In situ determination of the remote sensing reflectance : an inter-comparison

Inter-comparison of data products from simultaneous measurements performed with independent systems and methods is a viable approach to assess the consistency of data and additionally to investigate uncertainties. Within such a context the inter-comparison called Assessment of In Situ Radiometric Capabilities for Coastal Water Remote Sensing Applications (ARC) was carried out at the Acqua Alta Oceanographic Tower in the northern Adriatic Sea to explore the accuracy of in situ data products from various inand above-water optical systems and methods. Measurements were performed under almost ideal conditions, including a stable deployment platform, clear sky, relatively low sun zenith angles and moderately low sea state. Additionally, all optical sensors involved in the experiment were inter-calibrated through absolute radiometric calibration performed with the same standards and methods. Inter-compared data products include spectral waterleaving radianceLw (λ), above-water downward irradiance Ed(0,λ) and remote sensing reflectance Rrs(λ). Data products from the various measurement systems/methods were directly compared to those from a single reference system/method. Results for Rrs(λ) indicate spectrally averaged values of relative differences comprised between −1 and +6 %, while spectrally averaged values of absolute differences vary from approximately 6 % for the above-water systems/methods to 9 % for buoy-based systems/methods. The agreement between Rrs(λ) spectral relative differences and estimates of combined uncertainties of the inter-compared systems/methods is noteworthy.


Introduction
Climate studies largely rely on environmental indices de rived from remote sensing data (e.g.Behrenfeld et al., 2006: Achard et al., 2002: Kaufman and Tanré, 2002: Stroeve et al., 2007).Satellite ocean color data are also increasingly applied for coastal and inland water management, includ ing water quality monitoring, harmful algal bloom detection and sediment transport studies (Brando and Dekker, 2003: Stumpf and Tomlinson, 2005: Ruddick et al., 2008).How ever, the confident use of these data requires the quantifi cation of their uncertainties.This is generally accomplished through the comparison of satellite products with in situ ref erence measurements.In the case of satellite ocean color, the spectral remote sensing reflectance Rrs determined from topof-atmosphere radiance is the primary data product used for the generation of higher level products such as chlorophyll a concentration (Chi a).As a consequence, access to accurate in situ Rrs is essential for the assessment of primary data products from satellite ocean color missions.
In situ Rrs data are obtained through in-water and above water optical measurement systems.Both approaches rely on a number of methods frequently tied to a variety of instru ments characterized by different design and performances.This aspect together with a diverse implementation of mea surement methods, the application of different processing schemes, and the use of various sources and methods for the absolute radiometric calibration of field instruments may G. Zibordi et aí.: In situ determination of the remote sensing reflectance lead to unpredictable uncertainties affecting the assessment of satellite products.
The quantification and the successive reduction of uncer tainties for in situ measurements is thus a major challenge for ocean color scientists actively involved in field radiometry.Basic tasks include the precise implementation and ap plication of established measurement and analysis methods, and additionally an investigation and quantification of each source of uncertainty in primary data products.Best practice suggests the verification of each measurement and process ing step through inter-comparison exercises.
This work summarizes results from a radiometric inter comparison performed in the northern Adriatic Sea with the main objective of evaluating the agreement of in situ R rs products determined through the application of independent measurement systems and methods.

The inter-comparison
Inter-comparison activities are essential to evaluate the per formance of independent measurement methods and also the ability of individuals to properly implement them (e.g.Thome et al., 1998;Hooker et al., 2002a;Barton et al., 2004).A major requirement for field inter-comparisons is the need for performing measurements with different sys tems/methods under almost identical conditions.In the case of optical oceanography, this is better achieved with the use of fixed deployment platforms instead of ships.In fact, grounded platforms offer the major advantage of deploying instruments under controlled geometries not affected by su perstructure drift and roll.This favourable situation is easily achieved at the Aequa Alta Oceanographic Tower (AAOT) in the northern Adriatic Sea (e.g.Zibordi et al., 1999Zibordi et al., , 2009a;;Hooker and Zibordi, 2005).
The inter-comparison activity presented and discussed in this work focuses on a variety of measurement systems and methods applied to produce in situ data for the validation of marine primary radiometric products for the Medium Reso lution Imaging Spectrometer (MERIS) onboard the Envisat platform of the European Space Agency (ESA).The inter comparison, called Assessm ent o f In Situ Radiometric Ca pabilities for Coastal Water Remote Sensing Applications (ARC) was conceived within the framework of the MERIS Validation Team (MVT) and supported by ESA in the context of international activities promoted by the Working Group on Calibration and Validation (WGCV), Infrared and Visi ble Optical Systems (IVOS) subgroup of the Committee on Earth Observation Satellites (CEOS).
ARC activities comprise two successive phases carried out during July 2010.In the first phase, field measurements were carried out at the AAOT during four days character ized by favourable illumination and sea state conditions.In the second phase, the optical sensors previously deployed at the AAOT were inter-calibrated at the Joint Research Centre (JRC).This inter-calibration was achieved through the abso lute radiometric calibration of the optical sensors by using identical laboratory standards and methods, with the excep tion of one system (see Sect. 3.3.3)also calibrated at the JRC using the same standards and methods, but at a different time.Data products included in the inter-comparison were then all computed from data calibrated (or corrected) using consis tently determined radiometric coefficients.
The inter-comparison of data products from different mea surement systems and methods is here performed, relying on data from a single system/method considered as the refer ence because of its well documented performances and long standing application to the validation of satellite ocean color products.Due to the variety of multispectral and hyperspectral sensors included in the inter-comparison, the data anal ysis has been restricted to the center-wavelengths of major interest for satellite ocean color: 412, 443, 490, 510, 555, and 665 nm.The presentation of results is supported by un certainty budgets quantified for each system/method.

M easurement systems and methods
The ARC inter-comparison includes an assortment of in-and above-water measuring systems and methods.To rational ize their description, the basic elements common to generic methods (i.e.in-and above-water) are hereafter summarized, then details on each measurement system and method are provided.It is anticipated that the analysis of results is fo cused on Rrs determined according to its simplest defini tion (see Sect. 3.1) without applying any correction for the anisotropy of in-water radiance distribution (i.e. the bidirec tional effects).In fact, the objective of this work is to quan tify differences among fundamental radiometric products de rived from the application of various systems and methods; the use of the same scheme to account for bidirectional ef fects would not impact the comparison, while the application of different schemes is out of the scope of the study.In line with such a strategy, the dependence on the viewing geome try of above-water measurements (also depending on the in water radiance distribution) has been addressed by applying an identical correction scheme for all considered methods.

Overview on in-water measurements
In-water radiometiy relies on subsurface continuous or fixeddepth profiles of upwelling radiance Lu{z,X,t), downward irradiance E¿{z, X,t) and occasionally also upward irradi ance Eu{z,X,t) at depth z, wavelength X and time t.The above-water downward irradiance Ed(0+,X,t) is also mea sured to complement the in-water data.These latter data are used to extrapolate to 0" (i.e.just below the water surface) the radiometric quantities which cannot be directly measured because of wave perturbations.Above-water downward irra diance data are used to minimize the effects of illumination changes on in-water radiometric measurements during data collection.
In-water continuous profiles of radiometric quantities re sult generally from measurements performed with optical sensors operated on profiling systems (e.g.winched or freefall).Due to wave focusing and defocusing, the accuracy of sub-surface radiometric products largely depends on the sam pling depth interval and on the depth resolution (Zaneveld et al., 2001;D 'Alimonte et al., 2010).Thus, highly accurate in water radiometric products can only be determined by sam pling near the surface (especially in coastal regions due to possible vertical non-homogeneities in the optical properties of seawater), and by producing a large number of measure ments per unit depth not significantly affected by tilt (Zibordi et al., 2004a).
In-water fixed-depth profiles mostly result from the use of optical sensors operated on buoys at nominal depths.These buoy-based systems generally provide the capability of mea suring Lu(z,X,t), Ed{z,X,t) and possibly also Eu(z,X,t) at multiple depths (typically between 1 and 10 m ), in addition to £ 0 (0 +, X, t).By neglecting the effects of system tilt, the ac curacy of radiometric products determined with buoy-based systems is a function of the discrete depths selected for the optical sensors, the acquisition rate and the duration of log ging intervals (Zibordi et al., 2009a).
The same data reduction process is in principle applicable to both fixed-depth and continuous profile radiometric data S(z, X, t) (i.e.Lu(z, X, t), £ u(z, X, t) and £d(z, X,t)).The ini tial step, leading to minimization of perturbations created by illumination change during data collection, is performed ac cording to: where So(z, A,io) indicates radiometric values as if they were all taken at the same time io, and Zid(0-1-, A, io) speci fies the above-water downward irradiance at time io (with io generally chosen to coincide with the beginning of the acqui sition sequence).
Omitting the variable i, the sub-surface quantities So(0-, X ') (i.e.£ u(0_ , X), Eu(0~,X) and £ ¿ (0 " , A.)) are then determined as the exponentials of the intercepts resulting from the least-squares linear regressions of ln So (z, X) versus z within the extrapolation interval identified by zi < z < Z2 and chosen to satisfy the requirement of linear decay of lnSo(z, A.) with depth.The negative values of the slopes of the regression fits are the so-called diffuse attenuation co efficients K%(X) (i.e.Ki(X), KU (X) and Kd(X) determined from Lu(z,X,t), Eu(z,X,t) and Ed(z,X,t) values, respec tively, from the selected extrapolation interval).
The radiometric quantity of major relevance here is the so-called water-leaving radiance LW(X) in units of mW cm -2 pm -1 sr-1 .This is the radiance leaving the sea quantified just above the surface from: where the factor 0.543, derived assuming the seawater re fractive index is independent of wavelength (Austin, 1974), accounts for the reduction in radiance from below to above the water surface.
A second radiometric quantity central to this study is the remote sensing reflectance £ rs(A) in units of sr-1 , given by: with £d(0+ , X) in units of mW cm -2 pm -1 .£ rs(A) is thus a quantity corrected for illumination condi tions depending on sun zenith angle, Sun-Earth distance and atmospheric transmittance (Mueller et al., 2002).

Overview on above-water measurements
Above-water methods generally rely on measurements of (i) total radiance from above the sea Lj(9, A(p,X) (that includes water-leaving radiance as well as sky-and sunglint contributions); (ii) the sky radiance £ ;(0 ;, A (p,X)\ and (iii) usually also £d(0+ ,A).The measurement geometry is defined by the sea-viewing angle 0, the sky-viewing angle 6' and the difference between sun and sensor azimuth an gles, A (p = < P o-, P (Deschamps et al., 2004;Hooker et al., 2004;Zibordi et al., 2004b).The accurate determination of £ W(A) then depends on the capability of minimizing glint contributions through the use of suitable measurement ge ometries (Mobley, 1999), and additionally, the application of statistical filtering schemes to Lj (Hooker et al., 2002a;Zi bordi et al., 2002b), or physically-based correction methods relying on known reflectance properties of seawater in the near-infrared spectral region (Ruddick et al., 2006), or al ternatively, polarisers to directly reduce sky-and sun-glint (Fougnie et al., 1999).
In the case of non polarized systems, measurements of Lj(9, A(p,X) and Li{9',A(p,X) for the determination of £ W(A) are generally performed at 0 = 40° and 9' = 140°, with A (p chosen between + 90° and +135° or alternatively -90° and -135°.The value of A< $ = ±135° is considered the most appropriate (see Mobley, 1999).However, its ap plication must be regarded with special care because it may more likely lead to measurements significantly affected by the shadow cast by the deployment superstructure in the anti solar region (i.e.nearby the sea area seen by the sensor).

WiSPER
The Wire-Stabilized Profiling Environmental Radiometer (WiSPER) is a winched system deployed through a custombuilt profiling rig at a speed of 0.1 m s -1 at 7.5 m away from the main structure of the AAOT.The Lu, Eu and Ed optical sensors are mounted at approximately the same depth (see Zibordi et al., 2004a).The rigidity and stability of the rig is maintained through two taut wires anchored between the tower and the sea bottom.The immovability of the AAOT and the relatively low deployment speed ensure an accurate optical characterization of the subsurface water layer.WiSPER sensors include three OCE200 for Eu(z,X,t), Ed(z, X, t) and EjEO-1-, X, t), and one OCR-200 for Lu(z, X, t) measurements.These sensors, manufactured by Satlantic Inc. (Halifax, Canada), provide data at 6 Hz in seven spectral bands 10 nm wide centered at 412, 443, 490, 510, 555, 665 and 683 nm.The L u sensor has approximately 18° in-water full-angle field of view (FAFOV).Each WiSPER measure ment sequence includes data from down-and up casts.
WiSPER data are processed in agreement with the scheme presented in Sect.3.1.Radiometric products for ARC inter comparison have been determined choosing an extrapola tion interval of 0.3-3.0m .Additional processing includes the application of corrections for superstructure perturba tions (Doyle and Zibordi, 2002), self-shading of L u and Eu sensors (Gordon and Ding, 1992;Zibordi and Ferrari, 1995;Mueller et al., 2002), and non-cosine response of the above water Ed sensor (Zibordi and Bulgarelli, 2007).In addition to the diameter of the sensors, the application of these correc tions requires spectral values of the above-water diffuse to direct irradiance ratio (r), and subsurface seawater absorp tion (a) and beam-attenuation (c) coefficients (all regularly measured during each WiSPER deployment).geometry identified by 0, A < fi, sun zenith 0o, and of the sea state conveniently expressed through the wind speed W.
The water-leaving radiance LW(X) for a nadir-view direc tion is then determined by: where 34(0, W) and 34o (i.e.34(0, W) at 0 = 0) account for the sea surface reflectance and refraction, and depend mainly on 0 and W (Morel et al., 2002).The spectral quantities ß (0 , A(f>, 0o, X, t a, IOP) and <2«($o, A, ra, IOP) are the Qfactors at viewing angle 0 and at nadir (i.e.0 = 0), respec tively, describing the anisotropic distribution of the in-water radiance.Publically available Q-factors (Morel et al., 2002) have been theoretically determined as a function of 0, A < f> , 0o, the atmospheric optical properties (conveniently expressed through the aerosol optical thickness ra, even though as sumed constant), and the seawater inherent optical proper ties IOPs (conveniently expressed through Chi a for oceanic waters).
The remote sensing reflectance is then computed from Eq. (3) using measured or theoretical values of £ (](0+ , A).

Details on individual measurement systems and methods
Systems and methods included in the ARC inter-comparison are listed in Table 1 together with the institutes respon sible for data collection, processing and quantifying sys tem/method uncertainties.Additionally, Table 2 provides de tails for each system in conjunction with the main input pa rameters required for data processing.
TRIOS-B Above-water manned hy perspectral data in the 400-900 nm spectral region with 10 nm resolution 7° (in air) 0.1 Hz, 250 ms (typical for Lj{&, Aip,X) during ARC) TRIOS-E Above-water manned hy perspectral data in the 400-900 nm spectral region with 10 nm resolution 7° (in air) 0.1 Hz, 250 ms (typical for Lj{&, Aip,X) during ARC) An analysis of uncertainties of WiSPER R rsW from ARC measurements, performed assuming each contribution inde pendent from the others, indicates values in the range of ap proximately 4-5 % in the selected spectral region (see Ta ble 3).The uncertainty sources considered here are (i) un certainty of the absolute radiance calibration (Hooker et al., 2002b) and immersion factor (Zibordi, 2006) for the L u sen sor (i.e.2.7% and 0.5% , respectively, composed statisti cally): (ii) uncertainty of the correction factors applied for removing self-shading and tower-shading perturbations com puted as 25 % of the applied corrections: (iii) uncertainty of the absolute irradiance calibration of the above-water Ed sen sor (Hooker et al., 2002b) and uncertainties of the correction applied for the non-cosine response of the related irradiance collectors (Zibordi and Bulgarelli, 2007) (i.e.2.3% and 1 %, respectively, composed statistically): (iv) uncertainty in the extrapolation of sub-surface values due to wave perturbations and changes in illumination and seawater optical properties during profiling cumulatively quantified as the average of the variation coefficient of R rsW from replicate measurements.
It is noted that the proposed uncertainty analysis accounts for fully independent calibrations of E d and L u sensors (i.e. as obtained with different lamps and laboratory set-ups).The use of the same calibration lamp and set-up leads to a reduc tion of approximately 1 % of the quadrature sum of spectral uncertainties for WiSPER Rrs(E).
It is additionally noted that the bottom effects were not in cluded in the uncertainty analysis being assumed to be negli gible for the measuring conditions characterizing the ARC inter-comparison.In fact, despite the shallow water depth at the AAOT (i.e.17m), an evaluation of bottom perturba tions based on the scheme proposed by Zibordi et al. (2002a) www.ocean-sci.net/8/567/2012/Ocean Sei., 8, 567-586, 2012 indicates maximum values smaller than 0.5 % for Rrs at the 555 nm center-wavelength.The quality of W iSPER radiometric products is traced through quality-indices determined during data processing.These include (i) temporal changes in illumination condi tions as caused by cloudiness and quantified through the standard deviation of £ d(0+ , X,t) at each X; (ii) poten tial difficulties in the determination of subsurface extrap olated quantities flagged by a relatively small number of measurements per unit depth, significant differences between Eu(z, X, to)/Lu(z, X, to) at different depths in the extrapola tion interval, and large differences between Eo(0~,X, to) and £d(0+ , X, to)', and (iii) poor illumination conditions, result ing from high sun zenith angles or cloudiness, both quanti fied through values of the diffuse to direct irradiance ratio r(X) exceeding a threshold.These quality-indices, recorded as an integral part of the radiometric data set, are used to comprehensively qualify data products.The low deployment speed of WiSPER and the almost ideal sky and sea state con ditions characterizing the ARC measurements made all the collected data applicable for the inter-comparison.

TACCS
The Tethered Attenuation Chain Colour Sensors (TACCS) manufactured by Satlantic Inc. consist of an above-water E¿ sensor mounted on a buoy, an L u upwelling radiance sensor at depth zo = 0-5 m, and a chain of four in-water E¿ sensors at increasing depths zi-A weight suspended at the bottom of the chain stabilises the system against wave action.TACCS offers the advantage of easy deployment from small boats and the possibility of being operated at distances minimizing ship perturbations.Additionally, Lu(zo, X, t) data taken rela tively close to the surface can be averaged over time to mini mize the effects of wave focussing and defocusing.The main disadvantage is the reduced depth resolution with respect to profilers, requiring a careful quality check of data to exclude cases affected by near-surface vertical non-homogeneities.
Individual measurement sequences comprise collection of Lu(zo,X,t), Eo(zi,X,t) and Eo(0+,X,t) during intervals of three minutes.Measurement sequences are retained and cor rected using Eq. ( 1) for the effects of illumination change during data collection when the variability of iid(0+ , M 0 is no greater than 2.5 % with sea state 0-1, 3.0 % with sea state 1-2, or 4 % with sea state 4 (essentially, the variability should be consistent with wave action rather than with changes in illumination which have a higher frequency).Derived Lu(zo,X,to) and £d(zi, X, to) are then averaged over the three minute interval to determine time-averaged Lu(zo, X, to) and Ed(zi, X, to), respectively.
Log transformed £d(zi, X, to) are then applied to compute Ko(X) through least-squares linear regressions.Because of the similarity of Äi(A.) and K¿(X) values (Mobley, 1994), subsurface L u(0 , X) is then obtained from: Quality checks for L u(0_ ,A) include the evaluation of R2 determined from the regression of Eo(zi,X,to) at depths Z i and the visual inspection of É¿(zi, 490, io) profile data.If R2 and the vertical profile of log-transformed E¿(zu 490, to) in dicate non-homogeneity of the optical properties in the water column, then the lowest depth(s) are removed from the pro cessing.These steps aim to ensure the validity of the hypoth esis of homogeneous seawater optical properties between the surface and at least the second measurement depth.
Self-shading corrections of Lu(0~,X) data are performed following the methodology detailed by Mueller et al. (2002).Input quantities are (i) the total seawater absorption coeffi cient a(X), on a first approximation assumed equal to Kd(X) (Mobley, 1994) directly determined from E¿(zi,X,t) val ues; (ii) the diameter of the L u sensor (by neglecting the marginal effects of the surface float (Moore et al., 2010)); and (iii) the diffuse to direct irradiance ratio r(X) calculated from simulated data using the model of Bird and Riordan (1986) with extra-atmospheric sun irradiance from Thuillier et al. (2003) and aerosol optical thickness ra(X) from col located sun-photometric measurements.Comparison of self shading corrections determined for ARC measurement con ditions with the former 2-D scheme (where the system is assumed a disk with diameter equal to the case of the L u sensor) and corrections from a 3-D scheme developed by Leathers et al. (2001) for an equivalent buoy system indi cates differences well within the 35 % uncertainty declared for the 2-D based scheme (see the following subsections).
Two TACCS systems were deployed during the ARC inter comparison: one owned and managed by Stockholm Univer sity in collaboration with Bio-Optika (identified as TACCS-S), and the second by Sagremarisco Lda also in collaboration with Bio-Optika (identified as TACCS-P).Although the two TACCS systems have different radiometric configurations, the mechanical design is almost identical.
During the ARC activities both TACCS were operated at a few meters from each other at approximately 30 m from the AAOT.

TACCS-S
TACCS-S measures £'(](0+ ,A ,i) at 443, 490 and 670 nm, and E¿(zi, X, t) at 490 nm at the nominal depths of 2, 4, 6 and 8 m.Measurements of L u(zo, X, t) are performed at 412, 443, 490, 510, 560, 620 and 6 7 0 nm at the nominal depth zo = 0.5 m with an in water FAFOV of approximately 20°.All sensors have a 10 nm bandwidth.The acquisition rate is approximately 1 Hz.TACCS-S does not have tilt sensors, but when carefully balanced in water, combined x-y tilt of the above-water E¿ sensor remains below 5° at sea state 0-1.Since £ ,d(0+ ,/C i) is only measured at 443, 490 and 670 nm, simulated irradiances (computed using the same model utilized for the determination of r ) are normalized to the actual E¿(0+,X,t) to determine values at 412, 510, 560 and 620 nm.
Similarly, since Ad (A.) is only measured at 490 nm, spec tral values of K¿(X) at the relevant center-wavelengths are determined from measurements of a(X) and c(X) performed with an AC-9 (WET Labs, Philomath, USA) following Kirk (1994) with: where b(X) = c(X) -a(X), ßo is the mean cosine of the re fracted solar beam just below the sea surface, and gi and g2 constants depend on the scattering phase function.For the processing of ARC data, constant values are ßo = 0.86, g\ = 0.425, and ^2 = 0.19 corresponding to the Petzold (1972) phase function.It is assumed that these parameters provide the correct spectral shape of Ad (A), although its absolute value may be biased due to dependence of ßo on 0o-The analysis of uncertainties for TACCS-S Ars(A) from ARC measurements indicates values in the range of approx imately 7-8 % (see Table 4).Considered uncertainty sources are (i) uncertainty of the absolute radiance calibration and immersion factor, computed as for WiSPER; (ii) uncertainty of the correction factors applied for removing self-shading perturbations in Lu(0~,X) computed as 35% of the ap plied corrections (the higher expected values with respect to WiSPER are explained by the assumption of a(X) = K¿(X)); (iii) uncertainty of the absolute irradiance calibration of the above-water E¿ sensor (Hooker et al., 2002b) and non-cosine response of the related irradiance collectors (Zibordi and Bulgarelli, 2007) (i.e.2.3% and 2% , respectively, com posed statistically); (iv) uncertainty in the determination of £ 0 (0 +, X, t) at missing center-wavelengths estimated by cal culating Ed(0+,X,t) using the model of Bird and Riordan (1986) with ra(500) = 0.45 (average for measurements per formed during the field activities) and by bracketing the Angstrom exponent at 0.0 and 2.0; (v) uncertainties due to the assumption of K\(X) = Ad (A) resulting from the quadra ture sum of 1.7%, average difference between Ad (A) and ARA) determined through Hydrolight (Mobley, 1998) sim ulations using the specific TACCS E¿ sensor depths, and of approximately 1.7% per 100 nm due to spectral extrapola tion as estimated from actual measurements; (vi) uncertain ties due to geometrical effects estimated from simulations, assuming tilt of 5° for the above-water E¿ sensor, relative sun-sensor azimuth of 180°, do = 45°, r(X) computed with ra(500) = 0.45 and Angstrom exponent equal to 1.39 as re sulting from measurements performed during field activities; and (vii) uncertainty in the extrapolation of sub-surface val ues, computed as for WiSPER.
Uncertainties do not take into account potential shading of the in-water E¿ sensors by the cable.This is supported by the assumption that this perturbation similarly affects mea surements at all depths and thus does not significantly influ ence the determination of Ad(A).No uncertainty has been as signed to the nominal depths of in-water E¿ sensors assumed to be within ± 2 cm under calm sea.
Finally, in view of the inter-comparison analysis, it is anticipated that differences between TACCS-S centerwavelengths at 560 and 670 nm with respect to the reference ones at 555 and 665 nm are neglected.

TACCS-P
TACCS-P has hyperspectral sensors for E¿(0+ ,X,t) and Lu(zo,X,t) measurements with spectral range of 350 800 nm and resolution of 11 nm.The L u sensor has in-water FAFOV of approximately 18°.E¿(zi,X,t) is measured at 412, 490, 560 and 665 nm with a bandwidth of 10 nm at nominal depths of 2, 4, 8 and 16 m.Sampling rate is typi cally 2 Hz, although it may vary depending on illumination conditions.Tilt and compass sensors provide information on the levelling and orientation of the radiometer utilized for Ed(0+,X,t) measurements.
Since Kd(X) is only determined at 412, 490, 560 and 665 nm, at the other relevant center-wavelengths it is deter mined with the following scheme.The value of Chi a is esti mated from Ad (490) by inverting Eq. ( 9) from Morel and An toine (1994), duly taking into account the diffuse attenuation coefficient of pure seawater.Then the same equation with the estimated Chi a is applied to determine the diffuse atten uation coefficient of seawater (pure seawater excluded).The derived Ad (7.) spectrum is subsequently normalised to the ex perimental values determined at 412, 490, 560 and 665 nm.£d(0+ A, t) is calculated by two methods depending on tilt values during the sampling period.The value of £d(0+ , A 0 is kept unchanged if the combined x-y tilt value is less than 2°.Otherwise a correction is applied by assuming that the diffuse irradiance is unaffected by tilt (i.e. by ignoring the sky radiance distribution) according to: where £d(0 , V i, 0S) indicates data uncorrected for tilt and f(9o,Os) is given by: ƒ ( 0 0 ,0s) = COS (0S) COS( 00)' with 0S the apparent angle of the sun to the collector plane of the irradiance sensor.This correction, however, only applies to tilts less than 8° (chosen on the basis of trials performed under stable illumi nation conditions).In fact, when the tilt becomes high the radiance from the sea surface may add large perturbations, especially in the anti-solar direction.
The analysis of uncertainties for TACCS-P Rrs(A) from ARC measurements indicates values in the range of approx imately 6-7 % (see Table 5).Considered uncertainty sources are (i) uncertainty of the absolute radiance calibration and immersion factor of the L u sensor, computed as for W iS PER: (ii) uncertainty in the correction factors applied for re moving self-shading perturbations in Lu(zo, 7-, t), computed as for TACCS-S; (iii) uncertainty of the absolute irradiance calibration of the above-water sensor and the non-cosine response of the related irradiance collectors, computed as for TACCS-S: (iv) uncertainties due to the assumption of K\(X) = Kd(X), computed as for TACCS-S: (v) uncertainties due to geometrical effects computed as for TACCS-S: and (vi) uncertainty due to the extrapolation of sub-surface val ues, computed as for WiSPER.

SeaPRISM
The SeaWiFS Photometer Revision for Incident Sur face Measurements (SeaPRISM) is a modified CE-318 sun-photometer (CIMEL, Paris) that has the capabil ity of performing autonomous above-water measurements.SeaPRISM is regularly operated at the AAOT from a de ployment platform located in the western corner of the su perstructure at approximately 15m above the sea level (Zi bordi et al., 2009c).Measurements performed with a FAFOV of 1.2° every 30m in for the determination of LW(X) at a number of center-wavelengths including 412, 443, 488, 531, 551, 670 nm (Zibordi et al., 2009c) are (i) the direct sun irra diance Es(9o,(po,X) acquired to determine the atmospheric optical thickness ra(X) used for the theoretical computa tion of Zid (0-1-, A), and (ii) a sequence of 11 sea-radiance measurements for determining Lj (9, A<f>, X) and of 3 skyradiance measurements for determining L¡(9', A<fi, X).These sequences are serially repeated for each X with A < f> = 90°, 9 = 40° and 9' = 140°.The larger number of sea measure ments, when compared to sky measurements, is required be cause of the higher environmental noise (mostly produced by wave perturbations) affecting the former measurements dur ing clear sky.
Values of Rrs(A) are determined from SeaPRISM mea surements in agreement with basic principles provided in Sect.3.2.An additional element is the need to minimize the effects of glint perturbations in Lj(9, A< f> ,X) and possibly the effects of cloud perturbations in Li(9', A<f>, X).This is achieved by deriving these values from the average of inde pendent measurements satisfying strict filtering criteria (Zi bordi et al., 2009c: Zibordi, 2012).
Finally, as already anticipated, the value of £'(](0+ , X) is quantified theoretically under the assumption of clear sky.Specifically, where D2 accounts for the variations in the Sun-Earth dis tance as a function of the day of the year (Iqbal, 1983), td(X) is the atmospheric diffuse transmittance computed from mea sured values of ra(A.)(Gordon and Clark, 1981), and Eo(X) is the average extra-atmospheric sun irradiance (Thuillier et al., 2003).Quality flags are applied at the different processing lev els to remove poor determinations of Rrs(X).Quality flags include checking for (see Zibordi et al., 2009c) cloud con tamination, high variance of multiple sea-and sky-radiance measurements, elevated differences between pre-and post deployment calibrations of the SeaPRISM system, and spec tral inconsistency of the normalized water-leaving radiance Lwn(X) given by: ^w n (A.) -Rrs(E)Eo(X). (11) It is recalled that SeaPRISM data, handled through the Ocean Color component of the Aerosol Robotic Network (AERONET-OC, Zibordi et al., 2009c), are mostly intended to support satellite ocean color validation activities.Because of this, to minimize the effects of differences in centerwavelengths between the satellite and SeaPRISM data prod ucts a band-shift correction scheme has been developed for the latter.These corrections are performed relying on a biooptical model requiring Chi a and IOP values estimated through regional empirical algorithms applied to spectral ra tios of Lwn(X) (Zibordi et al., 2009b).Band-shift corrections have then been applied to SeaPRISM data products con tributing to the ARC inter-comparison to match the reference center-wavelengths.
SeaPRISM is the only system deployed during the ARC experiment that was not immediately inter-calibrated.This is justified by its continuous operation for periods of 6-12 months at the AAOT.However, pre-and post-deployment calibrations performed at the JRC with the same standards and methods applied during ARC indicated differences typi cally within 0.6 % during a 9 month period.Estimated uncertainties of SeaPRISM Rrs(k) data for the ARC experiment are approximately 4-5 % in the blue-green spectral regions and 10% in the red (see Table 6).These have been determined accounting for contributions from (i) uncertainty of the absolute radiance calibration (Hooker et al., 2002b) for Lj and L¡ sensors, but neglecting sensi tivity changes during deployment which should contribute less than 0.2 % when assuming a linear change with time between pre-and post-deployment calibrations: (ii) uncer tainty of corrections for the off-nadir viewing geometry com puted as 25 % of the applied correction factors (these rela tively large percent values are expected to account for un certainties due to the intrinsic assumption of Case 1 w a ter at the AAOT) ; (iii) variability in specific parameters re quired for the determination of Rrs(k) (taken from Zibordi et al., 2009c, and estimated from multi-annual measure ments accounting for changes in wind speed, sea surface re flectance, and atmospheric diffuse transmittance): (iv) uncer tainty in Eo (A) estimated by assuming ± 1 nm uncertainty in center-wavelengths: and finally, (v) environmental perturba tions (e.g.wave effects, changes in illumination and seawater optical properties during measurements) quantified as the av erage of the variation coefficient obtained from Rrs(A) values from replicate measurements.
The uncertainty related to band-shift corrections has not been accounted for in the overall budget.However, an eval uation of band-shift corrections applied to SeaPRISM data to match center-wavelengths of various satellite sensors indi cated average values of a few percent (Zibordi et al., 2006).Thus, the uncertainty affecting these values is expected to be a small fraction of the applied corrections and consequently to not significantly impact the uncertainty budget proposed for SeaPRISM Rrs(A).

TRIOS
Above-water TriOS (Rastede, Germany) Optical Sys tems (TRIOS) are composed of two RAMSES ARC-VIS hyperspectral radiometers measuring Lj(9, A ip,k) and Li(9r, A<p, k ) , and one RAMSES ACC-VIS for £d(0+ , k).Measurements are performed in the 400-900 nm spectral range with resolution of about 10 nm for the output data.The nominal FAFOV of radiance sensors is 7°.
The basic measurement method applied during ARC is that developed by Ruddick et al. (2006, see the main pa per and web appendices) based on the generic M ethod 1 de scribed in the Ocean Optics Protocols (Mueller et al., 2002).
L t and Lj sensors are simultaneously operated on the same frame with identical azimuth plane, and 6 = 40° and 0 '= 140°, respectively.Measurement sequences are per formed with user-definable intervals and frequencies, and in tegration time varying automatically between 8 ms and 4 s depending on the brightness of the target.During ARC, the deployment frame was adjusted for each measurement se quence to satisfy the requirement of Aip = 1 3 5 ° (or occa sionally of Aip = 90°, chosen to avoid superstructure pertur bations) .
Details on data processing, including measurement selec tion, averaging and quality checks, are described in Ruddick et al. (2006) (web appendix 1: http://aslo.org/lo/toc/vol_51/issue_2/1167al.pdf).A few elements on data processing are however provided here for completeness.
(13) Minimization of perturbations due to wave effects is then achieved through the so-called turbid water near-infrared (NIR) similarity correction (Ruddick et al., 2005) by deter mining the departure from the NIR similarity spectrum with: where wavelengths Aí and A 2 are chosen in the near infrared and the constant a is set accordingly from where the correction is assumed spectrally invariant.The cor responding NIR similarity corrected water-leaving radiance is calculated as: A number of data products (i.e. 5) are then averaged to obtain the NIR similarity corrected Lw(9, Aip, X).
For ARC measurements a viewing angle correction is also applied to Lw(9, Aip,X) in agreement with Eq. ( 5) to deter mine LW(X).The values of Chi a required for such a cor rection were estimated using a regional band-ratio algorithm (Berthon and Zibordi, 2004).
Two TRIOS systems were deployed at the AAOT adjacent to the SeaPRISM during the ARC experiment: one owned and handled by the Management Unit of the North Sea M ath ematical Models (identified as TRIOS-B) and the other by Tartu Observatory (identified as TRIOS-E).The two sys tems are equivalent, but measurements have been performed independently and reduced by applying slightly different schemes, corresponding to the standard practices of the two institutions and with some differences in the approach for un certainty estimate.These elements are separately presented in the following subsections.
Data for inter-comparisons have been constructed by lin early interpolating quality checked products at the reference center-wavelengths.

TRIOS-B
£ d(0+ ,A.), Lí(9, Aip, X) and Li(9', Aip, X) are simultane ously acquired for 10 min taking measurements every 10 s.Calibrated data are quality checked for incomplete and for in dividual measurements differing by more than 25 % from the neighbouring ones.In the case of ARC data, quality check ing led to the rejection of 1 % of measurements.The NIR similarity correction is then performed using M = 780 nm, Xï = 8 7 0 nm, and a = 1.91 (Ruddick et al., 2006).
Estimated uncertainties of Rrs(^) for TRIOS-B approxi mately vaty between 4 and 6 % in the spectral range of in terest (see Table 7).The considered uncertainty sources are (i) uncertainty of system calibration determined assuming the same irradiance standard is utilized for the absolute cal- ibration of the E¿, Lí, and sensors, and thus only ac counting for effects of mechanical setup, inadequate baf fling and reference plaque uncertainties (see Hooker et al., 2002b); (ii) uncertainty due to straylight effects quantified through the application of laboratory characterizations per formed for RAMSES E¿, Lí and sensors (Ansko, un published): (iii) polarization effects quantified as the max imum sensitivity to polarization determined through labo ratory characterizations for RAMSES Lí and sensors (Ruddick, unpublished): (iv) effects of non-cosine response of the above-water E¿ collector determined from labora tory measurements (Ruddick, unpublished): (v) uncertainty in sky light correction quantified in agreement with Ruddick et al. (2006) as a function of the uncertainty in p'(W); and (vi) uncertainty in the correction for off-nadir viewing angle quantified as 25 % of the applied corrections, and exhibit ing different values than those proposed for SeaPRISM be cause of the diverse viewing geometry generally relying on Aip = 135° instead of Aip = 90°.
It is noted that the uncertainty for the sky glint correction is highly dependent on sea state, and the relative percent value of this uncertainty is inversely proportional to Rrs(0, Aip, X) (see web appendix 2 of Ruddick et al., 2006).The values given here have, therefore, been calculated very specifically accounting for the sea state recorded during the ARC activ ities and the observed water-leaving radiances (see Sect. 4).Measurements performed in different waters or sea state con ditions may lead to different uncertainties.

TRIOS-E
The Li(9, Aip,X), Li(9', Aip, X), and iid(0+ ,^) measure ment sequences are simultaneously recorded every 10 sec onds for approximately 6 min, commonly using Aip = 135°.The NIR similarity correction is performed with M = 7 2 0 nm, X2 = 7 8 0 nm and a = 2.35 (Ruddick et al., 2006).The rationale for choosing this wavelength pair, different from that applied for TRIOS-B, is the higher signal to noise ratio characterizing measurements at the shorter wave lengths.Quality checks rely on the mode of R rs( 555) for each mea surement sequence.Data deviating by more than 10% from the mode value are rejected; actually none of the clear sky data included in the ARC inter-comparison was discarded.

Ocean
Estimated uncertainties of Rrs{X) from TRIOS-E vary ap proximately within 4-6 % (see Table 8).The considered un certainty sources are (i) uncertainty of system calibration, computed as for TRIOS-B; (ii) uncertainty due to straylight effects, computed as for TRIOS-B; (iii) polarization effects, computed as for TRIOS-B; (iv) uncertainty in the turbid w a ter NIR similarity correction quantified accounting for 25 % of the applied corrections; (v) uncertainty in the correction for off-nadir viewing angle (also estimated as 25 % of the applied corrections); (vi) effects of non-cosine response of the Ed collector guessed from published data (Zibordi and Bulgarelli, 2007); and (vii) environmental perturbations esti mated from the variation coefficient of Rrs{X) from the same measurement sequence.

Data analysis and results
The inter-comparison analysis has been performed using matchups (i.e.pairs of data products from different sys tems) constructed by setting ± 1 5 m in maximum difference between measurements from the two systems/methods to be compared.Matchup analysis has been performed through the average of relative and of absolute values of percent differ ences.Specifically, the average of relative percent differences (RD) is computed as: while the average of absolute values of percent differences (AD) is given by: where N is the number of matchups, n is the matchup index, superscript C indicates the quantity to be compared, and su perscript R indicates the reference.While RD is applied as an index to measure biases, AD is applied to quantify scattering between compared values.The root mean square of differences (RMS), is also included in the analysis as a statistical index to quan tify differences in absolute units.Data products from WiSPER are applied as the refer ence.This choice is only supported by the confidence ac quired with the system and the related measurement method.WiSPER data for ARC inter-comparisons comprise measure ments from 36 independent casts performed under clear sky conditions from 21 to 24 July 2010.Derived LW{X) spec tra are given in Fig. 1.The shape of spectra suggests a w a ter type characterized by moderate concentrations of phyto plankton and colored dissolved organic matter, as shown by the decrease of spectra from 555 nm toward 412 nm, and ad ditionally, moderate concentration of total suspended mat ter, as shown by non-negligible values at 665 nm.An evalu ation of the water type made in agreement with Loisel and Morel (1998) indicates the presence of Case 2 water dur ing the whole field experiment.Values for relevant quanti ties describing measurement conditions are reported in Ta ble 9. Specifically, measurements performed on water sam ples collected during ARC activities at the AAOT indicate average Chi a values of 0.9 ± 0.3 p g l-1 , concentrations of total suspended matter (TSM) of 1.8 ± 0.4 g I-1 , and absorp tion coefficient by colored dissolved organic matter ay at 412nm of 0.17 ± 0.03 m-1 .However, despite the relative constancy of near surface quantities, the analysis of a{X) and c{X) profile data collected simultaneously to WiSPER mea surements with an AC-9 showed occasionally marked optical stratifications at depths comprised between 5 and 13 m.The exclusion from data processing of the measurements related to these depths has minimized potential inconsistencies in the inter-comparison of products likely affected by the non linear decay with depth of log-transformed L u(j,X ,io) and Ed(z,XJo) data.
By recalling that the objective of the inter-comparison is the evaluation of the overall performance of different sys tems/methods regularly applied for satellite ocean color vali dation activities, and not a detailed investigation of any indi vidual method, a summary of inter-comparison results is pre sented through scatter plots in Figs.2-4 for LW(X), iid(0+ , X) and Rrs(X), respectively.The different number of matchups included in the analysis for the various systems/methods is explained by practical deployment issues for various systems on some days, such as the application of the ± 1 5 min thresh old not always being reached because of inadequate synchro nization of the start of measurement sequences, or like in the case of SeaPRISM data, justified by the automatic and fully asynchronous (when compared to ARC activities) ex ecution of measurements.It is however reported that most of the TRIOS-B and TRIOS-E measurements used to con struct matchups are within ± 1 min from WiSPER measure ments, while most of TACCS-S and TACCS-P measurements are within ± 3 min.
The inter-comparison results of £d(0+ , X), shown in Fig. 3, also exhibit quite good results when considering the variety of instruments and also methods applied.In particu lar, RMSs are close to 5 mW cm -2 pm -1 for the above-water systems/methods and between 8 and 10 mW cm -2 pm -1 for TACCS-S and TACCS-P, respectively.The different perfor mances of TRIOS and TACCS systems are explained by the diverse deployment methods: TRIOS £d(0+ , X) mea surements benefit from a fixed deployment platform while TACCS measurements are affected by the buoy motion adding geometric perturbations as a function of sea state.The RMS value determined for SeaPRISM is comparable to that obtained for TRIOS.This result acquires particular relevance when considering that SeaPRISM £d(0+ , A) data are deter mined theoretically from experimental values of ra(A), a very different approach from actual measurements applied for all other systems/methods.Values of RD for £d(0+ , A) are ap proximately within ± 3 % while values of AD are close to 3 % for the above-water systems (e.g.SeaPRISM, TRIOS-B and TRIOS-E), but reach 5 -7 % for the buoy-based sys tems/methods (i.e.TACCS-S and TACCS-P).Similarly, R2 vary between 0.87 and 0.92 for the above-water systems, and exhibit much lower values for TACCS-S and TACCS-P (i.e.R2 equal to 0.81 and 0.65, respectively).
The inter-comparison shown in Fig. 4 for Rrs{X) data exhibits results obviously depending on those obtained for LW(A) and £ ,d(0+ ,A) data.Specifically, lower RMS values (i.e.0.0002 sr-1 ) are shown for TRIOS-B and TRIOS-E, and the highest (i.e.0.0004 sr-M for TACCS-P.RD values vary from -1 to + 6 %, while AD values are approximately 6 % for the above-water systems and reach 9 % for the buoybased systems.All R2 vary between 0.95 and 0.99 with the lowest values again displayed by the TACCS-S and TACCS-P Rrs(A) as a result of the lower R2 shown by iid(0+ .A).
The former analysis efficiently summarizes the general performances of the various systems/methods, but limits the possibility of evaluating the spectral performances at the se lected center-wavelengths.The inter-comparison analysis is then completed with the presentation of spectral statistical indices for each system/method in Tables 10-12 for LW(A), £d(0+ ,A) and R rs{A), respectively.These data show various peculiarities.For instance, R 2 determined from spectral val ues of L W(A) and R rs{A) are much lower than those com puted with spectrally combined data (e.g.note the striking values for SeaPRISM L w at 4 4 3 nm).This is undoubtedly explained by the small range characterizing the spectral val ues of L W(A) due to the low variability of the seawater biooptical properties (see Table 9).W hen looking at Table 10, also relevant are the biases affecting TACCS-S and TACCS-P (i.e.-20% and +21 %, respectively) and also TRIOS-B and TRIOS-E (i.e.+ 1 2 % and + 1 0 % , respectively) at 665 nm.These are likely explained by the difficulty in determining near surface K¿(665) for TACCS and by imperfect sky-glint removal for TRIOS.
The evaluation of £d(0+ , A) data shows the highest values of RMS, RD and AD for TACCS-P, which is likely explained by wave perturbations.Statistical results for R rs{A) reflect those already presented for LW{X) and £'(](0+ , X), mainly in dicating significant biases at 665 nm for most of the consid ered methods/systems.An investigation of reasons for the observed differences is, however, beyond the scope of the work and likely out of the capabilities offered by the rela tively small ARC data set tied to specific measurement con ditions.

Discussion
Results for the ARC inter-comparison illustrate the best that can be achieved with the considered systems/methods under almost ideal measurement circumstances driven by favourable deployment capabilities as offered by the stability of the AAOT platform (i.e.making £d(0+ , A.) measurements unaffected by tilt, when performed from the main superstruc ture), almost ideal environmental conditions characterized by relatively low sun zenith angles, clear sky and moder ately low sea state, and finally inter-calibration of measure ment systems.By solely considering this latter element, it is recalled that the inter-calibration removes potential biases in derived radiometric products generated by out-of-date or inaccurate calibrations.The comparison of absolute coeffi cients obtained at the JRC during the inter-calibration with those previously applied for the various systems included in ARC has shown minimum differences of 1-2 % but also values exceeding 4 % for individual radiometers.These sec ond relatively high differences, if not removed, would signifi cantly degrade the inter-comparison for one of the considered systems/methods.Processing of data from in-water systems/methods re quires values of a(X) and c (A.).Differently, processing of data from above-water systems/methods requires values W and Chi a.The impact of uncertainties of these input quanti ties is accounted for in the Rrs(X) uncertainty budget for each system/method.It is however of interest to evaluate the im pact of important quantities such as Chi a utilized to correct for the off-nadir viewing geometry of Lw(0, A <p,X).In the present exercise Chi a was determined for all systems using a regional algorithm (see Berthon and Zibordi, 2004) applied to Rrs (A) ratios.The average and the standard deviation of values computed for ARC measurements are 1.9 ± 0.2 pg 1" 1.The corresponding values for actual concentrations deter mined from water samples through High Performance Tiquid Chromatography (HPTC) are 0.9 ± 0.3 p g l-1 .The analysis of TRIOS-B data indicates that the different Chi a estimates give viewing angle corrections differing by less than 1 % for A < p = 135° and varying between 1 and 4 % for A < p = 90°.However, the overall effect on Rrs{X) inter-comparisons is well within the assumed uncertainties.In fact, when using measured Chi a instead of the computed values, TRIOS-E, TRIOS-B and SeaPRISM results indicate an increase of 0.5% , 0.9% and 1.2%, respectively, for the spectrally av eraged RD, and no significant change for the other statistical quantities.Differences among spectrally averaged RD for the various systems/methods are explained by the different mea surement sequences included in the inter-comparison com prising diverse viewing geometries.
In order to evaluate the consistency of the overall inter comparison results illustrated in Sect.4, Table 13 displays spectral AD values determined for R rsW at the 443, 555 and 665 nm center-wavelengths for the various systems/methods with respect to WiSPER, and the combined spectral uncer tainties (CU) determined from the statistical composition of uncertainties quantified for WiSPER R rsW and for each other inter-compared system/method.Recognizing that the computed CU values are overesti mated by at least 1 % due to the inter-calibration of the var ious systems, the comparison is a way to evaluate the con sistency of the uncertainty budgets quantified for each sys tem/method.The agreement between AD and CU values adds confidence to the uncertainty values estimated for each system/method.As expected, the largest differences between AD and CU values are observed at 665 nm for a few sys tems/methods (see underlined values in Table 13).By point ing out that the low values of RT S (X) at 665 nm (on the aver age 6 times lower than those observed at 555 nm) might eas ily lead to higher percent differences in the inter-comparison results with respect to shorter wavelengths, the largest AD Spectrally averaged values of the absolute differences are ap proximately 6 % for the above-water systems/methods, and increase to 9 % for the buoy-based systems/methods.The general agreement of this latter spectral Rrs(f) uncertainty index with the combined uncertainties of inter-compared systems/methods is notable.This result undoubtedly con firms the consistency of the evaluated data products and provides confidence in the capability of the considered sys tems/methods to generate radiometric products within the de clared range of uncertainties.However, it must be recalled that all measurements were performed under almost ideal conditions and for a limited range of environmental situa tions.Additionally, all the optical sensors benefitted from a common laboratory radiometric inter-calibration.These ele ments are specific to the ARC activity, and there is no as surance of achieving equivalent results with the considered systems and methods when using fully independent abso lute radiometric calibrations, performing deployments from ships rather than grounded platforms (where applicable), or carrying out measurements during more extreme environ

Fig. 1 .
Fig. 1.L W(X) spectra from WiSPER produced during the ARC ex periment at the AAOT.

Fig. 2 .
Fig. 2. Scatter plots o f L W(X ) from the various systems/methods versus L W(X ) from W iSPER (ALL indicates merged data from all individual inter-comparisons).RMS indicates the spectrally averaged root mean square o f relative differences, while RD and AD in % indicate spectrally averaged values o f relative differences and o f absolute values o f relative differences, respectively.N is the number o f matchups, all obtained assuming a ± 15 min maximum difference between measurements.Diverse colors indicate data at different center-wavelengths.
where p(9, A < fi, 90 , W) is the sea surface reflectance that can be theoretically determined as a function of the

Table 2 .
Summary of ARC systems/methods details and of main input quantities required for data processing (symbols r, a and c indicate the above-water diffuse to direct irradiance ratio, the seawater absorption and beam attenuation coefficients, respectively).

Table 3 .
Uncertainty budget (in percent) for Rrs determined from

Table 4 .
Uncertainty budget (in percent) for Rrs determined from TACCS-S data.

Table 5 .
Uncertainty budget (in percent) for Rrs determined from TACCS-P data.

Table 6 .
Uncertainty budget (in percent) for Rrs determined from

586, 2012 576 G. Zibordi et al.: In situ determination of the remote sensing reflectanceTable 7 .
Uncertainty budget (in percent) for Rrs determined from TRIOS-B data.

Table 8 .
Uncertainty budget (in percent) for Rrs determined from TRIOS-E data.

Table 9 .
Values of major quantities characterizing the measurement conditions during ARC activities at the AAOT.

Table 10 .
Spectral values o f the statistical indices (i.e.RMS, RD, AD and R2) quantifying the inter-comparison results for L W(A) at the 443, 555 and 665 nm center-wavelengths for the various systems/methods with respect to WiSPER.

Table 12 .
As in Table 10 but for Rrs(X).

Table 13 .
Average values o f the absolute of relative percent dif ferences (AD) determined for RrS(A) at the 443, 555 and 665 nm center-wavelengths for the various systems/methods with respect to WiSPER, and combined uncertainties (CU) determined from the statistical composition of uncertainties quantified for RrS(A) derived from WiSPER and from each other inter-compared system/method.methods has been investigated within the framework of a field inter-comparison called Assess m ent o f In Situ Radiometric Capabilities for Coastal Wa ter Remote Sensing Applications (ARC), carried out in the northern Adriatic Sea.Taking advantage of the geometrically favourable deployment conditions offered by the Aequa Alta Oceanographic Tower, measurements were performed under almost ideal environmental conditions (i.e.clear sky, rela tively low sun zeniths and moderately low sea state) with a variety of measurement systems embracing multispectral and hyperspectral optical sensors as well as in-and above water methods.All optical sensors involved in the experi ment were inter-calibrated through absolute calibration per formed with the same standards and methods.Data prod ucts from the various measurement systems/methods were directly compared to those from a single reference sys tem/method.Overall, inter-comparison results indicate an expected better performance for systems/methods relying on stable deployment platforms and thus exhibiting lower uncer tainties in Zid(0-1-, A). Results for Rrs(f) indicate spectrally averaged relative differences generally within -1 and + 6 %.

586, 2012 mental
conditions (e.g.elevated sun zenith angles, high sea state, water column characterized by near-surface gradient of optical properties, partially cloudy sky).This final con sideration further supports the relevance and need for reg ular inter-comparison activities as best practice to compre hensively investigate uncertainties of measurements devoted to the validation of primary satellite ocean color products and mainly those that are going to be included in common repositories (e.g.MERIS Matchup In situ Database (MER MAID) and SeaWiFS Bio-optical Archive and Storage Sys tem (SeaBASS)).