Bidirectional Reflectance Function in Coastal Waters And its Application to the Validation of the Ocean Color Satellites Alexander Gilerson 1 , Soe Hlaing 1 , Tristan Harmel 1 , Alberto Tonizzo 1 , Robert Arnone 2 , Alan Weidemann 2 , Samir Ahmed 1 1 Optical Remote Sensing Laboratory, City College, New York 2 Naval Research Laboratory, Stennis Space Center
Bidirectional Reflectance Distribution Function ( BRDF )
Radiance emerging from the sea, in general, is not isotropic .
Varies directionally depending on viewing and illumination conditions.
Bi-directionality property depends on Inherent Optical Properties ( IOP ) of the water constituents which are highly variable, especially in coastal environment
This bidirectional effect needs to be corrected to get standardized parameters suitable for :
Oceanic and Coastal waters monitoring
Calibration-validation of ocean-color satellite data
Water Body Above water radiometer
Correction for Bidirectional Reflectance Distribution
Adjust the remote-sensing reflectance for
Hypothetical configuration of :
Sun at zenith
Current standard BRDF correction algorithm [ Morel & Gentili 2002 et. al ] :
Optimized for the open ocean water conditions.
Correction is based on the prior estimation of chlorophyll concentration.
But, inappropriate for typical coastal waters usually dominated by sediment or by colored dissolved organic matters (CDOM)
To analyze Case 2 BRDF, a dataset of remote sensing reflectances typical for coastal (Case 2) water conditions was generated through radiative transfer simulations for a large range of viewing and illumination geometries .
Based on this simulated dataset, a Case 2 water-focused remote sensing reflectance model is proposed to correct above-water and satellite water leaving radiance data for bidirectional effects.
Proposed model is validated with a one year time series of in situ above-water measurements acquired by collocated multi- and hyperspectral radiometers which have different viewing geometries.
With the use of proposed BRDF correction, match-up comparisons of in situ time series and the MODIS satellite data has been improved.
Theoretical Background Fundamental equation which relates Rrs to optical properties [Morel 2002 et. al] : merges reflection and refraction effects that occur when downward irradiance and upward radiance propagate through the air-water interface f relates the magnitude of the irradiance reflectance just below the surface to IOP Angular Coordinate Convention θ v ~ Viewing angle θ s ~ solar Zenith φ ~ solar-sensor relative azimuth BRDF correction: Set f and Q for Sun at zenith and nadir view Rrs ( W,IOP ) _corrected Q= bidirectional function W = wind speed ω = single back-scattering albedo ω = b b / ( a + b b ) determined by IOP
Bio-optical model and radiative transfer simulation 1053 sets of Viewing & illumination geometries Viewing angle ( θ v ) 0 o ~ 80 o solar Zenith ( θ s ) 0 o ~ 80 0 relative azimuth ( φ ) 0 o ~ 180 o Wavelength: 412,443, 491, 551, 668 nm Inherent Optical Properties (IOP) Range of input parameters [Chl] = 1 to 10mg/m 3 C NAP = 0.01 to 2.5mg/m 3 a CDOM = 0 to 2m -1 ω = b b / ( a + b b ) can be directly connected to Rrs through modeling 500 sets of IOP Obtain Rrs ( λ ) & equivalent ω ( λ ) from 500 sets of IOPs to investigate Rrs – ω relatioships for large sets of viewing and illumination geometries. Generated as random variables in the prescribe ranges typical for coastal water conditions Particle Scattering Phase Function Varied with particle Concentration & Composition Radiative transfer simulations (Hydrolight) Remote-sensing Reflectance Rrs ( λ )
Rrs ( λ ) vs Single back-scattering albedo ( ω ) at various illumination and viewing geometries
Rrs~f( ω ) relationship also depends on the viewing and illumination geometries.
Spectral dependency of the ω ~ Rrs relationship can be also observed [ Gilerson 2007 et.al ].
Rrs can be fitted to ω with a third order polynomial:
Rrs ~ function( ω ) with [ Gordon 1988, Lee 2002 & Park 2005 ]. coefficients are generated for each set of viewing / illumination geometries as well as for each wavelength. These coefficients are applicable to typical coastal water conditions.
CCNY-BRDF correction algorithm Optimized for typical Case-2 water conditions ω – single backscattering albedo θ s – Solar zenith angle θ v – Viewing zenith angle φ – Solar-sensor relative azimuth λ – Wavelengths
CCNY algorithm in 2 steps:
(1) From the measured Rrs ( θ s , θ v , φ , λ): S olve and retrieve ω (λ) with the use of the least mean square fitting & tabulated α i ( θ s , θ v , φ , λ) coefficients .
(2) Use the retrieved ω (λ ) in the equation with α i ( θ s =0, θ v =0, φ =0, λ ) coefficients for nadir viewing and illumination t o calculate the BRDF-corrected Rrs ( θ s =0, θ v =0, φ =0, λ )
Tabulated coefficients based on radiative transfer computation Use of third order polynomial parameterization based on radiative transfer computation for large range of optical properties generalized expression
Statistical Analysis/Comparison of the standard MG (Morel/Gentili) and proposed CCNY Algorithms Based on Simulated Dataset (1/2)
The standard use of Case 1-water based BRDF MG correction induce almost 10% uncertainty in the remote sensing reflectance retrieval in typical coastal waters.
The proposed algorithm permit to reduce this dispersion below 1% without adding any bias
Standard Algorithm CCNY Algorithm y = 0.93* x – 8.4e -5 (Standard) y = 1.00* x – 8.5e -6 (CCNY) Regression lines AAPD(Standard Algo)=9.5% AAPD(CCNY Algo)=0.6% Dispersion
Statistical Analysis of the Algorithms Based on Simulated Dataset (2/2)
Up to 26% in bi-directional variation is observed addressing the need for a BRDF correction.
Standard MG algorithm : helps, but 57% of the dataset still have relative percent difference more than 5% which is the required accuracy for Ocean Color Sensor
CCNY algorithm: ~98% of the cases have relative percent difference less than 5%
Important need to incorporate Case-2 water based BRDF correction in the current data processing
Possible suitability of CCNY-algorithm to fulfill the Ocean Color Radiometry requirements
CCNY algorithm Standard algorithm Without correction in %
ASSESSMENT OF BRDF-CORRECTION APPLICATION TO ABOVE-WATER DATA AT LONG ISLAND SOUND COASTAL OBSERVATORY
Identical measuring systems and protocols, calibrated using a single reference source and method, and processed with the same code;
Standardized products of exact normalized water-leaving radiance and aerosol optical thickness
LISCO Multispectral SeaPRISM system as part of AERONET – Ocean Color network [Zibordi et al., 2006] LISCO Site Characteristics LISCO
Water type: Moderately turbid and very productive (Aurin et al. 2010) Bathymetry : plateau at 13 m depth Location and Bathymetry LISCO Site Characteristics Depth in meters (GEBCO data)
LISCO Tower LISCO site Characteristics Platform : Collocated multispectral SeaPRISM and hyperspectral HyperSAS instrumentations since October 2009 12 meters Retractable Instrument Tower Instrument Panel
Direct Sun Radiance and Sky Radiance
Bands: 413, 443, 490, 551, 668, 870 and 1018 nm
Linear Polarization measurements
Hyperspectral: 180 wavelengths [305,900] nm
HyperSAS Instrument Data acquisition every 30 minutes for high time resolution time series LISCO Instrumentation
Instrument Panel Unique Capability of Making Near-Concurrent Water-Leaving Radiance From Different Viewing Geometries
Both instrument makes measurements with viewing angle, θ v = 40 o .
Thanks to the rotation feature of SeaPRISM , its relative azimuth angle, φ , is always set 90 o with respect to the sun (resulting in water scattering angle range of 132 ~ 145 o ).
HyperSAS instrument is fixed pointing westward position all the time, thus φ is changing throughout the day and resulting scattering angle range from 110 -175 o .
LISCO site instrumentations configuration permits to assess accuracy of the bi-directionality correction of the water leaving radiances.
Features of the LISCO site SeaPRISM HyperSAS N W
Above Water Signal decomposition Above-Water Data Processing Sun T otal radiance Sky radiance Water leaving radiance Sea surface reflectivity Sun glint radiance E d Rrs = L w / E d Down-welling Irradiance Remote-sensing reflectance: Needs to be corrected for the bidirectionality property L i L w θ θ L T = L w + ρ (W) L i + L g L i
Comparison of SeaPRISM and HyperSAS
Increased dispersion in the right figure is mainly due to BRDF
(filters exclude data from some geometries, specifically where relative azimuth angle, φ < 60° to eliminate glint effects)
For all the viewing geometries Both instrument pointing same direction (within ±10° in Azimuth) Rrs SeaPRISM [sr -1 ] Rrs SeaPRISM [sr -1 ] Rrs HyperSAS [sr -1 ] Rrs HyperSAS [sr -1 ]
Comparison between the Standard MG and Proposed CCNY Algorithm with the LISCO Dataset
Current MG algorithm does not reduce significantly the dispersion and induces a weaker correlation with R 2
The proposed CCNY algorithm reduce dispersion by 2% in absolute value and by more than 3% in relative values
Before BRDF Correction Corrected with MG Corrected with CCNY
APPLICATION TO OCEAN COLOR MODIS IMAGERY
Satellite Validation Satellite Pixel Selection for Matchup Comparison 3km×3km pixel box for matchup comparison Exclusion of pixel box if presence of cloud-contaminated pixels in this 9km×9km pixel box Validation of MODIS-Aqua against the LISCO Data Satellite Data Processing: Standard NASA Ocean Color Reprocessing 2009 Also exclusion of any pixel flagged by the NASA data quality check processing (Atmospheric correction failure, sun glint contamination,…)
Rrs Time series for the match-up comparison Comparison between LISCO and MODIS Ocean Color data Qualitative consistency in variations is observed between the in-situ and satellite data. How will the Satellite / in situ data comparison be improved by application of the CCNY BRDF-correction ?
Application to the Satellite Data
Application of the CCNY algorithm induces stronger correlation (0.926)
Spectral average absolute percent difference is improved by more than 3%.
Suitability of CCNY BRDF-correction to significantly improve OCR satellite data accuracy in coastal areas
Corrected with Standard Algo Corrected with CCNY AAPD (%) Wavelength (nm) 412 443 491 551 667 Standard 46.43 38.85 16.68 13.61 24.54 CCNY 42.40 34.16 14.93 10.99 21.89 Improvement 4.03 4.69 1.75 2.62 2.65
We proposed a new algorithm for BRDF correction of the remote-sensing reflectance based on extensive radiative transfer calculations for typical coastal ( case-2 ) waters conditions
Theoretical analysis showed that significant improvement are observed with the proposed algorithm reducing the uncertainty of this correction below 1%
This algorithm has been tested over the two years time series of LISCO observations.
It has been shown that the CCNY BRDF-correction algorithm improve the accuracy of the above-water data by more than 3%
Application of CCNY-algorithm to MODIS satellite data showed the same order of improvement. Suitability of CCNY BRDF-correction to significantly improve OCR satellite data accuracy in coastal areas
As a consequence of this work the operational application of this algorithm to current and future (VIIRS) OCR satellite is planned
ACKNOWLEDGMENTS NASA AERONET team for SeaPRISM calibration, data processing and support of the site operations NASA Ocean Color Processing Group for satellite imagery Partial support from: Office of Naval Research (ONR) National Oceanographic and Atmospheric Administration (NOAA)