Drone-based hyperspectral remote sensing of cyanobacteria using vertical cumulative pigment concentration in a deep reservoir

Remote Sensing of Environment 236 (2020) 111517

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Drone-based hyperspectral remote sensing of cyanobacteria using vertical cumulative pigment concentration in a deep reservoir

T

Yong Sung Kwona, JongCheol Pyoa, Yong-Hwan Kwonb, Hongtao Duanc, Kyung Hwa Choa,∗∗, Yongeun Parkd,∗ a

School of Urban and Environmental Engineering, Ulsan National Institute of Science and Technology, Ulsan, 44919, Republic of Korea Electronics and Telecommunication Research Institute (ETRI), 218 Gajeong-ro, Yuseong-gu, Deajeon, 305-700, Republic of Korea c Key Laboratory of Watershed Geographic Sciences, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing, 210008, China d School of Civil and Environmental Engineering, Konkuk University, Seoul, 05029, Republic of Korea b

A R T I C LE I N FO

A B S T R A C T

Keywords: Vertical cumulative pigment concentration Phycocyanin Subsurface remote sensing reflectance Drone-based hyperspectral imagery Bio-optical algorithm

The remote sensing of algal pigments is essential for understanding the temporal and spatial distribution of harmful algal blooms (HABs). In particular, the vertical distribution of cyanobacterial pigment (e.g., phycocyanin (PC)) is critical factor in remote sensing because the diel vertical migration of cyanobacteria may affect the spectral signals according to observational time. Although numerous studies have been conducted on the remote sensing of algal bloom using pigments, few studies considered the vertical distribution of the pigments for the remote sensing of cyanobacteria in inland waters. In this regard, the objective of this study was to develop an improved bio-optical remote-sensing method using in-situ remote-sensing reflectance (Rrs) at different water depths and cumulative PC and Chlorophyll-a (Chl-a) concentrations, which was cumulated from the surface to a 5-m water depth. The results showed that the bio-optical algorithm using surface Rrs and surface pigment concentration was more accurate than that using the subsurface Rrs and surface pigments. The bio-optical algorithm using subsurface Rrs showed the highest R-squared (R2) values (0.87–0.94) in each regression with the cumulative PC concentration from surface to each depth. The regressions between drone-based surface reflectance and cumulative PC concentration for each depth indicated a better performance than those between the reflectance and surface PC concentration; the highest R2 value of 0.82 was obtained from a bio-optical algorithm using drone-based reflectance and a 1.0-m cumulative PC concentration, which was the best-performing algorithm. The PC maps developed using the best bio-optical algorithm accurately described the spatial and temporal distributions of the PC concentrations in the reservoir. This study demonstrates that the application of vertical cumulative pigment concentration and subsurface Rrs measurement in bio-optical algorithms can improve their performance in estimating pigments, and that drone-based hyperspectral imagery is an efficient tool for the remote sensing of cyanobacterial pigments over a wide area.

1. Introduction Harmful algal blooms (HABs) cause crucial problems associated with freshwater aquatic ecosystems and human health, such as dissolved oxygen depletion and production of harmful toxins, which lead to the death of aquatic organisms (Carmichael and Boyer, 2016; Hilborn and Beasley, 2015; Huisman et al., 2018). To improve water quality management, it is important to better understand the spatial and temporal distribution of HABs and their influences on the aquatic ecosystem. Traditional field water sample collection, however, cannot adequately consider their spatial and temporal distribution because of ∗

the high spatial variability of HABs in waterbodies. In addition, substantial time and labor is required to monitor entire lakes or reservoirs (Power et al., 2005; Ritchie et al., 2003). Remote-sensing techniques could be an alternative solution because they allow broad spatial coverage and a short data-acquisition time (Brando and Dekker, 2003; Brezonik et al., 2005; Dekker, 1993; Hilton, 1984; Kallio et al., 2001). Many researchers have investigated the quantification of phytoplankton abundance or algal pigment, i.e., chlorophyll-a (Chl-a) concentration in inland waters (Mittenzwey et al., 1992; Pulliainen et al., 2001; Tyler et al., 2006; Yacobi et al., 2011). The concentration of phycocyanin (PC), an accessory pigment of

Corresponding author. Corresponding author. E-mail addresses: [email protected] (K.H. Cho), [email protected] (Y. Park).

∗∗

https://doi.org/10.1016/j.rse.2019.111517 Received 31 January 2019; Received in revised form 17 October 2019; Accepted 29 October 2019 0034-4257/ © 2019 Elsevier Inc. All rights reserved.

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

remote sensing signals (Kay et al., 2009; Kutser et al. 2009, 2013; Streher et al., 2014). Some researchers have investigated the effects of subsurface pigment-concentration profiles on the remote-sensing signal using Hydrolight-Ecolight (Mobley and Sundman, 2008; Xue et al., 2017), a subsurface light-field model Stramska and Stramski (2005); (Kutser et al., 2008; Pitarch et al., 2014; Xiu et al., 2008). Recently, Xue et al. (2015) documented a method to estimate the vertical distribution of chlorophyll-a class using remote sensing reflectance data. Previous studies, however, have typically focused on the analysis of relationships between the vertical pigment distributions from various optical properties of water bodies and the remote sensing signals. There are limited investigative studies on the comprehensive relation between the vertically cumulative pigment concentration and the vertical distribution of Rrs, and its application to bio-optical algorithms. To address these issues, the main objectives of this study are the following: (1) to measure the vertical pigment-concentration profile, portable sensor-based Rrs at the surface and three subsurface depths (i.e., subsurface, 0.5-m depth, and 1-m depth), and drone-based surface reflectance; (2) to compare the performance of the bio-optical algorithms using the Rrs at different sensing depths, surface reflectances, and vertical pigment-concentration profiles; and (3) to determine the spatial and temporal distribution of algal pigments through the bestperformance bio-optical algorithm using drone-based hyperspectral images.

cyanobacteria, has been used for the identification of cyanobacterial HABs (Kudela et al., 2015; Kutser, 2004; Kutser et al., 2006; Matthews et al., 2010; Park et al., 2017; Randolph et al., 2008). Multispectral satellite remote sensing has been widely used to estimate the pigment concentrations of HABs in both inland and coastal waters (Ahn et al., 2006; Gokaraju et al., 2011; Kutser, 2004; Kutser et al., 2006; Kwon et al., 2018; Proctor and Roesler, 2010; Wynne et al., 2010). There are numerous satellite-borne sensors of different spectral, spatial, and temporal resolutions. For example, the Landsat-8 operational land imager (OLI) and Sentinel-2 have been applied to various rivers and lakes to detect HABs because these sensors have relatively high spatial resolutions and spectral bands that detect algal pigment absorption peaks (Ansper and Alikas, 2019; Kwon et al., 2018; Mollaee, 2018; Toming et al., 2016; Zhao et al., 2017). However, the coarse spectral resolution of these sensors, may affect the performance of algal pigment detection (Palmer et al., 2015a, 2015b). Spectral sensors, which have a broad bandwidth from 60 to 80 nm, have difficulty discriminating the absorption features of algal pigments (Dekker et al., 1992). The seaviewing wide field-of-view sensor (SeaWiFS) and moderate resolution imaging spectroradiometer/aqua (MODIS-Aqua) have finer spectral resolutions than Landsat-8 and Sentinel-2, but the spatial resolution is insufficient to monitor the water quality in rivers and lakes. Recently, the medium resolution imaging spectrometer (MERIS) and Sentinel-3 OLCI, which have narrower spectral bands than SeaWiFS and MODIS, have been used for algal pigment remote sensing in inland waters (Blix et al., 2018; Kiefer et al., 2015; Odermatt et al., 2012; Palmer et al., 2015a, 2015b). Over the last few decades, airborne hyperspectral remote-sensing techniques have been widely used because of their high spectral, spatial, and temporal resolutions (Hunter et al., 2008; Jupp et al., 1994; Park et al., 2017; Roessner et al., 2001; Selige et al., 2006; Shippert, 2003). Hyperspectral imagery is an effective technique for pigment estimation and cyanobacteria bloom detection (Kudela et al., 2015; Kutser, 2004; Li et al., 2010; Wang et al., 2016; Wynne et al., 2010). Concentrations of PC and Chl-a have been estimated using apparent optical property (AOP) algorithms with multiple reflectance bands obtained from hyperspectral images (Hunter et al., 2010; Matsushita et al., 2015; Mishra et al., 2009; Pyo et al., 2018). Dekker (1993) introduced two-band ratio algorithms to estimate Chl-a and PC concentrations. With regards to the band selection, the reflectance from the blue-green band (440–560 nm) is used in coastal Chl-a estimation. The red and near-infrared wavelength range (650–800 nm) is mainly used for turbid waters because the colored dissolved organic matter (CDOM) absorbs the blue-green light while the mineral particles scatter the light (Gilerson et al., 2010; Gitelson et al. 2007, 2008; Mittenzwey et al., 1992; Pyo et al. 2016, 2017; Randolph et al., 2008). These hyperspectral imagery techniques of remote sensing of cyanobacteria focus mainly on reflectance on the water surface and the pigment concentrations of surface water (Kutser et al., 2006; Mishra et al., 2009; Randolph et al., 2008; Simis et al., 2005; Vincent et al., 2004). Using the surface reflectance may have a limitation in remotely detecting cyanobacterial pigments because of the vertical migration and distribution of cyanobacteria (Kromkamp and Mur, 1984; Nadeau et al., 1999; Oliver, 1994; Richardson and Castenholz, 1987; Visser et al., 1997; Wallace et al., 2000). In addition, this is closely related to photosynthesis via solar irradiation and wind-induced mixing. Hydrocarbon accumulation via photosynthesis results in cells sinking. However, mixing by wind may cause the cyanobacterial cells to disperse throughout the water column; the vertical distribution of cells is considerably affected by wind conditions during both the day and night (Ha et al., 2000; Wu and Kong, 2009). Moreover, a sun-glint effect based on the solar zenith angle has been documented since the 1980s, and Bhatti et al., (2009) found that sun glint caused by a surface wave resulted in a magnitude difference between the surface and subsurface reflectance in inland waters. Various studies have developed removal techniques of the sun-glint effect from

2. Methodology 2.1. Study area The Daechung Reservoir, constructed in June 1981 (N 36.35°–36.52°, E 127.48°–127.60°), is an artificial reservoir that supplies water to surrounding cities for domestic and industrial use. Its watershed and water surface area are 4134 and 72.8 km2, respectively. The length of the reservoir is 86 km and its storage volume is 1490 × 106 m3 (Shim et al., 2015). Three dominant algal genus, namely Anabaena, Microcystis, and Oscillatoria, were observed from early summer to late autumn (Shin et al., 1999). The Sookcheon region was chosen as the study area because previous researchers have regarded it as the main blooming area (Lee et al., 2014; Tae et al., 2014). The study area was divided into three sub-sections (A, B, and C) because of the limitations in the drone operation; the main monitoring section was section C (Fig. 1). 2.2. Data acquisition 2.2.1. Water sampling Seven surface-water sampling events were conducted from July 26 to October 25 in 2018. The observations are described in Table 1. Two different-sized water samples (i.e., 125 ml and 2 L) were collected for different purposes. The 2-L surface-water sample was collected for the Chl-a concentration calculation, whereas the 125-ml sample was used to calculate the PC concentration. A 2-L sterilized water sample bottle and 125-ml wide-mouthed bottle were used, respectively. The surfacewater samples were collected directly using a 5-L plastic beaker. A preconcentrated water sample was placed in a 125-ml bottle using a phytoplankton net of 20-μm mesh size. The volume of the water for preconcentration varied from 5 to 20 L according to the phytoplankton concentration. All water samples were stored in a box with ice and were moved to the laboratory for experimental investigations. 2.2.2. Pigment extraction in the laboratory We quantified the PC and Chl-a concentrations in the laboratory within 24 h of water sampling. PC is a pigment-protein complex that harvests sunlight for photosynthesis (Glazer, 1989; Jin et al., 2018). The PC extraction was based on methods from previous studies (Bennett and Bogorad, 1973; Sarada et al., 1999). The detailed procedures are 2

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 1. Study area and specific monitoring sections. Each field monitoring station are represented by markers according to monitoring events.

where, OD620 and OD652 are the optical densities and each subscript represents specific wavelengths, 620 nm and 652 nm, respectively. For the Chl-a extraction and quantification, a standard experimental method was followed (APHA , 2001). The extraction procedure of Chl-a is described in the following statements. (1) A volume of 250 ml water from the 2-L sample was filtered using a glass microfiber filter (Whatman) with a 0.7-μm pore size. (2) The filtered sample was only stored in a 15-ml conical tube, and then 10 ml of a 9:1 acetone and methanol solution was added to the sample. The resulting sample was placed in a -4-°C refrigerator for 24 h, without the grinding process to prevent the influence of ground filter paper. (3) The sample was centrifuged at 500 g and 20 °C for 20 min (4) The optical densities of the sample's supernatant were measured using a Cary-500 UV–Vis–NIR spectrophotometer. Then, the concentration of Chl-a was estimated using the following equation:

described in the following statements. (1) Pre-concentrated samples from netting were homogenized via an Ultra-Sonicator (Sonictopia Inc., Cheongju, Korea). (2) A total of 30 ml of homogenized sample was moved to a 50-ml conical tube, which was then centrifuged at 4000 rpm and 5 °C for 15 min (3) The supernatant was removed by pipetting the remaining liquid from the pellet. (4) A 5-ml phosphate buffer solution of pH 7.2 was added to the sample. (5) The PC was then extracted by applying a freezing-and-thawing method. The sample was frozen at −20 °C for 24 h and then thawed at room temperature to break the algal cells. (6) To release the PC pigment from the cyanobacterial cells, the samples were placed in a shaking incubator, which was then operated at 150 rpm for 15 min (7) The centrifugation process was repeated under the condition that was applied during the prior step (step 2) to analyze the supernatants. (8) Using a Cary-500 UV–Vis–NIR spectrophotometer (Agilent Inc., Santa Clara, CA, USA), the optical densities of the supernatant samples were measured throughout the 200- to 3300-nm wavelength range. Afterwards, the PC concentration was estimated using the following equation:

PC (mg / m3)

OD620 − (0.474 × OD652) = 5.34

Chl − a (mg / m3) =

(11.64 X1 − 2.16X2 + 0.10 X3) × V1 V2

(2)

where, X1, X2, and X3 represent (OD663-OD750), (OD645-OD750), and (OD630-OD750), respectively; V1 and V2 are the supernatant and filtered sample volumes, respectively.

(1)

Table 1 Information regarding drone operation and the number of in-situ observations. Sampling date

Number of surface water samples

Number of YSI-sensor measurements

Number of in-situ Rrs observation

Number of hyperspectral images

Sampling region

18/07/26–27 18/08/16–17 18/09/06 18/09/17 18/09/19 18/10/11 18/10/25

16 16 12 12 12 14 10

15 17 12 12 12 14 10

15 11 12 10 11 14 10

8 8 7 7 7 8 7

A, B and C A, B and C C C C C C

3

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

2.2.3. Field water quality measurement using EXO-2 YSI A YSI EXO-2 water-quality multi-probe (YSI, Yellow Springs, Ohio, USA) was used to monitor the vertical water-quality profile. The five water quality parameters, namely water temperature, dissolved oxygen, pH, blue-green algae PC (BGA-PC), and chlorophyll-a concentration, were continuously measured from the surface to a 5-m depth at a 0.5-s interval; the depth was determined by reducing the time gap between the hyperspectral sensing and field measurement. The continuous record of the vertical profile was converted to depth-versus-concentration data at a 0.05-m interval. To obtain the vertical cumulative-pigment concentration, the depth profile was integrated with a 50-cm interval using the following equation:

Fig. 2. Picture of drone and mounted hyperspectral sensor used in this study. MATRICE M600 Pro hexacopter drone (A) and Nano-Hyperspec® hyperspectral imaging sensor (B).

z/i

[vertical cumulative concentration] (mg / m2 ⋅ z ) =

∑ i⋅ Cn n=1

drone has 9.5 and 15.5 kg of weight and maximum takeoff weight, respectively. The sensor is a push-broom type sensor with an 8-mm lens which weighed 520 g. This type of sensor acquires the entire line of the image at once and captures the image continuously with the motion of drone (Kerekes and Schott, 2007). The wavelength ranges from 400 to 1000 nm (Hill and Clemens, 2015), whereas the minimum spectral resolution and maximum number of channels of the sensor are 2 nm and 270 channels, respectively (Beth, 2016). A sensor field of view (FOV) of 32.2° was set using an 8-mm lens. The number of hyperspectral images and sampling regions are described in Table 1. In this study, 112 bands were used to reduce the data size. The altitude of the drone hover was 150 m and the spatial resolution of the image was 20 cm, while the swath width was approximately 87 m at that altitude. The collected hyperspectral images were corrected using the SpectralView® software (Headwall Photonics Inc., MA, US). From the raw digital number data, radiance correction was conducted to calculate the radiance data from the raw digital number data using a dark reference, which was collected in the field with the lens cap cover and an internal calibration file that has optimal gains and offsets for digital numbers in each wavelength band (i.e., Radiance = Gain*Digital number + offset). After finishing the radiometric calibration, the surface reflectance was calculated with an in-scene reference, radiometric tarp. The reflectance tarp of 70% was located on flat ground to collect reference reflectance when the drone was flying. The internal calibrated reflectance spectra in the tarp file of the software was used as the reference for correcting the calibrated radiance to surface reflectance. The white reference data helped in providing the surface reflectance adjustment during reflectance generation. Prior to drone monitoring, the white reference data was collected using Spectralon; thus, the final output of the image-processing software was surface reflectance (Rdrone). Afterwards, ortho-rectification was applied using the lens, sensor position, and GPS information of roll, pitch, and yaw. The dronebased surface reflectance at each sampling station with the same GPS position as each water sampling station was extracted from the corrected images. The noisy remote-sensing signal can be corrected effectively using a filtering method (Zeng et al., 2017). In this study, Savitzky-Golay filtering was applied to minimize the noise of the dronebased surface reflectance spectra (Chen et al., 2004). A two-dimensional adaptive filter, Wiener2, was applied after the PC distribution map was generated (Lim 1990). Wiener2 filtering is a pixel-wise method using the local neighbors’ statistics of each image pixel; it is effective in reducing the speckle noise in remote-sensing images (Agrawal and Venugopalan, 2011). Atmospheric correction was not performed because the operational altitude of the drone was 150 m, which is very low compared to those of satellite and air-craft sensing (Adão et al., 2017). In water quality remote sensing, atmospheric correction has not been conducted because of the negligible atmospheric effect of the low-altitude, unmanned, aerial vehicle (UAV) operation (Kislik et al., 2018; Su and Chou, 2015).

(3)

where, z is the cumulative depth, i the depth profile interval, and Cn the concentration of n-th component of the depth profile. 2.2.4. In-situ Rrs measurement using a portable sensor Remote-sensing reflectance (Rrs) data were collected from 82 sampling stations using a FieldSpec HandHeld 2 spectroradiometer (ASD Inc., Boulder, CO, USA). The wavelength ranged from 325 to 1075 nm. This study focused on the wavelength region from 450 to 730 nm because the signals that were below the 450 nm and above the 730 nm spectra included sensing noise (Murphy et al., 2008; Olmanson et al., 2013; Uitz et al., 2015). The downwelling irradiance was measured using a detachable remote cosine receptor (RCR). The RCR was removed from the spectroradiometer after measuring the irradiance. Then, the sky and surface (Lw, surf) radiances were measured with a bare fiber. We measured three different subsurface radiances—below the water surface (Lw, sub, 0), 50 cm below the water surface (Lw, sub,50), and 100 cm below the water surface (Lw, sub, 100)—using a 1.5-m jumper cable as an accessory for the spectroradiometer; it is a submergible, waterproof, optical-fiber sensor that is detachable. The end of the fiber was bent to maintain a 40° viewing angle. The visibility of the waterbody was measured using a Secchi disk to check the depth of penetration. The average and standard deviation of the Secchi depth were 0.99 and 0.32 m, respectively, throughout the monitoring events. Throughout the in-situ measurements, the azimuth angle and zenith angle were maintained in the ranges of 130°–135° and 35°–40°, respectively, to reduce the sun-glint effect and shading interference (Li et al., 2013; Simis et al., 2007; Mobley, 1999). Finally, the surface Rrs and subsurface Rrs at three different depths were measured via in-situ measurement. Rrs is the ratio between the irradiance and radiance, which can be described as follows:

Rrs (λ) =

L w (λ ) Ed (λ ) −2

(4) −1

where Ed (W m nm ) is the downwelling irradiance from the sun and Lw (W m−2 nm−1 sr−1) is the radiance from the water. To minimize the influence of the sky, the radiance of the water, Lw, is defined as the following equation:

L w (λ ) = L w, r (λ ) − qLsky (λ ) −2

−1

(5) −1

where, Lw, r (λ) (W m nm sr ) is the water radiance with the sky influence; Lsky (λ) (W m−2 nm−1 sr−1) is the radiance from the sky; and q is the interference factor of the skylight between the water and air, defined as a constant of 0.025. The constant q was selected according to the reference with a 40° viewing angle and a wind speed of less than 5 m/s (Mobley, 1999). 2.2.5. Hyperspectral reflectance measurement using a drone-installed sensor Hyperspectral images were collected using a Nano-Hyperspec® hyperspectral imaging sensor (Headwall Photonics Inc., MA, US) installed on a MATRICE M600 Pro hexacopter (DJI Inc. China) (Fig. 2). The 4

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

2.3. Use of a band-ratio algorithm to estimate pigment concentration

PC and Chl-a.

A band-ratio algorithm was used to estimate the surficial and vertically cumulated pigment concentrations. Based on the literature, the two-band ratio algorithm for PC uses the 709 and 620 nm bands, whereas Chl-a uses the 709 and 665 nm bands (Duan et al., 2010; Gilerson et al., 2010; Gitelson et al., 2007; Liu et al., 2017). In this study, we used the 709, 620, and 665 nm bands to estimate PC and Chla because of the spectral resolution of the available hyperspectral sensor. The algorithms are shown in the following equations:

3.2. Relationship between portable-sensor hyperspectral Rrs and pigment concentration

PC (mg / m3 )  ∝   

Rrs (709) Rrs (620)

Chl − a (mg / m3 )  ∝   

3.2.1. Portable-sensor hyperspectral spectra Fig. S1 shows in-situ Rrs measured from the surface, subsurface, 0.5m subsurface, and 1.0-m subsurface. The Rrs intensities decreased as the measuring depth increased, whereas the noise in the Rrs signals increased as water depth increased. In particular, the intensities in the surface and subsurface measurements showed a remarkable decrease at the approximately 350 nm and above-750 nm bandwidths. Fig. 4 presents the Rrs intensity of the three main wavelengths used in this study, namely 620, 665, and 709 nm. For the three bands, approximately one order of magnitude difference in median Rrs intensity was observed between the surface and 1.0-m subsurface. Median intensities decreased exponentially as the depth increased, and the difference between the maximum and minimum intensity in each band decreased as the depth increased. Reductions in Rrs intensity between the surface and 1.0 msubsurface were 78% for 620 nm, 83% for 665 nm, and 84% for 709 nm.

(6)

Rrs (709) Rrs (665)

(7)

where, Rrs (710), Rrs (620), and Rrs (665) are the remote-sensing reflectances at 709, 620, and 665 nm, respectively. In this study, the band ratios were calculated from the reflectance data obtained from the portable and drone-installed sensors. These band ratios were used for the bio-optical algorithms. The coefficients in the algorithms were obtained from linear regressions between the band ratios and pigment concentrations. The coefficient of determination (Rsquared, R2) from the regressions were calculated; R2 values from the regression between the band ratio and sensor-measured pigment concentration were compared to those from the regression between the band ratio and experimental pigment concentration. The spatial and temporal distributions of PC were developed using the band-ratio algorithm with the highest R2 value.

3.2.2. Surface pigment concentrations and Rrs band ratios Figs. 5 and 6 describe the correlation between pigment concentrations as well as the surface and subsurface band ratio values. Linear regressions between the experimental PC and two band ratios showed low R2 values of 0.2340 and 0.2049 for the surface and subsurface band ratios, respectively. For the YSI-sensor-measured PC, the band ratios showed relatively higher R2 values of 0.5393 and 0.5038, respectively. In contrast to the PC shown in Fig. 5, the Chl-a concentrations in Fig. 6 revealed opposite trends. The experimental Chl-a concentration showed a higher correlation with the band ratios with R2 values of 0.4461 and 0.4070 for the surface and subsurface band ratios, respectively. In contrast, the YSI-sensor-measured Chl-a was found to have no correlation with the band ratios, yielding R2 values of 0.0350 and 0.0132 for the surface and subsurface band ratios, respectively.

3. Results 3.1. Pigment concentrations in surface water The descriptive statistics for pigment data in surface water are shown in Table 2, including the mean, minimum, maximum, and standard deviation. Overall, the experimental pigment concentrations are higher than the YSI-sensor-measured values. The highest concentrations of PC and Chl-a were observed on September 6th and September 19th, with concentrations of 803.01 and 402 mg/m3, respectively. Fig. 3 shows the temporal variations in the PC and Chl-a concentrations obtained from the lab experiment and YSI sensors. The experimental results showed a rapid decrease in PC and Chl-a concentrations between September and October. The YSI-sensor-measured PC concentration showed a similar trend to the experimental results between September and October. Correlations between the experimental and YSI-sensor-measured concentrations were very low for both

3.2.3. Cumulative pigment concentrations and Rrs band ratios Cumulative pigment concentrations at different depths were compared to the Rrs band ratio data collected by a portable sensor at different depths (Fig. 7 and Figs. S3–S8). Table 3 shows the correlation between cumulative PC concentrations and Rrs band ratio. The 0.5-m cumulative PC concentration showed the highest R2 performances of 0.8509, 0.9448, and 8315 for the surface, subsurface, and 1.0-m subsurface Rrs measurements, respectively. For the 0.5-m subsurface Rrs data, the highest R2 value of 0.8982 was found at the 1.0-m cumulative PC concentration. Overall, the subsurface ratio showed the best R2

Table 2 Statistical details of surface water measurement of pigment concentrations. Sampling

Measurement

Date 18/07/26–27 18/08/16–17 18/09/06 18/09/17 18/09/19 18/10/11 18/10/25

Phycocyanin (mg/m3) Mean

Experiment EXO-2 Experiment EXO-2 Experiment EXO-2 Experiment EXO-2 Experiment EXO-2 Experiment EXO-2 Experiment EXO-2

†

22.35 ± 20.04 0.99 ± 0.37 26.21 ± 16.56 1.32 ± 0.48 113.72 ± 228.43 3.18 ± 3.80 13.84 ± 8.09 3.12 ± 1.70 28.16 ± 10.69 5.67 ± 2.70 2.98 ± 1.44 0.63 ± 0.09 2.61 ± 1.89 0.44 ± 0.07

Chlorophyll-a (mg/m3) Max

Min

Mean

Max

Min

64.30 1.79 64.06 2.54 803.01 12.15 29.68 8.22 48.92 13.07 7.33 0.83 5.54 0.56

2.21 0.47 3.08 0.59 3.03 0.39 0.43 1.59 14.66 2.71 1.68 0.51 0.60 0.36

20.87 ± 10.65 9.60 ± 5.32 16.91 ± 5.76 13.78 ± 9.08 39.13 ± 31.99 14.73 ± 9.43 60.49 ± 30.35 5.06 ± 1.54 101.54 ± 99.06 8.26 ± 2.44 12.31 ± 2.96 8.41 ± 1.59 6.45 ± 2.18 4.27 ± 1.86

48.38 17.87 30.43 32.12 118.89 33.37 130.61 8.82 402.00 12.85 18.86 10.50 9.97 7.24

7.44 1.97 8.42 6.44 7.17 6.36 24.57 3.27 35.77 5.21 7.10 6.04 3.33 2.11

5

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 3. Temporal variation in PC and Chl-a concentrations measured via experiment(A, B and C), EXO-2 sensor (E, F and G) and scatter plots between two measurements (D and H). A and E are on the total dataset and B and F are the version with high concentration values removed.

comparison between the surface Rrs band ratio and 0.5-m cumulative Chl-a concentration. For the 1.0- to 1.5-m cumulative Chl-a concentrations, subsurface Rrs exhibited the highest and lowest R2 values of 0.0795 and 0.0365, respectively. Deeper cumulative Chl-a concentrations exhibited an R2 of under 0.01.

values among all Rrs measurement depths with cumulative PC concentrations. Fig. 8 and Figs. S6–S8 show comparisons between the surface to 1.0-m subsurface Rrs band ratio and cumulative Chl-a concentrations; the R2 values are shown in Table 4. Similar to the surfacesensor-measured concentrations (Fig. 6), Chl-a showed very low R2 values in the depth cumulation. The highest value was 0.1122 for the 6

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 4. Box plots for In-situ Rrs at 620 nm (A), 665 nm (B) and 709 nm (C) through surface to 1.0 m subsurface. Red line is median. Blue box represents range from 25th to 75th percentile. Dashed line is whisker of box and this represents max and minimum values. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 5. Comparison of experimentally and sensor-measured surface PC concentrations with surface Rrs band ratio (A and C) and subsurface Rrs band ratio (B and D). Each solid line represents linearly fitted curve. 7

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 6. Comparison of experimentally and sensor-measured surface Chl-a concentrations to surface Rrs band ratio (A and C) and subsurface Rrs band ratio (B and D). Each solid line represents linearly fitted curve.

3.3. Relationship between drone-based hyperspectral surface reflectance and pigment concentration

develop a PC distribution map. The linear regression-based empirical equations for the PC concentrations are as follows:

The raw and smoothed drone-base reflectance data at each sampling station are described in Fig. S2. Before the band-ratio bio-optical algorithm was applied, the drone reflectance spectra were compared with the in-situ Rrs at each monitoring station. The mean R2 between the insitu Rrs and drone reflectance was 0.85. Fig. 9 shows a comparison between the surface pigment concentrations and drone-based surfacereflectance band ratio. The R2 value of the YSI-sensor-measured PC (0.3436) was higher than the experimentally derived measurement (0.2570). The YSI-sensor-measured Chl-a concentration showed a negative relationship with the band ratio, resulting in an R2 value of 0.0067. For the experimental Chl-a, a relatively high R2 value of 0.3127 was found, whereas the sensor-measured PC showed a linear trend with reflectance. The cumulative concentrations of PC from 0.5 to 4.0 m were also compared to the drone-based surface-reflectance band ratio as shown in Fig. 10. The R2 values gradually decreased as the cumulation depth increased from 1.5 to 4.0 m.

(710) ⎞ R PC (mg m−3) = 5.8281 × ⎛ drone − 2.7126 ⎝ Rdrone (620) ⎠

(8)

(710) ⎞ R PC (mg m−2⋅1.0 m) = 4.882 × ⎛ drone − 2.8297 (620) R drone ⎝ ⎠

(9)

⎜

⎟

⎜

⎟

where, the PC concentrations are represented by the volume (m−3) and 1.0-m depth cumulation (m−2 ∙ 1.0 m). Rdrone (710) and Rdrone (620) are the drone-based surface reflectance at 709 and 620 nm, respectively. Figs. 11, S9, and S10 show the distribution maps of the surface and 1.0m cumulative PC concentrations in section C for September 6th, 17th, and 19th, respectively. As shown in Fig. 11, the surface PC concentration (A, B, and C) ranged approximately from 1.5 to 11 mg/m3 in all images, whereas the 1.0-m cumulative PC concentration (D, E, and F) ranged from 0.5 to 8 mg/m3. The PC distribution map developed using both surface and 1.0-m cumulative concentration, which were measured using a YSI sensor, showed a similar estimation pattern; however, the linear regression using surface PC concentration showed a higher slope than that of cumulative PC by approximately one. The high PC concentration regions were near the edge of the lake (B, C, E, and F in Fig. 11). A and D have regions with extremely high PC concentrations for both the surface and cumulative concentrations. A fence was installed in the region to prevent transport of HABs downstream. The fence may have caused an accumulation of phytoplankton and other particles near the surface, thereby resulting in high PC concentrations.

3.4. PC distribution map A PC distribution map based on the drone surface-reflectance data was developed using a linear-regression-based bio-optical algorithm with the YSI sensor-measured surface PC concentration and cumulative PC concentration. The 1.0-m cumulative PC concentration, which has the highest R2, was selected from eight different cumulative results to 8

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 7. Comparison of subsurface Rrs band ratio to cumulative PC concentrations. From A to F, each represent 0.5, 1.0, 1.5, 2.0, 2.5, and 3.0 m of cumulative depth, respectively. Each solid line represents linearly fitted curve.

were extracted from inside the algal cell using methanol extraction, whereas the YSI sensor measured pigments in a nondestructive manner by using the ability of the pigment to fluoresce. Past studies reported that in-situ fluorometers under-estimated the concentration of pigment compared to the experimental concentration measured by a spectrophotometric method (Gregor and Maršálek, 2004; Silva et al., 2016). Silva et al. (2016) documented the under-estimation factor, which ranged from 0.27 to 1.03 depending on the pigment extraction solvent. The Chl-a concentrations of the experiment and YSI were very poorly correlated below 0.1. Most Chl-a in cyanobacteria is located in photosystem I, i.e., weakly fluorescent. Only 10%–20% of Chl-a is located in photosystem II and the fluorescence probe excites this region and measures Chl-a (Beutler et al., 2003; Zamyadi et al., 2012). This is also one of the reasons that the YSI sensor-measured pigment concentration is lower than that measured in the experiment. The difference in the volume of the water sample of the experiment and EXO-2 sensing may cause this discrepancy. For example, in this study, water samples of 5 L and 500 ml were collected for the experimental measurement of the PC and Chl-a concentrations, respectively, whereas only 0.3 ml was scanned by the in-situ YSI sensor for measurement. In addition, Algal scum, which floats on the water surface, was often observed during field monitoring. Some scums may have been included in the water sample, resulting in high pigment concentration in the experiment. In contrast, the EXO-2 sensor measured only a small volume of water and the variability in the concentration may have been relatively high

Table 3 R-squared values in comparison to cumulative PC concentrations and band ratios at different depths. Cumulative Depth

0.5 m 1.0 m 1.5 m 2.0 m 2.5 m 3.0 m

Spectral measurement depth Surface

Subsurface

0.5 m subsurface

1 m subsurface

0.8509 0.8143 0.7908 0.7754 0.7702 0.7623

0.9448 0.9430 0.9311 0.9160 0.8955 0.8723

0.8864 0.8982 0.8901 0.8746 0.8607 0.8451

0.8315 0.8198 0.7993 0.7745 0.7456 0.7135

The highlighted areas in the black circles show the algal accumulation regions that were recognized as vegetation. 4. Discussion 4.1. Experimentally and YSI-sensor-measured pigment concentrations There are distinct discrepancies between the experimental and YSIsensor-measured pigment concentrations. The sensor-measured pigment concentration is substantially lower than the experimental pigment concentration. One reason for the discrepancy is the difference in measuring the pigments. In the laboratory experiment, the pigments 9

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 8. Comparison of subsurface Rrs band ratio to cumulative Chl-a concentrations. From A to F, each represent 0.5, 1.0, 1.5, 2.0, 2.5, and 3.0 m of cumulative depth, respectively. Each solid line represents linearly fitted curve.

contents of each cell, Chl-a concentration, and turbidity (Chang et al., 2012; Zamyadi et al., 2012). In Kong et al. (2014), a change in the taxonomical populations of cyanobacteria in the microalgal communities had a negative effect on the accuracy of the bio-optical algorithm.

Table 4 R-squared values in comparison to cumulative Chl-a concentrations and band ratios at different depths. Cumulative Depth

0.5 m 1.0 m 1.5 m 2.0 m 2.5 m 3.0 m

Spectral measurement depth Surface

Subsurface

0.5 m subsurface

1 m subsurface

0.1122 0.0602 0.0246 0.0112 0.0058 0.0020

0.1110 0.0795 0.0365 0.0172 0.0080 0.0026

0.0481 0.0672 0.0361 0.0187 0.0094 0.0045

0.0528 0.0629 0.0344 0.0190 0.0104 0.0055

4.2. In-situ Rrs and drone-based surface reflectance for pigment remote sensing A bio-optical algorithm using the in-situ Rrs data collected from the portable sensor showed a different estimation performance according to the pigment and measuring technique. In the case of PC estimation using sensor-measured concentration, it showed a strong correlation between the band ratio and pigment concentrations, whereas the Chl-a estimation showed a poor correlation. However, this is in contrast to the comparison of the experimental pigment concentrations and biooptical algorithm. The PC concentration obtained experimentally showed a relatively low estimation performance compared to that of the sensor-measured PC concentration. In contrast, the experimental Chl-a concentration had a higher R2 performance than that of the sensor-measured Chl-a. Duan et al. (2012) divided the PC concentration into two groups, less than and greater than 2 mg/m3. This resulted in a higher R2 value (0.7) for high PC concentration than that for the low PC concentration (0.62). In addition, they reported a very poor performance (R2 of 0.18) for the band ratio of an undivided PC concentration group. This implies that the bio-optical algorithm needs to consider

during measurement in the inhomogeneous water column. With regards to the statistical analysis, previous studies have showed R2 values greater than 0.6 in the linear regression between the YSI-sensor-measured PC and algal biomass during laboratory experiments (Bastien et al., 2011; Choo et al., 2018; Hodges et al., 2018). However, the field application study of the sensor showed a relatively low relationship between the total cyanobacterial biomass and the sensor-measured PC concentrations with an R2 value of 0.3 (Hodges et al., 2018). Bowling et al. (2016) reported R2 values ranging from 0.24 to 0.54 between experimentally and sensor-measured Chl-a. There are various potential causes of the disturbance of accurate measurement by the YSI EXO-2 sensor: colony size of the cyanobacteria, phycocyanin 10

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 9. Comparison of drone-reflectance band ratio to experimentally and sensor-measured surface PC (A, C) and Chl-a (B, D) concentrations. Each solid line represents linearly fitted curve.

radiation to detect suspended sediment in surface water. The result from this study showed a different correlation coefficient to that of the Landsat data according to the sun radiation angle. Unlike surface reflectance, the geometry between the sun and spectrophotometer is highly important in measuring in-situ Rrs. Mobley (1999) examined the influences of wind speed, sun angle, and viewing angle on Rrs measurement. For a near-40° viewing angle for the sensor, the sun angle showed a minimal effect on Rrs measurement. A sun zenith angle between 30° and 80° also had little effect on the measurement.

various water quality conditions. Subsurface and 0.5-m subsurface Rrs band ratios exhibited better R2 performances than that of the surface Rrs in the cumulative PC estimation. Previous works calculated the subsurface reflectance from the above-surface reflectance (Gons, 1999; Randolph et al., 2008; Rijkeboer et al., 1997; Simis et al., 2005). These early works showed reasonable R2 performances of between 0.7 and 0.9. Li et al. (2015) showed that surface and subsurface reflectance could be measured using separate above-surface submerged spectral sensors; the PC estimation performance had an approximate R2 of 0.7. In this study, the band ratio obtained from drone-based surface reflectance was directly introduced in the Rrs band-ratio algorithm. Both surface and cumulative pigment estimation using the drone-reflectance band ratio showed relatively low R2 performances. This low performance resulted from a discrepancy between the surface reflectance of the drone and the Rrs of in-situ monitoring. The surface reflectance is more dependent on incident light, sky condition, and wind speed than remote sensing reflectance; however, the spectral trend and features of the surface reflectance are similar to those of remote-sensing reflectance. This means that the ratio of the surface reflectance in estimating the pigment concentration could compensate for the magnitude difference of the surface reflectance spectra. From the early days to recent remote sensing studies, the surface-reflectance ratio has been utilized to estimate algal biomass (Gokul and Shanmugam, 2016; Gordon et al., 1983; Morel and Prieur, 1977). Previous research has reported that surface reflectance varied according to the sun angle (Gordon, 1989; Morel and Gentili, 1991; Morel and Maritorena, 2001; Oxford, 1976). Oxford (1976) measured the incident and reflected solar

4.3. Surface pigment and depth-cumulative pigment concentrations Previous studies have emphasized the necessity of monitoring the vertical distribution of cyanobacteria because of the rapid vertical migration of cyanobacteria and its effect on surface pigment detection (Binding et al., 2010; Hunter et al., 2008). Hunter et al. (2008) documented that the diel vertical migration of cyanobacteria can be detected through sequential airborne remote sensing of pigment concentration. However, the vertical monitoring data was not used to estimate pigment concentration. Binding et al. (2010) concluded that the vertical distribution of algae is required in algorithm development and validation, using vertical in-situ sampling. In addition, a recent study on algal vertical distribution demonstrated that remote sensing of the vertical distribution of algae using satellites is an unresolved challenge (Soontiens et al., 2019). This study continuously monitored the vertical profile of algal pigments in a deep lake using a YSI sensor. The vertical structure of PC was considered by using the vertical cumulative PC concentration. 11

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 10. Comparison of drone-reflectance band ratio and cumulative PC concentrations. From A to H, each represent 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, and 4.0 m of cumulative depth, respectively. Each solid line represents linearly fitted curve.

PC concentrations showed R2 values of greater than 0.8. The highest R2 value was obtained from each regression between the in-situ Rrs band ratio at all depths and the 0.5-m cumulative PC concentration, except for the case of the 0.5-m subsurface measurement which had the highest R2 value from the regression between the 0.5-m subsurface Rrs band ratio and the 1.0-m cumulative pigment concentration. However, regressions between the cumulative Chl-a concentrations and the band ratios from both in-situ Rrs and drone-based reflectance had very low

Several depth-cumulative pigment concentrations were regressed using in-situ and drone-reflectance band ratios. Compared to surface PC concentrations, the cumulative PC concentration showed a significantly high linear correlation with band ratios in both in-situ Rrs and dronebased reflectance measurements. The regression between the in-situ subsurface Rrs band ratio and 0.5-m cumulative PC concentration had a relatively high R2 value of greater than 0.9 (Table 3). Other regressions between the in-situ Rrs measurements and 0.5-m and 1.0-m cumulative 12

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Fig. 11. PC distribution map of section C on September 6th. A, B, and C are from the linear regression results using experimental surface PC. D, E, and F are linear regression results using YSI-sensor-measured 1.0-m cumulative PC.

R2 values of less than 0.12. Overall, the bio-optical algorithms using drone-based surface-irradiance reflectance performed slightly worse than those using in-situ Rrs. For the effective depth of cumulation, the drone-based reflectance band ratio was well-matched with the 1.0-m cumulative PC concentration in the bio-optical algorithm. This combination performed the best with the highest R2 value; a combination of the band ratio and 1.5m cumulative PC concentration performed slightly worse than the best combination. These results imply that the cumulation of pigment concentration in the depth direction can improve the performance of a biooptical algorithm in remotely estimating pigment concentrations in inland waters. Based on the map-based spatial and temporal distributions of PC concentration (Figs. 10, S7, and S8), the two bio-optical algorithms had

different concentration scales but similar distribution patterns for the sensor-measured surface and cumulative PC concentrations. Similar patterns in the spatial distribution and the relatively low concentration for vertical cumulative concentration can be explained by the vertical distribution of PC (Fig. S11). The highest PC concentration may be mainly near the surface (less than 1.0 m), and its concentration may gradually decrease as the depth increases. This implies that the PC distribution map developed using cumulative concentration followed the one using surface PC concentration. In the case of the 1.0-m cumulative PC, the concentration gradient was larger than that of the surface concentration, which signifies that the spatial occurrence of cyanobacteria can be readily detected. This study could be a novel application of hyperspectral sensing and the YSI device because synoptic and rapid monitoring are required to precisely detect harmful 13

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

cyanobacterial bloom. Song et al. (2013) concluded that in-situ monitoring using portable devices should be integrated with a remote sensing technique for a detailed interpretation of spatiotemporal cyanobacteria distribution.

Ansper, A., Alikas, K., 2019. Retrieval of chlorophyll a from sentinel-2 MSI data for the European Union water framework directive reporting purposes. Remote Sens. 11, 64. APHA (American Public Health Association), 2001. Standard Methods for the Examination of Water and Waste Water, twenty-first ed. APHA-AWWA-WPCF, Washington, DC. Bastien, C., Cardin, R., Veilleux, É., Deblois, C., Warren, A., Laurion, I., 2011. Performance evaluation of phycocyanin probes for the monitoring of cyanobacteria. J. Environ. Monit. 13, 110–118. Bennett, A., Bogorad, L., 1973. Complementary chromatic adaptation in a filamentous blue-green alga. J. Cell Biol. 58, 419–435. Beth, R.T., 2016. UAV based hyperspectral imaging of river ecosystems. PhD Dissertation In: School of Spatial Information Management. Carinthia University of Applied Sciences. Beutler, M., Wiltshire, K.H., Arp, M., Kruse, J., Reineke, C., Moldaenke, C., Hansen, U.-P., 2003. A reduced model of the fluorescence from the cyanobacterial photosynthetic apparatus designed for the in situ detection of cyanobacteria. Biochim. Biophys. Acta Bioenerg. 1604, 33–46. Bhatti, A.M., Rundquist, D., Schalles, J., Ramirez, L., Nasu, S., 2009. A comparison between above-water surface and subsurface spectral reflectances collected over inland waters. Geocarto Int. 24, 133–141. Binding, C.E., Greenberg, T., Jerome, J., Bukata, R., Letourneau, G., 2010. An assessment of MERIS algal products during an intense bloom in Lake of the Woods. J. Plankton Res. 33, 793–806. Blix, K., Pálffy, K., R Tóth, V., Eltoft, T., 2018. Remote sensing of water quality parameters over Lake Balaton by using sentinel-3 OLCI. Water 10, 1428. Bowling, L.C., Zamyadi, A., Henderson, R.K., 2016. Assessment of in situ fluorometry to measure cyanobacterial presence in water bodies with diverse cyanobacterial populations. Water Res. 105, 22–33. Brando, V.E., Dekker, A.G., 2003. Satellite hyperspectral remote sensing for estimating estuarine and coastal water quality. IEEE Trans. Geosci. Remote Sens. 41, 1378–1387. Brezonik, P., Menken, K.D., Bauer, M., 2005. Landsat-based remote sensing of lake water quality characteristics, including chlorophyll and colored dissolved organic matter (CDOM). Lake Reservoir Manag. 21, 373–382. Carmichael, W.W., Boyer, G.L., 2016. Health impacts from cyanobacteria harmful algae blooms: implications for the North American Great Lakes. Harmful Algae 54, 194–212. Chang, D.-W., Hobson, P., Burch, M., Lin, T.-F., 2012. Measurement of cyanobacteria using in-vivo fluoroscopy–Effect of cyanobacterial species, pigments, and colonies. Water Res. 46, 5037–5048. Chen, J., Jönsson, P., Tamura, M., Gu, Z., Matsushita, B., Eklundh, L., 2004. A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky–Golay filter. Remote Sens. Environ. 91, 332–344. Choo, F., Zamyadi, A., Newton, K., Newcombe, G., Bowling, L., Stuetz, R., Henderson, R.K., 2018. Performance evaluation of in situ fluorometers for real-time cyanobacterial monitoring. H2Open J. 1, 26–46. Dekker, A.G., 1993. Detection of Optical Water Quality Parameters for Eutrophic Waters by High Resolution Remote Sensing. Dekker, A., Malthus, T., Wijnen, M., Seyhan, E., 1992. The effect of spectral bandwidth and positioning on the spectral signature analysis of inland waters. Remote Sens. Environ. 41, 211–225. Duan, H., Ma, R., Xu, J., Zhang, Y., Zhang, B., 2010. Comparison of different semi-empirical algorithms to estimate chlorophyll-a concentration in inland lake water. Environ. Monit. Assess. 170, 231–244. Duan, H., Ma, R., Hu, C., 2012. Evaluation of remote sensing algorithms for cyanobacterial pigment retrievals during spring bloom formation in several lakes of East China. Remote Sens. Environ. 126, 126–135. Gilerson, A.A., Gitelson, A.A., Zhou, J., Gurlin, D., Moses, W., Ioannou, I., Ahmed, S.A., 2010. Algorithms for remote estimation of chlorophyll-a in coastal and inland waters using red and near infrared bands. Opt. Express 18, 24109–24125. Gitelson, A.A., Schalles, J.F., Hladik, C.M., 2007. Remote chlorophyll-a retrieval in turbid, productive estuaries: chesapeake Bay case study. Remote Sens. Environ. 109, 464–472. Gitelson, A.A., Dall'Olmo, G., Moses, W., Rundquist, D.C., Barrow, T., Fisher, T.R., Gurlin, D., Holz, J., 2008. A simple semi-analytical model for remote estimation of chlorophyll-a in turbid waters: Validation. Remote Sens. Environ. 112, 3582–3593. Glazer, A.N., 1989. Light guides. Directional energy transfer in a photosynthetic antenna. J. Biol. Chem. 264, 1–4. Gokaraju, B., Durbha, S.S., King, R.L., Younan, N.H., 2011. A machine learning based spatio-temporal data mining approach for detection of harmful algal blooms in the Gulf of Mexico. IEEE J. Sel. Top. Appl. Earth obs. Remote sens. 4, 710–720. Gokul, E.A., Shanmugam, P., 2016. An optical system for detecting and describing major algal blooms in coastal and oceanic waters around India. J. Geophys. Res.: Oceans 121, 4097–4127. Gons, H.J., 1999. Optical teledetection of chlorophyll a in turbid inland waters. Environ. Sci. Technol. 33, 1127–1132. Gordon, H.R., 1989. Dependence of the diffuse reflectance of natural waters on the sun angle. Limnol. Oceanogr. 34, 1484–1489. Gordon, H., Morel, A., Barber, R., Mooers, N., Bowman, M., Zeitzschel, B., 1983. A Review Lecture Notes on Coastal and Estuarine Studies. Remote Assessment of Ocean Color for Interpretation of Satellite Visible Imagery. Springer-Verlag NY. Gregor, J., Maršálek, B., 2004. Freshwater phytoplankton quantification by chlorophyll a: a comparative study of in vitro, in vivo and in situ methods. Water Res. 38, 517–522. Ha, K., Kim, H.-W., Jeong, K.-S., Joo, G.-J., 2000. Vertical distribution of Microcystis population in the regulated Nakdong river, Korea. Limnology 1, 225–230. Hilborn, E., Beasley, V., 2015. One health and cyanobacteria in freshwater systems:

5. Conclusions The main purposes of this study were to apply subsurface Rrs measurements and vertical cumulative pigments to bio-optical algorithms for the remote sensing of algal pigments, and to develop a PC distribution map using drone-based hyperspectral imagery. The major findings are as follows: (1) Both the experimental Chl-a and PC concentrations were higher than those measured by the sensor. Uncertainties in measurement conditions such as colony size, pigment content of each cell, turbidity, and taxonomical structure of the algal community may have caused this discrepancy. (2) The application of vertical cumulative PC concentration contributed to improvement in the overall performance of the bio-optical algorithm compared to the application of surface PC concentrations (experimental and sensor-measured). Unlike the PC concentration, the experimental Chl-a concentration showed a relatively weak correlation with the bio-optical algorithm, and there was no correlation between the sensor-measured PC concentration and the algorithms. (3) The drone-based surface reflectance had an R2 of 0.85 for in-situ Rrs, and its band ratio showed a slightly weak correlation with the cumulative pigment concentrations compared to the correlation between the in-situ surface Rrs band ratio and the cumulative pigment concentrations. The vertical cumulative concentration improved the bio-optical algorithm performance; the developed PC distribution map accurately describes the spatial and temporal distributions of the cumulative PC concentration. This study was able to demonstrate that the vertical cumulation concentration can improve the performance of bio-optical algorithms in remote sensing of cyanobacterial blooms in a lake. An option for future studies to improve the remote-sensing algorithm is the identification of the relationship between cumulation depth and light attenuation to estimate the vertical profile of the water-quality parameter. Acknowledgement This work was supported by the ICT R&D program of MSIT/IITP (2018-0-00219, Space-time complex artificial intelligence blue-green algae prediction technology based on direct-readable water quality complex sensor and hyperspectral image). This was also partially supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2019R1C1C1011366). Appendix A. Supplementary data Supplementary data to this article can be found online at https:// doi.org/10.1016/j.rse.2019.111517. References Adão, T., Hruška, J., Pádua, L., Bessa, J., Peres, E., Morais, R., Sousa, J.J., 2017. Hyperspectral imaging: a review on UAV-based sensors, data processing and applications for agriculture and forestry. Remote Sens. 9, 1110. Agrawal, N., Venugopalan, K., 2011. Speckle reduction in remote sensing images. In: Emerging Trends in Networks and Computer Communications (ETNCC), 2011 International Conference on. IEEE, pp. 195–199. Ahn, Y.-H., Shanmugam, P., Ryu, J.-H., Jeong, J.-C., 2006. Satellite detection of harmful algal bloom occurrences in Korean waters. Harmful Algae 5, 213–231.

14

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Mishra, S., Mishra, D.R., Schluchter, W.M., 2009. A novel algorithm for predicting phycocyanin concentrations in cyanobacteria: a proximal hyperspectral remote sensing approach. Remote Sens. 1, 758–775. Mittenzwey, K.H., Ullrich, S., Gitelson, A., Kondratiev, K., 1992. Determination of chlorophyll a of inland waters on the basis of spectral reflectance. Limnol. Oceanogr. 37, 147–149. Mobley, C.D., 1999. Estimation of the remote-sensing reflectance from above-surface measurements. Appl. Opt. 38, 7442–7455. Mobley, C.D., Sundman, L.K., 2008. Hydrolight 5 Ecolight 5. Sequoia Scientific Inc. Mollaee, S., 2018. Estimation of Phytoplankton Chlorophyll-A Concentration in the Western Basin of Lake Erie Using Sentinel-2 and Sentinel-3 Data. University of Waterloo. Morel, A., Gentili, B., 1991. Diffuse reflectance of oceanic waters: its dependence on Sun angle as influenced by the molecular scattering contribution. Appl. Opt. 30, 4427–4438. Morel, A., Maritorena, S., 2001. Bio‐optical properties of oceanic waters: a reappraisal. J. Geophys. Res.: Oceans 106, 7163–7180. Morel, A., Prieur, L., 1977. Analysis of variations in ocean color 1. Limnol. Oceanogr. 22, 709–722. Murphy, R., Underwood, A., Tolhurst, T., Chapman, M., 2008. Field-based remote-sensing for experimental intertidal ecology: case studies using hyperspatial and hyperspectral data for New South Wales (Australia). Remote Sens. Environ. 112, 3353–3365. Nadeau, T.-L., Howard-Williams, C., Castenholz, R.W., 1999. Effects of solar UV and visible irradiance on photosynthesis and vertical migration of Oscillatoria sp. (Cyanobacteria) in an Antarctic microbial mat. Aquat. Microb. Ecol. 20, 231–243. Odermatt, D., Pomati, F., Pitarch, J., Carpenter, J., Kawka, M., Schaepman, M., Wüest, A., 2012. MERIS observations of phytoplankton blooms in a stratified eutrophic lake. Remote Sens. Environ. 126, 232–239. Oliver, R.L., 1994. Floating and sinking in gas‐vacuolate cyanobacteria1. J. Phycol. 30, 161–173. Olmanson, L.G., Brezonik, P.L., Bauer, M.E., 2013. Airborne hyperspectral remote sensing to assess spatial distribution of water quality characteristics in large rivers: the Mississippi River and its tributaries in Minnesota. Remote Sens. Environ. 130, 254–265. Oxford, M., 1976. Remote sensing of suspended sediments in surface waters. Photogramm. Eng. Remote Sens. 42, 1539–1545. Palmer, S.C., Kutser, T., Hunter, P.D., 2015a. Remote Sensing of Inland Waters: Challenges, Progress and Future Directions. Elsevier. Palmer, S.C., Odermatt, D., Hunter, P., Brockmann, C., Presing, M., Balzter, H., Tóth, V., 2015b. Satellite remote sensing of phytoplankton phenology in Lake Balaton using 10 years of MERIS observations. Remote Sens. Environ. 158, 441–452. Park, Y., Pyo, J., Kwon, Y.S., Cha, Y., Lee, H., Kang, T., Cho, K.H., 2017. Evaluating physico-chemical influences on cyanobacterial blooms using hyperspectral images in inland water, Korea. Water Res. 126, 319–328. Pitarch, J., Odermatt, D., Kawka, M., Wüest, A., 2014. Retrieval of vertical particle concentration profiles by optical remote sensing: a model study. Opt. Express 22, A947–A959. Power, M.E., Brozović, N., Bode, C., Zilberman, D., 2005. Spatially explicit tools for understanding and sustaining inland water ecosystems. Front. Ecol. Environ. 3, 47–55. Proctor, C.W., Roesler, C.S., 2010. New insights on obtaining phytoplankton concentration and composition from in situ multispectral Chlorophyll fluorescence. Limnol Oceanogr. Methods 8, 695–708. Pulliainen, J., Kallio, K., Eloheimo, K., Koponen, S., Servomaa, H., Hannonen, T., Tauriainen, S., Hallikainen, M., 2001. A semi-operative approach to lake water quality retrieval from remote sensing data. Sci. Total Environ. 268, 79–93. Pyo, J., Ha, S., Pachepsky, Y.A., Lee, H., Ha, R., Nam, G., Kim, M.S., Im, J., Cho, K.H., 2016. Chlorophyll-a concentration estimation using three difference bio-optical algorithms, including a correction for the low-concentration range: the case of the Yiam reservoir, Korea. Remote Sens. Lett. 7, 407–416. Pyo, J., Pachepsky, Y., Baek, S.-S., Kwon, Y., Kim, M., Lee, H., Park, S., Cha, Y., Ha, R., Nam, G., 2017. Optimizing semi-analytical algorithms for estimating chlorophyll-a and phycocyanin concentrations in inland waters in Korea. Remote Sens. 9, 542. Pyo, J., Ligaray, M., Kwon, Y., Ahn, M.-H., Kim, K., Lee, H., Kang, T., Cho, S., Park, Y., Cho, K., 2018. High-spatial resolution monitoring of phycocyanin and chlorophyll-a using airborne hyperspectral imagery. Remote Sens. 10, 1180. Randolph, K., Wilson, J., Tedesco, L., Li, L., Pascual, D.L., Soyeux, E., 2008. Hyperspectral remote sensing of cyanobacteria in turbid productive water using optically active pigments, chlorophyll a and phycocyanin. Remote Sens. Environ. 112, 4009–4019. Richardson, L.L., Castenholz, R.W., 1987. Diel vertical movements of the cyanobacterium Oscillatoria terebriformis in a sulfide-rich hot spring microbial mat. Appl. Environ. Microbiol. 53, 2142–2150. Rijkeboer, M., Dekker, A.G., Gons, H.J., 1997. Subsurface irradiance reflectance spectra of inland waters differing in morphometry and hydrology. Aquat. Ecol. 31, 313–323. Ritchie, J.C., Zimba, P.V., Everitt, J.H., 2003. Remote sensing techniques to assess water quality. Photogramm. Eng. Remote Sens. 69, 695–704. Roessner, S., Segl, K., Heiden, U., Kaufmann, H., 2001. Automated differentiation of urban surfaces based on airborne hyperspectral imagery. IEEE Trans. Geosci. Remote Sens. 39, 1525–1532. Sarada, R., Pillai, M.G., Ravishankar, G., 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 Biochem. 34, 795–801. Selige, T., Böhner, J., Schmidhalter, U., 2006. High resolution topsoil mapping using hyperspectral image and field data in multivariate regression modeling procedures. Geoderma 136, 235–244. Shim, M., Yoon, J., Lee, S., 2015. Water quality properties of tributaries of Daechung lake, Korea. Korean J. Environ. Ecol. 48, 12–25.

animal illnesses and deaths are sentinel events for human health risks. Toxins 7, 1374–1395. Hill, S.L., Clemens, P., 2015. Miniaturization of high spectral spatial resolution hyperspectral imagers on unmanned aerial systems. In: Next-Generation Spectroscopic Technologies, vol.VIII. pp. 94821E International Society for Optics and Photonics. Hilton, J., 1984. Airborne remote sensing for freshwater and estuarine monitoring. Water Res. 18, 1195–1223. Hodges, C.M., Wood, S.A., Puddick, J., McBride, C.G., Hamilton, D.P., 2018. Sensor manufacturer, temperature, and cyanobacteria morphology affect phycocyanin fluorescence measurements. Environ. Sci. Pollut. Control Ser. 25, 1079–1088. Huisman, J., Codd, G.A., Paerl, H.W., Ibelings, B.W., Verspagen, J.M., Visser, P.M., 2018. Cyanobacterial blooms. Nat. Rev. Microbiol. 16, 471. Hunter, P., Tyler, A., Willby, N., Gilvear, D., 2008. The spatial dynamics of vertical migration by Microcystis aeruginosa in a eutrophic shallow lake: a case study using high spatial resolution time‐series airborne remote sensing. Limnol. Oceanogr. 53, 2391–2406. Hunter, P.D., Tyler, A.N., Carvalho, L., Codd, G.A., Maberly, S.C., 2010. Hyperspectral remote sensing of cyanobacterial pigments as indicators for cell populations and toxins in eutrophic lakes. Remote Sens. Environ. 114, 2705–2718. Jin, C., Mesquita, M.M., Deglint, J.L., Emelko, M.B., Wong, A., 2018. Quantification of cyanobacterial cells via a novel imaging-driven technique with an integrated fluorescence signature. Sci. Rep. 8, 9055. Jupp, D.L., Kirk, J.T., Harris, G.P., 1994. Detection, identification and mapping of cyanobacteria—using remote sensing to measure the optical quality of turbid inland waters. Mar. Freshw. Res. 45, 801–828. Kallio, K., Kutser, T., Hannonen, T., Koponen, S., Pulliainen, J., Vepsäläinen, J., Pyhälahti, T., 2001. Retrieval of water quality from airborne imaging spectrometry of various lake types in different seasons. Sci. Total Environ. 268, 59–77. Kay, S., Hedley, J., Lavender, S., 2009. Sun glint correction of high and low spatial resolution images of aquatic scenes: a review of methods for visible and near-infrared wavelengths. Remote Sens. 1, 697–730. Kerekes, J.P., Schott, J.R., 2007. Hyperspectral imaging systems. In: Chang, C.-I. (Ed.), Hyperspectral Data Exploitation: Theory and Applications. John Wiley & Sons,Inc, Hoboken, NJ, pp. 19–45 ch. 12. Kiefer, I., Odermatt, D., Anneville, O., Wüest, A., Bouffard, D., 2015. Application of remote sensing for the optimization of in-situ sampling for monitoring of phytoplankton abundance in a large lake. Sci. Total Environ. 527, 493–506. Kislik, C., Dronova, I., Kelly, M., 2018. UAVs in support of algal bloom research: a review of current applications and future opportunities. Drones 2, 35. Kong, Y., Lou, I., Zhang, Y., Lou, C.U., Mok, K.M., 2014. Using an online phycocyanin fluorescence probe for rapid monitoring of cyanobacteria in Macau freshwater reservoir. Hydrobiologia 741, 33–49. Kromkamp, J.C., Mur, L.R., 1984. Buoyant density changes in the cyanobacterium Microcystis aeruginosa due to changes in the cellular carbohydrate content. FEMS Microbiol. Lett. 25, 105–109. Kudela, R.M., Palacios, S.L., Austerberry, D.C., Accorsi, E.K., Guild, L.S., Torres-Perez, J., 2015. Application of hyperspectral remote sensing to cyanobacterial blooms in inland waters. Remote Sens. Environ. 167, 196–205. Kutser, T., 2004. Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnol. Oceanogr. 49, 2179–2189. Kutser, T., Metsamaa, L., Strömbeck, N., Vahtmäe, E., 2006. Monitoring cyanobacterial blooms by satellite remote sensing. Estuar. Coast Shelf Sci. 67, 303–312. Kutser, T., Metsamaa, L., Dekker, A.G., 2008. Influence of the vertical distribution of cyanobacteria in the water column on the remote sensing signal. Estuar. Coast Shelf Sci. 78, 649–654. Kutser, T., Vahtmäe, E., Praks, J., 2009. A sun glint correction method for hyperspectral imagery containing areas with non-negligible water leaving NIR signal. Remote Sens. Environ. 113, 2267–2274. Kutser, T., Vahtmäe, E., Paavel, B., Kauer, T., 2013. Removing glint effects from field radiometry data measured in optically complex coastal and inland waters. Remote Sens. Environ. 133, 85–89. Kwon, Y., Baek, S., Lim, Y., Pyo, J., Ligaray, M., Park, Y., Cho, K., 2018. Monitoring coastal chlorophyll-a concentrations in coastal areas using machine learning models. Water 10, 1020. Lee, J.J., Yang, S.Y., Yoon, J.H., Lee, J.B., Tae, E.A., Lee, S.H., 2014. Occurrence of Cyanobacterial Toxins in the Daechung Reservoir. Korean Society af Water Quality, pp. 680–681 2014. Li, L., Sengpiel, R.E., Pascual, D.L., Tedesco, L.P., Wilson, J.S., Soyeux, E., 2010. Using hyperspectral remote sensing to estimate chlorophyll‐a and phycocyanin in a mesotrophic reservoir. Int. J. Remote Sens. 31, 4147–4162. Li, L., Li, L., Song, K., Li, Y., Tedesco, L.P., Shi, K., & Li, Z. (2013).t, 135, 150-166. Li, L., Li, L., Song, K., 2015. Remote sensing of freshwater cyanobacteria: an extended IOP Inversion Model of Inland Waters (IIMIW) for partitioning absorption coefficient and estimating phycocyanin. Remote Sens. Environ. 157, 9–23. Liu, G., Simis, S.G., Li, L., Wang, Q., Li, Y., Song, K., Lyu, H., Zheng, Z., Shi, K., 2017. A four-band semi-analytical model for estimating phycocyanin in inland waters from simulated MERIS and OLCI data. IEEE Trans. Geosci. Remote Sens. 56, 1374–1385. Lim, J.S., 1990. Two-dimensional Signal and Image Processing. Englewood Cliffs, NJ, Prentice Hall, 1990, 710 p.. pp. 548 equations 549.544–549.546. Matsushita, B., Yang, W., Yu, G., Oyama, Y., Yoshimura, K., Fukushima, T., 2015. A hybrid algorithm for estimating the chlorophyll-a concentration across different trophic states in Asian inland waters. ISPRS J. Photogrammetry Remote Sens. 102, 28–37. Matthews, M.W., Bernard, S., Winter, K., 2010. Remote sensing of cyanobacteria-dominant algal blooms and water quality parameters in Zeekoevlei, a small hypertrophic lake, using MERIS. Remote Sens. Environ. 114, 2070–2087.

15

Remote Sensing of Environment 236 (2020) 111517

Y.S. Kwon, et al.

Remote Sens. Environ. 171, 58–74. Vincent, R.K., Qin, X., McKay, R.M.L., Miner, J., Czajkowski, K., Savino, J., Bridgeman, T., 2004. Phycocyanin detection from LANDSAT TM data for mapping cyanobacterial blooms in Lake Erie. Remote Sens. Environ. 89, 381–392. Visser, P.M., Passarge, J., Mur, L.R., 1997. Modelling vertical migration of the cyanobacterium Microcystis. Hydrobiologia 349, 99–109. Wallace, B.B., Bailey, M.C., Hamilton, D.P., 2000. Simulation of vertical position of buoyancy regulating Microcystis aeruginosa in a shallow eutrophic lake. Aquat. Sci. 62, 320–333. Wang, G., Lee, Z., Mishra, D.R., Ma, R., 2016. Retrieving absorption coefficients of multiple phytoplankton pigments from hyperspectral remote sensing reflectance measured over cyanobacteria bloom waters. Limnol Oceanogr. Methods 14, 432–447. Wu, X., Kong, F., 2009. Effects of light and wind speed on the vertical distribution of Microcystis aeruginosa colonies of different sizes during a summer bloom. Int. Rev. Hydrobiol. 94, 258–266. Wynne, T.T., Stumpf, R.P., Tomlinson, M.C., Dyble, J., 2010. Characterizing a cyanobacterial bloom in western Lake Erie using satellite imagery and meteorological data. Limnol. Oceanogr. 55, 2025–2036. Xiu, P., Liu, Y., Tang, J., 2008. Variations of ocean colour parameters with nonuniform vertical profiles of chlorophyll concentration. Int. J. Remote Sens. 29, 831–849. Xue, K., Zhang, Y., Duan, H., Ma, R., Loiselle, S., Zhang, M., 2015. A remote sensing approach to estimate vertical profile classes of phytoplankton in a eutrophic lake. Remote Sens. 7, 14403–14427. Xue, K., Zhang, Y., Ma, R., Duan, H., 2017. An approach to correct the effects of phytoplankton vertical nonuniform distribution on remote sensing reflectance of cyanobacterial bloom waters. Limnol Oceanogr. Methods 15, 302–319. Yacobi, Y.Z., Moses, W.J., Kaganovsky, S., Sulimani, B., Leavitt, B.C., Gitelson, A.A., 2011. NIR-red reflectance-based algorithms for chlorophyll-a estimation in mesotrophic inland and coastal waters: lake Kinneret case study. Water Res. 45, 2428–2436. Zamyadi, A., McQuaid, N., Dorner, S., Bird, D.F., Burch, M., Baker, P., Hobson, P., PRÉvost, M., 2012. Cyanobacterial detection using in vivo fluorescence probes: managing interferences for improved decision‐making. J. Am. Water Work. Assoc. 104, E466–E479. Zeng, C., Richardson, M., King, D.J., 2017. The impacts of environmental variables on water reflectance measured using a lightweight unmanned aerial vehicle (UAV)based spectrometer system. ISPRS J. Photogrammetry Remote Sens. 130, 217–230. Zhao, J., Temimi, M., Ghedira, H., 2017. Remotely sensed sea surface salinity in the hyper-saline Arabian Gulf: application to landsat 8 OLI data. Estuar. Coast Shelf Sci. 187, 168–177.

Shin, J.K., Cho, K.J., Oh, I.H., 1999. Dynamics of water environmental factors and phytoplankton in taechong reservoir. Korean J. Environ. Ecol. 17, 529–541. Shippert, P., 2003. Introduction to hyperspectral image analysis. Online J. Space Commun. 3. Silva, T., Giani, A., Figueredo, C., Viana, P., Khac, V.T., Lemaire, B.J., Tassin, B., Nascimento, N., Vinçon-Leite, B., 2016. Comparison of cyanobacteria monitoring methods in a tropical reservoir by in vivo and in situ spectrofluorometry. Ecol. Eng. 97, 79–87. Simis, S.G., Peters, S.W., Gons, H.J., 2005. Remote sensing of the cyanobacterial pigment phycocyanin in turbid inland water. Limnol. Oceanogr. 50, 237–245. Simis, S.G., Ruiz-Verdú, A., Domínguez-Gómez, J.A., Peña-Martinez, R., Peters, S.W., Gons, H.J., 2007. Influence of phytoplankton pigment composition on remote sensing of cyanobacterial biomass. Remote Sens. Environ. 106, 414–427. Song, K., Li, L., Tedesco, L., Clercin, N., Hall, B., Li, S., Shi, K., Liu, D., Sun, Y., 2013. Remote estimation of phycocyanin (PC) for inland waters coupled with YSI PC fluorescence probe. Environ. Sci. Pollut. Control Ser. 20, 5330–5340. Soontiens, N., Binding, C., Fortin, V., Mackay, M., Rao, Y.R., 2019. Algal bloom transport in Lake Erie using remote sensing and hydrodynamic modelling: sensitivity to buoyancy velocity and initial vertical distribution. J. Gt. Lakes Res. 45 (3), 556–572. Stramska, M., Stramski, D., 2005. Effects of a nonuniform vertical profile of chlorophyll concentration on remote-sensing reflectance of the ocean. Appl. Opt. 44, 1735–1747. Streher, A.S., Barbosa, C.C.F., Galvão, L.S., Goodman, J.A., de Moraes Novo, E.M.L., Silva, T.S.F., 2014. Sunglint correction in airborne hyperspectral images over inland waters. Rev. Bras. Cartogr. 3. Su, T.-C., Chou, H.-T., 2015. Application of multispectral sensors carried on unmanned aerial vehicle (UAV) to trophic state mapping of small reservoirs: a case study of TainPu reservoir in Kinmen, Taiwan. Remote Sens. 7, 10078–10097. Tae, E.A., Hong, S.H., Lee, J.B., Lee, J.J., Lee, S.H., 2014. He Toxins Analysis of Daecheong Reservoir Using LC/MS Method. Korean Society af Water Quality, pp. 374–375 2014. Toming, K., Kutser, T., Laas, A., Sepp, M., Paavel, B., Nõges, T., 2016. First experiences in mapping lake water quality parameters with Sentinel-2 MSI imagery. Remote Sens. 8, 640. Tyler, A., Svab, E., Preston, T., Présing, M., Kovács, W., 2006. Remote sensing of the water quality of shallow lakes: a mixture modelling approach to quantifying phytoplankton in water characterized by high‐suspended sediment. Int. J. Remote Sens. 27, 1521–1537. Uitz, J., Stramski, D., Reynolds, R.A., Dubranna, J., 2015. Assessing phytoplankton community composition from hyperspectral measurements of phytoplankton absorption coefficient and remote-sensing reflectance in open-ocean environments.

16