
  transform (ET) cycling
• 20 perturbed long forecasts (16-d) in each cycle
• Replaced previous14 long forecasts in each cycle

4
6 hours ET cycle NCEP ensemble (ET)
Re-scaling
6hrs
Up to 16-d
Next T00Z
T00Z 80m
Re-scaling
T06Z 80m
Up to 16-d
Re-scaling
T12Z 80m
Up to 16-d
Re-scaling
T18Z 80m
Up to 16-d
5
Concept of ET KF

• ET KF Ensemble Transform Kalman Filter for short
• ET KF provides a framework for estimating the
  effect
• of observations on forecast error covariance
• ET KF uses ensemble transformation and a
  normalization
• to obtain the prediction error covariance matrix
  associated
• with a particular deployment of observational
  resources
• Linearity is assumed for ensemble transformation

6
ET KF formulation
7
ET KF formulation
8
Targeting methods - ETKF
Dropsondes to be made by G-IV
Storm
The ETKF spotted the target area
Expected error reduction propagation
9
MAIN THEME
Study the lifecycle of perturbations as they
originate from the tropics, Asia, and/or the
polar front, travel through the Pacific
waveguide, and affect high impact wintertime
weather events over North America and the Arctic
Influence of tropical Flare-ups in western
Pacific (IR) on deep cyclogenesis in northeast
Pacific captured by Ensemble Transform targeting
method
10
Better adaptive strategy if implemented (examples)
The optimal sampling region located in the jet
core
11
(2)Additions to OSSE

• Assess threat of high impact events based on
  ensemble automatically pick high impact events
  at 3-day leading time
• Run ET/ET KF targeting for each high impact case
• Dispatch observing systems/data processing
  resources (before and inside DA)
• Wind Lidar, UAV etc.
• Assimilate targeted data (carry out adaptive data
  processing)
• Evaluation (EXAMPLES NEXT FROM WSR)

12
Impact of Data
Surface pressure
Precipitation
Contours are 1000mb geopotential height, shades
are differences in the fields between two
experiments
500mb height
250mb height
13
Forecast verification
500mb height
Sea Level Pressure
Red contours show forecast improvement due to WSR
dropsondes, blue contours show forecast
degradation
250mb height
14
Forecast Verification for Temperature
(Measure by root-mean-square errors)
10-20 RMS error reduction in Temperature
60 hr forecast is equivalent to 48hr forecast
RMS error reduction vs. forecast lead time
15
(3)Link with THOPREX

• THORPEX A World Weather Research Program
  (WWRP)
• Accelerate improvements in skill/utility of 1-14
  day weather forecasts
• Long-term (10-yrs) global research program in
  areas of
• Observing system, data assimilation, numerical
  modeling/ensemble, socioec. appl.
• Strong link with operational Numerical Weather
  Prediction (NWP) centers
• International program under WMO

16
THORPEX evaluation metrics (1)

• Possible new probabilistic guidance products for
  high impact events
• Hydrometeorology
• Extreme hydro-meteorological events, incl. dry
  and wet spells (CONUS)
• Quantitative extreme river flow forecasting
  (OCONUS)
• Tropical / winter storm prediction
• Extreme surface wind speed
• Extreme precipitation (related to wet spells)
• Storm surges
• Aviation forecasting
• Flight restriction
• Icing, visibility, fog, clear air turbulence
• Health and public safety
• Hot and cold spells

17
THORPEX evaluation metrics (2)

• Legacy NCEP internal probabilistic scores to
  assess long-term progress
• General circulation
• Probabilistic 1000 500mb height forecasts
• Storm
• Strike probability for track
• Probability of intensity (central pressure or
  wind-based)

18
(4) T-PARC interestsGlobal optimal positioning
of observing systems in OSSEImprove forecast
accuracy
19
T-PARC PROPOSED OBSERVING PLATFORMS
Day 3-4 Radiosondes Russia
NA VR
Day 5-6 Radiosondes Tibet
CONUS VR
D 2-3 G-IV
D 1-2 C-130 UAS
D-1 UAS P-3
Day 3-4 GEMS Driftsondes Aerosondes
Extensive observational platforms during T-PARC
winter phase allow us to track the potential
storms and take additional observations as the
perturbation propagate downstream into Arctic and
US continents
20
Before and after field campaign

• Nature is defined as a series of states
  corresponding to the real atmosphere
• Generated by very high resolution model runs
  nudged by operational analysis (GDAS)
• Advantages
• Use T-PARC type OSE to calibrate OSSE system
  much easier to calibrate, community will be
  convinced if we can reproduce their OSE work
• Retrospective work after T-PARC T-PARC represent
  only one configuration of global observing
  system, with OSSE such defined, many other
  configuration can be tested
• This is an alternative

21
Advantages (more)

• Ease of calibration (one-to-one comparison, can
  quantitatively evaluate osse system based on a
  SINGLE (or few) case(s), instead of requiring a
  large sample of cases
• Close to realistic representation of model
  related uncertainty
• No need to painstakingly evaluate or amend osse
  nature run
• Can use humidity (cloud, moisture) observations
  from real world to decide if certain observations
  can be made or not in osse world - potentially a
  big contribution to making osse real life-like
• Same nature can be redone with higher resolution
  or other type of model (using operational
  analysis as forcing) - direct comparison of
  different OSSE systems possible
• Estimate how proposed new observing systems
  would help analysis/forecast for real life
  significant events (Katrina, etc)
• Post field campaign analysis Add significant
  value by osse testing of alternative deployments
  (after calibration in which actual and simulated
  field phase observations are assimilated and
  their impacts are compared in both OSE and OSSE
  framework

22
Concern

• Improved analysis might not mean improved
  forecast for individual cases
• We think statistically it will improve forecasts

23
OSSE strategy

• 1. Implement ET similarly as NCEP operational
  Ensemble forecast system
• Coding development
• Initial conditions (Data analysis
  from conventional data radiance data
  assimilation)
• 2. Targeting strategy similarly as WSR
• Identify typical storm cases in the Nature
  run
• use targeting strategy to find sensitive
  areas to target
• 1. Increase data resolution in sensitive
  areas (adaptive grid)
• 2. Direct observation

24
T-PARC interests (Ideas can be tested in OSSE)

• Rossby-wave plays a major role in the development
  of high impact weather events over North America
  and the Arctic on the 3-5 days forecast time
  scale
• Additional remotely sensed and in situ data can
  complement the standard observational network in
  capturing critical processes in Rossby-wave
  initiation and propagation
• Adaptive configuration of the observing network
  and data processing can significantly improve the
  quality of data assimilation and forecast
  products
• Regime dependent planning/targeting
• Case dependent targeting
• New DA, modeling and ensemble methods can better
  capture and predict the initiation and
  propagation of Rossby-waves leading to high
  impact events
• Forecast products, including those developed as
  part of the TPARC research, will have significant
  social and/or economic value

25

• Sequence of analysis fields
• Dynamically consistent NOT COMPLETELY
• Lack of consistency interferes with forecast
  evaluation
• Only analysis quality can be evaluated directly
• NATURE MODEL CAN BE RUN ALONG WITH OSSE FCST
• Dynamics/physics different from assimilating
  model MOST REALISTIC REPRESENTATION OF MODEL
  ERRORS?
• PERFECT MODEL SCENARIO NOT POSSIBLE
• Differences should correspond to difference
  between nature our models
• No difference means perfect model assumption,
  THORPEX interest
• Realistic - YES
• Climate stats matching reality - YES
• Moisture variables realistic so obs locations can
  be chosen realistically
• YES
• Same weather as in nature - YES
• Allows direct comparison between OSSE OSE
  results for reliable calibration using small
  amount of data - YES