HICO-Based NIR–Red Models for Estimating Chlorophyll-A Concentration in Productive Coastal Waters Wesley J

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 1 HICO-Based NIR–Red Models for Estimating Chlorophyll-a Concentration in Productive Coastal Waters Wesley J. Moses, Anatoly A. Gitelson, Sergey Berdnikov, Jeffrey H. Bowles, Vasiliy Povazhnyi, Vladislav Saprygin, Ellen J. Wagner, and Karen W. Patterson

Abstract—We present here results that demonstrate the poten- in water is a key indicator of the biophysical status of a water tial of near-infrared (NIR)–red models to estimate chlorophyll-a body (e.g., [1]) and is one of the primary water quality parame- (chl-a) concentration in coastal waters using data from the space- ters. The optical complexity of turbid productive coastal waters borne Hyperspectral Imager for the Coastal Ocean (HICO). Since the recent demise of the MEdium Resolution Imaging Spectrome- renders conventional blue–green algorithms unreliable for es- ter (MERIS), the use of sensors such as HICO has become critical timating chl-a concentration (e.g., [2]). Numerous algorithms for coastal ocean color research. Algorithms based on two- and based on analytical and semi-analytical spectral inversion tech- three-band NIR–red models, which were previously used very niques and band combinations in the red and near-infrared successfully with MERIS data, were applied to HICO images. (NIR) regions of the spectrum have been recently developed The two- and three-band NIR–red algorithms yielded accurate estimates of chl-a concentration, with mean absolute errors that and successfully validated for estimating chl-a concentration in were only 10.92% and 9.58%, respectively, of the total range of inland and coastal waters. Matthews [3] and Odermatt et al. [4] chl-a concentrations measured over a period of several months have provided comprehensive lists of such algorithms and their in 2012 and 2013 on the Taganrog Bay in Russia. Given the associated accuracy values in estimating chl-a concentration. uncertainties in the radiometric calibration of HICO, the results Many of these algorithms were regionally tuned using data illustrate the robustness of the NIR–red algorithms and validate the radiometric, spectral, and atmospheric corrections applied to from specific water bodies and are limited in their bio-geo-optic HICO data as they relate to estimating chl-a concentration in scope of application, whereas some algorithms have demon- productive coastal waters. Inherent limitations due to the charac- strated a potential for quasi-universal application to turbid and teristics of the sensor and its orbit prohibit HICO from providing productive inland and coastal waters from various locations anywhere near the level of frequent global coverage as provided [5]–[10]. by standard multispectral ocean color sensors. Nevertheless, the results demonstrate the utility of HICO as a tool for determining The aforementioned reports [3], [4] contain several algo- water quality in select coastal areas and the cross-sensor appli- rithms that were developed based on data from the MEdium cability of NIR–red models and provide an indication of what Resolution Imaging Spectrometer (MERIS). MERIS has been could be achieved with future spaceborne hyperspectral sensors a reliable tool for monitoring water quality in coastal waters. in estimating coastal water quality. The availability of a spectral channel at 708 nm made MERIS Index Terms—Chlorophyll-a, International Space Station (ISS), preferable over the MODerate resolution Imaging Spectrora- near-infrared (NIR)–red algorithms, productive coastal waters, diometer (MODIS) for estimating chl-a concentration in turbid remote sensing. productive waters, particularly at low-to-moderate chl-a concentrations, where the results from MODIS are unreliable (e.g., I. INTRODUCTION [5] and [6]). EMOTE sensing has become a very valuable and virtually The demise of MERIS in April 2012 has caused a gap in the R indispensable tool for determining water quality in inland availability of reliable ocean color data for coastal waters. This and coastal waters. The concentration of chlorophyll-a (chl-a) data gap is crucial because no hyperspectral or multispectral sensor with similar or better spectral, spatial, and temporal characteristics is scheduled to be launched in the immediate Manuscript received September 4, 2013; revised October 10, 2013; accepted future. The Ocean Land Colour Instrument, which is MERIS’ October 14, 2013. This work was supported in part by the Office of Naval Research, through the Karles Fellowship awarded by the U.S. Naval Research replacement and will have all the spectral channels of MERIS Laboratory to W. J. Moses, and by the NASA Land Cover Land Use Change in addition to a few extra channels, is scheduled to be launched Program extended to A. A. Gitelson. W. J. Moses, J. H. Bowles, E. J. Wagner, and K. W. Patterson are with the onboard the satellite Sentinel-3 in 2014. Several hyperspectral U.S. Naval Research Laboratory, Washington, DC 20375 USA. sensors, such as the Japanese mission Hyperspectral Imager A. A. Gitelson is with the Center for Advanced Land Management Infor- SUIte (HISUI), the Italian mission PRecursore IperSpettrale mation Technologies, School of Natural Resources, University of Nebraska- Lincoln, Lincoln, NE 68583 USA. della Missione Applicativa (PRISMA), and the German mis- S. Berdnikov, V. Povazhnyi, and V. Saprygin are with the Southern Scientific sion Environmental Mapping and Analysis Program (EnMAP), Center of the Russian Academy of Sciences, 344006 Rostov-on-Don, Russia. are either under design or development, with EnMAP being Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. the closest to a launch date, which is tentatively set around Digital Object Identifier 10.1109/LGRS.2013.2287458 2017–2018.

2 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS

TABLE I DATES OF IN SITU AND SATELLITE DATA COLLECTION

and the spatiotemporal dynamics of these water bodies make Fig. 1. Map of the Taganrog Bay, with screenshots of HICO images taken from the sensor’s ascending and descending orbits overlaid. The yellow dots them uniquely suitable proving grounds for developing remote represent the stations where in situ measurements were taken. sensing techniques for turbid and productive coastal and inland waters. Hyperion, which was launched in 2000, has been used Algal blooms are common in the bay and occur throughout for coastal water studies. However, its signal-to-noise ratio is the year, particularly in summer. The Taganrog Bay receives a very low [11], and the sensor is unreliable for quantitatively significant amount of freshwater input through runoff from the estimating water quality parameters due to problems such as Don River, the Kalmius River, the Mius River, and the Yeya radiometric instability [12]. River. The rapid industrial development in the watershed area The Hyperspectral Imager for the Coastal Ocean (HICO), in recent years has resulted in increased influx of terrigenous built by the U.S. Naval Research Laboratory as a low-cost pro- nutrients into the bay [15], leading to enhanced eutrophication. totype spaceborne hyperspectral sensor specifically designed Variations in the freshwater input through precipitation and for coastal waters, has been operational since October 2009. fluvial runoff affect the spatiotemporal dynamics of the algal Spaceborne hyperspectral sensors offer unique advantages over population and the species diversity in the bay [16]. Changes multispectral and airborne hyperspectral sensors for detailed in the algal biomass and the species composition have conse- coastal water analysis that can be of much help to environmen- quent effects on the fish population. Intense eutrophication has tal decision makers and managers of coastal systems. caused drastically adverse effects on the fish population and, Bearing in mind that HICO is a demonstration mission and consequently, on the local economy in the region [17]. Dedi- was not designed as an alternative to sensors such as MERIS, cated efforts have been undertaken by the local authorities to we have used a multitemporal dataset to assess the potential implement integrated watershed management plans to monitor of HICO, in the absence of MERIS, as a tool for quantita- and regulate the water quality in the Taganrog Bay and the Sea tively determining water quality in coastal waters. Previous of Azov. studies involving large datasets spanning multiple years have Researchers at the Southern Scientific Center of the Russian demonstrated the consistently high accuracy and operational Academy of Sciences, in Rostov-on-Don, Russia, undertook potential of algorithms based on reflectances in the red and NIR five in situ data collection campaigns on the Taganrog Bay in spectral channels of MERIS to estimate chl-a concentration 2012 and 2013, resulting in data from 37 stations (see Table I). in the Taganrog Bay in Russia [7]–[9]. Using a single image Water samples were collected at each station and filtered and in situ data from only eight stations, Gitelson et al. [13] through Whatman GF/F glass filters. Chl-a was extracted from showed that HICO can be used to quantitatively detect chl-a the filters using hot ethanol, and its concentration was deter- and accessory pigments such as phycocyanin and phycoerythrin mined spectrophotometrically using the trichromatic equations in water, albeit without any validation of their chl-a algorithm. of Jeffrey and Humphrey [18]. The in situ dataset did not In this study, we have used data collected from the Taganrog include reflectance measurements. Bay during multiple campaigns in 2012 and 2013 after the demise of MERIS to test the applicability of NIR–red models to HICO data to retrieve accurate quantitative estimates of chl-a B. HICO Data concentration. This study also provides an indirect qualitative HICO is a push broom sensor that captures data in the assessment of the atmospheric correction and radiometric cal- wavelength range 350–1080 nm, with a spectral resolution of ibration of HICO data, as assessed by the accuracy of chl-a 5.73 nm. At a nadir viewing angle, its cross-track and along- estimates across multiple dates. track ground coverage are 42 and 192 km, respectively, resulting in a total image area of approximately 8000 km2.The II. DATA AND METHODS ground sampling distance is approximately 90 m at nadir, which is fine enough to capture the spatial dynamics of heterogeneous A. Study Area and In Situ Measurements coastal waters (e.g., [19]). A complete description of the sensor The Taganrog Bay is located on the northeastern part of characteristics can be found in [20]. the Sea of Azov (see Fig. 1), which is a shallow inland sea As a low-cost instrument, HICO has certain inevitable data adjoined by Russia on the east and Ukraine on the west. The quality issues. The sensor does not have a filter to block second- Taganrog Bay and the Sea of Azov are characterized by low order light at shorter wavelengths from falling on the detector salinity, shallow bottom, and high influx of nutrients through pixels for longer wavelengths. Therefore, the data recorded in runoff from several rivers, resulting in conditions that are the spectral channels between roughly 700 and 1080 nm are typically encountered in eutrophic lakes [14]. The spatial extent contaminated by diffracted second-order light in the wavelength This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MOSES et al.: HICO-BASED NIR–RED MODELS FOR ESTIMATING CHL-a CONCENTRATION IN COASTAL WATERS 3 range of 350–540 nm. Li et al. [21] have developed an empirical capability of HICO helps maximize image acquisitions over method to correct for the effects of second-order light, which high-priority targets for continual (though sporadic) monitor- is applied routinely to all HICO images. HICO was calibrated ing, with images that may be acquired at different viewing and spectrally and radiometrically in the laboratory prior to its illumination geometries and at time intervals that can be as launch, but the sensor was not equipped with an onboard light short as a day or as long as a few weeks. source for post-launch calibration. Therefore, vicarious tech- Table I shows the dates of acquisition of the HICO images niques had to be adopted to monitor the spectral and radiometric used in this study. Three of the five images were acquired during stability of the sensor. an ascending pass and covered the entire bay (see Fig. 1). Gao et al. [22] reported post-launch shifts in the spectral and radiometric characteristics of the sensor. They used well- C. NIR–Red Algorithms defined prominent spectral features such as the oxygen absorption band at 765 nm to examine the spectral stability of the NIR–red algorithms take advantage of the reflectance trough sensor and observed that there were significant spectral shifts around 670 nm due to absorption of light by chl-a (e.g., [26]) in the data. The spectral shift was as high as 1.72 nm in the and the reflectance peak in the NIR region due to a minimum first few images and progressively decreased until it became in the combined absorption by chl-a and water (e.g., [27]). The fairly stable at approximately 0.9 nm. There was a significant magnitude and position of this reflectance peak vary with the decrease in the radiometric sensitivity of HICO after its launch. concentration of chl-a (e.g., [28]). The at-sensor radiances from HICO were consistently lower Numerous studies based on data collected using field spec- than the at-sensor radiances from concurrently acquired (with trometers, airborne sensors, and spaceborne sensors and data the time difference less than an hour) MODIS images and simulated using a radiative transfer model (e.g., [5]–[10]) have at-sensor radiances simulated using radiative transfer models, demonstrated that the two-band NIR–red model [28] particularly in the blue spectral region. Scaling factors were Chl-a ∝ R−1 × R (1) determined to radiometrically correct the at-sensor radiances λ1 λ2 [22]. However, even with the application of the radiometric scaling factors, the reflectances retrieved after atmospheric and the three-band NIR–red model [29] correction are often negative, particularly at wavelengths below −1 −1 Chl-a ∝ R − R × Rλ (2) 450 nm. Therefore, a minimalist approach, assuming very low λ1 λ2 3 atmospheric aerosol loading, was adopted in removing the (Rλ is the remote sensing reflectance at the spectral channel atmospheric effects from the HICO data, using the 6S [23] i centered at λi nm) yield highly accurate estimates of chl-a version of Tafkaa [24]. concentration in waters of varied bio-geo-optical properties. The focal plane array of HICO contains a back-illuminated NIR–red algorithms based on the red and NIR spectral channels charge-couple device (CCD). While the use of a back- of MERIS (λ1 = 665 nm; λ2 = 708 nm; λ3 = 753 nm) have illuminated CCD helps increase the sensitivity of the sensor, been previously shown to yield accurate estimates of chl-a particularly in the blue spectral region, it can also result in concentration in the Taganrog Bay [8], [9]. HICO does not have spectral etaloning effects due to multiple reflections of photons a spectral channel centered at 665 nm but has channels centered between the layers of the CCD, which causes undesired con- at 662 and 668 nm. Therefore, the average of the reflectances at structive and destructive interference fringes. The effects of 662 and 668 nm was taken as Rλ . HICO has spectral channels etaloning are more pronounced at the NIR wavelengths than 1 centered at 708 nm (λ2) and 754 nm (λ3). The following are at the visible wavelengths because the silicon in the CCD is the NIR–red models based on the spectral channels of HICO. increasingly transparent at longer wavelengths [25]. To min- Two-Band HICO NIR-red Model: imize the effects of etaloning, HICO data were smoothed using a Gaussian filter with 10-nm full-width at half-maximum −1 Chl-a ∝ R¯ × R708 . (3) (FWHM) for wavelengths shorter than 745 nm and a filter with 665 20-nm FWHM for wavelengths longer than 745 nm. Three-Band HICO NIR-red Model: Mounted on the International Space Station (ISS) platform, ∝ ¯−1 − −1 × HICO orbits around the Earth approximately 16 times a day. Chl-a R665 R708 R754 . (4) Due to constraints on data transmission at the ISS, only one image can be acquired during one full orbit, resulting in a R¯665 is the average of the reflectances at 662 and 668 nm. maximum of 16 images per day. The ground coverage of HICO, when viewing nadir, is limited to within 51.6◦ N and 51.6◦ S III. RESULTS AND DISCUSSION latitudes because of the inclination of the ISS’s orbit. HICO has off-nadir pointing capability (i.e., 30◦ to the right and 45◦ The data collected on July 27 and August 24, 2012 (totaling to the left), which extends the imaging range by a few degrees 16 data points) were used to calibrate the NIR–red models, and depending on the ISS altitude. Repeat coverage of the same area the data collected on August 27, 2012; September 18, 2012; happens intermittently. The ISS does not maintain a constant and February 28, 2013 (totaling 21 data points) were used altitude due to the drag from the Earth’s atmosphere and solar to validate the algorithms. The minimum, maximum, median, activity and therefore requires periodic boosts to reset its orbit. and mean values of the chl-a concentrations in the calibra- In spite of the orbital constraints of the ISS, the pointing tion dataset were, 27.06, 172.77, 103.03, and 95.92 mg m−3, This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS

Fig. 4. (a) Atmospherically corrected reﬂectance spectra from HICO. (b) Chl-a map generated from the August 25, 2012, HICO image for the Taganrog Bay Fig. 2. Plots of chl-a concentrations measured in situ versus the (a) two-band using the HICO-based two-band NIR–red algorithm (5). and (b) three-band NIR–red model values for the calibration dataset.

Fig. 3. Plots of chl-a concentrations measured in situ versus chl-a concentrations estimated using the HICO-based (a) two-band (5) and (b) three-band (6) NIR–red algorithms for the validation dataset. Fig. 5. Changes in the chl-a concentration in the Taganrog Bay between August 25 and 27, 2012, as estimated using the HICO-based two-band NIR–red algorithm [see (5)]. Positive values indicate higher chl-a concentrations, and respectively, whereas the corresponding numbers for the valida- negative values indicate lower chl-a concentrations on August 27. tion dataset were 8.1, 151.18, 83.48, and 72.26 mg m−3, respectively. Both NIR–red models showed strong linear relationships data. The current dataset contained higher and a wider range with in situ chl-a concentrations in the calibration dataset, with of chl-a concentrations than the datasets that were used for determination coefficients of 0.84 and 0.87 (see Fig. 2), leading analysis with MERIS data. The spatial and temporal variations to the following two algorithms. of chl-a concentration in the bay were also higher than in the Two-Band HICO NIR–Red Algorithm: previous years. Fig. 5 illustrates significant changes in the chl-a concentration between August 25 and 27, 2012, as estimated ¯−1 × − Chl-a = 318.33 R665 R708 278.15. (5) from HICO images using the two-band NIR–red algorithm (5). In situ measurements of chl-a concentrations confirmed such Three-Band HICO NIR–Red Algorithm: high temporal variations. Such high temporal variations make ¯−1 − −1 × the calibration and validation of algorithms inherently chal- Chl-a = 505.05 R665 R708 R754 +38.916. (6) lenging. For more than 90% of the stations in this study, the Both NIR–red algorithms yielded accurate estimates of chl-a time difference between in situ and satellite data was a day or concentration when applied to the validation dataset (see less (see Table I), which helped mitigate the effects of high Fig. 3). The estimates from the two-band NIR–red model had temporal variation. The results also demonstrate that the two- a root-mean-square error (RMSE) of 20.11 mg m−3 and a and three-band NIR–red algorithms are largely resistant to the mean absolute error (MAE) of 15.62 mg m−3, which cor- data-quality issues inherent in HICO data. responded to 14.06% and 10.92%, respectively, of the range of chl-a concentrations in the validation dataset. The three- IV. CONCLUSION band NIR–red model gave slightly better estimates, with an RMSE of 17.73 mg m−3 and an MAE of 13.71 mg m−3, which The accuracy of the results obtained from the multitemporal corresponded to 12.39% and 9.58%, respectively, of the range dataset, notwithstanding the issues with the radiometric quality of chl-a concentrations. The results also validate the quality and of HICO data and the assumptions contained in the atmospheric reliability of the atmospherically corrected reflectances from correction [22], demonstrates the robustness of the simple two- HICO [Fig. 4(a)] for retrieving chl-a concentration. Chl-a maps and three-band NIR–red models and the potential of HICO for [e.g., Fig. 4(b)] generated using these algorithms illustrated the estimating chl-a concentration in optically complex productive spatial patterns of chl-a distribution and matched quantitatively coastal waters. well with in situ measurements. It should be noted that the specific two-band (5) and three- Although the errors are higher in terms of absolute magni- band (6) algorithms developed here are primarily meant for tudes and percentages than what were obtained from MERIS demonstrative purposes and proper caution needs to be exer- data in previous years for the Taganrog Bay [8], [9], the cised in applying them to future HICO data from the Taganrog results are still remarkable, considering the range and temporal Bay or elsewhere. The validity of the coefficients of these variations of the chl-a concentration in the bay and the uncer- algorithms is contingent on the validity of the radiometric tainties in the radiometric stability and integrity of the HICO corrections and the assumptions contained in the atmospheric This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MOSES et al.: HICO-BASED NIR–RED MODELS FOR ESTIMATING CHL-a CONCENTRATION IN COASTAL WATERS 5

correction of the HICO data. It is conceivable that the current [10] K. Song, L. Li, L. Tedesco, S. Li, H. Duan, D. Lio, B. Hall, J. Du, atmospheric correction might fail drastically in coastal waters Z. Li, K. Shi, and Y. Zhao, “Remote estimation of chlorophyll-a in turbid inland waters: Three-band model versus GA-PLS model,” Remote Sens. that are adjoined by industrial areas with high and variable Environ., vol. 136, pp. 342–357, Sep. 2013. atmospheric aerosol loading. Moreover, the radiometric sta- [11] C. Hu, L. Feng, Z. Lee, C. O. Davis, A. Mannino, C. R. McClain, and bility of the instrument and, consequently, the validity of the B. A. Franz, “Dynamic range and sensitivity requirements of satellite ocean color sensors: Learning from the past,” Appl. Opt., vol. 51, no. 25, radiometric corrections need to be continually monitored. One pp. 6045–6062, Sep. 2012. has to bear in mind also that the ISS platform does not permit [12] National Research Council, Earth Science and Applications From Space: frequent repeat coverage of the same area at regular time inter- National Imperatives for the Next Decade and Beyond, Washington, DC, USA: The National Academic Press, 2007. vals. Image acquisitions of the same area are intermittent, with [13] A. A. Gitelson, B-C. Gao, R-R. Li, S. Berdnikov, and V. Saprygin, “Es- time intervals sometimes as short as a day or as long as several timation of chlorophyll-a concentration in productive turbid waters using weeks, which places a constraint on the ability to monitor short- a hyperspectral imager for the coastal ocean—The Azov Sea case study,” Environ. Res. Lett., vol. 6, no. 2, pp. 024023-1–024023-6, Apr.–Jun. 2011. term phytoplankton dynamics in coastal waters. Nevertheless, [14] O. Borysova, A. Kondakov, S. Paleari, E. Rautalahti-Miettinen, the results presented here demonstrate the utility of HICO in F. Stolberg, and D. Daler, “Eutrophication in the Black Sea region: Impact estimating water quality in productive coastal waters, which assessment and causal chain analysis,” in Global International Waters Assessment. Kalmar, Sweden: University of Kalmar, 2005. is of critical value, particularly with the demise of MERIS. [15] G. Matishov, “System approach to the water quality and bioproductivity The availability of spectral channels at 662, 668, and 708 nm of the Azov Sea basin,” in River Basin Management III; WIT Transac- on HICO makes it possible to obtain chl-a estimates using tions on Ecology and the Environment, C. A. Brebbia, Ed. Southampton, U.K.: WIT Press, 2005, pp. 347–357. the two- and three-band NIR–red algorithms with reasonably [16] I. V. Shokhin, M. V. Nabozhenko, S. V. Sarvillina, and E. P. Titova, “The good accuracies, which is not possible with MODIS or the Sea- present-day condition and regularities of the distribution of the bottom viewing Wide Field-of-view Sensor (SeaWiFS) data. The fine communities in Taganrog Bay,” Oceanology, vol. 46, no. 3, pp. 401–410, May/Jun. 2006. hyperspectral information contained in HICO data can be also [17] E. Leppäkoski, T. Shiganova, and B. Alexandrov, “European enclosed potentially used for water quality analysis beyond merely chl-a and semi-enclosed seas,” in Biological Invasions in Marine Ecosystems: estimation, such as algal species distinction and quantification Ecological Studies, G. Rilov and J. A. Crooks, Eds. Berlin, Germany: Springer-Verlag, 2009, ch. 30, pp. 529–547. of ancillary pigments in water. [18] S. W. Jeffrey and G. F. Humphrey, “New spectrophotometric equations for Continued work is planned to acquire more HICO images determining chlorophylls a, b, c1, c2 in higher plants, algae, and natural and in situ measurements on the Taganrog Bay to develop stable phytoplankton,” Biochem. Physiol. Pflanzen, vol. 167, pp. 191–194, 1975. [19] C. O. Davis, M. Kavanaugh, R. Letelier, P. W. Bissett, and D. Kohler, NIR–red algorithms that can be routinely applied to HICO data “Spatial and spectral resolution considerations for imaging coastal wa- from the bay and elsewhere. ters,” in Proc. SPIE, Characterization Variability Coast. Ocean, Compos. Bio-Opt. Properties II, 2007, vol. 6680, p. 66800. REFERENCES [20] R. L. Lucke, M. Corson, N. R. McGlothlin, S. D. Butcher, D. L. Wood, D. R. Korwan, R. R. Li, W. A. Snyder, C. O. Davis, and D. T. Chen, [1] J. H. Ryther and C. S. Yentsch, “The estimation of phytoplankton pro- “Hyperspectral imager for the coastal ocean: Instrument description and duction in the ocean from chlorophyll and light data,” Limnol. Oceanogr., first images,” Appl. Opt., vol. 50, no. 11, pp. 1501–1516, Apr. 2011. vol. 2, no. 3, pp. 281–286, 1957. [21] R-R. Li, R. Lucke, D. Korwan, and B-C. Gao, “A technique for removing [2] G. Dall’Olmo, A. Gitelson, D. Rundquist, B. Leavitt, T. Barrow, and second-order light effects from hyperspectral imaging data,” IEEE Trans. J. Holz, “Assessing the potential of SeaWiFS and MODIS for estimating Geosci. Remote Sens., vol. 50, no. 3, pp. 824–830, Sep. 2012. chlorophyll concentration in turbid productive waters using red and near- [22] B-C. Gao, R-R. Li, R. L. Lucke, C. O. Davis, R. M. Bevilacqua, infrared bands,” Remote Sens. Environ., vol. 96, no. 2, pp. 176–187, D. R. Korwan, M. J. Montes, J. H. Bowles, and M. R. Corson, “Vicarious May 2005. calibrations of HICO data acquired from the International Space Station,” [3] M. W. Matthews, “A current review of empirical procedures of remote Appl. Opt., vol. 51, no. 14, pp. 2559–2567, May 2012. sensing in inland and near-coastal transitional waters,” Int. J. Remote [23] E. F. Vermote, D. Tanre, J. L. Deuze, M. Herman, and J. J. Morcrette, Sens., vol. 32, no. 21, pp. 6855–6899, Nov. 2011. “Second simulation of the satellite signal in the solar spectrum, 6S: An [4] D. Odermatt, A. Gitelson, E. Brando, and M. Schaepman, “Review of overview,” IEEE Trans. Geosci. Remote Sens., vol. 35, no. 3, pp. 675– constituent retrieval in optically deep and complex waters from satellite 686, May 1997. imagery,” Remote Sens. Environ., vol. 118, pp. 116–126, Mar. 2012. [24] M. J. Montes, B. C. Gao, and Davis C. O., “A new algorithm for atmo- [5] W. J. Moses, A. A. Gitelson, S. Berdnikov, and V. Povazhnyy, “Esti- spheric correction of hyperspectral remote sensing data,” in Proc. SPIE, mation of chlorophyll-a concentration in case II waters using MODIS Geo-Spatial Image Data Exp. II, 2001, vol. 4383, pp. 23–30. and MERIS data—Successes and challenges,” Environ. Res. Lett.,vol.4, [25] Roper Scientific Inc.. (2000). Etaloning in back illuminated CCDs, no. 4, pp. 045005-1–045005-8, Oct.–Dec. 2009. Trenton, NJ, USA, Tech. Note. [Online]. Available: http://www. [6] D. Gurlin, A. A. Gitelson, and W. J. Moses, “Remote estimation of chl-a princetoninstruments.com/Uploads/Princeton/Documents/Whitepapers/ concentration in turbid productive waters—Return to a simple two-band etaloning.pdf NIR-red model?” Remote Sens. Environ., vol. 115, no. 12, pp. 3479–3490, [26] V. M. Albers and H. V. Knorr, “Absorption spectra of single chloroplasts Dec. 2011. in living cells in the region from 664 m μ to 704 m μ,” Plant Physiol., [7] A. Gilerson, A. Gitelson, J. Zhou, D. Gurlin, W. Moses, I. Ioannou, and vol. 12, no. 3, pp. 833–843, Jul. 1937. S. Ahmed, “Algorithms for remote estimation of chlorophyll-a in coastal [27] A. Vasilkov and O. Kopelevich, “Reasons for the appearance of the max- and inland waters using red and near infrared bands,” Opt. Exp., vol. 18, imum near 700 nm in the radiance spectrum emitted by the ocean layer,” no. 23, pp. 24 109–24 125, Nov. 2010. Oceanology, vol. 22, no. 6, pp. 697–701, 1982. [8] W. J. Moses, A. A. Gitelson, S. Berdnikov, and V. Povazhnyy, “Satellite [28] A. Gitelson, “The peak near 700 mm on radiance spectra of algae and estimation of chlorophyll-a concentration using the red and NIR bands water—Relationships of its magnitude and position with chlorophyll of MERIS—The Azov Sea case study,” IEEE Geosci. Remote Sens. Lett., concentration,” Int. J. Remote Sens., vol. 13, no. 17, pp. 3367–3373, vol. 4, no. 6, pp. 845–849, Oct. 2009. Nov. 1992. [9] W. J. Moses, A. A. Gitelson, S. Berdnikov, S. Saprygin, and [29] G. Dall’Olmo and A. Gitelson, “Effect of bio-optical parameter variability V. Povazhnyy, “Operational MERIS-based NIR-red algorithms for esti- on the remote estimation of chlorophyll-a concentration in turbid produc- mating chlorophyll-a concentrations in coastal waters—The Azov Sea tive waters: Experimental results,” Appl. Opt., vol. 44, no. 3, pp. 412–422, case study,” Remote Sens. Environ., vol. 121, pp. 118–124, Jun. 2012. Jan. 2005.