Open Access
Add to BibSonomyAbstract
The need to obtain ocean color essential climate variables (OC-ECVs) using hyperspectral technology has gained increased interest in recent years. Assessing ocean color on a large scale in high latitude environments using satellite remote sensing is constrained by polar environmental conditions. Nevertheless, on a small scale we can assess ocean color using above-water and in-water remote sensing. Unfortunately, above-water remote sensing can only determine apparent optical properties leaving the sea surface and is susceptible to near surface environmental conditions for example sky and sunglint. Consequently, we have to rely on accurate in-water remote sensing as it can provide both synoptic inherent and apparent optical properties of seawater. We use normalized water leaving radiance LWN or the equivalent remote sensing reflectance RRS from 27 stations to compare the differences in above-water and in-water OC-ECVs. Analysis of above-water and in-water RRS spectra provided very good match-ups (R2 > 0.97, MSE<1.8*10−7) for all stations. The unbiased percent differences (UPD) between above-water and in-water approaches were determined at common OC-ECVs spectral bands (410, 440, 490, 510 and 555) nm and the classic band ratio (490/555) nm. The spectral average UPD ranged (5 – 110) % and band ratio UPD ranged (0 – 12) %, the latter showing that the 5% uncertainty threshold for ocean color radiometric products is attainable. UPD analysis of these stations West of Greenland, Labrador Sea, Denmark Strait and West of Iceland also suggests that the differences observed are likely a result of environmental and instrumental perturbations.
©2013 Optical Society of America
1. Introduction
Greenland sustains a very fragile ecosystem which is a source of livelihood for the Nuuks [1]. As the ice melts coastal and marine environments undergo rapid transition which means urgent need for understanding and managing the Arctic marine environment. It is thus our goal to assess the effect of these changes focusing on the remote sensing of aquatic optical properties and their interactions with polar biogeochemistry. Shipborne hyperspectral sensing of polar oceans is a growing field and holds the potential to carry out extensive or complementary optical monitoring of the Greenland area. Obtaining valid and high quality measurements will allow accurate bio-optical modeling and assist in improving present models. However, cloud cover, fog and fjordal features hinders satellite and/or airborne remote sensing in sub-polar to polar waters [2, 3].
To determine trustworthy and manageable ocean color essential climate variables (OC-ECVs) it is necessary to apply quality control. Sea-truth radiometric quantities, for satellite validation, can be obtained using above-water or in-water radiometry. However, these methods are not without error due to different sensor calibration, deployment approaches, and quality control [4]. Above-water radiometry is sensitive to meteorological conditions and sea state which makes it obligatory to perform effective glint correction [5, 6]. In-water profile radiometric quantities are used to approximate light just below the sea surface, but they are influenced by illumination changes with depth, sensor deployment, self-shading, ship shadow, air bubbles and wave-induced perturbations [4, 7]. Additionally, in polar regions there are more external effects; the constant low solar elevation, ice reflectance contribution to remote sensing reflectance, colored dissolved organic matter (CDOM) absorption decoupling with chlorophyll a abundance, and continued presence of cloud and fog [3]. These external effects can result in over or underestimation of OC-ECVs.
In this report we compare reflectance from above-water and in-water radiometry from sub-polar and polar North Atlantic waters. The objective is to identify which glint correction approach is useful for above-water radiometry to match in-water radiometry. We further look at the unbiased percent difference to determine how above-water and in-water observations vary at different sites. The spectra shape of above-water and in-water OC-ECVs are used to classify the water bodies into Case 1 or Case 2, it allows us to evaluate how this (spectra based) classification method may vary for spectra from different platforms.
2. Materials and methods
2.1 Study site
Above-water and in-water hyperspectral radiometric observations were conducted during RV Maria S. Merian field campaign 21 leg 3 between 27 July and 08 August 2012. These observations were performed West of Greenland, starting from Nuuk heading north to Disco Bay and the Uummannaq Fjord then back towards the Labrador Sea proceeding towards Iceland along the Denmark Strait and ending in the West Iceland fjordal system. Figure 1 shows the map highlighting the above-water and in-water stations.
Fig. 1 Above-water (red) and in-water (green) sampled stations during RV Maria S. Merian field campaign 21 leg 3 between 27 July and 08 August 2012.
2.2 Measurement methods and instruments
The steps involved in obtaining optical measurements and auxiliary data for match-up analyses are shown Fig. 2.
Fig. 2 Flowchart showing how ocean color essential climate variables (water leaving radiance – LW, normalized water leaving radiance - LWN, and remote sensing reflectance – RRS) were collected and processed for the comparison task.
2.3 Above-water
A radiometer setup [6] with a RAMSES-ACC hyperspectral cosine irradiance meter for ES (λ) downwelling solar irradiance and two RAMSES-ARC hyperspectral radiance meters Lsfc (θsfc, Φ, λ) is upwelling water leaving radiance and Lsky (θsky, Φ, λ) sky leaving radiance was implemented (TriOS GmbH, Germany). The radiance sensors were positioned at zenith angle, θsfc = 30° and θsky = 150° and at 60° from the ships heading. The zenith angles were slightly changing due to wave induced roll and pitch of the ship. Hyperspectral measurements were collected at 15 minute intervals over a spectral range λ = 320 – 950 nm. These measurements of ES, Lsfc, and Lsky are freely available via the PANGAEA database of the World Data Center for Marine Environmental Sciences [8–10]. Water leaving radiance, LW (θsfc, Φ, λ), and remote sensing reflectance, RRS (θ, Φ, λ) [sr−1] were calculated according to Eq. (1) [11],
where ES (λ) is downwelling solar irradiance, Lsfc (θsfc, Φ, λ) is upwelling radiance, Lsky (θsky, Φ, λ) is the sky radiance, and ρair-sea is the skyglint correction factor. Φ denotes the relative azimuthal angle of the sensor system to the sun which is variable due to ship motions. A Differential Global Positioning System (DGPS) logged the ship’s position and heading automatically at Coordinated Universal Time (UTC). To be able to eliminate the differences introduced by variable measurement conditions we determine normalized water leaving radiance, LWNwhere F0 is the mean solar irradiance at the top of the atmosphere [12].2.4 In-water
A hyperspectral free-falling optical profiler, Profiler II (Satlantic Inc., Canada) was used to measure in-water light profiles over a spectral range λ = 349 – 801 nm. It measured the upwelling radiance, upward and downward irradiance within the water column. A reference irradiance sensor, HyperOCR (Satlantic Inc., Canada) was positioned at an elevated position of the RV. The Profiler II was deployed at the back of the ship and let to drift 30 – 50 m away from the ship to avoid ship perturbations. At each station 2 – 4 casts were performed with the Profiler II. The deployment position was unaffected by ship shadow or superstructure perturbations. The instrument setup, quality control, and data handling is consistent with prior studies [4, 13]. Level (1 - 4) data processing including raw data calibration, tilt angle filtering was performed using the ProSoft Software version 7.7.16 (Satlantic Inc., Canada) with default constants i.e. reflection albedo = 0.043, reflective index = 0.021, refractive index = 1.345 and the mean solar irradiance [12]. The level 4 processed measurements of LW, LWN and RRSare freely available via the PANGAEA database of the World Data Center for Marine Environmental Sciences [14–16].
2.5 Match-up and comparison of above-water and in-water spectra
The above-water platform had an automated 15-minute sampling rate and therefore to obtain measurements in close proximity of in-water observations at time T, we use above water observations at time T - 1 and T + 1 as illustrated in Fig. 3. Each above-water observation (corrected for sky and/or sunglint [5, 17–19]) within the period (T - 1 and T + 1) is individually compared using linear regression with each in-water observation.
Figure 4 shows sample reflectance spectra for 4 in-water casts and a single above-water estimation using different sky and sunglint correction approaches [5, 17–19]. We did also test a new simple and robust glint correction approach [20]. The Kutser et al. [20] corrected spectra were in most cases negative or overcorrected i.e. < 0 sr−1. The Mobley [12] approach uses a constant skyglint correction factor 0.028 and Ruddick et al. [11] correction factor is a product of cloud cover and wind speed. Gould et al. [13] and Lee et al. [5] use a more complicated correction approach which assumes Fresnel reflectance and a residual glint component. Additionally, for comparison purposes we use 5 nm binning.
Fig. 4 Sample plot for RRS (sr−1) of 4 in-water casts and a single above-water estimation using different sky and sun glint correction approaches used in the match-up analysis. Note that the in-water casts only reach ~610 nm as a result of high of light attenuation with depth.
Match-up analysis of above-water and in-water observations utilized linear regression statistical quantities determination coefficient R2 and parameters. In our opinion, none of the approaches is superior to the other nor gives a ‘true’ measurement of the reflectance. We implement a comparing technique called unbiased percent differences (UPD) like other similar studies [4, 13, 21]. To determine the UPD [21] between two approaches A (above-water) and B (in-water) at time t,
where X represents the OC-ECVs e.g. LW, LWN or RRS and λ is any discrete wavelength. The spectral average UPD for N spectral bands (410, 440, 490, 510 and 555) nm, e.g. here will N = 5, is calculated by summing and weighting Eq. (3),Furthermore, we determine band ratio UPD for the classic 490/555 nm,where in our case is the ratio RRS (c = 490)/ RRS (d = 555) from above-water.3. Results and discussion
3.1 Glint correction
Above-water radiometric observations are corrected for glint after Mobley [18], Gould et al. [19], Ruddick et al. [17], and Lee et al. [5]. In Fig. 5 spectra from all 27 stations are presented to show how each correction model performs. Additionally, for comparison a red line is used to indicate zero reflectance, so as to determine if a correction model does overcorrect for glint. It is a challenge to determine if a glint correction model under corrects, but we can see overcorrection i.e. corrected spectra < 0 sr−1. Therefore, to rank the glint correction models we (i) did a visual inspection of Fig. 5 whereby we counted the number of spectra < 0 sr−1 for each model, and then (ii) fitted a straight line between each above-water observation (corrected for sky and/or sunglint [5, 17–19]) and each in-water observation. Based on our criterion we ranked the correction models; Gould (Best) > Lee > Ruddick > Mobley (Worst).
3.2 Match-up analysis
To perform the match-up analysis we used the Gould et al. [19] correction algorithm as it produced the best linear regression statistics summarized in Table 1. We assume that during the campaign we have uniform solar distribution as we observed 70 – 100% cloud cover at 24 stations of the 27 stations (Table 1).
Table 1. Best statistical match-up results for above-water vs. in-water RRS. N is the number of matching wavebands for both above-water and in-water observations. MSE is the mean square error [1/N(y (i)-x (i))2]. The solar zenith was computed using the solar position algorithm [22]. Cloud cover [23] was approximated from visual inspection and absolute wind speed was obtained from ship weather station.
3.3 Unbiased percent difference (UPD)
Hooker et al. [21] propose a comparison strategy that can be implemented to determine sources of uncertainty and differences between two methods, for instance here above-water and in-water radiometry. In this strategy, it is assumed a single parameter obtained from two approaches with none of them taken as the best approach, one can use UPD to see how measurements of this parameter vary. Whereby time and space frames are assumed to be similar. We apply this strategy after the match-up analysis (Table 1) and Table 2 summarizes our findings at each station. Here we did not compute UPD for the in-water methods because from the 2 – 4 available casts we retrieve the in-water OC-ECV from the cast with the best matching above-water OC-ECV as in Table 1.
Table 2. Summary of unbiased percent differences (UPD) between above-water and in-water measured RRS at chosen discrete wavelengths (410, 440, 490, 510, and 555) nm and their average spectral UPD with the standard deviation and the band ratio (490/555) nm UPD.
At each station it is clear that the UPD varies (Table 2) in some cases very high/bad e.g. average spectral UPD at station 506 (110.4%) or very low/best at station 535 (5.4%). As for the discrete wavelength UPD there is also great variability at each station and at each wavelength. Uncertainties can be expressed, but not exclusively by,
i. Sensor stability and calibration methods – In this study we use radiometers from different commercial suppliers Satlantic Inc. (Canada) and TriOS (Germany). There is therefore a calibration and sensor stability uncertainty from each vendor and the instrument overtime. Future works aim to have in-house calibration exercises so as to understand the extent of uncertainty.
ii. Environmental perturbations – Changes in meteorological, sea surface conditions and local optically active seawater properties have been known to introduce uncertainties [13, 21]. In Table 1 we show that in most case we have overcast - fully overcast at most stations which would suggest uniform light. In one case station 536 (Table 1 and Table 2) were we have fully overcast skies – 100% and low winds speed ~6 m/s the average spectral UPD is 5.4% and band ratio UPD is 0.67% and vice versa for station 506 and with high UPD values. However, station 504 has fully overcast skies – 100% and high winds speed ~11 m/s but lower UPD value < 50%. Solar zenith angle for station 506 = 79° and 504 = 59° can be assumed to be the influencing environmental perturbation rather than wind speed in when looking at station 506 and 536.
iii. Data processing – The in-water measurements were processed using ProSoft Software version 7.7.16 (Satlantic Inc., Canada) and as for the above-water measurement only the glint correction was different. Eliminating the glint correction models is justified in Fig. 4, whereby we show that glint correction can result in negative spectra. Some errors can be introduced as we extrapolate in-water radiometric quantities to the sea surface [24].
3.4 Ocean color product
The classification of water bodies [25] into Case 1 and Case 2 was also implemented. Case 2 water is expected to be 443/555 < 443/510 < 490/555 < 490/510 < 1 [26]. Kowalczuk et al. [13] tested this classification in an effort to see how their above-water and in-water observations differed. In their study the two measurements were consistent. Figure 6 shows all the spectra from above-water and in-water observations with the classification summarized in Table 3. Case 1 water will have lower RRS with maxima more in the blue whilst Case 2 water will have enhanced RRS in the green wavelengths compared to the blue wavelengths driven by high levels of seawater color producing agents; colored dissolved organic matter, phytoplankton and minerals [26].
Fig. 6 All spectra above-water (red) and in-water (blue) observations used distinguish station water type into case 1 or case 2.
Table 3. Ouillon and Petrenko [26] Case 1 and Case 2 water classification method applied to above-water and in-water observations.
Ouillon and Petrenko [26] Case 1 and Case 2 water classification method applied to above-water and in-water observations agreed very well at all stations except for one, station 530. Above-water reflectance classifies the station as Case 1 whilst the in-water reflectance classifies it as Case 2. However, this classic method of identifying water bodies as Case 1 and Case 2 can be ambiguous and a challenge [27]. The method cannot be fully relied on as no generic (for all oceanic regions) set of parameters have been suggested in literature for example, in Table 3, looking at station 530 there is a variation in the classification even though the UPD values are low < 34%. It is difficult to distinguish the cause of such a difference as all other stations agree well.
4. Conclusions and outlook
We evaluate OC-ECVs namely RRS obtained from two different platforms above-water (shipborne) and in-water (free-fall profiler). These two platforms are equipped with hyperspectral irradiance and radiance radiometer sensors. In-water radiometric quantities are measured as a function of depth and extrapolated to the sea surface. The extrapolated sea surface (in-water) RRS is compared with above-water estimated RRS to determine UPD. We determine UPDs at common OC-ECVs spectral bands (410, 440, 490, 510 and 555) nm and the classic band ratio (490/555) nm. The spectral average UPD ranged (5 – 110) % and band ratio UPD ranged (0 – 12) %, the latter showing that the 5% uncertainty threshold for ocean color radiometric products is attainable. Using band ratioing we compress the reflectance from above and in-water into Case 1 or Case 2 water body classes. The resulting above-water and in-water Case 1/Case 2 classes are assessed. The comparison approach presented here requires iterative and extensive information thereby also improving instrument deployment, uncertainty assessment, and bio-optical model sensitivity.
Using underwater technology to determine RRS provides a validation procedure for above-water remote sensing. It eliminates the major external effects namely sunglint and fog effects. However, the in-water free falling deployment does have its drawbacks, namely the tilt changes driven by upper layer turbulence, which therefore have to be accounted for in estimating error sources [7]. Nevertheless, a better understanding of uncertainties in above-water optical sensing can be achieved by optical closure with in-water measurements and solving radiative transfer equations. This optical closure is best done using accurate and comprehensive ancillary measurements including absorption, attenuation and scattering properties also known as Inherent Optical Properties (IOPs). The IOPs will then be used as input into radiative transfer models e.g. Hydrolight (Sequoia Scientific, Inc, USA). Without these IOPs, the best way to evaluate above-water RRS is to use in-water derived RRS utilizing free-fall hyperspectral irradiance and radiance sensors.
The plague of sky and sunglint is still to be resolved and in this study we present four approaches. We observe that the best approach for the investigated region should correct the water leaving radiance for skyglint and sea surface reflected glint. Future works need to further evaluate how best to eliminate glint contamination using radiative transfer modeled information integrated with optical closure. It will also be important to compare near surface observations with satellite derived observations.
Acknowledgments
The authors would like to thank Rohan Henkel, Lars Holinde, Daniela Meier, Daniela Voß, cruise chief scientist Allan Cembella (Alfred-Wegener-Institute for Polar and Marine Research), and crew of the RV Maria S. Merian for making field campaign a success. We also extent our gratitude to David G. Bowers, Guiseppe Zibordi, Dirk Aurin and the two anonymous reviewers for their suggestions and feedback.
References and links
1. A. Calbet, K. Riisgaard, E. Saiz, S. Zamora, C. Stedmon, and T. Nielsen, “Phytoplankton growth and microzooplankton grazing along a sub-Arctic fjord (Godthabsfjord, west Greenland),” Mar. Ecol. Prog. Ser. 442, 11–22 (2011). [CrossRef]
2. S. Bélanger, J. K. Ehn, and M. Babin, “Impact of sea ice on the retrieval of water-leaving reflectance, chlorophyll a concentration and inherent optical properties from satellite ocean color data,” Remote Sens. Environ. 111(1), 51–68 (2007). [CrossRef]
3. H. M. Dierssen and R. C. Smith, “Bio-optical properties and remote sensing ocean color algorithms for Antarctic Peninsula waters,” J. Geophys. Res. 105(C11), 26301–26312 (2000). [CrossRef]
4. G. Zibordi, J. F. Berthon, F. Mélin, and D. D'Alimonte, “Cross-site consistent in situ measurements for satellite ocean color applications: The BiOMaP radiometric dataset,” Remote Sens. Environ. 115(8), 2104–2115 (2011). [CrossRef]
5. Z. Lee, Y.-H. Ahn, C. Mobley, and R. Arnone, “Removal of surface-reflected light for the measurement of remote-sensing reflectance from an above-surface platform,” Opt. Express 18(25), 26313–26324 (2010). [CrossRef] [PubMed]
6. S. P. Garaba, J. Schulz, M. R. Wernand, and O. Zielinski, “Sunglint detection for unmanned and automated platforms,” Sensors (Basel) 12(9), 12545–12561 (2012). [CrossRef] [PubMed]
7. T. Suresh, M. Talaulikar, E. Desa, S. G. Prabhu Matondkar, T. S. Kumar, and A. Lotlikar, “A simple method to minimize orientation effects in a profiling radiometer,” Mar. Geod. 35(4), 441–454 (2012). [CrossRef]
8. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Spectral sky radiance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810739.
9. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Downward irradiance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810741.
10. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Upward radiance radiance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810740.
11. J. L. Mueller, C. Davis, R. Arnone, R. Frouin, K. Carder, Z. P. Lee, R. G. Steward, S. Hooker, C. D. Mobley, and S. McLean, “Above-water radiance and remote sensing reflectance measurement and analysis protocols,” in Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 4, NASA/TM-2003–21621, J. L. Mueller, G. S. Fargion, and C. R. McClain eds. (2003), pp. 21–31.
12. H. Neckel and D. Labs, “The solar radiation between 3300 and 12500 Å,” Sol. Phys. 90(2), 205–258 (1984). [CrossRef]
13. P. Kowalczuk, M. J. Durako, W. J. Cooper, D. Wells, and J. J. Souza, “Comparison of radiometric quantities measured in water, above water and derived from seaWiFS imagery in the South Atlantic Bight, North Carolina, USA,” Cont. Shelf Res. 26(19), 2433–2453 (2006). [CrossRef]
14. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Normalized water leaving radiance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810858.
15. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Water leaving radiance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810857.
16. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Remote sensing reflectance during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810742.
17. K. G. Ruddick, V. De Cauwer, Y. J. Park, and G. Moore, “Seaborne measurements of near infrared water-leaving reflectance: The similarity spectrum for turbid waters,” Limnol. Oceanogr. 51(2), 1167–1179 (2006). [CrossRef]
18. C. D. Mobley, “Estimation of the remote-sensing reflectance from above-surface measurements,” Appl. Opt. 38(36), 7442–7455 (1999). [CrossRef] [PubMed]
19. R. W. Gould, R. A. Arnone, and M. Sydor, “Absorption, scattering, and, remote-sensing reflectance relationships in coastal waters: Testing a new inversion algorithm,” J. Coast. Res. 17, 328–341 (2001).
20. T. Kutser, E. Vahtmäe, B. Paavel, and T. Kauer, “Removing glint effects from field radiometry data measured in optically complex coastal and inland waters,” Remote Sens. Environ. 133, 85–89 (2013). [CrossRef]
21. S. B. Hooker, G. Lazin, G. Zibordi, and S. McLean, “An evaluation of above- and in-water methods for determining water-leaving radiances,” J. Atmos. Ocean. Technol. 19(4), 486–515 (2002). [CrossRef]
22. I. Reda and A. Andreas, “Solar position algorithm for solar radiation applications,” Sol. Energy 76(5), 577–589 (2004). [CrossRef]
23. O. Zielinski, D. Voß, D. Meier, R. Henkel, L. Holinde, S. P. Garaba, and A. Cembella, “Cloud cover observations during Maria S. Merian cruise MSM21/3 (ARCHEMHAB)” (2013), http://doi.pangaea.de/10.1594/PANGAEA.810649.
24. Z. Lee, N. Pahlevan, Y.-H. Ahn, S. Greb, and D. O’Donnell, “Robust approach to directly measuring water-leaving radiance in the field,” Appl. Opt. 52(8), 1693–1701 (2013). [CrossRef] [PubMed]
25. A. Morel and L. Prieur, “Analysis of variations in ocean color,” Limnol. Oceanogr. 22(4), 709–722 (1977). [CrossRef]
26. S. Ouillon and A. Petrenko, “Above-water measurements of reflectance and chlorophyll-a algorithms in the Gulf of Lions, NW Mediterranean Sea,” Opt. Express 13(7), 2531–2548 (2005). [CrossRef] [PubMed]
27. C. D. Mobley, D. Stramski, W. P. Bissett, and E. Boss, “Optical modeling of ocean waters: Is the Case 1 - Case 2 classification still useful?” Oceanography (Wash. D.C.) 17(2), 60–67 (2004). [CrossRef]
