S. Ahmed1*, A. Gilerson1, T. Harmel2, S. Hlaing1, A. Tonizzo1, A. Weidemann3, R. Arnone3

1

Evaluation of atmospheric correction procedures for ocean color data processing using hyper- and

multi-spectral radiometric measurements from the Long Island Sound Coastal Observatory

1 Optical Remote Sensing Laboratory, City College, New York, NY 10031, United States2 Observatoire Océanologique de Villefranche sur Mer, France

3 Naval Research Laboratory, Stennis Space Center, MS 39529, United States

The Long Island Sound Coastal Observatory (LISCO) for Ocean Color data validation.

Hyper- and multi-spectral above water measurements, data processing and filtering procedures.

Representativeness of LISCO as OC data validation site.

Assessments of the atmospheric correction quality.

Impacts of the error in the atmospheric correction over water leaving radiance retrieval.

Conclusion

Contents

2

LISO

AERONET – Ocean Color: is a sub-network of the Aerosol Robotic Network (AERONET), relying on modified sun-photometers to support ocean color validation activities with highly consistent time-series of LWN() and a().

G.Zibordi et al. A Network for Standardized Ocean Color Validation Measurements. Eos Transactions, 87: 293, 297, 2006.

AERONET-Ocean Color

•Autonomous radiometers operated on fixed platforms in coastal regions;•Identical measuring systems and protocols, calibrated using a single reference source and method, and processed with the same code;•Standardized products of normalized water-leaving radiance and aerosol optical thickness.

Rationale:

Features of the LISCO site12

met

ers

Retractable Instrument Tower

Instrument Panel

LISCO Tower

Solar Panel

SeaPRISM instrument

HyperSAS InstrumentWater Leaving

RadianceSky Radiance and

Down Welling Irradiance

Hyper-Spectral 305 to 900 nm wavelength range.

Water Leaving Radiance

Direct Sun Radiance and Sky Radiance

Bands: 413, 443, 490, 551, 668, 870 and 1018 nm.

Co-located Multi- & Hyper-spectral instruments for spectral band matching with various current as well as future OC sensor.Data acquisition every 30 minutes for high time resolution time series

4 LISCO is the unique site in the world with collocated multi and

hyperspectral instrumentation for coastal waters monitoring

Instrument Panel

SeaPRISMHyperSAS

N

W

Both instrument makes measurements with viewing angle, θv = 40o.

Thanks to the rotation feature of SeaPRISM, its relative azimuth angle, φ, is always set 90o with respect to the sun.HyperSAS instrument is fixed pointing westward position all the time, thus φ is changing throughout the day.Both instruments point to the same direction when the sun is exactly at south. This instrument setup provides the ideal configuration to make assessments of the directional variation of the water leaving radiances. 5

Technical Differences between HyperSAS and SeaPRISMTwo Geometrical Configurations

Features of the LISCO site

Comparisons between HyperSAS and SeaPRISM data

6

Quantitative comparisons between the measurements made by two systems

For the comparison purpose, LT(λ), Li(λ) and Ed(λ) measurements of HyperSAS instruments are convolved with the spectral response function of SeaPRISM instrument in order to produce the multispectral data that is comparable to those of SeaPRISM.

dHf H(λ) - measurement made by HyperSAS instrument. ζ(λ) - spectral response function of SeaPRISM instrument. ν - specific center wavelengths of SeaPRISM. f (ν) - HyperSAS measurement quantity which is comparable to the measurements

of SeaPRISM at its center wavelengths.

Statistical estimators Absolute relative percent difference (ARPD). Unbiased relative percent different (URPD)

N

i i

ii

yxxy

NARPD

1

200

N

i ii

ii

yxxy

NURPD

1

200

ARPD provides the information regarding the dispersion. URPD can be used to assess the expected bias between the compared datasets.

7

Sky radiance measurements

Comparisons between HyperSAS and SeaPRISM average sky radiance, Li: (Left) relative azimuth angles for HyperSAS observations are restricted in

the 70o ≤ φ≤ 180o range; (Right) relative azimuth angles are restricted to 80o ≤ φ ≤ 100o range.

    Wavelength (nm)  Spectral

Average  412 443 491 551 668

R2 0.98 0.98 0.98 0.98 0.95 0.987

URPD (%) -2.01 1.54 3.03 3.22 4.04 1.964

ARPD (%) 3.97 4.55 5.61 6.62 6.8 5.510

Comparisons between HyperSAS and SeaPRISM data SeaPRISM Measurements are taken

pointing toward the direction perpendicular to the solar plane (φ is always set to 90°).

HyperSAS measurements are taken with varying φ values throughout the day resulting in values relatively lower than Li(Spr) for φ > 90o range and higher than Li(Spr) for φ < 90o.

variations in the intensity distribution of the sky radiance field can be clearly observed in Left figure.

The consistency between two measurement systems can be readily observed in the Right figure in which pointing directions of HyperSAS and SeaPRISM are within ±10o.

Comparison exhibits strong correlations and low discrepancy between the two systems.

8

Down-welling Irradiance

    Wavelength (nm)   Spectral Average

  412 443 491 551 668

R2 0.992 0.992 0.993 0.993 0.992 0.99

URPD (%) -2.14 -4.36 -1.32 -1.01 -0.35 -1.82

ARPD (%) 2.42 5.42 2.55 2.21 1.82 3.06

Comparisons between HyperSAS and SeaPRISM data Unlike HyperSAS, SeaPRISM does not

have the capability of directly making Ed measurements.

SeaPRISM, Ed has to be derived from the diffuse atmospheric transmittance, td(λ).

td is calculated from Rayleigh, aerosol, and ozone optical thicknesses (τR, τA, and τO).

HyperSAS system have been corrected for the effects of non-ideal cosine response [Zibordi 2007 et. al].

Observed negative bias can be at least partially explained by the possible presence of the absorbing aerosols [Ransibrahmanakul 2006 et. al].

Given the fact that the two systems acquire the Ed data in completely different methods, the observed discrepancy between the two data is not substantial.

)(cos),(),( 0*2 FtDE ssdsd

s

OAAARRdt

cos)()())()(1()())(1(

exp)(

9

Total sea radiance L*T

Inter-comparisons of HyperSAS and SeaPRISM sea radiance L*

T (in mWcm−2 sr−1 μm−1). for 80o ≤ φ ≤ 100o range.

    Wavelength (nm)  Spectral

Average  412 443 491 551 668

R2 0.942 0.952 0.963 0.983 0.962 0.989

URPD (%) 7.27 9.82 5.28 3.86 9.12 7.37

ARPD (%) 7.77 9.94 5.31 3.96 9.22 7.81

Comparisons between HyperSAS and SeaPRISM dataL*

T values of both systems are calculated by averaging the lowest (5% for HyperSAS & 20% for SeaPRISM) total sea radiance measurements.Different responses (integration time and field of view) of the HyperSAS and SeaPRISM to the excess sky glint perturbation removal procedure may have caused at least some of the observed discrepancies.This approach, clearly empirical, can certainly produce an overcorrection of sky glint perturbations [Zibordi 2009 et. al].Hyper SAS system's longer integration time may probably reduce the ability to filter out the rapidly changing sky glint perturbation effects and therefore existence of the offset background spectrum in the HyperSAS total sea radiance measurements is possible.Further investigations for the appropriateness of sky glint removal procedure should be granted.

10

Total sea radiance LT measurements

    Wavelength (nm)   Spectral Average

  412 443 491 551 668

R2 0.98 0.95 0.97 0.98 0.96 0.988

URPD (%) -2.83 1.62 0.89 1.64 4.66 1.2

ARPD (%) 2.19 1.99 1.79 1.86 3.18 2.21

R20.89

4 0.91 0.95 0.96 0.89 0.976

URPD (%) 0.27 4.32 2.51 2.24 5.3 2.93

ARPD (%) 5.72 6.04 4.77 4.96 7.89 5.88

Comparisons between HyperSAS and SeaPRISM data

We further carried out the comparison using the quality-assured Level 1.5 average total sea radiance, LT(Spr), data for SeaPRISM in lieu of the lowest ones.

L*T(HS) and LT(Spr) data can be considered

free of sun-glint perturbation effects (Sun-glint infected measurements have been effectively filtered out by taking the lowest values in the case of HyperSAS, and by using a measurement geometry and by accounting for field constraints in the case of SeaPRISM).

Significant statistical improvements are made by taking this step (6.1% reduction in URPD value).

11

Water leaving radiance Lw

    Wavelength (nm)   Spectral Average

  412 443 491 551 668

R20.91

50.94

60.97

80.98

20.97

4 0.991

URPD (%) -6.08 -0.27 -1.170.50

5 2.5 -0.904

ARPD (%) 10 6.02 3.75 3.39 5.58 5.75

R20.80

30.86

40.94

10.95

50.88

7 0.972

URPD (%) 14.5 13.4 5.78 3.82 8.04 9.1

ARPD (%) 21.1 16.3 7.93 5.82 10.9 12.4

Comparisons between HyperSAS and SeaPRISM dataSignificant positive bias is introduced for the comparison with the unrestricted relative azimuth range exhibiting URPD values more than 9%.

This drastic increase in the dispersion is mainly driven by the sky glint removal step in the shorter wavelengths where LT are usually low and Li measurements are high relative to the values at the longer wavelength.

Overall observed bias throughout the spectrum can be explained by directional variations in the bidirectional structure of the water leaving radiance field.

Radiative transfer simulations suggest that the water leaving radiances measured at the SeaPRISM geometry (i.e. relative azimuth angle φ = 90o) are usually lower than those measured at other relative azimuth angles for solar zenith angles θs greater than 30o which are the cases for more than 80% of the data shown in the figure.

Representativeness of LISCO as OC data Validation Site

An ideal site for the validation activity of satellite-derived parameters would provide ground truth data within a range and statistical distribution closely matching those of the satellite data.

Oceanic and atmospheric parameters are very variable from site to site and highly affect the measurements from space.

specificity of each site has to be preliminarily investigated in order to assess its representativeness and suitability for satellite validation activities.

12

Atmospheric parameters

o Aerosol optical thickness, τa, to make assessment of the atmospheric properties of the LISCO area.

Time-series analysis and matchup comparisons between satellite and in-situ data.

o To make assessment of uncertainties in the satellite OC data

Satellite Pixel Selection for Matchup Comparison

13

3km×3km pixel box for matchup comparison

Exclusion of pixel box if presence of cloud-contaminated pixels in this 9km×9km pixel box

Satellite Data Processing: Standard NASA Ocean Color Reprocessing 2009

Land & Cloud stray light contamination Atmospheric correction failure sun glint contamination

reduced or bad navigation qualitynegative Rayleigh-corrected radiance θv > 60°, θs > 70°τa(550 nm) > 0.4

Exclusion of any pixel flagged by the NASA data quality check processing

14

Quality of the atmospheric correctionComparisons of in situ and satellite retrieved aerosol optical thickness (τa)

Level 2 Satellite τa data were retrieved by applying the standard iterative-NIR atmospheric correction procedure. 108 satellite data points spanning 2 years period.Strong correlations between the in situ and satellite data are observed for every satellite missions.Most of the satellite data points fall in the within the uncertainty of the AERONET data which can be estimated by the equation 0.05× τa ±0.03.The satisfactory agreement in the retrievals of aerosol loading over LISCO area confirms the suitability of LISCO site for validation purposes.

15

Quality of the atmospheric correction

Histograms of in situ and satellite retrieved Angstrom exponent

1 11 2

2 2, log loga

aAngstrom

The quality of the atmospheric correction is highly sensitive to the spectral behavior of the aerosol optical properties.A simple, but robust, estimator of the spectral behavior of the aerosol optical properties is given by the Angstrom exponent.Low values of the Angstrom exponent indicate the predominance of coarse aerosol; conversely, high values indicate a predominance of fine aerosols.SeaPRISM data suggests that aerosols over LISCO site are typically dominated by fine aerosol particles.Limited set of aerosol models used in the atmospheric correction procedure may have implications in the estimation of water leaving radiances.

SeaPRISM

Satellite

Satellite retrieved Angstrom exponent are generally underestimated for LISCO location.

Time series of Remote Sensing Reflectance at LISCO

Rrs data exhibit significant seasonal variations in agreement between the three satellite missions and the field data.

In particular, a specific pattern of high water-leaving radiances is observable on March 17th 2010 resulting from an increase of sediment concentration following a significant storm event with higher riverine input and water body mixing.

seasonal changes are captured well by the satellite missions and the field instrumentation .

Impacts of Errors in the Atmospheric Parameter Derivation on the Rrs Retrieval

A direct consequence of underestimations of Angstrom exponent is to underestimate the aerosol radiance at the shortest wavelengths.

In turn, the water-leaving contribution will be overestimated.

At 667 nm wavelength which is close to those used for the NIR atmospheric correction, no significant impact due to the aerosol model selection is discernible.

Matchup Comparison of Diffuse Atmospheric Transmittance

Absorption and scattering characteristics of the aerosols can strongly impact the radiance distribution in the atmosphere and consequently the value of the actual atmospheric transmittance.Differences in the retrieved Diffuse Atmospheric Transmittance can also impact the Rrs retrievals.The impact is significant on the Rrs retrieval especially in the extreme blue part of the spectrum.

Conclusion Derivations of the aerosol loading by all satellite missions are satisfactory

exhibiting significant correlations between the field and satellite data.

However, misestimating aerosol models has the direct consequences on the retrievals of the remote sensing reflectances especially in the blue parts of the spectrum.

Degradation of the atmospheric correction performances due to erroneous aerosol model determination is also identified as a significant source of uncertainty on the Rrs retrieval.

The use of an enlarged set of aerosol models specifically adapted for coastal areas where fine or very fine aerosols can likely be transported from the continent is advocated

19