Journal of South American Earth Sciences xxx (2012) 1e10

Contents lists available at SciVerse ScienceDirect

Journal of South American Earth Sciences

journal homepage: www.elsevier.com/locate/jsames

A study of sediment transport in the Madeira River, Brazil, using MODIS

remote-sensing images
Raúl Espinoza Villar a, *, Jean-Michel Martinez a, b, c, e, Marie Le Texier c, d, Jean-Loup Guyot b, c,
Pascal Fraizy b, c, Paulo Roberto Meneses a, Eurides de Oliveira e
a
Instituto de Geociências, Universidade de Brasília, Campus Universitário Darcy Ribeiro, ICC Centro, 70910-900 Brasília, Brazil
b
GET, UMR5563, CNRS/IRD/Université Toulouse 3, 14 Avenue Edouard Belin, 31400 Toulouse, France
c
IRD, UMR 234, CP 7091 e Lago Sul, 71635-971 Brasilia, DF, Brazil
d
ENSEEIHT, INP-ENSEEIHT 2, Rue Charles Camichel, BP 7122, 31071 Toulouse Cedex 7, France
e
Agência Nacional de Águas, Setor Policial Sul, Área 5, Quadra 3, Bloco L, Sala 112, Brasília, DF, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The Madeira River may contribute nearly half of the Amazon River sediment discharge to the Atlantic
Received 11 January 2012 Ocean, showing the highest erosion rates in the Amazon Basin. However, few studies have assessed the
Accepted 15 November 2012 Madeira River sediment budget and the transport processes occurring in the main stem of the river. In
this study, MODIS space-borne sensors were used to analyze the suspended sediment transport
Keywords: processes along the main stem of the Madeira River. Field measurements of suspended sediment
Sediment transport
concentration, spectral radiometry and granulometry were performed during 10 cruises from 2007 to
Madeira river
2011. The relationship between the spectral reflectance and the surface suspended sediment concen-
Remote sensing
Modis
tration (SSSC) was analyzed using both field radiometric measurements and satellite data. Ten-day SSSC
Radiometry samples acquired by the HYBAM monitoring network were used to match satellite observations with
Hydrology field measurements performed from 2000 to 2011. Over 900 MODIS images of 6 different locations were
processed to monitor the SSSC dynamics in space and time. Satellite reflectance was found to be
significantly correlated with the SSSC. However, a seasonal dependency was demonstrated, most likely
caused by a variable granulometric distribution along the annual cycle. The ratio between the red and
near-infrared bands was found to be free of the seasonal dependency (r ¼ 0.79, N ¼ 282), and a SSSC
retrieval model was built from the satellite data using a bootstrap resampling technique. The satellite-
retrieved SSSC time series showed excellent accuracy over the 11-year period and at two different
stations located 800 km from each other. The satellite data were averaged to analyze the SSSC pattern
temporally and spatially along the entire Madeira River, which provided evidence of significant sedi-
mentation and resuspension. The backwater effect caused by the two-to-three-month lag between peak
water in the Madeira and Amazon Rivers was used to predict local sedimentation near the Madeira River
mouth. Our results facilitated a precise assessment of such sedimentation, which demonstrated an SSSC
decrease 400 km upstream from the MadeiraeAmazon confluence.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction on different time scales (Meade, 1994). The size of the basin and the
fact that the basin is moderately affected by anthropogenic activities
Sediment transport is of special significance in large tropical make it a valuable case study for understanding erosion and sedi-
basins, which are known to concentrate approximately 50% of the ment transport and their relation with past and present geomor-
world’s solid fluxes from continental lands to the oceans (Latrubesse phology (Baby et al., 2009) and climate (Martinez et al., 2009).
et al., 2005). Among these basins, the Amazon Basin exhibits high Most studies on the sediment budget of the Amazon Basin have
erosion rates in which most of the sediment is derived from the focused on the Amazon River main stem, where the availability of
Andes and from large floodplains that can store and release sediment hydrological data is better than in other areas of the catchment
(Filizola and Guyot, 2004; Gibbs, 1967; Martinez et al., 2009; Meade
et al., 1979). The Madeira River Basin drains approximately 25% of
* Corresponding author. Tel.: þ 5561 3307 2433; fax: þ5561 3272 4286. the Amazon Basin but may account for nearly half of the sediment
E-mail address: raulev@unb.br (R. Espinoza Villar). fluxes released to the ocean (Guyot et al., 1996). The Madeira

0895-9811/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jsames.2012.11.006

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
2 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10

drainage basin forms the southwestern boundary of the Amazon were analyzed to assess the sediment transport processes along the
Basin. It is limited to the southwest by the Cordillera Oriental, main stream of the Madeira River. Special attention was paid to the
which supplies a large quantity of sediment to the rest of the basin Madeira River mouth, where backwater effects originating from the
and to the northeast by the Brazilian craton (Baby et al., 2009). confluence with the Amazon River cause sedimentation.
Sediment yield data show that the upper basin hosts the most
elevated erosion rates in the Amazon Basin, which are approxi- 2. The study area
mately 3200 t  km2  year1 on average, with large variations
ranging from 50 to 50 000 t  km2  year1 (Guyot et al., 1996). With an area of 6.2  106 km2, the Amazon Basin is the
The total mass of suspended sediment exported from the eastern world’s largest catchment, delivering a water discharge of
Andes and sub-Andean drainage basins in Bolivia, excluding the 6600 km3  year1 (Molinier et al., 1996) and a sediment discharge
Peruvian Madre de Dios subcatchment, has been estimated to be of 800 million t  year1 based on measurements at the last gauged
approximately 500e600 million t  year1 (Guyot et al., 1996). station along the Amazon River that is not disturbed by sea tides
Using a network of sampling stations for river sediment discharge (Martinez et al., 2009). Sediments are transported principally from
assessment in Bolivia, Guyot et al. (1996) estimated that the current the Andes, which surround 12% of the Amazon Basin. One of the
sedimentation in the Madeira foreland basin may be approximately most important tributaries is the Madeira River, which contributes
270 million t  year1. 16% of the Amazon River water budget and approximately 50% of
Sediment transport monitoring is critical in the Madeira River the sediment discharge (Filizola, 1999). According to the Sioli
Basin because the monitoring provides valuable information for classification system (1957), the Madeira River is a white-water
economic activity that is dependent on river flows and sediment river that is rich in dissolved material and suspension solids.
discharge, such as ship transport. Interestingly, the construction of The Madeira River drains an area of approximately 1.4 106 km2
two large hydroelectric power-generation dams in the upper reach and has a mean annual discharge of 32 000 m3  s1 (Molinier
of the Madeira River, upstream of Porto Velho, may affect the natural et al., 1993). This river is among the ten largest rivers in the
sediment transport. Concerns about the impact of these dams and world and was classified as a mega river (Latrubesse, 2008). The
the future behavior of sediment discharge as a function of regional Madeira River is formed at the confluence of the Beni and Mamore
climate change (Espinoza Villar et al., 2009) require the develop- Rivers near the border between Brazil and Bolivia. These rivers are
ment of robust and cost-efficient monitoring methods adapted to of Andean origin and present high sediment loads of approximately
large rivers. It has been shown that the optical qualities of water are 690 and 280 mg  l1 at the river surface, respectively (Guyot et al.,
closely linked to certain quality parameters, such as turbidity 1996). The lower load of the Mamore River results from low sedi-
(Dekker et al., 2002; Mertes et al., 1993), algal pigment (Gohin et al., ment load tributaries that drain the lowlands (e.g., the Guapore
2002; Schalles et al., 1998) and organic matter (Vodacek et al., 1995). River) and strong sedimentation processes in the Andean Piedmont
Recent studies on the Amazon Basin have shown that space-borne (Guyot et al., 1996). Downstream from the Beni-Mamore conflu-
global monitoring sensors can be used to efficiently monitor river ence, all of the tributaries are characterized by low sediment
sediment discharge along the Amazon River (Martinez et al., 2009) concentrations because they drain the Amazonian lowlands on the
(Espinoza et al., 2012). In this study, we analyzed the optical prop- Brazilian Shield (Fig. 1). In Brazil, most of the tributaries from the
erties of the Madeira River waters to develop a surface suspended Abunã River down to the Aripuanã River confluence flow into the
sediment concentration (SSSC) retrieval algorithm based on MODIS Madeira River from its right bank as the Madeira River nears the
satellite data. The SSSC estimates acquired over an 11-year period Purus River catchment.

Fig. 1. A map of the Madeira River and the major tributaries showing the water-gauge stations along the river main stem maintained by the Brazilian Water Agency. The relief data
were extracted from the SRTM digital elevation model.

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10 3

The Madeira River sediment budget estimates show significant at the ABU, PV, HUM, MAN and FVA stations for the studied period.
variations from one study to another because some studies use We note well-defined monomodal hydrograph with peaks in
sediment discharge-water discharge relationships, whereas others MarcheApril and low water between August and September. At the
are based on a network of stations at which the river water was FVA station, the maximum water discharge recorded over the
sampled for particle concentration assessment but with different period is 60 000 m3  s1, the minimum is 3500 m3  s1, and the
sampling frequencies or sampling methods (i.e., integrated vs. average discharge is approximately 27 000 m3  s1.
surface sampling). Based on a limited set of sampling field The water discharge increase between the ABU and FVA stations
campaigns, Gibbs (1967) reported suspended sediment yields of is 62% on average with the inputs of the tributaries. The increase is
approximately 217  106 t  year1 at the mouth of the Madeira relatively larger in May, up to 100%, and smaller between October
River in the Amazon. Subsequent studies reevaluated the previous and December, down to 20%. For the main tributary of the Madeira
estimates and calculated a yield of 550 million t  year1 (Ferreira River, the ANA data show that the annual mean water discharge is
et al., 1988; Martinelli et al., 1993). In Bolivia, the results obtained 700, 1400, 1000 and 3300 m3  s1 for the Abunã, Ji-Parana, Mar-
by the Climatological and Hydrological Program of the Bolivian melos and Aripuana Rivers, respectively. These rivers have similar
Amazon basin (PHICAB) for the Upper Madeira Basin from 1983 hydrological cycles with a high water level between February and
to 1990 showed that the Beni and Mamore Rivers contribute a sus- April and a low water level in September, similar to the behavior of
pended sediment yield of 223 million t  year1 to the Madeira River the main stem of the Madeira River.
(Guyot et al., 1993; Roche and Fernandez Jauregui, 1988). Using Comparisons of water discharge data from the tributaries and
the quarterly integrated suspended sediment concentration data the river main stem show inconsistencies that may result from
from the Brazilian Water Agency, Guyot et al. (1996) calculated insufficiently accurate water level-river discharge relationships.
the sediment discharge at the Porto Velho station to be 320 The main problem occurs at the MAN station, where the flow is
million t  year1, which is significantly higher than the value much higher (i.e., 25% from June to December) than measured at
directly assessed upstream. Finally, using 10 years of sampling the next downstream station on the Madeira River, FVA.
records from the HYBAM network, Guyot et al. (2010) determined The large volume of the Amazon River affects the Madeira River
a sediment discharge of 411 million t  year1. However, the authors water level in the lower courses of the tributary, producing what is
suggest that the sediment discharge-water discharge relationships known as the backwater effect. This effect is amplified by the two-
they used for the Brazilian data should be disregarded for the to-three-month lag between peak water in the Madeira and
Madeira Basin because the relationships show strong dispersion. Amazon Rivers, resulting in a hysteresis effect in the stage-
discharge relationship, as shown in Fig. 3 for the FVA station.
3. Data acquisition and method While the backwater effect in the HUM station records is not
measurable, it can be detected downstream at the MAN station
3.1. General Hydrology (Meade et al., 1991) and becomes significant at the FVA station,
where the water level varies between 2 and 3 m for the same
Five water-gauge stations providing river-flow estimates are discharge amount, depending on the hydrological period.
currently maintained by the National Water Agency (ANA) along
the Madeira River in Brazil. Starting from the Bolivia-Brazil border, 3.2. The suspended sediment from the monitoring networks
the stations are Abunã (ABU; upstream of the Abunã River
confluence), Porto Velho (PV), Humaita (HUM), Manicore (MAN) The ANA protocol requires quarterly measurements of the sus-
and Fazenda Vista Alegre (FVA) (see www.ana.gov.br/hydroweb). pended sediment concentration at five stations in its network using
Near-real-time data are available at the PV and FVA stations, the depth integration method (using integrating USD-49 samplers).
whereas at the other stations, the river water level is measured However, the data are scarce for the 2000e2010 period. The
twice a day. The water level was calibrated to the water discharge HYBAM network maintained by the ANA and the Institut de
using the mechanical current-meter method at all of the stations Recherche pour le Développement (IRD) at the PV and FVA stations
except for the PV and FVA stations, where acoustic doppler current has been providing 10-day samples from the river surface since
profiler (ADCP) measurements were also used (Filizola and Guyot, 1995. To obtain the HYBAM data (see www.ore-network.org),
2004). Fig. 2 shows the Madeira River water discharge time series a local operator samples the SSSC every 10 days at a fixed point. The

Fig. 2. The monthly water discharges for the ABU, PV, HUM, MAN, and FVA stations along the main stem of the Madeira River recorded during 2000e2009, which were provided by
the Brazilian water agency.

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
4 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10

The suspended sediment samples were processed using the same

protocol as that used for the 10-day HYBAM measurement network.
The measurements were performed with three TriOS-RAMSES
hyperspectral spectroradiometers, two measuring radiance and
one measuring downwelling irradiance. The instruments were
mounted on a steel frame that was fixed to the prow of the ship,
facing forward to minimize ship shadow and reflection. A cosine
irradiance sensor detected the incident daylight. One radiance
sensor detected the reflected light from the upper water column at
a nadir angle of 30 , and another measured the diffuse radiance
originating from the region of the sky that reflects into the water-
viewing radiance sensor. The acquisition geometry recommended
by Mobley (1999) from numerical modeling was followed. The
results are provided as the remote-sensing reflectance, which is the
ratio of reflected light from the upper water column (upwelling
radiance) to incident light from the sky (downwelling solar irra-
diance). Field reflectance data were used to simulate the MODIS
surface reflectance in the red and near-infrared (NIR) channels.
Fig. 3. The relationship between water level and discharge at the FVA station, the
last water-gauge station maintained by the Brazilian Water Agency before the 3.4. Satellite images
AmazoneMadeira River confluence. The measurements were performed with the
mechanical current-meter method or the acoustic Doppler current profiler.
The Collection 5 atmospherically corrected surface reflectance
products from the Terra and Aqua MODIS space-borne sensors were
sample bottles are stored and collected every 6 months to be pro- used in this study. The MODIS data products MOD09Q1 (Terra on-
cessed at a laboratory at Amazonas State University in Manaus, board sensor) and MYD09Q1 (Aqua on-board sensor) provide
Brazil. The samples are filtered using 0.45-mm cellulose acetate calibrated reflectance for two radiometric bands measured at a 250-
filters (Millipore) which are previously dried for 24 h at 60  C and m pixel resolution while offering near-daily time coverage over
weighed. After filtration, the filters are dried for 24 h at 60  C and tropical areas (http://modis.gsfc.nasa.gov). Bands 1 and 2 are
weighed again. The suspended matter concentration is computed centered at 645 and 858.5 nm, respectively (Fig. 5). The MODIS
as the difference between the weight of the filter after and before surface reflectance 8-day composite data were acquired for March
filtration divided by the volume of water collected. In this study, we 2000 to July 2011 from the NASA Warehouse Inventory Search Tool
will use the HYBAM data, provided that this dataset furnishes (WIST) data gateway. We chose composite images for three
surface estimates that match what the satellite measures (because reasons: i) the 8-day composite images are compatible with the 10-
sunlight penetrates only a thin surface layer of the water column) day field measurement sampling frequency; ii) the 8-day composite
and provided that the HYBAM data enable access to SSSC temporal images reduce the quantity of data to be analyzed because persis-
variability at both intra- and inter-annual scales. tent cloud cover renders a large number of daily images unusable;
The SSSC time series from the HYBAM database exhibit and iii) the 8-day composite images significantly reduce the
a behavior similar to that of the hydrological cycle, with an increase directional reflectance effects and atmospheric artifacts. For each
in the SSSC during the rising-water period and decreasing values date, the composites from Terra and Aqua are automatically scan-
down to the low-water period. However, high SSSC values usually ned, and the image with the lowest cloud cover is selected. When
occur from December to February, one or two months prior to the both composites exhibit low cloud cover, the composite acquired
flood peak, highlighting that a simple sediment discharge-water with the lowest satellite viewing angle is preferred.
discharge relationship may not be applicable to the Madeira River The retrieval of river stream reflectance using the MODIS data is
(Filizola and Guyot, 2009). The sediment discharge peak for the hampered by the low spatial resolution that may result in few
Madeira River at Fazenda Vista Alegre occurs in March, whereas the pure (non-mixed) water pixels, depending on the river width and
water discharge peak occurs between April and May (Filizola and image acquisition geometry. Spectral mixing has been described
Guyot, 2009). The SSSC exhibits a strong variation from 1.1 to
1950 mg  l1 at the PV station and between 4.5 and 823 mg  l1
downstream at the FVA station. The SSSC exhibits a downstream
decrease. At the PV and FVA stations, the SSSC averages 440 and
202 mg  l1, respectively. Similar values were reported by Guyot
et al. (1996), who analyzed ANA data acquired from 1978 to 1994
and found an average concentration of 483 mg  l1 at PV (23
samples) and 181 mg  l1 at FVA (35 samples). Considering the
HYBAM data, the mean rate of SSSC decrease between both stations
over one year is 35%. However, this decrease varies with the season,
being higher in July (51%) and lower in October (6%).

3.3. Field-sampling campaigns

From 2007 to 2011, hyperspectral radiometric measurements

were performed using spectroradiometers during ten cruises that
were conducted to acquire field data along the Madeira River at Fig. 4. The particle volume distribution for surface samples collected at the HUM and
different hydrological periods. During these cruises, spectral MAN stations on two cruises during the rising-water stage (November) and flood peak
reflectance data and water samples were collected simultaneously. (April).

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10 5

The measurements consisted of the simultaneous recording of

spectroradiometric data and surface-water sampling. Three field
campaigns were conducted to visit most of the stations in the
downstream direction (i.e., the PV station to the FOZ station) in
November 2009, MarcheApril 2010 and July 2011. The stations
more easily accessible from Manaus (i.e., the FOZ and FVA stations)
were visited more often, namely up to 10 times. At the FVA station,
multiple sampling was performed during 3 campaigns. The 40 SSSC
records show a large range of values, from 25 to 622 mg  l1.
Particle-size distribution analyses were performed on 5-liter
surface-suspended sediment samples collected during the 2009
rising-water cruise and the 2010 flood-peak cruise. Granulometric
data were obtained using a laser grain-size measuring device (a
Mastersizer 2000 with a sample dispersion unit) at the CPRM
Laboratory for Sediment and Water Quality. Fig. 4 shows the grain-
size distribution at the HUM and MAN stations during the 2009 and
2010 cruises. Fine-grained silt dominated the surface samples (99%
Fig. 5. The remote-sensing reflectance of the Madeira River water measured during of the grains are finer than 63 mm). However, the distribution
three field sampling campaigns in November 2009 (rising water), April 2010 appeared to shift toward larger sizes during the flood peak. The
(flood peak) and July 2011 (low water). The gray curves denote the spectral response median particle-size distribution (D50) was 9 mm in April and 5 mm
functions of the MODIS red and near-infrared bands.
in November at the MAN station, whereas the D50 decreased from
7 mm in April to 5 mm in November at the HUM station.
extensively in the literature and occurs when different materials
are present in the same pixel. In the context of this study, spectral 4.2. The optical properties of the Madeira River
mixing may occur between water, riverine vegetation or sand
banks. Accordingly, specific algorithms have been developed that The analysis of the spectroradiometric measurements enables
enable the derivation of the water endmember reflectance in each the registering of the variations of the Madeira River spectra at
image. First, river pixels are partitioned into homogeneous clusters different periods during the hydrological cycle. Fig. 5 shows the
using the K-means algorithm (Martinez et al., submitted, 2009). remote-sensing reflectance as a function of wavelength at 3
Next, the fraction of each endmember in each cluster is obtained by different periods at the PV station. The sediment-loaded Madeira
applying a least squares technique to minimize the un-modeled River waters present low reflectance values in the blueegreen
residual. The testing of every cluster as a possible pure water spectra, followed by high reflectance values between 560 and
endmember is performed to solve the set of linear equations. The 705 nm and a strong decrease in the reflectance at longer wave-
cluster leading to the lowest residual is retained as the water lengths. The reflectance level seems to be directly related to the
endmember (Martinez et al., submitted). SSSC, with higher values occurring during the rising-water period,
The MODIS surface reflectance was retrieved for six different particularly in the red and infrared parts of the spectra. Fig. 5 shows
locations along the main stem of the river (Fig. 1) to create 11-year the location of the MODIS red and NIR bands. The MODIS bands
time series. For each station, 936 8-day composites were processed match the parts of the spectra where the reflectance shows greater
from the MODIS Terra and Aqua sensors. The locations match the sensitivity as a function of the SSSC.
five ANA water-gauge stations. An additional virtual station (i.e., the The low values in the blue and green parts of the spectra are
FOZ station) was added to analyze the SSSC near the Madeira River characteristic of the light absorption by CDOM and suspended
mouth at the confluence with the Amazon River. Data from the PV sediment that is known to decrease exponentially from ultraviolet
and FVA stations were used to analyze reflectance-SSSC matchups wavelengths toward higher wavelengths (Kirk, 1994). In the red
with the HYBAM network data, whereas the other stations enabled part of the spectrum, light scattering and absorption are controlled
the study of the SSSC behavior in space and time. by the concentration of inorganic suspended matter, and the
reflectance reaches its highest value, which seems to depend on the
4. Results TSS load. The contribution of pure water absorption increases
rapidly in the near infrared (NIR) region of the spectrum, which
4.1. The SSSC and granulometry results in a significant decrease in reflectance beyond 705 nm.
Although the concentrations in November 2009 and April 2010
Table 1 summarizes the measurements acquired during the 10 were similar (i.e., a 5% variation), there is a significant difference
field-sampling campaigns at each station along the Madeira River. between both spectra, indicating a possible seasonal dependence.
Additionally, the light attenuation in the water column was
measured by assessing downwelling irradiance as a function of
Table 1 depth in the euphotic zone. The data show that the light penetra-
Field data in Madeira River from 2007 to 2011 and suspended sediment rangue data tion in the water column in the red and NIR is directly linked to the
for three hydrological periods. SSSC, with the euphotic zone varying between 20 cm and 1 m. This
Stations Sample number SSSC range (mg/l) result confirms that the riverbed does not affect the water’s spectral
Low water Peak water Rising water Total
response because the main channel is on average more than 15 m
deep.
ABU 1 1 105
PV 2 1 3 6 106e401
HUM 2 1 3 110e286 4.3. The seasonal control of the SSSC-reflectance relationship
MAN 1 1 1 3 61e418
FVA 4 3 10 17 43e300
The field spectroradiometric measurements collected along the
FOZ 4 2 4 10 25e622
Madeira River were used to simulate the MODIS radiometric band

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
6 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10

month during the 2000e2011 period as a function of the average

SSSC over the same period. The error bars indicate the standard
deviation of the reflectance and of the SSSC of all of the samples
collected during each month. The data are plotted for the PV and
FVA stations. The seasonal dependency is obvious, with a stronger
reflectance during the rising-water period (from October to
January) than during the flood-peak period (FebruaryeApril). The
SSSC-reflectance relationship seems to vary similarly for both
stations. However, the effect seems to be less pronounced down-
stream following the sediment dilution by the sediment-poor
waters from the lowland tributaries between the PV and FVA
stations. Separate studies performed on the Amazon River did not
show this seasonal pattern, neither upstream in the Peruvian
Amazon River tributaries (Espinoza et al., 2012) nor downstream
near the Amazon River mouth at the Obidos station (Martinez et al.,
2009).
Fig. 6. The NIR MODIS band simulated from field radiometric measurements as A significant number of studies have been conducted concern-
a function of the SSSC, which was assessed during 10 sampling campaigns. The ing SSSC and remote-sensing reflectance, mostly over oceanic and
numbers denote the sampling month.
coastal waters. Forget et al. (1999) showed that the way in which
the light is scattered and absorbed by inorganic suspended sedi-
behavior as a function of the SSSC using the MODIS spectral ment is a function of particle concentration, size distribution and
response functions. Fig. 6 shows the simulation of the MODIS near- the refraction index. Therefore, for the same SSSC level, a difference
infrared band as a function of the SSSC, where each point is labeled in the measured reflectance may indicate a differential suspended
with the month of collection. Samples were collected during 10 sediment type. Guyot et al. (1999) measured the suspended sedi-
field sampling cruises that covered a large range of SSSCs (i.e., 25e ment grain size from the Andes downstream near the Benie
622 mg  l1). Low concentration or reflectance values correspond Mamore confluence. They showed that, downstream of the
to the low-water period (MayeOctober), whereas higher values Andean Piedmont where strong sedimentation processes occur for
were found during the rising-water and flood-peak period the larger grains, the suspended sediment granulometric distri-
(NovembereApril). The SSSC-remote-sensing reflectance relation- bution exhibits reduced variability at different flood stages. Our
ship appears to be linear for most of the SSSC range except during results suggest the same behavior but with certain slight variations
the rising-water period (i.e., the November samples). For similar between the hydrological phases. The silt-dominated suspended
SSSC levels, the November samples exhibit a significantly higher sediment exhibited a finer distribution during rising water than
reflectance than the flood-peak samples (principally, the March during the flood-peak stage. Analyzing the spectral reflectance
and April samples). This difference confirms our analysis in the from different soil textures, Han and Rundquist (1996) showed that
previous section, where the reflectance in the red and NIR part of as SSSC increased, the reflectance from the finest particles
the spectrum was much higher in November than in April for the increased faster than that from the coarser soil. Additionally,
same range of SSSC. However, the limited number of field radio- mineralogical differences in the suspended sediment may exhibit
metric samples makes a full assessment of this seasonal depen- different reflectance responses because the river flow may be
dency difficult. dominated by several different upstream sources. Guyot et al.
We now analyze the SSSC-near-infrared reflectance matchups (2007) suggested that the sediment from the Andes is enriched in
between the MODIS images and the HYBAM network samples to smectite as the river flows through the lowlands during the flood
improve our understanding of the relationship variation for the peak. Analyzing the HYBAM samples acquired during one hydro-
entire SSSC range and in all of the periods of the hydrological cycle. logical cycle, Viers et al. (2008) found that the Solimões and the
Fig. 7 shows the average reflectance derived from MODIS for each Madeira Rivers exhibit distinct suspended sediment chemical
compositions, although the same researchers did not detect
seasonal variations in the mineralogical composition of the rivers.
Nevertheless, the same authors found significant strontium
isotopic composition variations over time in the suspended sedi-
ments, with a difference between the high-water period and the
remainder of the year. This seasonal variation in the Sr isotopic
composition remains unclear and may be caused by increasing
physical weathering during the rainy season, i.e., when landslides
and riverbank erosion may facilitate the input of radiogenic sedi-
ments not easily mobilized during periods of low water. These
observations highlight a limited but significant variability in sus-
pended sediment type and size distribution. This variability may be
used to explain the seasonal dependency of the reflectance over
time. However, more research is necessary, using optical modeling,
to fully assess the phenomenon.

4.4. The satellite-based SSSC retrieval model

Fig. 7. The MODIS surface reflectance in the NIR as a function of the SSSC from the
HYBAM monitoring network. The black circles represent the PV station data, and the
gray circles represent the FVA station data. The measurements represent monthly
Light scattering by inorganic particles varies little as a function of
average values, and the error bars indicate the standard deviation for each month. The wavelength. However, the light absorption decreases exponentially
numbers denote the months from January (1) to December (12). toward higher wavelengths (Kirk, 1994). Therefore, the optical

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10 7

properties of the suspended sediment may vary from one part of the values, with a ¼ 1020  150 (95% confidence interval) and
spectrum to another. Different authors (Doxaran et al., 2002; b ¼ 2.94  0.37 computed using 1000 bootstrap samples.
Froidefond et al., 2002; Topliss et al., 1990) have proposed that the
use of the ratio of the reflectance in at least two different channels 4.5. SSSC time series from satellite data vs. those from conventional
should reduce the overall sensitivity of the remote-sensing reflec- networks
tance to the inorganic particle type. We computed the ratio between
the NIR and the red MODIS bands for the entire dataset. Fig. 8 shows The SSSCs were retrieved from the MODIS 8-day time composite
the matchups between the MODIS NIR/red surface reflectance ratios time series using the model defined in the previous section at the
and the HYBAM samples from the PV and FVA stations. The PV and FVA stations. Fig. 9 compares the satellite-derived and
matchups were selected when the delay between the MODIS network SSSC samples at PV and FVA over the 2000e2011 period.
acquisitions and the field samples was less than 4 days. The SSSC Good agreement was found, which shows that satellite data enable
exhibited a wide range, from 4 to 1832 mg  l1 with a mean value of the monitoring of the temporal variability regardless of cloud cover,
300 mg  l1, for 282 field samples. There is a significant positive even when the cloud cover is persistent during the rainy period.
relationship (r ¼ 0.79, p < 0.05) between the SSSC and the reflec- Over the common period of analysis between February 2000 and
tance ratio. The determination coefficient shows that 61% of the March 2009, MODIS 8-day composites enabled the calculation of
variation in the reflectance ratio is explained by the variation in the 372 SSSC estimates at the PV station, representing a 90% time
SSSC. Most of the residual dispersion may be attributed to the coverage, which is the same percentage that the HYBAM 10-day
differences in the spatial scale between the remote-sensing data, network achieved with 296 samples. At the FVA station, the
averaged over tens of pixels and representing several square kilo- MODIS coverage outperformed the HYBAM network by more than
meters, and the field samples acquired at the riverside in 250-ml 10%, with an 81% time coverage over the 2000e2009 period. These
bottles. Additional dispersion may be caused by the time delay results demonstrate the potential use of satellite time series to
between the sampling and the closest cloud-free satellite image, study the spatial and temporal variability in sediment transport.
which rarely matches the exact day of sampling. Finally, natural The PV and FVA stations are located 880 km from each other but
SSSC heterogeneity across the river reach is likely to increase the exhibit similar temporal behavior. However, we noted a significant
dispersion in the relationships between the satellite and field SSSC decrease from upstream to downstream, which we will
samples. To provide a better view of the impact of these different analyze in the next section.
dispersion factors, Fig. 8 shows the reflectance ratio-SSSC matchups
based on the field spectroradiometric measurements that should be 4.6. An analysis of the SSSC seasonal pattern along the Madeira
free of most of the scale problems, provided that the spectroradio- River based on MODIS images
metric measurements and water sampling were performed simul-
taneously from a boat. For these field spectroradiometric The sediment concentration at the virtual stations was calcu-
measurements, 40 matchups are available, with an SSSC variation of lated using MODIS 8-day composites. The SSSC estimates were
25e622 mg  l1 and a mean value of 198 mg  l1. The correlation averaged for each month of the year, enabling study of the sea-
factor is 0.96 and 91% of the variation in the field spectroradiometric sonality of sediment transport along the main stem of the river
measurements are explained by the SSSC variations. during 2000e2011. Fig. 10 shows the variation of the SSSC along the
The best-fit model was assessed using least squares regression. Madeira River from the ABU station downstream to the Madeira
To assess the variability in the model parameters, we used the River mouth at the FOZ station in 3 periods: rising water, flood peak
bootstrap resampling technique (Wehrens et al., 2000). This tech- and decreasing water. The SSSC exhibited a decreasing trend from
nique proves particularly useful in cases where the sample data are upstream to downstream in all of the periods. However, strong
limited. Its main uses are the estimation of accuracy measures and spatial variations were detected from one part of the seasonal cycle
the construction of confidence sets. Least squares regression shows to another. For the rising-water period (October to January), we
that a power law model, SSSC ¼ a. (RNIR/RRed)b, provides the best-fit noted the following: 1) a consistent increase in the SSSC level as the
water level rose and 2) a sharp SSSC decrease from upstream to
downstream. The SSSC decrease was considerable, particularly in
the upper reaches (i.e., between the ABU, PV and HUM stations),
with a mean decrease of approximately 30% between ABU and PV
from October to January. Downstream, the SSSC decrease was weak,
with a diminution of 4% between the FVA and FOZ stations in
December and January.
During the flood peak, from February to May, the overall SSSC
level was lower and the relative decrease was reduced compared
with the rising-water period. The concentration diminished by
approximately 12% between the ABU and PV stations on average
from February to May. The strongest SSSC decrease occurred
between the PV and HUM stations, with a reduction of approxi-
mately 26% on average over the same period. Downstream from the
PV-to-HUM reach, the SSSC decrease was lower, with a reduction of
approximately 22% between the HUM and FVA stations. In the last
Fig. 8. The NIR/red surface reflectance band ratio as a function of SSSC. The black dots reach, between the FVA and FOZ stations, the decrease was on
represent matchups between field spectroradiometric measurements and simulta- average 8%. During the low-water period (i.e., June to September),
neously acquired water samples (N ¼ 40) from 2007 to 2011. The spectroradiometric the SSSC level was much lower along the entire Madeira River, and
measurements were convolved to match the MODIS radiometric bands. The gray dots the SSSC variation along the river main stem was reduced
represent matchups between the MODIS data and the HYBAM SSSC data at the PV and
FVA stations during 2000e2009 with a 4-day acquisition maximum delay criterion
upstream. The SSSC decrease between the ABU and FVA stations
(N ¼ 282). The model was fitted for satellite data using least squares regression with was on average 51% between June and August and 28% in
1000 bootstrap samples. September. However, downstream from the MAN station, the SSSC

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
8 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10

Fig. 9. SSSC time series retrieved from 8-day composite MODIS images (black dots) and 10-day HYBAM samples during 2000e2011 (white dots) at the PV station (above) and the
FVA station (below). The error bars indicate the standard deviation of the satellite-derived SSSC estimate.

decrease was strong; namely, there was a 50% decrease from the a pronounced SSSC decrease from PV to FVA, followed by stabili-
MAN to FOZ stations from June to August. This decrease was the zation downstream at the FOZ station in January.
strongest for that reach of the river in all of the periods; namely, In the Madeira catchment, previous studies have shown that the
from June to August, the SSSC decrease between the FVA and FOZ SSSC is closely related to the average concentration in a river reach
stations was 22%, whereas the decrease was limited to 5% on (Guyot et al., 1996). Therefore, the SSSC data can be used as a proxy
average from October to May. to detect sedimentation or resuspension along the main stem of
a river. Our satellite-derived SSSC climatology data enable the
5. Discussion definition of three different regions exhibiting distinct sediment
transport processes.
5.1. Hydro-sedimentary processes along the Madeira River
1) Upstream, from the ABU station to the PV station, where the
The Madeira River sediment budget may include different water discharge inputs from the local tributaries are limited,
inputs, such as bed load, suspended sediment load, sediment from there is a sedimentation zone during the rising-water period
tributaries and bank erosion. The suspended sediment load is with a 30% SSSC decrease from October to January (Fig. 9a),
delivered almost entirely by the Andean tributaries, whereas the while the Madeira River discharge exhibits a very low increase
local tributaries drain mostly forested lowlands with reduced (þ1% from the ANA data). For the remainder of the period, the
erosion. From the quarterly suspended sediment concentration SSSC decrease is consistent with the river-water discharge
data acquired in Brazil between 1981 and 1994, Guyot et al. (1996) increase resulting from the inputs of the low sediment-load
estimated that the Ji-Parana and Aripuana sediment discharge tributaries (Abunã, Jaci parana).
masses may not exceed 1.2 million and 2.9 million t  year1, 2) Between the PV and FVA stations, dilution and sedimentation
respectively, representing approximately 1% of the Madeira River appear to occur during the rising-water period; the SSSC
sediment discharge that the same authors assessed at the PV decreases by 53%, whereas the water discharge increases by 17%.
station. The measurements performed during our field sampling However, resuspension is likely to occur during the flood peak
trips suggest that this budget may not have changed since then. from February to May; the SSSC decreases by 42%, while the
Few studies have been conducted on riverbed transport in the water discharge increases by 59%. In April, when the Madeira
Amazon. However, studies on the Amazon River main stem suggest River discharge is strongest, approximately 53 000 m3  s1 on
that this transport varies depending on the season but may not average at the FVA station, the SSSC decrease between the PV
exceed more than 5% of the total sediment budget of the river and FVA stations is 34%, whereas the river discharge increases
(Mertes, 1985; Strasser et al., 2004). by 78% with the inputs of the Marmelos and Aripuanã Rivers.
The SSSC decrease assessed by satellite data agrees with our During the period of decreasing water, resuspension continues
field observations and with previous results, such as those obtained but at a slower rate than during other periods.
by Martinelli et al. (1993) during two cruises from PV to FOZ in 3) Downstream, the backwater effect caused by the Amazon River
January 1986 and April 1984. Although these authors used a depth- water (Meade et al., 1991) has been predicted to produce
integrating sampler, which prevents the direct comparison of the temporal sedimentation as the slope of the water line dimin-
SSSC levels, the trends obtained from the satellite data and from ishes near the Madeira-Amazon confluence (Martinelli et al.,
these field measurements exhibit similar behavior. Both datasets 1993). The systematic satellite observations demonstrate the
demonstrate a much higher SSSC level in January than in April and following: 1) the backwater effect is significant from May to

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10 9

measurements and satellite observations to be closed. However,

this topic remains beyond the scope of the present study.

5.2. Remote-sensing monitoring

Although there is a limited amount of research on the remote-

sensing monitoring of the Amazon Basin waters, the advantage of
using satellite data for such a task has been foreseen for more than
3 decades. Bradley et al. (1979) reported the first study to
demonstrate the possibility of detecting the main water type in the
Amazon Basin from Landsat imagery. Several years later, Mertes
et al. (1993) estimated sediment concentration using spectral
mixture analysis of Landsat images. In that study, spectral end-
members were calibrated from a set of spectroradiometric labora-
tory measurements of Mississippi River water samples. Mertes et al.
(1995) proposed a qualitative retrieval algorithm to identify five
types of water depending on the concentration of sediments in
three Landsat scenes and at three different locations along the
Amazon River but without quantitative retrieval. More generally,
a significant quantity of research exists on the determination of the
optical properties of suspended sediment in oceanic and coastal
waters. Forget et al. (1999) presented an analytical reflectance
model showing the dependency of the optical properties on the
granulometry, mineralogy and concentration of turbid coastal
waters. However, such a model cannot be inverted for operational
retrieval because the model depends on a large set of input data
that are difficult to measure in different locations and periods.
Instead, the use of empirical or semi-analytical algorithms derived
from the simplification of the radiative transfer theory applied to
the interaction of sunlight and a water column has been proposed
(Gordon et al., 1983). Using SPOT satellite data for the Gironde
estuary, Doxaran et al. (2002) used the empirical relationship
between the visible SPOT bands of the near-infrared and red
channels to reduce the effects of sediment type (grain size and
refractive index) on reflectance. Nechad et al. (2010) developed
a semi-analytical generic model to retrieve the SSSC in coastal
waters from superspectral ocean color sensors. In contrast to
Fig. 10. The satellite-derived monthly average SSSC assessed along the Madeira main
previous studies on the Amazon Basin, our work uses the latest
stem from the Abunã station to the river’s mouth at the FOZ station. The data are
presented for 3 hydrological periods: a) rising water (September to January); b) peak available satellite data that enables both global and frequent
water (February to May); and c) low water (June to August). The distances were monitoring. Following the results presented by Martinez et al.
calculated from the ABU station. The monthly averages were computed over the (2009) and Espinoza (2012), the present study confirms that the
2000e2011 period.
MODIS time series enables robust monitoring of the SSSC and
facilitates the integration of this information in conventional
August, with a 22% decrease in the SSSC in the last reach of this monitoring networks.
river; and 2) sedimentation affects river reaches far from the
confluence, more than 400 km upstream between the MAN 6. Conclusions
and FVA stations, where the SSSC decreases by 36% during the
months of June, July and August. The SSSC increases slightly The relationship between optical properties and suspended
between the FVA and FOZ stations in September and October sediment was investigated using field spectroradiometric
by approximately 12%. This increase can occur because of the measurements and remote-sensing images. Remote-sensing
resuspension of recently deposited sediments during the early reflectance proved to be a robust predictor of SSSC. However,
stage of a flood, when the Amazon River reaches its low level, a seasonal control was detected that most likely resulted from the
generally in October. differential granulometric distribution of the suspended sediment.
Using the red and near-infrared reflectance channels to retrieve
To determine a precise budget of the sedimentation and resus- SSSC, a retrieval model was developed that demonstrated precise
pension occurring along a stream, accurate estimates of each trib- accuracy with field spectral radiometry (r ¼ 0.95 e N ¼ 40) and
utary flow near its confluence with the Madeira River and a vertical satellite data (r ¼ 0.78 e N ¼ 282). Time series of surface samples
distribution model for suspended sediment concentration in the from the HYBAM network were compared at two stations and
water column are necessary. However, the number of gauge showed excellent agreement. The 11-year MODIS time series were
stations providing river flow information is reduced along the main processed at six different locations along the Madeira River to
stem, with a mean between-station distance along the stream of analyze the monthly average SSSC at each station. The statistics
approximately 200 km. Furthermore, only three tributaries are enabled the identification of two depositional areas, namely the
gauged for river flow, and their stations are, in two cases, remote upper reach of the Madeira River in Brazil and an area near the
from the Madeira confluence (i.e., the Aripuana and Ji-Parana Madeira River mouth in a location influenced by the backwater effect
Rivers). Water-flow modeling may enable the gap between field of the confluence with the Amazon River. This study demonstrates

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006
10 R. Espinoza Villar et al. / Journal of South American Earth Sciences xxx (2012) 1e10

the utility and robustness of remote-sensing data for sediment- Guyot, J.L., Filizola, N., Laraque, A., Seyler, P., 1999. La variabilité saisonnière des flux
sédimentaires dans le bassin de l’Amazone, Conference on hydrological and
transport monitoring. In particular, this new tool may be useful for
geochemical processes in large-scale river basins. Manaus, Brazil.
the monitoring of future changes resulting from the construction of Guyot, J.L., Jouanneau, J.M., Soares, L., Boaventura, G.R., Maillet, N., Lagane, C., 2007.
two large hydroelectric dams along the Madeira River. Clay mineral composition of river sediments in the Amazon basin. Catena 71,
340e356.
Guyot, J.L., Vauchel, P., Cochonneau, G., De Oliveira, E., Filizola, N., Fraizy, P., Fuertes,
Acknowledgments O., Martinez, J., Molina, J., Noriega, L., Sondag, F., 2010. A new estimation of the
suspended sediment yields in the Madeira River basin, from the Andes to the
Amazon. AGU Meeting 2010: The Meeting of the Americas. Foz de Iguaçu Aout
This work was supported by the Brazilian Water Agency (ANA), 2010.
CNPq, MCT (the Ministry of Science and Technology), PEC-PG (the Han, L., Rundquist, D.C., 1996. Spectral characterization of suspended sediments
generated from two texture classes of clay soil. International Journal of Remote
Program for Graduate Students Agreement) in Brazil, IRD, the
Sensing 17, 643e649.
“Environmental Changes Observatory in the Amazon Region” Joint Kirk, J.T.O., 1994. Light and Photosynthesis in Aquatic Ecosystems. Cambridge
International Laboratory, INSU/Relief and CNES/TOSCA programs in University Press.
France. Additionally, we extend our appreciation to the HYBAM Latrubesse, E.M., 2008. Patterns of anabranching channels: the ultimate
end-member adjustment of mega rivers. Geomorphology 101, 130e145.
program partners in Brazil, namely the University of Brasilia (UnB), Latrubesse, E.M., Stevaux, J.C., Sinha, R., 2005. Tropicalrivers. Geomorphology 70,
the Amazonas State Federal University (UFAM) and the Mineral 187e206.
Resources Research Company (CPRM). The authors are grateful to Martinelli, L., Forsberg, B.R., Victoria, R.L., Devol, A.H., Mortatti, J., Ferreira, J.R.,
Bonassi, J.A., Oliveira, E., 1993. Suspended sediment load in the Madeira river.
the ship crews who accompanied them during the surveys. Mitteilungen aus dem Geologisch-Paläontologischen Institut der Universität
Hamburg 74, 41e54.
Martinez, J.M., Guyot, J.L., Filizola, N., Sondag, F., 2009. Increase in suspended
References sediment discharge of the Amazon River assessed by monitoring network and
satellite data. Catena 79, 257e264.
Baby, P., Guyot, J.L., Hérail, G., 2009. Tectonic control of erosion and sedimentation Martinez, J.M., Guyot, J.L., Cochonneau, G., Espinoza Villar, R., Filizola, N., Seyler, F.,
in the Amazon Basin of Bolivia. Hydrological Processes 23, 3225e3229. Submitted. Surface water Quality Monitoring of Rivers with MODIS Data e
Bradley, J., Eden, M.J., Rice-Evans, P., 1979. Remote sensing of suspended sediment Application to the Amazon River. Remote Sensing of Environment, submitted.
in Amazonian rivers from Landsat imagery. In: Allen, J.A., Harris, R. (Eds.), Meade, R.H., 1994. Suspended sediments of the modern Amazon and Orinoco
Remote Sensing and National Mapping. The Remote Sensing Society, University Rivers. Quaternary International 21, 29e39.
of Reading, Reading, UK, pp. 110e116. Meade, R.H., Nordin, C.F., Curtis, W.F., Costa Rodrigues, F.M., Do Vale, C.M.,
Dekker, A.G., Voz, R.J., Peters, S.W.M., 2002. Analytical algorithms for lake water Edmond, J.M., 1979. Sediment loads in the Amazon River. Nature 278, 161e163.
TSM estimation for retrospective analyses of TM and SPOT sensor data. Meade, R.H., Rayol, J.M., ConceizÆo, S.C., Natividade, J.R.G., 1991. Backwater effects
International Journal of Remote Sensing 23, 15e35. in the Amazon river basin of Brazil. Environmental Geology and Water Sciences
Doxaran, D., Froidefond, J.M., Lavender, S., Castaing, P., 2002. Spectral 18, 105e114.
signature of highly turbid waters. Application with SPOT data to quantify Mertes L.A.K., 1985. Floodplain development and sediment transport in the
suspended particulate matter concentrations. Remote Sensing of Environment Solimoes-Amazon River, Brazil. MSc thesis. University of Washington.108
81, 149e161. Mertes, L.A.K., Smith, C.T., Adams, J.B., 1993. Estimating suspended sediment
Espinoza, R., Martinez, J.M., Guyot, J.L., Frazy, P., Armijos, E., Crave, A., Bazán, H., concentrations in surface waters of the Amazon river wetlands from landsat
Vauchel, P., Lavado, W., 2012. Integration of field measurements and satellite images. Remote Sens.Environ 43, 281e301.
observations to derive river solid load in poorly monitored basins. Journal of Mertes, L.A.K., Daniel, D.L., Melack, J.M., Nelson, B., Martinelli, L.A., Forsberg, B.R.,
Hydrology 444e445, 221e228. 1995. Spatial patterns of hydrology, geomorphology, and vegetation on the
Espinoza Villar, J.C., Ronchail, J., Guyot, J.L., Cochonneau, G., Filizola, N., Lavado, W., floodplain of the Amazon River in Brazil from a remote sensing perspective.
De Oliveira, E., Pombosa, R., Vauchel, P., 2009. Spatio-temporal rainfall Geomorphology 13, 215e232.
variability in the Amazon basin countries (Brazil, Peru, Bolivia, Colombia and Mobley, C.D., 1999. Estimation of the remote-sensing reflectance from
Ecuador). International Journal of Climatology. http://dx.doi.org/10.1002/ above-surface measurements. Applied Optics 38, 7442e7455.
joc.1791. Molinier, M., Guyot, J.L., Calléde, J., Oliveira, E., GuimarÆes, V., Cudo, K.J.,
Ferreira, J.R., Devol, A.H., Martinelli, L., Forsberg, B.R., Victoria, R., Richey, J.E., Aquino, M.C., 1993. In: Roche, M.A., Bourges, J., Salas, E., Diaz, C. (Eds.),
Mortatti, J., 1988. Chemical composition of the Madeira river: seasonal trends Hidrologia de la cuenca amazonica brasileira, HiBAM. Primeros resultados sobre
and total transport. Mitteilungen aus dem Geologisch-Paläontologischen la cuenca del rio Madeira. PHICAB, La Paz, pp. 155e164.
Institut der Universität Hamburg 66, 63e75. Molinier, M., Guyot, J.L., Oliveira, E., Guimaraes, V., 1996. In: Chevallier, P.,
Filizola, N., 1999. O fluxo de sedimentos em suspensão nos rios da Bacia Amazônica Pouyaud, B. (Eds.), Les regimes hydrologiques de l’Amazone et de ses affluents.
brasileira. Publ. ANEEL, 63 pp. AIHS, Paris, pp. 209e222.
Filizola, N., Guyot, J.L., 2004. The use of Doppler technology for suspended sediment Nechad, B., Ruddick, K., Park, Y., 2010. Calibration and validation of a generic
discharge determination in the River Amazon. Hydrological Sciences Journal 49, multisensor algorithm for mapping of total suspended matter in turbid waters.
143e153. Remote Sensing of Environment 114, 854e866.
Filizola, N., Guyot, J.L., 2009. Suspended sediment yields in the Amazon basin: an Roche, M.A., Fernandez Jauregui, C., 1988. Water resources, salinity and salt yields of
assessment using the Brazilian national data set. Hydrological Processes 23, the rivers of the Bolivian Amazon. Journal of Hydrology 101, 305e331.
3207e3215. Schalles, J.F., Gitelson, A., Yacobi, Y.Z., Kroenke, A.E., 1998. Chlorophyll estimation
Forget, P., Ouillon, S., Lahet, F., Broche, P., 1999. Inversion of reflectance spectra of using whole seasonal, remotely sensed high spectral-resolution data for an
nonchlorophyllous turbid coastal waters. Remote Sensing of Environment 68, eutrophic lake. Journal of Phycology 34, 383e390.
264e272. Sioli, H., 1957. Sedimentation im Amazonasgebiet. Geologische Runsdchau 45,
Froidefond, J.M., Lavender, S., Laborde, P., Herbland, A., Lafon, V., 2002. SeaWiFS data 608e633.
interpretation in a coastal area in the Bay of Biscay. International Journal of Strasser, M.A., Vinzon, S.B., Borba Mascarenhas, F.C., 2004. Transporte de
Remote Sensing 23, 881e904. sedimentos de arrastre no Rio Amazonas., Third LBA Scientific Conference,
Gibbs, R.J., 1967. Amazon rivers: environmental factors that control its dissolved and Brasilia, Brazil.
suspended load. Science 156, 1734e1737. Topliss, B.J., Almos, C.L., Hill, P.R., 1990. Algorithms for remote sensing of high
Gohin, F., Druon, J.N., Lampert, L., 2002. A five channel chlorophyll concentration concentration, inorganic suspended sediment. International Journal of Remote
algorithm applied to SeaWiFS data processed by SeaDas in coastal waters. Sensing 11, 947e966.
International Journal of Remote Sensing 23, 1639e1661. Viers, J., Roddaz, M., Filizola, N., Guyot, J.-L., Sondag, F., Brunet, P., Zouiten, C.,
Gordon, H.R., Clark, D.K., Brown, J.W., Brown, O.B., H Evans, R., Broenkow, W.W., 1983. Boucayrand, C., Martin, F., Boaventura, G.R., 2008. Seasonal and provenance
Phytoplankton pigment concentrations in the Middle Atlantic Bight: comparison controls on Nd-Sr isotopic compositions of Amazon rivers suspended sediments
of ship determinations and CZCS estimates. Applied Optics 22, 20e36. and implications for Nd and Sr fluxes exported to the Atlantic Ocean. Earth and
Guyot, J.L., Quintanilla, J., Calliconde, M., Calle, H., 1993. In: Roche, M.A., Bourges, J., Planetary Science Letters 274, 511e523.
Salas, E., Diaz, C. (Eds.), Distribuci¢n regional de la hidroqu¡mica en la cuenca Vodacek, A., Hoge, F.E., Swift, R.N., Yungel, J.K., Peltzer, E.T., Blough, N.V., 1995. The
amazonica de Bolivia. PHICAB, La Paz, pp. 135e144. in situ and airborne fluorescence measurements to determine UV absorption
Guyot, J.L., Filizola, N.P., Quintanilla, J., Cortes, J., 1996. Dissolved Solids and coefficients and DOC concentrations in surface waters. Limnology and
Suspended Sediment Yields in the Rio Madeira Basin, from the Bolivian Andes Oceanography 40, 411e415.
to the Amazon, Erosion and Sediment Yield: Global and Regional Perspectives. Wehrens, R., Putter, H., Buydens, L., 2000. The bootstrap: a tutorial. Chemometrics
IAHS Publications, Exeter, pp. 55e63. and Intelligent Laboratory Systems 54, 35e52.

Please cite this article in press as: Espinoza Villar, R., et al., A study of sediment transport in the Madeira River, Brazil, using MODIS remote-
sensing images, Journal of South American Earth Sciences (2012), http://dx.doi.org/10.1016/j.jsames.2012.11.006