Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

20 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO.

1, MARCH 2010

Estimating River Depth From Remote Sensing Swath


Interferometry Measurements of River
Height, Slope, and Width
Michael Durand, Ernesto Rodríguez, Douglas E. Alsdorf, and Mark Trigg

Abstract—The Surface Water and Ocean Topography (SWOT) vation (WSE) measurements taken nearly continuously to pe-
mission is a swath mapping radar interferometer that would riodic measurements of flow velocity and channel cross-sec-
provide new measurements of inland water surface elevation tional area, from which instantaneous river discharge is derived.
(WSE) for rivers, lakes, wetlands, and reservoirs. SWOT WSE
estimates would provide a source of information for character- Stream gages provide the WSE or stage and are located sparsely
izing streamflow globally and would complement existing in situ at individual points along channels. From these instantaneous
gage networks. In this paper, we evaluate the accuracy of river measurements of discharge and concurrent stage measurements,
discharge estimates that would be obtained from SWOT measure- a “rating curve” is developed that relates discharge to WSE. The
ments over the Ohio river and eleven of its major tributaries within rating curve, or stage-discharge relationship, is then applied to
the context of a virtual mission (VM). SWOT VM measurements
are obtained by using an instrument measurement model coupled the continuous stage measurements to produce discharge as a
to simulated WSE from the hydrodynamic model LISFLOOD-FP, continuous derived measurement.
using USGS streamflow gages as boundary conditions and vali- As discussed by Alsdorf et al. [1], there have been a variety
dation data. Most model pixels were estimated two or three times of attempts to characterize river discharge via remote sensing
per 22-day orbit period. These measurements are then input into measurements. One approach is to use airborne [2], [3] or space-
an algorithm to obtain estimates of river depth and discharge. The
algorithm is based on Manning’s equation, in which river width borne [4] measurements to estimate the fluvial surface velocity.
and slope are obtained from SWOT, and roughness is estimated a Another approach relates river width or inundated area to river
priori. SWOT discharge estimates are compared to the discharge discharge, either at gaging stations [5], [6] or combined with es-
simulated by LISFLOOD-FP. Instantaneous discharge estimates timates of shoreline elevations [7]–[9]. A third approach is used
over the one-year evaluation period had median normalized root in studies that relate water elevations measured from radar al-
mean square error of 10.9%, and 86% of all instantaneous errors
are less than 25%. timeters to discharge, e.g., [10]. A fourth approach is to derive
estimates of river discharge from spaceborne measurements of
Index Terms—Hydrology, interferometry, radar, remote sensing. gravity fluctuations from the Gravity Recover And Climate Ex-
periment (GRACE) measurements [11]–[13].
I. INTRODUCTION The Surface Water and Ocean Topography (SWOT) mission
is a swath mapping radar interferometer that would provide
measurements of inland WSE for rivers, lakes, wetlands and
IVER discharge denotes the volume flowrate of water
R moving through a fluvial channel. River discharge has tra-
ditionally been estimated in situ by relating water surface ele-
reservoirs. SWOT has been recommended by the National Re-
search Council (NRC) Decadal Survey [14] to measure ocean
topography as well as WSE over land; the proposed launch date
timeframe recommended by the NRC is between 2013–2016.
In contrast with traditional radar altimeters, SWOT will directly
Manuscript received March 20, 2009; revised August 25, 2009.First pub- measure fluvial inundated area as well water elevation, with
lished November 10, 2009; current version published February 24, 2010. This spatial pixels on the order of 10 s of meters. Average revisit
work was supported in part by the NASA Terrestrial Hydrology Program; in part
by the NASA Physical Oceanography program; in part by the Jet Propulsion times will depend upon latitude, with two to four revisits at
Laboratory, California Institute of Technology, Pasadena, CA, under a contract low to mid latitudes and up to ten revisits at high latitudes per
with NASA; in part by the Climate, Water, and Carbon Program of The Ohio 22-day orbit repeat period. Although SWOT WSE estimates
State University; and in part by the U.K. Natural Environment Research Council
(Grant NER/S/A/2006/14062). will provide a source of information for characterizing stream-
M. Durand is with the Byrd Polar Research Center, The Ohio State University, flow globally, SWOT is not designed to replace stream gages.
Columbus, OH 43210 USA (e-mail: durand.8@osu.edu).
E. Rodríguez is with the Jet Propulsion Laboratory, California In-
Stream gauges can supply timely measurements (e.g., daily
stitute of Technology, Pasadena CA 91109 USA (e-mail: ernesto.ro- or even real-time) and measure discharge in small tributaries
driguez@jpl.nasa.gov). draining headwater catchments; thus, a close tie to rainfall gen-
D. E. Alsdorf is with the Byrd Polar Research Center, School of Earth Sci-
ences, and the Climate, Water, and Carbon Program, The Ohio State University,
erated streamflow. In contrast, the SWOT orbit will not supply
Columbus, OH 43210 USA (e-mail: alsdorf.1@osu.edu). daily data nor will the mission operate in real-time mode. The
M. Trigg is with the School of Geographical Sciences, University of Bristol, instrument’s spatial resolution limits the ability of SWOT to
Bristol, BS8 1SS, U.K. (e-mail: mark.trigg@bristol.ac.uk). estimate discharge in rivers having a small width. Streamflow
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. estimates derived from SWOT and gages will be complemen-
Digital Object Identifier 10.1109/JSTARS.2009.2033453 tary. Whereas in situ gages have real-time capability at a single
1939-1404/$26.00 © 2010 IEEE

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 21

point, SWOT measurements will span all rivers, and measure II. MODELS, METHODS, AND DATA
elevations between gages, but will only provide a measurement
weekly, not daily, for most locations. A. LISFLOOD-FP Model
Methods for estimating discharge from SWOT are still The LISFLOOD-FP model [17], [20] uses a 1-D finite differ-
being developed. Andreadis et al. [15] used the ensemble ence hydrodynamic scheme to solve for water depth, velocity,
Kalman filter (EnKF) to update discharge within the context and discharge in channel flow. Although LISFLOOD-FP also
of a data assimilation scheme. The Andreadis et al. approach includes functionality for floodplain inundation, in this study we
(conducted in the context of what has been termed the SWOT only utilize the channel solver in order to focus on developing
“virtual mission” or VM) was executed as follows: syntheti- a retrieval algorithm for in-channel flow. In order to achieve a
cally generated SWOT measurements were assimilated into the parsimonious implementation, a rectangular river cross section
variable infiltration capacity (VIC) hydrologic model [16] and is assumed. The model uses the diffusion wave approximation to
the LISFLOOD-FP hydrodynamic model [17] for a reach of the full Saint Venant equations of flow. If this is written in terms
the Ohio river. River depth was estimated as a state variable, of discharge, where Manning’s equation is used to describe ve-
assuming that channel bathymetry and roughness were known locity for a rectangular cross section and large width-to-depth
a priori.. Another approach, also conducted in the context of ratio, we have
the SWOT VM, used a data assimilation system to estimate the
slope of the channel bed elevation from seasonal measurements
of inundated area [18]. Once the bathymetry was estimated, (1)
the river depth and discharge could be estimated. While data
assimilation algorithms have the advantage of bringing as much where is the roughness coefficient, is the river width, is
ancillary data to bear on the problem of estimating discharge, the flow depth, is the bed slope, and is the distance along
they typically rely on ensemble-based estimates of these inputs the channel. Thus, the final term in (1) represents the slope of
and thus require significant computational resources. This the water surface. By taking this term into account, the diffusion
high computational expense may have implications for global wave approximation can be used to model the effects of changes
application of assimilation algorithms. A simpler approach that in flow downstream on the flow conditions upstream, including
is less computationally expensive has been outlined based on backwater effects. Such changes are manifested in changes in
using SRTM estimates of water elevation and slope in conjunc- water surface slope, which will be of crucial importance to our
tion with Manning’s equation by [19], assuming that the river estimates of channel depth.
depth is known a priori.
In this paper, our goal is to present a method of estimating B. Study Area, Model Setup and Inputs
river depth directly from SWOT measurements, since a priori. Our study area is the Ohio river basin, with a total drainage
depth estimates will not be available globally. We will perform area of approximately 529,000 km [21]. We chose a total of
the further step of evaluating the sensitivity of this method- eleven of the major Ohio river tributaries to include in the model;
ology to differences in SWOT orbits. At this time, it is expected the eleven tributaries and their respective drainage areas from
that SWOT will have two operational phases. During the “fast [21] are listed in Table I. From Table I, the contributing area
phase”, the SWOT orbit will repeat every three days, but spatial of these eleven tributaries comprises a total of 401,012 km , or
coverage will be limited; the duration of the fast phase will be 76% of the drainage area of the mainstem Ohio. The remaining
between three and six months. During the “nominal phase”, the 24% of the drainage area is comprised by smaller rivers observ-
SWOT orbit will repeat every twenty-two days, but spatial cov- able by SWOT, and by streams and lateral inflow that SWOT
erage will be global; the nominal phase will constitute the rest of will not be able to measure. We have chosen to work with these
the mission lifetime. Note that because of the swath sampling, eleven rivers in order to achieve a parsimonious model setup,
revisits to any given location will occur at least twice during a and to demonstrate proof-of-concept. Follow-on studies will in-
repeat cycle. clude all rivers, and examine which will be able to be charac-
Our overarching goal in this study is to characterize how ac- terized by SWOT measurements. In order to model these eleven
curately depth and discharge will be estimated. We test this ap- tributaries in LISFLOOD-FP, estimates of the river centerlines,
proach as part of the SWOT hydrology VM. Our methodology channel bed elevation along the centerline, and channel width
proceeds as follows: 1) a model of the “true” discharge, water are needed.
elevations, widths, and slopes is constructed using a hydrody- 1) LISFLOOD-FP Channel Inputs: The Hydro1K dataset
namic model; 2) synthetic SWOT observations are generated was used to provide estimates of the river centerline and channel
with realistic (to first order) error characteristics; 3) these ob- bed elevation. Hydro1K was derived from the GTOPO30 dig-
servations are used to estimate river depth and to calculate dis- ital elevation model (DEM) at 30 arc-second resolution [22].
charge; and 4) the SWOT measurements of discharge are com- Stream centerlines are derived from the DEM at approximately
pared with the true discharge to characterize SWOT discharge 1-km spatial resolution as described in [22], and represented as
accuracy. a series of sequential location points (i.e., latitude and longi-
tude). Some of these points have additional data describing the
river cross section geometry: width, bed elevation and rough-
ness. The DEM elevation will be used in this study to represent
the channel bed elevation. The Hydro1K data for the Ohio river

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
22 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO. 1, MARCH 2010

TABLE I
DRAINAGE AREA OF EACH OF THE ELEVEN TRIBUTARIES INCLUDED IN THE MODEL AS ESTIMATED BY [21], AS WELL AS INFORMATION FROM THE USGS GAGE
USED AS A BOUNDARY CONDITION FOR EACH TRIBUTARY: GAGE ID, CONTRIBUTING AREA, AND DISCHARGE AVERAGED OVER THE STUDY PERIOD

Landsat5 and Landsat7 imagery [24]. The NLCD classifica-


tion was used along with the algorithm developed Smith and
Pavelsky [25] to extract the river width for all of the rivers in
the Ohio basin, resulting in a raster image of the river width.
This raster image was intersected with streamlines from the
Hydro1K: all river width estimates from the raster image that
fell near each segment in a streamline were averaged to obtain
an estimate of the width for that segment. The resulting widths
are shown in Fig. 1.
2) LISFLOOD-FP Boundary Conditions: The diffusion
wave implementation of LISFLOOD-FP requires upstream
discharge boundary conditions for each tributary, as well as
a downstream depth boundary condition on the mainstem.
Note that the mainstem itself provides the downstream depth
boundary condition for the tributaries. In this study, we use
United States Geologic Survey (USGS) gages to provide the
Fig. 1. Map of the Ohio river basin is shown; tributaries included in the model depth and discharge boundary conditions from 1 June 1991–31
are shown in blue, while excluded tributaries are shown in grey. Streamlines are
from the Hydro1K dataset. River widths are shown by the relative thickness of May 1992. The location of each gage used as a boundary
the blue lines. USGS gages used for boundary conditions are shown as circles. condition is shown in Fig. 1. The USGS gage information
and mean annual discharge are shown in Table I. Gages were
chosen as far downstream as possible, in order to represent
basin is shown in Fig. 1. The LISFLOOD-FP model interpo- the total flow as completely as possible for each tributary; see
lates the latitude and longitude data from Hydro1K to a regular Fig. 1. From Table I, the gages represent between 67% and 99%
grid; in this case, we use a spatial resolution of 1 km. Thus, the of the drainage area of each tributary. As a whole, the gages
blue streamlines in Fig. 1 show the extent of the study area that represent a total of 342,857.4 km , which is approximately
is simulated in LISFLOOD-FP. The channel roughness (Man- 65.2% of the total Ohio river basin drainage area of 525,767.6
ning’s ) was assumed to be 0.03 for each channel. km from Table I. The downstream boundary condition thus
The LISFLOOD-FP model requires an estimate of river width represents hydrologic processes operating on the entire basin,
for each channel segment. For this study, we use the National of which only 65.2% is represented by our boundary condi-
Land Cover Dataset (NLCD) 2001 [23], which was derived from tions. In order to deal with this issue, we calculate a reduced

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 23

downstream water depth boundary condition from the roughness estimation at the end of this section. Equation (4) can
original boundary condition be recast such that it is linear in and by recognizing that ,
, and are known and defining
(2)
(5)
where is the sum of the mean annual discharge at the eleven
gages used as upstream boundary conditions (Table I) and resulting in
is the mean annual discharge from downstream boundary con-
ditions. Equation (2) follows from assuming that Manning’s (6)
equation holds, and calculating the reduction in depth from a
given reduction in discharge, given that width and roughness This can be rearranged to yield
are constant. Presumably, a significant fraction of the 35.8% of
the drainage area that is not included in this model would pro- (7)
duce runoff in channels large enough to be measured by SWOT,
though they are not included in this simplified model, as noted where the vector contains the unknown initial depths
in Section II-B.
(8)
C. Depth Estimation Algorithm
Our goal in this paper is to explore a method of estimating the matrix contains combinations of the observed , with a
river depth from SWOT measurements. The LISFLOOD-FP number of rows corresponding to the total number of measure-
model assumes a rectangular cross-section, which implies that ment times,
width is time-invariant. We proceed by making two assump-
tions: 1) most of the time, for most rivers, the discharge at
one point along the channel will not likely be significantly
different than the discharge at another point along the channel .. .. (9)
. .
a small distance (i.e., several kilometers) away, assuming no
major changes in contributing area (i.e., no major tributaries
between the two points); 2) most of the time, for most rivers, and the vector contains combinations of and
the effects of downstream changes in flow will not have a
significant effect on the flow upstream or downstream; in other
words, the kinematic wave approximation holds. We will refer
to these two assumptions as “the continuity assumption”, and .. (10)
.
the “the kinematic assumption,” hereafter; moreover, we will
investigate how well they hold for our model setup, below.
Note that the continuity assumption is subject to errors due to Assuming that there are more than two independent measure-
lateral inflows entering a river through channels that are too ment times, (7) represents an overconstrained set of linear equa-
narrow to be accurately characterized by SWOT, as discussed tions, which can be readily solved by finding the value of
in Section II-B2. Note also that we do not invoke the continuity that minimizes the least-squares differences for equations. In
assumption if a tributary joins the river that is wide enough to order for (7) to be solvable by this method, there must be more
be accurately characterized by SWOT. than two linearly independent rows in (9); otherwise, will be
Given the continuity assumption, we can equate the discharge singular, and will not be solvable. As roughness and width
at pixel and pixel at every time are time-invariant, all temporal variability in will be due to
temporal variability in slope. Slope time series variability in the
(3) model will be discussed in our results, below. It should be noted
that roughness could also be solved for using this approach, al-
Given the kinematic assumption, we have
though the minimization of residuals required to solve (7) would
then be over a set of nonlinear equations, which would be more
complex.
(4) The depth estimation analysis in (7) will be applied only
where is the depth at some initial measurement time, and between pixels if slope time series variability as measured by
is the change in water depth at time . Width, slope, and the coefficient of variation is greater than some arbitrary
roughness are defined as for (1), and roughness is assumed not threshold , and if the matrix is nonsingular. The latter
to vary in time. Note that , , and are all SWOT ob- condition will be evaluated by the matrix condition number
servables, and and are unknown. The depth at any time in the norm; is defined as the ratio of the largest singular
is given from and . Since our objective in this study is value of to the smallest singular value [27] as determined by
to estimate depth, we will make the assumption that roughness a singular value decomposition of . The analysis will only be
can be adequately estimated from ancillary data; we will discuss performed if is less than some arbitrary threshold.

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
24 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO. 1, MARCH 2010

The depth estimation analysis laid out in this section proceeds


as follows. For each pixel in the model domain, we first test
whether or not the slope time series coefficient of variation that
pixel exceeds . If not, we will estimate depth using
the interpolation methods presented in Section II-D. Otherwise,
we search the 5-km neighborhood of pixel and define pixels
within the neighborhood that satisfy the condition that ex-
ceeds . (The “5-km neighborhood” is defined as five pixels
upstream and five pixels downstream of pixel ). If no pixels are
found that satisfy this condition, we will estimate depth using
the interpolation methods presented in Section II-D. If one or
more pixels are found that satisfy the condition, then we imple-
ment (6) by choosing pixel from the 5-km neighborhood of
pixel . If one pixel is found that satisfies the condition, then
pixel is chosen to be that pixel. If multiple pixels are found
that satisfy the condition, then pixel is chosen to be the pixel
with the greatest value of . Thus

(11)

where is the number of pixels within the 5 km neighborhood


of pixel that meet the required slope time series condition.
Note that in order to avoid violation of the continuity assump- Fig. 2. Ohio basin rivers (streamlines) as represented in the LISFLOOD-FP
tion, depth estimation will only be performed for two pixels model are shown, along with representative SWOT measurement swaths from
and along the mainstem if there is not a tributary joining in the 22-day 78-degree orbit for ascending (a) and descending (b) passes. The
labels near the bottom of each swath indicate the day on which the measurement
between the two pixels. occurred. For simplicity, only measurements on the first seven days of the orbit
are shown.
D. Interpolation of Depth Estimates
There are two issues that could pose a difficulty to the method
laid out above. For the purpose of clarifying discussion, we de- drainage area for all rivers and river sections that have unique
fine a river “section” as a part of a river without tributaries en- values of drainage area. We then use this power law to predict
tering. Within the context of this study, each of the eleven trib- the discharge in the sections of the river where no depth esti-
utaries constitute a single section. In contrast, the mainstem is mates could be obtained using the depth algorithm described in
composed of twelve sections, where each section runs from the the previous section. The depth at the initial time is then calcu-
inflow of one tributary to the next. First, it is to be expected that lated from the discharge estimate obtained from the power law.
some—but not all—pixels of a given river section will meet the
slope timeseries variability condition described above. How will E. Obtaining SWOT Observations
the depth be estimated for pixels that do not meet this condition? SWOT observations are obtained by overlaying SWOT
The second issue is that for some sections of the mainstem Ohio swaths on the LISFLOOD-FP pixels. This is done by first
river, there may not be any pixels that meet the slope timeseries generating a SWOT ground track based on orbital elements; the
variability condition. How will depth be estimated within these ground track is represented as a series of latitude, longitude, and
river sections? spacecraft heading as a function of time, as measured from the
To deal with the first issue, we perform an interpolation over beginning of the orbit period. From this ground track, a swath
all pixels for a given section of the river where depth estimates of SWOT ground coverage polygons is generated from geomet-
for some pixels were obtained from the algorithm described rical considerations, and the specified swath width of 140 km.
in Section II-C. First, the discharge at the initial time is esti- As an example, Fig. 2 shows the SWOT coverage of the Ohio
mated for all locations where measurements exist from Man- river basin model as represented by the LISFLOOD-FP model
ning’s equation. Second, the discharge estimates obtained in described above. After overlaying the swaths, a spatial inter-
the first step are averaged together. Third, it is assumed that section of each LISFLOOD-FP pixel and each of the ground
the discharge for all locations where no depth estimate is avail- coverage polygons is performed to determine the precise time
able is equal to the average discharge obtained in the second at which each pixel is measured.
step. Fourth, the initial depth for locations where no depth es- After the measurement times are determined, synthetic
timate is available is calculated from the discharge assumed in SWOT measurements of river slope, river width, and river
the third step. To deal with the second issue, we have available height are generated at each model pixel by using the LIS-
the depth and discharge estimates at the initial time for sections FLOOD-FP output as the “true” model states and corrupting
of the river where depth was successfully estimated. We first the true states with measurement error. In this study, we include
construct a power law between discharge at the initial time and only measurement error of river height, as the spatial resolution

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 25

of the LISFLOOD-FP model at 1 km resolution observations derived from model results (3.2). We then evaluate
precludes accurate representation of SWOT errors of slope and the extent to which our assumptions hold (3.3). Finally, we
width. In reality, SWOT pixels will have a spatial resolution in present results from the depth estimation (3.4), the resulting
the cross-track direction that ranges from 10 m in the far swath discharge errors, and the sensitivity of the latter to various orbit
to 60 m in the near swath. Spatial resolution in the along-track configurations (3.5).
direction in the best-case scenario will be 2 m due to synthetic
aperture processing. Along-track spatial resolution may have A. LISFLOOD-FP Model Evaluation
the potential to degrade slightly due to temporal decorrelation Because our primary goal in this study was to evaluate a
of the scene [26], however; studies to investigate these effects methodology for estimating river depth from SWOT observa-
are ongoing. In this study, we make the very conservative tions, we did not perform any model calibration. In order to
assumption that SWOT spatial resolution in both along-track verify that the LISFLOOD-FP model set up is producing reason-
and cross-track will be approximately 50 m, and that able results, we compare the discharge predicted at the down-
each of these 50 50 m pixels will be characterized by zero stream model outlet and the discharge observed at the most
mean Gaussian random errors with a standard deviation of downstream USGS gage available. Note that the drainage-area-
0.5 m. SWOT errors for river height are simulated by adjusted water depth data from this gage are used as the down-
noting that the dominant errors in river height measurements stream boundary condition for LISFLOOD-FP. The gages used
derive from thermal noise, are additive in nature, and can for the upstream boundary conditions represent only 65.2% of
therefore be modeled as the total Ohio river drainage area, and the USGS gage obvi-
ously represents all of the drainage area. Thus, in order to assess
(12) the timing of streamflow predicted by LISFLOOD-FP, we have
scaled the USGS discharge time series by the ratio of the sum of
where is the number of SWOT pixels that would be con- the mean annual discharge for each upstream gage to the mean
tained within a LISFLOOD-FP model pixel; is calculated annual discharge for the downstream gage. The modeled and
individually for each pixel from the river width , , measured discharge time series are shown in Fig. 3; the mod-
and eled discharge clearly matches the observed discharge to first
order. Discharge for both the model and gage ranges from 2,000
(13) m s to 4,000 m s between June and December. During
December there is a significant increase in discharge to approx-
imately 17,000 m s for the gage; this peak is somewhat un-
In order to solve for the initial depth as described above, es-
derestimated by the model. Both model and gage decrease to
timates of are obtained between successive SWOT mea-
approximately 4,000 m s in mid-February, before showing
surement times from the sum of true water heights and randomly
two peaks in March and April. From this, we conclude that the
generated based on (12).
model is adequately representative of reality to investigate our
F. Error Metrics and Evaluation methods of estimating depth and discharge.

Depth errors will be evaluated by comparing the estimated B. SWOT Observations


and true values of at each pixel on each river. Two types of Examples of SWOT observations derived from the LIS-
discharge errors are considered here: 1) the difference between FLOOD-FP model results are shown in Fig. 4. The simulated
estimated and true values at the time of a SWOT measurement elevation of the water surface for the Wabash river is shown
will be referred to as instantaneous discharge errors, hereafter; along with SWOT measurements derived from these measure-
2) the difference between the monthly average of all instanta- ments as described in Section II-E. Synthetic SWOT measure-
neous discharge estimates and the monthly average of the true ments are shown on two different days, on 11 November and on
discharge from LISFLOOD-FP will be referred to as monthly 24 November. Water elevations along a flow distance of 228 and
discharge errors, hereafter. For instance, the normalized root 212 km are measured on the two days, respectively, indicating
mean square error (NRMSE) for instantaneous discharge error that the 140-km swath is at an angle less than perpendicular
will be evaluated for pixel as for the flow direction at the point where the river was crossed.
We would expect that the monthly sampling error ultimately
derived from these measurements would be closely tied to the
(14) number of times each pixel in the domain is sampled in each
measurement cycle or in each month. Fig. 5 shows a histogram
of the number of times each pixel was measured within the
SWOT measurement cycle. Most pixels are measured either
where is the total number of SWOT observations, and is
two or three times in the 22-day cycle, while 818 of the 5,860
the total number of days simulated.
pixels (14.1%) were measured four times. Over the one-year
period, the average number of measurements per month is
III. RESULTS AND DISCUSSION 3.75, indicating approximately weekly sampling. Note that
We first present results from the evaluation of the LIS- sampling regimes are latitude-dependent, with (on average)
FLOOD-FP model (3.1) and show examples of the SWOT higher latitudes being sampled more frequently.

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
26 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO. 1, MARCH 2010

Fig. 3. LISFLOOD-FP modeled discharge at the downstream model outlet is shown (solid line) as well as the discharge from the USGS gage at the farthest point
downstream on the Ohio river (dashed line). Note that the discharge from the USGS gage has been adjusted by multiplying by the ratio of the sum of the mean
discharge observed at the upstream boundary condition gages to the mean discharge observed at this downstream gage.

Fig. 5. Number of times each of the 5,860 model pixels is measured in a 22-day
cycle with 78-degree inclination angle is shown.

calculated using (14), most pixels have very small error due
Fig. 4. Examples of the synthetic SWOT height observations (circles) gener- to the continuity assumption. For 94.8% of all pixels from all
ated from the true water elevation simulated by LISFLOOD-FP (lines) used for tributaries the NRMSE is less than 5%. As a further test, we
the Cumberland river on day 163 (a) and day 175 (b) of the simulation period. calculated the errors due to both assumptions separately during
the first and second half of the year (i.e., during both high-
and low-flow conditions), and obtained very similar results.
C. Evaluation of Assumptions Errors associated with the continuity assumption are greater
As described above, our algorithm to estimate depth is de- than those for the kinematic assumption; this may be due to the
pendent upon the assumption that the kinematic approximation fact that where tributaries join the mainstem, two pixels may
can be used to represent the flow processes, and that continuity have significantly different contributing areas; this is dealt with
holds between pixels separated by small spatial distances. We as described in Section II-C. Note that we have assumed that
evaluated the kinematic wave assumption by assuming that the streamflow does not increase due to runoff processes in for a
change in water depth with distance is zero in (1), and using given river section between tributaries. Future studies of this
the true model simulated values of , , , and to estimate methodology should examine the sensitivity of the continuity
discharge (referred to as “kinematic discharge,” hereafter), assumption to these runoff processes. Within the context of this
then comparing the kinematic discharge with the true model study, based on the extent to which the continuity and kine-
discharge. For both absolute discharge RMSE, and discharge matic assumptions hold, it is expected that the depth estimation
NRMSE calculated using (14), most pixels have very low levels algorithm described above will be accurate.
of error due to the kinematic assumption. Indeed, for 98.9% Another potential limitation of our method is that it relies on
of all pixels from all tributaries, the NRMSE is less than 5%. slope time series variability to obtain accurate estimates of river
We evaluated the continuity assumption by comparing the depth. In Fig. 6(a), the values of the slope time series coefficient
discharge at each pixel with the discharge at a lag of five pixels. of variation is shown for the Tennessee river. The low values
For both absolute discharge RMSE and discharge NRMSE of for chainage 0–200 km and from 400–1000 km imply that

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 27

Fig. 6. Coefficient of variation  of the slope time series at various points along the Tennessee river is shown (a). The histogram of  across all model pixels
is shown, for the 96% of model pixels with  less than 0.75 (b).

it will be difficult to estimate depth for those pixels. The larger


values of between 200–400 km are likely caused by changes
in the channel slope and channel width at these parts of the river;
the large values of near the confluence of the mainstem after
1000 km chainage are likely due to the effects of the mainstem
on the Tennessee tributary. Based on the fact that some pixels
at least have significant slope time series variability, it is ex-
pected that the depth estimation algorithm described above will
achieve accuracy consistent with the underlying assumptions. A
histogram of values is shown in Fig. 6, for all model pixels
with less than 0.75 (95.8% of the pixels). The mean value of
is 0.1586, and 1293 pixels (22%) have a value of greater
than 0.25.

D. Estimating Depth
There was adequate slope time series variability to estimate
initial depth for a total of 227 pixels. The depth estimate for the
Cumberland river is shown in Fig. 7(a). Depth estimates were
derived from SWOT observables using the algorithm described
in Section II-C for a number of pixels near the downstream por-
tion of the river, where there was adequate temporal variability
in the slope coefficient of variation. For five sections on the
mainstem Ohio, and for the Licking river and Kentucky river,
there were no initial depth estimates; the power law approach
was used to estimate depth for these pixels (see Section II-D).
Relative depth error over all 5,860 pixels from all rivers is shown
in Fig. 7(b). The depth errors show a very slight positive bias,
with a mean relative error of 4.1%. The standard deviation of the
depth error is 11.2%. There are several outliers, with maximum
and minimum errors of 89% and 56%, respectively.
As noted above, the expected mission lifetime will be greater
than three years. The method we have presented used one year
of data to achieve the results described above. It is potentially
of great value to understanding the availability of data products
to know how many measurements are required before the depth
Fig. 7. Initial depth profile for the Cumberland river is shown (a). The true
will be able to be estimated. In Fig. 8, we show the standard depth at the initial time is taken from model output (solid line), the depth esti-
deviation of the depth error as a function of how many days of mates are derived from SWOT observables as described in Section II-C (circles),
data were used to calculate the depth. As the time series be- and interpolated to the pixels where SWOT observations are not available as de-
scribed in Section II-D (dashed line). The relative depth error for depth at the
comes longer, the accuracy of the depth estimate improves. The initial time for all pixels is shown in (b).
standard deviation of the error after 12 months is approximately

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
28 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO. 1, MARCH 2010

Fig. 9. Spatial profiles of discharge along the Kanawha river for four days are
shown. The true discharge profiles are shown as lines, where the solid, dashed,
dotted, and dash-dot lines refer to days 285, 286, 287, and 288, respectively.
The discharge estimates derived from SWOT measurements on days 285 and
288 are shown as points.

Fig. 8. Sensitivity of the standard deviation of the relative error in the initial
depth is shown as a function of the number of simulation days used to obtain
the depth estimate. namics were obtained from the synthetic SWOT measurements.
The large amount of scatter in the discharge estimates is due to
the random error added to the SWOT height observations. These
one half the error after 11 months. The depth algorithm was per- errors could be partially mitigated by utilizing a low-pass filter;
formed for only 204 pixels for the 11-month case, and for 227 e.g., a polynomial could be fitted to the height measurements,
pixels for the 12-month case. For the Cumberland river, the al- following the approach of [19].
gorithm was performed for 16 pixels for the 12-month case (see There are 366 days of simulation time and a total of 5,860
Fig. 7), but was not performed for any pixels for the 11-month pixels, resulting in many spatial and temporal series of dis-
case. This is due to highly variable river slopes during May charge to examine. We summarize these errors by calculating
1992: for the Cumberland river during May 1992 was greater the NRMSE of the discharge time series at each pixel using
than averaged over the other eleven months by a factor 5.02. (14). Both instantaneous discharge errors and monthly dis-
Thus, for the 11-month case, the 648 pixels for the Cumber- charge errors (as described in Section II-F) are shown in
land river were estimated via the interpolation algorithm de- Fig. 10. Instantaneous discharge errors compare only estimated
scribed in Section II-D. The mean of the relative depth error discharge to true discharge only during the measurement times,
for the Cumberland river for the 12-month case was 2.23 cm, with the initial depth estimated using the algorithm presented
but was 62.7 cm for the 11-month case. This bias in the Cum- in Sections II-C and II-D. Monthly discharge errors use true
berland river depth estimates is the reason for the change in the discharge at the SWOT measurement times to calculate a
overall error from the 11-month to the 12-month case shown in monthly discharge estimate, and compare with the true monthly
Fig. 8. These results indicate that the method accuracy for esti- discharge; thus, no depth error is included in the monthly
mating depth from the slope time series is generally better than discharge errors. The median instantaneous discharge error is
that using the interpolation algorithm if no pixel in a given river 10.9%, with 86% of all instantaneous errors less than 25%.
is measured. Moreover, the model results for the Cumberland Similarly, the median monthly discharge error is 14.7%, with
river, and the sensitivity of the depth estimates to slope vari- 87% of all monthly errors less than 25%. As a final analysis,
ability, indicates a need for future studies to explore the possible we combined both error due to temporal sampling and depth
seasonal dependence of . error, and found that the median error from both error sources
combined was 22%.
E. Estimating Discharge As noted above, the mission will consist of a fast phase (3 day
Example discharge results for the Kanawha river are shown period), and a nominal phase (22 day period). During the fast
in Fig. 9. The true discharge shows a large flood wave propa- phase, spatial coverage is not global; only 2,264 model pixels
gating through the river channel over the course of four days: (39%) were sampled. During the fast phase, however, pixels are
285, 286, 287, 288. On day 285, discharge is increasing from measured at least once every three days, or approximately ten
0–200 km, but is constant from 200–400 km. On day 286, dis- times per month. As noted above, pixels are measured on av-
charge at the upstream boundary has peaked, and discharge is erage 3.75 times per month in the nominal phase, which is far
decreasing along the remainder of the river. On day 287, the less frequently. The monthly discharge errors reflect this; the
flood wave peak is around 375 km, and on day 288, discharge is median NRMSE from Table II is 3% for the fast phase, and
increasing along the entire course of the Kanawha river. SWOT 14.7% for the nominal phase, with a 78 degree inclination angle.
measurements occurred on day 285 and 288, and different parts We also tested whether or not the temporal sampling was sen-
of the river were sampled. On day 285, the low flow condition sitive to the inclination angle of the orbit. The 74 degree incli-
was measured, and on day 288, the increasing discharge pro- nation angle is the minimum required to capture the outlets of
file was measured. Thus, a partial picture of the discharge dy- the major Arctic rivers. The 78 degree inclination angle would

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 29

mates. We can examine this more rigorously by using a first-


order Taylor series expansion to approximate the sensitivity of
discharge to depth, which yields an estimate of (instantaneous)
discharge errors due to depth errors

(15)

Normalizing this expression by discharge leads to a relative sen-


sitivity of discharge to depth error

(16)

The implication of this equation is that any errors in depth are


multiplied by a factor of 1.67; thus, to attain discharge accuracy
of 10%, depth errors must be limited to 6%. This sensitivity is
illustrated as the red line in Fig. 11. However, estimates of dis-
charge anomaly will be much less sensitive than absolute dis-
charge to errors in the initial depth. SWOT will yield highly ac-
curate measurements of water height anomaly and (thus) depth
anomaly . Defining discharge anomaly as the differ-
ence between discharge at time and time 1, we have

(17)

The sensitivity of this expression to error in the initial depth is


given by
Fig. 10. Histograms of instantaneous NRMSE discharge error (top) and
monthly NRMSE discharge error (bottom) are shown.
(18)

TABLE II Normalizing this by and rearranging gives


MONTHLY SAMPLING ERROR FOR BOTH THE NOMINAL PHASE (22-DAY
PERIOD) AND THE FAST PHASE (THREE-DAY PERIOD) OF THE MISSION, AND
FOR TWO INCLINATION ANGLES, ASSUMING PERFECT DISCHARGE ESTIMATES
(19)

Note that we have removed the dependence of the error metric


on slope in order to clearly show the differences between the ex-
pressions for absolute and relative discharge error due to depth:
(16) and (19) are identical except for the term in square brackets
in (19). The term in square brackets is a function only of the ratio
between the change in depth at time and the initial depth; given
this ratio, the error in discharge anomaly due to depth is a linear
function of the error in depth, as shown in Fig. 11. Discharge
anomaly is most sensitive to depth error for less than zero;
this is intuitive, since if the depth at time is much larger than
permit further oceanographic study of the Arctic Ocean circula-
the initial depth, the effect of the initial condition will be mini-
tion, and is the maximum allowable inclination angle, due to
mized. From Fig. 11, discharge anomaly is much less sensitive
other considerations. We calculated monthly discharge errors
to depth error than absolute discharge. For instance, to achieve a
for the nominal phase and for the fast phase for a 74 and a 78
discharge anomaly accuracy of 10%, depth errors must be lim-
degree inclination. During the fast phase, the monthly discharge
ited to 10.7%, for a relative increase of depth of 25% over the
error was identical for both inclination angles. During the nom-
initial time.
inal phase, the median NRMSE was 14.7% and 15.8% for the
78 degree and 74 degree orbits, respectively; the error associ-
ated with the 74 degree orbit was thus 7.4% greater. Thus, in IV. SUMMARY AND CONCLUSIONS
the context of this study, the monthly sampling error was not An algorithm for estimating river depth from SWOT mea-
sensitive to the inclination angle of the orbit. surements was presented and tested using LISFLOOD-FP
The instantaneous discharge results of Fig. 10 (top panel) are model output. The algorithm uses the time series of SWOT
essentially an experiment in examining how the depth errors cal- measurements to obtain an estimate of river depth at an arbitrary
culated from SWOT observables propagate into discharge esti- initial time. River depth at other times can then be estimated

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
30 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 3, NO. 1, MARCH 2010

the Ohio river basin. River depth at the initial simulation time
was successfully estimated for the 5,860 model pixels with a
mean (standard deviation) relative error of 4.1% (11.2%). From
these depth estimates and SWOT observables, discharge was
estimated, assuming that roughness was known. Instantaneous
discharge estimates over the one-year evaluation period had
median NRMSE of 10.9%, and 86% of all instantaneous errors
were less than 25%. As a separate experiment, we sampled the
true discharge time series at the SWOT measurement times, and
used only the sampled estimates to calculate monthly discharge.
The median monthly discharge error is 14.7%, and 87% of all
monthly errors are less than 25%. Combining both error due to
temporal sampling and depth error, the median error was 22%.
From this preliminary analysis, we conclude that the depth
algorithm presented here has potential for use in developing an
estimate of river depth from SWOT measurements. In contrast
to depth estimation approaches based on data assimilation pre-
sented by [15] and [18], no ensemble hydrodynamic simulations
are required, which significantly reduces the computational ex-
pense and may make this approach more feasible for global ap-
plication. A method to interpolate the SWOT estimates of dis-
charge such as an Optimal Interpolation (OI) scheme [28] has
the potential to improve averaged estimates of discharge from
the instantaneous estimates, such as the monthly discharge er-
rors analyzed here. Future work will investigate methods to es-
timate the roughness coefficient and depth simultaneously, will
explore the spatiotemporal characteristics of slope timeseries
variability, and will explore the role of slope and width errors
in discharge estimates.

ACKNOWLEDGMENT
The authors would like to thank the two anonymous reviewers
who provided helpful comments and improved the quality of the
paper.

REFERENCES
[1] D. E. Alsdorf, E. Rodriguez, and D. P. Lettenmaier, “Measuring surface
water from space,” Rev. Geophys., vol. 45, no. 2, pp. 1–24, 2007.
[2] J. E. Costa, K. R. Spicer, R. T. Cheng, F. P. Haeni, N. B. Melcher, E. M.
Thurman, W. J. Plant, and W. C. Keller, “Measuring stream discharge
by non-contact methods: A proof-of-concept experiment,” Geophys.
Res. Lett., vol. 27, no. 4, 2000, DOI:10.1029/1999GL006087.
[3] J. E. Costa, R. T. Cheng, F. P. Haeni, N. Melcher, K. R. Spicer, E.
Hayes, W. Plant, K. Hayes, C. Teague, and D. Barrick, Use of Radars
to Monitor Stream Discharge by Noncontact Methods vol. 42, no.
W07422, 2006, DOI:10.1029/WR004430.
[4] R. Romeiser, H. Breit, M. Eineder, H. Runge, P. Flament, K. de Jong,
and J. Vogelzang, “Current measurements by SAR along-track interfer-
ometry from a space shuttle,” IEEE Trans. Geosci. Remote Sens., vol.
43, no. 10, pp. 2315–2324, Oct. 2005.
Fig. 11. Sensitivity of absolute discharge (dashed line) and discharge anomaly [5] L. C. Smith, B. L. Isacks, A. L. Bloom, and A. B. Murray, “Estima-
are shown in the top panel; the circles, squares, diamonds, and triangles refer tion of discharge from three braided rivers using synthetic aperture
1 0
to zz values of 0.25, 0.25, 0.75, and 1.25, respectively (a). Absolute dis- radar (SAR) satellite imagery: Potential application to ungaged basins,”
charge (b) and discharge anomaly (c) are shown for a pixel near the mainstem Water Resour. Res., vol. 32, pp. 2021–2034, 1996.
Ohio downstream boundary condition, where the circles are discharge estimates [6] L. C. Smith and T. M. Pavelsky, Estimation of River Discharge,
and the line is the true discharge. Propagation Speed, and Hydraulic Geometry From Space: Lena River,
Siberia vol. 44, no. W03427, 2008, DOI:10.1029/WR006133.
[7] D. M. Bjerklie, S. L. Dingman, C. J. Vorosmarty, C. H. Bolster, and R.
G. Congalton, “Evaluating the potential for measuring river discharge
from the time variability of the SWOT height measurements from space,” J. Hydrol., vol. 278, pp. 17–38, 2003.
[8] D. M. Bjerklie, D. Moller, L. Smith, and L. Dingman, “Estimating dis-
and the depth at the initial time. The LISFLOOD-FP model charge in rivers using remotely sensed hydraulic information,” J. Hy-
was integrated for one year over the eleven largest tributaries of drol., vol. 309, pp. 191–209, 2005.

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.
DURAND et al.: ESTIMATING RIVER DEPTH FROM REMOTE SENSING SWATH INTERFEROMETRY MEASUREMENTS 31

[9] G. R. Brakenridge, S. V. Nghiem, E. Anderson, and S. Chien, “Space- [27] E. Anderson, Z. Bai, C. Bischof, S. Blackford, J. Demmel, J. Dongarra,
based measurement of river runoff,” Eos Trans. AGU, vol. 86, no. 19, J. Du Croz, A. Greenbaum, S. Hammarling, A. McKenney, and D.
pp. 185–188, 2005. Sorensen, LAPACK User’s Guide, 3rd ed. Philadelphia, PA: SIAM,
[10] C. M. Birkett, L. A. K. Mertes, T. Dunne, M. H. Costa, and M. J. 1999.
Jasinski, “Surface water dynamics in the Amazon Basin: Applications [28] O. Talagrand, “Bayesian estimation. Optimal interpolation. Statistical
of satellite radar altimetry,” J. Geophys. Res.—Atmospheres, vol. 107, linear estimation,” in Data Assimilation for the Earth System, R. Swin-
no. D20, p. 8059, 2002, DOI:10.1029/2001JD000609. bank, Ed. Dordrecht, The Netherlands: Kluwer, 2003, pp. 21–36.
[11] T. H. Syed, J. S. Famiglietti, J. Chen, M. Rodell, S. I. Seneviratne, P.
Viterbo, and C. R. Wilson, “Total Basin discharge for the Amazon and
Mississippi River Basins from GRACE and a land-atmosphere water
balance,” Geophys. Res. Lett., vol. 32, p. L24404, 2005, DOI:10.1029/
2005GL024851. Michael Durand received the B.S. degree in
[12] T. H. Syed, J. S. Famiglietti, V. Zlotnicki, and M. Rodell, “Contempo- mechanical engineering and biological systems
rary estimates of Pan-Arctic freshwater discharge from GRACE and re- engineering from Virginia Polytechnic Institute,
analysis,” Geophys. Res. Lett., vol. 34, p. L19404, 2007, DOI:10.1029/ Blacksburg, in 2002, and the M.S. and Ph.D. degrees
2007GL031254. in civil engineering from the University of Cali-
[13] T. H. Syed, J. S. Famiglietti, and D. Chambers, “GRACE-based es- fornia, Los Angeles, in 2004 and 2007, respectively.
timates of terrestrial freshwater discharge from basin to continental He is currently a Postdoctoral Researcher with the
scales,” J. Hydrometeorol., vol. 10, no. 1, pp. 22–40, 2009, DOI:10. Byrd Polar Research Center, The Ohio State Univer-
1175/2008JHM993.1. sity, Columbus.
[14] National Research Council, Earth Science and Applications From
Space: National Imperatives for the Next Decade and Beyond, 418
pp., Nat. Acad. Washington, DC, 2007.
[15] K. Andreadis, E. A. Clark, D. P. Lettenmaier, and D. E. Alsdorf,
“Prospects for river discharge and depth estimation through assimi-
Ernesto Rodríguez received the Ph.D. degree in
lation of swath-altimetry into a raster-based hydrodynamics model,”
physics from the Georgia Institute of Technology,
Geophys. Res. Lett., vol. 34, p. L10403, DOI:10.1029/2007GL029721.
Atlanta, in 1984.
[16] X. Liang, D. P. Lettenmaier, E. F. Wood, and S. J. Burges, “A simple
Since 1985, he has been with the Radar Science
hydrologically based model of land-surface water and energy fluxes for
and Engineering Section, Jet Propulsion Laboratory,
general-circulation models,” J. Geophys. Res.—Atmospheres, vol. 99,
California Institute of Technology, Pasadena. His
no. D7, pp. 14415–14428, 1994, DOI:10.1029/94JD00483.
research interests include radar interferometry,
[17] P. Bates and A. P. J. De Roo, “A simple raster-based model for flood
altimetry, sounding, terrain classification, and EM
inundation simulation,” J. Hydrol., vol. 236, pp. 54–77, 2000.
scattering theory.
[18] M. Durand, K. M. Andreadis, D. E. Alsdorf, D. P. Lettenmaier,
D. Moller, and M. Wilson, “Estimation of bathymetric depth and
slope from data assimilation of swath altimetry into a hydrody-
namic model,” Geophys. Res. Lett., vol. 35, p. L20401, 2008,
DOI:10.1029/2008GL034150.
[19] G. LeFavour and D. Alsdorf, “Water slope and discharge in the Amazon Douglas E. Alsdorf received the M.Sc. degree
River estimated using the shuttle radar topography mission digital ele- in geophysics from The Ohio State University,
vation model,” Geophys. Res. Lett., vol. 32, p. L17404, 2005, DOI:10. Columbus, in 1991, and the Ph.D. degree in geo-
1029/2005GL023836. physics from Cornell University, Ithaca, NY, in
[20] M. A. Trigg, M. D. Wilson, P. D. Bates, M. S. Horritt, D. E. Alsdorf, 1996.
B. R. Forsberg, and M. C. Vega, “Amazon flood wave hydraulics,” J. He is currently an Associate Professor with the
Hydrol., 2009. School of Earth Sciences at The Ohio State Univer-
[21] A. C. Benke and C. E. Cushing, Rivers of North America. Burlington, sity and Director of the Climate, Water, and Carbon
MA: Elsevier, 2005, 1168 pp.. Program and the Interim Directory of the Institute
[22] Hydro1K Documentation [Online]. Available: http://edc.usgs.gov/ for Energy and the Environment at The Ohio State
products/elevation/gtopo30/hydro/readme.html University.
[23] National Land Cover Dataset 2001 [Online]. Available: http://www.
epa.gov/mrlc/nlcd-2001.html
[24] C. Homer, J. Dewitz, J. Fry, M. Coan, N. Hossain, C. Larson, N.
Herold, A. McKerrow, J. N. Van Driel, and J. Wickham, “Completion
of the 2001 National Land Cover Database for the conterminous Mark Trigg received the B.Eng. degree in mechan-
United States,” Photogramm. Eng. Remote Sens., vol. 74, no. 4, pp. ical engineering from the University of Surrey, U.K.,
337–341, 2007. in 1991, and the M.Sc. degree in soil and water engi-
[25] T. M. Pavelsky and L. C. Smith, “RivWidth: A software tool for the cal- neering in 1997 from Cranfield University, U.K. He
culation of river widths from remotely sensed imagery,” IEEE Geosci. is currently pursuing the Ph.D. degree in geography
Remote Sens. Lett., vol. 5, no. 1, pp. 70–73, Jan. 2004. with the Hydrology Research Group at the Univer-
[26] D. Moller, E. Rodríguez, and M. Durand, “Temporal decorrelation and sity of Bristol, U.K. The topic of his current research
topographic layover impact on Ka-band swath altimetry for surface is Amazon flood wave hydraulics and floodplain dy-
water hydrology,” Eos Trans. AGU, vol. 89, no. 53, 2008, Fall Meet. namics.
Suppl., Abstract H41B-0877.

Authorized licensed use limited to: The Ohio State University. Downloaded on March 29,2010 at 13:14:41 EDT from IEEE Xplore. Restrictions apply.

You might also like