Sediment Balance Estimation of the ‘Cuvette Centrale’ of the Congo River Basin Using the SWAT Hydrological Model

: In this study, the SWAT hydrological model was used to estimate the sediment yields in the principal drainage basins of the Congo River Basin. The model was run for the 2000–2012 period and calibrated using measured values obtained at the basins principal gauging station that controls 98% of the basin area. Sediment yield rates of 4.01, 5.91, 7.88 and 8.68 t km − 2 yr − 1 were estimated for the areas upstream of the Ubangi at Bangui, Sangha at Ouesso, Lualaba at Kisangani, and Kasai at Kuto-Moke, respectively—the ﬁrst three of which supply the Cuvette Centrale. The loads contributed into the Cuvette Centrale by eight tributaries were estimated to be worth 0.04, 0.07, 0.09, 0.18, 0.94, 1.50, 1.60, and 26.98 × 10 6 t yr − 1 from the Likouala Mossaka at Makoua, Likouala aux Herbes at Botouali, Kouyou at Linnegue, Alima at Tchikapika, Sangha at Ouesso, Ubangi at Mongoumba, Ruki at Bokuma and Congo at Mbandaka, respectively. The upper Congo supplies up to 85% of the ﬂuxes in the Cuvette Centrale, with the Ubangi and the Ruki contributing approximately 5% each. The Cuvette Centrale acts like a big sink trapping up to 23 megatons of sediment produced upstream (75%) annually.


Introduction
The most important agent transporting sediment across the land is river transport, while relief and runoff are the most important factors determining the sediment load of rivers [1] (Hay, 1998). The suspended sediment load transported by a river or stream is a mixture of sediment derived from different areas in a catchment that comprise source types with different erosional processes depending on the time and place [2,3] (Walling, 2006;Haddadchi et al., 2013). Sediment fluxes and their variability are important and merit study because drastic changes in their amounts and patterns usually indicate additional human impacts on the basin [4,5] (NKounkou & Probst, 1987;Walling, 2009).
Wetlands are complex ecosystems that act as an interface between terrestrial and aquatic ecosystems. Wetlands receive water and sediments from external sources [6] (Mitsch & Gooselink, 2015). The hydrodynamic characteristics of wetland ecosystems, including the water inputs, outputs, water flow, and seasonality, can influence organic deposition and material fluxes [7] (Gosselink & Turner, 1978). [8] Novitzki (1979) characterized wetlands with regard to water source and landform as surface or groundwater depressions or slopes, which are defined by whether or not they receive a greater amount  [26] (after Laraque et al., 2020), and the wetlands of the Cuvette Centrale were extracted from the global wetland database [35] (Lehner & Doll, 2004).  [26] (after Laraque et al., 2020), and the wetlands of the Cuvette Centrale were extracted from the global wetland database [35] (Lehner & Doll, 2004).  Table 1.
The main drainage systems of the basin are the Ubangi and Sangha catchments in the north and the Lualaba and Kasai catchments in the south, which are separated by the equator (Figure 1). Their tributaries converge and join the Congo main stem in the middle of the basin. Precipitation and hydrological regimes over the basin are bimodal and typical of equatorial regions, with the main flood from November to January and a lesser flood from April to June [28] (Moukandi et al., 2021). Annual average rainfall in the Cuvette Central are over 2000 mm yr −1 and decrease to as low as 1000 mm yr −1 in the northern and southern portions of the basin ( [33] Alsdorf et al., 2016, see Figure 2d). The Cuvette Centrale is estimated to accumulate up to 9 km of sediment over much of geological time [36] (Crosby et al., 2010). The lithology of the basin comprises chiefly of alluvium, metamorphic and clastic sediments with most of the groundwater stored in permeable unconsolidated alluvium that could reach 400 m deep [37,38]

Hydro-Sedimentary Measurements in the Congo River Basin
The Congo River Basin (CRB) is among the least studied river basins in the world [26,33] Table 1.
The main drainage systems of the basin are the Ubangi and Sangha catchments in the north and the Lualaba and Kasai catchments in the south, which are separated by the equator (Figure 1). Their tributaries converge and join the Congo main stem in the middle of the basin. Precipitation and hydrological regimes over the basin are bimodal and typical of equatorial regions, with the main flood from November to January and a lesser flood from April to June [28] (Moukandi et al., 2021). Annual average rainfall in the Cuvette Central are over 2000 mm yr −1 and decrease to as low as 1000 mm yr −1 in the northern and southern portions of the basin ( [33] Alsdorf et al., 2016, see Figure 2d). The Cuvette Centrale is estimated to accumulate up to 9 km of sediment over much of geological time [36] (Crosby et al., 2010). The lithology of the basin comprises chiefly of alluvium, metamorphic and clastic sediments with most of the groundwater stored in permeable unconsolidated alluvium that could reach 400 m deep [37,38]

Hydro-Sedimentary Measurements in the Congo River Basin
The Congo River Basin (CRB) is among the least studied river basins in the world [26,33]  The compilation showed that TSS concentrations remained relatively high during the first half of the century and through the wet decades of the 1960s, ranging from 12 mg L −1 to 58 mg L −1 from 1926 to 1986. [40] Olivry et al. (1988) attributed the dispersed results of earlier studies to the imprecision of measurements. [19] Probst et al. (1992) used stochastic and deterministic methods to estimate 41.61 × 10 6 t yr −1 and 3.41 × 10 6 t yr −1 of dissolved material exported by the Congo and Ubangi rivers 40 km upstream of Brazzaville and at Bangui respectively. They noted that the monthly concentrations of dissolved elements vary inversely with discharge. [13] Moukolo et al. (1993), on the other hand, derived values of 61.1 × 10 6 t yr −1 for the basin outlet from 1987 to 1991, using another method of calculation. For the next few years between 1987 and 1996, monthly sampling and analyses were performed and velocity weighted concentrations of TSS fluxes were computed [41,42] (Olivry et al., 1995;Coynel et al., 2005). The range of TSS fluxes was from 8.3 to 11 t km −2 yr −1 further highlighting the limited interannual variability of material fluxes in the CRB.
Laraque et al. [9,43] (2009, 2013a) gives a concise review of material export rates at the sub-catchment and whole catchment level respectively. The lack of data for some drainage units was overcome using the principle of similarity-a method that relies on a sound knowledge of the physiographic functioning of the basin. Mean TSS concentrations were also within the range of previous assessments at 25.0 mg L −l . Similarly, [43] Laraque et al. (2013a) obtained, using linear interpolation methods for missing data, average TSS concentrations of 25.4 mg L −l at the basin outlet. [27] Moukandi N'kaya et al. (2020) made a comparison of the variability between two measurement periods: the Programme d'étude de l'Environnement et de la Géosphère Intertropicale-Operation Grands Bassins Fluviaux (PEGI-GBF), and the Observation Service For The Geodynamical, Hydrological and Biogeochemical Control of Erosion/Alteration and Materials Transport in the Amazon, Orinoco, and Congo Basins (SO-HYBAM). They noted a 7.5% increase of TSS (from 30.2 to 33.6 × 10 6 t yr −1 ), a 28% increase of DOC, and a 15% decrease in TDS concentration in the current period. They attributed these variations to the increase in water flows since the PEGI period. While these studies have shown the low variability of the material fluxes of the CRB, largely due to its stable and regular hydrological regime, they noticed an increase in the year-to-year variability of TSS concentration, from 25.3 ± 4.6 mg L −1 in 1987-1993 to 27.2 ± 7.9 mg L −1 in 2006-2012. However, there is the need for a more regular monitoring of the river networks with emphasis on the spatial distribution. For now, the only reliable gauge network is at the basin outlet with other sources of data, mainly independent studies conducted by workers for short periods. The obvious impact on this work will be the uncertainties that will arise from this lack of spatial and temporal representation.

Data and Methodology
The TRMM rainfall product (TMPA) 3B42 V.7 was used to force the SWAT hydrological model. The TRMM product has a spatial resolution of 0.25 • and has been validated over the Congo basin by Munzimi et al. (2015) using independent datasets of historical isohyets. Water discharge and suspended sediment measurements from the five principal drainage basins of the Congo basin were obtained from the SO/HYBAM Observatory and its collaborators; notably the CICOS (Commission Internationale du bassin Congo-Oubangui-Sangha), the GIE-SCEVN (Groupement d'Intérêt Economique-Service Commun d'Entretien des Voies Navigables), the RVF (Régie des Voies Fluviales), the PEGI/GBF project, and the Marien N'gouabi University. The 90m resolution SRTM digital elevation model from the consortium from spatial information was used. This resolution was considered considering the size of the basin and the uncertainties related to DEMs [44] (Wechsler et al., 2007). Meteorological data from 1979 to 2014 was provided by the Climate Forecast System Reanalysis (CFSR) and is available at https://swat.tamu.edu/ (accessed on 6 June 2018). Other spatial Water 2021, 13, 1388 6 of 31 data include the 1 km resolution Harmonized World Soil Database [45] (HWSD), while the land cover and land use information was derived from the Global Landcover database 2000. Further details of these inputs and other supporting data used are shown in Table 1.

Modeling Approach SWAT Model Overview
The Soil and Water Assessment Tool (SWAT) is a semi-distributed, physically-based model capable of integrating various spatial environmental data, including climate, soil, land cover and topographic features. SWAT simulates various hydrological processes e.g., surface runoff, infiltration, evapotranspiration (ET), lateral flow, percolation to shallow and deep aquifers and channel routing [48] (Arnold et al., 2012). SWAT simulates the hydrology of a catchment in two ways: (i) the land phase of the hydrologic cycle and (ii) the channel or routing phase of the hydrological cycle. The first phase controls the amount of water, sediment, nutrients and pesticides loadings to the main channel in each subbasin. The second phase is related to the movement of water, sediments, nutrient and pesticide through the channel network of the watershed to the outlet [49] (Neitsch et al., 2011). More details on the hydrological setup of SWAT for the Congo basin can be found in [50] Datok et al. (2021) and further details of the hydrological processes and equations used for hydrological simulation can be found in the SWAT documentation (https://swat.tamu.edu/; accessed on 6 June 2018).

SWAT Hydrological Model Setup
The catchment was delineated based on the dominant land use, soil and slope classes apportioning one sub-basin per Hydrologic Response Unit (HRU). This delineation performed within ESRI ArcGIS 10.4 using the ArcSWAT interface (www.esri.com; accessed on 6 June 2018) resulted in 272 subbasins and HRUs with 20 land uses and 14 soil classes (Figure 2a,b). Reservoirs were integrated into the upper part of the Lualaba subbasin to take into account the series of lakes and swamps that act as storages and affect downstream flows. The reservoirs were placed in the outlet of the sub-basins that receive flow from Lakes Tanganyika, Upemba, and Mweru. Reservoir parameters were set according to [50] Datok et al. (2021). Surface runoff was simulated using a modification of the soil conservation service Curve Number (CN) method. The runoff from each sub-basin was routed through the river network to the main basin outlet using the variable storage method. We modelled Evapotranspiration using the Penman-Monteith method. The model was calibrated with measured data on a monthly time scale from 1998 to 2012; comprising calibration (2000)(2001)(2002)(2003)(2004)(2005)(2006), validation (2006)(2007)(2008)(2009)(2010)(2011)(2012), and a two-year warm-up period (1998)(1999)(2000), to enable the model to achieve a quasi-steady state condition. The criteria of efficiency (NSE) and Kling-Gupta efficiency [51] (Gupta et al., 2009) in 60% of the calibrated and validated stations were above 0.6 and 0.7, respectively, while the percent bias (PBIAS) values ranged from −8 to 8 in all the stations. This result further gave confidence that the hydrology was well simulated, therefore guaranteeing a solid base for estimating soil erosion and sediment dynamics. Further details on the hydrological simulations performed in this case study can be found in [50] Datok et al. (2021).

SWAT Sediment Model
SWAT uses a Modified Universal Soil Loss Equation (MUSLE) developed by [52] Williams (1975) to simulate sediment yield: where, Sed is the sediment yield on a given day (metric tons day −1 ); Q surf = Surface runoff volume (mm day −1 ); q peak = runoff peak discharge (m 3 s −1 ); A hru = HRU area (ha); C, P, K, LS and F, are non-dimensional factors associated with the crop cover, management practice, soil erodibility, slope and stoniness [53] (Wischmeier and Smith, 1978). The stoniness refers to the fraction of rock fragments at the surface and must be distinguished from rock fragments below the surface. Surface rock fragments affect the C-cover factor, while below surface rock fragments affect the soil erodibility. The MUSLE replaced the rainfall factor (R) of the original USLE [54] (Wischmeier and Smith 1965) with instantaneous peak flows and total runoff to predict soil erosion. Two processes control the sediment transport in the channel network: deposition or erosion. The channel dimensions can either be kept constant so that SWAT can compute deposition/erosion. Otherwise, channel degradation can be activated which would allow channel dimensions to change and be constantly updated as a result of downcutting and widening [49] (Neitsch et al., 2011). In this study, we selected the former option, as the latter has not been proven yet. Just like the hydrological phase, the sediment phase also consists of the landscape and channel component. SWAT routes different particle sizes through grassed waterways, wetlands/ponds and vegetative filter strips in the land phase. Therefore, the sediment yield reaching the stream channel is the sum of the total sediment yield calculated by MUSLE minus the lag and the impediments listed before [49] (Neitsch et al., 2011). To calculate sediment balance, Sediment from upland sub-basins are routed and added to downstream reaches. This depends on the sediment transport capacity of the reach, which in turn depends on the stream power at the reach. A modified Bagnold's equation is employed to calculate the maximum concentration of sediments that can be transported in the reach: where Conc max = the maximum daily concentration of sediments (t m −3 ) that can be transported in the reach; prf = the sediment peak rate adjustment factor; q ch =average rate of peak flow (m 3 s −1 ); A ch = channel cross-sectional area (m 2 ). The coefficients C sp and spexp regulate the linear and exponential relationship between the stream power and the peak velocity and are customized by the user during model calibration. Sediments are deposited when sediment concentration exceeds Conc max (aggradation). Conversely, when sediment concentrations are below the stream transport capacity, nothing is deposited and bed sediment could be eroded if channel erodibility is activated. In this study, channel erodibility was applied sparingly during calibration in selected subbasins. More details of sediment transport equations and processes simulated in SWAT can be found in the SWAT theoretical documentation [49] (Neitsch et al., 2011).

SWAT Sediment Setup
SWAT2012 and ArcGIS 10.4 were used in this study. As stated earlier, the whole basin was divided into 272 subbasins with 14 land-use classes ( Figure 2a) and 20 soil classes ( Figure 2b). Only one HRU was defined for each subbasin, based on the combination of the dominant land use and soil type. This was done considering the size of the basin and the computational requirements needed. Furthermore, the MUSLE must be applied to a hydrologically isolated unit for a correct assessment of the water balance [55] (Vigiak et al., 2015). The final range of the HRU area was 0.085-61,775 km 2 with a median value of 9726 km 2 . We considered this delineation fit for our purpose, with consideration of our inputs, computational resources and modelling priorities to avoid the risk of misleading results [56] (Jha et al., 2004).
The USLE cover and management factor (C USLE ) values were converted from the 14 land uses (Figure 2a) by SWAT based on the amount of soil cover specified for a particular land use, and were extracted from the soil database during model set up. The agricultural area represents a small part of the land use in the basin (<2%) (see Figure 2a), therefore the C USLE and support practice (P USLE ) factors were not calibrated and set at default values over this land use. The topographic factor LS USLE is automatically derived from the DEM while the coarse fragment factor (F CRFG ) is calculated from the percent of rock in the first soil layer. When there is no rock in the first layer, the F CRFG value is 1.
The soil erodibility factor (K USLE ) was carefully calculated over each HRU. This was important in order not to expect too much from default model parameters, and in data scarce basins such as the CRB; a hydrological model should be setup with as much information as available. The default model parameters are based on the Harmonized World Soil Database, and SWAT commonly uses the most dominant soil series in that soil association. The information contained in the database is variable [45] (HWSD), and thus may or may not be representative of the soils in the locality.
The K USLE erodibility factor K was estimated using the freely available software Kuery 1.4 [57,58] (Borselli et al., 2009(Borselli et al., , 2012. It is based on the equations developed by [59,60] Torri et al. (1997,2002) from observations of a global dataset of 240 soils where erodibility is expressed as: where C is the fraction of total clay content, OM is organic matter as a percentage, and D g is the logarithm of the geometric mean D of the particle size distribution. In cases such as our own, where data of textural components of soil exist only as a percentage; sand (S), loam (L) and clay(C), D g is calculated according to [61] Shirazi et al. (1988) as: As compared to a homogeneous soil, K is affected by the percentage of rock fragment (R fg ) in the soil, which is needed as input in Kuery 1.4, as well as the climate classification  Sanchis et al., 2008). In addition, the logarithm of the geometric standard deviation of the geometric mean of the particle size distribution (Sg) [61] (Shirazi et al., 1988) is needed: The technical report of the soil and terrain database of the Democratic Republic of the Congo [63] (Engelen & Verdoodt, 2006), the Soil and Terrain Database for Central Africa (SOTERCAF) and the Harmonized World Soil Database [45] (HWSD) were consulted for the basic inputs D g , S g , OM, and percentage rock content (R fg ). The means of the resulting probability distribution was implemented and spatialized in the SWAT model based on the dominant soil types in the basin (Figure 2b). Table 2 shows the textural components in the percentage of ferralsols used as inputs to the Kuery 1.4 model. Ferralsols are the dominant soil class in the CRB covering 58% of the watershed area as delineated by the SWAT model. Textural components of the ferralsols were obtained based on the soil characteristics of eight locations covered by these soils. Other dominant soil classes with their textural components are shown in Table 3. These eight soil classes make up 40% of the watershed area with watershed size ranging from 0.65% to 17.41%. The textural component percentages of these eight soil classes were obtained from characteristics of these soils and generalized across the basin ( Table 3). The climate zone for our study area corresponds to the tropical classification of the Köppen-Geiger climate map of [64] Peel et al. (2007) and identified as such in our study. The generated K values frequency distribution allowed us to choose a more representative value for a given soil based on its characteristics given in our soil databases. These characteristics linked mainly to soil structural stability are derived from the description of soil profiles in the databases and literature.

Sediment Balance
We replicate the approach used by [50] Datok et al. (2021) in their hydrological study to estimate the sediment balance in the Cuvette Centrale. 11 points were selected in the model; eight upstream sub-catchments of River stations with their outlets feeding into the wetlands of the Cuvette Centrale (stations 1 to 8, Figure 3). They were completed by three additional stations: one at the outlet of the Kasai subbasin (st. 10), one at the gauging station of Brazzaville/Kinshasa (st. 11), and one intermediate station in the Congo River downstream of the wetlands (st. 9). The total input of sediment for the Centrale Cuvette is

Sediment Balance
We replicate the approach used by [50] Datok et al. (2021) in their hydrological study to estimate the sediment balance in the Cuvette Centrale. 11 points were selected in the model; eight upstream sub-catchments of River stations with their outlets feeding into the wetlands of the Cuvette Centrale (stations 1 to 8, Figure 3). They were completed by three additional stations: one at the outlet of the Kasai subbasin (st. 10), one at the gauging station of Brazzaville/Kinshasa (st. 11), and one intermediate station in the Congo River downstream of the wetlands (st. 9). The total input of sediment for the Centrale Cuvette is given by I = ∑ 8 =1 , the output is considered as the flux at Brazzaville/Kinshasa minus the contribution of Kasai O = S11 -S10, and the sediment balance of the Cuvette Centrale is:

Calibration and Validation of the Model
The hydrological model was validated daily. The model was then calibrated on a daily time-step with suspended sediment (SS) concentration time series (mg L −1 ). This time series comprises a one to three times monthly sampling obtained from the HYBAM observatory located at the Brazzaville/Kinshasa gauging station. The period of the calibration was 2006 to 2009, while validation was performed from 2010 to 2012. The other only source of SS data that was reliable for use was that measured on the Ubangi River [65,66]

Calibration and Validation of the Model
The hydrological model was validated daily. The model was then calibrated on a daily time-step with suspended sediment (SS) concentration time series (mg L −1 ). This time series comprises a one to three times monthly sampling obtained from the HYBAM observatory located at the Brazzaville/Kinshasa gauging station. The period of the calibration was 2006 to 2009, while validation was performed from 2010 to 2012. The other only source of SS data that was reliable for use was that measured on the Ubangi River [65,66]  . Spatialization of the calibration was highly limited due to the lack of gauging stations; therefore, we relied on literature data to calibrate broadly. Manual calibration with the aid of expert knowledge was also used for calibration. Since we lacked critical data for calibration, we first ran our model as simple as possible. We assumed zero net erosion or deposition in the channel by setting the exponential and linear parameters to eliminate deposition and the channel routing parameters to eliminate erosion. We then adjusted the support practice factor that controls the upland sediment yields and calibrated parameters controlling channel deposition and erosion in an upstream-downstream manner. Details are given in the next section.
Due to the paucity of SS data in this region, the simulated SS concentrations in the period 2000-2012 were validated using historical records measured at the Brazzaville/Kinshasa gauging station from the PEGI/GBF program (1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994). The justification for this method is based on the works of [67,68] Laraque et al. (2001, 2013b) and [28] Moukandi et al. (2021), who using statistical tests highlighted the low variability of hydrological regimes during different homogenous periods in the CRB. We selected stations that had at least five months of readings and compared the means noting the assumptions made in comparing the model performance objectively due to the lack of a common frequency of measurement.

Statistical Analysis
Literature has described the CRB as flat [31] (Daly et al., 1992), with the absence of mountain ranges and thick vegetation cover, resulting in low erosion rates [26]

Hydrological Responses in the Main Tributaries
Generally, the model exhibited an acceptable correspondence between the observed and modelled flows, showing its ability to capture the general shape of the hydrograph and magnitude of both high and low flows satisfactorily, at the outlet of the basin as well as in its sub-basins [50] (Datok et al., 2021). The validation of the daily simulations highlighted that SWAT can be calibrated at a monthly time step and validated at a daily time step. The differences in the evaluation criteria show the effect of the regionalization in the calibration as well as the level of complexity of each sub-basin. The lower performance of the daily model compared to the monthly simulation can be linked to the inadequate spatial representation of rainfall as well as errors in other inputs and measurements [71]  However, to have an overview, we summarize the main points that are essential for the application to sediment modelling.
The Northern sub-basin of the Ubangi, which controls about 16% of the basin, gave good to satisfactory outputs for the evaluation criteria used since it had the most reliable observed data amongst all sub-basins. Therefore, most of the processes operating in the basin were well captured.
The Sangha sub-basin, which is situated west of the Ubangi sub-basin (Figure 1), was also calibrated on a monthly time scale. The model was not able to capture the minor flood phases in July and August, which were amplified on a daily scale. This was put down to the effect of wetland processes and the inadequacies of the model structure to deal with this. Nevertheless, it reproduced satisfactorily the timing of the hydrograph as well as the water balance components.
In the southwestern Kasai sub-basin, performance ratings ranged from satisfactory to good, with the hydrograph dynamics well captured. The percent bias (PBIAS) in this sub-basin was low, which could be deceptive as the model underpredicted as much as it over-predicted. However, other evaluation criteria (the NSE, R 2 ), gave a good account of the results here, and the discharge predicted was within three per cent of the observed values.
The challenge was in attenuating the flows from the wetland regions of the basin, specifically in the upper parts of the Lualaba and the central basin. Reservoirs were added to the Lualaba basin to help in trapping the water generated by the large water bodies there. However, we were able to capture satisfactorily the magnitude of flows within one per cent of the observed value.
At the basin outlet, the effect of the convergence of flows from all the contributing tributaries was evident. We were able to reproduce the magnitude of the flows to within 9% of the observed flows. The objective of the study is not to achieve a perfect calibration of the discharge at the outlet, as the challenges have been well documented [72][73][74][75]  . Rather, the objective is to find a balance between the flows received by the central basin before being impacted by the "Cuvette Centrale" and the flows measured at the basin outlet. Moreover, to achieve this aim using a mass balance approach, observed measurements are available at the outlet for comparison. An analysis of the parameters showed that groundwater parameters were influential all across the basin, with subsurface processes dominating the Sangha and the central basin. The central basin also had the least response to recharge, while the Kasai and Ubangi basins are drawing up more water from the shallow aquifer to the overlying unsaturated zones through deep-rooted vegetation. Some optimal parameters like the canopy interception compared well with physical observations, being higher in areas with more vegetation and lower in less vegetated areas.
In estimating the water balance of the Cuvette Centrale, [  The hydrological regime of the Ubangi is unimodal, with the main flood period from September to November. There is a noticeable lag in the suspended sediments as they follow the discharge, a form associated with a higher concentration-discharge ratio at the rising limb than the falling limb [76] (Megounif et al., 2013), creating a clockwise hysteresis (Figure 4b), described by [77] Laraque and Olivry (1996). The sediment decline at the peak flood is usually attributed to diminishing sediment available in the stream channels [78] (Wang et al., 1998). It must also be noticed that the variability of monthly-suspended sediment concentration (SSC) values is high in the Ubangui basin (from 2.4 to 33.2 mg L −1 in 2010-2011).
rising limb than the falling limb [76] (Megounif et al., 2013), creating a clockwise hysteresis (Figure 4b), described by [77] Laraque and Olivry (1996). The sediment decline at the peak flood is usually attributed to diminishing sediment available in the stream channels [78] (Wang et al., 1998). It must also be noticed that the variability of monthly-suspended sediment concentration (SSC) values is high in the Ubangui basin (from 2.4 to 33.2 mg L −1 in 2010-2011).  (1996) noted the absence of the erosion-transport-alluvial phases that are associated with the Bangui station. Instead, the Brazzaville/Kinshasa outlet is noted for a single transport phase. This is due to the low seasonal variations in its particulate concentrations and its flows due to the all-year-round contributions of different sub-basins of the watershed to its flow. The hysteresis in Figure 5b displays a figure-eight pattern that combines a clockwise and anti-clockwise stage. This is due to the influence of the two equatorial regimes consisting of the main flood peak and a lesser one. The clockwise stage involves an increase of concentrations from the erosion of initially deposited stream sediments and thereafter a dilution effect with increased discharge [78,79] Laraque and Olivry (1996) noted the absence of the erosion-transport-alluvial phases that are associated with the Bangui station. Instead, the Brazzaville/Kinshasa outlet is noted for a single transport phase. This is due to the low seasonal variations in its particulate concentrations and its flows due to the all-year-round contributions of different sub-basins of the watershed to its flow. The hysteresis in Figure 5b displays a figure-eight pattern that combines a clockwise and anti-clockwise stage. This is due to the influence of the two equatorial regimes consisting of the main flood peak and a lesser one. The clockwise stage involves an increase of concentrations from the erosion of initially deposited stream sediments and thereafter a dilution effect with increased discharge [78,79] Table 4 shows some parameters used to calibrate sediment dynamics in the SWAT model. The parameters that proved important in the calibration of the sediment yield and transport are related to the channel processes and processes in the Hydrological Response Units (HRUs). The simulation reveals that the exponential parameter for sediment routing in the channel was the main parameter among the general basin-wide parameters with influence. The values of channel erodibility (CH_COV1) and sediment transport coeffi-  Table 4 shows some parameters used to calibrate sediment dynamics in the SWAT model. The parameters that proved important in the calibration of the sediment yield and transport are related to the channel processes and processes in the Hydrological Response Units (HRUs). The simulation reveals that the exponential parameter for sediment routing in the channel was the main parameter among the general basin-wide parameters with influence. The values of channel erodibility (CH_COV1) and sediment transport coefficient (SPCON), set to their default values reveal the dominance of channel depositional processes within the basin. The calibration of these factors was done in an upstream-downstream manner. A filter was also implemented in sub-basin 168 corresponding to the confluence of the Kasai with the main river. This was important, as this is a site of deposition due to a change in velocity [17] (Mushi et al., 2019). The length of the filter simulated is 28 m and caused a reduction of about 4% of the total sediment yield at the basin outlet. It is also important to note that the filter strip reduces sediment but does not affect surface runoff [49] (Neitsch et al., 2011). Calibration was done at a daily time step at the Bangui station outlet using measured data from the study of [65,66] (Figure 6b). At the Bangui station, the model was able to reproduce the general shape of the hydrograph, capturing both periods of high exports and low exports, a reflection of the good responses already documented during the hydrological calibration and validation at this station [50] (Datok et al., 2021). Statistical metrics used to assess the models' performance are displayed in Table 5. The coefficient of regression of 0.69 showed how much of the variability of the suspended sediments are related to the discharge as the hydrological model has good skill scores. The Nash-Sutcliffe Coefficient of Efficiency (NSE) value of 0.35 is a satisfactory representation of the dynamics captured here when the temporal range of the observed data, as well as the sampling frequency, is considered. In Figure 6a, it can be seen that the simulated load does not follow the observed load during the rising phase of the hydrograph but is more correlated with the discharge at the falling limb. The abnormally high PBIAS value of 49.10 is because of this under-prediction of the rising limb where the magnitudes were not properly captured. Statistics were better in the validation period, with a satisfactory PBIAS value of 24.2%. It is also worth noting that in this sub-basin, the sediment discharge varies over more than two orders of magnitude (2.4-33.2 mg L −1 in TSS and 470-8540 m 3 s −1 in Q, resulting in a 101-24,508 t day −1 range for sediment discharge).

Calibration Performance
does not follow the observed load during the rising phase of the hydrograph but is more correlated with the discharge at the falling limb. The abnormally high PBIAS value of 49.10 is because of this under-prediction of the rising limb where the magnitudes were not properly captured. Statistics were better in the validation period, with a satisfactory PBIAS value of 24.2%. It is also worth noting that in this sub-basin, the sediment discharge varies over more than two orders of magnitude (2.4-33.2 mg L −1 in TSS and 470-8540 m 3 s −1 in Q, resulting in a 101-24,508 t day −1 range for sediment discharge).    Table 5). The NSE is sensitive to extreme flows and reflects the under-prediction of peak flows as well as the influence of the Cuvette Centrale. The low values of regression coefficients also do not give a satisfactory indicator of the model performance at this outlet as the model did not accurately pick the timing of flows. This is in agreement with what has already been discussed about the failure of models to adequately slow the flow of water across the "Cuvette Centrale" [50,73] (Tshimanga et al (Figure 6a) also rely on much smaller ranges at Brazzaville (less than one order of magnitude) than at Bangui (more than two orders of magnitude in sediment load). The PBIAS of 9% in the validation period is adjudged satisfactory, however, showing that the model predicted well the average magnitudes although showing a bias towards underestimation of the fluxes.
From the analysis of our calibration dataset and the model calibration, it can be seen that SWAT sediment routing performs better on basins or sub-basins with low or moderate hysteresis (e.g., the Ubangi) than those with high hysteresis (e.g., at Brazzaville/Kinshasa). At Brazzaville, we observe a complex relationship between Q and SSC (figure-eight hysteresis), while the Bagnold formula considers a maximum concentration, increasing with increasing water discharge (or increasing discharge peaks). The difference may be attributed to processes of deposition/erosion over the land (especially in wetlands) or at the riverbed that are very difficult to calibrate correctly over a large basin with a very scarce dataset and natural complexities. To overcome these difficulties, there will be a need for additional measurements in the future at different stations of the basin to improve the model performance. In its state, the model was able to reproduce concentration values and sediment fluxes at a satisfactory level (with PBIAS < 50). It enables the first assessment of sediment balance in the Cuvette Centrale-which is the main target of the paper.

Additional Comparison of the Model Simulations with Historical Datasets
We compared our simulated fluxes with data from the PEGI-BGF program on the 1987-1994 period since there has not been any sustained sampling program over the entire basin. The justification of this comparison is based on the regularity of measured fluxes in the basin over the past century. For instance, [ Table 6. The PEGI observations of SSC and Q were sampled once every month, with some months not sampled at all in some stations, while we used the full time series for the SWAT simulations. More detailed statistical analysis could not be carried out because the frequency of measurement was not the same.
Average SSC values were very well estimated with the model at Bouatali (Likouala aux Herbes River), Ouesso (Sangha River) and Makoua (Likouala Mossaka River), with differences less than 7%, while it was over-estimated by 31.2% at BRZ/Kinshasa (Congo R.), and by 110% at Alima (Tchikapika) ( Table 6). Of course, these quantities are not perfectly comparable since simulations ran over years while measurements were performed a limited number of times per station. However, this result confirms the ability of the model to assess a first sediment budget of the Cuvette Centrale.

Surface Runoff and Sediment Yield
We analysed the spatial variation of sediment yield across the basin (Figure 7). Generally, it can be observed that high surface runoff (Figure 7b) corresponds to high sediment yields (Figure 7a) across the subbasins. Surface runoff also followed the rainfall pattern of the basin with the central areas receiving some of the highest rainfall and above-average sediment yields. The sediment yields are generated from individual HRUs and from Figure 7a, it can be deduced that the Cuvette Centrale area with its various land uses including row cropped fields or pasture is characterized by local erosion. These sediments are generated and then accumulate here, aided by the low slopes and low velocities of water, which combine to restrict the transport of these sediments. The mapped sediment yield rates compare to the ones mapped by [9] Laraque et al., (2009). We estimated sediment yield rates of 4.01, 5.91, 7.88 and 8.68 t km −2 yr −1 respectively for the Ubangi sub-basin at Bangui, Sangha at Ouesso, Lualaba at Kisangani, and Kasai at Kutu-Moke respectively. The sources of these sediments are mainly from the higher slopes of the eastern highlands, the southern parts of the basin as well as sources from the Angolan highlands to the southwest. These loads are conveyed in the river from the upper course at an altitude of 1400-1500 m and they gather more mass as the river exits the numerous swamps and wetlands in the Lualaba basin. There is a noticeable reduction in river load as it makes its way through the anastomosing river channels of the middle Congo River. These channels are up to 16 km wide at some places [82] (Runge, 2007) and are characterized by sand and silt bars indicating deposition. Effective river width of around 5 km, measured 700 km upstream from Brazzaville/Kinshasa, reduce to less than 3 km at 1600 km upstream as measured by [83] Carr et al. (2019). At the equatorial regions, tributaries from the northern and southern hemispheres join the main river and increase the loads. Some of these loads are contributed by the Ubangi catchment due to the erosion of grazing land of the Sudano Sahelian environment [81] (Nguimalet & Orange, 2013). Water surface slopes are higher here with a reduced velocity. This river stretch is multi-channel with constrictions that reduce the widths considerably [83] (Carr et al., 2019). The Kasai River then joins the main stem at Kwamouth. The Kasai is noted to have a large suspended sediment load [9] (Laraque et al., 2009), contributing to the main river. This is consistent with measurements by [83] Carr et al. (2019) where they found water surface slopes to be very low in this area (2 cm km −1 ). The loads are then transported and reduced through the Malebo pool with some loads produced at the urban settlement of Brazzaville/Kinshasa. here with a reduced velocity. This river stretch is multi-channel with constrictions that reduce the widths considerably [83] (Carr et al., 2019). The Kasai River then joins the main stem at Kwamouth. The Kasai is noted to have a large suspended sediment load [9] (Laraque et al., 2009), contributing to the main river. This is consistent with measurements by [83] Carr et al. (2019) where they found water surface slopes to be very low in this area (2 cm km −1 ). The loads are then transported and reduced through the Malebo pool with some loads produced at the urban settlement of Brazzaville /Kinshasa.

Land Use/Land Cover and Sediment Yield
We investigated the relationship between land cover, surface runoff and sediment yield in the CRB sub-basins over different land uses (Figure 8). The dominant land uses are forests, which occur in all the sub-basins while agriculture, pastoralism, and rangeland are the land uses with the most sediment yield.

Land Use/Land Cover and Sediment Yield
We investigated the relationship between land cover, surface runoff and sediment yield in the CRB sub-basins over different land uses (Figure 8). The dominant land uses are forests, which occur in all the sub-basins while agriculture, pastoralism, and rangeland are the land uses with the most sediment yield.
The Ubangi sub-basin is dominated by forests of the mixed kind, comprising small trees and shrubs, which are interspersed with larger broadleaved trees at the northern limits of the basin. At the lower and wetter parts of the sub-basin are the boundaries of the evergreen forest. The highest yields of 4.5 t km 2 yr −1 was in the mixed forests located on higher slopes, while the lowest yield of 2.16 t km −2 yr −1 was in the evergreen forests. The Ubangi sub-basin presented the lowest specific sediment yield of the principal basins studied. However, in terms of load contribution, it ranks second of the sub-basins that contribute to the Cuvette Centrale after the Lualaba sub-basin. This is mainly because the discharge measured at the Bangui station is the largest of the Northern tributaries. Low sediment yields could also be attributed to the largely stable nature of the Ubangi tree-covered peneplains [17] (Mushi et al., 2019).
The Sangha basin drains the northwestern limits of the CRB; it has relatively high sediment yields compared to its size-5.91 t km −2 yr −1 (this study) and 8.52 t km −2 yr −1 [9] (Laraque et al., 2009). The highest yield of sediment is found in pastureland (12.22 t km −2 yr −1 , see Figure 8) while the evergreen forests produced the least sediment of 3.97 t km −2 yr −1 . Soils of the exposed pasturelands are more susceptible to erosion than those of the evergreen forests, which are protected by dense tree cover. This is a view held by [84] Ndomba et al. (2011), who also pointed out that agriculture and pastoralism are the main sources of sediment in catchments.
In the Lualaba sub-basin, agriculture is the largest land use by size with the highest sediment yields. Runoff is low though suggesting infiltration of water through the soil profile. Rangeland has a relatively high specific yield of 9.17 t km −2 yr −1 considering that this land-use ranks fourth in the area behind agriculture, evergreen, and deciduous forests. This may in part be due to overgrazing or other anthropogenic activities that expose the soil. Deciduous forests yielded 7.71 t km −2 yr −1 of sediment while the evergreen forests yielded 9.68 t km −2 yr −1 of sediment. The evergreen forests yield the least sediment in the other sub-basins studied. [85] Mayaux et al. (2003) described the evergreen swamp forests as "edaphic", alluding to the influence of the sediments on their sustenance. The trees are also mostly dense and multilayered, helping to slow the agents of erosion. The exception in the Lualaba may be due to the relatively higher slopes in some areas, which induce more yields. Non-forested wetlands yielded 1.25 t km −2 yr −1 of sediment while wheatgrass yielded 7.39 t km −2 yr −1 of sediment, noting that, while it is not as high as agricultural output, it shows the effect of different management practices on sediment yield.
The Kasai sub-basin is known to be heavily mineralized, due to the many mining activities taking place there. These anthropogenic activities are very impactful on the land. Therefore, it is not surprising that the highest sediment yields in the CRB are attributed to this sub-basin. We simulated specific sediment yields of 3.8 t km −2 yr −1 , 11 t km −2 yr −1 , and 12.7 t km −2 yr −1 for evergreen forests, deciduous forests, and rangeland, respectively ( Figure 8). The sediment yield is more pronounced in the southwestern portion of this sub-basin probably due to anthropogenic activity mentioned earlier and/or less stable natural terrain [17] (Mushi et al., 2019). The Ubangi sub-basin is dominated by forests of the mixed kind, comprising small trees and shrubs, which are interspersed with larger broadleaved trees at the northern limits of the basin. At the lower and wetter parts of the sub-basin are the boundaries of the evergreen forest. The highest yields of 4.5 t km 2 yr −1 was in the mixed forests located on higher slopes, while the lowest yield of 2.16 t km −2 yr −1 was in the evergreen forests. The Ubangi sub-basin presented the lowest specific sediment yield of the principal basins studied. However, in terms of load contribution, it ranks second of the sub-basins that contribute to the Cuvette Centrale after the Lualaba sub-basin. This is mainly because the discharge measured at the Bangui station is the largest of the Northern tributaries. Low sediment yields could also be attributed to the largely stable nature of the Ubangi treecovered peneplains [17] (Mushi et al., 2019).
The Sangha basin drains the northwestern limits of the CRB; it has relatively high Land cover in the Cuvette is 57% Evergreen forest, 30% Deciduous forest, 8% Wetland forest, 2% Mixed forest with Pasture, Rangeland and Wheatgrass making up the balance. The soils are primarily composed of gleysols, which are affected by groundwater [86] (Ngongo et al., 2014). They are not well-drained and in fact, oversaturated causing relatively high surface runoff. We simulated the highest yields for the Pasture land, Deciduous forests, Wetland forests, Mixed forests, Evergreen forests, and Rangeland of 69, 14.3, 12.13, 12.09, 10.13, and 8.64 t km −2 yr −1 respectively. In the basin as a whole, high surface runoff did not necessarily mean high sediment yields, as is the case with wheatgrass. Similarly, mixed forests with the lowest yields (6.5 t km −2 yr −1 ) have only the second-lowest surface runoff rate (66 mm yr −1 ) (Figure 8b). To put Figure 8a and b in perspective, we plotted the average loads produced by each land use in Figure 8c. From the plot, pastureland, which otherwise has a very high specific yield, actually produced one of the least loads commensurate with its overall percentage of basin area. Likewise for agriculture, mixed forest, and non-forested wetlands.  (Figure 9d) is similar to theirs. The results showed that portions of the middle CRB are less affected by erosion, while high relief areas are the primary sources of sediment. It is important to note also the Cuvette central area, although located on low slopes with low erodible soils, yield relatively high sediment within its tributaries because of their high content in organic matter, which are mostly made up of fine particles that are transported easily even at low velocities. The coarser particles, most likely from stream bank erosion, will travel only shorter distances and be deposited, helping to create anastomosing structures observed along the middle reach [83,87,88]

Key Factors Affecting Sediment Yield
A principal component analysis (PCA) was performed on parameters that affect the sediment yield in the various subbasins of the catchment. Five key parameters were analysed: the slope in the HRUs (HRU_slp), precipitation (PCP), soil erodibility (KUSLE), channel vegetation cover (CH_COV2), and surface runoff (SUR_Q). PC1 explained 33.7% of

Key Factors Affecting Sediment Yield
A principal component analysis (PCA) was performed on parameters that affect the sediment yield in the various subbasins of the catchment. Five key parameters were analysed: the slope in the HRUs (HRU_slp), precipitation (PCP), soil erodibility (K USLE ), channel vegetation cover (CH_COV2), and surface runoff (SUR_Q). PC1 explained 33.7% of the data variability, while PC2 explained 24.4%. Together, PCA1 and PCA2 accounted for 56.1% of the variability in the data ( Table 7). The eigenvalues with the most magnitude in PC1 were 0.66 and 0.58, corresponding to surface runoff and precipitation, respectively. Figure 10a demonstrates this; as values increase positively along the PC1, so does precipitation and surface runoff. In addition, erodibility and slope are moderate and there is less channel cover. Similarly, the eigenvalues with the most magnitude in PC2 are erodibility and the slope in the HRUs with values of −0.62 and −0.59, respectively. Fig. 10b shows that as the values along with PC2 increase negatively, there is a corresponding increase in slope as well as the erodibility of the soils. At the same time, channel cover is low and there is no real difference between precipitation and surface runoff. To buttress these results, we used the Pearson correlation coefficient to compare the relationship between the variables and our two principal components that explain more than 50% of the data variability. Table 7 shows the highest positive correlation along PC1 of 0.86 associated with surface runoff. This implies that as the value of the observation increases along PC1 so does the surface runoff. Likewise, the next highest value is precipitation (0.75), further reinforcing the dominance of these variables in the CRB sediment processes. The reverse is also true for negative coefficient values; as the value of the principal components increase, the associated variables decrease. Low negative values would also mean negligible relationships between variables as observed with channel vegetation cover in PC1 and PC2.
Our results compare favourably with the literature cited in this study [9,17,39,69,82]  the erosional processes in the highland regions, which surround the basin as well as parts of the main drainage units of the basin. The PCA is also able to group main drainage units with similar characteristics. By condensing our variables into two linear combinations of PC1 and PC2, it will be possible to separate our sub-basins based on the similarity of characteristics. Further work is therefore needed to separate the sub-basins based on the factors influencing erosion. This would need more detailed and extensive data coupled with modelling for a deeper understanding of the sediment dynamics within the CRB.

Sediment Balance in the Cuvette Centrale
The Cuvette Centrale acts as a sediment trap slowing the flow of water, nutrients and pollutants [9] (Laraque et al., 2009). To quantify the effects of the Cuvette Centrale on upstream sediment loads, we carried out an upstream-downstream mass balance as described in Section 2.5 and as demonstrated in the hydrological balance of [50] Datok et al. (2021). Eight tributaries from both the northern and equatorial regions were considered and the results of the balance are presented in Table 8 below.   Table 8). The Sangha and the other principal tributaries on the right bank contribute three and two per cent to the Cuvette Centrale, respectively. Compared to the values of 33.7 × 10 6 t yr −1 of material measured by the SO-HYBAM at the basin outlet (station 11) from 2006 to 2017, our simulated value of 37 × 10 6 t yr −1 for 2000 to 2012 compares well. Next, we calculated the fluxes at the confluence of the Kasai River (at Kwamouth; station 10) with the Congo at 10.5 million tons, which was then subtracted from the flux of 37 million tons simulated at the basin outlet resulting in 26.6 million tons of material annually at the basin outlet. With respect to equation 6 in Section 2.7, we estimated the sediment balance as the difference between the fluxes coming in from stations 1 to 8 (~31Mt) and the fluxes at station 11 (~26 Mt) for a balance of~5 Mt. To verify the balance, we calculated the fluxes downstream of the Cuvette at Bolobo, which is 7.6 million tons of material. By subtracting the value at Bolobo from the inputs into the Cuvette Centrale gave a value of over 23 million tons, we confirmed that there is a massive reduction in the fluxes exiting the Cuvette Centrale. These high loads retained by the cuvette can be justified by the sediment yields simulated in Figure 7  The seasonal dynamics of the loads contributed to the Cuvette Centrale ( Figure 11) show that the right bank tributaries (stations 1-6; Table 8) contribute more load during the September-October-November (SON) floods associated with the equatorial regions. Disregarding the contributions from the main-stem Congo River (station 7); it is obvious that the right bank tributaries are a very important source of sediment to the Cuvette Centrale. This is especially so since unlike the Amazon which has numerous connecting pathways or "furos" between the main river and the floodplains, there is limited connectivity in the case of the CRB [33] (Alsdorf et al., 2016). The left bank tributary (station 8) also makes a modest contribution in the December-January-February (DJF) period that coincides with the small dry season of the southern basins and the pronounced dry season of the northern basins.

Comparison with the Amazon and the Orinoco
To put this study in context and appreciate the functioning of the CRB, we compare the material fluxes of the CRB with two other large world drainage basins: the Amazon and the Orinoco. This comparison is important because, after the Amazon, the Congo is the second largest area of tropical rainforest at 1.8 million km² [92] (Haensler et al., 2013); and its average discharge of 36,150 m 3 s −1 (this study) is second to the Amazon with 169,000 m 3 s −1 [93] (Filizola & Guyot, 2009) (Table 9). In comparison to the Orinoco, their mean annual discharges are similar, with the Orinoco discharging 31,400 m 3 s −1 . Together, the three river basins combine an average discharge of approximately 284 × 10 3 m 3 s −1 of water into the Atlantic [43]  . The sediment fluxes reported in Table 9 for the Amazon are taken from the Obidos station that controls 80% of the basin. In terms of the fluxes, both the Amazon and the Orinoco have much higher amounts discharged to the oceans. This is principally due to the difference in relief compared to the Congo as the South American rivers are underlain in part by the Andes [94] (Meade, 1984). In the Congo, chemical weathering dominates over physical weathering with dissolved matter superseding suspended sediments [27,43]

Comparison with the Amazon and the Orinoco
To put this study in context and appreciate the functioning of the CRB, we compare the material fluxes of the CRB with two other large world drainage basins: the Amazon and the Orinoco. This comparison is important because, after the Amazon, the Congo is the second largest area of tropical rainforest at 1.8 million km 2 [92] (Haensler et al., 2013); and its average discharge of 36,150 m 3 s −1 (this study) is second to the Amazon with 169,000 m 3 s −1 [93] (Filizola & Guyot, 2009) (Table 9). In comparison to the Orinoco, their mean annual discharges are similar, with the Orinoco discharging 31,400 m 3 s −1 . Together, the three river basins combine an average discharge of approximately 284 × 10 3 m 3 s −1 of water into the Atlantic [43]  . The sediment fluxes reported in Table 9 for the Amazon are taken from the Obidos station that controls 80% of the basin. In terms of the fluxes, both the Amazon and the Orinoco have much higher amounts discharged to the oceans. This is principally due to the difference in relief compared to the Congo as the South American rivers are underlain in part by the Andes [94] (Meade, 1984). In the Congo, chemical weathering dominates over physical weathering with dissolved matter superseding suspended sediments [27,43]

Uncertainties Related to Sediment Estimation
As with any large scale hydrological modelling study in ungauged basins, uncertainties are unavoidable. In this study, a lack of critical data for calibrating the sediment fluxes caused the model to be calibrated in a lumped manner. There is an unavoidable loss of information caused by the lack of spatial representation. However, the temporal variability was well captured and the magnitude of fluxes matched the literature. It should also be noted that concerns have been raised about the misrepresentation of wetland processes due to gaps in the model structure [50] (Datok et al., 2021). Therefore, this study was undertaken assuming that the processes occurring in the wetlands are straightforward. There is also uncertainty related to input data. For example, the soil database used to construct the SWAT lookup tables was primarily prepared for agricultural use with soil depths not exceeding 100 cm. Therefore, peak flows could be over-simulated in deeper soils resulting in an overestimation of sediment. The DEM resolution for computing the hillslope length also has implications for runoff and erosion computations. The Bagnold equation used to calculate the maximum amount of sediment transported in a stream segment has its limitations [96] (Wei et al., 2020). The algorithm does not differentiate between particle sizes and offers no distinction between channel and flood plain deposition [49] (Neitsch et al., 2011). Precipitation input uncertainty and the uncertainties of parameters of the central basin all have to be taken into account too.
In estimating the sediment balance of the Cuvette Centrale, an analysis of stream hydraulics will also help in determining if the widths and depths of the tributaries studied can process the amounts of sediment over the period considered. Settling rates of sediment in the Cuvette Centrale will also need to be studied. This is important in order not to overlook the possibility that the Cuvette Centrale is subsiding at a rate equivalent to the trapping of the sediments. This is an area of future work that will need exploring.
The existing dams that could affect our results are the Mobayi Dam downstream of the Uele River with an installed capacity of 11.5 MW and the Imboulou Dam on the Lefini River with an installed capacity of 120 MW. In the case of the former, it is upstream of the Bangui station and for the latter, it is downstream of the Cuvette Centrale. The Lualaba basin has swamps and lakes that serve as natural reservoirs with a few man-made ones. However, reservoirs were also integrated into the model to account for these structures. Considering the scale of the study, these effects although negligible, will be a source of uncertainty.

Conclusions
Understanding the sediment dynamics in wetlands is important to assess changes taking place over time correctly. These changes could be natural or human-made and could have important ecological as well as socio-economic consequences. We used the SWAT hydrological model in this study to investigate and quantify sediment fluxes and budgets from seasonal to annual time scales using freely available datasets. The SWAT model satisfactorily captured the dynamics of sediment yields and fluxes when compared to past studies. In summary, we have been able to use the SWAT model to do the following: Run the SWAT model on a daily timescale, calibrating the hydrology and sediment dynamics for the 2000 to 2012 period. Five principal catchments of the CRB were calibrated for hydrology based on previous work [50] (Datok et al., 2021). Due to lack of data, we were only able to calibrate one station in the upper part of the basin as well as the outlet of the basin for the sediments. We achieved acceptable results based on the performance of the evaluation criteria used in the study. For hydrology, the coefficients of efficiency were above 70% in 60% of the sub-basins calibrated, while other evaluation criteria also gave a good picture of the model performance. II. An assessment of the model outputs revealed that agriculture and pasture were the main land uses with the potential to contribute the most sediment yield. The absence of any major infrastructure like dams in most of the tributaries, to trap sediment even temporarily, makes this an important source of sediments. Generally, the steepest areas of the basin produced more sediment and runoff. Our map of degradation hotspots also revealed that the middle basin is least affected by erosion with degradation more pronounced in the outer rim of the Cuvette Centrale. At the same time, the Cuvette Centrale area showed the lowest values of soil erodibility mainly because of the soil properties. Thus, the erosion within this area is mainly local with corresponding sediment accumulation. The highest specific sediment yields by sub-basin were from the Kasai, aided by intense anthropogenic disturbances that lead to deforestation and degradation of uncovered land. The Ubangi subbasin, on the other hand, an old erosional surface, has the lowest specific sediment yields. In terms of load carried in stream channels, the heaviest loads were carried from the lakes at the southeastern parts of the basin and the confluence of the Kasai with the main Congo River. III. Principal component analysis revealed the main factors that aid in sediment transport.
They include surface runoff, precipitation, the slope in the HRU and the erodibility of the soils. These factors are closely related to the climate and physiography of the catchment with high relief areas more prone to erosion and the high precipitation causing increased surface runoff in the central basin.
We estimated the balance of fluxes entering and exiting the Cuvette Centrale from eight principal tributaries that contribute to the main river. Our estimates suggest that the Cuvette Centrale is indeed a giant reservoir that retains at least 23 megatons of material annually. Sediment from the upper Congo is the main contributor with 85% of the fluxes, while the principal tributary of the right bank-the Ubangi-contributes 5%. The Ruki also contributes about 5%, while the other right bank affluents contribute the remainder. With the lack of a robust database over the entire basin, the results over those areas were measured data are available, provide a higher degree of confidence. More advanced validations are needed over the ungauged portions especially at Mbandaka station as the upper Congo is the main supplier of sediments to the Cuvette Centrale.
This work also showed that the estimation of soil erodibility should not only depend on lithological information or a geological map. This is because their reliability is variable and could lead to over or under-estimates of the K values. Future improvements to the model will focus on the spatial representativeness of the sedimentation processes as well as the use of high-resolution input data. Incorporation of a wetland module will also aid in a proper understanding of the processes in the central ungauged basin (e.g., [97][98][99]  This study is a follow-up to [50] Datok et al., (2021), where a water balance of the Cuvette Centrale was made. They found that as much water passed through the Cuvette Centrale from the upper Congo as efficient precipitation falling over the area. Similarly, we estimated that a majority of the fluxes come from the upper Congo. The implications are that sediment, nutrients and pollutants are supplied to the Cuvette Centrale provided there are connecting pathways or overflows from the banks [33] (Alsdorf et al., 2016). The relatively higher water surface slopes in the middle reach passing through the Cuvette Centrale make this a real possibility. Sediment loads and yields are majorly influenced by land cover and land use, which alters upland erosion processes [96,100] (Wei et al., 2020;Sok et al., 2020). Rapid urbanization, hydroelectric dam construction and/or future water diversion schemes would also have a significant effect on the fluxes of material in the Cuvette Centrale. The consequences on downstream communities of wetland loss and nutrient imbalance from sedimentation are severe and should be given more attention.
For the first time, we have estimated the fluxes trapped by the Cuvette Centrale using an upstream-downstream mass balance approach. Our model can be used as a baseline for future sediment studies of the CRB. Different modelling approaches can also be initiated for the Cuvette Centrale from the point scale to the macro-level and vice versa to understand the processes operating within. Concerted joint efforts by stakeholders are also required if progress is to be made in fully appreciating the functioning of this still, relatively pristine watershed.