Ocean color atmospheric correction methods in view of usability for different optical water types

Ocean colo a mosphe ic co ec ion me hods in iew o usabili y o di e en op ical wa e ypes Ma in Hie onymi 1 *, Shun Bi 1 , Dagma Mülle 2 , Eike M. Schü 1,3 , Daniel Beh 1 , Ca s en B ockmann 2 , Ca ole Leb e on 2 , F anc¸ois S einme z 4 , Ke s in S elze 2 and Quin en Vanhellemon 5 1 Depa men o Op ical Oceanog aphy, Ins i u e o Ca bon Cycles, Helmhol z-Zen um He eon, Gees hach , Ge many, 2 B ockmann Consul GmbH, Hambu g, Ge many, 3 Ea h Obse a ion and Modelling, Depa men o Geog aphy, Kiel Uni e si y, Kiel, Ge many, 4 HYGEOS, Lille, F ance, 5 Royal Belgian Ins i u e o Na u al Sciences, Ope a ional Di ec o a e Na u al En i onmen s, B ussels, Belgium Sa elli e emo e sensing allows la ge-scale global obse a ions o aqua ic ecosys ems and ma e fluxes om he sou ce h ough i e s and lakes o coas s, ma ginal seas in o he open ocean. Fuzzy logic classifica ion o op ical wa e ypes (OWT) is inc easingly used o op imally de e mine wa e p ope ies and enable seamless ansi ions be ween wa e ypes. Howe e , e ec i e exploi a ion o his me hod equi es a success ul a mosphe ic co ec ion (AC) o e he en i e spec al ange, i.e., he ups eam AC is sui able o each wa e ype and always deli e s classifiable emo e-sensing eflec ances. In his s udy, we compa e fi e di e en AC me hods o Sen inel-3/OLCI ocean colo image y, namely IPF, C2RCC, A4O, POLYMER, and ACOLITE-DSF (all in he 2022 cu en e sion). We e alua e hei esul s, i.e., emo e-sensing eflec ance, in e ms o spa ial exploi abili y, indi idual flagging, spec al plausibili y compa ed o in si u da a, and OWT classifiabili y wi h ou di e en classifica ion schemes. Especially he esul s o A4O show ha i is beneficial i he pe o mance spec um o he a mosphe ic co ec ion is ailo ed o an OWT sys em and ice e sa. The s udy gi es hin s on how o imp o e AC pe o mance, e.g., wi h espec o homogenei y and flagging, bu also how an OWT classifica ion sys em should be designed o global deploymen . KEYWORDS a mosphe ic co ec ion, ocean colo , op ical wa e ypes, sa elli e emo e sensing, essen ial clima e a iable, Sen inel-3/OLCI F on ie s in Ma ine Science on ie sin.o g01 OPEN ACCESS EDITED BY Ja ie A. Concha, Eu opean Space Resea ch Ins i u e (ESRIN), I aly REVIEWED BY Jona han J. She man, Na ional Oceanic and A mosphe ic Adminis a ion (NOAA), Uni ed S a es Ila ia Cazzaniga, Join Resea ch Cen e, I aly Su ya P akash Tiwa i, King Fahd Uni e si y o Pe oleum and Mine als, Saudi A abia *CORRESPONDENCE Ma in Hie onymi [email p o ec ed] RECEIVED 22 Decembe 2022 ACCEPTED 22 June 2023 PUBLISHED 20 July 2023 CITATION Hie onymi M, Bi S, Mülle D, Schü EM, Beh D, B ockmann C, Leb e on C, S einme z F, S elze K and Vanhellemon Q (2023) Ocean colo a mosphe ic co ec ion me hods in iew o usabili y o di e en op ical wa e ypes. F on . Ma . Sci. 10:1129876. doi: 10.3389/ ma s.2023.1129876 COPYRIGHT © 2023 Hie onymi, Bi, Mülle , Schü , Beh , B ockmann, Leb e on, S einme z, S elze and Vanhellemon . This is an open-access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion License (CC BY). The use, dis ibu ion o ep oduc ion in o he o ums is pe mi ed, p o ided he o iginal au ho (s) and he copy igh owne (s) a e c edi ed and ha he o iginal publica ion in his jou nal is ci ed, in acco dance wi h accep ed academic p ac ice. No use, dis ibu ion o ep oduc ion is pe mi ed which does no comply wi h hese e ms. TYPE O iginal Resea ch PUBLISHED 20 July 2023 DOI 10.3389/ ma s.2023.1129876 1 In oduc ion Ocean Colo (OC) has been iden ified as an Essen ial Clima e Va iable (ECV), because o i s capabili y o obse e a ious aspec s o he ma ine en i onmen synop ically a global scales (GCOS, 2011;Hollmann e al., 2013). The colo o he ocean is de e mined by abso p ion and sca e ing in e ac ions o sunligh wi h wa e , ee-floa ing pa icles and dissol ed subs ances in he uppe wa e laye (cu en s a e o esea ch on his is summa ized by Bi e al., 2023). Colo , o mo e specifically he emo e-sensing eflec ance, R s ,isdefined as he spec al (back-sca e ed) wa e -lea ing adiance, L w , in p opo ion o he o al down-welling plane i adiance, E d . The e e ence poin lies di ec ly abo e he sea su ace a he bo om-o -a mosphe e (BOA). The spec al ange o R s includes no only he isible (VIS) ange, which is pe cei ed as colo o en defined o wa eleng hs om 380 o 760 nm, bu also pa s o he ul a iole (UV) and nea -in a ed (NIR) spec al ange; i is p ima ily de e mined by he pu e wa e abso p ion (e.g., Bi e al., 2023). Space-bo ne ocean colo senso s, howe e , measu e spec al adiances, L TOA , a he op-o -a mosphe e (TOA) om he gi en iewing di ec ion. This signal is s ongly influenced by ligh in e ac ions in he a mosphe e, like sca e ing by ai molecules, and ae osols o abso p ion by a mosphe ic gases, bu also by ligh eflec ions a he sea su ace (e.g., IOCCG, 2010;F ouin e al., 2019). Mo eo e , whi ecaps and ai bubbles in wa e , no ela ed o he ac ual ocean colo , con ibu e o he wa e -lea ing signal (e.g., Die ssen, 2019). The p ocess o e ie ing unobs uc ed emo e- sensing eflec ance a su ace le el om TOA adiance is ypically e e ed o as a mosphe ic co ec ion (AC). Spec al emo e-sensing eflec ance is he undamen al pa ame e om which biogeo-op ical p ope ies and co esponding concen a ions o op ically ac i e wa e cons i uen s can be de i ed. The concen a ion o he pigmen chlo ophyll-a in wa e , Chl,iswidely used as a p oxy o he phy oplank on biomass in he uppe wa e laye ; Chl is also conside ed as an ECV as i is linked o he ma ine ca bon- cycle. The Global Clima e Obse ing Sys em (GCOS, 2011)defines a a ge accu acy equi emen o R s (s ic ly speaking o he wa e - lea ing adiance) o 5% specifically o he blue and g een wa eleng hs and 30% o Chl. This applies o so-called Case-1 (C1) wa e s whose inhe en op ical p ope ies (IOPs) p ima ily depend on phy oplank on, i s abundance and i s deg ada ion p oduc s; his is gene ally he case o open oceans. In con as , all “op ically complex”wa e s o ma ginal seas, coas al and inland wa e bodies a e summa ized as Case-2 (C2) whe e addi ional wa e cons i uen s such as non-algal pa icles (NAP) and colo ed dissol ed o ganic ma e (CDOM) conside ably influence he wa e colo (Mo el and P ieu , 1977;Bi e al., 2023). CDOM is p ima ily leached om decaying de i us and e es ial o ganic ma e , bu i can also be yielded om p ecipi a ion wi h ele a ed CDOM le els in con inen ally influenced ainwa e (Kiebe e al., 2006). The accep ed unce ain ies o R s and subsequen ocean colo p oduc s a e conside ably highe o Case-2 wa e s and GCOS ecommends he implemen a ion o specifically ailo ed algo i hms. Based on his a ionale, EUMETSAT o example o e s wo independen Chl p oduc s (based on di e en AC me hods) om he ope a ional Ocean and Land Colo Ins umen (OLCI) on boa d he Sen inel-3 sa elli es, namely CHL_OC4ME o Case-1 and CHL_NN o Case-2 wa e s. Use consul a ions, howe e , e eal a clea p io i y o ocean colo algo i hms ha wo kac ossC1-C2wa e s,o a leas ha dema ca e he bounda y be ween he wo; mo eo e , app op ia e and s eady ocean colo p oduc s a e equi ed o clima e change s udies (Sa hyend ana h e al., 2017). The usage o b anching and blending o specialized algo i hms o seamless ansi ion and case-op imized phy oplank on es ima es has inc eased o e he cou se o he ecen yea s. Smi h e al. (2018) and Kajiyama e al. (2018) o example ha e de eloped OLCI- specific bipa i e swi ching algo i hms o egionally op imized Chl e ie als. Mo e holis ic app oaches in ol e a p e-classifica ion o R s spec a in o se e al op ical wa e ypes (OWT) in o de o display he ull spec al di e si y o oceanic, coas al, and inland wa e s (e.g., Moo e e al., 2001;Ma in T ayko ski and Sosik, 2003; Van epo e e al., 2012;Shi e al., 2013;Moo e e al., 2014;Melin and Van epo e, 2015;Minu e al., 2016;Ele eld e al., 2017; Hie onymi e al., 2017;Jackson e al., 2017;Spy akos e al., 2018; Soome s e al., 2019;Uudebe g e al., 2020;Jia e al., 2021;Wei e al., 2022). Howe e , e ec i e exploi a ion o his me hod p esumes a success ul a mosphe ic co ec ion o e he en i e spec al ange. Residual e o s om impe ec a mosphe ic co ec ion, which a e no ep oducible by combina ion o mean OWT eflec ance spec a, can esul in e y low o al membe ships and he e o e, p o e he unfi ness o he p ocessing cons ella ion o his case. This leads o he need ha he ups eam AC me hod is wi hin he scope o each wa e ype and ha i deli e s always-su ficien o al membe ships. The e a e a ious senso -specific AC me hods, which supply emo e-sensing eflec ance mos ly op imized o ei he oceanic, coas al o inland wa e s, e.g., desc ibed in IOCCG (2010) o F ouin e al. (2019). The co esponding AC pe o mance can di e significan ly depending on he selec ed e alua ion da a, op ical wa e ypes, applied flagging, senso p ope ies like came a bounda ies, he p esence o anspa en clouds o sun glin (e.g., Goyens e al., 2013;Mülle e al., 2015a;Mülle e al., 2015b;Qin e al., 2017;Tils one e al., 2017;Mog ane e al., 2019). F ouin e al. (2019) lis ed a numbe o significan issues o a mosphe ic co ec ion including clouds, adjacency e ec s, whi ecaps, he Ea h a mosphe e’s cu a u e, mul iple sca e ing, and pola iza ion. Mo eo e , a mosphe ic co ec ions ha e se ious di ficul ies in cases wi h high CDOM o NAP concen a ions in wa e , i.e., e y da k o b igh , so called ex eme Case-2 wa e s (Hie onymi e al., 2016;Hie onymi e al., 2017). Abso p ion o dissol ed o ganic ma e causes an exponen ial educ ion o he eflec ance especially in he blue; his is om a TOA- eflec ance poin o iew, a compa able spec al e ec as Rayleigh sca e ing by ai molecules and hence ambiguous. Abso bing o ex emely abso bing Case-2 wa e s (C2A, C2AX) a e cha ac e ized by low spec al R s wi h maximum in he g een and in cases wi h e y high CDOM-con en (i.e., a CDOM (440) >1 m -1 ) in he yellow, ed, o e en NIR spec al ange. Pa icles in wa e abso b, bu abo e all also sca e ligh , which leads o inc eased eflec ance a highe concen a ions, pa ly also in he NIR. The spec al abso p ion Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g02 and much highe sca e ing o non-algae pa icles also ha e an app oxima ely exponen ial cou se, as does he Rayleigh influence. A ela i ely high NAP concen a ions o 1 g m -3 , one speaks o sca e ing Case-2 wa e s (C2S); a NAP > 100 g m -3 o ex emely sca e ing wa e s (C2SX) espec i ely. Fu he mo e, AC p oblems a ise in he p esence o e y high concen a ions o phy oplank on and floa ing scum wi h non-negligible NIR eflec ance (e.g., Reina and Ku se , 2006). Clea ly, a combina ion o di e en AC algo i hms can po en ially imp o e an all-wa e - ype-emb acing R s - e ie al; examples a e gi en in Shi and Wang (2009);Au in e al. (2013);Bi e al. (2018);Liu e al. (2019), and Sch oede e al. (2022). Howe e , p og amma ic linking o undamen ally di e en AC algo i hms can be challenging and swi ching may lead o spa ial inconsis ency o a e ac s in he e ie als. Se e al AC me hods exis o ocean colo image y o Sen inel-3/ OLCI. Howe e , hei ange o alidi y is no always clea and hey do no always ulfil all equi emen s o unlimi ed usabili y o OWT-based wa e algo i hms like he ONNS algo i hm by Hie onymi e al. (2017). In his s udy, we compa e fi e concep ually di e en a mosphe ic co ec ion me hods o Sen inel-3/OLCI (specified in Table 1): 1) he s anda d (baseline) Le el-2 AC –Ins umen P ocessing Facili y (IPF), 2) he al e na i e Le el-2 AC C2RCC, 3) a no el a mosphe ic co ec ion o di e se op ical wa e ypes (A4O) by Hie onymi e al. (in p ep.), 4) POLYMER by S einme z e al. (2011), and 5) he Da k Spec um Fi ing (DSF) implemen ed in ACOLITE by Vanhellemon and Ruddick (2021). The e a e also o he me hods a ailable ha can be applied o OLCI (e.g., Guan e e al., 2010;Gossn e al., 2019; Sch oede e al., 2022), bu we ocus on hese fi e ACs as ep esen a i e examples o di e se app oaches. Based on op ically di e se Sen inel-3/OLCI images, we compa e he capaci y o da a exploi a ion, he spa ial plausibili y and homogenei y (noise), and analyze he AC ou pu , namely R s , in iew o di e en OWT classifica ion schemes. Mo eo e , we show compa isons wi h in si u ma ch-up da a. We a e he eby a emp ing o dema ca e he scope o applica ion o each AC me hod and iden i y po en ials o u u e imp o emen s. 2 Applied me hods and e alua ion da a 2.1 A mosphe ic co ec ion me hods unde conside a ion 2.1.1 IPF The Eu opean Space Agency (ESA), oge he wi h he Eu opean O ganisa ion o he Exploi a ion o Me eo ological Sa elli es (EUMETSAT), ope a es he Sen inel se ies o sa elli es om he Eu opean Union Cope nicus P og amme. EUMETSAT p o ides Le el-2 (L2) s anda d wa e p oduc s o Sen inel-3/OLCI. Ou wo k e e s o da a o he ocean colo “baseline a mosphe ic co ec ion” om he Ins umen P ocessing Facili y (IPF), which has been ope a ional since 2021 (OLCI Collec ion-3). The eflec ances p o ided a e he basis o he es ima ion o he chlo ophyll-a concen a ion in Case-1 wa e , CHL_OC4ME. The AC was de eloped o he open ocean and is based on wo k o Go don and Wang (1994); u he de elopmen s o his me hod we e summa ized by Go don (2021).Significan u he de elopmen s ega ding MERIS and OLCI a e based on An oine and Mo el (1998), and An oine and Mo el (1999);Moo e e al. (1999), and Nobileau and An oine (2005). Majo upda es o IPF ha e been in oduced in he Sen inel-3/OLCI L2 epo o baseline collec ion (EUMETSAT, 2021); he epo includes se e al compa isons wi h in si u da a and e e ence missions, and lis s he ecommended flags. Pa icula ly no ewo hy is he ecen ly implemen ed e ision o he so-called b igh pixel co ec ion wi hin he AC, which is applied e e ywhe e, bu b ings imp o emen s especially in NAP-domina ed coas al wa e s. 2.1.2 C2RCC The OLCI L2 p ocessing includes a second “al e na i e”AC whose esul s a e no p o ided, bu hey o m he basis o he L2 Case-2 wa e p oduc s like chlo ophyll-a concen a ion, CHL_NN. The AC uses neu al ne wo ks (NN) o he e ie al o R s and also TABLE 1 Examined a mosphe ic co ec ion me hods o Sen inel-3/OLCI ocean colo p ocessing wi h AC-specific masking (plus INVALID and LAND o all). AC Full name and Ve sion O iginal scope Flags o in alid pixel exp ession Addi ional wa ning flags IPF IPF L2-WFR OLCI Collec ion-3 OL_L2M.003.00 C1 CLOUD, CLOUD_AMBIGUOUS, CLOUD_MARGIN, COSMETIC, SATURATED, SUSPECT, HISOLZEN, HIGHGLINT, SNOW_ICE, AC_FAIL, ADJAC, WHITECAPS, RWNEG_[O2-O8] TURBID_ATM, TIDAL, MEGLINT, AC_FAIL, WHITE_SCATT, LOWRW, HIGHRW, ANNOT, RWNEG_[O1, O9-12, O16-18, O21] C2R C2RCC 1.7 including IPF gains C2S, C2A RHOW_OOR, IDEPIX_CLOUD, IDEPIX_CLOUD_BUFFER, IDEPIX_CLOUD_SHADOW, IDEPIX_SNOW_ICE, RTOSA_OOR RTOSA_OOS, CLOUD_RISK A4O A4O 0.23 (2022-01-19) C1, C2S/X, C2A/X CLOUD_RISK, SEA_ICE FLOATING, SUSPECT, GLINT_RISK, ADJACENCY, RTOA_EXCESS POL POLYMER 4.14 (2021-12-17) C1, C2S, C2A CLOUD_BASE, OUT_OF_BOUNDS, EXCEPTION, THICK_AEROSOL, HIGH_AIR_MASS NEGATIVE_BB, EXTERNAL_MASK, CASE2, INCONSISTENCY DSF ACOLITE-DSF 2022-10-25.0 C2S/X NIR_SWIR_THRES, CIRRUS, TOA_THRESH, NEGATIVE, EXTENT The e a e some imes addi ional flags o subsequen wa e algo i hms ha a e no shown he e. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g03 goes back o he MERIS he i age wi h wo ks o Doe e and Schille (2007). The o iginal Case-2 Regional (C2R) algo i hm, which con ains AC and wa e algo i hms, was op imized o coas al wa e s o he No h Sea. The algo i hm was u he de eloped in he Coas Colou p ojec (ESA) and is now known as C2RCC (B ockmann e al., 2016). C2RCC is a ailable in he Sen inel Toolbox (SNAP). The neu al ne wo ks used in he OLCI L2 p ocessing and hose o C2RCC a e iden ical. Howe e , he e a e small di e ences be ween OLCI ope a ional NN p oduc s and ou pu s om he SNAP C2RCC p ocessing due o some di e en p e-p ocessing s eps. In his s udy, he IPF-de i ed SVC gains ( om Collec ion 3) a e used o C2RCC p ocessing di ec ly on OLCI L1B da a, which is done sligh ly di e en in he OLCI L2 g ound segmen NN p ocessing (EUMETSAT, 2021). The applica ion o senso -specific and AC-specific sys em ica ious calib a ion (SVC) gains may ha e he bigges impac also in compa ison wi h p e ious s udies; in some s udies, such as Cazzaniga e al. (2023), he same SVC gains a e applied, in ea lie s udies han 2021, o he SVC gains we e used in some cases (e.g., Giannini e al., 2021). The pixel iden ifica ion ool IdePix was used o cloud de ec ion and co esponding addi ional flagging (B ockmann e al., 2013). Agains usual ecommenda ions o use equal p ocessing le els o ma ch-up analysis, he non-no malized R s p oduc o C2RCC is used, which has a b oade spec al ange in he NIR necessa y o some OWT models. 2.1.3 A4O In he cou se o he las ew yea s, Hie onymi e al. (in p ep.) de eloped a no el a mosphe ic co ec ion o di e se op ical wa e ypes (A4O). The basis was C2RCC, bu wi h undamen al concep ual e ision o op imize classifiabili y wi h he OWT amewo k implemen ed in he OLCI Neu al Ne wo k Swa m (ONNS) wa e algo i hm (Hie onymi e al., 2017). The aim o A4O is o be applicable o all na u al wa e s, om Case-1 o ex emely sca e ing o abso bing Case-2 wa e s. Special a en ion was dedica ed o phy oplank on di e si y. A4O applies an ensemble o di e en neu al ne wo ks and p o ides ully no malized R s .In addi ion, he e a e o he di e ences o C2RCC; hese include he specifica ion o wa e empe a u e and salini y using global clima ological da a, he ea men o ocean whi ecaps, he expansion o ea u es in he NN aining da a, flagging, and an op ion o spec al and spa ial smoo hing o he signal. The IPF- SVC gains a e also aken in o accoun he e p ima ily o compensa e o senso -specific di e ences, i.e., he ins umen s on Sen inel-3A and -3B. The in alid pixel exp ession e e s p ima ily o an own cloud masking, all isible wa e a eas a e alid in p inciple (non- physical nega i e eflec ance is ne e deli e ed). Howe e , he e a e a numbe o wa ning flags, e.g., o pixels wi h possible land influence o s ong sun glin signal, whe e esul s migh be aul y. I is planned o publish A4O and ONNS in SNAP in he medium e m. 2.1.4 POLYMER POLYMER is an AC algo i hm o iginally de eloped o oceanic and coas al wa e s (S einme z e al., 2011;S einme z and Ramon, 2018). I uses a spec al fi ing scheme ha elies on wo models: a polynomial-like model o a mosphe ic eflec ance and a model o wa e eflec ance. I was de eloped p ima ily o co ec ing sun- glin con amina ion on images o he MERIS senso , and has hen been applied o se e al mul ispec al and hype spec al senso s including OLCI. In addi ion o sun glin co ec ion, i is also obus o ae osol con amina ion and o he a mosphe ic and su ace e ec s such as hin clouds and adjacency e ec s (S einme z and Ramon, 2018;Zhang e al., 2019). POLYMER is he only me hod in his s udy ha does no use he IPF-SVC gains because all bands a e used simul aneously o a mosphe ic co ec ion. Thus, specific gains a e used, gene a ed by a dedica ed spec ally coupled SVC scheme. 2.1.5 ACOLITE-DSF The Da k Spec um Fi ing (DSF) algo i hm as implemen ed in ACOLITE, was o iginally de eloped o aqua ic applica ions o sa elli e da a wi h high spa ial esolu ion in he me e o decame e scale, e.g., he Landsa se ies, Sen inel-2/MSI, Pleiades, and Plane Scope (Vanhellemon and Ruddick, 2018;Vanhellemon , 2019a;Vanhellemon , 2019b;Vanhellemon , 2020). Vanhellemon and Ruddick (2021) adap ed he AC o Sen inel-3/OLCI especially o mapping o suspended pa icula e ma e and chlo ophyll-a concen a ion in u bid coas al wa e s. Thus, he main scope o ACOLITE-DSF is o aqua ic applica ions o inland and coas al wa e s, bu i can also be used o e clea e wa e s and e en land. The gains om IPF-SVC a e also being conside ed he e. 2.2 Re e ence sa elli e and ma ch-up da a 2.2.1 Selec ed scenes o spa ial analysis Ten ull- esolu ion OLCI (Le el-1) scenes we e selec ed o analysis o he spa ial AC pe o mance (pixel size 300 m a nadi , swa h wid h app oxima ely 1270 km). They co e a wide a ie y o op ical wa e ypes, egions, sun ele a ions, and senso - iewing angles ela i e o he sun (Table 2;Appendix Figu e A1). App oxima ely 47% o he obse ed Ea h su ace in he images is co e ed by wa e . O hese wa e a eas, 36% a e flagged o cloud- isk and 9% o sun-glin acco ding o he A4O designa ion. Fo a ep esen a i e analysis o hese scenes, common masks we e used whe e all 5x5 pixels a ound a cen al pixel mus be alid. This is o elimina e possible cloud a e ac s, cloud shadows, sun glin , and land adjacency e ec s as much as possible. The eely isible and in p inciple un es ic ed wa e a eas we e isually checked. Howe e , many o hese wa e pixels a e masked by he indi idual AC me hods; especially IPF masks la ge a eas because i p oduces nega i e R s alues he e. The selec ed ee wa e a eas co e 31.5 million pixels. Inland wa e s accoun o 4%. Abou 0.6% o he pixels show a cha ac e is ic ed edge inc ease o TOA eflec ance caused by floa ing biomass a he sea su ace and a e labelled as FLOATING in A4O. Hie onymi e al. (2016) sugges ed a defini ion o ex emely sca e ing wa e s wi h R s (865) ≥0.005 s -1 ; hus, he co e age o b igh pixels depends on he AC me hod and is up o 4%. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g04 2.2.2 Ma ch-up da a om in si u measu emen s and sa elli e obse a ions 2.2.2.1 AERONET-OC Independen alida ion was ca ied ou o ma ch-ups be ween OLCI image y and AERONET-OC in si u measu emen da a (Zibo di e al., 2009) om 2016 o 2020 dis ibu ed h ough he ESA OC-CCI in si u da abase (Valen e e al., 2022). The da a se was limi ed o OLCI bands (± 2 nm). All R s measu emen s a e no malized ollowing Pa k and Ruddick (2005). The s a ions a e widely dis ibu ed geog aphically, bu o en nea coas s o in inland wa e s (GLO –Glo ia, Black Sea; GDT –Gus a Dalen Towe , Bal ic Sea; HLH –Helsinki Ligh house, Bal ic Sea; LIS –LISCO, Long Island Sound; LUC –Lucinda, Eas Coas o Aus alia; MVC – MVCO, US Eas Coas ; PAL –Palg unden, Lake in Sweden; VEN – Venice, Ad ia ic Sea; WAV –Wa ecis_si e_csi_6, Gul o Mexico). The e o e, he wa e ypes a e e y simila and he da a a e no ep esen a i e o he ull ange o all na u al wa e s. In he cases whe e he en i e spec a a e a ailable, he maximum eflec ance lies a 560 nm in 89% cases o he da a, only 11% ha e he maximum a 490 o 510 nm; he e is no in si u da a included wi h he maximum in blue bands<490 nm o a bands >560 nm. The as majo i y o he da a coun s as Case-2 wa e . Fo band-wise compa isons, howe e , da a om Case-1 wa e s a e also included. Some o he AERONET-OC da a om he Bal ic Sea and he Black Sea ep esen dis inc blooms o cyanobac e ia o coccoli hopho es (e.g., Cazzaniga e al., 2021;Zibo di e al., 2022;Cazzaniga e al., 2023). Howe e , o a compa ison o AC esul s a all 16 (ou o 21) OLCI bands, in si u da a a e o en missing, especially in ed and NIR bands. In gene al, band-shi ing me hods can be used o de i e OLCI spec a om di e en band configu a ions, and he mean pe cen age e ie al e o in he spec al ange be ween 400 and 600 nm is usually less han 5%, bu o ed and NIR bands he unce ain ies a e much la ge (Hie onymi, 2019). Fo his eason, addi ional band shi ing was no used in his wo k, since he main pu pose o he ma ch-up compa ison is o show he spec al plausibili y o he AC esul s. 2.2.2.2 O he in si u da a In o de o be able o udimen a y quan i y he spa ial scenes in he ansi ion om coas al wa e ypes and also o con ex ualize e y u bid wa e s ha a e no co e ed in AERONET-OC, exempla y u he in si u measu emen da a a e conside ed. Fi s ly, eflec ance measu ed by Hie onymi e al. in he No h Sea/Ge man Bigh (OLCI ma ch-up wi h scene #2) wi h a p o ocol desc ibed in Tils one e al. (2020) and no malized wi h Pa k and Ruddick (2005). Secondly, OLCI ma ch-ups wi h he PANTHYR sys em (Vans eenwegen e al., 2019) ha is loca ed in u bid coas al wa e s in Belgium. The da a a e p o ided by Vanhellemon and Ruddick (2021); ACOLITE-DSF was specially designed o hese wa e s and a compa ison wi h he AC candida es (albei in di e en e sions o ACOLITE-DSF, IPF, and C2RCC, bu wi hou A4O) was discussed in hei o iginal pape . App oxima ely hal o he PANTHYR da a a e conside ed as ex emely sca e ing wa e s using he abo e-men ioned defini ion, he o he a e C2S. 2.2.2.3 Ma ch-up p ocedu e The Cal alus sys em (Fom e a e al., 2012) was used o iden i y OLCI image ma ches wi h in si u da a wi hin h ee hou s o he sa elli e o e pass. Al oge he , he e a e 2545 ma ch-ups be ween 2016 and 2020 o he nine AERONET-OC s a ions and 62 o PANTHYR (2019-2020) o OLCI-A & B. Fo some s a ions, he e a e only a ew spec al bands o he compa ison and he ma ch-up numbe a ies o each AC acco ding o he fil e ing o alid da a poin s. Duplica ed-flagged alues a e no used. Mini-scenes o abou 10x10 pixels in size we e selec ed a IPF, C2RCC, and A4O, and 5x5 mac o-pixels we e ex ac ed om hem. In he case o POLYMER and ACOLITE-DSF, he comple e scenes we e p ocessed fi s and he mac o-pixels ex ac ed om hem. ACOLITE-DSF can be a he sensi i e o size o he scene o sub- scene, and i is usually ecommended o use a spa ially limi ed s udy a ea wi h a single ae osol e ie al. Fo la ge scenes, as used he e, he ae osol e ie al is iled and in e pola ed o he ull ex en . Indi idual ile con en s may skew he esul s be ween ile cen e s. TABLE 2 Selec ed es scenes wi h la ge cloud- ee a eas ha co e high op ical di e si y (shown in Appendix Figu e A1). Scene Senso -Da e-UTC Region Special ea u es #1 S3A-20160720-092821 Ba en s Sea High la i udes, bloom o coccoli hopho es #2 S3A-20160720-093421 No h Sea, Wadden Sea Mode a ely o ex emely sca e ing wa e s, idal a eas, in si u da a #3 S3A-20170114-130626 Sou h A lan ic Ocean, Rio de la Pla a es ua y Ex emely sca e ing wa e s, clea oceanic wa e s, sun glin , Sou h A lan ic Anomaly #4 S3A-20170527-015236 Yellow Sea, Eas China Sea, Yang ze, Lake Taihu Ex emely sca e ing wa e s, idal a eas, la ge i e s, abso bing ae osols, sun glin #5 S3A-20170529-092334 Medi e anean Sea La ge a eas wi h clea wa e s, sun glin #6 S3A-20170913-080730 Black Sea, Aegean Sea Clea and abso bing wa e s #7 S3A-20180715-093613 No h Sea, Bal ic Sea In ense bloom o cyanobac e ia pa ly wi h scum #8, #9 S3A-20200601-092517, S3B-20200601-084546 No h Sea, Bal ic Sea In e -compa ison o S3A and S3B wi h di e en obse a ion angles, abso bing wa e s #10 S3B-20200406-093801 No h Sea, Bal ic Sea High OWT di e si y Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g05 The agg ega ion o he 5x5 mac o-pixel ollows mos ly he p ocedu e desc ibed in Mülle e al. (2015a). The alid pixel exp essions o each AC (Table 1) a e applied; all alid pixels a e sc eened o ou lie s pe band using a h eshold o 2.5 s anda d de ia ions. F om he emaining alid pixels hei mean alue, m, and s anda d de ia ion, s,iscalcula edand he numbe o alid obse a ions (excluding he ou lie s) is eco ded. Based on he pe cen age coe ficien o a ia ion, CV,ama ch-upis conside ed in u he analysis, i he spa ial homogenei y is high o he pa icula band and he e o e CV =s/m× 100% < 15%. Second, a leas hal o he pixels in he mac o-pixel mus be alid. These c i e ia a e checked o each da a poin and band independen ly, so ha AC solu ions wi h some noise in a pa o he spec al ange may lose good ma ch-ups he e bu e ain pa o he spec um in o he spec al egions. The numbe o ma ch-ups will he e o e a y pe band, which allows some in e p e a ion in e ms o spa ial noise. To compa e he pe o mance o he AC me hods, we use he ma ch-up s a is ics ecommended by EUMETSAT (2022). Besides he well-known linea eg ession s a is ics wi h he co ela ion coe ficien ( ), we use he oo -mean-squa e-e o (RMSE), median absolu e de ia ion (mdAD), median absolu e pe cen age de ia ion (mdAPD), he spec al angle mappe (SAM), and he Chi-squa ed es (c²). 2.3 Op ical wa e ype amewo ks The classifica ion o na u al wa e s in o op ical wa e ypes se es he pu pose o compa abili y and, in he case o la ge-scale sa elli e image p ocessing, he selec ion and blending o esul s o sui able algo i hms. Basically, cha ac e is ic R s -spec a and hei co a iance a e gi en o define a class. An OWT algo i hm ies o combine class-specific spec a in such a way ha he inpu R s - spec um can be ep oduced, whe eby weigh s a e assigned o he con ibu ing classes. The numbe o defined classes, shape and ampli ude o he mean spec a, as well as he ma hema ical de e mina ion o he class weigh s can a y g ea ly in he di e en app oaches (see Figu e 1). In o de o e alua e esul s o he fi e AC me hods wi h ega d o OWT, ou OWT classifica ion me hods we e selec ed wi h di e en emphases, e.g., ocusing on ma ine o inland wa e s. Fo he selec ion o he OWT me hods, i was necessa y o conside he deg ee o a filia ion o he clus e cen e s. The e o e, me hods based on uzzy logic clus e ing and using he Mahalanobis dis ance and c²-dis ibu ion o calcula e he o al membe ship alues we e chosen (Moo e e al., 2001;Moo e e al., 2014). Fu he mo e, only hype spec al o a leas OLCI band-based OWT me hods we e selec ed, bu no me hods using band a ios o concen a ion h esholds. Fo he selec ion, i was also impo an o ep esen a wide a ie y o spec al o ms ha a e conside ed impo an in he di e en me hods. The e o e, in gene al, o he classifica ion app oaches could be conside ed ha migh p o ide mo e obus esul s o he AC me hods unde conside a ion o ha a e no oo ocusedonei he ma ineo inlandwa e s.TheusedOWT classifica ion me hods a e: 1. J17 (Jackson e al., 2017) is an OWT me hod ha was de eloped in he ame o ESA’s Ocean Colou Clima e A B D C FIGURE 1 Spec al eflec ance o op ical wa e ypes om ou amewo ks by (A) Jackson e al. (2017),(B) Moo e e al. (2014),(C) Hie onymi e al. (2017), and (D) Bi e al. (2019), and Bi e al. (2021). The line deno es he o iginal spec al cen oid o each wa e ype and he shaded ibbon deno es he s anda d de ia ion om espec i e aining da ase s. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g06 Change Ini ia i e (OC-CCI). Millions o pixels om me ged sa elli e da a we e selec ed o clus e ing. 11 spec al ypes o ma ine wa e s we e iden ified, and h ee addi ional “highly- u bid”coas al spec a om Moo e e al. (2014) we e also adop ed. The o iginal publica ion e e ed o he OC-CCI da ase 2 wi h SeaWIFS bands; in 2020, new op ical wa e class se we e defined o he da ase 5 o MERIS- e e enced da a wi h POLYMER ( 4.12) as he a mosphe ic co ec ion (Sa hyend ana h e al., 2021). Thus, he adap ed OWT me hod uses 14 classes and six OLCI bands be ween 412 and 665 nm. 2. M14 (Moo e e al., 2014) uses hype spec al R s be ween 400 and 800 nm ha a e p ima ily ep esen a i e o coas al egions and lakes, whe e he cen oids we e ained based on in si u measu emen s. The app oach dis inguishes se en classes, bu ac ually no blue (oceanic) wa e s. Thei o iginal OWT analysis ac ually e e s o he unde wa e emo e- sensing a io, s , which can be ans e ed abo e-wa e o R s . 3. H17 (Hie onymi e al., 2017) is a mo e holis ic app oach o OWT classifica ion as i aims o co e “mos na u al wa e s”, om he open ocean o ex emely abso bing o sca e ing wa e s. The basis o H17 a e adia i e ans e simula ions wi h Hyd oligh (Mobley, 1994), which is a common app oach wi h he AC me hods C2RCC and A4O. The la e was e en op imized in e ms o OWT classifiabili y wi h H17. The OWT scheme uses 11 OLCI bands om 400 o 865 nm and dis inguishes 13 classes. In o de o a oid conflic wi h possible nega i e eflec ances, he spec a a e ans o med by log 10 (R s +1)and b igh ness-no malized, so ha he classifica ion is based on he shape o he spec um alone. 4. B21 is an ex ended OWT amewo k based on he wo ks o Bi e al. (2019), and Bi e al. (2021), de eloped specifically o inland wa e s. The hype spec al aining da a, which we e esampled o 15 OLCI bands om 400 o 865 nm, we e mos ly measu ed a la ge lakes, ese oi s, and i e s ac oss China. The app oach di e en ia es 17 classes including eu ophic and hype ophic cases wi h high biological p oduc i i y and e en su ace scum. The spec a a e no malized by di iding hem by hei in eg als because, acco ding o hei easoning, he composi ion o inland wa e s a ies g ea ly, which changes he shape o he eflec ance spec um a he han he magni ude. The selec ed OWT amewo ks ha e di e en app oaches o classi ying he spec a. In H17 and B21 he spec a a e no malized (albei in di e en ways) o highligh di e ences in spec al shapes be ween ypes, while in J17 and M14 di e ences in he magni ude o he spec a a e aken in o accoun . The e o e, i is expec ed ha he in e p e a ion o a mosphe ically co ec ed da a will depend in pa on he egion obse ed by he sa elli e, as he di e en wa e s o which hese me hods we e ini ially de eloped a e e y di e en . Fo example, B21 will no be able o ep esen oceanic wa e due o he lack o “blue ypes”, while J17 will ha e di ficul y dis inguishing eu ophic inland wa e s, which a e no o eseen in he ma ine model o POLYMER, on which J17 is based. In addi ion o he selec ed OWT amewo ks, we also use he (OLCI) wa eleng h o he R s maximum as a di ec and in ui i e indica ion o wa e ypes; a simila app oach using he spec ally-weigh ed Appa en Visible Wa eleng h has been shown o be e ec i e o di e en op ical condi ions (Vande meulen e al., 2020). In gene al, he maximum eflec ance in clea seawa e is a sho e wa eleng hs (mo e blue o g een), whe eas in u bid wa e he maximum is shi ed owa ds longe wa eleng hs (mo e g een, b own, and ed). 2.4 E alua ion o he classifiabili y In op ical uzzy logic classifica ion, he class membe ship is calcula ed by he cumula i e c²dis ibu ion wi h ndeg ees o eedom (band numbe ) and he Mahalanobis dis ance be ween he spec um and he OWT cen oid, no malized by he OWT s anda d de ia ion (see calcula ion de ails in Moo e e al., 2001). To assess he classifiabili y o an AC-de i ed spec um, we calcula e he o al membe ship o he OWT classifica ion scheme, u . An ideal classifica ion esul should gi e u close o (o e en sligh ly highe han) one. A lowe u , he classifica ion is pe o ming poo ly wi h a h eshold on o ally non-classifiable defined as u ≤10 -8 . Such cases can occu ei he because o insu ficien ype ep esen a ion in he amewo k o because o e o s o he spec al shape o in ensi y i sel , i.e., unde pe o mance o a mosphe ic co ec ion, unco ec ed influences om adjacency e ec s o bo om eflec ions, e c. (Moo e e al., 2014). Jackson e al. (2017) also men ioned ha u should no be much la ge han one in he ideal classifica ion esul ei he , which indica es o e lap and edundancy be ween ypes. Howe e , in his s udy, we allow u o be g ea e han one, because using amewo ks ac oss di e en wa e a eas will ine i ably induce o e lap be ween ypes. We define fi e le els o classifiabili y as shown in Table 3. A spec um is no classifiable i no OWT can be assigned, whe eas OWT membe ships a e dis ibu ed be ween he classes a he o he ou le els. E alua ion c i e ia ha e been discussed in a ious publica ions, e.g., Melin e al. (2011);Van epo e e al. (2012),o Hie onymi e al. (2017); he chosen le els a e a bi a y, bu wo k easonably well o he e alua ion o he classifica ion. A e all, he pe cen ages o classifiable alues in he di e en wa e ypes as well as in he en i e da a se a e calcula ed. The highe he pe cen age o high o medium alues, he be e he classifiabili y o R s . TABLE 3 Classifica ion le els ela ed o he o al membe ship om all classes. Assignable le els u anges Non-classifiable u ≤10−8≈0 Below- h eshold 0<u <10−4 Low 10−4≤u <0:3 Medium 0:3≤u <0:8 High u ≥0:8 Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g07 3 Resul s 3.1 Spa ial homogenei y and plausibili y o sa elli e da a The a ious a mosphe ic co ec ion me hods p o ide indi idual masks a di e en le els indica ing pe o mance limi s and unce ain ies (Table 1). Flagging is usually a ade-o be ween limi ed alidi y wi h suspec esul s a some spec al bands and s ill use ul esul s in ano he spec al ange. Many ocean colo algo i hms u ilize only one o a ew bands o which he AC esul s can be adequa e. O he in-wa e algo i hms use many bands ac oss he spec um, e.g., p inciple componen analysis o some neu al ne wo ks. Fo OWT applica ions, he whole spec um is impo an . O e co ec ion o an AC mani es s o en in nega i e R s , usually ei he in blue (especially IPF) o NIR bands; in any case, his is no a physically plausible esul and may be an in alid inpu o he in- wa e algo i hm. Looking a he whole spec um, IPF and POLYMER p oduce e y la ge a eas wi h nega i e eflec ances, bo h abou hal o he ee wa e a ea (albei he alues a e o en e y close o ze o). The IPF exp ession o alid pixels equi es posi i e eflec ances a leas in he cen al VIS ange (412-665 nm), which canno be sa isfied o e la ge pa s and is he main eason o >50% in alid masking (Table 4). POLYMER does no ha e his es ic i e flagging, so e e y hing emains alid. Depending on he p ocessing se ings, ACOLITE-DSF does no ou pu nega i e eflec ances, bu i s flagging esul s as NaN in he ou pu files, which is he main con ibu o o he 20% in alid flagging ( hese cases also occu in C2SX wa e s, o which ACOLITE-DSF was designed, e.g., isible in Figu es 2-A5,C5). C2RCC and A4O apply neu al ne wo ks o app oxima e log- ans o med R s di ec ly om R TOA wi hou sub ac ing indi idual con ibu ions om Rayleigh sca e ing o glin . Resul ing nega i e eflec ances a e uled ou , because o he log- ans o ma ion and he alue ange o he NN aining. This is an impo an ad an age wi h ega d o con inuous usabili y o he esul s wi h di e en ypes o wa e and allows R s es ima ion e en o e y small alues close o ze o wi h less noise. The sligh ly mo e sensi i e cloud de ec ion in C2RCC p ocessing wi h IdePix esul s in an addi ional 1% masking o he wa e a eas. Figu e 2 shows ex ac s o sa elli e images (#3, #7, #2, and #1; Table 2;Appendix Figu e A1) o he AC esul s o R s (560) wi h espec i e in alid flagging. Spa ial noise usually ans e s o he ocean colo p oduc s and is hus an indica o o AC pe o mance. In his con ex , he Sou h A lan ic Anomaly (SAA) a ea (Figu e 2A) is special; clea spec al ou lie s o indi idual bands occu he e in isola ed pixels and he peaks a e usually no iceably highe a longe wa eleng hs. Some AC me hods succeed in smoo hing he pixel spec um, he eby educing spa ial discon inui ies. C2RCC p oduces he mos isible noise in his a ea (Figu e 2-A2), which is p obably due o he use o neu al ne wo ks ha a e e y sensi i e o small spec al changes. A4O also uses NNs, bu has significan ly lowe spa ial noise due o a ious p ocessing s eps, including a dedica ed spec al smoo hing o suspec ou lie s and a e aging o he esul s o di e en NNs (Figu e 2-A3). Mo eo e , an op ion is ecommended o A4O ha applies a Gaussian fil e o e 3x3 mac o-pixels, which smoo hs esul s o wa e a eas, a enua es cloud a e ac s, and ea s down came a bounda ies. ACOLITE-DSF, as applied he e, in e pola es a mosphe ic pa ame e s o e a la ge spa ial egion, which e ec i ely educes he AC-induced noise le el. Looking a he spa ial homogenei y c i e ion (CV) a di e en wa eleng hs o homogeneous a eas o 100x100 pixels (Appendix Figu e A1), we see a low and compa able noise le els o he AC- inpu adiance a TOA o Case-1 and -2 wa e s; in he SAA a ea, CV alues a e abou wice as high (Table 4). In Case-1 wa e in he SAA (scene #3, Appendix Figu e A1), we see he bigges di e ences o CV(R s )be ween A4O and C2RCC, wi h A4O ha ing he leas noise o all he me hods. In ano he (p esumably clea e ) Case-1 wa e sea a ea in he Medi e anean Sea (eas o he island Sa dinia, TABLE 4 E alua ion o selec ed spa ial ea u es o 31.5 million ee wa e pixels in en es scenes o he fi e a mosphe ic co ec ion models. Fea u e L TOA IPF C2R A4O POL DSF In alid flagged wa e a ea 51.3 1.0 0 0 20.5 R s (412)< 0 18.5 0 0 0.5 0 R s (865)< 0 40.2 0 0 46.7 0 CV(412) in Case-1 wa e s 0.4 4.3 5.3 0.9 7.8 2.0 CV(560) in Case-1 wa e s 0.8 9.3 4.9 1.0 4.6 3.6 CV(665) in Case-1 wa e s 1.3 55.4 7.0 3.6 27.0 7.3 CV(412) in Case-1 wa e s (SAA) 0.9 9.3 16.0 5.3 17.8 9.7 CV(560) in Case-1 wa e s (SAA) 1.8 14.0 133.6 3.8 8.7 12.4 CV(665) in Case-1 wa e s (SAA) 4.0 129.1 >1000 9.5 63.0 36.1 CV(412) in Case-2 wa e s 0.4 >1000 15.5 2.6 33.0 2.9 CV(560) in Case-2 wa e s 0.6 7.6 18.9 1.5 6.2 3.6 CV(665) in Case-2 wa e s 1.0 24.7 16.5 1.8 11.1 10.2 The coe ficien s o a ia ions o AC-de i ed R s e e o homogeneous subse s o 100x100 pixels; co esponding alues o ini ial TOA adiance a e included o compa ison (see Appendix Figu e A1). All alues ha e he uni [%]. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g08 scene #5, Appendix Figu e A1), he noise o C2RCC is significan ly lowe and compa able o he o he me hods, IPF and POLYMER ha e he highes noise in he ed band a 665 nm abo e he alid- ma ch-up h eshold o 15%. In his e y clea blue wa e , R s (665) becomes e y small and app oaches ze o. In ac , he a iabili y o R s (665) in case o IPF and POLYMER is pu e andom noise, in A4O wa e mass s uc u es a e s ill clea ly isible and de e mine CV (665), and in C2RCC one can see weak noisy s uc u es as well. ACOLITE-DSF,whichisno designed o suchclea wa e , p o ides an R s (665) image wi h much highe alues compa ed o he o he ACs ( ac o 10 highe ). Because ACOLITE-DSF does no pe o m pixel-by-pixel a mosphe ic co ec ion, i shows clea FIGURE 2 Subse s om OLCI images (see Appendix Figu e A1). The op ow shows RGB images o L1 adiance a op-o -a mosphe e (A–D); poin s o spec al compa isons a e ma ked he e (see Figu e 3). The fi e ows below show he esul s o R s (560) o he compa ed AC me hods: IPF (A1-D1), C2RCC (A2-D2), A4O (A3-D3), POLYMER (A4-D4), and ACOLITE-DSF (A5-D5). A eas o AC-specific in alid pixel exp essions a e highligh ed anspa en ly o wi h NaN. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g09 membe ships a e. Only hal o he POLYMER eflec ances can be classified as ha ing weigh s abo e he h eshold, bu o al membe ship emains mos ly low. Insu ficien membe ships a e usually ound in highly sca e ing o p oduc i e wa e s, o when POLYMER p o ides nega i e eflec ances in Case-1 wa e s. A4O ma ches all defined classes, bu has low membe ships o p oduc i e wa e s OWTs 7-8, ha a e masked wi h BLOOM. The eason o low membe ships is likely he pa icula ly high a iance o na u al R s a NIR bands, which is no well cap u ed by he H17 c²- dis ibu ion. Howe e , i is impo an ha he class is iden ified co ec ly, which enables pos -classifica ion adap a ion o op imal wa e algo i hm selec ion. All o he ACs do no deli e such spec al shapes; (w ong) C2RCC can be ela i ely well classified. The majo i y o spec a p o ided by IPF, POLYMER, o ACOLITE- DSF wi h he maximum in he sho wa eleng hs (<560 nm) a e no classifiable wi h H17, b igh pixel spec a o IPF and ACOLITE- DSF, howe e , a e o en well classifiable. This shows ha low eflec ance alues play a majo ole in he log- ans o med classifica ion and ha he associa ed noise-le el o some bands leads o shape a ia ions no expec ed by H17. Me hod B21 dis inguishes mos classes bu has a ocus on inland and coas al wa e s wi h li le ega d o he ocean. In addi ion, he shape is also gi en mo e conside a ion he e, and he allowed a ia ions a e ai ly limi ed. None o he AC me hods succeeds in p o iding comp ehensi e spec a ha can be classified wi h he me hod o B21. Fo C2RCC, ne e heless, hal o he pixels a e classifiable wi h u abo e he h eshold (>10 -4 ). Fo all ACs, a leas 85% o he classifiable cases a e dis ibu ed among he fi s FIGURE 7 Same OWT classifiabili y o AC esul s as in Figu e 6, bu a X-axis wi h he OLCI wa eleng h o he R s maximum and co esponding pe cen age dis ibu ion no ed a he op ( his dis ibu ion is independen o he OWT me hod and he e o e he same o all). Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g16 h ee OWTs; he o he 14 classes a e spa sely used. C2RCC, A4O, and ACOLITE-DSF yield >90% usable spec a o BLOOM-labelled pixels. Again, C2RCC p o ides a highe pe cen age o well- classifiable esul s, bu hese a e no in he in ended classes (OWTs 14-17). A4O p o ides such spec a, he majo i y o which ha e use ul membe ships. Figu e 7T shows sligh ad an ages o he classifiabili y o ACOLITE-DSF spec a wi h he maximum in sho e wa eleng hs. 4 Discussions and ou look 4.1 E alua ion o AC me hods In e -compa ison esul s a e o en a snapsho in ime, as bo h AC and wa e algo i hms unde go con inuous e olu ion. This pape e e s o he mos ecen AC e sions (as o Oc obe 2022) and is au ho ed by some o hei main de elope s. I is clea ha he me hods a e a di e en ma u i y le els and ha some ha e been op imized using obse a ional da a, which is also eflec ed in he e o o unce ain y p oduc s and flagging. A4O by Hie onymi e al. is a u he de elopmen o C2RCC, bu is no ye publicly a ailable and he e is no o ficial e e ence o i as well. IPF is used in ope a ional se ice, bu one mus also app ecia e he con inuous de elopmen s, whe e wi h he OLCI Collec ion-3 (since 2021) imp o emen s ha e been achie ed, e.g., o coas al wa e s (Zibo di e al., 2022). One canno say ha his is a Case-1 ocean colo specific algo i hm anymo e, because he compa isons wi h Case-2 domina ed ma ch-up da a documen good ag eemen o e mos o he spec um (wi h specific p oblems desc ibed he e). Ou compa isons wi h AERONET-OC and o he da a show be e ag eemen s o IPF han p e iously epo ed (especially also wi h ega d o he p e ious IPF e sion Collec ion 2), e.g., Liu e al., 2021; Tils one e al., 2021;Vanhellemon and Ruddick, 2021;Li e al., 2022;o Windle e al., 2022. One influencing ac o is ce ainly he conside a ion o ecommended flags and he use o he same IPF- SVC gains o all AC me hods (excep o POLYMER). Ideally, AC- specific SVC gains should be used, bu hese a e no ye a ailable o C2RCC, A4O, and ACOLITE-DSF; specially fi ed SVC would ha e he po en ial o significan ly imp o e hei esul s. In he men ioned s udies, likewise o he e sions o C2RCC, POLYMER, and ACOLITE-DSF a e used; ne e heless, some simila obse a ions can be confi med, like he p incipal sui abili y o C2RCC and POLYMER o Case-2 wa e s especially o he cen al isible ange. A4O and ACOLITE-DSF ha e pa ly less a o able a ings compa ed o AERONET-OC da a, bu bo h p ocedu es a e cu en ly unde going a g ea e dynamic in hei de elopmen ( hey ha e unde gone se e al upda es in 2022). Fo all ACs, sui able me hods mus be ound in he u u e o be e iden i y ob ious ou lie s in o de o achie e be e spa ial and s a is ical e alua ions. This also includes e en be e cloud iden ifica ion. Conside ing he s ic in alid flagging o IPF, howe e , one po en ially loses conside able amoun s o obse a ional da a, which should be econside ed. Spa ial homogenei y, which has a s ong impac on he numbe o ma ch-ups, should be gi en mo e a en ion in u u e. Fo his pu pose, measu es o homogenize a mosphe ic p ope ies a mac o- pixel le el (A4O & ACOLITE-DSF) as well as he log- ans o ma ion o he R s e ie al o e y small alues (A4O & C2RCC) ha e p o en o be e ficien . In combina ion wi h spec al smoo hing (as in A4O), his is also ad an ageous o la ge a eas a ec ed by he Sou h A lan ic Anomaly. One may a gue ha using a non-s ic pixel-by-pixel a mosphe ic co ec ion limi s he high spa ial esolu ion (o up o 300 m), howe e , ele an a mosphe ic and oceanog aphic ea u es a e usually la ge in a ea and AC- induced noise is a significan sou ce o unce ain y o ocean colo p oduc s. High accu acy o e all magni udes o e ie ed R s is expec ed o e he en i e spec al ange o a ious applica ions. Recen e iews summa ize he equi emen s o ocean colo emo e sensing and especially a mosphe ic co ec ion, e.g., in e ms o de i ing inhe en op ical p ope ies o wa e (We dell e al., 2018), phy oplank on di e si y (B ache e al., 2017), ca bon con en (B ewin e al., 2023), and essen ial biodi e si y a iables (Mulle - Ka ge e al., 2018)–and his goes beyond he OLCI bands, also o u u e hype spec al applica ions. The selec ed AC me hod has o en a significan influence on he de i ed ocean colo p oduc s, e.g., he es ima e o he concen a ion o ca boninwa e o hephy oplank on biomass wi h co esponding p ima y p oduc ion. Juhls e al. (2022) o example compa ed in si u da a wi h OLCI ma ch-up esul s om IPF, C2RCC, and POLYMER and, mo eo e , di e en models o he es ima ion o CDOM abso p ion. This was done in o de o in es iga e fluxes o ela ed dissol ed o ganic ca bon om a la ge i e ac oss he u bid coas al zone in o he clea A c ic Ocean, hus in high la i udes (he e, POLYMER is iden ified as he mos sui able). The s onges op ical e ec o CDOM is isible in he blue bands, whe e, acco ding o ou s udy, C2RCC and A4O ha e sligh ad an ages also in e ms o noise and spec al beha io ; ACOLITE-DSF has no iceable p oblems. In his example, he ac ual pe o mance may be inconsis en along he op ical g adien , especially a sho wa eleng hs; OWT-op imized wa e algo i hms could po en ially con ibu e o educing he unce ain ies (i he classifica ion is success ul). Concen a ions o phy oplank on in he o de o Chl >1mgm -3 a e usually necessa y o hype -spec ally dis inguish special pigmen abso p ion ea u es and he eby phy oplank on di e si y; mo eo e , he cen al isible ange (450 o 650 nm) is pa icula ly impo an o ha (e.g., Xi e al., 2015;Xi e al., 2017;Bi e al., 2023). The in ensi y and spec al shape o he eflec ance in he case o “mode a e”algal blooms a e gene ally well ep oduced by all AC me hods in es iga ed (e.g., Figu e 3C). Resul s om he cu en e sion o A4O, howe e , mos ly show an unde es ima ion (which may also ha e o do wi h influences o he angle no maliza ion ha s ill need o be cla ified). A highe Chl (>10 mg m -3 ), he ed edge abso p ion ea u e becomes impo an in he Chl e ie al (e.g., Gons, 1999;Ruddick e al., 2001). High concen a ions o cyanobac e ia wi h possible scum a he wa e su ace, which is a equen phenomenon in inland wa e s and he Bal ic Sea, a e a pa icula challenge o AC. Spec a om IPF, C2RCC, and POLYMER a e mos ly un us wo hy he e and he esul s a e pa ly no su ficien ly accompanied by wa nings (Figu e 3D). A4O, which has a specific wa ning flag o his, p o ides a plausible spec al Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g17 shape and indica es enhanced R s unce ain ies in co esponding p oduc s (which is also easoned by he usual small-scale he e ogenei y o such blooms). The spec a om A4O can be assigned o he designa ed wa e classes in H17 and B21, bu o en wi h low membe ships. In he shown example (Figu e 3D), he shape o ACOLITE-DSF is also plausible excep o he fi s wo bands ha a e likely o e es ima ed and may be impac ed by smile co ec ion a e ac s ( he spec a a e usually no well-classifiable in H17 o B21). Howe e , he e is a possible ad an age o he da k-spec um-fi ing app oach in he ange 500-700 nm, which can be help ul o phycocyanin ea u e de ec ion (a ma ke o cyanobac e ia). The o he example wi h a bloom o coccoli hopho es (Figu e 3G) shows compa able spec al shapes deli e ed by all ACs, bu also clea di e ences in he b igh ness o he e ie ed eflec ance (al hough all esul s a e o a ealis ic o de o magni ude, e.g., Cazzaniga e al., 2021). Me hods o emo ely sense pa icula e ino ganic ca bon ocus on op ical de ec ion o coccoli hopho es, e.g., wi h a colo index made om a ios o g een, ed, and NIR bands (Mi chell e al., 2017;B ewin e al., 2023); he e significan di e ences would occu depending on he AC used. Rega ding he exploi a ion o ed and NIR bands in ocean-wa e algo i hms (also impo an o he es ima ion o he fluo escence line heigh ), he spa ial homogenei y and nega i e eflec ances a e imp o able o IPF and POLYMER, and he emo al o a e ac s om small-scale a mosphe ic a iabili y o ACOLITE-DSF. 4.2 Discussion on OWT amewo ks The dis inc ion o op ical wa e ypes is impo an o many aspec s o ma ine biology, physical oceanog aphy, unde wa e isibili y, e c., and he defini ion o specific p ope ies has a long adi ion (e.g., Je lo , 1976). Cu en esea ch aims o de e mine eliable wa e quali y cha ac e is ics om sa elli e da a o he en i e aqua ic con inuum o land-coas -ocean. Howe e , a balance be ween e o and benefi mus be ound he e and ca e mus be aken in sa elli e images o ensu e no unwan ed discon inui ies a ise. The e may be specific challenges o oceanog aphic o limnological ques ions, e.g., wi h ega d o wa e cons i uen s, sun glin , whi ecaps, shallow wa e , o adjacency e ec s, bu om an op ical emo e sensing poin o iew, i does no make much sense o educe onesel o one applica ion. This common disconnec ion ac ually hinde s eliable s udies on ma e ans e om land o he sea, which is impo an o he ca bon cycle, o example. The lack o classes wi h cha ac e is ic op ical ea u es is a p oblem o all OWT me hods ha we e examined, e.g., classes ep esen a i e o oligo ophic ocean, e y high NAP concen a ions, o hype -eu ophic wa e s a e o en missing. On he o he hand, he e may be spec al classes ha a e di ficul o explain om an IOP pe spec i e. Especially inland wa e OWT amewo ks a e o en based on clus e ing o la ge in si u da a collec ions, which include po en ial measu emen e o s such as adjacency e ec s, bo om eflec ions o inadequa e sky-glin co ec ion. Consequen ly, classes wi h ques ionable mean eflec ances can also be defined. Some OWT amewo ks a e p ima ily used o e alua e he quali y o R s spec a (e.g., Wei e al., 2016). An independen con ol is he Quali y Wa e Index Polynomial (QWIP) me hod o Die ssen e al. (2022). The QWIP sco e o hype spec al da a should no exceed 0.2, o mul ispec al da a as o OLCI he nominal h eshold can be elaxed o 0.3, alues abo e he h eshold should be subjec o addi ional checks. In ac , he QWIP me hod does no include “g een ypes”wi h R s maximum in he NIR, such as defined by B21. Howe e , ew classes o B21, e.g., hei OWT 2, ecei e a QWIP sco e close o 0.2 (no e ha some OWT amewo ks like Spy akos e al. (2018) define classes wi h highe sco es ha possibly ail he QWIP quali y con ol). The B21 OWT 2 class-mean spec um has a local minimum a 440 nm (Figu e 1D). Ou OWT analysis shows ha B21-classifiable spec a o IPF, C2RCC, A4O, and POLYMER a e in his OWT 2 wi h less han 1%, whe eas 80% o ACOLITE-DSF spec a all in o his class. The compa isons wi h AERONET-OC indica e an unde es ima ion o he a mosphe ic signal o ACOLITE-DSF in blue bands; u he mo e, he e is eason o conclude ha adjacency e ec s, e.g., om b igh clouds, play a ole (Bulga elli and Zibo di, 2018). Indeed, QWIP can be used di ec ly o quali y con ol o sa elli e- de i ed R s , e.g., Tu ne e al. (2022) compa ed esul s om ACOLITE-DSF and POLYMER (in o he e sions) as well as he s anda d NASA SeaDAS algo i hm o OLCI (L2gen) o an es ua y a he US Eas Coas finding POLYMER o be he p e e ed app oach. Applying he QWIP sco e o he AC esul s o ou s udy o alid ee wa e pixels in he scenes and assuming a h eshold o ≤0.2 gi es 100% eliable R s o A4O and C2RCC, 99% o POLYMER, 81% o IPF, and 45% o ACOLITE-DSF. Wi h a less s ingen h eshold o ≤0.3, ACOLITE-DSF achie es abou 88% quali y-assu ed R s. Wi h a e y s ic QWIP sco e o ≤0.1, A4O s ill eaches 99.4%. This means ha i ually all esul s om A4O, C2RCC and POLYMER pass he QWIP quali y con ol wi h sligh ad an ages o A4O. Bu as men ioned, he e ie ed R s can ac ually ha e he “w ong”shape. The abili y o fill all classes and gene ally good classifiabili y o eflec ances om POLYMER in he J17 amewo k o om A4O in H17 shows he g ea ad an ages o ma ching AC and OWT amewo ks. As desc ibed, howe e , he e is a dange o o e - aluing alse spec a om he AC o measu emen s/simula ions. Ne e heless, i has also p o en ine ec i e no o allow la ge a iances om he expec ed spec um, i.e., po en ial e o s o he AC. Ob iously good spec a om IPF o POLYMER, bu also om A4O, canno be classified well wi h H17. This is especially ue o B21, whe e in p inciple he esul s o all ACs do no ulfil he expec a ions. A comp ehensi e e alua ion o he OWT sys ems and o he pe o mance o di e en a mosphe ic co ec ions is di ficul because he ac ual a eas o applica ion and alidi y o e lap some imes only sligh ly, i.e., inland wa e s s. ocean. La ge a eas o inland wa e s a e in alid flagged o a leas ha e wa ning flags aised, so i is no su p ising ha almos all da a all in o one o only a ew designa ed ocean classes o M14 o B21. Howe e , some o he AC me hods gi e plausible and usable esul s o inland wa e s, which is pa ly e iden in he compa ison wi h AERONET-OC. Lea ing aside he ac ha he e a e also e oneous es ima es o he R s shape, C2RCC and A4O p oduce classifiable esul s o a leas 95% o he cases in he OWT amewo ks J17, M14, and H17, whe e A4O co e s mo e Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g18 in ended classes. POLYMER also achie es his classifiabili y a e o J17 and M14, bu only 70% o H17. Conside ing he ecommended flags, he sui abili y o IPF and ACOLITE-DSF in he in es iga ed classifica ions is insu ficien . The wo k by Liu e al. (2021) also compa es IPF, C2RCC, POLYMER, and o he OLCI AC me hods in con ex wi h he op ical wa e ype and quali y con ol amewo k o Wei e al. (2016), which di e en ia es 22 classes; hey conclude ha POLYMER has bes pe o mance ollowed by C2RCC and IPF. Figu e 3A illus a es a emaining p oblem, namely ha undamen ally di e en spec al shapes o he de i ed R s can o en occu in he ansi ion om coas o sea, when he eshwa e CDOM concen a ion is dilu ed. In some cases, he e a e ea u es in he TOA signal ha can be used o flag po en ial unce ain ies, e.g., a ed-edge enhancemen (Figu e 3D). The ambigui ies o he op ical e ec s o di e en componen s in he wa e , a he ai -sea in e ace, and in he a mosphe e a e ela i ely la ge o spec ally smoo h TOA eflec ance wi h colo nuances o blue. Wi hou sys ema ic compa isons wi h sui able in si u da a, we ha e no means o de e mining which spec al shape is co ec , i.e., which OWT is p esen . Fo his pu pose, mo e hype spec al fiducial e e ence measu emen s especially wi h maximum R s a wa eleng hs ≤510 nm a e needed. 5 Conclusions Fi e a mosphe ic co ec ion me hods o Sen inel-3/OLCI ocean colo image y we e compa ed in e ms o spa ial and spec al esul s and indi idual flagging. The models unde in es iga ion a e he mos ecen e sions o OLCI L2 baseline a mosphe ic co ec ion (IPF), C2RCC, a new me hod A4O, POLYMER, and ACOLITE-DSF. The ex en o which AC me hods p o ide use ul and con inuous esul s o a wide a ie y o na u al wa e s was in es iga ed. Fo his pu pose, he sa elli e- de i ed emo e-sensing eflec ances we e e alua ed in ou op ical wa e ype schemes. Flagging leads in some cases o majo limi a ions in da a exploi a ion e en o clea ly isible wa e a eas; IPF ecommends e y s ic c i e ia, esul ing in 50% less co e age in ou sa elli e image y. Ou pu o R s wi h nega i e alues is a majo issue he e. Howe e , we ha e also shown ha many cases a e inadequa ely flagged by he AC me hods; an example a e high concen a ions o cyanobac e ia a he sea su ace. Only A4O has a dedica ed wa ning flag o floa ing algae, bu A4O is alid he e and deli e s as he only one easonable R s o e he en i e spec um. Ne e heless, a e ision o he indi idual flags wi h espec o spa ial and spec al inconsis encies is ecommended o all AC me hods. Cloud and cloud shadow de ec ion also need o be imp o ed o all me hods, as co esponding deficiencies a e eflec ed in he de i ed wa e quali y p oduc s. Pixel-based app oxima ion o a mosphe ic p ope ies and eflec ance leads o AC-induced spa ial noise. High spa ial he e ogenei y, especially a low eflec ance (and o e co ec ed nega i e alues), leads o conside able losses o possible ma ch- ups wi h in si u measu emen da a. The noise le el can be e ec i ely educed by means o log- ans o ma ion in he R s e ie al p ocess and app op ia e smoo hing, which is bo h applied in A4O. Mainly because o i s high spa ial homogenei y, A4O achie es significan ly mo e ma ch-ups wi h AERONET-OC da a han all o he me hods, namely a leas wice as many poin s in he blue and NIR bands. The numbe o ma ch-ups achie ed also a ec s he s a is ical e alua ion o R s e ie al pe o mance. Compa ison wi h in si u da a, which a e mo e ep esen a i e o coas al and inland wa e s, shows ha he spec al shape and magni ude o R s is essen ially well ep oduced by IPF, C2RCC, and POLYMER, a leas in he cen al isible ange. The cu en e sion o A4O mos ly gi es a easonable shape o R s , bu o en sligh ly lowe alues han obse ed. ACOLITE-DSF p o ides good ma ches o b igh pixel, i.e., highly sca e ing wa e s, bu has significan defici s o low wa e eflec ance in pa icula in he sho wa eleng hs. Hype spec al in si u da a in he 400 o 865 nm ange a e un o una ely no a ailable o all wa e ypes, especially clea oceanic and hype -eu ophic cases a e missing; howe e , his would be impo an o ha e o u u e OWT- ela ed alida ion o AC me hods. Op ical wa e ype classifica ion is used o he selec ion o app op ia e wa e quali y algo i hms and seamless blending o hei esul s. This equi es good classifiabili y o he AC-de i ed R s and i is ad an ageous i all spec al o ms o R s can be ep oduced. Compa ison o he fi e AC me hods shows ha A4O p o ides he g ea es op ical flexibili y. A4O p o ides mo e han 95% usable esul s o h ee OWT amewo ks, namely by Jackson e al. (2017); Moo e e al. (2014), and Hie onymi e al. (2017); u he mo e, A4O popula es mos classes, including hype -eu ophic cases. C2RCC also achie es >95% use ul esul s o he h ee OWT amewo ks, bu has ailing e ie als o in ense cyanobac e ial blooms. Fo he OWT me hod by Jackson e al. (2017), he eflec ances o POLYMER a e bes classifiable; his OWT scheme was de eloped on he basis o such da a. POLYMER also gi es mos ly well- classifiable esul s o M14, bu alls o o H17. The gene al classifiabili y o R s om IPF is compa able o POLYMER, bu conside ing he ecommended alid-pixel-exp ession, he sui abili y o IPF o OWT classifica ion is insu ficien . ACOLITE-DSF is e y ocused on wa e s wi h high concen a ions o non-algal pa icles; he e a e significan p oblems a low ma ine eflec ances, limi ing b oad applica ion in he OWT con ex . The esul s o all AC me hods, o he mos pa , could no be well-classified using he OWT sys em o Bi e al. (2019), and Bi e al. (2021), which has i s ocus o applica ion on inland wa e s; ye compa isons wi h in si u da a sugges ha he ough shape o R s is well ep oduced by mos ACs. So a , OWT algo i hms ha e ocused oo much ei he on ma ine o limnological applica ions; o a comp ehensi e usabili y, missing classes should be added. The classifica ion schemes o Hie onymi e al. (2017) p o ides a good basis, as i includes ep esen a i e classes o ocean, coas al and inland wa e s. Howe e , his me hod in pa icula shows ha e o ole ances should be inc eased in o de o achie e be e classifiabili y o AC esul s, which is he basis o a ully comp ehensi e exploi a ion o an OWT sys em. The ocusing o an OWT sys em on he spec al shape, h ough log- ans o med no maliza ion, inc eases he sensi i i y o noise and small inaccu acies, and hus leads o educed classifica ion pe o mance. I is gene ally ad an ageous i Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g19 he OWT classifica ion sys em is aligned wi h he pe o mance spec um o he a mosphe ic co ec ion and ice e sa. Da a a ailabili y s a emen The en OLCI scenes used in his wo k we e sa ed as Ne CDF files along wi h all he esul s o he fi e a mosphe ic co ec ion models and made a ailable a he open eposi o y Zenodo (Hie onymi e al., 2023). The da a can be ound online a : h ps:// doi.o g/10.5281/zenodo.7567534. Au ho con ibu ions MH concep ualized he s udy, p epa ed, and w o e he o iginal d a . MH, ES, DB, CL, and DM we e in cha ge o he da a cu a ion and sa elli e da a p ocessing. SB pe o med independen OWT analysis. DM pe o med independen ma ch-up analysis. MH and ES conduc ed he spa ial-spec al s udies. MH, KS, CB, FS, QV, and DM deli e ed backg ound in o ma ion on a mosphe ic co ec ion me hods. All au ho s con ibu ed o he a icle and app o ed he submi ed e sion. Funding The Helmhol z Associa ion wi h he esea ch p og am Ea h and En i onmen (PoF IV) unded his s udy. Addi ional suppo was p o ided by he He eon-I 2 Bp ojec Phy oDi eand he Eu opean Cope nicus Ma ine En i onmen Moni o ing Se ice (EU, 77-CMEMS-TAC-OC). Mo eo e , his wo k benefi ed om achie emen s o he ollowing p ojec s: Coas Colou (ESA), OC- CCI (ESA), SEOM-C2X (ESA), OC-BPC (EUMETSAT), and WEnMAP (BMWi & DLR, 50EE1718). Acknowledgmen s This wo k is based on ee and open sa elli e da a om he Eu opean Union’s Cope nicus P og amme p o ided by ESA and EUMETSAT. In addi ion, in si u da a om AERONET-OC we e used, o whichwe hank hePIs:G.Zibo di,S.Ahmed,A.Gile son,S. K a ze ,T.Sch oede ,H.Feng,H.M.Sosik,A.Weidemann,B.Gibson, and R. A none. Mo eo e , D. Vans eenwegen, he Flemish Ma ine Ins i u e, and POM Wes -Vlaande en a e hanked o he ins alla ion, ope a ion and p o ision o PANTHYR da a. We also hank R. Rö ge s,H.K asemann,C.Maze an,M.Pe e s,M.Bö che ,andV. B ando o inspi ing discussions and suppo . Finally, we would like o hank he edi o , J.A. Concha, and h ee expe s o hei ho ough e iew o he pape and help ul commen s. Conflic o in e es Au ho s KS, DM, CL, and CB a e employed by he company B ockmann Consul GmbH, Ge many. Au ho FS is employed by he company HYGEOS, F ance. The emaining au ho s decla e ha he esea ch was conduc ed in he absence o any comme cial o financial ela ionships ha could be cons ued as a po en ial conflic o in e es . The handling edi o JC decla ed a pas co-au ho ship wi h he au ho QV and e iewe IC decla ed a pas collabo a ion wi h he au ho DM o he handling edi o . Publishe ’s no e All claims exp essed in his a icle a e solely hose o he au ho s and do no necessa ily ep esen hose o hei a filia ed o ganiza ions, o hose o he publishe , he edi o s and he e iewe s. Any p oduc ha may be e alua ed in his a icle, o claim ha may be made by i s manu ac u e , is no gua an eed o endo sed by he publishe . Re e ences An oine, D., and Mo el, A. (1998). Rela i e impo ance o mul iple sca e ing by ai molecules and ae osols in o ming he a mosphe ic pa h adiance in he isible and nea -in a ed pa s o he spec um. Appl. Op . 37 (12), 2245–2259. doi: 10.1364/ AO.37.002245 An oine, D., and Mo el, A. (1999). A mul iple sca e ing algo i hm o a mosphe ic co ec ion o emo ely sensed ocean colou (MERIS ins umen ): p inciple and implemen a ion o a mosphe es ca ying a ious ae osols including abso bing ones. In . J. Remo e Sens. 20 (9), 1875–1916. doi: 10.1080/014311699212533 Au in, D., Mannino, A., and F anz, B. (2013). Spa ially esol ing ocean colo and sedimen dispe sion in i e plumes, coas al sys ems, and con inen al shel wa e s. Remo e Sens. En i on. 137, 212–225. doi: 10.1016/j. se.2013.06.018 Bi, S., Hie onymi, M., and Rö ge s, R. (2023). Bio-geo-op ical modelling o na u al wa e s. F on . Ma . Sci. 10. doi: 10.3389/ ma s.2023.1196352 Bi, S., Li, Y., Liu, G., Song, K., Xu, J., Dong, X., e al. (2021). Assessmen o algo i hms o es ima ing chlo ophyll-a concen a ion in inland wa e s: a ound- obin sco ing me hod based on he op ically uzzy clus e ing. IEEE T ans. Geosci. Remo e Sens. 60, 1– 17. doi: 10.1109/TGRS.2021.3058556 Bi, S., Li, Y., Wang, Q., Lyu, H., Liu, G., Zheng, Z., e al. (2018). Inland wa e a mosphe ic co ec ion based on u bidi y classifica ion using OLCI and SLSTR syne gis ic obse a ions. Remo e Sens. 10 (7), 1002. doi: 10.3390/ s10071002 Bi, S., Li, Y., Xu, J., Liu, G., Song, K., Mu, M., e al. (2019). Op ical classifica ion o inland wa e s based on an imp o ed uzzy c-means me hod. Op . Exp ess. 27, 34838– 34856. doi: 10.1364/OE.27.034838 B ache , A., Bouman, H. A., B ewin, R. J., B icaud, A., B o as, V., Cio i, A. M., e al. (2017). Ob aining phy oplank on di e si y om ocean colo : a scien ific oadmap o u u e de elopmen . F on . Ma . Sci. 4. doi: 10.3389/ ma s.2017.00055 B ewin, R. J. W., Sa hyend ana h, S., Kulk, G., Rio, M.-H., Concha, J. A., Bell, T. G., e al. (2023). Ocean ca bon om space: cu en s a us and p io i ies o he nex decade. Ea h-Science Re . 240, 104386. doi: 10.1016/j.ea sci e .2023.104386 B ockmann, C., Doe e , R., Pe e s, M., Ke s in, S., Embache , S., and Ruescas, A. (2016). “E olu ion o he C2RCC neu al ne wo k o sen inel 2 and 3 o he e ie al o ocean colou p oduc s in no mal and ex eme op ically complex wa e s,”in P oc. li ing plane symposium., ol. SP-740. (P ague, Czech Republic: ESA), 1–6. B ockmann, C., Pape in, M., Danne, O., and Ruescas, A. (2013). “Mul i-senso cloud sc eening and alida ion: IdePix and PixBox,”in P oc. li ing plane symposium., ol. SP-722. (Edinbu gh, UK ESA), 9–13. Bulga elli, B., and Zibo di, G. (2018). On he de ec abili y o adjacency e ec s in ocean colo emo e sensing o mid-la i ude coas al en i onmen s by SeaWiFS, MODIS-a, MERIS, OLCI, OLI and MSI. Remo e Sens. En i on. 209, 423–438. doi: 10.1016/ j. se.2017.12.021 Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g20 Cazzaniga, I., Zibo di, G., and Melin, F. (2021). Spec al a ia ions o he emo e sensing eflec ance du ing coccoli hopho e blooms in he Wes e n black Sea. Remo e Sens. En i on. 264, 112607. doi: 10.1016/j. se.2021.112607 Cazzaniga, I., Zibo di, G., and Melin, F. (2023). Spec al ea u es o ocean colou adiome ic p oduc s in he p esence o cyanobac e ia blooms in he Bal ic Sea. Remo e Sens. En i on. 287, 113464. doi: 10.1016/j. se.2023.113464 Cazzaniga, I., Zibo di, G., Melin, F., Kwia kowska, E., Talone, M., Dessailly, D., e al. (2022). E alua ion o OLCI neu al ne wo k adiome ic wa e p oduc s. IEEE Geosci. Remo e Sens. Le . 19, 1–5. doi: 10.1109/LGRS.2021.3136291 Die ssen, H. M. (2019). Hype spec al measu emen s, pa ame e iza ions, and a mosphe ic co ec ion o whi ecaps and oam om isible o sho wa e in a ed o ocean colo emo e sensing 7, 14. doi: 10.3389/ ea .2019.00014 Die ssen, H. M., Vande meulen, R. A., Ba nes, B. B., Cas agna, A., Knaeps, E., and Vanhellemon , Q. (2022). QWIP: a quan i a i e me ic o quali y con ol o aqua ic eflec ance spec al shape using he appa en isible wa eleng h. F on . Remo e Sens. 3. doi: 10.3389/ sen.2022.869611 Doe e , R., and Schille , H. (2007). The MERIS case 2 wa e algo i hm. In . J. Remo e Sens. 28 (3-4), 517–535. doi: 10.1080/01431160600821127 Ele eld, M. A., Ruescas, A. B., Homme som, A., Moo e, T. S., Pe e s, S. W., and B ockmann, C. (2017). An op ical classifica ion ool o global lake wa e s. Remo e Sens. 9 (5), 420. doi: 10.3390/ s9050420 EUMETSAT (2021). Sen inel-3 OLCI L2 epo o baseline collec ion OL_L2M_003.. EUMETSAT (2022). Recommenda ions o sen inel-3 OLCI ocean colou p oduc alida ions in compa ison wi h in si u measu emen s –ma chup p o ocols.. Fom e a, N., Bö che , M., Zühlke, M., B ockmann, C., and Kwia kowska, E. (2012). “Cal alus: ull-mission EO cal/ al, p ocessing and exploi a ion se ices,”in P oc. IEEE in e na ional geoscience and emo e sensing symposium (IGARSS’12).(Munich, Ge many), 5278–5281. F ouin, R. J., F anz, B. A., Ib ahim, A., Knobelspiesse, K., Ahmad, Z., Cai ns, B., e al. (2019). A mosphe ic co ec ion o sa elli e ocean-colo image y du ing he PACE e a. F on . Ea h Sci. 7. doi: 10.3389/ ea .2019.00145 GCOS (2011). “Sys ema ic obse a ion equi emen s o sa elli e-based p oduc s o clima e,”in 2011 upda e., ol. 154. (WMO GCOS Rep.), 127. Giannini, F., Hun , B. P., Jacoby, D., and Cos a, M. (2021). Pe o mance o OLCI sen inel-3A sa elli e in he no heas pacific coas al wa e s. Remo e Sens. En i on. 256, 112317. doi: 10.1016/j. se.2021.112317 Gons, H. J. (1999). Op ical elede ec ion o chlo ophyll a in u bid inland wa e s. En i on. Sci. Technol. 33 (7), 1127–1132. doi: 10.1021/es9809657 Go don, H. R. (2021). E olu ion o ocean colo a mosphe ic co ec ion: 1970–2005. Remo e Sens. 13 (24), 5051. doi: 10.3390/ s13245051 Go don, H. R., and Wang, M. (1994). Re ie al o wa e -lea ing adiance and ae osol op ical hickness o e he oceans wi h SeaWiFS: a p elimina y algo i hm. Appl. Op . 33 (3), 443–452. doi: 10.1364/AO.33.000443 Gossn, J. I., Ruddick, K. G., and Doglio i, A. I. (2019). A mosphe ic co ec ion o OLCI image y o e ex emely u bid wa e s based on he ed, NIR and 1016 nm bands and a new baseline esidual echnique. Remo e Sens. 11 (3), 220. doi: 10.3390/ s11030220 Goyens, C., Jame , C., and Sch oede , T. (2013). E alua ion o ou a mosphe ic co ec ion algo i hms o MODIS-aqua images o e con as ed coas al wa e s. Remo e Sens. En i on. 131, 63–75. doi: 10.1016/j. se.2012.12.006 Guan e , L., Ruiz-Ve du, A., Ode ma , D., Gia dino, C., Simis, S., Es elles, V., e al. (2010). A mosphe ic co ec ion o ENVISAT/MERIS da a o e inland wa e s: alida ion o Eu opean lakes. Remo e Sens. En i on. 114 (3), 467–480. doi: 10.1016/ j. se.2009.10.004 Hie onymi, M. (2019). Spec al band adap a ion o ocean colo senso s o applicabili y o he mul i-wa e biogeo-op ical algo i hm ONNS. Op . Exp ess. 27 (12), A707–A724. doi: 10.1364/OE.27.00A707 Hie onymi, M., Bi, S., Mülle , D., Schü , E. M., Beh , D., B ockmann, C., e al. (2023). Supplemen a y da ase o he publica ion by Hie onymi e al.: "Ocean colo a mosphe ic co ec ion me hods in iew o usabili y o di e en op ical wa e ypes". Zenodo. doi: 10.5281/zenodo.7567534 Hie onymi, M., K asemann, H., Mülle , D., B ockmann, C., Ruescas, A. B., S elze , K., e al. (2016). “Ocean colou emo e sensing o ex eme case-2 wa e s,”in P oc. Li ing plane symposium, ol. SP-740. (P ague, Czech Republic: ESA), 1–5. Hie onymi, M., Mülle , D., and Doe e , R. (2017). The OLCI neu al ne wo k swa m (ONNS): a bio-Geo-Op ical algo i hm o open ocean and coas al wa e s. F on . Ma . Sci. 4. doi: 10.3389/ ma s.2017.00140 Hollmann, R., Me chan , C. J., Saunde s, R., Downy, C., Buchwi z, M., Cazena e, A., e al. (2013). The ESA clima e change ini ia i e: sa elli e da a eco ds o essen ial clima e a iables. Bull. Am. Me eo ological Soc. 94 (10), 1541–1552. doi: 10.1175/ BAMS-D-11-00254.1 Hun e , P. D., Ma hews, M. W., Ku se , T., and Tyle , A. N. (2016). “Remo e sensing o cyanobac e ial blooms in inland, coas al, and ocean wa e s,”in Handbook o cyanobac e ial moni o ing and cyano oxin analysis.. Eds. J. Me iluo o, L. Spoo and G. A. Codd, 89–99. doi: 10.1002/9781119068761.ch9 IOCCG (2010). A mosphe ic co ec ion o emo ely-sensed ocean-colou p oduc s. Vol. 10. Ed. M. Wang (Da mou h, Canada: Repo s o he In e na ional Ocean-Colou Coo dina ing G oup), 84. doi: 10.25607/OBP-101 IOCCG (2019). Unce ain ies in ocean colou emo e sensing. Vol. 18. Ed. F. Melin (Da mou h, Canada: Repo s o he In e na ional Ocean-Colou Coo dina ing G oup), 164. doi: 10.25607/OBP-696 Jackson, T., Sa hyend ana h, S., and Melin, F. (2017). An imp o ed op ical classifica ion scheme o he ocean colou essen ial clima e a iable and i s applica ions. Remo e Sens. En i on. 203, 152–161. doi: 10.1016/j. se.2017.03.036 Je lo , N. G. (1976). Ma ine op ics. (Else ie ). A ailable a : h ps://books.google.de/ books?hl=de&l =&id= zwg nW_lYC&oi= nd&pg=PP1&o s=23jJaYPV6i&sig= K7jZxQsizMxC_QOZ1 S2 QeL72o# =onepage&q& = alse. Jia, T., Zhang, Y., and Dong, R. (2021). A uni e sal uzzy logic op ical wa e ype scheme o he global oceans. Remo e Sens. 13 (19), 4018. doi: 10.3390/ s13194018 Juhls, B., Ma suoka, A., Lizo e, M., Becu, G., O e duin, P. P., El Kassa , J., e al. (2022). Seasonal dynamics o dissol ed o ganic ma e in he Mackenzie del a, Canadian A c ic wa e s: implica ions o ocean colou emo e sensing. Remo e Sens. En i on. 283, 113327. doi: 10.1016/j. se.2022.113327 Kajiyama, T., D’Alimon e, D., and Zibo di, G. (2018). Algo i hms me ging o he de e mina ion o chlo ophyll-a concen a ion in he black Sea. IEEE Geosci. Remo e Sens. Le . 16 (5), 677–681. doi: 10.1109/LGRS.2018.2883539 Kiebe , R. J., Whi ehead, R. F., Reid, S. N., Willey, J. D., and Sea on, P. J. (2006). Ch omopho ic dissol ed o ganic ma e (CDOM) in ainwa e , sou heas e n no h Ca olina, USA. J. A mo. Chem. 54 (1), 21–41. doi: 10.1007/s10874-005-9008-4 Li, Q., Jiang, L., Chen, Y., Wang, L., and Wang, L. (2022). E alua ion o se en a mosphe ic co ec ion algo i hms o OLCI images o e he coas al wa e s o qinhuangdao in bohai Sea. Regional S ud. Ma . Sci. 56, 102711. doi: 10.1016/ j. sma.2022.102711 Liu, H., He, X., Li, Q., Hu, X., Ishizaka, J., K a ze , S., e al. (2021). E alua ion o ocean colo a mosphe ic co ec ion me hods o sen inel-3 OLCI using global au oma ic in si u obse a ions. IEEE T ans. Geosci. Remo e Sens. 60, 1–19. doi: 10.1109/TGRS.2021.3136243 Liu, H., Hu, S., Zhou, Q., Li, Q., and Wu, G. (2019). Re isi ing e ec i eness o u bidi y index o he swi ching scheme o NIR-SWIR combined ocean colo a mosphe ic co ec ion algo i hm. In . J. Appl. Ea h Obs. Geoin . 76, 1–9. doi: 10.1016/j.jag.2018.10.010 Ma in T ayko ski, L. V., and Sosik, H. M. (2003). Fea u e-based classifica ion o op ical wa e ypes in he No hwes A lan ic based on sa elli e ocean colo da a. J. Geophys. Res. Oceans. 108 (C5). doi: 10.1029/2001JC001172 Melin, F., and Van epo e, V. (2015). How op ically di e se is he coas al ocean? Remo e Sens. En i on. 160, 235–251. doi: 10.1016/j. se.2015.01.023 Melin, F., Van epo e, V., Cle ici, M., D’Alimon e, D., Zibo di, G., Be hon, J.-F., e al. (2011). Mul i-senso sa elli e ime se ies o op ical p ope ies and chlo ophyll-a concen a ion in heAd ia icSea.P og. Oceanog . 91, 229–244. doi: 10.1016/j.pocean.2010.12.001 Minu, P., Lo like , A. A., Shaju, S. S., Ash a , P. M., Kuma , T. S., and Meenakuma i, B. (2016). Pe o mance o ope a ional sa elli e bio-op ical algo i hms in di e en wa e ypes in he sou heas e n A abian Sea. Oceanologia. 58 (4), 317–326. doi: 10.1016/ j.oceano.2016.05.005 Mi chell, C., Hu, C., Bowle , B., D apeau, D., and Balch, W. M. (2017). Es ima ing pa icula e ino ganic ca bon concen a ions o he global ocean om ocean colo measu emen s using a eflec ance di e ence app oach. J. Geophys. Res. Oceans. 122 (11), 8707–8720. doi: 10.1002/2017JC013146 Mobley, C. D. (1994). Ligh and wa e : adia i e ans e in na u al wa e s. (Academic p ess). Mog ane, M. A., Jame , C., Loisel, H., Van epo e, V., Me iaux, X., and Cau in, A. (2019). E alua ion o fi e a mosphe ic co ec ion algo i hms o e F ench op ically- complex wa e s o he sen inel-3A OLCI ocean colo senso . Remo e Sens. 11 (6), 668. doi: 10.3390/ s11060668 Moo e, G. F., Aiken, J., and La ende , S. J. (1999). The a mosphe ic co ec ion o wa e colou and he quan i a i e e ie al o suspended pa icula e ma e in case II wa e s: applica ion o MERIS. In . J. Remo e Sens. 20 (9), 1713–1733. doi: 10.1080/ 014311699212434 Moo e, T. S., Campbell, J. W., and Feng, H. (2001). A uzzy logic classifica ion scheme o selec ing and blending sa elli e ocean colo algo i hms. IEEE T ans. Geosci. Remo e Sens. 39, 1764–1776. doi: 10.1109/36.942doi Moo e, T. S., Dowell, M. D., B ad , S., and Ve du, A. R. (2014). An op ical wa e ype amewo k o selec ing and blending e ie als om bio-op ical algo i hms in lakes and coas al wa e s. Remo e Sens. En i on. 143, 97–111. doi: 10.1016/j. se.2013.11.021 Mo el, A., and P ieu , L. (1977). Analysis o a ia ions in ocean colo . Limnology Oceanog aphy. 22 (4), 709–722. doi: 10.4319/lo.1977.22.4.0709 Mülle , D., K asemann, H., B ewin, R. J., B ockmann, C., Deschamps, P. Y., Doe e , R., e al. (2015a). The ocean colou clima e change ini ia i e: i. a me hodology o assessing a mosphe ic co ec ion p ocesso s based on in-si u measu emen s. Remo e Sens. En i on. 162, 242–256. doi: 10.1016/j. se.2013.11.026 Mülle , D., K asemann, H., B ewin, R. J., B ockmann, C., Deschamps, P. Y., Doe e , R., e al. (2015b). The ocean colou clima e change ini ia i e: II. spa ial and empo al homogenei y o sa elli e da a e ie al due o sys ema ic e ec s in a mosphe ic co ec ion p ocesso s. Remo e Sens. En i on. 162, 257–270. doi: 10.1016/j. se.2015.01.033 Mulle -Ka ge , F. E., Hes i , E., Ade, C., Tu pie, K., Robe s, D. A., Siegel, D., e al. (2018). Sa elli e senso equi emen s o moni o ing essen ial biodi e si y a iables o coas al ecosys ems. Ecol. Appl. 28 (3), 749–760. doi: 10.1002/eap.1682 Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g21 Nobileau, D., and An oine, D. (2005). De ec ion o blue-abso bing ae osols using nea in a ed and isible (ocean colo ) emo e sensing obse a ions. Remo e Sens. En i on. 95 (3), 368–387. doi: 10.1016/j. se.2004.12.020 Pa k, Y.-J., and Ruddick, K. G. (2005). Model o emo e-sensing eflec ance including bidi ec ional e ec s o case 1 and case 2 wa e s. Appl. Op . 44 (7), 1236– 1249. doi: 10.1364/AO.44.001236 Qi, L., Hu, C., Duan, H., Cannizza o, J., and Ma, R. (2014). A no el MERIS algo i hm o de i e cyanobac e ial phycocyanin pigmen concen a ions in a eu ophic lake: heo e ical basis and p ac ical conside a ions. Remo e Sens. En i on. 154, 298–317. doi: 10.1016/j. se.2014.08.026 Qin, P., Simis, S. G., and Tils one, G. H. (2017). Radiome ic alida ion o a mosphe ic co ec ion o MERIS in he Bal ic Sea based on con inuous obse a ions om ships and AERONET-OC. Remo e Sens. En i on. 200, 263–280. doi: 10.1016/j. se.2017.08.024 Reina , A., and Ku se , T. (2006). Compa ison o di e en sa elli e senso s in de ec ing cyanobac e ial bloom e en s in he Bal ic Sea. Remo e Sens. En i on. 102 (1- 2), 74–85. doi: 10.1016/j. se.2006.02.013 Ruddick, K. G., Gons, H. J., Rijkeboe , M., and Tils one, G. (2001). Op ical emo e sensing o chlo ophyll a in case 2 wa e s by use o an adap i e wo-band algo i hm wi h op imal e o p ope ies. Appl. Op . 40 (21), 3575–3585. doi: 10.1364/AO.40.003575 Sa hyend ana h, S., B ewin, R. J., Jackson, T., Melin, F., and Pla , T. (2017). Ocean- colou p oduc s o clima e-change s udies: wha a e hei ideal cha ac e is ics? Remo e Sens. En i on. 203, 125–138. doi: 10.1016/j. se.2017.04.017 Sa hyend ana h, S., Jackson, T., B ockmann, C., B o as, V., Cal on, B., Chup in, A., e al. (2021). ESA Ocean colou clima e change ini ia i e (Ocean_Colou _cci): e sion 5.0 da a. (NERC EDS Cen e o En i onmen al Da a Analysis). doi: 10.5285/ 1dbe7a109c0244aaad713e078 d3059a Sch oede , T., Schaale, M., Lo ell, J., and Blondeau-Pa issie , D. (2022). An ensemble neu al ne wo k a mosphe ic co ec ion o sen inel-3 OLCI o e coas al wa e s p o iding inhe en model unce ain y es ima ion and senso noise p opaga ion. Remo e Sens. En i on. 270, 112848. doi: 10.1016/j. se.2021.112848 Shi, K., Li, Y., Li, L., Lu, H., Song, K., Liu, Z., e al. (2013). Remo e chlo ophyll-a es ima es o inland wa e s based on a clus e -based classifica ion. Sci. To . En i on. 444, 1–15. doi: 10.1016/j.sci o en .2012.11.058 Shi, W., and Wang, M. (2009). An assessmen o he black ocean pixel assump ion o MODIS SWIR bands. Remo e Sens. En i on. 113 (8), 1587–1597. doi: 10.1016/ j. se.2009.03.011 Smi h, M. E., Lain, L. R., and Be na d, S. (2018). An op imized chlo ophyll a swi ching algo i hm o MERIS and OLCI in phy oplank on-domina ed wa e s. Remo e Sens. En i on. 215, 217–227. doi: 10.1016/j. se.2018.06.002 Soome s, T., Uudebe g, K., Jako els, D., Zaga s, M., Reina , A., B auns, A., e al. (2019). Compa ison o lake op ical wa e ypes de i ed om sen inel-2 and sen inel-3. Remo e Sens. 11 (23), 2883. doi: 10.3390/ s11232883 Spy akos, E., O’Donnell, R., Hun e , P. D., Mille , C., Sco , M., Simis, S. G. H., e al. (2018). Op ical ypes o inland and coas al wa e s. Limnology Oceanog aphy. 63 (2), 846–870. doi: 10.1002/lno.10674 S einme z, F., Deschamps, P. Y., and Ramon, D. (2011). A mosphe ic co ec ion in p esence o sun glin : applica ion o MERIS. Op . Exp ess. 19 (10), 9783–9800. doi: 10.1364/OE.19.009783 S einme z, F., and Ramon, D. (2018). “Sen inel-2 MSI and sen inel-3 OLCI consis en ocean colou p oduc s using POLYMER,”in P oc. SPIE emo e sensing o he open and coas al ocean and inland wa e s., ol. Vol. 10778. (Honolulu, USA), 46– 55. doi: 10.1117/12.2500232 Tils one, G., Dall’Olmo, G., Hie onymi, M., Ruddick, K., Beck, M., Ligi, M., e al. (2020). Field in e compa ison o adiome e measu emen s o ocean colou alida ion. Remo e Sens. 12 (10), 1587. doi: 10.3390/ s12101587 Tils one, G., Mallo -Hoya, S., Gohin, F., Cou o, A. B., Sa, C., Goela, P., e al. (2017). Which ocean colou algo i hm o MERIS in no h Wes Eu opean wa e s? Remo e Sens. En i on. 189, 132–151. doi: 10.1016/j. se.2016.11.012 Tils one, G. H., Pa do, S., Simis, S. G., Qin, P., Selmes, N., Dessailly, D., e al. (2021). Consis ency be ween sa elli e ocean colou p oduc s unde high colou ed dissol ed o ganic ma e abso p ion in he Bal ic Sea. Remo e Sens. 14 (1), 89. doi: 10.3390/ s14010089 Tu ne , K. J., Tzo ziou, M., G une , B. K., Goes, J., and She man, J. (2022). Op ical classifica ion o an u banized es ua y using hype spec al emo e sensing eflec ance. Op . Exp ess. 30 (23), 41590–41612. doi: 10.1364/OE.472765 Uudebe g, K., Aa as e, A., Kõks, K. L., Anspe , A., Uusõue, M., Kang o, K., e al. (2020). Op ical wa e ype guided app oach o es ima e op ical wa e quali y pa ame e s. Remo e Sens. 12 (6), 931. doi: 10.3390/ s12060931 Valen e, A., Sa hyend ana h, S., B o as, V., G oom, S., G an , M., Jackson, T., e al. (2022). A compila ion o global bio-op ical in si u da a o ocean colou sa elli e applica ions– e sion h ee. Ea h Sys em Sci. Da a. 14 (12), 5737–5770. doi: 10.5194/essd-14-5737-2022 Vande meulen, R. A., Mannino, A., C aig, S. E., and We dell, P. J. (2020). 150 shades o g een: using he ull spec um o emo e sensing eflec ance o elucida e colo shi s in he ocean. Remo e Sens. En i on. 247, 111900. doi: 10.1016/j. se.2020.111900 Vanhellemon , Q. (2019a). Adap a ion o he da k spec um fi ing a mosphe ic co ec ion o aqua ic applica ions o he landsa and sen inel-2 a chi es. Remo e Sens. En i on. 225, 175–192. doi: 10.1016/j. se.2019.03.010 Vanhellemon , Q. (2019b). Daily me e-scale mapping o wa e u bidi y using CubeSa image y. Op . Exp ess. 27 (20), A1372–A1399. doi: 10.1364/OE.27.0A1372 Vanhellemon , Q. (2020). Sensi i i y analysis o he da k spec um fi ing a mosphe ic co ec ion o me e-and decame e-scale sa elli e image y using au onomous hype spec al adiome y. Op . Exp ess. 28 (20), 29948–29965. doi: 10.1364/OE.397456 Vanhellemon , Q., and Ruddick, K. (2018). A mosphe ic co ec ion o me e-scale op ical sa elli e da a o inland and coas al wa e applica ions. Remo e Sens. En i on. 216, 586–597. doi: 10.1016/j. se.2018.07.015 Vanhellemon , Q., and Ruddick, K. (2021). A mosphe ic co ec ion o sen inel-3/ OLCI da a o mapping o suspended pa icula e ma e and chlo ophyll-a concen a ion in Belgian u bid coas al wa e s. Remo e Sens. En i on. 256, 112284. doi: 10.1016/j. se.2021.112284 Vans eenwegen, D., Ruddick, K., Ca ijsse, A., Vanhellemon , Q., and Beck, M. (2019). The pan-and- il hype spec al adiome e sys em (PANTHYR) o au onomous sa elli e alida ion measu emen s–p o o ype design and es ing. Remo e Sens. 11 (11), 1360. doi: 10.3390/ s11111360 Van epo e,V.,Loisel,H.,Dessailly,D.,andMe  iaux, X. (2012). Op ical classifica ion o con as ed coas al wa e s. Remo e Sens. En i on. 123, 306–323. doi: 10.1016/j. se.2012.03.004 Wei, J., Lee, Z., and Shang, S. (2016). A sys em o measu e he da a quali y o spec al emo e-sensing eflec ance o aqua ic en i onmen s. J. Geophys. Res. Oceans. 121 (11), 8189–8207. doi: 10.1002/2016JC012126 Wei, J., Wang, M., Mikelsons, K., Jiang, L., K a ze , S., Lee, Z., e al. (2022). Global sa elli e wa e classifica ion da a p oduc s o e oceanic, coas al, and inland wa e s. Remo e Sens. En i on. 282, 113233. doi: 10.1016/j. se.2022.113233 We dell, P. J., McKinna, L. I., Boss, E., Ackleson, S. G., C aig, S. E., G egg, W. W., e al. (2018). An o e iew o app oaches and challenges o e ie ing ma ine inhe en op ical p ope ies om ocean colo emo e sensing. P og. Oceanog aphy. 160, 186–212. doi: 10.1016/j.pocean.2018.01.001 Windle, A. E., E e s-King, H., Lo eday, B. R., Ond usek, M., and Silsbe, G. M. (2022). E alua ing a mosphe ic co ec ion algo i hms applied o OLCI sen inel-3 da a o Chesapeake bay wa e s. Remo e Sens. 14 (8), 1881. doi: 10.3390/ s14081881 Xi, H., Hie onymi, M., K asemann, H., and Rö ge s, R. (2017). Phy oplank on g oup iden ifica ion using simula ed and in si u hype spec al emo e sensing eflec ance. F on . Ma . Sci. 4. doi: 10.3389/ ma s.2017.00272 Xi,H.,Hie onymi,M.,Rö ge s,R.,K asemann,H.,andQiu,Z.(2015). Hype spec al di e en ia ion o phy oplank on axonomic g oups: a compa ison be ween using emo e sensing eflec ance and abso p ion spec a. Remo e Sens. 7 (11), 14781–14805. doi: 10.3390/ s71114781 Zhang, M., Hu, C., and Ba nes, B. B. (2019). Pe o mance o POLYMER a mosphe ic co ec ion o ocean colo image y in he p esence o abso bing ae osols. IEEE T ans. Geosci. Remo e Sens. 57 (9), 6666–6674. doi: 10.1109/TGRS.2019.2907884 Zibo di, G., Kwia kowska, E., Melin, F., Talone, M., Cazzaniga, I., Dessailly, D., e al. (2022). Assessmen o OLCI-a and OLCI-b adiome ic da a p oduc s ac oss Eu opean seas. Remo e Sens. En i on. 272, 112911. doi: 10.1016/j. se.2022.112911 Zibo di, G., Melin, F., Be hon, J. F., Holben, B., Slu ske , I., Giles, D., e al. (2009). AERONET-OC: a ne wo k o he alida ion o ocean colo p ima y p oduc s. J. A mosphe ic Oceanic Tech. 26 (8), 1634–1651. doi: 10.1175/2009JTECHO654.1 Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g22 Appendix 1 APPENDIX FIGURE A1 O e iew o all en OLCI scenes used (Table 2) in e ical nea -side pe spec i e. RGB images c ea ed om L TOA . Ma ked in ed a e he 100x100 pixel a eas o es ima ing spa ial homogenei y (Table 4). The image sec ions in Figu e 2 a e shown in o ange. Hie onymi e al. 10.3389/ ma s.2023.1129876 F on ie s in Ma ine Science on ie sin.o g23