JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 107, NO. D20, 10.1029/2001JD000740, 2002
Long-term simulations of discharge and floods in the Amazon Basin
Michael T. Coe,1 Marcos Heil Costa,2 Aurélie Botta,1 and Charon Birkett3
Received 13 April 2001; revised 21 August 2001; accepted 18 December 2001; published 23 August 2002.
[1] A terrestrial ecosystem model (integrated biosphere simulator (IBIS)) and a
hydrological routing algorithm (HYDRA) are used in conjunction with long time
series climate data to simulate the river discharge and flooded area of the Amazon/
Tocantins River basin over the last 60 years. Evaluating the results of this modeling
exercise over the entire basin yields three major results: (1) Observations at 121
stations throughout the basin show that discharge is well simulated for most tributaries
originating in Brazil. However, the discharge is consistently underestimated, by greater
than 20%, for tributaries draining regions outside of Brazil and the main stem of the
Amazon. The discharge underestimation is most likely a result of underestimated
precipitation in the data set used as model input. (2) A new flooding algorithm within
HYDRA captures the magnitude and timing of the river height and flooded area in
relatively good agreement with observations, particularly downstream of the
confluence of the Negro and Solimões Rivers. (3) Climatic variability strongly impacts
the hydrology of the basin. Specifically, we find that short (3–4 years) and long
(28 years) modes of precipitation variability drive spatial and temporal variability in
river discharge and flooded area throughout the Amazon/Tocantins River
INDEX TERMS: 1833 Hydrology: Hydroclimatology; 1860 Hydrology: Runoff and
basins.
streamflow; 3322 Meteorology and Atmospheric Dynamics: Land/atmosphere interactions; 9360
Information Related to Geographic Region: South America
1. Introduction
[2] The Amazon/Tocantins River system of South America is the largest river system on the planet. It covers about
6.7 million km2 and transports about 20% of the world’s
river discharge. Although the Amazon Basin is relatively
undisturbed today, rates of land conversion are increasing
rapidly throughout the basin [Nepstad et al., 1999; Skole
and Tucker, 1993; Skole et al., 1994]. Additionally, increasing atmospheric CO2 concentrations threaten to alter the
water budget through changes in temperature and the
physiological responses of plants. Therefore, it is important
to gain a clear understanding of how the Amazon River
system behaves on seasonal to interannual timescales in
order to gauge how future changes may impact the water
budget of the basin.
[3] Water balance and water transport models provide a
means of investigating the water balance of the Amazon
Basin because they are able to derive spatially and
temporally consistent estimates of the energy and water
budget from simple climatological data (such as precip1
Center for Sustainability and the Global Environment, Gaylord Nelson
Institute for Environmental Studies, University of Wisconsin, Madison,
Wisconsin, USA.
2
Department of Agricultural Engineering, Federal University of Viçosa,
Viçosa, Minas Gerais, Brazil.
3
ESSIC, University of Maryland at College Park, Greenbelt, Maryland,
USA.
Copyright 2002 by the American Geophysical Union.
0148-0227/02/2001JD000740$09.00
LBA
itation and temperature). Previous modeling studies of the
Amazon Basin include a study by Vorosmarty et al.
[1989], which used water balance and water transport
models at 1/2 degree spatial resolution to demonstrate
the feasibility of large-scale simulation of the mean discharge and flooding in the Amazon Basin. That study
compared the simulated discharge to observations at six
locations within the basin. Costa and Foley [1997] used a
coupled land surface and water transport model (also at 1/
2 degree spatial resolution) to simulate the discharge of the
basin and compare it to 56 discharge locations throughout
the basin.
[4] Recently, a number of new long time series data sets
for model input and validation have become available. In
addition, more powerful computers have made higher
resolution, time-transient simulations possible. Therefore,
the objective of this study is to simulate the hydrology of
the Amazon River basin at 5-minute horizontal resolution
(about 9 km) and to evaluate the simulations with diverse
data throughout the Amazon River basin. This study is an
extension of previous simulations in the resolution and
complexity of the models used, the time-transient nature
of the simulations, and the spatial extent and diverse range
of data used for evaluation.
[5] To simulate the river discharge and seasonal flooding
throughout the Amazon River system over the last 60 years
we use the integrated biosphere simulator (IBIS) [Kucharik
et al., 2000] and the hydrological routing algorithm
(HYDRA) [Coe, 2000] with long-term mean monthly
climate data provided by the Climate Research Unit of the
University of East Anglia, Norwich [New et al., 2000]. We
11 - 1
LBA
11 - 2
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
validate our simulations against observed river discharge,
satellite observed water height and estimates of flooded
area, which are available for discontinuous periods in the
1950s through 1990s throughout the Brazilian portion of the
basin. The validation against spatially extensive data allows
us to more thoroughly investigate the causes of discrepancies between simulated and observed river discharge on
the regional scale.
[6] Finally, we present a limited analysis of the simulated
discharge, water height, and flooded area for the period
1939 – 1998. The analysis of the long-term simulation
allows us to link the observed modes of variability in the
atmospheric state (precipitation and temperature) to the land
surface hydrology, for which continuous observations are
not available. A more comprehensive analysis of the water
cycle, as simulated by these models will be presented in a
subsequent paper by J. A. Foley et al., (The El Niño/
Southern Oscilliation and the Climate, Ecosystems and
Rivers of Amazonia, submitted to Global Biogeochemical
Cycles, 2002, hereinafter referred to as Foley et al., submitted manuscript).
2. Methods
2.1. Model Descriptions
[7] We use two models developed at the University of
Wisconsin, IBIS and HYDRA, to simulate the water balance of the Amazon River system between 1939 and 1998.
IBIS is used to derive estimates of the land surface water
balance, from long-term climate data. The IBIS simulations
of runoff (surface runoff and subsurface drainage) are used
as input to the HYDRA model to estimate changes in river
discharge, and the volume of water stored in the floodplain
of the river system. Both IBIS and HYDRA are thoroughly
described in previous publications [Coe, 2000; Donner et
al., 2002; Foley et al., 1996; Kucharik et al., 2000], therefore only a brief description of the models and recent
improvements are provided below.
[8] IBIS represents land surface processes (energy, water,
and momentum exchange among soil, vegetation, and the
atmosphere), canopy physiology (canopy photosynthesis
and conductance), vegetation phenology (bud burst and
senescence), and long-term ecosystem dynamics (vegetation
dynamics and carbon cycling). These processes are organized in a hierarchical framework and operate at different
time steps, ranging from 60 min to 1 year. This allows for
explicit coupling among ecological, biophysical, and physiological processes occurring on different timescales.
[9] HYDRA simulates the time-varying flow and storage
of water in terrestrial hydrological systems, including rivers,
wetlands, lakes, and human-made reservoirs [Coe, 1998,
2000]. This model currently operates on the global scale on
a 5-minute latitude by longitude grid (9 km at the equator)
and with a 1-hour time step. HYDRA requires the following
boundary conditions: topography (from digital elevation
models), potential evaporation (estimated from climate data,
using a simple Penman energy balance model), surface
runoff (supplied by IBIS), base flow (drainage from the
soil column, supplied by IBIS), and precipitation (from
climate data).
[10] HYDRA derives potential lake and wetland volumes
from digital elevation model (DEM) representations of the
land surface. River paths are prescribed from the Amazon
Basin river directions defined by Costa et al. [2002]. The
physical land surface of HYDRA is coupled to a linear
reservoir model to simulate (1) the discharge of river
systems, (2) the spatial distribution (and volume) of large
permanent lakes, and (3) the flux and concentration of
nitrogen in surface water. Rivers and lakes are defined as
a continuous hydrologic network in which locally derived
runoff accumulates and is transported across the land surface via rivers, it fills lakes and wetlands, and is eventually
transported to the ocean or is evaporated from an inland
water body.
[11] The linear reservoir model used to simulate the
transport of water in the river system is the same as that
used by Coe [2000] and is based on those used in numerous
other large-scale hydrology studies [e.g., Miller et al., 1994;
Vorosmarty et al., 1989]. The linear reservoir model simulates water transport in terms of prescribed river routing
directions derived from the local topography, residences
times of water within a grid cell, and effective flow
velocities.
[12] The total water entering the hydrologic network at
each grid cell is the sum of the land surface runoff (Rs),
subsurface drainage (Rd), precipitation (Pw) and evaporation
(Ew) over the surface waters, and flux of water from
upstream grid cells (Fin, all in m3/s). The water transport
is represented by the time dependent change of three water
reservoirs. First, the river system reservoir (WR), which
contains the sum of upstream and local water in the river
system. Second, the surface runoff pool (Ws), which contains water that has run off the surface locally and is flowing
toward a river. Third, the subsurface drainage pool (Wd),
which contains water that has drained through the local soil
column and is flowing toward a river. All reservoirs are
represented in m3 and flow is governed by the following
differential equations.
dðWs Þ
Ws
¼ Rs dt
Ts
d ðWd Þ
Wd
¼ Rd Td
dt
X
d ðWR Þ Ws Wd
WR
ð1 Aw ÞþðPw Ew Þ Aw þ
Fin
þ
¼
Ts Td
TR
dt
[13] Aw is the fractional water area in the grid cell; from
1 (lake, wetland, or reservoir covers entire cell) to 0 (no
water present) and is predicted by HYDRA. Ts, Td, and TR
are the residence times (s) of the water in each of the
reservoirs. Pw and Ew are the precipitation and evaporation
rates (m3/s) over the surface water, respectively and Fin
is the sum of the fluxes of water (m3/s) from the upstream
cells.
[14] The local surface and subsurface residence times (Ts
and Td) are set to globally constant values for simplicity. In
this application Td and Ts are set to 2 hours, similar to the
value for Ts used by Costa and Foley [1997] to simulate
large-scale flow in the Amazon Basin and by Coe [2000] to
simulate global river flow. In this application we have set Td
equal to Ts because the subsurface drainage is provided by
the IBIS model, which explicitly calculates the transfer of
water through a 4-meter soil column. Therefore, the
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
increased residence time of subsurface flow compared to
surface flow is calculated in IBIS and does not need to be
accounted for by HYDRA.
[15] The streamflow residence time, TR, is defined as the
ratio of the distance between centers of the local and
downstream grid cells (D) and the effective velocity of
the water (u). The effective velocity (u) is calculated as by
Miller et al. [1994] and is proportional to a ratio of the
downstream topographic gradient (ic, in m/m) and a reference gradient io (= 0.5 104 in m/m):
u ¼ uo1 ðic =io Þ
0:5
11 - 3
It enters a separate but parallel set of equations and is
calculated only to understand the spatial extent and depth of
flooding at any given time.
[18] The storage and transport of water on the floodplain
is represented by the following differential equations.
X
d Wf
Wf
Ffi ¼ Vf þ
TR
dt
Vf ¼
d ðWRcf Wrma Þ
dt
Vf ¼ 0
uo1 is the minimum effective velocity of the river (0.8 m/s)
and is allowed to vary between 0.3 and 3 m/s.
[16] Although, the version of HYDRA described above
calculates the flux of water in rivers and the storage in lakes,
it is unable to capture the significant seasonal flooding on
the river floodplain. Therefore, in this study an algorithm to
diagnose the time-transient extent of flooded area adjacent
to rivers has been added to HYDRA. The algorithm is based
on the same sets of equations as the water transport
described above and is similar to methods developed by
Vorosmarty et al. [1989] to simulate the discharge of the
Amazon River basin and by Bates and De Roo [2000] for an
application to a reach of the Meuse River in the Netherlands. The method of Vorosmarty et al. [1989] was developed for the Amazon Basin as a coupled model in which
flooding was a dynamic part of the river system and
contributed to the evaporation from and timing of the river
flow. The model of Vorosmarty uses a flood initiation
parameter to define the threshold volume at which flooding
occurs in a river channel. This is useful where data on the
basin geomorphology is limited and because it is a general
solution that can be applied globally. However, in their
method floodwaters cannot be transported outside the grid
cell where the flooding originated and cannot be applied to
time transient solutions. Bates and De Roo [2000] developed a model that explicitly simulates water transport across
the floodplain based on water head. That model accurately
simulates flood extent and height but requires detailed
knowledge of the basin geomorphology, which is often
not available for large river systems.
[17] Our inundation method combines aspects from both
of the models described above to diagnose the time transient
flooded area at a one-hour time step. In our method, water
in excess of a prescribed maximum river channel volume
(floodplain initiation parameter) is transported from the
river onto the floodplain as was done by Vorosmarty
[1989]. We use a flood initiation parameter (described
below) because we do not have detailed topographic data
for the basin. Once on the floodplain, water flows across the
land surface to neighboring grid cells, as stated by Bates
and De Roo [2000], to simulate large flood events. The
direction and velocity of the flow across the floodplain is
controlled by the difference in water elevation between
neighboring grid cells. There is only a one-way coupling
of the floodplain inundation to the river discharge. Water
enters the floodplain reservoir from the stream but it is not
subtracted from the river transport, does not reenter the
stream system, attenuate the river discharge hydrograph
during floods, or contribute to evaporation from the river.
LBA
WR > cf Wrma
WR cf Wrma
[19] Where Wf is the floodplain reservoir (m3). The
change with time of the floodplain reservoir is the sum of
the fluxes of water to the floodplain from the river (Vf) and
from all upstream floodplain grid cells (Ffi), minus that
transported to the downstream floodplain grid cell (Wf/TR,
all in m3/s). TR is the same residence time as that in the river
transport calculation. Wrma is the mean annual volume of
water in the river (m3). The flood initiation parameter (cf) is
a unitless multiplier at which the river channel is considered
to be full. In this study cf is set to a constant value of 2.5. In
reality the flood initiation parameter should differ for each
grid cell based on local physical conditions. Future studies
will be needed to investigate whether cf can be derived from
existing data such as digital elevation models or limited
observations of stream characteristics.
[20] The height of the floodwaters (Hf, in m) is derived
from the volume of water on the floodplain, the land area of
the grid cell (At, in m2), and the land surface elevation (Z, in
m).
Hf ¼
Wf
At
þZ
[21] The fractional area of a grid cell inundated by the
floodwaters (Af) is set to 100% of the grid cell if Wf /At is
greater than or equal to 1 meter. For Wf /At less than 1 meter
the area inundated is set to Wf /At *(1/1m). For example, if
Wf /At = 0.3m the inundated area is set to 0.3 (30%) of the
grid cell.
[22] Starting with an initial value of 0 for WR, Ws, Wd,
and Wf, HYDRA is forced with 0.5° 0.5° estimates of
monthly mean runoff, precipitation, and surface water
evaporation (for the period 1939 – 1998) converted to daily
values and linearly interpolated to the 50 50 grid of
HYDRA. The model solves the equations with a time step
of 1 hour. The predicted river discharge and flooded area
represent the surface hydrology in equilibrium with the
prescribed climate.
[23] HYDRA is a general model that has previously been
used at global and regional scales. For example, it has been
used to simulate global lake area [Coe, 1998] and river
discharge [Coe, 2000]. The model has also been used to
evaluate the performance of general circulation model
simulations of paleoclimate in the tropics and northern
Africa [Coe and Harrison, 2002; de Noblet-Ducoudré
et al., 2002]. IBIS and HYDRA together have been used
to evaluate the simulated hydrology of the National Center
for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) climate reanalysis for the
LBA
11 - 4
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Figure 1. The Amazon Basin with river discharge station
ID numbers [Costa et al., 2002].
period 1963 – 1995 over the continental United States [Lenters et al., 2000] Recently, the models have been used to
quantify the impact of human water management practices
and climate variability on the Lake Chad basin in northern
Africa [Coe and Foley, 2001] and to evaluate the impact of
climate variability on nitrate transport within the Mississippi
River basin [Donner et al., 2002].
2.2. Experiment Design
[24] Long-term climatic data from the Climate Research
Unit of the University of East Anglia, Norwich [New et al.,
2000] (hereinafter referred to as CRU05) are used as
climatological forcing to IBIS and HYDRA. CRU05 is a
global, monthly mean data set of temperature, precipitation,
humidity, and cloudiness, at 0.5° by 0.5° latitude/longitude
resolution, for the period 1901 – 1998.
[25] IBIS was run on a 0.5° by 0.5° latitude/longitude
grid, extending over the entire Amazon River basin (21°S–
6°N; 45°W – 80°W). Unfortunately, the precipitation and
temperature data for the years 1921 – 1932 are in error,
therefore we limited our IBIS simulation to the period
1935 – 1998. The specific IBIS simulations are described
in more detail by Foley et al. (submitted manuscript). The
IBIS results extending from 1939 to 1998 were used in the
HYDRA simulations along with the climate data (precipitation and estimated lake surface evaporation). The hourly
output from HYDRA was then averaged to monthly mean
values for comparison to observations.
2.3. Validation Data
[26] The simulated river discharge is compared to a data
set of mean monthly river discharge at 121 locations in the
Brazilian portion of the river basin (Figure 1 and Table 1).
The original daily river discharge data set was obtained
from ANEEL, the Brazilian National Agency for Waters
and Electrical Energy. The data has been averaged to
monthly means and described by Costa et al. [2002].
The river discharge is calculated from a measurement of
river water level and converted into a discharge volume
using a rating curve, which is updated several times per
year. The major sources of error in calculating river
discharge probably result from direct measurement and
the use of the rating curve, which assumes a constant
stream cross-sectional area [Cogley, 1989]. The accuracy of
the discharge measurements is not given in the original
data. However, analysis of the potential error in river
discharge measurements suggests that 10– 15% is a reasonable estimate of the error in observed annual mean
discharge [Cogley, 1989].
[27] The height of the simulated floodwaters is compared
to water height measured by the NASA radar altimeter
aboard the TOPEX/POSEIDON satellite. A time series of
mean monthly relative surface water height was constructed
for 10 locations on the main stem of the Amazon for the
period 1992– 1998 from an about 10 day temporal resolution data set created by Birkett et al. [2002]. The altimeter
emits a series of microwave pulses at 13.6 GHz at the land
surface. The surface height is calculated from the time delay
between pulse emission and echo reception. Each height
value is an average of all surface heights found within the
footprint of the altimeter. The effective diameter of the
footprint depends on the surface roughness, but can typically range between 200 m (for open pools of water in calm
conditions) to a few kilometers (open water with surface
waves). This measurement technique has been applied to a
number of rivers and wetlands in several test-case studies
and validated against surface observations of water height
[Birkett, 1998, 2000] (Å. Rosenqvist et al., Using satellite
altimetry and historical gauge data for validation of the
hydrological significance of the JERS1 SAR (GRFM)
mosaics in Central Africa, IJRS GRFM, in review). The
results demonstrate that submonthly, seasonal, and interannual variations in surface water height can be monitored to
accuracies of 10s of cm RMS for rivers. The seasonal water
height varies by about 10 meters or less therefore the total
measurement error is probably less than 10%.
3. Evaluation of Simulated Surface Hydrology
[28] In this section we present the results of the simulated
river discharge, surface water level, and seasonally flooded
area. Comparison is made to satellite and ground-based
observations where data is available.
3.1. Discharge
3.1.1. Mean Annual
[29] The simulated mean annual discharge is well correlated with the observations for the 121 sites (r2 = 0.99). It
is within ±20% of the observations for only 45 of the 121
stations (Table 1 and Figure 2a) and is within 40% of the
observations for 92 of the 121 sites (Figure 2a). In general,
simulation of the mean annual discharge is difficult because
it depends upon the input precipitation data set and
calculation of the evapotranspiration in IBIS (itself a
complex function of the input data and simulated radiative
properties, vegetation and soil characteristics). In these
simulations we have not tuned the model to produce
results in agreement with the observations.
[30] An advantage of having a large number of discharge
stations to compare to the simulation is that we can begin to
pinpoint where in the basin, and possibly why, discrepancies between simulated and observed discharge are occurring. For example, the simulated mean annual discharge at
Óbidos, the furthest downstream station (Table 1 and Figure
1, #33), is about 25% less than the observed discharge for
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 5
Table 1. Observed and Simulated Annual Mean Discharge and % Error for 121 Stations (the station ID, name and locations are in
columns 1 – 4)a
Station ID
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
Name
Latitude
Longitude
Observed
Simulated
Error (%)
Years
Javari at Estirão do Repouso
Solimões at Teresina
Solimões at São Paulo de Olivença
Iça at Ipiranga Velho
Solimões at Santo Antônio do Içá
Juruá at Cruzeiro do Sul
Tarauacá at Envira
Juruá at Gavião
Japurá at Acanauı́
Solimões at Itapeuá
Purus at Seringal Providência
Purus at Seringal da Caridade
Acre at Floriano Peixoto
Purus at Seringal Fortaleza
Ituxi at São Gregório
Purus at Labréa
Purus at Arumã-Jusante
Solimões at Manacapuru
Uaupés at Taraquá
Negro at Curicuriari
Negro at Serrinha
Branco at Caracaraı́
Guaporé at Pedras Negras
Mamoré at Guajará-Mirim
Madeira at Abunã
Abunã at Morada Nova
Madeira at Porto Velho
Ji-Paraná at Ji-Paraná
Ji-Paraná at Tabajara
Madeira at Humaitá
Madeira at Manicoré
Aripuanã at Prainha
Amazonas at Óbidos
Arinos at Porto dos Gaúchos
Teles Pires at Cachoeirão
Teles Pires at Indeco
São Manoel at Três Marias
Tapajós at Barra São Manoel
Tapajós at Jatobá
Xingu at São Felix do Xingu
Xingu at Belo Horizonte
Curuá at Mouth
Irirı́ at Pedra do Ó
Xingu at Altamira
Tocantins at São Felix
Paranã at Ponte Paranã
Fresco at Boa Esperança
Paranã at Paranã
Tocantins at Peixe
Tocantins at Porto Nacional
Tocantins at Miracema
Sono at Porto Real
Tocantins at Tupiratins
Tocantins at Carolina
Tocantins at Tocantinópolis
Tocantins at Tucuruı́
Curuca at Santa Maria
Ituı́ at Seringal do Ituı́
Juruá at Eirunepe-Montante
Acre at Xapuri
Acre at Rio Branco
Purus at Valparaı́so
Mucuim at Cristo
Cuniua at Bacaba
Negro at São Felipe
Uaupés at Uaracu
Uraricoera at Mocidade
Uraricoera at Faz. Pássaro
Mucajaı́ at Fé e Esperança
Guaporé at Mato Grosso
Madeira at Palmeiral
4.42
4.33
3.50
3.00
3.17
7.67
7.33
4.92
1.83
4.08
9.00
9.08
9.08
7.75
7.58
7.33
4.75
3.33
0.17
0.25
0.50
1.75
12.92
10.83
9.75
9.92
8.83
10.92
9.00
7.58
5.83
7.33
1.92
11.67
11.83
10.17
7.67
7.33
5.17
6.67
5.42
5.75
4.58
3.25
13.58
13.33
6.75
12.58
12.08
10.75
9.58
9.25
8.25
7.42
6.33
3.83
4.75
4.75
6.75
10.67
10.00
8.75
7.33
6.42
0.33
0.50
3.42
3.17
2.75
15.08
9.58
70.92
69.67
68.67
69.50
67.92
72.67
70.17
66.67
66.50
63.00
68.58
68.50
67.33
66.92
64.92
64.75
62.08
60.50
68.50
66.75
64.75
61.08
62.92
65.33
65.33
65.50
63.83
61.92
62.08
63.00
61.25
60.33
55.42
57.33
55.75
55.50
57.83
58.00
56.75
52.00
52.83
54.42
54.00
52.17
48.08
47.17
51.75
47.83
48.50
48.42
48.33
48.00
48.08
47.42
47.33
49.67
71.42
70.25
69.83
68.58
67.75
67.33
64.17
64.83
67.25
69.08
60.92
60.58
61.25
59.92
64.75
2503
44,000
46,546
7046
55,074
913
1297
4780
13,922
81,541
806
1346
590
3681
722
5569
10,469
98,969
2735
11,752
16,054
2865
915
8467
18,523
730
19,357
713
1407
21,829
24,726
3400
171,504
764
847
1178
3980
8339
10,795
4627
5324
862
2663
8665
904
357
837
753
2007
2225
2579
807
3500
4042
4566
11,704
942
787
1834
220
344
2112
258
1503
7406
2462
1236
1308
276
143
20,789
1552
20,847
22,712
3898
28,194
1166
1681
5311
9671
57,991
1383
2100
805
4724
900
6560
11,296
73,514
1754
8136
11,795
2539
1180
5695
13,545
576
14,737
669
1440
18,544
19,880
3727
128,543
962
877
1527
4984
10,573
13,024
6936
7783
1270
4291
13,162
960
339
1276
712
1936
2700
3042
643
3724
4189
4613
14,078
626
540
2260
292
552
3088
229
1230
4426
1488
896
992
472
140
17,353
38
53
51
45
49
28
30
11
31
29
72
56
36
28
25
18
8
26
36
31
27
11
29
33
27
21
24
6
2
15
20
10
25
26
4
30
25
27
21
50
46
47
61
52
6
5
52
6
4
21
18
20
6
4
1
20
34
31
23
33
60
46
12
18
40
40
27
24
71
2
17
15
17
22
14
22
27
16
22
22
10
7
28
28
29
7
29
13
12
19
19
19
28
15
26
20
4
29
17
17
22
27
23
27
21
17
17
12
15
21
21
20
20
16
26
27
15
20
18
14
35
13
15
27
34
24
17
8
10
15
23
28
11
13
17
19
19
10
18
22
19
8
LBA
11 - 6
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Table 1. (continued)
Station ID
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
Name
Latitude
Longitude
Observed
Simulated
Error (%)
Years
Jamari at Ariquemes
Jamari at São Carlos
Jamari at São Pedro
Jamari at Cachoeira do Samuel
Candeias at Santa Isabel
Pimenta Bueno at Cachoeira Primavera
Pimenta Bueno at Pimenta Bueno
Aripuana at Boca do Guariba
Sucunduri at Santarém Sucunduri
Araguaia at Xambioá
Curua at Boca do Inferno
Teles Pires at Porto Roncador
Teles Pires at Teles Pires
Verde at Lucas
Curua-Una at Barragem-Jusante
Maicuru at Arapari
Paru de Este at Fazenda Paquira
Iriri at Laranjeiras
Jari at São Francisco
Maranhão at Ponte Quebra Linha
Almas at Ceres
Almas at Colônia dos Americanos
Maranhão at Porto Uruacu
Paranã at Flores de Goiás
Paranã at Nova Roma
Paranã at Montante Barra do Palma
Palma at Rio da Palma
Palma at Barra do Palma
Tocantins at Fazenda Angical
Santa Tereza at Colonha
Santa Terza at Jacinto
Manuel Alves at Porto Jerônimo
Manuel Alves at Fazenda Lobeira
Sono at Jatoba
Sono at Novo Acordo
Balsas at Porto Gilandia
Balsas at Rio das Balsas
Perdida at Dois Irmãos
Manuel Alves Grande at Goiatins
Tocantins at Descarreto
Tocantins at Itaguatins
Claro at Montes Claros de Goiás
Vermelho at Travessão
Cristalino at Barra do Forquilinha
Mortes at Toriqueje
Mortes at Xavantina
Mortes at Trecho Médio
Mortes at Santo Antônio do Leverger
Araguaia at Torixoréu
Araguaia at Barra do Garças
10.00
9.75
9.00
8.83
8.83
11.92
11.67
7.75
6.83
6.42
1.58
13.67
13.00
13.17
2.83
1.83
0.42
5.75
0.75
15.00
15.33
14.58
14.58
14.58
13.83
12.67
12.42
12.58
12.33
12.33
12.00
11.75
11.58
10.17
10.08
10.75
10.08
9.33
7.75
5.83
5.75
16.00
15.58
12.92
15.25
14.75
13.50
12.08
16.25
15.92
63.00
63.08
63.25
63.42
63.67
61.17
61.17
60.25
58.92
48.50
54.75
55.25
55.83
55.92
54.25
54.33
53.67
54.17
52.50
48.67
49.50
49.08
49.00
47.00
46.83
47.83
47.08
47.75
48.25
48.58
48.67
47.83
48.25
47.25
47.75
47.75
47.92
47.75
47.25
47.42
47.42
51.25
50.67
50.83
52.92
52.33
51.42
50.83
52.42
52.17
173
229
324
346
320
220
206
1445
432
5685
129
260
360
114
135
114
490
1235
1019
155
167
335
529
80
213
626
246
273
1793
131
187
157
206
341
349
96
252
192
166
4939
4526
146
85
118
362
501
791
904
353
600
182
260
343
441
296
141
174
1603
548
6424
516
164
249
123
477
424
756
2231
1550
143
207
362
569
99
238
645
175
236
1979
187
268
195
245
203
274
152
245
125
125
4521
4204
149
96
110
203
302
605
817
161
317
5
14
6
27
8
36
15
11
27
13
300
37
31
7
252
271
54
81
52
8
24
8
8
24
12
3
29
14
10
42
44
24
19
41
22
58
3
35
25
8
7
2
13
7
44
40
24
10
54
47
27
10
10
12
21
4
15
19
22
27
21
21
11
13
20
25
16
12
30
23
19
15
17
9
13
7
11
19
10
8
17
10
20
11
12
12
8
6
23
23
8
24
22
12
26
28
16
24
24
24
a
The simulated discharge is averaged for the same years as the observed data. The number of years averaged is in the last column. The location of each
station is shown on Figure 1.
the period 1970 to 1996 (128,543 versus 171,504 m3/s).
This underestimation is primarily a result of a negative bias
associated with the tributaries that drain Colombia, Ecuador,
Bolivia, and Peru. The negative bias summed for the four
major tributaries when they enter Brazil (Table 2, stations 5,
9, 21, and 25 on the Solimões, Japurá, Negro, and Madeira
rivers respectively) is about 40,000 m3/s. This accounts
for about 95% of the 43,000 m3/s difference between the
simulated and observed discharge on the main stem at
Óbidos (Table 2, station #33 in Table 1). In fact, the furthest
upstream station on the main stem (Station #2, Solimões
River at Teresina) accounts for more than half of the error at
Óbidos (23,000 m3/s).
[31] The magnitude of the simulated discharge generated
within Brazil is in good agreement with the observations for
all of the major tributaries of the main stem, including those
with a strong negative bias. To illustrate that there is no
substantial negative difference between the simulated and
observed discharge on these tributaries within Brazil, we
subtracted the discharge at the border from the downstream
discharge in each of the tributaries. The simulated in-stream
discharge (downstream discharge minus the furthest
upstream discharge) is within about 15% of the observed
discharge for the six major tributaries contributing to the
discharge at Óbidos (Table 3, Solimões, Juruá, Purus,
Negro, Branco, and Madeira). Therefore, the large underestimation of the discharge rate shown in Tables 1 and 2 for
Óbidos and other stations (Solimões, Japurá, Negro, and
Madeira rivers) is transferred from the Brazilian border to
downstream stations as a nearly constant bias. Note also that
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 7
Table 2. Difference Between Simulated and Observed Annual
Mean Discharge (sim obs)a
Station
ID
Discharge
Óbidos
Madeira
Japurá
Negro
Solimões
Sum
33
25
9
21
5
42,961
4846
4251
4259
26,880
40,236
a
Mean discharge values are expressed in m3/s. Sum is the sum of the
discharge of all stations in this table other than Óbidos.
Figure 2. (a) Histogram of percent error in simulated
annual mean river discharge (i.e., 100 [sim obs]/obs)
for the 121 stations listed in Table 1. (b) Histogram of
percent error in simulated long-term seasonal cycle. Percent
error is calculated as 100 (sima obsa)/obs, where sima
and obsa are the deviation from the mean monthly simulated
and observed discharge and obs is the observed mean
annual discharge. Sample size is 1452 (121 stations 12
months). (c) Similar to Figure 2b, but for interannual
discharge anomalies, excluding years with data gaps. Here,
sima and obsa are annual rather than monthly anomalies and
the sample size is 1951 (121 stations 16 years of data
for each station, the number of years depends on that
available for the observations).
the nearby Juruá and Purus rivers that do not drain large
regions outside Brazil, do not show this strong negative bias
(Table 1 and Figure 1 stations 8 & 17).
[32] Costa and Foley [1997] found a similar negative bias
in their simulated discharge on the Amazon main stem
(using different models and precipitation data from this
study). In that study the authors noted that the simulated
runoff ratios for stations draining the Andes were unusually
low compared to those inside Brazil. Therefore, the strong
spatial coherency of the error in our simulated discharge and
similar errors in the independent study of Costa and Foley
[1997], in a region for which precipitation estimates are
very difficult to obtain, suggest that this large negative
difference is likely associated with errors in the precipitation
data set outside Brazil rather than with the calculation of
evapotranspiration in IBIS. As pointed out by numerous
authors [e.g., Leemans and Cramer, 1991] orographically
induced rainfall is often underestimated because the spatial
distribution of rain gauges is not sufficient to capture the
small space scale (but large magnitude) differences in
precipitation.
[33] The simulated mean annual discharge at the furthest
downstream stations on the tributaries in the eastern portion
of the basin (Table 1; Tapajós #39, Xingu #44, and
Tocantins/Araguaia #56) is generally overestimated compared to the observations. The location of the error can be
pinpointed by looking at the discharge on individual sections of the rivers. For the Tapajós the overestimation occurs
in the middle and upper reaches of the river (Table 3). The
discharge generated between stations 38 and 39 on the
Tapajós is in excellent agreement with the observations
(Table 3) while between stations 34 and 38, and 36 and
37 it is more than 25% greater than the observations. For the
Xingu River the overestimation occurs throughout the lower
and middle portions of the basin. The simulated discharge
generated between station 44 and its upstream stations (#43
Table 3. Total Discharge From Individual Reaches of Major
Amazon Tributaries (m3/s) and the % Difference Between Simulated
and Observed Reach Discharge
River
Reach
Observed
Simulated
Percent
Difference
Solimões
Juruá
Purus
Negro
Branco
Madeira
Tapajós
Tapajós
Tapajós
Xingu
Tocantins
5–4–2
8 – 59
17 – 62
21 – 66 – 65
26 – 68 – 69
31 – 25
39 – 38
37 – 36
38 – 34
44 – 43 – 40
56 – 55 – 81
4027
2946
8357
6186
1280
6203
2455
2802
7493
1375
1453
3449
3051
8208
5881
1075
6336
2451
3458
9696
1935
3041
14
4
2
5
16
2
0
23
29
41
109
The reach discharge is calculated as the discharge at a given station
minus the discharge from one or more upstream stations. The stations used
to calculate the reach discharge are listed in column 2. See Table 1 and
Figure 1 for the location of the stations.
LBA
11 - 8
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Figure 3. Scatter diagram of simulated mean monthly discharge versus observations in m3/s. The
sample size is 23,412 (121 locations 16 years of monthly data for each station). The one-to-one
agreement line is shown for comparison.
& 40) is about 40% greater than observed. On the Tocantins/Araguaia a large portion of the discrepancy occurs in
the lowest reach (Table 3, station 56 minus 55 and 81).
[34] On many of the small headwater tributaries of the
Tapajós and Tocantins/Araguaia Rivers the simulated discharge is more than 25% less than the observations. This
underestimation is small in magnitude but spatially consistent (Table 1, see stations 83, 84, 105, 106, 109, 110, 115–
118). It may be related to basin topography which is not
well captured at the 1/2 degree resolution of IBIS or to
differences between the simulated and observed vegetation.
For example, the IBIS simulated vegetation in these upland
regions is dominated by broadleaf evergreen forest and
savannah. However, the actual vegetation in these regions
has been highly modified for agriculture and grazing
[Cardille et al., 2002]. As a result, our surface hydrologic
budget is based on a land surface with far different radiative
and hydrological properties from the observations. Future
simulations will include land use changes.
3.1.2. Seasonal Cycle
[35] The seasonal cycle of the simulated river discharge is
in fairly good agreement with the observations throughout
the basin. The coefficient of correlation (r2) between the
simulated and observed discharge for the 23,412 months of
the observations is 0.97 (Figure 3). The clustering of points
below the 1:1 line clearly indicates the bias toward
underestimation of the river discharge indicated in the
mean annual discharge.
[36] The simulated anomalies (from the annual mean) of
the monthly mean discharge are within 20% of the observed
anomalies for almost 50% of the 1452 station-months (12
months 121 stations) and within 40% of the observed
monthly anomaly for 77% of the station-months (Figure 2b).
The monthly discharge anomaly is in better agreement with
the observations for stations with high discharge rates. For
example, for the stations with discharge greater than 10,000
m3/s the difference between the simulated monthly anomaly
is within 20% of the observed for 70% of the 204 stationmonths (12 months 17 stations, not shown) and is within
40% for 97% of the station-months.
[37] The most obvious sources of potential error in the
simulated monthly discharge anomaly are; (1) the accuracy
of the input data sets to IBIS (such as precipitation,
cloudiness, and temperature), (2) the calculation of runoff,
soil moisture, and subsurface drainage within IBIS, and (3)
the calculation of the water transport within HYDRA. Any
one of these sources of error is potentially large. Because of
the complexity of the models and the large amount of data
used as input, it is difficult to assess the individual sources
of error. However, simulation of the soil column physics
and the generation of surface and subsurface runoff requires
high resolution data on soil characteristics (such as texture,
porosity, and hydraulic conductivity) and simple parameterizations of complex and poorly understood coefficients for
soil hydraulic characteristics. These parameterizations are
often based on research in midlatitude sites only and
therefore, may not be well suited to the tropics. Additionally, our previous work with IBIS and HYDRA in the
Mississippi River basin [Lenters et al., 2000; Donner et al.,
2002] suggests that our simple soil data and the parameterizations of soil column physics within IBIS may not
adequately simulate the transport of water through the soil
column.
[38] A second potentially large source of error is the
calculation of effective velocity within HYDRA. Currently,
the model does not adequately include the impact of flooding
on the velocity nor the physical characteristics of the land
surface such as the river channel sinuosity. In fact, river flow
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 9
Figure 4. Interannual variability of simulated (dashed line) and observed (black line) river discharge.
The variability is shown as the percent difference from the respective means so that the bias in the
simulated discharge is removed. (a) Tocantins River #56, (b) Xingu River #44, (c) Óbidos #33, (d)
Solimões #5, (e) Madeira #31, and (f ) Purus #17. Stations locations are shown on Figure 1. The length of
comparison is determined by the number of years of observed discharge available (see Table 1).
velocities are parameterized using a global function, instead
of empirical functions specific for the Amazon Basin.
3.1.3. Interannual Variability
[39] The agreement of the simulated interannual variability
of the discharge with the observations is, in general, much
better than for the seasonal variability and annual mean
(Figure 2c compared to Figures 2a and 2b). The anomaly
of the simulated annual discharge from the observed mean is
within 20% of the observations for 70% of the 1951 stationyears (121 stations 16 years of data for each station) and
within 40% of the observations for 93% of the station-years
(Figure 2c).
[40] The magnitude and timing of the simulated variation
from the mean annual discharge is in very good agreement
with the observations for those stations not directly draining
regions outside Brazil. For example, the percent year-toyear variation in simulated discharge closely mimics the
observations at Óbidos (on the main stem) and on the Purus,
Xingu, Tocantins (Figures 4c, 4f, 4b, and 4a) and Tapajós
(not shown) Rivers. However, the relative variability on the
upper reaches of the Madeira, Solimões, (Figures 4e and 4d)
and Japurá Rivers (not shown) is more extreme than the
observations. Year-to-year simulated variations of 30% are
common in these locations compared to only 10– 15% in
the observations. On the Madeira after 1988 there is little
correlation of the simulated variability to the observed
(Figure 4). The poor agreement with observations on the
tributaries draining regions outside of Brazil is consistent
with the conclusion that the precipitation data set is poor in
these regions.
LBA
11 - 10
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Table 4. Comparison of Simulated and Observed Water Height at 10 Locations
ID
Lat
Lon
r2
OBS std dev
Sim std dev
Months
OBS ann dev
Sim ann dev
a
b
c
d
e
f
g
h
I
j
all
2.540
2.540
3.210
3.125
3.790
4.100
3.290
2.540
3.040
4.290
56.540
56.960
58.875
59.875
62.875
63.200
64.625
65.540
67.875
69.710
0.83
0.79
0.81
0.86
0.79
0.70
0.72
0.70
0.60
0.53
0.79
1.9
2.5
2.8
3.5
2.7
2.3
2.9
2.3
1.6
2.8
2.9
2.3
2.1
2.2
1.7
2.4
3.0
3.1
1.9
1.2
1.1
2.7
72
69
64
72
51
55
56
47
38
29
553
0.90
1.02
1.02
1.13
0.90
0.53
0.71
0.67
0.91
1.60
0.94
0.56
0.76
0.83
0.41
0.56
0.63
0.64
0.46
0.33
0.39
0.56
Column 1 contains the location designation shown on Figure 6. Latitude and longitude of locations are in columns 2 and 3. The coefficient of correlation
is in column 4. The standard deviations of the mean monthly observed and simulated height from the mean are in columns 5 and 6. The number of months
for comparison at each location is in column 7. The deviations of the mean annual height from the mean of all years are in columns 8 and 9.
3.2. Floodplain Inundation
[41] In addition to the river discharge, we diagnose the
mean monthly water height and flooded area throughout the
Amazon River basin. In its present form the floodplain
inundation is not a fully dynamic part of HYDRA. The
wetlands created do not impact the river water balance or
velocity; they are merely diagnosed from the water available
above the prescribed bank full volume. In this section we
compare the simulated flooded area to two data sets; 1)
satellite microwave observations of surface water height for
the period 1992 – 1998, and 2) estimates of flooded area for
the period 1979– 1987 derived by Sippel et al. [1998] from
satellite observations.
3.2.1. Height
[42] We have chosen 10 locations, on the main stem of
the Amazon River for comparison (Table 4). The
simulated water height is the height of the water above
Figure 5. Scatter diagram of simulated versus observed
mean annual water height above flood stage for 10 locations
on the main stem for the period September 1992 to
September 1998. The sample size is 553 (10 locations about 55 months above flood stage within the period). The
one-to-one agreement line is shown for comparison.
flood stage. The height of the water below flood stage is
not simulated. The observed height of the water is the
height (relative to the best pass of the satellite over the
site) at all times, regardless of whether the river is above
or below flood stage. Therefore, in this section we
compare the standard deviation of the monthly and
annual mean water height only for the 553 months for
which the simulated river is above flood stage at the 10
locations during the period 1992 –1998.
[43] The simulated mean monthly water height is in
relatively good agreement with the observations. The coefficient of correlation (r2) between the simulated and observed river height (for the 553 months that the simulated
river is above flood stage at the 10 locations) is 0.79 (Table 4
and Figure 5). The simulated standard deviation of the
monthly water height for the entire period is 2.7 m compared to the observed deviation of 2.9 m.
[44] The best agreement with observations (r2 from 0.79
to 0.83) occurs at the five locations downstream of the
confluence of the Negro and Solimões rivers (Table 4 and
see Figure 6, locations a– e). Upstream of the confluence
(Table 4, locations f – j) the coefficient of correlation begins
Figure 6. The Amazon Basin with the 12 reaches defined
by Sippel et al. [1998] numbered 1 – 12. Also shown are the
10 locations at which simulated river height is compared to
the TOPEX/POSEIDON observations (labeled a– j).
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 11
Figure 7. Time series of simulated (dashed line) and observed (black line) relative mean monthly water
height above flood stage for locations c and d (see Figure 6 for locations). The observed height is relative
to the height at the best pass of the satellite. The simulated height is relative to the mean height for the
period and has been arbitrarily shifted to coincide with the observed.
to drop, as far as about 0.53 at about 70° W (location j).
The progressive decrease in the correlation west of the
confluence is consistent with the poor simulation of the
discharge on the Solimões River discussed in the previous
section.
[45] The model simulates the timing and magnitude of
the seasonal changes in relative water height in good
agreement with the observations, particularly for those
locations downstream of the confluence of the Negro and
Solimões Rivers (Table 4 and Figure 7). The standard
deviation of the monthly water height from the mean of
the months is well simulated for all locations except
locations d and j. At locations d and j the deviation of the
observed height is large in comparison to neighboring
locations (3.5 & 2.8 m respectively) and in both cases the
model underestimates the standard deviation by greater than
50% (Table 4).
[46] The interannual variability of the water height is also
relatively well simulated, relatively low water years (1995
and 1998) and relatively high years (1992, 1997) agree with
the observations (Figure 7). The standard deviation of the
annual height from the mean of all 553 months (Table 4) is
generally lower in the model (0.5 m) than the observations
(1 m).
LBA
11 - 12
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Table 5. Mean Annual Floodplain Inundation (km2) Summed for
Each Reach (see Figure 6 for locations of reaches)
Reach
1
2
3
4
5
6
7
8
9
10
11
12
sum
Sippel et al.
Simulated
2044
3532
4890
4651
2837
4879
2364
2403
5593
5595
4996
3342
47,124
1433
2083
2937
3692
3357
4491
3115
7637
5349
5400
4041
3611
47,147
3.2.2. Flooded Area
[47] Evaluation of the accuracy of the simulated annual
mean inundated area is difficult. No large-scale ground based
measurements of flooded area are available for comparison
to the simulation. However, Sippel et al. [1998] (hereafter
referred to as Sippel) used mean monthly passive microwave
observations (from SMMR on Nimbus-7) of surface
brightness temperature combined with an empirical model
to calculate mean monthly flooded area within 12 reaches
(segments) of the Amazon River main stem for the period
1979– 1986 (see Figure 6 for location of reaches). Because
these estimates are not strictly an observation they cannot be
used as direct validation for our simulated values. However,
assuming that the scale of the flooding estimated by Sippel is
Figure 8. HYDRA simulated (gray) and Sippel (black) estimated flooded area. (a) Mean annual
inundated area for the period January 1979 to December 1987 summed for each of the 12 reaches (see
Figure 6 for reach locations) and (b) flooded area for each year (1979 – 1987) summed for all 12 reaches.
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 13
Figure 9. Times series of flooded area (km2) for the HYDRA simulation (dashed line) and the Sippel
estimate (solid line) for 1979 – 1987 for Reach 5 (see Figure 6 for reach location).
reasonable, comparison of the two estimates of flooded area
is instructive.
[48] There are at least three major sources of potential
error in our simulated flooded area. First, the accuracy of
the simulated river discharge impacts the amount of water
available to flood surrounding areas. As discussed above,
the models underestimate the river discharge on the main
stem of the Amazon. As a result we can expect that our
estimates of flooded area in this simulation will be too
low. Second, the digital elevation model (DEM) determines how water can spread across the land surface. There
is significant error in the digital elevation model in the
Amazon due, at least in part, to inaccuracies in the
operational navigation charts used to derive the DEM.
Additionally, the 9 km horizontal resolution of the DEM
used in HYDRA is too coarse to represent small-scale
variations in topography. Therefore, it is likely that the
topography introduces some error. Thirdly, the choice of
the value of the flood initiation parameter determines the
threshold at which water can leave the simulated river
channel. In the present model the value is universally
constant and does not represent the fundamental physical
characteristics of the grid cell. Therefore, it is likely that
the flood initiation parameter introduces error into the
inundation simulation.
[49] The simulated mean annual inundated area summed
for all 12 reaches (41,826 km2) is about 10% less than the
estimate of Sippel (46,197 km2). For the individual reaches
the simulated mean annual area is generally less than the
Sippel estimates, consistent with the underestimation of
discharge (Table 5). The simulated area is within 20% of
the Sippel estimates for 7 of the 12 reaches (Figure 8a). Best
agreement occurs on the downstream reaches (5, 7, 9 – 12)
where the bias in the simulated discharge was least. The
model simulates about 50% less inundated area for the
upstream reaches on the Solimões (1 – 4), which is consistent
with the about 30% or greater underestimation of discharge
on the Solimões (Table 1). The simulated mean annual
flooded area is significantly greater than the Sippel estimate
only on reach 8 (twice as large as the Sippel estimate). It is
unclear why the simulated flooded area is so much greater
than the observed at reach 8. However, since the simulated
discharge is not greater than the observations at this location,
it suggests that the topography may not be particularly well
represented in HYDRA. The topography in the model is very
flat in this region and it is possible that the flooded area is
exaggerated for this segment.
[50] For most of the 12 reaches the seasonality of the
flooding is in relatively good qualitative agreement with the
Sippel estimates. The month of peak flooding occurs generally in April – May and the length of the flooded season is
about 4 – 5 months (for example, Figure 9).
[51] The interannual variability of the annual flooded area
(summed for all 12 reaches) agrees with the Sippel estimates
for the years 1979 – 1986. The variation of the total simulated flooded area is within 25% of the Sippel estimates for
6 of the 8 years (Figure 8b). The model simulates 35 and
45% less flooded area in 1985 and 1986, respectively
(Figure 8b) compared to Sippel. The relatively small simulated flooded area in 1985 and 1986 is consistent with a
very strong underestimation of the discharge for the same
period on the tributaries coming from outside of Brazil (see
Figures 4d and 4e, Solimões and Madeira).
4. Simulated Water Balance: Long-Term Spatial
and Temporal Variability
[52] The results of section 3 suggest that although there
are differences in the magnitude of the simulated discharge
and flooded area compared to observations, the seasonal
and interannual variability is relatively well simulated.
Therefore, in this section we present a limited analysis of
the simulated results for the period 1939 – 1998. In this way
we can investigate the spatial and temporal variability of the
hydrologic cycle throughout the Amazon Basin, which is
not possible using observations alone. A more complete
analysis of the simulation is in preparation and will be
presented in the future.
[53] A singular spectral analysis of the modes of variability of the Amazon climate system by Botta et al. [2002]
LBA
11 - 14
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Figure 10. Interannual variability of simulated river discharge, shown as percent difference from the
mean for the period 1939 – 1998. (a) Óbidos River station #33, (b) Japurá River #9, (c) Negro near
confluence with Solimões, (d) Purus #17, (e) Madeira #31, and (f) Tapajós #38. Stations locations are
shown on Figure 1 except for the Negro which does not correspond to a station location.
indicates that precipitation for the period 1935– 1998 has
two major modes of variability; 28 year and 3 – 4 years. The
3 – 4 year mode of variability has been associated with the
El Niño/Southern Oscillation phenomena (ENSO) by a
number of authors [Kousky et al., 1984; Marengo, 1992;
Richey et al., 1989; Zeng, 1999]. The cause of the long
mode of variability is uncertain but is consistent with
independent observations of the modes of variability of
temperature [Victoria et al., 1998] and river discharge at
Manacapuru [Richey et al., 1989].
4.1. Discharge
[54] The deviation of the simulated annual discharge
from the mean for the period 1939 –1998 reflects the control
of the long and short-term (ENSO) variability on the water
balance of the Amazon Basin (Figures 10a – 10f ). The
simulated discharge at Óbidos (Figure 10a) clearly illustrates the long timescale variability. Relatively wet years are
clustered in the 1940s – 50s and 1970s, dry years in the
1960s and 1980 – 90s, consistent with a similar pattern in the
observed river height and discharge described by Marengo
[1995] and Marengo et al. [1998] for much of the basin. In
our simulation the long mode of variability is not limited to
a particular region of the basin. It is expressed on all of the
major tributaries throughout the basin including the north
and western rivers (e.g., Japurá, Negro, Purus, and Madeira,
Figures 10b – 10e) and the eastern basins (e.g., Tapajós,
Figure 10f ).
[55] A number of authors have shown that El Niño years
are correlated with dryer conditions in the Amazon Basin,
La Niña years with wet conditions [Marengo, 1992; Marengo et al., 1993; Richey et al., 1989; Zeng, 1999]. In the
simulated discharge the ENSO variability is embedded
within the longer mode of variability throughout most of
the basin. Strong El Niño years show up as a negative
deviation from the simulated mean at Óbidos and many of
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
LBA
11 - 15
Figure 11. Time series of simulated mean monthly flooded area in the Amazon River basin for the
period 1939 – 1998. Shading indicates the fraction of the 50 grid cell covered with water; black being
100% flooded.
the larger tributaries (Figures 10a – 10f, e.g., 1941 – 42,
1951 – 52, 1982 – 83, 1986 – 87, and 1992 – 93).
[56] The impact of La Niña on the simulated river
discharge is also apparent throughout much of the basin.
Many La Niña years (e.g., 1945 – 46, 1950, 1955 –56, 1962,
1974 – 75, 1988 – 89) coincide with high discharge rates at
Óbidos, in the western basins (e.g., Purus and Japurá) and in
the east (e.g., Tapajós, Figure 10f ).
4.2. Flooding
[57] The simulated mean monthly flooded area (Figure 11,
mean of 1939– 1998) indicates least flooding at the end of
the dry season in November. The flooded area increases from
February to April, in the southern portions of the basin. The
maximum simulated mean monthly flooded extent occurs in
April and May. The flooding shifts to the northern portions
of the basin late in the wet season (May – July), and finally
decreases throughout the basin (after August).
[58] As with the simulated discharge, there is considerable year-to-year variation in the simulated mean annual
flooded extent (Figure 12). Consistent with the discharge,
the deviation of the annual flooded area from the mean is
greater in the 1940s– 50s and 1970s, less in the 1960s and
1980s– 90s. The coefficient of variation is 18% (standard
Figure 12. Simulated mean annual flooded area in km2 summed for the entire Amazon River basin.
LBA
11 - 16
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
deviation/mean) with the strongest negative departure from
the mean (about 50%) occurring in 1992 coincident with
the dry period of the 1980s– 90s and the strong El Niño of
1992 – 93 (Figure 12). The maximum departure from the
mean annual flooded area for the basin (about +30%)
occurs in 1949 at the peak of the long wet period of the
1940s – 50s.
5. Summary and Conclusions
[59] The IBIS ecosystem model and the HYDRA surface hydrology model were used together to simulate the
river discharge and flooded area from historical climate
records from 1939 – 1998. Evaluation of the results
against diverse observations indicates that estimates of
precipitation are likely greatly underestimated outside of
the Brazilian portion of the Amazon Basin. As a result,
simulated discharge and flooded area are consistently
underestimated for watersheds with significant input from
non-Brazilian portions of the basin. However, despite
poor input precipitation in portions of the basin, the
seasonal and interannual variability of the river discharge
is relatively well simulated for most of the large watersheds.
[60] The flooding algorithm within HYDRA simulates
the behavior of seasonal flooding on the Amazon well.
The monthly and interannual deviation of the simulated
river water height from the mean agrees well with the
observations for the period 1992 – 1998 on the main stem
of the Amazon River (r2 = 0.79), particularly downstream
of the confluence of the Negro and Solimões Rivers. The
simulated flooded area on the main stem of the Amazon
River is also in relatively good agreement with independent estimates of water area for the period 1979 – 1986
(within 25% of estimates for 6 of the 8 years available for
comparison).
[61] The simulated discharge and flooding for the period
1939 – 1998 show results consistent with previous examinations of observed discharge data [Marengo et al., 1998;
Richey et al., 1989; Zeng, 1999]. Discharge and flooding are
increased relative to the mean for the period in the 1940s –
50s and 1970s, and decreased in the 1960s and 1980s– 90s.
El Niño years are associated with generally decreased
simulated discharge (e.g., 1941 – 42, 1951 – 52, 1982 – 83,
1986 – 87, and 1992– 93) and La Niña years with increased
discharge (e.g., 1945– 46, 1950, 1955 – 56, 1962, 1974– 75,
1988 – 89).
[62] Future studies of the Amazon/Tocantins River basin
could be improved through: (1) more accurate input data
sets, such as precipitation, river channel geometry, current
and historical land use patterns, and soil type and texture;
(2) better characterization of model parameters, such as
soil hydraulic properties; and (3) improvements to the
models themselves. For example, more accurate and
higher-resolution digital elevation models may improve
the simulation of the ratio of water height to flooded area
by better defining the basin topography. More accurate
precipitation data will improve the mean annual simulation. Furthermore, analysis of the simulated soil characteristics and discharge velocity should improve the simulation
of the water budget within IBIS and HYDRA. Finally,
making the flooding algorithm a fully dynamic component
of the model should improve the simulated discharge and
flooded area.
[63] Acknowledgments. We would like to thank John Melack,
Christine Delire, and three anonymous reviewers for suggesting numerous
improvements to this manuscript. We would also like to thank Steve
Hamilton for providing the flooding estimates for comparison. This work
was supported by an EOS Interdisciplinary Science grant from the NASA
Office of Earth Science, by a cooperative agreement with the NASA
Goddard Institute for Space Sciences at Columbia University, and by
NASA through the LBA-Ecology program. M. H. Costa is a CNPq
fellow.
References
Bates B. C., and A. P. J. De Roo, A simple raster model for flood inundation simulation, J. Hydrol., 236, 54 – 77, 2000.
Birkett C. M., Contribution of the TOPEX NASA radar altimeter to the
global monitoring of large rivers and wetlands, Water Resour. Res., 34(5),
1223 – 1239, 1998.
Birkett C. M., Synergistic remote sensing of Lake Chad: Variability of basin
inundation, Remote Sens. Environ., 72, 218 – 236, 2000.
Birkett C. M., L. Mertes, T. Dunne, M. H. Costa, and M. J. Jasinski,
Surface water dynamics in the Amazon Basin: Application of satellite
radar altimetry, J. Geophys. Res., 107, 10.1029/2001JD000609, in press,
2002.
Botta A., N. Ramankutty, and J. A. Foley, Long-term variations of climate
and carbon fluxes over the Amazon basin, Geophys. Res. Lett., 29(9),
10.1029/2001GL013607, 2002.
Cardille J. A., et al., Characterizing patterns of land use and land cover in
Amazonia by merging satellite images and agricultural censuses, Global
Biogeochemical Cycles, 16, 10.1029/2000GB001386, 2002.
Coe M., and J. Foley, Human and natural impacts on the water resources of
the Lake Chad basin, J. Geophys. Res., 106, 3349 – 3356, 2001.
Coe M. T., A linked global model of terrestrial hydrologic processes: Simulation of modern rivers, lakes, and wetlands, J. Geophys. Res., 103,
8885 – 8889, 1998.
Coe M. T., Modeling terrestrial hydrological systems at the continental
scale: Testing the accuracy of an atmospheric GCM, J. Clim., 13,
686 – 704, 2000.
Coe M. T., and S. P. Harrison, Simulating the water balance of northern
Africa during the mid-Holocene: An evaluation of the 6 ka BP PMIP
experiments, Clim. Dyn., 19, 155 – 166, 2002.
Cogley G., Runoff from the World’s Landmasses: Amounts and Uncertainties at 2-Degree Resolution, p. 25, Trent Univ., Trent, Ont., 1989.
Costa M. H., and J. A. Foley, Water balance of the Amazon Basin: Dependence on vegetation cover and canopy conductance, J. Geophys. Res.,
102, 23,973 – 23,989, 1997.
Costa M. H., C. H. C. Oliveira, R. G. Andrade, T. R. Bustamante, F. A.
Silva, and M. T. Coe, A microscale hydrological data set of river flow
routing parameters for the Amazon Basin, J. Geophys. Res., 107,
10.1029/2001JD000309, in press, 2002.
de Noblet-Ducoudré N., E. Poutou, J. Chappellaz, M. T. Coe, and G.
Krinner, Indirect relationship between surface water budget and wetland
extent, Geophys. Res. Lett., 29(4), 51 – 54, 2002.
Donner S. D., M. T. Coe, J. D. Lenters, T. E. Twine, and J. A. Foley,
Modeling the impact of hydrological changes on nitrate transport in the
Mississippi River Basin from 1955 – 1994, Globa Biogeochemical Cycles, 16, 10.1029/2001GB001396, in press, 2002.
Foley J. A., C. I. Prentice, N. Ramankutty, S. Levis, D. Pollard, S. Sitch,
and A. Haxeltine, An integrated biosphere model of land surface processes, terrestrial carbon balance, and vegetation dynamics, Glob. Biogeochem. Cycles, 10(4), 603 – 628, 1996.
Kousky V. E., M. T. Kagano, and F. A. Cavalcanti, A review of the southern
oscillation: Oceanic-atmospheric circulation changes and related rainfall
anomalies, Tellus, 36A, 490 – 504, 1984.
Kucharik C. J., J. A. Foley, C. Delire, V. A. Fisher, M. T. Coe, J. D. Lenters,
C. Young-Molling, N. Ramankutty, J. M. Norman, and S. T. Gower,
Testing the performance of a Dynamic Global Ecosystem Model: Water
balance, carbon balance, and vegetation structure, Glob. Biogeochem.
Cycles, 14(3), 795 – 825, 2000.
Leemans R. and W. P. Cramer, The IIASA Database for Mean Monthly
Values of Temperature, Precipitation, and Cloudiness on a Global Terrestrial Grid, p.62, Int. Inst. Appl. Syst. Anal., Laxenburg, Austria,
1991.
Lenters J. D., M. T. Coe, and J. A. Foley, Surface water balance of the
continental United States, 1963 – 1995: Regional evaluation of a terrestrial biosphere model and the NCEP/NCAR reanalysis, J. Geophys. Res.,
105, 22,393 – 22,425, 2000.
COE ET AL.: LONG-TERM SIMULATIONS OF DISCHARGE AND FLOODS
Marengo J. A., Interannual variability of surface climate in the Amazon
Basin, Int. J. Climatol., 12, 853 – 863, 1992.
Marengo J. A., Variations and change in South American streamflow, Clim.
Change, 31, 99 – 117, 1995.
Marengo J. A., L. M. Druyan, and S. Hastenrath, Observational and modeling studies of Amazonia interannual climate variability, Clim. Change,
23, 267 – 286, 1993.
Marengo J. A., J. Tomasella, and C. R. Uvo, Trends in streamflow and
rainfall in tropical South America: Amazonia, eastern Brazil, and northwestern Peru, J. Geophys. Res., 103, 1775 – 1783, 1998.
Miller J. R., G. L. Russell, and G. Caliri, Continental-scale river flow in
climate models, J. Clim., 7, 914 – 928, 1994.
Nepstad D. C., et al., Large-scale impoverishment of Amazonian forests by
logging and fire, Nature, 398(6727), 505 – 508, 1999.
New M., M. Hulme, and P. D. Jones, Representing 20th century space-time
climate variability, II, Development of 1901 – 1996 monthly terrestrial
climate fields, J. Clim., 13, 2217 – 2238, 2000.
Richey J. E., C. Nobre, and C. Deser, Amazon river discharge and climate
variability: 1903 to 1985, Science, 246, 101 – 103, 1989.
Sippel S. J., S. K. Hamilton, J. M. Melack, and E. M. M. Novo, Passive
microwave observations of inundation area and the area/stage relation in
the Amazon River floodplain, Int. J. Remote Sens., 19(16), 3075 – 3096,
1998.
Skole D., and C. Tucker, Tropical deforestation and habitat fragmentation in
the amazon: Satellite data from 1978 to 1988, Science, 260, 1905 – 1910,
1993.
LBA
11 - 17
Skole D. L., W. H. Chomentowski, W. A. Salas, and A. D. Nobre, Physical
and human dimensions of deforestation in Amazonia, BioScience, 44(5),
314 – 322, 1994.
Victoria R. L., L. A. Martinelli, J. M. Moraes, M. V. Ballester, and A. V.
Krusche, Surface air temperature in the Amazon region and its borders
during this century, J. Clim., 11, 1105 – 1110, 1998.
Vorosmarty C. J., B. Moore, A. L. Grace, M. P. Gildea, J. M. Melillo, B. J.
Peterson, E. B. Rastetter, and P. A. Steudler, Continental scale models of
water balance and fluvial transport: An application to South America,
Glob. Biogeochem. Cycles, 3(3), 241 – 265, 1989.
Zeng N., Seasonal cycle and interannual variability in the Amazon hydrologic cycle, J. Geophys. Res., 104, 9097 – 9106, 1999.
A. Botta and M. T. Coe, Center for Sustainability and the Global
Environment, Institute for Environmental Studies, University of Wisconsin,
1225 West Dayton St., Madison, WI 53706, USA. (coe@phoebus@
meteor.wisc.edu)
M. H. Costa, Federal University of Viçosa, Viçosa, Minas Gerais, 36571000, Brazil.
C. Birkett, ESSIC, University of Maryland at College Park, NASA/
GSFC, Mail Code 923, Greenbelt, MD 20771, USA.