FROM OLIGO TO EUTROPHIC INLAND WATERS: ADVANCEMENTS AND 

CHALLENGES FOR BIO-OPTICAL MODELING 

 

 

 

 

 

 

 

 

 

 

Presidente Prudente 

2017  

UNIVERSIDADE ESTADUAL PAULISTA “JÚLIO DE 

MESQUITA FILHO” 

CAMPUS DE PRESIDENTE PRUDENTE 

FACULDADE DE CIÊNCIAS E TECNOLOGIA 

Programa de Pós-Graduação em Ciências Cartográficas 



THANAN WALESZA PEQUENO RODRIGUES 

 

 

 

 

 

 

 

 

 

 

FROM OLIGO TO EUTROPHIC INLAND WATERS: ADVANCEMENTS AND 

CHALLENGES FOR BIO-OPTICAL MODELING 

 

 

 

 

 

 

 

 

 

 

Presidente Prudente 

2017  

Thesis presented to the Department of 

Cartography of the Faculty of Science and 

Technology of São Paulo State University 

as part of the requirements for obtaining 

the PhD degree in Cartographic Sciences. 

 

Advisor: Prof. Dr. Nilton Nobuhiro Imai  

Co-Advisor: Prof. Dr. Enner Herenio de 

Alcântara 



 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

FICHA CATALOGRÁFICA 
 

  

Rodrigues, Thanan Walesza Pequeno. 

R617f From oligo to eutrophic inland waters: advancements and challenges for 

bio-optical modeling / Thanan Walesza Pequeno Rodrigues. - Presidente 

Prudente : [s.n.], 2017 

 145 f. : il. 

  

 Orientador: Nilton Nobuhiro Imai 

Coorientador: Enner Herenio de Alcântara 

 Tese (doutorado) - Universidade Estadual Paulista, Faculdade de 

Ciências e Tecnologia 

 Inclui bibliografia 

  

 1. Modelagem bio-óptica. 2. Águas interiores. 3. Sensoriamento remoto 

da água. 4. Sistema de reservatórios em cascata. I. Rodrigues, Thanan 

Walesza Pequeno. II. Imai, Nilton Nobuhiro. III. Alcântara, Enner Herenio 

de. IV. Universidade Estadual Paulista. Faculdade de Ciências e Tecnologia. 

V. Título. 

  

  



  



 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

To God. 

To my beloved parents Waldete 

and Salatiel 

To my beloved husband Ulisses 

To my dear sisters and brother 

Suzan, Rejane and Walesson. 



AGRADECIMENTOS 

 

Agradeço antes de tudo à Deus por ter me concedido grandes oportunidades e me dado 

forças para sempre prosseguir. A caminhada durante o doutorado foi árdua e sem ele nada teria 

valido a pena. Nessa jornada contei com ajuda e colaboração de muitas pessoas e a elas eu 

expresso meus sinceros agradecimentos em especial: 

A meus pais que tanto amo Waldete e Salatiel por me fazerem chegar onde estou hoje 

sem medirem esforços, agradeço a dedicação incondicional, amor e ensinamentos passados ao 

longo de toda minha vida. 

Aos meus irmãos Waleska, Walesson e Rejane pelo companheirismo, amizade e amor. 

A meu marido e amigo Ulisses que sempre esteve presente na minha caminhada 

acadêmica e profissional me incentivando e apoiando as decisões mais difíceis. 

Aos meus tios e primos queridos: Cris, Walmir, Waldemir, Heloisa, Amandinha, Adonis 

e Ágatha. 

Ao meu orientador Dr. Nilton Imai pelas contribuições e por ter possibilitado meu 

ingresso no Programa de Pós-Graduação da Unesp. 

Ao meu co-orientador Dr. Enner Alcântara, profissional que realmente se dedica à 

pesquisa e ciência, que me incentivou desde o princípio a escrever e acreditar no meu potencial 

como pesquisadora. 

Ao Dr. Deepak Mishra da Universidade da Geórgia, Estados Unidos, pelo exemplo de 

pesquisador, e que me ajudou, durante o período sanduíche, com discussões sobre a tese e 

artigos e que me inseriu no seu grupo de pesquisa em Athens. Agradeço ainda o Departamento 

de Geografia da UGA por ter concedido ajuda de custo para participar de um evento na cidade 

de São Francisco. 

Agradeço a todos os meus colegas e grandes parceiros do grupo de pesquisa em 

Sensoriamento Remoto da água, em especial: Fernanda, Luiz, Nariane, Alisson, Bruno, Stela, 

Carol Ambrósio e todos aqueles que direta ou indiretamente fizeram com que as análises e 

campos ocorressem. Agradeço ainda a todos os meus colegas de Pós-Graduação que me 

ajudaram a sobreviver no primeiro ano de doutorado seja em sala de aula ou jogando vôlei de 

areia ou ainda futebol. 

Agradeço ao Dr. Cláudio Clemente do INPE pelos equipamentos emprestados nos 

primeiros campos. Agradeço ainda o Renato e Daniel pela ajuda nos campos e nas análises 

laboratoriais. 



Ao Dr. Edivaldo Velini e técnicos da Unesp Botucatu, por nos deixar utilizar os 

equipamentos para análises laboratoriais. 

A todos os professores do PPGCC por todo ensinamento e discussões, em especial: as 

Professoras Lourdes, Ivana e Vilma pelas contribuições na tese e em artigos escritos com 

colaboração. 

Aos meus colegas Ike e Kumar por ter tornado minha vida mais fácil nos EUA, e por 

terem me ajudado com discussões relacionados aos artigos. Agradeço ainda a Dipanwita e 

Shuvo pela gentileza e por terem me apresentado a cultura indiana. Agradeço também ao 

Benjamin pelas discussões relacionadas a programação ou mesmo com o idioma inglês. 

Agradeço a CAPES pela concessão da bolsa de doutorado (Processo n° 1193768), ao 

CNPq pela concessão da bolsa sanduíche (Processo nº 200152/2015-7 - SWE) e por financiar 

projetos (Processos nº 472131/2012-5 e 400881/2013-6) e à FAPESP pelos projetos (Processos 

n° 2012/19821-1 e 2015/21586-9). 

Por fim agradeço à Unesp e o PPGCC pela infraestrutura e suporte para estudo. 

Agradeço ainda os funcionários da Unesp, em especial a Cinthia, André e Ivonete da Secretaria 

da Pós-Graduação. 

Obrigada a todos! 

  



 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

“We must have perseverance and 

above all confidence in ourselves. 

We must believe that we are gifted 

for something and that this thing 

must be attained.” 

(Marie Curie) 



RESUMO 

 

O presente trabalho teve como objetivo realizar um levantamento detalhado das características 

bio-ópticas nos reservatórios de Barra Bonita (BB) e Nova Avanhandava (Nav) com o intuito 

de avaliar o desempenho de uma única abordagem voltada para a estimativa das propriedades 

ópticas inerentes (POIs), assim como, a concentração de totais sólidos suspensos (TSS). A 

investigação foi realizada utilizando dados coletados no campo entre 2014 e 2016, incluindo, 

as POIs, componentes opticamente significativos (COSs) e reflectância de sensoriamento 

remoto (𝑅𝑟𝑠). Os dados apresentados dos COSs confirmaram que BB é um ambiente mais 

túrbido que Nav por apresentar maior produção fitoplanctônica em função do recebimento de 

altas cargas de nutrientes provenientes da bacia de drenagem. Por outro lado, Nav é um 

ambiente mais transparente e com maior influência de material inorgânico, o que favorece o 

surgimento de macrófitas submersas. A concentração de clorofila-a (Chl-a) em BB alcançou 

máximo de 797.8 µg l-1 em outubro/2014, enquanto Nav apresentou máximo de 38.6 µg l-1 em 

maio/2016. A variabilidade nos COS esteve altamente vinculada a frequência de chuvas, sendo 

que no ano de 2014, ocorreu um evento extremo de seca alterando as características 

biogeoquímicas dos ambientes. BB reagiu de forma mais abrupta que Nav por apresentar um 

sistema de operação do tipo acumulação e por estar mais próxima das regiões potencialmente 

poluidoras, diferente de Nav que apresenta um sistema fio-d’água em que não há acumulação 

e sim fluxo constante da água. Além disso, no âmbito óptico, a absorção em Nav apresentou 

maior influência do particulado não-algal (NAP) enquanto que em BB, a absorção foi dominada 

por fitoplâncton. Com base nesses resultados pode-se concluir que os dois ambientes 

apresentam não só diferenças na qualidade da água, mas também nas propriedades ópticas, o 

que leva a afirmação de que um modelo único baseado nos dois ambientes pode não ter um 

bom resultado quando se pretende utilizar uma abordagem empírica. Um algoritmo quase-

analítico (QAA) parametrizado para as condições de Nav (QAAOMR) apresentou resultados 

significativos com erros (erro médio percentual absoluto – MAPE) inferiores a 17% para o 

coeficiente de absorção total (𝑎𝑡), 19% para o coeficiente de absorção orgânico detrital (𝑎𝐶𝐷𝑀) 

e 47% para o coeficiente de absorção do fitoplâncton (𝑎𝜙). O respectivo modelo foi utilizado 

para verificar seu desempenho em um ambiente eutrofizado como BB e a versão parametrizada 

por Watanabe et al. (2016) e denominada QAABBHR foi aplicada aos dados de Nav. Como 

resultado, observamos que as duas versões foram adequadas para estimar 𝑎𝑡 com erros 

inferiores a 40%, no entanto, existe ainda a necessidade de melhorar as etapas para estimativa 

de 𝑎𝐶𝐷𝑀 e 𝑎𝜙. No caso de se aplicar um modelo empírico de única abordagem para estimar 

concentração de TSS para ambos os reservatórios, observamos que essa abordagem não 

apresentou resultados satisfatórios, portanto, modelos específicos baseados na banda do 

vermelho do MODIS foram utilizados para mapear TSS em cada um dos reservatórios. Pode-

se concluir então, que o conhecimento acerca das propriedades ópticas da água se mostrou 

determinante para a modelagem bio-óptica, principalmente no que diz respeito aos ambientes 

altamente contrastantes como BB e Nav. 

 

Palavras-chave: modelagem bio-óptica, águas interiores, sensoriamento remoto da água, 

sistema de reservatórios em cascata 

 

 

  



ABSTRACT 

 

The objective of the present work was to perform a detailed survey of the bio-optical 

characteristics of the reservoirs of Barra Bonita (BB) and Nova Avanhandava (Nav) in order to 

evaluate the performance of a single approach aimed at estimating the inherent optical 

properties (IOPs), as well as the concentration of total suspended solids (TSS). The research 

was carried out using data collected in the field between 2014 and 2016, including the IOPs, 

optically significant components (OSCs) and remote sensing reflectance (𝑅𝑟𝑠). The data 

presented from the OSCs confirmed that BB is more turbid than Nav because it presents higher 

phytoplankton production due to the input of high nutrient loads from the drainage basin. On 

the other hand, Nav is more transparent with greater influence of inorganic matter, which favors 

the appearance of submerged macrophytes. The concentration of chlorophyll-a (Chl-a) in BB 

reached a maximum of 797.8 μg l-1 in October/2014, while Nav presented a maximum of 38.6 

μg l-1 in May/2016. The variability in the COS was highly related to the frequency of rainfall, 

in the year 2014, an extreme drought event occurred, altering the biogeochemical 

characteristics. BB reacted more abruptly than Nav because it presented an accumulation type 

operation system and because it is closer to the potentially polluting region. Nav presents a 

water system in which there is no accumulation but constant flow of water. In addition, in the 

optical context, the absorption in Nav presented greater influence of the non-algal particulate 

(NAP) while in BB, the absorption was dominated by phytoplankton. Based on these results, it 

can be concluded that the two environments present not only differences in water quality but 

also in optical properties, which leads to the assertion that a single model based on the two 

environments may not have a good result when it is intended to use empirical approach. A 

quasi-analytical algorithm (QAA) parameterized for Nav conditions (QAAOMR) presented 

significant results with errors (mean absolute percentage error - MAPE) lower than 17% for the 

total absorption coefficient (𝑎𝑡), 19% for the carbon detrital matter absorption coefficient 

(𝑎𝐶𝐷𝑀) and 47% for the absorption coefficient of phytoplankton (𝑎𝜙). The respective model 

was used to verify its performance in a eutrophic environment such as BB and the version 

parameterized by Watanabe et al. (2016) and named QAABBHR was applied to the Nav data. 

Thus, we note that the two versions were suitable for estimating 𝑎𝑡 with errors (MAPE) less 

than 40%, however, improvements must be carried out for estimating 𝑎𝐶𝐷𝑀 and 𝑎𝜙. In the case 

of applying a single empirical model to estimate TSS concentration for both reservoirs, we 

observed that it did not present satisfactory results, so specific models based on the MODIS red 

band were used to map TSS in each of the reservoirs. It can be concluded, therefore, that 

knowledge about the optical properties of water has proved to be determinant for the bio-optical 

modeling, especially with respect to highly contrasting environments such as BB and Nav. 

 

Keywords: bio-optical modeling, inland waters, remote sensing of water, reservoirs in cascade 

system 

 

  



LIST OF FIGURES 

 

Figure 2.1. Graphic representation of the study area emphasizing (a) Brazil’s territory, (b) Tietê River 

located at São Paulo State and the location of the reservoirs from upstream to downstream: Barra Bonita, 

Bariri, Ibitinga, Promissão, Nova Avanhandava and Três Irmãos, (c) sampling location of Nav and (d) 

BB. ........................................................................................................................................................ 25 

Figure 2.2. Flowchart showing the methodological scheme for sampling stations definition. 𝜌𝑇𝑂𝐴 

represents the value of reflectance at the top of atmosphere (TOA), SD stands for the standard deviation, 

NDWI is the Normalized Difference Water Index, and PCA is the Principal Component Analysis. The 

box 1 refers to the multispectral images; box 2 stands for the reservoir’s delimitation and box 3, stratified 

sampling. ............................................................................................................................................... 27 

Figure 2.3. Rainfall data from the period between 2011 to 2016 (boxplots) highlighting the years of 

2014, 2015 and 2016 for (a) Nav and (b) BB. Nav1 (28 April – 2 May/2014), Nav2 (23 – 26 

September/2014) and Nav3 (9 and 14 May/2016). ............................................................................... 30 

Figure 2.4. Water level data from the period between 2011 to 2016 (boxplots) highlighting the years of 

2014, 2015 and 2016 for (a) Nav and (b) BB. Different axis y was used due to distinct magnitudes. Nav1 

(28 April – 2 May/2014), Nav2 (23 – 26 September/2014) and Nav3 (9 and 14 May/2016). .............. 31 

Figure 2.5. Water flow data from the period between 2011 to 2016 (boxplots) highlighting the years of 

2014, 2015 and 2016 for (a) Nav and (b) BB. Nav1 (28 April – 2 May/2014), Nav2 (23 – 26 

September/2014) and Nav3 (9 and 14 May/2016). ............................................................................... 32 

Figure 3.1. Relationship between water quality and optical parameters considering data from all field 

campaigns. (a) TSS versus Chl-a, (b) 𝑎𝜙(443) versus Chl-a, (c) 𝑎𝑁𝐴𝑃(443) versus TSS, (d) 

𝑎𝐶𝐷𝑂𝑀(443) versus TSS and (e) 𝑎𝐶𝐷𝑂𝑀(443) versus Chl-a. ............................................................... 41 

Figure 3.2. Ternary plot depicting the absorption coefficient budget of both Nav and BB reservoirs at 

three wavelengths: (a) 443 nm, (b) 560 nm and (c) 665 nm. ................................................................ 42 

Figure 3.3. Variability of 𝑎𝐶𝐷𝑂𝑀by field trip. (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 and (f) the 

average value of 𝑎𝐶𝐷𝑂𝑀(𝜆) for each field trip. Different y axes were applied for Nav and BB due to 

magnitude discrepancies. ...................................................................................................................... 43 

Figure 3.4. Variability of 𝑎𝜙by field trip. (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 and (f) the 

average value of 𝑎𝜙(𝜆) for each field trip. Different y axes were applied for Nav and BB due to 

magnitude discrepancies. ...................................................................................................................... 44 

Figure 3.5. Variability of 𝑎𝑁𝐴𝑃(𝜆) in all field trips (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 and 

(f) the average value of 𝑎𝑁𝐴𝑃(𝜆) for each field trip. ............................................................................. 45 

Figure 3.6. Variability of 𝑎𝑝(𝜆) in all field trips.  (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 and 

(f) the average value of 𝑎𝑝(𝜆) for each field trip. Different y axles were applied for Nav and BB due to 

magnitude discrepancies. ...................................................................................................................... 46 

Figure 3.7. (a) Variation of 𝑎𝜙(𝜆)/𝑎𝜙(443) ratio as a function of wavelength considering the average 

values of BB and Nav and the mass-specific absorption of phytoplankton, 𝑎𝜙
∗(𝜆), for all field 

campaigns (b) Nav1, (c) Nav2, (d) Nav3, (e) BB1 and (f) BB2. .......................................................... 49 

Figure 3.8. (a) Variation of 𝑎𝑁𝐴𝑃(𝜆)/𝑎𝑁𝐴𝑃(443) ratio as a function of wavelength considering the 

averages values of BB and Nav and (b) the relationship between 𝑆𝑁𝐴𝑃and OSS (mg l-1) from BB. .... 51 

Figure 4.1. Reciprocal remote sensing reflectance (𝑅𝑟𝑠
−1) data from the (a) first, (b) second and (c) third 

field trips. Pure water absorption (𝑎𝑤) (red line) is shown for reference. ............................................ 63 



Figure 4.2. Ternary plots displaying the relative contribution of CDOM, phytoplankton and detritus to 

the total absorption at different wavelengths, (a) 443 nm, (b) 560 nm, (c) 665 nm and (d) the relative 

contribution of water, particulate and CDOM to absorption at 709 nm. ............................................... 64 

Figure 4.3. Relationship between estimated and measured 𝑎𝑡(𝜆) using existing QAAs: (a) QAALv5, (b) 

QAALv6 and (c) QAAM14. The colored circles represents the band centers of OLCI sensor. ................ 66 

Figure 4.4. Scatter plot between measured and estimated (a) 𝑎𝑡(𝜆), (b) 𝑎𝐶𝐷𝑀(𝜆) and (c) 𝑎𝜙(𝜆) at OLCI 

spectral bands. ....................................................................................................................................... 68 

Figure 4.5. Taylor diagrams for 𝑎𝑡(𝜆) at (a) 443 nm, (c) 560 nm, (e) 665 nm. Target diagrams for 𝑎𝑡(𝜆) 

at (b) 443 nm, (d) 560 nm and (f) 665 nm. Color symbols indicate the following: QAALv5 (grey dot), 

QAALv6 (yellow dot), QAAM14 (blue dot) and the QAAOMR (green dot), reference observation (red dot). 

The black circle in the Target diagrams (M0) corresponds to a normalized total RMSD of 1.0, so all 

points between this circle and the origin are positively correlated. ...................................................... 70 

Figure 4.6. Taylor diagrams for 𝑎𝐶𝐷𝑀(𝜆) at (a) 443 nm, (c) 560 nm, (e) 665 nm. Target diagrams for 

𝑎𝐶𝐷𝑀(𝜆) at (b) 443 nm, (d) 560 nm and (f) 665 nm. Symbols indicate the following: QAALv5 (grey dot), 

QAALv6 (yellow dot), QAAM14 (blue dot) and the QAAOMR (green dot), reference observation (red dot). 

The black circle in the Target diagrams (M0) corresponds to a normalized total RMSD of 1.0. .......... 73 

Figure 4.7. Taylor diagrams for 𝑎𝜙(𝜆) at (a) 443 nm, (c) 560 nm, (e) 665 nm. Target diagrams also for 

𝑎𝜙(𝜆) at (b) 443 nm, (d) 560 nm and (f) 665 nm. Symbols indicate the following: QAALv5 (grey dot), 

QAALv6 (yellow dot), QAAM14 (blue dot) and the QAAOMR (green dot), reference observation (red dot).

 ............................................................................................................................................................... 76 

Figure 4.8. Validation result showing the scatter plot between measured and estimated (a) 𝑎𝑡(𝜆), (b) 

𝑎𝐶𝐷𝑀(𝜆), (c) 𝑎𝜙(𝜆) and the comparison of spectral shape of average measured and estimated (d) 𝑎𝑡(𝜆), 

(e) 𝑎𝐶𝐷𝑀(𝜆) and (f) 𝑎𝜙(𝜆). ................................................................................................................... 77 

Figure 4.9. Graphic depicting the (a) rainfall, (b) runoff, (c) water level and (d) discharge variability 

along the period between October 2013 and August 2016. The green diamonds represent at(443) 

retrieved based on QAAOMR from the three field trips carried out in Nav during April/2014, 

September/2014 and May/2016. Rainfall data was acquired in NASA’s GIOVANNI database based on 

TRMM data with 0.25° of spatial resolution (http://giovanni.sci.gsfc.nasa.gov/giovanni/). Runoff data 

was also acquired through NASA’s GIOVANNI database. Water level as well as discharge were 

downloaded from the Water National Agency of Brazil through the website (http://sar.ana.gov.br/). . 80 

Figure 4.10. Relationship between optical properties and water quality parameters: (a) 𝑎𝑡(443) versus 

TSS; (b) 𝑎𝐶𝐷𝑀(443) versus TSS; (c) 𝑎𝜙(443) versus TSS; (d) 𝑎𝑡(443) versus Chl-a; (e) 𝑎𝐶𝐷𝑀(443) 

versus Chl-a and (f) 𝑎𝜙(443) versus Chl-a. ......................................................................................... 82 

Figure 5.1. Scatter plots in log-scale between measured and estimated 𝑎𝑡(𝜆), 𝑎𝐶𝐷𝑀(𝜆) and 𝑎𝜙(𝜆) at 

OLCI spectral bands. ............................................................................................................................. 89 

Figure 5.2. Spectral shape acquired through (a) QAABBHR and (b) Nav in situ 𝑎𝜙 acquired from 

laboratory analysis. The continuous red line stands for the average spectral 𝑎𝜙 from Nav in situ data.

 ............................................................................................................................................................... 90 

Figure 5.3. Scatter plots in log-scale between measured and estimated 𝑎𝑡(𝜆), 𝑎𝐶𝐷𝑀(𝜆) and 𝑎𝜙(𝜆) at 

OLCI spectral bands. ............................................................................................................................. 91 

Figure 5.4. Spectral shape acquired through (a) QAAOMR and (b) in situ 𝑎𝜙 from BB acquired from 

laboratory analysis. The continuous red line stands for the average spectral 𝑎𝜙 from BB in situ data. 93 

Figure 6.1. Schematic diagram showing the (a) viewing angle (𝜃𝑣) of the sensors to avoid specular 

scattering (Mobley, 1999) (b) Geometry of the sensor relative to the sun used for radiometric 



measurements, represented by zenith (𝜃𝑠), azimuthal (𝜙) and nadir (𝑛) angles, (c) sensors collecting total 

upwelling radiance 𝐿𝑡(𝜆), the incident sky radiance 𝐿𝑠(𝜆) and the downwelling irradiance 𝐸𝑑(𝜆), (d) 

sensors collecting upwelling radiance 𝐿𝑢(𝜆), upwelling irradiance 𝐸𝑢(𝜆) and the downwelling 

irradiance 𝐸𝑑(𝜆). ................................................................................................................................... 99 

Figure 6.2. Cumulative percentage of TSM concentration and the respective frequency of the calibration 

dataset from BB (a) and Nav (b) reservoirs. ....................................................................................... 104 

Figure 6.3. Relationship between TSM (mg l-1) and Chl-a (mg m-3) in (a) BB and (b) Nav. ............. 106 

Figure 6.4. Ternary graphs depicting the absorption budget at three wavelengths in the visible region (a) 

blue – 443 nm, (b) green – 555 nm and (c) red – 645 nm. .................................................................. 107 

Figure 6.5. 𝑅𝑟𝑠(𝜆)𝑓𝑖𝑒𝑙𝑑 spectra representing different TSM concentrations overlapped with MODIS 

Terra spectral response function for bands at (A) 469 nm, (B) 555 nm and (D) 645 nm. .................. 108 

Figure 6.6. Correlation between the original 500 m MODIS B3 and the downscaled 250 m MODIS B3 

for (a) May and (b) October, 2014. ..................................................................................................... 109 

Figure 6.7. Spatio-temporal distribution of TSM over the main body of Nav based on MODIS 8-day 

composite images for months of May and October from 2000 to 2015. ............................................. 114 

Figure 6.8. Spatio-temporal distribution of TSM over the main body of BB based on MODIS 8-day 

composite images for the months of May and October from 2000 to 2015. The blank region represent 

the locations where the model extrapolated the TSM values over 150 m l-1. ...................................... 115 

Figure 6.9. TSM variability during the (a) beginning of the dry season and (b) end of the dry season in 

Nav; (c) beginning of the dry season and (d) end of the dry season in BB at different locations along the 

reservoir. SBS: Santa Barbara Stream, BR: Bonito River, PR: Piracicaba River, TR: Tietê River, TZ1: 

Tranzition Zone 1, TZ2: Transition Zone 2. (e) TSM (mg l-1) concentration from field campaigns carried 

out in Tietê River (Barra Bonita – BB, Bariri – B, Ibitinga – I, Promissão – P, Nova Avanhandava – 

Nav) except for Três Irmãos (TI) reservoir in 2000 (CAVENAGHI et al., 2003) (solid lines); including 

all reservoirs along the Tietê River during the years 2001 and 2002 (ZANATA, 2005) (dotted lines) and 

in the lower Tietê basin during 2008 and 2009 (dashed lines) (SANTOS, 2010). .............................. 118 

 

  



LIST OF TABLES 

 

Table 3.1. Descriptive statistics of water quality and optical parameters. The notations Min-

Max, Aver, SD, CV and n stand for minimum-maximum, average, standard deviation, 

coefficient of determination and number of samples. .............................................................. 39 

Table 4.1. QAA steps comparing the version 5 from Lee et al. (2002) and the QAAOMR proposed 

in this study. .............................................................................................................................. 58 

Table 4.2. Descriptive statistic of the water quality variables used for calibration and validation. 

SD: standard deviation, CV: coefficient of variation and n is the number of samples. ........... 61 

Table 4.3. Comparative band-specific errors related to the re-parametrization of 𝑎𝑡(𝜆) based on 

MAPE (%) and RSMD (m-1) metrics. ...................................................................................... 68 

Table 4.4. Comparative band-specific errors related to the re-parameterization of 𝑎𝐶𝐷𝑀(𝜆) 

based on MAPE (%) and RSMD (m-1) metrics. ....................................................................... 72 

Table 4.5. Comparative band-specific errors related to the re-parametrization of 𝑎𝜙(𝜆) based 

on MAPE (%) and RSMD (m-1) metrics. ................................................................................. 75 

Table 5.1 QAA steps comparing the QAABBHR from Watanabe et al. (2016) and the QAAOMR 

proposed in the last chapter. ..................................................................................................... 87 

Table 5.2. Comparative band-specific errors related to the IOPs retrieved by QAABBHR based 

on MAPE (%), RSMD (m-1) and bias (m-1) metrics. ................................................................ 89 

Table 5.3. Comparative band-specific errors related to the IOPs retrieved by QAAOMR based on 

MAPE (%), RSMD (m-1) and bias (m-1) metrics. ..................................................................... 92 

Table 6.1. TSM empirical models: model fit, TSM concentration range (mg l-1), and geographic 

location. .................................................................................................................................. 102 

Table 6.2. Descriptive statistics of environmental dataset in both reservoirs. ....................... 105 

Table 6.3. Calibration results of the models using the single and two band indexes. Models with 

R² below 0.50 were not displayed here. ................................................................................. 110 

 

 

  



LIST OF ABBREVIATIONS AND ACRONYMS 

 

𝜌𝑇𝑂𝐴 Reflectance at the top of atmosphere 

𝜁 𝑎𝜙(412)/𝑎𝜙(443) 

𝜉 𝑎𝐶𝐷𝑀(412)/𝑎𝐶𝐷𝑀(443) 

𝜂 Spectral power for backscattering coefficient 

𝜆0 Reference wavelength 

𝛿13𝐶 Delta thirteen carbon 
σ* Normalized standard deviation 

ap Total particulate absorption coefficient 

a or at Total absorption coefficient 

𝑎𝑤 Absorption coefficient of pure water 

𝑎𝑁𝐴𝑃 Absorption coefficient of non-algal particle 

𝑎𝑑 Absorption coefficient of detritus 

𝑎𝐶𝐷𝑂𝑀 Absorption coefficient of colored dissolved organic matter 

𝑎𝐶𝐷𝑀 Absorption coefficient of CDOM and detritus 

𝑎𝜙 Absorption coefficient of phytoplankton 

𝑎𝑡−𝑤 Total absorption budget without the water fraction 

𝑎𝜙
∗  Specific absorption coefficient of phytoplankton 

𝑎𝜙
+ Normalized spectral absorption of phytoplankton 

𝑏𝑏 Backscattering coefficient 

𝑏𝑏𝑝 Particulate backscattering coefficient 

𝑏𝑏𝑤 Pure water backscattering coefficient 

B* Normalized bias 

CDOM Colored dissolved organic matter 

Chl-a Chlorophyll-a 

𝐸𝑢 Upwelling irradiance 

𝐸𝑑 Downwelling irradiance 

IOP Inherent optical properties 

ISS Inorganic suspended solids 

𝐿𝑢 Upwelling radiance 

𝐸𝑑(0+) Downwelling irradiance incident onto the water surface 

𝐿𝑠𝑘𝑦 Incident sky radiance 

𝐿𝑡 Total upwelling radiance 
LULC Land use and land cover 
𝑙 Path length 

MAPE Mean absolute percentage error 

MODIS Moderate Resolution Imaging Spectroradiometer 

NAP Non-algal particle 

NDWI Normalized Difference Water Index 

OSC Optically significant components 

OLCI Ocean and Land Colour Instrument 

OSS Organic suspended solids 

OD Optical density 

ODsample Optical density of the sample 

ODreference Optical density of the reference 

QAA Quasi-analytical algorithm 

𝑅 Irradiance reflectance 



𝑅𝑟𝑠 Remote sensing reflectance 

𝑟𝑟𝑠 Subsurface remote sensing reflectance 

𝑅𝑟𝑠𝑠𝑎𝑡
 Water surface remote sensing reflectance 

𝑅𝑟𝑠𝑠𝑖𝑚
 Simulated bands from MODIS/Terra 

𝑅𝑟𝑠𝑓𝑖𝑒𝑙𝑑
 In situ remote sensing reflectance 

𝑅𝑟𝑠
−1 Reciprocal remote sensing reflectance 

RMSD Total root mean square difference 

RMSE Total root mean square error 

𝑆𝐶𝐷𝑀 Spectral slope of colored detrital matter absorption coefficient 

𝑆𝐶𝐷𝑂𝑀 Spectral slope of colored dissolved organic matter 

𝑆𝑁𝐴𝑃 Spectral slope of non-algal particle 

SIOP Mass-specific inherent optical properties 

SRF Spectral response function 

TSS Total suspended sediment or solids 

TSM Total suspended matter 

𝑢 Ratio of backscattering coefficient to the sum of backscattering and 

absorption coefficients 

uRMSD* Normalized unbiased RMSD 

 

  



CONTENTS 

 

CHAPTER 1: INTRODUCTION ......................................................................................... 19 

1.1 Background ................................................................................................................... 19 

1.2 Motivation ...................................................................................................................... 20 

1.3 Hypothesis ...................................................................................................................... 22 

1.4 Objectives ...................................................................................................................... 23 

1.5 Outline of the Thesis ..................................................................................................... 23 

CHAPTER 2: STUDY AREA AND FIELD CAMPAIGNS ............................................... 24 

2.1 General characteristics ................................................................................................. 24 

2.2 Strategy for sampling design ....................................................................................... 26 

2.3 Field campaigns ............................................................................................................. 29 

2.4 Field data acquisition .................................................................................................... 32 

CHAPTER 3: ABSORPTION PROPERTIES OF TWO OPTICALLY DIFFERENT 

RESERVOIRS SITUATED ALONG THE CASCADE SYSTEM OF TIETÊ RIVER .. 34 

3.1 Introduction ................................................................................................................... 34 

3.2 Data and Methods ......................................................................................................... 36 

3.2.1 Water quality parameters ......................................................................................... 36 

3.2.2 Inherent optical properties ........................................................................................ 37 

3.3 Results ............................................................................................................................ 38 

3.3.1 General characteristics of water quality parameters and optical properties ............. 38 

3.3.2 Absorption coefficient budget .................................................................................. 41 

3.3.3 CDOM absorption .................................................................................................... 42 

3.3.4 Phytoplankton absorption ......................................................................................... 43 

3.3.5 NAP absorption ........................................................................................................ 44 

3.3.6 Particle absorption .................................................................................................... 45 

3.4. Discussion ..................................................................................................................... 46 

3.5 Conclusion ..................................................................................................................... 51 

CHAPTER 4: RETRIEVAL OF INHERENT OPTICAL PROPERTIES FROM 

OLIGO-TO-MESOTROPHIC INLAND WATER USING THE QUASI-ANALYTICAL 

ALGORITHM ........................................................................................................................ 53 

4.1 Introduction ................................................................................................................... 53 

4.2 Data and methods ......................................................................................................... 55 

4.2.1 Water quality parameters ......................................................................................... 55 

4.2.2 In-Situ Radiometric Data .......................................................................................... 56 

4.2.3 In Situ Inherent Optical Properties ........................................................................... 57 



4.2.4 QAA General Context .............................................................................................. 58 

4.2.5 Re-parametrization /Validation and Accuracy Assessment ..................................... 59 

4.3 Results and Discussion .................................................................................................. 61 

4.3.1 Biogeochemical characterization ............................................................................. 61 

4.3.2 Bio-optical characterization ..................................................................................... 62 

4.3.3 OSC relative contribution ........................................................................................ 63 

4.3.4 Performance of existing QAA .................................................................................. 65 

4.3.5 Re-parametrization of QAA to derive 𝑎𝑡 ................................................................. 66 

4.3.6 Re-parametrization of QAA to derive 𝑎𝐶𝐷𝑀 ............................................................ 70 

4.3.7 Re-parametrization of QAA to derive 𝑎𝜙 ................................................................ 73 

4.3.8 Model Validation...................................................................................................... 76 

4.3.9 Linking IOPs variability to physical and meteorological variation ......................... 78 

4.3.10 Factors influencing optical changes in the reservoir .............................................. 81 

4.3.11 Implications of QAAOMR for water resource management .................................... 83 

4.4 Conclusion ..................................................................................................................... 83 

CHAPTER 5: EVALUATION OF QAAOMR AND QAABBHR PERFORMANCES IN 

DERIVING THE IOPS IN BOTH BB AND NAV RESERVOIRS ................................... 85 

5.1 Introduction ................................................................................................................... 85 

5.2 Data and Methods ......................................................................................................... 86 

5.3 Results and Discussion .................................................................................................. 88 

5.3.1 Use of QAABBHR in Nav’s dataset ............................................................................ 88 

5.4.2 Use of QAAOMR in BB’s dataset .............................................................................. 91 

5.4. Conclusion .................................................................................................................... 93 

CHAPTER 6: LONG-TERM MONITORING OF TOTAL SUSPENDED MATTER IN 

TROPICAL RESERVOIRS WITHIN A CASCADE SYSTEM WITH WIDELY 

DIFFERING OPTICAL PROPERTIES .............................................................................. 95 

6.1 Introduction ................................................................................................................... 95 

6.2 Data and Methods ......................................................................................................... 98 

6.2.1 Field data .................................................................................................................. 98 

6.2.2 Remote sensing reflectance (𝑅𝑟𝑠) ............................................................................. 98 

6.2.3 Inherent Optical Properties (IOPs) ......................................................................... 100 

6.2.4 Satellite Data and Processing ................................................................................. 101 

6.2.5 Bio-optical model calibration and validation ......................................................... 102 

6.3 Results .......................................................................................................................... 103 

6.3.1 Long-term spatio-temporal monitoring .................................................................. 104 

6.3.2 Water quality characterization ............................................................................... 104 



6.3.3 Bio-optical properties description .......................................................................... 106 

6.3.4 Downscaling MODIS image procedure ................................................................. 108 

6.3.5 TSM algorithm calibration and validation ............................................................. 109 

6.3.6 TSM spatio-temporal variability ............................................................................ 113 

6.3.7 Factors affecting optical changes in the reservoirs ................................................ 119 

6.4 Conclusion ................................................................................................................... 120 

CHAPTER 7: CONCLUSION AND FUTURE RECOMMENDATIONS ..................... 122 

7.1 Conclusion ................................................................................................................... 122 

7.2 Future Recommendations .......................................................................................... 124 

REFERENCES ..................................................................................................................... 126 



 

 

Rodrigues, T.W.P. 

19 

CHAPTER 1: INTRODUCTION 

 

1.1 Background 

 

The reservoirs of the Tietê River placed in the State of São Paulo, are exposed by a 

series of pollution sources such as sugar cane plantation, pastures and urban centers from which 

a large amount of pollutants reaches the aquatic system. On the one hand, the construction of 

these reservoirs supports the agriculture and industrial development of the region; on the other 

hand, it generates a series of negative impacts including intensive deforestation, eutrophication 

due to high loads of nutrients and suspended sediment from various activities, sedimentation 

and contamination of water bodies (BARBOSA et al. 1999). 

According to Fracácio et al. (2002), all the reservoirs of the Tietê River were classified 

as eutrophic for the parameter chlorophyll-a (Chl-a); for total phosphorus (TP), the reservoirs 

of Barra Bonita (BB), Bariri (B) and Ibitinga (Ib) were considered mesotrophic; and the 

Promissão (Pr) and Nova Avanhandava (Nav), oligotrophic. Differences in the mass of water 

composition between BB e Nav were reported and the first reservoir was attested to be more 

polluted than Nav. BB also has the highest concentration of suspended solids, thus higher 

turbidity when compared to Nav (CAVENAGHI et al. 2003). These results clearly show the 

existence of a trophic gradient between the cascade reservoirs. 

Although some studies have shown that the cascade reservoirs have a trophic and 

spectral gradient (WACHHOLZ et al., 2009; PEREIRA FILHO et al. 2009), just recently 

studies have investigated the cascade system effects on optical properties (ALCÂNTARA et 

al., 2016, RODRIGUES et al., 2016a). In terms of water resource management, the traditional 

methods for water monitoring are time consuming, expensive and demands in situ collection. 

Besides, these techniques present low spatial and temporal representativeness, therefore, the 

optical properties of the in-water constituents (Chl-a, non-algal-particle – NAP, total suspended 

sediment – TSS, colored dissolved organic matter – CDOM) and here known as optical 

significant components (OSCs), can be used in bio-optical models in order to estimate the 

concentrations of the OSCs and support the water resource management considering large areas 

and short time mapping. 

 

 

 



 

 

Rodrigues, T.W.P. 

20 

1.2 Motivation 

 

Environments such as reservoirs designed in cascade system cause water quality 

modifications from the upstream to downstream reducing the turbidity and increasing the 

transparency of water (BARBOSA et al. 1999). The longitudinal variability is not the only 

factor influencing the water quality dynamic, but the watershed controls several mechanisms 

of reservoirs functioning. Watersheds subjected to extensive use by human activities can 

produce high loads of nutrients, suspended matter or even toxic substances to the reservoir 

(JORGENSEN et al., 2012). 

Built to attend single or more specific purpose, the reservoirs have multiple uses and 

therefore, demands for complex strategies of management. The desired phases for management 

systems include the integration of geographic information system and remote sensing in order 

to provide a rapid or a large-scale comprehension of reservoirs (TUNDISI and MATSUMURA-

TUNDISI, 2011). Remote sensing is also important for offering spatial and temporal views of 

water quality surface parameters, which is limited from in situ collections. Its applicability 

becomes useful when the water degradation is caused by OSC, such as the green algae pigments 

as Chl-a, TSS, CDOM and NAP, thus producing detectable effects by optical remote sensing 

instruments (GIARDINO et al., 2010). 

The spectral region where these components can be detected is limited to a narrow range 

of optical wavelengths, in general, restricted between 400 to 850 nm (DEKKER, 1993). In 

inland waters, NAP and CDOM absorb at shorter wavelengths showing an exponential decay 

toward longer wavelengths, while Chl-a presents absorption peaks in the blue and red 

wavelengths (ROESLER et al., 1989; BRICAUD et al., 1981). However, in waters with low 

phytoplankton concentration the absorption peak at the blue region is not detected (WU et al., 

2011), nevertheless, significant contributions of Chl-a lead to the enhancement of the 

absorption peak at ~675 nm allowing its detection on the remote sensing reflectance (𝑅𝑟𝑠) signal 

in very productive inland waters (DOXARAN et al., 2006).  

Waters dominated by sediment, especially NAP and pure water showed satisfactory 

results with the empirical relationships 𝑅𝑟𝑠(850)/𝑅𝑟𝑠(550) and 𝑅𝑟𝑠(850)/𝑅𝑟𝑠(650) for TSS 

retrieval in Gironde estuary (DOXARAN et al., 2006). In waters comprised by mineral 

suspended sediment (MSS), the red-to-green ratio showed to be applicable, however, with 

increasing influence of phytoplankton and CDOM this relationship was found to break down, 

assuming the limited potential of the model (BINDING et al., 2003). In the case of CDOM 

absorption estimation, D’Sa and Miller (2003) showed that the ratio between two visible bands 



 

 

Rodrigues, T.W.P. 

21 

𝑅𝑟𝑠(510)/𝑅𝑟𝑠(555) was suitable for the turbid waters of Mississippi River, while Doxaran et 

al. (2006) found the ratio 𝑅𝑟𝑠(400)/𝑅𝑟𝑠(700) more appropriate for Tamar estuary. For Chl-a 

retrieval, Dall’Olmo et al. (2003, 2005) showed that a three-band model based on NIR-Red 

wavelengths produce more accurate values in turbid waters, however, they also noticed that 

both Chl-a fluorescence and Chl-a mass-specific absorption coefficient (𝑎𝐶ℎ𝑙𝑎
∗ ) variability can 

introduce more uncertainties in the estimative. 

In other words, the knowledge about the optical water properties can improve the 

formulation of models for OSC concentration retrieval, however, algorithms based on empirical 

assumptions often fail when applied to other study sites (RITCHIE et al., 2003). Besides, these 

models are also limited for a specific range of OSC concentration. Kumar et al. (2016) for 

example, found a simple correlation between MODIS 𝑅𝑟𝑠(645 𝑛𝑚) and TSS concentration in 

Chilika Lagoon, India. Their model showed to be limited to the range between 6.5 and 200 mg 

l-1, with increasing error in very low TSS (< 6.54 mg l-1). Ogashawara et al. (2013) evaluated 

the influence of Chl-a absorption on the performance of several empirical algorithms using data 

from environments with very low and very high cyanobacteria. As result, they verified a 

decrease of model’s sensitivity in high phycocyanin concentration environments. Chen et al. 

(2015) studied a wide range of TSS (58 – 577.2 mg l-1) in estuary and coastal waters using an 

improved model based on the log-ratio (𝑅𝑟𝑠(𝑁𝐼𝑅)/𝑅𝑟𝑠(𝑅𝑒𝑑)), however, they had to consider 

a constraint based on TSS concentration below and above 31 mg l-1. 

In general, the empirical methods work well for turbid inland waters, however, in 

oligotrophic environments the performance decrease considerably using the same band 

combination. Gons et al. (2008) for example, showed that a red-to-NIR band Ch-a algorithm 

proved to be applicable for eutrophic to hypereutrophic waters, however the accuracy dropped 

considerably for oligotrophic waters. They also reported that remote sensing of Chl-a is limited 

in low concentration range and not the contrary. The alternative for low concentrations would 

be to use the blue-to-green band ratio (GONS and AUER, 2004; GONS et al., 2008).  

Aiming to address the empirical issue, semi-analytical approaches use the IOP and 

apparent optical properties (AOP) to model the reflectance and vice versa. These properties are 

then used in analytical methods to retrieve the water constituents. The main limitation regards 

the insufficient knowledge about the IOPs used in the equations (MOREL and GORDON, 1980; 

DEKKER, 1993). The example of a more robust model is the quasi-analytical algorithm (QAA) 

developed to derive the absorption and backscattering coefficients in open ocean and coastal 

waters (LEE et al., 2002). After that, several initiatives were carried out aiming to fit this model 



 

 

Rodrigues, T.W.P. 

22 

to productive inland waters (LE et al., 2009a; YANG et al., 2013, MISHRA et al., 2014; 

WATANABE et al., 2016a). A version based on CDOM dominated waters was also developed 

(OGASHAWARA et al., 2016), however, a model adapted to inorganic matter dominated water 

was not addresses yet. A semi-analytical algorithm using the QAA was successfully applied for 

retrieving the water clarity based on new theoretical model to interpret the Secchi disk depth, 

𝑍𝑆𝐷 (LEE et al., 2015). For validation, the authors used data covering oceanic, coastal and 

inland waters (lake) and as result they got an average absolute difference of ~18%, highlighting 

the accomplishment of a model using a wide range of data. 

As previously mentioned, the comprehension about the bio-optical properties of 

different environments helps to indicate the suitable approach for water quality parameters 

estimation. Many efforts have shown the success of empirical approaches in deriving the water 

optical properties in very turbid inland waters, however, when data from two widely trophic 

state (oligo to eutrophic) reservoirs is supposed to be mixed, we expect the models to perform 

poorly, indicating the limitation of a single approach in retrieving the OSCs. Up to now, 

incipient initiatives showed the influence of cascade system in the optical properties, which 

means that a lot of efforts still need to be done. 

 

1.3 Hypothesis 

 

In a cascading arrangement where the water flows with different composition through 

the system, the bio-optical modeling can be challenging. As we learned from the literature such 

systems range from hyper-to-oligotrophic, from organic-to-inorganic dominated waters, and 

consequently the development of an algorithm to map the water optical properties needs to deal 

with such contrast. Moreover, the literature showed that empirical models are constrained by 

the local optical properties and the range of OSC concentration, while semi-analytical 

approaches tend to fail where the phytoplankton is not the dominant component and where 

some empirical steps are calibrated using synthetic data. Thus, our hypothesis bases on the fact 

that due to the organic and inorganic nature of both BB and Nav reservoirs, the empirical 

approach won’t be able to map the OSC accurately using a single model, however, recalibrating 

the empirical steps using in situ data, the quasi-analytical algorithm will increase the retrieval 

of water optical properties and therefore will be able to map operationally the water quality 

from remotely sensed images in a cascade system. 

 



 

 

Rodrigues, T.W.P. 

23 

1.4 Objectives 

 

According to the thesis hypothesis, we aimed to investigate the water optical properties 

of both BB and Nav in order to evaluate the performance of a single model built for retrieving 

the inherent optical properties as well as total suspended sediment concentration of both 

reservoirs. For this, specific objectives were designed: 

 

- Characterize the absorption properties in both BB and Nav reservoirs; 

- Re-parameterize the QAA algorithm based on OLCI/Sentinel-3 bands using in situ data 

from Nav; validate the algorithm using an independent dataset collected from the same 

area in a different season; 

- Assess the performance of a single QAA version (QAABBHR from Watanabe et al. 2016 

and QAAOMR proposed here) highlighting the improvements and fragilities from each 

version supposing to choose a single approach for mapping the absorption properties in 

the entire cascade; 

- Assess the performance of empirical models for retrieving TSS concentration in both 

BB and Nav using single or separate models. 

 

1.5 Outline of the Thesis 

 

This thesis is organized in 7 chapters. Chapter 1 introduces the theme, highlighting the 

problem of the research followed by the questions we intend to answer and the objectives we 

expect to achieve. Chapter 2 describes the study areas focusing on their physical and 

environmental characteristics; description of the method designed to sampling definition and 

the field trips. Chapter 3 describes the IOPs of both reservoirs and their relationships with 

optical water parameters. In Chapter 4, a parametrization of QAAOMR was carried out using 

data from Nav and validate it with an independent dataset collected in a different season. 

Chapter 5 compares and evaluates the performance of both QAAOMR and QAABBHR in retrieving 

the IOPs using data from BB and Nav, respectively. In Chapter 6 empirical models were 

formulated aiming to retrieve TSS in both reservoirs. Lastly, Chapter 7 pointed out the main 

findings of the research and it makes some recommendations for future works. 

  



 

 

Rodrigues, T.W.P. 

24 

CHAPTER 2: STUDY AREA AND FIELD CAMPAIGNS 

 

2.1 General characteristics 

 

The reservoirs of Barra Bonita (BB) and Nova Avanhandava (Nav) (Figure 2.1) are 

situated in the middle and lower portion of the Tietê River, São Paulo State, respectively. BB 

(22°31′10″S, 48°32′3″W) is a storage system and began its operation in 1963 flooding an area 

of 310 km2, with a dam length of 480 m, 90.3 days of residence time, being formed from the 

damming of Tietê and Piracicaba Rivers. The regional climate transits between tropical and 

subtropical, and the annual seasons are not well marked. According to the Koppen 

classification, the climate is mesothermal type - CWA, with a dry winter and a hot summer 

(PRADO, 2004). Pastures and sugar cane monoculture predominantly comprised the land 

cover. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 

 

Rodrigues, T.W.P. 

25 

Figure 2.1. Graphic representation of the study area emphasizing (a) Brazil’s territory, (b) Tietê 

River located at São Paulo State and the location of the reservoirs from upstream to 

downstream: Barra Bonita, Bariri, Ibitinga, Promissão, Nova Avanhandava and Três Irmãos, 

(c) sampling location of Nav and (d) BB. 

 

 

On the other hand, Nav (21°7′1″S, 50°12′6″W) is a run-of-river reservoir and was 

created in 1982, flooding an area of 210 km² (at its maximum quota), with a dam length of 

2,038 m and mean residence time of the water around 46 days (TORLONI, et al., 1993). The 

reservoir is part of a region whose influence of Continental Tropical and Polar Antarctic air 

masses are marked. The first mass is hot and dry and occurs mainly in the summer (24 and 30° 



 

 

Rodrigues, T.W.P. 

26 

C), while the second is cold and damp, and, despite being-active all year, its occurrence is more 

intense in winter, causing a decrease of temperature (22 to 14° C) (CBH-BT, 1999).  

BB reservoir is an ecosystem characterized as polymictic and eutrophic, with high loads 

of nutrients, whose contribution leads to the development of blooms of cyanobacteria during 

the summer, and Bacillariophyceae during the winter (DELLAMANO-OLIVEIRA et al. 2007). 

The Piracicaba and Tietê Rivers, which along their courses are subject to the carrying of organic 

and inorganic origin waste, arising from agricultural, urban and industrial activities, affecting 

water quality. Nav is an oligo-to-mesotrophic reservoir with the upper layer of the water column 

well oxygenated and pH ranging from slightly acid to alkaline (6.47 – 8.2), conductivity 

relatively high (83 – 150 µS cm-1) and low concentrations of nutrients (Total N: 0.05 – 0.23 µg 

l-1 and Total P: 18.02 – 32.33 µg l-1) (RODGHER et al. 2005; SMITH et al. 2014). The high 

transparency of the water often leads to the growth of submerged macrophytes (e.g. Egeria sp. 

– Elodea) (SMITH et al. 2014), although, during sample collections we avoided those areas. 

The catchment basin surrounding the reservoir receives input from non-point source of 

pollution such as sugar cane and citric plantation (orange and lemon) and cattle breeding. 

There is evidence that the downstream reservoirs of Tietê River have improved water 

quality to those further upstream providing an enhancement of chemical constitution of water 

limiting the development of floating vegetation representative of a eutrophic environment 

(BARBOSA et al., 1999; CAVENAGHI et al. 2003). The cascade system generates significant 

modifications along the river changing aspects such as heterogeneity, connectivity and the 

coarse / fine particulate organic material. Moreover, this type of system influences the water 

quality, the composition and structure of phytoplankton community, and increases the 

eutrophication process in the upstream reservoirs (BARBOSA et al., 1999). 

 

2.2 Strategy for sampling design 

 

Based on the sampling design, a random stratified sampling method was used to 

determine the location for collecting water and optical data in BB and Nav (RODRIGUES et 

al., 2016b). Therefore, statistical and computational procedures were established aiming to 

provide parameters that subsidize the sampling design definition (Figure 2.2). For this purpose, 

the Operational Land Imager (OLI)/Landsat-8 imagery referring to an annual cycle was 

acquired at the USGS – U.S. Geological Survey website <http://earthexplorer.usgs.gov/> and 

the criterion of less cloud cover and absence of glint effect, visually detected, was applied.  



 

 

Rodrigues, T.W.P. 

27 

The images hosted at the USGS website are geometrically corrected and available for 

users in the orthorectified level (L1T) in which the reference data are terrain control points and 

altitude based on digital elevation model (DEM), such as the SRTM (Shuttle Radar Topography 

Mission) (USGS, 2014). This procedure assures the overlapping between images with 

geometric error below one pixel providing the high matching considering different periods. The 

methodological approach is displayed in Figure 2.2. 

 

Figure 2.2. Flowchart showing the methodological scheme for sampling stations definition. 

𝜌𝑇𝑂𝐴 represents the value of reflectance at the top of atmosphere (TOA), SD stands for the 

standard deviation, NDWI is the Normalized Difference Water Index, and PCA is the Principal 

Component Analysis. The box 1 refers to the multispectral images; box 2 stands for the 

reservoir’s delimitation and box 3, stratified sampling. 

 

 

A set of images available for a year was radiometrically calibrated. In this process, the 

digital numbers (DN) of each pixel are rescaled for radiance or reflectance at the top of the 

atmosphere (TOA), 𝜌𝑇𝑂𝐴, using gain and offset parameters provided by the metadata of the 

image (USGS, 2013). This procedure removes the effects caused by the difference of 



 

 

Rodrigues, T.W.P. 

28 

illumination geometry (COLLETT et al., 1998). Studies carried out in aquatic systems showed 

that TOA products yielded similar or better results from those atmospherically corrected data 

(OLMANSON et al., 2011, KUTSER, 2012).  

In view of the hydrological dynamics and other anthropic or natural interventions that 

the reservoirs undergo over time, it is important to consider the variations in the chemical and 

biological characteristics of the water body, with the input of nutrients from agricultural 

activities around the reservoir, and still insertion of organic and inorganic substances carried 

longitudinally along the entire cascade system. Therefore, two statistical metrics were 

considered to analyze the variations: mean and standard deviation (SD). After the radiometric 

correction process, a set of images referring to one year (2013 - 2014), each consisting of 6 

spectral bands (bands 2 to 7), were submitted to the calculation of the mean and subsequent 

standard deviation. Bands corresponding to the same wavelength were compressed and then the 

mean and standard deviation for the studied months were calculated, with each pixel of the final 

image having the mean value of spectral 𝜌𝑇𝑂𝐴, as well as standard deviation.  

For delimitation of the boundary of each reservoir a set of processes were established, 

such as the selection of a reference image, application of the NDWI, slice, conversion raster-

vector e finally, the creation of the mask. The image of reference was based on the season with 

low rainfall. In this case, the image from July/2013 was chosen. It is worth mentioning that Nav 

does not present a notable variation in water level due to its operational system, on the contrary, 

BB varies seasonally. Aiming to separate water from other targets, NDWI from McFeeters 

(1996) was applied as follows: 

 

𝑁𝐷𝑊𝐼 =
𝐺𝑟𝑒𝑛𝑛 − 𝑁𝐼𝑅

𝐺𝑟𝑒𝑒𝑛 + 𝑁𝐼𝑅
 (2.1) 

 

where Green and NIR are the bands 3 and 5 from OLI/Landsat-8, respectively. This ratio aims 

to maximize the water reflectance in the visible spectral region and minimize it in the infrared.  

There are modifications of this index, such as the MNDWI proposed by Xu (2006) who 

used a ratio between the green and middle infrared (MIR) bands aiming to enhanced the features 

of water in regions which the contamination by other targets are evident like noises related to 

built-up land, soil and vegetation overestimating the water targets. However, we kept the 

Equation 2.1, since the method employed here predicts the use of a buffer in order to avoid the 

borders with other targets but water. The separation of water from the other targets was obtained 

by slicing between values representative of each target. Thus, pixels with values assigned with 



 

 

Rodrigues, T.W.P. 

29 

water were allocated in a class while the others in another class, resulting in two classes: water 

and non-water.  

For stratified sampling, the SD images were transformed in Principal Component 

Analysis (PCA) in a Geographic Information System (GIS) environment and then sliced 

according to intervals empirically defined. The selected image for this step was based on the 

component with higher variability from the set of 6 SD bands. The slice step allowed the 

creation of stratus for further sampling points definition. A buffer with 70 m was created for 

each reservoir to prevent the edges formed by land. Using the tool Hawth’s Tools developed by 

Hawthorne Beyer (http://www.spatialecology.com/) and compatible to a GIS environment, the 

random stratified sampling was possible to be carried out. Initially, more than a required number 

of samples were created in order to remove the non-spatialized stations. The minimal distance 

of 1 km between samples was established to prevent clusters. As result, 20 samples were 

acquired for each field campaign except for the third campaign in BB that included 4 samples 

(totalizing 24 samples) and in Nav 20 samples in different locations were added for the first and 

second field trips and 19 for the third field trip. 

 

2.3 Field campaigns 

 

The water quality parameters, as well as their spatial distribution are seasonal 

dependent, therefore, we considered at least two periods of the year for water and optical data 

collection. The water samples were collected in six field campaigns (Figure 1). For Nav, the 

field trips occurred during austral autumn (Nav1: 28 April – 2 May/2014 and Nav2: 23 – 26 

September/2014) and austral spring (Nav3: 9 and 14 May/2016). For BB, the field trips were 

carried out in austral autumn (BB1: 5 – 9 May/2014) and austral spring (BB2: 13 – 16 

October/2014 and BB3: 13 – 15 September/2015). Both seasons are known for intermediate 

precipitation amounts between summer (wet period) and winter (dry period). Data collections 

were avoided in the summer due to high rainfall rates making difficult to obtain cloud free 

satellite images. 

As depicted in Figure 2.3, the rainfall data in BB and Nav showed distinct values. In 

Nav, for example, the month of January and February from 2011 to 2016 showed the highest 

averages values of 197.24 and 196.92 mm, respectively. BB, in turn, presented 222.59 and 

209.16 mm in January and February, respectively. These values were expected because they 

occurred in the wet period (summer). The year of 2014 showed the lowest values for the 

respective months presenting 130.75 and 141.81 mm in Nav and 168.51 and 146.24 mm in BB, 



 

 

Rodrigues, T.W.P. 

30 

respectively. The lowest average values were observed in July and August corresponding to the 

dry period (winter). Nav showed in July and August averages of 42.51 and 12.79 mm, 

respectively, whilst BB presented averages of 72.85 and 46.29 mm. In 2014, the lowest rainfall 

in Nav were observed in June (24.00 mm) and August (6.80 mm) while in BB, the lowest ones 

were seen in July (40.26 mm) and August (38.83 mm). The months chosen for the field works 

varied below and above the mean measures (2011 – 2016), but in the austral summer of 

2013/2014, the values were below the average and this happened due to an extreme event of 

drought classified as exceptionally dry (COELHO et al., 2015). The authors also described the 

2014/2015 summer as atypical and dry, however, in lower magnitude as the 2013/2014 summer. 

The months of April/May of 2014 represented the period of the first data collection in 

Nav and the rainfall data showed that in both months the values (98.41 and 69.00 mm, 

respectively) were below the average between 2011 and 2016 (99.67 and 95.05 mm). The 

month of September/2014 related to the second field collection presented value (141.40 mm) 

above the average (86.14 mm) and the same happened to the third field trip (May/2016) which 

showed a value of 197.49 mm above the average (95.05 mm). 

 

Figure 2.3. Rainfall data from the period between 2011 to 2016 (boxplots) highlighting the 

years of 2014, 2015 and 2016 for (a) Nav and (b) BB. Nav1 (28 April – 2 May/2014), Nav2 (23 

– 26 September/2014) and Nav3 (9 and 14 May/2016). 

 

Source: NASA/GIOVANNI (https://giovanni.gsfc.nasa.gov/giovanni/). 

 

The first field trip of BB carried out in May/2014 presented a value of 56.31 mm, below 

the average of 98.54 mm. The second field trip that occurred in October/2014, also exhibited a 

value (56.19 mm) below the average (164.14 mm). On the other hand, the third field trip 

(September/2015) displayed a value (192.08 mm) above the average (101.35 mm). The rainfall 

variability and the special event of dry in the austral summer (2013/2014) implied in severe 



 

 

Rodrigues, T.W.P. 

31 

impacts in water availability for public consumption, hydropower generation and agriculture 

(COELHO et al., 2015).  

Data of water level also showed the impact of the dry summer in 2014. A slight 

variability between 2011 to 2016 was observed in Nav (Figure 2.4), with an average interval of 

357.60 ± 0.30 to 357.77 ± 0.32 m from January to December. The year of 2014 displayed the 

lowest water level except in January that showed a value of 357.56 m. The water level fluctuated 

between 0.13 m in 2011 to 0.69 m in 2015, in 2014 the amplitude was 0.50 m. Nav is a run-of-

river reservoir, therefore, the water level variation with low amplitude is expected (PERBICHE-

NEVES et al., 2013). 

 

Figure 2.4. Water level data from the period between 2011 to 2016 (boxplots) highlighting the 

years of 2014, 2015 and 2016 for (a) Nav and (b) BB. Different axis y was used due to distinct 

magnitudes. Nav1 (28 April – 2 May/2014), Nav2 (23 – 26 September/2014) and Nav3 (9 and 

14 May/2016). 

 

Source: National Water Agency – ANA (http://sar.ana.gov.br/MedicaoSIN) 

 

The water level in BB (Figure 2.4b) varied from an average of 447.85 ± 1.21 m in 

December and 451.10 ± 0.51 m in June. The year of 2014 was also the one with water level 

excepted for the months of January and February that was lower in 2015. The amplitude ranged 

between 2.11 m in 2013 to 9.84 m in 2015. 2014 presented an amplitude of 3.85 along the year. 

The storage or accumulation reservoir of BB generates a higher variation in water level than 

the run-of-river system (PERBICHE-NEVES et al., 2013). 

The water flow (Figure 2.5) also showed to vary as a function of the dry event in 2014, 

showing the lowest values in almost the entire year, except for January of 2015. Regarding the 

magnitude, Nav and BB presented significant differences, the variability in Nav, for instance, 

ranged from 232.09 m3/s in October to 683.34 m3/s in December, while in BB, the range was 



 

 

Rodrigues, T.W.P. 

32 

between 113.00 m3/s in October and 361.50 m3/s in December. The water flow was 

considerably reduced in BB aiming to keep the water retained in the system during the dry 

events, on the other hand, in Nav, the regulation of water flow, theoretically, is not necessary 

due to the absence of storage area, so the water flows naturally according to the river regime 

(EGRÉ and MILEWSKI, 2002). In addition, the regulation of Nav showed to be influenced by 

BB because of the variation in phase observed at Figure 2.5, highlighting the co-oscillation 

between them. 

 

Figure 2.5. Water flow data from the period between 2011 to 2016 (boxplots) highlighting the 

years of 2014, 2015 and 2016 for (a) Nav and (b) BB. Nav1 (28 April – 2 May/2014), Nav2 (23 

– 26 September/2014) and Nav3 (9 and 14 May/2016). 

 

Source: National Water Agency – ANA (http://sar.ana.gov.br/MedicaoSIN) 

 

2.4 Field data acquisition 

 

For this study both water quality and optical data were acquired for each sampling 

station. The amount of water collected for BB and Nav was established according to the total 

particulate matter capable to be filter without saturate it. Therefore, a total of 5L of water was 

collected for both reservoirs, however, for each TSS and Chl-a concentrations 250 – 500 mL 

were filtered considering one filter/replica in BB and 750 – 1000 mL were used for Nav. The 

water samples were stored in polyethylene containers and refrigerated for laboratory analysis. 

Water transparency was measured by a Secchi disk (30 cm diameter), turbidity by a portable 

turbidimeter, model Hanna HI 93414, and dissolved oxygen by a portable meter, model Hanna 

HI 9146-04. 

The in situ spectroradiometric measurements are very important for bio-optical 

characterization and information extraction from remote sensing data, as they act as a bridge 



 

 

Rodrigues, T.W.P. 

33 

between optical measurements in laboratory and measures taken at the orbital or airborne level. 

This kind of measure permits the elimination of some undesirable effects such as atmospheric 

influence and the scale effect. Thus, radiometric measurements were carried out through 

RAMSES-ARC (radiance sensors) and RAMSES-ACC (irradiance sensors) 

spectroradiometers. The RAMSES instruments provide simultaneous measurements of 

upwelling (𝐸𝑢) and downwelling (𝐸𝑑) irradiance and upwelling radiance (𝐿𝑢) of the water 

column as well as measurements of downwelling irradiance incident onto the water surface 

(𝐸𝑑(0+)), incident sky radiance (𝐿𝑠𝑘𝑦) and total upwelling radiance (𝐿𝑡) above the surface of the 

water. 

The determination of the total particulate absorption coefficient (ap = NAP + 

phytoplankton) was accomplished by the use of an integrating sphere module present in the 

double-beam Shimadzu UV-2600 UV-Vis spectrophotometer, with spectral sampling from 280 

nm to 800 nm and spectral sampling of 1 nm. The pure water absorption coefficient is a constant 

obtained by Pope and Fry (1997). 

  



 

 

Rodrigues, T.W.P. 

34 

CHAPTER 3: ABSORPTION PROPERTIES OF TWO OPTICALLY DIFFERENT 

RESERVOIRS SITUATED ALONG THE CASCADE SYSTEM OF TIETÊ RIVER 

 

3.1 Introduction 

 

Optical properties of water are the basis of watercolor retrieval algorithms. Variations 

of these properties can change the model parameters, optimal spectral bands and the accuracy 

of retrieval algorithms. In order to improve the performance of bio-optical models (semi-

analytical approaches), a better understanding about the variability in the absorption (a, units 

in m-1) and backscattering (bb, units in m-1) coefficients is crucial. These coefficients influence 

the magnitude and the spectral distribution of the water-leaving reflectance. 

The a(λ) denotes the contribution of all components present in the aquatic system and 

is commonly represented into four additive elements, including the pure water (𝑎𝑤), NAP 

(𝑎𝑁𝐴𝑃), CDOM (𝑎𝐶𝐷𝑂𝑀) and phytoplankton (𝑎𝜙), all dependent on wavelength (BABIN et al., 

2003; BINDING et al., 2008). In case of pure water, the absorption starts to increase from green 

to near-infrared (NIR) wavelengths (SMITH and BAKER, 1981). CDOM can remove the blue 

light in the first layer of the water and exhibit an exponential decrease with increasing 

wavelength; however, the shape is not equal for all waters (DEKKER, 1993). In order to 

describe the spectral CDOM absorption, Bricaud et al. (1981) came up with the following 

model: 

 

𝑎𝐶𝐷𝑂𝑀(𝜆) = 𝑎𝐶𝐷𝑂𝑀(𝜆0)𝑒−𝑆𝐶𝐷𝑂𝑀(𝜆−𝜆0) (3.1) 

 

where 𝜆0 is a reference wavelength, 𝑎𝐶𝐷𝑂𝑀(𝜆0) is the absorption estimate at a reference 

wavelength, and 𝑆𝐶𝐷𝑂𝑀 is the spectral slope of the 𝑎𝐶𝐷𝑂𝑀(𝜆). The former provides insights 

about the characteristics of CDOM in terms of chemistry, source and diagenesis (HELMS et 

al., 2008) and is also a proxy for CDOM composition such as the ratio of fulvic to humic acids 

and molecular weight (SHANMUGAM et al., 2011). Besides, it is strongly dependent on the 

wavelength interval chosen for analysis (ROESLER et al., 1989; LOISELLE et al., 2009). The 

NAP also presents an exponential decrease from short to longer wavelengths and is represented 

by the spectral slope, 𝑆𝑁𝐴𝑃, which is related to the fraction of organic and inorganic matter 

(BABIN et al., 2003). Meanwhile, phytoplankton presents two distinct absorption features at 

443 and 675 nm. Its composition is clearly defined by the water quality and its abundance is 

essentially determined by the underwater light and nutrient loading (DEKKER, 1993). 



 

 

Rodrigues, T.W.P. 

35 

The characterization of IOP in ocean and coastal waters have been massively studied in 

the last four decades, therefore, the comprehension about the relationship between in-water 

constituents and optical properties are well known (MOREL and MARITORENA, 2001; 

BABIN et al., 2003). In open ocean, for example, the influence of terrigenous matter is 

negligible and is assumed that only phytoplankton domain the optical properties as well as the 

pure water (MOREL and BRICAUD, 1986). In contrast, the inland waters are considered 

complex environments due to the increased number of color-producing agents and the high 

variability in their concentrations, limiting the use of empirical oceanic algorithms for Chl-a 

retrieval, for example (BUKATA, 2005). More robust models such as the semi-analytical and 

analytical approaches require the knowledge about the mass-specific inherent optical properties 

(SIOPs), like those used by Lee et al. (2002) with the QAA and the Garver-Siegel-Maritorena 

algorithm (GARVER and SIEGEL, 1997; MARITORENA et al., 2002). The SIOPs are 

retrieved by normalizing the absorption coefficient of a certain OSC and the respective 

concentration. The mass-specific phytoplankton absorption (𝑎𝜙
∗ ), for example, is related to 

phytoplankton cell size, accessory pigments and package effect (BRICAUD et al., 1995; 

BABIN et al., 2003).  

Registers of the (S)IOPs in inland waters are available for specific regions around the 

world such as in Babin et al. (2003) who studied the coastal waters around Europe; Binding et 

al. (2008) produced a data set of absorption coefficients in Lake Erie; Perkins et al. (2009) 

documented the spectral features, magnitudes and variability of the IOPs in Finger Lakes of 

New York; Le et al. (2009b) investigated the variations of absorption and total specific 

absorption coefficient of phytoplankton in Lake Taihu, China; Campbell et al. (2011) described 

the SIOPs of three sub-tropical and tropical waters reservoirs in Australia; Wu et al. (2011) 

estimated the absorption and backscattering coefficients in Poyang Lake, China and analyzed 

their relations to the main water constituents; Matthews and Bernard (2013) presented the 

absorption properties of phytoplankton, CDOM and NAP for three small optically-diverse 

South African inland waters; Ylöstalo et al. (2014) characterized variations of different 

absorption components between different lake types and seasons in various lakes in the boreal 

region, in turn, Riddick et al. (2015) showed the spatial variability of absorption coefficients 

over a biogeochemical gradient in Lake Balaton, Hungary. 

Many efforts have been done from the last decade regarding the studies of (S)IOPs in 

inland waters, however, much more is expected to be done in order to cover a variety of 

environments and provide subsides to understand how these properties influence the 



 

 

Rodrigues, T.W.P. 

36 

parameterization of bio-optical models aiming to assist the remote-sensed-based monitoring of 

water quality. Environments such as reservoirs designed in a cascade system, for example, 

causes limnological modifications from the upstream to downstream reducing the turbidity and 

increasing water clarity (BARBOSA et al., 1999). The (S)IOPs can be somehow modulated by 

biogeochemical filtration from the first to the last reservoirs. The current study aims to 

characterize two optically different reservoirs situated along the cascade system of Tietê River, 

whose influence are dictated by the watershed dynamic, in order to improve the knowledge 

about bio-optical properties of a system covering two distinct trophic states (oligo to 

hypereutrophic) and to raise the main sources of spatial variability in the IOPs. 

 

3.2 Data and Methods 

 

For this study, data from all the field trips were used. The campaigns of BB were named 

as BB1, BB2 and BB3 standing for the sequence of field campaigns realization. For Nav, the 

campaigns were identified as Nav1, Nav2 and Nav3. The number of samples used for this 

chapter is displayed in Table 3.1. 

 

3.2.1 Water quality parameters 

 

To determine the TSS concentration we applied the method described by APHA (1998). 

The water volume was filtered on the same day of collection through pre-ashed and pre-weighed 

Whatmam fiberglass GF/F filter with a nominal porosity of 0.7 μm and then stored at the 

refrigerator until analysis. The filters were dried in 100° C oven for 12h, and then weighted in 

an analytical balance. To retrieve the inorganic suspended sediment (ISS), the filter dried and 

weighed in the last step was subjected to a muffle furnace for 75 min at 550° C and weighted 

again. As result, we determined the TSS, ISS and to estimate the organic suspended sediment 

(OSS), the last weighted filter was subtracted to the original filter weight after first drying. 

The analysis to determine the Chl-a concentration was based on Golterman et al. (1978). 

The specified volume of water was filtered under low vacuum pressure through a Whatman 

fiberglass GF/F filter with a porosity of 0.7 μm, and then frozen for laboratory analysis for no 

longer than 1 week. The chlorophyll was extracted by maceration in 90% acetone, then stored 

in 20 mL tubes and centrifuged. Afterward, one sample was placed in 1 cm quartz cuvette and 

to represent the reference a volume of acetone was placed in another cuvette. The readings were 

made in a spectrophotometer at 663 nm indicating the maximum absorption of chlorophyll in 



 

 

Rodrigues, T.W.P. 

37 

acetone and at 750 nm characterizing the absorption by chlorophyll and pheophytin. After this 

reading, 0.1N hydrochloric acid was added to the sample in order to remove magnesium from 

the chlorophyll and convert it into pheophytin and another reading was taken. 

 

3.2.2 Inherent optical properties 

 

To estimate the CDOM absorption coefficient, 𝑎𝐶𝐷𝑂𝑀(𝜆), a volume of water was 

filtered through a fiberglass Whatman GF/F with porosity of 0.7 μm, and then re-filtered under 

low vacuum pressure using a nylon membrane filter with porosity of 0.2 μm. The readings were 

performed using Shimadzu UV-2600 UV-VIS spectrophotometer (SHIMADZU, Japan) in 

absorbance mode, and the samples were placed in 10 cm quartz cuvettes. For each set of 

measurements, we performed a reference reading containing Milli-Q water, and for each read 

sample (𝑂𝐷𝑠𝑎𝑚𝑝𝑙𝑒), the reference absorbance value was subtracted (𝑂𝐷𝑟𝑒𝑓). The measured 

optical densities (𝑂𝐷𝑠𝑎𝑚𝑝𝑙𝑒) were converted to absorption coefficient by multiplying by 2.303 

and dividing by the path length (l = 0.1 m for a 10 cm cuvette). Therefore, the 𝑎𝐶𝐷𝑂𝑀(𝜆) was 

estimated following the equation (TILSTONE et al., 2002): 

 

𝑎𝐶𝐷𝑂𝑀(𝜆) = 2.303
𝑂𝐷𝑠𝑎𝑚𝑝𝑙𝑒

𝑙
= 2.303

𝑂𝐷𝑠𝑎𝑚𝑝𝑙𝑒

0.1
     (𝑚−1) (3.2) 

 

The 𝑎𝐶𝐷𝑂𝑀(𝜆) follows an exponential function decreasing from the visible to longer 

wavelengths. The data adopted to fit to the nonlinear regression ranged from 350 and 500 nm 

(BABIN et al., 2003). Twardowski et al. (2004) highlighted the main issues related to the 

spectral variability found in the results of many studies and attribute to the spectral range used 

in the adjustment. The 𝑆𝐶𝐷𝑂𝑀 can be used to understand the composition of CDOM (GREEN 

and BLOUGH, 1994, TWARDOWSKI et al., 2004). Absorption measures may contain errors 

related to the scattering of small particles or colloids that can cross the filters, leading some 

researchers to use equations to correct these effects (BRICAUD et al., 1981, GREEN and 

BLOUGH, 1994, BABIN et al., 2003). In order to turn the results comparable, the procedure 

described by Babin et al. (2003) were adopted in this study. 

For particulate absorption coefficients, water samples were filtered at low vacuum 

pressure using a 47 mm diameter fiberglass Whatman GF/F filter. The filters were wrapped in 

aluminum foil and frozen until laboratory analysis. A white filter wetted with ultrapure water 

was used as reference and the filter containing the particulate was placed on the integrating 



 

 

Rodrigues, T.W.P. 

38 

sphere module presented in the double-beam Shimadzu UV-2600 UV-VIS spectrophotometer 

(SHIMADZU, Japan) with spectral sampling ranging from 280 – 800 nm, to measure their 

optical density (OD). The T-R (Transmittance-Reflectance) method described by Tassan and 

Ferrari (1995, 1998) was employed to derive the total particulate absorption coefficient (𝑎𝑝). 

The particulate filter was submitted to depigmentation by oxidation in 10% sodium 

hypochlorite (NaClO), ensuring free phytoplankton influence to obtain 𝑎𝑁𝐴𝑃. The 𝑎𝜙 was 

retrieved by subtracting the 𝑎𝑁𝐴𝑃 from the 𝑎𝑝. 

 

𝑎𝜙(𝜆) = 𝑎𝑝(𝜆) − 𝑎𝑁𝐴𝑃(𝜆) (3.3) 

 

The 𝑎𝜙
∗ (𝜆) was obtained by normalizing the absorption due to phytoplankton by the 

Chl-a concentration. Babin et al. (2003) highlight that 𝑎𝜙 includes absorption related to all 

pigments, incorporating phaeopigment associated with other particles other than living 

phytoplankton. The 𝑎𝑁𝐴𝑃 can be adjusted by Roesler et al. (1989): 

 

𝑎𝑁𝐴𝑃(𝜆) = 𝑎𝑁𝐴𝑃(𝜆0)𝑒−𝑆(𝜆−𝜆0) (3.4) 

 

The fit of the model was carried out between 350 and 800 nm, excluding the interval 

between 400 – 480 and 620 – 710 nm related to the absorption of some type of pigment interval 

that may have remained after depigmentation (BABIN et al., 2003). 

 

3.3 Results 

 

3.3.1 General characteristics of water quality parameters and optical properties 

 

The water quality parameters as well as the optical properties were all analyzed in terms 

of field trips (Table 3.1). Thus, the average turbidity between BB1 and BB2 was statistically 

different (p-value < 0.05) and the same happened between BB2 and BB3. On the other hand, 

data from BB1 and BB3 were not statistically different (p-value > 0.05). The same result was 

also observed for Chl-a and TSS concentrations. The increase of Chl-a concentration in October 

of 2014 was due to the drought effect leading to flow reduction and longer retention time 

(WATANABE et al., 2016b). High values of Chl-a for BB revealed the eutrophication status 

of the water, mainly in the winter (LUZIA, 2004). In Nav, the average turbidity from Nav1 and 



 

 

Rodrigues, T.W.P. 

39 

Nav2 was not considered statistically different (p-value > 0.05), however, the average Chl-a 

concentration from all three field trips was statistically different. The average TSS from Nav1 

and Nav2 was not different, but this was not true when data from Nav2 was compared to Nav3 

and Nav1 with Nav3. Assessing both reservoirs, we clearly observe that BB is more turbid due 

to high contribution of phytoplankton mainly in the BB2 that achieved a Chl-a value of 797.80 

μg l-1 close to the dam. On the contrary, Nav presented low concentrations of Chl-a, where the 

average did not exceed 9.0 μg l-1 for Nav1 and Nav2 but reached 26.4 μg l-1 in Nav3, where the 

samples were concentrated at the upstream region of the reservoir. The variation of water 

quality parameters was strictly related to rainfall condition as well as the location of each 

reservoir in the cascade system. The analysis of individual reservoirs cannot provide the 

necessary information about the water quality dynamic, because this kind of structure cause 

discontinuity in the physical and biological features (BARBOSA et al., 1999). 

 

Table 3.1. Descriptive statistics of water quality and optical parameters. The notations Min-

Max, Aver, SD, CV and n stand for minimum-maximum, average, standard deviation, 

coefficient of determination and number of samples. Blank spaces represent lack of data. 

  Nav1 Nav2 Nav3 BB1 BB2 BB3 

Chl-a 

(μg l-1) 

Min–Max 2.5–12.6 4.5–20.5 15.8–38.6 17.8–279.9 263.2 – 797.8 62.8–245.7 

Aver ± SD 6.2±2.5 9.0±4.1 26.4±6.7 120.4±68.5 428.7±154.5 127.1±51.3 

CV (%) 40.0 45.5 25.3 56.9 36 40.4 

n 20 19 10 20 20 24 

TSS 

(mg l-1) 

Min–Max 0.1–2.6 0.5–2.8 1.9–5.3 3.6–16.3 10.8–44.0 1.6–8.4 

Aver ± SD 1.0±0.6 1.0±0.6 3.1±1.0 7.2±3.9 22.0±7.0 5.6±1.8 

CV (%) 61.7 56.2 32.4 44.1 32.1 32.0 

n 17 19 10 20 20 24 

OSS 

(mg l-1) 

Min–Max - - - 2.8–14.7 10.2–30.4 - 

Aver ± SD - - - 6.1±2.6 18.2±3.5 - 

CV (%) - - - 41.9 19.3 - 

n - - - 20 19 - 

Turbidity 

(NTU) 

Min–Max 1.0–2.5 1.0–2.6 - 1.7–12.5 11.6–33.2 3.1–6.8 

Aver ± SD 1.7±0.4 1.7±0.4 - 5.8±2.4 18.6±5.3 4.2±0.8 

CV (%) 25.4 22.9 - 45.8 28.3 20.3 

n 20 19 - 20 20 24 

SDD 

(m) 

Min–Max 2.3–4.8 2.5–4.7 1.9–3.8 0.8–2.3 0.4–0.8 1.0–1.6 

Aver ± SD 3.2±0.6 3.4±0.6 3.0±0.6 1.5±0.4 0.6±0.1 1.3±0.2 

CV (%) 20.0 17.6 21.0 28.2 17.2 16.6 

n 20 19 18 20 20 24 

Depth 

(m) 

Min–Max 5.3–30.0 - 7.4–32.9 10.0–30.0 8.0–18.5 9.6–26.0 

Aver ± SD 18.0±8.3 - 22.9–6.8 15.4±4.1 13.0±2.7 16.3±3.7 

CV (%) 46.2 - 29.8 26.5 21.0 22.5 

n 20 - 18 19 20 24 

𝒂𝝓(𝟒𝟒𝟑) 

(m-1) 

Min–Max 0.1–0.4 0.1–0.4 0.1–0.6 0.3–2.4 - - 

Aver ± SD 0.2±0.1 0.2±0.1 0.3±0.1 1.1±0.6 - - 

CV (%) 39.1 40.8 43.5 52.1 - - 

n 20 20 19 20 - - 

Min–Max 0.1–0.5 0.1–0.7  0.4–0.8 0.3–0.8 - - 



 

 

Rodrigues, T.W.P. 

40 

𝒂𝑵𝑨𝑷(𝟒𝟒𝟑) 

(m-1) 

Aver ± SD 0.3±0.1 0.3±0.1 0.6±0.1 0.5±0.1 - - 

CV (%) 25.8 52.0 19.6 25.4 - - 

n 20 20 19 20 - - 

𝒂𝑪𝑫𝑶𝑴(𝟒𝟒𝟑) 

(m-1) 

Min–Max 0.2–0.3 0.2–0.5 0.06–0.09 0.6–1.1 - - 

Aver ± SD 0.3±0.0 0.3±0.1 0.08±0.01 0.7±0.1 - - 

CV (%) 8.5 17.0 8.9 14.2 - - 

n 20 19 19 20 - - 

 

Water quality properties such as TSS and Chl-a as well as the IOPs at 443 nm were 

plotted against each other aiming to provide the main constitution of both reservoirs. No 

significant linear relationship (p-value > 0.05) was observed between TSS and Chl-a (Figure 

3.1a) for Nav considering a log scale (Nav1: R = 0.44; Nav2: R = -0.18; Nav3: R = 0.42), which 

means that Nav is not dominated by phytoplankton. However, for BB, there is a significant 

relationship for all three field trips (BB1: R = 0.75, p-value < 0.001; BB2: R = 0.60, p-value < 

0.05; BB3: R = 0.74, p-value < 0.001) showing that phytoplankton was the main contributor of 

TSS. 𝑎𝜙(443) and Chl-a concentration (Figure 3.1b) did not show significant relationship (p-

value > 0.05) for Nav1 (R = 0.42) and Nav2 (R = -0.22) on the other hand, for Nav3, we 

observed a strong linear relationship (R = 0.81, p-value < 0.05). For BB, the correlation was 

statistically significant for both BB1 (R = 0.92, p-value < 0.001) and BB2 (R = 0.52, p-value < 

0.05). As reported by Le et al. (2015), the strong correlation between 𝑎𝜙(443) and Chl-a is 

associated with high loads of nutrients in the aquatic system increasing phytoplankton 

production. According to Luzia (2004), the limiting nutrient that controls the growth of aquatic 

plants in majority of fresh waters is the inorganic form of phosphorus.  

Considering the relationship between 𝑎𝑁𝐴𝑃(443) and TSS (Figure 3.1c) for Nav, it was 

difficult to notice any relationship (Nav1: R = 0.27; Nav2: R = -0.11; Nav3: R = 0.33, p-value 

> 0.05). For BB, the non-correlation was also observed for both field trips (BB1: R = -0.05; 

BB2: R = -0.37, p-value > 0.05). Now considering the relationship between 𝑎𝐶𝐷𝑂𝑀(443) and 

Chl-a (Figure 3.1d) for Nav, again, no correlation was observed in any of the three campaigns 

(Nav1: R = 0.14; Nav2: R = 0.15; Nav3: R = 0.17, p-value > 0.05), while for BB1 the 

relationship was significant (R = 0.77, p-value < 0.001) and weak for BB2 (R = 0.25, p-value 

> 0.05). Regarding the relationship between 𝑎𝐶𝐷𝑂𝑀(443) and TSS (Figure 3.1e), Nav did not 

show any statistical correlation (Nav1: R = -0.06; Nav2: R = -0.18; Nav3: R = 0.14, p-value > 

0.05) whilst BB1 showed to be moderately (R = 0.53, p-value < 0.05) and weakly correlated 

for BB2 (R = 0.19, p-value > 0.05). The non-correlation combined with the low Chl-a in Nav 

can indicate that TSS and CDOM were possibly originated from land-based sources and not 

from phytoplankton degradation. 



 

 

Rodrigues, T.W.P. 

41 

Figure 3.1. Relationship between water quality and optical parameters considering data from 

all field campaigns. (a) TSS versus Chl-a, (b) 𝑎𝜙(443) versus Chl-a, (c) 𝑎𝑁𝐴𝑃(443) versus 

TSS, (d) 𝑎𝐶𝐷𝑂𝑀(443) versus TSS and (e) 𝑎𝐶𝐷𝑂𝑀(443) versus Chl-a. 

 

 

3.3.2 Absorption coefficient budget 

 

The relative contribution of phytoplankton, CDOM and NAP relative to the total 

absorption without the water fraction (𝑎𝑡−𝑤) can be seen in Figure 3.2. The wavelengths chosen 

for analysis (443, 560 and 665 nm) characterize the light interaction with dissolved organic 

matter and particulate matter (BABIN et al., 2003; LE et al., 2013). The ternary plots show the 

contribution and representativeness of certain OSC in the water and can be used to assist the 

estimation of these components using proper algorithms (MISHRA et al., 2014; RIDDICK et 

al., 2015). Figure 3.2(a) shows that at 443 nm, Nav1 was dominated by 𝑎𝑁𝐴𝑃 with 43.24 ± 6.52 

%, Nav2 by 𝑎𝐶𝐷𝑂𝑀 with 37.89 ± 9.22 %, Nav3 by 𝑎𝑁𝐴𝑃 62.17 ± 9.39 %. At 560 nm (Figure 

3.2b), Nav1, Nav2 and Nav3 were all dominated by 𝑎𝑁𝐴𝑃 with 48.57 ± 7.42 %, 46.87 ± 10.11 

% and 72.14 ± 13.70 %, respectively. At 665 nm (Figure 3.2c), all three field trips were 

dominated by 𝑎𝜙 (Nav1: 55.00 ± 8.95 %, Nav2: 39.72 ± 12.06 % and Nav3: 65.16 ± 12.39 %). 

As expected, the most dominant OSC in the absorption budget considering both field 

trips and wavelengths in BB was the phytoplankton. At 443 nm (Figure 3.2a), BB 1 and BB 2 



 

 

Rodrigues, T.W.P. 

42 

were dominated by 𝑎𝜙 with a percentage of 45.72 ± 12.71 % and 72.38 ± 9.36 %. At 560 nm 

(Figure 3.2b), the 𝑎𝜙 of BB 1 and BB 2 presented a percentage of 41.98 ± 13.43 % and 63.33 

± 13.75 %, respectively while at 665 nm (Figure 3.2c), the contribution was over 80% (BB 1 = 

85.38 ± 6.80 % and BB 2: 87.90 ± 6.38 %). 

 

Figure 3.2. Ternary plot depicting the absorption coefficient budget of both Nav and BB 

reservoirs at three wavelengths: (a) 443 nm, (b) 560 nm and (c) 665 nm. 

 

 

3.3.3 CDOM absorption 

 

The 𝑎𝐶𝐷𝑂𝑀 spectra approached to zero close to 700 nm and showed an exponential 

decrease pattern from shorter to higher wavelengths (BRICAUD et al., 1981) with values at 

443 nm ranging between 0.06 to 0.45 m-1 in Nav and between 0.62 to 2.34 m-1 in BB (Figure 

3.3). The magnitude differs from one reservoir to another as well as from one season to another. 

Values from BB are like those found by Matthews and Bernard (2013), who stated an interval 

between 0.63 to 4.13 m-1 in three South African reservoirs. Wu et al. (2011) found values 

ranging between 0.33 to 1.01 m-1 in Poyang Lake, China, considered a suspended inorganic 

matter dominated water with low Chl-a concentration (1.47 – 24.65 μg l-1) while Campbell et 

al. (2011) reported an interval of 0.36 and 1.59 m-1 in three Australian reservoirs. Zhang et al. 

(2007) observed in Lake Taihu values ranging between 0.27 to 2.36 m-1. Binding et al. (2008) 

reported values at 440 nm ranging between 0.08 to 0.75 m-1 in Lake Erie (Canada/USA) that 

was close to that found in Nav. The mean value for the spectral slope of CDOM (𝑆𝐶𝐷𝑂𝑀) was 

0.018 nm-1 for BB1 and 0.016 nm-1 for BB2, whilst for Nav1 the mean 𝑆𝐶𝐷𝑂𝑀 was 0.020 nm-1 

and 0.018 nm-1 for Nav2 and Nav3. Riddick et al. (2015) also found a mean 𝑆𝐶𝐷𝑂𝑀 of 0.018 

nm-1 in Kis-Balaton (wetland system) and 0.020 nm-1 in the Lake Balaton (freshwater lake in 

central Europe). On the other hand, Babin et al. (2003) reported a narrow range around 0.0176 



 

 

Rodrigues, T.W.P. 

43 

nm-1 in coastal waters around Europe. The authors highlighted that diverse protocols to retrieve 

𝑆𝐶𝐷𝑂𝑀 can make it difficult to compare values from different regions. 

 

Figure 3.3. Variability of 𝑎𝐶𝐷𝑂𝑀 by field trip. (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 

and (f) the average value of 𝑎𝐶𝐷𝑂𝑀(𝜆) for each field trip. Different y axes were applied for Nav 

and BB due to magnitude discrepancies. 

 

 

3.3.4 Phytoplankton absorption 

 

Figure 3.4 depicts the 𝑎𝜙 spectra between 400 to 700 nm. The magnitude and shape are 

different when we compare both reservoirs and this can be due to the diversity of trophic states 

and phytoplankton assemblages (MATTHEWS and BERNARD, 2013). The peaks at the blue 

and red regions are consistent with the typical diagnostic features of phytoplankton absorption 

(WU et al., 2011). At 440 nm, accessory pigments and Chl-a contribute with high absorption, 

while at 675 nm, Chl-a plus phaeophytin are primarily responsible for high absorption 

(SATHYENDRANATH et al., 1987; LE et al., 2009b). Other pigments such as carotenoids 

absorb from 460 to 490 nm, however, this feature is not prominent in Nav neither BB.  

The absolute intervals of absorption at 443 nm are 0.05 – 0.57 m-1 for Nav and 0.27 – 

10.34 m-1 for BB. Roesler et al. (1989) found values at 400 nm between 0.03 to 0.58 m-1 in 

inland marine waters in the USA, which was very close to that found in Nav. High values were 

reported in many turbid inland waters with characteristics close to that presented in BB (WU et 



 

 

Rodrigues, T.W.P. 

44 

al., 2011, MATTHEWS and BERNARD, 2013, MISHRA et al., 2014). 𝑎𝜙 at 443 nm exhibited 

an increased trend from the first to the second field campaigns, and the same happened for Nav 

considering the first to the third field trips.  

Mean 𝑎𝜙(620) ranged from 0.33 to 1.47 m-1 in BB and 0.04 to 0.10 m-1 in Nav. Chl-a 

concentration showed a strong relationship with 𝑎𝜙(620) and BB1 (R² = 0.80, p < 0.001, not 

presented here) but no correlation was found with BB2 (R² = 0.11, p > 0.05) as well as with 

Nav considering all three field trips (p > 0.05). The pigment phycocyanin, when present, can 

be detected through the 𝑎𝜙 spectrum especially at 620 nm. This pigment is a marker for 

cyanobacteria in eutrophic inland waters (SIMIS et al., 2005). Figure 3.1(b) showed the non-

relationship between 𝑎𝜙(443) and Chl-a in Nav, suggesting that Chl-a was not the only 

pigment affecting 𝑎𝜙, but other accessory pigments, such as carotenoid (WU et al., 2011). 

 

Figure 3.4. Variability of 𝑎𝜙 by field trip. (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) BB2 and 

(f) the average value of 𝑎𝜙(𝜆) for each field trip. Different y axes were applied for Nav and BB 

due to magnitude discrepancies. 

 

3.3.5 NAP absorption 

 

The spectral shape of 𝑎𝑁𝐴𝑃 (Figure 3.5) is quite similar of 𝑎𝐶𝐷𝑂𝑀 with an exponential 

decrease from 400 to 700 nm. Visually, the seasonal pattern between field campaigns and 

reservoirs did not show remarkable changes as reported previously. Taking into account the 

data from Nav, we observed that at 443 nm the values varied from 0.13 to 0.82 m-1 with averages 

varying from 0.33 m-1 in Nav1, 0.27 m-1 in Nav2 and 0.61 m-1 in Nav3, showing that 𝑎𝑁𝐴𝑃(443) 



 

 

Rodrigues, T.W.P. 

45 

increased in Nav3. Sample 5 from Nav2 stood out from the dataset because of the presence of 

sand dredging activity in that region leading to sediment resuspension. Meanwhile, BB showed 

a range of 0.23 to 1.67 m-1 with averages varying from 0.45 m-1 in BB1 and 0.47 m-1 in BB2. 

Samples 17 to 20 placed at Tietê River and confluence of Tietê and Piracicaba Rivers presented 

the highest values for 𝑎𝑁𝐴𝑃(443). These regions receive high loads of sediment from the 

metropolitan region of São Paulo. The mean slope (𝑆𝑁𝐴𝑃) of BB1 and BB2 was 0.007 and 0.008 

nm-1, respectively, ranging between 0.006 to 0.01 nm-1 while for Nav the mean values were 

0.009, 0.006 and 0.007 nm-1, respectively, ranging between 0.003 to 0.011 nm-1. 

 

Figure 3.5. Variability of 𝑎𝑁𝐴𝑃(𝜆) in all field trips (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) 

BB2 and (f) the average value of 𝑎𝑁𝐴𝑃(𝜆) for each field trip. 

 

 

3.3.6 Particle absorption  

 

The Figure 3.6 displays the spectral behavior of 𝑎𝑝 from 400 to 700 nm. It shows the 

contribution of phytoplankton and non-algal particle. Besides the magnitude, the shapes of 𝑎𝑝 

from BB and Nav are clearly different, showing a smooth feature between 400 to 450 nm in 

Nav and a marked feature in BB. These are due to the high contribution of NAP and CDOM at 

this wavelength range, as also supported by Figure 3.1. On the contrary, BB was highly affected 

by photosynthetic pigments at the blue and red regions. The pattern from Nav was also observed 

by Meler et al. (2016) in Gulf of Gdánsk during winter and in river mouths, Baltic Sea during 



 

 

Rodrigues, T.W.P. 

46 

all seasons except summer. Wu et al. (2011) did not observe any feature at the blue and red 

regions due to high loads of inorganic sediment, which masked the phytoplankton contribution 

along the spectra. 

 

Figure 3.6. Variability of 𝑎𝑝(𝜆) in all field trips.  (a) Nav1, (b) Nav2, (c) Nav3, (d) BB1, (e) 

BB2 and (f) the average value of 𝑎𝑝(𝜆) for each field trip. Different y axles were applied for 

Nav and BB due to magnitude discrepancies. 

 

 

3.4. Discussion 

 

As reported in many studies, BB as the first reservoir of the cascade system receives 

high loads of sediments coming from the metropolit