EarthSun System Division National Aeronautics and Space Administration

Earth-Sun System Division National Aeronautics and Space Administration AIRS Profile Evaluation and Assimilation Bradley Zavodsky Gary Jedlovec Shih-hung Chou November 21, 2005 SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Motivation for Evaluating AIRS Profiles Ø AIRS Science Team has done extensive global validation Ø Quantify regional errors in the AIRS soundings (collaborative with OU EOS activity) Ø Apply knowledge of localized errors to tune parameters for regional data assimilation High-End Methodology Ø RMS errors between regional RAOBs and collocated AIRS soundings for January 2004 (supports January 14 -17 case study) Ø Analyze differences between quality indicators in the AIRS V 4. 0 retrievals SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration AIRS Data – January 14 -17, 2004 Temperature and moisture profiles Ø ≈ 50 km spacing Ø 1 K in 1 km layer for T Ø 20% in 2 km layers for RH Ø profiles assigned quality values by science team Quality indicators Ø identify retrieval process Ø layer quality checks Ø only temperature for V 4. 0 Retrieval QI Flags (Vers. 4. 0) 2329 UTC 2141 UTC Full retrieval Sfc flagged Sfc+Bot+Mid flagged All levels flagged No retrieval IR image 15 January 2004 0030 UTC Distribution of AIRS Data by QI SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Application to Analysis Systems: AIRS Vertical Layers 700 h. Pa All retrievals Full Sfc + Bot flagged Sfc + Bot + Mid flagged Sfc flagged All Levels flagged Ø Collaborative work with Keith Brewster (CAPS, OU) has shown that AIRS data should be spread vertically over 1500 m layers in analysis systems Ø Results for January 2004 compares favorably Ø QIs have congruent trends but different magnitudes SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Application to Analysis Systems: Error Tables How do we use 1500 m this information layer in average ADAS? All retrievals Full Sfc + Bot flagged Sfc + Bot + Mid flagged Sfc flagged All Levels flagged Ø RMS errors improve as data quality improves Ø Errors slightly larger than AIRS Science Team values (0. 9 -1. 3 K, 20% globally) Ø Temperature trend as expected; moisture influenced by marine inversion Ø Generate analysis error tables using these results SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration AIRS Data Assimilation in WRF Establish assimilation methodology and demonstrate short term weather forecast improvement with AIRS profiles FY 04: Initial case studies over SEUS (reported on last year) o LAPS/MM 5 (previous experience with surface fields) o o AIRS Vers. 3. 6 un-validated soundings (mainly over land) Limited quality flags Previous work: Limited impact (mainly upper level temperature) FY 05: Adapt methodologies for ADAS / WRF - west coast US winter-time storm system (14 -16 January 2004) o ADAS (flexibility, tunable for unique datasets) o o o WRF- operational use by NCEP AIRS Vers. 4. 0 validated soundings – T & q ocean profiles only T - quality flags important for proper data assimilation SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration January 14 -17, 2004 Case Study Slow moving synoptic system off west coast – inadequate forecasts with conventional models Case selection o weather system over ocean varied cloud coverage from AIRS – multiple assimilation times o availability of AIRS version 4. 0 profiles o applicable to SPo. RT SEUS situations (data void over Gulf) o o 2141 UTC TC 9 U 232 Infrared image on 15 January 2004 SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration ADAS Bratseth Method Used iteratively to update a first-guess (or background) field provided by a model forecast. The correction, , at each grid point is given by where x(k+1) is the analysis for the kth iteration, x(k) is the analysis value at the grid point (background value if k =1), [ iobs - i(k)] is the value of the innovations (obs. - bckgrd), and xi is the weighting function. The xi is a function of observation and background error variances (error tables), distance of the observations from the grid point and is proportional to where rij and Δzij - horizontal / vertical distances between obs. and grid R and Rz - horizontal and vertical scaling factors. SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration ADAS Horizontal and Vertical Resolution Factors Vertical Resolution Factor Changes Resolution factors can control influence of AIRS data on resulting assimilated field • select factors consistent with AIRS vertical and horizontal resolution • relative magnitude w. r. t other assimilated data is important Influence of AIRS varies with ADAS constraints ADAS Resolution Factors used with AIRS Profiles SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Influence of Data Type in ADAS While error variances useful to quantify data errors, “representativeness” of the data type is important to establish relative weights of each data input • Vertical resolution and accuracy of AIRS – varies between T, q • Interplays with vertical/horizontal influence factors Data source weights used in ADAS – no raob Moisture Temperature Background AIRS Background If raobs are also assimilated, relative weights of AIRS data should be reduced (i. e. , AIRS error table values should be increased) AIRS values taken from V 4. 0 validation results SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration ADAS and AIRS Data Assimilation ADAS example: WRF 4 h forecast (upper left) used as background for 2200 UTC assimilation AIRS assimilated 850 mb T at 2200 UTC on 14 January 2004 - 4 h WRF as background o analysis of AIRS data (lower left) with selected QIs o AIRS and MADIS data assimilated to produce new analysis field (upper right) o ADAS Background Bckgrd+AIRS+MADIS AIRS analysis Impact of DA impact of assimilation (mainly AIRS data seen in difference field (lower right) – 0. 5 -1. 0 C changes o AIRS changes the analysis – does it change the forecast? SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration SPo. RT Research WRF for AIRS Assimilation Ø 30 km domain with 37 vertical levels Ø Initialized with NCEP 1° GFS grids, with 6 -h forecasts used as LBC Ø Numerical experiments: Control (CNTL): Conventional meteorological data (NO AIRS) AIRS Best Quality (FULL): Conventional AND full-retrieval AIRS Surface & Bottom Fail (SFBT): Surface- and bottom-failed AIRS SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Assimilation Cycles ADAS 4 h WRF Forecast 18 UTC 1/14 48 h WRF Forecast initialized from ADAS analysis at 00 UTC ADAS 2 h WRF Forecast 22 UTC 00 UTC 1/15 Validation every 12 h 00 UTC 1/17 Validation region 2329 UTC AIRS Swath 2141 UTC AIRS Swath SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Impact of AIRS Data on Temperature Forecast 24 -hour Forecast Valid at 00 UTC January 16 2004: 500 h. Pa L L CNTL 500 h. Pa Temperature AIRS FULL minus CNTL Ø AIRS impact on simulation + 2 ºC (significant differences) Ø Weaker thermal gradient along 500 h. Pa front SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Impact of AIRS Data on Moisture Forecast 24 -hour Forecast Valid at 00 UTC January 16 2004: 700 h. Pa 1. 6 L 1. 2 0. 8 L 0. 4 -0. 8 -1. 2 -1. 6 CNTL 700 h. Pa Mixing Ratio -2. 0 AIRS FULL minus CNTL Ø AIRS impacts moisture +2 g/Kg ØLess intense frontal boundary—consistent with 500 h. Pa temp analysis SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Verification Statistics 24 h fcst valid 00 UTC 16 Jan 2004 CNTL Ø Temp RMSE increases with height Ø Moisture RMSE decreases with height FULL (16%) Ø Slight T improvement 600 to 250 h. Pa Ø Degrades q at 700 h. Pa and below SFBT (60%) Ø More significant T improvement between 600 and 250 h. Pa Ø Improved moisture 700 to 500 h. Pa but degradation below 700 h. Pa SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration NEWT Error Tables Ø Reduce impact of lower quality AIRS by increasing error values Ø Retain impact of higher quality AIRS above failed quality layer SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration Verification Statistics 24 h fcst valid 00 UTC 16 Jan 2004 CNTL Ø Temp RMSE increases with height Ø Moisture RMSE decreases with height FULL (16%) Ø Slight T improvement 600 to 250 h. Pa Ø Degrades q at 700 h. Pa and below SFBT (60%) Ø More significant T improvement between 600 and 250 h. Pa Ø Improved moisture 700 to 500 h. Pa but degradation below 700 h. Pa NEWT Ø Improves moisture forecast near the surface Ø The suppression of poor-quality lowlevel information are not fully achieved – need a new error table SPo. RT SAC Nov 21 -22, 2005

Earth-Sun System Division National Aeronautics and Space Administration ADAS/WRF Summary Assimilation of AIRS profiles can have a positive impact on short-term weather forecasts • Make prudent use of AIRS profiles based on quality indicators (QIs) • Adapt ADAS parameters to represent AIRS vertical / horizontal resolution • Construct error profiles tailored for AIRS and relative to other data sets Working closely with AIRS science team (Susskind) on QIs • quality at each level • separate indicators for T and q Need for more case studies Evaluate use of AIRS vers. 5. 0 retrievals Smart data thinning based on QIs, meteorological gradients, etc. SPo. RT SAC Nov 21 -22, 2005