Influence of phytoplankton pigment composition on remote ... - cedex

Remote Sensing of Environment 106 (2007) 414 – 427 www.elsevier.com/locate/rse

Influence of phytoplankton pigment composition on remote sensing of cyanobacterial biomass Stefan G.H. Simis a,⁎, Antonio Ruiz-Verdú b , Jose Antonio Domínguez-Gómez b , Ramón Peña-Martinez b , Steef W.M. Peters c , Herman J. Gons a a

Netherlands Institute of Ecology (NIOO-KNAW), Centre for Limnology, Rijksstraatweg 6, 3631 AC Nieuwersluis, The Netherlands b Centre for Hydrographic Studies (CEDEX), Paseo Bajo de la Virgen del Puerto 3, E-28005, Madrid, Spain c Institute for Environmental Studies (IVM), Vrije Universiteit, De Boelelaan 1087, 1081 HV Amsterdam, The Netherlands Received 27 March 2006; received in revised form 29 July 2006; accepted 9 September 2006

Abstract An extensive field campaign was carried out for the validation of a previously published reflectance ratio-based algorithm for quantification of the cyanobacterial pigment phycocyanin (PC). The algorithm uses band settings of the Medium Resolution Imaging Spectrometer (MERIS) onboard ENVISAT, and should accurately retrieve the PC concentration in turbid, cyanobacteria-dominated waters. As algae and cyanobacteria often co-occur, the algorithm response to varying phytoplankton composition was explored. Remote sensing reflectance and reference pigment measurements were obtained in the period 2001–2005 in Spain and the Netherlands using field spectroradiometry and various pigment extraction methods. Additional field data was collected in Spain in May 2005 to allow intercalibration of spectroradiometry and pigment assessment methods. Two methods for extraction of PC from concentrated water samples, and in situ measured PC fluorescence, compared well. Reflectance measurements with different field spectroradiometers used in Spain and the Netherlands also gave similar results. Residual analysis of PC predicted by the algorithm showed that overestimation of PC mainly occurred in the presence of chlorophylls b and c, and phaeophytin. The errors were strongest at low PC relative to Chl a concentrations. A correction applied for absorption by Chl b markedly improved the prediction. Without such a correction, the quality of the PC prediction still increased markedly with estimates N50 mg PC m− 3, allowing monitoring of the cyanobacterial status of eutrophic waters. The threshold concentration may be lowered when a high intracellular PC:Chl a ratio or cyanobacterial dominance is expected. Below the limit, predicted PC concentrations should be considered as the highest estimate. We evaluated that remote sensing of both PC and Chl a would allow assessment of cyanobacterial risk to water quality and public health in over 70% of our cases. © 2006 Elsevier Inc. All rights reserved. Keywords: Remote sensing; Phytoplankton; Cyanobacteria; Phycocyanin; Chlorophyll; MERIS

1. Introduction The management of inland water quality is a growing concern wherever anthropogenic eutrophication affects water bodies. Potentially toxic cyanobacterial blooms pose a health threat and are detrimental to the economic and environmental value of many lakes and reservoirs. Regular monitoring of water bodies is necessary to provide timely warnings in case of cyanobacterial bloom. Remotely sensed water quality products

⁎ Corresponding author. Tel.: +31 294 239327; fax: +31 294 232224. E-mail address: [email protected] (S.G.H. Simis). 0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2006.09.008

have the potential to provide high spatial and temporal coverage, while conventional sampling methods lack both. Detection of cyanobacterial biomass can be based on the absorption feature of the pigment phycocyanin (PC), used for light harvesting and only present in considerable concentrations in cyanobacteria. The absorption signal of most modifications of cyanobacterial PC is strongest around 615 nm (Bryant, 1981) and can be detected from reflectance data of eutrophic, turbid water bodies (Dekker et al., 1991; Gons et al., 1992; Jupp et al., 1994; Kutser et al., 2006). A number of reflectance spectra, collected for this study, from which the PC absorption feature can be observed have been plotted in Fig. 1. This figure also shows the location of wavebands used in pigment retrieval

S.G.H. Simis et al. / Remote Sensing of Environment 106 (2007) 414–427

Fig. 1. Rrs(k) spectra representing the variation in spectral shape and magnitude of reflectance data encountered in the dataset. MERIS bands used to obtain PCRAD are indicated by vertical bars (centered at 620, 665, 708.75 and 778.75 nm). The trough around 625 nm is primarily caused by pigments PC and Chl a. The trough around 675 nm is attributed to Chl a. Spectra from L. 1' Albufera (Spain) corresponded with pigment measurements PCFL=120–728 and Chl a 29–439 mg m−3. PCFL and Chl a measurements for the Rosarito spectra displayed here were both in the 35–80 mg m−3 range. PCFT and Chl a measurements of the displayed L. IJsselmeer samples ranged 20–329 and 26–109 mg m−3, respectively.

algorithms described below. The first attempts at quantifying PC from spectral reflectance were semi-empirical baseline (Dekker, 1993) and band ratio (Schalles & Yacobi, 2000) algorithms, targeting the absorption feature caused by the presence of PC and chlorophyll a (Chl a) in the 620–625 nm region. Variable PC:Chl a ratios are not accounted for by these algorithms, and current satellite sensors with global coverage do not offer all required wavebands. The Medium Resolution Imaging Spectrometer (MERIS) onboard the ENVISAT mission is the first sensor to offer a combination of several narrow wavebands to target both Chl a and accessory pigment absorption in the red spectral region, at a spatial resolution (260 m across track) sufficient for mediumsized water bodies, and with a satisfactory signal-to-noise ratio. Algorithms based on ratios of reflectance in the red and nearinfrared (NIR) spectral region can be used to retrieve photosynthetic pigment concentrations in turbid inland water bodies, where often the most severe water quality problems are experienced. For optical remote sensing of the concentration of the main photosynthetic pigment chlorophyll a, the ratio of a NIR band (around 700–710 nm) over a band near the red absorption maximum of Chl a at 675 nm (Mittenzwey et al., 1992) has been successfully applied to a wide range of turbid water bodies for N 10 mg Chl a m− 3 (Gons, 1999; Gons et al., 2000). The band ratio targeting Chl a absorption thus serves as an indicator for phytoplankton biomass (algae and cyanobacteria) in productive water bodies, and can be used with MERIS imagery (Gons et al., 2005). Recently, a semi-analytical, nested band ratio-based algorithm for the retrieval of PC was proposed, using only MERIS

415

bands (Simis et al., 2005; refer to Eqs. (5A) and (5B) below). This PC algorithm attributed the absorption signal derived from the 620-nm band to two absorption features that are dominant in cyanobacteria-rich waters: the absorption peak of PC and the shoulder to red Chl a absorption located around 623–628 nm (Bidigare et al., 1990; Ficek et al., 2004; Sathyendranath et al., 1987). The absorption by Chl a in this waveband was obtained from a nested band ratio of bands at 708.75 and 665 nm, subsequently using a fixed conversion factor describing the in vivo absorption of Chl a between 620 and 665 nm. In this way, the absorption at 620 nm could be corrected for Chl a and the remaining absorption was attributed to PC. Absorption by dissolved substances, suspended sediments, and detrital matter could contribute a nearly flat background absorption to the 620 nm band. Spectrally neutral absorption by these substances is not expected to be a major influence on the retrieval of absorption through a band ratio-based algorithm, as long as the used bands are located sufficiently close together, and the reflectance spectrum has a high amplitude. However, phytoplankton pigments that absorb light in the red bands but not in the NIR band are prone to have an influence on the estimation of absorption at 620 nm, and inherently, on the estimation of PC. This effect should be particularly clear at relatively low PC concentrations. All chlorophyllous pigments and their degradation products absorb light in the red spectral region. The discussed PC retrieval algorithm currently only corrects for absorption by Chl a. For most pigments, absorption around 620 nm only amounts to a fraction of their absorption at longer wavelengths. However, the diatom pigments chlorophyll c1 and c2 [Chl c] have absorption maxima on both sides of the 620-nm band (Ficek et al., 2004; Jeffrey et al., 1997). Chl c is therefore likely to lead to overestimations of the PC concentration as it is not corrected for by the algorithm. The pigment chlorophyll b [Chl b] is another major accessory photosynthetic pigment in inland water bodies, mainly present in green algae and prochlorophytes. Chl b has a broad absorption maximum around 650 nm and a minor feature around 600 nm (Ficek et al., 2004; Jeffrey et al., 1997; Sathyendranath et al., 1987) and could lead to overestimates of the absorption in both the 665-nm and 620-nm wavebands. Degraded chlorophyll forms may also contribute to absorption in these bands. During the development of the PC algorithm, overestimations of the PC concentration were indeed observed mainly at mixed-phytoplankton sites, but could not be attributed to any particular phytoplankton composition due to lack of sufficient data (Simis et al., 2005). This paper reports on the influence of the presence of major phytoplankton groups on the retrieval of PC by the mentioned reflectance band ratio-based algorithm. Field spectroradiometric and pigment data were gathered from a range of water bodies in Spain and the Netherlands with a varying share of PC-containing cyanobacteria in the phytoplankton. Errors in PC retrieval were regressed against the presence of marker pigments for different algal groups. The aim of the study was to identify situations where the PC algorithm for MERIS bands can be applied, and situations where additional information is needed because of interference by red-absorbing algal pigments. The results may be

416


used in the spectral design of future sensors, to define the boundary conditions for operational monitoring of cyanobacteria using remote sensing, and to aid the development of improved or novel algorithms from (hyper)spectral reflectance data. 2. Method 2.1. Study sites An overview of sampling sites and the instrumentation that was used at each site is given in Table 1. The Spanish water bodies were visited in the period 2001–2005 and represent a range of 57 lakes and reservoirs from all parts of the country. The water bodies ranged widely in trophic state (observed Chl a range 0.5–305 mg m− 3), surface area (b 1–103 km2), depth (0.9– 104 m, measured at the sampling location), and transparency (secchi disk depth 0.2–9.6 m). Many of the deeper lakes were vertically stratified, so care was taken to take samples that were representative of the layer up to the first optical depth, the maximum penetration depth of 90% of the remotely sensed light (Gordon & McCluney, 1975). In turbid lakes (secchi disk depth b 1 m), samples were taken from the surface layer. Monospecific blooms of algae or cyanobacteria are common in several of the water bodies from spring onward, but several deep and clear lakes without substantial cyanobacterial growth were also included. Many of the lakes are subject to eutrophication, while serving a function of drinking water supply and irrigation resource. In the Netherlands, Lake IJsselmeer and Lake Ketelmeer were visited 6 times for 2-day cruises in 2004 and 2005. This turbid, eutrophic lake system is the largest (1190 km 2 ) freshwater body of Western Europe and exhibits a seasonal phytoplankton distribution where cyanobacteria are more abundant towards the end of summer, but substantial biomass can be present in diatoms, green algae, prochlorophytes, and cryptophytes. Cyanobacteria, mainly species of Aphanizomenon and Microcystis can build up bloom patches at the water surface of L. IJsselmeer during calm summer days, but the shallow lake (average depth 4.4 m) is usually well mixed by wind which causes resuspension of settled organic material and sediments and low transparency (average secchi disk depth 1.0 m). It is common to find large horizontal differences in phytoplankton pigment composition. The smaller L. Ketelmeer serves as the input of nutrient-rich river water to L. IJsselmeer and is characterized by high suspended sediment loading and the presence of diatom species. 2.2. Radiometric measurements Remote sensing reflectance Rrs(λ) was defined here as the MERIS level-2 standard product ‘normalized water-leaving reflectance’ (Montagner, 2001): Rrs ðkÞ ¼ ½qw N ð0þ ; kÞ ¼ kLw ð0þ ; kÞ=Ed ð0þ ; kÞ

ð1Þ

where Lw(0+,λ) is water-leaving radiance corrected for diffuse sky light reflected at the water surface, Ed(0+,λ) is downward

irradiance, depth 0+ points to the situation just above the water surface and wavelength dependence is denoted by λ. These quantities can all be obtained from shipboard measurements. In that case, Lw(0+,λ) is calculated from Lw ð0þ ; h; u; kÞ ¼ Lwu ð0þ ; h; u; kÞ−½rðhÞLsky ð0þ ; h; u; kÞ

ð2Þ

where Lwu(0+,θ,φ,λ) is water-leaving radiance uncorrected for reflectance of downwelling light at the water surface; this correction is given by the bracketed term. Sky radiance is denoted Lsky(0+,θ,φ,λ) and measured at an angle θ away from the zenith axis. For small instrument acceptance angle there is negligible dependence of Lwu on viewing angle θ (the same angle from zenith as for Lsky, but mirrored on the horizontal plane) and azimuth angle φ (away from the sun's azimuth), as long as θ b 42°, and for φ ranging 90°–135° (Morel & Gentili, 1993; Tyler, 1960). The measurement geometry was kept within these limits and sun zenith angles N60° were avoided in all cases. The skylight correction factor r(θ) can be approached by a constant value for calm water surfaces, or obtained as a function of wind speed and cloud cover (Mobley, 1999). The irradiance Ed(0+,λ) can be acquired with a cosine collector, or by measuring the radiance Ld from an intercalibrated spectrally neutral diffuse plate. Different field spectroradiometers were used to acquire the Spanish and Dutch datasets, and each differed in the way the radiance quantities in Eqs. (1) and (2) were measured. Details on the different optical configurations, instrument characteristics and measurement parameters can be found in Table 2. For both measurement methods, a minimum of 3 Rrs spectra was produced. After removal of invalid Rrs spectra due to the capture of sun glint or variable cloud cover during the measurement cycle, remaining spectra were averaged. 2.3. Pigment analysis Samples for pigment analysis were always taken from the surface water layer in shallow, turbid lakes, and from the first optical depth layer (Gordon & McCluney, 1975) in vertically stratified likes. For phycobilin pigments there is no standard method of extraction. Two different methods were used for cell disruption and extraction of PC and phycoerythrin (PE), and concurrent measurements are available in the current dataset for their intercalibration. In addition, for samples taken and analysed in Spain, PC fluorescence was measured in situ, providing a third PC reference to the spectroradiometric approach. The first method for phycobilin extraction was based on repeated freezing and thawing cycles of samples. PC concentrations resulting from this method will be referred to as PCFT. The procedure was based on the work by Sarada et al. (1999) and was adapted for the current study from Simis et al. (2005). In short: fresh water samples stored at 0 °C for b48 h were concentrated by high-speed centrifugation, suspended in a phosphate buffer of pH 6.7, and frozen and thawed nine times at − 20 °C and room temperature, respectively, while kept in the dark. Note that the phosphate buffer used here was of pH 6.7 rather than pH 7.4 as used before (Simis et al., 2005), which resulted in up to 28% higher extraction yield from cultured Limnothrix sp. and

Table 1 Summary of sampled lakes and reservoirs Location name

Position

Depth

Lat.

Lon.

d.dd

d.dd

m (avg.)

Surface area 2

km

Elevation m a.s.l.

Chl a

Secchi depth −3

mg m (avg.)

Visitsa

m (avg.)

Samples

Spectroradiometry samples

Total

PR-650

Pigment samples ASD-FR

HPLC

PC FT

MG

FL

52.73 52.6

5.39 5.72

4.4 3.0

1190 35

0 0

43.9 12.5

0.8 1.0

6 6 6

290 256 34

289 256 33

0 0 0

203 187 16

201 186 15

75 68 7

0 0 0

Spain Aguilar R. Alarcón R. L. l'Albufera Alcántara R. Alcorlo R. Almendra R. Los Arroyos R. El Atazar R. Beniarrés R. Bornos R. Brovales R. Buendía R. Burguillo R. El Campillo Lgn. Canelles R. Castrejón R. Castro R. Cernadilla R. Cijara R. Contreras R. Cortes R. Ebro R. Entrepeñas R. Finisterre R. Giribaile R.

42.81 39.6 39.34 39.73 41.02 41.24 40.59 40.91 38.81 36.82 38.35 40.4 40.43 40.32 42.01 39.83 39.81 42.02 39.34 39.61 39.25 42.98 40.51 39.65 38.09

− 4.31 − 2.16 − 0.35 − 6.61 − 3.03 − 6.28 − 4.05 − 3.49 − 0.36 − 5.72 − 6.7 − 2.77 − 4.57 − 3.50 0.64 − 4.3 − 3.75 − 6.47 − 4.93 − 1.53 − 0.93 − 4.04 − 2.73 − 3.65 − 3.49

14.1 17.4 0.9 47.0 27.0 71.3 n/a 49.5 14.0 13.1 n/a 40.0 30.5 n/a 104.0 n/a n/a 38.0 28.5 27.0 15.0 4.9 25.0 n/a 26.0

18 69 24 103 7 79 b1 10 2 22 1 80 9 b1 16 5 b1 13 74 27 4 62 34 5 23

942 806 1 218 920 730 845 870 318 104 303 712 729 538 506 425 559 889 428 669 326 838 718 686 346

6.6 3.5 304.9 4.4 1.6 13.2 24.8 4.5 57.0 6.2 53.8 0.5 18.9 37.2 1.0 144.8 48.2 3.1 1.4 1.6 1.5 52.8 2.1 14.7 2.9

1.4 2.4 0.2 3.5 2.1 2.7 0.0 7.2 1.0 0.9 0.6 9.6 2.1 0.7 6.0 0.5 0.3 3.3 5.2 2.6 4.0 0.9 4.5 0.7 2.3

89 2 2 4 1 1 2 1 3 1 1 1 1 2 1 1 1 1 1 1 1 1 2 1 1 1

193 4 4 20 2 1 3 1 4 1 2 1 1 4 2 4 1 1 1 2 2 1 3 1 1 1

23 0 0 7 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0

170 4 4 19 2 1 3 0 4 1 2 0 1 4 0 4 0 0 1 2 2 1 3 1 0 1

169 4 4 16 2 1 3 1 4 1 2 1 1 4 2 4 1 1 1 2 2 1 2 1 1 1

16 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 1 1 0 0 0 0 0 0 1 0

122 4 4 14 2 1 3 1 4 1 2 1 1 3 2 0 1 1 1 2 2 1 3 1 1 1

168 4 4 15 2 1 3 0 3 1 2 1 1 4 0 4 0 0 1 2 2 1 3 1 0 1


The Netherlands L. IJsselmeer L. Ketelmeerb

(continued on next page)

417

418

Table 1 (continued) Location name

Depth

Lat.

Lon.

d.dd

d.dd

36.66 38.17 36.96 39.8 37.27 38.24 40.53 40.04 37.57 40.76 40.33 40.94 40.30 40.30 40.31 41.96 42.96 41.61 40.1 40.38 42.12 40.61 40.72 38.9 42.07 42.21 42.93 39.8 40.55 41.98 38.31 38.76

− 5.73 − 3.47 − 4.83 − 4.09 − 4.34 − 3.95 − 3.94 − 5.12 − 2.9 − 3.62 − 4.25 − 3.73 − 3.53 − 3.52 − 3.52 1.22 − 5.05 − 5.94 − 5.3 − 4.33 − 6.72 − 5.61 − 3.84 − 5.19 0.89 0.95 − 2.59 − 5.5 − 4.05 − 6.29 − 6.66 − 3.79

m (avg.)

29.5 22.0 21.0 n/a 44.3 56.0 n/a n/a 28.5 n/a n/a 13.1 n/a n/a n/a 14.2 31.5 32.9 7.9 43.5 47.0 34.0 11.0 31.8 5.4 22.3 13.7 29.6 24.5 25.0 n/a n/a

Surface area 2

km

11 11 8 2 25 10 b1 9 25 4 b1 4 1 b1 b1 18 21 54 13 6 4 24 11 40 3 8 15 62 7 11 2 3

Elevation m a.s.l.

102 350 362 605 421 360 632 370 638 828 1040 1089 535 534 531 430 1100 684 307 580 998 886 894 352 372 501 547 315 831 833 297 639

Chl a

Secchi depth −3

mg m (avg.)

m (avg.)

1.0 2.5 106.5 12.0 32.2 2.9 53.6 53.0 1.1 21.6 9.0 24.3 38.4 39.7 n/a 8.7 10.6 5.9 61.9 13.1 1.3 3.5 41.3 2.9 0.9 2.4 3.3 15.8 28.8 5.0 37.3 41.0

3.6 1.0 1.3 0.7 2.5 5.5 0.0 0.6 4.4 0.0 2.8 1.6 0.6 0.6 0.5 2.0 3.4 3.2 0.7 5.0 5.4 3.0 1.0 4.3 0.6 4.6 4.0 2.7 1.7 3.0 0.6 0.7

Visitsa

Samples

Spectroradiometry samples

Total

PR-650

Pigment samples ASD-FR

HPLC

PC FT

1 1 1 1 2 1 1 1 1 1 1 3 1 1 1 1 1 2 12 1 1 1 2 2 1 1 1 3 3 2 1 1

2 1 2 1 4 2 1 3 2 1 1 4 1 1 1 4 2 4 49 2 2 2 4 4 2 4 2 5 6 3 2 1

0 0 0 1 0 0 1 3 0 1 1 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 2 1

2 1 2 0 4 2 0 3 2 0 0 4 0 0 0 4 2 4 48 2 2 2 3 4 2 3 2 4 4 3 0 1

2 1 2 1 3 2 1 3 2 1 1 4 1 1 0 4 2 4 35 2 2 2 4 4 1 4 2 4 6 2 2 1

0 0 0 1 0 0 1 0 0 1 1 0 1 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 2 0 0 0

MG 2 1 2 1 4 2 1 3 2 1 1 4 1 1 0 0 2 4 2 2 2 2 4 4 0 0 2 4 6 2 2 1

FL 2 1 2 0 4 2 0 3 2 0 0 4 0 0 0 4 2 4 44 2 2 2 3 4 1 4 2 5 4 3 2 1

Depth refers to echo soundings at sampling stations. Symbols and abbreviations: n/a — no data; avg. — average; m.a.s.l. — metres above sea level; lat. — latitude; lon. — longitude; FT — freeze/thaw cycles method; MG — mechanical grinding method; FL — in situ fluorescence; R. — reservoir; L. — Lake; Lgn. — Lagoon. a Visits spanning 2 days counted as 1. bIncludes samples of lower River IJssel.


Guadalcacín R. Guadalén R. Guadalteba R. Guajaraz R. Iznájar R. Jándula R. Molino de la Hoz R. Navalcán R. Negratín R. Pedrezuela R. Picadas R. Pinilla R. El Porcal-1 Lgn. El Porcal-2 Lgn. El Porcal-3 Lgn. Rialb R. Riaño R. Ricobayo R. Rosarito R. San Juan R. L. Sanabria Santa Teresa R. Santillana R. La Serena R. Terradets R. Tremp R. Ullívarri R. Valdecañas R. Valmayor R. Valparaíso R. Valuengo R. Vega de Jabalón R.

Position

S.G.H. Simis et al. / Remote Sensing of Environment 106 (2007) 414–427 Table 2 Details on field spectroradiometric configurations used to collect Rrs spectra in Spain and The Netherlands Dataset

The Netherlands

Instrument Manufacturer

PR-650 Photo Research, Inc. Chatsworth, CA, USA Light collector Lens Acceptance angle 1° Spectral interval 4 nm Full-width half- 8 nm maximum Spectral range 380–780 nm Viewing angle θ 42° Azimuth angle φ 90° Skylight correction 0.029 factor r(θ) Reflectance panel 100% (white) (for Ed) Spectralon Exposure time Dynamic Sensor saturation 95% Neutral density prevention filter Exposures per 10 measurement

Spain ASD-FR Analytical Spectral Devices, Inc. Boulder, CO, USA Lens on glass fibre (Ø 0.17 cm) 8° 1.4 nm 3 nm 350–1000 nm 40° 135° Function of wind speed, cloud cover (Mobley, 1999) 25% (gray) Spectralon Manual Shorten exposure time 20

accordingly a specific absorption coefficient for PC in the 620nm band [a⁎PC(620)] that was lowered to 0.007 m2 mg− 1. The second method for phycobilin pigment extraction was based on mechanical grinding (PCMG) of samples concentrated on glass fibre filters (Whatman GF/F) and resuspended in glycerol (Quesada & Vincent, 1993; Wyman & Fay, 1986a,b). After collection, filters were frozen in liquid nitrogen and later stored at − 80 °C. The filters were thawed, cut into small pieces, and one volume of glycerol was added to the filter remains placed in a centrifuge tube. A close-fitting Teflon pestle connected to an electric drill was used to homogenize the sample in the tube at 500–1000 rpm while avoiding the sample to heat up. The samples were kept in the dark for approximately 2 h after which 9 volumes of distilled water were added to induce osmotic shock. The samples were briefly subjected to another round of homogenization in the pestle-tube combination. After extraction with either the PCFT or PCMG method, the phycobilin pigment concentrations were computed from the absorption spectra of supernatants of centrifuged samples, according to the equations published in Bennett and Bogorad (1973). The PCMG extraction was repeated on the pelleted material and the results summed to give the final concentration, after correction for the initially filtered sample volume. Finally, the in situ PC quantification by fluorescence (PCFL) was performed using a Minitracka II PA Fluorometer Model HB202 (Chelsea Instruments Ltd., Surrey, UK) with excitation centred at 590 nm (35 nm band width) and recorded emission around 645 nm (35 nm band width). The instrument was factory-calibrated in the 0.03–100 mg m− 3 range using PC (Sigma Chemicals) dissolved in pH 7 phosphate buffer. 2.4. HPLC pigments Pigments that could be extracted with organic solvents were analysed using gradient HPLC based on the protocols in Jeffrey

419

et al. (1997). Samples collected and analysed in the Netherlands were concentrated on Schleicher and Schuell GF6 filters, frozen in liquid nitrogen and stored in a − 80 °C freezer. Subsequently filters and cells were disrupted in a bead beater, adding 90% acetone as a solvent. After centrifugation, the pigment content of the supernatant was separated using a reversed-phase column (Waters Nova-pak C18 column; Waters Millennium HPLC system), to which gradient mixing pumps delivered three mobile-phase solvents: methanol/ammonium acetate, 90% acetonitril and 100% ethyl acetate (Rijstenbil, 2003 and references therein). Pigments were identified against commercially available standards (DHI Water and Environment, Hørsholm, Denmark) using a fluorescence detector (Waters 474 Scanning Fluorescence Detector) and a photodiode array absorption detector (Waters 996 Photodiode Array Detector). The Spanish samples were treated in a similar way, except that samples were concentrated on Whatman GF/F filters and kept in liquid nitrogen until analysis, cell disruption and pigment extraction was achieved by sonication followed by 24h extraction at 4 °C, and the column was equipped with a Waters Spherisorb ODS-2 and HP Agilent 1050 diode array system. The Spanish samples were spiked with a known concentration of canthaxanthin to calibrate the instrument response. The concentrations of the following pigments were acquired for all samples in both countries: Chl a, Chl b, phaeophytin, peridinin, neoxanthin, violaxanthin, alloxanthin, lutein, zeaxanthin, and fucoxanthin. 2.5. PC retrieval from radiometry Rrs(λ) data at either 1 or 4-nm intervals (from ASD-FR and PR-650, respectively) were transformed to MERIS band

Fig. 2. Comparison of remote-sensing reflectance [Rrs(MERIS channel, band center, band with)] derived from the instrumentations used in Spain (ASD-FR) and The Netherlands (PR-650). Only reflectances for the bands used in Eqs. 3 and 5 are plotted. The measurements were carried out on a number of Spanish water bodies in May 2005. Regression results are shown by the solid line and dashed 95% confidence limits.

420


resolution. Current MERIS band settings are, denoted Rrs (MERIS channel, midpoint, band width): Rrs(6, 620, 10), Rrs(7, 665, 10), Rrs(9, 708.75, 10), and Rrs(12, 778.75, 15). Weighted averaging (weight depending on the overlap with the MERIS band) was used to derive the bands from the 4-nm resolution data. Band Rrs(778.75) derived from PR-650 data was based on the 771.25–780 nm range as 780 nm was the upper measurement limit of the instrument. For 1-nm data the wavelength ranges, rounded to the nearest nm, were averaged. Calculation of the backscattering coefficient [bb] from Rrs (778.75) followed Gons et al. (2005): bb ð778:75Þ ¼ 1:61Rrs ð778:75Þ= f0:082−½0:6Rrs ð778:75Þg

ð3Þ

The PC algorithm from Simis et al. (2005) that is given below assumes that bb is spectrally neutral over the used wavelength range, that absorption in the 665-nm band can be attributed to Chl a and water [aw(λ)], and absorption in the 620-nm band to water, Chl a, and PC. Absorption by phytoplankton pigments [aph(λ)] is obtained from the ratio of two reflectance bands, in fact a ratio of the Gordon reflectance model (Gordon et al., 1975) for both bands, which results for bands λ = 620 or 665 nm in: aph ðkÞ ¼ pðkÞ−1 ðf½Rrs ð709Þ=Rrs ðkÞ ½aw ð709Þ þ bb g−aw ðkÞ−bb Þ

ð4Þ

where p(λ) is an empirical factor to compensate for a weaker absorption signal retrieved from radiometric data compared to absorption measurements with a spectrophotometer, as was found during the development of the PC algorithm (Simis et al., 2005). Substitution of aw(620) = 0.281, aw(665) = 0.401, and aw(708.75) = 0.727 from Buiteveld et al. (1994), and a⁎PC(620) = 0.007 m2 mg− 1, p(620) = 0.84, and p(665) = 0.68 from Simis

Fig. 4. Comparison of PC values derived from the instrumentations used in Spain (ASD-FR) and The Netherlands (PR-650). The measurements were carried out on a number of Spanish water bodies in May 2005. The boxed area excludes points from L. l'Albufera where cyanobacterial biomass was exceptionally high. Solid regression line and dashed 95% confidence limits represent the regression results through all plotted points.

et al. (2005) gives the PC algorithm that is validated in this paper: ð5aÞ

aph ð665Þ ¼ 1:47 ðf½Rrs ð709Þ=Rrs ð665Þ ½0:727 þ bb g−0:401−bb Þ PCRAD ¼ 170 ððf½Rrs ð709Þ=Rrs ð620Þ

½0:727 þ bb g−bb −0:281Þ−½e aph ð665ÞÞ ð5bÞ where ε is used to derive absorption by Chl a at 620 nm from the pigment absorption at 665 nm, which is assumably dominated Table 3 Results for linear regression between PC reference measurement methods, between PCRAD and PC reference methods, and between PCRAD and PCREF, the average of reference PC measurements Figure y-method x-method Country n

Fig. 3. Comparison of backscattering values derived from the instrumentations used in Spain (ASD-FR) and The Netherlands (PR-650). Concurrent measurements were carried out on selected Spanish water bodies in May 2005. The boxed area excludes points from L. l'Albufera where cyanobacterial biomass was exceptionally high. The solid regression line and dashed 95% confidence limits represent the regression results through all plotted points.

7A

PCMG

PCFT

7B 8A

PCMG PCRAD

8B

PCRAD

PCFL PCFT PCFT PCMG PCMG PCFL PCREF PCREF PCREF PCREF

NL S S NL S NL S S NL S NL + S NL + S

73 15 106 200 8 75 114 162 202 171 373 373

a

b

r2

p

1.18 1.55 0.96 b0.01 1.13 28.01 0.85 b0.01 0.89 17.82 0.59 0.489 0.68 29.17 0.77 b0.001 0.34 241.93 0.00 0.034 0.58 29.37 0.74 0.450 1.09 25.8 0.65 0.049 0.99 15 0.53 0.164 0.63 29.75 0.75 b0.001 1.18 − 3.23 0.77 0.117 1.06 9.23 0.74 b0.001 1.09 0 a b0.001

Legend: n = data points; regression model parameters y = ax + b; NL = samples from The Netherlands, S = Spanish samples; bold-face p-values are significant at 99% confidence after applying Bonferroni's correction. a Intercept forced to zero.

S.G.H. Simis et al. / Remote Sensing of Environment 106 (2007) 414–427 Table 4 Multiple regression results of residual errors of PCRAD versus PCREF with intercept forced to zero (see text) PCREF b 50 mg m− 3

Country

n Multiple R2 Adjusted R2 SE of estimate whole-model p β Chl b β Fucoxanthin β Phaeophytin β Alloxanthin β Peridinin

PCREF N 50 mg m− 3

NL + S

NL

S

NL + S

NL

S

234 0.366 0.352 14.90 0.000 0.499 0.210 0.131 − 0.026 − 0.041

141 0.564 0.551 9.37 0.000 0.723 0.202 0.065 − 0.139 n/a

93 0.205 0.160 15.00 0.001 0.296 0.304 −0.095 0.115 0.020

101 0.084 0.046 45.50 0.073 0.219 − 0.057 0.146 − 0.093 n/a

56 0.069 −0.004 41.69 0.442 −0.002 −0.193 −0.006 −0.106 n/a

45 0.152 0.067 49.43 0.150 0.299 0.011 0.130 − 0.087 n/a

The residuals were regressed against the marker pigments Chl b (green algae), peridinin (dinoflagellates), alloxanthin (cryptophytes), and fucoxanthin (diatoms), as well as the Chl degradation product phaeophytin. The analysis was performed for all samples simultaneously, for subsets by country, and for the PCREF range split at 50 mg m− 3. Significant (N99% confidence) p and β values are printed in bold face. Prior to analysis, outliers (residuals more than ±2 times standard deviation) were removed. Symbols and abbreviations: NL — Samples from The Netherlands; S — Spanish samples; n — number of valid cases; SE — standard error.

by Chl a. From optimization of PCRAD versus extracted PC, it was found that ε = 0.24 for cyanobacteria-dominated waters (Simis et al., 2005). 2.6. Data screening and analysis Concurrent spectroradiometric measurements with the ASDFR and PR-650 instruments were available for 13 samples from 4 Spanish lakes, visited in May 2005. Linear regression analysis was used to evaluate the correlation between data subsets of the spectroradiometric products Rrs, bb, and PC (Figs. 2–4). The reference PC methods (PC extraction and in situ fluorescence) were also compared for samples where more than one technique was used (Fig. 7, Table 3). The availability of the different data types for each of the water bodies is given in Table 1. For all regression results presented in this study, results marked as significant passed Bonferroni's correction at 99% confidence level unless stated otherwise.

421

HPLC and phycobilin pigment data were subjected to a principal component analysis (PCA) to identify marker pigments for phytoplankton composition (Fig. 6). Two samples from the Guadalteba reservoir featured a high-biomass dinoflagellate bloom with extremely high values of peridinin (11 and 47 mg m − 3 ). As these high peridinin values obscured the relations between the other pigments in the pigment matrix, they were omitted from PCA and further regression analyses. PCRAD was compared with reference PC methods for the full dataset and subsets separated by country (Fig. 8A). The correlation between the radiometric approach and the reference measurements was analysed using paired t-tests (Table 4). PCRAD was also compared to the average of the available reference PC measurements for each sample, denoted PCREF (Fig. 8B). It is noted that averaging reference PC measurements is not optimal, as in several cases extraction of phycobilin pigments clearly failed (zero values for PCMG, discussed below) while cyanobacterial presence was evident. In those cases choosing the highest extraction value would eliminate part of the error caused by low extraction yield. On the other hand, the PCFL method is most likely to yield positive rather than negative errors, as it relies on the sensor sensitivity to PC among fluorescence of other substances (dissolved organic matter, algal pigments). In cases where PCMG extraction failed, a concurrent PCFL measurement would probably be closer to the ‘real’ PC concentration. However, since combinations of reference PC methods were not available for all samples and no comparison was available between PCFT and PCFL, it was considered most straightforward to define PCREF as the average of all available reference PC values. To evaluate prediction errors by the PC algorithm, a regression with zero intercept was fitted through the plot of PCRAD against PCREF measurements. The residuals of this regression were analysed in a multiple linear regression against chlorophyllous HPLC pigments including phaeophytin, and substituting fucoxanthin for Chl c as the latter was not measured for the whole dataset. The multiple regression was carried out separately for subsets of the data split at PCREF = 50 mg m − 3 . These analyses were made for the dataset in full as well as separated by country (Table 4).

Fig. 5. Boxplots of PCREF and Chl a concentrations as well as the PCREF:Chl a ratio in the Spanish (S) and Dutch (NL) water bodies.

422


3. Results The Rrs(λ) derived from ASD-FR and PR-650 measurements could be compared for a subset of 13 locations at 4 Spanish lakes where both instruments were used. The intercalibration measurements were all carried out in May 2005 in Spain and 7 samples were from the hypertrophic L. l'Albufera which stood out in terms of high Rrs(λ) spectra, near-infrared backscattering, and pigment concentrations. These aspects strongly limit the representativeness of the calibration set for the full dataset. Regression of the Rrs data in the used MERIS wavebands yielded acceptable agreement between the used methods (Fig. 2: r2 = 0.94; p b 0.01; n = 56). The Rrs(λ) values obtained with the ASD-FR setup were on average 17% higher than PR-650 values in all wavebands. These reflectances lead to 30% higher estimates of bb (Fig. 3; r2 = 0.94; p b 0.01; n = 13). ASD-FR data nevertheless resulted in only 10% higher PCRAD values (Fig. 4; r2 = 0.97; p b 0.01, significant at 95% confidence; n = 13). The correlation coefficients for products Rrs and PCRAD remained high when the 7 samples from L. l'Albufera were excluded from the regression (r2 = 0.87 and 0.97 for bb and PC respectively), but the relationship was no longer significant for bb. We did not proceed to calibrate the Rrs of one dataset to another, as the impact of differences in the measurement of Rrs on the retrieval of PCRAD was limited, and the number and timespan of observations were highly limited compared to the full dataset. Considerable differences in pigment composition were found between the Dutch and Spanish water bodies. Boxplots of PCREF (the average of available PC extractions and PC fluorescence results), Chl a, and the ratio of PCREF to Chl a showed that the sampling covered a wide range of pigment concentrations in both countries, with the most even distribution of pigment concentrations and the highest PCREF:Chl a ratios in Spain (Fig. 5). The PCREF:Chl a ratio is a possible indicator of the cyanobacterial share in total phytoplankton biomass. Of the samples taken and analysed in the Netherlands, 84% had a ratio PCREF:Chl a ≤ 1.25, suggesting relatively low intracellular PC: Chl a ratios, a limited share of cyanobacteria in the phytoplankton, or both. In contrast, 57% of the samples collected and analysed in Spain had a ratio PCREF:Chl a ≥ 1.25. PCA based on pigment correlations showed good separation of cyanobacterial and green algal pigments in the full dataset as well as subsets divided by country (Fig. 6). The diatom pigment fucoxanthin grouped with cyanobacterial presence (phycobilin pigments and zeaxanthin) in Spanish samples while the influence of fucoxanthin on pigment variance was weak in the Fig. 6. PCA factor scores based on pigment correlations for (A) the full dataset, (B) Dutch samples, and (C) Spanish samples. The following pigments were included to represent the indicated phytoplankton groups: alloxanthin — cryptomonads; phycocyanin, phycoerythrin, zeaxanthin — cyanobacteria; fucoxanthin — diatoms; peridinin — dinoflagellates; phaeophytin — not specified; chlorophyll b, neoxanthin, violaxanthin — green algae. Peridinin was not found in Dutch samples. The variance that was explained by each PCA axis is given in the axis labels. Two samples taken during a high-biomass dinoflagellate bloom in Guadalteba reservoir were omitted from PCA due to very high (11 and 47 mg m− 3) peridinin values that dominated the pigment matrix.

Dutch subset. Indeed, fucoxanthin concentrations were generally low in the Dutch samples. Fucoxanthin was always correlated with phaeophytin which could indicate that these samples were influenced by resuspension of sedimented diatom


cells. Cryptomonads in the phytoplankton community, indicated by the alloxanthin pigment, were weakly represented regarding the full dataset, but prominent in the Dutch subset. Cryptomonad presence was not correlated to PC, even though cryptomonads are known to contain low concentrations of the latter pigment. Two samples from a single high-biomass dinoflagellate bloom were omitted from PCA and subsequent multiple regression analyses, as they otherwise dominated the pigment matrix. Subsequently, peridinin presence was restricted to some samples in the Spanish data and did not contribute much to total variance. The PCA yielded the following marker pigments, to serve as proxies for the phytoplankton groups with a suspected influence on absorption in the red spectral area: fucoxanthin for diatoms, Chl b for green algae and prochlorophytes, alloxanthin for cryptomonads, and peridinin for dinoflagellates. Reference measurements for PC were derived from extraction of the phycobilin pigments through cell disruption (PCFT and PCMG), or from in situ PC fluorescence (PCFL). Fig. 7 shows comparisons of these methods, which were available for subsets of the data (Table 1). Regression results are given in Table 3. For all comparisons of methods there was a high degree of scatter. Between PCMG and PCFT high correlation was found with on average slightly higher values of PCMG, although the number of concurrent measurements was low for Spanish water bodies (Fig. 7A). Note that the PCMG method can suffer from poor separation of Chl a from the supernatant due to the presence of submicron cell debris, which can easily lead to a 10% elevated PC quantification (unpublished results). Between PCMG and PCFL methods (Fig. 7B) the linear regression equation was not significant, which is primarily explained by a high number of zero PCMG values (plotted at 0.11 mg m− 3). These zero values represent cases where the PCMG method was unsuccessful in extracting any phycobilin pigments while the pigment could be measured following PCFT extraction or by

423

fluorescence. Extraction of the pigment from tough, very small, or mucilage-covered cells may be problematic with this method, but we did not find an obvious reason for the poor extraction yield. Unfortunately no PCFL data was available for the Dutch samples, which would have completed the comparison. As the reference PC quantification methods showed a high degree of scatter between them and no trends strongly deviating from unity, we were not able to identify a standard method, and it was decided that no intercalibration of used methods would be carried out. Instead, where multiple measurements were available, the average value (denoted PCREF) was used. Comparison of PCRAD values with reference PC measurements (Fig. 8A) only gave a significant correlation (regression results in Table 3) between PCRAD and PCFT. The regression slope of this comparison was low, caused by a substantial number of relatively high PCRAD values in the lower PCFT range. The correlation between PCRAD and PCREF was significant for the full dataset and the Dutch subset, but not for Spanish samples alone. Positive errors of PCRAD compared to PCREF occurred most frequently at Dutch sites and were again concentrated at the lower end of the PCREF range. In general, the correlation between PCREF and PCRAD improved with PCRAD N 50 mg m− 3. Of all reference PC measurements, only PCFL measurements were often higher than the PCRAD values. An idealized regression (intercept forced through zero) of PCRAD versus PCREF reveals under- and overestimations of PCRAD that are not systematic. The regression slope was 1.09 (Table 3). The deviation from unity of this slope is relatively low in comparison to the scatter that was found when comparing the reference methods for PC quantification, and is therefore interpreted as support for the PC algorithm's current parameterization. The relative residual errors calculated as (PCRAD − 1.09 PCREF) / PCREF were highest at low PCREF concentrations, especially when Chl a biomass was not as low

Fig. 7. Log–log plot of reference PC measurements for samples where multiple methods were applied (A) Mechanical grinding (PCMG) vs. Freeze/Thaw (PCFT) method, separated by country. (B) PCMG extraction method vs. PCFL fluorescence method, for samples taken and analysed in Spain. Regression results are listed in Table 3.

424


Fig. 8. Comparison of PC assessment methods. (A) PCRAD plotted against PCFT, PCMG, and PCFL, samples separated by country. (B) PCRAD plotted against PCREF (the average of available reference methods), separated by country. Note the scale breaks at 200 mg m− 3 in both panels. Regression results are listed in Table 3.

(Fig. 9A,B). This suggests an influence of other phytoplankton groups on the quality of PC retrieval in the lower PC concentration range. At higher (relative) PC concentrations (50–350 mg PC m− 3 and PC:Chl a approximately N1.75, visible as a scattered cluster of samples at the high PC concentration end in Fig. 9B), negative relative errors of PCRAD were visible. At PC concentrations N350 mg m− 3 these negative errors were no longer visible. Possibly, the negative errors in the intermediate range are related to a pigment package effect, causing a nonlinear relation between pigment concentration and observed absorption, while the effect is no longer visible at the highest concentrations due to increased reflectance in the NIR bands, but there is no evidence to support this hypothesis.

To identify the effects of phytoplankton pigment composition on the estimation of PC by the reflectance algorithm (Eq. (5b)), multiple linear regression analyses were carried out on the residual errors from regression of PCRAD against PCREF with zeroed intercept (calculated from PCRAD − 1.09 PCREF). The multiple regression was carried out against the marker pigments for the main algal groups (identified from PCA) and phaeophytin. Table 4 lists the multiple regression results carried out for the full dataset, the Dutch and Spanish samples separately, and each part split into subsets of PCREF higher or lower than 50 mg m− 3. Significant overestimations caused by Chl b and fucoxanthin were found in the lower concentration range, where the residual errors were relatively high. A positive effect of phaeophytin on

Fig. 9. Relative residual errors of the regression model PCRAD = 1.09 × PCREF, plotted (shaded areas) as a function of PCREF and Chl a, based on distance-weighted least squares interpolation of the errors. PCREF and Chl a concentrations of each sample are plotted as circles to indicate the algorithm performance of each sample in relation to cyanobacterial and total phytoplankton biomass. (A) Results for the 0–800 mg m− 3 PC and Chl a range. (B) Same results, but for the 0–200 mg m− 3 pigment concentration range (note a different gray scale for the relative residual errors).


PCRAD was found to be significant only for the full dataset b 50 mg m− 3. Only in the Spanish subsets an overestimation of PCRAD was significantly correlated with fucoxanthin or diatom presence. This corresponds with the differences in the pigment matrix that were found between Spain and the Netherlands through PCA, i.e. a weak representation of diatom pigment in the Dutch dataset, but co-occurrence of diatom and cyanobacterial pigments in the Spanish subset. Residual PCRAD errors could not be explained by the regression model for the concentration range N 50 mg m− 3 so the presence of accessory pigments could not explain the underestimation of some PC values in the intermediate range, as observed from Fig. 9. 4. Discussion The performance of a semi-analytical, nested reflectance ratio-based algorithm for the quantification of PC (Eq. (5b)) was tested with field spectroradiometric and pigment data collected in the Netherlands and Spain, in the period 2001–2005. The influence of phytoplankton pigment composition on the performance of the algorithm was explored using HPLC marker pigments for different phytoplankton groups. Analysis of PC: Chl a ratios and PCA of the HPLC and phycobilin pigment matrix suggested that co-occurrence of phytoplankton groups was common in Dutch samples while cyanobacterial predominance occurred in a number of Spanish water bodies. The PC algorithm results exhibited a positive trend with PC concentrations that ranged several orders of magnitude. Closer investigation showed that most PCRAD errors were overestimates that occurred in the low (PCREF b 50 mg m− 3) concentration range. The errors were most evident with Dutch samples and were primarily associated with the presence of Chl b, indicating the presence of green algae or prochlorophytes. Fucoxanthin, indicating diatom presence, and phaeophytin also correlated with overestimation of PC in parts of the dataset. It may be assumed that the effect of fucoxanthin represents the influence of red absorption by Chl c, even though Chl c itself was not measured for this dataset. There is no standardized method for the extraction of PC from water samples, while in situ fluorescence probes for PC are increasingly popular for their ease of use and possibility to use them in unattended measurement systems. We compared a fluorometric approach and two extraction methods in order to compile the full dataset. The three methods for PC quantification showed similar values for concurrent measurements, however with a high degree of scatter. Besides a number of samples for which the PCMG method was ineffective, comparison of PCMG with the other methods suggested no strong methodological bias within the current dataset. Nevertheless, PCFL measurements were not available for the Dutch dataset, so the comparison remains incomplete. It is noted that the quality of PC extraction for Dutch sites was improved since the development of the PC algorithm, thereby reducing the overestimation that was previously found (cf. Fig. 6 of Simis et al., 2005). In general, the high degree of scatter between the methods suggest that PC ground truth measurements should be used with scrutiny. In vivo PC fluorescence methods are perhaps most promising for routine

425

measurement of PC, as sample preservation and pigment extraction steps are not needed. The potential of PC fluorescence probes to monitor cyanobacterial biomass has been reported in several studies (Asai et al., 2001; Izydorczyk et al., 2005; Lee et al., 1994, 1995). It was previously found that bulk pigment absorption can be retrieved from turbid water reflectance, seemingly regardless of suspended sediments and coloured dissolved organic matter (Simis et al., 2005). Unfortunately, absorption spectra were not available for the dataset that was analysed here, so evaluation of the empirically defined factors γ and δ to relate aph(665) and aph(620) from the absorption obtained from reflectance ratios (Eqs. (5A) and (5B) respectively) was not possible. Additionally, it was assumed that errors in the PC estimation from field spectroradiometric data can be directly related to the presence of accessory photosynthetic pigments and degradation products of Chl a. It is noted that the presence of these pigments is not necessarily the cause of the observed error; this could also be a related factor such as differences in light scattering or absorption between the phytoplankton groups that the diagnostic pigments represent. Other errors that are not considered here may yet exist, e.g. those that are inherent to the assumptions that were made in algorithm development: an assumed negligible influence of the spectral shape of bb, insensitivity to CDOM and suspended sediments, and the ability to correct for Chl a absorption at 620 nm through a constant fraction ε of aph(665). NIR/red band ratio-based algorithms should be relatively insensitive to spectrally neutral (‘white’) errors, when these algorithms are applied to turbid water bodies that already have a relatively bright reflectance. The influence of both suspended sediments and CDOM can be considered spectrally neutral between the used bands. The backscattering may not be neutral between the ratioed bands and the 778.75-nm bands, but there is no evidence that NIR/red band ratios are strongly affected by this when applied to turbid water bodies (cf. Gons, 1999; Gons et al., 2000). Studies of in vivo pigment absorption, decomposed into multiple Gaussian curves for each pigment through multiple regression, show that the main chlorophyllous pigments (Chl a, b, c, and phaeophytin) absorb light throughout the 600–675 nm region (Ficek et al., 2004; Hoepffner & Sathyendranath, 1993; Sathyendranath et al., 1987). Indeed, the present study shows that Chl b, phaeophytin, and fucoxanthin as a proxy for Chl c were related to overestimation of PC. Specifically at low PC concentrations, the absorption by these pigments could increase absorption in the 620-nm band significantly. The only correction for chlorophyllous pigment absorption in the original PC algorithm was for Chl a absorption. The influence of this pigment in the 620-nm band was derived from absorption in the 665-nm band, which was fully attributed to Chl a, using a fixed factor ε (Eq. (5b)). The optimized value of ε for cyanobacteriadominated sites was set at ε = 0.24 (Simis et al., 2005), i.e. lower than the value of 0.3–0.4 for Chl a in algal species, and much lower than the specific in vivo absorption of Chl b with ε = 0.5– 0.6 and Chl c with ε = 1.7–3.7 (Bidigare et al., 1990; Ficek et al., 2004; Hoepffner & Sathyendranath, 1993; Sathyendranath et al., 1987). Obviously, with increasing concentrations of Chl b and c,

426


the ε-correction of 0.24 is increasingly inadequate and the derived PC absorption will be too high. There are no MERIS bands that specifically target the accessory chlorophylls, and it is yet unknown whether hyperspectral information could provide a robust index for the presence of Chl b and c. A correction for the presence of Chl b, using a Chl b-specific absorption value at 620-nm band of 0.05 m2 (mg Chl b)− 1 , efficiently removed most of the overprediction in the whole dataset (PCRAD = 0.83 × PCREF + 1.71; r2 = 0.90; p = 0.002; n = 223). This specific absorption value is considerably higher than reported values for in vivo Chl b absorption in this waveband, which range 0.002–0.013 m2 (mg Chl b)− 1. This suggests that the presence of Chl b in our dataset weighs more heavily on the PC prediction than would be expected from its absorption in the 620–665 nm area alone. As discussed above, other factors that influence PCRAD when Chl b containing species are present may account for this difference, such as different scattering behaviour or pigment packaging, or a different expression of in vivo Chl a absorption in green algae, as indicated by a higher ε factor reported in various literature studies. Absorption spectra should be obtained and analysed along with the pigment matrix of chlorophyllous and phycobilin pigments to resolve this issue. Additional corrections for Chl c (through the fucoxanthin proxy) or phaeophytin did not improve the PC prediction even though our multiple regression results indicated a significant effect of their presence. Correction for these pigments had a limited effect because they were present in very low concentrations in most samples. If the pigment balance of Chlorophyllous pigments and PC in our dataset reflects the situation in most eutrophic inland water bodies, then detection of the contribution of Chl b to the absorption envelope can be considered the most important next step in algorithm development. Despite the demonstrated influence of Chl b and other accessory pigments on the PC predictions, PCRAD values above 50 mg m− 3 were increasingly reliable. Additionally, with higher PC:Chl a ratios the reliability of PCRAD should extend below 50 mg m− 3, which is no surprise as background absorption at 620 nm should become less important when PC starts to dominate the absorption envelope. To be safe, predictions below 50 mg m− 3 should be taken as the high estimate, i.e. the actual concentration could be lower, but not likely higher than the estimated value. Evaluating the current dataset, 57% of all samples had such ambiguous PCRAD values below 50 mg m− 3. However, we should also consider the Chl a concentrations of these particular samples, since low Chl a concentrations will not give rise to an immediate need for PC monitoring, from the perspective of water quality management. The World Health Organization (WHO) dictates a guideline of approximately 50 mg Chl a m− 3 in cyanobacteria as a moderate health warning level (WHO, 2003), which corresponds to PC concentration of 50 to 100 mg m− 3 for common intracellular PC:Chl a ratios of 1–2. A lower guideline Chl a concentration may be given for some cyanobacterial species that are known to exhibit relatively high toxin levels, e.g. Planktothrix agardhii which at a Chl a concentration of 12–25 mg Chl a m− 3 may reach the daily tolerance intake level for microcystin (WHO, 2003). When no shortage of nitrogen exists, an intracellular PC:Chl a ratio of at

least 1.5 may be expected, so that sites with PC concentrations of 18–37 mg m− 3 should optimally be distinguishable from sites with lower PC concentrations. This criterion was not yet met with the used PC algorithm. Of all samples in our dataset, 29% had a Chl a concentration higher than 12 mg m− 3 (regardless of whether this was present in cyanobacteria) in combination with a PCRAD value below 50 mg m− 3. Therefore, in 29% of our sites the cyanobacterial state of a water body would not be indicated with confidence while the risk of toxic cyanobacteria is present. In all other cases, the site could either be considered a priori harmless as the Chl a concentration would be too low to suspect problems, or the PCRAD estimate would be above the 50 mg m− 3 threshold and could be used as an indicator for the cyanobacterial biomass. Considering both Chl a and PC estimates thus significantly increases, up to 71% of sites, the value of remotely sensed information compared to using only Chl a (no information on cyanobacterial state) or PC (only 43% of sites has a reliable estimate). We will not discuss the quality of current Chl a algorithm at this point, but Chl a charting from remotely sensed imagery is already carried out for many water bodies, including eutrophic lakes and reservoirs. Remote sensing of cyanobacterial biomass relies not only on validated algorithms for PC and Chl a, but also on the quality of remotely sensed imagery. Currently only MERIS provides the necessary spectral bands to estimate both PC and Chl a at a suitable temporal and spatial resolution. MERIS was initially designed to observe the oceans, land, and the coastal zone (Rast et al., 1999). Inland water bodies provide challenges that have not been met by the ocean colour community, such as the effects of adjacent land on the radiance measured at the sensor, and elevated NIR reflectance from dense phytoplankton blooms, which may interfere with atmospheric correction schemes. Our current findings warrant further research into the quality of reflectance products for NIR/red band ratio algorithms for turbid inland water, both from MERIS and other (satellite or airborne) sensors, and not only for Chl a retrieval but also for diagnostic accessory pigments. As a final consideration it is noted that the physiological state of cyanobacteria can be such that very little or no PC is produced (Tandeau de Marsac, 1977), or it may be broken down to serve as a nitrogen resource in times of shortage (Bogorad, 1975; Grossman et al., 1993). In such situations mapping the PC concentration will yield little information on the cyanobacterial state of a water body, while cyanobacterial biomass and toxin concentrations could still be substantial. Also, the ratio of PC to Chl a, which could be used as an indicator for the share of cyanobacteria in the phytoplankton, is susceptible to variability in the intracellular PC:Chl a ratio of the cyanobacteria at the species level, and to variability due to physiological conditions. Ultimately, remote sensing offers the possibility to increase the spatial coverage of existing monitoring programmes, and to strategically direct conventional monitoring efforts—replacing them should not be the goal. Acknowledgments The authors thank the drivers and field technicians at the Centre for Hydrographic Studies (CEDEX, Spain) and the crew of R/V Luctor (NIOO, The Netherlands). Antonio Quesada (Universidad


Autonoma, Madrid, Spain) is thanked for advice and the use of lab facilities for PC extraction during our data intercalibration efforts. Further, thanks are due to Nicole Dijkman (NIOO) for advice on phytoplankton group representation in the pigment matrix, and to Ana Alonso and Covadonga Alonso (CEDEX) and Cobie Kleppevan Zetten (NIOO) for HPLC pigment analysis. Critical reading by three anonymous reviewers helped to improve the manuscript. Funding for SGHS was provided through grant EO-053 from the User Support Programme managed by the programme office External Research of the Netherlands Organization for Scientific Research-National Institute for Space Research (NWO-SRON). A one-month stay for SGHS at CEDEX was covered by a grant from the Schure-Beijerinck Popping fund (KNAW, The Netherlands). This is publication 3936 NIOO-KNAW Netherlands Institute of Ecology, Centre for Limnology. References Asai, R., Horiguchi, Y., Yoshida, A., McNiven, S., Tahira, P., Ikebukuro, K., et al. (2001). Detection of phycobilin pigments and their seasonal change in Lake Kasumigaura using a sensitive in situ fluorometric sensor. Analytical Letters, 34, 2521−2533. Bennett, A., & Bogorad, L. (1973). Complementary Chromatic Adaptation in a filamentous blue-green alga. The Journal Of Cell Biology, 58, 419−435. Bidigare, R. R., Ondrusek, M. E., Morrow, J. H., & Kiefer, D. A. (1990). In vivo absorption properties of algal pigments. SPIE Ocean Optics X, 1302, 290−302. Bogorad, L. (1975). Phycobiliproteins and complementary chromatic adaptation. Annual Review of Plant Physiology, 26, 369−401. Bryant, D. A. (1981). The photoregulated expression of multiple phycocyanin species. European Journal of Biochemistry, 119, 425−429. Buiteveld, H., Hakvoort, J. H. M., & Donze, M. (1994). The optical properties of pure water. SPIE Ocean Optics XII, 2258, 174−183. Dekker, A. G. (1993). Detection of optical water quality parameters for eutrophic waters by high resolution remote sensing. Phd Thesis. Amsterdam: Vrije Universiteit. Dekker, A. G., Malthus, T. J., & Seyhan, E. (1991). Quantitative modeling of inland water-quality for high-resolution mss systems. IEEE Transactions on Geoscience and Remote Sensing, 29, 89−95. Ficek, D., Kaczmarek, S., Ston-Egiert, J., Wozniak, B., Majchrowski, R., & Dera, J. (2004). Spectra of light absorption by phytoplankton pigments in the Baltic; conclusions to be drawn from a Gaussian analysis of empirical data. Oceanologia, 46, 533−555. Gons, H. J. (1999). Optical teledetection of chlorophyll a in turbid inland waters. Environmental Science and Technology, 33, 1127−1132. Gons, H. J., Kromkamp, J., Rijkeboer, M., & Schofield, O. (1992). Characterization of the light field in laboratory scale enclosures of eutrophic lake water (Lake Loosdrecht, The Netherlands). Hydrobiologia, 238, 99−109. Gons, H. J., Rijkeboer, M., Bagheri, S., & Ruddick, K. G. (2000). Optical teledetection of chlorophyll a in estuarine and coastal waters. Environmental Science and Technology, 34, 5189−5192. Gons, H. J., Rijkeboer, M., & Ruddick, K. G. (2005). Effect of a waveband shift on chlorophyll retrieval from MERIS imagery of inland and coastal waters. Journal of Plankton Research, 27, 125−127. Gordon, H. R., Brown, O. B., & Jacobs, M. M. (1975). Computed relationships between the inherent and apparent optical properties of a flat homogeneous ocean. Applied Optics, 14, 417−427. Gordon, H. R., & McCluney, W. R. (1975). Estimation of the depth of sunlight penetration in the sea for remote sensing. Applied Optics, 14, 413−416. Grossman, A. R., Schaffer, M. R., Chiang, G. G., & Collier, J. L. (1993). Environmental-effects on the light-harvesting complex of cyanobacteria. Journal of Bacteriology, 175, 575−582. Hoepffner, N., & Sathyendranath, S. (1993). Determination of the major groups of phytoplankton pigments from the absorption-spectra of total particulate matter. Journal of Geophysical Research-Oceans, 98, 22789−22803.

427

Izydorczyk, K., Tarczynska, M., Jurczak, T., Mrowczynski, J., & Zalewski, M. (2005). Measurement of phycocyanin fluorescence as an online early warning system for cyanobacteria in reservoir intake water. Environmental Toxicology, 20, 425−430. Jeffrey, S. W., Mantoura, R. F. C., & Wright, S. W. (1997). Phytoplankton pigments in oceanography: Guidelines to modern methods. Paris: Unesco. Jupp, D. L. B., Kirk, J. T. O., & Harris, G. P. (1994). Detection, identification and mapping of cyanobacteria— Using remote-sensing to measure the optical-quality of turbid inland waters. Australian Journal of Marine and Freshwater Research, 45, 801−882. Kutser, T., Metsamaa, L., Strombeck, N., & Vahtmae, E. (2006). Monitoring cyanobacterial blooms by satellite remote sensing. Estuarine, Coastal and Shelf Science, 67, 303−312. Lee, T., Tsuzuki, M., Takeuchi, T., Yokoyama, K., & Karube, I. (1994). In-vivo fluorometric method for early detection of cyanobacterial waterblooms. Journal of Applied Phycology, 6, 489−495. Lee, T. Y., Tsuzuki, M., Takeuchi, T., Yokoyama, K., & Karube, I. (1995). Quantitative-determination of cyanobacteria in mixed phytoplankton assemblages by an in-vivo fluorometric method. Analytica Chimica Acta, 302, 81−87. Mittenzwey, K. H., Ullrich, S., Gitelson, A. A., & Kondratiev, K. Y. (1992). Determination of chlorophyll-a of inland waters on the basis of spectral reflectance. Limnology and Oceanography, 37, 147−149. Mobley, C. D. (1999). Estimation of the remote-sensing reflectance from abovesurface measurements. Applied Optics, 38, 7442−7745. Montagner, F. (2001). Reference model for MERIS Level 2 processing. European Space Agency, document PO-TN-MEL-GS-0026. Morel, A., & Gentili, B. (1993). Diffuse-reflectance of oceanic waters. 2. Bidirectional aspects. Applied Optics, 32, 6864−6879. Quesada, A., & Vincent, W. F. (1993). Adaptation to the light regime within Antarctic cyanobacterial mats. Verhandlungen Internationale Vereinigung für theoretische und angewandte Limnologie, 25, 960−965. Rast, M., Bezy, J. L., & Bruzzi, S. (1999). The ESA medium resolution imaging spectrometer MERIS - a review of the instrument and its mission. International Journal of Remote Sensing, 20, 1681−1702. Rijstenbil, J. W. (2003). Effects of UVB radiation and salt stress on growth, pigments and antioxidative defence of the marine diatom Cylindrotheca closterium. Marine Ecology. Progress Series, 254, 37−47. Sarada, R., Pillai, M. G., & Ravishankar, G. A. (1999). Phycocyanin from Spirulina sp.: Influence of processing of biomass on phycocyanin yield, analysis of efficacy of extraction methods and stability studies on phycocyanin. Process Biochemistry, 34, 795−801. Sathyendranath, S., Lazzara, L., & Prieur, L. (1987). Variations in the spectral values of specific absorption of phytoplankton. Limnology and Oceanography, 32, 403−415. Schalles, J. F., & Yacobi, Y. Z. (2000). Remote detection and seasonal patterns of phycocyanin, carotenoid and chlorophyll pigments in eutrophic waters. Archiv fur Hydrobiologie Special Issues Advances in Limnology, 55, 153−168. Simis, S. G. H., Peters, S. W. M., & Gons, H. J. (2005). Remote sensing of the cyanobacterial pigment phycocyanin in turbid inland water. Limnology and Oceanography, 50, 237−245. Tandeau de Marsac, N. (1977). Occurrence and nature of chromatic adaptation in cyanobacteria. Journal of Bacteriology, 130, 82−91. Tyler, J. E. (1960). Radiance distribution as a function of depth in an underwater environment. Bulletin of the Scripps Institution of Oceanography, 7, 363−411. World Health Organization (2003). Guidelines for safe recreational water environments. Coastal and fresh waters, Vol. 1. ( pp. 136−158). Geneva: World Health Organisation. Wyman, M., & Fay, P. (1986a). Underwater light climate and the growth and pigmentation of planktonic blue-green-algae (cyanobacteria). 1. The influence of light quantity. Proceedings of the Royal Society of London. Series B, Biological Sciences, 227, 367−380. Wyman, M., & Fay, P. (1986b). Underwater light climate and the growth and pigmentation of planktonic blue-green-algae (cyanobacteria). 2. The influence of light quality. Proceedings of the Royal Society of London. Series B, Biological Sciences, 227, 381−393.