AEROCENTER Fall Seminar Series October 2 nd 2007

AEROCENTER Fall Seminar Series, October 2 nd, 2007 Radiative Transfer Codes for Atmospheric Correction and Aerosol Retrievals: Intercomparison Study Svetlana Y. Kotchenova & Eric F. Vermote The study is being performed in collaboration with: Robert Levy, Alexei Lyapustin, and Omar Torres

Project Description The project is devoted to the comparison and detailed evaluation of five atmospheric RT codes incorporated in different Monte Carlo (benchmark) Svetlana & Eric RT 3 (vector) Robert 6 SV 1. 1 (vector) satellite data processing only molecular atmosphere algorithms VPD (vector) Omar Svetlana & Eric SHARM (scalar) Alexei Coulson’s tabulated values (benchmark) MODTRAN (scalar) Svetlana only molecular atmosphere 2

Applications of the codes § 6 SV 1. 1 (Second Simulation of a Satellite Signal in the Solar Spectrum, Vector, version 1. 1 ): MODIS atmospheric correction and internal aerosol inversion § RT 3 (Radiative Transfer 3): MODIS coarse resolution (10 -km) aerosol retrieval § VPD (Vector Program D): TOMS (Total Ozone Mapping Spectrometer) aerosol inversion § SHARM (Spherical Harmonics): MAIAC (Multi-Angle Implementation of Atmospheric Correction for MODIS) § MODTRAN (Moderate Resolution Atmospheric Transmittance and Radiance Code): AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) atmospheric correction 3

Description of the codes: 6 SV 1. 1 Author: E. Vermote (University of Maryland, USA) Modified: E. Vermote et al. Language: Fortran 77, 95 Spectrum: 350 to 3750 nm Features: Ground surface: homogeneous & non-homogeneous with & without directional effect (10 BRDF + 1 user-defined models) Aerosol atmosphere: 6 code-embedded & 4 user-defined models & AERONET http: //6 s. ltdri. org Publications + Interface to create input files Molecular atmosphere: 6 codeembedded & 2 user-defined models Instruments: AATSR, ALI, ASTER, AVHRR, ETM, GLI, GOES, HRV, HYPBLUE, MAS, MERIS, METEO, MSS, TM, MODIS, POLDER, Sea. Wi. FS, VIIRS, & VGT – 19 in total 4

Description of the codes: RT 3 Author: F. Evan (Colorado State University) Language: Fortran 77 Input: Disadvantages: 1) pre-computed sets of output angles (interpolation might be needed) 2) no embedded MIE-code (combination with a MIE-code is needed to simulate aerosols) 5

Description of the codes: SHARM Author: T. Muldashev (Space Research Institute, Kazakhstan) Modified: A. Lyapustin Language: C/C++ Input: 1. atm. MIE config. par 1 L. sfc Advantages: very fast, simultaneous simulations for multiple geometries and wavelengths 6

Description of the codes: MODTRAN Author: Berk et al. (Air Force Research Laboratory) Language: Fortran 77 Modeling Features: molecular atmospheres (a lot of effort is put into gas absorption!), aerosols (with the help of DISORT at 16 Gaussian angles), clouds, surface Input: in the form of formatted “cards” (quite painful!) tape 5 – molecular atmosphere card 2 card 1 a Output: single geometry but for a range of wavelengths 7

Project History 2005 6 SV & VPD & RT 3 & SHARM Coulson’s tables discussions, calculations, Web site creation. . . 2006 Monte Carlo 6 SV & MODTRAN VPD 2007 & RT 3 & SHARM 2008 Why do you ignore MODTRAN? 8

Goals of the project § to evaluate the accuracy of each code based on the comparison with standard benchmark references such as Coulson’s tabulated values and a Monte Carlo approach § to illustrate differences between individual simulations of the code § to determine how the revealed differences influence on the accuracy of aerosol optical thickness and surface reflectance retrievals § to create reference (benchmark) data sets that can be used in future code comparison studies 9

Presentation of the results All results will be put on the Internet and summarized in a manuscript titled “Radiative Transfer Codes for Atmospheric Correction and Aerosol Retrievals: Intercomparison Study” which will be submitted to Applied Optics. http: //rtcodes. ltdri. org 10

Characterization of a RT code In regard to remote sensing applications. . . 1. Versatility 6 SV 1. 1, SHARM, MODTRAN, VPD & RT 3 (RT 3 needs to be combined with a MIE-code) 2. Accuracy to be determined 3. User-friendliness 6 SV 1. 1 & SHARM, RT 3, MODTRAN, VPD (VPD is not publicly available) 4. Speed to be determined 11

Code Accuracy The general atmospheric RT code accuracy requirement for pure simulation studies is 1%. Reference: Muldashev et al. , Spherical harmonics method in the problem of radiative transfer in the atmosphere-surface system, Journal of Quantitative Spectroscopy and Radiative Transfer, 61(3), 393 -404, 1999. Will violation of this requirement have a significant effect on the resulting satellite product? Step 1: comparison with benchmarks to see if there is violation Step 2: evaluation of the impact of violation 12

Benchmarks: Coulson’s tables Coulson’s tabulated values represent the complete solution of the Rayleigh problem for a molecular atmosphere. Reference: Coulson et al. , Tables related to radiation emerging from a planetary atmosphere with Rayleigh scattering (1960). 13

Benchmarks: Monte Carlo The code is written by F. M. Bréon (le Laboratoire des Sciences du Climat et de l'Environnement, France) based on the Stokes vector approach. Languages: Fortran, C. Limitations: large amounts of calculation time and angular space discretization. 14

Comparison Procedure 1. Molecular Atmosphere ( surf = 0. 0; 0. 25) direction of incident light 2. Aerosol Atmosphere ( surf = 0. 0) The same procedure was used in the previous comparison study: A. Lyapustin “Radiative transfer code SHARM-3 D for radiance simulations over a non-Lambertian nonhomogeneous surface: intercomparison study”, Applied Optics, 41(27), 5607 -5615. larger particles direction of incident light 3. Mixed Atmosphere ( surf = 0. 0; 0. 25) surf is the reflectance of a Lambertian surface 15

Molecular Atmosphere: Conditions All RT codes are compared to the Coulson’s tabulated values. mol ( , nm) 0. 1 (530) 0. 25 (440) 0. 5 (360) surf 0. 0 0. 25 θs, deg. 23. 0739 53. 1301 78. 4630 0. 0 36. 8699 66. 4218 θv, deg. as in Coulson’s tables φ, deg. 0. 0; 90. 0; 180 * mol is the molecular optical thickness is the wavelength surf is the surface reflectance θs is the sun zenith angle θv is the view zenith angle φ is the relative azimuth ** Monte Carlo is used only as an auxiliary means here. 16

Molecular Atmosphere: Results We calculate the absolute values of average relative differences: 17

Aerosol Atmosphere: Conditions 6 SV 1. 1 is compared to Monte Carlo and then the other codes are compared to 6 SV 1. 1. . . Model Biomass Burning Amazonian forest, Brazil (1993 -1994), Bolivia (1998 -1999) African savanna, Zambia (1995 -2000) 0. 1 ≤ (440) ≤ 1. 0 0. 1 ≤ (440) ≤ 3. 0 0. 1 ≤ (440) ≤ 1. 5 0. 2; 0. 8; 2. 0 0. 2; 0. 8 Real and imaginary parts of refractive index 1. 41 -0. 03 (440); 0. 003 1. 47; 0. 0093 1. 51; 0. 021 SSA at = 412/440/670 nm 0. 97/0. 98/0. 97 0. 94/0. 93 0. 88/0. 84 r. Vf , m 0. 12+0. 11 (440) 0. 14+0. 013 (440) 0. 12+0. 025 (440) f, m 0. 38 0. 40 r. Vc, m 3. 03+0. 49 (440) 3. 27+0. 58 (440) 3. 22+0. 71 (440) c, m 0. 75 0. 79 0. 73 CVf , m 3/ m 2 0. 15 (440) 0. 12 (440) CVc, m 3/ m 2 0. 01+0. 04 (440) 0. 05 (440) 0. 09 (440) Location Range of τaer Values of τaer selected for this study Urban-Industrial and Mixed GSFC, Greenbelt, MD (1993– 2000) 18

Aerosol Atmosphere: Results (compared to MC). . . 6 SV 1. 1 can be used as benchmark because it demonstrates good agreement with MC θs = {0. 0°, 23. 0°, 50. 0°} black soil 19

Aerosol Atmosphere: Results (compared to 6 SV) We calculate the absolute values of average relative differences: * maer is the selected aerosol model 20

Mixed Atmosphere: Conditions We simply added a molecular atmosphere to all considered aerosol models. Profiles: Mixture: 20 Km O 2, CO 2 Trace Gases Ozone, Stratospheric Aerosols 8 Km Molecules (Rayleigh Scattering) Molecular optical thickness: 2 -3 Km = 412 nm - mol = 0. 303 H 2 O, Tropospheric Aerosol = 440 nm - mol = 0. 232 = 670 nm - mol = 0. 042 Ground Surface 21

Mixed Atmosphere: Results (compared to MC) 6 SV 1. 1 demonstrates relatively good agreement with MC (within 0. 85%) Molecular + Urban-industrial aerosol aer = 0. 2, θs = 0. 0° θs = {0. 0°, 23. 0°, 50. 0°} mol = 0. 303 aer = 0. 8, θs = 0. 0° black soil 22

Mixed Atmosphere: Results (compared to 6 SV) Again, we calculate the absolute values of average relative differences: 23

Accuracy vs. Speed Time for 1 run (the case of a mixed atmosphere (λ = 440 nm, AF, aer = 0. 8) + surface): SHARM: ≈ 5. 6 s (7. 3 s for a number of angles 6 x 16 x 3) 6 SV 1. 1: ≈ 3 s (this time x number of SZA) Monte Carlo: ≈ 45 min (for one SZA) Time is important: code comparison like this one Time is not that important: calculation of LUTs Accuracy depends on many factors: SHARM: the number of harmonics 6 SV 1: the number of Legendre coefficients, calculation layers and angles 24

VPD for a molecular atmosphere Molecular atmosphere by VPD: Aerosol atmosphere by MODTRAN: Good results for a molecular atmosphere do not mean that the accuracy of aerosol simulations will be satisfactory! 25

Error on AOT Retrieval: Theory 1) is the TOA reflectance of a vector code, is the TOA reflectance of a scalar code, is the error of a scalar code, 2) From (1) and (2) we can calculate the AOT retrieval error: , where The accuracy of 6 SV retrievals ? 1% (compared to MC) Assumption: TOA reflectance is a linear function of AOT 26

Error on AOT retrieval: Results Molecular + Aerosol (African Savanna, aer = 0. 2): SHARM: aer = 0. 2 ± 0. 14 6 SV: aer = 0. 2 ± 0. 01 27

Error on AOT retrieval: Results (Cont. ) Molecular + Aerosol (African Savanna, aer = 0. 8): SHARM: aer = 0. 8 ± 0. 15 6 SV: aer = 0. 8 ± 0. 05 28

AOT Retrieval from MODIS data MODIS Land Surface Reflectance algorithm by Vermote et al. Multi-Angle Implementation of Atmospheric Correction for MODIS by Lyapustin & Wang 490 nm 470 nm 443 nm 412 nm 0. 2 0. 5 470 nm 670 nm 0. 4 AOT Ex. : AERONET site Alta Floresta, day 197 of 2003 Ex. : Part of Arabian Peninsula, day 207 of 2005 29

Error on SR Retrieval: Theory The same procedure as for AOT retrievals, but is replaced by surface reflectance L (L = 0. 05, d. L = 0. 01) 1) 2) The SR retrieval error: , where 30

Error on SR Retrieval: Results Molecular + Aerosol (Amazonian Forest, aer = 0. 2) + Surface (Lambertian, surf = 0. 05): SHARM: surf = 0. 05 ± 0. 01 6 SV: surf = 0. 05 ± 0. 002 31

Error on SR Retrieval: Results (Cont. ) Molecular + Aerosol (Amazonian Forest, aer = 0. 8) + Surface (Lambertian, surf = 0. 05): SHARM: surf = 0. 05 ± 0. 01 6 SV: surf = 0. 05 ± 0. 003 32

Vector ? Scalar for Remote Sensing Is it important to use a vector code? 1) AOT (+ other aerosol properties) retrievals: + 2) surface reflectance retrievals: important, from a theoretical point of view pre-assigned set of aerosol models: Smoke LABS Smoke HABS Urban POLU Urban CLEAN important The accuracy of LUTs directly depends on the RT code simulations. The best solution is to calculate a product error budget. 33

Reference data set The goal is to use 6 SV 1 to create a reference data set for further code comparison studies. 34

. . . Thank you for your attention! Questions? . . skotchen@ltdri. org 35