ELSEVIER i Content lists available at ScienceDirect = Remote Sensing of Environment journal homepage: www .elsevier comilocatelrse Sentinel-2 MultiSpectral Instrument (MSI) data processing for aquatic science applications: Demonstrations and validations Qe NN. Pahlevan“'”*, S. Sarkar*®, B.A. Franz’, $.V. Balasubramanian”, J. He® NASA Cord Spare Hhe Ce 800 Gren Roa rene MO 20771, USA © Scans Sten ond Apts ne, TOT Crean Rend Sue 60, Lanham, MD 2076, USA Uno of trond Deparmen of Gesprapha Sees Coe Pak 20740, USA ARTICLE INFO ABSTRACT ‘The Baropean Space Ageney* Sentine/-2A mission with the MulSpecal Instrument (WSD onboard was lure shed in 2015, inating s new erin hgh-osmoderstoresoltion Ge, 10 60m) imaging of Fath’ resources ‘This manuerpt deseriber the implementations af MSI processing into the SeaWiFS Data Analyse System (SeaDAS) and provider qualitative and quantitative analyser of remote sensing rellectanee prog (R), hich ‘and imager (OL bth indicated rezanable produc eartal/inland waters Following vieariour alibraions ‘squared dilerences (RMSD) found forthe matchup analyses were, < 7% and = 0.0012 I/s, cespectively, for the bie and green bands. With preliminary indesions af consistency wth the OL produits and very good agreements with in is data, a Umeverles plot of total suspended solids (TSS) product derived from both ‘sions was produced and analyzed for ap flandsjstem is surmised that trequent madeate 30 LIC images over spatially uniform and cleat waters. Fig. 1 shows the mean SNR computed by taking the average of locally estimated SNRs using the ratio of mean (y) to standaré deviation (@), Le, Hy. The calculations were performed for 7 x 7-elementwin- dows in & similar fashion as performed for OL SNR calculations SNR OSD SNR OLD, SNR 446 497 $60 664 704 740 TH 803 86S ‘MST band center (am) Fig, 1 The imagedarwedsgnabtonoive rate SNR) of te Mespetl Iran (as cleate over dew lors odes of water The SRG ate emp fre save spl elution 0 MSI petal band (fr piel rade Fund ove Clear water bodes, While, i sce the MSI SNR te compare fo eve of OL (acon eta, 2043) the Ot arto let Nher SNR a nae 30m eon, (Gee Tole 1) Te allow far compre OLEMST pot gain, to MS product ny (Pablevan etal, 20140), The computed SNRs are compared with the coreesponding SNR values for the OL For fir intercomparison, the OLI's SNRs were sealed (Pahlevan ef a, 20148) to the mean TOA ra diance levels lized in tis study (Vale 1). The mean SNRs caleulated for the MSCs native spatial resolutions for mean radiances (Ls wm=%u" 77") observed at mean SZA of 49 are tabulated in Table 1 eis inferred tht the OLT's SNR forthe three visible bands (497,560, and 664 nm) are 2X to 3X better than those of MSI, whereas, MSI SNR ‘444 nm ie 40% higher than that ofthe OLL. Recognizing tht the SNR ie proportional tothe squaze root ofthe area ofa pixel (Scho, 2007), can be surmised that aggregated 20 or 30 m MSI bands should offer similar or better radiometric quality in the aforementioned three visible bands (ee Section 6). Similarly, aggregated 60.m OL 448m images (isble 1) offer 30% higher SNR in the corresponding MSI band ‘Therefore, the overall radiometric quality of MSI and OL1 products is comparable, and both are found to offer more precise measurements ‘than those ofthe heritage Landsat missions over aquatic systems (ran. etal, 2015; Gerace etal, 2013; Hu etal, 2012: Pahlevan and Schott 2015). Note also that OL!s SNRS in the SWI bands are 3X to 4X beterPola than those of MSI Table 1). These differences inthe noise level will ‘have implications on the pesformance of atmospheric correction (any, ‘and Wel, 2012), In general, one should keep in mind thatthe on-orbit [SNR performance of OLI is 2X to 3X beter than the requirements Irons fal, 2012; Pablevan etal, 2014b) due to design trades for minimizing, on-uniformites (striping) (Knight, 2017). 3. Methodology For the implementation ofthe etmospheric correction, we followed the standard methodology introduced in Gordon and Wang (1994) in ‘Which the TOA reflestance is formulated as below 242) = Hg) + 188) + 8A) + ADE ‘here tis the diffuse transmission, p isthe Rayleigh reflectance in the absence of aerosol, p isthe aerosol radiance, and guy isthe radiance arising from Raylegh-aerosol multiple scattering, and py is the water- leaving reflectane just above water, which ean readily be eonverced to Ry, (iobley er, 2015). The goal i to retrieve R, whichis the eiiel parameter forthe retrievals of the inherent optical properties (OP) and ‘the biogeochemical variables, Note that in the above notation, we have ropped contributions from the sunglint and whitecaps for brevity ‘Also, the above equation can be interchangeably expressed in radiance domain (L). A vector radiative transfer simulation (Ahmad and Fraser, 1982) was performed, in combination with the MSI spectral response functions to develop the common aerosol and Rayleigslook-up-tables ‘used as part ofthe SeaWiFS Daca Analysis System (SeaDAS). Per-pixel, (10, 20, and 60m) and perchand viewing angle coeficiente (i.e, VZA, and VAA) were re-created using coarse resolution (5 km x 5 kr) angle Information provided in metadata files (Catt and Bertolini, 2013). The per-pel reconstruction of angle coefficients was carried out ina few steps: convert the given angles to observations of directions to the Sentinel-2A spaceeraft (ie, line of sigh), reconstruct the orbit fom the viewing angles and calculate the time foreach grié cell, we the cal= culated observation times to construct a line/sample vers time model for each band and detector module, and generate the angle les for ‘each spectral band (see Appendix A fora detailed destription provided bby James Storey and the corresponding code provided as supplemen- ‘ary material in Appendix B). Therefore, the viewing angle coefficients for 10, 20, and 60'm resolution hands are avilable inthe processing ‘On the other hand, the fine-resalution solar angles (SZA and $AZ) were created simply by replieating the course-esolution angles. 'A full deseription of NASA' standatd almospherie correction ema Sng of Erne 201 (2017) 47-56 process is given in (Mobley eta, 2016), The process begins with re- ‘moving the Rayleigh contcbution wsing ancillary data (including digital clevation models) and pre-computed LUTs, followed by the aerosol removal. The band ratio of Rayleigh-orrected radiance, ie, L-Lp ly ly for two bands where Ry +0, is then computed and used (o infer an aerosol type. This band ratio is extrapolated to the visible binds and subtracted from the Rayleigh-correced radiance. Due ‘onon-neglgible water-leaving radiances inthe NIR band (Siegel et. 2000) in mast coasta/inland waters, we allow uiiliztions of the ‘terative NIR-signal removal method (Ssiley eal, 2010) that is eur. tenly available in SeaDAS. Further, for simplicity in processing, we re sampled the 10m resolution bands to & common grid with 20: grid cells via arithmetic averaging. The 444m hand data (at 60m) are replicated for underlying 20-m grid ells. In addition, similar tothe OL processing (Franz etal. 2015), to minimize the noise effets, the MST p, mages were smoothed using a 9 X S-element (180m x 180:m) aver- aging filter, Effectively, the 442 nm band is averaged by a 2 x ale- ‘ment window and the other visible binge are averaged by an 18 X 18- clement window. For intercomparisons with OLI produets (Section 4, (LI products were smoothed using a 6 x element windows “The performance of the atmospheric correction was further eval- uated against the in siu radiometric measurements made af the acean color component ofthe AErosel RObotic NETWork (AERONET-O6) sites (ibord el, 2006), For the image statistics, median Ry, computed for 7 x 7-element windows was sed, The 3% d-element center pixels were discarded 0 avoid adjacency effects induced by the AERONET-OC. platforms (Pahlevan et al, 2016). Note further that SeaDAS applies a broadband to narrowband (10 nm) conversion (Wang et al, 2001) 10 create Ry values centered at band centers specified in Table 1, We ‘examined three different band combinations for the removal of aerosol contributions to evaluate che relative performanees (ran? etal, 2015; ‘Wang and Shi, 2007). These inelude the 865-1613 nm, 865-2200 nm, ‘and 1513-2200 nm. The in situ AERONET-OC data (/iborei etal. 2009) collected within a +/— 1h interval of satelite overpasses were considered for matchup analyses, This is 4 more restrieted matehop analysis than that suggested fr global ocean color missions Bailey a= ‘Wendel, 2006), to accommodate the dynamic nature ofthe exchisvely coastal in sity locations employed, 1g. 2: Theisidetied Ry ne ue Dds compared co ance 2016/08/39) In genera ot SeaDAS proceed ‘he pinay scopes ae the ferences Ip the rel bight ant actface mere renee in the 37) fr MSt a OU respecte Te deen ip snasogsomerses cn eli ome of the diene io ‘he we produce (Fearon of he referees to ‘ele in sige legend been eee oe web ‘esi of i ate)