Use of Moon Observations for Characterization of Sentinel-3B Ocean and Land Color Instrument

remote sensing

Article Use of Moon Observations for Characterization of Sentinel-3B Ocean and Land Color Instrument

Maciej Neneman 1,* ,Sébastien Wagner 2, Ludovic Bourg 3, Laurent Blanot 3, Marc Bouvet 1, Stefan Adriaensen 4 and Jens Nieke 1

1 ESA-ESTEC—Keplerlaan 1, PB 299, 2200 AG Noordwijk, The Netherlands; [email protected] (M.B.); [email protected] (J.N.) 2 European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT), 64297 Darmstadt, Germany; [email protected] 3 ACRI-ST, 06901 CEDEX Sophia Antipolis, France; [email protected] (L.B.); [email protected] (L.B.) 4 Flemish Institute for Technological Research (VITO), 2400 Mol, Belgium; [email protected] * Correspondence: [email protected]

Received: 3 July 2020; Accepted: 5 August 2020; Published: 7 August 2020

Abstract: During the commissioning of the Sentinel-3B satellite, a single lunar observation was performed to assess the possible use of the moon for characterization and validation of onboard instruments. The observation was carried out in stable orientation after a roll maneuver, allowing the moon to be imaged by the Earth view of instruments. Data acquired by the Ocean Land Color Instrument (OLCI) allowed inflight verification of stray-light correction (SLC) performed by the Mission Performance Centre (MPC), and assessment of radiometric behavior of instrument in comparison with lunar irradiance models performed in cooperation between European Space Research and Technology Centre (ESTEC) and European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT). This paper describes the results of those activities along with the proposed update of stray-light correction developed with the use of lunar data.

Keywords: moon; radiometry; characterization; OLCI; lunar calibration; stray-light correction; ROLO; GIRO; LIME

1. Introduction The Sentinel-3 (S3) mission is a part of the Copernicus Space Component (CSC)—the European Commission’s Earth Observation program. The main objectives of the S3 mission are measurements of sea surface topography, sea and land surface temperature as well as ocean and land color to support forecasting systems and environmental and climate monitoring. S3 payloads include Ocean and Land Colour Instrument (OLCI), Sea and Land Surface Temperature Radiometer (SLSTR), Synthetic Aperture Radar Altimeter (SRAL) and MicroWave Radiometer (MWR) along with instruments used for orbit determination. OLCI is a visible-near-infrared pushbroom imaging spectrometer based on the design of the ENVISAT MERIS instrument (Medium Resolution Imaging Spectrometer). The instrument has a 68.9-degree ﬁeld of view (1270 km swath) tilted by 12.6 degrees west to avoid sun glint, while minimizing Observation Zenith Angle (OZA) at the edge of swath. OLCI swath is built from ﬁve cameras and images the surface of the Earth with 300 m spatial resolution in 21 spectral bands (with a spectral resolution between 2.5 and 40 nm) described in Table1 and [1].

Remote Sens. 2020, 12, 2543; doi:10.3390/rs12162543 www.mdpi.com/journal/remotesensing Remote Sens. 2020, 12, 2543 2 of 20

Table 1. Ocean Land Color Instrument (OLCI) spectral bands.

Central Wavelength Band Width Band Function (nm) (nm) Oa01 400 15 Aerosol correction, improved water constituent retrieval Oa02 412.5 10 Yellow substance and detrital pigments (turbidity) Chlorophyll absorption maximum, biogeochemistry, Oa03 442.5 10 vegetation Oa04 490 10 High Chlorophyll Oa05 510 10 Chlorophyll, sediment, turbidity, red tide Oa06 560 10 Chlorophyll reference (Chlorophyll minimum) Oa07 620 10 Sediment loading Chlorophyll (2nd Chlorophyll absorption maximum), Oa08 665 10 sediment, yellow substance/vegetation For improved fluorescence retrieval and to better account Oa09 673.75 7.5 for smile together with the bands 665 and 680 nm Oa10 681.25 7.5 Chlorophyll fluorescence peak, red edge Oa11 708.75 10 Chlorophyll fluorescence baseline, red edge transition Oa12 753.75 7.5 O2 absorption/clouds, vegetation Oa13 761.25 2.5 O2 absorption band/aerosol correction Oa14 764.375 3.75 Atmospheric correction Oa15 767.5 2.5 O2A used for cloud top pressure, fluorescence over land Oa16 778.75 15 Atmos. corr./aerosol corr. Atmospheric correction/aerosol correction, clouds, pixel Oa17 865 20 co-registration Water vapor absorption reference band, vegetation Oa18 885 10 monitoring Water vapor absorption/vegetation monitoring Oa19 900 10 (maximum reflectance) Water vapor absorption, Atmospheric correction/aerosol Oa20 940 20 correction Oa21 1020 40 Atmospheric correction/aerosol correction

Currently, there are two Sentinel-3 satellites in orbit: Sentinel-3A (S3A) and Sentinel-3B (S3B), launched in February 2016 and April 2018, respectively. During the S3B commissioning phase (summer 2018) a single lunar observation was performed to study the viability of collected data for calibration of instruments and to verify the possibility of routine moon observations for S3C and S3D satellites [2]: By nature, the moon presents several advantages for postlaunch characterizing and monitoring of the radiometric performances of space-borne instruments aboard spacecraft in the Earth’s orbit. In particular, it is a relatively bright object against a very dark and cold background, which allows, for instance, an assessment and potential adjustments of stray-light corrections as performed by the on-ground image processing chains, e.g., [3]. In addition, the exceptional stability in time of the moon’s surface and the predictability of its surface reflectance makes the moon an ideal radiometric source for the calibration of space-borne sensors in Earth’s orbit [4,5], in the reflective part of the solar spectrum. It supposes the use of a lunar radiometric model able to reproduce the lunar signal consistently under the full range of lunar phases and liberations. Due to the complexity and the spatial non-homogeneity of the moon’s surface, lunar models are often integrating the moon’s signal to derive a total lunar irradiance. The viability of lunar calibration was already proved for many Earth-observing instruments on polar orbiters (e.g., [6–8]) and on geostationary satellites (e.g., [9]), and it is routinely used for monitoring the radiometric stability of sensors acquiring images within the spectral interval (350 nm, 2500 nm) (e.g., [10–13]). However, whereas radiometers aboard geostationary satellites can regularly see the moon in their field of view, polar satellites require a maneuver in the absence of a dedicated space-view port. OLCI-B observed the moon on 27 July 2018 at 05:22 UTC, during the commissioning phase of S3B. The moon was observed in stable orientation after performing a 160◦ roll (relative to nominal pointing). This allowed the moon to be acquired with Camera 4 at nadir (Figure1), similarly to an Earth scene Remote Sens. 2020, 12, x FOR PEER REVIEW 3 of 21

and it is routinely used for monitoring the radiometric stability of sensors acquiring images within the spectral interval (350 nm, 2500 nm) (e.g., [10–13]). However, whereas radiometers aboard geostationary satellites can regularly see the moon in their field of view, polar satellites require a Remotemaneuver Sens. 2020, in12 ,the 2543 absence of a dedicated space‐view port. 3 of 20 OLCI‐B observed the moon on 27 July 2018 at 05:22 UTC, during the commissioning phase of S3B. The moon was observed in stable orientation after performing a 160° roll (relative to nominal viewedpointing). by the This instrument. allowed Thethe moon so-called to be “Moon acquired Maneuver” with Camera was 4 at performed nadir (Figure during 1), similarly the eclipse to a (whenn the satelliteEarth scene is in viewed a shadow by the of instrument. the Earth; The during so‐called “night”). “Moon The Maneuver” Moon Maneuver was performed occupies during the the whole durationeclipse of (when the eclipse. the satellite It needs is in to a shadow accommodate of the Earth; stabilization during “night”). periods The after Moon a roll Maneuver (and roll-back) occupies of the satellite.the whole Execution duration of this of the maneuver eclipse. It does needs not to accommodate inﬂuence the sta continuitybilization ofperiods OLCI after data; a roll the (and instrument roll‐ is back) of the satellite. Execution of this maneuver does not influence the continuity of OLCI data; the not collecting data during the eclipse (unlike thermal infrared and microwave instruments). The moon instrument is not collecting data during the eclipse (unlike thermal infrared and microwave observation was performed during the commissioning phase to prove the viability of this maneuver and to investigate possible beneﬁts for the mission. Encouraging results of activities described below became proof of concept for lunar calibration in the Sentinel-3 mission (planned maneuver of S3A).

FigureFigure 1. Basic 1. Basic Ocean Ocean Land Land Color Color Instrument Instrument (OLCI) geometry. geometry. Field Field of ofview view (FOV) (FOV) is tilted is tilted westward westward to avoidto avoid sun sun glint. glint.

ThisThis paper paper provides provides insight insight into into the the moon moon acquisitionacquisition performed performed by by OLCI OLCI aboard aboard Sentinel Sentinel-3B‐3B during the commissioning phase. At first, it details the stray‐light performance analysis done with during the commissioning phase. At first, it details the stray-light performance analysis done with the the moon image. A potential improvement in stray‐light correction in certain bands is proposed. In moon image. A potential improvement in stray-light correction in certain bands is proposed. In the the second part, the methods that were deployed to infer parameters, such as oversampling factors secondand part, pixels the solid methods angles, that are presented were deployed and documented. to infer parameters, An accurate such estimate as oversampling of those parameters factors is and pixelsrequired solid angles, before are the presented lunar data and can documented.be used to investigate An accurate the radiometric estimate of behavior those parameters of OLCI further. is required beforeFinally, the lunar a preliminary data can analysis be used of to the investigate radiometric the behavior radiometric of the instrument behavior ofis presented OLCI further. using Finally,two a preliminarymodels for analysis simulating of the the radiometric lunar irradiance behavior at the instrument of the instrument level: the is GIRO presented model: using Global two Space models‐ for simulatingBased Inter‐ theCalibration lunar irradiance System (GSICS) at the Implementation instrument level: of the the Robotic GIRO Lunar model: Observatory Global Space-Based(ROLO) Inter-Calibration[10,14], and the System newly (GSICS)developed Implementation Lunar Irradiance of Model the Robotic European Lunar Space Observatory Agency (LIME) (ROLO) [15]. The [10 ,14], and theanalysis newly focuses developed on interband Lunar behavior Irradiance of instrument, Model European effects of Space stray‐ Agencylight correction (LIME) on [15 radiometry,]. The analysis focuses on interband behavior of instrument, effects of stray-light correction on radiometry, and residual stray-light in different stray-light correction (SLC) implementations. It further demonstrates the added value of lunar observations in analyzing the radiometric performance of imaging spectrometers in space, and the potential that regular lunar observations could offer to monitor their performance. Remote Sens. 2020, 12, x FOR PEER REVIEW 4 of 21

and residual stray‐light in different stray‐light correction (SLC) implementations. It further demonstrates the added value of lunar observations in analyzing the radiometric performance of Remote Sens. 2020, 12, 2543 4 of 20 imaging spectrometers in space, and the potential that regular lunar observations could offer to monitor their performance. 2. Materials and Methods 2. Materials and Methods 2.1. OLCI Moon Observation Data 2.1. OLCI Moon Observation Data AfterAfter the rollthe roll of the of the spacecraft spacecraft was was stabilized, stabilized, thethe instrument was was turned turned on onin nominal in nominal Earth Earth ObservationObservation mode. mode. The The instrument instrument was collecting collecting samples samples for for2 minutes 2 min with with the the moon moon coming coming into into view roughlyview roughly in the in middlethe middle of thisof this period period with with a a phase phase angleangle of ‐6.45°.6.45 The. The timeline timeline of the of maneuver the maneuver − ◦ is presentedis presented in Figure in Figure2 . The2. The calibration calibration window window was selected selected to to make make sure sure that that the themoon moon was wasas as closeclose to the to middle the middle of the of the swath swath as as possible possible to to avoid avoid possible negative negative effects eﬀects that that may may occur occur close close to to edgesedges of the of camera the camera ﬁeld field of viewof view (FOV). (FOV). Additionally, Additionally, one of of the the objectives objectives was was to capture to capture all possible all possible stray-lightstray‐light pixels. pixels.

Figure 2. Timeline of OLCI-B. Moon Observation.

Figure 2. Timeline of OLCI‐B Moon Observation. Raw data (in Digital Numbers—DNs) acquired during the observation needs to be corrected and calibratedRaw to physical data (in Digital values—undergo Numbers – DNs) Level-1 acquired processing. during the observation During processing, needs to be raw corrected counts and of the detectorcalibrated are converted to physical to radiancevalues—undergo as follows: Level‐1 processing. During processing, raw counts of the detector are converted to radiance as follows: 1 1 h i Lb,k,m = 1 NL− Xb,k,m Smb,k,m L k,m + SL(L , , ) Cb,k,m SL(L , , ) (1) A b,m,k − ∗ ∗ ∗ ∗ − − ∗ ∗ ∗ 𝐿,, b,k,m 𝑁𝐿,,𝑋,,𝑆𝑚,, 𝐿∗, 𝑆𝐿𝐿∗,∗,∗ 𝐶,,𝑆𝐿𝐿∗,∗,∗ (1) 𝐴,,

where:where: b, k, m subscripts𝑏, 𝑘, 𝑚 subscripts stand for stand the spectral for the band,spectral the band, spatial the pixel, spatial and pixel, the cameraand the identifiers,camera identifiers, respectively. Whenrespectively.is used, it When means ∗ thatis used, all itemsit means within that all the items range within are involved;the range are involved; ∗ 𝐿,, is the scene radiance as seen by the pixel 𝑏, 𝑘, 𝑚; Lb,k,m is the scene radiance as seen by the pixel b, k, m; 𝑋,, is the OLCI raw sample – in the equation, it is corrected for nonlinearity; Xb,k,m is the OLCI raw sample—in the equation, it is corrected for nonlinearity; 𝑁𝐿,, is a nonlinear function representing the nonlinearities within the video chain. It also NLb,k,dependsm is a nonlinear on the VAM function programmable representing gain the𝑔; nonlinearities within the video chain. It also depends on the VAM𝐴,, programmable is the instrument gain radiometricg; gain expressed in counts per radiance unit. It is determined Ab,k,minis flight the instrument through the radiometricperiodic radiometric gain expressed calibration in (performed counts per with radiance solar diffuser); unit. It is determined in flight through𝑆𝐿 is the stray periodic‐light radiometric radiance contribution. calibration It may (performed result from with all solarthe wavelengths diffuser); the detector is SL issensitive the stray-light to, from radiancethe whole contribution.camera FOV (across It may‐track) result as well from as from all the out wavelengthsof the FOV (i.e., the from detector FOV is sensitive to, from the whole camera FOV (across-track) as well as from out of the FOV (i.e., from FOV of other modules, but also from FOV along-track, i.e., from scenes that have been previously observed or that will be further observed, hence the subscript L , , ; ∗ ∗ ∗ Smb,k,m is the smear contribution (light integration during Charge-Coupled Device (CCD) frame transfer);

Cb,k,m is the dark signal of the b, k, m pixel during the observation.

The radiometric model is described in technical guides available at [16]. Remote Sens. 2020, 12, 2543 5 of 20

Since the radiometric model is valid regardless of the observed scene, this calibration approach holds for an OLCI moon observation. However, the processing chain was designed for Earth Observation without plans to perform moon observations routinely. Some steps of Level-1 processing are not applicable to data acquired from the moon scan. An obvious example of such a processing step is the georeferencing of samples. For that reason, a development platform of the Instrument Processing Facility (IPF-D) was used to process the moon data. It was configured to output partial Level-1 products, such as calibrated instrument radiances (before and after stray-light correction). In the case of a wide field of view imaging spectrometer, such as the OLCI, in some observation scenarios (cloud-free ocean pixels close to clouds or land covered by vegetation), stray-light might significantly contribute to radiances. This issue is addressed with stray-light correction performed during Level-1 processing. The moon seen by Sentinel-3 instruments is (in visible bands) a bright circle on a nearly complete black background. This kind of sharp dark bright dark transition can be the → → perfect target to analyze stray-light [17–19]. Moon observation has proven to be a great opportunity to assess stray-light correction on-orbit.

2.2. OLCI Stray-Light and Its Correction According to the OLCI design, the two main contributors to stray-light are the two main optical components, the ground imager and the spectrometer. Stray-light occurring within the ground imager consists of a two-dimensional process, related to the two spatial dimensions, namely the along-track and the across-track directions. There is an important assumption of lack of exchange between wavelengths, as there is no identified process implying energy transfer in the spectral dimension of the incoming radiance field inside the ground imager (GI). Contribution from the spectrometer is also a two-dimensional process with only one spatial dimension—the entrance slit aligned with the across-track dimension, canceling the along-track one and the spectral one introduced by the light dispersion device, the grating. In contrast with the ground imager, the spectrometer stray-light also includes a contribution from the sensor—a CCD—within which scattering of photoelectrons occurs and becomes significant at wavelengths higher than 900 nm. OLCI stray-light correction was designed based on preliminary characterization showing moderate levels of stray-light—a few percent of the total signal. Therefore, it is assumed that the fundamental structure of the signal is preserved (even if it is slightly blurred by the stray-light) both from a spatial and spectral point of view. This allows using a robust correction method based on a second degradation of the signal. It is assumed that the second degradation has roughly the same impact on the (already) degraded signal as the actual sensing had on the original signal. As the system response is known, it is possible to degrade a second time the measured signal and, by subtraction, to estimate the degradation itself and correct for it. This is equivalent to the well-known expansion:

(1 + ε)2 = 1 + 2ε + ε2 1 + 2ε (2) ' Rewriting this for a stray-light degradation (linear) operator, the degraded version of the incoming signal x can be written as the sum of the original signal and the degradation itself: xˆ = x + ex. The second degradation applied to the result gives:

xˆ = (x + ex) + (^x + ex) = x + 2ex +ex (3)

If the degradation operator can be considered as a perturbation (in the physics sense)—that is verifying: energy(ex) << energy(x)—it follows: xˆ x + 2ex asex can be neglected and x can be retrieved by: ≈ x xˆ 2xˆ (4) ≈ − This method is sometimes referred to as the “second degradation method”, described in detail in OLCI Level 1 Algorithm Theoretical Basis Document [20]. Remote Sens. 2020, 12, x FOR PEER REVIEW 6 of 21

𝑥𝑥 2𝑥 (4)

This method is sometimes referred to as the ʺsecond degradation methodʺ, described in detail in OLCI Level 1 Algorithm Theoretical Basis Document [20]. Remote Sens. 2020, 12, 2543 6 of 20 2.2.1. The Moon, a Great Target for Investigation on SL Correction.

2.2.1.The The OLCI Moon,‐B a moon Great Targetobservation for Investigation has been used on SL to Correction assess the performance of the stray‐light correction. TheFor several OLCI-B reasons, moon observation the moon hasis appropriate been used to for assess stray the‐light performance analysis. First, of the there stray-light is in all correction. spectral bandsFor a sharp several transition reasons, between the moon bright is appropriate and dark forzones stray-light (respectively, analysis. moon First, and there deep is space), in all spectral which bandsis very a important sharp transition to analyze between the impact bright andof SL dark correction zones (respectively, close to the bright/dark moon and deep transitions. space), whichSecond, is verythe dark important zone (deep to analyze space) the is impacta homogeneous of SL correction absolute close reference to the brightat zero/dark radiance, transitions. not perturbed Second, theby darkany geophysical zone (deep space)signal. is Third, a homogeneous the (near) absolutefull moon reference image allows at zero selecting radiance, dark/bright not perturbed interfaces by any geophysicalaligned either signal. with Third, the ALT the (near)(along full‐track) moon direction image allowsor with selecting the ACT dark (across/bright‐track) interfaces direction, aligned as eitherillustrated with in the Figure ALT (along-track)3. This is important direction to help or with to separate the ACT the (across-track) impact of GI direction, (ground imager) as illustrated SL from in Figurethe impact3. This of isSP important (spectrometer) to help stray to separate‐light, since the impact in the ofALT GI (grounddirection, imager) only GI SL stray from‐light the impact is active. of SPNeglecting (spectrometer) the curvature stray-light, of the since moon in theover ALT a few direction, lines or onlya few GI columns stray-light allows is active. averaging Neglecting these lines the curvatureor columns of to the reduce moon the over noise. a few Finally, lines or the a fewstray columns‐light correction allows averaging performance these in lines the ALT or columns and ACT to reducedirections the can noise. be Finally,assessed the on stray-light both sides correction of the moon performance disk (above in and the ALTbelow, and left ACT and directions right), which can be is assessedimportant on since both the sides stray of the‐light moon kernels disk (aboveare not and systematically below, left and symmetrical right), which (especially is important some since ghosts the stray-lightwhich are present kernels only are not on systematically one side). symmetrical (especially some ghosts which are present only on oneThese side). conditions, well adapted for stray‐light analysis, are difficult to find on Earth observation scenes.

Figure 3. Synthetic representation of the OLCI-BOLCI‐B moon observation showing the interfaces between bright andand darkdark zone, zone, which which are are aligned aligned with with along-track along‐track (ALT) (ALT) direction direction (blue (blue arrows) arrows) or across-track or across‐ (ACT)track (ACT) direction direction (red arrows). (red arrows).

2.2.2.These Potential conditions, SL Correction well adapted Improvements for stray-light Tested analysis, on the are Moon difficult Data to find on Earth observation scenes.

2.2.2.The Potential appropriateness SL Correction of the Improvements moon scene Testedfor the onstray the‐light Moon investigations Data allowed us to test two potential improvements in the stray‐light correction: The appropriateness of the moon scene for the stray-light investigations allowed us to test two  2 potentialAn iterative improvements method, in a the solution stray-light to the correction: limitation that appeared necessary for Near Infrared (NIR) bands  AnAdditional iterative spectrometer method, a solution SL kernels to the atε 2wavelengthslimitation that above appeared 1000 nm, necessary improving for Near correction Infrared of • (NIR)channel bands Oa21, for which photoelectrons scattering inside the CCD is currently underestimated. Additional spectrometer SL kernels at wavelengths above 1000 nm, improving correction of • Iterativechannel Method Oa21, for which photoelectrons scattering inside the CCD is currently underestimated. As mentioned in Section 2.2, the stray‐light correction relies on the “second degradation Iterative Method method”. This method is expected to work well when the stray‐light is very small compared to the uncorruptedAs mentioned signal. in In Section the NIR 2.2 ,(typically the stray-light spectral correction bands relies Oa19 on to the Oa21), “second the degradationphotoelectron method”. scatter Thisstray method‐light level is expected becomes to workhigh, welland when the “second the stray-light degradation is very smallmethod” compared results to thein a uncorrupted stray‐light signal.overcorrection. In the NIR (typically spectral bands Oa19 to Oa21), the photoelectron scatter stray-light level becomes high, and the “second degradation method” results in a stray-light overcorrection. In theory, this limitation of the “second degradation method” (also referred to as “ε2 limitation”) for high stray-light levels can be improved by an iterative approach. It consists after a ﬁrst estimation of the stray-light corrected radiance in re-estimating the stray-light by applying the SL kernel on this SL-corrected radiance, providing a better stray-light estimation since it is obtained from an image that has less stray-light. This re-estimated stray-light is then removed from the original image. This process Remote Sens. 2020, 12, x FOR PEER REVIEW 7 of 21

In theory, this limitation of the “second degradation method” (also referred to as “ 2 Remotelimitation”) Sens. 2020, 12 ,for 2543 high stray‐light levels can be improved by an iterative approach. It consists after a7 of 20 first estimation of the stray‐light corrected radiance in re‐estimating the stray‐light by applying the SL kernel on this SL‐corrected radiance, providing a better stray‐light estimation since it is obtained can befrom performed an image iteratively, that has less each stray iteration‐light. This delivering re‐estimated a better stray‐ estimationlight is then ofremoved the stray-light from the original that should convergeimage. toward This process the true can stray-light. be performed iteratively, each iteration delivering a better estimation of the Practically,stray‐light that this should “iterative” converge method toward can the be true achieved stray‐light. with no-extra computational cost by using a simple re-formulationPractically, this of “iterative” the kernels method in the can Fourier be achieved transform with no domain.‐extra computational cost by using a simple re‐formulation of the kernels in the Fourier transform domain. Tests on the moon data have shown, as expected, signiﬁcant improvement in the NIR bands. This is Tests on the moon data have shown, as expected, significant improvement in the NIR bands. illustrated for the band Oa20 (Figure4) in which the overcorrection due to the “second degradation method” has been signiﬁcantly reduced, even though a slight overcorrection remains, which may be due to other problems than the ε2 limitation, such as characterization problem or interpolation problem.

FigureFigure 4. Oa20 4. Oa20 moon moon radiance radiance(left (left) without) without stray-light stray‐light correction, correction, (middle (middle) after) after nominal nominal correction, correction, (right()right after) after correction correction using using the the iterative iterative method. method. ColorColor scale scale is is linear, linear, centered centered on onzero zero (deep (deep space space value)value) and limitedand limited to stray-light to stray‐light radiance radiance levels levels (i.e.,(i.e., the moon moon disk disk is issaturated); saturated); red red is for is positive, for positive, whitewhite for close for close to 0, to blue 0, blue for for negative negative radiance. radiance. OnOn the left left plot, plot, a ablack black circle circle has has been been drawn drawn to to visualizevisualize the moon the moon disk disk better. better.

UsageUsage of Additional of Additional Spectrometer Spectrometer Stray-Light Stray‐Light Kernel Kernel AboveAbove 1000 1000 nm nm In theIn NIRthe NIR domain, domain, one one strong strong contributor contributor toto the total total stray stray-light‐light is isthe the scattering scattering of the of the photoelectronsphotoelectrons within within the CCD the CCD (so-called (so‐called “NIR “NIR scatter” scatter” stray-light). stray‐light). This scatteringThis scattering is not is an not electronic an cross-talkelectronic but an cross optical‐talk dibutff usionan optical in the diffusion silicon layerin the of silicon CCD. layer Photons of CCD. with Photons longer wavelengths with longer are wavelengths are weakly absorbed within CCD material and diffuse to neighboring pixels. The NIR weakly absorbed within CCD material and diffuse to neighboring pixels. The NIR scatter stray-light is scatter stray‐light is completely negligible below ≈900 nm but strongly and gradually increases completely negligible below 900 nm but strongly and gradually increases toward longer wavelengths, toward longer wavelengths,≈ reaching a maximum around 1028 nm, followed by a slight decrease reachinguntil a 1040 maximum nm, the around upper limit 1028 of nm, OLCI’s followed spectral by range. a slight The decrease NIR scatter until stray 1040‐light nm, contribution the upper limit is of OLCI’saccounted spectral for range. in the The spectrometer NIR scatter stray stray-light‐light kernels, contribution together with is accountedthe optical contributions for in the spectrometer (scatter stray-lightand ghost). kernels, together with the optical contributions (scatter and ghost). The nominalThe nominal SL correction SL correction uses uses spectrometer spectrometer kernels characterized characterized at atfive five different different wavelengths. wavelengths. The lastThe onelast one is located is located at at 973 973 nm, nm, while the the center center of band of band Oa21 Oa21 is around is around 1020 nm. 1020 Thus, nm. its stray Thus,‐ its stray-lightlight correction correction fully fully relies relies on on the the kernel kernel at at973 973 (there (there is isno no extrapolation extrapolation beyond beyond characterization characterization wavelengths).wavelengths). Consequently, Consequently, the the NIR_SCATTER NIR_SCATTER stay-lightstay‐light is is strongly strongly underestimated underestimated for channel for channel Oa21, resulting in significant residual NIR scatter stray‐light in the corrected radiances. Oa21, resulting in significant residual NIR scatter stray-light in the corrected radiances. To solve this problem, 2 stray‐light kernels at 1028 nm and 1039 nm were added, corresponding Torespectively solve this to problem, the maximum 2 stray-light of the NIR kernels scatter at curve 1028 nmand and to the 1039 upper nm wavelength were added, edge corresponding of band respectivelyOa21. to the maximum of the NIR scatter curve and to the upper wavelength edge of band Oa21. Stray-lightStray‐light correction correction was was tested tested with with these these two two additional kernels kernels on on the the moon moon data, data, and andvery very promisingpromising results results were were obtained, obtained, especially especially when when coupled coupled with with the iterativethe iterative method method (described (described above). As illustratedabove). As in illustrated Figure5: in the Figure green 5: ACT the green pattern ACT present pattern onpresent both on sides both of sides the moonof the moon (left plot, (left plot, left and right of the moon disk) is residual NIR scatter stray-light: it almost completely disappears when using seven kernels and iterative method (right plot).

2.2.3. Possible Improvements to Earth Observation Data In Section 2.2.2, the positive impact on moon data of the SL correction improvements (iterative method and additional kernels) was presented. Even though stray-light correction performance Remote Sens. 2020, 12, 2543 8 of 20 assessment is more diﬃcult (especially quantitatively) on Earth Observation (EO), the new methods on EO scenes were applied and provided evidence of the expected improvements. Figure6 illustrates the improvement brought to Oa20 by the iterative method applied on an Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 21 OLCI-A Earth Observation. It shows that, as for the moon observation, overcorrection next to the ACT bright/dark transition (here water/land) was signiﬁcantly reduced.

FigureFigure 5. Oa21 5. Oa21 moon moon radiance radiance (left (left) without) without stray-light stray‐light correction, (center (center) after) after nominal nominal correction, correction, (right)(right after) after correction correction using using seven seven kernels kernels and and iterative iterative method. Plotted Plotted radiances radiances are are normalized normalized to to maximum and log‐scaled to highlight the stray‐light impact on the deep‐space background around maximumRemote Sens. and 2020 log-scaled, 12, x FOR PEER to highlight REVIEW the stray-light impact on the deep-space background around9 of 21 the moonthe disk. moon It cannot disk. It becannot compared be compared to that to of that Figure of Figure4. 4.

2.2.3. Possible Improvements to Earth Observation Data In Section 2.2.2, the positive impact on moon data of the SL correction improvements (iterative method and additional kernels) was presented. Even though stray‐light correction performance assessment is more difficult (especially quantitatively) on Earth Observation (EO), the new methods on EO scenes were applied and provided evidence of the expected improvements. Figure 6 illustrates the improvement brought to Oa20 by the iterative method applied on an OLCI‐A Earth Observation. It shows that, as for the moon observation, overcorrection next to the ACT bright/dark transition (here water/land) was significantly reduced.

Figure 6. Earth Observation stray-light corrected Oa20 radiance over part of the lake of Michigan where dark/brightFigure (water 6. Earth/land) Observation transition stray occurs‐light (OLCI-A corrected scene, Oa20 27 radiance/07/2017 over 15h35), part (ofupper the lake left of) without Michigan iteration, where dark/bright (water/land) transition occurs (OLCI‐A scene, 27/07/2017 15h35), (upper left) (upper right) with iterations. The red arrows highlight some improvements brought by the iterative without iteration, (upper right) with iterations. The red arrows highlight some improvements method.brought (bottom by) the ACT iterative cross-section method. ( inbottom the region) ACT cross of the‐section image in indicated the region byof the the image rectangle indicated (a few by frames were averagedthe rectangle to reduce (a few the frames noise). were The averaged curves to were reduce normalized the noise). The by theircurves radiance were normalized value at by detector their index 660 locatedradiance in the value water at detector part at aindex significant 660 located distance in the from water the part water at a/ landsignificant transition. distance The from data the without stray-lightwater/land correction transition. has been The added data without in black. stray‐light correction has been added in black.

Figure 7 illustrates the improvement brought to band Oa21 by the two additional kernels when applied on an OLCI‐B Earth observation. It shows that, as for the moon observation, ACT residual NIR scatter stray‐light (thus an SL undercorrection) next to the ACT bright/dark transition (here land/water and cloud/water) was significantly reduced when considering the additional kernels (combined with the iterative method). One can see that the additional kernels at 1028 nm and 1039.75 nm allowed decreasing the residual NIR‐scatter stray‐light. Indeed, one can see in this image that:  The location of the transition between the bright and dark pixels in the land/water interfaces was now closer to the coastline (see blue circle as an example).  The transition between bright and dark pixels in the cloud/water interface became sharper (see orange circle as an example).

Remote Sens. 2020, 12, 2543 9 of 20

Figure7 illustrates the improvement brought to band Oa21 by the two additional kernels when applied on an OLCI-B Earth observation. It shows that, as for the moon observation, ACT residual NIR scatter stray-light (thus an SL undercorrection) next to the ACT bright/dark transition (here land/water and cloud/water) was signiﬁcantly reduced when considering the additional kernels (combined with the iterative method). One can see that the additional kernels at 1028 nm and 1039.75 nm allowed decreasing the residual NIR-scatter stray-light. Indeed, one can see in this image that: The location of the transition between the bright and dark pixels in the land/water interfaces was • now closer to the coastline (see blue circle as an example). The transition between bright and dark pixels in the cloud/water interface became sharper • Remote(see Sens. orange 2020, 12, circle x FOR asPEER an REVIEW example). 10 of 21

FigureFigure 7. 7. EarthEarth Observation Observation stray stray-light‐light corrected corrected Oa21 Oa21 radiance radiance in incamera camera 4, in 4, ina region a region of the of the Gulf Gulf of Saintof Saint-Laurence‐Laurence where where dark/bright dark/bright (sea/land (sea/land and and sea/cloud) sea/cloud) transitions transitions occur. occur. This Thismeasurement measurement was was done on 25 MAR 2020 14h57 with OLCI-B. (top) nominal processing, (bottom) processing with done on 25 MAR 2020 ≈14h57≈ with OLCI‐B. (top) nominal processing, (bottom) processing with iterations, and seven kernels. The green line represents the coastline. The yellow and blue circles give iterations, and seven kernels. The green line represents the coastline. The yellow and blue circles give an example of where the improvement brought by the additive kernels (combined with the iterative an example of where the improvement brought by the additive kernels (combined with the iterative method) happens. The orange and blue circles give examples of where the improvement happened method) happens. The orange and blue circles give examples of where the improvement happened (note, however, that it happened at all ACT bright/dark transition pixels). (note, however, that it happened at all ACT bright/dark transition pixels). 2.3. Oversampling of Moon Data 2.3. Oversampling of Moon Data OLCI has been designed to observe the Earth from a specific orbit, at a specific spatial resolution, withOLCI a specific has been internal designed clock for to observe the acquisitions. the Earth The from distance a specific between orbit, at the a spacecraftspecific spatial and theresolution, moon is withmuch a largerspecific than internal the distance clock for between the acquisitions. the satellite The and distance the Earth, between which causes the spacecraft the lunar and acquisitions the moon to isbe much oversampled larger asthan the satellitethe distance flies at between its nominal the speed. satellite Oversampling and the Earth, (Figure which8) is a causes phenomenon the lunar that acquisitionsoccurs when to the be distance oversampled between as the centerssatellite of flies neighboring at its nominal pixels speed. (ground Oversampling sampling distance—GSD) (Figure 8) is ais phenomenon smaller than thethat pixel occurs instantaneous when the distance field of view between (IFOV). the centers of neighboring pixels (ground samplingOLCI distance—GSD) moon observation is smaller data is than oversampled the pixel instantaneous in flight (along-track—ALT) field of view (IFOV). direction; the moon appears as elongated ellipse (Figure9). Since the instrument is essentially a scanner, the along-track dimension of the image is built by satellite motion; the sampling time must be tuned to “match” the velocity of the satellite. This does not hold when observing the moon in nominal Earth Observation mode—the FOV diverges, while the orbital velocity and integration time (contributing to pixel pitch) remain the same. Since image in the across-track (ACT) direction is formed by the same array of detectors, their angular FOV and pixel separation remain the same, as during Earth Observation—there is no ACT oversampling. This instance of oversampling effect could be mimicked if we used an

Figure 8. Different sampling conditions. Three ground pixels are equally spaced, however, depending on instantaneous field of view (IFOV) detectors, this scene might be over/under sampled.

OLCI moon observation data is oversampled in flight (along‐track—ALT) direction; the moon appears as elongated ellipse (Figure 9). Since the instrument is essentially a scanner, the along‐track dimension of the image is built by satellite motion; the sampling time must be tuned to “match” the velocity of the satellite. This does not hold when observing the moon in nominal Earth Observation mode—the FOV diverges, while the orbital velocity and integration time (contributing to pixel pitch) remain the same. Since image in the across‐track (ACT) direction is formed by the same array of detectors, their angular FOV and pixel separation remain the same, as during Earth Observation— there is no ACT oversampling. This instance of oversampling effect could be mimicked if we used an

Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 21

RemoteFigure Sens. 2020 7. Earth, 12, x ObservationFOR PEER REVIEW stray‐ light corrected Oa21 radiance in camera 4, in a region of the Gulf of11 of 21 Saint‐Laurence where dark/bright (sea/land and sea/cloud) transitions occur. This measurement was office scanner to scan a printed image of the moon, while slowly pulling the paper in the direction of done on 25 MAR 2020 ≈14h57 with OLCI‐B. (top) nominal processing, (bottom) processing with scanners motion. iterations, and seven kernels. The green line represents the coastline. The yellow and blue circles give Anan example accurate of estimate where the of improvement the oversampling brought factor by the is additiveneeded kernelsto both (combined correct the with image the iterativeand analyze the radiancemethod) values.happens. This The estimate orange and can blue be inferred circles give from examples the orbit of andwhere instrument the improvement characteristics. happened Since the spacecraft completes one orbit in 100 minutes, the average angular velocity is 1.047 . (note, however, that it happened at all ACT bright/dark transition pixels). During 44 ms (sampling time), the instrument covers an angular distance of 46 μrad. Compared to the2.3. distanceOversampling between of Moon S3 and Data the Moon, the linear displacement of the satellite during this period is negligible,OLCI hasand been this designedangle can to be observe treated the as Earth the moonfrom a observation specific orbit, angular at a specific sampling spatial step resolution, in ALT Remotedirection.with a Sens. specific 2020When, 12 internal, 2543the instrument clock for looksthe acquisitions. at the Earth, The the distance subsatellite between point the (SSP) spacecraft moves andby a the distance10 moon of 20 ofis 46much μrad larger ⋅ 6380 thankm the 294 distance m. This linearbetween pixel the displacement satellite and seen the fromEarth, 814 which km orbitcauses forms the anglelunar . acquisitions𝛽 to 360 be oversampledμrad. The oversampling as the satellite factor flies is theat its ratio nominal of those speed. two: Oversampling 7.83 (Figure. 8) is oﬃce scanner to scan a printed image of the moon, while slowly pulling the paper in the direction of a phenomenonThis calculation that occurs can whenbe expressed the distance as: between the centers of neighboring pixels (ground scanners motion. sampling distance—GSD) is smaller than the pixel instantaneous field of view (IFOV). 𝜔⋅𝑇 ⋅𝑅 𝑅 𝑂𝑆 7.83 (5) 𝐻 ⋅𝜔⋅𝑇 𝐻

where 𝜔 is the angular velocity of the spacecraft, 𝑇 is the sampling time, 𝑅 is the mean Earth radius, and 𝐻 is the orbit height (above sea level). Most of the terms cancel out, and oversampling becomes a function of orbit height. This calculation of the oversampling ratio is assuming perfect sampling of Earth data form given orbit height. Based on a range of values of the calculated oversampling ratio derived with different parameters, the overall uncertainty is estimated at 2.5% of the calculated value. It is important to notice that the instrument is assumed to sample from its orbit perfectly and that FiguretheFigure moon 8. 8. DiDifferent isﬀerent assumed samplingsampling to be conditions. conditions. sufficiently ThreeThree far groundground to neglect pixelspixels aredisplacementare equallyequally spaced,spaced, of the however,however, instrument depending depending between consecutiveonon instantaneous instantaneous frames. ﬁeld field of of view view (IFOV) (IFOV) detectors, detectors, this this scene scene might might be be over over/under/under sampled. sampled.

FigureFigure 9. 9. Radiance imagette fromfrom OLCIOLCI cameracamera 4,4, bandband Oa1Oa1 moon scan shows the oversampling eeffectﬀect onon moon moon data. data. ImageImage isis 300300 × 300 pixels with ACT direction on X-axis.X‐axis. × AnThe accurate oversampling estimate factor of the is sometimes oversampling estimated factor is using needed an ellipse to both fitting correct algorithm the image to and fit the analyze moon theshape. radiance However, values. this approach This estimate is not can recommended be inferred [18] from even the for orbit high and illumination instrument conditions characteristics. as the Since the spacecraft completes one orbit in 100 min, the average angular velocity is 2π = 1.047 mrad . 6000s s During 44 ms (sampling time), the instrument covers an angular distance of 46 µrad. Compared to the distance between S3 and the Moon, the linear displacement of the satellite during this period is negligible, and this angle can be treated as the moon observation angular sampling step in ALT direction. When the instrument looks at the Earth, the subsatellite point (SSP) moves by a distance of 46 µrad 6380 km = 294 m. This linear pixel displacement seen from 814 km orbit forms angle · = 0.294km = µ β = 360 = β 814km 360 rad. The oversampling factor is the ratio of those two: α 46 7.83. Remote Sens. 2020, 12, 2543 11 of 20

This calculation can be expressed as:

ω Tsampl Rearth R OS = · · = earth = 7.83 (5) f actor H ω T H orbit · · sampl orbit where ω is the angular velocity of the spacecraft, Tsampl is the sampling time, Rearth is the mean Earth radius, and Horbit is the orbit height (above sea level). Most of the terms cancel out, and oversampling becomes a function of orbit height. This calculation of the oversampling ratio is assuming perfect sampling of Earth data form given orbit height. Based on a range of values of the calculated oversampling ratio derived with different parameters, the overall uncertainty is estimated at 2.5% of the calculated value. It is important to notice that the instrument is assumed to sample from its orbit perfectly and that the moon is assumed to be sufficiently far to neglect displacement of the instrument between consecutive frames. The oversampling factor is sometimes estimated using an ellipse fitting algorithm to fit the moon shape. However, this approach is not recommended [18] even for high illumination conditions as the fitting procedure parameters (e.g., the threshold on the quantization), as well as potential artifacts in the image, such as stray-light or ghosts, may lead to erroneous estimates of the oversampling factor. During this activity, ellipse fitting was only used to locate the moon and to automatically find the region of interest on the lunar scan and delimit the area of irradiance integration to the moon disc (with an additional fringe to ensure all moon pixels were included).

2.4. Irradiance Calculation To assess the inflight radiometric performance of an instrument that has acquired lunar imagery, actual observations have to be compared to models. Such models spectrally simulate the moon signal for the specific geometrical conditions of the acquiring instrument at the time of the observation. Community reference models like ROLO) or GIRO estimate the reflectance spectrum of the moon as a function of time and satellite position and infer the lunar irradiance at instrument entry based on the solar irradiance. The observed lunar irradiance is estimated by integrating the lunar radiances recorded by the instrument. Spectral irradiance is the integral of spectral radiance over a given surface. Z Eλ = Lλ,x,y dΩ (6) Ω where Eλ is the spectral irradiance coming from an area delimited by Ω—solid angle of the integrated surface, Lλ,x,y is the spectral radiance coming from point (x, y). In the case of discrete spectral radiances recorded by OLCI, integration is replaced by summation. Summation is performed over a range of points n = 1, 2, 3 ... that contribute to surface Ω. The infinitesimal solid–angular increment dΩ is replaced by ωn—IFOV solid angle related to sample n.

1 X E = L ω (7) λ OS λ,n n corr n ·

A spectral radiance product of IPF-D was used in this study. It provided radiance values for each pixel in each scan of each band, corrected for eﬀects normally inﬂuencing readouts during any observation (e.g., smear, nonlinearity, and stray-light). Values were corrected for oversampling (OScorr) and appropriate IFOV values were assigned to each CCD pixel (ωn). Remote Sens. 2020, 12, 2543 12 of 20

The IFOV solid angle of a pixel can be estimated based on the ground pixel size. If we assume nadir pixel is 272 m ACT and 294 m ALT, and spacecraft is on 810 km orbit, the IFOV can be estimated as surface area divided by distance square:

PixACT PixALT 7 ω = · = 1.22 10− sterad (8) nadir R2 · Valid for the central pixel of the CCD. A more accurate value can be calculated based on the detector pixel pitch and focal length. If we take focal length f = 67.3 mm and pixel pitch ν = 0.0225mm = v (from instrument speciﬁcation), we can calculate FOV in both directions: ACT—ax 2 atan 2 f , · v ALT—ay = atan f . 7 This results in a solid angle: ω = ax ay = 1.12 10 sterad nadir · · − ACT IFOV of pixel n (counting from the middle of CCD) can be calculated as ACT IFOV of n v v pixels minus ACT IFOV of n-1 pixels: ax = atan n atan (n 1) . ALT IFOV was limited by · f − − · f spectrometer slit and was assumed to be constant—simulated diﬀerences were lower than 0.2%. The most accurate value of detector IFOVs was based on ground tests and simulations performed duringRemote instrument Sens. 2020, 12 assembly, x FOR PEER and REVIEW acceptance (Figure 10). The uncertainty of the IFOV solid angle was13 of 21 estimated based on test requirements at 1.4%.

Figure 10. Variation of pixel IFOV along CCD. In observation performed in June 2018, the moon was imagedFigure by pixels10. Variation between of red pixel lines IFOV marked along on CCD. the chart.In observation performed in June 2018, the moon was imaged by pixels between red lines marked on the chart. After accounting for oversampling and solid angle, the scene irradiance can be calculated and veriﬁed againstAfter accounting a lunar irradiance for oversampling model. It and is worth solid pointing angle, the out scene that irradiance OLCI data can is radiometrically be calculated and calibratedverified by against a ground a lunar data irradiance processor model. (IPF) that It is applies worth calibrationpointing out inferred that OLCI from data an onboardis radiometrically solar diﬀuser,calibrated providing by a ground Level 1b data radiance processor product. (IPF) that At applies instrument calibration level, calculationinferred from of thean onboard observed solar irradiancediffuser, reduces providing to a summationLevel 1b radiance of radiances product. inside At the instrument moon disc level, (multiplied calculation by respective of the observed solid angles)irradiance and correcting reduces to for a summation oversampling. of radiances With this inside approach, the moon any disc discrepancy (multiplied between by respective modeled solid irradianceangles) andand calculatedcorrecting instrumentfor oversampling. irradiance With can this indicate approach, imperfection any discrepancy in processing between chain modeled or instrumentirradiance radiometry and calculated (within instrument the uncertainty irradiance range ofcan modeled indicate irradiance). imperfection in processing chain or instrument radiometry (within the uncertainty range of modeled irradiance).

2.5. The GIRO Model The GIRO model [10] has been developed by EUMETSAT in collaboration with United States Geological Survey (USGS). It has been endorsed by the Lunar Calibration Community, which encompasses the GSICS Research Working Group and the Comitee on Earth Observation Satellites (CEOS) Working Group on Calibration & Validation (WGCV) Infrared and Visible Optical Sensors Subgroup (IVOS), as the common lunar calibration reference model [17,18]. It is an implementation of the ROLO model [5] where the lunar disk reflectance, within the interval [350 nm, 2500 nm], is described by the following equation: ln A ag bΦ c θ cϕcΦθ cΦϕ de de ()(9) gp d cos p

where 𝑘 denotes the wavelength band, 𝑔 is the absolute phase angle, 𝜃, 𝜙 are the selenographic latitude and longitude of the observer (spacecraft), and 𝛷 is the selenographic longitude of the sun. Parameters 𝑎, 𝑏, 𝑐, 𝑑,𝑝 were retrieved from a fitting procedure making use of the extensive dataset acquired by the ROLO telescopes over more than eight years. The GIRO implemented the coefficients available in [5]. Reflectance is converted to spectral irradiance as follows:

Remote Sens. 2020, 12, 2543 13 of 20

2.5. The GIRO Model The GIRO model [10] has been developed by EUMETSAT in collaboration with United States Geological Survey (USGS). It has been endorsed by the Lunar Calibration Community, which encompasses the GSICS Research Working Group and the Comitee on Earth Observation Satellites (CEOS) Working Group on Calibration & Validation (WGCV) Infrared and Visible Optical Sensors Subgroup (IVOS), as the common lunar calibration reference model [17,18]. It is an implementation of the ROLO model [5] where the lunar disk reﬂectance, within the interval [350 nm, 2500 nm], is described by the following equation:

X3 X3 g g ! i 2j 1 g p3 lnA = a g + b Φ + c θ + c φ + c Φθ + c Φφ + d e− p1 + d e− p2 + d cos − (9) k ik jk − 1 2 3 4 1k 2k 3k p i=0 j=1 4 where k denotes the wavelength band, g is the absolute phase angle, θ, φ are the selenographic latitude and longitude of the observer (spacecraft), and Φ is the selenographic longitude of the sun. Parameters a, b, c, d, p were retrieved from a fitting procedure making use of the extensive dataset acquired by the ROLO telescopes over more than eight years. The GIRO implemented the coefficients available in [5]. Reflectance is converted to spectral irradiance as follows:

ES E = A Ω k (10) λ k · · π where Ω is the solid angle of the moon, ESk denotes solar spectral irradiance in band k (corresponding to wavelength λ). The GIRO was validated against the ROLO on an extensive set of data acquired by the SEVIRI instrument aboard Meteosat-8, 9, and 10. Additional comparisons were successfully made with instruments such as MODIS aboard Terra and Aqua or Suomi NPP VIIRS for the First Joint GSICS/IVOS Lunar Calibration Workshop to consolidate this validation [17]. The GIRO provides the lunar irradiance for the reﬂective solar bands of the processed radiometer at the instrument level. The range of phases covered by the model was |2, 92| degrees, as for the ROLO. The current uncertainties of the GIRO model for absolute calibration were between 5% and 10%, following the assessment made for the ROLO model [21]. However, for stability monitoring, the estimated uncertainty achieved less than 0.5% and even closer to 0.1% for long time series with consistent high illumination conditions [7,21].

2.6. Lunar Irradiance Model ESA—LIME In a recent project led by the National Physical Laboratory (NPL) in collaboration with the University of Valladolid and Flemish Institute for Technological Research (VITO) (funded by ESA), new lunar irradiance model—LIME (Lunar Irradiance Model ESA)—was developed. LIME aimed at deriving an improved lunar irradiance model with sub-2% absolute radiometric uncertainty, SI traceable, based on new lunar observations carried out with a lunar photometer operated from the ground at the Pico Teide in Tenerife (Spain). The instrument was calibrated and characterized at NPL and operated from March 2018 to June 2019, providing around 150 useful nights of lunar observations that allowed deriving the ﬁrst version of LIME. Additionally, wavelength dependency was introduced for several coeﬃcients. The uncertainty analysis of the model output in the lunar photometer spectral bands led to an uncertainty of about 2% (k=2). Initial comparison to the GIRO model and a time series of EO satellite measurements (PROBA-V and Pléïades) was performed. A detailed description of the model and results of the comparison are available at 15. The photometer is still in operation. The model will be updated on a yearly basis based on an extended set of measurements. Remote Sens. 2020, 12, 2543 14 of 20

In the current implementation of LIME, it provides lunar reﬂectance and irradiance at normalized distances. To compare instrument irradiance with LIME, it should be normalized to LIME geometry by following: 2 2 DS M DV M E = f E0 , where f = − − (11) λ d λ d 1 AU 384400 km where Eλ denotes spectral irradiance in LIME observation geometry, Eλ0 denotes spectral irradiance in satellite observation geometry, DS M denotes the distance between the sun and the moon, DV M − − denote the distance between the vehicle and the moon. This normalization is performed externally to reduce the number of regression parameters in the model.

3. Results and Discussion This section will refer to three datasets obtained in diﬀerent ways. The three datasets are:

OLCI new SLC—dataset obtained with the improved stray-light correction described in Section 2.2. • It is important to notice that radiometric gains in this variant were obtained using ﬁve kernel (old) implementation of SLC. The update is currently being implemented for gains calculation. The diﬀerence will be limited to band Oa21 and will result in approximately +2%–+3% higher gains. This change might improve results in Oa21. OLCI old SLC—dataset obtained with the current operational implementation of SLC. • OLCI no SLC—dataset obtained as breakpoint mid-product before SLC correction. This is, however, • still corrected for nonlinearity, dark offset, smear, and radiometrically scaled.

The detailed description of level 1 processing is provided in [22]. Instrument requirements specify radiometric uncertainty at 2% for wavelengths below 900 nm, 5% for longer wavelengths. An additional contributor to uncertainty was the uncertainty of the oversampling factor (estimated at 2.5%) and uncertainty of solid angle (IFOV) of detectors (speciﬁed at 1.4%). Overall, the uncertainty of measured irradiance was estimated at 3.5% for bands below 900 nm (Oa01-Oa18), 5.8% for bands above 900 nm (Oa19-Oa21). LIME model delivers irradiance normalized to reference distances along with scaling factor that can be used to convert irradiance between observation distance and normalized distance. For the GIRO model, this conversion is already performed on output irradiance (it is provided as seen from the position of the satellite). For consistency, LIME values were scaled to satellite position by dividing them by fd from Equation (10).

3.1. Deep Space Oﬀset Since the moon was considered the only source of illumination of the acquired scene, we assumed the entire signal recorded by the instrument was coming from the moon—radiance contribution from stars was negligible. Therefore, additional veriﬁcation of stray-light correction could be made based on the distribution of the signal on the full imagette. Ideally, the deep space signal (noise) should average to zero over any region of interest, and the sum of radiances of the whole imagette should be equal to the sum of the radiances inside the moon disc. However, it was noticed that the mean deep space level varies between bands (Figure 11). While the relative level of deep space noise was very low, it is important to notice that there were around 4200 pixels inside the moon disk and 1.6 M pixels outside. To avoid bias, the inband average value of deep space pixels was subtracted from every pixel.

3.2. Spectral Behaviour Radiances were limited to those inside the Moon disc, which was defined by an ellipse fitting algorithm and converted to mask array. Values in this mask array were one for every pixel inside ellipse-fitted moon disc; zero for every pixel outside. Then, this array was morphologically diluted with 5 5 kernel filled with ones, with three iterations. This caused the elliptical mask to grow in size, ensuring × Remote Sens. 2020, 12, x FOR PEER REVIEW 15 of 21

The detailed description of level 1 processing is provided in [22]. Instrument requirements specify radiometric uncertainty at 2% for wavelengths below 900 nm, 5% for longer wavelengths. An additional contributor to uncertainty was the uncertainty of the oversampling factor (estimated at 2.5%) and uncertainty of solid angle (IFOV) of detectors (specified at 1.4%). Overall, the uncertainty of measured irradiance was estimated at 3.5% for bands below 900 nm (Oa01‐Oa18), 5.8% for bands above 900 nm (Oa19‐Oa21). LIME model delivers irradiance normalized to reference distances along with scaling factor that can be used to convert irradiance between observation distance and normalized distance. For the GIRO model, this conversion is already performed on output irradiance (it is provided as seen from the position of the satellite). For consistency, LIME values were scaled to satellite position by dividing

them by 𝑓 from Equation (10).

3.1. Deep Space Offset

RemoteSince Sens. 2020 the, 12moon, 2543 was considered the only source of illumination of the acquired scene,15 ofwe 20 assumed the entire signal recorded by the instrument was coming from the moon—radiance contribution from stars was negligible. Therefore, additional verification of stray‐light correction allcould moon be pixelsmade werebased included on the distribution in calculations. of the Calculated signal on instrument the full imagette. spectral irradiance Ideally, the was deep compared space tosignal LIME (noise) and GIRO should models average (Figure to zero 12) to over investigate any region the overallof interest, radiometric and the behavior sum of ofradiances the instrument of the further.whole imagette The models should used be theequal instrument to the sum mean of the spectral radiances response inside functions the moon (SRF) disc. of However, each channel it was as providednoticed that in [ the22] insteadmean deep of the space SRF level corresponding varies between to the bands pixel viewing (Figure the11). moon center on Camera 4.

Remote Sens. 2020, 12, x FOR PEER REVIEW 16 of 21

with 5×5 kernel filled with ones, with three iterations. This caused the elliptical mask to grow in size, ensuring all moon pixels were included in calculations. Calculated instrument spectral irradiance was compared to LIME and GIRO models (Figure 12) to investigate the overall radiometric behavior of the instrument further. The models used the instrument mean spectral response functions (SRF) of each channel as provided in [22] instead of the SRF corresponding to the pixel viewing the moon center on CameraFigure 4. 11. Mean deep space levels relative to mean level inside moon disc. Figure 11. Mean deep space levels relative to mean level inside moon disc.

While the relative level of deep space noise was very low, it is important to notice that there were around 4200 pixels inside the moon disk and 1.6 M pixels outside. To avoid bias, the inband average value of deep space pixels was subtracted from every pixel.

Figure 12. Comparison of absolute irradiance at satellite level with Lunar Irradiance Model ESA (LIME)Figure and12. Comparison GSICS Implementation of absolute of irradiance Robotic Lunar at satellite Observatory level (GIRO)with Lunar models. Irradiance Dash–dot Model envelopes ESA indicate(LIME) expectedand GSICS uncertainty Implementation ranges of of the Robotic model. Lunar Observatory (GIRO) models. Dash–dot envelopes indicate expected uncertainty ranges of the model.

Both models predicted OLCI irradiance within their respective uncertainty ranges. There was a slight wavelength dependence between two models for short waves, maybe resulting from the fact that LIME introduced a number of wavelength dependencies to certain parameters. This interband behavior is better illustrated in Figure 13, where results were normalized to band Oa17. This way, all curves intersected at 865 nm with a relative value of one—bias between results was removed. Results showed a difference of about 5% between the models and the observations for bands Oa6 and Oa7 that were beyond the uncertainties on the oversampling and the solid angle calculation. It is also important to note that the change in curvature in band Oa20 (penultimate point—940 nm) visible in the models was absent from instrument data. Band Oa15 (767.5mn) also showed a different behavior between the models, which had a drop in irradiance, and the observed irradiance that had a much smoother trend. This behavior is not explained for now. To consolidate those results or to assess the temporal behavior of the instrument, more observations would be required.

Remote Sens. 2020, 12, 2543 16 of 20

Both models predicted OLCI irradiance within their respective uncertainty ranges. There was a slight wavelength dependence between two models for short waves, maybe resulting from the fact that LIME introduced a number of wavelength dependencies to certain parameters. This interband behavior is better illustrated in Figure 13, where results were normalized to band Oa17. This way, all curves intersected at 865 nm with a relative value of one—bias between results was removed. Results showed a diﬀerence of about 5% between the models and the observations for bands Oa6 and Oa7 that were beyond the uncertainties on the oversampling and the solid angle calculation. It is also important to note that the change in curvature in band Oa20 (penultimate point—940 nm) visible in the models was absent from instrument data. Band Oa15 (767.5mn) also showed a diﬀerent behavior between the models, which had a drop in irradiance, and the observed irradiance that had a much smoother trend. This behavior is not explained for now. To consolidate those results or to assess the temporal behavior Remote Sens. 2020, 12, x FOR PEER REVIEW 17 of 21 of the instrument, more observations would be required.

Figure 13. Irradiance normalized to band Oa17 (865 nm). Figure 13. Irradiance normalized to band Oa17 (865nm). 3.3. Effect of Residual Stray-Light on Radiometric Accuracy 3.3. EffectThe relativeof Residual bias Stray between‐Light on models Radiometric and observations Accuracy was calculated for the three mentioned datasetsThe to relative assess thebias influence between of models the stray and light observations correction on was the calculated integrated observedfor the three lunar mentioned irradiance. Thedatasets stray-light to assess model the forinfluence OLCI is of defined the stray as not light energy-conservative. correction on the Strayintegrated light isobserved characterized lunar irradiance.by the use ofThe ray stray tracing‐light for model two cases: for OLCI ideal instrumentis defined as with not all energy photons‐conservative. not following Stray direct light path is characterizedexcluded and by actual the use instrument of ray tracing including for two all cases: photons ideal on instrument the direct with path, all as photons well as scatterednot following and directdiffused path photons. excluded Stray-light and actual is defined instrument as the diincludingfference between all photons those on two the cases. direct This path, means as thatwell the as scatteredsum of irradiances and diffused in all photons. bands (total Stray energy)‐light is before defined and as afterthe difference SLC will bebetween different. those To two achieve cases. energy This meansconservation that the in sum this comparison,of irradiances radiometric in all bands gains (total for energy) imagette before before and SLC after should SLC will be derived be different. from Todata achieve without energy stray-light. conservation in this comparison, radiometric gains for imagette before SLC should be derivedCalculated from irradiances data without were stray also‐light. compared to models after dividing the bias by modeled irradiance: Calculated irradiances were also compared to models after dividing the bias by modeled Irr Irr irradiance: modeled − measured 100% (12) Irrmodeled · 𝐼𝑟𝑟 𝐼𝑟𝑟 This way,the interband behavior of instrument data with⋅ different 100% SLC can be analyzed (Figures 14 and()(12) 15 ). 𝐼𝑟𝑟

This way, the interband behavior of instrument data with different SLC can be analyzed (Figures 14 and 15).

Remote Sens. 2020, 12, x FOR PEER REVIEW 18 of 21

Remote Sens. 2020, 12, 2543 17 of 20

Figure 14. Comparison of OLCI calculated irradiances (with different stray‐light correction (SLC)) relative to the GIRO model.

The results with SLC (new and old) were very similar, with slight differences at higher wavelengths. Irradiance uncorrected for SLC was systematically higher—that is a result of the mentioned lack of energy conservation between those. Figure 14. Comparison of OLCI calculated irradiances (with di erent stray-light correction (SLC)) Figure 14. Comparison of OLCI calculated irradiances (with ﬀdifferent stray‐light correction (SLC)) relative to the GIRO model. relative to the GIRO model.

Figure 15. Comparison of OLCI calculated irradiances (with diﬀerent SLC) relative to the LIME model. Figure 15. Comparison of OLCI calculated irradiances (with different SLC) relative to the LIME model. The results with SLC (new and old) were very similar, with slight diﬀerences at higher wavelengths. Irradiance uncorrected for SLC was systematically higher—that is a result of the mentioned lack of energy conservation between those. Measured irradiances were aligned closer to the LIME model. The majority of them were within the 2% uncertainty range. An overcorrection in Oa21 was likely causing a much lower level of this band. Figure 15. Comparison of OLCI calculated irradiances (with different SLC) relative to the LIME model.

Remote Sens. 2020, 12, x FOR PEER REVIEW 19 of 21

Measured irradiances were aligned closer to the LIME model. The majority of them were within the 2% uncertainty range. An overcorrection in Oa21 was likely causing a much lower level of this

Remoteband. Sens. 2020, 12, 2543 18 of 20 Residual stray‐light can be measured by comparing irradiance inside of the lunar disc to irradiance of complete imagette, so that: Residual stray-light can be measured by comparing irradiance inside of the lunar disc to irradiance 𝐼𝑟𝑟 𝐼𝑟𝑟 of complete imagette, so that: 𝑅𝑒𝑠𝑖𝑑𝑢𝑎𝑙 ⋅ 100% ()(13) 𝐼𝑟𝑟 Irrdisc Irrimg Residual = − 100% (13) For this analysis, the integration disk wasIrr tightenedimg · only to include moon pixels. Stray‐light correction algorithms are optimized for the best performance of ground data processing. The very For this analysis, the integration disk was tightened only to include moon pixels. Stray-light high contrast of the lunar scene causes strong overcorrection in pixels very close to the edge of the correction algorithms are optimized for the best performance of ground data processing. The very high moon. This region (called no‐requirement zone) was excluded by tightening the disc. This way, a contrast of the lunar scene causes strong overcorrection in pixels very close to the edge of the moon. comparison of plot Figure 16 shows the improvement in SLC in longer wavelengths. This region (called no-requirement zone) was excluded by tightening the disc. This way, a comparison of plot Figure 16 shows the improvement in SLC in longer wavelengths.

Figure 16. Residual stray-light in different SLC variants. Strong overcorrection of the longest wavelengths Figure 16. Residual stray‐light in different SLC variants. Strong overcorrection of the longest is replaced by slight undercorrection (negative values). wavelengths is replaced by slight undercorrection (negative values). 4. Conclusions 4. Conclusions While lunar observations were not envisioned as part of Sentinel-3 operations, a single observation was performedWhile lunar by the observations 3-B unit during were commissioning not envisioned to allowas part assessment of Sentinel of the‐3 operations, potential benefits a single fromobservation this kind was of maneuver. performed The by acquiredthe 3‐B unit data during set was commissioning subject to a number to allow of internalassessment studies of the performedpotential at benefits the European from this Space kind Research of maneuver. and Technology The acquired Centre data (ESTEC), set was ACRI, subject and to EUMETSAT a number of thatinternal are presented studies performed here. While at the some European aspects Space of the Research instrument and can Technology be characterized Centre (ESTEC), by a single ACRI, observationand EUMETSAT (stray-light), that are the presented assessment here. of the While radiometric some aspects performances of the instrument and the temporal can be characterized drift of the instrumentby a single would observation require regular (stray observations.‐light), the assessment However, singleof the observations radiometric stillperformances contribute toand the the overalltemporal understanding drift of the of instrument the geometric would and require radiometric regular behavior observations. of the instrument.However, single Assessment observations of stray-lightstill contribute correction to performedthe overall with understanding the moon as targetof the allowedgeometric improvement and radiometric in SLC behavior methodology. of the Thisinstrument. improvement Assessment could be preciselyof stray‐light tested correction on lunar observationperformed datawith thanksthe moon to the as characteristics target allowed of theimprovement scene (simple in SLC image methodology. with high This contrast). improvement Such conditions could be doprecisely not occur tested on on the lunar Earth observation scenes. Improveddata thanks SLC mainlyto the characteristics affects the long-wavelength of the scene (simple channels, image most with notably high significantlycontrast). Such improving conditions the do performancenot occur ofon band the Oa21.Earth Calculatedscenes. Improved instrument SLC irradiance mainly affects was compared the long to‐wavelength lunar irradiance channels, models. most Fornotably most bands, significantly the calculated improving irradiances the performance were within of band a 2% Oa21. uncertainty Calculated range instrument of the LIME irradiance model. was Currently, Sentinel-3A is preparing for a similar Moon Maneuver as for Sentinel-3B, i.e., performed in similar illumination and geometrical conditions (with the nadir Camera 4). Acquired data will provide Remote Sens. 2020, 12, 2543 19 of 20 an independent snapshot of the instrument radiometric status in the middle of its designed lifetime and may give an interesting insight into the aging of the subsystems.

Author Contributions: Conceptualization, M.N., S.W., L.B. (Ludovic Bourg), L.B. (Laurent Blanot), and J.N.; Data curation, M.N., S.W., and S.A.; Formal analysis, J.N.; Funding acquisition, M.N. and J.N.; Methodology, M.N., S.W., L.B. (Ludovic Bourg), L.B. (Laurent Blanot), and J.N.; Project administration, J.N.; Resources, S.W., M.B., and S.A.; Software, M.N. and S.W.; Supervision, J.N.; Validation, L.B. (Ludovic Bourg), M.B., and J.N.; Writing—original draft, M.N., S.W., and L.B. (Ludovic Bourg); Writing—review and editing, L.B. (Ludovic Bourg), L.B. (Laurent Blanot), M.B., and J.N. All authors have read and agreed to the published version of the manuscript. Funding: The essential part of the research was carried out under the ESA Young Graduate Trainee Programme (MN). Acknowledgments: The authors acknowledge Nic Mardle, Johannes Frerick, David Sanchez-Cabezudo, Benedikt Guldimann (ESA-ESTEC), and rest of ESTEC Sentinel-3 team for continuous support, Montserrat Pinol Sole (ESA-ESTEC) for the definition of observation geometry, Maciej Grochowicz and Volkan Salma (ESA-ESTEC) for ground processing of instrument data and the LIME team for providing LIME data (Africa Barreto (Observatory Izaña), Alberto Berjón (Universidad Valladolid), Claire Greenwell (National Physical Laboratory of UK), Sarah Taylor (National Physical Laboratory of UK), Carlos Toledano (Universidad Valladolid), and Emma Woolliams (National Physical Laboratory of UK)). Conflicts of Interest: The authors declare no conflict of interest.

References

1. Nieke, J.; Mavrocordatos, C.; Donlon, C.; Berruti, B.; Garnier, T.; Riti, J.B.; Delclaud, Y. Ocean and Land Color Imager on Sentinel-3. In Optical Payloads for Space Missions; John Wiley & Sons, Ltd.: Hoboken, NJ, USA, 2015; Chapter 10; pp. 223–245. [CrossRef] 2. Nieke, J.; Dransfeld, S.; Donlon, C.; Berruti, B.; Mecklenburg, S. SENTINEL-3 A and B optical payload: Early results from commissioning and tandem flight activities. In Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), Valencia, Spain, 22–27 July 2018; Institute of Electrical and Electronics Engineers Inc.: Piscataway, NJ, USA, 2018; Volume 2018, pp. 1098–1099. [CrossRef] 3. Ando, S.; Tanaka, K. In-orbit stray-light characterization and correction of optical radiometer using the moon image. Mech. Eng. J. 2019, 6, 19-00098. [CrossRef] 4. Kieffer, H.; Wildey, R. Establishing the Moon as a Spectral Radiance Standard. J. Atmos. Ocean. Technol. 1996, 13, 360–375. [CrossRef] 5. Kieffer, H.H.; Stone, T.C. The Spectral Irradiance of the Moon. Astron. J. 2005, 129, 2887. Available online: http://iopscience.iop.org/1538-3881/129/6/2887/pdf/204408.web.pdf (accessed on 26 June 2020). [CrossRef] 6. Godden, G.; McKay, C. A Strategy for Observing the Moon to Achieve Precise Radiometric Stability Monitoring. Can. J. Remote Sens. 1997, 23, 333–341. [CrossRef] 7. Barnes, R.A.; Eplee, R.E.; Patt, F.S.; Kieffer, H.H.; Stone, T.C.; Meister, G.; Butler, J.J.; McClain, C.R. Comparison of SeaWiFS Measurements of the Moon with the U.S. Geological Survey Lunar Model. Appl. Opt. 2004, 43, 5838–5854. [CrossRef][PubMed] 8. Xiong, X.; Sun, J.; Chiang, K.; Xiong, S.-J.; Barnes, W.L. MODIS on-orbit characterization using the moon. Proc. SPIE 2003, 4881, 299–307. 9. Stone, T.C.; Kieffer, H.H.; Grant, I.F. Potential for Calibration of Geostationary Meteorological Satellite Imagers using the Moon. Proc. SPIE 2005, 5882.[CrossRef] 10. Viticchiè, B.; Wagner, C.S.; Hewison, T.J.; Stone, T.C.; Nain, J.; Gutierrez, R.; Müller, J.; Hanson, C. Lunar Calibration of MSG/SEVIRI Solar Channels. In Proceedings of the AMS—EUMETSAT User Conference, Vienna, Austria, 16–20 September 2013. 11. Oh, T.H.; Kim, D. COMS Visible Channel Calibration Using Moon Observation Data. Remote Sens. 2018, 10, 726. [CrossRef] 12. Sun, J.; Xiong, X.; Barnes, W.; Guenther, B. MODIS Reflective Solar Bands On-orbit Lunar Calibration. IEEE Trans. Geosci. Remote Sens. 2007, 45, 2383–2393. [CrossRef] 13. Choi, T.; Shao, X.; Cao, C. On-orbit radiometric calibration of Suomi NPP VIIRS reflective solar bands using the Moon and solar diffuser. Appl. Opt. 2018, 57, 9533–9542. [CrossRef][PubMed] 14. EUMETSAT Website. Moon as Reference for Long Term Stability Assessment of Optical Sensors. Available online: https://www.eumetsat.int/website/home/Data/ScienceActivities/ScienceStudies/ Moonasreferenceforlongtermstabilityassessmentofopticalsensors/index.html (accessed on 26 June 2020). Remote Sens. 2020, 12, 2543 20 of 20

15. CEOS Cal/Val Portal, Lunar Irradiance Model ESA: LIME. Available online: http://calvalportal.ceos.org/lime (accessed on 26 June 2020). 16. Sentinel online Website. OLCI Technical Guide. Available online: https://sentinel.esa.int/web/sentinel/ technical-guides/sentinel-3-olci (accessed on 26 June 2020). 17. Wagner, S.T.; Hewison, T.; Stone, T.; Lacherade, S.; Fougnie, B.; Xiong, X. A summary of the joint GSICS—CEOS/IVOS lunar calibration workshop: Moving towards intercalibration using the Moon as a transfer target. SPIE Remote Sens. 2015, 9639.[CrossRef] 18. GSICS Wiki. Second Joint GSICS/IVOS Lunar Calibration Workshop in Xi’an (China), 13–17 November 2017. Available online: http://129.2.124.227/bin/view/Development/20171106 (accessed on 26 June 2020). 19. Summary of the Second Joint GSICS/IVOS Lunar Calibration Workshop, 13–17 November 2017, Xi’an, China. Available online: http://gsics.atmos.umd.edu/pub/Development/20171106/Summary%20-%20Second% 20Lunar%20Calibration%20Workshop%20-%20Xian%202017.pdf (accessed on 26 June 2020). 20. Bourg, L. “OLCI Level 0, Level 1b Algorithm Theoretical Basis Document”, S3-ACR-TN-007 Issue 5.0. 10 December 2014. Available online: https://sentinels.copernicus.eu/web/sentinel/user-guides/sentinel-3- olci/document-library/-/asset_publisher/hkf7sg9Ny1d5/content/sentinel-3-olci-level-0-and-level-1b-atbd (accessed on 26 June 2020). 21. Stone, T.C.; Kieffer, H.H. Assessment of uncertainty in ROLO lunar irradiance for on-orbit calibration. In Proceedings of the SPIE 5542, Earth Observing Systems IX, Denver, CO, USA, 2–6 August 2004. [CrossRef] 22. Sentinel online Website. OLCI Instrument Specifications. Available online: https://sentinel.esa.int/web/ sentinel/technical-guides/sentinel-3-olci/olci-instrument/specifications (accessed on 26 June 2020).

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).