A novel remote sensing algorithm to quantify phycocyanin in cyanobacterial algal blooms

We present a novel three-band algorithm (PC3) to retrieve phycocyanin (PC) pigment concentration in cyanobacteria laden inland waters. The water sample and remote sensing reflectance data used for PC3 calibration and validation were acquired from highly turbid productive catfish aquaculture ponds. Since the characteristic PC absorption feature at 620 nm is contaminated with residual chlorophyll-a (Chl-a) absorption, we propose a coefficient (ψ) for isolating the PC absorption component at 620 nm. Results show that inclusion of the model coefficient relating Chl-a absorption at 620 nm–665 nm enables PC3 to compensate for the confounding effect of Chl-a at the PC absorption band and considerably increases the accuracy of the PC prediction algorithm. In the current dataset, PC3 produced the lowest mean relative error of prediction among all PC algorithms considered in this research. Moreover, PC3 eliminates the nonlinear sensitivity issue of PC algorithms particularly at high PC range (>100 μg L−1). Therefore, introduction of PC3 will have an immediate positive impact on studies monitoring inland and coastal cyanobacterial harmful algal blooms.


Introduction
Cyanobacterial harmful algal blooms (CyanoHABs) are becoming a frequent phenomenon in inland lakes and reservoirs, estuaries, and coastal waters all over the world due to various factors including nutrient loading from agricultural and urban runoff, increased frequency of drought and associated reduction in water quantity, and ecological regime shifts triggered by external events. Increase in nutrient loading has been linked with severe CyanoHABs in numerous water bodies around the world including Lake Taihu, China , Patos Lagoon, Brazil (Yunes et al 1996), Gulf of Finland (Bianchi et al 2000), Lake Erie, USA (Vincent et al 2004), Lake Pontchartrain, USA (Mishra and Mishra 2010) eventually leading to anoxia and harmful impacts on aquatic life and human health. CyanoHABs have been broadly recognized as a human and animal health problem because of the wide varieties of toxins (cyanotoxins) associated with them. Some cyanotoxins are strong neurotoxins (anatoxin-a and saxitoxins), others are primarily hepatotoxins (microcystins, nodularin, and cylindrospermopsin). The hazard to human health caused by cyanotoxins can be estimated from toxicological knowledge in combination with information on their occurrence (World Health Organization 1999). Quantitative surveys on Cyano-HABs occurrence are lacking, and therefore, incidence of cyanotoxin exposure through drinking water or during recreational water activities is largely unknown. A recent study reports that climate change may even worsen this environmental problem as rise in global temperature could affect the physiological and molecular changes in Environmental Research Letters Environ. Res. Lett. 9 (2014) 114003 (9pp) doi: 10.1088/1748-9326/9/11/114003 Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. cyanobacteria and consequently increasing hepatotoxin production (El-Shehawy et al 2012). Therefore, effective and accurate monitoring solutions should be developed for early warning systems or analyzing the triggering mechanism of a CyanoHABs event.
For remote sensing applications, researchers have extensively used the accessary photopigment, phycocyanin (PC), as the marker pigment for estimating the presence of cyanobacteria (Dekker 1993, Simis et al 2005, Mishra et al 2009. PC absorption feature at ∼620 nm in the remote sensing reflectance, R rs (λ), spectrum has been extensively used to develop empirical and semi/quasi-analytical algorithms (QAAs) to detect PC concentration in inland water bodies (Dekker 1993, Simis et al 2005, Hunter et al 2010, 2014. Over the years, five broad types of algorithms have been proposed to quantify PC from remote sensing reflectance data based on its absorption feature (a trough in reflectance) at 620 nm. They include (1) reflectance band ratio empirical algorithms (Schalles and Yacobi 2000, Vincent et al 2004, Mishra et al 2009 (2) a semi-analytical baseline subtraction algorithm (Dekker 1993); and (3) a nested semi-analytical band ratio algorithm (Simis et al 2005); (4) a three band semi-analytical algorithm (Hunter et al 2010); and (5) a QAA (Mishra et al , 2014. Existing algorithms (Dekker 1993, Simis et al 2005, 2007 have reported the effect of accessory photopigments on the PC prediction accuracy. Dekker (1993) documented that other than Chl-a, accessory photopigments such as Chl-b, Chl-c, and allophycocyanin-b also absorb light in the photosysthetically active region of the electromagnetic spectrum with the in vitro absorption peaks at 643-645 nm, 628 nm, and 618 nm respectively which overlap the PC absorption peak. Therefore, presence of accessory pigment would lead to overestimation of PC concentrations. Similarly, Simis et al (2007) reported that accessory pigments such as Chl-b and fucoxanthin could significantly increase absorption at 620 nm consequently overestimating PC estimations especially at low PC concentrations (PC < 50 μg L −1 ). To address this issue, semi/QAAs have focused on isolating PC absorption components by resolving absorption and particulate back-scattering from non-PC optically active constituents in the water (Dekker 1993, Simis et al 2005, 2014. Our previously proposed QAA  showed an improved performance compared to the nested semi-analytical algorithm (Simis et al 2005). QAA produced a median relative error of 22.08%, whereas, the nested semianalytical algorithm produced a model error of 45.96% in the same dataset . Most of the model uncertainty in case of QAA was believed to be resulting from the inversion step where the R rs (λ) signal is inverted to total absorption signal using bio-optical models and radiative transfer theory. Because of the physical nature of QAA algorithm, errors associated with the bio-optical inversion can be attributed to the total absorption and particulate backscattering coefficient modeling error at the reference wavelength, a t (λ 0 ) (see Lee et al 2002). Through a systematic error analysis,  reported that, for coastal waters, errors from the modeled total absorption at the reference wavelength will have larger effect, as compared to particlebackscattering coefficient, b bp (λ), modeling errors on the total absorption at any other wavelengths, a t (λ). It was also found that for strongly absorbing waters, when a t (440) approaches 0.5 m −1 , uncertainty in retrieving a t (440) goes up to ± 37% . In case of turbid productive waters similar to our current study site, bio-optical inversion accuracy of total absorption coefficient decreases and eventually reduces the PC estimation accuracy. There could be even higher uncertainties in PC pigment absorption, a φ (λ), and absorption by colored detrital matter, a CDM (λ), estimations because of the physical nature of the decomposition of a t (λ) signal into their individual components (refer to Mishra et al 2013 for a detailed description of the algorithm and modeling uncertainty). Therefore, a simplified yet accurate approach is still needed where the model uncertainty from bio-optical inversion could be avoided while addressing the noise from the Chl-a absorption at the PC absorption maximum (620 nm).
The proposed algorithm is based on a three-band algorithm conceptualized by Gitelson et al (2003) to estimate Chla concentration in higher plant leaves. The algorithm was designed to isolate the absorption coefficient of pigment of interest i.e., Chl-a by using reflectance measurements at three specific bands as below.  (2010) developed a three-band algorithm to quantify PC concentration by setting the λ 1 , λ 2 , and λ 3 at 620 nm, 600 nm, and 725 nm. However, during cyanobacterial bloom conditions, PC absorption (a PC ) at 600 nm and 620 nm are often very close ( figure 1(a)). For example, in our dataset, a PC (620)/a PC (600) varied between 0.94 and1.12 with an average value of 1.02. Therefore, it is not advisable to use 600 nm as λ 2 because R rs (600) is almost equally sensitive to PC concentration as R rs (620). Because of the fact that Chl-a has significant absorption in the spectral region where PC has its absorption maximum (620 nm), developing a three-band algorithm for PC quantification is not straightforward. It is evident from existing literature that the removal of Chl-a contribution at 620 nm would be critical for the successful isolation of PC contribution (Simis et al 2005, Mishra et al 2009, Ogashawara et al 2013. In this research, we developed a conceptual three-band PC algorithm (PC 3 ) using field radiometric and pigment dataset collected from highly turbid and productive aquaculture ponds dominated by cyanobacteria. We proposed a coefficient (ψ) in order to remove the confounding effects of Chl-a absorption and to isolate the PC absorption component at 620 nm in order to enhance PC predictive accuracy. The specific objective of this research was to calibrate and validate the conceptual algorithm using in situ dataset with very high Chl-a and PC content to demonstrate the sensitivity of the model at extremely high PC concentrations level.

Data and methods
In situ data comprising of water samples for pigment extraction and R rs (λ) were collected from 15 aquaculture ponds at the Delta Research Extension Center located near Stoneville, MS, USA during 13-16 July 2010 and 28-29 April 2011.

Pigment measurements
Water samples for Chl-a and Chl-b analysis were simultaneously collected in 1L Niskin bottles and immediately filtered onto GF/F filters (Whatman, 0.7 μm pore size) under low vacuum (<5 inch of Mercury). Samples were extracted in triplicates using 90% acetone and concentrations were measured using high performance liquid chromatography following the Environmental Protection Agency method 447 (Arar 1997). Water samples for PC analysis were filtered immediately after collection through a 0.2 μm nucleopore membrane filters (Milipore) under low vacuum. PC pigments were extracted in 50 mM phosphate buffer and measured using spectrophotometric method. A detailed description of the pigment extraction method has been published in Mishra et al (2013).

Spectral measurements
In situ radiometric data were collected within a three-hour window centering the local solar noon. Sky condition was clear during the field sampling days. A dual sensor system with two inter-calibrated Ocean Optics (Ocean Optics, Dunedin, FL, USA) spectroradiometers were used to collect R rs data in the range of 400-900 nm with a sampling interval of 0.3 nm as given in Mishra et al (2013). Radiometer 1, equipped with a 25°field-of-view optical fiber measured the upwelling radiance just below the air water interface; whereas, radiometer 2, equipped with an optical fiber and cosine diffuser (yielding a hemispherical field of view) acquired above surface downwelling irradiance. To match their transfer functions, inter-calibration of the radiometers was accomplished by measuring the upwelling radiance of a white Spectralon reflectance standard (Labsphere, North Sutton, NH) simultaneously with incident irradiance. The two radiometers were inter-calibrated immediately before and after measurements at each field site. After the data acquisition, R rs was calculated as below: rs 2 u d where, t is the transmittance at the air-water interface (0.98); n is the refractive index of water (1.34); L u (λ) and E d (λ) are the upwelling radiance measured below the air-water interface and downwelling irradiance measured above-surface. For each station, six consecutive scans were recorded and further averaged to calculate a representative R rs (λ) spectrum (figure 1(b)).

Conceptual algorithm development
Key pigment spectral features generally observed in reflectance data from a cyanobacteria dominated turbid productive Absorption coefficients of phytoplankton pigments, and tripton were measured in vivo through quantitative filtration technique, whereas phycocyanin and CDOM absorption coefficients were measured using spectrophotometric method. Water absorption coefficients are from Pope and Fry (1997) and Smith and Baker (1981); the gray dotted, dashed, and solid line at 600 nm, 620 nm, and 665 nm respectively highlights the phytoplankton and PC absorption at the key spectral regions. Note that chlorophyll absorption (≈phytoplankton absorption-phycocyanin absorption) is quite significant at 620 nm, the commonly used wavelength for phycocyanin absorption maxima; (b) remote sensing reflectance, R rs (Sr −1 ) measured at study sites. Gray lines show the location of 600 nm, 620 nm, and 665 nm bands.
system have been used in the algorithm development and are presented in figure 1(b). Reflectance trough near 620 nm and 665 nm appears due to strong absorption by PC and Chl-a pigments respectively (Schalles and Yacobi 2000, Dall'Olmo et al 2003, Mishra et al 2009. The spectral peak around 650 nm appears because of prominent absorption by chlorophylls and PC on both sides of the peak and PC fluorescence emission at 640 nm (Schalles andYacobi 2000, Mishra et al 2009). Generally, the reflectance trough (i.e., the absorption feature) observed at 620 nm is assumed to be entirely due to PC absorption in empirical band ratio models (Schalles andYacobi 2000, Ogashawara et al 2013). However, we have found that the PC absorption feature at 620 nm is contaminated by Chl-a absorption and that interference is quite significant (Simis et al 2005, 2007, Mishra et al 2009. It is quite evident in figure 1(a) that the difference in phytoplankton absorption (≈Chl-a absorption) and PC absorption is quite significant at 620 nm highlighting the residual Chl-a absorption at the PC absorbing wavelength region. Therefore, elimination of residual Chl-a absorption signal is required in order to estimate PC concentration accurately. In this research, we systematically develop the PC 3 algorithm in two steps: 2.3.1. The optimal PC 3 model. For the optimal PC 3 algorithm development, we used the three-band algorithm framework proposed by Gitelson et al (2003) and adapted by Hunter et al (2010). In order to design the optimal PC 3 algorithm, we carried out best band analysis to select λ 1 , λ 2 , and λ 3 at three successive steps. At step 1, we took the initial value of λ 1 and λ 3 as 620 nm and 778 nm. Initially, we picked 620 nm for λ 1 as it is considered as the PC absorption maxima and hence highly sensitive to PC concentration. Similarly, based on the three-band architecture of the PC 3 , λ 3 is required to be at a wavelength region where pigment absorption is negligible so that it can be used to compensate for the residual backscattering of the medium. We selected 778 nm as λ 3 because it matched ESA's medium resolution imaging spectrometer(MERIS) band architecture (MERIS band 12). Although out of commission, MERIS has a massive archive of data for spatio-temporal studies of cyanobacterial blooms.
Since the ultimate goal was to make the algorithm readily applicable to satellite data, whenever possible even during optimization, initial bands selection for PC 3 was performed keeping MERIS or MERIS-like sensors in consideration. The three-band algorithm was calculated iteratively by systematically selecting the value of λ 2 from 400-800 nm at 1 nm interval. Standard error was calculated for each iteration by least-square linear regression between the model, , and PC concentration. Optimized λ 2 was selected where the model produced minimum standard error of the estimates. At step 2, initial value of λ 1 (620 nm) and optimal value of λ 2 from the step 1 was used to estimate the optimal λ 3 . Similarly at step 3, two optimal values of λ λ and 2 3 were used to estimate the optimal value of λ 1 . From three steps, optimal band selection procedure returned 629 nm, 659 nm, and 724 nm as λ 1 , λ 2 , and λ 3 . Figure 2 shows the standards errors normalized to their range for λ 1 , λ 2 , and λ 3 . PC 3 in the optimal form can be written as: MERIS-like sensors can be written as: However, it should be noted that 665 nm spectral region is highly sensitive to Chl-a absorption, and therefore, subtraction of PC PC ratio, ψ 2 , and filter-pad measured phytoplankton pigment absorption, ϕ a (620) and ϕ a (665) as below .
where, ψ is the Chl-a pigment absorption ratio from 665 nm to 620 nm, a a (665)/ (620) chl chl , and ψ 2 is the PC pigment absorption ratio from 665 nm to 620 nm, a a (665)/ (620) PC PC . The detailed method explaining the pigment absorption and retrieval of ψ is available in .
For remote sensing applications, ψ was empirically modeled from a reflectance band ratio, R R . It should be noted that R rs (560) is sensitive to the presence of cyano-phycoerythrin bearing algae in the water body. However in this study, cyanophycoerythrin will have negligible effect on modeled ψ due to its minimal concentration as compared to abundant presence of chloropyllous pigments and PC. In other water bodies, if significant cyano-phycoerythrin is present than a different reference wavelength needs to be used for modeling ψ.
Finally, we modify PC 3 algorithm in equation (4) by incorporating ψ to remove the confounding effect of Chl-a at 620 nm and to isolate the PC contribution. PC 3 in the final form can be written as: represents the proxy of absorption by Chl-a at 620 nm making the right side of the equation (5) proportional to PC absorption at 620 nm. PC 3 assumes that b b within the spectral range of 620-778 nm is spectrally flat and difference in absorption by colored dissolved organic matter (CDOM) and detrital matter at 620 and 665 nm is negligible. Note that the inclusion of ψ in the model is not a complete analytical treatment for retrieval of a PC (620). However, assuming − R 665 1 to be a proxy of a b [ (665)/ ] b chl , we consider that inclusion of ψ would enable PC 3 to compensate for Chl-a contribution at 620 nm.   To explore the performance of PC 3 using optimal bands, those were selected from the best band analysis, we used all available data points and analyzed it successively in steps.

PC 3 model for MERIS-like sensors
As discussed in the previous section, PC 3 was also designed using the channels available in MERIS-like sensors. This analysis was performed in order to make the model applicable to MERIS data or any other sensor with a band at 620 nm.  4(a)). However, it is evident from the regression output that the model was not sensitive to PC concentration below 300 μg L −1 (figure 4(a)). Low sensitivity in this range can be attributed to the low PC:Chl-a ratio (mean = 0.94) as compared to higher PC:Chl-a ratio (mean = 1.97) in samples with PC concentration greater than 300 μg L −1 Lower PC:Chl-a ratio indicates that the samples in the lower range are dominated by Chl-a. We addressed this issue in PC 3 by using model coefficient ψ (equation (5)). After incorporating ψ in the PC 3 , relationship between PC concentration and PC 3 substantially increased (R 2 = 0.98, p < 0.0001) ( figure 4(b)). The model produced a standard error of 83.7 μg L −1 , which is almost one third of the error from the model before the incorporation of ψ. Results from the regression analysis produced evidence that the model coefficient ψ is able to address the issue of interference of Chl-a absorption at the PC absorption band (620 nm) and enables accurate isolation of PC component.
In order to calibrate and validate the PC 3 model, we randomly divided the dataset into two groups. 70% of dataset was used for model calibration and the rest was used for model validation. In the calibration dataset, the model in the form of PC ( μg L −1 ) = 480.92*PC 3 + 123.23 accounted for 98% of the variance in PC concentration (p < 0.0001) ( figure 5(a)). For PC 3 accuracy assessment, predicted PC values were compared with analytically measured PC concentrations. Results showed that the model produced a mean relative error, ∑ − ⎡ ⎣ ⎤ ⎦ (( measured PC predicted PC )/measured PC)*100 n n 1 1 of 30.71%. The slope of the best-fit line between predicted and measured PC was 0.97 (R 2 = 0.99) ( figure 5(b)). Overall, after correcting for the confounding effects of Chl-a at 620 nm PC 3 performed significantly better than all PC models considered in this research (table 2). Mean relative error of PC 3 was ∼12% lower than Hunter et al (2010) and ∼4% lower than the quasi-analytical PC algorithm by Mishra et al (2013). Similarly, improvement in accuracy was ∼52% higher as compared to the semi-analytical algorithm (Simis et al 2005) in the current dataset (table 2). We also assessed the performance of PC 3 at the lower range of PC concentration (<300 μg L −1 ) through a systematic validation analysis and comparison with existing PC algorithms considered in this research. Results showed that PC 3 performed most accurately compared to other existing algorithms when validated at the low range (below 300 μg L −1 ). Overall, PC 3 , Mishra et al (2013), and Hunter et al (2010) showed high prediction accuracy even in the lower range by producing 38.18%, 40.26%, and 53.67% of MRE respectively (table 2). It is apparent from the validation analysis that accurate PC prediction for waters with low to moderate PC concentration is still an issue. We believe that the mediocre performance of PC 3 and other algorithms at low to moderate range is mainly due to interference from other optically active constituents in the water primarily Chl-a. Since relative PC absorption is low for water bodies with low to moderate PC concentrations, Chl-a interference plays a dominant role and is often difficult to isolate from PC absorption at 620 nm. This is due to the fact that at low PC:Chl-a scenarios, confounding signal from Chl-a absorption at 620 overpowers the weak PC signal thereby decreasing the signal to noise ratio and increasing the uncertainty of PC absorption component retrieval. Although PC 3 performs relatively well as compared to other algorithms, the model error is considered to be high (38.18%) for the low to medium PC range. This is probably because the accuracy of the empirical model we used to derive ψ (figure 3) decreases for low to moderate PC concentrations. As ψ is the critical parameter that addresses Chl-a interference, any error with ψ retrieval would be carried over to the final PC prediction stage. In our future research, we will parameterize ψ for the low PC concentration scenarios in order to improve the prediction accuracy of PC 3 for low to moderate PC concentration. We believe that inclusion of data points in the 0-70 μg L −1 PC range would improve the model prediction accuracy making it widely applicable.
Reflectance band ratio models often show lower sensitivity to pigment of interest because of saturation in absorption at high pigment concentration. Figure 6 explains this phenomena and shows the nonlinear sensitivity of a band ratio, R rs (620)/R rs (708), to PC concentration. The nonlinearity is most likely due to the saturation in pigment absorption at high concentration range and it is an inherent difficulty  Based on the error analysis reported by  where the same dataset was used for model calibration and validation. associated with band-ratio algorithms. To its advantage, PC 3 showed a linear trend within the entire PC range even at very high PC concentration unlike its band ratio counterparts (figure 4(a)), even when compared to PC 3 before the introduction of ψ (equation (4)). So far semi-analytical and QAAs (Simis et al 2005, Mishra et al 2013) needed a reflectance band at 709 nm, which is only available in limited sensors, to estimate PC. However, in this research we have proposed an innovative method to address the overlapping Chl-a absorption issue at 620 nm by using reflectance data measured at 620 nm, 665 nm, and at a longer NIR band, 778 nm. Therefore, application of PC 3 can be extended to any sensor even without the 709 nm band.

Conclusion
In this research, we have developed a three-band PC algorithm, PC 3 , to accurately retrieve PC concentration in inland water bodies. Using best band selection analysis, we selected three optimal bands centered at 629 nm, 659 nm, and 724 nm for λ 1 , λ 2 , and λ 3 respectively. The optimal PC 3 model showed strong PC predictive ability in the current dataset (STE = 150.38 μg L −1 ). However, for practical applicability of the algorithm, PC 3 was also designed using the channels available in MERIS-like sensors i.e., 620 nm, 665 nm and 778 nm as λ 1 , λ 2 , and λ 3 respectively. PC 3 for MERIS-like sensors produced relatively lower prediction accuracy (STE = 232.41 μg L −1 ) than the optimal PC 3 algorithm. We concluded that the high standard error observed in both variants of the PC 3 algorithm was mainly due to the interference of Chl-a absorption at PC absorption maximum (620 nm). To address the lack of sensitivity of PC 3 at low to moderate PC range (<300 μg L −1 ), we introduced a model coefficient, ψ, to remove the effects of Chl-a absorption at 620 nm. In the final form, PC 3 for MERIS-like sensors produced highest PC prediction accuracy (MRE = 30.71%) compared to all algorithms considered in this research. PC 3 model prediction accuracy somewhat decreased when validated for the low to moderate PC range, although it produced the minimum MRE among all algorithms. It is evident from the results that: (a) PC 3 can be applied to retrieve cyanobacterial PC concentration in inland waters with considerably higher accuracy than existing algorithms; (b) inclusion of ψ in the model compensates for the confounding effect of Chl-a absorption at the PC absorption band, i.e. 620 nm; and (c) unlike reflectance band ratios, PC 3 addresses the nonlinearity issue at higher PC range. Development of remote sensing algorithms for quantifying cyanobacterial PC is a challenging task primarily because specific absorption coefficient of PC and Chl-a varies significantly with change in phytoplankton species composition. In addition, Chl-a absorption at the PC absorbance maxima causes significant interference that adds to the model complexities (Dekker 1993, Simis et al 2005. Therefore, modeling errors from all PC algorithms considered in this research were quite high (∼30-204%). Results from this analysis supports the previous findings that PC algorithms perform poorly when cyanobacteria population in the phytoplankton community is relatively smaller as documented in Hunter et al (2010), Ruiz-Verdu et al (2008), . Despite the modeling complexities, the newly developed PC 3 algorithm for MERIS-like sensors performed best among all PC algorithms considered in this research. It should also be noted that PC 3 does not require the 709 nm band, which is used in the semi-analytical algorithm (Simis et al 2005) and QAA  to retrieve PC absorption coefficients. As PC 3 does not require the 709 nm band, we anticipate that it can be widely applicable to other sensors, i.e., sensors without the 709 nm band and hence another advantage over many existing PC algorithms.
In this research, model calibration and validation analysis was performed on a small dataset with extreme values of PC concentration. It should be noted that the model parameter, ψ, plays a critical role in the model configuration and separation of Chl-a interference. Therefore, in future the empirical model that retrieve ψ from the reflectance band ratio should be considered for recalibration specifically for low PC concentrations for robust performance of PC 3 . It should also be noted that PC 3 has been calibrated and validated using a dataset with PC concentration varying from 68.13 to 3032.47 μg L −1 . Hence the algorithm should be applied as is to datasets within this PC range. Our future research effort will use a comprehensive dataset including data particularly in the lower range (0-70 μg L −1 ) of PC concentration to recalibrate and validate the algorithm for wider applicability. A large dataset with a wide range of PC concentration will also help to validate PC 3 , and systematically compare its performance with other existing PC algorithms. Findings from this research can be leveraged to any current and future satellite or airborne sensor with the specific bands to monitor cyanobacterial algal blooms. Therefore, introduction of PC 3 will have an immediate positive impact on studies monitoring inland and coastal CyanoHABs.