Monitoring satellite-derived SSS concentration

1. Introduction

The Brazilian National Water Agency (ANA) and the French National Research Institute for Sustainable Development (IRD), through an agreement with the Brazilian Agency for Cooperation, have been developing since 2009 the technical cooperation project Hydrological Monitoring of Large Basins (Quantity and Quality) – Project “MEG-HIBAM”, which aimed to demonstrate the possibility of monitoring hydrological parameters by using spatial sensors.

Under the MEG-HIBAM Project, activities were developed to produce river and reservoir level estimates using altimetric radar sensors and to evaluate water quality, using quality parameters extracted from spatial imaging sensors. Tools were developed for the processing of massive data, producing time series of hydrological parameters in the Amazon and Northeast basins.

These tools were integrated into an online cartographic diffusion system (HidroSat) for the visualization of estimates of sediment and algae pigment concentrations at “virtual” stations in the Amazon and the Northeast region. Aiming to systematize the knowledge generated by the activities carried out in the “MEG-HYBAM” project, ANA and IRD prepared this page, so that users without prior knowledge have access to technical information that enabled the processing of satellite images for the extraction of water quality parameters that are available in the HidroSat system. The purpose of this page is to present:

Essas ferramentas foram integradas num sistema de difusão cartográfico online (HidroSat) para a visualização das estimativas de concentração de sedimentos e de pigmentos algais em estações “virtuais” da Amazônia e da região Nordeste. Visando sistematizar os conhecimentos gerados pelas atividades realizadas no Convênio, a ANA e o IRD elaboraram esta pagina, de forma que usuários sem conhecimento prévio tenham acesso a informações técnicas que possibilitaram o processamento de imagens de satélite para a extração de parâmetros de qualidade de água e que são disponibilizados no sistema HidroSat. O objetivo desta pagina é apresentar:
1) theoretical concepts of the spectral behavior of water;
2) information about the orbital sensor imageador MODIS;
3) the different steps in the processing of MODIS images until the availability of hydrological information in the Hidrosat system.
This page is under development and will be completed during the project. Data are subject to inconsistencies.

2. Optical properties of water


Radiometry is the quantitative measure of the intensity of any of the known types of radiation, such as electromagnetic radiation, emitted by the sun or by an artificial source such as a lamp (Meneses & Netto, 2001).

Electromagnetic radiation has the property of propagating in empty space or vacuum, which allows its use for remote sensing purposes, since it can be detected by sensors from photographic cameras or multispectral imagers on board of airplanes or satellites.

In the region of the electro-magnetic spectrum that extends from the wavelength range of the visible to the infrared of short waves, from 400 to 2500 nm (0.4 to 2.5 µm), electromagnetic radiation receives the denomination of optical, whose main property is to be able to be reflected by the surfaces of objects according to the optical laws of reflection (Meneses & Netto, 2001).


Each material or object interacts spectrally differently from the illuminating energy coming from the sun. When electromagnetic energy reaches the surface of an object, exchanges of energy will occur which will result in the absorption, reflection and/or transmission of part of the incident energy. The amount of energy for each of these forms of interaction is related to the physical, chemical and biological properties of the object, and to external properties such as measurement geometry (Meneses & Netto).

Reflectance is the percentage ratio between the intensity of energy reflected by the object and the intensity of incident energy.

Numerous systematic studies of reflectance measurements, with the detailing of the spectral behavior of the targets and characterization through spectral curves of the reflectance patterns of different objects, have generated fundamental knowledge for the extraction of information about the nature of the objects contained in the pixels of the images, providing resources to researchers and users for interpretation of multispectral images.


Reflectance spectroradiometry is a technique that measures at different wavelengths the electromagnetic energy reflected from the surface of objects and is represented in the form of a graph that is called the spectral reflectance curve (Meneses & Netto, 2001).

The absorption coefficient of pure water is minimal in the region between 400 and 600 nm, rapidly increasing in the near infrared region. On the contrary, the pure water scattering coefficient is maximum in the blue range and decreases exponentially towards the infrared. Estimating the spectrum of water reflection by the ratio between the coefficient of backspreading (BB) and the coefficient of absorption (a), the energy reflected by pure water is maximum in the range of blue and decreases towards red, tending to present the blue color when observed from a satellite.


The spectral behavior of continental aquatic systems is related to the concentration of optically active components in water. These components are substances in suspension or solution in water that cause changes in the color of pure water depending on its concentration and nature. There are three large groups of optically active components:

* Inorganic and organic particles in suspension in water (sediments);
* Organic components in solution in water (humic and fulvic acids);
* Pigments related to the presence of living organisms (phytoplankton) such as chlorophyll-a.

Influence of living organisms: The pigments responsible for the photosynthesis of phytoplankton – chlorophylls, carotenoids and biliproteins – cause the selective absorption of electromagnetic radiation that penetrates the water surface. Most plants have chlorophylls a, b, c and more rarely d, and chlorophyll is the one that occurs in greater abundance. Among algae, however, the concentration of chlorophyll varies widely. The ratio between chlorophyll concentration a and b also varies from species to species, and may reach minimum values around 1.0 mg/l in marine species up to 6.0 mg/L for Euglenophyta (Meneses & Netto, 2001).

The blue and red regions are the main ranges of absorption of phytoplankton pigments (Table 1). Thus, an increase in the concentration of algae in the water mainly implies a reduction in water reflectance in the blue region (Meneses & Neto, 2001).


Pigments Absorption bands (nm)
Chlorophyll-a 435/675
Chlorophyll b 480/650
Chlorophyll c 440/645
Carotenoids 425/450/500
Biliproteins 498/553/555/562/568/585/620/650/670

Table 1. Absorption bands of photosynthetic pigments.
Source: (Meneses & Netto, 2001).

As the concentration of pigments increases, there is a constant decrease in the energy reflected by water in the pigment absorption bands. The maximum reflectance of water gradually passes from the blue to the green region, while at the same time an increase in the energy reflected in the region of around 680-700 nm begins to occur, which corresponds to the emission region associated with fluorescence by chlorophyll. Therefore, the waters with high concentrations of pigments will tend to present green color, since the maximum reflected energy occurs in the green region (Meneses & Netto, 2001).

Martinez et al. (2011) studied the eutrophication processes of 4 large reservoirs in the Northeast region of Brazil using MODIS images from 2000 to 2010. During the research, field campaigns were carried out during the dry and wet seasons to measure various water quality parameters, including pigments and nutrient concentrations at different depths in the water column, and hyperspectral radiometric measurements, in order to relate these data to reflectance data from the MODIS sensor. The study found a strong contrast of the spectral behavior between phytoplankton bloom-labeled waters and sediment-loaded waters. Near the length of where 440 nm was observed a minimum reflectance that is caused by the maximum absorption of chlorophyll-a. A prominent 560nm wavelength reflectance peak in the green range represents the minimum absorption of all algae pigments and is related to the scattering caused by suspended inorganic matter and the response of phytoplankton cell walls. In the range 620 to 630 nm, a marked decline in reflectance caused by the absorption of phycobillin pigment from cyanobacteria is noted. A reflectance hole near the wavelength of 670 nm is due to the maximum absorbance by chlorophyll-a in the red range of the spectrum. The research proved the effectiveness of using MODIS images for monitoring eutrophication processes in the four reservoirs studied, demonstrating that the spectral bands of the MODIS sensor are suitable for monitoring eutrophication by quantifying the minimum absorption of chlorophyll-a near the 560 nm band.

Influence of inorganic suspended particles: Suspended particles, especially inorganic ones, can also increase the water absorption coefficient. According to Meneses & Netto (2001), indirect measures show that the particle absorption coefficient tends to have a spectral behavior similar to that of dissolved organic matter, with maximum absorption in blue.

The main effect of particulate matter is to increase the water scattering coefficient; the higher the concentration of total suspended solids in the water, the higher the water scattering coefficient.

Experiments with edaphic material with organic and inorganic fractions showed an increase in the reflectance of the water volume. Laboratory studies have shown that the inorganic fraction is the largest responsible for the increase in water reflectance due to its higher refractive index. With a higher concentration of suspended sediments in the water there is a displacement of the maximum of reflectance of the water towards longer wavelengths, in addition, there is an enlargement of the spectral region in which this maximum occurs, which starts to behave with a level of reflectance almost constant between 500 and 700 nm. There is also an expressive growth of reflectance in the infrared region.

According to Martinez et al. (2009), numerous studies concern the sensitivity of remote sensor reflectance to suspended sediment concentrations in oceans and inland waters. A significant number of researchers have reported a strong positive correlation between surface suspended sediment (SSS) concentrations and spectral radiance, and have observed that the relationship may depend on the concentration range, on the water types and on the origin of the suspended matter. Most studies agree that the best correlation between reflectance and SSC is between 700 and 800 nm in turbid inland waters.

Martinez et al. (2009) quantified the sediment balance in the Amazon River using reflectance data derived from 554 MODIS (MODerate-resolution Imaging Spectroradiometer) spatial sensor images from the period 2000 to 2009, and monthly data on sediment load suspended from the HYBAM monitoring network from the period 1995 to 2007. The results show an increase in the reflectance of water extracted from the MODIS infrared band according to measurements of the concentration of suspended sediments on the surface. The study also presented an equation relating surface suspended sediment and mean suspended sediment, the latter obtained in 18 sampling campaigns conducted between 1995 and 2003. Thus, the study determined the mean solid discharge of the Amazon River in Óbidos in the period 1996-2007, contributing to the quantification of erosion processes in one of the main remaining natural ecosystems. With respect to the surface reflectance of the waters of the Amazon River, the reflectance data appear to be robust with the concentration of sediments suspended to the river surface over a large concentration range and for several consecutive hydrological cycles. With the combination of excellent temporal resolution and fine calibration, MODIS data can be used operationally with field observations to provide more information to poorly assessed basins as well as large basins.

Influence of the dissolved organic substances in the water: Kirk (1994) shows an exponential increase of the coefficient of absorption of the water towards the longer wavelengths. Mobley (1994) suggests that the presence of colloids associated with dissolved organic matter promotes an increase in the coefficient of backspreading towards longer wavelengths, i.e., with an increase in the concentration of materials dissolved in water, the reflectance of the blue region decreases to a point where the maximum reflectance starts to occur in the green and red region, which gives the water a yellow color.

According to Meneses (2001), when the concentration of dissolved organic substances is very high, as in the case of the Rio Negro in the Amazon, the absorption supersedes the scattering and there is practically no energy spread backwards by the volume of water, which acquires a black appearance when in large volume.

Mantovani (1993) shows, in laboratory simulations, that the increase in the concentration of dissolved organic matter is associated with the reduction of water reflectance in the region of blue and green. In the red region (around 650 nm), the reflectance practically does not change with the variation of the concentration of organic matter and, in the infrared region, the reflectance of water expands with the dissolved organic matter. With the increase in the concentration of dissolved organic matter in the water, the water will tend first to a yellow colour, and then become darker red according to the low levels of reflectance.

3. Post-processing of MODIS images


MOD3R (MODIS Reflectance Retrieval over Rivers) is a MODIS image post-processing program developed by IRD in JAVA language for the extraction of MODIS image reflectance time series from water bodies. It is a functionality developed to be easily used by people working with sediment or phytoplankton concentration measurements in water bodies. The same software is implemented in the Hidrosat system to automatically process MODIS images.

The algorithm developed for the program accurately and consistently determines over time the pixels of pure water in an image, or its best candidates, regardless of the types of river morphology. By extracting the reflectance values from the red and infrared bands of MODIS images, it is possible to determine the surface concentrations of sediments and phytoplankton in water.

The program presents a simple interface for selecting the images and the parameters required for the calculations. Several output files are generated with summaries of the processed images. The result for each image is accompanied by a quality indicator coded between one and three, giving a sense of confidence that can be attributed to the value found.

The information that can be extracted from satellite images for research in hydrology depends both on the geomorphology of the river (mainly the width of the section of the river studied), the meteorology and the image acquisition geometry (the final resolution of the pixels depends on the angle of view of the satellite and the position of the sun).

The MOD3R is a version for individual use with a specific interface. The same program is being automatically operated in Hidrosat.


The MOD3R software was developed in JAVA programming language, object oriented language, robust, free and widely used all over the world, which allows the creation of runnables independent of the type of machine, providing good portability. These features will allow the execution of quick and easy updates for future software development.


Martinez et. al (review article) has developed an algorithm to separate within a satellite image pixels that present a water spectral response from other pixels in which the response is mixed with other targets, such as vegetation and soils on the banks of water bodies. The process of spectral mixing depends on the spatial resolution of the pixel that changes with each acquisition, since the angle of acquisition in a MODIS image can range from -53 degrees to +53 degrees. This variability, together with the variation of the water mirror itself as a function of the hydrological period, makes it impossible, in most continental water bodies (rivers, lakes), to ensure that a given pixel is always a representative pixel of water.

The MOD3R program allows to automatically process a large number of images only informing the area of interest in the image through the selection of a mask. If there are several points of interest in an image (river, lake, etc…), it will be necessary to create as many masks as there are points of interest. MOD3R only processes MODIS MOD/MYD09Q1 and MOD/MYD09A1 products in HDF format that have been downloaded from the GETMODIS image bank within the HidroSat system. These images were produced from a spatial cutout of the MOD09/MYD09 products and a selection of radiometric bands and metadata.

The extraction of the reflectance of the water is performed by the MOD3R in 4 steps:
* Selection of the optimal processing quality pixels in the image according to the information provided in the MOD/MYD09A1 product quality band;
* Extraction of pixels located in the area of interest defined by the mask;
* Segmentation of the selected pixels into N homogeneous groups (“clusters”) by the K-means method from a Monte Carlo-type random sampling performed M times (“loop”);
* Calculation of an inverse model of spectral mixture for the determination of water pixels.

Selecting the area of interest in the image: The area of interest will be extracted in each image from the area delimited by the mask selected in the “Choose mask file” function of the MOD3R Main window.

Estimation and classification of pixels in homogeneous groups: Through the statistical algorithm K-means are created homogeneous groups (clusters) of pixels. This method of unsupervised classification seeks to form K groups, represented by the center of the group (in the spectral space) and the dispersion of the values of the other pixels in the given group.

Application of a linear model of spectral mixture: Once the groups have been defined in the previous step, the one that best represents the water endmember Refágua is selected. As this endmember a priori is not known, it is necessary to test each Gi group to determine which is the one that best describes the process of spectral mixing in all the other Gj groups. The spectral mixing equation is written for each radiometric band and for each group the result of i segmentation with 1 ≤ i ≤ K. Thus, there are only two unknowns: the fractions i e i for each “j” group. With two equations (for the red band and the infrared band) for each Gj group the system of equations is solved producing estimates of i, i and εi. After solving the system of equations, a condition of positivity of the fractions is added to force the algorithm to produce realistic estimates of the fractions. This process is repeated for each Gj group to store finally the total residue t that is associated with the candidate to endmember Gi water. In an interactive way, each group is tested as candidate to water endmember Gi and the group that produces the smallest value of εt is selected as the final water endmember.


The application can generate up to 4 products:
1. File in jpg format for each processed image where the pixels used in image processing are shown in color. In red are shown the pixels that were not selected for the extraction of reflectance, and in blue the pixels selected for the extraction of reflectance;
2. Spreadsheet with information of the good quality pixels selected in the mask;
3. Spreadsheet that presents more accurate information on the clusters generated by the program. The file generated has information on the average reflectance of each cluster generated in the red and infrared bands and the pixel number of the cluster. It also shows the best clusters selected and the residue value calculated by the algorithm for each cluster. The smaller the residue, the closer the behavior of the cluster will be to the behavior of pure water.
4. Automatically generated data sheet with the results (Table 1). Columns 3 to 7 show the mean of the reflectance in each band and columns 8 to 12 the standard deviation of the reflectance. When no data is available within the 8-day period (due to the persistent presence of clouds, for example), the row is filled with zeros.

Column 1: Date
Column 2: Filename
Column 3: Band #1 Mean reflectance
Column 4: Band #2 Mean reflectance
Column 5: Band #3 Mean reflectance
Column 6: Band #4 Mean reflectance
Column 7: Band #5 Mean reflectance
Column 8: Band #1 Reflectance Standard Deviation
Column 9: Band #2 Reflectance Standard Deviation
Column 10: Band #3 Reflectance Standard Deviation
Column 11: Band #4 Reflectance Standard Deviation
Column 12: Band #5 Reflectance Standard Deviation
Column 13: Buffer size
Column 14: Viewing Zenith Angle
Column 15: Viewing Zenith Angle Standard Deviation
Column 16: Sunglint angle
Column 17: Number of valid pixel within the buffer
Column 18: Quality flag

Table 2 – Description of the fields in the data sheet.

Column 18 shows a classification of the image quality level into 4 levels:

0: no data;

1: optimal quality data;

2: average quality data;

3: poor quality data.

Poor quality data should be discarded after processing. The most common occurrences of “3” quality are associated with the reflection of sunlight by the water mirror causing the “Sun glint” process. Quality data 2 corresponds to data that were acquired in adverse conditions (angle of incidence above 45 degrees or strong cloud cover). Therefore, for a quick estimate of the quality parameters it is advisable to automatically discard the quality data 2 and 3. In a second step the quality data “2” can be considered to increase the quantity of measurements and should be criticized depending on the data available in the near future.

4. Bibliographic list on the use of satellite images and spectro-radiometers to monitor water quality parameters

Dekker A.G., R.J.V., S.W.M. Peters (2002). Analytical algorithms for lake water TSM estimation for retrospective analyses of TM and SPOT sensor data. International journal of remote sensing, 23, 15-35

Aranuvachapun, S., & Walling, D.E. (1988). Landsat-MSS Radiance as a measure of suspended sediment in the lower Yellow River (Hwang Ho). Remote Sensing of Environment, 25, 145-165

Bhargava, D.S., & Mariam, D.W. (1991). Effects of suspended particle size and concentration on reflectance measurements. Photogrammetric Engineering Remote Sensing, 57, 519-529

Chen, Z., Hu, C., & Muller-Karger, F. (2007). Monitoring turbidity in Tampa Bay using MODIS/Aqua 250-m imagery. Remote Sensing of Environment, 109, 207-220

Dekker, G.A., & Malthus, T.J. (1991). Quantitative Modeling of Inland Water Quality for High-Resolution MSS Systems. IEEE Transactions on Geoscience and Remote Sensing, 29, 89-95

Doxaran, D., Froidefond, J.M., Lavender, S., & Castaing, P. (2002). Spectral signature of highly turbid waters. Application with SPOT data to quantify suspended particulate matter concentrations. Remote Sensing of Environment, 81, 149-161

Hu, C., Chen, Z., Clayton, T.D., Swarzenski, P., Brock, J.C., & Muller-Karger, F.E. (2004). Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: Initial results from Tampa Bay, FL. Remote Sensing of Environment, 93, 423-441

Justice, C.O., Townshend, J.R.G., Vermote, E.F., Masuoka, E., Wolfe, R.E., Saleous, N., & al., e. (2002). An overview of MODIS land data processing and product status. Remote Sensing of Environment, 83, 3-15

Kirk, J.T.O. (1994). Light and Photosynthesis in Aquatic Ecosystems: Cambridge University Press

Martinez J.M., JL Guyot, G Cochonneau, F Seyler. Surface water quality monitoring in large rivers with MODIS data – Application to the amazon basin. Proceedings of International Geoscience and Remote Sensing Symposium IGARSS 2007. Barcelona, July 2007

Martinez J.M., L. Maurice-Bourgoin, P. Moreira-Turcq, F. Seyler and J.L. Guyot. Amazon floodplain water quality monitoring using MERIS and MODIS data. Proceedings of the ENVISAT & ERS Symposium, 6-10 September 2004, Salzburg, Autriche, ESA publication SP-572, (CD-ROM), 10 p., 2004

Mertes, L.A.K., Smith, C.T., & Adams, J.B. (1993). Estimating Suspended Sediment Concentrations in Surface Waters of the Amazon River Wetlands from Landsat Images. Remote Sens.Environ., 43, 281-301

Miller, R.L., & McKee, B.A. (2004). Using MODIS Terra 250 m imagery to map concentrations of total suspended matter in coastal waters. Remote Sensing of Environment, 93, 259-266

Novo, E.M.L.M., Hamsom, J.D., & Curran, P.J. (1989a). The effect of sediment type on the relationship between reflectance and suspended sediment concentrations. International journal of remote sensing, 10, 1283-1289

Novo, E.M.L.M., Hamsom, J.D., & Curran, P.J. (1989b). The effect of viewing geometry and wavelength on the relationship between reflectance and suspended sediment concentration. International journal of remote sensing, 10, 1357-1372

Thiemann, S., & Kaufmann, H. (2002). Lake water quality monitoring using hyperspectral airborne data – a semiempirical multisensor and multitemporal approach for the Mecklenburg Lake District, Germany. Remote Sensing of Environment, 81, 228-237

Toole, D.A., D.A., S., Menzies, D.W., Neumann, M.J., & Smith, R.C. (2000). Remote-sensing reflectance determinations in the coastal ocean environment: impact of instrumental characteristics and environmental variability. Applied Optics, 39, 456-469