The middle reach of the Congo remains one of the most difficult places to access, with ongoing conflicts and a lack of infrastructure. This has resulted in the Congo being perhaps the least understood large river hydraulically, particularly compared to the Amazon, Nile, or Mississippi. Globally the Congo River is important; it is the largest river in Africa and the basin contains some of the largest areas of tropical forests and wetlands in the world, which are important to both the global carbon and methane cycles. This study produced the first detailed hydraulic characterization of the middle reach, utilizing mostly remotely sensed data sets. Using Landsat imagery, a 30 m resolution water-mask was created for the middle reach, from which effective river widths and the number of channels and islands were determined. Water surface slopes were determined using ICESat observations for three different periods during the annual flood pulse, and while the overall slope calculated was similar to previous estimates, greater spatial variability was identified. We find that the water surface slope varies markedly in space but relatively little in time and that this appears to contrast with the Amazon where previous studies indicate that time and spatial variations are of equal magnitude. Five key hydraulic constraints were also identified, which play an important role in the overall dynamics of the Congo. Finally, backwater lengths were approximated for four of these constraints, with the results showing that at high water, over a third of the middle reach is affected by backwater effects.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Little is known about the hydraulics of the Congo Basin (Figure 1), despite it being the second largest river in terms of catchment area (3,680,000 km2) and average discharge (41,800 m3 s−1). For many Congolese, the Congo is the major transportation route and lifeline, especially the middle reach, which is fully navigable by boat. Of all the major rivers in the world, the Congo River has been studied least, especially when compared to the Amazon. This is particularly surprising as the Congo Basin has the second largest area of tropical forest with over 2 million km2 [Laporte et al., 1998]. The majority of this globally important forest is found in the central basin of the Congo, which is either drained directly by the middle reach of the Congo River or its many tributaries. The Congo Basin also contains a large percentage of the world's wetlands with over 185,000 km2 in the Central Basin alone, a number which includes the internationally important Tumba-Ngiri-Maindombe wetland, the area of which is 65,695 km2. The role of wetlands has been recognized for a long time [Naiman et al., 1998; Segers, 1998] in many diverse areas such as global climate systems; biodiversity conservations; water supply; quality regulation; and food supply. Tropical and subtropical wetlands, in general, are identified as an important source of methane, with Matthews  estimating that these areas contribute 50–75% of the annual methane emissions from natural wetlands.
 Very few research studies have been undertaken on the Congo, relative to other large river basins, with most carried out in the mid 20th century and little since. This is mainly due to the conflict that has plagued the region, especially the Democratic Republic of Congo (DRC), which is the largest country in sub-Saharan Africa and contains the majority of the middle reach of the Congo River, with the remainder of the middle reach acting as the border between the Democratic Republic of Congo and the Republic of Congo. Prior to and since independence from Belgium, the DRC has suffered from long periods of violence, especially the past 20 years with the first Congo War (November 1996–May 1997), the second Congo War (August 1998–June 2003), and ongoing conflicts with rebels.
 Of the published work available on the Congo River, very few studies look specifically at its hydraulic characteristics. Where hydraulic characteristics are discussed, it is in support of related research such as fisheries [Balon and Stewart, 1983; Roberts and Stewart, 1976], the movement of animals [Colyn et al., 1991], or how the Congo Basin was formed [Crosby et al., 2010; Deffontaines and Chorowicz, 1991; Lezzar et al., 1996]. Despite these studies not specifically researching hydraulic characteristics, they do give some valuable insights into the workings of the river and also emphasize the need for hydraulic information in support of many other fields of research. Deffontaines and Chorowicz  suggested that the river network may be affected by local patterns that are generally thought to be due to the structure of the bedrock. The few papers that look at the Congo Basin in specifically hydrological or hydraulic terms can be divided into two groups: (1) papers that consider the modeling of either the hydrology [Beighley et al., 2011] or the floodplain hydraulics [Jung et al., 2010; Lee et al., 2011] and (2) papers that consider the hydrology [Laraque et al., 2001] or hydraulics [Marlier, 1973; Runge, 2007] more generally for the entire basin and no particular region in any great detail. Neither Beighley et al.  nor Lee et al.  looked at the river widths in any detail, but instead assumed a relationship between channel width and upstream drainage area based on data from 38 locations such that in their model widths increased monotonically downstream. Beighley et al.  did note that this method would likely underestimate the effective channel widths and that a more detailed study is needed.
 The use of remote sensing now allows the Congo Basin to be studied in greater detail. Remote sensing has proved valuable in expanding our understanding of other large basins such as the Amazon where, for example, Landsat imagery has been used to study the hydraulic characteristics [Mertes et al., 1996; Trigg et al., 2012]; altimetry data from remote sensing platforms [Birkett et al., 2002; Hall et al., 2012] has been used to look at water surface slopes and/or the correction of datum values for water level gauges; and Synthetic Aperture Radar (SAR) [Alsdorf, 2003] has been used to measure water storage in the floodplain. One of the main advantages of the remote sensing that we rely on here is its ability to provide good spatial data for large areas, which is not often possible using field studies. This is especially true for the Congo Basin, where conflict and the lack of modern infrastructure make field studies logistically difficult, costly, and a perceived risk.
 In this paper, we present the first detailed study of the river hydraulics of the 1600 km middle reach of the Congo River using remotely sensed data. Given no preexisting information, a basic hydraulic characterization for any river might consist of spatially rich data on river width, depth, planform, and the time and space variation of water surface slope, velocity, and discharge. Whilst using only remotely sensed information restricts the set of basic variables that can be measured to width, planform, and water slope, taken together, these do allow key hydraulic variables, such as backwater lengths, to be calculated. The four main objectives of this paper are therefore to: (1) create a detailed water-mask for the entire middle reach of the Congo River using Landsat imagery; (2) analyze how the width of the Congo River changes along its length, based on the detailed water-mask; (3) quantify the number of islands, their location and size along the middle reach; and (4) produce a detailed surface water slope profile of the middle reach of the Congo with the use of ICESat data.
 These objectives lead to the overall aim of this paper, which is to quantify important hydraulic characteristics of the middle reach of the Congo River. This will improve our understanding of this complex and unique river system, and will provide a basic quantification of characteristics not currently available, thus allowing improved hydrological/hydraulic modeling of this region.
2.1. Study Area
 The source of the Congo River lies in the southeast of the Democratic Republic of Congo (DRC) at an altitude between 1400 and 1500 m and consists of a number of small streams, swamps, and lakes [Runge, 2007]. The course of the Congo River is often divided into three reaches, based on the locations of the two largest rapids: the Boyoma Falls and the Livingstone Falls. These rapids separate the river into: (i) the upper reach known as the Lialaba River, which flows from the source to the Boyoma Falls, immediately upstream of Kisangani; (ii) the middle reach from the Boyoma Falls to the Livingstone Falls, just downstream of Kinshasa; (iii) and the lower reach from Kinshasa to the Atlantic Ocean (Figure 1).
 The Congo River drops 1500 m in elevation along its 4000 km course from its source to the sea. The majority of this drop occurs along the upper and lower reaches, as unlike the rest of the Congo River, the middle reach flows in a large arc and is uninterrupted by major falls or rapids. Runge  described the middle reach as “dominated by gentle, broad and anastomosing sand-silt stretches,” with Marlier  suggesting that these stretches were due to the erosive power of the upper reaches of the river that flow into the middle reach and also due to the “slow currents and flat morphology of the basin.” The slope of the middle reach of the Congo has been estimated in a number of studies. Roberts  calculated the slope of the Congo from source to mouth and determined that the slope between Kisangani and Kinshasa was 6.6 cm km−1. However, there is some variability in the published slope estimates, with Roberts and Stewart  suggesting that the slope was never greater than 8cm km−1, and Marlier  suggesting that the slope was always greater than 5 cm km−1.
 Along its course, the middle reach meets a number of large tributaries that drain: (i) the highlands, with the Ubangai, Mongala, and the Sangha rivers joining on the northern banks and (ii) the middle basin with the Kasai, Lulonga, and Lomami rivers joining on the southern banks. Some of these tributaries could be classified as large rivers in their own right, especially the Ubangi and the Kasai whose catchment areas equal 777,000 km2 and 900,000 km2, respectively. The annual variability of the Congo discharge/level is low. Runge  noted an annual variability of the discharge at Kinshasa equal to a factor of 2.8. This is very similar to the annual variability of discharge of the Mississippi at Vicksburg of 3.1. Bultot and Dupriez  also commented that the discharge of the Congo remained “a relatively constant flow around the year.” Roberts and Stewart  noted that the difference between high and low water marks was seldom more than 3 m along the entire Congo River. Marlier  proposed two reasons for the small fluctuation in water level: (1) lakes and swamps on the upper course damp the effects of heavy rain and (2) the very flat topography which will lead to significant wave attenuation. Low water level fluctuations are also likely to be due in part to the network hydrology of the basin, with contributing tributaries discharging their flood waves into the main stem at different times through the year such that these do not fully superimpose. Little is also known about the bathymetry of the middle reach, except that it is mostly a very shallow river with maximum depths of approximately 15 m and minimum depths at low water often less than 2.5 m [Marlier, 1973]. The exception to this is the section of river known as “The Channel”: the narrow, 210 km long passage upstream of Malebo Pool and Kinshasa. Marlier  suggested that at the passage the river narrows to between 800 and 1000 m from a width greater than 13,000 m upstream and the depth varies between 23 and 30 m.
2.2. Gauge Data
 Stage data were obtained for the three stations on the middle reach, Kinshasa, Mbandaka, and Kisangani, from the International Commission for Congo-Ubangui-Sangha Basin. The periods of records vary for each station: Kinshasa (1902–2010); Mbandaka (1913–1984); and Kisangani (1967–2010). Using these long-term data sets, stage duration curves and hydrographs were produced. High flow periods are classified when the discharge was greater than or equal to the 25% exceedance rate based on the flow duration curves (corresponding to the 75th percentile flow/level). Figure 2 shows the long-term daily mean hydrographs for the three gauge locations with the standard deviation and the high flow threshold. From this analysis, the high flow periods for the gauging stations were determined as: Kisangani, 16 November to 29 December; Mbandaka, 2 November to 11 January; and Kinshasa, 30 October to 20 January.
 The Landsat data used were from the Landsat-7 Enhanced Thematic Mapper Plus (ETM+) and were downloaded from the Earth Explorer website (available at earthexplorer.usgs.gov). Reflective bands 2 (0.53–0.61 µm) and 4 (0.78–0.90 µm) were used. Both of these bands have a spatial resolution of 30 m. Using the ENVI software package, the images were calibrated for atmospheric correction using internal average relative reflectance (IARR), and the Normalised Difference Vegetation Index (NDVI) was applied to bands 2 and 4 to identify water (NDVI < 0). Each Landsat image is 185 km × 170 km, defined in a Worldwide Reference System of path (ground track parallel) and row (latitude parallel) coordinates [Arvidson et al., 2001]. To reduce the possibility that the smaller channels would be dry in the imagery, which would reduce the connectivity of the river system, we used only scenes from high flow periods as determined by the gauge data analysis (Figure 1 and Table 1), Landsat scenes were also chosen with minimal cloud coverage and were collected prior to the Landsat ETM+ scan line corrector failure. As it was sometimes difficult to obtain cloud free images for all locations, multiple scenes were used, when required, to compensate for cloud cover. In this study, secondary images were used twice to compensate for clouds covering approximately 60 km of the main channel or 4% of the total length. These secondary images were only used for the area of each image where cloud cover was an issue. These secondary images were also chosen so that they too fell within the high flow period. This meant that they were often for different years; however, it was checked that all images used were not from anomalous years, and where cloud free portions of images overlapped, visual inspection showed that there were no differences in the river width between the primary and secondary images in the overlap.
Secondary image used to correct for area of cloud cover in primary image.
 ICESat Geoscience Laser Altimeter System (GLAS) was the first Earth orbiting laser altimeter and started its mission in 2003 and ended in 2009, when the third laser reached the end of its lifespan. ICESat GLAS had a surface footprint of ∼65 m and made observations every 172 m along its track [Schutz et al., 2005]. This small footprint and along track observation distance makes it more attractive for measuring river water levels than radar altimeters, such as TOPEX/Poseidon (T/P), as smaller rivers can be observed and contamination due to bank vegetation is reduced. For example, T/P has a footprint of ∼1 km and along track observation distance of approximately 596 m [Leon et al., 2006]. The ICESat GLAS product used was GLA14 Land Elevation Product, Release 33 and was downloaded from the Reverb website (available at reverb.echo.nasa.gov). The data were extracted using the Interactive Data Language (IDL) code provided by the National Snow and Ice Data Centre (NSIDC). This code also converted the extracted data to the World Geodetic System of 1984 (WGS84) ellipsoid. Suitable observations were selected by use of the elevation-use flag and the saturation index was used to remove/correct saturated observations. The same criteria as used by Hall et al.  were implemented, in which observations with a saturation index of zero or one were used without correction, observations with an index of two were corrected for saturation and observations above two were excluded. The selected observations were then converted to the vertical datum Earth Gravitational Model of 1996 (EGM96) using the geoid conversion tool, F477, available from the National Geospatial-Intelligence Agency (NGA).
2.5. NDVI to River Width
 After NDVI was performed on the Landsat images, they were mosaicked into a single raster image. Using GIS software, this image was then converted into two different binary images, each using different threshold values to separate wet and dry pixels. A range of NDVI thresholds were tested and the two thresholds that produced the best results were selected. These thresholds were chosen so as to ensure connectivity of the smaller channels and reduce contamination of the water-mask by clouds or dry pixels. The first threshold, equal to −0.25, was chosen to remove the boundary effects and the second slightly higher threshold, equal to −0.2, was selected to discriminate between wet and dry pixels. To disconnect floodplain areas that are not part of the main channel, wet pixels not connected to the main channel were removed. By comparing these two binary images, a single binary image was created. The outer edges of the river were then determined by the outermost wet pixels and a river centerline was created automatically that was equidistant between the river edges. From here onwards, we refer to the main channel as the full river envelope that includes all channels between the outer edges. Main channel river cross sections were then created from this binary water-mask with HEC-GeoRAS [Ackerman, 2009] following the same process used by Trigg et al.  to derive floodplain widths. The use of HEC-GeoRAS provides a semiautomated method of creating cross sections similar to other semi and fully automated methods, and allows the user to apply quality assurance to correct any overlapping cross sections caused by sharp angle changes in the automatically created river centerline vectors. This problem of overlapping cross sections affects other automated methods, such as RivWidth [Pavelsky and Smith, 2008]. In total, 6585 cross sections were created with an interval of 250 m covering 1646.5 km of the middle reach (Figure 3). The 250 m longitudinal distance was chosen to avoid repetitive calculations associated with smaller interval sizes and to accurately measure rapid changes in the river width, which larger intervals may miss. For each cross section, we calculated: (1) the effective river width; (2) the maximum width of a single channel; (3) the number of channels and the number of islands; (4) the total width of the cross section between the outer banks; and finally (5) the error associated with the effective width. This error was based on pixel size and the number of channels per cross section. The errors are due to issues in determining the boundary between wet and dry pixels. Hence, the error per cross section is equal to twice the pixel size multiplied by the number of channels. Using the binary water-mask, the number and sizes of the islands were determined, along with the location of the centroid of each island along the chainage of the river. The term chainage refers to the distance along the river centerline (in kilometers) from the furthest downstream cross section, located just downstream of Kinshasa.
2.6. ICESat to Water Surface Slope
 Once the ICESat GLA14 observations were processed, suitable observations within the river area were extracted using the binary water-mask created from Landsat images. In total, 4449 individual observations were extracted. However, observations that were within a 65 m buffer zone of land were excluded to avoid signal contamination from land or vegetation. This 65 m buffer zone takes into account the 30 m boundary error in the binary water-mask and the 35 m footprint radius of the ICESat observation. After the buffer zone was implemented, 4356 observations remained for 280 different time/location stamps (Figure 1). The observations were averaged for each of the 280 time/location stamps and their midpoint location between banks was determined. The 280 time/location stamps were then classified further into three distinct flow periods, based on timing of observations, which centered on the following months: March, June, and November. These three periods correspond to the falling limb of the flood wave, the low flow period, and the rising limb of the flood wave, respectively. When this classification was applied, the number of averaged observations for each period were: March (114), June (59), and November (107). For each of these classification periods, the water surface slope and ICESat measurement error were calculated. For the measurement error calculation, a number of assumptions were made. First, the measurement error for each ICESat observation was 11.5 cm, based on research by Urban et al.  for inland rivers under partly cloudy conditions. Second, the errors are normally distributed and that for each intersection the measurement error is inversely proportional to the square root of the number of observations.
 The observations were then sorted by date and location and if the observations were within 2 days of each other and the second observation was downstream of the first, the slope was calculated for this pairing. This was done for all periods. As there were overlapping slopes for some locations, the entire middle reach of the Congo was divided into interval bins, each 25 km long. The bin size was chosen to be larger than the minimum distance between pairings (∼22 km) to avoid the introduction of artifacts into the data. For each interval bin and classification period, the mean slope and standard deviation were calculated.
3.1. River Width
 The width of the middle reach of the Congo varies greatly along its course (Figure 4a). At its narrowest, around Kisangani just below the Boyoma Falls (formerly known as Stanley Falls) its effective width is only 512 m and width slowly increases downstream to chainage 1000 (1000 km upstream of Kinshasa), where it narrows slightly for 200 km. Approximately from chainage 800 to chainage 250 km, the middle reach undergoes a general rapid expansion in its width, with its effective width increasing from approximately 4000 m to nearly 10,000 m. During this rapid expansion, three large visible short-length contractions in width are encountered, at chainages of: 315, 485, and 555 km, and with approximate lengths of 7.5, 5, and 6.5 km, respectively. A longer constriction in the effective width occurs between 250 and 50 km upstream of Kinshasa, when the Congo narrows to an average of 906 m as it flows through “The Channel.” The Congo then widens just upstream of Kinshasa as it enters Malebo Pool and narrows again at the crossing between Kinshasa and Brazzaville and before the Livingstone Falls. The width of the largest single channel stays fairly constant along the majority of the middle reach, with the exception of the areas around the above-mentioned contractions, where the width decreases. The number of river channels is generally greater than 1 (Figure 4b); however, there are a number of locations where the number of channels reduces greatly: at the contractions; as it flows through “The Channel”; and after Malebo pool. The largest number of channels occurs at Malebo Pool. The ratio of the effective river widths to the total cross-sectional widths indicates the size and number of islands at each cross section (Figure 4c), where a ratio equal to 1 indicates that there is only water between the outer banks and a ratio less than 0.5 indicates that there is more land than water at those cross sections. There is only a small percentage (20%) of the middle reach where the ratio is equal to 1, these areas occur upstream at Kisangani and as the river flows through “The Channel” upstream of Kinshasa. From a chainage of 1600 km to approximately 900 km, there is an overall decrease in the ratio of effective width to total cross-sectional width, with the smallest ratio of 0.175 occurring at chainage 910 km. Between this chainage and approximately 250 km, there is an overall increase in the ratio between effective river and overall cross-sectional widths. Despite these overall trends, there are large variations in the ratio over smaller spatial scales, with the largest variations occurring at Malebo Pool, between chainages 45 and 11 km, with the ratio decreasing from 1 to 0.34 before returning to 1 at Kinshasa.
3.2. Island Analysis
 The distribution and sizes of islands are calculated for 10 km intervals along the reach based on the location of the centroid of each island (Figure 5). If an island crosses more than one interval, its attributes are assigned to the interval its centroid falls within. The number of islands per interval remains fairly constant along the entire middle reach, with the number seldom greater than 50 per 10 km. However, there are two intervals where the number of islands is greater than 50: between chainages 300 and 480 km and 10 and 40 km. There are also instances where the number of islands per interval is visibly lower than the surrounding intervals. In the downstream direction, the first decrease in the number of islands occurs roughly at chainage 560 km, the second occurs at chainage 500 km, the third occurs at chainage 320 km, and the fourth occurs at chainage 10 km. There is also a section of river where the number of islands per interval remains zero or close to zero, this section is known as “The Channel,” roughly between chainages 50 and 250 km. The size of the islands follows a pattern similar to that of the number of islands. However, the total area of islands is not directly correlated to the number of islands. Between chainages 10 and 40 km (Malebo Pool), where the largest number of islands were encountered, the corresponding total area was not the greatest. Upstream from this area, between chainages 940 and 1260 km, there are four large peaks in the total area of islands per 10 km, with the total area in each case being larger than 200 km2. These are the only cases where the total area was greater than 200 km2, and island area for the majority of 10 km intervals does not exceed 100 km2.
3.3. Water Surface Slope
 Water surface slopes with standard deviations and measurement error were calculated for the entire middle reach in 25 km lengths for each of the three ICESat classification periods (Figure 6 and Table 2). The March observation period has the lowest minimum and maximum slope values and the smallest standard deviations. Whilst the November observation period has the highest minimum and maximum slope values and the largest standard deviation. However, the November period has the shallowest average slope of all three periods. It was found using t tests that the slopes for all three flow periods were not statistically different (p < 0.05). The slopes also follow a similar pattern downstream from Kisangani to Kinshasa. Downstream from Kisangani to approximately chainage 900 km, the slope of the middle reach of the Congo remains fairly constant through the annual flood pulse with the slope always greater than 6 cm km−1 and with a standard deviation of 3.5 × 10−6 cm km−1. Between chainages 650 and 900 km, the slope of the middle reach decreases to just under 6 cm km−1 and from chainages 650 to 300 km, decreases again in stages from 6 to approximately 3cm km−1. The slope increases slightly between chainages 250 and 40 km to between 3 and 3.5 cm km−1. Downstream of chainage 40 km the middle reach slope again increases up to between 5.5 and 9.6 cm k−1m depending on observation period, with the greatest slopes occurring in the November and June observation periods.
Table 2. ICESat Slope Summaries for the Three Observation Periods
Mean (cm km−1)
Maximum (cm km−1)
Minimum (cm km−1)
Standard Deviation (cm km−1)
Measurement Error ± (cm/km)
 From our results, it is clear that the empirical method of estimating the effective river width used by Beighley et al.  and Lee et al.  does not account for the large variability that exists in the middle reach of the Congo River. Using the first-order method implemented in these papers, where the effective channel width at any point is related its upstream drainage area, results in the effective channel width increasing monotonically from 1440 m at Kisangani to 3264 m at Kinshasa. These values are comparable to the observed effective widths determined in this study for Kisangani (1014 m) and Kinshasa (3000 m); however, this may be because these two locations were used to calibrate the empirical method and therefore it may not be a surprise that the method works well at these locations. However, the empirical method produces large errors in effective widths between these locations as anticipated by Beighley et al. . These errors are particularly large between chainages 250 and 800, where the effective main channel width increases to beyond 6000 m for 550 km. While the method of estimating effective river width based on upstream drainage area may be acceptable in large-scale hydrological models, for hydraulic models to be capable of accurately predicting inundation patterns, the storage in both the river and floodplain and the timings of the flood wave and occurrence of inundation need to be spatially accurate. When the effective width is underestimated, this leads to an underestimation of storage in the channel, which results in either incorrect wave travel time and inundation onset or overestimation of floodplain storage, or both.
 From the results, five key hydraulic constraints to the flow were identified (Figures 7a–7e). These constraints were identified when all three of the following measures were true: (1) the effective width and maximum single channel width are equal; i.e., there are no islands; (2) the equivalence between effective width and maximum single channel width exists only for a short distance, chosen here as less than 10 km; and (3) the ratio of change in effective width and minimum effective width over 10 km was greater than 1. The method of estimating the effective width based on drainage area fails to identify these important locations. These five key constraints act as hydraulic controls or chokes on the middle reach of the Congo. A hydraulic constraint/control is a location where the river is forced to flow through a reduced width and/or depth relative to the channels upstream width/depth causing the upstream water elevation to increase, commonly also causing a backwater effect for some distance upstream as the choke point acts as a base level control. The effect that hydraulic controls have on upstream depths is more pronounced during periods of high flow, as these structures can only convey a certain amount of water at any given moment. As a result, these are important structures to include in a hydraulic model, especially when looking at main channel/floodplain interactions. To provide a first estimate of the magnitude of this effect, the backwater lengths for constraints B, C, D, and E were determined using equation (1) proposed by Samuels :
where L is the backwater length, D is the depth at the constriction, and So is the bed slope. Two assumptions were used: (1) the bed slope is equal to the water surface slope and (2) as no bathymetry data exists for this reach of the Congo, backwater lengths were calculated for a low water depth of 2.5 m and high water depth of 15 m as suggested by Marlier .
 From downstream to upstream, the first hydraulic constraint (Con. A) occurs just downstream of Malebo Pool at chainage ∼6 km (Figure 7a). Here, the effective width decreases from over 13,000 m to less than 2000 m over a 20 km distance and at one point decreases from 8500 to 3000 m width in less than 3 km. At this constraint, the number of islands reduces from over 25 down to 0. The effect this constraint has on the water surface slope cannot be determined as there were no water level observations available from ICESat downstream of this constraint. This constraint occurs between the cities of Kinshasa (DRC) on the southern bank of the Congo River and Brazzaville in the Republic of Congo (RC) on the northern bank. From Figure 7a and SRTM data, we believe that the constraint is due to the river cutting through the mountain chain known as the Atlantic Rise.
 The second constraint, Con. B (Figures 6 and 7b), occurs at chainage 260 km and again is due to the river cutting through an upland area, in this case, the Bateke plateau. The magnitude of the decrease in the effective river width is not as dramatic as with the first constraint, with the river contracting from 7000 m to less than 3000 m in 4 km distance. This constraint occurs at the upstream end of “The Channel,” where the river is forced through a narrow gorge. Looking at the water surface slope around the second constraint, there is a visible increase in slope downstream that is especially noticeable in the March and June observation periods. As these periods correspond to the falling limb of the flood wave and the low flow period, it is not surprising that the constraint would have more impact than during the peak or rising limb (November). The backwater lengths estimated for constriction 2 range from 51.5 km at low water to a maximum of 350 km at high water.
 The third constraint, Con. C (Figures 6 and 7c), occurs at roughly chainage 315 km and again seems to be due to the river cutting through an outcrop of high land. However, this outcrop is only visible in the SRTM DEM on the southern bank of the river. At this constraint, the effective width of the river narrows from over 8000 to 4900 m and expands rapidly to over 13,000 m in less than 15 km distance. The effect of this constraint on the water surface slope is a steepening in slope downstream of the constraint/expansion. At low water, the backwater length caused by this constriction is approximately 58 km and at high water 357 km.
 The fourth constraint, Con. D (Figures 6 and 7d), occurs at chainage 485 km. This constraint is similar to the third constraint in that the river width contracts and expands rapidly. The SRTM shows a small terrain outcrop on the southern bank; however, no conclusions could be drawn on the formation of this constraint. Here, the river narrows from an effective width of approximately 9000 to 2600 m and then expands to over 10,000 m in only 16 km distance with the constraint being nearly 5 km long itself. The effect of the fourth constraint on the water surface slope is discussed below along with the effect of the fifth constraint. Backwater lengths are approximately 44 km at low water and 300 km at high water.
 The fifth constraint, Con. E (Figures 6 and 7e), occurs just downstream of the confluence of the Ubangi and Congo Rivers at a chainage of approximately 555 km. From the SRTM, there is a terrain outcrop on the northern bank of the constraint but there is nothing visible on the southern bank. At this constraint, the river goes from a wide braided river to a narrow channel and back to a wide braided river, all within 10−20 km. Ignoring the Ubangi River just upstream of the constraint, the effective main channel width is over 8000 m and narrows to under 4000 m for over 5 km before expanding to over 12,000 m downstream. Looking at the effect of the fourth and fifth constraints on the water surface slope, there is a clearly visible pattern for each of the three observation periods. These two constraints, after Malebo Pool, have the biggest impact on the water surface slope. Between these two constraints, there is a pronounced shallowing of slope on the middle reach of the Congo, with the slope reducing from 5 to 3.65 cm km−1 for the June observations, from 4.9 to 3.2 cm km−1 for the November observations, and from 5.4 to 3.4 cm km−1 for the March observations.
 At approximately chainage 600 km, there is an increase in water surface slope. This increase is largest for the March and June observations and lowest for the November observations. This increase is not due to any visible constriction in the effective width and we believe it is due to a backwater effect from the confluence of the Ubangi and Congo Rivers downstream. For constriction E, the approximate backwater lengths were 35 km for low water and 240 km at high water.
 This work highlights the spatial complexity of the water surface slope, which previously had not been studied for the Congo at this level of detail. The water surface slope was determined for the three different periods, rising and falling limbs of the flood wave and a low flow period. It was found that the slope remained approximately constant in time between the three periods with the average slope for the falling limb equal to 5.52 cm km−1, low water equal to 5.56 cm km−1, and rising limb equal to 5.41 cm km−1; however, all these values are marginally lower than the previous average slope estimate of 6.6 cm km−1 by Roberts . The maximum slope calculated was for the rising limb (9.57 cm km−1), and this is greater than the 8 cm km−1 suggested by Roberts and Stewart  as the maximum for the middle reach. The minimum slope calculated for all three periods of 2.66 cm km−1 was also shallower than the minimum slope of 5 cm km−1 suggested by Marlier . This study found that the minimum slopes occurred before the Congo enters “The Channel” and were less than 3 cm km−1 for all three periods studied. Moreover, there was at least 500 km of the middle reach where the slope was less than the Marlier  minimum suggested slope. These differences in calculated slopes are mainly due to both greater detail and accuracy in water elevation data available to this study, compared to previous investigations. When the surface water slope is compared to the Amazon River, it is clear that both rivers behave very differently. The water surface slope for the Congo varies markedly in space and these variations persist throughout the flood hydrograph, whilst for the Amazon, previous studies by Dunne et al. , Mertes et al. , and Trigg et al.  indicate that the variation in the temporal and spatial water surface slopes are of a similar scale. Dunne et al.  and Mertes et al.  found that for the middle reach of the Amazon (between 700 and 2000 km downstream from Iquitos) the range in water surface slope was approximately 1.5 cm km−1, while Trigg et al.  found temporal variations in slope of similar magnitudes. The variations in the Amazon's temporal and spatial water surface slopes were also identified by Birkett et al. , whose water surface slopes were derived from the satellite radar altimetry. As we identify here for the Congo, Birkett et al.  also noted that for the Amazon changes in water surface slopes occurred at confluences and geomorphic constraints such as tectonic arches in the craton underlying the basin.
 It was clear that for the middle reach of the Congo, the number and size of islands along the reach could also be used to identify constrictions in the channel, with a large decrease in both size and number of islands at each of the constraints. These observations could be used to infer an increase in the sediment transport capacity at the constraints, which would also be consistent with a likely increase in flow velocity at the constrictions.
 As mentioned above, there were a number of locations where the total area of islands per 10 km interval was greatly larger than that for the surrounding intervals. Each of these was due to a large island spanning multiple intervals but only being counted in the interval in which its centroid lay. For example, the largest total island area per 10 km occurred between chainages 930 and 940 km, with the total area equal to 542 km2. One island, “Île Sumba,” accounts for over 500 km2 of the total and spans roughly between chainages 885and 975 km. Around this chainage, there is a corresponding decrease in slope of the river across all three time periods. This suggests that Île Sumba is a result of sedimentary deposition due to the change in water surface slope that occurs in the area. It is also important to quantify the number of islands, as the number of islands is closely related to the number of river channels. This will have implications for the hydrodynamics of the Congo, as an increase in the number of islands will result in an increase in the wetted perimeter, and the friction of the channel, which is directly proportional to the wetted perimeter.
 The Congo River remains a relatively under-studied river basin, despite its size and importance at both African and global scales. In this paper, we presented the first detailed hydraulic characterization of the middle reach of the Congo, utilizing mostly remotely sensed data sets. Using Landsat imagery, a high water binary water-mask for the middle reach was created at 30 m resolution. Using this binary water-mask, river widths were determined for every 250 m and an analysis of the number and sizes of islands in the middle reach was performed. The effective width of the Congo varies greatly along its middle reach, with a mean width of 3911 m and maximum and minimum widths of 13,435 and 513 m, respectively. These results show that the methods to parameterize river width previously used in basin studies [Beighley et al., 2011; Lee et al., 2011] are, as expected, unable to represent the large variations in the effective river width and are also unable to identify hydraulic constraints which we show are likely to greatly influence flow processes. Better representation of width and slope will be important for future modeling of the hydraulics of the river system. For example, if river width is underestimated, the volume of storage in the channel is consequently underestimated, and this results in both an overestimation of the storage of water on the floodplain and incorrect prediction of the timing and location of floodplain inundation. In turn, this will result in a poor representation of flood wave travel time and dynamics through the reach. To ensure that the volume in the channel is correct, the depth would need to be increased and this would cause errors in the timing of inundation and possible errors in the bed slope, which would affect channel-floodplain interactions. For global-scale hydrological modeling, it is also crucial to accurately represent the river width [Decharme et al., 2012; Yamazaki et al., 2011]. The method of determining river widths used in this study could be used in global-scale modeling and would have greater detail than current automated methods. However, when compared to automated methods, it is time intensive to apply. The analysis presented here provides key hydraulic data and understanding for the construction of a first large-scale hydrodynamic model of the Congo.
 With the use of ICESat altimetry data, the water surface slope was studied. The ICESat data were grouped into three periods that corresponded to the falling limb (March), low water (June), and the rising limb (November), with slopes remaining approximately constant across the three periods. This study produced a more detailed water surface slope than previous studies and it was discovered that the average slope of the middle reach has been previously overestimated. We also uncovered much greater spatial variability in water surface slopes than has previously been identified. These findings highlight that the Congo and the Amazon rivers behave very differently, as the slope of the Amazon shows similar variability in time and space whilst more variability in space rather than time is observed for the Congo. These findings will hopefully lead to better understanding of the surface water dynamics of the Congo and better hydraulic modeling.
 Finally, by studying the water-mask and the ICESat derived slopes, five key hydraulic constraints were identified and their effects on the water surface slope were estimated. These five constraints have wide implications for the hydraulics of the middle reach of the Congo, as they are base level controls and affect both the upstream water surface levels and downstream flows. Overbank flow due to this backwater effect could be a major source of water in the floodplain, as the connectivity of the main channel and the floodplain seem to be very different from the Amazon, whose floodplain and main channel are highly connected via visible floodplain channels as well as through high water diffuse overbank flows. These hydraulic constraints may also be the cause of the marked variation in Congo River water surface slope in space but not in time compared to the Amazon.
 Approximate backwater lengths were calculated using an equation proposed by Samuels . This indicated that, at low water, over 11% of the total reach is affected by backwater effects. During high water, the backwater effect is far greater, with over 33% of the total length affected, including the entire length of the middle Congo between the second constraint, at chainage 260 km, and fifth constraint, at chainage 555 km.
 It should be emphasized that this work has only looked at the channels that make up the middle reach of the Congo River. The bathymetry of the channels, the width of the floodplain and its effect on the hydraulics of the Congo were not investigated during this study. Further work is needed to assess the importance of the floodplain on the hydraulic characteristics of the middle reach and vice versa, whose behavior is likely to be very different to that of the Amazon.
 The work in this paper was supported by a research grant from the Leverhulme Trust. The work by G. Schumann was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The authors would like to thank Ir. Georges Gulemvuga, Director of Water Resources for the International Commission for the Congo-Ubangui-Sangha Basin, for providing the stage level data used in this study. Mark Trigg is funded by the Willis Research Network. Finally, the authors would like to thank Doug Alsdorf and other anonymous reviewers for their constructive comments.