Int. J. Environ. Res. Public Health 2015, 12, 10391-10417; doi:10.

3390/ijerph120910391
OPEN ACCESS

International Journal of
Environmental Research and
Public Health
ISSN 1660-4601
www.mdpi.com/journal/ijerph

Article

Estimation of Chlorophyll-a Concentration and the Trophic

State of the Barra Bonita Hydroelectric Reservoir Using
OLI/Landsat-8 Images
Fernanda Sayuri Yoshino Watanabe 1,*, Enner Alcântara 1,*, Thanan Walesza Pequeno Rodrigues 1,
Nilton Nobuhiro Imai 1, Cláudio Clemente Faria Barbosa 2 and Luiz Henrique da Silva Rotta 1

1
Department of Cartography, Sao Paulo State University, Cep. 19060-900, Presidente Prudente,
SP 19060-900, Brazil; E-Mails: [email protected] (T.W.P.R.); [email protected] (N.N.I.);
[email protected] (L.H.S.R.)
2
Image Processing Division, National Institute for Space Research, São José dos Campos,
SP 12227-010, Brazil; E-Mail: [email protected]

* Authors to whom correspondence should be addressed;

E-Mails: [email protected] (F.S.Y.W.); [email protected] (E.A.);
Tel.: +55-18-3229-5519 (F.S.Y.W. & E.A.); Fax: +55-18-3229-1534 (F.S.Y.W. & E.A.).

Academic Editor: Yu-pin Lin

Received: 26 June 2015 / Accepted: 19 August 2015 / Published: 26 August 2015

Abstract: Reservoirs are artificial environments built by humans, and the impacts of these
environments are not completely known. Retention time and high nutrient availability in
the water increases the eutrophic level. Eutrophication is directly correlated to primary
productivity by phytoplankton. These organisms have an important role in the
environment. However, high concentrations of determined species can lead to public health
problems. Species of cyanobacteria produce toxins that in determined concentrations can
cause serious diseases in the liver and nervous system, which could lead to death.
Phytoplankton has photoactive pigments that can be used to identify these toxins.
Thus, remote sensing data is a viable alternative for mapping these pigments, and
consequently, the trophic. Chlorophyll-a (Chl-a) is present in all phytoplankton species.
Therefore, the aim of this work was to evaluate the performance of images of the sensor
Operational Land Imager (OLI) onboard the Landsat-8 satellite in determining Chl-a
concentrations and estimating the trophic level in a tropical reservoir. Empirical models
Int. J. Environ. Res. Public Health 2015, 12 10392

were fitted using data from two field surveys conducted in May and October 2014 (Austral
Autumn and Austral Spring, respectively). Models were applied in a temporal series of OLI
images from May 2013 to October 2014. The estimated Chl-a concentration was used to
classify the trophic level from a trophic state index that adopted the concentration of this
pigment-like parameter. The models of Chl-a concentration showed reasonable results, but
their performance was likely impaired by the atmospheric correction. Consequently, the
trophic level classification also did not obtain better results.

Keywords: chlorophyll-a; bio-optical models; case-2 waters; remote sensing;

multispectral image

1. Introduction

The construction and use of hydroelectric dams and reservoirs change the hydrodynamics of rivers,
causing diverse impacts to terrestrial and aquatic systems [1]. The increase in residence time leads to
the availability of nutrients for a longer time and, consequently, to eutrophication [2]. Eutrophication is
a natural process in aquatic systems, characterized by the over-enrichment of waters with nutrients,
such as phosphorus and nitrogen [3,4].
However, this process can be accelerated by human inputs of nutrients [5], through the discharge of
domestic, agricultural, and industrial effluents [6], known as cultural eutrophication [7]. Eutrophication
increases primary productivity, represented by algae communities, such as phytoplankton [4,5].
Phytoplankton has an important role in aquatic systems. Furthermore, phytoplankton is responsible for
the ocean being considered a carbon sink and an ally against global warming [8–10].
On the other hand, reservoirs have been appointed as potential emitters of greenhouse gases [11],
although there are reservoirs with high phytoplankton concentrations [2,12]. Studies of Brazilian
hydroelectric reservoirs showed that greenhouse gases were emitted by these aquatic systems due to
high eutrophication levels [12–15]. Furthermore, some species present pigments that increase the
efficiency of light and nitrogen absorption, such as cyanobacteria, favoring their predominance in
different aquatic environments [16]. Some of these species produce toxic metabolites that can cause
minor illnesses in the liver and nervous system [17–20]. Hence, the monitoring and classification of
trophic levels are important for identifying the waters’ health.
Photosynthetically active pigments, such as chlorophylls and carotenoids, can be used as proxies for
results these organisms. Chlorophyll-a (Chl-a) is the main pigment found (usually in larger amounts)
in all phytoplankton species, and the Chl-a concentration is used for primary productivity studies and
the determination of phytoplankton biomass. Due to a high correlation between the Chl-a and nutrient
concentrations, this pigment was used as a proxy of trophic status. To determine the trophic status of
rivers, lakes, and reservoirs, researchers created different ways of classifying them. Trophic State
Index (TSI) [21–23] uses phosphorus concentration, Secchi disk transparency, and Chl-a concentration
as parameters of the equation.
The sampling scheme based on punctual measurement collection is a time, money and human
resources-consuming method that does not represent spatial variability. One way to overcome these
Int. J. Environ. Res. Public Health 2015, 12 10393

limitations is by using remote sensing [24]. Remote sensing offers a significant source of information
that can be used in methods for the operational large-scale monitoring of water quality [25].
Remote sensing has great potential to detect pigments, such as Chl-a, because their constituents are
photo-active and their reflected energy can be measured by photosensitive sensors [26,27].
In the last decades, several bio-optical models were developed to estimate pigment concentrations
and other optically active constituents (OAC) from hyperspectral (HICO, Hyperion) or moderated
spectral resolution images (MERIS, MODIS and SeaWiFis) [28–34]. Previous studies [35–38] used
multispectral images, such as Landsat, to estimate phytoplankton pigments. A trophic state
classification can be applied using the estimated Chl-a. [39] evaluated the potential of a field
spectrometer and IRS-1C satellite image to monitor the Chl-a content and trophic state of a lake in
Germany. This lake belonged to a complex aquatic system and presented oligotrophic characteristics,
with algal blooms during specific periods of the year.
The Operational Land Imager (OLI) sensor was launched onboard the Landsat-8 satellite in
February 2013. The OLI sensor has shown the potential for applications in assessing water quality,
and it is freely available. In [40] its authors conducted an analysis of the OLI’s radiometric performance
in applications of the aquatic sciences, such as the retrieval of optically active constituents and benthic
mapping. This study showed that OLI derived water-leaving radiance and reflectance presented some
discrepancy when compared to in situ measurements or MODIS Aqua bands.
In [41] OLI images were used to study offshore suspended sediment concentrations. Their results
showed that the OLI spatial resolution and signal noise ratio was suitable to study the offshore
suspended sediment concentration. A study in the Amazon [42] used a time-series (from 1973 to 2013)
of Landsat-MSS/TM/OLI images to monitor the impacts of mining actives near the Tapajós River
(Amazon, Brazil) from the retrieval of the total suspended sediment (TSS). They obtained satisfactory
results, showing that TSS concentration was in sync with the mining activities.
The main goal of this work was to map the seasonal trophic state of a tropical reservoir.
Two approaches were used. The first approach was based on the Chl-a concentration retrieved from
OLI/Landsat-8 images. The second approach was based on the Chl-a concentration and Secchi disk
data collected during the two field surveys. The results of the second approach were considered as the
ground truth. Both of the approaches used methods based on the classification adopted by the
Environment Protection Agency in Sao Paulo State (CETESB) [43].
Two field campaigns were conducted to collect field spectroscopy data and water samples.
These data were used to fit an empirical model to estimate the Chl-a concentration that was applied on
time series of OLI/Landsat-8 images from May 2013 to October 2014. The classification of the trophic
level was carried out for each OLI scene using the retrieved Chl-a data. The Chl-a retrieved in the
laboratory and the Secchi disk transparency data collected in situ were used to classify the trophic level
at each sampling spot.

2. Study Area

The study area was the Barra Bonita hydroelectric reservoir—BBHR (22°31′10″ S and 48° 32′3″ W),
located in the middle course of the Tietê River, São Paulo State, Brazil (Figure 1). The BBHR is used not
only for power generation but also for many other uses, such as recreation and tourism, fisheries, and
Int. J. Environ. Res. Public Health 2015, 12 10394

aquaculture [44]. The BBHR is the first of six reservoirs cascading in the Tietê River. It was built in
1963, flooding an area of 310 km2 with a volume of 3.622 × 106 m3. It is one of the two accumulation
reservoirs in the dam chain and presents a considerable variation of quota from 439.5 m to 451.5 m [45].
The BBHR is located in a transitional region between tropical to subtropical [2], characterized by a dry
period between May–October and a wet period between November and April [46] (Figure 2).

Figure 1. Study area-BBHR, Tietê River, São Paulo State. (a) Sampling spot locations;
(b) OLI/Landsat-8 image of October 13, 2014, colored composition RGB-432, and Barra
Bonita reservoir localization inside of Brazil.

Figure 2. Plot of average monthly precipitation from January 2009 to December 2013
(5 years) associated to the Barra Bonita automatic station.
Int. J. Environ. Res. Public Health 2015, 12 10395

According to [45] and [44], the maximum depth is 25 m, with an average of 10.2 m, and the water
retention time varies from 30 days (austral summer) to 180 days (austral winter). The flow range is
over 1500 m3·s−1 in the summer (wet season) and approximately 200 m3·s−1 in the winter (dry season).
The BBHR is formed by the confluence of the Piracicaba and Tietê Rivers. The rivers are both
located in highly populated and industrial areas, contributing high load of nutrients and pollution into
the water. The BBHR is considered a eutrophic water body, mainly due to the discharge of wastewater
and agricultural fertilizers [46]. In [12] it was shown that the eutrophication process caused greenhouse
gas emissions, such as CH4 (methane), CO2 (carbon dioxide) and N2O (nitrous oxide) in the BBHR.
These processes occurred due to the decomposition of the organic matter, releasing CH4 and CO2 [11],
and denitrification, releasing NO and N2O to the atmosphere [12].
Historically, the BBHR presents a species richness and high concentration of phytoplankton [47,48].
Reference [47] investigated the phytoplankton species diversity in the reservoir during two years
(1993–1994). The authors verified the predominance of the species Microcystis aeruginosa in summer,
with high diversity in the winter. In [48] the authors conducted a taxonomic analysis during two years
(2002–2004) and recorded nine groups. The most abundant species and M. aeruginosa and free cells of
Microcystis sp. (Cyanophyceae), and Aulacoseira granulata filaments (Bacillariophyceae). The variation
of phytoplankton communities was mainly associated with vertical mixing and residence time.

3. Data and Methods

3.1. Field Survey

Two field campaigns were conducted in distinct seasons of the year. The first field survey was
accomplished from 5 to 9 May 2014 (Austral Autumn–end of the wet season). The second field survey
was conducted from 13 to 16 October 2014 (Austral Spring–end of the dry season). The field survey
dates were chosen to coincide with the Landsat satellite overpass in periods of low rainfalls and during
the ends of the flood and drought periods. Unfortunately, there was high cloud cover during the
satellite overpass for the survey in May.
Radiometric and limnological data, inherent optical properties (IOPs), and water samples were
collected in 20 sampling spots (Figure 1). The sampling spots were randomly distributed along the
reservoir. A semi-automatic approach was applied for selecting the points, adapted from [49]. First, the
standard deviation was calculated from an annual image series composed of nine ETM+/Landsat
7 scenes (without clouds) from 17 August 2002 to 14 June 2003. A stratified random sampling was
applied to the standard deviation using Hawth’s Analysis Tools (GME, Brisbane, Australia) for
ArcGIS software (ESRI, Redlands, CA, USA). A 1 km buffer from margin was created to avoid
optically shallow waters. The points within this buffer were discarded. This last step was especially
important because the drought of that year considerably reduced the level of water.

Field Measurements

Radiometric data were measured using three RAMSES spectroradiometers. One spectroradiometer was
an irradiance sensor (ACC-VIS), and the other two were radiance sensors (ARC-VIS) with a 7°
field-of-view. ARC-VIS and ACC-VIS RAMSES work in a wavelength range of 320 nm to 950 nm and
Int. J. Environ. Res. Public Health 2015, 12 10396

spectral sampling of 3.3 nm. The operation temperature varied from −10 °C to +50 °C, while the
integration time range was 4 ms to 8 s [50]. The cosine collector was pointed upward and coupled to a 1.5
m rod for collecting the incident sky radiation, E s (λ ) (W·m−2). Meanwhile, a radiance sensor was pointed
in the upward direction of 135° to measure the incident sky radiance Ls (λ ) (W·m−2·sr−1). The second
radiance sensor was pointed in the downward direction of 45° to measure the total radiance, Lt (λ )
(W·m−2·sr−1). Lt is the sum of the water-leaving radiance Lw (λ ) (W·m−2·sr−1) plus any Ls (λ )
(W·m−2·sr−1) that was reflected by the water surface in the sensor direction, Lr (λ ) (W·m−2·sr−1) [51]. All of
the sensors simultaneously collect the measurements. The geometry of acquisition has considered [51],
and [52]. These data were used to calculate the remote sensing reflectance, Rrs (sr−1) [50]:
( Lt − ρLs )
Rrs = (1)
Es
where Lt (λ ) is the measurement obtained by the sensor pointed at the water surface; Ls (λ ) is the
incident sky radiance; Es (λ ) is the incident sky irradiance; and ρ is a reflectance factor related to
direction, wavelength, wind speed, sensor FOV, and sky radiance distribution [51,52]. Furthermore,
optical water quality parameters were collected in situ, such as turbidity and Secchi disk depth at each
sampling station. In the first field survey, 19 radiometric samples were successfully collected, and in
20 radiometric samples were successfully collected in the second survey.

3.2. Laboratory Analysis

Water samples were collected at each sampling spot and filtered through a Whatman GF/F glass
fiber filter, 47 mm diameter and 0.7 μm pore size, to estimate the Chl-a concentration in the
laboratory. The filter was frozen and kept in the dark until further analysis. The Chl-a concentration
was estimated using the extraction by acetone method [53].
Water samples were filtered through a Whatman GF/F glass fiber filter (47 mm diameter and 0.7μm
pore size) and stored frozen and in the dark to estimate TSS. In this case, the filters used in the
filtration were previously ignited at 550 °C for 30 min and weighed. In the laboratory, the filters were
dried in an oven at 105 °C for 12 h, desiccated and weighed to obtain the TSS. After, the filters were
ignited at 550 °C for 30 min in the muffle furnace, desiccated and weighed to acquire the inorganic
suspended solids (ISS). Subtracting the ISS from the TSS obtained the organic suspended solids
(OSS), and dividing each component of solids determined their concentrations [54].
The absorption coefficient of each AOC was determined by spectrophotometry. Water samples
were filtered with Whatman GF/F glass fiber filters, kept frozen and in the dark until the analysis to
estimate the absorption coefficient by phytoplankton, detritus and total particulate material.
The measurements were acquired over the 280–800 nm spectral range in 1 nm by using a 2600 UV-Vis
spectrophotometer (Shimadzu, Kyoto, Japan) with dual beam and integrating sphere. The optical
density of the particulate material was obtained from the first reading. After, the pigment in the filter
was extracted with sodium chloride and the filter was measured again to determine the optical density
of the detritus. These optical densities were corrected for multiple scattering effects caused by the
glass-fiber filter [55]. The particles and detritus absorption coefficients were estimated from the
corrected optical densities [56]. The phytoplankton absorption coefficient was obtained by subtracting
the detritus absorption coefficient from the particles absorption coefficient.
Int. J. Environ. Res. Public Health 2015, 12 10397

Water samples were filtered through a Whatman nylon membrane filter with 0.22 μm pore size and
47 mm diameter to measure the CDOM optical density. The filtrate water sample was stored, kept cool
and in the dark until the analysis. Samples were measured at room temperature in a spectral range of
280–800 nm using a quartz cuvette with 10 cm optical path in a Shimadzu 2600 UV-Vis
spectrophotometer with a single beam. Milli-Q water was used as blank reference. The CDOM
absorption coefficients were calculated from Equation (2) [57]:
ACDOM (λ )
aCDOM = 2.3 (2)
r
where ACDOM (λ ) is the optical density at wavelength ( λ ) and r is the cuvette path length in meters.
This value was corrected for particles backscattering, using 700 nm as the reference wavelength [57].

3.3. OLI Image Processing

The empirical models for estimating Chl-a concentration were fitted to be applied on multispectral
images from OLI/Landsat-8, freely available from the United States Geological Survey [58]. Most of
the OLI bands have a terrestrial application. However, the bands with applications on vegetation and
photosynthetic pigments can be suitable for estimating phytoplankton pigment concentrations in highly
productive water bodies. The advantages of using OLI/Landsat-8 images are the acquisition of data
every 16 days and their free availability.
The study area was located in the path/row 220/76. In this work, available OLI/Landsat-8 images
were used since its launch until October 2014. After a preliminary analysis of the images, only those
without cloud cover were selected. Eight scenes from 19 May 2013 to 13 October 2014 were used.
Dates of the scenes used in this work were: 19 May 2013; 4 June 2013; 23 August 2013; 8 September
2013; 13 December 2013; 30 January 2014; 11 September 2014; and 13 October 2014. The first day of
the field survey overlapped with the overpass of the OLI sensor on 13 October 2014.
The radiometric calibration was conducted to convert digital numbers into top-of-atmosphere
radiance, using the metadata released with the images. The retrieval of the at-surface reflectance was
accomplished using the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH),
an atmospheric correction module, implemented in the ENVI software. This tool adopted the
MODerate resolution atmospheric TRANsmission (MODTRAN4), an atmospheric radioactive transfer
code [59]. Agricultural activities, such as sugar cane, orange, and coffee cultures, were the main
activities in the region of the BBHR. Ethanol production is the main industrial activity. Therefore, the
tropical atmospheric model and rural aerosol model were selected for correction.
The cirrus band was included in the atmospheric process. The cirrus band was used to detect clouds
and water vapor contents that were not detectable in the other bands [58]. Atmospheric aerosols can be
liquid or solid particles suspended in a gas [60]. Rayleigh atmospheric scattering (aerosol scattering)
affects the direction of short wavelengths, resulting in haze in the blue and green bands [61].
The inclusion of the cirrus band helped to eliminate the estimation of negative at-surface reflectance by
FLAASH at short wavelengths (blue region). Negative values were associated with the overestimation
of aerosol reflectance [62–64].
The surface reflectance values were divided by π [65,66] to convert them into Rrs. Five samples
(see P1, P2, P4, P5, and P6 in Figure 1) were collected at the time of the Landsat-8 overpass
Int. J. Environ. Res. Public Health 2015, 12 10398

(13 October 2014) and were used to evaluate the performance of the atmospheric correction. Optical
closure was conducted between the Rrs obtained of the image and the simulated OLI Rrs figured out
from the field spectroscopy data.

3.4. Model Calibration

Only two-band models were tested due to the limited number of OLI/Landsat-8 bands that could be
used to estimate Chl-a. The structure proposed by [67] was adopted for fitting an empirical model for
the retrieval Chl-a concentrations using OLI/Landsat-8 bands and in situ data (Equation (3)):
−1 (λ1 ) × Rrs (λ 2 )
Chl-a concentration ∝ Rrs (3)
In this work, empirical models were fitted from the simulated OLI bands. The Rrs for OLI bands
were simulated (RrsL8k) using OLI’s spectral response [68], from the field radiometric data set
composed of 39 measurements collected in the two field campaigns, as Equation (4):
λ2 k
 Lw (λ ) S k (λ ) dλ
RrsL8k = λ1k (4)
λ2 k
 Ed (λ ) S k (λ ) dλ
λ1k

where Sk (λ) is the radiometric sensitivity of band k that extends from lower wavelength λ1k up to upper
wavelength λ2k. SWIR-1 and SWIR-2 bands were not simulated because their wavelengths exceeded
the spectral range collected by the RAMSES spectroradiometer [50].
The proposed of this work to fit a unique model that could be applied to time series of OLI/Landsat-8
images of the BBHR. Therefore, samples gathered from two field campaigns (May and October) were
used to calibrate the empirical model. The five samples collected at the time of the satellite overpass
(13 October 2014) were used for validation. From the other 34 samples, it was done selection of the
calibration samples. Samples must be within the prediction interval, considering a significance level of
0.95. Twenty-six samples were used to calibrate the empirical bio-optical model.
Parameterization was conducted using the Interactive Correlation Environment (ICE) [69]. ICE is a
statistical web tool for data collinearity analysis that determines the correlation of a dependent variable
over the ratio between two independent variables. In this case, the dependent variable was the Chl-a
concentration and the independent variables were OLI/Landsat-8 bands. The bands ratio (Rrs) that was
more correlated with the Chl-a concentration was used to develop a simple two-band model. The
models were fit using the least-squared method and the data set composed of 26 samples. Linear,
polynomial, and exponential fits were tested.

3.5. Tuning Models and Application of Existing Models

Other bands algorithms proposed by different researchers were tested. The bands selection of these
algorithms were based on different sensors with specific bands for water, such as MERIS and MODIS,
and made it possible to test models with more than two bands. First, band combinations proposed by
different authors were tuned from the calibration using the spectroscopy data collected in the two field
campaigns. The models tested were the two-band and three-band algorithms developed by [67],
Int. J. Environ. Res. Public Health 2015, 12 10399

originally for terrestrial application, with bands selected by Moses et al. [70] and the Normalized
Difference Chlorophyll Index (NDCI) developed by [32]. The applied band combinations, references,
and adopted abbreviations are shown in Table 1.

Table 1. Structure and bands adopted by two-band and three-band algorithms, and NDCI.
Abbreviation Bands Combination Reference
2B Rrs (708) × Rrs (665) −1 [70]
3B ( Rrs (665) −1 − Rrs (708) −1 ) × Rrs (753) [70]
NDCI ( Rrs (665) − Rrs (708)) ( Rrs (665) + Rrs (708)) [32]

The calibrations proposed in the literature [32,70–72] were also tested. MERIS bands were used to
compose the two-band and three-bands algorithms. Table 2 shows the coefficients obtained by
different researchers. The researchers proposed linear, polynomial of second degree, and exponential
fits. These models were calibrated using a large dataset of different seasons and aquatic environments.

Table 2. Coefficients of calibration proposed in the literature using 2B and 3B and NDCI.
Reference Index a* b* c* d*
2B −37.94 61.324 − −
[70]
3B 23.174 232.29 − −
2B −19.3 35.75 − 1.124
[71]
3B 16.45 113.36 − 1.124
2B −15.18 14.85 25.28 −
[72]
3B 25.66 215.95 315.5 −
[32] NDCI 14.039 86.115 194.325 −
* Linear fit y = a + bx; polynomial fit y = a + bx + cx ; Exponential y = (a + bx)d.
2

3.6. Validation

The algorithms performance was evaluated using the following statistic metrics: normalized root
mean squared error (NRMSE) (Equation (5)); the RMSE (root mean squared error) (Equation (6));
mean absolute percentage error (MAPE) (Equation (7)); bias (Equation (8)); mean normalized bias
(MNB) (Equation (9)); normalized root mean squared error (NRMS) (Equation (10)); and
determination coefficient (R2):
RMSE
NRMSE = (5)
( xmaximo
measured measured
− xminimo )
1 n estimated
RMSE =  (x
n i =1 i
− ximeasured ) 2 (6)

n  xiestimated − ximeasured 
  x measured

 (7)
MAPE =  
i =1 i
n
1 n estimated
bias = 
n i =1
( xi − ximeasured ) (8)
Int. J. Environ. Res. Public Health 2015, 12 10400

MNB = mean (ε i ) %
(9)
NRMS = stdev (ε i ) %
(10)
where ε i is the systematic and random error (Equation (11)):

ε i = 100 ×
(xiestimated − ximeasured )
(11)
ximeasured

3.7. Trophic State Classification

The trophic state classification approach used was an adaptation from the method proposed by [18],
conducted for CETESB [23,43], whose Equations (12), (13) and (14) correspond to the calculation
from the Secchi disk transparency (S, m), Chl-a concentration (Chl-a, mg·m−3) and total phosphorus
(Pt, mg·L−1), respectively:
TSI ( S ) = 10 × {6 − [(ln S ) ln 2]} (12)

TSI (Chla ) = 10 × {6 − [(0.92 − 0.34 × (ln Chla )) ln 2]} (13)

TSI ( Pt ) = 10 × {6 − [1.77 − 0.42 × (ln Pt ) ln 2]} (14)

Each TSI was separated into six tropic classes associated to a TSI range. Table 3 shows the
classification of the reservoir’s trophic levels based on the TSI parameters and equivalent to total
phosphorus, Chl-a, and Secchi disk transparency proposed by [23].

Table 3. Trophic state classification adopted by CETESB according to TSI and

equivalence to the total phosphorus concentration (mg·L−1), Chl-a concentration (mg·m−3),
and Secchi disk transparency (m) parameters.
Secchi Disk
Trophic State Total Phosphorus (mg·L−1) Chl-a (mg m−3) TSI
Transparency (m)
Ultraoligotrophic Pt ≤ 0.008 Chl-a ≤ 1.17 S ≥ 2.4 TSI ≤ 47
Oligotrophic 0.008–0.019 1.17–3.24 2.4–1.7 47–52
Mesotrophic 0.019–0.052 3.24–11.03 1.7–1.1 52–59
Eutrophic 0.052–0.12 11.03–30.55 1.1–0.8 59–63
Supertrophic 0.12–0.2333 30.55–69.05 0.8–0.6 63–67
Hypertrophic 0.233 < Pt 69.05 < Chl-a 0.6 > S TSI > 67

Reference [21] suggested that the choice of the indicator of the trophic state depends on the lake and
seasonal period. According to [21], the use of Secchi disk transparency is not recommended for lakes
with high nonalgal particles concentration, in highly colored lakes, and in extremely clear lakes.
The main advantages of using a Secchi disk is the operational simplicity and low cost, and it usually
has similar TSI values to Chl-a.
Reference [21] also suggested using the Chl-a concentration as the trophic state indicator because
its values had less interference from other environment parameters. On the other hand, the accurate
total phosphorus concentration as an indicator depends on it being the limiting factor for algal growth
Int. J. Environ. Res. Public Health 2015, 12 10401

and being correlated to the algal biomass. Thus, the authors suggested classifying using a biological
parameter, such as Chl-a in summer and total phosphorus in spring and fall. In this work,
the classification was conducted using the Chl-a parameter and applied on all seasons.

4. Results

Figure 3 shows Rrs spectra computed from in situ data collected on May 2014 (Figure 3a) and
October 2014 (Figure 3b). The Rrs curves presented characteristics of productive waters both in May and
October. The Chl-a features were quite accentuated even in the spectra with lower Chl-a concentration
(19.1 mg·m−3). All of the curves showed high absorption at the blue and red spectral regions,
highlighting the reflection peak at the green region. Another feature associated with the presence of
Chl-a was the peak at the beginning of near infrared (NIR) (region associated to red edge), at
approximately 710 nm to 720 nm. Due to the high Chl-a concentration observed in October, the peak at
NIR nm is highlighted in relation to the visible spectrum, and shifted slightly to the right. Another
reflection feature was observed at 810 nm associated with both chlorophyll and organic matter [73].

Figure 3. Rrs spectra related to the field surveys conducted in (a) 5–9 May 2014, and (b)
13–16 October 2014.

In addition to Chl-a features, every Rrs spectrum presented an accentuated absorption feature at
approximately 620 nm to 630 nm associated with the presence of phycocyanin, a pigment present in
cyanobacteria [16,74]. The reflectance peak at approximately 650 nm to 660 nm was associated with
phycocyanin fluorescence [16]. Phycocyanin has the characteristics of low absorption in blue and
green regions [16]. Historically, Microcystis was the predominant species in the BBRH [47,48], which
explained the phycocyanin and Chl-a spectral features. Moreover, high phycocyanin concentrations are
a concern from the environmental and public health point of view. The BBHR has multiple uses of
direct and indirect contact, such as recreation and fish intake.
There was an increase considerable of the high phytoplankton, CDOM, and particles absorption
coefficients from May to October. The average particles absorption coefficients were 1.6 m−1 in May
and 2.8 m−1 in October (Table 4). The average Chl-a concentration was 122.5 mg·m−3 in May, while in
Int. J. Environ. Res. Public Health 2015, 12 10402

October the average was 428.7 mg·m−3 (Table 4). The Chl-a concentrations ranged from 19.1 mg·m−3
to 293.3 mg·m−3 in May, and from 263.2 mg·m−3 to 797.8 mg·m−3 in October. It is important to note
that the minimum Chl-a concentration measured in October (263.2 mg·m−3) was near the maximum
concentration collected in May (293.2 mg·m−3).

Table 4. Descriptive statistics of the optical water quality parameters measured in situ or
determined in the laboratory: Chl-a concentration; Secchi disk depth; turbidity; TSS
concentration; OSS concentration; ISS concentration; OSS/TSS ratio; ISS/TSS ratio;
phytoplankton absorption coefficient at 665nm; CDOM absorption coefficient at 440nm;
total particulate material absorption coefficient at 440nm; and wind speed. Statistics
metrics used: minimum value (Min), maximum value (Max), median, median, standard
deviation (SD) and coefficient of variation (CV) in percentage (%), that is
CV = (SD / mean) × 100.
a. Dataset measured on 5–9 May 2014 at 19 stations
Parameter Min Max Mean Median SD CV (%)
−3
Chl-a, mg m 19.1 293.2 122.5 105.9 72.3 60
Secchi disk depth, m 0.8 2.3 1.5 1.4 0.4 30
Turbidity, NTU 1.7 12.5 5.2 5 2.5 50
−1
TSS, mg L 3.6 16.3 7.0 6.5 3.2 50
OSS, mg L−1 2.8 14.7 5.9 4.8 3.1 50
−1
ISS, mg L 0.2 4.4 1.1 0.8 0.9 80
OSS/TSS 0.45 0.98 0.83 86 11.6 10
ISS/TSS 0.02 0.55 0.17 14 11.6 70
aphy (665), m−1 0.2 1.3 0.6 0.5 0.3 50
−1
aCDOM (440), m 0.6 1.1 0.8 0.8 0.1 10
−1
ap (440), m 0.7 2.9 1.6 1.5 0.7 40
−1
Wind, m s 0.6 4.9 1.8 1.6 1.1 60

b. Dataset measured on 13–16 October 2014 at 20 stations

Parameter Min Max Mean Median SD CV(%)
−3
Chl-a, mg m 263.2 797.8 428.7 368.9 154.5 40
Secchi disk depth, m 0.4 0.8 0.6 0.6 0.1 20
Turbidity, NTU 11.6 33.2 18.6 17.6 5.3 30
−1
TSS, mg L 10.8 44 22 21.2 7 30
−1
OSS, mg L 10.2 30.4 18.2 18.4 4.8 30
−1
ISS, mg L 0.6 3.8 2.6 2.8 1 40
OSS/TSS 0.8 1 0.9 0.9 0.1 10
ISS/TSS 0.04 0.2 0.1 0.1 0.4 10
−1
aphy (665), m 0.6 2.1 1.1 1 0.4 40
−1
aCDOM (440), m 0.9 2.4 1.3 1.3 0.3 30
ap (440), m−1 1.6 5.6 2.8 2.6 1 40
Wind, m s−1 0 5 1.5 1.1 1.5 100

The TSS concentration was mainly composed of organic particles (OSS) (approximately 90%),
indicating that TSS could be directly related to phytoplankton. Nevertheless, the complexity of the
Int. J. Environ. Res. Public Health 2015, 12 10403

BBHR must not be neglected during the development of the models. An ISS of approximately
4 mg·L−1 can mask some features of pigments, mainly in the visible region.
As expected, increasing turbidity influenced the Secchi disk transparency. Maximum turbidity was
12.5 NTU (Nephelometric Turbidity Unit) in May and 33.2 NTU in October. The Secchi disk depth
range was 0.8 m to 2.3 m in the first field campaign, while the Secchi disk depth did not exceed 0.8 m in
the second field campaign. Furthermore, the elevation of the TSS and pigment concentration reflected
the particles absorption coefficient (ap) and phytoplankton absorption coefficient (aphy), respectively.
The collection of radiometric measurements was not affected by the influence of waves in both field
campaigns. Except for a few sampling spots, the wind speed was close to 0 m·s−1.
Simulation of the OLI bands was accomplished using the spectra plotted in Figure 3. Figure 4
shows the simulated Rrs curves. As expected, the features associated with the absorption by
phycocyanin (approximately 630 nm) and reflectance by chlorophyll-a (near 720nm) disappeared. The
high absorption at the green region and low absorption at the blue region, associated to chlorophyll-a
presence, were maintained. Simulated bands were used to select and calibrate the empirical models.

Figure 4. Rrs simulated for OLI/Landsat-8 bands of the spectra collected in (a) May 2014
and (b) October 2014.

The bands were selected using the web tool ICE, and the result is presented in Figure 5.
The 2-Dimensional correlogram shows the correlation between the bands ratio and Chl-a
concentration. As expected, the NIR-Red ratio showed the highest positive correlation (0.865) with
Chl-a. The Red-NIR ratio presented the highest negative correlation of -0.8161. The NIR-Green and
NIR-Blue also showed good correlation (greater than 0.8).
These bands were tested in the development of an empirical model. The models calibration was
accomplished by a least-squared regression. Linear and polynomial fits were tested, obtaining
satisfactory results (Figure 6). The adjustment line of regression and R2 showed that there were little
differences between the linear and polynomial fits (Figure 6a). The NIR-Red ratio had the best
calibrations, with R2 of 0.7537 to linear fit and R2 of 0.7555 to polynomial calibration. The NIR-Green
(Figure 6b) and NIR-Blue (Figure 6c) ratios also presented satisfactory fits. The NIR-Blue ratio
obtained R2 lower than 0.70, except for the linear fit.
Int. J. Environ. Res. Public Health 2015, 12 10404

Figure 5. 2-Dimensional plot of the correlation coefficients (R) between bands ratio and
Chl-a concentration.

Figure 6. Two-band models developed from the OLI bands simulated using field
radiometric data, using (a) NIR-Red ratio, (b) NIR-Green ratio, and (c) NIR-Blue ratio.
Int. J. Environ. Res. Public Health 2015, 12 10405

Table 5 shows the calibration parameters, determination coefficient of fit, and p-value statistics for
each calibration using OLI/Landsat-8 bands. Table 5 shows the same parameters for tuned models
using MERIS wavelengths proposed by different authors (Table 2). Among all of the calibrations
tested, the best fits were obtained by the NIR-Red ratio using OLI/Landsat-8 bands. Among the models
using MERIS wavelengths, the model 3B showed the best fit both linear and polynomial. The p-value
statistics of the regressions were equal to zero or negligible; that is, the coefficients of the regression
models were suitable to explain the Chl-a concentration.

Table 5. Fit parameters of the two-band models obtained from the simulated bands of the
OLI/Landsat-8 sensor and bands algorithms tuned for bands of the MERIS and MODIS
sensors (2B; 3B and NDCI): equations coefficients (a, b and c), R2 (determination
coefficient), and p-value statistic.
Index a b c R2 p-value
NIR/Red (Linear) −77.16 925.001 − 0.7537 0.00000
NIR/Green (Linear) −18.96 1289.84 − 0.6858 0.00000
NIR/Blue (Linear) −58.90 742.33 − 0.7085 0.00000
NIR/Red (Polynomial) −120.87 1179.72 −323.81 0.7555 0.00000
NIR/Green (Polynomial) −149.72 2557.24 −2565.99 0.7085 0.00000
NIR/Blue (Polynomial) −123.35 1058.23 −335.47 0.7156 0.00000
2B (Linear) −124.72 213.73 − 0.3910 0.0006
3B (Linear) 71.212 603.15 − 0.5794 0.00001
NDCI (Linear) 53.361 767.2 - 0.3945 0.0006
2B (Polynomial) −214.41 321.62 −30.63 0.3926 0.0006
3B (Polynomial) 33.95 918.36 −465.68 0.5963 0.00001
NDCI (Polynomial) 56.818 724.88 93.472 0.3946 0.0006
Linear fit y = a + bx; polynomial fit y = a + bx + cx2

Table 6 shows the validation results of the fitted models from simulated OLI bands (Table 6a), the
tuned models from the bands combinations proposed by literature (Table 6b), and the same bands
combinations using calibrations proposed by different authors [31,64] (Table 6c). Among the models
for OLI data (Table 6a), the polynomial NIR-Green model presented the best results in the validation.
This model presented a NRMSE of 82.39%; MAPE of 36.02%; bias of -126.47; MNB of 35.13%; and
R2 of 0.1929. The polynomial NIR-Blue model presented the second best results with a NRMSE of
86.54%; MAPE of 45.69%; bias of -137.90; MNB of 43.28%; NRMS of 31.47%; and R2 of 0.3441.
Overall, the models developed using simulated OLI bands did not have satisfactory performances.
Tuned models yielded better results than the fitted models using simulated OLI/Landsat-8 data
(Table 6b). Among the tuned models, polynomial and linear 3B models had the best performances with
NRMSE of 16.71% and 18.70%, respectively. On the other hand, the NDCI model yielded the worst
estimation, with MNB of 83.8%. However, the tuned NDCI performance was still better than the fitted
models. Performances of the tuned 2B models were next to the tuned NDCI models.
Among the models proposed by the literature, the 3B model proposed by [72] showed the best
performance, with NRMSE of 80.45% and MAPE of 43.57%. The 2B model proposed by [72] had the best
Int. J. Environ. Res. Public Health 2015, 12 10406

results when compared to other 2B models tested (Table 6c). The NDCI calibration proposed by [32]
presented the lowest performance, with NRMSE of 157.99% and MAPE of 81.72%.
Figure 7 shows plots 1:1 between the validation samples and Chl-a concentration estimated from
the models using simulated OLI/Landsat-8 bands that had the best results in the validation, i.e.,
NIR-Green and NIR-Blue models with polynomial fit. Overall, the polynomial NIR-Green model was
the best model, with a satisfactory fit and the best results in the validation when compared to other
fitted models using simulated OLI/Landsat-8 bands. Therefore, the bio-optical model was applied to
the temporal series of OLI images from May 2013 to October 2014.

Table 6. Validation of bio-optical models applied for estimating Chl-a concentration from
the NRMSE, MAPE, bias, MNB, NRMS, and R2.
a. Models calibrated using ICE [62] from OLI bands simulated and data collected in the study area.
NRMSE MAPE
Model Bias MNB (%) NRMS (%) R2
(%) (%)
NIR/Red (Linear) 170.01 91.96 307.56 91.96 44.17 0.1921
NIR/Green (Linear) 190.91 103.90 348.61 103.90 47.00 0.2823
NIR/Blue (Linear) 100.78 52.32 163.16 50.84 36.12 0.3301
NIR/Red (Polynomial) 144.17 77.92 259.62 77.92 37.71 0.1934
NIR/Green (Polynomial) 82.39 36.02 −126.47 −35.13 32.38 0.1929
NIR/Blue (Polynomial) 86.54 45.69 137.90 43.28 31.47 0.3441
b. Tuning bands combinations proposed by literature.
NRMSE MAPE MNB NRMS
Model Bias R2
(%) (%) (%) (%)
2B (Linear) 32.09 13.60 −51.71 −13.60 8.39 0.816
3B (Linear) 18.70 8.90 0.75 −0.47 11.62 0.7953
NDCI (Linear) 33.84 12.48 −49.56 −12.44 9.59 0.867
2B (Polynomial) 32.49 12.96 −50.26 −12.96 8.86 0.7918
3B (Polynomial) 16.72 7.67 −3.19 −0.34 9.69 0.7724
NDCI (Polynomial) 33.57 12.48 −49.54 −12.48 9.47 0.7759
c. Models proposed by authors, calibrated using data from other environments
NRMSE MAPE MNB NRMS
Model Bias R2
(%) (%) (%) (%)
2B [70] 146.82 75.82 −274.69 −75.82 2.35 0.816
3B [70] 120.13 62.88 −226.17 −62.88 4.56 0.867
2B [71] 146.19 75.64 −273.87 −75.64 2.37 0.821
3B [71] 127.51 66.81 −240.23 −66.81 4.35 0.8723
2B [72] 128.11 66.68 −240.63 −66.68 3.84 0.841
3B [72] 80.45 43.57 −151.14 −43.57 13.62 0.8975
NDCI [32] 157.99 81.72 −295.85 −81.72 1.85 0.7978

An atmospheric correction was applied to OLI images so the model could be applied. Figure 8
shows the validation of the atmospheric correction. Plotting the OLI imagery Rrs versus simulated OLI
Rrs showed that Rrs was overestimated (Figure 8a). Simulated OLI Rrs and OLI imagery Rrs spectra of
the sampling point P1 are shown in Figure 8b. There is a difference in the magnitude between the Rrs
Int. J. Environ. Res. Public Health 2015, 12 10407

curves, but the shapes of both are very similar. Likely, difference in the magnitude was associated to
the glint effect or aerosols that were not completely removed by the atmospheric inversion model.
In models of only one band, the magnitude difference could underestimate or overestimate the
estimations. However, the bands ratios eliminated this problem because the shape remained equal.

Figure 7. Two-band models validation, using (a) polynomial NIR-Green ratio and
(b) polynomial NIR-Blue ratio.

Figure 8. Validation of atmospheric correction. Optical closure of (a) FLAASH satellite

Rrs versus simulated Rrs for OLI data at Blue, Green, Red, and NIR bands and (b)
comparison of Rrs calculated for OLI imagery (dotted line) with simulated OLI Rrs (solid
line) at the time of the overpass.

Figure 9 shows the maps of the Chl-a concentration estimated by the empirical model using the
NIR-Green ratio. Many pixels were estimated with negative values, showing that the model greatly
underestimated the Chl-a concentration. These pixels were not represented in the maps (Figure 9a–h),
but it can clearly see in Figure 9f (30 January 2014 image). January had extremely high Chl-a
concentrations when compared to the other months, and the model was not able to suitably estimate
high concentrations. The model yielded negative values in other months as well. Despite the model
having underestimated high Chl-a concentration, variations of concentration were consistently
observed throughout the reservoir. When the points in the map were compared to the field data, it
verified that there was coherence in the areas of major and minor concentrations of Chl-a.
Int. J. Environ. Res. Public Health 2015, 12 10408

Chl-a Concentration (mg·m−3)

Figure 9. Maps of Chl-a concentration (mg·m−3) retrieved from OLI/Landsat-8 based on

NIR-Green algorithm for (a) May 2013, (b) June 2013, (c) August 2013, (d) September
2013, (e) December 2013, (f) January 2014, (g) September 2014, and (h) October 2014.
Int. J. Environ. Res. Public Health 2015, 12 10409

Figure 10 shows maps representing the points of the trophic state for May (Figure 10a,b) and
October (Figure 10c,d) 2014. Each point represents the trophic level of a sample of Secchi disk
transparency (Figure 10a,c) and Chl-a concentration (Figure 10b,d) collected in the field. It was clear
that Secchi disk parameter tended to classify the water for lower trophic levels when compared to
classification from Chl-a in both periods.
This difference was more visible in May than in October, when the trophic levels were lower.
If there really was an underestimation, then sanitation issues may be of concern. In this situation, if the
classification was conducted by Secchi disk, it would be regarded as the suitable for multiple uses.
Nevertheless, this could be a mistake that could endanger public health.

Figure 10. Trophic state classification: from Secchi disk transparency and Chl-a
concentration. Classification for May according to (a) Secchi and (b) Chl-a, and October
(c) Secchi and (d) Chl-a.

Despite the classification from the Secchi disk in May that indicated an environment with lower
trophic levels, other classifications showed the opposite. The predominance of hypereutrophic points
from the classification by Chl-a was observed in both May and October. Reference [21] suggested that
these parameters yielded similar trophic states in conditions of non-clear waters and the predominance
of algal particles. Although algal particles were predominant in the BBHR, the classification using the
Secchi disk and Chl-a yielded quite different results.
Int. J. Environ. Res. Public Health 2015, 12 10410

5. Discussion

Chl-a measurements collected in situ corroborated with the classifications of water quality that were
classified by CETESB. From the beginning of monitoring, water quality was considered bad due to
high levels of eutrophication. Chl-a concentrations were very high in May and October. Chl-a
concentrations were far above the minimum value of hypertrophic waters, which has been determined
as 69.05mg·m−3 [23,43]. This situation worsened due to the historic elevated proliferation and
predominance of M. aeruginosa and Microcystis sp, species of cyanobacteria known to produce
hepatoxin [47,48]. The BBHR is used for activities such as recreation and fishing. Direct contact with
water or ingestion of fish can cause problems, from minor skin problems to microcystin
toxin contamination.
The fitted model in this work was able to represent the seasonal dynamic in the BBHR when there
was no scum. Nevertheless, this model cannot accurately estimate the Chl-a concentration and,
consequently, the trophic state. Analyzing Figure 9a–h, the lacustrine zone (the wide area before the
river narrowing and next to the dam) [44] of the BBHR had high Chl-a concentrations when compared
to other points of the reservoir, mainly in the Austral Spring and Austral Summer. Increasing residence
time favored the development of algae in this region of the BBHR. The Tietê River, before the
confluence with the Piracicaba River, also had high Chl-a concentrations when compared to other
points of the reservoir. In this section of the BBHR, the eutrophication caused by a high discharge of
pollutions coming from the metropolitan area of São Paulo city was the main reason for the
proliferation of algae [2,12,44,45,75].
There were lower Chl-a concentrations in the Piracicaba River during periods of lower temperature,
such May, June, and August (Austral Autumn and Winter) than in the other regions of the reservoir.
This behavior was confirmed by the reports of Surface Water Quality in São Paulo State [43].
According to the data of three CETESB monitoring spots (located in Piracicaba River, Tietê River and
the BBHR), the Piracicaba River was less eutrophic, while the Tietê River was more eutrophic [43].
In the Austral Spring the situation was different, and it was possible to identify the maximum
concentration in some regions. However, at the beginning of the rainy season (Austral Summer), the
dilution decreased the Chl-a concentration in this region again.
The trophic state classification (Figure 10) showed that the classification by the Chl-a parameter
was more rigorous than Secchi disk parameter. In May, before the confluence, the Piracicaba and Tietê
Rivers had higher trophic levels that decreased along the path. Two points approaching the dams were
considered mesotrophic. However, as the channel narrows, the trophic level starts to rise. In the nearest
point nearest to the dam, its classification was already hypereutrophic. In October, the classification by
Chl-a concentration considered all of points as hypereutrophic.
The classification using Chl-a obtained similar results with the TSI of the three CETESB
monitoring spots inserted into the BBHR [43], with predominantly supertrophic and hypertrophic
environments. In inland waters, the classification using Secchi disk must be used with caution because
the transparency was associated not only with algal particles but also with inorganic particles [76–78].
The inclusion of the cirrus band improved the atmospheric correction results, eliminating negative
values, mainly in the blue region. Typically, the atmospheric correction of Landsat images using
FLAASH results in negative values for dark targets, such as water bodies. Negative values are
Int. J. Environ. Res. Public Health 2015, 12 10411

associated with the overestimation of aerosol reflectance [62–64] and harm the performance of models
that use blue bands [79]. Validation of the atmospheric correction showed that there was a difference
in the magnitude between the simulated Rrs and OLI imagery Rrs. Likely, all of the influence of the
glint effect or aerosol was not removed. Differences in the magnitude were eliminated when bands
algorithms were used. On the other hand, differences in the magnitude negatively influenced the
models that used only one band.

6. Conclusions

Overall, satisfactory fits were obtained from simulated OLI/Landsat-8 bands. The NIR-Red,
NIR-Green and NIR-Blue ratios yielded an R2 greater than 0.70. These results showed that OLI bands
were sensitive enough to detect Chl-a concentration. Nevertheless, validation of the models showed
that fitted models did not have accurate results for estimating the Chl-a concentration. The NIR-Red
ratio yielded the best fits, both linear and polynomial. The NIR-Green ratio obtained minor errors and
did not achieve satisfactory results.
All of the fitted models showed problems of overestimation, except the polynomial NIR-Green
model that underestimated the Chl-a concentration. Validation showed that polynomial NIR-Green
model underestimated high concentrations; however, it also overestimated low concentration that
impaired the classification of the trophic state. The fitted empirical models had estimated values
greater than the threshold for hypereutrophic water (69.05 mg·m−3). Hence, the products of Chl-a
supplied for these models were unsuitable for low trophic environments. The models were not able to
include the range of variations in the Chl-a concentration observed in the BBHR. Consequently, maps
of the trophic levels yielded from the Chl-a concentration estimated by these models would not present
the reliable variation that can be seen in the representation by points.
Although the fitted models did not yield accurate estimates of the Chl-a concentration, due the good
sensitivity of the OLI/Landsat-8 with Chl-a, it is still possible that OLI/Landsat-8 bands are suitable
for other aquatic systems with characteristics different than the BBHR.
Among the bands-based algorithms proposed by the literature and tested, the 3B models obtained
satisfactory results when it they were tuned using the field data. Validation results were greater than
the fitted models using simulated OLI bands. Unfortunately, we did not have an image of the study
area with bands compatible for applying the model and having a spatial representation Chl-a
concentration of the whole reservoir. However, the models would likely yield satisfactory spatial
representations of the Chl-a concentration variations along the BBHR.
Although the models were calibrated from MERIS sensor (not operational) wavelengths, satellite
data of sensors with equivalent wavelengths can be adopted, such as MODIS, Sentinel-2 (both
operational), and Sentinel-3 (the first satellite is scheduled for launch in the fourth quarter of 2015).
Bands of these sensors likely have suitable sensitivity for estimating the Chl-a concentration and,
consequently, the trophic state. Furthermore, these sensors present a band at approximately 620 nm,
allowing the estimation of the phycocyanin pigment. Thus, periodic monitoring can be accomplished
and alerts could be issued for the population.
The calibrations proposed in the literature did not obtain satisfactory results. The calibration
proposed by [72] for 3B combination had similar results when compared to the best results obtained by
Int. J. Environ. Res. Public Health 2015, 12 10412

the fitted models. The unsatisfactory performance of these calibrations showed that was not possible to
obtain a global empirical model (suitable to other time periods and regions), even when using a large
calibration dataset.
Finally, more studies must be completed to improve the models in the BBHR. The inherent optical
properties must be investigated better and applied to other approaches, such as semi-analytical and
quasi-analytical models for estimating the Chl-a concentration.

Acknowledgments

The authors thank the FAPESP Project (Process N° 2012/19821-1), CNPq Projects (400881/2013-6
and 472131/2012-5), and PPGCC/UNESP for financial support, and CAPES for scholarship.

Author Contributions

Fernanda Watanabe wrote the article, helped organize and conduct both field campaigns, and
processed the data. Enner Alcântara is coordinator of the FAPESP Project (Process N° 2012/19821-1)
and CNPq Project (Process N° 472131/2012-5). He gave several ideas for the article helped organize
the field campaigns and reviewed the article. Thanan Rodrigues helped organize and conduct the field
surveys, and completed the laboratory analyses. Nilton Nobuhiro Imai is Fernanda Watanabe’s
advisor. He helped organize both field surveys and reviewed the article. Cláudio Barbosa gave ideas
for this article and reviewed it. Luiz Rotta helped conduct the field campaigns and helped with the
laboratory analyses.

Conflicts of Interest

The authors declare no conflict of interest.

References

1. Liu, Y.; Islam, M.A.; Gao, J. Quantification of shallow waters quality parameters by means of
remote sensing. Prog. Phys. Geogr. 2003, 27, 111–117.
2. Calijuri, M.C.; Santos, A.C.A.; Jati, S. Temporal changes in the phytoplankton community
structure in a tropical and eutrophic reservoir (Barra Bonita, S.P.–Brazil). J. Plankton Res. 2002,
24, 617–634.
3. Smith, V.H. Eutrophication of freshwater and coastal marine ecosystems: a global problem.
Environ. Sci. Pollut. 2003, 10, 126–139.
4. Bennett, E.M.; Carpenter, S.R.; Caraco, N.F. Human impact on erodable phosphorus and
eutrophication: a global perspective. BioScience 2001, 51, 227–234.
5. Cai, W.J.; Hu, X.; Huang, W.J.; Murrel, M.C.; Lehrter, J.C.; Lohrens, S.E.; Chou, W.C.; Zhai, W.;
Hollibaugh, J.T.; Wang, Y.; et al. Acidification of subsurface coastal waters enhanced by
eutrophication. Nat. Geosci. 2011, 4, 766–770.
6. Rabalais, N.N.; Turner, R.E.; Díaz, R.J.; Justic, D. Global change and eutrophication of coastal
waters. ICES J. Marine Sci. 2009, 66, 1528–1537.
Int. J. Environ. Res. Public Health 2015, 12 10413

7. Smith, V.H. Cultural eutrophication of inland, estuarine and coastal waters. In Successes,
Limitations and Frontiers in Ecosystems Science; Pace, M.L., Groffman, P.M., Eds.;
Springer-Verlag: New York, NY, USA, 1998; pp. 7–49.
8. Blain, S.; Quéguiner, B.; Armand, L.; Belviso, S.; Bombled, B.; Bopp, L.; Bowie, A.; Brunet, C.;
Brussaard, C.; Carlotti, F.; et al. Effect of natural iron fertilization on carbon sequestration in the
Southern Ocean. Nature 2007, 446, 1070–1075.
9. Boyd, P.W.; Jickells, T.; Law, C.S.; Blain, S.; Boyle, E.A.; Buesseler, K.O.; Coale, K.H.;
Cullen, J.J.; Baar, H.J.W.; Follows, M. et al. Mesoscale iron enrichment experiments 1993–2005
synthesis and future directions. Science 2007, 315, 612–617.
10. Sayre, R. Microalgae: The potential for carbon capture. BioScience 2010, 60, 722–272.
11. Giles, J. Methane quashes green credentials of hydropower. Nature 2006, 444, 524–525.
12. Abe, D.S.; Matsumura-Tundisi, T.; Rocha, O.; Tundisi, J.G. Denitrification and bacterial
community structure in the cascade of six reservoirs on a tropical river in Brazil. Hydrological
2003, 504, 67–76.
13. Roland, F.; Vidal, L.O.; Pacheco, F.S.; Barros, N.O.; Assireu, A.; Ometto, J.P.H.B.;
Cimbleris, A.C.P.; Cole, J.J. Variability of carbon dioxide flux from tropical (Cerrado)
hydroeletric reservoirs. Aquat. Sci. 2010, 72, 283–293.
14. Kemenes, A.; Forsberg, B.R.; Melack, J.M. CO2 emissions from a tropical hydroeletric reservoir
(Balbina, Brazil). J. Geophys. Res. 2011, 116, doi:10.1029/2010JG001465.
15. Barros, N.; Cole, J.J.; Tranvik, L.J.; Prairie, Y.T.; Bastviken, D.; Huszar, V.L.M.; Giorgio, P.;
Roland, F. Carbon emission from hydroelectric reservoirs linked to reservoir age and latitude.
Nat. Geosci. 2011, 4, 593–596.
16. Weaver, E.C.; Wrigley, R. Factors Affecting the Identification of Phytoplankton Groups by Means
of Remote Sensing; NASA Ames Research Center: Moffett Field, CA, USA, 1994.
17. Pearson, L.; Mihali, T.; Moffit, M.; Kellmann, R.; Neilan, B. On the chemistry, toxicology and
genetics of the cyanobacterial toxins, microcystin, nodularin, saxitoxin and cylindrospermopsin.
Mar. Drugs. 2010, 8, 1650–1680.
18. Neilan, B.A.; Dittmann, E.; Rouhiainen, L.; Bass, R.A.; Schaub, V.; Sivonen, K.; Börner, T.
Nonribosomal peptide synthesis and tosigenicity of cyanobacteria. J. Bacteriol. 1999, 181,
4089–4097.
19. McGregor, G.B.; Stewart, I.; Sendall, B.C.; Sadler, R.; Reardon, K.; Carter, S.; Wruck, D.;
Wickramasinghe, W. First report of a toxic Nodularia spumigena (Nostocales/Cyanobateria)
blooms in sub-tropical Australia. I. Phycological and public health investigations. Int. J. Environ.
Res. Public Health 2012, 9, 2396–2411.
20. Ju, J.; Saul, N.; Kochan, C.; Putschew, A.; Pu, Y.; Yiu, L.; Steinberg, C.E.W. Cyanobaterial
xenobiotcs as evaluated by a Caenorhabditis elegans neurotoxicity screening test. Int. J. Environ.
Res. Public Health. 2014, 11, 4589–4606.
21. Carlson, R.E. A trophic state index for lakes. Limnol. Oceanogr. 1977, 22, 361–369.
22. Walker, W.W. Predicting lake water quality. In The lake and reservoir restoration guidance
manual, 1st ed.; Lynn, M., Kent, T., eds.; EPA: Washington, DC, USA, 1998;
pp. 1–23.
Int. J. Environ. Res. Public Health 2015, 12 10414

23. Lamparelli, M.C. Graus de trofia em corpos d’água do Estado de São Paulo. Ph.D. Thesis,
University of São Paulo, State of São Paulo, Brazil, 3 Sepetmber 2004.
24. Palmer, S.C.J.; Kutser, T.; Hunter, P.D. Remote sensing of inland waters: Challenges, progress
and future directions. Remote Sens. Environ. 2015, 157, 1–8.
25. Koponen, S.; Pullainen, J.; Kallio, K.; Hallikainen, M. Lake water quality classification with
airborne hyperspectral spectrometer and simulated MERIS data. Remote Sens. Environ. 2002, 79,
51–59.
26. Kirk, J.T.O. Light and Photosynthesis in Aquatic Ecosystems, 3rd ed.; Cambridge University
Press: Cambridge, UK, 2011.
27. Bukata, R.P.; Jerome, J.H.; Kondratyev, K.Y.; Pozdnyakov, D.V. Optical Properties and Remote
Sensing of Inland and Coastal Waters; CRC Press: Boca Raton, FL, USA, 1995.
28. Carder, K.L.; Chen, F.R.; Lee, Z.P.; Hawes, S.K.; Kamykowski, D. Semianalytic
Moderate-Resolution Imaging Spectrometer algorithms for chlorophyll a and absorption with
bio-optical domains based on nitrate-depletion temperatures. J. Geophys. Res. 1999, 104, 5403–5421.
29. Brando, V.E.; Dekker, A.G. Satellite hyperspectral remote sensing for estimating estuarine and
coastal water quality. IEEE Trans. Geosci. Remote Sens. 2003, 41, 1378–1387.
30. Doeffer, R.; Schiller, H. The MERIS Case 2 water algorithm. Int. J. Remote Sens. 2007, 28,
517–535.
31. Gitelson, A.A.; Schalles, J.F.; Hladik, C.M. Remote chlorophyll-a retrieval in turbid, productive
estuaries: Chesapeake Bay case study. Remote Sens. Environ. 2007, 109, 464–472.
32. Mishra, S.; Mishra, D.R. Normalized difference chlorophyll index: A novel model for remote
estimation of chlorophyll-a concentration in turbid productive waters. Remote Sens. Environ.
2012, 117, 394–406.
33. Huang, C.; Zou, J.; Li, Y.; Yang, H.; Shi, K.; Li, J.; Wang, Y.; Chena, X.; Zheng F. Assessment of
NIR-red algorithms for observation of chlorophyll-a in highly turbid inland waters in China.
ISPRS J. Photogramm. 2014, 93, 29–39.
34. Matsushita, B.; Yang, W.; Yu, G.; Oyama, Y.; Yoshimura, K.; Fukushima, T. A hybrid algorithm
for estimating the chlorophyll-a concentration across different trophic states in Asian inland
waters. ISPRS J. Photogramm. Remote Sens. 2015, 102, 28–37.
35. Mayo, M.; Gitelson, A.; Yacobi, Y.Z.; Ben-Avraham, Z. Chlorophyll distribution in Lake
Kinneret determined from Landsat Thematic Mapper data. Int. J. Remote Sens. 1995, 16,
175–182.
36. Yacobi, Y.Z.; Gitelson, A.; Mayo, M. Remote sensing of chlorophyll in Lale Kinneret using higj
spectral-resolution radiometer and Landsat TM: Spectral features of reflectance and algorithm
development. J. Plankton Res. 1995, 117, 2155–2173.
37. Vincent, R.K.; Qin, X.; McKay, M.L.; Miner,, J.; Czajkowski, K.; Savino, J.; Bridgeman, T.
Phycocyanin detection from LANDSAT TM data for mapping cyanobacterial blooms in Lake
Erie. Remote. Sens. Environ. 2004, 89, 381–392.
38. Sun, D.; Hu, C.; Qiu, Z.; Shi, K. Estimating phycocyanin pigment concentration in productive
inland waters using Landsat measurements: A case study in Lake Dianchi. Opt. Express. 2015, 23,
3055–3074.
Int. J. Environ. Res. Public Health 2015, 12 10415

39. Thiemann, S.; Kaufmann, H. Determination of chlorophyll content and trophic state of lakes using
field spectrometer and IRS-1C satellite data in the Mecklenburg Lake District, Germany.
Remote Sens. Environ. 2000, 73, 227–235.
40. Pahlevan, N.; Lee, Z-P.; Wei, J.; Schaaf, C.B.; Schott, J.R.; Berk, A. On-orbit radiometric
characterization of OLI (Landsat-8) for applications in aquatic remote sensing. Remote Sens.
Environ. 2014, 154, 272–284.
41. Vanhellemont, Q.; Ruddick, K. Turbid wakes associated with offshore wind turbines observed
with Landsat 8. Remote Sens. Environ. 2014, 145, 105–115.
42. Lobo, F.L.; Costa, A.P.F.; Novo, E.M.L.M. Time-series analysis of Landsat-MSS/TM/OLI
images over Amazonian waters impacted by gold mining activities. Remote Sens. Environ. 2015,
157, 170–184.
43. CETESB (Companhia Ambiental do Estado de São Paulo). Águas superficiais. CETESB: São
Paulo, Brasil. Available online: http://www.cetesb.sp.gov.br/agua/aguas-superficiais/
35-publicacoes-/-relatorios (accessed on 23 April 2015).
44. Tundisi, J.G.; Matsumura-Tundisi, T.; Abe, D.S. The ecological dynamics of Barra Bonita (Tietê
River, SP, Brazil) reservoir: Implications for its biodiversity. Brazilian J. Biol. 2008, 68,
1079–1098.
45. AES Tietê. Barra Bonita. 2013. Available online: http://www.aestiete.com.br/usinas/Paginas/
BarraBonita.aspx\ (accessed on 19 May 2015).
46. Matsumura-Tundisi, T.; Tundisi, J.G. Plankton richness in a eutrophic reservoir (Barra Bonita
reservoir, SP, Brazil). Hydrobiologia 2005, 542, 367–378.
47. Calijuri, M.C.; Santos, A.C.A. Short-term changes in the Barra Bonita reservoir (São Paulo,
Brazil): Emphasis on the phytoplankton communities. Hydrobiologia 1996, 330, 163–175.
48. Dellamano-Oliveira, M.J.; Vieira, A.A.H.; Rocha, O.; Colombo, V. Phytoplankton taxonomic
composition and temporal changes in a tropical reservoir. Funda. Appl. Limnol. 2008, 171, 27–38.
49. Rodrigues, T.W.P.; Guimarães, U.S.; Rotta, L.H.S.; Watanabe, F.S.Y.; Alcântara, E.; Imai, N.N.
Delineamento amostral em reservatórios utilizando imagens Landsat-8/OLI: um estudo de caso no
reservatório de Nova Avanhandava (Estado de São Paulo, Brasil). Boletim de Ciências
Geodésicas 2015, submitted.
50. msda_xe—Analysis and Control Software. Available online: http://trios-science.com/
index.php?option=com_content&view=article&id=104&catid=65&Itemid=91&lang=en (accessed
on 26 June 2015).
51. Mobley, C.D. Estimation of the remote-sensing reflectance from above-surface measurements.
Appl. Opt. 1999, 38, 7442–7455.
52. Mueller, J.L. In-water radiometric profile measurements and data analysis protocols. In Ocean
Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 4, Volume III:
Radiometric Measurements and Data Analysis Protocols; Muller, J.L., Fargion, G.S.,
McClain, C.R., Eds.; NASA Goddard Space Flight Space Center: Greenbelt, MD, USA, 2003;
pp. 7–20.
53. Golterman, H.L. Developments in Water Science 2. In Physiological Limnology: An Approach to
the Physiology of Lake Ecosystems; Elsevier: Amsterdam, The Netherlands, 1975.
Int. J. Environ. Res. Public Health 2015, 12 10416

54. APHA. Standard Methods for the Examination of Water and Wastewater, 20th ed.; American
Public Health Association (APHA), American Water Works Association (AWWA), Water
Environmental Federation (WEF): Washintong, DC, USA, 1998.
55. Cleveland, J.S.; Weidemann, A.D. Quantifying absorption by aquatic particles: A multiple
scattering correction for glass-fiber filters. Limnol. Oceanogr. 1993, 38, 1321–1327.
56. Tassan, S.; Ferrari, G.M. Measurement of light absorption by aquatic particles retained on filters:
Determination of the optical pathlength amplification by the “transmittance-reflectance” method.
J. Plankton Res. 1998, 20, 1699–1709.
57. Bricaud, A.; Morel, A.; Prieur, L. Absorption by dissolved organic matter of the sea (yellow
substance) in the UV and visible domains. Limnol. Oceanogr. 1981, 26, 43–53.
58. Frequently Asked Questions About the Landsat Missions. Available online:
landsat.usgs.gov/best_spectral_bands_to_use.php (accessed on 23 April 2015).
59. Atmospheric Correction Module: QUAC and FLAASH User’s Guide. Available online:
https://www.exelisvis.com/portals/0/pdfs/envi/Flaash_Module.pdf (accessd on 26 June 2015).
60. Aerosols and Climate Change. Available online: http://earthobservatory.nasa.gov/Features/
Aerosols/what_are_aerosols_1999.pdf (accessed on 29 July 2015).
61. Mishra, D.R.; Narumalani, S.; Rundquist, D.; Lawson, M. Characterizing the vertical diffuse
attenuation coefficient for downwelling irradiance in coastal waters: Implications for water
penetration by high resolution satellite data. ISPRS J. Photogramm. 2005, 60, 48–64.
62. Gordon, H.R.; Brown, J.W.; Evans, R.H. Exact Rayleigh scattering calculations for use with the
Nimbus-7 Coastal Zone Color Scanner. Appl. Opt. 1988, 27, 862–871.
63. Ruddick, K.G.; Ovidio, F.; Rijkeboer, M. Atmospheric correction of SeaWiFS imagery for turbid
coastal and inland waters. Appl. Opt. 2000, 39, 897–912.
64. Vanhellemont, Q.; Ruddick, K. Advantages of high quality SWIR bands for ocean colour
processing: Examples from Landsat-8. Remote Sens. Environ.2015, 161, 89–106.
65. Moses, J.M.; Gitelson, A.A.; Perk, R.L.; Gurlin, D.; Rundquist, D.C.; Leavitt, B.C.; Barrow, T.M.;
Brakhage, P. Estimation of chlorophyll-a concentration in turbid productive waters using airborne
hyperspectral data. Water Res. 2012, 46, 993–1004.
66. Moses, W.J.; Bowles, J.H.; Corson, M.R. Expected improvements in the quantitative remote
sensing of optically complex waters with the use of an optically fast hyperspectral
spectrometer—A modeling study. Sensors 2015, 15, 6152–6173.
67. Gitelson, A.A.; Gritz, Y.; Merzlyak, M.N. Relationships between leaf chlorophyll content and
spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant
leaves. J. Plant Physiol. 2003, 160, 271–282.
68. Barsi, J.A.; Lee, K.; Kvaran, G.; Markham, B.L.; Pedelyy, J. The spectral response of the Landsat-8
Operational Land Imager. Remote Sens. 2014, 6, 1023 –1025.
69. Ogashawara, I.; Curtarelli, M.P.; Souza, A.F.; Augusto-Silva, P.B.; Alcântara, E.H.; Stech, J.L.
Interactive Correlation Environment (ICE)—A statistical web tool for data collinearity analysis.
Remote Sens. 2014, 6, 3059–3074.
70. Moses, W.J.; Gitelson, A.A.; Berdnikov, S.; Povazhnyy, V. Satellite estimation of chlorophyll-a
concentration using the red and NIR bands of MERIS–the Azov Sea case study. IEEE Geosci.
Remote S. 2009, 6, 845–849.
Int. J. Environ. Res. Public Health 2015, 12 10417

71. Gilerson, A.A.; Gitelson, A.A.; Zhou, J.; Gurlin, D.; Moses, W.; Ioannou, I.; Ahmed, S.A.
Algorithms for remote estimation of chlorophyll-a in coastal and inland waters using red and near
infrared bands. Opt. Express 2010, 18, 109–125.
72. Gurlin, D.; Gitelson, A.A.; Moses, W.J. Remote sensing of chl-a concentration in turbid
productive waters—Return to a simple two-band NIR-red model? Remote Sens. Environ. 2011,
115, 3479–3490.
73. Rundquist, D.C.; Han, L.; Schalles, J.F.; Peake, J.S. Remote measurement of algal chlorophyll in
surface waters: The case for the first derivative of reflectance near 690 nm. Photogramm. Eng.
Remote Sensing 1996, 62, 195–200.
74. Detection of Optical Water Quality Parameter for Eutrophic Water by High Resolution Remote
Sensing. Available online: http://dspace.ubvu.vu.nl/handle/1871/12714 (accessed on 26 June
2015).
75. Soares, A.; Mozeto, A.A. Water quality in the Tietê river reservoirs (Billings, Barra Bonita, Bariri
and Promissão, SP-Brazil) and nutrient fluxes across the sediment-water interface (Barra Bonita).
Acta Limnol. Bras. 2006, 18, 247–266.
76. Carlson, R.E. More complications in the chlorophyll-secchi disk relationship. Limnol. Oceanogr.
1980, 25, 379–382.
77. Lorenzen, M.W. Use of chlorophyll-Secchi disk relationships. Limnol. Oceanogr. 1980, 25, 371–372.
78. Megard, R.O.; Settles, J.C.; Boyer, H.A. Light, Secchi disk, and trophic states. Limnol. Oceanogr.
1980, 25, 373–377.
79. Mannino, A.; Russ, M.E.; Hooker, S.B. Algorithms development and validation for satellite-derived
distributions of DOC and CDOM in the U.S. Middle Atlantic Bight. J. Geophys. Res. 2008, 13,
doi:10.1029/2007JC004493.

© 2015 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article
distributed under the terms and conditions of the Creative Commons Attribution license
(http://creativecommons.org/licenses/by/4.0/).