ACIX-III AQUA: EVALUATION OF ATMOSPHERIC CORRECTION PROCESSORS FOR HYPERSPECTRAL SATELLITE OVER INLAND AND COASTAL WATERS

ACIX-III AQUA EVALUATION OF ATMOSPHERIC CORRECTION PROCESSORS FOR HYPERSPECTRAL SATELLITE OVER INLAND AND COASTAL WATERS A comparative analysis of atmospheric correction methods applied to PRISMA and EnMAP using in situ reflectance data CNR-IREA, Milan, November 2025 ISBN 979-12-985355-2-7 ACIX-III Aqua report 2 Author list and affiliations Claudia Giardino1, Nima Pahlevan2, Alice Fabbretto1,3, Lodovica Panizza1, Andrea Pellegrino1,4, Ryan Vandermeulen5, Marco Gianinetto6, Stefan Adriaensen7, Joram Agten7, Hendrik Bernert8, Liesbeth De Keukelaere7, Tristan Harmel9, Thomas Heege8, Anders Knudby9, Karin Schenk8, François Steinmetz10, Sindy Sterckx7, Quinten Vanhellemont12, Yulun Wu10, Federica Braga13, Vittorio E. Brando14, Mariano Bresciani1, Ana Dogliotti15, Susanne Kratzer16, Salvatore Mangano1, Daniel Odermatt17, Gian Marco Scarpa13, Thomas Schroeder18, Mortimer Werther17, Ettore Lopinto19, Kevin Alonso20, Noelle Cremer20, Georgia Doxani20, Ferran Gascon20 Astrid Bracher21,22, Mariana A. Soppa21, Avotramalala Najoro Radrianalisoa21 and additionally all other coauthors from Soppa et al. (2024) who are not listed above: Maximilian Brell23, Sabine Chabrillat24, Leonardo M. A. Alvarado25, Peter Gege26, Stefan Plattner24, Ian Somlai-Schweiger24, Nicole Pinnel24, Simone Collela14, Dieter Vansteenwegen25, Maximilian Langheinrich26, Emiliano Carmona26, Martin Bachmann26, Miguel Pato26, Sebastian Fischer27 1 Institute of Electromagnetic Sensing of the Environment (IREA), CNR, Milan, Italy. 2 NASA Goddard Space Flight Center, Greenbelt, MD, USA. 3 Department of Remote Sensing, Tartu Observatory, University of Tartu, Tõravere, Estonia 4 Department of Engineering, University of Sapienza, Rome, Italy. 5 NOAA Fisheries, Silver Spring, MD, USA. 6 Politecnico di Milano, Milan, Italy. 7 Flemish Institute for Technological Research (VITO), Mol, Belgium. 8 EOMAP GmbH & Co. KG, Seefeld, Germany. 9 Earth Observation Unit, Magellium, Ramonville-Saint-Agne, France. 10 University of Ottawa, Ottawa, Canada. 11 HYGEOS, Lille, France. 12 Royal Belgian Institute of Natural Sciences (RBINS), Brussels, Belgium. 13 Institute of Marine Sciences (ISMAR), CNR, Venice, Italy. 14 Institute of Marine Sciences (ISMAR), CNR, Rome, Italy. 15 Instituto de Astronomía y Física del Espacio (IAFE), Buenos Aires, Argentina. 16 Department of Ecology, Environment and Plant Sciences (DEEP), Stockholm University, Stockholm, Sweden. 17 Swiss Federal Institute of Aquatic Science and Technology (EAWAG), Dübendorf, Switzerland. 18 Commonwealth Scientific and Industrial Research Organisation (CSIRO), Brisbane, Australia. 19 Italian Space Agency (ASI), Rome, Italy. 20 European Space Research Institute, European Space Agency (ESA), Frascati, Italy. 21 Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, 27570 Bremerhaven, Germany 22 Institute of Environmental Physics, University of Bremen, Otto-Hahn-Allee 1, 28359 Bremen, Germany 23 GFZ German Research Centre for Geosciences, Potsdam, Germany21 CSIRO, Brisbane, Australia 24 German Aerospace Center (DLR), Remote Sensing Technology Institute, Weßling, Germany 25 Flanders Marine Institute (VLIZ), Oostende, Belgium 26 Earth Observation Center (EOC), German Aerospace Center (DLR), Weßling, Germany 27 German Aerospace Center (DLR), Bonn, Germany ACIX-III Aqua report 3 TABLE OF CONTENTS 1 ABSTRACT .............................................................................................................................. 10 2 KEY POINTS ............................................................................................................................ 11 3 INTRODUCTION ...................................................................................................................... 12 3.1 REPORT AIMS ................................................................................................................. 15 4 MATERIALS AND METHODS ..................................................................................................... 17 4.1 IN SITU DATA .................................................................................................................. 17 4.1.1 AERONET-OC NETWORK ............................................................................................ 17 4.1.2 COMMUNITY VALIDATION DATABASE ......................................................................... 18 4.2 ATMOSPHERIC CORRECTION MODELS ........................................................................... 18 4.2.1 ACOLITE ..................................................................................................................... 19 4.2.2 hGRS .......................................................................................................................... 20 4.2.3 POLYMER ................................................................................................................... 21 4.2.4 iCOR .......................................................................................................................... 21 4.2.5 MIP ............................................................................................................................ 22 4.2.6 PACO-WASI ................................................................................................................ 22 4.2.7 ACOLITE-T-Mart .......................................................................................................... 23 5 PRISMA DATA ANALYSIS ......................................................................................................... 25 5.1 FEATURES OF THE PRISMA DATASET USED IN THE EXERCISE .......................................... 25 5.2 MATCH-UPS ANALYSIS ................................................................................................... 26 5.2.1 DATASET .................................................................................................................... 27 5.2.2 PERFORMANCE ASSESSMENT .................................................................................... 27 5.2.3 STATISTICS ................................................................................................................. 28 5.2.4 QUALITY WATER INDEX POLYNOMIAL ......................................................................... 29 5.3 RESULTS ........................................................................................................................ 31 5.3.1 OVER THE SITES ......................................................................................................... 31 5.3.2 OVER THE SPECTRAL BANDS ...................................................................................... 37 5.3.3 OPTICAL WATER TYPES ............................................................................................... 50 5.4 INTERPRETING THE QWIP OUTPUT ................................................................................. 55 6 EnMAP DATA ANALYSES .......................................................................................................... 61 6.1 DATASET AND MATCH-UP ANALYSIS ............................................................................... 61 6.2 IN SITU DATA .................................................................................................................. 61 6.3 EnMAP DATA .................................................................................................................. 63 6.4 MATCH-UP ANALYSIS ..................................................................................................... 63 ACIX-III Aqua report 4 6.5 PERFORMANCE ASSESSMENT ........................................................................................ 65 6.5.1 MULTISPECTRAL PERFORMANCE ............................................................................... 65 6.5.2 HYPERSPECTRAL PERFORMANCE .............................................................................. 72 7 CONCLUSIONS ...................................................................................................................... 79 8 ACKNOWLEDGMENTS ............................................................................................................ 81 9 REFERENCES ......................................................................................................................... 82 10 PRISMA DATA REPOSITORY ................................................................................................ 90 11 PRISMA GALLERY ............................................................................................................... 91 ACIX-III Aqua report 5 LIST OF FIGURES Figure 1 Dataset of the globally distributed sites and number of available match-ups between PRISMA and field data. The colour of the dots indicates the number of available match-ups ranging from a minimum of 1 to a maximum of 17 for every study area. The Location of AERONET-OC are indicated with circles only, the black dots indicate CVD sites; Rio de la Plata and AAOT (Venice) are both AERONET-OC and CVD. The bottom-right bar plot shows the frequency distribution of Rrs on the y-axis, with the values in sr⁻¹ reported on the x-axis. This is shown for both multispectral (AERONET-OC, light blue) and hyperspectral (CVD, green) in situ data collection. The darker blue area shows the overlap between the two distributions.................................................................. 27 Figure 2 Reference OWTs for the different AC models performance assessment [79]. ...................................... 28 Figure 3 Spectral comparisons between PRISMA L2C and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of PRISMA L2C is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ... 32 Figure 4 Spectral comparisons between ACOLITE and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ... 33 Figure 5 Spectral comparisons between hGRS and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of hGRS is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ...................... 34 Figure 6 Spectral comparisons between POLYMER and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ... 34 Figure 7 Spectral comparisons between iCOR and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of PRISMA iCOR is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ... 35 Figure 8 Spectral comparisons between MIP and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ...................... 36 Figure 9 Spectral comparisons between ACOLITE-T-Mart and in situ data in the 21 globally distributed AERONETOC sites. The variability across the dataset in the mean spectra of ACOLITE-T-Mart is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ....................................................................................................................................................... 36 Figure 10 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. .......................................................................................... 38 Figure 11 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β;) for all the match-ups combined. The dashed line corresponds to a 30% threshold [18]. ............................................ 39 Figure 12 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined (equal size samples). The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. .................................................. 40 Figure 13 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β;) for all the match-ups combined (equal size samples). The dashed line corresponds to a 30% threshold [18]. ............. 41 ACIX-III Aqua report 6 Figure 14 Spectral comparisons between PRISMA L2C and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of PRISMA L2C is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ... 42 Figure 15 Spectral comparisons between ACOLITE and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ...................... 43 Figure 16 Spectral comparisons between hGRS and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of hGRS is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ................................ 44 Figure 17 Spectral comparisons between POLYMER and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ...................... 44 Figure 18 Spectral comparisons between iCOR and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of iCOR is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ................................ 45 Figure 19 Spectral comparisons between MIP and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ................................ 46 Figure 20 Spectral comparisons between ACOLITE-T-Mart and hyperspectral in situ data in the 8 sites of the CVD. The variability across the dataset in the mean spectra of ACOLITE-T-Mart is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ................................................................................................................................................................... 46 Figure 21 Overall performance of AC processors using the CVD match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. ............................................................................................................... 48 Figure 22 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β) for all the match-ups combined. The dashed line corresponds to a 30% threshold. ........................................................ 48 Figure 23 Overall performance of AC processors using the CVD match-ups with all the satellite data combined (equal size samples). The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. .......................................................................... 49 Figure 24 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β) for all the match-ups combined (equal size samples). The dashed line corresponds to a 30% threshold [18]. .................. 50 Figure 25 Performance assessments as determined by the spectral Median Symmetric Accuracy (ε) and Bias (β) for the AC methods. The dashed-dotted line corresponds to the 30% accuracy threshold suggested by GCOS and adopted also by ACIX-Aqua [18]. ............................................................................................................ 50 Figure 26 Mean values of PRISMA-derived Rrs spectra for each AC method, by OWT. For in situ data, AERONETOC and CVD have been aggregated and plotted as mean values with standard deviations, from 443 to 667 nm. N provides the number of averaged in situ spectra per OWT. ............................................................................. 52 Figure 27 Aggregation of pairwise inter-comparisons (heatmaps) obtained from the MdSA (%) reported in each cell. Processors with lighter colours (white/yellow) are likely to generate high quality for a given OWT and band. ................................................................................................................................................................... 54 ACIX-III Aqua report 7 Figure 28 An example of QWIP analysis output for a scene at Ariake Tower AERONET-OC site on May 11, 2020, processed using the hGRS AC approach. (A) Map of AVW, (B) Map of the QWIP score, (C) scatter plot of QWIP analysis, and (D) frequency distribution of incremental QWIP scores. ............................................................. 57 Figure 29 An example of QWIP analysis output for a scene at Ariake Tower AERONET-OC site on May 11, 2020, processed using the hGRS AC approach. These plots display the mean normalised spectral shapes meeting various criteria (A) |QWIP| < 0.2, (B) 0.2 > |QWIP| > 0.4, (C) QWIP > 0.4, (D) QWIP < -0.4, and (E) out of range water type. ........................................................................................................................................................... 58 Figure 30 Results obtained for a Lake Trasimeno scene considering all five AC models (A: ACOLITE, B: hGRS, C: POLYMER, D: iCOR, E: MIP) and T-Mart product (F). On the left it is shown the QWIP score map and on the right the scatterplots of the QWIP analysis. ........................................................................................................... 60 Figure 31 Dataset of the globally distributed sites and number of available match-ups between EnMAP and field data. The colour of the dots indicates the number of available match-ups ranging from a minimum of 1 to a maximum of 13 for every study area. The Location of AERONET-OC are indicated with circles only, the black dots indicate CVD sites. AAOT (Venice) is both Hyperspectral (CVD sites) and AERONET-OC. ................................. 62 Figure 32 Frequency distribution of reflectance levels the hyperspectral (CVD) and multispectral in situ measurements (all wavelengths). ................................................................................................................. 65 Figure 33 Spectral comparisons between EnMAP L2A water products (MIP) and in situ data in the 11 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ....................................................................................................................................................... 66 Figure 34 Spectral comparisons between ACOLITE and in situ data in the 9 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ........ 67 Figure 35 Spectral comparisons between POLYMER and in situ data in the 10 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ........ 68 Figure 36 Spectral comparisons between PACO-WASI and in situ data in the 12 globally distributed AERONETOC sites. The variability across the dataset in the mean spectra of PACO-WASI is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ... 69 Figure 37 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in Table 11 along with the statistics. The black lines refer to the 1:1 line. ............................................................................................................... 70 Figure 38 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and median symmetric Bias (β) for all the match-ups combined. The dashed line corresponds to a 30% threshold [18]. ...... 72 Figure 39 Spectral comparisons between EnMAP L2A standard products (MIP) and hyperspectral in situ data in the 6 sites of the CVD. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ....................................................................................................................................................... 74 Figure 40 Spectral comparisons between ACOLITE and hyperspectral in situ data in the 5 sites of the CVD. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ........................... 75 Figure 41 Spectral comparisons between POLYMER and hyperspectral in situ data in the 5 sites of the CVD. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue ACIX-III Aqua report 8 area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ........................... 76 Figure 42 Spectral comparisons between PACO-WASI and hyperspectral in situ data in the 6 sites of the CVD. The variability across the dataset in the mean spectra of PACO-WASI is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ........ 77 Figure 43 Spectra of ϵ, β, RMSE and MdAPE for MIP, Polymer, ACOLITE and PACO-WASI for coincident hyperspectral match-ups during the operational mission phase (N=7). The grey shaded area represents the required uncertainty (RMSE) outside of strong atmospheric absorption regions as defined by the EnMAP Ground Segment and for AOT at 550 nm lower than 0.4 [22]. ...................................................................................... 78 ACIX-III Aqua report 9 LIST OF TABLES Table 1 A section of traceability matrix in the ‘Inland and Coastal’ water market area as defined by the CHIME MRD (cf. Table 4-1) showing link between user needs and requirements, defined on the basis of high-level EU policies and the CHIME mission requirements............................................................................................... 13 Table 2 In situ data used in the study, percentage of location in inland and coastal waters (the % are computed with respect to the number of sites for each of the instruments). * highlights in situ data only used for matchups in the EnMAP data evaluation. ...................................................................................................................... 17 Table 3 Table 3 Main characteristics of the AC models and T-Mart model. Notably ACOLITE, POLYMER and MIP are applied to both PRISMA and EnMAP; hGRS, iCOR and T-Mart only to PRISMA; PACO-WASI only to EnMAP. .. 18 Table 4 Band settings in the different datasets: the nominal band used in the match-up analysis and the corresponding band settings of AERONET-OC and PRISMA. ........................................................................... 26 Table 5 Conceptual description of OWTs [79]. ............................................................................................... 28 Table 6 Band names ..................................................................................................................................... 37 Table 7 AC models performance by wavelength and OWT (including also the T-Mart model). ............................ 54 Table 8 Average SA results for each OWT class. N indicates the size of each class. .......................................... 55 Table 9 List of study sites, location, type of in situ measurements and number of images (C: Commissioning, O: Operational mission phase, LQ: Low Quality). ............................................................................................... 61 Table 10 Band settings in the different datasets: the nominal band used in the match-up analysis and the corresponding band settings of AERONET-OC and EnMAP. The first part of the table refers to the band setting used for the entire dataset, while the second part refers to the additional bands considered for the CVD dataset only. ............................................................................................................................................................ 64 Table 11 Summary of the five statistical metrics for the EnMAP matchup data shown in Figure 37. N represents the total number of samples for each band. .................................................................................................. 71 Table 12 Statistics of MIP, Polymer, ACOLITE and PACO-WASI for two wavelength ranges using the common hyperspectral dataset (both mission phases). ............................................................................................... 78 ACIX-III Aqua report 16 mission is much younger than PRISMA, less matchups with in situ data are available and only four different AC methods are considered in this report. The evaluation of the data by in situ data takes advantage mostly of the same sites as used for the PRISMA evaluation. The data, methods, results and discussion of the EnMAP AC evaluation is based on the study by [22] which focuses on 2022-2023 matchups only and was complemented by the recently developed atmospheric correction method PACO-WASI. In addition to the presented Section 2, the report is composed by Section 3, which presents the in situ and satellite data and the AC models applied to both PRISMA and EnMAP. In Section 4 and Section 5 the report presents the results of PRISMA and EnMAP separately, for the sake of clarity. Section 6 is based on the conclusions and combining the results of PRISMA and EnMAP. ACIX-III Aqua report 17 4 MATERIALS AND METHODS 4.1 IN SITU DATA In order to validate satellite mission data, it is essential to have a large amount of in situ data, collected at the same time as the satellite overpasses [18]. This section introduces the in situ data used to carry out this work: global networks, fixed radiometer data and fieldwork data, as presented in Table 2. The data are presented similarly to those provided in ACIX-Aqua, for purposes of continuity across the two exercises. Table 2 In situ data used in the study, percentage of location in inland and coastal waters (the % are computed with respect to the number of sites for each of the instruments). * highlights in situ data only used for matchups in the EnMAP data evaluation. Network In situ instrument Example of instrument use Characteristics Location and representativeness of the system AERONET-OC CIMEL CE318T [23] Fixed radiometer, multispectral data 85% coastal waters 15% inland waters COMMUNITY VALIDATION DATABASE (CVD) WISPStation [11] Fixed radiometer, hyperspectral data 100% inland waters PANTHYR [24] Fixed radiometer, hyperspectral data 100% coastal waters THETIS [25] Fixed radiometer, hyperspectral data 100% inland waters HYPSTAR [26] Fixed radiometer, hyperspectral data 67% coastal waters 34% inland waters WISP-3 [4] Ad-hoc campaigns, hyperspectral data 75% inland waters 25% coastal water Hyper-Pro* [27] Ad-hoc campaigns, hyperspectral data 100% oligotrophic water Satlantic HyperOCR* [28] Fixed radiometer, hyperspectral data 100% coastal water OOSS* [22,29] Ad-hoc campaigns, hyperspectral data 100% inland waters 4.1.1 AERONET-OC NETWORK Several protocols can be followed for validation activities, the most common of which is the global multispectral network AERONET-OC [23,30]. The network covers over 40 aquatic sites worldwide, each with different optical characteristics. The main advantages of the network are autonomous operation and data availability within hours of collection, high quality and consistency of data due to standardised collection methodology, annual instrument calibrations and data re-processing techniques, and open access products through a specific data policy. AERONET-OC provides measurements of normalised water radiance (LWN) measured by modified CIMEL (Paris, France) CE318T solar photometers installed on fixed offshore platforms, multispectral data in the spectral range 400-1020 nm (~10 nm bandwidth). Previously, AERONET-OC supported the radiometric characterisation of Landsat-8 OLI and Sentinel-2 MSI for aquatic applications, with the previous ACIX- ACIX-III Aqua report 18 Aqua exercise [18]; in this report it is used for the assessment of PRISMA and EnMAP hyperspectral data. 4.1.2 COMMUNITY VALIDATION DATABASE The global AERONET-OC network valuably provides a global coverage of water sites (mostly marine, with lakes are still underrepresented) in a multi-spectral setting, while the availability of equivalent data in a hyperspectral band setting remains rather limited. As for ACIX-Aqua, in this work, the dataset collected by the CVD was hence exploited along with the AERONET/OC. The CVD is the result of a collaboration of different scientific teams that provided in situ hyperspectral data to validate the different AC models. In particular, the following fixed position autonomous radiometers deployed to collect either above or underwater measurements were used in the study: WISPStation [11], Hypstar [26,31], Panthyr [24,32], HyperOCR [28] and Thetis [25]. Furthermore, CVD includes also data collected by the following hand-held spectroradiometers used during fieldworks: Hyper-Pro [27], OOSS [29] and WISP-3 [4,33]. All data were supplied by providers in physical units of above water Rrs (sr-1). 4.2 ATMOSPHERIC CORRECTION MODELS This section presents the characteristics of the different AC models. A summary of the main features is given in Table 3. Table 3 Table 3 Main characteristics of the AC models and T-Mart model. Notably ACOLITE, POLYMER and MIP are applied to both PRISMA and EnMAP; hGRS, iCOR and T-Mart only to PRISMA; PACO-WASI only to EnMAP. AC model ACOLITE hGRS POLYMER iCOR MIP PACO-WASI T-Mart (adjacencyeffect correction only) Gaseous O2, O3 (OMI) O2, O3, O4, NO2, CO, CO2, CH4 (from CAMS) O3 and NO2 O2, O3, NO2, CO, CO2, CH4 (O3 climatology) O2, O3, CO, CO2, NO, NO2, CH4, SO2 H2O, O3, CO2, CO, CH4, N2O, O2, NH3, NO, NO2, So2, HNO3 MERRA2 Water vapor NCEP ECMWF/CAMS No ECMWF (OLI) [34] No Yes [34] MERRA2 Sun-glint Fit to residuals at ρrc(1609) & ρrc(2200) Optimal estimation from SWIR bands Treated as bulk signal Subtraction of minimum Implicit [35] No Sky-glint [36] through radiative transfer LUT No [37] [35] No No Adjacency effects No No No SIMEC [38] [39] Yes [40] [41] Aerosol Dark target approach (area-based) Optimal estimation with OPAC models Polynomial fitting (perpixel) Dark target and AOT multiparameter inversion Pixel-wise coupled retrieval Dark Dense Vegetation and dark target approach (area / scene based) MERRA2 ACIX-III Aqua report 19 (areabased) Rayleigh LUT 6SV [42] Coupled Rayleighaerosol diffuse radiance LUT SOS [43] MODTRAN 5.0 [44] Contained in fully coupled model MODTRAN 5.4 [44] Not applicable Geometry Scene center for OLI and 5-km grids for MSI Per-pixel Per-pixel Per-pixel Per-pixel Per-pixel Scene center Aerosol model Continental/maritime aerosols [42] [42] OPAC No MODTRAN rural models [44] modified Shettle Fenn MODTRAN rural models [44] Linearly mixed continental/maritime aerosols [42] Cloud masking ρt(1609) > 0.0215 Thresholds on several spectral normalized indexes and bands around 2200 nm ρt (865) > 0.2 Cloud mask layers are provided [45] Decision Tree based on spectral and spectral indice thresholds Thresholds on several spectral normalized indexes Not applicable Output grid cell size (m) 10 30 10/20/60 60 30 Same as L1 data Same as input Assumptions on bio-optical conditions No No Yes [46] No No Yes [47] No Version 20231023.0 - - v3.2 2.3.798 EnMAP 01.05.02 (PACO), 7 (WASI) 2.2.2 Open source access Yes (ACOLITE) No (Under construction) Yes (POLYMER ) No No No (PACO) Yes (WASI) Yes (T-Mart) Organizations RBINS Magellium HYGEOS VITO EOMAP DLR University of Ottawa References [48,49] [50] [51,52] [45] [39] [47,53,54] [41] 4.2.1 ACOLITE ACOLITE, developed at Royal Belgian Institute of Natural Sciences, is a software developed for the AC of satellite images for aquatic applications. It is appliable to images acquired by numerous satellite sensors, among which Landsat 8/9 [55], Sentinel-2 MSI [48,56,57] and Sentinel-3 OLCI [58] and to data acquired by several hyperspectral missions, including PRISMA [24] and EnMAP [22]. ACOLITE is available on GitHub 2 and in this study the version 20231023.0 has been used, without and with the option of glint correction. ACOLITE is based on the Dark Spectrum [48,57] in which multiple dark targets in the scene or subscene are chosen to construct a dark spectrum. This is then used to estimate the AOT at 550 and atmospheric path reflectance according to the best-fitting aerosol model. The AOT and aerosol type were imposed to be fixed over the 30 × 30 km PRISMA acquisition. ACOLITE used the Continental and Maritime aerosol models from 6SV. For ACOLITE DSF processing both a PRISMA L1 image and the matching L2C data are required as inputs. At runtime, band specific 2 https://github.com/acolite/acolite ACIX-III Aqua report 20 Gaussian relative spectral response functions are generated for both VNIR and SWIR detectors, using the central wavelength and FWHM information provided in the HDF5 metadata. Geometry information is currently not included in the L1 data, and hence the per-pixel sun and view geometry (i.e. sun and view zenith and relative azimuth angles) is extracted from the matching L2C file. Rrs is output by dividing the corrected water reflectance by π. Hereafter we will refer to Rrs obtained from the ACOLITE AC of PRISMA L1 as ACOLITE DSF. In addition to the DSF procedure, ACOLITE has an optional imagebased sun glint correction [48]. This correction is performed by estimating the interface reflectance signal in the SWIR from the average ρs between 1500 and 2400 nm - i.e. assuming zero water leaving radiance in this spectral range. To extend this average SWIR observation towards the VNIR, the interface reflectance is modelled with OSOAA [59] for the current scene average sun-sensor geometry, the estimated aerosol model and AOT for a high (20 m s-1) wind speed. Remote sensing reflectance is output by dividing the air–water interface reflectance corrected water reflectance by π. 4.2.2 hGRS The processor hGRS is a hyperspectral adaptation of the previous Glint Removal for Sentinel-2-like (GRS) algorithm which was developed to handle proper correction for the sunglint signal [50] and was evaluated through the previous intercomparison exercise ACIX-Aqua [18]. The new hyperspectral algorithm hGRS provides a two-step algorithm with (i) a correction for the gaseous absorption followed by (ii) a coupled correction of the atmospheric effect (with aerosol estimation) and the reflection of the sun on the water surface (sunglint). Note that this correction for sunglint is of utmost importance for high-spatial resolution and nadir-viewing sensors such as PRISMA for which the sunglint signal might be severely pronounced. The proposed implementation for the gaseous absorption is based on the LibRadTran package. The hyperspectral atmospheric transmittance was precomputed at 0.1-nm spectral resolution for various concentrations of each absorbing gas (O3, O2, water vapor, CH4, CO2…). Based on those pre-computed transmittances the final gaseous transmittance is computed from the gas concentration values taken from the CAMS database and then convoluted with the spectral response functions of the PRISMA sensor. As to the aerosol component of the signal, the radiative modeling is based on the models from the Optical Properties of Aerosol and Clouds (OPAC), version 4. Those aerosol models are built upon the OPAC database [60] encompassing spectral complex refractive index, size distribution, hygroscopicity changes and nonspherical particles. Aerosol models are generated based on a mixture of pure single components. One advantage of those models is to consider the spectral variation of the complex refractive index of the individual aerosols. the scattering matrix based on Mie theory for homogeneous sphere, T-matrix for ellipsoidal/spheroidal particles; those computations were extended for large particle sizes with the “Improved Geometric Optics Method” [61]. The computations were performed through the MOPSMAP package (Modeled optical properties of ensembles of aerosol particles) based on pre-computed lookup tables [62]. These aerosol optical properties LUT were used to generate the intrinsic reflectance of the atmosphere and the total Rayleigh-aerosol transmittances through the full vector radiative transfer code OSOAA [59]. The aerosol content, water vapor and sunglint component are finally ACIX-III Aqua report 21 retrieved from the optimal estimation using the Levenberg-Marquardt approach providing the final information for AC and associated uncertainty pixel-wise. 4.2.3 POLYMER The POLYMER processor, developed by HYGEOS, is an AC algorithm purposely developed for retrieving the ocean colour when the observation is contaminated by the sun glint [51,52]. It has been developed for MERIS on Envisat and is currently applicable to images acquired by other sensors: MODIS Aqua, SeaWiFS, VIIRS, Sentinel-3 OLCI, Sentinel-2 MSI, PRISMA, HICO and EnMAP. POLYMER is available on GitHub 3 and in this study the version 4.17beta2 has been used. The POLYMER algorithm uses the image bands in the whole spectral range from the blue to the near infrared, to decouple the atmospheric and surface components of the signal from the water reflectance. The bands near strong atmospheric absorption features are filtered out. The algorithm relies on two models, an analytical model for the atmosphere which uses a second-order polynomial formulation with respect to the wavelength, and a water reflectance model based on two parameters, one for the chlorophyll concentration and one for the backscattering coefficient. The latter is based on the biooptical model of [46]. Water-leaving reflectance is further normalised to the nominal wavelength and for the water-leaving reflectance bidirectional effects (BRDF), such that the radiometric output of Polymer is the fully normalised water-leaving reflectance spectrum corrected for bidirectional effects. Detailed information on the pre-correction (e.g., estimation of the gaseous transmittance, Rayleigh scattering, and pre-glint correction), spectral matching, and the atmospheric model was presented in [51,52]; information on the water-leaving reflectance model can be found in [46] and its modifications, water-leaving reflectance normalisation, and BRDF correction were described in [52]. Rrs is output by dividing the corrected normalised water-leaving reflectance by π. 4.2.4 iCOR iCOR (image CORrection for atmospheric effects) is a MODTRAN-5-based AC tool designed to process satellite data collected over coastal, inland, or transitional waters and land [45]. Basic iCOR versions for Landsat-8, Sentinel-2, and OLCI-Sentinel-3 are publicly available as free plugins for the SNAP toolbox. For ACIX-III-Aqua, an advanced version of iCOR has been developed to enable the correction of hyperspectral PRISMA data. Starting from PRISMA TOA data, iCOR identifies land, water, and cloud pixels through a sensor-specific thresholding approach. The AOT values are then retrieved from the image using iCOR’s image-based AOT retrieval method. When cloud cover is too extensive, or the scene lacks sufficient spectral variability for accurate AOT retrieval, an automatic fallback mechanism uses Near-Real Time (NRT) AOT values from the Copernicus Atmospheric Monitoring Service (CAMS). Water vapor retrieval is performed directly from PRISMA imagery using the Atmospheric Pre-corrected Differential Absorption (APDA) approach [34]. Additionally, SIMEC adjacency correction [38] is applied to minimize adjacency effects, especially near land-water 3 https://github.com/hygeos/polymer ACIX-III Aqua report 22 boundaries. Following AOT retrieval and adjacency correction, spectral bands are atmospherically corrected using pre-calculated MODTRAN-5 look-up tables (LUTs). These LUTs incorporate (1) solar and viewing geometry, (2) terrain altitude from the GLOBE DEM, (3) the retrieved AOT values, (4) water vapor concentrations, and (5) monthly ozone climatology data from the Total Ozone Mapping Spectrometer (TOMS). Finally, for water pixels, a Fresnel reflectance correction is applied, while land pixels are modelled as Lambertian surfaces. To address uncertainties in Fresnel correction related to glint and haze, the minimum water-leaving reflectance in selected NIR-SWIR bands is calculated for each water pixel. This value, if greater than zero, is subtracted from the retrieved water-leaving reflectance values across all spectral bands. 4.2.5 MIP MIP is an operational processing chain that is in principle sensor-independent. It is optimized to most accurately predict in water properties and uses a coupled atmosphere-water model stored in a lookup-table. An inversion algorithm searches for the most likely set of model parameters that explain the observed sensor radiances. Those parameters include surface altitude, geometry, aerosol properties, water ingredient concentrations and water surface roughness. Except for the adjacency correction, the parameters are retrieved pixel-wise. Reflectance products are not directly used in the retrieval process, but are a side product that can be reconstructed using the retrieved atmospheric and water surface properties and propagating those through the radiative transfer simulation. To produce water reflectance data, the chain creates a pixelwise classification into land/water/cloud. It removes the adjacency effect that land pixels have on water and then retrieves all relevant parameters for each water pixel. From those parameters the reflectance products can be created using the method mentioned above. 4.2.6 PACO-WASI For the EnMAP satellite data only, a combination of two software tools, PACO and WASI, was applied. The EnMAP Level-2A data can be downloaded as water product, processed using MIP, as land product, processed using PACO, or as a combined product. PACO-WASI refers to the postprocessing of the land product with the software WASI to convert surface reflectance to remote sensing reflectance. PACO (Python-Based Atmospheric Correction, [53]) is the Python version of the software package ATCOR [40,63] which is optimized for the atmospheric correction of multiand hyperspectral satellite and aerial images over land surfaces. It makes use of precomputed lookup tables, generated with Modtran 5.4, and accounts for ground effects such as terrain elevation and adjacency effects. The resulting Level-2A product is Hemispherical Directional Reflectance (HDR) at bottom of atmosphere, commonly called surface reflectance. WASI (Water Color Simulator) is a software designed for the simulation and data analysis of spectral measurements in and above water. It can be downloaded from [54]. Originally developed for field spectrometers [64], WASI has since been adapted to process multiand hyperspectral image data over water from any sensor [47]. The input ACIX-III Aqua report 23 data must be atmospherically corrected and can be in units of reflectance or radiance. The simulation of remote sensing reflectance is based on the analytical models of [65] for optically deep and shallow waters, which had been developed using radiative transfer simulations with Hydrolight [66]. The reflections at the water surface are simulated using the model of [67] which decomposes the sky radiance reflected in sensor direction into three spectrally different components originating from the sun (sun glint), scattering at molecules (Rayleigh glint) and scattering at particles (aerosol glint). The wavelength-dependent glint components are parameterized in terms of aerosol optical thickness, Angström exponent of aerosol scattering, water vapor and ozone scale height using the irradiance model of [68]. All model parameters are experimentally accessible and all files specifying the optical properties of the components in the water and the atmosphere can be exchanged easily, hence WASI allows the simulation of measurements for a wide range of environmental conditions. Data analysis applies inverse modelling (IM) to spectral measurements, and in case of image data, it additionally employs a neural network (NN) that is trained image-specifically using the IM results from a representative subset of several hundred pixels. The flexibility of WASI allows a regional fine-tuning of the simulations and of data processing, which can provide more accurate results compared to global algorithms, yet the optimization of IM for the actual conditions requires expert knowledge. Since the same physical processes are responsible for path radiance and sky glint, IM cannot distinguish between path radiance errors from atmospheric correction and sky glint, which has the advantage for the derivation of remote sensing reflectance that both effects are corrected together. The glint correction algorithm is described in [35]. In the combined PACO-WASI workflow, PACO is first applied to correct the atmosphere, followed by WASI, which processes the surface reflectance output of PACO to correct for sun glint, sky glint and path radiance errors. It is important to note that PACOWASI is not yet an operational workflow as the step of inverse modelling requires manual fine-tuning of WASI for each individual scene by an experienced user. This tuning process involves iterative adjustments to the IM settings – including the selection of fit parameters, initial values of fit parameters, and spectral range used for data processing – until two criteria are fulfilled: (1) the modelled spectra fit the surface reflectance spectra as close as possible, (2) the noise caused by spectral ambiguities is minimized. Spectral ambiguities, which are a common problem in optically complex waters [69], are identified using correlation plots of pixels that have been processed using both IM and NN approaches; these diagnostic plots are generated automatically by WASI. An automatization of these fine-tuning steps is currently under development. 4.2.7 ACOLITE-T-Mart As mentioned earlier, in addition to the five AC models, the correction of the adjacency effect using TMart combined with ACOLITE was included. In particular, in this exercise, a pre-processing step using T-Mart [41,70] was applied to correct for the adjacency effect at the TOA level, followed by ACOLITE AC (version: 20231023) with the same setting as the ACOLITE-only implementation. T-Mart adjusts the TOA reflectance for each spectral band to satisfy the homogeneous-surface assumption, i.e. pixels are corrected to the TOA reflectance they would exhibit if each pixel were surrounded by pixels of ACIX-III Aqua report 24 identical reflectance. The correction process begins by calculating the atmospheric point spread function (PSF) specific to an image’s solar-sensor geometry, followed by convolving the image with the PSF as the kernel. The difference between the original and convolved images is scaled, based on radiative transfer simulations, by a factor that approximates the ratio of upward diffuse to upward direct transmittance. The scaled difference is then subtracted from the original image, followed by a matrix multiplication that accounts for variations in surface-level irradiance due to surface heterogeneity using a scene-specific look-up table, to finalize the adjacency effect correction [41]. When computing the PSFs and scaling factors in T-Mart, the atmospheric composition, including ozone, water vapor, and aerosol properties from the GMAO MERRA2 reanalysis data [71], is used to characterize the optical properties of the atmosphere at various altitudes via the 6S radiative transfer model [72]. ACIX-III Aqua report 25 5 PRISMA DATA ANALYSIS PRISMA (PRecursore IperSpettrale della Missione Applicativa) is an advanced Earth observation satellite funded and operated by the Italian Space Agency (ASI). Launched on 22nd March 2019 aboard the Vega rocket, PRISMA orbits in a low Earth Sun-synchronous orbit at an altitude of approximately 614.8 km [24,73–77]. It is designed to revisit the same area every 29 days, though this period can be reduced to less than one week using off-nadir pointing capabilities. This flexibility allows PRISMA to capture specific areas more frequently, making it suitable for diverse environmental monitoring applications. The PRISMA payload is a combination of two key instruments: a hyperspectral sensor with a Ground Sampling Distance (GSD) of 30 meters, and a panchromatic (PAN) camera with a GSD of 5 meters. The hyperspectral sensor captures data in 238 contiguous spectral bands ranging from 400 to 2500 nm, enabling precise detection of features across the visible, nearinfrared (VNIR), and short-wave infrared (SWIR) regions. This imaging spectrometer creates a hyperdata cube, allowing for detailed analysis of spectral and spatial features in a wide range of environmental contexts. Meanwhile, the PAN camera provides high-resolution context for the hyperspectral data, aiding in applications that require finer-scale mapping. PRISMA's system can acquire, downlinking, and archiving up to 200,000 km² of hyperspectral and panchromatic images daily, covering areas between 70°S and 70°N in latitude and from 180°W to 180°E in longitude. In a single pass, it can capture images up to 1000 km apart by adjusting its viewing angles. This flexibility enhances its temporal resolution, which is crucial for tracking dynamic environmental changes over time. PRISMA products range from Level 0 (L0) to Level 2 (L2): L0 represents formatted data products, Level 1 (L1) represents radiance data radiometrically corrected and calibrated in physical units, and L2 represents the result of the conversion of Top-of-Atmosphere (TOA) spectral radiance measurements into Bottom-of-Atmosphere (BOA) remote sensing reflectance measurements. L2 data are divided into: ● L2B (geolocated ground spectral radiance product); ● L2C (geolocated at-surface reflectance product); ● L2D (the geocoded version of L2C products). The AC process is described in section 2.2.1 PRISMA mission is recognized for its potential in various domains, including agriculture, forestry, mining, and aquatic remote sensing. Its data is particularly valuable for algorithm development and innovative environmental monitoring techniques. 5.1 FEATURES OF THE PRISMA DATASET USED IN THE EXERCISE The data used in the exercise are the PRISMA L1 data (version 4.1-0) and PRISMA L2C data (version 02.05). The quality of L1 data was assessed in [73] by comparison with TOA radiances simulated with a radiative transfer code, from in situ measurements, obtained from a series of fixed-position autonomous radiometers. The results showed that PRISMA provides TOA radiances with the same amplitude and shape as those simulated in situ with slightly larger uncertainties at shorter ACIX-III Aqua report 32 ● In the Gustav Dalen Tower and Irbe Lighthouse sites, the spectral signatures are similar: peak Rrs (< 0.005 sr-1) around 550 nm, which then tends to decrease to zero. ● In the Galata platform site, the spectrum is associated with optically complex waters, with Rrs maxima typically observed in the 490-560 nm range, properties common to the Section 7 site). ● In the Lucinda site (in the tropical coastal waters of the Great Barrier Reef lagoon near the Herbert River estuary), the spectrum shows a peak just before 550 nm and a standard deviation of ± 0.005 sr-1. PRISMA L2C In all cases, an overestimation of the Rrs PRISMA L2C spectra compared to the in situ data can be observed; better agreement between satellite data and in situ data is seen in more eutrophic waters, where the water-leaving signal is higher. Thus, the lower SA value (8.92°) was identified in the case of the Bahia Blanca site. In more clear waters where there is a low signal, PRISMA's performance is often sub-optimal, as has also been reported in other studies [24]: mean SA ~ 18°. For the Lake Erie site, the Rrs PRISMA spectra presents the highest standard deviation (±0.015 sr-1). In Lake Okeechobee, where pronounced absorption and reflection peaks are observed, there is a low level of agreement (SA=32.86°). The unrealistic spectral variability at some sites (e.g., Casablanca, Galata platform, Socheongcho, USC SeaPrism) should also be noted. This non-physical variability indicates the presence of image artefacts (e.g., scattered light), particularly in the blue part of the spectrum and at sites with low-magnitude spectra (i.e. blue or organic-rich waters). The SA ranges from 8.92° (Bahia Blanca site) to 47.49° (Palgrunden site). Figure 3 Spectral comparisons between PRISMA L2C and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of PRISMA L2C is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ACIX-III Aqua report 33 ACOLITE A comparison between ACOLITE and in situ data shows a good fit between the data. Particularly in the case of the Bahia Blanca (SA=2.93°), Casablanca platform (SA=8.21°), Lake Erie (SA=10.68°) and Zeebrugge (SA=9.37°) sites. The highest standard deviation values were recorded in the cases of LISCO (±0.002 sr-1), San Marco platform (±0.007 sr-1) and South Greenbay (±0.01 sr-1). The SA ranges from 2.93° (Bahia Blanca site) to 36.98° (Irbe Lighthouse site). Figure 4 Spectral comparisons between ACOLITE and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. hGRS Here, the products obtained by applying hGRS model to satellite data; SA was lower than 20° for the 62% of the study areas. There is high concordance in the NIR region, and some difference with in situ data in the blue region. The highest agreement was reached in Bahia Blanca (SA=5.1°), followed by Zeebrugge and Casablanca (both with SA=8.9°). Least agreement was noted in Kemigawa (SA=46.1°), while a good performance can be noted in the case of the South Greenbay site, where a hypereutrophic condition (super nutrient richness) is present. In San Marco platform and Lake Erie a high standard deviation was registered (±0.015 sr-1 and ±0.008 sr-1 respectively). ACIX-III Aqua report 34 Figure 5 Spectral comparisons between hGRS and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of hGRS is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. POLYMER In the case of POLYMER, there is a good level of agreement with in situ measurements, particularly in the case of the Ariake Tower, Casablanca platform, Lake Erie, Lucinda and Rio de la Plata sites, with mean SA = 3.5°. Only in a few cases is the blue spectrum somewhat noisy and there is an overestimation of the Rrs spectrum compared to the in situ data. In the case of Lake Okeechobee, the result of the comparison is weak. The SA varies from 2.54° (Casablanca site) to 73.33° (Lake Okeechobee site). Figure 6 Spectral comparisons between POLYMER and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ACIX-III Aqua report 35 iCOR The Rrs spectra derived from the correction with iCOR showed a good agreement for all the study areas with in situ data. The highest agreement was reached in Zeebrugge (SA=4.87°), followed by Bahia Blanca (SA=6.88°), and Ariake Tower (SA=7.34°), while lower concordance was found in Palgrunden (SA=31.01°). In Lake Erie, LISCO and South Greenbay the reflectance peak, due to the eutrophic conditions of the water, can be noticed (0.007 sr-1, 0.003 sr-1 and 0.012 sr-1 respectively). A standard deviation of about ±0.010 sr-1 was registered in San Marco Platform and Lake Okeechobee. Figure 7 Spectral comparisons between iCOR and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of PRISMA iCOR is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. MIP The Rrs spectra derived from the correction with MIP showed a good agreement for all the study areas with in situ data (SA lower than 20° for the 75% of the study areas), particularly in the blue region, although this area, together with the NIR spectral region presented some unusual peaks (around 700 nm). The spectral feature at 760 nm is probably due to the oxygen absorption. The SA ranges from 5.02° (Zeebrugge site) to 33.01° (Palgrunden site). In San Marco platform high standard deviation was registered (±0.009 sr-1). ACIX-III Aqua report 36 Figure 8 Spectral comparisons between MIP and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ACOLITE-T-Mart Similar results (respect to ACOLITE) obtained from the ACOLITE-T-Mart model, there is a slight improvement at sites closer to the coast, i.e. those that may be more subject to the adjacency effect. In fact, the improvement is most noticeable in the cases of LISCO and South Greenbay (distance from the coast of 1.6 and 0.5 nautical miles), SA improved by about 20%. The SA ranges from 2.81° (Bahia Blanca site) to 32.00° (Irbe Lighthouse site). Figure 9 Spectral comparisons between ACOLITE-T-Mart and in situ data in the 21 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of ACOLITE-T-Mart is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in orange. SA represents the Spectral Angle and N represents the number of the images. ACIX-III Aqua report 37 5.3.2 OVER THE SPECTRAL BANDS After performing the spectral comparison, satellite and in situ data from all 21 sites were aggregated by AC model and resampled to the spectral configuration of AERONET-OC, selecting the PRISMA bands closest to the corresponding AERONET-OC bands. Table 6 shows the name assigned to each corresponding band that is used from now on. Table 6 Band names Name b_443 b_490 b_560 b_620 b_667 b_709 Band 443 nm 490 nm 560 nm 620 nm 667 nm 709 nm Figure 10 shows the scatterplots in the 6 selected bands together with the sample size value (N) and statistics (RMSD, MAD, MAPD, Mean bias, Slope). In general, more dispersion was noted in the UV/blue region, particularly at b_443 (RMSD ranged 0.0042 - 0.0128 sr-1) and b_490 (RMSD ranged 0.0030 - 0.0101 sr-1). Good fitting at b_560 (RMSD ranged 0.0023 - 0.0073 sr-1) and b_620 (RMSD ranged 0.0018 - 0.0076 sr-1). Slight dispersion in the case of the red band at b_667 (RMSD ranged 0.0014 - 0.0065 sr-1). The dispersion at b_709 may also be due to the presence of few samples (RMSD ranged 0.0019 - 0.0099 sr-1). For all bands POLYMER showed the lowest RMSD values, followed by MIP (better performance at b_443). Slightly higher RMSD values in the case of ACOLITE-T-Mart and hGRS (lowest values at b_490). The sample size (N) changes at different wavelengths due to the different configurations of the in situ instruments; for the same wavelength, it also changes depending on the model due to the presence of negative values (removed from the analysis) and the complete nonreturn in output of the product. ACIX-III Aqua report 38 Figure 10 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. (Image adapted from [79]). Figure 11 shows the two statistics MdSA and Bias used previously in the ACIX-Aqua project paper in bar plot format. In general, the noisiest band, where the highest error values were recorded, was b_443, while a lower error was found at b_560 and b_620. Lower values of MdSA and Bias, hence better agreement between the data, were obtained in the case of MIP. Furthermore, for almost all bands MIP showed MdSA values slightly higher than the 30% threshold. Higher uncertainties shown in the case of PRISMA. ACIX-III Aqua report 39 Figure 11 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β;) for all the match-ups combined. The dashed line corresponds to a 30% threshold [18]. (Image taken from Figure 4 [79]). To perform a more robust analysis, the statistical comparison was also carried out by considering samples with equal size: to do this, the records of each model were removed when even one model had negative values or did not provide the output. In this case, POLYMER showed again the lower mean bias, followed by MIP (from b_443 to b_667) and hGRS at b_709. Slightly higher values in the case of ACOLITE-T-Mart and iCOR; while the first one performed particularly well at b_620, b_667 and b_709, iCOR showed better results at b_443 and b_490. Overall, the error values reported in the bar plot (Figure 13) are slightly decreased and MIP has MdSA error values slightly under the 30% threshold. ACIX-III Aqua report 40 Figure 12 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined (equal size samples). The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. ACIX-III Aqua report 41 Figure 13 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β;) for all the match-ups combined (equal size samples). The dashed line corresponds to a 30% threshold [18]. 5.3.2.1 CVD Figure 14 - Figure 20 show the spectral comparisons between satellite and in situ data for the seven CVD sites. The selected study areas are four lakes (three in Italy and one in Switzerland), Venice (lagoon and AAOT), and Rio de La Plata (Argentina). The study areas have significant optical and morphometric differences; for all of them, in situ radiometric data were available for calibration and validation purposes. For Lake Garda the analysis has been performed differentiating between optically deep and shallow waters (2 m average depth, close to Sirmione peninsula). A brief characterisation of the sites follows: ● Lake Garda (oligo-mesotrophic): is a deep subalpine glacial lake, where the Rrs spectrum is around 0.007 sr-1 with a small standard deviation; instead, in the shallow waters the Rrs spectrum is about double (peak around 550 nm) and has a larger standard deviation. ● Lake Geneve (oligo-mesotrophic): is a deep subalpine glacial lake, the Rrs spectrum is around 0.006 sr-1 with a standard deviation of about 0.004 sr-1. ● Rio de la Plata (eutrophic): is a basin estuary with a typical spectral signature of brown water, the Rrs peak is about 0.030 sr-1 (around 600 nm). ● Lake Trasimeno (meso-eutrophic): is a shallow tectonic lake, with the Rrs peak around 0.0025 sr-1 in the green region, absorption and reflectance peaks (around 670-700 nm) related to the presence of Chl-a are noticeable from the spectral signature. ● Lake Varese (meso-eutrophic): is a shallow glacial lake, presenting a spectral signature similar to Lake Trasimeno with slightly lower Rrs peak values. ● Venice (AAOT) (oligo-mesotrophic): is a transitional region between open sea and coastal waters, the Rrs peak is about 0.007 sr-1 (around 500 nm), the spectral signature tends roughly to zero after the reflectance peak. ● Venice Lagoon (meso-eutrophic): is a shallow coastal environment, the Rrs peak is about 0.020 sr-1, with high standard deviation (about 0.010 sr-1). ACIX-III Aqua report 48 Figure 21 Overall performance of AC processors using the CVD match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. (Image adapted from [79]) Figure 22 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β) for all the match-ups combined. The dashed line corresponds to a 30% threshold 5 . The statistical comparison was repeated with the same modalities considering samples with equal size, as did for AERONET-OC. The results can be found in Figure 23 and Figure 24. As in the previous analysis, POLYMER showed the lowest MAPD values in all the bands, except for b_709 where MIP (MAPD=37%) and hGRS (MAPD=42%) agreed more with in situ data. MIP is performing well also at b_ 620 and b_667 (MAPD=38%), while hGRS at b_443 and b_490 (MAPD=59% and 42% respectively). 5 https://gcos.wmo.int/en/essential-climate-variables/lakes/ecv-requirements ACIX-III Aqua report 49 Figure 23 Overall performance of AC processors using the CVD match-ups with all the satellite data combined (equal size samples). The number of matchups per processor and per band is reported in the scatterplots (N) along with the statistics. The black lines refer to the 1:1 line. ACIX-III Aqua report 50 Figure 24 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and Bias (β) for all the match-ups combined (equal size samples). The dashed line corresponds to a 30% threshold [18]. Deploying hyperspectral data, which provides a wider range of bands, a more detailed analysis was carried out considering the entire spectrum (Figure 25). Concerning all AC methods, higher errors occurred in the blue and red spectral regions, as previously demonstrated for both PRISMA (Pellegrino et al., 2023) and EnMAP [22]. PRISMA L2C has higher ϵ and β values than the other AC methods; POLYMER provides lower ϵ values in the blue band, while hGRS provides lower ϵ values in the red-NIR bands. Figure 25 Performance assessments as determined by the spectral Median Symmetric Accuracy (ε) and Bias (β) for the AC methods. The dashed-dotted line corresponds to the 30% accuracy threshold suggested by GCOS and adopted also by ACIX-Aqua [18]. (Image taken from Figure 6 [79]). 5.3.3 OPTICAL WATER TYPES The pairwise intercomparison strategy of in situ Rrs resulted in eight OWTs (selected among the 10 proposed by [80]) that were then used to classify the corresponding PRISMA-derived Rrs spectra from the AC methods (Figure 26). In the first OWT classes (OWT 2, 3a and 3b), representing blue waters with increasing loads of suspended and dissolved matters, the number of in situ Rrs spectra to be used in the pairwise intercomparison with PRISMA ranges from 10 for OWT 2 to 57 for OWT 3b. For OWT 2, the spectral shape of PRISMA-derived Rrs values generally conforms to the in situ measurements for all ACIX-III Aqua report 51 AC processors except PRISMA L2C. In the case of OWT 3a and 3b PRISMA-derived Rrs values exhibit systematic overestimation at wavelengths shorter than 450 nm for all AC methods. The most represented class of in situ Rrs was the OWT 4, with a higher predominance of OWT 4b (82 spectra) than OWT 4a (22 spectra). This class is typical of greenish water, with high phytoplankton biomass and with low reflectance at the shortest wavelengths due to absorption by particles and yellow substances. Generally, PRISMA-derived Rrs spectra maintain the features of the in situ data collection with some exceptions in the blue bands, mostly from PRISMA L2C. OWT 5 is also representative of green eutrophic water, with substantially higher phytoplankton biomass and a local Rrs maximum around 700 nm; this is also captured by PRISMA as corrected with all the AC methods. OWT 6 is instead characteristic of bright water with high detritus concentrations, which has a high reflectance determined by scattering. These waters are also captured by PRISMA, with some exceptions for MIP and ACOLITE-T-Mart, which underestimate the shortest wavelengths. OWT 7, typical for dark water dominated by absorption from high concentration of yellow substances, shows high divergence in spectral shape for PRISMA L2C and POLYMER, while the other AC methods provide spectra more uniform with the in situ spectra. ACIX-III Aqua report 52 Figure 26 Mean values of PRISMA-derived Rrs spectra for each AC method, by OWT. For in situ data, AERONET-OC and CVD have been aggregated and plotted as mean values with standard deviations, from 443 to 667 nm. N provides the number of averaged in situ spectra per OWT. (Image taken from Figure 7 [79]). ACIX-III Aqua report 53 OWTs were then used to build heatmaps (Figure 27) to illustrate per band and per OWT performances estimated via win rates, as computed by [18]. The cells of the heatmaps for each AC model have been colored according to the normalized index (Norm(ε)); the AC methods that can generate the highestquality products for a given OWT and a given band are highlighted with brighter colors. The absolute value of ε (%) is also shown in each cell for completeness. The heatmaps are computed only for the first five of the six pre-defined bands due to the low number of samples in the band at 709 nm. The PRISMA L2C provides the lowest performance across all OWTs and bands, while brighter colors generally emerge for MIP, POLYMER and ACOLITE-T-Mart. A closer inspection of the heatmaps reveals that, depending on both the OWT and the band, each AC method performed differently. For example, at 443 nm, hGRS showed the lowest value of ε (33%) for OWT 2, POLYMER for OWTs 3a, 3b, 4b, 5a and 6, iCOR for OWT 4a (22%) and ACOLITE-T-Mart for OWT 7 (9%). The Table 7 provides an overview of the method’s performance, based on ε, for each OWT in the five significant bands. ACIX-III Aqua report 54 Figure 27 Aggregation of pairwise inter-comparisons (heatmaps) obtained from the MdSA (%) reported in each cell. Processors with lighter colours (white/yellow) are likely to generate high quality for a given OWT and band. (Image taken from Figure 8 [79]). Table 7 AC models performance by wavelength and OWT (including also the T-Mart model). (Table taken from Figure 8 [79]). OWT 443 nm 490 nm 560 nm 620 nm 667 nm 2 hGRS hGRS ACOLITE MIP MIP 3a POLYMER MIP MIP ACOLITE-T-Mart MIP 3b POLYMER hGRS hGRS ACOLITE-T-Mart POLYMER 4a iCOR MIP POLYMER MIP POLYMER 4b POLYMER MIP hGRS hGRS MIP 5a POLYMER POLYMER POLYMER POLYMER ACOLITE-T-Mart ACIX-III Aqua report 55 6 POLYMER POLYMER POLYMER POLYMER POLYMER 7 ACOLITE-T-Mart ACOLITE-T-Mart MIP ACOLITE-T-Mart MIP Additionally, the following Table 8 shows the SA results for each OWT class. The mean SA for each class is considered, N indicates the class size. Table 8 Average SA results for each OWT class. N indicates the size of each class. AC Models 2 (N=10) 3a (N=26) 3b (N=57) 4a (N=22) 4b (N=82) 5a (N=16) 6 (N=14) 7 (N=12) PRISMA 14.3° 19.3° 23.2° 10.0° 26.4° 22.1° 14.4° 34.0° ACOLITE 31.4° 20.8° 18.9° 7.0° 20.1° 17.4° 21.5° 20.2° hGRS 9.2° 13.1° 19.6° 6.5° 22.4° 26.9° 11.9° 35.5° POLYMER 9.3° 12.0° 12.6° 10.3° 13.3° 18.5° 15.2° 74.4° iCOR 12.6° 14.9° 24.2° 7.3° 17.1° 14.7° 11.7° 18.4° MIP 10.3° 12.0° 14.9° 10.7° 16.1° 15.4° 18.0° 15.1° ACOLITE-T-Mart 30.5° 19.2° 19.3° 6.9° 18.1° 15.6° 17.5° 16.6° 5.4 INTERPRETING THE QWIP OUTPUT The intent of utilising the QWIP in the ACIX-III Aqua was not to quantify the comparative skill of AC approaches, but instead to provide a scene-by-scene conceptualization of spectral reflectance behaviour and a set of diagnostic tools to provide spatial/spectral context to algorithm performance. The QWIP does not provide metrics that convey the absolute accuracy of spectral returns, but it does convey whether the spectral returns are outputting reasonable shapes that resemble typical in situ reflectance patterns. A known caveat is that QWIP is largely insensitive to random noise or small residual atmospheric features in the spectra. There may be limitations in regards to the representativeness of the dataset used to develop the QWIP, however, there appears to be an underlying mechanistic rationale for optical properties to behave somewhat reliably (i.e. some relative covariance of absorption and backscatter) in natural waters, which likely explains the consistent tendency for the spectral shape of reflectance to closely follow the NDI (490,665) over a broad range of water types. Nevertheless, false positives and negatives may be present, and the spectral reflectance output should be considered in the interpretation of results. The supporting online materials provide nine figures for each independent scene and AC approach. Figure 28 (A, B, C, D) visually display results of the QWIP analysis, while Figure 29 (A, B, C, D, E) provide output of the extracted spectral reflectance matching a given quality control criteria defined by QWIP scores. Interpretation guidelines, limitations, and caveats of the results and figures are listed below. Detailed explanations of each plot are given as follows: ● Figure 28 (A) - Map of Apparent Visible Wavelength (AVW). The magnitude of values corresponds to the weighted harmonic mean of the reflectance spectra. Conceptually, this represents the relative balance point of the visible reflectance spectrum. Typically, far redshifted values (>580 nm) are characterised by high sediment/CDOM or very high biomass and far blue-shifted values (440 –450 nm) represent clear, oligotrophic conditions. Note, the ACIX-III Aqua report 56 colour-scale used in Figure 29 that of the AVW image, to help conceptualise where in the image the spectra are coming from. ● Figure 28 (B) - Map of the QWIP score. The Quality Water Index Polynomial itself (Figure 28 (C), black line) represents an empirical relationship between a 2-band Normalised Difference Index (NDI) and the AVW. The difference between a spectrum’s NDI and the QWIP equals its QWIP score. Spectra corresponding to lower absolute QWIP scores (|QWIP| ≤ 0.2) tend to exhibit features that are more closely aligned with in situ data (and are coloured in grayscale here) while higher absolute QWIP scores (red/yellow shading; |QWIP| > 0.2) are flagged as potentially suspect data. The threshold criteria are based on analysis of in situ data, however, these are arbitrary guidelines to aid in conceptualization and should not be considered absolute. ● Figure 28 (C) - This scatter plot displays how each data point in the image compares to the QWIP. Divergent values typically exhibit a proportional reduction in spectral quality. Using a decision tree based on reflectance values, each point is additionally classified as one of three water types [82]. Type I (blue-green), Type II (green), and Type III (brown) waters typically have AVW ranges of 440 –530 nm, 500 –600 nm, and 540 –600 nm, respectively. Spectra are independently flagged if they fall out of a given water type range (see Figure 29 (E)). While not a universal rule, spectra with higher scores (QWIP > 0.4; Figure 29 (C)) tend to exhibit more egregious spectral anomalies relative to lower scores (QWIP < 0.4; Figure 29 (D)). ● Figure 28 (D) - A histogram showing the frequency distribution of QWIP scores, binned in |0.1| increments. The total number of pixels falling into each bin are also shown. Spectral quality can vary along a continuum of QWIP scores, i.e. exercise caution in designating an absolute pass/fail criteria. ● Figure 29 (A) – The mean normalised spectral shapes corresponding to |QWIP| ≤ 0.2, as a function of AVW. For example, all spectra meeting the criteria of 550 nm < AVW < 551 nm & QWIP < |0.2| are averaged and displayed as one representative spectra, the spectra is colourcoded as AVW = 550 nm, and so forth for incremental AVW values ranging from AVW = 440 – 600 nm. Note, all spectra are normalised by the trapezoidal integration of the area under the reflectance curve prior to averaging. These QWIP criteria tend to yield spectral reflectance shapes that correspond most closely to in situ data, relative to higher absolute QWIP scores. This is not always the case, hence the availability of these figures as a qualitative check. These spectra correspond to the grey-scale areas in Figure 28 (B), and the colour coding of the spectra correspond to the AVW map in Figure 28 (A). ● Figure 29 (B) - The mean normalised spectral shapes corresponding to an intermediate range of QWIP scores (0.2 > |QWIP| > 0.4), as a function of AVW. Some of these spectra are still in the range of the developmental dataset for QWIP (i.e. they’re not all ‘bad’ spectra), but some obvious processing failures may start to present in this range. These are their own categories representing a mix of ‘good’ and ‘bad’ spectra. ● Figure 29 (C) - The mean normalised spectral shapes corresponding to QWIP > 0.4, as a function of AVW. While not a unilateral rule, QWIP scores meeting this criterion tend to ACIX-III Aqua report 57 capture some evident AC failures. Of all the QWIP output, this particular criterion tends to consistently flag problematic spectra, with few Type II exceptions. ● Figure 29 (D) - The mean normalised spectral shapes corresponding to QWIP < -0.4, as a function of AVW. While this criteria will often capture AC failures, sometimes otherwise goodlooking spectral shapes may meet this criteria if a large portion of the spectrum contains slightly negative or near-zero values, i.e. be on the lookout for reasonable shapes that were perhaps just overcorrected, usually in the far-blue or far-red portion of the spectrum. ● Figure 29 (E) – The mean normalised spectral shapes corresponding to spectra with an AVW – water type mismatch. According to [19], sometimes a spectrum will have a low QWIP score (indicating better quality), but the water type [82] seems to fall out of range of the expected AVW value. For example, in this instance there were several “green” waters that were shifted much too far towards the blue end of the spectrum compared to what we typically see with in situ data. These criteria tend to be robust in terms of flagging suspect spectra. Figure 28 An example of QWIP analysis output for a scene at Ariake Tower AERONET-OC site on May 11, 2020, processed using the hGRS AC approach. (A) Map of AVW, (B) Map of the QWIP score, (C) scatter plot of QWIP analysis, and (D) frequency distribution of incremental QWIP scores. ACIX-III Aqua report 64 excluded from the MIP and Polymer retrievals in the AC intercomparison exercise. For the relative comparison between the ACs, it can be expected that all processors are affected to a similar extent by the instrument uncertainties [18]. To check for the high temporal variability of the in situ measurements close to the EnMAP overpass time, we calculated the mean and standard deviation at the ± 15 minute window when data were available. For three days at LJCO the in situ reflectance showed high temporal variability, and we opted to compare EnMAP data to the mean in situ reflectance instead of using the closest measurement in time, as for all other in situ matchups. Table 10 Band settings in the different datasets: the nominal band used in the match-up analysis and the corresponding band settings of AERONET-OC and EnMAP. The first part of the table refers to the band setting used for the entire dataset, while the second part refers to the additional bands considered for the CVD dataset only. Nominal band AERONET-OC band setting EnMAP band setting AERONET-OC & CVD datasets 443 nm 443 ± 5 nm 444.7 ± 3.0 nm 490 nm 490 ± 5nm 491.8 ± 2.9 nm 560 nm 560 ± 5 nm 561.1 ± 3.2 nm 620 nm 620 ± 5 nm 622.9 ± 3.6 nm 667 nm 667 ± 5 nm 666.6 ± 3.8 nm EnMAP retrievals were compared to the in situ measurements following the OLCI match-up protocol [94] and using a 3 x 3-pixel box and a time window of ± 01:15 h. Most of the measurements were available within ± 15 minutes from the overpass (36 out of 50); 17 during the overpass. EnMAP-MIP pixels were excluded from analysis when they were masked as cirrus, cloud, cloud shadow, and haze flags. For Polymer, we found that the default flags did not always accurately correspond to invalid match-ups, so we decided not to apply any flags and investigate this issue further. For ACOLITE we applied the default recommended flags: NIR or SWIR threshold test, CIRRUS threshold test, TOA threshold test, NEGATIVE surface reflectance test and EXTENT test. Most of the pixels of the 3 x 3-pixel box were retained in the end, usually 7 to 9 pixels, for the valid match-ups. The EnMAP spectrum represents the median of the pixel box after the flagged and outlier pixels were removed. The number of valid match-ups of each AC processor for hyperand multispectral in situ data was 22 and 24 for MIP, 21 and 20 for Polymer, 11 and 17 for ACOLITE without and 3 and 8 for ACOLITE with glint correction applied, and 21 and 30 for PACO-WASI, respectively. The frequency distribution of the Rrs levels of the in situ matchup data are provided in Figure 32. As for the PRISMA evaluation, we followed the statistical metrics detailed in section 2.4.3. We also analysed the spectral shape of EnMAP using the spectral angle mapper (SAM, θ◦) to determine the similarity between an EnMAP and an in situ reference spectrum. . Spectral Angles (SA) were calculated on the mean spectra. ACIX-III Aqua report 65 Figure 32 Frequency distribution of reflectance levels the hyperspectral (CVD) and multispectral in situ measurements (all wavelengths). 6.5 PERFORMANCE ASSESSMENT In this section, results from the comparison of EnMAP standard L2A (MIP) products and ACOLITE, POLYMER and PACO-WASI AC models are presented for the comparison to AERONET-OC and to the hyperspectral matchups. Firstly, the matchups at individual sites are discussed. Then the statistical results over all common matchups for the four AC models MIP, POLYMER, ACOLITE and PACO-WASI are presented and intercompared. The study [22] additionally presents statistical results for all matchups of EnMAP-MIP to hyperspectral and multispectral in situ data which were much more (22 and 24, respectively) then the ones in common for all three AC models (11 and 14, respectively). 6.5.1 MULTISPECTRAL PERFORMANCE Figure 33 - Figure 36 show the spectral comparisons between satellite and in situ data for the nine common AERONET-OC sites among the four EnMAP AC models. For some sites several matchups were available, but only one is shown as example. For EnMAP MIP data, two more matchups were available (MVCO and Irbe Lighthouse), and also one more for EnMAP Polymer (MVCO). Example spectra for these sites are shown as well. Overall, the Rrs spectra of AERONET-OC peaks slightly lower than 0.002 sr-1 in the blue region to maxima of around 0.025 sr-1 in the green region. A brief characterization of the sites and their example matchup follow: ● Venice (AAOT) and Socheongcho sites, where Rrs spectra are higher in the blue region and tend to decrease after about 505 nm, represent the typical shape of clear water spectra. Rrs is very low for Socheongcho (<0.004 sr-1) and shows rather large standard deviation (~20%). ● Bahia Blanca, Banana River, Chesapeake Bay, Irbe Lighthouse, San Marco platform, South Greenbay site, and Kemigawa sites, where Rrs spectra are higher in the green region (around 560 nm), represent the typical shape of turbid water spectra. While Kemigawa shows the ACIX-III Aqua report 66 lowest Rrs values (<0.002 sr-1), the Bahia Blanca site presents the peak in the green region with the highest value of Rrs (<0.025 sr-1); this site is characterised as an area subject to diffuse erosion and strong tidal currents which are responsible for the typically high suspended loads in the channel, also the standard deviation is larger which is similar for Chesapeake Bay. ● In the South Greenbay site, the peak around 710 nm is very evident, compared to the other cases, this could be due to the fact that this site is characterised by a condition of hypereutrophication (super-richness of nutrients). A less noticeable peak around 680 nm is also visible in the case of LISCO site. ● In the Lucinda site (in the tropical coastal waters of the Great Barrier Reef lagoon near the Herbert River estuary), the spectrum shows a peak just below 550 nm and a small standard deviation of ± 0.0005 sr-1. MIP Except for Kemigawa at all bands and for Lucinda in the blue, EnMAP-MIP matches well the AERONETOC bands magnitudes and shape (Figure 33, SA <20°). For Kemigawa there is at all bands an overestimation by EnMAP-MIP, while at Lucinda for bands 400 to 490 nm EnMAP-MIP is lower than AERONET-OC. The standard deviation in Rrs measured by certain in situ measurements reflecting higher variability in the environmental conditions is similarly obtained by the EnMAP-MIP retrievals standard deviation within the 3x3 matchup pixel box. Figure 33 Spectral comparisons between EnMAP L2A water products (MIP) and in situ data in the 11 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of MIP is displayed as blue curves, with the shaded ACIX-III Aqua report 67 blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. ACOLITE EnMAP-ACOLITE overestimates Rrs at all sites significantly (Figure 34). Deviations in the spectral shape are partly large (SA 19°-40°, South Greenbay, Kemigawa, Chesapeake Bay and Banana River), but also rather low for the rest (SA 4.5-<15°). The standard deviation in Rrs measured by certain in situ measurements is similar to the EnMAP-ACOLITE retrievals standard deviation within the 3x3 matchup pixel box for Chesapeake Bay while it is much higher for the three other sites. Overall, the spectral shape agrees with the AERONET-OC observations; however, except for Lucinda, Kemigawa, and Bahía Blanca, the deviations are considerably larger than those observed for EnMAP-MIP. Figure 34 Spectral comparisons between ACOLITE and in situ data in the 9 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of ACOLITE is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. POLYMER EnMAP-Polymer deviates significantly in Rrs at all sites, except for Bahia Blanca and Chesapeake Bay where the agreement is within the standard deviation to AERONET-OC (Figure 35). Mostly the values are much higher in the blue, and much lower in the red compared to AERONET-OC. Deviation in the spectral shape are the largest among the four AC models: SA range from ~9° for Bahia Blanca and Lucinda, to ~15°-25° for Venice (AAOT), San Marco Platform, Socheongcho and MVCO and between 40° to 67° for the remaining four sites. The standard deviation in Rrs measured by certain in situ ACIX-III Aqua report 68 measurements of AERONET-OC is only slightly larger for the EnMAP-ACOLITE retrievals standard deviation within the 3x3 matchup pixel box. Figure 35 Spectral comparisons between POLYMER and in situ data in the 10 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of POLYMER is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. PACO-WASI Except for Kemigawa and MVCO, the reflectance derived by PACO-WASI matches well the AERONETOC measurements in magnitude and shape (Figure 36). For Kemigawa, PACO-WASI overestimates reflectance in all bands strongly, and for MVCO it overestimates it from 400 to 550 nm moderately. With Rrs < 0.001 sr-1, Kemigawa has from all sites by far the lowest reflectance in the blue, which makes atmospheric correction particularly challenging. More than one matchup was available for Bahia Blanca, Chesapeake Bay, Socheongcho and Lucinda. For the first three, the standard deviation of the in situ measurements is high, reflecting a high temporal variability in the environmental conditions. A very similar variability is obtained from PACO-WASI. The overlapping standard deviations of both datasets indicate a good correspondence of PACO-WASI reflectance with the in situ measurements. ACIX-III Aqua report 69 Figure 36 Spectral comparisons between PACO-WASI and in situ data in the 12 globally distributed AERONET-OC sites. The variability across the dataset in the mean spectra of PACO-WASI is displayed as blue curves, with the shaded blue area representing the standard deviation. The mean and standard deviation of the in situ data are equivalently shown in red. SA represents the Spectral Angle and N represents the number of the images. Summary The common multispectral dataset for EnMAP contains 14 match-ups with AERONET-OC data from nine study sites: one at AAOT, four at Bahia Blanca, one at Banana River, two at Cheasepeake Bay, one at Kemigawa Offshore, one at San Marco Platform, one at Socheongcho, one at South Greenbay and two at LJCO. Figure 37 shows the overall performance of AC processors using all the AERONETOC match-ups for each EnMAP AC model separately with all the EnMAP satellite data combined. Table 11 summarizes the statistical metrics of the AC model performance for each reference band as plotted in Figure 37. Regarding RMSD and MAD, the best performance, i.e. the lowest values at the five bands, was obtained for PACO-WASI, closely followed by MIP. The two other AC models show much higher values. For MAPD, MIP is best (except at 620 nm where POLYMER is slightly better), followed by POLYMER and PACO-WASI (which is even slightly better than MIP at 667 nm). The mean bias is best at 443 nm and 490 nm for MIP; at 560 nm for POLYMER and PACO-WASI; at 620 nm for PACO-WASI, and at 667 nm for POLYMER (very closely followed by MIP). Generally, the mean bias is quite high at 443 nm for all four methods (worst for PACO WASI with 131%). While ACOLITE performed worst for the previous metrics, it performed similarly well to PACO-WASI and even better than PACO-WASI at 443 ACIX-III Aqua report 70 nm. MIP showed the weakest results here. Concerning the slope, PACO-WASI is closest to 1 for all bands, except at 443 nm, where ACOLITE is slightly better. For all methods, the regression at 443 nm is much worse than for all other bands, ranging from anti-correlation in case of POLYMER (slope = - 0.24) to not more than 0.75 (ACOLITE). POLYMER and MIP still show no convenient regression at 490 nm (slopes of 0.61 and 0.71, respectively), but ACOLITE and PACO-WASI are close to 1. All methods reveal slopes between 0.75 and 1.09 at the remaining bands at 560 nm, 620 nm and 667 nm. Figure 37 Overall performance of AC processors using the AERONET-OC match-ups with all the satellite data combined. The number of matchups per processor and per band is reported in Table 11 along with the statistics. The black lines refer to the 1:1 line. ACIX-III Aqua report 71 Table 11 Summary of the five statistical metrics for the EnMAP matchup data shown in Figure 37. N represents the total number of samples for each band. Figure 38 summarizes the results for the statistical parameters ϵ and β. In terms of Median Symmetric Accuracy (ϵ, Figure 38 left panel), PACO-WASI performs best, followed by MIP. PACO-WASI meets the 30% threshold [18] very well for all bands except 443 nm (38%), and MIP meets the 30% threshold well for the green and red bands, while the error is much higher at 443 nm (110%) and slightly above the threshold for 490 nm (32%). ACOLITE and POLYMER exhibit higher errors across all five bands, most notably for the blue bands (>95% and >75%, respectively). POLYMER shows higher accuracy than ACOLITE but has the largest error at 443 nm among the four AC models. Regarding the Median Symmetric Bias (β, Figure 38, right panel), PACO-WASI shows the lowest median symmetric Bias, which is positive (5-10%) for all bands except 443 nm (-7%). MIP shows the second lowest values for β, which are negative for all bands (largest for 443 nm with 65%, lower for the other bands with 5-15 %). POLYMER shows a higher positive β than MIP’s negative β for the blue bands, and similar negative bias for the other bands. ACOLITE obtains highest β values which are always positive. One has to keep in mind that in this intercomparison, PACO-WASI is based on an improved version of the EnMAP L1 and L2A data as compared to the other three AC methods which may explain some of the better performance. Furthermore, processing was not done automatically as for the other softwares but required time-consuming optimization of inversion by experts. In addition, the number of valid matchups varies among the different AC methods, but also among the different wavelengths. ACIX-III Aqua report 72 Comparisons only considering the same matchups are preferred for a ranking of methods, as has been provided for EnMAP-MIP, -POLYMER and -ACOLITE in [22]. Figure 38 Performance assessments as determined by the Median Symmetric Accuracy (ϵ) and median symmetric Bias (β) for all the match-ups combined. The dashed line corresponds to a 30% threshold [18]. 6.5.2 HYPERSPECTRAL PERFORMANCE Figures 39, 40, 41 and 42 show examples of the spectral comparisons between satellite and in situ data for the four-common hyperspectral in situ data sites (AAOT, Lake Constance, Lampedusa, Lucinda) among the four EnMAP AC models. Additionally, for EnMAP-MIP two more matchups were available (Oostende and Lake Trasimeno), while for EnMAP-Polymer only matchups were available for Lake Trasimeno and for EnMAP-ACOLITE only for Oostende. Some of these sites had several matchups, but only one is shown as example. Overall, the Rrs spectra of hyperspectral in situ peaks slightly lower than 0.007 sr-1 in the blue region to maxima of around 0.027 sr-1 in the green region. A brief characterisation of the sites and their example matchup follows: ● Lampedusa where Rrs spectra are highest at 400nm and continuously decreasing to the red, showing very clear (open ocean) water spectra, although the water depth at the station is just 20 m. Rrs is quite low (<0.007 sr-1) but not as low as found for the AERONET-OC site Socheongcho (Chapter 4.2.1). ● Venice (AAOT), where Rrs spectra peak at 500-530 nm and are a bit higher in the green than the blue region, decrease a lot in the red, but still have some contribution at 600 to 700 nm, showing more mesotrophic with little CDOM but some sediment. It shows less clear water conditions as opposed to the matchup shown for AAOT in 4.2.1. There is quite a large standard deviation (>20%) indicating dynamic (most likely suspended sediment) conditions during matchup. ● Lake Constance: a rather deep (mean depth 100m), subalpine glacial lake, where Rrs spectra peak at 535 nm, are much higher in the green than the blue region, indicating significant CDOM ACIX-III Aqua report 73 loading as compared to sediments. There is a strong decrease in the red part, but still some Rrs left. It indicates mesotrophic water conditions with rather low chlorophyll-a concentration. ● Lucinda is in the tropical coastal waters of the Great Barrier Reef lagoon near the Herbert River estuary; the spectrum shows a peak at 560 nm and also a rather large standard deviation up to 50% indicating highly variable suspended loads during the matchup. ● Oostende is a coastal site where Rrs spectra are higher in the green region (around 560 nm), represent the typical shape of the turbid water spectra, but also significant contribution of CDOM absorption. It presents the peak in the green region with the highest value of Rrs (<0.027 sr-1); this site is characterised as an area subject to diffuse erosion and strong tidal currents which are responsible for the typically high suspended loads in the channel. ● Lake Trasimeno (meso-eutrophic): is a shallow tectonic lake, with the Rrs peak around 0.0025 sr-1 in the green region. The spectrum shows reflectance peaks and dips in the range 670-700 nm related to the presence of Chl-a. MIP Overall, we observe very good agreement between EnMAP-MIP and in situ spectra for the different water sites and reflectance magnitudes and the spectral shape is well kept (SA between 2.9 to 8.2°; Figure 39). The highest disagreement was observed at Oostende, a region with typical turbid and very turbid waters [48]. The underestimation of MIP at Oostende might be improved by the new EnMAP processor version (introduced in March 2025) that should fix the issue of the different water types in the MIP processor, as now all products are retrieved using the clear water option. Three low-quality images of LJCO were acquired on dates (July 08, August 15, and August 16, 2022) when in situ reflectance exhibited high temporal variability (highlighted by the shaded green region). As previously mentioned, on these days, we compared the EnMAP data to the mean of the 15-minute window (indicated by the dashed green line). In the red-NIR region, there are still atmospheric absorption features that are not completely removed, in particular the O2 band around 760 nm. Overall, spectral noise, specially at shorter wavelengths, is observed in the EnMAP-MIP product and it is mostly due to an insufficient sampling in the convolution to compute water look-up-tables in MIP. This issue was recently fixed and will be implemented in the next EnMAP processor version. These band-to-band spectral variations are also observed in the PRISMA data by [24] and [76] and CHRIS-PROBA by [95], possibly a result of inter-band calibration issues. ACIX-III Aqua report 80 was slightly higher than for AERONET-OC; indeed, for CVD most of the AC methods produced data that satisfied the 30% accuracy threshold recommended by the GCOS for Rrs. Additionally, each atmospheric correction processor showed varying degrees of accuracy depending on the Optical Water Type (OWT); these data were in fact representative of eight OWTs that encompass blue, phytoplankton-rich, turbid and humic waters in marine and lacustrine ecosystems. Analysing the heatmaps helped illustrate how method performance depends on water properties, indicating the AC method with the lowest ε for each combination of OWT and spectral band. OWT analysis was not conducted for EnMAP due to the lower number of samples. To conclude, ACIX-III Aqua has provided a valuable framework for understanding how state-of-the-art atmospheric correction methods perform when applied to PRISMA and EnMAP, which have been offering relevant observations for water applications. In particular, the study provides an overview of the comparison of fieldand satellite-derived Rrs values for specific PRISMA bands over 26 globally distributed inland and coastal water sites, using a dataset spanning from 2020 to 2024. Nonetheless, some elements have to be considered for improving the exercise: • Some of the AC methods are also providing estimates of AOT; a comparison of such estimation with reference data from AERONET and CVD networks (e.g., Microtops II) would provide further elements to evaluate the AC methods. • The QWIP analysis, presented here just as an example, would merit further investigation to obtain a vision across the entire image (30 km by 30 km) rather than the direct comparison in a limited portion (i.e. 90 m x 90 m); this would allow to appreciate how AC methods perform in shallow waters, with clouds etc. • A deeper analysis should be undertaken to consider the uncertainties (e.g., geo-location) associated with both in situ and satellite data, and to evaluate the impact on the estimation of biophysical parameters, as performed in previous ACIX-Aqua studies [18]. • Both PRISMA and EnMAP have not been specifically designed to meet the observational requirements (e.g., high signal to noise ratio) for developing aquatic applications so that the AC results are of course depending on the properties of these satellite data. • Further analysis is required to understand how the variations in AC performance are related to the increased hyperspectral information content of PRISMA/EnMAP and how the results of this exercise compare to those obtained with other satellites (e.g. PACE). The analysis is hence not ending and we do hope that such work will prosecute also to support the generation of improved products for aquatic applications from hyperspectral data collected by future mission, such as CHIME and SBG. ACIX-III Aqua report 81 8 ACKNOWLEDGMENTS This work would not have been possible without access to high-quality in situ data. We are deeply grateful to the AERONET-OC Principal Investigators: J. Ishizaka, K. Arai, H. Loisel, P. Pratolongo, R. Frouin, M. Talone, F. Mélin, B. Bulgarelli, G. Zibordi, H. Higa, T. Moore, S. Ruberg, M. Wang, S. Ahmed, A. Gilerson, T. Schroeder, Y.J. Park, B. Jones, M. Ragan, D. Aurin, H. Sosi, S. Ladner, and D. Van der Zande. We also acknowledge the contributions of the Joint Research Centre (JRC) of the European Commission, JAXA’s GCOM-C project, the Integrated Marine Observing System (IMOS)—enabled by the National Collaborative Research Infrastructure Strategy (NCRIS)—and the HYPERNETS H2020 project for providing hyperspectral data from AAOT and Lampedusa. Special thanks go to Mauro Musanti and Salvatore Mangano for their fieldwork in Italy; Mara Gomes, Milad Niroumand-Jadidi, the Phytooptics group members, and LUBW/ISF for their support during the Lake Constance EnMAP validation campaigns. We are grateful to Angelo Amodio and Luigi Agrimano from Planetek for they valuable support in handling PRISMA products. We thank Roberta Pozzi and Simone Lella from CNRIREA for supporting data handling and storage. We are also grateful to all contributors and supporters of the EnMAP Cal/Val Water Project. We sincerely thank Shun Bi for valuable discussions on optical water types. This work was supported by: the Italian Space Agency through the PRISCAV Project (grant no. 2019-5-HH.0) and PANDA-WATER Project (contract ASI no. 2022-15-U.0); the European Space Agency (contract no. 4000139081/22/I-EF, HYPERNETS-POP); the Horizon 2020 HYPERNETS project (grant agreement no. 775983); the Bundesministerium für Wirtschaft und Klimaschutz (grants 50EE1915, 50EE1923, 50EE2401); the DLR for funding the EnMAP mission; The Swiss National Science Foundation (SNSF) under the Lake3P project (grant no. 204783); The Swedish National Space Agency (Dnr. 2021-00050); the Belgian Science Policy Office through the TERRASCOPE project (contract no. CB/67/12). ACIX-III Aqua report 82 9 REFERENCES 1. Strong, A.E. Remote Sensing of Algal Blooms by Aircraft and Satellite in Lake Erie and Utah Lake. Remote Sens Environ 1974, 3, 99–107, doi:10.1016/0034-4257(74)90052-2. 2. Shi, W.; Wang, M. Three‐dimensional Observations from MODIS and CALIPSO for Ocean Responses to Cyclone Nargis in the Gulf of Martaban. Geophys Res Lett 2008, 35, doi:10.1029/2008GL035279. 3. Stumpf, R.P.; Wynne, T.T.; Baker, D.B.; Fahnenstiel, G.L. Interannual Variability of Cyanobacterial Blooms in Lake Erie. PLoS One 2012, 7, e42444, doi:10.1371/journal.pone.0042444. 4. Bresciani, M.; Adamo, M.; De Carolis, G.; Matta, E.; Pasquariello, G.; Vaičiūtė, D.; Giardino, C. Monitoring Blooms and Surface Accumulation of Cyanobacteria in the Curonian Lagoon by Combining MERIS and ASAR Data. Remote Sens Environ 2014, 146, 124–135, doi:10.1016/j.rse.2013.07.040. 5. Matthews, M.W.; Odermatt, D. Improved Algorithm for Routine Monitoring of Cyanobacteria and Eutrophication in Inland and Near-Coastal Waters. Remote Sens Environ 2015, 156, 374– 382, doi:10.1016/j.rse.2014.10.010. 6. Tyler, A.N.; Hunter, P.D.; Spyrakos, E.; Groom, S.; Constantinescu, A.M.; Kitchen, J. Developments in Earth Observation for the Assessment and Monitoring of Inland, Transitional, Coastal and Shelf-Sea Waters. Science of The Total Environment 2016, 572, 1307–1321, doi:10.1016/j.scitotenv.2016.01.020. 7. Pahlevan, N.; Schott, J.R.; Franz, B.A.; Zibordi, G.; Markham, B.; Bailey, S.; Schaaf, C.B.; Ondrusek, M.; Greb, S.; Strait, C.M. Landsat 8 Remote Sensing Reflectance (Rrs) Products: Evaluations, Intercomparisons, and Enhancements. Remote Sens Environ 2017, 190, 289–301, doi:10.1016/j.rse.2016.12.030. 8. Hedley, J.D.; Roelfsema, C.; Brando, V.; Giardino, C.; Kutser, T.; Phinn, S.; Mumby, P.J.; Barrilero, O.; Laporte, J.; Koetz, B. Coral Reef Applications of Sentinel-2: Coverage, Characteristics, Bathymetry and Benthic Mapping with Comparison to Landsat 8. Remote Sens Environ 2018, 216, 598–614, doi:10.1016/j.rse.2018.07.014. 9. Pahlevan, N.; Balasubramanian, S. V.; Sarkar, S.; Franz, B.A. Toward Long-Term Aquatic Science Products from Heritage Landsat Missions. Remote Sens (Basel) 2018, 10, 1337, doi:10.3390/rs10091337. 10. Cazzaniga, I.; Bresciani, M.; Colombo, R.; Della Bella, V.; Padula, R.; Giardino, C. A Comparison of Sentinel-3-OLCI and Sentinel-2-MSI-Derived Chlorophylla Maps for Two Large Italian Lakes. Remote Sensing Letters 2019, 10, 978–987, doi:10.1080/2150704X.2019.1634298. 11. Bresciani, M.; Pinardi, M.; Free, G.; Luciani, G.; Ghebrehiwot, S.; Laanen, M.; Peters, S.; Della Bella, V.; Padula, R.; Giardino, C. The Use of Multisource Optical Sensors to Study Phytoplankton Spatio-Temporal Variation in a Shallow Turbid Lake. Water (Basel) 2020, 12, 284, doi:10.3390/w12010284. ACIX-III Aqua report 83 12. Bresciani, M.; Rossini, M.; Morabito, G.; Matta, E.; Pinardi, M.; Cogliati, S.; Julitta, T.; Colombo, R.; Braga, F.; Giardino, C. Analysis of Withinand between-Day Chlorophyll-a Dynamics in Mantua Superior Lake, with a Continuous Spectroradiometric Measurement. Mar Freshw Res 2013, 64, 303, doi:10.1071/MF12229. 13. Hestir, E.L.; Brando, V.E.; Bresciani, M.; Giardino, C.; Matta, E.; Villa, P.; Dekker, A.G. Measuring Freshwater Aquatic Ecosystems: The Need for a Hyperspectral Global Mapping Satellite Mission. Remote Sens Environ 2015, 167, 181–195, doi:10.1016/j.rse.2015.05.023. 14. Dekker, A.G.; Vos, R.J.; Peters, S.W.M. Comparison of Remote Sensing Data, Model Results and in Situ Data for Total Suspended Matter (TSM) in the Southern Frisian Lakes. Science of The Total Environment 2001, 268, 197–214, doi:10.1016/S0048-9697(00)00679-3. 15. Dierssen, H.M.; Ackleson, S.G.; Joyce, K.E.; Hestir, E.L.; Castagna, A.; Lavender, S.; McManus, M.A. Living up to the Hype of Hyperspectral Aquatic Remote Sensing: Science, Resources and Outlook. Front Environ Sci 2021, 9, doi:10.3389/fenvs.2021.649528. 16. Striebel, M.; Behl, S.; Diehl, S.; Stibor, H. Spectral Niche Complementarity and Carbon Dynamics in Pelagic Ecosystems. Am Nat 2009, 174, 141–147, doi:10.1086/599294. 17. Doxani, G.; Vermote, E.; Roger, J.-C.; Gascon, F.; Adriaensen, S.; Frantz, D.; Hagolle, O.; Hollstein, A.; Kirches, G.; Li, F.; et al. Atmospheric Correction Inter-Comparison Exercise. Remote Sens (Basel) 2018, 10, 352, doi:10.3390/rs10020352. 18. Pahlevan, N.; Mangin, A.; Balasubramanian, S. V.; Smith, B.; Alikas, K.; Arai, K.; Barbosa, C.; Bélanger, S.; Binding, C.; Bresciani, M.; et al. ACIX-Aqua: A Global Assessment of Atmospheric Correction Methods for Landsat-8 and Sentinel-2 over Lakes, Rivers, and Coastal Waters. Remote Sens Environ 2021, 258, 112366, doi:10.1016/j.rse.2021.112366. 19. Dierssen, H.M.; Vandermeulen, R.A.; Barnes, B.B.; Castagna, A.; Knaeps, E.; Vanhellemont, Q. QWIP: A Quantitative Metric for Quality Control of Aquatic Reflectance Spectral Shape Using the Apparent Visible Wavelength. Frontiers in Remote Sensing 2022, 3, doi:10.3389/frsen.2022.869611. 20. Storch, T.; Honold, H.-P.; Chabrillat, S.; Habermeyer, M.; Tucker, P.; Brell, M.; Ohndorf, A.; Wirth, K.; Betz, M.; Kuchler, M.; et al. The EnMAP Imaging Spectroscopy Mission towards Operations. Remote Sens Environ 2023, 294, 113632, doi:10.1016/j.rse.2023.113632. 21. Chabrillat, S.; Foerster, S.; Segl, K.; Beamish, A.; Brell, M.; Asadzadeh, S.; Milewski, R.; Ward, K.J.; Brosinsky, A.; Koch, K.; et al. The EnMAP Spaceborne Imaging Spectroscopy Mission: Initial Scientific Results Two Years after Launch. Remote Sens Environ 2024, 315, 114379, doi:10.1016/j.rse.2024.114379. 22. Soppa, M.A.; Brell, M.; Chabrillat, S.; Alvarado, L.M.A.; Gege, P.; Plattner, S.; Somlai-Schweiger, I.; Schroeder, T.; Steinmetz, F.; Scheffler, D.; et al. Full Mission Evaluation of EnMAP Water Leaving Reflectance Products Using Three Atmospheric Correction Processors. Opt Express 2024, 32, 28215, doi:10.1364/OE.523813. 23. Zibordi, G.; Mélin, F.; Berthon, J.-F.; Holben, B.; Slutsker, I.; Giles, D.; D’Alimonte, D.; Vandemark, D.; Feng, H.; Schuster, G.; et al. AERONET-OC: A Network for the Validation of Ocean Color Primary Products. J Atmos Ocean Technol 2009, 26, 1634–1651, doi:10.1175/2009JTECHO654.1. ACIX-III Aqua report 84 24. Braga, F.; Fabbretto, A.; Vanhellemont, Q.; Bresciani, M.; Giardino, C.; Scarpa, G.M.; Manfè, G.; Concha, J.A.; Brando, V.E. Assessment of PRISMA Water Reflectance Using Autonomous Hyperspectral Radiometry. ISPRS Journal of Photogrammetry and Remote Sensing 2022, 192, 99–114, doi:10.1016/j.isprsjprs.2022.08.009. 25. Wüest, A.; Bouffard, D.; Guillard, J.; Ibelings, B.W.; Lavanchy, S.; Perga, M.; Pasche, N. LéXPLORE: A Floating Laboratory on Lake Geneva Offering Unique Lake Research Opportunities. WIREs Water 2021, 8, doi:10.1002/wat2.1544. 26. González Vilas, L.; Brando, V.E.; Concha, J.A.; Goyens, C.; Dogliotti, A.I.; Doxaran, D.; Dille, A.; Van der Zande, D. Validation of Satellite Water Products Based on HYPERNETS in Situ Data Using a Match-up Database (MDB) File Structure. Frontiers in Remote Sensing 2024, 5, doi:10.3389/frsen.2024.1330317. 27. Casey, K.A.; Rousseaux, C.S.; Gregg, W.W.; Boss, E.; Chase, A.P.; Craig, S.E.; Mouw, C.B.; Reynolds, R.A.; Stramski, D.; Ackleson, S.G.; et al. A Global Compilation of in Situ Aquatic High Spectral Resolution Inherent and Apparent Optical Property Data for Remote Sensing Applications. Earth Syst Sci Data 2020, 12, 1123–1139, doi:10.5194/essd-12-1123-2020. 28. IMOS AODN Lucinda Jetty Coastal Observatory: HyperOCR Data Available online: https://thredds.aodn.org.au/thredds/catalog/IMOS/SRS/OC/LJCO/catalog.html (accessed on 7 March 2025). 29. Klotz, P. Assessment of the Suitability of PlanetScope SuperDove Imagery for Aquatic Remote Sensing at Lake Junin in Central Peru, Ludwig-Maximilians-Universität München, 2024. 30. Zibordi, G.; Holben, B.N.; Talone, M.; D’Alimonte, D.; Slutsker, I.; Giles, D.M.; Sorokin, M.G. Advances in the Ocean Color Component of the Aerosol Robotic Network (AERONET-OC). J Atmos Ocean Technol 2021, 38, 725–746, doi:10.1175/JTECH-D-20-0085.1. 31. Ruddick, K.G.; Brando, V.E.; Corizzi, A.; Dogliotti, A.I.; Doxaran, D.; Goyens, C.; Kuusk, J.; Vanhellemont, Q.; Vansteenwegen, D.; Bialek, A.; et al. WATERHYPERNET: A Prototype Network of Automated in Situ Measurements of Hyperspectral Water Reflectance for Satellite Validation and Water Quality Monitoring. Frontiers in Remote Sensing 2024, 5, doi:10.3389/frsen.2024.1347520. 32. Vansteenwegen, D.; Ruddick, K.; Cattrijsse, A.; Vanhellemont, Q.; Beck, M. The Pan-and-Tilt Hyperspectral Radiometer System (PANTHYR) for Autonomous Satellite Validation Measurements—Prototype Design and Testing. Remote Sens (Basel) 2019, 11, 1360, doi:10.3390/rs11111360. 33. Hommersom, A.; Kratzer, S.; Laanen, M.; Ansko, I.; Ligi, M.; Bresciani, M.; Giardino, C.; BeltránAbaunza, J.M.; Moore, G.; Wernand, M.; et al. Intercomparison in the Field between the New WISP-3 and Other Radiometers (TriOS Ramses, ASD FieldSpec, and TACCS). J Appl Remote Sens 2012, 6, 063615, doi:10.1117/1.JRS.6.063615. 34. Schläpfer, D.; Borel, C.C.; Keller, J.; Itten, K.I. Atmospheric Precorrected Differential Absorption Technique to Retrieve Columnar Water Vapor. Remote Sens Environ 1998, 65, 353–366, doi:10.1016/S0034-4257(98)00044-3. 35. Gege, P.; Grötsch, P. A Spectral Model for Correcting Sunglint and Skyglint. In Proceedings of the Ocean Optics XXIII 2016; 2016. ACIX-III Aqua report 85 36. Gordon, H.R. Ocean Color Remote Sensing Systems: Radiometric Requirements.; Slater, P.N., Ed.; October 12 1988; p. 151. 37. Haan, J.F.; Kokke, J.M.M. Remote Sensing Algorithm Development: Toolkit I: Operationalization of Atmospheric Correction Methods for Tidal and Inland Waters. Netherlands Remote Sensing Board (BCRS) 1996. 38. Sterckx, S.; Knaeps, S.; Kratzer, S.; Ruddick, K. SIMilarity Environment Correction (SIMEC) Applied to MERIS Data over Inland and Coastal Waters. Remote Sens Environ 2015, 157, 96– 110, doi:10.1016/j.rse.2014.06.017. 39. Kiselev, V.; Bulgarelli, B.; Heege, T. Sensor Independent Adjacency Correction Algorithm for Coastal and Inland Water Systems. Remote Sens Environ 2015, 157, 85–95, doi:10.1016/j.rse.2014.07.025. 40. Richter, R.; Schläpfer, D.; Müller, A. An Automatic Atmospheric Correction Algorithm for Visible/NIR Imagery. Int J Remote Sens 2006, 27, 2077–2085, doi:10.1080/01431160500486690. 41. Wu, Y.; Knudby, A.; Pahlevan, N.; Lapen, D.; Zeng, C. Sensor-Generic Adjacency-Effect Correction for Remote Sensing of Coastal and Inland Waters. Remote Sens Environ 2024, 315, 114433, doi:10.1016/j.rse.2024.114433. 42. Kotchenova, S.Y.; Vermote, E.F.; Matarrese, R.; Klemm, Jr., F.J. Validation of a Vector Version of the 6S Radiative Transfer Code for Atmospheric Correction of Satellite Data Part I: Path Radiance. Appl Opt 2006, 45, 6762, doi:10.1364/AO.45.006762. 43. Lenoble, J.; Herman, M.; Deuzé, J.L.; Lafrance, B.; Santer, R.; Tanré, D. A Successive Order of Scattering Code for Solving the Vector Equation of Transfer in the Earth’s Atmosphere with Aerosols. J Quant Spectrosc Radiat Transf 2007, 107, 479–507, doi:10.1016/j.jqsrt.2007.03.010. 44. Berk, A.; Anderson, G.P.; Acharya, P.K.; Bernstein, L.S.; Muratov, L.; Lee, J.; ...; Shettle, E.P. MODTRAN5: 2006 Update. In Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XII; 2006; Vol. 6233, pp. 508–515. 45. De Keukelaere, L.; Sterckx, S.; Adriaensen, S.; Knaeps, E.; Reusen, I.; Giardino, C.; Bresciani, M.; Hunter, P.; Neil, C.; Van der Zande, D.; et al. Atmospheric Correction of Landsat-8/OLI and Sentinel-2/MSI Data Using ICOR Algorithm: Validation for Coastal and Inland Waters. Eur J Remote Sens 2018, 51, 525–542, doi:10.1080/22797254.2018.1457937. 46. Park, Y.-J.; Ruddick, K. Model of Remote-Sensing Reflectance Including Bidirectional Effects for Case 1 and Case 2 Waters. Appl Opt 2005, 44, 1236, doi:10.1364/AO.44.001236. 47. Gege, P. WASI-2D: A Software Tool for Regionally Optimized Analysis of Imaging Spectrometer Data from Deep and Shallow Waters. Comput Geosci 2014, 62, 208–215, doi:10.1016/j.cageo.2013.07.022. 48. Vanhellemont, Q. Adaptation of the Dark Spectrum Fitting Atmospheric Correction for Aquatic Applications of the Landsat and Sentinel-2 Archives. Remote Sens Environ 2019, 225, 175–192, doi:10.1016/j.rse.2019.03.010. ACIX-III Aqua report 86 49. Vanhellemont, Q.; Ruddick, K. Atmospheric Correction of Metre-Scale Optical Satellite Data for Inland and Coastal Water Applications. Remote Sens Environ 2018, 216, 586–597, doi:10.1016/j.rse.2018.07.015. 50. Harmel, T.; Chami, M.; Tormos, T.; Reynaud, N.; Danis, P.-A. Sunglint Correction of the MultiSpectral Instrument (MSI)-SENTINEL-2 Imagery over Inland and Sea Waters from SWIR Bands. Remote Sens Environ 2018, 204, 308–321, doi:10.1016/j.rse.2017.10.022. 51. Steinmetz, F.; Deschamps, P.-Y.; Ramon, D. Atmospheric Correction in Presence of Sun Glint: Application to MERIS. Opt Express 2011, 19, 9783, doi:10.1364/OE.19.009783. 52. Steinmetz, F.; Ramon, D. Sentinel-2 MSI and Sentinel-3 OLCI Consistent Ocean Colour Products Using POLYMER. In Proceedings of the Remote Sensing of the Open and Coastal Ocean and Inland Waters; Frouin, R.J., Murakami, H., Eds.; SPIE, October 30 2018; p. 13. 53. de los Reyes, R.; Langheinrich, M.; Schwind, P.; Richter, R.; Pflug, B.; Bachmann, M.; Müller, R.; Carmona, E.; Zekoll, V.; Reinartz, P. PACO: Python-Based Atmospheric Correction. Sensors 2020, 20, 1428, doi:10.3390/s20051428. 54. IOCCG (2025) Available online: https://ioccg.org/resources/software/ (accessed on 20 October 2025). 55. Vanhellemont, Q.; Ruddick, K. Turbid Wakes Associated with Offshore Wind Turbines Observed with Landsat 8. Remote Sens Environ 2014, 145, 105–115, doi:10.1016/j.rse.2014.01.009. 56. Vanhellemont, Q.; Ruddick, K. Acolite for Sentinel-2: Aquatic Applications of MSI Imagery. In Proceedings of the 2016 ESA Living Planet Symposium; Prague, Czech Republic, 2016. 57. Vanhellemont, Q. Sensitivity Analysis of the Dark Spectrum Fitting Atmospheric Correction for Metreand Decametre-Scale Satellite Imagery Using Autonomous Hyperspectral Radiometry. Opt Express 2020, 28, 29948, doi:10.1364/OE.397456. 58. Vanhellemont, Q.; Ruddick, K. Atmospheric Correction of Sentinel-3/OLCI Data for Mapping of Suspended Particulate Matter and Chlorophyll-a Concentration in Belgian Turbid Coastal Waters. Remote Sens Environ 2021, 256, 112284, doi:10.1016/j.rse.2021.112284. 59. Chami, M.; Lafrance, B.; Fougnie, B.; Chowdhary, J.; Harmel, T.; Waquet, F. OSOAA: A Vector Radiative Transfer Model of Coupled Atmosphere-Ocean System for a Rough Sea Surface Application to the Estimates of the Directional Variations of the Water Leaving Reflectance to Better Process Multi-Angular Satellite Sensors Data over the Ocean. Opt Express 2015, 23, 27829, doi:10.1364/OE.23.027829. 60. Hess, M.; Koepke, P.; Schult, I. Optical Properties of Aerosols and Clouds: The Software Package OPAC. Bull Am Meteorol Soc 1998, 79, 831–844, doi:10.1175/15200477(1998)079<0831:OPOAAC>2.0.CO;2. 61. Yang, P.; Feng, Q.; Hong, G.; Kattawar, G.W.; Wiscombe, W.J.; Mishchenko, M.I.; Dubovik, O.; Laszlo, I.; Sokolik, I.N. Modeling of the Scattering and Radiative Properties of Nonspherical Dust-like Aerosols. J Aerosol Sci 2007, 38, 995–1014, doi:10.1016/j.jaerosci.2007.07.001. 62. Gasteiger, J.; Wiegner, M. MOPSMAP v1.0: A Versatile Tool for the Modeling of Aerosol Optical Properties. Geosci Model Dev 2018, 11, 2739–2762, doi:10.5194/gmd-11-2739-2018. ACIX-III Aqua report 87 63. Richter, R. A Spatially Adaptive Fast Atmospheric Correction Algorithm. Int J Remote Sens 1996, 17, 1201–1214, doi:10.1080/01431169608949077. 64. Gege, P. The Water Color Simulator WASI: An Integrating Software Tool for Analysis and Simulation of Optical in Situ Spectra. Comput Geosci 2004, 30, 523–532, doi:10.1016/j.cageo.2004.03.005. 65. Albert, A.; Mobley, C. An Analytical Model for Subsurface Irradiance and Remote Sensing Reflectance in Deep and Shallow Case-2 Waters. Opt Express 2003, 11, 2873, doi:10.1364/OE.11.002873. 66. Mobley, C.D.; Gentili, B.; Gordon, H.R.; Jin, Z.; Kattawar, G.W.; Morel, A.; Reinersman, P.; Stamnes, K.; Stavn, R.H. Comparison of Numerical Models for Computing Underwater Light Fields. Appl. Opt. 1993, 32. 67. Gege Analytic Model for the Direct and Diffuse Components of Downwelling Spectral Irradiance in Water. Appl Opt 2012, 51, 1407, doi:10.1364/AO.51.001407. 68. Gregg, W.W.; Carder, K.L. A Simple Spectral Solar Irradiance Model for Cloudless Maritime Atmospheres. Limnol Oceanogr 1990, 35, 1657–1675, doi:10.4319/lo.1990.35.8.1657. 69. Defoin‐Platel, M.; Chami, M. How Ambiguous Is the Inverse Problem of Ocean Color in Coastal Waters? J Geophys Res Oceans 2007, 112, doi:10.1029/2006JC003847. 70. Wu, Y.; Knudby, A.; Lapen, D. Topography-Adjusted Monte Carlo Simulation of the Adjacency Effect in Remote Sensing of Coastal and Inland Waters. J Quant Spectrosc Radiat Transf 2023, 303, 108589, doi:10.1016/j.jqsrt.2023.108589. 71. Global Modeling and Assimilation Office MERRA-2 Tavg1_2d_chm_Nx: 2d,1-Hourly,TimeAveraged,Single-Level,Assimilation,Carbon Monoxide and Ozone Diagnostics V5.12.4. Greenbelt, MD, USA, Goddard Earth Sciences Data and Information Services Center (GES DISC) [Dataset] 2015. 72. Vermote, E.F.; Tanre, D.; Deuze, J.L.; Herman, M.; Morcette, J.-J. Second Simulation of the Satellite Signal in the Solar Spectrum, 6S: An Overview. IEEE Transactions on Geoscience and Remote Sensing 1997, 35, 675–686, doi:10.1109/36.581987. 73. Giardino, C.; Bresciani, M.; Braga, F.; Fabbretto, A.; Ghirardi, N.; Pepe, M.; Gianinetto, M.; Colombo, R.; Cogliati, S.; Ghebrehiwot, S.; et al. First Evaluation of PRISMA Level 1 Data for Water Applications. Sensors 2020, 20, 4553, doi:10.3390/s20164553. 74. Niroumand-Jadidi, M.; Bovolo, F.; Bruzzone, L. Water Quality Retrieval from PRISMA Hyperspectral Images: First Experience in a Turbid Lake and Comparison with Sentinel-2. Remote Sens (Basel) 2020, 12, 3984, doi:10.3390/rs12233984. 75. Bresciani, M.; Giardino, C.; Fabbretto, A.; Pellegrino, A.; Mangano, S.; Free, G.; Pinardi, M. Application of New Hyperspectral Sensors in the Remote Sensing of Aquatic Ecosystem Health: Exploiting PRISMA and DESIS for Four Italian Lakes. Resources 2022, 11, 8, doi:10.3390/resources11020008. 76. Pellegrino, A.; Fabbretto, A.; Bresciani, M.; de Lima, T.M.A.; Braga, F.; Pahlevan, N.; Brando, V.E.; Kratzer, S.; Gianinetto, M.; Giardino, C. Assessing the Accuracy of PRISMA Standard ACIX-III Aqua report 88 Reflectance Products in Globally Distributed Aquatic Sites. Remote Sens (Basel) 2023, 15, 2163, doi:10.3390/rs15082163. 77. Fabbretto, A.; Bresciani, M.; Pellegrino, A.; Alikas, K.; Pinardi, M.; Mangano, S.; Padula, R.; Giardino, C. Tracking Water Quality and Macrophyte Changes in Lake Trasimeno (Italy) from Spaceborne Hyperspectral Imagery. Remote Sens (Basel) 2024, 16, 1704, doi:10.3390/rs16101704. 78. Bailey, S.W.; Werdell, P.J. A Multi-Sensor Approach for the on-Orbit Validation of Ocean Color Satellite Data Products. Remote Sens Environ 2006, 102, 12–23, doi:10.1016/j.rse.2006.01.015. 79. Giardino, C.; Pahlevan, N.; Fabbretto, A.; Panizza, L.; Pellegrino, A.; Vandermeulen, R.; Gianinetto, M.; Adriaensen, S.; Agten, J.; Bernert, H.; et al. ACIX-III Aqua: Evaluation of Atmospheric Correction for Hyperspectral PRISMA Imagery over Inland and Coastal Waters. Int J Remote Sens 2025, 1–25, doi:10.1080/01431161.2025.2574517. 80. Bi, S.; Hieronymi, M. Holistic Optical Water Type Classification for Ocean, Coastal, and Inland Waters. Limnol Oceanogr 2024, 69, 1547–1561, doi:10.1002/lno.12606. 81. Vandermeulen, R.A.; Mannino, A.; Craig, S.E.; Werdell, P.J. 150 Shades of Green: Using the Full Spectrum of Remote Sensing Reflectance to Elucidate Color Shifts in the Ocean. Remote Sens Environ 2020, 247, 111900, doi:10.1016/j.rse.2020.111900. 82. Balasubramanian, S. V.; Pahlevan, N.; Smith, B.; Binding, C.; Schalles, J.; Loisel, H.; Gurlin, D.; Greb, S.; Alikas, K.; Randla, M.; et al. Robust Algorithm for Estimating Total Suspended Solids (TSS) in Inland and Nearshore Coastal Waters. Remote Sens Environ 2020, 246, 111768, doi:10.1016/j.rse.2020.111768. 83. Bracher, A.; Randrianalisoa, A.N.; Soppa, M.A.; Chabrillat, S.; Scheffler, D.; Steinmetz, F.; Vanhellemont, Q.; Gege, P.; ...; Vansteenwegen, D. Evaluation of ENMAP Water Leaving Reflectance Products from Three Atmospheric Correction Methods by Validation with Hyperspectral and Multispectral in Situ Measurements from Inland and Coastal Water around the World. In Proceedings of the ESA Living Plant Symposium; Vienna, Austria, 2025. 84. Goyens, C.; Lavigne, H.; Dille, A.; Vervaeren, H. Using Hyperspectral Remote Sensing to Monitor Water Quality in Drinking Water Reservoirs. Remote Sens (Basel) 2022, 14, 5607, doi:10.3390/rs14215607. 85. Peters S.; Laanen M.; Groetsch P.; Ghezehegn S.; Poser K.; Hommersom A.; De Reus E.; Spaias L. WISPstation: A New Autonomous above Water Radiometer System. In Proceedings of the Ocean Optics XXIV Conference; Dubrovnik, Croatia, 2018; pp. 7–12. 86. Brando, V.; Keen, R.; Daniel, P.; Baumeister, A.; Nethery, M.; Baumeister, H.; Hawdon, A.; Swan, G.; Mitchell, R.; Campbell, S.; et al. The Lucinda Jetty Coastal Observatory’s Role in Satellite Ocean Colour Calibration and Validation for Great Barrier Reef Coastal Waters. In Proceedings of the OCEANS’10 IEEE SYDNEY; IEEE, May 2010; pp. 1–8. 87. Tilstone, G.; Dall’Olmo, G.; Hieronymi, M.; Ruddick, K.; Beck, M.; Ligi, M.; Costa, M.; D’Alimonte, D.; Vellucci, V.; Vansteenwegen, D.; et al. Field Intercomparison of Radiometer Measurements for Ocean Colour Validation. Remote Sens (Basel) 2020, 12, 1587, doi:10.3390/rs12101587. ACIX-III Aqua report 89 88. Zibordi, G.; Voss, K.J.; Johnson, B.C.; Mueller, J.L. Ocean Optics and Biogeochemistry Protocols for Satellite Ocean Colour Sensor Validation, Volume 3.0: Protocols for Satellite Ocean Colour Data Validation: In Situ Optical Radiometry. International Ocean Colour Coordinating Group (IOCCG) Dartmouth, NS, Canada 2019. 89. Lee, Z.P.; Du, K.; Voss, K.J.; Zibordi, G.; Lubac, B.; Arnone, R.; Weidemann, A. An InherentOptical-Property-Centered Approach to Correct the Angular Effects in Water-Leaving Radiance. Appl Opt 2011, 50, 3155, doi:10.1364/AO.50.003155. 90. Thuillier, G.; Hersé, M.; Labs, D.; Foujols, T.; Peetermans, W.; Gillotay, D.; Simon, P.C.; Mandel, H. The Solar Spectral Irradiance from 200 to 2400 Nm as Measured by the SOLSPEC Spectrometer from the Atlas and Eureca Missions. Sol Phys 2003, 214, 1–22, doi:10.1023/A:1024048429145. 91. Scheffler, D.; Brell, M.; Bohn, N.; Alvarado, L.; Soppa, M.A.; Segl, K.; Bracher, A.; Chabrillat, S. EnPT – an Alternative Pre-Processing Chain for Hyperspectral EnMAP Data. In Proceedings of the IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium; IEEE, July 16 2023; pp. 7416–7418. 92. Jakimow, B.; Janz, A.; Thiel, F.; Okujeni, A.; Hostert, P.; van der Linden, S. EnMAP-Box: Imaging Spectroscopy in QGIS. SoftwareX 2023, 23, 101507, doi:10.1016/j.softx.2023.101507. 93. Vanhellemont, Q. Daily Metre-Scale Mapping of Water Turbidity Using CubeSat Imagery. Opt Express 2019, 27, A1372, doi:10.1364/OE.27.0A1372. 94. EUMETSAT Recommendations for Sentinel-3 OLCI Ocean Colour Product Validations in Comparison with in Situ Measurements – Matchup Protocols. . 95. Lavigne, H.; Ruddick, K. Inter-Band Calibration for Hyperspectral Water Remote Sensing: Demonstration for Chris-Proba. In Proceedings of the 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS; IEEE, July 11 2021; pp. 7771–7774. 96. Doxaran, D.; ElKilani, B.; Corizzi, A.; Goyens, C. Validation of Satellite-Derived Water-Leaving Reflectance in Contrasted French Coastal Waters Based on HYPERNETS Field Measurements. Frontiers in Remote Sensing 2024, 4, doi:10.3389/frsen.2023.1290110.