Aerosol retrieval using polarimetric observations: THEORY and PRACTICE. What users should expect from polarization?

Transcription

1 Aerosol retrieval using polarimetric observations: THEORY and PRACTICE What users should expect from polarization? GRASP O. Dubovik1, P. Litvinov1, D. Tanré1, T. Lapyonok1, A. Fedorenko1, A. Lopatin1, F. Ducos1, M. Spetsberger2, A. Coman2, W. Planer2, C. Federspiel2 1- University of Lille 1, CNRS, France 2 - Catalysts GmbH, High Performance Computing, Linz, Austria Sensitivity of polarization to aerosol; Advantages and possible challenges; GRASP algorithm; PARASOL retrievals reality, potential ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

2 POLDER Available measurements of polarization: GLOBAL: every 2 days SPATIAL RESOLUTION: 5.3km 6.2km VIEWS: N Q = 16: (80 0 Q ) INTENSITY: N t l=6: (0.44, 0.49, 0.56, 0.67, 0.865, 1.02 mm) POLARIZATION: N P l=3: (0.49, 0.67, mm) AERONET New generation of Cimel multi-spectral polarimeters VIEWS: N Q = ~ 30: (2 0 Q ) INTENSITY + POLARIZATION: N l =8: (0.34, 0.38, 0.44, 0.50, 0.67, 0.87, 1.02, 1.64 mm) lidar Multi-spectral backscattering and depolarization profiles VIEWS: N Q = 1: (Q = ) INTENSITY backscattering profiles : N t l= ~ 3: (0.355, 0.532, mm) DEPOLARIZATION profiles : N P l=1: (0.532 mm)

3 SINDLE SCATTERING What polarization may add? OPTIMISTIC view P 11 - characterizes intensity, P 12, P 22, P 33, P 34, P 44 - characterize state of polarization add sensitivity to all aerosol properties (size, shape, absorption, refraction) MULTIPLE SCATTERING 1.Polarization of land surface is very weak, spectrally neutral and quite homogeneous; add sensitivity to aerosol over bright land surfaces 2. Molecular scattering strongly polarizes scatted light and it vertically very stable; add sensitivity to aerosol vertical distribution (in passive observations)

4 Accounting for polarization in radiation transmitted through the atmospheric Total: I Q;l = m 0 exp m 0 exp m 1 m 0 m 1 L 1 ; L 2 - rotation matrices 0 L 2 P Q;l L 1 I 0 mult. scat. I = I Q U V - Stokes vector F(Q;l) = F 11 F F 12 F F 33 F F 34 F 44 phase matrix!!! 1 0 I 0 = 0 0 I Q;l ~ 0 F 11 Q;l mult. scat. P Q;l ~ - F 12 Q;l /F 11 Q;l mult. scat. - Intensity -Linear Polarization

5 Phase Function Degree of Linear Plarization (-F 12 /F 11 ) Sensitivity of to particle size Log-normal monomodal dv(r)/dlnr : s v = 0.5, m = 0.44 mm, n = 1.4, k = SPHERES Rv = 0.1 Rv = 0.12 Rv = 0.14 Rv = 0.2 Rv = 0.4 Rv = 0.6 Rv = 1.0 Rv = 2.0 Rv = SPHERES Rv = 0.1 Rv = 0.12 Rv = 0.14 Rv = 0.2 Rv = 0.4 Rv = 0.6 Rv = 1.0 Rv = 2.0 Rv = Scattering Angle ( degrees) Scattering Angle ( degrees)

6 Spheres Sensitivity to real part of refractive index for fine mode aerosol Log-normal monomodal dv(r)/dlnr : s v = 0.5, m = 0.44 mm, k = 0.005

7 Spheres Sensitivity to to real part of refractive index for coarse mode aerosol Log-normal monomodal dv(r)/dlnr : s v = 0.5, m = 0.44 mm, k = 0.005

8 Real aerosol Modeled spherical: Randomly oriented spheroids : AERONET model

9 Degree of Linear Plarization (-F 12 /F 11 ) Degree of Linear Plarization (-F 12 /F 11 ) Sensitivity of Linear Polarization to particle size Log-normal monomodal dv(r)/dlnr : s v = 0.5, m = 0.44 mm, n = 1.4, k = SPHERES Rv = 0.1 Rv = 0.12 Rv = 0.14 Rv = 0.2 Rv = 0.4 Rv = 0.6 Rv = 1.0 Rv = 2.0 Rv = SPHEROIDS Rv = 0.1 Rv = 0.12 Rv = 0.14 Rv = 0.2 Rv = 0.4 Rv = 0.6 Rv = 1.0 Rv = 2.0 Rv = Scattering Angle ( degrees) Scattering Angle ( degrees)

10 Sensitivity of Linear Polarization of fine mode aerosol to real part of refractive index Log-normal monomodal dv(r)/dlnr : s v = 0.5, m = 0.44 mm, k = 0.005

11 Sensitivity of polarization to particle shape Coarse aerosol

12 What polarization may add? SINDLE SCATTERING P 12 - important for passive observation of polarization add sensitivity to - particle shape and - size distribution and real ref. index of small and large spherical particles COMPLICATIONS: particles can be inhomogeneous, and have very complex shape MULTIPLE SCATTERING 1.Polarization of land surface is very weak, spectrally neutral and quite homogeneous; add sensitivity to aerosol over bright land surfaces COMPLICATIONS: accurate modeling of surface polarization is complicated 2. Molecular scattering strongly polarizes scatted light and it vertically very stable; REALISTIC view add sensitivity to aerosol vertical distribution (in passive observations) COMPLICATIONS: accurate assumption of aerosol vertical properties is complicated

13 independent POLDER/PARASOL measurements : GLOBAL: every 2 days SPATIAL RESOLUTION: 5.3km 6.2km VIEWS: N Q = 16: (80 0 Q ) INTENSITY: N t l=6: (0.44, 0.49, 0.56, 0.67, 0.865, 1.02 mm) POLARIZATION: N P l=3: (0.49, 0.67, mm) SINGLE OBSERVATION: a lot!!! as much as AERONET N t l N P l N Q = (6+3) 16= 144

14 Present aerosol retrieval from PARASOL: Over Ocean (Herman et al., 2005): - Uses look-up tables - Fits both intensity and polarizations at 0.67 and 0.87 mm -Retrieves: AOT of fine and coarse mode, size information, nonsphericity, some height information. -Issues: does not always provide consistency with other channels Over Land (Deuzé et al., 2001): - Uses look-up tables - Fits only polarizations at 0.67 and 0.87 mm using look-up tables - Retrieves: AOT of fine mode only, some size information - Issues: quite limited

15 QUESTION: Why the PARASOL ( ) polarimeter data did not result in а great aerosol product until now? ALGORITM is not adequate???

16

17 GRASP objectives: Accurate, Versetile and Fast Inversion scheme: - search in continues space of solution; - optimization as Multi-term LSM; - simultaneous multi-pixel retrieval; GRASP Forward model: - applicable to diverse remote sensing observations; - accurate modeling using direct on-line computations; Software implementation: - advanced highly parralized CPU and GPU programinng; - public, open source aerosol retrieval code;

18 Smoke Desert Dust AERONET retrievals are driven by 31 variables : PARASOL: + surface BRDF+ aerosol height Particle Size Distribution: 0.05 mm R (22 bins) 15 mm dv/lnr - size distribution (22 values); n(l) and k(l) - ref. index (4 +4 values) C spher (%) - spherical fraction (1 value) Complex Refractive Index at l = 0.44; 0.67; 0.87; 1.02 mm Maritime

19 Forward Model Vector of retrieved parameters : a aer - aerosol properties GRASP Vertical distribution shape a aer a surf - surface properties a surf Aerosol single scattering Surface reflectance (l h), 0 (l h), P(l,Q h) BRDF BPDF Multiple scattering effects All calculations are on line!!! Radiative Transfer: F(l,Q ) Simulated observations:

20 Statistically Optimized Minimization Multi-Term LSM Dubovik and King 2000, Dubovik 2004, Dubovik et a Measurements: A priori restrictions consistency Indicator l, i f i f i x i weighting f i a f i x (N total - N x ) ˆ i i Lagrange multipliers 0 2 Radiances (ground/plane/space); - Pol. radiances (ground/plane/space); - (l) (e.g. AERONET); - b (l h lidar backscattering; Restrictions on derivatives of: - size distr. Variability; AERONET - n spectral variability; - k spectral variability; -their covariances (should depend on l and Q) -lognormal error distributions - BRDF and BPDF spectral variability; - C(h) vertical variability; Lidar Surface

21 Time-Variability Constraints The concept of multi-pixel retrieval POLDER/PARASOL ( t 3 ; x ; y ) t ( t 2 ; x ; y ) z X ( t 1 ; x ; y ) x X X-Variability Constraints

22 GRASP: Generalized Retrieval of Aerosol and Surface Properties AERONET POLDER lidar AERONET laboratory MISR??? GRASP single scattering Surface BRDF, BPDF shape

23 Tests over dark surface («Blind» Test) Kokhanovsky, et al, The intercomparison of major satellite aerosol retrieval., Atmos. Meas. Tech., 3, , ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

24 retrieved AOT retrieved AOT POLDER: LOA-2(Dubovik) algorithm (BRDF) nm 443nm 490nm 565nm 675nm 870nm 1020nm LOA nm 443nm 490nm 565nm 675nm 870nm 1020nm LOA reference AOT reference AOT Case 1 Case 2

25 retrieved AOT MSPI (JPL): I+Q Lambertian unknown surface nm 558nm 672nm 866nm MISR Case reference AOT

26 Conclusion from the numerical retrieval tests: Positive: GRASP provides rather accurate aerosol retrievals; Negative: GRASP is essentially slower than traditional aerosol algorithms; (originally ~ 10 sec per pixel) ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

27 GRASP is too slow? Acceleration using GPU Overall Timing Trends Now: ~ 0.1 sec per pixel, it is not over CATALYSTS GMBH, LABORATOIRE D'OPTIQUE ATMOSPHÉRIQUE

28 GRASP Open Source Code distributed via web 28

29 Aerosol Retrieval LEGO? Game principles: - basic blocs are solid, easy to use, replaceable, can be connected in different combinations; - final construction is reliable. Benefits: - variety of applications; - different designs are possible; - testing of alternative basic blocks ; - testing various connections Results are better!!! Progress is faster!!!

30 GRASP Nephelometer Sun-photometer Diverse applications of GRASP Lidar AERONET PARASOL INVERSION INVERSION INVERSION Instrument design Instrument design Instrument design Aerosol (l h), 0 (l h),p(l,q h All modules are fully consistent!!! vertical profiles vertical profiles radiative transfer calculations

31 GRASP applications: GRASP Real Data: - PARASOL over land and ocean (CNES, ESA); - polar polarimeter - MERIS over land (ESA); - polar radiometer - AERONET+ lidar ( FP-7 Europe) - ground-based - airborne nephelometer (UMBC) - airborne Numerical test with perfect observations: - 3MI / EPS-SG ; - polar polarimeter - Sentinel 4 (GOCI real data ESA); - geostationary radiometer - MERIS, AATSR (ESA); - polar radiometer - MISR; - polar radiometer - GICOM project (JAXA); - polar bi-viewing polarimeter - CHINA GLORY + 3MI project - polar polarimeter

32 Comparison of PARASOL with AERONET Beiging Kanpur Banizoumbou Mongu In cooperation with ESA - CCI Banizoumbou: January, February, 2008 Surface: Grassland. Aerosol: Coarse mode is dominated. Mongu: August, September, 2008 Surface: Savanna. Aerosol: Fine mode is dominated. Beijing: April, December, 2008 Surface: Urban. Aerosol: Mixture of Fine and Coarse Kanpur: October-December, 2008 Surface: Urban. Aerosol: Mixture of Fine and Coarse The IGBP (International Geosphere Biosphere Program) land type specification was used

33 Regional maps (1800 x 1800 km). Banizoumbou, AOD 670 nm Strong spatial and temporal variation of AOD

34 Regional maps (1800 x 1800 km). Banizoumbou, SSA 670 nm Essential temporal variation of SSA

35 Daily variation of AOD and SSA at 670 nm. Banizoumbou (Jan., Febr. 2008) Frascati March 12, 2013

36 AOD. POLDER/AERONET

37 Angstrom Exponent. POLDER/AERONET Frascati March 12, 2013

38 SSA. POLDER/AERONET

39 Importance of Polarization? Synthetic measurements simulation We simulated 2 months of PARASOL measurements Aerosol and surface properties and aerosol concentration are taken from the retrieval of PARASOL measurements over Banizoumbou in January, February ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

40 Inversion of synthetic measurements Intensity + Polarization (I, Q, U) ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

41 41

42 42

43 Inversion of synthetic measurements Intensity only (I) ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

44 44

45 45

46 Inversion of synthetic measurements Polarization only (Q, U) ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

47 47

48 48

49 Inversion of REAL PARASOL measurements Different Scenarios: I+Q+U; I; Q+U ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

50 I,Q,U retrieval vs I-retrieval and Q,U-retrieval: AOD (Mongu. August, September 2008)

51 I,Q,U retrieval vs I-retrieval and Q,U-retrieval: SSA (Banizoumbou)

52 I,Q,U retrieval vs I-retrieval and Q,U-retrieval

53 I,Q,U retrieval vs I-retrieval and Q,U-retrieval: total reflectance fits ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

54 Conclusion from the tests: Polarization is a great addition to multi-angular intensity observations ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan

55 Differences in Radiances (%) Differences in Radiances (%) PARASOL observations under correction by CNES now Spectral changes in radiances Banizoumbou Mongu Changes in Radiances Banizoumbou, February, 2008 Changes in Radiances Banizoumbou, February, Scattering Angle 175 Scattering Angle Scattering Angle 115 Scattering Angle Wav elenght (mm) Wav elenght (mm)

56 Radiances (I( Q)/I 0 ) Radiances (I( Q)/I 0 ) Radiances (I( Q)/I 0 ) Radiances (I( Q)/I 0 ) Radiances (I( Q)/I 0 ) Radiances (I( Q)/I 0 ) Changes in radiances PARASOL Spectral Radiances Banizoumbou, February, 2008 PARASOL Spectral Radiances Banizoumbou, February, 2008 PARASOL Spectral Radiances Banizoumbou, February, ORIGINAL 440 NEW ORIGINAL 670 NEW ORIGINAL 1020 NEW Scattering angle (degree) PARASOL Spectral Radiances Banizoumbou, February, Scattering angle (degree) Fit of new radiances PARASOL Spectral Radiances Banizoumbou, February, Scattering angle (degree) PARASOL Spectral Radiances Banizoumbou, February, FIT 440 MES NEW FIT 670 MES NEW FIT 1020 MES NEW Scattering angle (degree) Scattering angle (degree) Scattering angle (degree)

57 0.44), PARASOL 0.44), PARASOL Changes in retrieved AOD (440) - DUST Banizoumbou (original) Banizoumbou ( new ) PARASOL v ersus AERONET (original) Banizoumbou, January - February, 2008 PARASOL v ersus AERONET (new ) Banizoumbou, January - February, y = x R= y = x R= ), AERONET 0.44), AERONET

58 0.44), PARASOL 0.44), PARASOL Changes in retrieved AOD (440) (high loading) Mongu ( original ) Mongu ( new ) PARASOL versus AERONET (original) Mongu, August - September, 2008 PARASOL versus AERONET (original) Mongu, August - September, y = x R= y = x R= ), AERONET 0.44), AERONET

59 0 (0.44), PARASOL 0 (0.44), PARASOL Changes in retrieved SSA (440) BIOMAS BURNING (high loading) Mongu (original) Mongu ( new ) PARASOL v ersus AERONET (original) Mongu, August - September, 2008 (0.44) > PARASOL v ersus AERONET (original) Mongu, August - September, 2008 (0.44) > y = x R= y = x R= (0.44), AERONET (0.44), AERONET

60 PARASOL (GRASP algorithm) 0 (l), PARASOL Changes in retrieved SSA (4wl) (high loading) 4 sites ( original ) Mongu + Banizoumbou ( new ) SSA (Ross-Li+Maignan model) PARASOL v ersus AERONET (new ) M ongu, August - September, 2008 Banizoumbou, January - February, 2008 (0.44) > Mongu Banizoumbou nm 670 nm 865 nm 1020 nm AERONET y = x R= (l), AERONET 0 y = x R= y = x R= y = x R= y = x R=

61 GRASP retrieval. Regional maps (1800 x 1800 km). Europe, SALB 670 and 865 nm Big contrast between 670 and 865 nm because of vegetation!!!

62 GRASP retrieval. Regional maps (1800 x 1800 km). Europe, AOD and SSA 670 nm Strong spatial variation of AOD and SSA

63 GRASP retrieval. Regional maps (1800 x 1800 km). Beijing, SSA 670 nm Strong spatial and temporal variation of SSA

64 GRASP retrieval. Regional maps (1800 x 1800 km). Mongu, SSA 670 nm NASA Global Fire Maps (detected by MODIS) Small SSA correspond to biomass burning!

65 Dubovik et al., AMT, 2011 GRASP Status: Software is our passion GRASP Core Algorithm is developed and performs well: - uses very elaborated aerosol and RT models; - inversion is based on rigorous statistical optimization; - performs well in numerical test (Dubovik et al. 2011, Kokhanovsky et al. 2010); - has a lot of flexibility - GRASP is under acceleration on GPU; - Cloud retrieval is to be integrated into GRASP; Promises: - Operational PARASOL aerosol product (SSA, etc.) in 2014; - Open Source GRASP code is to be available to a wide community; - Using GRASP for other sensors has some potential ICAP working group meeting, 5-8 November, 2013, Tsukuba, Japan