Abstract
The polarimetric characteristics of a class of hyperspectral radiometers commonly applied for above-water radiometry have been investigated by analyzing a sample of sensors. Results indicate polarization sensitivity increasing with wavelength and exhibiting values varying from sensor to sensor. In the case of radiance sensors, the maximum differences increase from approximately 0.4% at 400 nm to 1.3% at 750 nm. In the case of irradiance sensors, due to depolarizing effects of the diffusing collector, the maximum differences between horizontal and vertical polarization sensitivities vary from approximately 0.3% at 400 nm to 0.6% at 750 nm. Application of the previous results to above-water radiometry measurements performed in sediment dominated waters indicates that neglecting polarization effects may lead to uncertainties not exceeding a few tenths of a percent in remote sensing reflectance determined in the 400–570 nm spectral interval. Conversely, uncertainties spectrally increase toward the near infrared, reaching several percent at 750 nm in the case of oligotrophic waters.
© 2016 Optical Society of America
1. INTRODUCTION
The systematic and global monitoring of the sea through satellite ocean color is of major importance for climate and environmental activities. In fact, it allows for the quantification of optically significant constituents like pigment concentrations used as a proxy for phytoplankton biomass. Because of this, ocean color has been included among essential climate variables [1,2] of the global climate observing system (GCOS): this implies uninterrupted observations as well as continuous development of methods to minimize uncertainties in data products.
Current requirements for satellite ocean color radiometry indicate uncertainties lower than 5% in the blue–green spectral regions and radiometric stability of 0.5% per decade [2,3]. These ambitious goals entail continuous efforts in the calibration of space sensors and in the validation of data products. It is therefore essential that in situ measurements directed to support calibration and validation activities are performed with state-of-the-art technology and by applying consolidated measurement methods.
Optical radiometers, commonly used to produce in situ reference measurements in support of indirect calibration (i.e., vicarious calibration) or validation activities are often assumed to be insensitive to the polarization of light. This hypothesis, however, does not always reflect actual performance for sensors utilizing gratings, slits, beam splitters, or mirrors [4].
Within such a general context, the present study focuses on the analysis of the polarimetric characteristics of a class of hyperspectral radiometers widely used by the ocean color community. Specifically, a set of 11 commercial radiometers (i.e., RAMSES) manufactured by TriOS Mess-und Datentechnik GmbH (Rastede, Germany) were analyzed and their polarimetric characteristics quantified in view of determining the uncertainties affecting in situ data products from above-water radiometry when polarization sensitivity is neglected.
2. BACKGROUND ON THE STOKES VECTOR AND THE MUELLER MATRIX
The polarimetric characteristics of a light source can be conveniently described through the Stokes vector and Mueller matrix representations. In particular, by omitting the explicit dependence on wavelength, a source (either radiance or irradiance) can be expressed by the following four-element Stokes vector [5]:
where elements , and , alternatively indicated as , and , designate the intensity of the radiance or irradiance (i.e., ), the difference between horizontal and vertical components (i.e., ), the difference between the two linearly polarized components at 45° and 135° (i.e., ), and the difference between the two circularly polarized right and left components (i.e., ), respectively. It is noted that the formalism used in the right term of Eq. (1) makes explicit the parameters defining the polarimetric characteristics of the light source: , , and .Consistently with the Stokes vector representation, any optical system can be described by the transmittance of its components through the Mueller matrix as [3]
where indicates the fraction of the th component of the input signal transmitted to the th component of the output signal.The result from the interaction of a light source defined by the Stokes vector with an optical system determined by the Mueller matrix is indicated as and is given by the matrix multiplication :
When considering the specific case of a radiometer, it can be ideally modeled as two separate components (see Fig. 1): the first representing the optics of the sensor, solely accounting for its polarimetric characteristics; and the second representing the detector, only sensitive to the intensity of the optics output.
Defining the optics and the detector through their respective Mueller matrices and , the measured value (DN) is given by
where the single element composing the matrix , , is the detector responsivity defined by the variation of the radiometer output per unit of incident radiance or irradiance. Underlying assumptions are the spectral invariance of in each radiometer spectral band and the homogeneity of the incident flux in the radiometer field of view.From Eq. (4), is determined by
where indicates the responsivity of the radiometer to unpolarized light, and parameters , , and define its polarimetric characteristics.3. INSTRUMENTS, METHODS, AND DATA
The determination of the polarimetric characteristics of radiometers is performed assuming that all the optical components involved exhibit almost-ideal performance (i.e., nonidealities are limited to a few percent [6]). This assumption has been verified within the limits of the measurement setup. Assessments are presented in the Appendix for both radiance and irradiance sources, polarizers, and depolarizers. The description of the radiometers and the methodology applied for their characterization are presented in the next section. In addition, possible uncertainties induced by the measurement setup are also addressed and discussed.
A. Radiometers
Eleven TriOS RAMSES hyperspectral radiometers have been characterized: eight RAMSES-ARC radiance sensors (with serial numbers SAM-82CD, SAM-82CF, SAM-8313, SAM-8346, SAM-84C2, SAM-82C3, SAM-8508, and SAM-850D) and three RAMSES-ACC irradiance sensors (with serial numbers SAM-835C, SAM-82C1, and SAM-84C0).
These radiometers rely on ZEISS (Oberkochen, Germany) monolithic miniature spectrometers (MMS-1). This spectrometer, which builds on a 256 channel silicon photodiode array coupled to a grating, has a spectral resolution better than 10 nm in the 320–950 nm spectral interval with an average spectral sampling of 3.3 nm per element of the detector array and a spectral accuracy of 0.3 nm. Specific to MMS-1 spectrometers is the fiber optics connector for light supply, the entrance slit, the concave image grating, and the photodiode array, which are all permanently glued to each other to increase ruggedness and ensure permanent alignment. Additionally, light sensitivity is optimized through a fiber bundle cross-section converter with single fibers arranged in a linear configuration to form the entry slit (http://www.zeiss.com/spectroscopy/products/spectrometer-modules/monolithic-miniature-spectrometer-mms.html).
B. Polarimetric Characterization
A FEL-1000 W lamp has been used as a point source for the execution of measurements. The determination of the polarimetric characteristics of irradiance sensors has been obtained by pointing the radiometers normally to the lamp. Differently, the characterization of radiance sensors has been performed by pointing the radiometers at 45 deg incidence angle to a 99% reflectance plaque illuminated by a FEL-1000 W lamp. The stability of the source has been monitored by measuring the current flowing through the lamp using a shunt in series with the lamp. In particular, the typical maximum change observed in the 8 A current applied has been 0.3 mA. This implies variations of 0.03% in the lamp flux at 400 nm, decreasing to less than 0.02% at 700 nm [7]. All characterizations have been restricted to the 400–750 nm spectral region, which is of major interest for ocean color applications but also imposed by the spectral features of the depolarizers applied in the study.
The experimental methodology applied for the polarimetric characterization of radiometers, which implies the determination of parameters , , and , is comprehensively described in [5,6,8] and summarized hereafter. In general terms, the radiometer to be characterized needs to be pointed toward a stable source with a linear polarizer placed in between.
Specifically, the configurations used for radiance and irradiance sensors are displayed in the top and bottom panels of Fig. 2, respectively.
The orientation of the polarizer is kept constant with respect to the source, whereas the sensor is rotated around its main axis to perform successive measurements at different rotating angles. This solution allows the sensor to look at a linearly polarized light realized by coupling a source and a polarizer.
Following the flow diagram illustrated in Fig. 3, the signal resulting from the interaction of the source with the polarizer is given by the product of the respective Mueller matrices and . The generic Mueller matrix for a polarizer is defined by Eq. (6), where indicates the rotation angle between the source and the polarizer, and parameters , , , and are functions of the maximum and minimum attenuations ( and , respectively) and of the retardance of the polarizer. Specifically, where:
andAssuming the polarizer is oriented with respect to , the matrix reduces to
and When is rotated by an angle by applying the Mueller matrix the resulting matrix is The value measured by the radiometer is expressed by the product : Assuming the polarimetric parameters of both the radiometer (, , and ) and the source (, , and ) are of the order of a few percent, thus allowing to assume that any second-order addend can be safely neglected being below the instrument noise (i.e., tentatively below of the maximum output value of RAMSES radiometers), Eq. (15) reduces to The parameters and can be determined through optimal fitting (in this study, it has been performed by using the Powell function described in Section 10.5 of [9]). Parameter is neglected under the assumption that the sensitivity to circular polarization is weak or that its contribution is commonly negligible for sky radiance and relatively small for sea radiance [10–12].C. Inaccuracy of the Measurement Setup
All measurements were performed in a temperature/humidity-controlled environment where average excursions of temperature and relative humidity have been within 2°C and 20%, respectively. Sensors as well as all the other optical components have been installed on an optical bench screened with curtains and baffles to minimize ambient stray light.
Inaccuracies of the measurement setup, which may lead to measurement uncertainties, have been quantified by clearing the optical path between the sensor and the source, through the process illustrated by the flow diagram in Fig. 4.
As illustrated in Fig. 4, the signal related to the source is virtually rotated through the matrix and transmitted to the radiometer , and the measured value is
By neglecting second-order addends, Eq. (17) reduces toAny residual signal could be due to sensor–source misalignment, spatial inhomogeneity or temporal instability of the source, or inaccuracy of the derived . Two examples of residual signal are displayed in Fig. 5, as determined with SAM-84C3 but with independent installations of the radiometer (i.e., using different 99% Spectralon reflectance plaques together with a complete reinstallation of the measurement setup). In Fig. 5, stars represent the actual measurements while curves are the fitted values. Measurements and fits are only displayed for a selected set of wavelengths, namely, 412, 443, 490, 555, 670, and 750 nm. Equivalent results have been obtained for other RAMSES-ARC radiance and also RAMSES-ACC irradiance sensors.
Results from the analysis of the measurement inaccuracy, which are systematically applied for the correction of background values from measurement sequences during data processing, indicate values of the order of 0.1%–0.2% without any marked spectral feature. This latter finding may indicate a geometric dependence of the residual signal.
It is mentioned that in addition to the correction applied to remove the residual signal, before and after each characterization, multiple ambient measurements have been performed and their average has also been removed from each measurement in view of correcting for laboratory stray light.
4. RESULTS AND DISCUSSION
Main results from the characterization of the various radiometers are presented and discussed hereafter together with an evaluation of the effects of polarization sensitivity on above-water measurements. It is mentioned that handling of RAMSES radiometer data is comprehensively described in Talone et al. [13].
A. Polarization Sensitivity of the Considered Class of Radiometers
Polarimetric characteristics of the eleven RAMSES radiometers have been determined as described in Section 1. Outcomes are separately presented for radiance and irradiance sensors.
1. RAMSES-ARC Radiance Sensors
Results for RAMSES-ARC radiance sensors are displayed in Fig. 6 as a function of wavelength. Radiometers are grouped two-by-two according to the level of field exploitation: SAM-8508 and SAM-850D manufactured in 2016, never used; SAM-84C2 and SAM-84C3 manufactured in 2015, only once used in the field; SAM-82CD and SAM-82CF manufactured in 2010, and SAM-8313 and SAM-8346 manufactured in 2012, all intensively used during several field campaigns. It is specified that the custom radiometers SAM-82CD and SAM-82CF have full-angle field of view of 3 deg different from the 7 deg of any other RAMSES-ARC sensor.
Radiometers have been reciprocally aligned to maximize parameter (at the expense of ) by commonly adjusting source-to-sensor orientation. For this reason, results are presented in terms of only. Equivalent formulations can be obtained including both and by changing the relative source and sensor orientations.
As shown in Fig. 6, the difference between horizontal and vertical polarization sensitivities increases with wavelength. Additionally, assuming the radiometers were all manufactured with identical optical components, the polarization sensitivity appears to increase with the radiometer age and use. In particular, exhibits a maximum value lower than 0.4% at 750 nm for new radiance sensors but reaches 1.5% for the most used units. However, such a finding does not appear supported by additional investigations performed on radiometers SAM-8346 and SAM-84C2 after recent field deployments (i.e., pre- and post-field characterizations did not show any appreciable change in polarization sensitivities).
Applying Eq. (16) and assuming for the specific measurement geometry, the percent change of the measured signal as a function of with respect to the mean value can be approximated as
The percent change is shown in Fig. 7 for the particular case of RAMSES SAM-8346.
Figure 7 indicates that a better optimal fitting of measurements is achieved at the longest wavelengths. Differently, poorer results are obtained below 500 nm, which are likely due to the low values of (i.e., lower than 0.1%) competing with the accuracy of polarimetric characterizations.
2. RAMSES-ACC Irradiance Sensors
Parameters determined for the three RAMSES-ACC irradiance sensors are displayed in Fig. 8 as a function of wavelength.
Data indicate that the difference between horizontal and vertical polarization sensitivity for the considered irradiance sensors is lower than approximately 0.6%, with the highest values at the longest wavelengths. Values obtained for SAM-835C and SAM-82C1 are always lower that 0.3% and are very similar to each other. SAM-84C0 of more recent production when compared to the previous sensors, instead presents higher values comprised between 0.3% and 0.6%. The angular percent change is illustrated in Fig. 9 for SAM-82C1.
Similar to the case of radiance sensors, a better optimal fitting of measurements is obtained at the longer wavelengths exhibiting higher values of .
Previous results confirm the depolarizing effect of the collector at the entrance of the radiometer as opposite to radiance sensors.
B. Effects on Field Measurements
RAMSES radiometers are commonly applied to perform above-water radiometric measurements in support of satellite ocean color validation activities. By neglecting the nonistrophic distribution of the in-water light field and also high glint perturbations, the remote sensing reflectance is determined from
where is the downward irradiance at the sea surface, is the water-leaving radiance computed from (total radiance from the sea) and (sky radiance), both measured with appropriate geometries [14], and finally is the reflectance of the surface ideally determined accounting for sky radiance polarization and sea surface effects [15,16].When neglecting the polarization sensitivity and consequently the sensor orientation with respect to the preferred polarization plane, polarization perturbations in , , and measurements likely affect computed . This source of uncertainty in has been investigated through the application of parameter and additionally accounting for the degree of linear polarization (DoLP) of light [17]:
Ignoring the polarization perturbations in measurements because of the relatively small values of determined for RAMSES irradiance sensors, the perturbations in and have been estimated through additive contributions and to and , respectively. In agreement with a viewing geometry commonly applied for and measurements (i.e., relative azimuth of 90° with respect to the Sun, and viewing angles of 40° and 140° for and , respectively [14]), representative DoLP values of 0.25, 0.50, and 0.75 for (i.e., ) and of 0.40 for (i.e., ) have been considered [15–18]. The sensitivity analysis has been performed using field measurements collected with clear-sky conditions in oligotrophic waters in the Western Mediterranean Sea and in sediment-dominated waters in the northern Adriatic Sea. Measured and , and computed spectra, are displayed in Figs. 10 and 11.In view of focusing the analysis on those values maximizing perturbations due to polarization sensitivity, results are only presented for the largest spectral values of displayed in Fig. 6. The percent differences with respect to of both and are displayed as a function of wavelength in Figs. 12 and 13. A tentative value of equal to 0.028 has been applied for both case studies.
Results from the analysis of perturbations induced by neglecting polarization sensitivity are displayed in Figs. 14 and 15 through percent differences of the values displayed in Figs. 10 and 11 with respect to those determined accounting for the additive contributions shown by the solid lines and represented by the error bars.
Results, still limited to two case studies, consistently indicate that spectral differences in are limited to few tenths of a percent below 570 nm. Diversely, they may more importantly impact at longer wavelengths as a function of the water type. Specifically, as shown in Figs. 14 and 15, increases with wavelength up to 5% between 600 and 650 nm and then exceeds 10% above 700 nm in case of oligotrophic waters (see Fig. 14), while it never exceeds 1.4% in the case of sediment-dominated waters (see Fig. 15). In view of better evaluating results, the individual and contributions are separately investigated. When solely considering , polarization effects on are always lower than 0.1% for sediment-dominated waters (i.e., comparable to the uncertainties driving the determination of the related polarimetric characteristics). However, in case of oligotrophic waters, the contribution of to the polarization effects on increases above 570 nm, exceeding 5% beyond 730 nm. Higher effects are observed for whose contribution to can be up to several times larger than that quantified for . Thus, when analyzing , differences in due to polarization sensitivity are mostly driven by below 570 nm. At longer wavelengths, the contribution is only negligible in the case of sediment-dominated waters, while it may become a large fraction of for oligotrophic waters.
When considering uncertainty requirements for satellite ocean color radiometry [2], the previous results indicate that the polarization sensitivity in the blue–green spectral region of the considered class of radiometers can be disregarded at the expense of a moderate spectral increase in the measurement uncertainty as a function of illumination conditions and water type. Definitively, polarimetric sensitivity should be considered when highly accurate measurements above approximately 570 nm are required, especially in oligotrophic and mesotrophic waters.
5. CONCLUSIONS
The objective of this work was the determination of the polarimetric characteristics of a class of hyperspectral radiometers in view of quantifying their effects on in situ from above-water measurements of , , and . The characteristics of eleven hyperspectral radiometers were investigated (i.e., three RAMSES-ACC irradiance and eight RAMSES-ARC radiance units).
Results indicate that the polarimetric characteristics may appreciably vary from unit to unit. For all instruments, the polarization sensitivity exhibits an increase with wavelength. In the case of radiance sensors, the difference at 400 nm exhibits values varying from below 0.1% to 0.3%, while at 750 nm differences vary from approximately 0.4% up to 1.3%. In the case of irradiance sensors, the largest differences between horizontal and vertical polarization sensitivities vary from 0.3% at 400 nm to approximately 0.6% at 750 nm.
Concerning in situ above-water radiometric measurements, sensor orientation with respect to its preferred polarization plane can be reasonably neglected for measurements. Differently, in the case of and , it can lead to appreciable polarization effects affecting the determination of as a function of water type and the state of polarization of sky and in-water light fields. An attempt to quantify such perturbations indicates differences in computed accounting and alternatively neglecting polarization sensitivity always lower than approximately 0.7% up to 570 nm (0.3% for sediment-dominated waters). Above 570 nm, these differences slightly increase with the wavelength, reaching the maximum value of 1.4% at 750 nm in case of sediment-dominated waters, while they can exceed 10% beyond 730 nm for oligotrophic waters.
The previous results, which provide valuable uncertainty estimates, do not necessarily indicate the need for the application of measurement and data reduction methods accounting for polarization sensitivity of RAMSES hyperspectral radiometers in the blue–green spectral region. Nonetheless, this should be envisaged when highly accurate measurements are required beyond approximately 570 nm, especially in oligotrophic and mesotrophic waters.
APPENDIX A: CHARACTERIZATION OF THE OTHER OPTICAL COMPONENTS
A. Light Sources
The polarimetric characterization of the irradiance (i.e., a FEL-1000 W) and radiance (i.e., a FEL-1000 W combined with a 99% reflectance plaque) sources have been performed by fixing a polarizer on the radiometer in front of the fore optics (see Fig. 16). The radiometer was then rotated around its axis and successive measurements were performed at different angles.
Figure 17 illustrates the flow diagram of the characterization process. In this case, the applied solution allows us to look at the source by selecting different linear polarization states as a function of the rotation angle .
Equivalent to the determination of the polarimetric characteristics of a radiometer, by following the scheme in Fig. 17, we find
and Consequently, the measured value is By neglecting the second-order addends, Eq. (A3) reduces to Parameters and for the source can be obtained by optimal fitting, while parameter is neglected in agreement with assumptions already applied for .Results from the polarimetric characterization of a FEL-1000 W lamp, as obtained from independent measurements performed with three irradiance sensors, are shown in Fig. 18 as a function of wavelength, in terms of mean value and standard deviation of parameter . Similar to the radiometer characterization, parameter is set to zero by properly adjusting the measurement geometry.
In agreement with previous studies [5,19], results indicate a difference between horizontal and vertical polarizations (i.e., ) varying between 2.5% and 2.7%, slightly increasing with wavelength.
As mentioned, the determination of is obtained by optimal fitting of measurements acquired under different rotation angles (i.e., 21 with 18 deg steps). The percent change of the measured signal as a function of , as determined with respect to the mean value , is shown in Fig. 19 for the particular case of characterizations performed with the RAMSES SAM-82C1 irradiance sensor.
Spectralon reflectance plaques from Labsphere (North Sutton, New Hampshire, USA) are commonly used to produce radiance fields from irradiance sources obtained with FEL-1000 W lamps [14]. This solution leads to less pronounced polarimetric characteristics of the radiance source when compared to those of the irradiance. This is highlighted in Fig. 20 through the spectral mean values and standard deviations of parameter from measurements performed with eight RAMSES-ARC radiance sensors. The average value of increases with wavelength from 0.15% to 0.25%, confirming the results obtained in specific investigations on polarization sensitivity of Spectralon plaques [20,21]. Specifically, the expected DOLP is approximately 3%–4% of that characterizing the light incident at the plaque [20,21]. This suggests expected values of approximately 0.1% in the specific case of this study. Differences between expected (i.e., 0.1%) and quantified (i.e., 0.15%–0.25%) values are likely explained by the uncertainties intrinsic of the measurement setup.
As with the characterization of the lamp, the percent change as a function of the rotation angle is shown in Fig. 21 for the case of characterizations performed with the SAM-8346 radiance sensor.
B. Polarizers
Three linear polarizers have been used in this study. A 100 mm diameter dichroic sheet polarizer from Melles Griot (Rochester, New York, USA) was applied for the characterization of radiometers, and two 50 mm diameter dichroich glass polarizers and from Edmund Optics (Barrington, New Jersey, USA) were used for the characterization of sources. The characteristics of these polarizers have been investigated based on the method proposed by Nicodemus [6]. Specifically, the determination of parameters and is avoided because they do not affect the characterization of both sensors and sources [see Eqs. (15) and (A3)].
Consistent with Section 1, Mueller matrices of the three polarizers are denoted as , , and and their respective parameters as , , and .
The measurement setup is illustrated in Fig. 22 considering the specific case of polarizers and . It is composed of an ideally unpolarized light source obtained by illuminating a reflectance plaque with the FEL-1000 W lamp, a radiometer assumed insensitive to polarization [in this case, realized by coupling a depolarizer to the fore optics of a RAMSES-ARC radiance sensor (i.e., SAM-8346)], and the two polarizers placed in the optical path between the radiometer and the source. It is anticipated that the polarimetric characteristics of the radiance source (i.e., the lamp flux reflected by the plaque) are considered negligible when compared to those of the polarizer.
Each measurement sequence has been performed according to the following steps:
- i. First, parameters are determined by placing one polarizer at a time between the source and the sensor and measuring the radiance with and without the polarizer. If the source were unpolarized, the measurements should be independent of the rotation angle. To minimize this effect, measurements are performed with different rotation angles of the polarizer and then averaged. Parameters are calculated as , where and are the radiances measured with and without the polarizer, respectively.
- ii. Parameters are retrieved using two coupled polarizers. While the orientation of the first polarizer is kept constant with respect to the sensor, the other polarizer is rotated. In this case, by solely considering polarizers and , the radiance measured by the sensor is , where is the angle between the two polarizers. The product is determined from measurements performed at different angles and from optimal fitting of . The parameters for each polarizer are calculated by combining the products obtained with the different configurations , , and .
The spectral values of parameters determined for polarizers , , and applying the previous scheme are shown in Fig. 23(a). Extinction ratios defined as are displayed in Fig. 23(b) for the same polarizers. These latter values roughly range between and , while specifications indicate values of . Differences are explained by uncertainties of approximately 1% affecting the experimental estimates of extinction ratios as a result of uncertainties of 0.1% impacting the individual determinations of and .
C. Depolarizer
A quartz-wedge achromatic depolarizer from Thorlabs (Newton, New Jersey, USA) has been used to realize the unpolarized sensor required for the characterization of polarizers. Its performance has been verified by determining the polarimetric sensitivity of the same radiometers (both radiance and irradiance sensors) with and without coupling the depolarizer to their fore optics. Results are summarized in Fig. 24 through parameters determined from radiometers without the depolarizer (denoted by ) and alternatively with the depolarizer (denoted by ).
The same test has been performed using a quartz Lyot depolarizer from Edmund Optics. It is noted that these depolarizers are only applicable over wide spectral bands, tentatively larger than 50 nm. Results (not shown) indicate that after spectral averaging, differences between values of obtained with the different depolarizers are generally lower than 0.1%.
Funding
Joint Research Centre (JRC).
Acknowledgment
The authors would like to thank the anonymous reviewers whose comments and suggestions largely improved the quality of this work. Davide D’Alimonte is duly acknowledged for his willingness to make available his RAMSES radiometers for this analysis.
REFERENCES
1. S. Bojinski, M. Verstraete, T. C. Peterson, C. Richter, A. Simmons, and M. Zemp, “The concept of essential climate variables in support of climate research, applications, and policy,” Bull. Am. Meteorol. Soc. 95, 1431–1443 (2014). [CrossRef]
2. “Implementation plan for the global observing system for climate in support of the UNFCCC,” (2010, update) WMO GCOS Report 138, [available online at http://www.wmo.int/pages/prog/gcos/Publications/gcos-138.pdf].
3. G. Zibordi, F. Melin, K. J. Voss, C. Johnson, B. A. Franz, E. Kwiatkowska, J.-P. Huot, M. Wang, and D. Antoine, “System vicarious calibration for ocean color climate change applications: requirements for in situ data,” Remote Sens. Environ. 159, 361–369 (2015). [CrossRef]
4. C. L. Wyatt, Radiometric System Design (MacMillan, 1987).
5. H. J. Kostkowski, Reliable Spectroradiometry (Spectroradiometry Consulting, 1997).
6. National Bureau of Standards (NBS), “Self-study manual on optical radiation measurements,” in Technical Notes 910-1 through 910-8, F. E. Nicodemus, ed. (U.S. Government Printing Office, 1976–1985).
7. E. Early, A. Thompson, C. Johnson, J. DeLuisi, P. Disterhoft, D. Wardle, E. Wu, W. Mou, Y. Sun, T. Lucas, T. Mestechkina, L. Harrison, J. Berndt, and D. S. Hayes, “The 1995 North American interagency intercomparison of ultraviolet monitoring spectroradiometers,” J. Res. Natl. Inst. Stand. Technol. 103(1), 15–62 (1998). [CrossRef]
8. D. H. Goldstein, Polarized Light, 3rd ed. (CRC Press, 2016).
9. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes in C: The Art of Scientific Computing, 2nd ed. (Cambridge University, 1992).
10. R. Walraven, “Polarization imagery,” Opt. Eng. 20, 200114 (1981). [CrossRef]
11. A. Ivanoff and T. H. Waterman, “Elliptical polarization of submarine illumination,” J. Mar. Res. 16, 255–282 (1958).
12. G. W. Kattawar and X. Xu, “Detecting Raman scattering in the ocean by use of polarimetry,” Proc. SPIE 2258, 222–233 (1994). [CrossRef]
13. M. Talone, G. Zibordi, I. Ansko, A. C. Banks, and J. Kuusk, “Stray light effects in above-water remote-sensing reflectance from hyperspectral radiometers,” Appl. Opt. 55, 3966–3977 (2016). [CrossRef]
14. G. Zibordi and K. J. Voss, “In situ optical radiometry in the visible and near infrared,” in Optical Radiometry for Oceans Climate Measurements, G. Zibordi, C. Donlon, and A. Parr, eds., Experimental Methods in the Physical Sciences (Elsevier and Academic, 2014), Vol. 47.
15. D. D’Alimonte and T. Kajiyama, “Effects of light polarization and waves slope statistics on the reflectance factor of the sea surface,” Opt. Express 24, 7922–7942 (2016). [CrossRef]
16. C. D. Mobley, “Polarized reflectance and transmittance properties of windblown sea surfaces,” Appl. Opt. 54, 4828–4849 (2015). [CrossRef]
17. J. R. Schott, Fundamentals of Polarimetric Remote Sensing (SPIE, 2009), Vol. TT81.
18. K. J. Voss and N. Souaidia, “POLRADS: polarization radiance distribution measurement system,” Opt. Express 18, 19672–19680 (2010). [CrossRef]
19. K. J. Voss and L. Belmar da Costa, “Polarization properties of FEL lamps as applied to radiometric calibration,” Appl. Opt. 55, 8829–8832 (2016). [CrossRef]
20. A. Bhandari, B. Hamre, Ø. Frette, L. Zhao, J. J. Stamnes, and M. Kildemo, “Bidirectional reflectance distribution function of Spectralon white reflectance standard illuminated by incoherent unpolarized and plane-polarized light,” Appl. Opt. 50, 2431–2442 (2011). [CrossRef]
21. D. A. Haner, B. T. McGuckin, and C. J. Bruegge, “Polarization characteristics of Spectralon illuminated by coherent light,” Appl. Opt. 38, 6350–6356 (1999). [CrossRef]