This document discusses strategies for adaptive weather observation to improve forecasting. It presents a case study on Hurricane Floyd simulations with different observation strategies: 1. A control simulation with high-resolution observations matched the actual storm track well. 2. A coarse simulation deviated significantly due to lower resolution. 3. Targeted observations around the storm (654 observations) significantly improved the coarse forecast track. Even fewer observations (100 or 50) led to similar improvements. 4. Observations along the predicted track were less effective than initial targeted observations around the storm location. 5. Adaptive observation strategies that focus on high-impact initial observations could dramatically improve forecasts compared to regular observation arrays.

1. Adaptive Observation Strategies
for Advanced Weather Prediction
Center for Atmospheric Physics
David P. Bacon, Zafer Boybeyi
Center for Atmospheric Physics
Science Applications International Corporation
Michael Kaplan
Dept. of Marine, Earth, & Atmos. Sci.
North Carolina State University
October 30, 2001
David P. Bacon
(703)676-4594
david.p.bacon@saic.com

2. Weather Forecasting – A Primer
Center for Atmospheric Physics
• Weather forecasting relies on a system of systems:
– In-situ observations (surface, balloon, & aircraft)
– Remotely-sensed observations (satellite-based)
– Data assimilation (creating a physically consistent 3-D dataset)
– Prognostic models (extrapolation into 4-D)

3. Center for Atmospheric Physics
In-situ Observations
• Irregular spatial distribution
– Where the people (and planes) are
• Quasi-regular temporal distribution
– Surface observations
» Hourly with special reports
for significant weather
– Balloon observations
» Twice-daily at 0000Z & 1200Z
– Aircraft observations
» Regular intervals

4. Remote Sensing Observation Strategies
Center for Atmospheric Physics
• The beginning:
– TIROS-1 (April 1, 1960)
» Television and InfraRed Operational Satellite
– ATS-1 (December 6, 1966)
» Applications Technology Satellite – Geostationary
» Communications system test bed
• Raster based systems
– Vidicon tubes to CCDs

5. Center for Atmospheric Physics
Data Assimilation
• Goal: A physically consistent 3-D representation
of the atmosphere at an instant of time
• Requirements: Rationalize a disparate set of data that is a
mix of irregular and grid point information at
synoptic (0000Z & 1200Z) and asynoptic times
• Modify large-scale effects to reflect small-scale features
– Capture steep gradients critical to severe weather

6. Center for Atmospheric Physics
Prognostic Models
• Most models are based on a rectangular grid structure
– Intuitively obvious
– Simplest tiling
– Simple operator decomposition
• Higher resolution is obtained by nesting
– Conceptually easy
– Numerically tricky
» Reflective internal boundaries
» Differing surface conditions
» Scale interactive information exchange
• Initial conditions obtained from data assimilation system
– Generally at synoptic times

7. Center for Atmospheric Physics
A Vicious Circle
• Increased horizontal (α) and vertical (β) resolution in models leads to:
– Higher resolution initial conditions: α2β
– Increased run-time: α2β
– Increased output volume: α2β max(αααα,,,, ββββ)
• Higher resolution ICs requires higher resolution observations

8. Dynamic Adaptation of Data Assimilation
Center for Atmospheric Physics
and Model Resolution
• Initial grid for an
OMEGA simulation of
Hurricane Floyd and
the grid (inset) at 72
hours

9. OMEGA Forecast of Hurricane Floyd
Center for Atmospheric Physics

10. Center for Atmospheric Physics
Control Simulation
• Initialization:
– 1200Z September 13
– NOGAPS Analysis
• Boundary Conditions:
– NOGAPS Forecast
• Grid Resolution:
– 5 - 15 km

11. Verification of Control Simulation
• Storm Tracks
– White: Observations
– Red: Control
Center for Atmospheric Physics

12. Center for Atmospheric Physics
Coarse Resolution Simulation
• Initialization:
– 0000Z September 14
– NOGAPS Analysis
• Boundary Conditions:
– NOGAPS Forecast
• Grid Resolution:
– 75 - 120 km

13. Verification of Coarse Resolution Simulation
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
• Significant westward track
deviation
Center for Atmospheric Physics

14. Center for Atmospheric Physics
Adaptive Observations
• Control Run provides a
source of pseudo-observations
for OSSEs
• Control cell centroid closest to
the Coarse cell centroid used
to provide pseudo-sounding

15. Case #1: 654 Targeted Observations
• Coarse Resolution
Configuration
• Added 654 observations
– All cell centroids in box
around storm
Center for Atmospheric Physics

16. Verification of Case #1 (654 Targeted Observations)
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
– Orange: Case #1 (654)
• Noticeable improvement in
forecasted track
Center for Atmospheric Physics

17. Case #2: 100 Targeted Observations
• Coarse Resolution
Configuration
• Added 100 observations
– Regular 10 x 10 array
around storm
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
– Orange: Case #1 (654)
– Green: Case #2 (100)
• Virtually identical track to
Case #1 with only 20% of
the targeted observations
Center for Atmospheric Physics

18. Case #3: 11 Targeted Observations
• Coarse Resolution
Configuration
• Added 11 observations
– Around initial storm
location
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
– Cyan: Case #3 (11)
• Very minor difference from
Coarse resolution simulation
Center for Atmospheric Physics

19. Case #4: 50 Targeted Observations
• Coarse Resolution
Configuration
• Added 50 observations
– Along forecasted track
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
– Pink: Case #4 (50)
• While this case improved the
forecast early on, it did not
improve the track at later
times as much as Case #1 or
Case #2
Center for Atmospheric Physics

20. Case #5: 25 Targeted Observations
• Coarse Resolution
Configuration
• Added 50 observations
– Along forecasted track
• Storm Tracks
– White: Observations
– Red: Control
– Yellow: Coarse Res
– Cyan: Case #5 (25)
• Track nearly identical to
Case #4.
Center for Atmospheric Physics

21. Center for Atmospheric Physics
Conclusions and Ramifications
• Targeted observations can have a dramatic impact on storm forecasts
– Improvement in initial storm conditions has the largest payoff
– Other scenarios may have different requirements
• Greatest improvement will come in identifying and obtaining key
observations for developing convective storms
– The critical scales are so small and hence the volume of regular
arrays of observations is so large that either the communications
become a dominant problem or the extraction of a “signal” from
the “noise” of data bits prevents utilization
• The routine utilization of targeted as opposed to general satellite
observations has significant impact on satellite operations
– Timeliness is key ⇒Communication & data mining issues
• A tightly linked forecast and observational system can address these
issues as well