Assimilation of AIRS Data at NRL Assimilation of AIRS Data at NRL Benjamin Ruston, Clay Blankenship, William Campbell, Rolf Langland, and Nancy Baker Naval Research Laboratory, Monterey, CA, USA
Data Use Data Use • AIRS-324 channel subset, U1 and U2 alternate golfballs • Co-located with AMSU/A sensor, simultaneous assimilation • Thinned to approximately 300km resolution • Channels with sensitivity above model top (4mb) rejected • Ozone sensitive channels rejected • Near-infrared channel rejected in daytime • Approximately 4million observations per 6 hour watch before thinning and quality control
Channel Selection Channel Selection NOGAPS model top 4hPa, and no prognostic O 3 additional tools: ECMWF & MSC operational lists, adjoint senstivities
Data Thinning Data Thinning thinned to ~300km resolution
Quality Control Quality Control • Radiances modeled using JCSDA-CRTM • Histograms of observed minus simulated (ob-background) • Slope change in wavelength vs. (ob-background)* • Gross check on 2 window and 2 water vapor channels • Individual checks for each channel based on ob error *A. McNally and P. W atts, 2 0 0 3 , A cloud detection alg orithm for hig h-spectral- resolution infrared sounders. Q. J. R. Meteorol. Soc. , 2 0 9 , pp. 3 4 1 1 -3 4 2 3 .
Forecast Impacts Forecast Impacts Anomaly Correlations Anomaly Correlations AQUA impacts AQUA impacts (ocean only) (ocean only) AIRS+AMSU SH better AIRS+AMSU NH worse SH better NH slightly worse AMSU only ocean only improves SH slightly worse NH performance NH better
Forecast Impacts Forecast Impacts Tropical Cyclone Tropical Cyclone = = = = A LL TRO PI CA L STO RM S FO R 2 0 0 6 0 7 2 6 0 0 TO 2 0 0 6 0 9 0 2 0 0 = = = = N O A A A M SU N O A A A M SU + A Q U A A M SU /A I RS # storm # strom # storm # storm Tau dsl p1 ddi s dsl p2 ddi s2 s s s s 0 0 . 0 0 1 7 2 3 2 . 5 8 1 7 2 0 . 0 0 1 7 2 2 8 . 9 5 1 7 2 1 2 1 . 3 4 1 5 1 5 1 . 3 2 1 5 4 1 . 2 3 1 5 3 4 8 . 0 9 1 5 6 2 4 1 . 8 8 1 3 4 7 0 . 9 4 1 3 8 1 . 6 2 1 3 2 6 4 . 1 2 1 3 7 better 3 6 2 . 2 1 1 1 6 9 3 . 5 9 1 2 2 1 . 9 3 1 1 3 8 4 . 7 2 1 1 9 4 8 2 . 2 7 9 7 1 1 6 . 9 5 1 0 5 1 . 8 0 9 5 1 0 5 . 2 9 1 0 3 6 0 1 . 9 7 8 0 1 3 6 . 4 3 9 0 1 . 6 6 7 9 1 2 8 . 0 8 8 9 7 2 1 . 8 4 6 8 1 5 0 . 8 0 7 7 1 . 4 3 6 7 1 4 7 . 7 3 7 6 8 4 1 . 5 4 5 7 1 6 9 . 4 6 6 6 1 . 0 9 5 6 1 6 7 . 8 7 6 5 9 6 1 . 1 8 4 8 1 9 1 . 7 8 5 7 0 . 5 6 4 8 1 9 7 . 4 8 5 7 worse 1 0 8 0 . 6 9 4 0 2 2 4 . 0 6 4 9 0 . 0 1 4 0 2 3 7 . 9 7 4 9 1 2 0 0 . 2 7 3 3 2 6 3 . 8 5 4 1 -0 . 5 6 3 3 2 8 0 . 2 6 4 1
Adjoint Sensitivities Sensitivities Adjoint •Sensitivity to radiances assessed with adjoints of NAVDAS & NOGAPS •Energy-weighted forecast error norm (moist TE-norm) C = matrix of energy-weighting coefficients f = NOGAPS forecast t = verifying NAVDAS / NOGAPS analysis x = NOGAPS state vector (u, v, θ , q, p t ) e f has units of J kg -1 〈 , 〉 = scalar inner product Langland and Baker (Tellus, 2004), slide courtesy of Rolf Langland
Ob Impact Calculation Ob Impact Calculation Sensitivity gradient in observation space NAVDAS adjoint 0.5 deg, current to ops version of NAVDAS Observation Impact (J kg -1 ) Innovations assimilated for Xa Langland and Baker (Tellus, 2004), slide courtesy of Rolf Langland
Observation Impact Observation Impact 30 e 0.0 the observatio n is BENEFICIAL � < 24 30 e 0.0 the observatio n is NON - BENEFICIAL � > 24 30 e is an approximat ion of e e � � � 24 24 30 n Ob sensitivity summary: Ob sensitivity summary: Aug 17-31, 2006 Aug 17-31, 2006
Observation Impact Observation Impact Ob sensitivity summary: Ob sensitivity summary: Aug 15-26, 2006 Aug 15-26, 2006
Observation Impact Observation Impact good good Ob sensitivity summary: Ob sensitivity summary: Aug 15-26, 2006 Aug 15-26, 2006 spatial distribution spatial distribution bad good shows strong impacts shows strong impacts are generally outliers are generally outliers beneficial channels have beneficial channels have slightly positively slightly positively skewed distributions skewed distributions
1DVAR preprocessor 1DVAR preprocessor GOAL : improve atmospheric profiling over land improve the Land Surface Temperature (LST) reduce rejects over desert & elevated terrain ISSUES : - true background land emission (validation) - uncertainty in land surface temperature analysis - behavior of surface emission characteristics (scales) How do the emission characteristics vary ? - canopy properties - soil properties - surface roughness effects
1DVAR preprocessor 1DVAR preprocessor • Combined microwave and infrared • AMSU/A, AMSU/B, and HIRS/3 • AMSU, AIRS under development – Retrieve • profiles of T,q • Land Surface temperature (LST) – single value • Surface emissivity (spectrally) – a priori • Atmospheric profiles – NOGAPS 3,6,9 hr forecast interpolated to observation – error covariance global • Infrared Emissivity – Indexed surface type to spectral library – error covariance from retrieval statistics, NO channel correlation • Microwave Emissivity – JCSDA MEM – error covariance from retrieval statistics
Ancillary Data Ancillary Data • Vegetation Data (1km – global) – static (based largely on AVHRR 1992) – needed for indexing and input to JCSDA MEM • Soil Data (1/12 th degree – global) – static, modeled in many regions – needed for indexing and input to JCSDA MEM • Snow cover – Air Force product • Sea Ice – NRL in-house analysis • ASTER spectral library – Infrared spectra of soils and vegetation
Initial emissivity emissivity estimate estimate Initial Infrared : Land Databases indexed to spectral library Microwave : JCSDA MEM
1DVAR retrieval 1DVAR retrieval
1DVAR retrieval 1DVAR retrieval
Scan Dependence Scan Dependence mean from retrieval investigation indicates weak dependence Δ ε ( mean – LUT*) empirical correction possible *LUT: look-up-table, emissivities from ASTER spectral library indexed to soil and vegetation databases
Summary Summary - AIRS assimilation cycling in NRL NAVDAS/NOGAPS system - forecast impacts mixed, continuing to broaden study, examine quality control, bias correction - observation sensitivities helping to guide channel selection assess sensor performance Future Work Future Work - AIRS experience to help guide transition to and use of IASI - 1dvar preprocessor used to estimate effective land surface emissivity and assimilate sounding channels over land - NAVDAS operational with COAMPS, test regional assimilation - Cloudy radiance assimilation with COAMPS (5km inner nest) - NAVDAS-AR (4DVAR) better scalability with increase spectral spation observation density
Scan dependence Scan dependence greatest at lower freq dampens with increasing freq & vegetation
Scan dependence Scan dependence MEM generally good, slight overcorrection at edge of scan IR small dependence
Recommend
More recommend
