Toward Long-Term Aquatic Science Products from Heritage Landsat Missions

Article Toward Long-Term Aquatic Science Products from Heritage Landsat Missions

Nima Pahlevan 1,2,*, Sundarabalan V. Balasubramanian 1,3, Sudipta Sarkar 1,2 and Bryan A. Franz 1 ID

1 NASA Goddard Space Flight Center, 8800 Greenbelt Road, Greenbelt, MD 20771, USA; sundarabalan.velaudhaperumalbalasubramanian@nasa.gov (S.V.B.); [email protected] (S.S.); [email protected] (B.A.F.) 2 Science Systems and Applications, Inc., 10210 Greenbelt Road, Suite 600 Lanham, MD 20706, USA 3 Department of Geographical Sciences, University of Maryland, College Park, MD 20740, USA * Correspondence: [email protected]; Tel.: +1-301-614-6684

Received: 29 January 2018; Accepted: 15 August 2018; Published: 22 August 2018

Abstract: This paper aims at generating a long-term consistent record of Landsat-derived remote sensing reflectance (Rrs) products, which are central for producing downstream aquatic science products (e.g., concentrations of total suspended solids). The products are derived from Landsat-5 and Landsat-7 observations leading to Landsat-8 era to enable retrospective analyses of inland and nearshore coastal waters. In doing so, the data processing was built into the SeaWiFS Data Analysis System (SeaDAS) followed by vicariously calibrating Landsat-7 and -5 data using reference in situ measurements and near-concurrent ocean color products, respectively. The derived Rrs products are then validated using (a) matchups using the Aerosol Robotic Network (AERONET) data measured by in situ radiometers, i.e., AERONET-OC, and (b) ocean color products at select sites in North America. Following the vicarious calibration adjustments, it is found that the overall biases in Rrs products are significantly reduced. The root-mean-square errors (RMSE), however, indicate noticeable uncertainties due to random and systematic noise. Long-term (since 1984) seasonal Rrs composites over 12 coastal and inland systems are further evaluated to explore the utility of Landsat archive processed via SeaDAS. With all the qualitative and quantitative assessments, it is concluded that with careful algorithm developments, it is possible to discern natural variability in historic water quality conditions using heritage Landsat missions. This requires the changes in Rrs exceed maximum expected uncertainties, i.e., 0.0015 [1/sr], estimated from mean RMSEs associated with the matchups and intercomparison analyses. It is also anticipated that Landsat-5 products will be less susceptible to uncertainties in turbid waters with Rrs(660) > 0.004 [1/sr], which is equivalent of ~1.2% reflectance. Overall, end-users may utilize heritage Rrs products with “fitness-for-purpose” concept in mind, i.e., products could be valuable for one application but may not be viable for another. Further research should be dedicated to enhancing atmospheric correction to account for non-negligible near-infrared reflectance in CDOM-rich and extremely turbid waters.

Keywords: Landsat; coastal/inland waters; atmospheric correction; vicarious calibration; validation; water quality; time-series applications

1. Introduction The Landsat program is among the longest Earth-observing programs on record, with continuous operations dating from 1972. The primary goal for the Landsat series is to monitor Earth resources from space using best-available technology. Following successful launches and operations of Landsat 1, 2, and 3, which carried analogue cameras [1], Landsat-4 and -5 carried the Thematic Mapper (TM),

Remote Sens. 2018, 10, 1337; doi:10.3390/rs10091337 www.mdpi.com/journal/remotesensing Remote Sens. 2018, 10, 1337 2 of 23 which, at the time, had been considered a technological breakthrough, i.e., 30 m spatial sampling, six spectral bands, 8-bit radiometric resolution, and a single thermal channel. A similar technology with some improvements [2] was utilized for the Enhanced Thematic Mapper Plus (ETM+) launched in 1999.Remote The TMSens. and2018, 10 ETM+, x FOR PEER were REVIEW whiskbroom imagers with 16 detectors laid out in the2 along-track of 23 direction [3]; thus, with each mirror scan, ~480 m is swept on the ground. The short dwell-time together with 8-bitsampling, radiometric six spectral resolution, bands, however,8-bit radiometric yields resolution, low signal-to-noise and a single ratiosthermal (SNR) channel. for A dark similar surfaces such astechnology water bodies with[ 4some,5]. Theimprovements two instruments [2] was utilized carried for stimulation the Enhanced lamps Thematic to allowMapper for Plus onboard (ETM+) launched in 1999. The TM and ETM+ were whiskbroom imagers with 16 detectors laid out in radiancethe calibrations along-track [ 6direction–8]. In addition[3]; thus, with to onboard each mirro calibrationr scan, ~480 observations, m is swept on the multiple ground. desert The short sites have historicallydwell-time been usedtogether to monitorwith 8-bit long-termradiometric stabilityresolution, of however, these missions. yields low For signal-to-noise reference, theratios relative spectral responses(SNR) for dark (RSR) surfaces for TM such and as water ETM+ bodies in the [4,5]. visible The two and instruments the near infrared carried stimulation (NIR) regions lamps alongside to those of theallow most for recentonboard Landsat radiance sensor, calibrations i.e., the [6–8]. Operational In addition Land to Imager onboard (OLI), calibration are illustrated observations, in Figure 1 . Currentlymultiple available desert sites ocean have historically color satellites been used are to not monitor suitable long-term for long-term stability of monitoring these missions. of For estuarine reference, the relative spectral responses (RSR) for TM and ETM+ in the visible and the near infrared environments. First, well-calibrated, dedicated ocean color sensors (e.g., the Sea-viewing Wide (NIR) regions alongside those of the most recent Landsat sensor, i.e., the Operational Land Imager Field-of-view(OLI), are Sensor illustrated (SeaWiFS)), in Figure have1. only become available since the late 1990s, so longer term historical studiesCurrently [9] available are not ocean possible. color satellites Second, are ocean not suitable color missionsfor long-term are monitoring generally of equippedestuarine with 1 km nadirenvironments. ground sampling First, well-calibra distanceted, [ 10dedicated]; therefore, ocean lackcolor sufficientsensors (e.g., spatial the Sea-viewing resolution Wide to Field- capture the spatial variabilityof-view Sensor necessary (SeaWiFS)), to observe have only critical become features, available such since as the upwelling, late 1990s, so small-scale longer term eddies/blooms, historical river plumes,studiestidal [9] are effects, not possible. and local Second, gradients ocean color in inlandmissions and are nearshoregenerally equipped waters. with Third, 1 km due nadir to steep ground sampling distance [10]; therefore, lack sufficient spatial resolution to capture the spatial gradients and high variability in optical properties as well as interfering optical signals from high variability necessary to observe critical features, such as upwelling, small-scale eddies/blooms, river concentrationsplumes, of tidal particles, effects, and chlorophyll-a local gradients (Chl), in inland and and colored nearshore dissolved waters. organic Third, due matter to steep (CDOM) gradients in these regions, aand clear high delineation variability ofin these optical water properties quality as properties well as isinterfering difficult withoutoptical signals careful from calibration high and validationconcentrations with in situ of data particles, [11]. Moreover,chlorophyll-a although (Chl), and standard, colored dissolved operational organic ocean matter color (CDOM) algorithms in are well-suitedthese for regions, ocean colora clear imagers delineation (e.g., of SeaWiFS) these water for quality relatively properties clear ocean is difficult waters, without to examine careful inland calibration and validation with in situ data [11]. Moreover, although standard, operational ocean and nearshore coastal waters, the development of algorithms customized to specific aquatic systems of color algorithms are well-suited for ocean color imagers (e.g., SeaWiFS) for relatively clear ocean interest iswaters, necessary to examine for Landsat-type inland and nearshore (broad-band) coastal wa sensors.ters, the development of algorithms customized Theto long specific record aquatic of systems Landsat of interest observations is necessary provides for Landsat-type unprecedented (broad-band) synoptic sensors. views of Earth resources; a recordThe long that record has beenof Landsat substantially observations under-utilized provides unprecedented in aquatic studies.synoptic Recognizingviews of Earth that the TM andresources; ETM+ sensors a record (a) that were has been equipped substantially with onlyunder three-utilized visible in aquatic bands studies. (Figure Recognizing1), (b) lack that sufficientthe SNR [5,12TM], and and ETM+ (c) suffer sensors from (a) were artifacts equipped (e.g., with memory only three effects, visible bands striping) (Figure [13 1),], (b) they lack stillsufficient have been SNR [5,12], and (c) suffer from artifacts (e.g., memory effects, striping) [13], they still have been found found useful in some aquatic studies. These studies have primarily revolved around quantifying useful in some aquatic studies. These studies have primarily revolved around quantifying concentrationsconcentrations of total of total suspended suspended solids solids (TSS) (TSS) [14–23] [14–23. ].Attempts Attempts have have also been also made been to madequantify to Chl quantify Chl [24–27[24–27]] and and the the absorption absorption byby CDOM CDOM [28–30]. [28–30 ].

Figure 1.FigureThe relative 1. The relative spectral spectral response response (RSR) (RSR) for for the the ThematicThematic Mapper Mapper (TM), (TM), the Enhanced the Enhanced Thematic Thematic Mapper Plus (ETM+), and the Operational Land Imager (OLI) with their corresponding nominal Mapper Plus (ETM+), and the Operational Land Imager (OLI) with their corresponding nominal center center wavelength. The largest difference between TM/ETM+ and OLI is in the near infrared (NIR) wavelength.channel The where largest OLI difference band was designed between narrow TM/ETM+ to avoid and water-vapor OLI is in absorption the near in infrared the atmosphere. (NIR) channel where OLI band was designed narrow to avoid water-vapor absorption in the atmosphere.

Remote Sens. 2018, 10, 1337 3 of 23

Further applications include analyzing bottom properties and bathymetry [31–35]. Many studies have also processed and examined time-series of images to assess feasibility of using Landsat for monitoring applications [9,27,36–45]. Although the utilization of the Landsat archive has recently increased (since 2008 when the entire archive was made available to public at no cost [46]), the invaluable historic image data have not yet been fully exploited. To date, every retrospective study has followed a different strategy for data processing and/or atmospheric corrections [9,47]. Some studies have not even applied atmospheric corrections and used Level-1 (top-of-atmosphere; TOA) radiance (Lt) to evaluate water quality [36]. In general, over-water TM and ETM+ images are expected to contain relatively large uncertainties (compared to those of ocean color missions, like the Moderate Resolution Imaging Spectroradiometer (MODIS) [48–50]); however, with a common processing system and vicarious corrections to Level-1B data, we anticipate that not only can retrospective studies be conducted but also current and future research can be traced back to the past. Owing to the successful integrations of Landsat-8 [51] and Sentinel-2 [52] processing in the SeaWiFS Data Analysis System (SeaDAS), we intend to expand this capability to Landsat-5 and -7 (to the extent possible within the limitations of instrument radiometric performances). More specifically, we implement a physics-based and extensively validated atmospheric correction method to generate Landsat products for the science community; enabling long-term studies of water quality in inland and nearshore coastal areas. Therefore, we first implement the atmospheric correction method (Section 2.1) through which non-negligible water-leaving radiance (Lw) in the near-infrared (NIR) band [53] is accounted for using a slightly different strategy than what is currently built into SeaDAS for Landsat-8 and Sentinel-2 [54]. Then, to reduce existing calibration biases in TOA observations of TM and ETM+, we perform vicarious calibrations (Section 2.2) using reference ocean color data [55,56]. Due to uncertainties in the vicarious calibrations, intercomparisons/cross-calibrations are further conducted (Section 2.3) to minimize residual biases and ensure consistency in Level-2 products, i.e., remote sensing reflectance (Rrs), with those derived from more recent missions. Sections 3.1 and 3.2 provide qualitative and quantitative assessments of TM- and ETM+-derived Rrs products. The quantitative evaluation is carried out using the Aerosol Robotic Network (AERONET) data measured by in situ ocean color radiometers, i.e., AERONET-OC. Here, the remote sensing reflectance is defined as the ratio of Lw to the total downwelling irradiance just above the water surface. Since AERONET-OC sites do not adequately represent aquatic/atmospheric regimes in inland areas, the heritage Rrs products are further cross-validated against those derived from ocean color sensors over relatively large inland waters. Further, time-series of Rrs products over 12 aquatic systems across the continental United States are analyzed (Section 3.3) to examine the utility of products derived from the three Landsat missions, i.e., TM, ETM+, and OLI. Further details on the limitations of heritage Landsat missions for retrospective water quality studies in inland and nearshore coastal areas are provided in Section4.

2. Materials and Methods Although the atmospheric correction methodology [57,58] and its integration [51] have been described in previous studies, we provide a brief description of how Lw(NIR) is accounted for in the TM and ETM+ processing (Section 2.1). This section continues with computing vicarious calibration gains for TM and ETM+ data. This is an essential step towards minimizing biases in products over bodies of water [55,56], which has proven to enhance the overall quality of Landsat-8 and Sentinel-2A Rrs products [52,59]. The derived gains are further slightly adjusted using cross-calibrations against ocean color products derived from OLI [59] and MODIS [48–50].

2.1. Atmospheric Correction Similar to the implementations for Landsat-8 and Sentinel-2, per-pixel geometry data, i.e., view zenith/azimuth angles and solar zenith/azimuth angles, were utilized in the processing. The current methodology to correct for Lw(NIR) 6= 0 is based on a bio-optical model, which uses Rrs(443) data [54]. Remote Sens. 2018, 10, x FOR PEER REVIEW 4 of 23 Remote Sens. 2018, 10, 1337 4 of 23

subsurface remote sensing reflectance [60]. The empirical method for estimating 𝜂 (referred to as

More𝜂/ specifically,) is as follows this model utilizes rrs(443)/rrs(561) to estimate the spectral slope (η) of particulate backscattering (b ), which is applied in the retrieval of Rrs(NIR),.  where() rrs is the subsurface remote bp 𝜂 =211.2  𝑒𝑥𝑝 / ( ) (1) sensing reflectance [60]. The empirical method for estimating η(referred to as η443/561) is as follows The TM and ETM+ spectral band centers are tabulated in Table 1. Throughout this manuscript, we 0.9rrs(443) refer to their visible bands asη4443/561 blue and= 2 green1 − 1.2 bandsexp −unless specific bands for each instrument (1)are rrs(561) discussed. Since the 443 nm channel is not available for TM and ETM+, we developed a slightly differentThe TM strategy. and ETM+ To do spectral so, in situ band radiometric centers are da tabulatedta collected in Table over1 various. Throughout inland this waters manuscript, [61] were weused refer to toexamine their visible the correlations bands as bluebetween and green𝑟(443 bands)⁄𝑟( unless561) and specific 𝑟( bands482)⁄𝑟 for(561 each) instrumentfor OLI. We arefound discussed. a strong Since relationship the 443 nm between channel the is two not ratios available (Figure for TM 2 (left)). and ETM+, Hence, we the developed ratio in Equation a slightly (1) differentcan potentially strategy. be To replaced do so, inby situ𝑟( radiometric482)⁄𝑟(561 data) , which collected results over in various the equation inland below waters [61] were [ ( ) ( )] [ ( ) .( )] used to examine the correlations between rrs 443 /rrs 561 and rrs(482) /rrs 561 for OLI. We found 𝜂/ = 2 2.2 ∗ 𝑒𝑥𝑝(0.9 × 0.7554 ) (2) a strong relationship between the two ratios (Figure2 (left)). Hence,( the) ratio in Equation (1) can potentially be replaced by [rrs(482)/rrs(561)], which results in the equation below The performance of the newly estimated 𝜂/ was tested against 𝜂/ in Figure 2 (right). Similar relationships were then built for TM and ETM+. This assessment! further suggests that r (482) 1.3994 Equation (2) can be utilizedη to= estimate2 − 2.2 ∗𝑅exp(NIR)−0.9 in× an0.7554 iterativers fashion for processing TM and ETM+(2) 482/561 r (561) as implemented for OLI. rs

Figure 2. The strong relationship (left) between the band ratios of subsurface remote sensing Figure 2. The strong relationship (left) between the band ratios of subsurface remote sensing reﬂectance reflectance in the blue and green channels, i.e., 𝑟(482/561) and 𝑟(443/561). The relationship follows in the blue and green channels, i.e., rrs(482/561) and rrs (443/561). The relationship follows the the power-law function (A = 0.7554, B = 1.3994) and provides a high correlation coefficient, i.e., 𝑅 = power-law function (A = 0.7554, B = 1.3994) and provides a high correlation coefﬁcient, i.e., R2 = 0.987. 0.987. The scatterplot (right) shows the cross-validation of 𝜂 derived using two different band ratios, The scatterplot (right) shows the cross-validation of η derived using two different band ratios, i.e., rrs i.e., 𝑟(443 /561) and 𝑟(482/561). (443 /561) and rrs (482/561).

Table 1. The OLI, ETM+ and TM spectral band centers (nm) computed in SeaDAS. Table 1. The OLI, ETM+ and TM spectral band centers (nm) computed in SeaDAS. Blue [nm] Green [nm] Red [nm] NIR [nm] SWIR-I [nm] SWIR-II [nm] OLI Blue482 [nm] Green 561 [nm] Red 655 [nm] NIR 865 [nm] SWIR-I 1609 [nm] SWIR-II 2201 [nm] OLIETM+ 482478 561 560 655 661 865835 1648 1609 22052201 ETM+TM 478485 560 569 661 660 835840 1676 1648 22232205 TM 485 569 660 840 1676 2223 2.2. Vicarious Calibration The performance of the newly estimated η482/561 was tested against η443/561 in Figure2 (right). In this exercise, the TM and ETM+ data were vicariously calibrated using different reference data Similar relationships were then built for TM and ETM+. This assessment further suggests that to reduce systematic biases in the three visible bands (Table 1). Theoretically, all the Earth-observing Equation (2) can be utilized to estimate Rrs(NIR) in an iterative fashion for processing TM and ETM+ satellites passing over one of the existing vicarious calibration systems, such as the Marine Optical as implemented for OLI. Buoy (MOBY) [62] or BOUSSOLE [63], can be vicariously calibrated. The ETM+ data, with regular 2.2.image Vicarious acquisitions Calibration over both sites, can thus be vicariously calibrated. However, due to the limitations in the downlink capacity of Landsat-5, TM images were acquired neither over the MOBY nor over In this exercise, the TM and ETM+ data were vicariously calibrated using different reference data the BOUSSOLE site, which became operational in 2003. Hence, we employed MODIS data to to reduce systematic biases in the three visible bands (Table1). Theoretically, all the Earth-observing vicariously calibrate, i.e., cross-calibrate, TM data.

Remote Sens. 2018, 10, 1337 5 of 23 satellites passing over one of the existing vicarious calibration systems, such as the Marine Optical Buoy (MOBY) [62] or BOUSSOLE [63], can be vicariously calibrated. The ETM+ data, with regular image acquisitions over both sites, can thus be vicariously calibrated. However, due to the limitations in the downlink capacity of Landsat-5, TM images were acquired neither over the MOBY nor over the BOUSSOLE site, which became operational in 2003. Hence, we employed MODIS data to vicariously calibrate, i.e., cross-calibrate, TM data.

2.2.1. Landsat-7 (ETM+) The time-series of all USGS ETM+ (Collection-1) images over the MOBY and BOUSSOLE sites were obtained and processed using SeaDAS to compute the vicarious gains [56]. The calibration is performed in two steps. In Step I, the NIR and the Short-wave Infrared (SWIR) bands are calibrated. This is carried out by selecting scenes with low aerosol loading, i.e., aerosol optical thicknesses in the NIR (AOT(835)) < 0.15. The goal is to minimize propagations of systematic errors (e.g., calibration biases) in aerosol path radiance estimates for these channels when generating Rrs products [64]. With the calibration gains derived for these bands, the vicarious gains for the three visible bands were derived in Step II. Overall, 58 cloud-free images over the MOBY (n = 30) and BOUSSOLE (n = 28) sites were utilized in the analyses. While MOBY hyperspectral Lw resampled to ETM+ multispectral bands were used, the BOUSSOLE narrow-band, multispectral Lw were converted to ETM+ multispectral bands using a trained deep neural network [61]. The vicarious gains were initially produced using three different band combinations to account for aerosol contributions. The band combinations for ETM+ were 835–1648, 835–2205, and 1648–2205 (Table1). Similar to the results in Reference [59], preliminary analyses of the Rrs retrievals at AERONET-OC [65] sites indicated that the most robust matchups were produced with 835 and 1648 nm bands (Table1); therefore, we only present results derived with this band pair. The median gains within 5 × 5-element windows centered over the exact coordinates are utilized to represent spectral gains for each ETM+ image. The median statistics were chosen to suppress image artifacts, including striping. The median statistics were found to be nearly identical (∼ 10−4) to the average computed within the interquartile range [66]. Furthermore, a sensitivity analysis on the window size, i.e., deriving gains for 3 × 3-, 7 × 7-, 9 × 9-, and 11×11-element kernels, indicated minimal differences, i.e., <10−3, in the derived gains as a function of the window size. The time-series of vicarious gains for the visible bands derived using the 835–1648 combination are shown in Figure3. The median gains and one standard deviation (in round brackets) are reported in Table2. The MOBY and BOUSSOLE median gains in the blue, green, and red channels were 1.033, 1.011, 1.021 and 1.027, 1.009, 1.033, respectively. The gains listed in Table2 are the average of the two. The relatively large difference in the red channel can be attributed to uncertainties in the spectral band adjustments for the multispectral ﬁeld observations at the BOUSSOLE site. While the standard deviations (σ = 2.3%) were found comparable to the gains derived from BOUSSOLE data for SeaWiFS, they are twice as large as those derived from MOBY data [66]. The high uncertainties suggest that there may be residual errors in the gains that can be identiﬁed and accounted for using further cross-calibration (see Section 2.3). Therefore, we refer to these gains as preliminary. Radiometric artifacts (e.g., detector striping) and low SNRs in the NIR and SWIR bands contribute to large uncertainties in the gains [64], as do environmental factors like cirrus clouds or surface waves.

Table 2. The vicarious gains for the ETM+ and TM bands. One-standard deviations are presented in round brackets. Note that these gains are derived from the NIR-SWIR-I band combination. The gains for these two bands are derived in step I.

Sensor Blue Green Red NIR SWIR-I SWIR-II ETM+ 1.030 (0.023) 1.01 (0.021) 1.0278 (0.022) 1.002 (0.008) 0.964 (0.12) 0.991 (0.15) TM 1.047 (0.02) 1.011 (0.024) 1.0581 (0.032) 0.956 (0.011) 1.095 (0.13) 1.0151 (0.072) Remote Sens. 2018, 10, 1337 6 of 23 Remote Sens. 2018, 10, x FOR PEER REVIEW 6 of 23

Figure 3. The vicarious gains (n = 58) for ETM+ visible bands. Each data point corresponds to a Figure 3. The vicarious gains (n = 58) for ETM+ visible bands. Each data point corresponds to a computed gain over the MOBY and/or BOUSSOLE site. The temporally median gains and (one) computed gain over the MOBY and/or BOUSSOLE site. The temporally median gains and (one) standard deviation applied to ETM+ data are provided in Table 2. Note that the gains are derived standard deviation applied to ETM+ data are provided in Table2. Note that the gains are derived after after vicariously calibrating NIR and SWIR-I bands in Step I. vicariously calibrating NIR and SWIR-I bands in Step I. 2.2.2. Landsat-5 (TM) 2.2.2. Landsat-5 (TM) Due to the absence of TM images over the calibration sites, we used normalized 𝐿 (𝑛𝐿 ) Dueproducts to the derived absence from of the TM MODIS images observations over the co calibrationllected onboard sites, both we Aqua used and normalized Terra platformsLw (nL w) productssince derived 1999. For from this exercise, the MODIS near-concurrent observations (defined collected as image-pair onboard collections both Aqua on andthe same Terra day) platforms TM since 1999.and MODIS For this images exercise, over a near-concurrent number of preselected (deﬁned locations as image-pair were analyzed. collections The selected on the locations same day) were those considered favorable sites for potential future vicarious calibration systems, due to factors TM and MODIS images over a number of preselected locations were analyzed. The selected such as frequency of cloud cover and temporal and spatial stability of atmosphere and water optical locationsproperties were those[67]. In considered effect, we cross-calibrated favorable sites TM fordata potential with MODIS future products vicarious at these calibration stable locations. systems, due toFirst, factors the suchcloud- as and frequency glint-free of USGS cloud TM cover (Colle andction-1) temporal images and were spatial visually stability identified. of atmosphere The and watercorresponding optical propertiesLevel-2 [MODIS67]. In effect,products we cross-calibratedwere obtained TMfrom data NASA’s with MODISweb portal: products at thesehttps://oceancolor.gsfc.nasa.gov/. stable locations. First, the Note cloud- that and th glint-freeese standard USGS products TM (Collection-1) have undergone images very were visuallyconservative identiﬁed. pre-processing The corresponding procedures. Level-2 Areas MODISwith low productsChl, i.e., a median were obtained of 0.25 mg from⁄ m, NASA’sAOT(869) web portal:< https://oceancolor.gsfc.nasa.gov/0.15, and within 60° viewing zenith. angles, Note that were these selected standard to perform products the vicarious have undergone calibration. very conservativeThese products pre-processing are assumed procedures. to carry Areasminimal with unce lowrtainties Chl, i.e.,in the a median atmospheric of 0.25 correctionmg/m3 ,process AOT(869) [66]. The median Rrs spectra were extracted from MODIS 3×3-element windows. The sites include < 0.15, and within ±60◦ viewing zenith angles, were selected to perform the vicarious calibration. West Australia (n = 60), US West Coast (n = 65), São Miguel Island (n = 25), and the Northern These products are assumed to carry minimal uncertainties in the atmospheric correction process [66]. Mediterranean Sea (n = 46). Note that the Mediterranean site is centered around the existing × The medianBOUSSOLERrs spectra site. The were TM data extracted over the from other MODIS suggested3 calibration3-element sites windows. (e.g., Puerto The Rico) sites were include scarce West Australiaand (thusn = 60),not used. US West A total Coast of 196 (n =matchups 65), São were Miguel identified Island for (n =this 25), vicarious and the calibration Northern exercise. Mediterranean The

Sea (n MODIS-derived= 46). Note that 𝑅 the were Mediterranean then adjusted site for is differences centered around in spectral the existingbands to BOUSSOLE create TM-equivalent site. The TM data overMODIS the otherdata (𝑅 suggested ) [61]. calibration sites (e.g., Puerto Rico) were scarce and thus not used. A total of 196 matchupsThe TM were images identiﬁed were processed for this vicariousin the vicarious calibration calibration exercise. mode The in MODIS-derivedSeaDAS by supplyingRrs were 𝑅 spectra. Given MODIS viewing zenith angles, TM image products were spatiallyMOD sampled−TM then adjusted for differences in spectral bands to create TM-equivalent MODIS data (Rrs )[61]. Thesuch TM that images they accurately were processed represent the in extent the vicarious “viewed” calibration by MODIS. Similar mode into the SeaDAS ETM+ processing, by supplying the gain calculations were carried out in two steps (see Section 2.1.1) and we only present gains RMOD−TM spectra. Given MODIS viewing zenith angles, TM image products were spatially sampled rs derived from the NIR and SWIR-I, i.e., 840–1676. Here, we assume a black ocean in these two bands, such that they accurately represent the extent “viewed” by MODIS. Similar to the ETM+ processing, which is valid for these relatively clear waters. Figure 4 illustrates time-series of gains derived for the the gainTM calculations data. It should were also carried be noted out that in twoTerra-MODIS steps (see data Section collected 2.2.1 only) and prior we onlyto 2003 present were used gains in derived the from thevicarious NIR and calibration SWIR-I, [68]. i.e., The 840–1676. derived Here, gains weare assumecolor-coded a black differently ocean infor these the four two sites. bands, Table which 2 is valid forcontains these median relatively gains clear and waters. (one) standard Figure4 illustratesdeviations. time-series Similar to the of gainsETM+ derived gain derivations, for the TM the data. It should also be noted that Terra-MODIS data collected only prior to 2003 were used in the vicarious calibration [68]. The derived gains are color-coded differently for the four sites. Table2 contains median gains and (one) standard deviations. Similar to the ETM+ gain derivations, the median statistics were Remote Sens. 2018, 10, 1337 7 of 23

Remote Sens. 2018, 10, x FOR PEER REVIEW 7 of 23 very similar to the mean value calculated within the interquartile range. It can be inferred that the TM radiometricmedian statistics responses were in very the visiblesimilar to bands the mean are biasedvalue calculated low. Analyses within the of gainsinterquartile computed range. at It eachcan site indicatedbe inferred noticeable that the differences TM radiometric in the responses blue gains, in th i.e.,e visible 1.031, bands 1.046, are 1.054, biased and low. 1.056 Analyses for the of gains four sites. Furthermore,computed TMat each images site indicated contain morenoticeable systematic differences errors in the (e.g., blue nonlinearity gains, i.e., 1.031, in instrument 1.046, 1.054, response) and 1.056 for the four sites. Furthermore, TM images contain more systematic errors (e.g., nonlinearity in than those of ETM+. Uncertainty in spectral band adjustments is another source of error in gain instrument response) than those of ETM+. Uncertainty in spectral band adjustments is another source computations. Nevertheless, the vicarious gains should largely minimize the biases in the long-term of error in gain computations. Nevertheless, the vicarious gains should largely minimize the biases Landsat-5in the datalong-term record Landsat-5 (see Section data record3). (see Section 3).

FigureFigure 4. The 4. The vicarious vicarious gains gains for TM TM visible visible bands. bands. Each Each data point data corresponds point corresponds to a TM-MODIS to a TM-MODIS near- near-concurrentconcurrent matchup, matchup, i.e., i.e.,same-day same-day overpasses, overpasses, over various over variouscalibration calibration sites. The temporally sites. The median temporally mediangains gains and and(one) (one) standard standard deviation deviation applied applied to TM da tota TM are dataprovided are provided in Table 2. in Note Table that2. Notethe gains that the are derived after vicariously calibrating NIR and SWIR-I bands. gains are derived after vicariously calibrating NIR and SWIR-I bands. 2.3. Gain Adjustments 2.3. Gain Adjustments Considering the uncertainties (i.e., instrument artifacts and environmental noise), there remain Consideringresidual biases the in both uncertainties TM and ETM+ (i.e., gains. instrument Therefore, artifacts these gains and environmentalonly provide initial noise), estimates there for remain residualthe biases and in both slight TM adjustments and ETM+ may gains. be Therefore, inevitable. theseIn this gains section, only we provide will further initial examine estimates the for the biasesrobustness and slight of adjustments these gains may by beevaluating inevitable. TM- In and this section,ETM+-derived we will 𝑅 further products examine against the robustnessthose measured/derived from AERONET-OC and OLI-derived 𝑅 products and fine-tune them as of these gains by evaluating TM- and ETM+-derived Rrs products against those measured/derived needed. Because of their orbit configurations, Landsat-5 and -7 and Landsat-7 and -8 do not create from AERONET-OC and OLI-derived Rrs products and ﬁne-tune them as needed. Because of their opportunities for direct intercomparisons (of TOA products) for which near-simultaneous overpasses orbit conﬁgurations, Landsat-5 and -7 and Landsat-7 and -8 do not create opportunities for direct are necessary [69–71]. Inherently, the satellite pairs observe a target at 8-day intervals. It is thus intercomparisons (of TOA products) for which near-simultaneous overpasses are necessary [69–71]. imperative to choose sites with optically stable waters within this time window and assess the 𝑅 Inherently,products the (Level-2). satellite We pairs chose observe Crater Lake a target and atLake 8-day Tahoe intervals. representing It is oligotrophic thus imperative and mesotrophic to choose sites withwater optically bodies, stable respectively waters within [69]. Here, this timethe OLI window data products and assess (in theoperationRrs products since 2013 (Level-2). and in good We chose Crateragreement Lake and with Lake MODIS Tahoe [59]) representing are regarded oligotrophic as references and for mesotrophic evaluating waterETM+ bodies,products, respectively which are [69]. Here,then the utilized OLI data to evaluate products TM (in products operation within since the 20131999–2012 and inperiod. good These agreement sites, however, with MODIS allow only [59]) are regardedfor intercomparisons as references for over evaluating a limited ETM+ signal products,range in the which blue, arei.e., then 0.002 utilized < Rrs (485) to evaluate< 0.008 [1/sr], TM productsand withineven the smaller 1999–2012 in period.the green These and sites,red channels. however, To allow extend only that for intercomparisonsrange, we also performed over a limited intercomparisons at AERONET-OC sites. Note that in both analyses, uncertainties in (a) the signal range in the blue, i.e., 0.002 < Rrs (485) < 0.008 [1/sr], and even smaller in the green and red atmospheric correction and (b) OLI products are unavoidable. channels. To extend that range, we also performed intercomparisons at AERONET-OC sites. Note that in both analyses, uncertainties in (a) the atmospheric correction and (b) OLI products are unavoidable. To begin, TM and ETM+ data over these sites (i.e., all available matchups at AERONET-OC stations, Crater and Tahoe lakes) were processed using the computed vicarious gains (Table2). Initial Remote Sens. 2018, 10, 1337 8 of 23

analyses of the data revealed that the TM- and ETM+-derived Rrs products are slightly biased low (i.e., Remote Sens. 2018, 10, x FOR PEER REVIEW 8 of 23 <0.0016 [1/sr]) in the blue and red channels. The difference was found largest in the TM’s blue band. As an example,To begin, Figure TM5 illustrates and ETM+ thedata correspondingover these sites (i.e., scatterplots all available for matchups the OLI-ETM+ at AERONET-OC and TM-ETM+ intercomparisonsstations, Crater in the and blue Tahoe bands. lakes) were The processed root-mean-square using the computed differences vicarious (RMSD), gains (Table the median 2). Initial relative 𝑅 differencesanalyses (MRD), of the and data median revealed biases that the (Bias) TM- forand theETM+-derived OLI-ETM+ matchups products canare slightly be computed biased low to gauge (i.e., <0.0016 [1/sr]) in the blue and red channels. The difference was found largest in the TM’s blue the relativeband. differences: As an example, Figure 5 illustrates the corresponding scatterplots for the OLI-ETM+ and TM- ETM+ intercomparisons in the rblue bands. The root-mean-square differences (RMSD), the median h i2 relative differences (MRD),(λ) and= medianN= 216biasesETM (Bias)+(λ for) − theOLI OLI-ETM+(λ) matchups[ ] can be computed RMSD ∑1 Rrs(i) Rrs(i) /N 1/sr (3) to gauge the relative differences:

 ETM+(λ) OLI(λ)  R () − R() (3) RMSD(λ)= ∑ rs𝑅(i)() 𝑅rs()(i) 𝑁 [1/sr] MRD(λ) = median  × 100 [] (4) ()AERONET()(λ) Rrs(()i) () MRD(λ) = median () × 100 [ ] (4) h () i (λ) = ETM+(λ) − OLI(λ) [ ] Bias median Rrs(()i) R()rs(i) 1/sr (5) Bias(λ) = median𝑅() 𝑅() [1/sr] (5) TheThe RMSD, RMSD, Bias, Bias, andand the slope slope for for the the linear linear fit are fit reported are reported in Figure in Figure5. After5 .running After a running few a few experimentsexperiments to to minimize minimize thethe differencesdifferences in in TM TM and and ETM+ ETM+ products products with withrespects respects to those to of those of AERONET-OCAERONET-OC and OLI, and finalOLI, final vicarious vicarious adjustments adjustments werewere determined. determined. Table Table 3 contains3 contains these thesefinal final gains. Itgains. should It should be noted be noted that that due due to largeto large uncertainties uncertainties in the the gain gain computations computations for the for SWIR-I the SWIR-I band band (see Table(see2), Table the largest 2), the largest adjustments adjustments were were made made in in these these bandsbands to to compensate compensate for potential for potential residual residual differences/biases in the visible bands [64]. The goodness of these final gains is independently tested differences/biases in the visible bands [64]. The goodness of these final gains is independently tested at Level-2 products against those of Aqua-MODIS (MODISA) and Terra-MODIS (MODIST) over a at Level-2handful products of sites against across Nort thoseh America of Aqua-MODIS (see Section 3.2.2). (MODISA) and Terra-MODIS (MODIST) over a handful of sites across North America (see Section 3.2.2).

Figure 5. The scatterplots obtained from intercomparisons of OLI-ETM+ and TM-ETM+ Rrs products Figure 5. The scatterplots obtained from intercomparisons of OLI-ETM+ and TM-ETM+ Rrs products over optically stable waters of Crater and Tahoe lakes. The gains tabulated in Table 2 were supplied over opticallyto SeaDAS stable to generate waters these of Crater products. and Final Tahoe gains lakes. are tabulated The gains in Table tabulated 3. Note in that Table the 2image were pairs supplied to SeaDASwere to acquired generate 8 days these apart. products. Final gains are tabulated in Table3. Note that the image pairs were acquired 8 days apart. Table 3. The final vicarious gains for after slight adjustments made to gains enlisted in Table 2. The Table 3. adjustmentsThe ﬁnal were vicarious minimal gains for the for green after bands. slight Note adjustments that no change made was to gainsmade to enlisted the SWIR-II in Table 2. channels. The adjustments were minimal for the green bands. Note that no change was made to the SWIR-II channels. Sensor Blue Green Red NIR SWIR-I ETM+ 1.034 1.012 1.039 0.992 1.12 SensorTM Blue 1.059 Green 1.011 1.058 Red 0.931 NIR 1.21 SWIR-I

ETM+ 1.034 1.012 1.039 0.992 1.12 TM 1.059 1.011 1.058 0.931 1.21

Remote Sens. 2018, 10, 1337 9 of 23

Remote Sens. 2018, 10, x FOR PEER REVIEW 9 of 23 3. Results 3. Results After implementing the vicarious gains (Table3), the generated Rrs products are ﬁrst evaluated for qualitativeAfter andimplementing quantitative the vicarious rigor followed gains (Table by analyzing 3), the generated the time-series 𝑅 products products. are first evaluated for qualitative and quantitative rigor followed by analyzing the time-series products. 3.1. Qualitative Assessment of Rrs Products 3.1. Qualitative Assessment of Rrs Products Two examples of products processed via SeaDAS are illustrated in Figures6 and7. The goal is to qualitativelyTwo examples demonstrate of products the processed TM and via ETM+ SeaDAS products are illustrated and visually in Figures compare 6 and 7. The them goal to is those to qualitatively demonstrate the TM and ETM+ products and visually compare them to those obtained from OLI. Note that these products are associated with different seasons/times; therefore, obtained from OLI. Note that these products are associated with different seasons/times; therefore, this analysisthis analysis is not is not suitable suitable to to gauge gauge consistency, consistency, whichwhich will will be be discussed discussed in inSections Sections 3.2.2 3.2.2 and and3.3. 3.3. Note thatNote athat7 × a7 7×7-element-element (uniform) (uniform) averaging averaging ﬁlter filter is is applied applied inin thethe aerosol aerosol removal removal process. process. Figure Figure 6 illustrates6 illustrates the Rrs theproducts 𝑅 products derived derived from from TM(left), TM (left), ETM+ ETM+ (middle), (middle), and and OLI OLI (right) (right) over over thethe turbidturbid and shallowand northern shallow northern San Francisco San Francisco Bay area Bay (i.e., area San (i.e., Pablo San Bay). Pablo The Bay). natural The natural variability variability (e.g., gradients(e.g., in opticalgradients regimes) in optical in water-column regimes) in water-column is clearly capturedis clearly captured in TM andin TM ETM+ and ETM+ images. images. Small Small eddies, local resuspensions,eddies, local resuspensions, and upwelling and up eventswelling can events visually can visually be discerned be discerned in the in TM- the andTM- ETM+-derivedand ETM+- derived 𝑅 products. Nevertheless, artifacts are pronounced in TM and ETM+ products when Rrs products. Nevertheless, artifacts are pronounced in TM and ETM+ products when compared to thosecompared of OLI. The to those artifacts of OLI. in TM The products artifacts in (in TM particular products in(in the particular blue band) in the are blue due band) to memory are due to effect, memory effect, coherent noise, and striping [13]. Due to OLI’s higher SNR and absence of major coherent noise, and striping [13]. Due to OLI’s higher SNR and absence of major artifacts, the quality artifacts, the quality of OLI products is noticeably better than those of TM and ETM+. The scan-line of OLIcorrector products issue, is noticeably i.e., SLC-off, better [3] associated than those with of TMETM+ and retrievals ETM+. Theare masked, scan-line i.e., corrector white stripes. issue, i.e., SLC-off,Furthermore, [3] associated it should with be ETM+ noted retrievalsthat, although are TM masked, and ETM+ i.e., whitemeasure stripes. spectral Furthermore, radiances in three it should be notedbroad that, bands, although these bands TM and capture ETM+ relatively measure distinct spectral spectral radiances signatures in three of near-surface broad bands, constituents these bands captureand relatively their optical distinct properties. spectral signatures of near-surface constituents and their optical properties.

Figure 6. The Rrs products derived from TM (left), ETM+ (middle), and OLI (right) over the northern Figure 6. The Rrs products derived from TM (left), ETM+ (middle), and OLI (right) over the northern San FranciscoSan Francisco Bay Bay area. area. The The band band centers centers shown shown to to thethe right refer refer to to TM TM bands. bands. The The scan-line scan-line corrector corrector issues (i.e., SLC-off) associated with ETM+ retrievals are masked, i.e., white stripes. Varieties of issues (i.e., SLC-off) associated with ETM+ retrievals are masked, i.e., white stripes. Varieties of physical physical phenomena, including upwelling events and sediment resuspensions are captured by phenomena, including upwelling events and sediment resuspensions are captured by heritage Landsat heritage Landsat missions. Note that a 7×7-element (uniform) averaging filter is applied in the missions. Note that a 7 × 7-element (uniform) averaging ﬁlter is applied in the aerosol removal process. aerosol removal process.

Another demonstration of the natural variability in inland and nearshore coastal waters captured in Rrs(569) and Rrs(560) products derived from TM and ETM+ is illustrated in Figure7. The images Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 23 Remote Sens. 2018, 10, 1337 10 of 23 Another demonstration of the natural variability in inland and nearshore coastal waters captured in Rrs(569) and Rrs(560) products derived from TM and ETM+ is illustrated in Figure 7. The wereimages acquired were eight acquired days aparteight days in September apart in Septembe 2006 overr 2006 the easternover the North eastern Carolina North Carolina inland watersinland and shorelines.waters Theand shorelines. scan-line correctorThe scan-line issues corrector (i.e., SLC-off) issues (i.e., associated SLC-off) associated with ETM+ with retrievals ETM+ retrievals are masked (shownare withmasked white (shown stripes). with white The TM stripes). and ETM+The TM clearly and ETM+ reveal clearly various reveal water various types, water such types, as relativelysuch as relatively clear waters in the coastal shelf area, nearshore turbid waters, and CDOM-rich river clear waters in the coastal shelf area, nearshore turbid waters, and CDOM-rich river waters denoted waters denoted with arrows, i.e., Tar/Pamlico River. Resuspension events can also be inferred in withshallower arrows, i.e., waters. Tar/Pamlico These demonstrations River. Resuspension suggest the eventsremarkable can value also beof archived inferred Landsat in shallower data and waters. Thesetheir demonstrations utility enabled suggest via SeaDAS. the remarkable The negative value retrievals, of archived i.e., failure Landsat of the dataatmospheric and their correction, utility enabled in via SeaDAS.CDOM-rich The river negative waters retrievals, are a known i.e., problem failure requiring of the atmospheric further study correction, and investigations in CDOM-rich into the river watersfuture. are a known problem requiring further study and investigations into the future.

FigureFigure 7. Scene-wide 7. Scene-wide Rrs R(569)rs(569) and and R Rrsrs(560)(560) productsproducts derived derived from from TM TM (left (left) and) and ETM+ ETM+ (right (right) over) overthe the NorthNorth Carolina Carolina shorelines shorelines and and Pamlico Pamlico Sound Sound (USA). (USA). The The imagesimages were acquired eight eight days days apart apart on on 18 September18 September (TM) and (TM) 26 and September 26 September 2006 at2006 about at about 15:30 15:30 GMT. GMT. Large Large gradients, gradients, i.e., i.e., relatively relatively clear clear coastal shelfcoastal waters, shelf eddies, waters, resuspensions, eddies, resuspensions, CDOM-dominated CDOM-dominated waters andwaters bottom and features,bottom features, are captured are in the twocaptured images. in the The two CDOM-rich images. The watersCDOM-rich of Tar/Pamlico waters of Tar/Pamlico Rivers are Rivers indicated are indicated with arrows. with arrows. Note that Note that negative retrievals do exist in such challenging waters in the ETM+ product [72] and a negative retrievals do exist in such challenging waters in the ETM+ product [72] and a 7 × 7-element 7×7-element (uniform) averaging filter was applied in the aerosol removal process. (uniform) averaging ﬁlter was applied in the aerosol removal process. 3.2. Validation 3.2. Validation For a quantitative assessment of heritage Landsat products, we utilized two data sources; (a) ForAERONET-OC a quantitative data and assessment (b) Rrs products of heritage obtained Landsat from products,MODIS. This we was utilized to provide two dataas much sources; (a) AERONET-OCevidence as possible data to and ensure (b) Rthatrs products these products obtained were, from on average, MODIS. in Thisreasonable was toagreement provide with as much evidencestandard as possible products. to ensure that these products were, on average, in reasonable agreement with standard products. 3.2.1. AERONET-OC 3.2.1. AERONET-OCSince the early 2000s, the AERONET-OC data (https://aeronet.gsfc.nasa.gov) have been used as standard datasets to evaluate the quality of ocean color data products. Although the sites do not Since the early 2000s, the AERONET-OC data (https://aeronet.gsfc.nasa.gov) have been used represent global in-water conditions, they provide adequate and easily accessible matchups to as standardevaluate datasetsthe performance to evaluate of the both quality atmosphe of oceanric correction color data and products. instrument’s Although radiometric the sites do not representcharacteristics. global The in-water scatterplots conditions, for ETM+ they and TM provide data for adequate all the bands and easilyare shown accessible in Figure matchups 8. For to evaluatethe matchups, the performance hazy images of both and atmospheric areas affected correction by partial andsunglint instrument’s were removed radiometric from the characteristics. analyses. The scatterplotsA total of 216 for and ETM+ 79 matchups and TM were data found for for all ETM+ the bands and TM are data, shown respectively. in Figure For8. ForTM and the ETM+, matchups, hazy>90% images and and >65% areas of the affected matchups by partial correspond sunglint to modera were removedtely turbid from waters the of analyses. the Venise A total(located of 216in and 79 matchupsNorthern were Adriatic found Sea) for and ETM+ Martha’s and Vineyard TM data, (situated respectively. in southern For TMcoastal and waters ETM+, of Massachusetts) >90% and >65% of the matchupssites. The statistics correspond were to generated moderately in a turbidsimilar watersfashion ofas thein Equations Venise (located (3)–(5). in Northern Adriatic Sea) and Martha’s Vineyard (situated in southern coastal waters of Massachusetts) sites. The statistics were generated in a similar fashion as in Equations (3)–(5). Remote Sens.Remote2018 Sens., 10, 2018 1337, 10, x FOR PEER REVIEW 11 of 23 11 of 23

Figure 8. The scatterplots for TM and ETM+ matchups at AERONET-OC sites. The low median biases Figure 8.(TableThe 4) scatterplots indicate that forthe final TM vicarious and ETM+ gains, matchups on average, at have AERONET-OC enhanced the overall sites. consistency The low of median biases (TableLandsat-derived4) indicate products. that the The ﬁnal matchups vicarious exclude gains, retrievals on average, in areas have with enhancedpartial glints. the The overall consistencyAERONET-OC of Landsat-derived matchups (N products. = 216) enable The a robust matchups analysis exclude of the ETM+ retrievals products, in areas whereas with the partial glints. Thequality AERONET-OC of TM products matchups require fu (Nrther= 216) scrutiny enable (see aSection robust 3.2.2). analysis of the ETM+ products, whereas the quality of TM products require further scrutiny (see Section 3.2.2). The computed statistics after implementing gains (Table 3) are provided in Table 4. For reference, MRD and root-mean-square error (RMSE) computed with unity gains, i.e., prior to Theimplementing computed statistics the gains, after are given implementing in round brackets. gains Note (Table that3) RMSE are provided was computed in Table as in4. Eq. For 3; reference, yet, MRD andhere, root-mean-square we treat the AERONET-OC error (RMSE) data as computed reference standards. with unity Comparing gains, i.e., MRD prior and to RMSE implementing statistics the gains, areindicate given inthat round the vicarious brackets. gains, Note onthat average, RMSE have was significantly computed reduced as in Equation median biases (3); yet, in products here, we treat (in particular for TM); thereby enhancing the overall consistency in long-term Landsat-derived the AERONET-OC data as reference standards. Comparing MRD and RMSE statistics indicate that the products. The slope and intercepts for TM matchups allude to potential non-linearity at low radiance vicariouslevels. gains, Collectively, on average, the have statistics signiﬁcantly suggest that reduced TM products median should biases be inhandled products with (in care particular when for TM); therebyaccurate/precise enhancing products the overall (e.g., consistencyChl) are desired. in long-term The best performances Landsat-derived are, in general, products. found The in slopethe and interceptsgreen for TMchannels. matchups The relatively allude tolarge potential slopes in non-linearity ETM+ analyses at lowimplicate radiance potential levels. nonlinearity Collectively, in the statisticsinstrument suggest that responses. TM products We further should computed be handled the median with 𝑅 care spectra, when i.e., accurate/precise 𝑅, from AERONET-OC products (e.g., Chl) aredata desired. to estimate The bestpercent performances RMSE ( 𝑅𝑀𝑆𝐸/𝑅 are,) in retrievals. general, The found 𝑅 in for the ETM+ green and TM channels. data were The 0.0049, relatively 0.00519, 0.00146 [1/sr] and 0.0050, 0.00371, 0.0013 [1/sr], respectively. The percent RMSEs were large slopes in ETM+ analyses implicate potential nonlinearity in instrument responses. We further estimated as 36%, 28%, 62%, and 46%, 35%, 84% for ETM+ and TM’s blue, green, and red bands, computedrespectively. the median AlthoughRrs spectra, the quality i.e., of ETM+Rrs, are from adequately AERONET-OC evaluated here data using to estimateAERONET-OC percent data, RMSE (RMSE/thereRrs) inis still retrievals. a need to ThebetterR assessrs for TM ETM+ prod anducts over TM a data broader were range 0.0049, of water 0.00519, bodies. 0.00146 [1/sr] and 0.0050, 0.00371, 0.0013 [1/sr], respectively. The percent RMSEs were estimated as 36%, 28%, 62%, and 46%, 35%,Table 84% 4. forThe ETM+statistics andassociated TM’s with blue, matchup green, analys andesred of Landsat-7 bands, and respectively. -5 at AERONET-OC Although sites. the quality The values in round brackets refer to statistics derived prior to implementing the gains listed in Table of ETM+ are3. adequately The large RMSEs evaluated and deviations hereusing of slopes AERONET-OC from unity are associated data, there with random is still and a need systematic to better assess TM productserrors over in a TM broader and ETM+ range observations. of water The bodies. Landsat-5 matchups are available only for the last eight years of the mission. Table 4. The statistics associated with matchup analyses of Landsat-7 and -5 at AERONET-OC sites. Band Center [nm] MRD [%] RMSE [1/sr] Slope Intercept Bias [1/sr] 𝑹𝟐 The values in round brackets refer to statistics derived prior to implementing the gains listed in Table3. Landsat-7 (ETM+) The large RMSEs478 and deviations9.81 of(−33.8) slopes from0.0018 unity (0.0022) are associated 1.25 with−0.000016 random and0.0006 systematic 0.79 errors in TM and ETM+560 observations.3.4 The(−7.9) Landsat-5 0.0013 matchups (0.0015) are1.10 available −0.000024 only for the0.000099 last eight 0.81 years of the mission. 661 1.07 (−46.5) 0.00076 (0.0011) 1.18 −0.000014 0.0000111 0.64 Landsat-5 (TM) Band Center485 [nm] MRD13.6 [%] (−85.5) RMSE0.0020 [1/sr] (0.006) Slope1.17 Intercept0.000013 Bias0.00072 [1/sr] R0.562 569 11.9 (−14.7) Landsat-70.0016 (0.002) (ETM+) 1.04 0.00101 0.00082 0.85 478660 9.81 (10.1−33.8) (−116.2) 0.0018 0.00101 (0.0022) (0.003) 1.25 1.14 −0.0000160.0002 0.00060.00011 0.790.59 560 3.4 (−7.9) 0.0013 (0.0015) 1.10 −0.000024 0.000099 0.81 661 1.07 (−46.5) 0.00076 (0.0011) 1.18 −0.000014 0.0000111 0.64 Landsat-5 (TM) 485 13.6 (−85.5) 0.0020 (0.006) 1.17 0.000013 0.00072 0.56 569 11.9 (−14.7) 0.0016 (0.002) 1.04 0.00101 0.00082 0.85 660 10.1 (−116.2) 0.00101 (0.003) 1.14 0.0002 0.00011 0.59 Remote Sens. 2018, 10, 1337 12 of 23

3.2.2. MODIS-Derived Rrs Products Since the AERONET-OC data do not provide a full representation of aquatic conditions in inland and nearshore coastal waters, we further evaluated TM/ETM+ products against (same-day) MODISA and MODIST products over the central basin of Lake Mead (n = 100/101), San Francisco (SF) Bay (n = 63/75), San Pablo Bay (n = 66/69), and the eastern basin of Lake Erie (n = 19/29) in North America for which a rich archive of TM data is available (limitations in the data downlink during periods of TM operations preclude a robust global analysis). The number of matchups (in round brackets) refer to the number of MODIST-ETM+ and MODISA-TM matchups, respectively. These sites collectively enable a fair assessment of TM and ETM+ products over a wider range of aquatic conditions (e.g., areas with high CDOM and/or sediment loads) and are sufficiently spatially extensive for valid ocean color products from MODIS observations. Landsat-5, Landsat-7, and Terra were/are all morning satellites, while Aqua is in an afternoon orbit. Our initial assessments indicated noticeable uncertainties in intercomparisons due to the large time differences between ETM+ and MODISA overpasses. This was, in particular, observed over SF Bay influenced by significant tidal forcing. Therefore, for cross-validating ETM+ products, MODIST products were applied. In addition, because the two missions are in very similar orbits, near-nadir MODIST observations/products are available for cross-validating ETM+ products. On the other hand, Landsat-5 and Terra’s orbits are different such that only MODIST observations at the edge of the swaths are available for valid, same-day intercomparisons with those of TM. Such products are represented with large footprint sizes; therefore, larger uncertainties in products are expected. It was then decided to utilize MODISA products for investigating TM products, recognizing that there is two-to-three-hour time gap between data collections, which increase uncertainties in intercomparisons. The Level-1B MODIS data were processed to Level-2 products using a NIR-SWIR band combination [73]. The rest of the parameter-setting was identical to the current NASA Ocean Biology Processing Group’s standard processing. The spatial sampling for Aqua-MODIS Rrs products were conducted by taking the median values within 3×3-element windows to avoid contaminations from land. The equivalent sample area for Landsat products (i.e., median within 3 × 3 km regions) were adjusted according to MODIS viewing geometries. Further, the differences in the spectral bands were accounted for [61]. Figure9 shows the scatterplots with the statistical analyses tabulated in Table5. Overall, there is a reasonable correlation in MODIST-ETM+ intercomparisons. The scattered matchups correspond to artifacts in TM and ETM+ products, uncertainties in the atmospheric correction, the spectral band adjustments, and possible changes in water-column conditions (in particular for TM-MODISA intercomparisons over SF Bay and San Pablo Bay influenced by strong tides). The uncertainties due to the time difference in overpasses were also found over Lake Mead and Lake Erie when both MODIST-ETM+ and MODISA-ETM+ intercomparisons were examined. Further site-specific analyses suggested larger uncertainties in TM-derived blue and red bands in Lake Mead and Lake Erie where Rrs(485) and Rrs(660) were, on average, 0.0034 [1/sr] and 0.0022 [1/sr], respectively.

Table 5. The statistics associated with TM-MODIS (n = 274) and MODIS-ETM+ (n = 249) over select sites across North America. Uncertainties associated with the intercomparisons are attributed to atmospheric correction and spectral band adjustments.

Band Center [nm] MRD [%] RMSD [1/sr] Slope Intercept Bias [1/sr] R2 Landsat-7 (ETM+) 478 −9.56 0.00152 0.923 −0.000017 −0.00044 0.91 560 −3.29 0.00103 1.007 −0.00021 −0.0002 0.97 661 −8.15 0.00102 0.939 −0.000057 −0.000039 0.98 Landsat-5 (TM) 485 −7.11 0.002283 0.984 −0.00025 −0.00033 0.76 569 −1.83 0.001708 1.012 −0.000048 −0.00017 0.93 660 1.97 0.001905 0.981 0.000228 0.0000361 0.91 Remote Sens. 2018, 10, 1337 13 of 23 Remote Sens. 2018, 10, x FOR PEER REVIEW 13 of 23

Figure 9. The scatterplots for the TM-MODISA and MODIST-ETM+ intercomparisons. Considering Figure 9. The scatterplots for the TM-MODISA and MODIST-ETM+ intercomparisons. Considering the dynamic range in the signal and the uncertainties in the intercomparisons, TM and ETM+ products the dynamic range in the signal and the uncertainties in the intercomparisons, TM and ETM+ products are in good agreement with those of MODIS. The largest scatter is found in TM-derived Rrs(485) are in goodproducts. agreement The larger with uncertainties those of in MODIS. the TM-MODIS The largestA analyses scatter have is aroused found infrom TM-derived two-to-three-hour Rrs(485) products.time The difference larger uncertainties in their respective in the overpass TM-MODISA times. analyses have aroused from two-to-three-hour time difference in their respective overpass times. 3.3. Time-Series Applications

3.3. Time-SeriesIn order Applications to examine the long-term record of Landsat-derived 𝑅 products, we analyzed seasonal composites generated for 12 aquatic systems across the continental USA. The seasonal In order to examine the long-term record of Landsat-derived Rrs products, we analyzed seasonal composite plots were derived from locally processed Landsat data over 12 sites, three of which are composites generated for 12 aquatic systems across the continental USA. The seasonal composite presented in this manuscript. The median 𝑅 computed in each season represents seasonal plots werecomposite derived products, from locally which processedwere produced Landsat to minimi dataze over noise 12 in sites, the analyses. three of The which three are plots presented shown in this manuscript.in Figures 10–12 The median correspondRrs tocomputed the Indian in River each Lagoon season (IRL) represents (27.8355N, seasonal 80.4515W), composite Pamlicoproducts, Sound which were(35.1963N, produced 76.2670W), to minimize and San Joaquin noise in River the analyses.(38.09182N, The121.6297W). three plots The shownother sites in were Figures Boston 10– 12 correspondHarbor to (MA), the IndianSuisan RiverBay (CA), Lagoon Neuse (IRL) River (27.8355N,(NC), Upper 80.4515W), Chesapeake PamlicoBay (MD), Sound Potomac (35.1963N, River 76.2670W),(MD), and San San Francisco Joaquin Bay River (CA), (38.09182N, Onondaga Lake 121.6297W). (NY), Lake The Tahoe other (CA), sites and were Crater Boston Lake. Harbor The only (MA), Suisan Bayproducts (CA), evaluated Neuse River were (NC),𝑅 for Upper the three Chesapeake visible bands Bay (MD),(Table Potomac1). Figure River10 shows (MD), these San products Francisco Bay (CA),for Onondagathe IRL in the Lake vicinity (NY), of LakeSebastian Tahoe Inlet (CA), State and Park Crater (FL). Lake. The only products evaluated were Our approach developed for compensating for non-zero Lw(NIR) (Section 2.1) significantly Rrs for the three visible bands (Table1). Figure 10 shows these products for the IRL in the vicinity of increased the number of valid retrievals, i.e., more than 40% of the retrievals were negative in Rrs(485) Sebastian Inlet State Park (FL). products. Since data were collected eight days apart, potentially yielding biases in seasonal Ourcomposites, approach it developedis difficult to for infer compensating consistency in forthe non-zerodata products. Lw(NIR) In addition, (Section seawater-intrusion 2.1) signiﬁcantly increasedand thetidal number exchanges of may valid complicate retrievals, such i.e., analyses. more thanSeasonal 40% 𝑅 of composites the retrievals over werePamlico negative Sound, in Rrs(485)North products. Carolina, Since are data illustrated were collectedin Figure 11. eight Some days of the apart, retrievals potentially may be yielding associated biases with inthe seasonal noise composites,in the itNIR is and difﬁcult SWIR-I to bands infer that consistency propagate in throug the datah the products.atmospheric In correction addition, [64]. seawater-intrusion Note that prior and tidalto accounting exchanges for may Lw(NIR) complicate the entire such record analyses. in the blue Seasonal band wasRrs invalid.composites The time-series over Pamlico plot for Sound, the North Carolina,San Joaquin are River illustrated (California) in Figure is shown 11. Some in Figure of the retrievals12 For this may riverine be associated system dominated with the noisewith in the NIRscattering and SWIR-I particles, bands Lw(NIR) that propagateshould be accounted through for the to atmospheric allow for valid correction and/or accurate [64]. retrievals Note that [54]. prior Similar results were found for a downstream system, i.e., Suisan Bay. The impact of the proposed to accounting for Lw(NIR) the entire record in the blue band was invalid. The time-series plot for methodology was most noticeable in such CDOM-dominated waters, i.e., frequent negative retrievals the San Joaquin River (California) is shown in Figure 12 For this riverine system dominated with in the blue and green were observed prior to its implementation. scattering particles, Lw(NIR) should be accounted for to allow for valid and/or accurate retrievals [54]. Similar results were found for a downstream system, i.e., Suisan Bay. The impact of the proposed methodology was most noticeable in such CDOM-dominated waters, i.e., frequent negative retrievals in the blue and green were observed prior to its implementation. Remote Sens. 2018, 10, 1337 14 of 23 Remote Sens. 2018, 10, x FOR PEER REVIEW 14 of 23 Remote Sens. 2018, 10, x FOR PEER REVIEW 14 of 23

Figure 10. The seasonal composites of Rrs products derived from Landsat-5 (solid circles), Landsat-7 Figure 10. The seasonal composites of Rrs products derived from Landsat-5 (solid circles), Landsat-7 Figure(triangle), 10. andThe Landsat-8seasonal composites (hollow circles) of Rrs over products the Indian derived River from Lagoon Landsat-5 (27.8355N, (solid 80.4515W), circles), Landsat-7 Florida. (triangle), and Landsat-8 (hollow circles) over the Indian River Lagoon (27.8355N, 80.4515W), Florida. (triangle),The composites and Landsat-8 were obtained (hollow by circles) taking over the thmediane Indian of River retrievals Lagoon in each(27.8355N, season. 80.4515W), The images Florida. were The composites were obtained by taking the median of retrievals in each season. The images were Theprocessed composites via SeaDAS. were obtained The methodology by taking the for median accounting of retrievals for non-negligible in each season. Lw(NIR) The imagessignificantly were processed via SeaDAS. The methodology for accounting for non-negligible L (NIR) signiﬁcantly processedimproved thevia numberSeaDAS. of The vali dmethodology (non-negative) for retrievals. accounting for non-negligible Lw(NIR)w significantly improvedimproved the numberthe number of validof vali (non-negative)d (non-negative) retrievals.retrievals.

Figure 11. Seasonal composites of Rrs products derived from Landsat-5 (solid circles), Landsat-7 Figure(triangle), 11. andSeasonal Landsat-8 composites (hollow of circles) Rrs products in Pamlico derived Sound, from North Landsat-5 Carolina (solid (35.1963N, circles), 76.2670W).Landsat-7 Figure 11. Seasonal composites of Rrs products derived from Landsat-5 (solid circles), Landsat-7 (triangle),The composites and Landsat-8 were obtained (hollow by circles) taking in thePamlico medi Sound,an of retrievals North Carolina in each (35.1963N, season. The 76.2670W). images (triangle), and Landsat-8 (hollow circles) in Pamlico Sound, North Carolina (35.1963N, 76.2670W). Theprocessed composites via SeaDAS. were Theobtained product by consistencytaking the canmedi bean inferred of retrievals in the overlapping in each season. periods The indicating images The composites were obtained by taking the median of retrievals in each season. The images processed processedthe utility viaof SeaDAS.long-term The Landsat product data consistency record for can aquatic be inferred science in thestudies. overlapping Note that periods because indicating of the via SeaDAS. The product consistency can be inferred in the overlapping periods indicating the utility thedifferences utility of in long-termthe temporal Landsat sampling data (i.e., record eight for da aquaticys) and science considering studies. various Note tidal that stages,because a robustof the of long-termdifferencesanalysis of Landsat productin the temporal data consistency record sampling for is not aquatic (i.e.,possible. eight science days) studies. and considering Note that various because tidal of the stages, differences a robust in the temporalanalysis sampling of product (i.e., consistency eight days) is not and possible. considering various tidal stages, a robust analysis of product consistency is not possible.

Remote Sens. 2018, 10, 1337 15 of 23 Remote Sens. 2018, 10, x FOR PEER REVIEW 15 of 23

Figure 12. Seasonal Rrs composites derived from Landsat-5 (solid circles), Landsat-7 (triangle), and Figure 12. Seasonal Rrs composites derived from Landsat-5 (solid circles), Landsat-7 (triangle), Landsat-8 (hollow circles) in San Joaquin River (38.09182N, 121.6297W), California. The composites and Landsat-8 (hollow circles) in San Joaquin River (38.09182N, 121.6297W), California. The composites were obtained by taking the median of retrievals in each season. The images were processed via were obtained by taking the median of retrievals in each season. The images were processed via SeaDAS. SeaDAS. The product consistency can be inferred in the cross-mission overlapping periods. Few The productnegative retrievals consistency are attributed can be inferred to imperfect in the corre cross-missionctions for the contributing overlapping water-leaving periods.Few radiance negative retrievalsin the areNIRattributed channel. to imperfect corrections for the contributing water-leaving radiance in the NIR channel. 4. Discussions: Product Quality 4. Discussions:The analyses Product thus Quality far have highlighted some of the potentials and limitations of Landsat archive forThe long-term analyses aquatic thus far science have highlightedstudies. Now, some with of a therigorous, potentials physics-based and limitations processing of Landsat system archivein for long-termplace, the science aquatic community science studies. can further Now, explore with th ae rigorous, utility of the physics-based Landsat data processingrecord in science system in studies, such as examining impacts of human-induced activities (e.g., landuse/landcover change) on place, the science community can further explore the utility of the Landsat data record in science water quality. We demonstrated that following the cross-calibration and vicarious calibration of TM studies, such as examining impacts of human-induced activities (e.g., landuse/landcover change) on and ETM+ data, the biases in Rrs products are significantly reduced. In this section, we will further water quality. We demonstrated that following the cross-calibration and vicarious calibration of TM elaborate on the limitations of TM and ETM+ Rrs products with regard to noise contributions and and ETM+processing data, flaws. the biasesA brief indiscussionRrs products on adjacency are signiﬁcantly effects is also reduced. provided. In this section, we will further elaborate on the limitations of TM and ETM+ Rrs products with regard to noise contributions and processing4.1. Signal-to-Noise ﬂaws. A brief Ratio discussion (SNR) on adjacency effects is also provided.

Although we showed that TM- and ETM+-derived Rrs products agree well, on average, with 4.1. Signal-to-Noisethose of OLI [59], Ratio OLI’s (SNR) SNR is three-to-four times higher than those of TM and ETM+ [5,12]. To better

understandAlthough we the showed implications that TM-of these and differences ETM+-derived in Rrs products,Rrs products a brief agree sensitivity well, onanalysis average, is carried with those of OLIout. [59 Here,], OLI’s a small SNR subset is three-to-four (10 × 10) of a times cloud-free higher OLI than image those (ID:LC80220402014227LGN00, of TM and ETM+ [5,12]. solar To better zenith angle of 26.3°) with low aerosol load, i.e., AOT(865) = 0.08, and over moderately turbid waters understand the implications of these differences in Rrs products, a brief sensitivity analysis is carried (Table 6) was processed by introducing various noise levels to simulate TM, ETM+, and OLI × out. Here, a small subset (10 10) of a cloud-free𝑛 OLI image (ID:LC80220402014227LGN00,𝐿 ⁄𝑆𝑁𝑅 𝐿 solar𝑅 zenith performances◦ [5,12]. For each sensor, noise ( ) is estimated via . The median and angle(derived of 26.3 from) with OLI low subset) aerosol are load, tabulated i.e., AOT(865)in Table 6. =The 0.08, SNRs and are over adopted moderately from References turbid waters [5,74].(Table 6) was processedNote that by introducing OLI’s SNR from various [74] noisewere scaled levels to simulatematch theTM, SNRs ETM+, derived and from OLI typical performances radiances [ 5,12]. For eachreported sensor, in Reference noise (ns [5]) is for estimated TM, ETM+, via andLt/ MODIS.SNR. The The median analysesL weret and conductedRrs (derived via Monte from OLI Carlo subset) are tabulatedsimulations in for Table all 6the. The bands SNRs (Table are 1) adopted except the from SWIR-II References band, [which5,74]. was not used in the aerosol removal.Note that To OLI’s provide SNR a fromreference, [74] werethe OLI scaled subset to match was also the SNRsprocessed derived with fromthe MODIS typical noise radiances reportedperformance in Reference [5] representing [5] for TM, a high-fidelity ETM+, and ocean MODIS. color sensor. The analyses To generate were various conducted scenarios, via 𝐿 Monte Carlowere simulations allowed to for vary all thewithin bands the (Table𝐿 𝑛1 )range except with the 𝑛 SWIR-II having band,a normal which distribution. was not usedAll the in the permutations resulted in ~7000 SeaDAS runs, which were repeated four times for TM, ETM+, OLI, aerosol removal. To provide a reference, the OLI subset was also processed with the MODIS noise and MODIS. performance [5] representing a high-ﬁdelity ocean color sensor. To generate various scenarios, Lt were allowed to vary within the Lt ± ns range with ns having a normal distribution. All the permutations resulted in ~7000 SeaDAS runs, which were repeated four times for TM, ETM+, OLI, and MODIS.

The uncertainties (σRrs ) provided in Table4 represent one standard deviation derived from the resultant Rrs distributions. The derived uncertainties provide guidance on expected random Remote Sens. 2018, 10, 1337 16 of 23 variability in heritage Landsat products compared to those derived from OLI and MODIS. In general, the uncertainties in TM are two-to-five times higher than those of OLI. Given the magnitude of the Rrs spectrum, uncertainties in TM-, ETM+-, OLI-, and MODIS-derived Rrs products are, on average, ~25%, ~27%, ~8%, and ~2%, in the blue and green bands, respectively. This implies that subtle changes in Rrs, i.e., < σRrs , leading to small changes in water quality parameters, may not be detected. In inland and nearshore coastal waters (e.g., bays, estuaries) where concentrations of TSS and Chl are often much higher than those in coastal ocean waters, highly precise concentrations (at levels expected from ocean color missions) may not be required. Therefore, end-users may utilize heritage Rrs products with the “fit-for-purpose” concept in mind, i.e., products could be valuable for one application but may not be viable for another. In addition, the choice of algorithm may alter how the noise is propagated to end products. For example, band-ratio algorithms [75,76] amplify the uncertainties. It is, thus, critical that users apply smoothing techniques, i.e., spatial filtering, (available to users by default via SeaDAS) to reduce the uncertainties. Assuming a 7 × 7-element (uniform) averaging window, σRrs in Table6 are reduced by a factor of 1/7. This analysis was carried out for TOA radiance. In practice, ns increases with increase in Lt leading to variable uncertainties across a scene. Note, however, that the median Lt considered in this analysis is higher than typical radiances encountered in Landsat scenes [74]. Hence, the computed values in Table6 provide an upper bound for (random component of) uncertainties. It should also be noted that although ETM+’s SNRs in the visible bands are higher than those of TM, uncertainties in TM-derived Rrs are slightly smaller. This is attributed to TM’s higher SNR in the NIR band employed in the aerosol removal [64].

4.2. Image Artifacts and Processing Flaws Heritage Landsat data, in particular TM images, often contain artifacts (e.g., striping, coherent noise, memory effects) suggesting that Rrs products should always be interpreted with care. Some of these artifacts are evident in Figure6. Further, we found anomalies in TM products within the 1990–94 interval. This is highlighted in Figure 13 where time-series of Rrs products over fairly optically stable inland waters [59], i.e., Crater Lake and Lake Tahoe, are illustrated. Further analyses of Lt and other intermediate products (e.g., Lr) indicated unusually high Lt in the red, NIR, and SWIR-I bands. These high radiometric responses yield invalid Rrs products [64]. No such anomalies were identiﬁed for ETM+ time-series, i.e., fewer artifacts present in ETM+ data. Nonetheless, occasional noisy retrievals were observed in time-series plots. Another observation was nonlinearity in response when adjusting vicarious gains in Section 2.3. This was, in particular, conﬁrmed in the ETM+ matchup analysis (Table5). Overall, the products may be used with caution in oligotrophic (e.g., Crater Lake), mesotrophic (e.g., Lake Tahoe), and CDOM-rich waters.

Table 6. The SNR, TOA radiance (Lt ), and the uncertainties in Rrs (σRrs ) associated with TM, ETM+,

OLI, and MODIS. σRrs represents one standard deviation corresponding to the Rrs distributions derived from many SeaDAS runs for a small OLI subset. Note that the aerosol removal was conducted using the 840–1676 band combination and OLI’s SNR from [74] were scaled to provide comparable SNRs estimated for typical radiances reported in Reference [5] for TM, ETM+, and MODIS. The units for Lt −2 −1 −1 −1 and σRrs are wm µ sr and sr , respectively. Band centers refer to TM channels.

Band (nm) 485 569 660 840 1676

SNR σRrs SNR σRrs SNR σRrs SNR TM 71.8 9.1 × 10−4 42.3 7.2 × 10−4 29.0 4.8 × 10−4 16.6 10.3 ETM+ 77.9 1.0 × 10−3 69.7 8.5 × 10−4 40.7 5.2 × 10−4 13.2 10.3 OLI 382.8 3.0 × 10−4 256.9 2.4 × 10−4 135.1 1.8 × 10−4 58.9 17.6 MODIS 2209.2 1.0 × 10−4 2401.8 6.9 × 10−5 1422.0 4.7 × 10−5 994.5 31.5 Lt 54.2 29.9 15.2 4.3 0.30 Rrs 0.0031 0.0028 0.0028 NA NA Remote Sens. 2018, 10, 1337 17 of 23

Remote Sens. 2018, 10, x FOR PEER REVIEW 17 of 23 To infer uncertainties in products, we quadratically added up systematic and random errors. It is, in general,To infer difﬁcult uncertainties to fully in characterize products, we systematic quadratically errors added for heritage up systematic sensors/products. and random However,errors. It is,according in general, to difficult the matchup to fully analyses characterize (Table systematic4), intercomparisons errors for heritage with MODIS sensors/products. (Table5), and However, SNR accordingassessments to (Tablethe matchup6), it is possibleanalyses to (Table speculate 4), in atercomparisons maximum uncertainty with MODIS estimate (Table of 0.0019, 5), and 0.0018, SNR assessmentsand 0.0012 [1/sr](Table in6), theit is blue, possible green, to speculate and red channelsa maximum of TMuncertainty and ETM+ estimate products, of 0.0019, respectively. 0.0018, andThese 0.0012 estimates [1/sr] in are the calculated blue, green, by takingand red the channe averagels of ofTM RMSEs and ETM+ in Tables products,4 and 5respectively.. Also, note These that estimatesthese maximum are calculated uncertainties by taking are the estimated average recognizing of RMSEs (a)in Tables implementation 4 and 5. Also, of smoothing note that ﬁltersthese maximumfor reducing uncertainties random noise are contributionestimated recognizing and (b) uncertainties (a) implementation in the matchup/intercomparison of smoothing filters for reducinganalyses. random The changes noiseand/or contribution anomalies and (b) larger uncert thanainties the above in the threshold matchup/intercomparison are expected to be analyses. detected Thein the changes heritage and/or products. anomalies According larger than to thethe matchupabove threshold analyses are (Section expected 3.2 to), be the detected quality in of the the heritage products. According to the matchup analyses (Section 3.2), the quality of the TM-derived TM-derived Rrs(660) were found to be reliable for relatively highly backscattering waters where Rrs(660) were found to be reliable for relatively highly backscattering waters where Rrs(660) > 0.004 Rrs(660) > 0.004 [1/sr]. Our processing system may also contribute to uncertainties. For instance, [1/sr].it is likely Our processing that the retrievals system overmay also Crater contribute and Tahoe to uncertainties. lakes are affected For instance, by imperfect it is likely corrections that the of retrievalswater-vapor over transmission. Crater and Bear Tahoe in mind lakes that are TM affected and ETM+’s by imperfect NIR channels corrections (Figure 1of), i.e.,water-vapor ~140 nm transmission.wide, contain Bear a water in mind absorption that TM featureand ETM+’s centered NIR aroundchannels ~820 (Figure nm, 1), whereas i.e., ~140 OLI’s nm wide, NIR contain band is arelatively water absorption narrow (Figure feature1). centered Furthermore, around some ~820 of thenm, existing whereas noise OLI’s in NIR the analyses band is mayrelatively be attributed narrow (Figureto failure 1). in Furthermore, cloud-screening/-masking some of the existing conducted noise using in the thresholds analyses may adopted be attributed from ocean to colorfailure data in cloud-screening/-maskingprocessing. There are also conducted inaccurate using retrievals threshol in periodsds adopted where from CDOM ocean loads color are data high processing. leading to There are also inaccurate retrievals in periods where CDOM loads are high leading to the the underestimation of Rrs. More research must be dedicated to these topics to enhance these historic underestimation of Rrs. More research must be dedicated to these topics to enhance these historic Rrs Rrs products. products. 4.3. Adjacency Effects 4.3. Adjacency Effects The adjacency effect (AE) is another prime topic when using Landsat data in inland waters. SuchThe signal adjacency complicating effect (AE)Rrs retrievals is another were prime pronounced topic when over using Onondaga Landsat data Lake in (NY). inland To waters. qualitatively Such signalevaluate complicating presence/absence 𝑅 retrievals of adjacency were pronounced effects, the Rayleigh-correctedover Onondaga Lake (Lrc (NY).) data To over qualitatively three sites, evaluateincluding presence/absence Onondaga Lake of (NY), adjace Chowanncy effects, River the (NC), Rayleigh-corrected and Boston Harbor (L) data (MA) over were three analyzed. sites, includingThe exact Onondaga locations forLake these (NY), sites Chowan are (43.0919N, River (NC), 76.2145W), and Boston (36.1763N, Harbor (MA) 76.7378W), were analyzed. and (42.3401, The exact70.9824W), locations respectively. for these To sites do so, are below (43.0919N, index (AE_Ind) 76.2145W), was computed(36.1763N, 76.7378W), and (42.3401, 70.9824W), respectively. To do so, below index (AE_Ind) was computed Lrc(NIR) − Lrc(Red) AE_Ind = ()() (6) AE_IndL =rc(NIR) + Lrc(Red ) (6) ()()

Figure 13. The time-series of Rrs products in the blue bands over optically stable waters of Crater Figure 13. The time-series of Rrs products in the blue bands over optically stable waters of Crater and and Tahoe lakes. The overall signal level is very similar in time. A potential calibration issue in TM Tahoe lakes. The overall signal level is very similar in time. A potential calibration issue in TM products is highlighted. products is highlighted.

Although AE_Ind (ranging from −1 to 1) includes aerosol contributions, it provides insights into whether the L spectrum deviates from expected trends, i.e., a monotonically decreasing one

Remote Sens. 2018, 10, 1337 18 of 23

Remote Sens. 2018, 10, x FOR PEER REVIEW 18 of 23 Although AE_Ind (ranging from −1 to 1) includes aerosol contributions, it provides insights into towards longer wavelengths. Therefore, the higher the AE_Ind, the stronger the NIR signal is, whether the Lrc spectrum deviates from expected trends, i.e., a monotonically decreasing one towards potentiallylonger wavelengths. alluding to Therefore,the presence the of higher adjacency the AE_Ind effects., The the strongertime-series the plots NIR of signal AE_Ind is, potentiallyextracted fromalluding Boston to theHarbor, presence Chowan of adjacency River, effects.and Onondaga The time-series Lake are plots illustrated of AE_Ind in extractedFigure 14. from From Boston the analyses of the Rrs data products (not shown here) over the Boston Harbor and the corresponding Harbor, Chowan River, and Onondaga Lake are illustrated in Figure 14. From the analyses of the Rrs time-seriesdata products shown (not in shown Figure here) 14, it over appears the Boston that there Harbor is negligible and the corresponding adjacency effect. time-series Considering shown the in plotsFigure associated 14, it appears with Boston that there Harbor is negligible as the reference, adjacency it is effect. found Considering that there is theAE plotsin some associated instances with of ChowanBoston Harbor River retrievals. as the reference, The impact it is found of AE that is theremuch is more AE in pronounced some instances over of Onondaga Chowan River Lake retrievals. where AE_IndThe impact is frequently of AE is very much close more to pronounced zero and even over positive Onondaga for some Lake instances. where AE_Ind Sinceis the frequently topography very aroundclose to Onondaga zero and even Lake positive and Chowan for some River instances. is fairly Sinceflat and the the topography two systems around are comparable Onondaga Lake in their and morphology,Chowan River i.e., is fairly< 2 km ﬂat wide, and the it twocan systemsbe conclu areded comparable that environmental in their morphology, conditions i.e., (e.g., <2 km wind, wide, humidity,it can be concluded turbulence that at environmentalland-atmosphere conditions interface) (e.g., play wind, a major humidity, role in turbulence the magnitude at land-atmosphere of adjacency signal.interface) play a major role in the magnitude of adjacency signal. WhileWhile with with only only six six spectral spectral measurements measurements of of TM TM and and ETM+, ETM+, it it is is nearly nearly impossible impossible to to remove remove adjacencyadjacency effects, effects, it it is is certainly certainly possible possible to to minimize minimize such such effects effects with with prior prior knowledge knowledge of of the the spectral spectral characteristicscharacteristics of of surrounding surrounding land land targets, targets, do dominantminant environmental environmental conditions conditions and and topography. topography. Therefore,Therefore, it it is is cautioned cautioned that that the the adjacency adjacency sign signalsals may may not not be be assumed assumed for for all all inland inland water water sites sites (simply(simply because because they they are are small small or or narrow narrow water water bodies). bodies). For For instance, instance, land land signals signals were were not not identified identiﬁed overover San San Joaquin Joaquin River River (CA), (CA), which which is is narrower narrower th thanan Onondaga Onondaga Lake Lake or or Chowan Chowan River. River. Extensive Extensive researchresearch is is required required to to identify identify water water bodies bodies su susceptiblesceptible to to these these effects, effects, understand understand the the dominant dominant environmentalenvironmental conditions, conditions, and and minimize minimize them them for th fore past the pastand existing and existing multispectral multispectral missions. missions. With futureWith futurehyperspectral hyperspectral missions missions planned/proposed planned/proposed for the fornext the decade, next decade, it may itbe may possible be possible to more to accuratelymore accurately account account for these for unwanted these unwanted signals. signals.

FigureFigure 14. 14. TheThe AE_Ind AE_Ind computed computed for for three three sites sites to to analyze analyze presence/absence presence/absence of of adjacency adjacency effects effects (AE). (AE). TheThe higher higher the the index, index, the the larger larger the the magnitude magnitude of of AE AE is. is. The The Onondaga Onondaga Lake Lake has has the the highest highest and and most most frequentfrequent presence presence of of AE AE whereas whereas the the effect effect is is deemed deemed to to be be negligible negligible or or absent absent over over Boston Boston Harbor. Harbor. NoteNote that that the the NIR NIR bands bands of of TM TM and ETM+ are much wider thanthan thatthat ofof OLI.OLI. ThisThis has has likely likely resulted resulted in insmaller smaller OLI-derived OLI-derived AE_Ind. AE_Ind.

5. Conclusions This work described the potentials and limitations of heritage Landsat missions for use in aquatic studies. Here, we demonstrated that the Landsat-5 and Landsat-7 data records, when

Remote Sens. 2018, 10, 1337 19 of 23

5. Conclusions This work described the potentials and limitations of heritage Landsat missions for use in aquatic studies. Here, we demonstrated that the Landsat-5 and Landsat-7 data records, when corrected for atmospheric effects and vicariously calibrated via the algorithms and software available in the SeaWiFS Data Analysis System (SeaDAS), yield remote-sensing reﬂectance (Rrs) products that, on average, agree with those derived from in situ measurements and/or other ocean color products. We further proposed a slightly different strategy to account for Lw(NIR) for TM and ETM+, which were/are missing the 443 nm channel. With careful regional or global algorithm developments, the processing system has enabled the creation of a long-term record of aquatic science products since 1984. Nevertheless, larger uncertainties in TM and ETM+ due to random and systematic noise, along with the availability of only three visible spectral bands, limit the choice of algorithms. Hence, algorithm developments for estimating turbidity and concentrations of TSS may be conducted with care. Following our analyses of several water bodies in North America, the TM and ETM+ data products are recommended to be developed and used in moderately to highly turbid waters (i.e., Rrs(660) > 0.004 [1/sr]), where instrument artifacts are less noticeable. Overall, although the uncertainties in TM and ETM+ products are higher than those of OLI, they are still valuable assets for retrospective studies. This may include assessing changes in water quality of lakes, reservoirs, water supplies, and coastal receiving waters due to short-/long-term landuse/landcover changes as well as examining climate variability (i.e., weather patterns) and its impacts on water quality (e.g., location, frequency, and persistence in historic harmful algal blooms). There is also a need for the community to measure and quantify optical properties of water constituents in these poorly known aquatic systems. A more extensive database of in-water optical properties will enable developing regional or global long-term aquatic science products. Future research should focus on (a) developing algorithms that are less prone to random noise, (b) enhancing atmospheric correction to account for non-negligible near-infrared reﬂectance in extremely turbid or CDOM-rich waters, and (c) characterizing adjacency effects under various environmental/climatic conditions.

Author Contributions: The analyses were all performed by S.V.B., N.P. led and wrote the manuscript, B.A.F. provided the necessary look-up-tables, and S.S. integrated the processing in SeaDAS. Funding: Nima Pahlevan has been funded under NASA ROSES #NNX16AI16G, USGS Landsat Science Team Award #140G0118C0011, and the Massachusetts Institute of Technology SeaGrant Award # 2015-R/RC-140. Acknowledgments: The authors are sincerely grateful to Wesley Moses for his comments and commitment to enhancing the quality this manuscript. The reviewers’ efforts that resulted in enhanced presentations of the discussions on product uncertainties are appreciated as well. We also would like to thank Landsat-8 project scientist Jim Irons, with NASA Goddard Space Flight Center, for his encouragements and support of this study. The authors also thank the MOBY Operations Team and the BOUSSOLE project for maintaining and disseminating in situ radiometric data. We also thank three anonymous reviewers for their very constructive comments. Conﬂicts of Interest: The authors declare no conﬂicts of interest.

References

1. Williams, D.L.; Goward, S.; Arvidson, T. Landsat. Photogramm. Eng. Remote Sens. 2006, 72, 1171–1178. [CrossRef] 2. Markham, B.L.; Storey, J.C.; Williams, D.L.; Irons, J.R. Landsat sensor performance: History and current status. IEEE Trans. Geosci. Remote Sens. 2004, 42, 2691–2694. [CrossRef] 3. Schott, J.R. Remote Sensing the Image Chain Approach, 2nd ed.; Oxford University Press: New York, NY, USA, 2007; p. 666. 4. Gerace, A.D.; Schott, J.R.; Nevins, R. Increased potential to monitor water quality in the near-shore environment with Landsat’s next-generation satellite. J. Appl. Remote Sens. 2013, 7, 073558. [CrossRef] 5. Hu, C.; Feng, L.; Lee, Z.; Davis, C.O.; Mannino, A.; McClain, C.R.; Franz, B.A. Dynamic range and sensitivity requirements of satellite ocean color sensors: Learning from the past. Appl. Opt. 2012, 51, 6045–6062. [CrossRef][PubMed] Remote Sens. 2018, 10, 1337 20 of 23

6. Helder, D.L.; Markham, B.L.; Thome, K.J.; Barsi, J.A.; Chander, G.; Malla, R. Updated Radiometric Calibration for the Landsat-5 Thematic Mapper Reflective Bands. IEEE Trans. Geosci. Remote Sens. 2008, 46, 3309–3325. [CrossRef] 7. Markham, B.; Goward, S.; Arvidson, T.; Barsi, J.; Scaramuzza, P. Landsat-7 long-term acquisition plan radiometry-evolution over time. Photogramm. Eng. Remote Sens. 2006, 72, 1129–1135. [CrossRef] 8. Markham, B.L.; Boncyk, W.C.; Helder, D.L.; Barker, J.L. Landsat-7 Enhanced Thematic Mapper Plus Radiometric Calibration. Can. J. Remote Sens. 1997, 23, 318–332. [CrossRef] 9. Barnes, B.B.; Hu, C.; Holekamp, K.L.; Blonski, S.; Spiering, B.A.; Palandro, D.; Lapointe, B. Use of Landsat data to track historical water quality changes in Florida Keys marine environments. Remote Sens. Environ. 2014, 140, 485–496. [CrossRef] 10. IOCCG. Status and Plans for Satellite Ocean-Colour Missions: Considerations for Complementary Missions; Yoder, J.A., Ed.; IOCCG: Dartmouth, NS, Canada, 1999. 11. IOCCG. Remote Sensing of Ocean Colour in Coastal, and Other Optically-Complex, Waters; Sathyendranath, S., Ed.; Reports of the International Ocean-Colour Coordinating Group (IOCCG); IOCCG: Dartmouth, NS, Canada, 2000. 12. Pahlevan, N.; Schott, J. Leveraging EO-1 to Evaluate Capability of New Generation of Landsat Sensors for Coastal/Inland Water Studies. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2013, 6, 360–374. [CrossRef] 13. Helder, D.L.; Ruggles, T.A. Landsat thematic mapper reflective-band radiometric artifacts. IEEE Trans. Geosci. Remote Sens. 2004, 42, 2704–2716. [CrossRef] 14. Ouillon, S.; Douillet, P.; Andréfouët, S. Coupling satellite data with in situ measurements and numerical modeling to study fine suspended-sediment transport: A study for the lagoon of New Caledonia. Coral Reef. 2004, 23, 109–122. 15. Doxaran, D.; Froidefond, J.M.; Castaing, P. Remote-sensing reflectance of turbid sediment-dominated waters. Reduction of sediment type variations and changing illumination conditions effects by use of reflectance ratios. Appl. Opt. 2003, 42, 2623–2634. [PubMed] 16. Doxaran, D.; Froidefond, J.M.; Castaing, P.; Babin, M. Dynamics of the turbidity maximum zone in a macrotidal estuary (the Gironde, France): Observations from field and MODIS satellite data. Estuar. Coast. Shelf Sci. 2009, 81, 321–332. [CrossRef] 17. 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. Sci. Total Environ. 2001, 268, 197–214. [CrossRef] 18. Baban, S.M. Detecting water quality parameters in the Norfolk Broads, UK, using Landsat imagery. Int. J. Remote Sens. 1993, 14, 1247–1267. [CrossRef] 19. Ritchie, J.C.; Cooper, C.M.; Schiebe, F.R. The relationship of MSS and TM digital data with suspended sediments, chlorophyll, and temperature in Moon Lake, Mississippi. Remote Sens. Environ. 1990, 33, 137–148. [CrossRef] 20. Mertes, L.A.; Smith, M.O.; Adams, J.B. Estimating suspended sediment concentrations in surface waters of the Amazon River wetlands from Landsat images. Remote Sens. Environ. 1993, 43, 281–301. [CrossRef] 21. Bourgoin, L.M.; Bonnet, M.P.; Martinez, J.M.; Kosuth, P.; Cochonneau, G.; Moreira-Turcq, P.; Guyot, J.L.; Vauchel, P.; Filizola, N.; Seyler, P. Temporal dynamics of water and sediment exchanges between the Curuaí floodplain and the Amazon River, Brazil. J. Hydrol. 2007, 335, 140–156. [CrossRef] 22. Stumpf, R.P.; Goldschmidt, P.M. Remote sensing of suspended sediment discharge into the western Gulf of Maine during the April 1987 100-year flood. J. Coast. Res. 1992, 8, 218–225. 23. Dekker, A.G.; Vos, R.; Peters, S. Analytical algorithms for lake water TSM estimation for retrospective analyses of TM and SPOT sensor data. Int. J. Remote Sens. 2002, 23, 15–35. [CrossRef] 24. Giardino, C.; Pepe, M.; Brivio, P.A.; Ghezzi, P.; Zilioli, E. Detecting chlorophyll, Secchi disk depth and surface temperature in a sub-alpine lake using Landsat imagery. Sci. Total Environ. 2001, 268, 19–29. [CrossRef] 25. Allan, M.G.; Hamilton, D.P.; Hicks, B.J.; Brabyn, L. Landsat remote sensing of chlorophyll a concentrations in central North Island lakes of New Zealand. Int. J. Remote Sens. 2011, 32, 2037–2055. [CrossRef] 26. Brezonik, P.; Menken, K.D.; Bauer, M. Landsat-based remote sensing of lake water quality characteristics, including chlorophyll and colored dissolved organic matter (CDOM). Lake Reserv. Manag. 2005, 21, 373–382. [CrossRef] 27. Pattiaratchi, C.; Lavery, P.; Wyllie, A.; Hick, P. Estimates of water quality in coastal waters using multi-date Landsat Thematic Mapper data. Int. J. Remote Sens. 1994, 15, 1571–1584. [CrossRef] Remote Sens. 2018, 10, 1337 21 of 23

28. Kutser, T.; Pierson, D.C.; Kallio, K.Y.; Reinart, A.; Sobek, S. Mapping lake CDOM by satellite remote sensing. Remote Sens. Environ. 2005, 94, 535–540. [CrossRef] 29. Nieke, B.; Reuter, R.; Heuermann, R.; Wang, H.; Babin, M.; Therriault, J. Light absorption and fluorescence properties of chromophoric dissolved organic matter (CDOM), in the St. Lawrence Estuary (Case 2 waters). Cont. Shelf Res. 1997, 17, 235–252. [CrossRef] 30. Griffin, C.G.; Frey, K.E.; Rogan, J.; Holmes, R.M. Spatial and interannual variability of dissolved organic matter in the Kolyma River, East Siberia, observed using satellite imagery. J. Geophys. Res. Biogeosci. 2011, 116, G3. [CrossRef] 31. Phinn, S.; Dekker, A.; Brando, V.E.; Roelfsema, C. Mapping water quality and substrate cover in optically complex coastal and reef waters: An integrated approach. Mar. Pollut. Bullet. 2005, 51, 459–469. [CrossRef] [PubMed] 32. Armstrong, R.A. Remote sensing of submerged vegetation canopies for biomass estimation. Int. J. Remote Sens. 1993, 14, 621–627. [CrossRef] 33. Lyzenga, D.R. Remote sensing of bottom reflectance and water attenuation parameters in shallow water using aircraft and Landsat data. Int. J. Remote Sens. 1981, 2, 71–82. [CrossRef] 34. Andréfouët, S.; Muller-Karger, F.E.; Hochberg, E.J.; Hu, C.; Carder, K.L. Change detection in shallow coral reef environments using Landsat 7 ETM+ data. Remote Sens. Environ. 2001, 78, 150–162. [CrossRef] 35. Capolsini, P.; Andréfouët, S.; Rion, C.; Payri, C. A comparison of Landsat ETM+, SPOT HRV, Ikonos, ASTER, and airborne MASTER data for coral reef habitat mapping in South Pacific islands. Can. J. Remote Sens. 2003, 29, 187–200. [CrossRef] 36. Olmanson, L.G.; Bauer, M.E.; Brezonik, P.L. A 20-year Landsat water clarity census of Minnesota’s 10,000 lakes. Remote Sens. Environ. 2008, 112, 4086–4097. [CrossRef] 37. Lavery, P.; Pattiaratchi, C.; Wyllie, A.; Hick, P. Water quality monitoring in estuarine waters using the Landsat Thematic Mapper. Remote Sens. Environ. 1993, 46, 268–280. [CrossRef] 38. Tyler, A.; Svab, E.; Preston, T.; Présing, M.; Kovács, W. Remote sensing of the water quality of shallow lakes: A mixture modelling approach to quantifying phytoplankton in water characterized by high-suspended sediment. Int. J. Remote Sens. 2006, 27, 1521–1537. [CrossRef] 39. Kutser, T. The possibility of using the Landsat image archive for monitoring long time trends in coloured dissolved organic matter concentration in lake waters. Remote Sens. Environ. 2012, 123, 334–338. [CrossRef] 40. Lobo, F.L.; Costa, M.P.; Novo, E.M. Time-series analysis of Landsat-MSS/TM/OLI images over Amazonian waters impacted by gold mining activities. Remote Sens. Environ. 2015, 157, 170–184. [CrossRef] 41. Lyons, M.B.; Roelfsema, C.M.; Phinn, S.R. Towards understanding temporal and spatial dynamics of seagrass landscapes using time-series remote sensing. Estuar. Coast. Shelf Sci. 2013, 120, 42–53. [CrossRef] 42. Wang, Y.; Xia, H.; Fu, J.; Sheng, G. Water quality change in reservoirs of Shenzhen, China: Detection using LANDSAT/TM data. Sci. Total Environ. 2004, 328, 195–206. [CrossRef][PubMed] 43. Erkkilä, A.; Kalliola, R. Patterns and dynamics of coastal waters in multi-temporal satellite images: Support to water quality monitoring in the Archipelago Sea, Finland. Estuar. Coast. Shelf Sci. 2004, 60, 165–177. [CrossRef] 44. Kallio, K.; Attila, J.; Härmä, P.; Koponen, S.; Pulliainen, J.; Hyytiäinen, U.-M.; Pyhälahti, T. Landsat ETM+ images in the estimation of seasonal lake water quality in boreal river basins. Environ. Manag. 2008, 42, 511–522. [CrossRef][PubMed] 45. Hui, F.; Xu, B.; Huang, H.; Yu, Q.; Gong, P. Modelling spatial-temporal change of Poyang Lake using multitemporal Landsat imagery. Int. J. Remote Sens. 2008, 29, 5767–5784. [CrossRef] 46. Wulder, M.A.; Masek, J.G.; Cohen, W.B.; Loveland, T.R.; Woodcock, C.E. Opening the archive: How free data has enabled the science and monitoring promise of Landsat. Remote Sens. Environ. 2012, 122, 2–10. [CrossRef] 47. Hu, C.; Muller-Karger, F.E.; Andrefouet, S.; Carder, K.L. Atmospheric correction and cross-calibration of LANDSAT-7/ETM+ imagery over aquatic environments: A multiplatform approach using SeaWiFS/MODIS. Remote Sens. Environ. 2001, 78, 99–107. [CrossRef] 48. Antoine, D.; d’Ortenzio, F.; Hooker, S.B.; Bécu, G.; Gentili, B.; Tailliez, D.; Scott, A.J. Assessment of uncertainty in the ocean reflectance determined by three satellite ocean color sensors (MERIS, SeaWiFS and MODIS-A) at an offshore site in the Mediterranean Sea (BOUSSOLE project). J. Geophys. Res. Ocean. 2008, 113.[CrossRef] Remote Sens. 2018, 10, 1337 22 of 23

49. Hu, C.; Feng, L.; Lee, Z. Uncertainties of SeaWiFS and MODIS remote sensing reflectance: Implications from clear water measurements. Remote Sens. Environ. 2013, 133, 168–182. [CrossRef] 50. IOCCG. Mission Requirements for Future Ocean-Colour Sensors; Reports of the International Ocean-Colour Coordinating Group; McClain, C.R., Meister, G., Eds.; IOCCG: Dartmuouth, NS, Canada, 2012; p. 102. 51. Franz, B.A.; Bailey, S.W.; Kuring, N.; Werdell, P.J. Ocean color measurements with the Operational Land Imager on Landsat-8: Implementation and evaluation in SeaDAS. J. Appl. Remote Sens. 2015, 9, 096070. [CrossRef] 52. Pahlevan, N.; Sarkar, S.; Franz, B.A.; Balasubramanian, S.V.; He, J. Sentinel-2 MultiSpectral Instrument (MSI) data processing for aquatic science applications: Demonstrations and validations. Remote Sens. Environ. 2017, 201, 47–56. [CrossRef] 53. Siegel, D.A.; Wang, M.; Maritorena, S.; Robinson, W. Atmospheric correction of satellite ocean color imagery: The black pixel assumption. Appl. Opt. 2000, 39, 3582–3591. [CrossRef][PubMed] 54. Bailey, S.W.; Franz, B.A.; Werdell, P.J. Estimation of near-infrared water-leaving reflectance for satellite ocean color data processing. Opt. Express 2010, 18, 7521–7527. [CrossRef][PubMed] 55. Gordon, H.R. In-orbit calibration strategy for ocean color sensors. Remote Sens. Environ. 1998, 63, 265–278. [CrossRef] 56. Franz, B.A.; Bailey, S.W.; Werdell, P.J.; McClain, C.R. Sensor-independent approach to the vicarious calibration of satellite ocean color radiometry. Appl. Opt. 2007, 46, 5068–5082. [CrossRef][PubMed] 57. Gordon, H.R.; Wang, M. Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm. Appl. Opt. 1994, 33, 443–452. [CrossRef][PubMed] 58. Ahmad, Z.; Franz, B.A.; McClain, C.R.; Kwiatkowska, E.J.; Werdell, J.; Shettle, E.P.; Holben, B.N. New aerosol models for the retrieval of aerosol optical thickness and normalized water-leaving radiances from the SeaWiFS and MODIS sensors over coastal regions and open oceans. Appl. Opt. 2010, 49, 5545–5560. [CrossRef][PubMed] 59. 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. [CrossRef] 60. Lee, Z.P.; Carder, K.L.; Arnone, R.A. Deriving inherent optical properties from water color: A multiband quasi-analytical algorithm for optically deep waters. Appl. Opt. 2002, 41, 5755–5772. [CrossRef][PubMed] 61. Pahlevan, N.; Smith, B.; Binding, C.; O’Donnell, D.M. Spectral band adjustments for remote sensing reflectance spectra in coastal/inland waters. Opt. Express 2017, 25, 28650–28667. [CrossRef] 62. Clark, D.K.; Murphy, M.Y.; Yarbrough, M.; Feinholz, M.; Flora, S.; Broenkow, W.; Bettye, C.J.; Steven, W.B.; Kim, Y.S.; Mueller, J. MOBY, a radiometric buoy for performance monitoring and vicarious calibration of satellite ocean color sensors: Measurement and data analysis protocols. In Ocean Optics Protocols for Satellite Ocean Color Sensor Validation; Mueller, J.L., Giulietta, G.S., McClain, C.R., Eds.; NASA/NOAA/USAF: Greenbelt, MD, USA, 2003; p. 32. 63. Antoine, D.; Guevel, P.; Deste, J.-F.; Bécu, G.; Louis, F.; Scott, A.J.; Bardey, P. The “BOUSSOLE” buoy-a new transparent-to-swell taut mooring dedicated to marine optics: Design, tests, and performance at sea. J. Atmos. Ocean. Technol. 2008, 25, 968–989. [CrossRef] 64. Pahlevan, N.; Roger, J.-C.; Ahmad, Z. Revisiting short-wave-infrared (SWIR) bands for atmospheric correction in coastal waters. Opt. Express 2017, 25, 6015–6035. [CrossRef][PubMed] 65. Zibordi, G.; Mélin, F.; Berthon, J.-F.; Holben, B.; Slutsker, I.; Giles, D.; D’Alimonte, D.; Vandemark, D.; Feng, H.; Schuster, G. AERONET-OC: A network for the validation of ocean color primary products. J. Atmos. Ocean. Technol. 2009, 26, 1634–1651. [CrossRef] 66. Bailey, S.W.; Hooker, S.B.; Antoine, D.; Franz, B.A.; Werdell, P.J. Sources and assumptions for the vicarious calibration of ocean color satellite observations. Appl. Opt. 2008, 47, 2035–2045. [CrossRef][PubMed] 67. Zibordi, G.; Mélin, F. An evaluation of marine regions relevant for ocean color system vicarious calibration. Remote Sens. Environ. 2017, 190, 122–136. [CrossRef][PubMed] 68. Franz, B.A.; Kwiatowska, E.J.; Meister, G.; McClain, C.R. Moderate Resolution Imaging Spectroradiometer on Terra: Limitations for ocean color applications. J. Appl. Remote Sens. 2008, 2, 023525. [CrossRef] 69. Pahlevan, N.; Schott, J.R. Characterizing the relative calibration of Landsat-7 (ETM+) visible bands with Terra (MODIS) over clear waters: The implications for monitoring water resources. Remote Sens. Environ. 2012, 125, 167–180. [CrossRef] Remote Sens. 2018, 10, 1337 23 of 23

70. Uprety, S.; Cao, C.; Xiong, X.; Blonski, S.; Wu, A.; Shao, X. Radiometric intercomparison between Suomi-NPP VIIRS and Aqua MODIS reﬂective solar bands using simultaneous nadir overpass in the low latitudes. J. Atmos. Ocean. Technol. 2013, 30, 2720–2736. [CrossRef] 71. Cao, C.; Weinreb, M.; Xu, H. Predicting Simultaneous Nadir Overpasses among Polar-Orbiting Meteorological Satellites for the Intersatellite Calibration of Radiometers. J. Atmos. Ocean. Technol. 2004, 21, 537–542. [CrossRef] 72. Atmospheric Correction for Remotely-Sensed Ocean-Colour Products; Reports and Monographs of the International Ocean-Colour Coordinating Group (IOCCG); Wang, M. (Ed.) International Ocean-Colour Coordinating Group: Dartmouth, NS, Canada, 2010. 73. Aurin, D.; Mannino, A.; Franz, B. Spatially resolving ocean color and sediment dispersion in river plumes, coastal systems, and continental shelf waters. Remote Sens. Environ. 2013, 137, 212–225. [CrossRef] 74. Pahlevan, N.; Lee, Z.; Wei, J.; Schaff, C.; Schott, J.; Berk, A. On-orbit radiometric characterization of OLI (Landsat-8) for applications in aquatic remote sensing. Remote Sens. Environ. 2014, 154, 272–284. [CrossRef] 75. Esaias, W.E.; Abbott, M.R.; Barton, I.; Brown, O.B.; Campbell, J.W.; Carder, K.L.; Clark, D.K.; Evans, R.H.; Hoge, F.E.; Gordon, H.R. An overview of MODIS capabilities for ocean science observations. IEEE Trans. Geosci. Remote Sens. 1998, 36, 1250–1265. [CrossRef] 76. Carder, K.; Chen, F.R.; Lee, Z.; Hawes, S.K.; Cannizzaro, J.P. MODIS Ocean Science Team: Algorithm Theoretical Basis Document-ATBD19; College of Marine Science, University of South Florida: Tampa, FL, USA, 2003.

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).