- Email: [email protected]
Satellite monitoring of coccolithophore blooms in the Black Sea from ocean color data
Remote Sensing of Environment 146 (2014) 113–123
Contents lists available at ScienceDirect
Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse
Satellite monitoring of coccolithophore blooms in the Black Sea from ocean color data O. Kopelevich a,⁎, V. Burenkov a, S. Sheberstov a, S. Vazyulya a, M. Kravchishina a, L. Pautova a, V. Silkin b, V. Artemiev a, A. Grigoriev a a b
Shirshov Institute of Oceanology, Russian Academy of Sciences (SIO RAS), Nakhimovsky Prospect 36, 117997 Moscow, Russia Southern Branch of the Shirshov Institute of Oceanology, Gelendjick, Russia
a r t i c l e
i n f o
Article history: Received 3 October 2012 Received in revised form 2 September 2013 Accepted 3 September 2013 Available online 11 October 2013 Keywords: Satellite monitoring Ocean color Black Sea Coccolithophore bloom Backscattering River runoff Two-parametric model Regional algorithm
a b s t r a c t Satellite observation is an eminently suitable tool for monitoring and study of coccolithophore blooms but quantitative estimation needs perfected algorithms. The current satellite algorithm using the data of ocean color scanners, such as SeaWiFS and MODIS, is based on retrieval of the backscattering coefficient and then estimation of the calcite concentration via an empirical relationship. The regional algorithm for the Black Sea is also an empirical relationship between the coccolithophore concentration and the particle backscattering coefficient, based on in situ measured data. The drawback of the both algorithms is that “non-coccolithophore” backscattering is not accounted for. This shortcoming is particularly significant for the Black Sea which is strongly influenced by river runoff. We analyze the problem of taking into account the “non-coccolithophore” backscattering by using the integrated approach with the comprehensive data set including satellite data and optical, hydrological and biogeochemical parameters measured in situ in the eastern part of the Black Sea. The main result obtained is a new algorithm for retrieval of coccolithophore concentration in the Black Sea from satellite ocean color data. The algorithm is based on the two-parametric model of the particle backscattering coefficient, which takes into account both coccolithophore bloom and river runoff. Application of the new algorithm to satellite ocean color data in 1998–2011 gave an opportunity to separate the changes associated with the abovementioned factors. Several new interesting results were obtained. It was shown that coccolithophore blooms occupied only a part of the area with high values of the particle backscattering coefficient, and there were the years with no coccolithophore bloom in the Black Sea in June even though the high values of the particle backscattering coefficient were observed. The other problems to be solved for improvement of the satellite algorithm, providing a quantitative assessment of the intensity of coccolithophore blooms in the Black Sea, were discussed. © 2013 Elsevier Inc. All rights reserved.
1. Introduction Coccolithophores are single-celled algae with a spherical cell covered by disk-shaped coccoliths, composed of calcium carbonate, CaCO3. Coccolithophore blooms (CB) can spread to vast areas in various oceans and seas, and have a significant impact on important physical and biogeochemical processes, in particular, the exchange of CO2 between the ocean and the atmosphere, and global climate change (Thierstein & Young, 2004). The long-term flux of coccoliths to the ocean floor contributes to the formation of chalk and limestone rocks. Plated cells and detached coccoliths have a strong non-selective light scattering, that makes it possible to detect CB from satellite color scanner. Our work deals with the problem of quantitative estimates of CB on the satellite data and development of a regional algorithm for the Black Sea, where data on coccolithophore and coccolith concentrations are available from field
⁎ Corresponding author. 0034-4257/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.rse.2013.09.009
measurements. The water-leaving radiance, recorded by satellite color scanner, is increasing dramatically in the Black Sea in June. In most cases, this is due to the CB, but a close relationship between the particle backscattering coefficient bbp, calculated from satellite data, and the intensity of the CB is not always observed. An example of a coccolithophore bloom in the Black Sea is given in Fig. 1; it is seen that the bright areas in Fig. 1A correspond to the high values of bbp in Fig. 1B (the bbp distribution was computed by the simplified algorithm described by Burenkov, Ershova, Kopelevich, Sheberstov, & Shevchenko, 2001). Satellite studies of coccolithophore blooms in the Black Sea began with SeaWiFS ocean color data (Cokacar, Kubilay, & Oguz, 2001). The particle backscattering coefficient bbp is a key parameter for quantitative assessment of the CB intensity; its value is derived from satellite data and then used for calculating the CB characteristics (http://optics. ocean.ru). The NASA SeaDAS software (http://oceancolor.gsfc.nasa. gov/seadas/) can derive calcite (CaCO3) concentration (particulate inorganic carbon — PIC) with the merged 2-band and 3-band algorithm, default for all sensors (Gordon & Balch, 1999; Gordon et al., 2001).
114
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 1. Coccolithophore bloom in the Black Sea on 12 June 2004 from MODIS-Aqua satellite scanner. A, MODIS-Aqua image in true color; B, spatial distribution of the particle backscattering coefficient bbp.
The obvious drawback of the above algorithm, which the authors themselves pointed out, is that all particulate backscatter is assumed to be calcite-related, and the non-calcite related backscattering is not taken into account. This shortcoming is particularly significant for the Black Sea which is strongly influenced by river run-off, and the concentration of particulate matter brought by rivers increases just in June, after the maximum of river discharge in the second half of May. This inadequacy is also related to the SIO RAS algorithm for retrieval of coccolithophore concentration Ncoc from satellite data on the particle backscattering bbp (http://optics.ocean.ru). Our present study aims to take into account the “non-coccolithophore” backscattering in the Black Sea, using a new approach. 2. Data and methods 2.1. Study area Our study is focused in the eastern part of the Black Sea, where the field measurements have been conducted since 2004. The Black Sea is classified with Case 2 waters because their seawater optical properties are determined not only by phytoplankton and by the material associated with it, as in Case 1 waters, but also by the particulate and dissolved matter brought with river runoff (International Ocean Color
Coordinating Group (IOCCG), 2000). Based on the position, bathymetry and influence of river runoff, the Black Sea can be divided into eight subregions (Fig. 2). Sub-regions #1–3 are located in the shelf area with depth of less than 50 m; region #1 is under strong influence of water discharge of the Dnieper, Dniester and Bug rivers, whereas region #2 is influenced mainly by the Danube River. Sub-regions #4 and 5 are the northern and southern areas of the outer shelf (depth of 50–200 m) in the western part of the Black Sea. In the eastern part of the basin, the southern, eastern and north-eastern shelf areas were considered as a single region (#8). Though most large rivers (Danube, Dniester, Southern Bug, and Dnieper) flow into the Black Sea in the north-west (their suspended sediment load is equal to 41.5 · 106 t/year or about 55% of the total load), the contribution from the north-eastern, eastern and southern rivers of Russia, Georgia and Turkey is also high — 33.8 · 106 t/year altogether (Mikhailov, Morozov, Cheroy, Mikhailova, & Zavyalova, 2008). Their manifestation is well seen on satellite images in the second half of May and in June. In 2004–2006, our field studies were conducted in both the coastal turbid and open clear waters; the stations were planned relying on the analysis of satellite data. The ship route in 2004 is shown on Fig. 3A. As seen, the ship route included three sections: from Gelendjik to the open part (~44°N, 37.5°E), along the shore (to ~43°N, 38.5°E), and then from open sea to Sochi. The values of the bio-optical characteristics varied over a wide range: Secchi depth from 2.5 to 14 m, the nearsurface beam attenuation coefficient from 0.4 to 3.4 m−1, chlorophyll concentration from 0.17 to 0.65 mg m−3, and suspended matter concentration from 0.5 to 4.1 mg/l. Nearly the same values of the parameters were measured in expeditions in 2005–2006. In 2006, the ship route was planned for studying the area of turbid waters in open sea. In 2007–2011, by technical reason, we conducted our studies in the coastal zone.
2.2. Field studies
Fig. 2. Sub-regions in the Black Sea: 1 — northern inner shelf; 2 — north-western inner shelf; 3 — south-western inner shelf; 4 — north-western outer shelf; 5 — south-western outer shelf; 6 — western open part; 7 — eastern open part; and 8 — eastern and southern shelves.
The field studies, in parallel with satellite observations, were carried out each June since 2004, to understand reasons of the enhanced values of bbp in the eastern part of the Black Sea. They included optical and biogeochemical measurements. As a rule, the following optical quantities were measured: the upwelling radiance just beneath the sea surface and the surface irradiance by a floating spectroradiometer (Artemiev et al., 2000), the surface and underwater irradiance (Khrapko, Kopelevich, Burenkov, Grigoriev, & Terekhova, 2007), vertical distribution of the beam attenuation coefficient and seawater temperature by a submersible transmissometer, and Secchi depth. The biogeochemical parameters included the concentrations of chlorophyll “a” and pheophytin “a”
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
115
Fig. 3. Field studies in the eastern part of the Black Sea. A, Area of the studies (the station numbers in 2004 are given); B, a floating spectroradiometer.
measured by a fluorometric method (Holm-Hansen & Riemann, 1978; EPA Method 445.0, 1997), total suspended matter (TSM), the concentrations of coccolithophore cell (Ncoc) and coccolith (Nc), specific composition of the marine phytoplankton and qualitative composition of particulate and dissolved matter. For development of an algorithm quantifying the intensity of coccolithophore blooms in the Black Sea, taking into account the “non-coccolithophore” backscattering, data on the spectral radiance reflectance and on the quantitative composition of phytoplankton and suspended matter were most important. 2.2.1. Floating spectroradiometer data Detailed description of the floating spectroradiometer (Fig. 1B), its calibration and data processing was given by Artemiev et al. (2000). The instrument measures continuously spectral values of the surface irradiance Ed(λ, 0+) and the upwelling radiance just below the sea surface L(λ, 0−) to avoid sun glints. The spectral range is 390–700 nm; spectral resolution — 2.5 nm; accuracy is about 5%. The measurements are performed at a distance of 20–30 m from the ship to avoid the ship influence on the results. The absolute calibration of irradiance and radiance was carried out in laboratory conditions before the field studies. The immersion factor was used to correct the calibration coefficient for L(λ, 0−) while the instrument in water (Artemiev et al., 2000). From the measured values of Ed(λ, 0+) and L(λ, 0−) the water radiance reflectance ρ(λ) is calculated h i − þ ; ρðλÞ ¼ πLðλ; 0 Þ= T E Ed λ; 0
ð1Þ
where TE, is the transmission of the sea surface for the downwelling irradiance. From ρ(λ), the parameter X = bb / (a + bb) is derived with the formula by Lee, Carder, Mobley, Steward, and Patch (1998) 0:753 ρðλÞ ¼ π 0:07 þ 0:155X X:
ð2Þ
values for two wavelengths 488 and 555 nm, corresponding to MODIS spectral bands. As usual, the low-parametric models are used for the absorption a(λ) and backscattering bb(λ) coefficients aðλÞ ¼ aw ðλÞ þ ag ðλÞ;
ð3Þ
bb ðλÞ ¼ bbw ðλÞ þ bbp ðλÞ;
ð4Þ
−n
bbp ðλÞ ¼ bbp ðλ=555Þ
;
ag ðλÞ ¼ ag ð440Þ exp½−0:015ðλ−440Þ;
ð6Þ
aw and bbw are the absorption and backscattering coefficients of pure seawater. Contribution of phytoplankton aph to the absorption is not considered, because we use the wavelengths of 488 and 555 nm, where the contribution of the pigments can be neglected compared with the contribution of yellow substance and pure seawater. Simple estimate for aph(488) and aph(555), by using the chlorophyll-specific absorption coefficients of natural phytoplankton (Bricaud, Babin, Morel, & Claustre, 1995) and the mean chlorophyll concentration from our field studies equal to 0.43 mg m−3, gives, respectively, 0.016 and 0.003 m−1; the total absorption at these wavelengths is typically 25–30 times greater. The spectral slope n for bbp(λ) is taken as 0.87; the ratio bbp(488)/ bbp(555) with this value is equal to ≅1.12 and agrees within 5% with values of this ratio when using other spectral slopes (n = 0.5 and 1.2 — see Section 3.2). To simplify the algorithm, it was also assumed that the ratio bb(488)/ bb(555) ≅ bbp(488)/bbp(555). This assumption is well satisfied when bbp ≫ bbw but even for bbp = 0.004 m− 1 (the lowest value of bbp during our June field studies) the error is less that 15%. Under the above assumptions, the ratio Y(488)/Y(555), where Y = X / (1 − X), is equal to Y ð488Þ=Y ð555Þ≅1:12að555Þ=að488Þ:
Then the particle backscattering coefficient bbp and the absorption coefficient of yellow substance ag are calculated by using a simplified semianalytic algorithm. The input parameters of the algorithm are the ρ
ð5Þ
ð7Þ
From Eq. (7), the parameter ag = ag(440) is calculated and then bb = bbp(555) is determined from X(555).
116
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
The bbp values derived by this algorithm were in good agreement with the bbp values computed with the simplified algorithm (http:// optics.ocean.ru). 2.2.2. Determination of the phytoplankton composition The seawater samples were taken by a water sampler from depths down to 15–20 m or by a bucket from the sea surface. The number of phytoplankton cells was counted in concentrates obtained by the filtration of 2–5 l of seawater samples through inverse filtering chambers with 1 μm nuclear filters (Mikaelyan, Silkin, & Pautova, 2011). The identification of the species and counting of the cells were carried out using a light microscope at 16 × 10 and 16 × 40 magnifications. The nanoand microplankton cells were counted using 0.05 ml Nageotte and 1 ml Naumann counting chambers. The small flagellates (2–4, 4–6, and 6–8 μm fractions), picoplankton (1–2 μm fraction), and coccoliths were counted using a Finuchs–Rosenthal counting chamber. The biomass was calculated using a volumetric method. 2.2.3. Measuring TSM and its components TSM concentration was determined by a standard geological technique. Seawater samples taken at different depths were filtrated through membrane nuclear filters 47 mm in diameter with a pore size of 0.45 μm, which previously were dried and weighed. At the same samples, the volume concentration of suspended matter was determined with Coulter counter (Multisizer 3). The water samples (from 1 to 5 l depending on the TSM concentration) were filtered under a vacuum of 0.4 atm. The TSM mass concentration was measured in laboratory conditions and its composition was determined (http://www.geotraces.org/). For determining the carbon, the water sample was filtered through fiberglass reinforced filters (GF/F of the Whatman Company) under a vacuum of 0.2 atm (http://www.geotraces.org/). The main TSM chemical components (Si, Al, and P) were determined, using nuclear filters, by the photometric technique in line with the method by Gel'man and Starobina (1976), modified in SIO RAS for the suspended particulate matter. The accuracy of the method is ±15%. Inter-calibration of this method with ICP-MS and the atomic absorption methods was conducted. It is assumed that the content of lithogenic material is directly proportional to the aluminum content: TSM_trg = Al · 100 / 8.58 (Taylor & McLennan, 1985). The carbon content in the suspended matter was automatically measured by the culonometric method using a Russian carbon analyzer AN 7560. Under carbon contents of 30–100 μg C/l, the accuracy was ±15% and the measuring limits were 5–500 μg C/l. 2.3. Satellite ocean color data Satellite data were collected for the Black Sea with the SeaWiFS sensor (1998–2010) and MODIS-Aqua (July 2002 to the present). They were used to construct long-term series of data on chlorophyll “a” concentration, the particle backscattering and the yellow substance absorption coefficients. These data are presented in an electronic atlas “Bio-optical characteristics of the Barents, White, Black, and Caspian seas from data of satellite ocean color scanners”; its last 6th issue, 2011 is available at web site http://optics.ocean.ru. Satellite data were obtained from NASA web site (http://oceancolor.gsfc.nasa.gov). The Atlas (http://optics.ocean.ru) contains color maps with the mean monthly distributions of chlorophyll concentration and the particle backscattering coefficient bbp in the Black Sea from January 1998 to December 2010 and the mean monthly distributions of the yellow substance absorption coefficient ag for warm season (May–September) in these years. There are the diagrams, showing a variability of the monthly means of the bio-optical characteristics in different regions, and tables with the seasonal (for chlorophyll concentration and the yellow substance absorption coefficient) and annual (for the particle backscattering coefficient) mean values with their standard deviations in different
regions. It was found that the spatial distributions of bbp in June are noticeably different from other months. Fig. 4 shows the variations of the monthly means of the particle backscattering coefficient in the north-western inner shelf (#2), western (#6) and eastern (#7) open parts, and the eastern shelf (#8). The algorithm description is given at http://optics.ocean.ru. One can see that the Black Sea grows turbid during June, and the increased values of the particle backscattering coefficient bbp are observed both in the western (#2, #6) and the eastern (#7, #8) sub-regions, as in the coastal zone (#2, #8) and in the open sea (#6, #7). During our field studies, the bbp values derived from data, measured by a floating spectroradiometer, were used for validation of the satellite retrieval; the agreement was quite good. 3. Development of a regional algorithm for retrieval of the coccolithophore concentration 3.1. Statistical relationships between Ncoc, TSM and bbp derived from the field data The field studies have provided us with direct evidence of influence of the coccolithophore blooms and the river runoff on the bbp values and their results have shown that the contributions in bbp vary markedly in different years. Fig. 5A demonstrates the statistical relationship between coccolithophore concentration Ncoc and the particle backscattering coefficient bbp derived from field data of 2004–2008. A considerable difference is observed between the slopes of regression lines in 2006 and in other years; for the whole data set the significant correlation exists between Ncoc and bbp (level of significance P b 0.005) but the correlation is weak (R2 = 0.14). The same also applies to the relationship between TSM and bbp: for the whole set R2 = 0.32, but the correlation sorted by year is much stronger (2004 — n = 19, R2 = 0.76; 2005 — n = 13, R2 = 0.95; 2007 — n = 13, R2 = 0.90; but 2006 — n = 16, R2 = 0.36). Appropriate to note that the June spatial distributions of bbp in 2004 and 2006 were quite different from each other (Burenkov et al., 2007). In June 2004, the area of turbid waters was adjacent to the eastern coastal zone of the Black Sea (Fig. 1), while in June 2006, the greatest values of bbp were mainly observed in open sea (Fig. 10 in Section 5). The latter was assumed to be associated with coccolithophore bloom because the area of high values of bbp in 2006 was separated from the coastal zone. To investigate the role of the river runoff and coccolithophore blooms in forming high values of bbp in the eastern part of the Black Sea in June, we collected all field data from 2004 to 2008 for the stations where the data from a floating spectroradiometer, direct determination of coccolithophore concentration, and matching data of satellite measurements were available (n = 72). Then we selected the stations with Ncoc N 0.5 · 106 cells/l (n = 55) and the stations in 2006 (n = 16). The four data sets are presented in Table 1. The following results should be noted from the data in Table 1: • The mean values of bbp for the set #3 and set #4 are almost the same, though the coccolithophore concentrations bNcocN for the set #3 is about 2.3 times more than for the set #4. • The mean value of the absorption coefficient of yellow substance bagN for the set #3 is about 1.3 times less than for the set #4. • A noticeable difference is seen for the correlation between changes of Ncoc and bbp, bbp and ag for the sets #3 and #4; for the set #3 a quite strong correlation is observed between changes of Ncoc and bbp but a weak correlation between bbp and ag; for the set #4 — on the contrary. • For the set #4, there is a significant correlation between changes of Ncoc and ag. These are important conclusions which help to understand the roles of coccolithophore blooms and river runoff in the formation of high values of bbp in June.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
117
Fig. 4. Changes of the monthly means of the particle backscattering coefficient bbp, m−1 in different sub-regions derived from satellite ocean color data in 1998–2011: north-western inner shelf (#2), western (#6) and eastern (#7) open parts, and the eastern shelf (#8) (http://optics.ocean.ru).
Fig. 5. Statistical relationships derived from field data in 2004–2008: A, coccolithophore concentration Ncoc vs. the particle backscattering coefficient bbp; total number of data is 55; crosses — data of 2004–2005 and 2007–2008 (n = 39), circles — 2006 (n = 16). Solid line is the regression line between Ncoc and bbp for all data, the dashed line — data of 2006, the dash-and-dot line — 2004–2005 and 2007–2008. B, Total suspended matter concentration TSM vs. bbp (n = 59). Triangles — 2004, circles — 2005, diamonds — 2006, and crosses — 2007.
118
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Table 1 Parameters of the selected data sets: n is number of data; bbbpN, bNcocN, and bagN are the means of bbp, Ncoc, and ag for each of the set; R2[Ncoc, bbp], R2[ag, bbp], and R2[Ncoc, ag] are the coefficients of determination for the linear regression between Ncoc and bbp, ag and bbp, and Ncoc and ag, respectively. Set
#1 (all data)
#2 (Ncoc N 0.5 · 106 cells/l)
#3 (2006)
#4 (2004, 2005, 2007, 2008)
n bbbpN, m−1 bNcocN, ×106 cells/l bagN, m−1 R2[Ncoc, bbp] R2[ag, bbp] R2[Ncoc, ag]
72 0.0170 1.47 0.111 0.209 0.684 0.036
55 0.0188 1.83 0.119 0.143 0.668 0.007
16 0.0189 3.04 0.098 0.578 0.048 0.110
39 0.0188 1.33 0.128 0.194 0.819 0.352
Table 2 The lowest monthly means of bbp and ag derived from satellite data over 2003–2010. Parameter
n
Interval
Mean value
Standard deviation
bbp, m−1 ag, m−1
24 21
0.0022–0.0028 0.021–0.060
0.0025 0.047
0.0002 0.013
The coefficients Kcoc and Kriv were determined from the statistical characteristics of the sets #3 and #4 in Table 1. As seen from Table 1, their statistical characteristics are significantly different because these data sets relate to different conditions occurring in the eastern part of the Black Sea. The following values of Kcoc and Kriv were obtained: Kcoc = 2.74 · 10−3, Kriv = 0.157.
3.2. Two-parametric model of the particle backscattering coefficient bbp in June
3.3. Algorithm for retrieval of coccolithophore concentration from satellite ocean color data
The results discussed in Section 3.1 clearly indicate that the high values of bbp in June, along with coccolithophore blooms, can be caused by suspended particles coming from the river runoff. These results point to the need for two-parametric model for bbp in June, including not only Ncoc but also a parameter characterizing the river runoff. Since our goal is the development of satellite algorithm, we choose our second parameter so that it can be determined from satellite data. Rivers carry a large amount of suspended particles and colored organic matter into the Black Sea. The latter is characterized by the absorption coefficient ag which can be derived from satellite data. Our idea of evaluation of the “non-coccolithophore” component of bbp (denote it as bb_riv) is based on an assumption that the values of ag and bb_river are correlated with each other, and bb_riv can be represented through ag. As it was shown above (Table 1), this assumption is supported by results of our field studies. The proposed two-parameter model is based on the following assumptions:
If the values of bbp and ag are known, it is easy to calculate from Eq. (11) the coccolithophore concentration Ncoc. In Section 2.2.1, it is shown how to calculate bbp and ag from data measured by a floating spectroradiometer. In the case of MODIS satellite data, the values of the remote sensing radiance reflectance Rrs for two MODIS spectral bands at 488 and 555 nm are taken as the input parameters. They are converted into values of the sub-surface radiance reflectance ρ(λ) by means of a formula by Lee et al. (1998):
• The increase of the particle backscattering coefficient bbp in the period of CB, regarding its background value bbp_bg, is caused by two factors: (i) the scattering by coccolithophore particles bb_сoc; (ii) the augmentation of scattering by particles coming from the river runoff Δbb_riv Δbbp ¼ bbp −bbpXbg ¼ bbXсoc þ ΔbbXriv :
ð8Þ
• The scattering by coccolithophore particles bb_сoc is proportional to a numerical concentration of coccolithophore bbXсoc ¼ K coc N coc ;
ð9Þ
where Ncoc is coccolithophore concentration, Kcoc is the coefficient of proportionality. • The increase of the particle backscattering arising from the river runoff Δbb_riv is proportional to the augmentation of the yellow substance absorption ag regarding its background value ag_bg ΔbbXriv
¼ K riv ag –agXbg ;
ð10Þ
where Kriv is the coefficient of proportionality between Δag and Δbb_riv. Thus, the model has the form: bbp −bbpXbg ¼ K coc Ncoc þ K riv ag –agXbg :
ð11Þ
As the background values, the lowest monthly means of bbp and ag, derived from satellite data over 2003–2010, were selected (Table 2). Interestingly, that 20 of 24 the selected values of bbp were in September–October, 4 — in March–April; 14 of 21 the selected values of ag occurred in June–July, 7 — in August–October.
ρðλÞ ¼ Rrs ðλÞ=½0:165 þ 0:497Rrs ðλÞ:
ð12Þ
Next, we use Eqs. (2)–(6) and (8)–(11), taking into account the spectral dependences of the backscattering of coccolithophore and river particles (Voss, Balch, & Kilpatrick, 1998; Kopelevich, 2012): −1:2
bbXсoc ðλÞ ¼ K coc ðλ=555Þ ΔbbXriv
N coc ;
ð13Þ
−0:5 ¼ K riv ðλ=555Þ ag –agXbg :
ð14Þ
With Rrs(488) and Rrs(555) as the input parameters and by using the above formulae, we can obtain a linear system of two equations with two unknowns ag and Ncoc f 1 ðλi Þag þ f 2 ðλi ÞNcoc ¼ g ðλi Þ;
i ¼ 1; 2
ð15Þ
where expressions for the functions f1(λi), f2(λi), and g(λi) are derived from Eqs. (3)–(6) and (9)–(14). Stability of the algorithm is provided by an appropriate choice of the spectral bands (in particular by large difference in the absorption coefficients of pure water — 0.0145 m−1 at 488 nm and 0.060 m−1 at 555 nm) and high values of the particle backscattering in our case. 4. Validation of the developed algorithm 4.1. Testing the bb_riv model One of two main assumptions, which are the basis of the proposed model and algorithm, is the relationship between bb_riv and the absorption coefficient of yellow substance ag derived from data on the spectral radiance reflectance (ρ(λ) measured by a floating spectroradiometer or Rrs(λ) from satellite data). It is important to clarify that for our problem we do not need a precise knowledge of absolute values of ag — we use ag as a parameter allowing us to evaluate bb_riv from remote sensing data. Since we cannot directly test the relationship between bb_riv and ag, the data on the terrigenous component TSM_trg of TSM, derived from the geochemical study (see Section 2.2.3), were used for the testing. Fig. 6 shows the regression line between TSM_trg and ag.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 6. Relationship between the terrigenous component TSM_trg and ag (n = 33, R2 = 0.86) derived from field data in 2005–2007.
Unfortunately, TSM_trg was not determined at all in 2004, and it was not determined in all samples in 2005–2007. But the considered set is sufficiently representative, because it involves the years with different conditions (see Fig. 5). The regression equation between TSM_trg and ag has the form TSMXtrg ¼ 13:8ag −1:21:
119
Fig. 8. The regression between the coccolithophoride concentration Ncoc (106 cells/l) and the particulate inorganic carbon PIC (mol/m3), derived from concurrent in situ and MODIS data in the eastern part of the Black Sea in June 2004–2008 (n = 35, R2 = 0.367).
The regression equations between Ncoc_calc and Ncoc_meas have the following form NcocXcalcXfield ¼ 0:83NcocXmeas þ 0:57;
ð17Þ
NcocXcalcXmodis ¼ 0:87N cocXmeas þ 0:55:
ð18Þ
ð16Þ
This equation is valid for ag N 0.088; TSM_trg was taken to be 0 at lower values of ag. The correlation coefficient between TSM_trg and ag equals 0.927; error of the regression sregr = 0.30 mg/l. Such value of sregr is quite acceptable for quantitative assessment of the terrigenous component TSM_trg using the parameter ag, derived through values of the spectral radiance reflectance.
4.2. Direct verification of Ncoc calculation with the developed algorithm Verification of the Ncoc values calculated with the developed algorithm was performed by using the in situ measured coccolithophore concentrations. Fig. 7 shows the scatterplots Ncoc_cacl vs. Ncoc_meas for two cases: the first — calculation with data measured by a floating spectroradiometer, the second — from the satellite data. All appropriate data of 2004–2008 were used.
As seen, Eqs. (17) and (18) are similar to each other; the correlation coefficients are equal to 0.764 and 0.730, respectively; standard errors 1.15 and 0.99 · 106 cells/l. The smaller error in the second case is related to lower concentration of Ncoc_meas: the average value in the first case was 1.42 · 106 cells/l (n = 47), in the second — 1.17 · 106 cells/l (n = 38). The accuracy of calculation is not high, but it is important that the calculation from satellite data is quite stable, and there is no significant increase in errors compared with calculation on the field data. One of the main reasons of high error in the calculated values of Ncoc is variability of the ratio Nlith/Ncoc between concentrations of the detached coccoliths Nlith and the plated cells Ncoc. This problem was considered for the Barents Sea, and it was shown that this ratio changed about 20 times but the coefficient Kcoc — only 3 times (http://www. sequoiasci.com/research/pie2012.cmsx). In the Black Sea, great variations of the ratio Nlith/Ncoc were also observed, and the coefficient Kcoc in Eq. (9) is its optimal value from the statistical evaluation (see Section 3.2).
Fig. 7. Scatterplots Ncoc_calc vs. Ncoc_meas: A, Ncoc_cacl_field is calculated from data on ρ(λ) measured by a floating spectroradiometer (n = 47, R2 = 0.583), B, Ncoc_cacl_modis — from MODISAqua data (n = 38, R2 = 0.534).
120
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 9. Comparison between the spatial distributions of Ncoc calculated with the new regional algorithm (A) and of PIC (B) derived from MODIS-Aqua data on 12 June 2004.
Fig. 10. Changes in the spatial distributions of the particle backscattering coefficient bbp, coccolithophoride concentration Ncoc, and the ratio rp = bb_riv / bbp in the Black Sea in May–June 2006 derived from MODIS-Aqua.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
121
Fig. 11. Changes of the June means of the coccolithophore concentration Ncoc (left) and the ratio rp = bb_riv / bbp (right) in the eastern (#7) open part and the eastern shelf sub-region (#8).
4.3. Comparison between the developed algorithm and Gordon, Balch algorithm
5.1. Changes in the spatial distributions of bbp, Ncoc and in the bb_riv/bbp ratio in May–June 2006
It is interesting to compare the results derived with our algorithm (Ncoc) and by Gordon, Balch algorithm (http://oceancolor.gsfc.nasa. gov/seadas/) providing calcite concentration (particulate inorganic carbon — PIC data product). Fig. 8 shows the relationship between Ncoc and PIC derived from concurrent in situ and MODIS data in the eastern part of the Black Sea in June 2004–2008 (n = 35, R2 = 0.367). First of all note, that the correlation coefficient is equal to ~0.61, and a significant correlation exists between Ncoc and bbp (level of significance P b 0.0005). The regression equation has the form
These changes are shown in Fig. 10. It is seen how the turbidity was growing since the first half May, mainly in the western (near Danube) and eastern (Caucasus rivers) coastal zones. In the period 6–15 May, the contribution bb_riv from the river runoff to bbp was more than 0.9 almost all over the Sea, and only in two small areas in the eastern part, where CB originated, it decreased below 0.6. In the period 6–15 June, on the contrary, a strong CB covered a large part of the Sea, and rp = bb_riv / bbp was below 0.4 and even 0.2. Comparing the distributions of bbp and Ncoc in this period with those for 2004 (Figs. 1В and 9А), one can see significant difference. First, in 2004, the highest values of bbp were observed in the sub-region #2, under the Danube runoff, and in the coastal zone of the eastern half of the Sea (Fig. 1В), whereas in 2006, they were observed in the open sea (Fig. 10). Second, in 2004, CB was limited to the eastern part of the Sea (Fig. 9А), whereas in 2006, it covered almost all of the Sea (Fig. 10). These distinctions for the eastern part of the Sea have already been noted in Section 3.1 according to the data of field measurements.
Ncoc ¼ 223PIC þ 0:06;
ð19Þ
the regression error Sregr is equal to ~0.7 · 106 cells/l. Eq. (19) allows us to estimate the PIC boundary value, corresponding the accepted lowest value of the coccolithophore bloom Ncoc = 1 · 106 cells/l; the value of PIC (with account of error) is in the range 0.001–0.007 mol/m3 with the mean value of 0.004 mol/m3. It is interesting that the close value is obtained from data of Balch, Kilpatrick, Holligan, Harbour, and Fernandez (1996) for central North Atlantic and Gordon et al. (2001) south of Plymouth; the boundary value of PIC was found to be ~0.004 mol/m3 (under realistic for the Black Sea assumption that the ratio Nlith/Ncoc is equal to 10). In Fig. 9, the spatial distributions of Ncoc, calculated with the new regional algorithm (A), and of PIC, derived from MODIS-Aqua data on 12 June 2004 (B), are compared. It is also interesting to compare both distributions with the bbp distribution in Fig. 1B. One can see that the PIC distribution agrees very well with the bbp distribution, whereas some distinction is observed with the Ncoc distribution (it should be noticed that the both algorithms failed in the western part of the Black Sea). If we accept the CB boundaries as 1 · 106 cells/l for Ncoc and 0.004 mol/m3 for PIC, it is seen that they are well matched. In the eastern part, the exceptions are observed near the Kerch Strait and in the eastern and, in particular, the southern coastal zones where the high values of PIC are not associated with coccolithophore blooms. 5. The algorithm application The main achievement of the new algorithm is that it can distinguish the changes, associated with coccolithophore blooms and with the river runoff effect, although the accuracy is not very high. Some examples of such separation are given below.
5.2. Inter-annual changes in the coccolithophore concentration and in the terrigenous component of suspended matter in June from satellite ocean color data in 1998–2011 Fig. 11 shows the changes in the June means of the coccolithophore concentration Ncoc and the ratio rp = bb_riv / bbp in the eastern (#7) open part and the eastern shelf sub-region (#8). As seen, there are years with no marked coccolithophore blooms (2001, 2003, and 2010); it was shown by Burenkov, Kopelevich, and Sheberstov (2011) that the inter-annual changes in the intensity of coccolithophore blooms can be linked to the winter sea surface temperature. But the higher values of bbp were observed in the above mentioned years (Fig. 4), and, according to our model, they should be attributed to the bb_riv contribution. It is seen from Fig. 11, the bb_riv contribution to the particle backscattering bbp in the sub-region #7 was more than 50% in all years except 2006 and 2008; in the sub-region #8 this contribution was more than 60% in all years. In open sea (sub-region #7) the lowest value of rp was in 2006 (less than 0.4), the highest in 2001 and 2003 (more than 0.9). Fig. 12 demonstrates variability of the mean June distributions of the coccolithophore concentration Ncoc, as well as, for comparison, of the particulate inorganic carbon PIC and of the ratio rp = bb_riv / bbp derived from satellite ocean color data with the new algorithm. We applied our algorithm to the whole basin while it was designed for the eastern part of the Black Sea. Unfortunately, we have not been able to
122
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 12. Changes of the June mean distributions of Ncoc, PIC and the ratio rp = bb_riv / bbp from satellite ocean color data in 1998–2011.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
validate it by data from field measurements in the western part, so the correspondence to reality of the constructed distributions in that part is questionable. Some evidence of that the presented distributions of Ncoc are true, at least qualitatively, follows from the results of comparison between the distributions of Ncoc and PIC (this issue was discussed in Section 4.3). 6. Conclusion The main result of the present work is development of a new algorithm for retrieval of coccolithophore concentration in the Black Sea from satellite ocean color data. The algorithm is based on the twoparametric model allowing take into account the “non-coccolithophore” share of the particle backscattering coefficient bbp. In June, when coccolithophore blooms are observed, an origin of such particles is mainly river runoff. Application of the new algorithm to satellite ocean color data in 1998–2011 gave an opportunity to separate changes associated with coccolithophore bloom and with the river runoff effect. It was found that there were the years with no marked coccolithophore bloom (2001, 2003, and 2010). The coccolithophore blooms are more changeable in comparison with the river runoff which is more stable. Possible reasons of the inter-annual changes in the intensity of coccolithophore blooms in the Black Sea were discussed before (Burenkov et al., 2011). The results obtained can be considered only as a first step, perhaps the most important, on the way to the creation of satellite algorithm for quantitative assessment of the intensity of coccolithophore blooms in the Black Sea. It is necessary to solve several serious problems, one of them is variability of the components of coccolithophore bloom. It includes coccoliths, plated cells, naked cells and others; these components have quite different specific backscattering coefficients, and the ratio of their contributions can be very changeable. One of possible solution is use of calcite concentration as a parameter to be determined, as it is done in the Gordon, Balch algorithm. Another problem is how to distinguish the coccolithophore bloom from the blooms of other mass algae, such as diatoms, by using satellite data. This is very hard task, but it is very important for assessment of the carbon cycle and activity of the biological pump in the Black Sea. Solution to these problems is impossible only on the satellite data; they must be combined with data from the direct measurements of optical and biogeochemical characteristics. Acknowledgment The work was supported by the Program for Fundamental Research No.23, Russian Academy of Sciences, by grants for the NATO project ESP.EAP.SFPP 982678 “Bio-Optical Characterization of the Black Sea for Remote Sensing Applications”, and by grant Nos. 10-05-00936 and 1305-00618 from the Russian Foundation for Basic Research. SeaWiFS and MODIS data used were obtained from the Goddard Distributed Active Archive Center under the auspices of the National Aeronautics and Space Administration. The authors thank A.A. Klyuvitkin for the data on terrigenous component of the total suspended matter, and two anonymous reviewers for very helpful comments. References Artemiev, V. A., Burenkov, V. I., Vortman, M. I., Grigoriev, A. V., Kopelevich, O. V., & Khrapko, A. N. (2000). Sea-truth measurements of ocean color: A new floating spectroradiometer and its metrology. Oceanology, 40, 139–145 (translated from Okeanologiya, 40, 148–155).
123
Balch, W. M., Kilpatrick, K., Holligan, P. M., Harbour, D., & Fernandez, E. (1996). The 1991 coccolithophore bloom in the central north Atlantic II: Relating optics to coccolith concentration. Limnology and Oceanography, 41, 1684–1696. Bricaud, A., Babin, M., Morel, A., & Claustre, H. (1995). Variability in the chlorophyll-specific absorption coefficients of natural phytoplankton: Analysis and parametrization. Journal of Geophysical Research, 100(C7), 13,135–13,142. Burenkov, V. I., Ershova, S. V., Kopelevich, O. V., Sheberstov, S. V., & Shevchenko, V. P. (2001). An estimate of the distribution of suspended matter in the Barents Sea waters on the basis of the SeaWiFS satellite ocean color scanner. Oceanology, 41, 622–628 (translated from Оkeanologiya, 2001, 41, 653–659). Burenkov, V. I., Kopelevich, O. V., & Sheberstov, S. V. (2011). Seasonal and interannual variations of the biooptical characteristics of the Black Sea from satellite data. Current problems of remote sensing from Space, 8(4), 190–199. Burenkov, V., Kopelevich, O., Sheberstov, S., Prokhorenko, O. V., Klyuvitkin, A. A., & Pautova, L. A. (2007). The June maxima of the particle backscatter in the Black Sea: Results of three years satellite and field studies. Proceedings of IV International Conference “Current Problems in Optics of Natural Waters (ONW'2007)” (pp. 157–160). Nizhny Novgorod: Institute of Applied Physics RAS. Cokacar, T., Kubilay, N., & Oguz, T. (2001). Structure of E. huxleyi blooms in the Black Sea surface waters as detected by SeaWiFS imagery. Geophysical Research Letters, 28, 4607–4610. EPA Method 445.0 (1997). In vitro determination of chlorophyll “a” and pheophytin “a” in marine and freshwater algae by fluorescence. National Exposure Research Laboratory, U.S. Environmental Protection Agency (Revision 1.2, Sept. 1997). Gel'man, E. M., & Starobina, I. Z. (1976). Photometric methods for determining the rock forming elements in ores, rocks and minerals. Moscow, SSSR: Central Laboratory, Sector of Chemical Analytic Methods GEOKhI AN [in Russian]. Gordon, H. R., & Balch, W. M. (1999). MODIS detached coccolith concentration. Algorithm theoretical basis document (30 pp.). Gordon, H., Boynton, G., Balch, W., Groom, S., Harbour, D., & Smyth, T. (2001). Retrieval of coccolithophore calcite concentration from SeaWiFS imagery. Geophysical Research Letters, 28, 1587–1590. Holm-Hansen, O., & Riemann, B. (1978). Chlorophyll-a determination: improvements in methodology. Oikos, 30, 438–447. International Ocean Color Coordinating Group (IOCCG) (2000). Remote sensing of ocean colour in coastal, and other optically-complex, waters. In S. Sathyendranath (Ed.), Reports of the International Ocean-Colour Coordinating Group, No. 3. IOCCG, Dartmouth, Canada (140 p.). Khrapko, A. N., Kopelevich, O. V., Burenkov, V. I., Grigoriev, A. V., & Terekhova, A. A. (2007). New instrument for measuring surface and underwater irradiances. Proceedings of IV International Conference “Current Problems in Optics of Natural Waters (ONW'2007)” (pp. 271–275). Nizhny Novgorod: Institute of Applied Physics RAS. Kopelevich, O. V. (2012). Application of data on seawater light scattering for the study of marine particles: A selective review focusing on Russian literature. Geo-Marine Letters. http://dx.doi.org/10.1007/S00367-012-0278-y.2012. Lee, Z., Carder, K. L., Mobley, C. D., Steward, R. G., & Patch, J. S. (1998). Hyperspectral remote sensing for shallow waters. 1. A semianalytical model. Applied Optics, 37, 6329–6338. Mikaelyan, A. S., Silkin, V. A., & Pautova, L. A. (2011). Coccolithophorids in the Black Sea: Their interannual and long-term changes. Oceanology, 51, 39–48 (published in Okeanologiya, 2011, 51, 45–53). Mikhailov, V. N., Morozov, V. N., Cheroy, N. I., Mikhailova, M. V., & Zavyalova, E. F. (2008). Extreme flood on Danube in 2006. Meteorology and Hydrology, 1, 80–89. Taylor, S. R., & McLennan, S. M. (1985). The continental crust: Its composition and evolution. Oxford: Blackwell (312 p.). Thierstein, H. R., & Young, J. R. (2004). Coccolithophores — From molecular processes to global impact. Springer (565 p.). Voss, K. J., Balch, W. M., & Kilpatrick, K. A. (1998). Scattering and attenuation properties of Emiliania huxleyi cells and their detached coccoliths. Limnology and Oceanography, 43, 870–876.
Web references Sampling and sample-handling protocols for GEOTRACES cruises. (Version 1.0, Edited by the 2010 GEOTRACES Standards and Intercalibration Committee). http://www. geotraces.org/. (December 2010). Feldman, G. C. (2012a). Ocean color products and algorithms. http://oceancolor.gsfc.nasa. gov/seadas/ Feldman, G. C. (2012b). Ocean color web. http://oceancolor.gsfc.nasa.gov Kopelevich, O. V., Burenkov, V. I., Sheberstov, S. V., & Vazyulya, S. V. (2012). Remote sensing of coccolithophore particles: Regional features. PiE 2012 Particles in Europe. Program and Abstract Volume. Barcelona, Spain: ICM (http://www. sequoiasci.com/research/pie2012.cmsx) Kopelevich, O. V., Burenkov, V. I., Vazyulya, S. V.(2011). Bio-optical characteristics of the Barents, White, Black, and Caspian seas from data of satellite ocean color scanners. http://optics.ocean.ru
Contents lists available at ScienceDirect
Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse
Satellite monitoring of coccolithophore blooms in the Black Sea from ocean color data O. Kopelevich a,⁎, V. Burenkov a, S. Sheberstov a, S. Vazyulya a, M. Kravchishina a, L. Pautova a, V. Silkin b, V. Artemiev a, A. Grigoriev a a b
Shirshov Institute of Oceanology, Russian Academy of Sciences (SIO RAS), Nakhimovsky Prospect 36, 117997 Moscow, Russia Southern Branch of the Shirshov Institute of Oceanology, Gelendjick, Russia
a r t i c l e
i n f o
Article history: Received 3 October 2012 Received in revised form 2 September 2013 Accepted 3 September 2013 Available online 11 October 2013 Keywords: Satellite monitoring Ocean color Black Sea Coccolithophore bloom Backscattering River runoff Two-parametric model Regional algorithm
a b s t r a c t Satellite observation is an eminently suitable tool for monitoring and study of coccolithophore blooms but quantitative estimation needs perfected algorithms. The current satellite algorithm using the data of ocean color scanners, such as SeaWiFS and MODIS, is based on retrieval of the backscattering coefficient and then estimation of the calcite concentration via an empirical relationship. The regional algorithm for the Black Sea is also an empirical relationship between the coccolithophore concentration and the particle backscattering coefficient, based on in situ measured data. The drawback of the both algorithms is that “non-coccolithophore” backscattering is not accounted for. This shortcoming is particularly significant for the Black Sea which is strongly influenced by river runoff. We analyze the problem of taking into account the “non-coccolithophore” backscattering by using the integrated approach with the comprehensive data set including satellite data and optical, hydrological and biogeochemical parameters measured in situ in the eastern part of the Black Sea. The main result obtained is a new algorithm for retrieval of coccolithophore concentration in the Black Sea from satellite ocean color data. The algorithm is based on the two-parametric model of the particle backscattering coefficient, which takes into account both coccolithophore bloom and river runoff. Application of the new algorithm to satellite ocean color data in 1998–2011 gave an opportunity to separate the changes associated with the abovementioned factors. Several new interesting results were obtained. It was shown that coccolithophore blooms occupied only a part of the area with high values of the particle backscattering coefficient, and there were the years with no coccolithophore bloom in the Black Sea in June even though the high values of the particle backscattering coefficient were observed. The other problems to be solved for improvement of the satellite algorithm, providing a quantitative assessment of the intensity of coccolithophore blooms in the Black Sea, were discussed. © 2013 Elsevier Inc. All rights reserved.
1. Introduction Coccolithophores are single-celled algae with a spherical cell covered by disk-shaped coccoliths, composed of calcium carbonate, CaCO3. Coccolithophore blooms (CB) can spread to vast areas in various oceans and seas, and have a significant impact on important physical and biogeochemical processes, in particular, the exchange of CO2 between the ocean and the atmosphere, and global climate change (Thierstein & Young, 2004). The long-term flux of coccoliths to the ocean floor contributes to the formation of chalk and limestone rocks. Plated cells and detached coccoliths have a strong non-selective light scattering, that makes it possible to detect CB from satellite color scanner. Our work deals with the problem of quantitative estimates of CB on the satellite data and development of a regional algorithm for the Black Sea, where data on coccolithophore and coccolith concentrations are available from field
⁎ Corresponding author. 0034-4257/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.rse.2013.09.009
measurements. The water-leaving radiance, recorded by satellite color scanner, is increasing dramatically in the Black Sea in June. In most cases, this is due to the CB, but a close relationship between the particle backscattering coefficient bbp, calculated from satellite data, and the intensity of the CB is not always observed. An example of a coccolithophore bloom in the Black Sea is given in Fig. 1; it is seen that the bright areas in Fig. 1A correspond to the high values of bbp in Fig. 1B (the bbp distribution was computed by the simplified algorithm described by Burenkov, Ershova, Kopelevich, Sheberstov, & Shevchenko, 2001). Satellite studies of coccolithophore blooms in the Black Sea began with SeaWiFS ocean color data (Cokacar, Kubilay, & Oguz, 2001). The particle backscattering coefficient bbp is a key parameter for quantitative assessment of the CB intensity; its value is derived from satellite data and then used for calculating the CB characteristics (http://optics. ocean.ru). The NASA SeaDAS software (http://oceancolor.gsfc.nasa. gov/seadas/) can derive calcite (CaCO3) concentration (particulate inorganic carbon — PIC) with the merged 2-band and 3-band algorithm, default for all sensors (Gordon & Balch, 1999; Gordon et al., 2001).
114
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 1. Coccolithophore bloom in the Black Sea on 12 June 2004 from MODIS-Aqua satellite scanner. A, MODIS-Aqua image in true color; B, spatial distribution of the particle backscattering coefficient bbp.
The obvious drawback of the above algorithm, which the authors themselves pointed out, is that all particulate backscatter is assumed to be calcite-related, and the non-calcite related backscattering is not taken into account. This shortcoming is particularly significant for the Black Sea which is strongly influenced by river run-off, and the concentration of particulate matter brought by rivers increases just in June, after the maximum of river discharge in the second half of May. This inadequacy is also related to the SIO RAS algorithm for retrieval of coccolithophore concentration Ncoc from satellite data on the particle backscattering bbp (http://optics.ocean.ru). Our present study aims to take into account the “non-coccolithophore” backscattering in the Black Sea, using a new approach. 2. Data and methods 2.1. Study area Our study is focused in the eastern part of the Black Sea, where the field measurements have been conducted since 2004. The Black Sea is classified with Case 2 waters because their seawater optical properties are determined not only by phytoplankton and by the material associated with it, as in Case 1 waters, but also by the particulate and dissolved matter brought with river runoff (International Ocean Color
Coordinating Group (IOCCG), 2000). Based on the position, bathymetry and influence of river runoff, the Black Sea can be divided into eight subregions (Fig. 2). Sub-regions #1–3 are located in the shelf area with depth of less than 50 m; region #1 is under strong influence of water discharge of the Dnieper, Dniester and Bug rivers, whereas region #2 is influenced mainly by the Danube River. Sub-regions #4 and 5 are the northern and southern areas of the outer shelf (depth of 50–200 m) in the western part of the Black Sea. In the eastern part of the basin, the southern, eastern and north-eastern shelf areas were considered as a single region (#8). Though most large rivers (Danube, Dniester, Southern Bug, and Dnieper) flow into the Black Sea in the north-west (their suspended sediment load is equal to 41.5 · 106 t/year or about 55% of the total load), the contribution from the north-eastern, eastern and southern rivers of Russia, Georgia and Turkey is also high — 33.8 · 106 t/year altogether (Mikhailov, Morozov, Cheroy, Mikhailova, & Zavyalova, 2008). Their manifestation is well seen on satellite images in the second half of May and in June. In 2004–2006, our field studies were conducted in both the coastal turbid and open clear waters; the stations were planned relying on the analysis of satellite data. The ship route in 2004 is shown on Fig. 3A. As seen, the ship route included three sections: from Gelendjik to the open part (~44°N, 37.5°E), along the shore (to ~43°N, 38.5°E), and then from open sea to Sochi. The values of the bio-optical characteristics varied over a wide range: Secchi depth from 2.5 to 14 m, the nearsurface beam attenuation coefficient from 0.4 to 3.4 m−1, chlorophyll concentration from 0.17 to 0.65 mg m−3, and suspended matter concentration from 0.5 to 4.1 mg/l. Nearly the same values of the parameters were measured in expeditions in 2005–2006. In 2006, the ship route was planned for studying the area of turbid waters in open sea. In 2007–2011, by technical reason, we conducted our studies in the coastal zone.
2.2. Field studies
Fig. 2. Sub-regions in the Black Sea: 1 — northern inner shelf; 2 — north-western inner shelf; 3 — south-western inner shelf; 4 — north-western outer shelf; 5 — south-western outer shelf; 6 — western open part; 7 — eastern open part; and 8 — eastern and southern shelves.
The field studies, in parallel with satellite observations, were carried out each June since 2004, to understand reasons of the enhanced values of bbp in the eastern part of the Black Sea. They included optical and biogeochemical measurements. As a rule, the following optical quantities were measured: the upwelling radiance just beneath the sea surface and the surface irradiance by a floating spectroradiometer (Artemiev et al., 2000), the surface and underwater irradiance (Khrapko, Kopelevich, Burenkov, Grigoriev, & Terekhova, 2007), vertical distribution of the beam attenuation coefficient and seawater temperature by a submersible transmissometer, and Secchi depth. The biogeochemical parameters included the concentrations of chlorophyll “a” and pheophytin “a”
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
115
Fig. 3. Field studies in the eastern part of the Black Sea. A, Area of the studies (the station numbers in 2004 are given); B, a floating spectroradiometer.
measured by a fluorometric method (Holm-Hansen & Riemann, 1978; EPA Method 445.0, 1997), total suspended matter (TSM), the concentrations of coccolithophore cell (Ncoc) and coccolith (Nc), specific composition of the marine phytoplankton and qualitative composition of particulate and dissolved matter. For development of an algorithm quantifying the intensity of coccolithophore blooms in the Black Sea, taking into account the “non-coccolithophore” backscattering, data on the spectral radiance reflectance and on the quantitative composition of phytoplankton and suspended matter were most important. 2.2.1. Floating spectroradiometer data Detailed description of the floating spectroradiometer (Fig. 1B), its calibration and data processing was given by Artemiev et al. (2000). The instrument measures continuously spectral values of the surface irradiance Ed(λ, 0+) and the upwelling radiance just below the sea surface L(λ, 0−) to avoid sun glints. The spectral range is 390–700 nm; spectral resolution — 2.5 nm; accuracy is about 5%. The measurements are performed at a distance of 20–30 m from the ship to avoid the ship influence on the results. The absolute calibration of irradiance and radiance was carried out in laboratory conditions before the field studies. The immersion factor was used to correct the calibration coefficient for L(λ, 0−) while the instrument in water (Artemiev et al., 2000). From the measured values of Ed(λ, 0+) and L(λ, 0−) the water radiance reflectance ρ(λ) is calculated h i − þ ; ρðλÞ ¼ πLðλ; 0 Þ= T E Ed λ; 0
ð1Þ
where TE, is the transmission of the sea surface for the downwelling irradiance. From ρ(λ), the parameter X = bb / (a + bb) is derived with the formula by Lee, Carder, Mobley, Steward, and Patch (1998) 0:753 ρðλÞ ¼ π 0:07 þ 0:155X X:
ð2Þ
values for two wavelengths 488 and 555 nm, corresponding to MODIS spectral bands. As usual, the low-parametric models are used for the absorption a(λ) and backscattering bb(λ) coefficients aðλÞ ¼ aw ðλÞ þ ag ðλÞ;
ð3Þ
bb ðλÞ ¼ bbw ðλÞ þ bbp ðλÞ;
ð4Þ
−n
bbp ðλÞ ¼ bbp ðλ=555Þ
;
ag ðλÞ ¼ ag ð440Þ exp½−0:015ðλ−440Þ;
ð6Þ
aw and bbw are the absorption and backscattering coefficients of pure seawater. Contribution of phytoplankton aph to the absorption is not considered, because we use the wavelengths of 488 and 555 nm, where the contribution of the pigments can be neglected compared with the contribution of yellow substance and pure seawater. Simple estimate for aph(488) and aph(555), by using the chlorophyll-specific absorption coefficients of natural phytoplankton (Bricaud, Babin, Morel, & Claustre, 1995) and the mean chlorophyll concentration from our field studies equal to 0.43 mg m−3, gives, respectively, 0.016 and 0.003 m−1; the total absorption at these wavelengths is typically 25–30 times greater. The spectral slope n for bbp(λ) is taken as 0.87; the ratio bbp(488)/ bbp(555) with this value is equal to ≅1.12 and agrees within 5% with values of this ratio when using other spectral slopes (n = 0.5 and 1.2 — see Section 3.2). To simplify the algorithm, it was also assumed that the ratio bb(488)/ bb(555) ≅ bbp(488)/bbp(555). This assumption is well satisfied when bbp ≫ bbw but even for bbp = 0.004 m− 1 (the lowest value of bbp during our June field studies) the error is less that 15%. Under the above assumptions, the ratio Y(488)/Y(555), where Y = X / (1 − X), is equal to Y ð488Þ=Y ð555Þ≅1:12að555Þ=að488Þ:
Then the particle backscattering coefficient bbp and the absorption coefficient of yellow substance ag are calculated by using a simplified semianalytic algorithm. The input parameters of the algorithm are the ρ
ð5Þ
ð7Þ
From Eq. (7), the parameter ag = ag(440) is calculated and then bb = bbp(555) is determined from X(555).
116
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
The bbp values derived by this algorithm were in good agreement with the bbp values computed with the simplified algorithm (http:// optics.ocean.ru). 2.2.2. Determination of the phytoplankton composition The seawater samples were taken by a water sampler from depths down to 15–20 m or by a bucket from the sea surface. The number of phytoplankton cells was counted in concentrates obtained by the filtration of 2–5 l of seawater samples through inverse filtering chambers with 1 μm nuclear filters (Mikaelyan, Silkin, & Pautova, 2011). The identification of the species and counting of the cells were carried out using a light microscope at 16 × 10 and 16 × 40 magnifications. The nanoand microplankton cells were counted using 0.05 ml Nageotte and 1 ml Naumann counting chambers. The small flagellates (2–4, 4–6, and 6–8 μm fractions), picoplankton (1–2 μm fraction), and coccoliths were counted using a Finuchs–Rosenthal counting chamber. The biomass was calculated using a volumetric method. 2.2.3. Measuring TSM and its components TSM concentration was determined by a standard geological technique. Seawater samples taken at different depths were filtrated through membrane nuclear filters 47 mm in diameter with a pore size of 0.45 μm, which previously were dried and weighed. At the same samples, the volume concentration of suspended matter was determined with Coulter counter (Multisizer 3). The water samples (from 1 to 5 l depending on the TSM concentration) were filtered under a vacuum of 0.4 atm. The TSM mass concentration was measured in laboratory conditions and its composition was determined (http://www.geotraces.org/). For determining the carbon, the water sample was filtered through fiberglass reinforced filters (GF/F of the Whatman Company) under a vacuum of 0.2 atm (http://www.geotraces.org/). The main TSM chemical components (Si, Al, and P) were determined, using nuclear filters, by the photometric technique in line with the method by Gel'man and Starobina (1976), modified in SIO RAS for the suspended particulate matter. The accuracy of the method is ±15%. Inter-calibration of this method with ICP-MS and the atomic absorption methods was conducted. It is assumed that the content of lithogenic material is directly proportional to the aluminum content: TSM_trg = Al · 100 / 8.58 (Taylor & McLennan, 1985). The carbon content in the suspended matter was automatically measured by the culonometric method using a Russian carbon analyzer AN 7560. Under carbon contents of 30–100 μg C/l, the accuracy was ±15% and the measuring limits were 5–500 μg C/l. 2.3. Satellite ocean color data Satellite data were collected for the Black Sea with the SeaWiFS sensor (1998–2010) and MODIS-Aqua (July 2002 to the present). They were used to construct long-term series of data on chlorophyll “a” concentration, the particle backscattering and the yellow substance absorption coefficients. These data are presented in an electronic atlas “Bio-optical characteristics of the Barents, White, Black, and Caspian seas from data of satellite ocean color scanners”; its last 6th issue, 2011 is available at web site http://optics.ocean.ru. Satellite data were obtained from NASA web site (http://oceancolor.gsfc.nasa.gov). The Atlas (http://optics.ocean.ru) contains color maps with the mean monthly distributions of chlorophyll concentration and the particle backscattering coefficient bbp in the Black Sea from January 1998 to December 2010 and the mean monthly distributions of the yellow substance absorption coefficient ag for warm season (May–September) in these years. There are the diagrams, showing a variability of the monthly means of the bio-optical characteristics in different regions, and tables with the seasonal (for chlorophyll concentration and the yellow substance absorption coefficient) and annual (for the particle backscattering coefficient) mean values with their standard deviations in different
regions. It was found that the spatial distributions of bbp in June are noticeably different from other months. Fig. 4 shows the variations of the monthly means of the particle backscattering coefficient in the north-western inner shelf (#2), western (#6) and eastern (#7) open parts, and the eastern shelf (#8). The algorithm description is given at http://optics.ocean.ru. One can see that the Black Sea grows turbid during June, and the increased values of the particle backscattering coefficient bbp are observed both in the western (#2, #6) and the eastern (#7, #8) sub-regions, as in the coastal zone (#2, #8) and in the open sea (#6, #7). During our field studies, the bbp values derived from data, measured by a floating spectroradiometer, were used for validation of the satellite retrieval; the agreement was quite good. 3. Development of a regional algorithm for retrieval of the coccolithophore concentration 3.1. Statistical relationships between Ncoc, TSM and bbp derived from the field data The field studies have provided us with direct evidence of influence of the coccolithophore blooms and the river runoff on the bbp values and their results have shown that the contributions in bbp vary markedly in different years. Fig. 5A demonstrates the statistical relationship between coccolithophore concentration Ncoc and the particle backscattering coefficient bbp derived from field data of 2004–2008. A considerable difference is observed between the slopes of regression lines in 2006 and in other years; for the whole data set the significant correlation exists between Ncoc and bbp (level of significance P b 0.005) but the correlation is weak (R2 = 0.14). The same also applies to the relationship between TSM and bbp: for the whole set R2 = 0.32, but the correlation sorted by year is much stronger (2004 — n = 19, R2 = 0.76; 2005 — n = 13, R2 = 0.95; 2007 — n = 13, R2 = 0.90; but 2006 — n = 16, R2 = 0.36). Appropriate to note that the June spatial distributions of bbp in 2004 and 2006 were quite different from each other (Burenkov et al., 2007). In June 2004, the area of turbid waters was adjacent to the eastern coastal zone of the Black Sea (Fig. 1), while in June 2006, the greatest values of bbp were mainly observed in open sea (Fig. 10 in Section 5). The latter was assumed to be associated with coccolithophore bloom because the area of high values of bbp in 2006 was separated from the coastal zone. To investigate the role of the river runoff and coccolithophore blooms in forming high values of bbp in the eastern part of the Black Sea in June, we collected all field data from 2004 to 2008 for the stations where the data from a floating spectroradiometer, direct determination of coccolithophore concentration, and matching data of satellite measurements were available (n = 72). Then we selected the stations with Ncoc N 0.5 · 106 cells/l (n = 55) and the stations in 2006 (n = 16). The four data sets are presented in Table 1. The following results should be noted from the data in Table 1: • The mean values of bbp for the set #3 and set #4 are almost the same, though the coccolithophore concentrations bNcocN for the set #3 is about 2.3 times more than for the set #4. • The mean value of the absorption coefficient of yellow substance bagN for the set #3 is about 1.3 times less than for the set #4. • A noticeable difference is seen for the correlation between changes of Ncoc and bbp, bbp and ag for the sets #3 and #4; for the set #3 a quite strong correlation is observed between changes of Ncoc and bbp but a weak correlation between bbp and ag; for the set #4 — on the contrary. • For the set #4, there is a significant correlation between changes of Ncoc and ag. These are important conclusions which help to understand the roles of coccolithophore blooms and river runoff in the formation of high values of bbp in June.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
117
Fig. 4. Changes of the monthly means of the particle backscattering coefficient bbp, m−1 in different sub-regions derived from satellite ocean color data in 1998–2011: north-western inner shelf (#2), western (#6) and eastern (#7) open parts, and the eastern shelf (#8) (http://optics.ocean.ru).
Fig. 5. Statistical relationships derived from field data in 2004–2008: A, coccolithophore concentration Ncoc vs. the particle backscattering coefficient bbp; total number of data is 55; crosses — data of 2004–2005 and 2007–2008 (n = 39), circles — 2006 (n = 16). Solid line is the regression line between Ncoc and bbp for all data, the dashed line — data of 2006, the dash-and-dot line — 2004–2005 and 2007–2008. B, Total suspended matter concentration TSM vs. bbp (n = 59). Triangles — 2004, circles — 2005, diamonds — 2006, and crosses — 2007.
118
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Table 1 Parameters of the selected data sets: n is number of data; bbbpN, bNcocN, and bagN are the means of bbp, Ncoc, and ag for each of the set; R2[Ncoc, bbp], R2[ag, bbp], and R2[Ncoc, ag] are the coefficients of determination for the linear regression between Ncoc and bbp, ag and bbp, and Ncoc and ag, respectively. Set
#1 (all data)
#2 (Ncoc N 0.5 · 106 cells/l)
#3 (2006)
#4 (2004, 2005, 2007, 2008)
n bbbpN, m−1 bNcocN, ×106 cells/l bagN, m−1 R2[Ncoc, bbp] R2[ag, bbp] R2[Ncoc, ag]
72 0.0170 1.47 0.111 0.209 0.684 0.036
55 0.0188 1.83 0.119 0.143 0.668 0.007
16 0.0189 3.04 0.098 0.578 0.048 0.110
39 0.0188 1.33 0.128 0.194 0.819 0.352
Table 2 The lowest monthly means of bbp and ag derived from satellite data over 2003–2010. Parameter
n
Interval
Mean value
Standard deviation
bbp, m−1 ag, m−1
24 21
0.0022–0.0028 0.021–0.060
0.0025 0.047
0.0002 0.013
The coefficients Kcoc and Kriv were determined from the statistical characteristics of the sets #3 and #4 in Table 1. As seen from Table 1, their statistical characteristics are significantly different because these data sets relate to different conditions occurring in the eastern part of the Black Sea. The following values of Kcoc and Kriv were obtained: Kcoc = 2.74 · 10−3, Kriv = 0.157.
3.2. Two-parametric model of the particle backscattering coefficient bbp in June
3.3. Algorithm for retrieval of coccolithophore concentration from satellite ocean color data
The results discussed in Section 3.1 clearly indicate that the high values of bbp in June, along with coccolithophore blooms, can be caused by suspended particles coming from the river runoff. These results point to the need for two-parametric model for bbp in June, including not only Ncoc but also a parameter characterizing the river runoff. Since our goal is the development of satellite algorithm, we choose our second parameter so that it can be determined from satellite data. Rivers carry a large amount of suspended particles and colored organic matter into the Black Sea. The latter is characterized by the absorption coefficient ag which can be derived from satellite data. Our idea of evaluation of the “non-coccolithophore” component of bbp (denote it as bb_riv) is based on an assumption that the values of ag and bb_river are correlated with each other, and bb_riv can be represented through ag. As it was shown above (Table 1), this assumption is supported by results of our field studies. The proposed two-parameter model is based on the following assumptions:
If the values of bbp and ag are known, it is easy to calculate from Eq. (11) the coccolithophore concentration Ncoc. In Section 2.2.1, it is shown how to calculate bbp and ag from data measured by a floating spectroradiometer. In the case of MODIS satellite data, the values of the remote sensing radiance reflectance Rrs for two MODIS spectral bands at 488 and 555 nm are taken as the input parameters. They are converted into values of the sub-surface radiance reflectance ρ(λ) by means of a formula by Lee et al. (1998):
• The increase of the particle backscattering coefficient bbp in the period of CB, regarding its background value bbp_bg, is caused by two factors: (i) the scattering by coccolithophore particles bb_сoc; (ii) the augmentation of scattering by particles coming from the river runoff Δbb_riv Δbbp ¼ bbp −bbpXbg ¼ bbXсoc þ ΔbbXriv :
ð8Þ
• The scattering by coccolithophore particles bb_сoc is proportional to a numerical concentration of coccolithophore bbXсoc ¼ K coc N coc ;
ð9Þ
where Ncoc is coccolithophore concentration, Kcoc is the coefficient of proportionality. • The increase of the particle backscattering arising from the river runoff Δbb_riv is proportional to the augmentation of the yellow substance absorption ag regarding its background value ag_bg ΔbbXriv
¼ K riv ag –agXbg ;
ð10Þ
where Kriv is the coefficient of proportionality between Δag and Δbb_riv. Thus, the model has the form: bbp −bbpXbg ¼ K coc Ncoc þ K riv ag –agXbg :
ð11Þ
As the background values, the lowest monthly means of bbp and ag, derived from satellite data over 2003–2010, were selected (Table 2). Interestingly, that 20 of 24 the selected values of bbp were in September–October, 4 — in March–April; 14 of 21 the selected values of ag occurred in June–July, 7 — in August–October.
ρðλÞ ¼ Rrs ðλÞ=½0:165 þ 0:497Rrs ðλÞ:
ð12Þ
Next, we use Eqs. (2)–(6) and (8)–(11), taking into account the spectral dependences of the backscattering of coccolithophore and river particles (Voss, Balch, & Kilpatrick, 1998; Kopelevich, 2012): −1:2
bbXсoc ðλÞ ¼ K coc ðλ=555Þ ΔbbXriv
N coc ;
ð13Þ
−0:5 ¼ K riv ðλ=555Þ ag –agXbg :
ð14Þ
With Rrs(488) and Rrs(555) as the input parameters and by using the above formulae, we can obtain a linear system of two equations with two unknowns ag and Ncoc f 1 ðλi Þag þ f 2 ðλi ÞNcoc ¼ g ðλi Þ;
i ¼ 1; 2
ð15Þ
where expressions for the functions f1(λi), f2(λi), and g(λi) are derived from Eqs. (3)–(6) and (9)–(14). Stability of the algorithm is provided by an appropriate choice of the spectral bands (in particular by large difference in the absorption coefficients of pure water — 0.0145 m−1 at 488 nm and 0.060 m−1 at 555 nm) and high values of the particle backscattering in our case. 4. Validation of the developed algorithm 4.1. Testing the bb_riv model One of two main assumptions, which are the basis of the proposed model and algorithm, is the relationship between bb_riv and the absorption coefficient of yellow substance ag derived from data on the spectral radiance reflectance (ρ(λ) measured by a floating spectroradiometer or Rrs(λ) from satellite data). It is important to clarify that for our problem we do not need a precise knowledge of absolute values of ag — we use ag as a parameter allowing us to evaluate bb_riv from remote sensing data. Since we cannot directly test the relationship between bb_riv and ag, the data on the terrigenous component TSM_trg of TSM, derived from the geochemical study (see Section 2.2.3), were used for the testing. Fig. 6 shows the regression line between TSM_trg and ag.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 6. Relationship between the terrigenous component TSM_trg and ag (n = 33, R2 = 0.86) derived from field data in 2005–2007.
Unfortunately, TSM_trg was not determined at all in 2004, and it was not determined in all samples in 2005–2007. But the considered set is sufficiently representative, because it involves the years with different conditions (see Fig. 5). The regression equation between TSM_trg and ag has the form TSMXtrg ¼ 13:8ag −1:21:
119
Fig. 8. The regression between the coccolithophoride concentration Ncoc (106 cells/l) and the particulate inorganic carbon PIC (mol/m3), derived from concurrent in situ and MODIS data in the eastern part of the Black Sea in June 2004–2008 (n = 35, R2 = 0.367).
The regression equations between Ncoc_calc and Ncoc_meas have the following form NcocXcalcXfield ¼ 0:83NcocXmeas þ 0:57;
ð17Þ
NcocXcalcXmodis ¼ 0:87N cocXmeas þ 0:55:
ð18Þ
ð16Þ
This equation is valid for ag N 0.088; TSM_trg was taken to be 0 at lower values of ag. The correlation coefficient between TSM_trg and ag equals 0.927; error of the regression sregr = 0.30 mg/l. Such value of sregr is quite acceptable for quantitative assessment of the terrigenous component TSM_trg using the parameter ag, derived through values of the spectral radiance reflectance.
4.2. Direct verification of Ncoc calculation with the developed algorithm Verification of the Ncoc values calculated with the developed algorithm was performed by using the in situ measured coccolithophore concentrations. Fig. 7 shows the scatterplots Ncoc_cacl vs. Ncoc_meas for two cases: the first — calculation with data measured by a floating spectroradiometer, the second — from the satellite data. All appropriate data of 2004–2008 were used.
As seen, Eqs. (17) and (18) are similar to each other; the correlation coefficients are equal to 0.764 and 0.730, respectively; standard errors 1.15 and 0.99 · 106 cells/l. The smaller error in the second case is related to lower concentration of Ncoc_meas: the average value in the first case was 1.42 · 106 cells/l (n = 47), in the second — 1.17 · 106 cells/l (n = 38). The accuracy of calculation is not high, but it is important that the calculation from satellite data is quite stable, and there is no significant increase in errors compared with calculation on the field data. One of the main reasons of high error in the calculated values of Ncoc is variability of the ratio Nlith/Ncoc between concentrations of the detached coccoliths Nlith and the plated cells Ncoc. This problem was considered for the Barents Sea, and it was shown that this ratio changed about 20 times but the coefficient Kcoc — only 3 times (http://www. sequoiasci.com/research/pie2012.cmsx). In the Black Sea, great variations of the ratio Nlith/Ncoc were also observed, and the coefficient Kcoc in Eq. (9) is its optimal value from the statistical evaluation (see Section 3.2).
Fig. 7. Scatterplots Ncoc_calc vs. Ncoc_meas: A, Ncoc_cacl_field is calculated from data on ρ(λ) measured by a floating spectroradiometer (n = 47, R2 = 0.583), B, Ncoc_cacl_modis — from MODISAqua data (n = 38, R2 = 0.534).
120
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 9. Comparison between the spatial distributions of Ncoc calculated with the new regional algorithm (A) and of PIC (B) derived from MODIS-Aqua data on 12 June 2004.
Fig. 10. Changes in the spatial distributions of the particle backscattering coefficient bbp, coccolithophoride concentration Ncoc, and the ratio rp = bb_riv / bbp in the Black Sea in May–June 2006 derived from MODIS-Aqua.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
121
Fig. 11. Changes of the June means of the coccolithophore concentration Ncoc (left) and the ratio rp = bb_riv / bbp (right) in the eastern (#7) open part and the eastern shelf sub-region (#8).
4.3. Comparison between the developed algorithm and Gordon, Balch algorithm
5.1. Changes in the spatial distributions of bbp, Ncoc and in the bb_riv/bbp ratio in May–June 2006
It is interesting to compare the results derived with our algorithm (Ncoc) and by Gordon, Balch algorithm (http://oceancolor.gsfc.nasa. gov/seadas/) providing calcite concentration (particulate inorganic carbon — PIC data product). Fig. 8 shows the relationship between Ncoc and PIC derived from concurrent in situ and MODIS data in the eastern part of the Black Sea in June 2004–2008 (n = 35, R2 = 0.367). First of all note, that the correlation coefficient is equal to ~0.61, and a significant correlation exists between Ncoc and bbp (level of significance P b 0.0005). The regression equation has the form
These changes are shown in Fig. 10. It is seen how the turbidity was growing since the first half May, mainly in the western (near Danube) and eastern (Caucasus rivers) coastal zones. In the period 6–15 May, the contribution bb_riv from the river runoff to bbp was more than 0.9 almost all over the Sea, and only in two small areas in the eastern part, where CB originated, it decreased below 0.6. In the period 6–15 June, on the contrary, a strong CB covered a large part of the Sea, and rp = bb_riv / bbp was below 0.4 and even 0.2. Comparing the distributions of bbp and Ncoc in this period with those for 2004 (Figs. 1В and 9А), one can see significant difference. First, in 2004, the highest values of bbp were observed in the sub-region #2, under the Danube runoff, and in the coastal zone of the eastern half of the Sea (Fig. 1В), whereas in 2006, they were observed in the open sea (Fig. 10). Second, in 2004, CB was limited to the eastern part of the Sea (Fig. 9А), whereas in 2006, it covered almost all of the Sea (Fig. 10). These distinctions for the eastern part of the Sea have already been noted in Section 3.1 according to the data of field measurements.
Ncoc ¼ 223PIC þ 0:06;
ð19Þ
the regression error Sregr is equal to ~0.7 · 106 cells/l. Eq. (19) allows us to estimate the PIC boundary value, corresponding the accepted lowest value of the coccolithophore bloom Ncoc = 1 · 106 cells/l; the value of PIC (with account of error) is in the range 0.001–0.007 mol/m3 with the mean value of 0.004 mol/m3. It is interesting that the close value is obtained from data of Balch, Kilpatrick, Holligan, Harbour, and Fernandez (1996) for central North Atlantic and Gordon et al. (2001) south of Plymouth; the boundary value of PIC was found to be ~0.004 mol/m3 (under realistic for the Black Sea assumption that the ratio Nlith/Ncoc is equal to 10). In Fig. 9, the spatial distributions of Ncoc, calculated with the new regional algorithm (A), and of PIC, derived from MODIS-Aqua data on 12 June 2004 (B), are compared. It is also interesting to compare both distributions with the bbp distribution in Fig. 1B. One can see that the PIC distribution agrees very well with the bbp distribution, whereas some distinction is observed with the Ncoc distribution (it should be noticed that the both algorithms failed in the western part of the Black Sea). If we accept the CB boundaries as 1 · 106 cells/l for Ncoc and 0.004 mol/m3 for PIC, it is seen that they are well matched. In the eastern part, the exceptions are observed near the Kerch Strait and in the eastern and, in particular, the southern coastal zones where the high values of PIC are not associated with coccolithophore blooms. 5. The algorithm application The main achievement of the new algorithm is that it can distinguish the changes, associated with coccolithophore blooms and with the river runoff effect, although the accuracy is not very high. Some examples of such separation are given below.
5.2. Inter-annual changes in the coccolithophore concentration and in the terrigenous component of suspended matter in June from satellite ocean color data in 1998–2011 Fig. 11 shows the changes in the June means of the coccolithophore concentration Ncoc and the ratio rp = bb_riv / bbp in the eastern (#7) open part and the eastern shelf sub-region (#8). As seen, there are years with no marked coccolithophore blooms (2001, 2003, and 2010); it was shown by Burenkov, Kopelevich, and Sheberstov (2011) that the inter-annual changes in the intensity of coccolithophore blooms can be linked to the winter sea surface temperature. But the higher values of bbp were observed in the above mentioned years (Fig. 4), and, according to our model, they should be attributed to the bb_riv contribution. It is seen from Fig. 11, the bb_riv contribution to the particle backscattering bbp in the sub-region #7 was more than 50% in all years except 2006 and 2008; in the sub-region #8 this contribution was more than 60% in all years. In open sea (sub-region #7) the lowest value of rp was in 2006 (less than 0.4), the highest in 2001 and 2003 (more than 0.9). Fig. 12 demonstrates variability of the mean June distributions of the coccolithophore concentration Ncoc, as well as, for comparison, of the particulate inorganic carbon PIC and of the ratio rp = bb_riv / bbp derived from satellite ocean color data with the new algorithm. We applied our algorithm to the whole basin while it was designed for the eastern part of the Black Sea. Unfortunately, we have not been able to
122
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
Fig. 12. Changes of the June mean distributions of Ncoc, PIC and the ratio rp = bb_riv / bbp from satellite ocean color data in 1998–2011.
O. Kopelevich et al. / Remote Sensing of Environment 146 (2014) 113–123
validate it by data from field measurements in the western part, so the correspondence to reality of the constructed distributions in that part is questionable. Some evidence of that the presented distributions of Ncoc are true, at least qualitatively, follows from the results of comparison between the distributions of Ncoc and PIC (this issue was discussed in Section 4.3). 6. Conclusion The main result of the present work is development of a new algorithm for retrieval of coccolithophore concentration in the Black Sea from satellite ocean color data. The algorithm is based on the twoparametric model allowing take into account the “non-coccolithophore” share of the particle backscattering coefficient bbp. In June, when coccolithophore blooms are observed, an origin of such particles is mainly river runoff. Application of the new algorithm to satellite ocean color data in 1998–2011 gave an opportunity to separate changes associated with coccolithophore bloom and with the river runoff effect. It was found that there were the years with no marked coccolithophore bloom (2001, 2003, and 2010). The coccolithophore blooms are more changeable in comparison with the river runoff which is more stable. Possible reasons of the inter-annual changes in the intensity of coccolithophore blooms in the Black Sea were discussed before (Burenkov et al., 2011). The results obtained can be considered only as a first step, perhaps the most important, on the way to the creation of satellite algorithm for quantitative assessment of the intensity of coccolithophore blooms in the Black Sea. It is necessary to solve several serious problems, one of them is variability of the components of coccolithophore bloom. It includes coccoliths, plated cells, naked cells and others; these components have quite different specific backscattering coefficients, and the ratio of their contributions can be very changeable. One of possible solution is use of calcite concentration as a parameter to be determined, as it is done in the Gordon, Balch algorithm. Another problem is how to distinguish the coccolithophore bloom from the blooms of other mass algae, such as diatoms, by using satellite data. This is very hard task, but it is very important for assessment of the carbon cycle and activity of the biological pump in the Black Sea. Solution to these problems is impossible only on the satellite data; they must be combined with data from the direct measurements of optical and biogeochemical characteristics. Acknowledgment The work was supported by the Program for Fundamental Research No.23, Russian Academy of Sciences, by grants for the NATO project ESP.EAP.SFPP 982678 “Bio-Optical Characterization of the Black Sea for Remote Sensing Applications”, and by grant Nos. 10-05-00936 and 1305-00618 from the Russian Foundation for Basic Research. SeaWiFS and MODIS data used were obtained from the Goddard Distributed Active Archive Center under the auspices of the National Aeronautics and Space Administration. The authors thank A.A. Klyuvitkin for the data on terrigenous component of the total suspended matter, and two anonymous reviewers for very helpful comments. References Artemiev, V. A., Burenkov, V. I., Vortman, M. I., Grigoriev, A. V., Kopelevich, O. V., & Khrapko, A. N. (2000). Sea-truth measurements of ocean color: A new floating spectroradiometer and its metrology. Oceanology, 40, 139–145 (translated from Okeanologiya, 40, 148–155).
123
Balch, W. M., Kilpatrick, K., Holligan, P. M., Harbour, D., & Fernandez, E. (1996). The 1991 coccolithophore bloom in the central north Atlantic II: Relating optics to coccolith concentration. Limnology and Oceanography, 41, 1684–1696. Bricaud, A., Babin, M., Morel, A., & Claustre, H. (1995). Variability in the chlorophyll-specific absorption coefficients of natural phytoplankton: Analysis and parametrization. Journal of Geophysical Research, 100(C7), 13,135–13,142. Burenkov, V. I., Ershova, S. V., Kopelevich, O. V., Sheberstov, S. V., & Shevchenko, V. P. (2001). An estimate of the distribution of suspended matter in the Barents Sea waters on the basis of the SeaWiFS satellite ocean color scanner. Oceanology, 41, 622–628 (translated from Оkeanologiya, 2001, 41, 653–659). Burenkov, V. I., Kopelevich, O. V., & Sheberstov, S. V. (2011). Seasonal and interannual variations of the biooptical characteristics of the Black Sea from satellite data. Current problems of remote sensing from Space, 8(4), 190–199. Burenkov, V., Kopelevich, O., Sheberstov, S., Prokhorenko, O. V., Klyuvitkin, A. A., & Pautova, L. A. (2007). The June maxima of the particle backscatter in the Black Sea: Results of three years satellite and field studies. Proceedings of IV International Conference “Current Problems in Optics of Natural Waters (ONW'2007)” (pp. 157–160). Nizhny Novgorod: Institute of Applied Physics RAS. Cokacar, T., Kubilay, N., & Oguz, T. (2001). Structure of E. huxleyi blooms in the Black Sea surface waters as detected by SeaWiFS imagery. Geophysical Research Letters, 28, 4607–4610. EPA Method 445.0 (1997). In vitro determination of chlorophyll “a” and pheophytin “a” in marine and freshwater algae by fluorescence. National Exposure Research Laboratory, U.S. Environmental Protection Agency (Revision 1.2, Sept. 1997). Gel'man, E. M., & Starobina, I. Z. (1976). Photometric methods for determining the rock forming elements in ores, rocks and minerals. Moscow, SSSR: Central Laboratory, Sector of Chemical Analytic Methods GEOKhI AN [in Russian]. Gordon, H. R., & Balch, W. M. (1999). MODIS detached coccolith concentration. Algorithm theoretical basis document (30 pp.). Gordon, H., Boynton, G., Balch, W., Groom, S., Harbour, D., & Smyth, T. (2001). Retrieval of coccolithophore calcite concentration from SeaWiFS imagery. Geophysical Research Letters, 28, 1587–1590. Holm-Hansen, O., & Riemann, B. (1978). Chlorophyll-a determination: improvements in methodology. Oikos, 30, 438–447. International Ocean Color Coordinating Group (IOCCG) (2000). Remote sensing of ocean colour in coastal, and other optically-complex, waters. In S. Sathyendranath (Ed.), Reports of the International Ocean-Colour Coordinating Group, No. 3. IOCCG, Dartmouth, Canada (140 p.). Khrapko, A. N., Kopelevich, O. V., Burenkov, V. I., Grigoriev, A. V., & Terekhova, A. A. (2007). New instrument for measuring surface and underwater irradiances. Proceedings of IV International Conference “Current Problems in Optics of Natural Waters (ONW'2007)” (pp. 271–275). Nizhny Novgorod: Institute of Applied Physics RAS. Kopelevich, O. V. (2012). Application of data on seawater light scattering for the study of marine particles: A selective review focusing on Russian literature. Geo-Marine Letters. http://dx.doi.org/10.1007/S00367-012-0278-y.2012. Lee, Z., Carder, K. L., Mobley, C. D., Steward, R. G., & Patch, J. S. (1998). Hyperspectral remote sensing for shallow waters. 1. A semianalytical model. Applied Optics, 37, 6329–6338. Mikaelyan, A. S., Silkin, V. A., & Pautova, L. A. (2011). Coccolithophorids in the Black Sea: Their interannual and long-term changes. Oceanology, 51, 39–48 (published in Okeanologiya, 2011, 51, 45–53). Mikhailov, V. N., Morozov, V. N., Cheroy, N. I., Mikhailova, M. V., & Zavyalova, E. F. (2008). Extreme flood on Danube in 2006. Meteorology and Hydrology, 1, 80–89. Taylor, S. R., & McLennan, S. M. (1985). The continental crust: Its composition and evolution. Oxford: Blackwell (312 p.). Thierstein, H. R., & Young, J. R. (2004). Coccolithophores — From molecular processes to global impact. Springer (565 p.). Voss, K. J., Balch, W. M., & Kilpatrick, K. A. (1998). Scattering and attenuation properties of Emiliania huxleyi cells and their detached coccoliths. Limnology and Oceanography, 43, 870–876.
Web references Sampling and sample-handling protocols for GEOTRACES cruises. (Version 1.0, Edited by the 2010 GEOTRACES Standards and Intercalibration Committee). http://www. geotraces.org/. (December 2010). Feldman, G. C. (2012a). Ocean color products and algorithms. http://oceancolor.gsfc.nasa. gov/seadas/ Feldman, G. C. (2012b). Ocean color web. http://oceancolor.gsfc.nasa.gov Kopelevich, O. V., Burenkov, V. I., Sheberstov, S. V., & Vazyulya, S. V. (2012). Remote sensing of coccolithophore particles: Regional features. PiE 2012 Particles in Europe. Program and Abstract Volume. Barcelona, Spain: ICM (http://www. sequoiasci.com/research/pie2012.cmsx) Kopelevich, O. V., Burenkov, V. I., Vazyulya, S. V.(2011). Bio-optical characteristics of the Barents, White, Black, and Caspian seas from data of satellite ocean color scanners. http://optics.ocean.ru