AIRS Profile Assimilation: RealTime Demonstration

1
AIRS Profile Assimilation Real-Time
Demonstration Brad Zavodsky Shih-hung Chou, Gary
Jedlovec, Bill Lapenta SPoRT Science Advisory
Committee June 13, 2007
2
Overview
Relevance to SPoRT Illustrates the ability to
assimilate AIRS L2 profile data into a numerical
weather prediction system in real time to aid
forecasting of sensible parameters for
short-term, regional, operational 6 48h
forecasts

• Motivation
• Insights From Past Case Study Work
• Real-Time Sample Timeline
• Example of Real-Time Web Interface
• Summary/Future Plans

3
Motivation

• Profiles may add value to WFOs running NWP
  systems not equipped to handle radiance data
  (e.g. MFL, MLB)
• moisture return over Gulf
• coastal processes
• weather features over oceanic regions
• Operationally, assimilation/forecasts must be
  completed in timely manner to be valuable for
  forecasters (e.g. available in AM for PM
  forecasts)
• demonstrate ability using real-time system with
  near-real-time data
• provide AIRS-enhanced initial conditions (ICs)
  to local WFOs
• Select new case studies and compile long-term
  statistics

4
Insights From Case Study Work

• Learned proper assimilation procedure for timely
  regional forecasts
• scale factors and error characteristics of
  background and observations
• use of quality indicators
• initialization time to take into account model
  spin-upassimilate only once
• Positive impact on the initial conditions
  varying results on regional forecasts
• Need a larger number of forecasts (i.e. larger
  set of statistics) to determine true value of
  adding AIRS profiles
• Must select proper case studies to show impact

5
Sample Timeline for Real Time Simulations

• 12-km WRF ADAS domains
• Non-parallel, ADAS analysis 15 nodes parallel
  WRF
• AIRS 48h forecast at 1500 UTC web products by
  1600 UTC
• 1600 UTC 1000 am CDT timely enough for
  forecasters to use for afternoon forecasts

Initialization by NAM 00Z analysis
Obtaining and using AIRS profiles in real-time is
not trivial!
6
Challenges Using Near-Real-Time AIRS Data

• Multiple NRT data sources through our
  collaborations
• NESDIS
• GES DISC
• UW direct broadcast

7
Sample Timeline for Real Time Simulations

• 12-km WRF ADAS domains
• Non-parallel, ADAS analysis 15 nodes parallel
  WRF
• AIRS 48h forecast at 1500 UTC web products by
  1600 UTC
• 1600 UTC 1000 am CDT timely enough for
  forecasters to use for afternoon forecasts

Initialization by NAM 00Z analysis
8
Real-Time Products to the Web

• Results posted to private, in-house website
• Daily Posting
• surface and pressure-level maps
• T, q, h, and V at 1000, 850, 700, 500 and 200 hPa
• difference fields T, q, h
• T, q, h at 1000, 850, 700, 500, and 200 hPa
• Delayed Posting
• precipitation difference fields
• verification statistics
• bias and RMSE for T and q against east coast
  RAOBs and NAM analysis
• qualitative precipitation forecasts (QPF)
• bias score and equitable threat score against
  Stage IV precipitation
• Website Real-Time AIRS Assimilation Website

Backup website
9
Summary

• Analysis/forecast system currently running in
  real-time for V4 AIRS profileswaiting for V5
• Assist in selection of case studies and
  calculation of long-term statistics of sensible
  weather parameters (e.g. precipitation)
• 48h forecast with AIRS data assimilation
  complete in 4 hours
• difference fields between control and
  AIRS-assimilated forecasts posted daily to web by
  1600 UTC
• in time to aid forecasters in making
  afternoon/evening forecasts using output
• WFOs can use AIRS-enhanced initial conditions
  for local model runs and help assess value of
  AIRS profiles in regional modeling

10
Future Work

• Continue collaboration with AIRS Science Team
  and University of Wisconsin to get V5 AIRS data
  onto the direct broadcast server as quickly as
  possible
• Monitor daily weather and AIRS impact to select
  case studies
• Calculate statistics over extended period of
  time for more thorough determination of AIRS
  impact
• AIRS-enhanced ICs to WFOsthey help assess value
• Optimize analysis/forecast
• introduce parallelized variational assimilation
  system
• explicit use of observation error in AIRS
  profiles
• parallelization speeds up analysis procedure
• improve analysis/forecast dynamic balance
  constraints
• use three-dimensional intelligent data thinning
  (IDT being developed through SPoRT partnership
  with FIT and UAH ITSC) to reduce data volume