Open Access
Add to BibSonomyAbstract
Retrieval of chlorophyll fluorescence magnitude using Fluorescence Height algorithms in coastal waters is more complicated than in the open ocean because of the strong deviations of elastic reflectance within the fluorescence band from the derived fluorescence baseline. We use results of our recently established relationship between fluorescence magnitude and concentrations of water constituents together with extensive HYDROLIGHT simulations, field and satellite data to analyze the performance and retrieval limitations of MODIS and MERIS FLH algorithms in the variety of coastal waters and to examine improvements for spectral band selection suitable for future sensors.
©2008 Optical Society of America
1. Introduction
Quantification of the sun-induced chlorophyll fluorescence component in the remote sensing reflectance measured by satellite sensors is necessary for a number of water quality applications. These include: improved retrievals of chlorophyll concentration [Chl] and quantum yield using both operational and more recently developed algorithms [1–4], detection and characterization of harmful algae blooms [5], improvement of [Chl] retrieval using measurements of the NIR spectral peak [6,7], and the analysis of photosynthetic and transspectral processes [8,9].
It is well recognized that operational Fluorescence Line Height (FLH) algorithms [3,10] based on the measurements of reflectance at three wavelengths in the fluorescence band are sufficient for fluorescence retrieval in the open ocean where atmospheric correction algorithms work well and elastic reflectance (reflectance excluding the fluorescence component) in the fluorescence band is well approximated by the baseline curve due to the relatively weak elastic scattering signal which depends on [Chl] alone. Clearly, this is not the case in coastal areas. While recent advances in atmospheric corrections suitable for coastal waters [11–13] make it possible to better isolate the water leaving radiances in the NIR spectral region, application of the FLH algorithms in the coastal waters is still significantly complicated by a peak in the underlying elastic reflectance which spectrally overlaps and contaminates any fluorescence retrieval. The structure and nature of this NIR peak is the result of a modulation of the particulate elastic spectrum (from both algal and non algal particles) by the combined phytoplankton and water absorption spectra, where the confluence of the decreasing phytoplankton absorption and the increasing absorption of water with wavelength results in a local absorption minimum. This absorption minimum leads to the maximum in the reflectance spectra which are inversely related to the total absorption [6, 14–16].
To compensate for the effects of this overlap of fluorescence and elastic spectra, and improve the operational FLH algorithms for coastal waters, it is clear that suitable models which attempt to take into account the larger impact of the spectral variation of the underlying elastic reflectance peak must be developed. These can then be coupled with studies on new optimized positioning of spectral channels to improve fluorescence retrievals and reduce errors which cannot be totally compensated by modeling. This work is of interest not only for current satellite sensors but also for the development of the new ones, as well as for the retrieval of fluorescence from airborne multispectral measurements, and even for in situ measurements where the advanced understanding can be used together with greater spectral availability to improve retrieval accuracy. These algorithms should be tested with simulated and field data of fluorescence retrieval including polarization discrimination technique [15,16] and inversion algorithms [17].
The present work stems from the preliminary evaluation we made of MODIS and MERIS FLH algorithms in coastal waters [18] based on synthetic datasets. This study was performed by modeling fluorescence as a Gaussian line shape (centered at 685 nm) with a peak magnitude established using an expression from Gower et al. [19] which is a function of [Chl] and then superimposing this Gaussian fluorescence feature onto 275 reflectances simulated by Lee [20] using HYDROLIGHT [21,22] for a range of [Chl]=1–30 mg/m3. We found that errors in the simulated fluorescence retrievals can exceed 100% for high [Chl] values. However, these estimates could not be considered sufficiently accurate or comprehensive since the IOCCG dataset had a relatively low spectral resolution of 10 nm and the concentrations of minerals and [Chl] did not cover the whole range of interest for coastal waters. To remedy these deficiencies we expanded, in Part 1 [23] of this work, the synthetic datasets relevant to this study to more detailed and comprehensive datasets, together with additional improvements in the parameterization of fluorescence peak magnitudes using data obtained from field measurements. In particular, it was shown that for coastal waters, the fluorescence amplitude can be substantially lower than in the open ocean because of the strong attenuation of light in the fluorescence excitation zone due to the absorption of phytoplankton, CDOM and minerals, non-algal particulate scattering, and decreasing specific chlorophyll absorption for higher [Chl] values. By comparing simulated reflectances with those observed in the field and by direct comparison of simulated fluorescence amplitudes with our fluorescence retrievals and fluorescence measurements in the other water areas we found that actual fluorescence quantum yield does not change significantly even over a wide variety of water conditions and is generally close to η=1%. This coincides with previous estimations for open ocean waters [2] and measurements near the water surface [24,25]. We also found that fluorescence amplitude typically represents 10–30 % of the total NIR reflectance if concentration of nonalgal particles does not exceed about 2 mg/l. For higher concentrations of nonalgal particles Cnap>5 mg/l, the contribution of fluorescence is very low and its detection does not appear practical because of added light attenuation in the excitation zone as well as the much higher component of elastic reflectance. Using these results, in Part 1, we proposed and tested parameterizations which connect the fluorescence amplitude directly to concentrations of the typical primary water constituents. Additionally, in the introduction to Part 1, more details are given on the studies of NIR reflectance peak and retrieval of fluorescence in coastal waters.
In the very recent paper [26] which was published almost simultaneously with [23], a similar conclusion that in coastal waters the fluorescence component can be substantially attenuated by nonalgal components in the water was stated; and synthetic testing of the MODIS FLH algorithms concluded that caution is required in the interpretation of FLH signals. While these results qualitatively agree with some of our results reported in [23] and with further results in the present paper, significant quantitative differences are noted: in [26] the fluorescence quantum yield in HYDROLIGHT simulations was assumed η0=2% which corresponds to the actual value η=η0/Q*a=4% if the reabsorption coefficient in the algae cells Q*a=0.5 is taken into account. As a consequence fluorescence amplitudes [26] are several times higher than the values retrieved from field measurements [23] and satellite data [19] which were shown [23] to be consistent with efficiencies η near 1% (η0=0.5% in HYDROLIGHT). Furthermore, in HYDROLIGHT simulations [26], the backscattering ratio for mineral particles was taken as b̃b nap=0.04 in comparison with average values around b̃b nap=0.02 [27,28], which was also assumed in our simulations [23]. These values resulted in a strong overestimation of backscattering effects and elastic reflectances with the increase of the mineral concentrations.
In the present paper we use our synthetic datasets, field data and established parameterization for the fluorescence amplitude [23] as well as satellite imagery to expand our earlier analysis of the fluorescence retrieval [18] including MODIS and MERIS FLH algorithms and to evaluate possibilities of retrieval improvement of the fluorescence component in the remote sensing reflectance using different band combinations for various conditions in coastal waters.
2. Main features of the synthetic and field data
Approximately 1500 reflectance spectra with and without fluorescence were simulated using HYDROLIGHT with 1 nm resolution for conditions typical of coastal waters: [Chl]=1–100 mg/m3, CDOM absorption at 400 nm 0–5 m-1, concentrations of nonalgal particles in the range 0–100 g/m3 and 5 different specific chlorophyll absorption shapes. All details and assumptions used for the simulation of the water parameters are given in Part 1 [23]. They were based on the findings of many authors [27–34] for IOP characteristics, and were similar to the assumptions used in the construction of the Lee IOCCG datasets [20]. Solar input was simulated using the Gregg and Carder [35] model with a cloud-free sky. The specific chlorophyll absorption spectra used in our simulations are critical for an understanding of our further simulations of FLH and are plotted in Fig.1.
The different datasets we use are defined as follows.
Dataset 1 500 runs were executed with randomly chosen [Chl]=1–100 mg/m3, concentration of nonalgal particles Cnap= 0–1 g/m3, absorption of CDOM at 400 nm ay(400)=0–5 m-1. A set of specific absorptions was taken according to Ciotti et al. [36] as a sum of specific absorptions of microplankton and picoplankton with different weights Sf where Sf=0.1–0.5 according to the expression (1)
These spectra are shown in Fig. 1. The chlorophyll absorption was considered proportional to [Chl] as is often assumed [21, 37] and given by achl(λ)=[Chl]·a*chl(λ).
Dataset 2 repeated dataset 1 but with concentrations of nonalgal particles in the range Cnap=1–10 g/m3.
Dataset 3 repeated dataset 1 but with concentrations of nonalgal particles in the range Cnap=10–100 g/m3.
All datasets were created first for reflectances without fluorescence and afterwards, the simulations were repeated with fluorescence assuming fluorescence quantum yield η0=0.5% This HYDROLIGHT quantum yield corresponds to the field quantum yield η=1% because HYDROLIGHT does not take into account the reabsorption coefficient for which we assumed a constant value Q*a=0.5 for our range of [Chl] (see [23] for details). All reflectances were simulated for the sun zenith angle θi=30° and nadir viewing.
Field cruise data and its processing were detailed in Part 1 [23]. Measurements were conducted in Chesapeake Bay in the summer of 2005. These included reflectance measurements just below the water surface, absorption and attenuation measurements using WET Labs ac-s instrument, as well as collection of water samples from which [Chl], CDOM absorption and TSS concentration were measured [38]. The range of parameters were [Chl]=9–354 mg/m3, total absorption at 400 0 nm a(400)=1.4–10.8 m-1, and concentrations of TSS=7–64.8 mg/l.
3. Basic relationships
Retrieval of fluorescence from hyperspectral reflectance spectra is possible using inversion models [17]. However, with only a few bands available on current MODIS and MERIS satellite sensors, Fluorescence Height over Baseline algorithms remain the main approach to estimate fluorescence magnitude [3,10,19]. Three bands on each of these sensors are used to determine FLH. The outer bands are meant to estimate the baseline or elastic reflectance (mostly outside the fluorescence spectrum) while the middle band determines the height. The baseline approach is graphically shown in Fig. 2(a). FLH is defined [3] as
where L 1, L 2 and L 3 are radiances at appropriate wavelengths λ1, λ2, λ3. On the satellite sensors these wavelengths are at the centers of the bands with bandwidths 7–10 nm. The central band is chosen usually to the left of 685 nm to avoid oxygen absorption at 687 nm [3]. And since fluorescence is not equal to zero at λ1 and λ3, only part of the fluorescence magnitude can be captured by the FLH approach, even in an ideal case and related measurements should be scaled appropriately. The bands used are 667, 678, 746 nm for MODIS and 665, 681, 709 nm for MERIS [19].
Fig. 2. (a) Fluorescence height over baseline, (b) Overlapping of fluorescence and elastic radiance peaks in NIR for two [Chl] values.
In case 1 waters, the FLH algorithms work reasonably well because elastic radiance for low [Chl]<10 mg/m3 is a close approximation to the baseline between λ1 and λ3 [4, 19] as can be seen in one of simulated examples in Fig. 2(b). There the lower trace for [Chl]=5.9 mg/m3 shows the elastic reflectance and the upper trace the total reflectance including fluorescence. As mentioned above, because of the choice of spectral bands, FLH does not capture the full fluorescence amplitude, so for MODIS in case 1 waters retrieved FLH values need to be multiplied by a correction coefficient k≈1.75 [19]. With increasing [Chl]=22.5 mg/m3 the peak in elastic radiance described earlier is also observed in the same wavelength range and overlaps the fluorescence spectrum. The elastic spectrum clearly deviates more from a simple baseline, which obviously affects FLH retrieval accuracy.
We found [23] that fluorescence amplitude can become significantly smaller in the presence of CDOM and minerals due to the increased attenuation in the excitation and emission zones and this makes fluorescence estimation on top of the elastic radiances even less reliable. Thus, expression (3) below was found to be a good approximation for the average fluorescence values with the specific absorptions given in Fig. 1, while expressions (4a, b, c) were subsequently found to be good approximations for a low mineral case (Cnap=0–1 g/m3), an average mineral case (Cnap=1–10 g/m3) and a high mineral case (Cnap=10–100 g/m3) with average ay(400)=2.5 m-1 and averageCnap respectively Cnap=0.5g/m3,Cnap=5 g/m3 and Cnap=50 g/m3 substituted ay(400) and Cnap in expression (3).
In expressions (3) and (4) the fluorescence amplitude Fl at 685 nm is given in radiance units (Wm-2sr-1 µm-1) and an equivalent fluorescence component in remote sensing reflectance can be obtained as a ratio of Fl and the irradiance at 685 nm (Ed=1.1 Wm-2 nm-1).
In the work reported below, we analyze the performance of FLH algorithms for both MODIS and MERIS using our synthesized datasets. However, rather than using the HYDROLIGHT code to simulate the fluorescence feature in the synthetic data, we found it more effective to superimpose the fluorescence shape, with its magnitude determined from Eq. (4), on top of HYDROLIGHT derived elastic radiances, since Eq. (4) showed to be good approximations for fluorescence magnitudes. These results were then compared with the values retrieved using FLH algorithms. To test the effect of the specific fluorescence spectral shape superimposed, we compared fluorescence shapes obtained in the lab using argon laser illuminated algae, with the traditional Gaussian one which used in the HYDROLIGHT simulations. While small differences in the spectral shape are found to exist at the red end of the spectrum, differences are too small to be an issue with current satellite spectral resolutions about 10 nm.
4. Analysis of the Fluorescence Height algorithms on the synthetic datasets
4.1 Results for low mineral particle concentrations
As a first step we estimate the impact of various specific absorption shapes on the performance of FLH algorithms. Elastic remote sensing reflectances from dataset 1, with concentrations of nonalgal particles Cnap<1mg/m3 were used in these simulations with the absorption shapes shown in Fig. 1. Remote sensing reflectances were multiplied by the downwelling irradiance spectra near the surface and fluorescence was superimposed on top of the upwelling radiances according to expression (4a). Radiances L 1, L 2 and L 3 for FLH estimation from Eq. (2) were calculated taking into account the Gaussian shape transmittance of MODIS and MERIS bands, and with the bandwidths in accordance with sensor specifications. We considered separately results for low Cnap=0–1 g/m3, average Cnap=1–10 g/m3 and high Cnap=10–100 g/m3, since, as will be shown below, the actual concentrations of nonalgal particles are crucial in determining the performance of FLH algorithms.
Superimposed and retrieved FLH for MODIS and MERIS band sets and reflectances from dataset 1 are shown in Fig. 3(a,b). Specific absorptions for dataset 1 (Fig. 1) [36] have relatively small variability in the NIR zone and some of them with low weight Sf have lower average values which can be more typical of waters with high [Chl] [39]. Results for both MODIS and MERIS are strongly dependent on absorption shapes. As shown in the inset of Fig. 3(a) this dependence becomes obvious if FLH values are multiplied by the correction coefficient k=1..75. While some correlation between FLH and [Chl] seems to occur at low [Chl]<10 mg/m3, after applying an appropriate amplitude correction coefficient, significant variability in the results is clearly seen.
Fig. 3. Performance of MODIS (a) and MERIS (b) FLH algorithms with reflectances simulated with specific absorptions Fig. 1. Solid line – superimposed fluorescence amplitude according to expr. (4a), signs – retrieved FLH. Different shapes of signs correspond to different specific chlorophyll absorption spectra. Inset in (a): same data but with correction coefficient k=1.75 applied.
We found that significant improvement can be achieved by using a 681 nm band in MODIS algorithm instead of the 678 nm when applied to the data from dataset 1. Results are shown in Fig. 4(a) where after applying an amplitude correction coefficient k=1.6, errors less than ±30% in retrieved FLH can be expected. The MERIS combination of 665, 681, 753 nm bands gave significantly higher errors but with the center band shifted to 685 nm and a correction coefficient of k=1.1 applied, errors were reduced to the range ±10% even for high [Chl] values as shown in Fig. 4(b). In our analysis, we focused primarily on optimizing the peak central band which is the band most sensitive to variations in spectral location. Optimization of the central band position was obtained by the minimization of the FLH retrieval errors for available baseline bands of both MODIS and MERIS.
As was mentioned before, in our datasets total phytoplankton absorption was considered proportional to [Chl] with the specific absorption spectra a*Chl(λ) constant for [Chl] =1–100 mg/m3. Another approach [2,39] is to consider specific absorption decreasing with [Chl]. This case was modeled by using specific absorptions from Fig. 1, with constant weights, Sf, applied for small [Chl] ranges. Thus Sf=0.5 for 0<[Chl]<20, Sf=0.4 for 20<[Chl]<40, Sf=0.3 for 40<[Chl]<60, Sf=0.2 for 60<[Chl]<80 and Sf=0.1 for 80<[Chl]<100 were used. In this case as shown in Fig. 5 FLH retrieval is quite accurate with a k=1.6 applied.
Fig. 4. Performance of FLH algorithm with the optimized position of the center band (a) MODIS with 667, 681, 748 set, k=1.6 (b) MERIS with 665, 685, 753 nm and k=1.1. Solid line – superimposed fluorescence amplitude according to expr. (4a), points – retrieved FLH.
Fig. 5. Performance of MODIS FLH algorithm with the optimized position of the center band at 681 nm and reflectances simulated for decreasing with [Chl] specific absorption (see text), k=1.6. Solid line – superimposed fluorescence amplitude according to expr. (4a), points – retrieved FLH.
Results very similar to those shown in Fig. 4(a) and Fig. 5 can be achieved with the center band at 684 nm without requiring a correction coefficient as we move directly into the center of the fluorescence peak.
Finally Fig. 6 shows the FLH – [Chl] relationship where all data for reflectances simulated with fluorescence and η=1% are presented for dataset 1 with low concentrations of nonalgal particles Cnap<1 g/m3. In Fig. 6(a) there are HYDROLIGHT simulated values, in Fig. 6(b) there are results of the FLH retrieval using MODIS algorithm. Obviously the FLH – [Chl] relationship is almost lost in the retrieval. Some correlation should exist only for the range [Chl]<10 mg/m3.
Fig. 6. (a) Simulated fluorescence amplitude for dataset 1, (b) Retrieval of FLH from reflectances simulated with fluorescence for dataset 1 using MODIS algorithm.
4.2 Results for average and high mineral particle concentrations
In Part 1 of the paper [23] we found that the relative contribution of the fluorescence component to the reflectance spectra peak in the NIR decreases rapidly with the increasing concentration of nonalgal particles mostly because of the relative increase of the elastic component to the total reflectance. This certainly affects performance of FLH retrieval algorithms. Retrievals using MODIS bands for waters with Cnap<5 g/m3 and Cnap<10 g/m3 were simulated using dataset 2 and results are shown in Fig. 7. Since expressions (4a) and (4b) do not differ significantly, for comparison convenience we still used (4a) to superimpose fluorescence on elastic reflectance. Inset in Fig. 7(a) shows FLH augmented by the factor k=1.75.
Fig. 7. Performance of MODIS FLH algorithms with reflectances simulated with (a) Cnap<5 g/m3 and (b) Cnap<10 g/m3. Solid line – superimposed fluorescence amplitude according to expr. (4a), points – retrieved FLH. Inset in (a): same data but with correction coefficient k=1.75 applied.
Comparing these results with simulations in Fig. 3(a) we see that the spread in retrieved values increases strongly even for Cnap<5 g/m3 and especially for Cnap<10 g/m3 making this algorithm almost ineffective. This can be partially improved by moving the center band closer to the fluorescence maximum. As shown in Fig. 8 with the center band at 684 nm points are better grouped around superimposed fluorescence values.
When the standard fluorescence shape with amplitude obtained from expression (4c) was superimposed on the elastic reflectances from the dataset 3 with high concentration of nonalgal particles Cnap=10–100 g/m3, the FLH was completely erratic for almost all of water compositions being mostly negative for [Chl]>20 mg/m3. This is due to the substantial change of the reflectance spectral shape caused by the relatively strongly increased elastic scattering from minerals and the much smaller relative fluorescence values, which are even further diminished by enhanced mineral absorption. Thus the greatly increased total reflectance is even less sensitive to the much smaller fluorescence component in it, making its impact more difficult to discern. These results for MODIS and MERIS are shown in Fig. 9. Changes in the band positions do not improve retrieval because magnitudes of fluorescence remain small in comparison with elastic reflectance.
Fig. 8. Performance of FLH algorithm with the optimized position of the center band (667, 681, 748 nm set). Solid line – superimposed fluorescence amplitude according to expr. (4a), points – retrieved FLH.
Fig. 9. Performance of MODIS (a) and MERIS (b) FLH algorithms with reflectances simulated for high mineral concentrations Cnap=10–100 g/m3. Solid line – superimposed fluorescence magnitude according to expr. (4c), points – retrieved FLH.
This confirms the conclusion [23] that fluorescence retrieval for high mineral concentrations is very difficult if possible at all.
5. Tests of Fluorescence Height algorithms with field and satellite data
5.1 Field measurements results
To test the FLH algorithm on field data, we show in Fig. 10(a) fluorescence amplitudes retrieved (using the procedure detailed in [23]) by fitting total elastic below surface remote sensing reflectances constructed from in situ absorption and attenuation measurements (ac-s, WET Labs, Inc.) and a superimposed Gaussian fluorescence component into measured in-situ reflectance spectra for 28 stations in Chesapeake Bay. Those values are then compared with the FLH retrieved by standard MODIS and MERIS algorithms applied to the in-situ reflectances. In Fig. 10(b) we compared fitting results with FLH when the 678 nm band for MODIS was replaced by 684 nm. Linear regression was also performed for the MODIS algorithm in Fig. 10(a) and data in Fig. 10(b).
Fig. 10. Test of FLH algorithms on Chesapeake Bay field data: (a) standard MODIS and MERIS algorithms; (b) 667, 684, 748 bands. Red lines – linear fits, (a) r2=0.1, (b) r2=0.36.
There is more of a correlation in Fig. 10(b) which is in contrast to little correlation when using standard algorithms in Fig. 10(a). In addition, Fig 10(b) shows that there is a systematic overestimation of FLH for field data where TSS values were mostly close to 10 g/m3. This is similar to the simulation results with the same bands shown previously in Fig. 8 for Cnap<10 g/m3.
5.2 Results from MODIS satellite data
We also analyzed relationships between FLH and [Chl] using satellite data of MODIS Aqua for August 4, 2005 processed through SeaDAS version 5.0.4. The main area of study was the Chesapeake Bay, but we also included results from the Delaware Bay and the zone along the coast in between these bays. We applied the new SWIR atmospheric correction algorithm which employs 1240 nm and 2130 nm spectral bands [12,13], together with the standard MODIS chlorophyll and FLH algorithms. It should be noticed that in the areas of strong CDOM concentrations, chlorophyll values determined by blue-green ratio algorithm will be overestimated. [Chl] distributions in the area of interest are shown in Fig. 11(a) and relationship between [Chl] and FLH are plotted in Fig. 11(b).
Comparison of Fig. 11(b) with our simulations in section 4 shows that higher values of FLH which occur for low [Chl]<10 mg/m3 could be explained by high mineral concentrations. To test this hypothesis, we applied a turbidity filter condition for a remote sensing reflectance at 670 nm Rrs(670)<0.0025 which is approximately 2 times higher than the value used in the operational MODIS turbid water flag (Rrs(670)<0.0012) [40] and corresponds to a concentration of minerals Cnap<5 g/m3. Applying the turbidity filter eliminated the Delaware Bay as well as some zones along the coast from the dataset. After the removal of areas of high turbidity, the relationship between FLH and [Chl] is shown in Fig. 12(a) which we see agrees quite well with the simulations of Fig. 6(b) where FLH is mostly in the range between -0.1–0.2 Wm-2sr-1 µm-1 and confirms our previous conclusion that MODIS FLH retrieval is not applicable to high turbidity waters.
Fig. 11. (a) [Chl] map for the area of Chesapeake Bay, Delaware Bay and coast between them using MODIS algorithm for August 4, 2005, (b) FLH over [Chl] for this area using MODIS FLH and [Chl] algorithms
On the other hand, if we limit the data to low [Chl]<10 mg/m3 values, we can observe a very good correlation between FLH and [Chl] as can be seen in Fig. 12(b) for [Chl]<4 mg/m3. Multiplying by a magnitude correction coefficient k=1.75, the same as for the open ocean, we can see that the FLH retrieved is in the range of simulated values in Fig. 6(a) for [Chl]≈4 mg/m3. Unfortunately, for higher [Chl] there is no correlation between FLH and [Chl]. Similar slopes of relationships between FLH and [Chl] on simulated data Fig. 6(b) and satellite data Fig. 12(b) for [Chl]<4 mg/m3 confirm that satellite [Chl] algorithm does not have significant errors for this range of concentrations.
Fig. 12. FLH over [Chl] for the area in Fig. 11(a) using MODIS FLH and [Chl] algorithms after applying low mineral condition: a – for all [Chl] values, b – [Chl]<10 mg/m3
Very similar results were obtained for the Gulf of Mexico on September 21, 2006 near Florida coast in Karenia Brevis bloom conditions with [Chl] up to 200 mg/m3 [41]. The [Chl] map and FLH – [Chl] relationship for [Chl]<10 mg/m3 are shown in Fig. 13. Some correlation in Fig. 13(b) also exists for [Chl]<4 mg/m3.
We can also use the satellite data to assess our previous results [23] that quantum efficiency is less variable than commonly believed. In particular, retrieved satellite FLH data using the MODIS algorithm from Fig. 12 were compared with the simulated data for three values of fluorescence quantum efficiency η=0.5, 1 and 2%. It is obvious from these simulations that the retrieved FLH values dependence on η is stronger than on type of species and thus is quite well separated by η. We note that the FLH values in Fig. 14(b) are clearly in the range of the simulated values mostly for η=1% in Fig. 14(a). This confirms our previous conclusion [23] based on simulations and field measurements that fluorescence quantum yield mostly does not change in a very wide range and on average is close to η=1% although lower and higher values are certainly possible for some species or conditions.
Fig. 13. (a) [Chl] map at the Florida west coast in conditions of algae bloom, (b) FLH over [Chl] for that area using MODIS FLH and [Chl] algorithms, [Chl]<10 mg/m3.
Fig. 14. (a) Simulated FLH using MODIS algorithm for dataset 1 and three values of fluorescence quantum efficiency η=0.5, 1 and 2 %, (b) FLH MODIS satellite data for the area from Fig. 11(a) and [Chl]<100 mg/m3.
Further accurate field tests in variable coastal water conditions are needed with simultaneous measurements of water optical properties and reflectances and further fluorescence retrieval to verify optimal band combinations for FLH algorithms.
6. Conclusions
Datasets with more than 1500 reflectances typical for coastal waters with high variability of [Chl], concentrations of CDOM and minerals, specific chlorophyll absorption spectra and spectral distributions of attenuation and scattering together with results of complex measurements in Chesapeake Bay and satellite data were used to analyze algorithms for retrieval of FLH in coastal waters.
The smaller values of fluorescence magnitudes in coastal waters due to attenuation by water components, complicates substantially their estimation using FLH algorithms which otherwise work successfully for the open ocean. Additionally, deviation of elastic reflectance spectra from the baseline in FLH with increasing [Chl] because of increasing changes in chlorophyll absorption in NIR makes FLH retrievals much more challenging.
Fluorescence retrieval is very sensitive to the mineral concentrations and specific chlorophyll absorption spectra. For mineral concentrations Cnap>4–5 g/m3 FLH algorithms are very erratic because of smaller fluorescence and enhanced elastic reflectance values proportional to Cnap.
Analysis of MODIS Aqua satellite images of the Chesapeake Bay and Florida coast of the Gulf of Mexico with [Chl] up to 200 mg/m3 confirmed this result: high false FLH values were observed at very low [Chl] concentrations and disappeared when pixels with high water turbidity (Cnap>5 g/m3) have been eliminated. Remaining data with low water turbidity showed that a relationship between FLH and [Chl] can be observed only for [Chl]<4 mg/m3. For higher [Chl] values it is found that FLH does not change substantially with increasing [Chl].
MERIS FLH retrievals have yet to be carried out, but we can already conjecture, based on our simulations and field data, that good performance of the MERIS algorithm in the coastal waters is unlikely, even for low [Chl] concentrations. While it may be thought that the MERIS center band at 681 nm, and therefore closer to the fluorescence peak at 685 nm, than the MODIS band at 678 nm, should help, larger errors are in fact introduced through the significantly higher baseline which connects the spectra at 665 nm and 709 nm and even makes retrieved FLH negative in many cases.
Generally, both sensors have significant problems in FLH retrieval in coastal waters. Our analysis concludes that the increasing departure of reflectance spectra from a linear baseline in the NIR with increasing [Chl] which is mostly due to the spectral changes in the total absorption is a problem at higher [Chl] values. In addition, the reflectance at 667 nm is very close to the reflectance at 678 nm in the MODIS FLH algorithm and is almost insensitive to the changes in [Chl] and fluorescence magnitude.
Significant improvements can be achieved if the central band is moved to 684 – 685 nm and right band is at 748–753 nm; errors due to the oxygen band at 687 nm should be properly estimated based on the position of the sensor (above water, airborne or satellite) and its spectral resolution.
Comparison of the satellite data with the simulations confirms our previous conclusion that fluorescence quantum yield in coastal waters does not change in a very wide range and on average is close to 1%.
Acknowledgments
We would like to thank Prof. J. Schalles for the data of chlorophyll concentrations and TSS measured from the water samples at Chesapeake Bay during our joint campaign. We are also very grateful to the reviewers of the paper whose comments and suggestions helped to improve it significantly. This research was supported partially by grants from NASA NCC-1- 03009, NOAA NA06OAR4810162 and ONR N00014-08-1-0325.
References and links
1. J. Fischer and U. Kronfeld, “Sun-stimulated chlorophyll fluorescence. 1. Influence of oceanic properties,” Int. J. Remote Sens. 11, 2125–2147 (1990). [CrossRef]
2. M. Babin, A. Morel, and B. Gentili, “Remote sensing of sea surface Sun-induced chlorophyll fluorescence: consequences of natural variations in the optical characteristics of phytoplankton and the quantum yield of chlorophyll a fluorescence,” Int. J. Remote Sens. 17, 2417–2448 (1996). [CrossRef]
3. R.M. Letelier and M.R. Abbott, “An Analysis of Chlorophyll Fluorescence Algorithms for the Moderate Resolution Imaging Spectrometer (MODIS),” Remote Sens. Environ. 58, 215–223 (1996). [CrossRef]
4. Y. Huot, C. A. Brown, and J. J. Cullen, “New algorithms for MODIS sun-induced chlorophyll fluorescence and a comparison with present data products,” Limnol. Oceanogr. Methods 3, 108–130 (2005). [CrossRef]
5. C. Hu, F.E. Muller-Karger, C.J. Taylor, K.L. Carder, C. Kelble, E. Johns, and C.A. Heil, “Red tide detection and tracing using MODIS fluorescence data: A regional example in SW Florida coastal waters,” Remote Sens. Environ. 97, 311–321 (2005). [CrossRef]
6. G. Dall’Olmo, A.A. Gitelson, D.C. Rundquist, B. Leavitt, T. Barrow, and J.C. Holz, “Assessing the potential of SeaWiFS and MODIS for estimating chlorophyll concentration in turbid productive waters using red and near-infrared bands,” Remote Sens. Environ. 96, 176–187 (2005). [CrossRef]
7. S. Ahmed, A. Gilerson, J. Zhou, J. Chowdhary, I. Ioannou, R. Amin, B. Gross, and F. Moshary, “Evaluation of the impact of backscatter spectral characteristics on Chl retrievals in coastal waters,” Proc. SPIE 6406 (2006). [CrossRef]
8. D. Pozdnyakov, A. Lyaskovsky, H. Grassl, and L. Petterson, “Numerical modeling of transspectral processes in natural waters: implications for remote sensing,” Int. J. Remote Sens. 23, No 8, 1581–1607 (2002). [CrossRef]
9. S. R. Laney, R. M. Letelier, and M. R. Abbott, “Parameterizing the natural fluorescence kinetics of Thalassiosira weissflogii,” Limnol. Oceanogr. 50, 1499–1510 (2005). [CrossRef]
10. J.F.R. Gower, R. Doerffer, and G.A. Borstad, “Interpretation of the 685 nm peak in water-leaving radiance spectra in terms of fluorescence, absorption and scattering, and its observation by MERIS,” Int. J. Remote Sens. 20, 1771–1786 (1999). [CrossRef]
11. R.A. Arnone, Z.P. Lee, P. Martinolich, B. Casey, and S.D. Ladner, “Characterizing the optical properties of coastal waters by coupling 1 km and 250 m channels on MODIS – Terra,” in Proc. Ocean Optics XVI, Santa Fe, New Mexico (2002).
12. M. Wang and W. Shi, “Estimation of ocean contribution at the MODIS near-infrared wavelengths along the east coast of the U.S.: Two case studies,” Geophys. Res. Lett. 32, L13606 (2005). [CrossRef]
13. B. Franz, “MODIS land bands for ocean remote sensing applications,” in Proc. Ocean Optics XVIII, Montreal, Canada (2006).
14. J.F. Schalles, Optical Remote Sensing Techniques to Estimate Phytoplankton Chlorophyll a Concentrations in Coastal Waters with Varying Suspended Matter and CDOM Concentrations, in Remote Sensing of Aquatic Coastal Ecosystem Processes: Science and Management Applications, L.L. Richardson and E.F. LeDrew, eds. (Springer, 2006), Chap. 3.
15. S. Ahmed, A. Gilerson, A. Gill, B. M. Gross, F. Moshary, and J. Zhou, “Separation of Fluorescence and Elastic Scattering from Algae in Seawater Using Polarization Discrimination,” Opt. Commun. 235, 23–30 (2004). [CrossRef]
16. A. Gilerson, J. Zhou, M. Oo, J. Chowdhary, B. Gross, F. Moshary, and S. Ahmed, “Retrieval of fluorescence from reflectance spectra of algae in sea water through polarization discrimination: modeling and experiments,” Appl. Opt. 45, 5568–5581 (2006). [CrossRef] [PubMed]
17. C.S. Roesler and M.J. Perry, “In situ phytoplankton absorption, fluorescence emission, and particulate backscattering spectra determined from reflectance,” J. Geophys. Res. 100, C7, 13279–13294 (1995). [CrossRef]
18. S. Ahmed, A. Gilerson, J. Zhou, I. Ioannou, B. Gross, and F. Moshary, “Impact of Apparent Fluorescence Shift on Retrieval Algorithms for Coastal Waters,” in Proc. Ocean Optics XVIII, Montreal, Canada (2006).
19. J.F.R. Gower, L. Brown, and G.A. Borstad, “Observation of chlorophyll fluorescence in west coast waters of Canada using the MODIS satellite sensor,” Can. J. Remote Sens. 30, 17–25 (2004). [CrossRef]
20. Z. P. Lee, http://www.ioccg.org/groups/OCAG_data.html.
21. C. D. Mobley, Light and Water. Radiative Transfer in Natural Waters (Academic Press, New York, 1994).
22. C. D. Mobley and L. K. Sundman, HYDROLIGHT 4.2, Sequoia Scientific, Inc. (2001).
23. A. Gilerson, J. Zhou, S. Hlaing, I. Ioannou, J. Schalles, B. Gross, F. Moshary, and S. Ahmed, “Fluorescence component in the reflectance spectra from coastal waters. Dependence on water composition,” Opt. Express 15, 15702–15721 (2007). [CrossRef] [PubMed]
24. S. Maritorena, A. Morel, and B. Gentili, “Determination of the fluorescence quantum yield by oceanic phytoplankton in their natural habitat,” Appl. Opt. 39, 6725–6737 (2000). [CrossRef]
25. J. R. Morrison, “In situ determination of the quantum yield of phytoplankton chlorophyll a fluorescence: a simple algorithm,observations, and a model,” Limnol. Oceanogr. 48, 618–631 (2003). [CrossRef]
26. D. McKee, A. Cunningham, D. Wright, and L. Hay,“Potential impacts of nonalgal materials on water-leaving Sun induced chlorophyll fluorescence signals in coastal waters,” Appl. Opt. 46, 7720–7729 (2007). [CrossRef] [PubMed]
27. M. S. Twardowski, E. Boss, J. B. Macdonald, W. Scott Pegau, A. H. Barnard, and J. V. Zaneveld, “A model for estimating bulk refractive index from the optical backscattering ratio and the implications for understanding particle composition in case I and case II waters,” J. Geophys. Res. 106, 14129–14142 (2001). [CrossRef]
28. A. Gilerson, J. Zhou, and R. Fortich, “Particulate Scattering in Coastal Waters: Chesapeake Bay Study,” Sea Technology 48, 15–18 (2007).
29. R. Pope and E. Fry, “Absorption spectrum (380 – 700 nm) of pure waters: II. Integrating cavity measurements,” Appl. Opt. 36, 8710–8723 (1997). [CrossRef]
30. R.P. Bukata, J.H. Jerome, K.Y. Kondratyev, and D.V. Pozdnyakov, Optical Properties and Remote Sensing of Inland and Coastal Waters (CRC Press, 1995).
31. D. Stramski., A. Bricaud, and A. Morel, “Modeling the inherent optical properties of the ocean based on the detailed composition of the planktonic community,” Appl. Opt. 40, 2929–2945 (2001). [CrossRef]
32. A. Morel, “Optical properties of pure water and pure sea water,” in Optical Aspects of Oceanography, N.G. Jerlov and E. S. Nielsen. eds. (Academic, New York, 1974).
33. M. Sydor and R. A. Arnone, “Effect of suspended particulate and dissolved organic matter on remote sensing of coastal and riverine waters,” Appl. Opt. 36, 6905–6912 (1997). [CrossRef]
34. K.J. Voss, “A spectral model of the beam attenuation coefficient in the ocean and coastal areas,” Limnol. Oceanogr. 37, 501–509 (1992). [CrossRef]
35. W.W. Gregg and K.L. Carder, “A simple spectral solar irradiance model for cloudless maritime atmospheres,” Limnol. Oceanogr. 35, 1657–1675 (1990). [CrossRef]
36. A. M. Ciotti, M. R. Lewis, and J. J. Cullen, “Assessment of the relationships between dominant cell size in natural phytoplankton communities and the spectral shape of the absorption coefficient,” Limnol. Oceanogr. 47, 404–417 (2002). [CrossRef]
37. P. Gege and A. Albert, “A tool for inverse modeling of spectral measurements in deep and shallow waters” in Remote Sensing of Aquatic Coastal Ecosystem Processes: Science and Management Applications, L.L. Richardson and E.F. LeDrew, eds. (Springer, 2006), Chap. 4. [CrossRef]
38. A.A. Gitelson, J.F. Schalles, and C.M. Hladik, “Remote chlorophyll-a retrieval in turbid, productive estuaries: Chesapeake Bay case study,” Remote Sens. of Environ. 109, 464–472, 2007. [CrossRef]
39. A. Bricaud, M. Babin, A. Morel, and H. Claustre, “Variability in the chlorophyll-specific absorption coefficients of natural phytoplankton: Analysis and parameterization,” J. Geophys. Res. 100, 13321–13332 (1995). [CrossRef]
