1. AERONET Inversions: Progress and Perspectives Oleg Dubovik (NASA / GSFC) Alexander Sinyuk (NASA / GSFC) Tatyana Lapyonok (NASA / GSFC) Brent Holben (NASA / GSFC) Tom Eck (NASA / GSFC) Alexander Smirnov (NASA / GSFC) Anne Vermeulen (NASA / GSFC) Teruyuki Nakajima (CCSR, Tokyo, Japan) Takashi Nakajima (CCSR, Tokyo, Japan) Charles Gatebe (NASA / GSFC) Michael King (NASA / GSFC) Francois-Marie Breon (CEA/DSM/LSCE, France) Michael Sorokin (NASA / GSFC) Ilya Slutsker (NASA / GSFC) AERONET/PHOTON… AERONET/PHOTON… (Word-Wide)

2. Observations Numerical inversion: -Accounting for noise -Solving Ill-posed problem - Setting a priori constraints Forward model: -Spectral and angular scattering by particles with different sizes, compositions and shapes - Accounting for multiple scattering in atmosphere aerosol particle sizes, refractive index, single scattering albedo single scattering albedo, etc. Retrieval scheme:  (Dubovik and King, JGR, 2000  (Dubovik and King, JGR, 2000)

3. Multiple Scattering Multiple scattering effects are accounted by solving scalar radiative transfer equation with assuming Lambertian ground reflectance (Nakajima – Tanaka code) Aerosol scattering Molecular scattering Gaseous absorption Surface reflection Multiple Scattering

4. Single Scattering by Single Particle Scattering and Absorption is modeled assuming aerosol particle as homogeneous sphere with spectrally dependent complex refractive index ( m( )= n( ) - i k( )) - “Mie particles” m( ) Radius P(  )- P(  )- Phase Function;    -single scattering albedo  ( ) - extinction optical thickness;  ( )     absorption  optical thickness I 0 ( ) I scat (  ) I trans ( ) Single Scattering

5. AERONET model of aerosol spherical: Randomly oriented spheroids : (Mishchenko et al., 1997) AERONET model of aerosol

6. Statistically optimized fitting:  (Dubovik and King, 2000) Measurements: i=1 - optical thickness i=2 - sky radiances -their covariances (should depend on and  ) -lognormal error distributions a priori restrictions on norms of derivatives of: i=3 -size distr. variability; i=4 -n spectral variability; i=5 -k spectral variability; Lagrange parameters consistency Indicator weighting Inversion

7. Fitting as a retrieval strategy

8. The averaged optical properties of various aerosol types (Dubovik et al., 2002, JAS) + _

9. AERONET inversion developments Forward model: - accounting for particle shape - using non-lambertian surface - modeling polarization Retrieval flexibility: - additional spectral channels - different geometries Inversion of combined data: - different geometries - combining with satellite - combining with aircraft Output improvements: - detailed phase function - degree of polarization - flexible separation of modes - fluxes and forcing - details of fitting (biases and random) Errors estimation: -for individual retrieval -for absorption optical thickness -for phase functions, etc. Perspectives: - assuming bi-component aerosols - combining with polarimetric satellite observations - retrieval of shape distribution

10. Almucantar:  ( ), I(  ) = 0.38, 0.44, 0.5, 0.67, 0.87, 1.02, 1.64,  m AERONET inversion scenarios Principal Plane:  ( ), I(  ) = 0.38, 0.44, 0.5, 0.67, 0.87, 1.02, 1.64,  m Polarized Principal Plane:  ( ), I(  ),P(  ) = 0.87  m spheres spheroids Inversion Products: dV/dln(r i ) n(  k(  BRDFerrors satellite, aircraft, etc.   (  P 11 (  P 12 (  fine & coarse, … fluxes, …

11. Utilizing additional spectral channels Potential enhancement of information Increased calibration efforts Desert Dust (Dhabi, UAI) dV/dln(r i ) n(  (( Almucantar:  ( ), I(  ) = 0.44, 0.67,0.87, 1.02, 1.64,  m

12. Utilizing additional spectral channels Potential enhancement of information Increased calibration efforts GSFC aerosol dV/dln(r i ) n(  (( Almucantar:  ( ), I(  ) = 0.38, 0.44, 0.67, 0.87, 1.02,  m

13. GSFC aerosol Dhabi dust + 0.38  m + 0.5, 1.64  m+ 1.64  m Fitting additional spectral channels + 0.02 - 0.02 ??? water vapor ?fine ?

14. Utilizing principal plane Enhanced range of scattering angles Sensitivity to vertical structure of aerosol Challenging cloud screening Desert Dust (Dhabi, UAI) dV/dln(r i ) n(  (( Principal Plane:  ( ), I(  ) = 0.44, 0.67, 0.87, 1.02, 1.64,  m

15. Utilizing principal plane Enhanced range of scattering angles Sensitivity to vertical structure of aerosol Challenging cloud screening dV/dln(r i ) n(  (((( Principal Plane:  ( ), I(  ) = 0.44, 0.5, 0.67, 0.87,1.02, 1.64,  m GSFC aerosol

16.  ( ), I(  ),P(  ) Numerical inversion: (F 11 ; -F 12 /F 11 !!!) -Accounting for uncertainty (F 11 ; -F 12 /F 11 !!!) - Setting a priori constraints aerosol particle sizes, refractive index, single scattering albedo AERONET Polarized Inversion Forward Model: Single Scat: Multiple Scat: DEUZE JL, HERMAN M, SANTER R, JQSRT, 1989 Successive Orders of Scattering Code

17. Utilizing polarization Enhanced range of scattering angles Sensitivity to vertical structure of aerosol Challenging cloud screening Calibration verification dV/dln(r i ) n(  (((( Principal Plane:  ( ), I(  ) = 0.44, 0.5, 0.67, 0.87,1.02, 1.64,  m Polarization :  ( ), I(  ),P(  ) = 0.87  m Cape Verde aerosol

18. Fitting polarization Enhanced range of scattering angles Sensitivity to vertical structure of aerosol Challenging cloud screening Calibration verification Radiance Linear Polarizartion Principal Plane:  ( ), I(  ) =0.87  m Polarization :  ( ), I(  ),P(  ) = 0.87  m Cape Verde aerosol

19. Fine and Coarse modes separations Radiance Beijing aerosol Flexible separation between fine and coarse modes (curently: ~0.6  m) 0.45  m

20. Retrieval using combinations of up-looking Ground-based and down-looking satellite observations Retrieved: Aerosol Properties: - size distribution - real ref. ind. - imag. ref. ind (AERONET sky channels) Surface Parameters: -BRDF (MISR channels) -Albedo (MODIS IR channels)

21. AERONET / POLDER-2 retrieval POLDER-2 fit Size distribution BRDF Biomass burning Mongu, Zambia, June, 2003

22. AERONET Ground-based Sun-sky radiometer:  ± 0.02 at 6 channels: 0.34, 0.38, 0.44, 0.67, 0.87, 1.02, 1.65  m   ± 0.05% at 4 channels: 0.38, 0.44, 0.67, 0.87, 1.02, 1.65  m 3° ≤ scattering angles ≤ ~70° - P  ± 0.02% at 0.87 MISR Reflectance at 4 channels: 0.45, 0.55, 0.67, 0.87  m 9 viewing angles: ±70.5 o, ± 60 o, ± 45.6 o, ± 26.1 o, 0 o MODIS Reflectance at 7 channels: 0.47, 0.55, 0.66, 0.87,1.2, 1.6, 2.1 AERONET/ MISR/ MODIS retrieval

23. Retrieval using combinations of up- and down-looking observations Retrieved: Aerosol above plane: - size distribution - real ref. ind. - imag. ref. ind Aerosol below plane: - size distribution - real ref. ind. - imag. ref. ind Surface Parameters: - BRDF, albedo, etc.

24. CAR - Cloud Absorption Radiometer 8 spectral channels: 0.34, 0.38, 0.47, 0.68, 0.87, 1.03, 1.19, 1.27  m Measures radiation transmitted* and reflected: 0° ≤ Obs. Zenith ≤ 180° 0° ≤ Obs. Azimuth ≤ 360° * Stray light problems for scattering angles ≤ 10 ° Flown by CV-580 aircraft at ~ 700 m above ground Univ. of Washington CV-580

25. Optical thickness  on September 6, 2000 AERONET daily variations AATS-14 AERONET AATS-14 versus AERONET

26. Aerosol retrieved from combined CAR - AERONET - AATS-14 obs.

27. Comparison of model retrieved BRDF with corrected direct BRDF Gatebe et al. 2003 BRDF constrains model: - positive and smooth; - PP symmetrical

28. Comparison of retrieved surface reflectance with other observations Mongu, September 6, 2000 Lambertian Lambertian approximation Mongu, Zambia

29. New Inversion Options: Almucatars (any numbers of spectral channels) - Almucatars (any numbers of spectral channels) - Principal planes (any numbers of spectral channels) - Polarized Principle Planes (0.87 mm) - Combined Principle Planes: Polarized (0.87 mm) + Regular (0.44 - 1.02mm) - Other Combined Cimel Data: -Almucantar + Principle Plane (?) - Several Almucantars + Principle Planes (??) - AERONET + satellite data (MODIS, MISR, POLDER …) - AERONET + aircraft (CAR) + …satellite - Spherical & Nonspherical model (for all retrievals) Perspectives: - assuming bi-component aerosols - combining with polarimetric satellite observations - retrieval of shape distribution

30. Sensitivity to instrumental offsets Offsets were considered in: - optical thickness: - sky-channel calibration: - azimuth angle pointing: - assumed ground reflectance: (bi - modal log-normal): Aerosol models considered (bi - modal log-normal): - Water-soluble aerosol for 0.05 ≤  (440) ≤ 1; - Desert dust for 0.5 ≤  (440) ≤ 1; - Biomass burning for 0.5 ≤  (440) ≤ 1; Results summary: -  (440) ≤ 0.2 - dV/dlnr (+), n( ) (-), k( ) (-),   ( ) (-) -  (440) > 0.2 - dV/dlnr (+), n( ) (+), k( ) (+),   ( ) (+) - Angular pointing accuracy is critical for dV/dlnr of dust (+) CAN BE retrieved (-) CAN NOT BE retrieved Offsets

31.  bias influence at  0  bias: Sky Radiance bias:

32. Random ERRORS in AERONET retrievals ASSUMPTIONS: - measurements have Normal Noise: - optical thickness:  0.01 - sky-radiances:  5% - sky-radiances:  5% CONCLUSIONS: - the retrievals stable - important tendencies outlined

33. Rigorous ERRORS estimates: General case : large number of unknowns and redundant measurements U - matrix of partial derivatives in the vicinity of solution Above is valid: - in linear approximation - for Normal Noise - no a priori constraints

34. ERRORS estimates with a priori constraints ISSUES: - in linear approximation - for Normal Noise - strongly dependent on a priori constraints - very challenging in most interesting cases

35. ERROR Factors: Important Factors: - Aerosol Loading - Scattering Angle Range - Number of Angles (homogeneity) - Aerosol Type etc.

36. Examples of error estimates high loading low loading

37. flaming combustion Rio Branco, Brazil smoldering combustion Quebec fires, July 2002 ABSORPTION of SMOKE