Presentation on theme: "New IR Surface Emissivity Estimates to be Evaluated for CRTM Ron Vogel (IMSG) MW / IR Emissivity Group Meeting July 7, 2008."— Presentation transcript:
New IR Surface Emissivity Estimates to be Evaluated for CRTM Ron Vogel (IMSG) MW / IR Emissivity Group Meeting July 7, 2008
Outline Brief review of new methodologies for estimating IR surface emissivity –1D-Var tuning of emissivity index (Ruston) –PC-based MODIS high spectral (Borbas) –Hyperspectral satellite retrievals –New reflectance spectra for surface-type classes Comparisons of emissivity methods –SARTA simulation of AIRS using Borbas, Li, & AIRS-standard emissivity (from Borbas) –CRTM simulation of AVHRR using NRL emissivity Current steps Future steps Discussion of priorities for IR emissivity development
1D-Var tuning of Indexed Emissivity (Ruston, NRL) Indexed Emissivity: Laboratory measured reflectances indexed to vegetation and soil databases Example: grid cell contains 50% Open Shrubland and 50% Wooded Grassland Rveg = 0.5 (1/8 Rconifer + 1/8 Rdeciduous + ¾ Rgrass) (1/4 Rconifer + 1/4 Rdeciduous + ½ Rgrass) Rsoil similar using sand/silt/clay reflectances Seasonal dynamics from temporal greenness variation Use 1D-Var cost function to minimize the fit of model radiances to HIRS or AMSU observations. Retrieves emissivities for HIRS/AMSU channels using Indexed Emissivity as first guess. NEXT STEP: Tune first-guess Indexed Emissivity to improve 1D-Var retrieved emissivity. Repeat tuning until error in retrieved emissivity is small. Use final tuned Indexed Emissivity in CRTM? –0.5 degree spatial resolution –0.01 μm spectral resolution in range 3-14 μm –No match-up of surface classes to GSI (emissivity determined by user’s input lat/lon of grid cell, greenness & wavelen) 4.3 μm 10.8 μm
PC-based MODIS high spectral emissivity (Borbas, U. Wisconsin) Baseline-fit emissivity (BF) fits MODIS emissivity at 6 wavebands to conceptual model of emissivity spectrum based on 123 spectra of lab- measured reflectance. Results in emissivity at 10 “hinge point” wavelengths. PC regression of baseline-fit emissivity using PC’s of 123 lab reflectance spectra to calculate High Spectral Resolution (HSR) emissivity –416 spectral points in range 3.6 – 14.3 μm –Monthly temporal resolution, 0.05 deg (5km) spatial resolution –Needs monthly 40MB BF emissivity file and PC algorithm to compute HSR emissivity Possibly will develop 5-year climatology of global monthly emissivity at 0.05 deg spatial resolution BF PC- HSR
Satellite Retrieval of IR Emissivity Hyperspectral Regression: IASI/AIRS (L. Zhou, NESDIS) –Will process IASI Mar 1 – Apr 15, 2008, instantaneous retrieval, all view angles AIRS – U. Wisconsin emissivity algorithm (Jun Li, UW) –Will process AIRS Jan 1 – Feb 15, 2004, instantaneous retrieval, all view angles AIRS – Standard emissivity algorithm (AIRS Sci Team) –Swath (45 km); Daily, 8-day, Monthly (1 deg grid) –Available from GSFC DAAC
IR emissivity from laboratory-measured reflectance spectra (Vogel, based on Ruston) Created new reflectance spectra for new NCEP surface classification (20 categories). Eliminates need to match CRTM surface types with GFS/GDAS surface types. Version 1 (pre-Beta): Utilizes NRL method for generating reflectance spectra for a particular classification scheme. –Uses reflectances from JPL spectral library –Simplification of NRL method for whole class (no grid or temporal variation) –Example: Open Shrub = ¼ (Rdecid) + ¼ (Rconifer) + ½ (Rsoil) NCEP wants CRTM to implement old GFS classes first, new classes later. Example new classes
Reflectance comparisons: CRTM vs new classes CRTM class New NCEP class CRTM class New NCEP class
Comparison of 4 emissivity methods in SARTA simulation of AIRS BT BT Residual (calc – obs) BT Residual (calc – obs) Longwave IR Shortwave IR 1.HSR emis V4 2.HSR emis V5 3.BF emis V4 4.BF emis V5 5.Std AIRS emis 6.UW AIRS emis Emis Methods: From E. Borbas, Wisc
Comparison of emissivity estimates using CRTM CRTM run with new emissivity estimates and compared to satellite observations 1.Match satellite pixel to GDAS grid location –GDAS Atmos profiles: temp, pressure, humidity, ozone 64 vertical layers, 768 x 384 global grid –Surface parameters: surf temp, snow depth, vegetation type 1152 x 576 global grid –Profile chosen closest to satellite pixel –6-hour GDAS field time-interpolated to satellite pixel time 2.Run CRTM with 2 emissivities for locations matched to satellite locations 1.current internal emissivity (control) 2.new emissivity added externally (test) 3.Compare CRTM brightness temp to satellite brightness temp for both control and test Satellites to compare: –AVHRR (NOAA-18, GAC) channels 3, 4, 5 (3.76, 11, 12 μm ) days: (winter), (summer) use cloud mask to select cloud-free pixels –HIRS channels 8, 10, 19 (11, 12, 3.76 μm ) days: none selected yet cloud mask?
Results: CRTM simulation of AVHRR using NRL emissivity Difference between CRTM and AVHRR is high: Up to -20 K (daytime, shown at right) for both CRTM current emissivity and NRL emissivity. Bias of -4 – +6 K (globally, day+night), depending on surface type, for both emissivity methods. RMSE between CRTM and AVHRR is high for both methods: 3 – 8 K Surface temperature input to CRTM is itself biased (from GDAS). Bias is too large to discern improvements in emissivity. Will need more accurate surface temperatures to run CRTM for emissivity comparisons. Control (CRTM emis) Test (NRL emis) CRTM minus AVHRR BT Jun 25, Scale: -20K - +20K
LST Difference: GDAS minus GOES Monthly Mean 18Z LST [K] July 2007o From J. Meng, EMC LST [K] Verification at SURFRAD Sites July 2007 GDAS surface temp is biased over dry land areas in daytime.
Sensitivity of CRTM to emissivity variation Broadleaf Evergreen Tree Broadleaf Shrub with Bare Soil Emissivity varied accord. to Snyder (1998) BRDF IR model min/max emissivity Emissivity input to CRTM CRTM sim of AVHRR Bright Temp Single pixel summer Result: For Bands 4 & 5, emissivity variability of 0.02 gives CRTM BT variability of 0.5 K for green surfaces and 1.5 K for bare surfaces.
Current Steps Created module to run test/control emissivities in CRTM Creating new reflectance spectra for existing GFS surface-type classes –To be implemented in CRTM to avoid surface-type matchups in GSI –Switch to new NCEP surface-type classes later Creating in-situ surface temperature testbed from Climate Reference Network (CRN) stations (and SurfRad stations?) –Utilize in-situ surface temperatures for evaluating new emissivity methods in CRTM vs Sat Obs –Assumes in-situ surface temperature is accurate SurfRad Locations
Future Steps Improved new LST is available from EMC/Land Modeling group –Study impact on CRTM vs AVHRR bias, over several regions In-situ surface temperatures –Evaluate new reflectance spectra: CRTM current classes with current reflectance spectra (control) GFS current classes with new reflectance spectra (test) NCEP new classes with new reflectance spectra (test) –Repeat evaluation of NRL emissivity Extend evaluations to include HIRS Work with Ben Ruston to tune emissivity index with 1DVar approach Directional effects on emissivity using IASI and AIRS (view-angle dependence of IR emissivity)
Discussion What is the priority order of testing new emissivities with CRTM? New reflectance spectra (GFS classes, new NCEP classes) NRL emissivities (repeat analysis using in-situ LST) PC-based MODIS high spectral Satellite-retrieved emissivity (IASI, AIRS) –A lot depends on overall direction of improvements to CRTM/GSI: New reflectance spectra is a small change in CRTM & GSI Other methods (NRL, PC-MODIS) require much higher effort to change CRTM & GSI Data assimilation systems could use satellite-retrieved emissivity, rather than calculating emissivity in CRTM? Which other evaluations are important? Impact of EMC/Land improved LST on CRTM bias Directional effects on IR emissivity using IASI and AIRS 1DVar tuning of NRL emissivity index Add HIRS as additional satellite for emissivity comparisons Surface types to evaluate Land first, then sea-ice, snow?
NPOESS reflectance LUT (current CRTM IR emissivity method) Static reflectance value for each of 24 surface types Spectral range: 0.2 – 15.0 μm Spectral resolution: μm in range 0.2 – 1.0 μm 0.05 μm in range 1.0 – 2.0 μm 0.5 μm in range 2.0 – 10.0 μm 1.0 μm in range 10.0 – 15.0 μm Spatial resolution: surface type map available at 0.17 deg resolution, but CRTM does not implement this. Temporal resolution: none, LUT is temporally static User input: –surface type (or lat/lon if implemented) –Wavelength Obtained from IPO, derivation unknown
Issues with NPOESS reflectance spectra NPOESS JPL Spec Lib NPOESS has lower spectral resolution than JPL NPOESS reflectance values too high
Implementation of CRTM’s NPOESS Surface Types in GDAS Gridpoint Statistical Interpolation (GSI) GFS Vegetation TypeMatched NPOESS Surface Type 0.Water 1.Broadleaf-Evergreen TreesBroadleaf Forest 2.Broadleaf-Deciduous TreesBroadleaf Forest 3.Broadleaf and Needleleaf TreesBroadleaf and Pine Forest 4.Needleleaf-Evergreen TreesPine Forest 5.Needleleaf-Deciduous TreesPine Forest 6.Broadleaf Trees with GroundcoverBroadleaf Brush 7.GroundcoverScrub 8.Broadleaf Shrubs with Perennial GroundcoverScrub 9.Broadleaf Shrubs with Bare SoilScrub-Soil 10.Dwarf Trees and Shrubs with GroundcoverTundra 11.Bare SoilCompacted Soil 12.CultivationsTilled Soil 13.Glacial IceCompacted Soil
NRL Indexed Emissivity (Ruston): Algorithm Summary JPL reflectance spectra are weighted according to surface type Example: Open Shrubland = 1/8 Rconifer + 1/8 Rdeciduous + ¾ Rgrass Surface types are combined for each grid cell based on % veg type and % soil type –% vegetation type per grid cell determined by spatially binning 1km surface type map to 0.5 deg grid Rveg = ∑ VegType% for all Veg Types that comprise the grid cell Example: grid cell contains 50% Open Shrubland and 50% Wooded Grassland Rveg = 0.5 (1/8 Rconifer + 1/8 Rdeciduous + ¾ Rgrass) (1/4 Rconifer + 1/4 Rdeciduous + ½ Rgrass) –% soil type per grid cell determined by spatially binning soil type map to 0.5 deg grid Rsoil = ∑ SoilType% for all Soil Types that comprise the grid cell Rveg further weighted with Vegetation Fraction (VF) and Greenness Fraction (GF) depending on vegetation and greenness conditions Example for greenness > 1%: Rveg = GF * Rveg + ( 1 – GF )[ VF * Rveg + ( 1 – VF ) * Rdry grass ] Rsoil+veg = VF * Rveg + ( 1 – VF ) * Rsoil Rland = snowindex * Rsnow + ( 1 – snowindex ) * Rsoil+veg Rtotal = Rland | Rice | Rwater User input: –Lat/Lon –Vegetation Fraction, Greenness Fraction –Ice Fraction, Snow Depth –Wavelength or sensor’s Spectral Response Function
1.Broadleaf Evergreen Tree8. Broadleaf Shrub w Perennial Groundcover 2.Broadleaf Deciduous Tree9. Broadleaf Shrub w Bare Soil 3.Broadleaf & Needleleaf Tree10. Dwarf Tree & Shrub w Groundcover 4.Needleleaf Evergreen Tree11. Bare Soil 5.Needleleaf Deciduous Tree12. Cultivations 6.Broadleaf Tree w Groundcover13. Glacial - Ice 7.Groundcover only Weizhong Zheng, EMC GFS Vegetation Types
1.Evergreen Needleleaf Forest11. Permanent Wetland 2.Evergreen Broadleaf Forest12. Cropland 3.Deciduous Needleleaf Forest13. Urban and Built-Up 4.Deciduous Broadleaf Forest14. Crop/Veg Mosaic 5.Mixed Forest15. Snow and Ice 6.Closed Shrubland16. Barren 7.Open Shrubland17. Water 8.Woody Savanna18. Woody Tundra 9.Savanna19. Mixed Tundra 10.Grassland20. Bare Ground Tundra Proposed NCEP 20-category classification (based on MODIS)
NPOESS surface types used in CRTM for estimating IR surface emissivity 1.Water13. Tundra 2.Old Snow14. Grass – Soil 3.Fresh Snow15. Broadleaf – Pine Forest 4.Compacted Soil16. Grass – Scrub 5.Tilled Soil17. Soil – Grass – Scrub 6.Sand18. Urban commercial 7.Rock19. Pine-brush 8.Irrigated Low Vegetation20. Broadleaf-brush 9.Meadow Grass21. Wet Soil 10.Scrub22. Scrub – Soil 11.Broadleaf Forest23. Broadleaf 70 – Pine Pine Forest24. New Ice