Experimental design
A hydrological-hydrodynamic model (MGB; (Collischonn et al., 2007)) is
set up for a case study in the Amazon (Purus River Basin) with a priori
parameter sets based on their variability as reported in literature
(references in Table S2). The study is then divided into two steps.
Firstly, a sensitivity analysis is performed to understand how different
parameter sets (river hydraulic, soil, vegetation) affect model output
variables (river discharge, flood extent, river water level, soil
moisture, evapotranspiration and terrestrial water storage).
Then, a calibration step is performed in which the model is calibrated
with the well-known MOCOM-UA optimization algorithm (Yapo et al.,
(1998)) considering six variables: (1) in-situ streamflow (one gauge at
the basin outlet), and RS freely available, state-of-the-art
observations of (2) water level (one satellite altimetry virtual
station), (3) flood extent (sum of flooded areas over the Lower Purus
River Basin), (4) terrestrial water storage (TWS), (5)
evapotranspiration, and (6) soil moisture. Variables (4), (5) and (6)
are averaged over the whole basin. The calibration of each variable is
performed individually (single-variable), and evaluated for all
variables. All calibration
experiments are repeated three times with differing initial parameter
sets to ensure that convergence is not dependent on the initial
parameter sets. Given limitations on the availability of simultaneous RS
time coverage, the model is calibrated for one time period (2009-2011),
and evaluated for: (i) the same time period of calibration; and (ii) for
a different period (2006–2008 for discharge, flood extent, TWS, ET and
2013–2014 for water level and soil moisture). To understand how lumped
calibration can retrieve the remotely sensed spatial patterns, a
qualitative evaluation is provided additionally. A final test is
performed in which two multi-variable calibration experiments are
conducted: (i) calibration with all analyzed variables, except
discharge; and (ii) calibration with two complementary variables (water
level and soil moisture), which are selected for simultaneous
calibration for being complementary and having led to satisfactory
calibration performance.
Study area: Purus River
Basin
The Purus River Basin (Figure 2) in Amazon has a drainage area of
approximately 236,000 km², and river discharge ranges from around 1,000
(June-December) to 12,000 m³/s (January-July) at Canutama gauge. Because
of its large area, it is compatible with the spatial resolution of RS
products (e.g., a pixel of GRACE presents spatial resolution of roughly
300-400 km). Purus river has minor anthropogenic influences, which
simplifies the modeling process. The climate is equatorial (Figure 2d),
and mean annual rainfall is 2147 mm/year (according to in-situ gauges).
Purus was selected because of its representativeness of tropical regions
as the Amazon basin, which is the largest river in the world (Holeman,
1968), and it is characterized by extensive floodplains (Junk, 1997).
For instance, on the lower Purus, the floodplain width is in the order
of 30 km, which corresponds to approximately 30 times the main channel
width (Paiva et al., 2011). These floodplains allow a satisfactory flood
extent monitoring by RS image classification, which contributes to the
suitability of Purus River Basin for this study.