SelfCalibrating Multivariate Precipitation Retrieval SCaMPR

1
Self-Calibrating Multivariate Precipitation
Retrieval (SCaMPR)

• Bob Kuligowski, Shuang Qiu, Jung-Sun Im
• NOAA/NESDIS/ORA

2
Theory / Schematic

• Algorithm Inputs (IR, MW, NWP)
• 6.9, 10.7, and 13.2-µm Tbs from GOES-12
• SSM/I rain rates from NRL
• AMSU rain rates from MSPPS (Ferraro)
• All GOES inputs are from GVAR feeds
• Algorithm Process (how the inputs are converted
  to rainfall estimates)
• Rain/no rain separation via discriminant analysis
• Rain rate estimation via stepwise linear
  regression
• IR Calibration Data Cube
• GOES inputs aggregated to SSM/I or AMSU
  footprints plus SSM/I or AMSU rain rates
• Most recent matched data used such that 5,000
  raining points are available (nonraining points
  are included, but the number does not matter)

3
SCaMPR Predictors
SCaMPR Predictands
GOES T6.9 µm
GOES T10.7 µm
GOES T13.3 µm
SSM/I Rain Rates
T10.7-T13.3
T10.7-T6.9
AMSU Rain Rates
S 0.568 (T10.7,min-217 K)
Derived Quantities
Gt-S (T10.7,avg-T10.7,min)-S
Aggregate to microwave footprint
Match predictor and predictand pixels (separate
sets for raining and non-raining predictor pixels)
Raining pixels
Dry pixels
Calibrate rain/no rain discrimination via
discriminant analysis
Calibrate rain rate estimation via multiple
regression
Apply to independent predictor data for rain rate
retrievals
4
Theory / Schematic

• Strengths and Weaknesses of Underlying
  Assumptions
• Only as reliable as the SSM/I and AMSU rain
  rates SCaMPR will perform poorly where they
  perform poorly
• Differences between SSM/I and AMSU rain rates
  make adjustments necessary before calibration
• Currently calibrated for CONUS as a whole, but
  geographic differences in Tb-rain rate
  relationships are inducing time-varying biases in
  SCaMPR rain rates
• Balance of having a long enough training period
  for statistically significant training, but a
  short enough training period to capture
  nonstationarity in relationships between
  predictors and rain rate no objective way to
  determine this

5
Theory / Schematic

• Planned Modifications / Improvements
• Currently transitioning from all-CONUS to
  regional calibration
• Soon to begin real-time ingest of NAM PW,
  1000-700 hPa mean RH, convective EL computed from
  T, q profile (already shown to have positive
  impact on case studies)
• Soon to begin real-time ingest of cloud-to-ground
  lightning from National Lightning Detection
  Network (NLDN)already shown to have positive
  impact on case studies
• Transition to global production planned for fall

6
Algorithm Output Information

• Spatial Resolution 4 km
• Spatial Coverage CONUS (2550N 125-65W)
• Update Frequency every 15 min
• Data Latency ???
• Source of Real-Time Data
• Soon to be made available on the Web (graphic
  images only) as a link from the Flash Flood Web
  page (http//www.orbit.nesdis.noaa.gov/smcd/emb/ff
  )

7
Algorithm Output Information

• Source of Archive Data
• Limited online archive at http//www.orbit.nesdis.
  noaa.gov/smcd/emb/ff/validation/validation.html
• Offline CONUS archive back to November 2004
• Capability of Producing Retrospective Data (data
  and resources required / available)
• SSM/I and AMSU rain rates can be reproduced from
  CLASS archive
• GOES data available to December 2003 from CLASS
  archive
• Eta data available from ORA tape archive back to
  1997