Improved Ocean-Color Remote Sensing in the Arctic Using the POLYMER Algorithm

Transcription

1 Improved Ocean-Color Remote Sensing in the Arctic Using the POLYMER Algorithm Robert Frouin* a, Pierre-Yves Deschamps a,b, Didier Ramon b, François Steinmetz b a Scripps Institution of Oceanography, La Jolla, California, USA b Hygeos, Euratechnologies, Lille, France ABSTRACT Atmospheric correction of ocean-color imagery in the Arctic brings some specific challenges that the standard atmospheric correction algorithm does not address, namely low solar elevation, high cloud frequency, multi-layered polar clouds, presence of ice in the field-of-view, and adjacency effects from highly reflecting surfaces covered by snow and ice and from clouds. The challenges may be addressed using a flexible atmospheric correction algorithm, referred to as POLYMER (Steinmetz and al., 2011). This algorithm does not use a specific aerosol model, but fits the atmospheric reflectance by a polynomial with a non spectral term that accounts for any non spectral scattering (clouds, coarse aerosol mode) or reflection (glitter, whitecaps, small ice surfaces within the instrument field of view), a spectral term with a law in wavelength to the power -1 (fine aerosol mode), and a spectral term with a law in wavelength to the power -4 (molecular scattering, adjacency effects from clouds and white surfaces). Tests are performed on selected MERIS imagery acquired over Arctic Seas. The derived ocean properties, i.e., marine reflectance and chlorophyll concentration, are compared with those obtained with the standard MEGS algorithm. The POLYMER estimates are more realistic in regions affected by the ice environment, e.g., chlorophyll concentration is higher near the ice edge, and spatial coverage is substantially increased. Good retrievals are obtained in the presence of thin clouds, with ocean-color features exhibiting spatial continuity from clear to cloudy regions. The POLYMER estimates of marine reflectance agree better with in situ measurements than the MEGS estimates. Biases are or less in magnitude, except at 412 and 443 nm, where they reach and 0.002, respectively, and root-mean-squared difference decreases from at 412 nm to less than at 620 and 665 nm. A first application to MODIS imagery is presented, revealing that the POLYMER algorithm is robust when pixels are contaminated by sea ice. Keywords: Ocean-color remote sensing, marine reflectance, chlorophyll concentration, MERIS, ICESCAPE 1. INTRODUCTION Ice cover plays a key role in the Arctic ecosystem. It affects water temperature and salinity, stratification, light distribution, and the transport of nutrients and carbon. Most Arctic marine species depend on the presence of sea ice. In winter, ice algae clings to the underside of the dark ice. In spring, as darkness ends, phytoplankton blooms develop beneath the ice. As the ice breaks up, the blooms spread into a belt several tens of kilometers wide surrounding the ice edge. This highly productive region is home to numerous invertebrates, which, in turn, are eaten by fish species. Organic matter released from the ice and ice-edge algae enriches the floor of the continental shelf, supporting benthic communities. Seals, walrus, and cetaceans follow the ice edge, which moves thousands of kilometers each year, taking advantage of the readily available food. The decrease of Arctic sea ice documented during the last decades (e.g., Serreze et al., 2007) is disruptive to the seaice dependent organisms, including various species of phytoplankton, zooplankton, crustaceans, the seals that feed on them, and the polar bears and humans that feed on the seals. As more solar radiation is allowed to penetrate the *SIO/UCSD, Climate, Atmospheric Science, and Physical Oceanography Division, 9500 Gilman drive, La Jolla, California ; rfrouin@ucsd.edu; phone Remote Sensing of the Marine Environment II, edited by Robert J. Frouin, Naoto Ebuchi, Delu Pan, Toshiro Saino, Proc. of SPIE Vol. 8525, 85250I 2012 SPIE CCC code: /12/$18 doi: / Proc. of SPIE Vol I-1

2 surface, the ocean heat content is increased, which tends to create a positive feedback by lengthening the ice-free season (Perovich et al., 2007). The changes in the physical environment might change the amount of nutrients in the water, preventing blooms from occurring or changing the type of phytoplankton species that bloom. The timing of ice retreat, by influencing the timing of the phytoplankton blooms, might favor the development of benthic organisms or provide more food to the pelagic food web depending on whether zooplankton are present or not (Michel et al., 2006). The annual extent and duration of the ice pack, by determining bottom temperatures, can also influence fish distributions (Wyllie-Echevarria, 1995). Pabi et al., (2008) and Arrigo et al. (2008) reported a 40% increase of Arctic marine production during the last decade (26% between 2003 and 2007), which they attributed to a decreased minimum summer ice extent and, more importantly, a longer growing season. The impacts on marine ecosystem structure and carbon export are potential, and some scenarios are hypothesized, but further study is required. Arrigo et al. (2008) also indicated that if the current trend in primary production is maintained, the resulting decrease in partial pressure of carbon dioxide would partially offset the increased out-gassing of carbon dioxide expected as Arctic waters warm during the coming years, constituting a negative feedback on climate change. Investigating and understanding the links between physical environmental changes and primary production and ecosystem structure in the Arctic, and the local and global implications associated with decreasing sea ice, requires adequate observations of the physical environment and lower level production, as well as ocean-biogeochemical models that can assimilate some of the observations and test and predict changes in the marine ecosystems. Ocean color variables such as chlorophyll concentration are essential to achieve those objectives. Current satellite ocean color sensors have the capability to provide information at the required space and time scales (e.g., to describe the timing of phytoplankton blooms), but the standard products are limited in accuracy and coverage. Atmospheric correction of ocean-color imagery in the Arctic requires taking into account some specific issues that the standard atmospheric correction algorithm does not address, namely low solar elevation, high cloud frequency, multi-layered polar clouds, presence of ice in the field-of-view, and adjacency effects from highly reflecting surfaces covered by snow and ice and from clouds. Adjacency effects, in particular, may yield erroneous chlorophyll concentrations at distances of 10 km, even 20 km from the ice edge (Santer and Schmechtig, 2000), i.e., in the productive region where phytoplankton blooms develop as the ice begins to break-up in the spring. In the Arctic, the complex interactions between the scattering by molecules, aerosols, clouds, and the surface reflection are not described accurately by the standard algorithm, which has the only flexibility to tune its aerosol model to fit the measurements. Large uncertainties are therefore associated with the standard ocean-color products, whose spatial and temporal coverage is considerably reduced by cloudiness. Even weekly composites show no information in many areas, as illustrated in Figure 1, which displays SeaWiFS-derived maps of chlorophyll concentration in the Chukchi and Beaufort seas for 26 June-03 July, July July, July, and 28 July- 04 August During those periods, 18%, 16%, 23%, 45%, and 56% of the open ocean pixels are missing, respectively, and the monthly variability is not easily inferred. Figure 2 indicates that 8-day composites may only cover less than 60% of the ice-free Chukchi and Beaufort seas during most of the year. The existing satellite ocean color products should be improved, and their spatial coverage increased, if one wants to determine the spatial and temporal characteristics of phytoplankton blooms, estimate accurately primary production and its changes, and understand and assess the impact of climate change on the Arctic ecosystem. To address the issues identified above, the satellite imagery is processed with a more flexible atmospheric correction algorithm referred to as POLYMER, originally developed to handle MERIS data in glitter-affected areas. This algorithm does not use a specific aerosol model but fits the atmospheric reflectance by a polynomial with (1) a non spectral term that accounts for any non spectral scattering (clouds, coarse aerosol mode) or reflection (glitter, whitecaps, small ice surfaces), (2) a spectral term with a power law in λ -1 (fine aerosol mode), and (3) a spectral term with a power law in λ -4 (molecular scattering, adjacency effects from clouds and white surfaces). The POLYMER algorithm is applied to selected MERIS imagery with thin clouds and adjacency effects acquired over the Chucki and Beaufort Seas. The derived ocean properties, i.e., marine reflectance and chlorophyll concentration, are compared with those obtained with the standard MEGS algorithm. Accuracy of the derived marine reflectance is quantified in comparisons with measurements performed during the ICESCAPE 2010 and 2011 field campaigns. Proc. of SPIE Vol I-2

3 26 Jun - 3 Jul, 2003 mailiega ` _.. 4 Jul - '11 Jul, Jul - 19 Jul, Jul - 27 Jul, 2003 Chla mg,ms >15.0 i1.5 < Jul - 4 Aug, 2003 Figure 1: 8-day level 3 composites of SeaWiFS-derived chlorophyll concentration in the Chukchi and Beaufort Seas during 26 June 04 August 2003 (standard processing). The percent of missing ocean pixels is 18%, 16%, 23%, 45%, and 56% for the periods 26 Jun-3 Jul, 4 Jul-11 Jul, 12 Jul -19 Jul, 20 Jul-27 Jul, and 28 Jul-4 Aug, 2003, respectively Julian Day, Figure 2: Percentage during June to October 2003 of the Chukchi and Beaufort Seas (ice-free area) for which chlorophyll concentration is retrieved by the standard SeaWiFS processing in 8-day composites. Proc. of SPIE Vol I-3

4 2. THE POLYMER ALGORITHM The POLYMER algorithm (Steinmetz, et al., 2011) is a spectral matching algorithm of the observed reflectance R by models of atmospheric scattering and glitter, R agm, and of marine reflectance, R w. Three parameters are used to describe the atmospheric scattering and glitter, two parameters are used to vary the marine reflectance as a function of chlorophyll concentration and particle backscattering coefficient. The observed reflectance is expressed as: R = (R m + R agm + T m T a R w )T g, (1) where R m is molecular reflectance, T m is molecular transmittance, T a is aerosol transmittance, T g is gaseous transmittance, and R m and T m are obtained from pre-computed look-up tables. 2.1 Aerosol and glitter model Aerosol scattering and glitter are modeled as a polynomial function of the wavelength: R agm = C 0 + C 1 λ -1 + C 2 λ -4. (2) The rationale behind the above model is that glitter has no spectral dependence, and the aerosol scattering has some wavelength dependence with a power law having a negative exponent, the so-called Angstrom coefficient, between 0 and 1. The two first terms of Equation (2) do account for glitter and aerosol scattering when molecular scattering is negligible. The third term of the polynomial approximates the complex coupling with molecular scattering. Note that no aerosol model is used in Equation (2), giving simplicity and robustness to the algorithm. 2.2 Marine reflectance model The marine reflectance is parameterized as a function of the chlorophyll concentration, [Chl], and a backscattering coefficient for non-phytoplankton particles, B bs : R w = f ([Chl], B bs ). (3) In the modeling, R w is computed between 400 and 700 nm as a function of [Chl] according to Morel and Maritorena (2001), but a variable B bs is added to their phytoplankton backscattering coefficient to represent more turbid waters. The marine reflectance spectrum is extended beyond 700 nm following the similarity spectrum of Ruddick et al. (2006). 2.3 Correction algorithm The gaseous transmittance T g is computed using the best available ozone vertical content. The molecular scattering terms R m and T m are pre-computed and adjusted using the best available meteorological data for sea level atmospheric pressure and wind speed. Then the simplex method (Nelder and Mead, 1965) is used to tune the five parameters, C 0, C 1, C 2, [Chl], and B bs by minimizing the difference between the reflectance observed and modeled by Equation (1). Next the top-of-atmosphere reflectance is corrected for atmospheric scattering and glitter using the correction defined by parameters C 0, C 1, and C 2, to retrieve the marine reflectance. 2.3 Application example Figure 3 gives a typical example of the result of processing MERIS data by the POLYMER algorithm in the Arctic Ocean. The MERIS image was acquired over the Beaufort Sea on July 1, A large part of the image is covered by sea ice. A bit mask was applied to reject ice/cloud pixels and invalid pixels (i.e., too negative backscattering coefficient, out-of-bound values). This mask is less conservative than the standard MEGS processing mask. In particular, retrievals are obtained for cloudy pixels with a top-of-atmosphere reflectance at 865, corrected for Proc. of SPIE Vol I-4

5 molecular scattering effects, R 865, less than The POLYMER values are realistic in the region affected by the ice environment (upper left part of the image), with no abnormal values near the ice edge. Rw443 no units 0.03! c. Chla mg/m Figure 3: Example of MERIS imagery (1-km resolution) processed with the POLYMER algorithm, Beaufort Sea, July 1, 2008, 20:15 GMT. Top left: R TOA - R m at 865 nm.top right: R w at 443 nm. Lower left: R w at 560 nm. Lower right: Chlorophyll concentration. 3. EVALUATION AGAINST IN-SITU MEASUREMENTS The marine reflectance derived by the POLYMER algorithm has been compared with in-situ measurements made during the 2010 and 2011 ICESCAPE field campaigns. The marine reflectance data were collected with a Biospherical Inc. PRR profiler (data provided courtesy of Dr. Greg Mitchell, Scripps Institution of Oceanography). Two types of match-up data sets were considered. Two types of comparisons were made. First only the satellite estimates within one pixel of the surface measurement location (nearest neighbor pixel) were used (first data set). Second the satellite estimates were obtained by averaging all the good pixels within a 5x5 pixel box centered on the surface measurement location (second data set). Only MERIS observations acquired within ±3 hours of the surface measurement were selected. Flags used in the MERIS standard algorithm (MEGS) were applied to select the good pixels (i.e., not contaminated by ice, clouds, etc.). This resulted in 18 and 14 match-ups, respectively. Table 1 gives the list of match-ups with date, time, location, and time difference between satellite observation and in situ measurement, and Figure 4 shows the geographic location of the match-ups. On some occasions, the same in situ measurement was used for consecutive orbit passes. Most of the match-ups of the first data set (nearest neighbor pixels) are included in the second data set (5x5 pixel box averages). Figure 5 displays, for selected match-ups, MERIS imagery at 865 nm showing where the match-ups are located with respect to the surrounding environment. In some cases, the location is near ice (July 9, 2010) or clouds (June 29, 2010). Proc. of SPIE Vol I-5

6 Table 1: Satellite/surface match-ups of marine reflectance during the 2010 and 2011 ICESCAPE field experiments. ΔTime is the time difference between satellite estimate and measurement, 5x5 refers to comparisons using a 5x5 pixel box, and NN refers to comparisons with the nearest neighbor. Match-ups are within ±3 hours. Year Month Day Time (h) ΔTime (h) Lat (deg.) Lon (deg.) 5x5 NN N ICESCAPE o 70N lat 65N 60N 170W 160W 155W 1S0W Figure 4: Geographic location of marine reflectance match-ups (±3 hours) during the 2010 and 2011 ICESCAPE field experiments. The red crosses correspond to match-ups used in the comparison with nearest neighbor, the blue circles to matchups used in the comparison using a 5x5 pixel box about the in situ measurement location. Proc. of SPIE Vol I-6

7 , v, 07/09/ :12 GMT ti.}' ü, 06/29/2011, 22:39 GMT Air _ 07/08/2010, 22:25 GMT 06/21/2010, 22:59 GMT 7.:: -, Figure 5: MERIS imagery at 865 nm showing the location of selected match-ups (red boxes) with respect to clouds, ice, or land... `,. The red boxes are 11x11pixels in size. Match-up selection was made based on flags used in the standard MERIS data processing. Table 2 gives the performance statistics for the first (nearest neighbor) match-up data set, and Figure 6 (left) displays estimated versus measured values. (Results obtained using 5x5 boxes, which are similar, are not presented here.) The estimates generally agree with the measurements. Biases are about or less in magnitude, except at 412 and 443 nm, where they reach and 0.002, respectively. The root-mean-squared difference decreases from at 412 nm to less than at 620 and 665 nm. Correlation coefficients (r 2 ) are below 0.5 at 412 and 443 nm and above 0.5 at other wavelengths. The relatively low correlation coefficients at the shorter wavelengths may be explained by small variability in the data (range of values). Table 2: Performance statistics of the POLYMER algorithm during ICESCAPE. Comparisons are made with the nearest neighbor pixel. λ(nm) Mean(sat) Mean(in situ) r 2 Bias RMSE N Proc. of SPIE Vol I-7

8 The marine reflectance estimated by the MEGS algorithm has also been compared with in situ measurements, in the same way as described above (data sets, procedures) for the POLYMER algorithm. The results are displayed in Table 3 and Figure 6 (right) for the first data set (nearest neighbor pixels. Compared with the POLYMER retrievals, the MEGS retrievals differ substantially from the measurements, with higher root-mean-squared differences and lower correlation coefficients. This suggests that the POLYMER algorithm handles better pixels located near clouds and sea ice, which yield erroneous values when processed by the MEGS algorithm. Table 3: Performance statistics of the MEGS algorithm during ICESCAPE. Comparisons are made with the nearest neighbor pixel. λ(nm) Mean(sat) Mean(in situ) r 2 Bias RMSE N < n 560 o =Pfi, RW, in situ I I RW, in situ Figure 6: Comparison of marine reflectance estimates by POLYMER (left) and MEGS (right) with in situ measurements during the ICESCAPE field campaigns. The comparison is made using the nearest neighbor pixel. Math-ups are within ±3 hours. 4. ROBUSTNESS IN THE PRESENCE OF THIN CLOUDS AND SEA ICE Several MERIS images have been processed with the POLYMER algorithm in the Arctic region. Figure 7 (left) displays an example for the Chukchi Sea, where one can see that POLYMER (top right) performs better than MEGS Proc. of SPIE Vol I-8

9 (top left) in the presence of thin clouds. Water reflectance under a large semi-transparent cloud (left of the image) is correctly retrieved by POLYMER, but not by MEGS. There is continuity between features under the cloud and adjacent to the cloud. This may be partly due to thin clouds interpreted as white aerosols in the spectral optimization. Figure 7 (right) displays an example of processing in the Beaufort Sea, at the delta of the Mackenzie River. The POLYMER processing (middle right, and bottom right) retrieves a consistent parameter over ice-free areas in the middle of the ice pack. The standard processing, on the other hand, does not account for environment effects (middle left and bottom left), leading to an anomalous increase of water reflectance. More precisely, near the ice edge MEGS underestimates the perturbing atmospheric signal at visible wavelengths (i.e., the coupling between molecular scattering and reflection by sea ice, which varies as λ -4, is not determined properly using wavelengths in the near infrared). v -. R.040W. ' " ` e, K4 TOA reflectance at 865 nm 110'.. z Water reflectance at 560 nm, MEGS processing OA= #.-.i `r;= A..\ Water reflectance at 560 nm, POLYMER processing Ns Water reflectance at 560 nm, MEG5 processing Water reflectance at 560 nm, POLYMER processing X:. reflectance at 865 nmytoa Chukchi Sea, , MERIS Water reflectance at 560 nm, Water reflectance at 560 nm, MEGS processing (detail) POLYMER processing (detail) Beaufort Sea, , MERIS Figure 7: (Left Panels) MERIS image of the Chukchi Sea showing that water reflectance is retrieved by POLYMER, but not by MEGS in the presence of a large semi-transparent cloud (left of the image). (Right) MERIS image of the Beaufort Sea, at the delta of the Mackenzie River, showing that the marine reflectance derived by POLYMER is consistent over ice-free areas in the middle of the ice pack, while the MEGS processing does not account properly for the ice environment, leading to an anomalous increase of retrieved reflectance. Proc. of SPIE Vol I-9

10 One example of a Level 3 product (, i.e., chlorophyll concentration) is presented in Figure 8, which displays 8-day composites of processing by MEGS, POLYMER, and standard MODIS in the bay of Baffin. Here, only the POLYMER processing applied to MERIS imagery (top left) is robust to the environment effect of the ice shelf, while the standard processing for MERIS and MODIS (top right and bottom right, respectively) reveals an anomalous decrease in chlorophyll concentration towards the edge of the ice shelf, Northwest of the ice-free main water body (denoted by A in Figure 8). The standard atmospheric correction underestimates the adjacency contribution, which varies spectrally as λ -4, as explained above, yielding too high marine reflectance in the blue and green, therefore too low chlorophyll concentration. This suggests that primary production estimates for the Arctic Ocean, obtained using standard satellite estimates of chlorophyll concentration, may be biased low. irslmi Bra 2 Figure 8: 8-day level 3 composites of [chl] in the bay of Baffin, from to The color scale unit is log10([chl]). Top-left: Standard processing of MERIS data, top-right: POLYMER processing of MERIS data, bottom-left: observation at 865 nm, and bottom-right: MODIS level 3 processing. The MERIS standard processing and the MODIS processing show an anomalous decrease of chlorophyll concentration in the vicinity of the ice shelf (marked by "A"), but not the POLYMER processing. The POLYMER algorithm has been adapted and applied to MODIS data. Results are preliminary (e.g., ozone transmission is not yet taken into account). The MODIS spectral band at 1240 nm is used to determine the atmospheric signal, but not the spectral bands at 1640 and 2135 nm, which will be used in a second version of the algorithm. Figure 9 displays the first MODIS image processed with the POLYMER algorithm. Region is Eastern Beaufort Sea and date is July 13, Land is displayed in green, and ice/clouds are masked in blue (OPBPG mask). The POLYMER algorithm is more robust when pixels are contaminated by sea ice, as abnormal white dots in the center of the left image (standard OBPG processing) do not appear in the right image (POLYMER processing). Proc. of SPIE Vol I-10

11 Figure 9: Example of 1-km MODIS imagery processed into Chlorophyll concentration with the POLYMER algorithm. Region is Eastern Beaufort Sea and date is June 13, Left: Standard OBPG processing. Right: POLYMER processing. Land is displayed in green, and ice/clouds are masked in blue (OPBPG mask). The POLYMER algorithm is more robust when pixels are contaminated by sea ice, as abnormal white dots in the center of the left image do not appear in the right image. 5. CONCLUSIONS The POLYMER algorithm (Steinmetz and al, 2011) has been evaluated on satellite imagery acquired over the Arctic seas. The derived ocean properties, i.e., marine reflectance and chlorophyll concentration, when compared with those obtained from MERIS data and the standard MEGS algorithm, are more realistic in regions affected by the ice environment, e.g., higher chlorophyll concentration near the ice edge. Agreement is better with in-situ measurements. Good retrievals are obtained in the presence of thin clouds, with ocean color features exhibiting spatial continuity from clear to cloudy regions. Spatial coverage is substantially increased on individual images, but not on weekly or monthly composites, because of the multiple orbits at high latitudes. The number of observations in the composites, however, is generally increased when using the POLYMER algorithm. The marine reflectance derived by the POLYMER and MEGS algorithms has been compared with in-situ measurements made during the 2010 and 2011 ICESCAPE field campaigns. The POLYMER estimates generally agree with the measurements. Biases are about or less in magnitude, except at 412 and 443 nm, where they reach and 0.002, respectively. The root-mean-squared difference decreases from at 412 nm to less than at 620 and 665 nm. Compared with the POLYMER retrievals, the MEGS retrievals differ substantially from the measurements, with higher root-mean-squared differences and lower correlation coefficients. The improved capability and accuracy of POLYMER in the presence of ice/clouds will help to determine the characteristics of Arctic phytoplankton blooms, which develop over relatively short time scales. It will also contribute to more accurate primary production estimates for the Arctic Ocean. Assimilation of the chlorophyll concentration data into numerical ocean-biogeochemical models will contribute to a better understanding and forecasting of air-sea exchanges of carbon dioxide and of taxonomic changes within phytoplankton communities associated with the decline of sea ice extent and environmental change. Proc. of SPIE Vol I-11

12 ACKNOWLEDGMENTS The National Aeronautics and Space Administration provided funding for this work under grant No. NNX10AH61G. The technical support of Mr. John McPherson from the Scripps Institution of Oceanography, University of California at San Diego, is gratefully acknowledged. REFERENCES [1] Arrigo, K.R., van Dijken, G. and Pabi, S. (2008), Impact of a shrinking Arctic ice cover on marine primary production, Geophysical Research Letters, 35, L [2] Michel, C., Ingram, R.G. and Harris, L.R. (2006), Variability in oceanographic and ecological processes in the the Canadian Arctic Archipelago, Progress in Oceanography, 71, pp [3]Morel, A. and Maritorena, S. (2001), Bio-optical properties of oceanic waters: a reappraisal, Journal of Geophysical Research, 106, pp [4] Nelder, J. A. and Mead, R. (1965), A Simplex Method for Function Minimization, Computer Journal, 7, pp [5] Pabi, S., van Dijken, G., and Arrigo, K.R. (2008), Primary production in the Arctic ocean, , Journal of Geophysical Research, bf 113, C08005, doi: /2007jc [6] Perovich, D.K., Light, B., Eicken, H., Jones, F., Runciman, K. and Nghiem, S.V. (2007), Increasing solar heating of the Arctic Ocean and adjacent seas, : Attribution and role in the ice-albedo feedback, Geophysical Research Letters, 34, L [7] Ruddick, K.G., De Cauwer, V., Park, Y. and Moore, G. (2006), Seaborne measurements of near infrared waterleaving reflectance: the similarity spectrum for turbid waters, Limnology and Oceanography, 51, pp [8] Santer, R., and Schechtig, C. (2000). Adjacency effects on water surfaces: primary scattering approximation and sensitivity study, Applied Optics, 38, pp [9] Serreze, M.C., Holland, M.M. and Stroeve, J. (2007), Perspectives on the Arctic's shrinking sea-ice cover, Science, 110, pp [10] Steinmetz, F., Deschamps, P.-Y.. and Ramon, D. (2011), Atmospheric correction on the presence of sun glint: application to MERIS, Optics Express, 19, [11] Wylllie-Echeverria, T. (1995), Sea-ice conditions and the distribution of walleye pollock (Chalcogramma theragra) on the Bering and Chukchi shelf. In: Beamish, R.J. (Ed.), Climate change and northern fish populations, Canadian Special Publication in Fisheries and Aquatic Science, 121, pp Proc. of SPIE Vol I-12