NWS Training Slide Set

1
Automated Geostationary Satellite
Nowcasting of Convective Initiation The
SATellite Convection AnalySis and Tracking
(SATCAST) System John Mecikalski1 Kristopher
Bedka2 1 University of Alabama-Huntsville 2
Cooperative Institute for Meteorological
Satellite Studies (CIMSS), UW-Madison NWS
Training Slides Prepared 6 August 2006
2
Motivation

• Numerical models have significant problems
  nowcasting location/intensity of convective
  weather phenomena in the 0-6 hour time frame
• This is especially true over oceanic regions
  where poor initialization results in incorrect
  location/intensity forecasts for convective
  storms
• Since little real-time satellite-derived data is
  available in airplane cockpits, coupled with NWP
  deficiencies, mid-flight convective storm
  initiation and growth represents a significant
  hazard for aviation interests
• A major portion of the accidents from aircraft
  turbulence encounters are within close proximity
  to atmospheric convection (Kaplan et al, 1999)
• The cost of diverted flight can be as high as
  150,000 and a cancellation close to 40,000,
  depending on the size of the plane (Irrgang and
  McKinney, 1992)

3
Motivation (contd)

• The NASA sponsored Advanced Satellite Aviation
  weather Product (ASAP) initiative was started to
  better infuse satellite data into FAA Aviation
  Weather Research Program (AWRP) product
  development teams' (PDT's) aviation weather
  diagnostics and forecasts
• Geostationary satellites provide excellent
  coverage (both spatial and temporal) of regions
  prone to convective storms (60 S 60 N)
• - Since one can see the development of convection
  in satellite imagery, we sought to develop an
  algorithm to identify pre-convective initiation
  signatures and nowcast new convective initiation
  in real-time
• - Convective Initiation The first detection of
  significant precipitation echoes (gt 30 dBz) from
  cumulus clouds by ground-based radar

4
Datasets

• USE McIDAS to acquire and process
• GOES-12 1 km visible and 4-8 km infrared imagery
  every 15 mins
• - CI nowcasting techniques can be applied to any
  high-resolution ( 4 km) geostationary
  satellite sensor where satellite-derived winds
  are available
• - IR data interpolated to the 1 km visible
  resolution for direct relationship between IR and
  VIS analysis techniques
• UW-CIMSS visible/IR satellite derived winds for
  cloud motion assessment
• - Winds used to track cumulus features back in
  time for cloud-top trend estimates
• WSR-88D base reflectivity composite used for
  real-time validation
• - Composite also interpolated to the 1 km VIS
  resolution (not shown)

5
Evaluation of Pre-CI Satellite Signatures

• Integrate GOES satellite and WSR-88D radar
  imagery
• - Identified GOES IR TB and multi-spectral
  technique thresholds and time trends present
  before convective storms begin to precipitate
• - Studied numerous real-time and archived
  convective events with diverse mesoscale forcing
  regimes and thermodynamic environments
  continental (U.S. Great Plains) to sub-tropical
  (S. Florida)
• - Leveraged upon documented satellite studies of
  convection/cirrus clouds Roberts and Rutledge
  (2003), Ackerman (1996), Schmetz et al. (1997),
  Inoue (1987)
• - After pre-CI signatures are established, test
  on other independent cases to assess algorithm
  performance

6
CI Interest Field Criteria
From RR03
         
7
May 4, 2003 Convective Event

• Slow-moving spring storm produced 90 tornadoes
  across Kansas, Missouri, Tennessee, and Arkansas
• Western KS and NE convection produced mainly
  wind/hail damage

8
Convective Cloud Mask

• The foundation of the CI nowcast algorithmonly
  calculate IR fields where cumulus are present
• Utilizes time of day/year dependent brightness
  thresholding, brightness gradients, and
  brightness standard deviation techniques
• Collaboration with Dr. Udaysankar Nair (UAH) to
  implement statistical pattern-recognition based
  cumulus detection method by summer 2004

9
Multi-Spectral Band Differencing

• Compared multi-spectral techniques with
  co-located WSR-88D imagery to identify difference
  thresholds for cumulus in a pre-CI state
• 3.9 - 10.7 technique for cloud-top microphysics
  (Ellrod WF 1995, Setvak and Doswell MWR 1991)
  not used due to variation of 3.9 µm radiance with
  solar angle

10
Per-Pixel Cloud-Top Cooling Estimates

• Study of colocated GOES-8 10.7 µm TB and radar
  reflectivity pixel trends for stationary
  convective clouds along the Colorado Front Range
• Found that - 4C/15 mins (- 8C/15 mins)
  corresponds to weak (vigorous) growth

• BBy monitoring via satellite both the cloud
  growth and the occurrence of subfreezing
  cloud-top temperatures, the potential for up to
  30 min advance notice of convective storm
  initiation (gt 35 dBz), over the use of radar
  alone, is possible

Roberts and Rutledge, Weather and Forecasting
(2003)
11
Satellite-Derived Offset Vector (SOV) Technique
t-15 mins
235º _at_ 10 ms-1
1 km
12
Satellite-Derived Wind Analysis
850 hPa Analysis (winds in kts)

• 4 images at 15 min frequency used for winds
  Visible, 6.5 µm, and 10.7 µm
• - Reduced effect of NWP model background to
  better capture unbalanced mesoscale flows (i.e.
  anvil expansion, lower tropospheric outflow
  boundaries)
• Barnes analysis used to interpolate winds to 1
  km visible resolution
• - Wind field over 3 layers established (1000-700,
  700-400, 400-100 hPa) height assignment based on
  10.7 µm TB and NWP model temperatures

13
Cloud-Top Cooling Estimates Moving Cumulus
1930 UTC
2000 UTC
14
CI Nowcast Algorithm

• Nowcasts captured convective development well
  across eastern and north-central Kansas
• Conservative cloud growth threshold (4 C/15
  mins) can lead to greater false alarm occurrences
• Detailed analysis reveals lead times up to 45
  mins

CI Threshold
15
CI Nowcast Algorithm June 12th IHOP

• Since 5 min GOES-11 data was used, time trend
  thresholds are cut in half, resulting in noisy
  nowcasts for quasi-stationary convection in New
  Mexico
• TX Panhandle/OK convective development captured
  well

16
CI Nowcast Algorithm August 3, 2003

• Complex convective forcing from upper-level cold
  core cyclone, combined with lake breeze
  circulation
• Although noisy at first glance, CI over
  central/western IL identified up to 1 hour in
  advance
• Objective validation methodology very difficult
  to develop

17
Conclusions

• Through 1) identification of VIS cumulus
  clouds,
• 2) calculation of IR multi-spectral
  techniques,
• 3) tracking of cumulus cloud movement, and
• 4) estimation of IR cloud-top time trends,
• We have demonstrated skill in nowcasting CI and
  identifying growing cumulonimbus at 30-45 min
  lead times using current generation geostationary
  imagery
• Mecikalski, J. M., and K. M. Bedka Forecasting
  Convective Initiation by Monitoring the Evolution
  of Moving Cumulus in Daytime GOES Imagery.
  Submitted to IHOP_2002 Convective Initiation
  Special Issue of Monthly Weather Review, April
  2004.
• Hyperspectral satellite data will provide an
  unprecedented resource for
• 1) characterizing the 3-D thermodynamic
  environment near air-mass/mesoscale
  boundaries
• 2) identifying pre-CI signatures for moving
  cumulus
• 3) diagnosing the intensity/severity of
  existing convective storms

18
Validation Method A Non-objective Visual
Comparsion
2000 UTC

• A visual comparison of the CI nowcast to future
  radar imagery would likely yield the
  qualitative skill descriptions provided above
• Although this method may be good enough for
  most users (e.g. operational forecasters), people
  always want to know exactly how good the product
  is (e.g. correct nowcasts of CI occurrence 82 of
  the time)

2030 UTC
CI Threshold
2100 UTC
19
Validation Method B
Objective Tracking of Radar
Satellite
Step 1 Radar Tracking Algorithm
Step 2 CI Nowcast Pixel Advection
2000 UTC

• Go back to the CI nowcast (at 2000 UTC in this
  case) and advect flagged (red) pixels forward
  along the radar motion vector
• Identify the radar reflectivity at the nowcast
  time and at the new location 30 mins in the
  future
• LLook for dBZ increases from below 30 dBZ to
  above 30 dBZ. These are good CI nowcasts

2000 UTC
2030 UTC

• Use sequential radar imagery from t to t30 mins
  (hopefully with greater than 30 mins resolution)
  to determine where radar echoes moved for the 30
  min period after the nowcast was made