Sediment Fingerprinting Enables the Determination of Soil Erosion Sources and Sediment Transport Processes in a Topographically Complex Nile Headwater Basin

Abstract

Sediment fingerprinting was utilized to identify potential hotspots of soil erosion and sediment transport pathways in the Nile Nyabarongo Upper Catchment (NNYU) in Rwanda, where rivers and reservoirs are suffering from alarmingly high levels of sedimentation. Sediment fingerprinting is a practical approach used to identify erosional hotspots and sediment transport processes in highly mountainous regions undergoing swift land use transformation. This technique involves a statistical comparison of the elemental composition of suspended sediments in river water with the elemental composition of soils belonging to different geological formations present in the catchment, thereby determining the sources of the suspended sediment. Suspended sediments were sampled five times over dry and wet seasons in all major headwater tributaries, as well as the main river channel, and compared with soils from respective delineated watersheds. Elemental composition was obtained using laser ablation inductively coupled plasma mass spectrometry, and elements were chosen that could reliably distinguish between the various geological types. The final results indicate different levels of sediment contribution from different geological types. A three-level intervention priority system was devised, with Level 1 indicating the areas with the most serious erosion. Potential sources were located on an administrative map, with the highest likely erosion over the study period (Level 1) occurring in Kabuga cell in the Mwogo sub-catchment, Nganzo and Nyamirama cells in the Nyagako sub-catchment and Kanyana cell in the NNYU downstream sub-catchment. This map enables the pinpointing of site visits in an extensive and rugged terrain to verify the areas and causes of erosion and the pathways of sediment transport. Sediment concentrations (mg L⁻¹) were the highest in the Secoko and Satinsyi tributaries. The composition of suspended sediment was seen to be temporally and spatially dynamic at each sampling point, suggesting the need for an adequate number of sampling locations to identify erosion hotspots in a large mountainous watershed. Apart from prioritizing rehabilitation locations, the detailed understanding of critical zone soil–land cover–climate processes is an important input for developing region-specific watershed management and policy guidelines.

1. Introduction

1.1. Soil Erosion—A Critical Problem in the Nile Headwaters

Soil erosion constitutes the largest source of nonpoint pollution to rivers in watersheds worldwide, with fine sediment being the most common pollutant [1,2,3]. Increasing sediment loads in rivers leads to the deterioration of water quality, degradation of aquatic ecosystems and the filling up of river channels and reservoirs, thereby increasing flooding susceptibility [4] as well as water and hydropower energy security. Furthermore, the irreversible loss of valuable topsoil from catchments creates barren land within just a few decades, severely limiting agricultural productivity [2,5].

In Rwanda and other mountainous areas within the upper Nile Basin, suspended sediments have been sharply increasing in water bodies since the 1990s [6,7,8,9,10]. The increase in soil erosion is a direct consequence of increasing human activities in catchments, the major ones being the clearing of native forest for agriculture, inadequate soil conservation measures on hillslopes, road-building, fires and mining [10,11]. However, apart from sporadic information on sediment loads in certain Rwandan rivers, there have been no studies that identify the sources and areas that contribute most of the sediment, which makes remediation an overwhelmingly impossible task, given the sheer size and relief of these inaccessible areas. Addressing the issue of river sedimentation requires controlling upland erosion, starting from the most affected areas, which in turn requires prior accurate identification of those areas contributing disproportionately to sediment loads in catchments [12]. Furthermore, the identification of specific eroding areas enables the investigation of the factors and processes that cause soil erosion and sediment transport in the Earth’s critical zone [13]. These include the soil surface, rhizosphere and land–water interfaces—instrumental to the effective design and implementation of catchment rehabilitation programs with limited resources.

1.2. Identifying Sources of Soil Erosion—Sediment Fingerprinting

The aim of sediment transport studies at the watershed scale is to understand the source, fate and transport of sediments mobilized within a catchment [14]. However, identifying areas with high soil erosion rates is challenging in a large and mountainous catchment, as soil erosion is not always a visible or constant process, and is also impacted by differences in climate, vegetation, topography, soil type and human disturbances. Traditional direct monitoring techniques using soil erosion pins (monitoring the change in the depth of soil along a pin stuck into the soil) or soil erosion troughs (quantifying amount of sediment in a trough brought in by uphill runoff over a year) have not been very successful in identifying sources because of the intensive amount of sampling all over the catchment at various times that is needed with such approaches [15]. The use of Google Earth imagery, while enabling a landscape-level perspective, lacks temporal resolution—the time gap between individual images is long (9–12 months or more)—and thus cannot yield a comprehensive look at soil erosion events that arise from localized rainfall and human activities. An alternative approach, sediment fingerprinting, has been increasingly used worldwide since the mid-1970s to understand sources of suspended sediment transport [16,17,18,19].

1.3. A Brief Review of Sediment Fingerprinting

Sediment fingerprinting is a method used to identify potential sediment sources in a catchment and allocate the amount of sediment contributed from each source through the use of natural tracer technologies with a combination of fieldwork, laboratory analyses of soils and sediments and statistical modeling techniques [3]. Sediment fingerprinting is essentially a two-step process: first, diagnostic physical and chemical properties (‘fingerprints’) that unambiguously differentiate between different soil groups (potential sources of sediment) in a catchment are selected, and secondly, these fingerprint properties are compared with those of suspended sediment samples taken from rivers (particle size < 62 µm) within the catchment. The approach is based on the assumption that some of the properties of sediment reflect those of the sources [15]. As long as soils consistently differ in some aspects within a catchment, the sediment fingerprinting approach can discriminate between different soil types and infer the relative contribution of each soil type to the suspended sediment load in rivers. Thereby, the areas contributing the largest amounts of suspended sediment can be prioritized for soil conservation and erosion control measures, which is an enormous undertaking, especially in a large, hilly catchment. Sediment fingerprinting thus offers a potentially valuable tool for understanding critical zone processes and improving watershed management in rivers that focusses on the assessment of suspended sediment sources to aid in developing efficient remediation strategies for soil-erosion-related problems.

Fingerprinting techniques began in the 1970s with the qualitative comparison of individual soil properties within a watershed, such as physical particle size, color, density, isotopic ratios, radiometric/mineralogic/chemical composition and organic properties (e.g., [16,17,18,19]). These single-component signatures have been successfully used to infer sediment origin from both spatial origin (based on lithological differences within the catchment) and source type (land cover/land use—agriculture, pasture, forest, etc.). However, such qualitative approaches were seen to have several possible limitations, as described in an early review by Collins et al. [15]. For instance, suspended sediment in a river may resemble a particular source in the catchment but could also result from a combination of several other sources in the catchment. Individual tracers can also be subject to physical and chemical changes from further weathering, color change and geochemical transformation during transport in streams and interaction with the environment. Koiter et al. [20] discuss this further. Walling et al. [19] showed that no single diagnostic sediment property could reliably distinguish between different sources. Hence, to overcome these problems of single-component signatures, the use of composite multiple signatures began so as to decrease ambiguity and improve accuracy of determination [15,17,19]. This was accompanied by the development of rigorous quantitative procedures that included the statistical verification of the ability of parameters to distinguish between potential sources, followed by the use of multivariate mixing models to determine the percentages of the various sources [15,19,21].

1.4. Selection of Tracers

Since the late 1990s, researchers from various disciplines have applied the sediment fingerprinting method to a wide range of watersheds globally [3], utilizing many different physical and biogeochemical tracers at a variety of landscape scales, from single-field-plot studies to large river basins, such as the 650,000 km² Murray–Darling basin in Australia [22]. Most studies employ a wide variety of tracers to be able to accurately and unambiguously distinguish between sediment sources. It is important to note that no single type of natural tracer can be used globally to infer sediment sources in all watersheds [15]; the choice of tracers in sediment fingerprinting is site-specific. This is because tracer properties depend on watershed geology, soil type and land cover/land use [23]. Davis et al. [3] give a comprehensive review of the types of tracers that have been employed over the past two decades. Tracers can be grouped into physical and biogeochemical categories:

(i)

Physical tracers include density, particle size and color. Their advantage is that these are readily identifiable and easily measurable in the field. However, physical tracers can be nonconservative; i.e., their properties can change during transport from their source to the river and further instream. For instance, color can change depending upon moisture content, particle size breakdown and subsequent chemical reactions with other natural elements in the catchment. Similarly, particle size can change due to aggregation and disaggregation during transport.

(ii)

Biogeochemical tracers include organic tracers, inorganic tracers and radionuclides. The availability of analytical laboratory techniques such as atomic and mass spectrometry has enabled studies to obtain the elemental or spectroscopic composition of soils and sediments, and thereby use a whole group of tracers if these have unique values for different soils in the catchment. Organic tracers include total organic carbon (TOC), total organic nitrogen (TON), total organic phosphorus (TOP), C:N ratio and stable isotopes of carbon (δC13) and nitrogen (δN15). Organic tracers are useful for distinguishing between different categories of land use, as they are affected by both vegetation type and soil exposure (tilling vs. no tilling) activity. However, organic tracers are not conservative; i.e., they can undergo biological transformations to other forms, as well as be taken up by plants (for N and P). Therefore, they are not suitable as catchment-scale tracers. Radionuclides (lead-210 and cesium-137) are present in soil from atmospheric fallout (either natural processes or nuclear weapons testing), and their concentrations can vary with soil depth; thus, they are used to infer the depth from which soil erosion might be happening. Inorganic tracers form a large group (e.g., Ag, Al, As, Ca, Ce, Co, Cr, Cu, Fe, K, La, Mg, Mn, Na, Ni, Pb, S, Si, Sr, Ti, Y, Zn, total inorganic carbon, total inorganic nitrogen, total inorganic phosphorus). Unlike organic and radionuclide tracers, which discriminate sources based on soil organic matter cycling and soil depth, respectively, inorganic tracers are less associated with specific environmental processes because of the large number of processes that affect elemental composition of sediments. Furthermore, studies have focused on multivariate methods to handle these large numbers of inorganic tracers, treating them more as a way to distinguish between sources and focusing less on emphasizing the mechanisms or explanations of the differences in a particular element (e.g., Cr) between different soil types or sources.

This study applied a sediment fingerprinting technique, using inorganic tracers, to investigate the primary sources of erosion occurring within an extensive, rugged catchment that has undergone rapid land use change since European colonial times. The aim was to identify the areas that are contributing the highest amounts of sediment to the Nyabarongo River, the main headwater tributary of the Nile in western Rwanda (Figure 1) that has been witnessing especially heavy sedimentation over the past two decades. Of particular importance was identifying the sources of sediment entering the NNYU Hydroelectric Reservoir and leaving the NNYU catchment.

2. Study Site and Methods

The process of identifying the major contributors of suspended sediments to the rivers in the Nile Nyabarongo Upper Catchment (NNYU) (Figure 1 and Figure 2) involved the sediment fingerprinting of the watersheds. This was followed by the prioritization of areas for action based upon geological type or formation. Here, geological type or formation refers to rock formations with different geological histories (Table 1). This process involved five basic steps, as follows:

• The collection of soil samples from areas of all geological types present in the catchment.
• The collection of suspended sediment samples from the river system.
• The laboratory analysis of soil and suspended sediment samples.
• The statistical analysis of laboratory results.
• The identification of potential hotspots for prioritization of rehabilitation.

2.1. The Nile Nyabarongo Upper Catchment (NNYU)

The NNYU (Figure 1 and Figure 2) catchment in western Rwanda drains the some of the furthest headwater streams of the Nile River and has a surface area of 3348 km², an average estimated annual surface water runoff of 385 mm yr⁻¹ (1289 MCM yr⁻¹) and an average annual rainfall of 1365 mm. Being part of the Albertine Rift zone, the terrain is very rugged, with slopes of 30–60% and a preponderance of geological upthrust metamorphic formations (Table 1, Figure 3) along with some igneous rock zones and alluvial soils in the valley bottoms. The land cover in the NNYU is primarily rain-fed agriculture (banana, maize, beans, tubers, corn and vegetables), pasture and agroforestry (dominated by Eucalyptus), along with mining activities and urban zones [11]. Native forests, including Nyungwe National Park (Figure 1), cover only about 7% of the catchment area [11]. Streams flowing east from the primary forest-covered mountains on the western border of the NNYU flow into the Nile Basin, while streams flowing west of this continental divide flow into the Congo Basin (Figure 1). East-flowing streams arising in Nyungwe National Park are typically clear and flow into the Rukarara River, which joins the Mwogo River, which flows through extensive Papyrus (Cyperaceae) wetlands before their confluence (Figure 1 and Figure 2). Further downstream, the Mwogo meets the Mbirurume River and forms the Nyabarongo River, which then flows northwards and is joined by the Kiryango, Nyagako and Secoko tributaries (and numerous smaller streams) before flowing into the Nyabarongo Hydropower Project Reservoir. After the reservoir, the Nyabarongo River continues northwards, is met by the Satinsyi River and leaves the NNYU, turning sharply southeast towards the densely populated center of Rwanda to join the Akagera River, bordering Tanzania, that then flows on towards Lake Victoria to form the Nile.

Figure 2. A map of the NNYU. The sub-catchments are delineated from suspended sediment sampling stations (orange circles) in the Nyabarongo River and its tributaries, and are labeled here by tributary or cumulative drainage area name. Terrestrial soil sampling locations are indicated by black dots and black squares, representing two different sampling campaigns. The blue arrows indicate the general northward direction of river flow. This map was adapted from technical report [24].

Figure 2. A map of the NNYU. The sub-catchments are delineated from suspended sediment sampling stations (orange circles) in the Nyabarongo River and its tributaries, and are labeled here by tributary or cumulative drainage area name. Terrestrial soil sampling locations are indicated by black dots and black squares, representing two different sampling campaigns. The blue arrows indicate the general northward direction of river flow. This map was adapted from technical report [24].

Table 1. Geological types or formations in Nile Nyabarongo Upper Catchment. Table adapted from technical report [24].

Geological Type	Description
Bb/Ng	Bumbogo formation. Alternate sandstone or quartzite and schist to phylite with fine sandstone dominance in small benches. Nyabugogo formation. Benches of fine quartzite to very coarse and alternation of fine sandstone beds.
Bu	Butare complex. Alternation of granites, gneissic granites, quartzitic metasediments, micaschist and amphibolites.
Gd	Granitoides diverse (miscellaneous granitoides): at Nyabisindu/Kigali, generally foliated. Presence of metasedimentary enclaves, sometimes mylonites.
Gi	Granites indiferencies—all terrain with granitic lithology in the Butare complex.
Gt	Gatwaro superformation (700–900 m). Mountain peak (Nyabidahe formation) with finely laminated dark gray light quartzophylite. Sandy base (Gatwaro formation) essentially clear quartzite of motley various origins.
Gdm	Granite 2 Micas: Porphyric (Mutara, Bukora) or equigranular (Masango) homogenous massifs, essentially intrusive.
Ho	Holocene and Pleistocene undifferentiated. Alluvial or valley sediment, middle and lower terraces, cones of dejection.
Ka	Kaduha formation. Areno-peltic (sandstone-shale): centrimetric to decentimetric alternation of light gray quartzite and dark gray to black quartzophylite to phylite. Mountain peak with decametric to metric levels of yellow quartzophylite.
Nw	Nyungwe formation (200–400 m) Areno-peltic: Centimetric to metric sequential alternation of red quartzite and quartzophylite to black phylite—levels of probable volcanic origins.
Nz	Ndiza formation (120 m): Benches of quartzite and fine to medium sandstone, regular alternation of sandstone and schist. Shyorongi formation (200 m). Homogenous schist and regular alternation of fine sandstone and schist beds.
Q1–Q3	Isolated outcrops, granitoid, basic rocks and pegmatite.
Sk	Sakinyaga formation (350–700 m) with a sandy dominant character. Greenish to whitish quartzite, usually laminated and finely laminated dark gray, light quartzophylites. Gravel dominated towards the mountain peaks.
St	Satinsyi complex: Phylite (biotite, chloritoid, garnet), chloritoschist, quartzite, probable metavolcanic and numerous basic intrusions (amphobilized).
Uw/Cr	Uwinka formation (500 m). Dominance of peltic geology (metamorphosis of mudstones/shale): bicolor banded quartzophylite, generally black/light gray to black/red. Cyurugeyo superformation (1110–1500 m). Pelitic mountain peak (Kibuye formation) finely laminated with gray quartzophylite.

Table 1. Geological types or formations in Nile Nyabarongo Upper Catchment. Table adapted from technical report [24].

Geological Type	Description
Bb/Ng	Bumbogo formation. Alternate sandstone or quartzite and schist to phylite with fine sandstone dominance in small benches. Nyabugogo formation. Benches of fine quartzite to very coarse and alternation of fine sandstone beds.
Bu	Butare complex. Alternation of granites, gneissic granites, quartzitic metasediments, micaschist and amphibolites.
Gd	Granitoides diverse (miscellaneous granitoides): at Nyabisindu/Kigali, generally foliated. Presence of metasedimentary enclaves, sometimes mylonites.
Gi	Granites indiferencies—all terrain with granitic lithology in the Butare complex.
Gt	Gatwaro superformation (700–900 m). Mountain peak (Nyabidahe formation) with finely laminated dark gray light quartzophylite. Sandy base (Gatwaro formation) essentially clear quartzite of motley various origins.
Gdm	Granite 2 Micas: Porphyric (Mutara, Bukora) or equigranular (Masango) homogenous massifs, essentially intrusive.
Ho	Holocene and Pleistocene undifferentiated. Alluvial or valley sediment, middle and lower terraces, cones of dejection.
Ka	Kaduha formation. Areno-peltic (sandstone-shale): centrimetric to decentimetric alternation of light gray quartzite and dark gray to black quartzophylite to phylite. Mountain peak with decametric to metric levels of yellow quartzophylite.
Nw	Nyungwe formation (200–400 m) Areno-peltic: Centimetric to metric sequential alternation of red quartzite and quartzophylite to black phylite—levels of probable volcanic origins.
Nz	Ndiza formation (120 m): Benches of quartzite and fine to medium sandstone, regular alternation of sandstone and schist. Shyorongi formation (200 m). Homogenous schist and regular alternation of fine sandstone and schist beds.
Q1–Q3	Isolated outcrops, granitoid, basic rocks and pegmatite.
Sk	Sakinyaga formation (350–700 m) with a sandy dominant character. Greenish to whitish quartzite, usually laminated and finely laminated dark gray, light quartzophylites. Gravel dominated towards the mountain peaks.
St	Satinsyi complex: Phylite (biotite, chloritoid, garnet), chloritoschist, quartzite, probable metavolcanic and numerous basic intrusions (amphobilized).
Uw/Cr	Uwinka formation (500 m). Dominance of peltic geology (metamorphosis of mudstones/shale): bicolor banded quartzophylite, generally black/light gray to black/red. Cyurugeyo superformation (1110–1500 m). Pelitic mountain peak (Kibuye formation) finely laminated with gray quartzophylite.

2.2. Sediment Fingerprinting

2.2.1. Soil and Suspended Sediment Sampling Approach

The purpose of sediment fingerprinting was to identify the major sources of suspended sediments in the catchments. Soil sampling sites were chosen on the basis of prior knowledge of erosion-prone areas (provided by the Rwanda Ministry of Natural Resources and local village communities), satellite maps and geographic information system (GIS) analysis combining topography, river drainage, geological types and land cover/land use. Examining the hydrologic behavior of the watershed in conjunction with land cover/land use enabled us to anticipate the erosional processes and the various possible sources of sediment, such as topsoil (farms, pasture, deforestation), gullies, unpaved roads, streambanks and re-mobilized riverbed sediment. These considerations then determined the number and locations of the soil sampling sites. While increasing both the amount and frequency of sampling is desirable, logistic issues of difficulty in accessing the rugged, roadless terrain, together with the financial cost of soil analysis, limited the sample number.

2.2.2. Soil Sampling

The NNYU catchment has fourteen geological types or formations (Figure 3, Table 1). Its land use is primarily small-scale agriculture. Five samples per geological type were taken as far apart as possible in each sub-catchments (Figure 2), with the exception of the geological type ‘Ho’, which was not collected. ‘Ho’ (from the Holocene era) represents alluvial deposits along river floodplains (Figure 3) that could have originated from various areas upstream in the catchment and thus was not useful for pinpointing sediment sources. A total of 73 soil samples were taken—65 covering the remaining thirteen geological types and some extra samples. Each sample was a composite of soil taken from five locations within a 50 m radius, collected in a clear polythene Ziploc bag. The purpose of this was to avoid the possibility of the nugget effect, i.e., a single soil sampling site being contaminated by animal dung/urine or other pollutants. Soil was scooped from the top 2–3 cm as usually the topsoil erodes naturally, except in cases of excavation for road building, mining or plowed fields. A plastic trowel was used to avoid possible contamination with elements from metal shovels. The trowel was first rinsed with bottled drinking water, gently scrubbed and then rinsed with deionized water to avoid contamination of the next sample. Photographs, global positioning system (GPS) locations and field notes were taken at each site for each sampling campaign. Sample bags were labeled with the date, geologic type and GPS location, subsequently air-dried in the laboratory at the Rwanda Integrated Water Security Program (RIWSP) headquarters in Kigali and shipped to the Trace Evidence Analysis Facility (TEAF) at Florida International University (FIU) for elemental analysis. Soil samples were collected in two separate campaigns in the dry season, a month apart (Figure 2). The second campaign included additional sites based on information received later regarding erosion hotspots.

2.2.3. Suspended Sediment Sampling

To select appropriate suspended sediment sampling sites, the drainage network of the NNYU catchment was examined in context of the river continuum concept [25] in order to predict sediment transport from headwater streams and tributaries to the downstream outlet of the entire NNYU catchment past Satinsyi (northern region of NNYU). Suspended sediment composition in a river system can change both spatially and temporally. As a river flows, some sediment settles out on slow-flowing zones, such as river bends or flow obstructions, while new sediment comes in through tributaries downstream. The sediment loading of rivers also depends on the season (rainfall), as well as anthropogenic activities in the watershed. Accordingly, locations were selected on each major tributary at the closest accessible site upstream of the confluence with the Nyabarongo River. Locations were also selected on the Nyabarongo after confluence with major tributaries, as well as directly before the Nyabarongo Hydroelectric Project Reservoir to infer the sources and composition of the sediments entering the reservoir. Furthermore, a sampling location was also selected just downstream of the reservoir to understand sedimentation in the reservoir. The last location was selected at the outlet of the NNYU, for a total of 14 locations throughout the NNYU catchment (Figure 2). Suspended sediment samples for fingerprinting analysis were collected using a previously published method from East Africa [26].

Five suspended sediment sampling campaigns were held at these fourteen locations from January 2016 to April 2016, covering both the short dry season and the main rainy season that exerts a dominant influence on soil runoff. Therefore, a total of 70 suspended sediment samples were taken over the study. During each campaign, one water sample was collected in a 1 L Nalgene bottle from a well-mixed section of the river, typically mid-channel, to avoid having a disproportionate amount of runoff from adjacent banks. This was achieved by using a telescopic extensible water sampler, or from a bridge if present at a site. A 250 mL volume of water was then filtered through a pre-cleaned and pre-weighed cellulose nitrate filter paper (pore size 0.45 μm, diameter 47 mm, Whatman, GE Healthcare Life Sciences, Pittsburgh, PA, USA). A filter apparatus operated with a manual handpump was used to create suction to draw water through the filter paper to facilitate the filtration process. Forceps were used to handle the filter paper by the edges. The filtering equipment and measuring cylinder were rinsed thrice with deionized water in between successive samples to avoid contamination from one sample to another. The filter paper with suspended sediment was allowed to air dry and then stored inside a sealed Petri dish in the RIWSP laboratory and subsequently shipped to the TEAF lab for elemental analysis.

Sediment-laden filters were then dried at 60 °C until reaching constant mass and then weighed. The initial filter weight was then subtracted from the total weight of the sediment-laden filter to calculate the total suspended sediment on the filter. Sediment concentrations in river water were calculated from the mass of sediment on the filter from 250 mL of sampled river water.

2.2.4. Sub-Catchment Delineation from Suspended Sediment Sampling Points

In order to determine the watershed of origin for a suspended sediment, catchment delineation was carried out using ArcMap 10.2, with the sampling point as the outlet of that catchment (pour point analysis) [27]. This analysis was carried out for all 14 suspended sediment sampling points. Figure 2 shows the map of these catchments, which are referred to from now on as sub-catchments, indicating that these were part of the overall NNYU catchment. Note that these sub-catchments were delineated from the sampling point and are not strictly hydrological catchments of the tributaries. For example, the Rukarara sub-catchment as depicted on this map demarcates the catchment for the sampling point on the Rukarara and not the entire catchment for the Rukarara River, because the sampling point was taken a few hundred meters upstream of the confluence of the Rukarara and Mwogo Rivers.

Sub-catchments were analyzed in order flow direction, from headwater to downstream (the river flows in a south to north direction), as follows: the Rukarara River sub-catchment, Upper Mwogo River sub-catchment, Mwogo River sub-catchment (consisting of the catchment of the entire Mwogo—Upper, Middle and Lower—River before its confluence with the Mbirurume), Mbirurume River sub-catchment, Nyabarongo headwaters (including the Mbirurume and the Mwogo sub-catchments), Kiryango River sub-catchment, Nyagako River sub-catchment, Secoko River sub-catchment, Hydropower sub-catchment (including the Kiryango, Nyagako and Secoko sub-catchments and the sub-catchment Nyabarongo sub-catchment upstream of the Hydropower Reservoir), Nyabarongo downstream of reservoir sub-catchment (constituting the entire Nyabarongo catchment before the confluence with Satinsyi), Satinsyi river sub-catchment and finally the NNYU outlet site, draining the entire NNYU.

2.3. Laboratory Methodology

Soil and sediment samples were shipped to the TEAF lab for elemental analysis via laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) using a method previously developed for the preparation and analysis of soil and filter-bound sediments [28], summarized here. Soil samples were dried in an oven for 48 h at 60 °C and then sieved to 64 µm using disposable nylon meshes to minimize contamination. The fraction smaller than 64 µm was spiked with an internal standard and milled using a titanium-carbide ball mill (Retsch, Newtown, PA, USA). The filter paper and sediment were likewise milled with an internal standard, as it was difficult to separate all the suspended sediment from the filter paper [28]. Part of the resulting powder was pressed into pellets using stainless steel dies in a vacuum-assisted Carver pellet press (Carver, MA, USA). The internal standards used were indium and scandium single-element standard solutions (Ricca Chemical Company, Arlington, TX, USA). Each pellet was then subjected to laser ablation, and the resulting particles were swept into the ICP-MS apparatus with a He carrier gas for the detection of elemental concentrations. The ICP-MS equipment used in this study was a quadrupole ELAN DRC II (Perkin Elmer Sciex, Framingham, MA, USA) coupled to a 266 nm J200 laser ablation unit (Applied Spectra, West Sacramento, CA, USA). Four replicate analyses were performed on each pellet in spot mode using a 200 µm spot at 3 Hz and 30% of the maximum energy output for 75 s following a gas blank of 20 s. Reference material (RM) pellets of soil and sediment standards with certified elemental concentrations were prepared and analyzed in the same manner as the sample pellets [28].

The mass spectrometric data was then assessed for quality control and processed with GLITTER software 4.4 (GEMOC, Macquarie University, Sydney, Australia) [28]. This software integrates the ablation signal, subtracts the gas blank signal and normalizes the data using scandium as the internal standard. It also calculates the concentration (µg g⁻¹) of the samples based on the certified concentration values in the database of RMs used as calibrators. Minimum detection limits (MDLs) for each element were calculated at the 99% confidence level for each replicate. More details on the process are included in [28,29]. The following elements passed the TEAF laboratory analytical quality control checks and were evaluated for their use as tracers: Li, Mg, Al, P, K, Ti, V, Cr, Mn, Fe, Cu, Zn, Ga, As, Rb, Mo, Sn, Ba, La, Ce, Nd, Sm, Eu, Dy, Th and U.

2.4. Statistical Analysis

The first part of the statistical analysis identified a set of elements that could reliably distinguish between the geological types (soil sources) in the NNYU catchment. The second part employed a mixing model that compared the elemental composition of a suspended sediment sample to those of the geological types in the watershed. The product of this step was a probabilistic distribution of each geological unit contribution within a particular sample.

The mixing model analysis was performed on each sample per collection campaign. This analysis was also performed on each sub-catchment, using all the suspended sediment samples across sampling campaigns taken from that sub-catchment outlet in the mixing model at once. Because of soil erosion and contribution to sediment loads being caused by rainfall and human activities, and the localized and random nature of rainfall, together with certain activities that may happen at specific times only (road building, burning of fields, deforestation and excavation for mining), sources in a catchment usually vary over time. Therefore, statistical analysis was performed at the modeled sub-catchment level for individual sampling events (Analysis A) as well as for the pool of all samples (composite) from all sampling campaigns (Analysis B). Note that the composite was not the average of the 5 sampling campaigns; it was obtained by pooling together the analytical results of all the samples across the sampling events. All statistical analysis was performed in the statistical computing language R 3.4.3 [30].

A range test was first used to ensure that the downstream elemental values were possible given the potential sources within the sub-catchments. Li and Ga failed the range test and were excluded from consideration as tracers. All remaining elemental tracers were approximately normally distributed and appropriate for input into the MixSIAR mixing model. The Kruskal–Wallis H test was then used to identify elements that showed significant differences between the potential source types [31]. A stepwise discriminant function analysis (DFA) based on the minimization of Wilks’ Lambda was then used to determine the optimum combinations of elements capable of discriminating between source types [32,33]. In addition, a jackknifed DFA was used to assess the discriminatory power of the multi-element source signature through a leave-one-out cross-validation procedure [32]. This was performed by running the discriminant function analysis as many times as there were samples, leaving a different sample out each time and assessing the degree that the sources may be confused with other sources [34]. The optimum set of tracers for each sub-catchment varied between 3 elements (Nd, Sm and La) for the Secoko, with a cumulative error rate of 17%, and 21 elements (Al, Ba, Zn, V, As, Sn, Ce, Mn, Mg, Rb, K, Dy, Eu, Sm, Fe, La, Th, Cr, P, U, Nd) for the Lower Nyabarongo, with a cumulative error rate of 13% (Supplementary Table S1). The highest cumulative error rate, 42%, was found for the Upper Mwogo using six elements (Sn, Al, V, Ba, Cr and Fe). Prediction accuracy determined using leave-one-out jackknifed discriminant function analysis quantified the uncertainty in predicting each of the potential sources within each of the individual sub-catchments as between 0% (unable to accurate predict the Nw geologic type in the several of the sub-catchments) and 83% (high prediction ability for the Gi geologic type in the Satinsyi sub-catchment).

In order to ascertain the source proportions of a suspended sediment sample, a mixing model with Bayesian inference was utilized [35]. Specifically, the MixSIAR Mixing Model, which allows for all sources of uncertainty to be propagated through the model [35,36], has been successfully used in sediment fingerprint studies in East Africa [26,28]. The model is fit by a Markov Chain Monte Carlo routine.

Due to soil erosion and contribution to sediment loads being caused by rainfall and human activities, the localized and random nature of rainfall, together with certain activities that may happen at specific times only (road building, burning of fields, deforestation and excavation for mining), sources in a catchment usually vary over time. Two types of analyses were run. In Analysis A, one model was run per sediment collection campaign (January–May). Analysis was performed at the sub-catchment level on each sediment sample per collection campaign. In Analysis B, one model was run per sub-catchment, with all the sediment campaigns pooled into one. Analysis was performed on each sub-catchment using all the suspended sediment samples input into the mixing model at once. Note that pooling was not the average of the five sampling campaigns; it was obtained by pooling together the analytical results of all the samples across the sampling events.

2.5. Prioritization of Focal Areas for Catchment Rehabilitation

The initial categorization of the level of importance or dominance of geological types as sources within a sub-catchment was based on the proportion of suspended sediment contributed. The categories were as follows:

Level 1: >40% of suspended sediment represented by that geological type.

Level 2: 20–40%.

Level 3: 10–20%.

Geological types contributing less than 10% were not assigned any priority level and were considered relatively insignificant. Determination of the erosional hotspots in a sub-catchment was arrived at by examining suspended sediment composition in the river at the sub-catchment outlet and then locating where the dominant geological types in the suspended sediment were located in the sub-catchment. When scaling up to the entire NNYU, sub-catchment-level analyses were conducted starting from the headwater sub-catchments and progressing downstream along the river in a cumulative manner, as described earlier in Section 2.2.4 and depicted in Figure 4.

Sediment transport processes (flow-dependent settling and resuspension, and the addition of further sediment from tributaries) can change sediment composition both spatially along the river as well as temporally (season, anthropogenic events). Therefore, it is possible that a sediment source that is a major contributor in an upstream catchment is no longer as dominant in river water downstream. To account for this dynamic change in suspended sediment composition as one goes downstream, a further prioritization strategy was taken as follows:

Level 1: Assigned to a geological type that retained its dominance in sediment contribution downstream (>40%), as seen from the sediment composition results at a downstream point on the river.

Level 2: Geological types that were Level 1 in a headwater catchment but decreased in contribution level downstream (20–40%).

Level 3: Geological types that were Level 2 in a headwater catchment and decreased to Level 3 or less (<20%).

This process was repeated for results from each downstream sampling point, with a cumulative increase in catchment area, until the entire NNYU was covered. These results were overlaid over an administrative map to identify cells, sectors and districts occurring in the identified areas of potentially high erosion. This map was instrumental in deciding field locations to visit in order to verify erosion, ascertain the reasons for erosion, examine the pathways of sediment transport from eroding areas to the rivers and thereby develop appropriate catchment rehabilitation approaches.

3. Results

3.1. Sediment Sources in Each Sub-Catchment

The mixing model yielded the proportions of sediment arising from each geological type present in each sub-catchment. Figure 5 depicts results for Analysis B (each sub-catchment, pooling all five sampling events), thereby representing an estimate of the situation over the short dry and main wet seasons. The dominant sources in each sub-catchment are summarized in

Table 2. Geological types that were the dominant sources of suspended sediments in ten sub-catchments of the NNYU over January–April 2016. Sources < 10% are not included here.

Sub-Catchment	Potential Sources over Entire Sampling Period
Rukarara	Gi (40%), Ka (25%), Q1Q3 (19%)
Upper Mwogo	Q1Q3 (63%)
Mbirurume	Ka (40%), Q1Q3 (28%)
Nyabarongo Headwaters	Nw (40%), Bu (21%), Ka (15%)
Kiryango	Gdm (70%)
Nyagako	Bb/Ng (60%), Gdm (23%)
Secoko	UwCr (80%)
Reservoir	Q1Q3 (43%), Gd (14%) and BbNg (12%)
Satinsyi	Bu (60%), UwCr (32%)
NNYU Outlet	BbNg (50%), Nz (30%)

. Results for ten out of the fourteen sub-catchments are shown here, because the sub-catchments of the Middle and Lower Mwogo, reservoir downstream and Nyabarongo downstream are included in the ten listed sub-catchments. The results from these four sub-catchments can be found in the Supplemental Materials. Figure 6 depicts results for Analysis A; sediment fingerprinting results for each sediment sampling campaign are presented for the NNYU outlet (i.e., the entire NNYU Catchment). The full set of results with figures for Analysis A for each sub-catchment can be referred to in the technical report for this project [24].

In some sub-catchments, a large proportion of sediment was found to be associated with just a few geological types. In other catchments, the sediment sources had contributions from almost all geological types. This is especially the situation in the downstream reaches of the Nyabarongo, which retain some amount of suspended sediment from far upstream. The furthest headwaters of the Nyabarongo arise in the primary rainforests of Nyungwe National Park (Figure 1); the clear streams (personal observation over duration of study) then flow halfway down the sub-catchment through tea plantations to form the Rukarara, which further flows through agroforestry-dominated valleys until its confluence with the Mwogo. Even though types Gi and Ka were the largest contributors (Table 2), all the other geological types were also represented in suspended sediment over the sampling period (Figure 5), indicating soil loss possibly occurring throughout the sub-catchment. In the Upper Mwogo sub-catchment, Q1Q3 was the dominant sediment source (63%) at the outlet of the sub-catchment and remained in the suspended sediment downstream of the confluence with the Rukarara that brings in Q1Q3 and Bu. Nw was also noticed in the Middle Mwogo sub-catchment (Supplementary Materials) before the confluence with the Mbirirume, carrying Ka (40%), Q1Q3 and Bu as the dominant sources. The sediment composition downstream of that confluence (named Nyabarongo headwaters) had Nw, Bu and Ka as the dominant geological units. The suspended sediment from Kiryango was dominated by Gdm (70%), Nyagako by Bb/Ng (60%) plus Gdm, and Secoko by Uw/Cr (80%). The Nyabarongo River then enters the Hydroelectric Reservoir, where considerable settling of suspended sediment occurs because of the decrease in flow velocity. Water leaving the Reservoir had Q1Q3, Gd and Bb/Ng as its major constituents. The Satinsyi flows in with Bu and Uw/Cr as its dominant types, just before the Nyabarongo turns sharply southeast (Figure 1) and out of the NNYU.

Figure 5. Geological types (sources) by proportion in suspended sediment samples, with the five sampling time points pooled together for each sub-catchment (Analysis B; plates (A–J)). Only sources present in the specific catchments are shown as source proportions in those respective catchments. The range of each sample in the box plot represents the 95% confidence interval, and the dot for each source represents the most likely value for that source (mean). See Table 1 for a description of each source (geological type) abbreviation. This figure was adapted from the technical report [24].

Figure 5. Geological types (sources) by proportion in suspended sediment samples, with the five sampling time points pooled together for each sub-catchment (Analysis B; plates (A–J)). Only sources present in the specific catchments are shown as source proportions in those respective catchments. The range of each sample in the box plot represents the 95% confidence interval, and the dot for each source represents the most likely value for that source (mean). See Table 1 for a description of each source (geological type) abbreviation. This figure was adapted from the technical report [24].

Figure 6. Sediment composition by proportion of geological type in the NNYU (at the NNYU outlet suspended sediment sampling point). Panels (A–E) represent sampling events on 21–25 January, 9–13 February, 7–11 March, 28 March–1 April and 21–24 April, 2016 respectively (Analysis A). The black dots represent the mean contribution from each potential source. The colored boxes represent the 95% confidence intervals. See Table 1 for a description of each source (geological type) abbreviation. This figure was adapted from the technical report [24].

Figure 6. Sediment composition by proportion of geological type in the NNYU (at the NNYU outlet suspended sediment sampling point). Panels (A–E) represent sampling events on 21–25 January, 9–13 February, 7–11 March, 28 March–1 April and 21–24 April, 2016 respectively (Analysis A). The black dots represent the mean contribution from each potential source. The colored boxes represent the 95% confidence intervals. See Table 1 for a description of each source (geological type) abbreviation. This figure was adapted from the technical report [24].

3.2. The Geological Type Origin of Suspended Sediment Leaving the Nile Nyabarongo Upper Catchment (NNYU)

At the point where the Nyabarongo River leaves the NNYU, the geological types Bb/Ng and Nz accounted for ranges of 25–51% and 10–44% of the suspended sediment composition, respectively, across the study period. About 20 km upstream, Bb/Ng was found as major contributor in the Nyagako and Nyabarongo Hydroelectric Reservoir sub-catchments, while Nz was found to be a major contributor in the Nyabarongo downstream sub-catchment (Supplementary Material). The five plots in Figure 6 indicate the geological sources of sediment at each sampling event at the NNYU outlet over January–April. The last plot (J) in Figure 5 depicts the results of pooling all five sampling rounds (Analysis B) to create an overall picture of the suspended sediment trends over the entire sampling period.

3.3. Suspended Sediment Concentration Differences Across NNYU

The Secoko River had the highest suspended sediment concentration in every sampling campaign (Figure 7 and Figure 8, Supplementary Table S1), followed by the Satinsyi River. These results corroborate our personal observations and photographs of highly turbid waters flowing between heavily sedimented riverbanks in these two rivers in all five sampling visits [24]. The reservoir’s suspended sediment concentration values are relatively lower, possibly because the water is stationary, allowing the settling down of sediment particles. Figure 8 illustrates the dynamic nature of suspended sediment concentrations in the NNYU river system. While the Secoko and Satinsyi had the highest concentrations across the entire study period, as shown by their high values, other tributaries and stations along the main Nyabarongo River saw considerable variation in their suspended sediment concentrations from the dry to wet season. It should be noted that these are suspended sediment concentrations in milligrams of sediment per liter of river water, and do not signify the sediment loads in the rivers. For that, river discharge data are necessary.

3.4. Identifying Priority Rehabilitation Focal Areas in NNYU

The prioritization analysis was carried out starting from the headwater sub-catchments—Rukarara, Upper Mwogo, Mwogo and Mbirurume—and then following sediment composition progressively downstream to the outlet, with the catchment increasing in size (Figure 4). A map of the NNYU catchment sediment source status (Figure 9) shows the three levels of intervention priority—1 (red), 2 (yellow) and 3 (green)—with Level 1 indicating the areas that are potentially contributing the highest amounts of sediment. Level 1 areas were identified to occur in the Kabuga cell in the Mwogo sub-catchment, Nganzo and Nyamirama cells in the Nyagako sub-catchment and Kanyana cell in the NNYU downstream sub-catchment. In interpreting such a map, however, it should be kept in mind that sediment arising from distant headwater areas can settle out. For example, in Figure 9, much of the Rukarara and Mwogo catchments are at Level 3, implying that these furthest headwater areas contribute relatively less sediment to water sampled at the NNYU outlet as compared to areas at Levels 1 and 2. This does not mean, however, that the Rukarara and Mwogo sub-catchments have low levels of erosion—it just indicates that, at the NNYU outlet, the amount of sediment from these headwater catchments is present in relatively low amounts as compared to sediment from closer to the outlet.

4. Discussion

This pioneering study in Rwanda used sediment fingerprinting to determine the most probable soil erosion hotspots in the rugged and extensive catchments of the Nile Nyabarongo River. The results can help to focus limited resources in a timely manner to locate sites with the heaviest erosion, which is otherwise an overwhelming endeavor in this region, much of which is inaccessible by road. The results of the study also enable further research into soil erosion–land cover change linkages and sediment transport processes by pinpointing the actual areas that are likely contributing sediment. An improved understanding of critical zone processes can then be incorporated into the active management of watersheds and restoration of ecosystems that intrinsically aid in trapping sediment and regulating hydrology, such as gallery forests and natural wetlands in floodplains.

4.1. Interpreting the Results—Inferring the Sources and Causes of Soil Erosion

As mentioned in our discussion of the methods, the examination of suspended sediment composition results was carried out from the river continuum perspective [25]. The analysis started from the headwater sub-catchments and followed the river downstream, noting how the suspended sediment composition changed as tributaries joined the main river further downstream, the influence of settling in the Hydropower Reservoir and finally the composition of the sediments as the river exited the NNYU on its journey eastwards into central Rwanda.

The furthest headwater tributary, the Rukarara River, arises in primary forest-clad mountains in Nyungwe National Park (Figure 1) and can be seen to flow with clear water halfway through the sub-catchment past tea plantations until it reaches agricultural areas. By the time it joined the Mwogo River, the sediment concentration had increased and was similar to other tributaries in the Upper Nyabarongo system, thereby indicating significant soil erosion in the lower part of the Rukarara sub-catchment. The composition of geological types in suspended sediment is preserved as the river flows downstream, with the largest contributors from Mwogo, Rukarara and Mbirurume—Q1/Q3, Gi and Ka—still represented as large sources even after these tributaries combine to form the Nyabarongo (Nyabarongo headwaters). The finding that almost all the geological types in each sub-catchment were represented in the suspended sediment indicates that soil from a wide area across these sub-catchments is entering the river system. This suggests diffuse nonpoint pollution, characteristic of agriculture that is widespread over the catchment, where every year some soil is washed off. Despite widespread soil conservation measures in existence such as terraces, dykes and furrows, soil loss from the steep hillsides is still significant when compared to areas with montane evergreen forests (the original land cover), whose thick multi-layered canopy shields the ground from the erosive force of the intense rain events characteristic of the tropics.

Further downstream in the lower Nyabarongo sub-catchments such as Kiryango, Nyagako, Secoko and Satinsyi, one or two geological types dominate the suspended sediment composition (>60%). The dominance of one or two geological types suggests localized intense soil disturbances in areas belonging to that geological type, releasing a heavy amount of soil from activities such as mining and road building. The region has numerous mining operations for tin, cobalt and tungsten—raw materials used in electronic components. It is also possible that subsistence slash-and-burn agriculture without effective soil conservation [5] is prevalent in areas belonging to these geological types. It was also noticed that dominant geological types arising from one sub-catchment, such as in Kiryango (Gdm), Secoko (Uw/Cr) or Satinsyi (Uw/Cr), are not seen downstream in any significant proportion. The possible reasons for this are that such sediments may settle out on river bends, deposit on the river bed/shore, be intercepted by emergent bank vegetation or decrease in relative proportion through additional sediments entering through tributaries and direct runoff downstream. This study clearly indicates the dynamic nature of suspended sediment composition both in terms of space (Figure 5) and time (Figure 8) along the river continuum. Still, certain geological types appear to persist over long distances, such as Bb/Ng, which interestingly occupies only a small area in the NNYU that has considerable mining activity. Further studies can investigate why certain types remain prominent and others decrease in proportion.

While the proportion of a geological type may decrease relative to other types, the amount of sediment arising from that source may still remain the same if the overall sediment concentration increases—which can either result from increased soil wasting or reduced river flow. Most of the sampling stations saw large differences in concentration between sampling events (Figure 7, Supplementary Table S1). While this could be attributed to a number of factors, such as rainfall events (seasonality) or sporadic human activities such as soil excavation that could lead to localized soil runoff events, changes in sediment concentration could also arise because of changes in flow volume. Hence, river discharge data is necessary to further interpret sediment concentration changes across time, as well as to enable sediment transport modeling across the catchment.

The settling of sediments in slow-flowing river zones, such as in the reservoir and in wetlands, bears further study, albeit with more frequent sampling across space and time. For instance, there are extensive natural wetlands that are dominated by Papyrus sedges in the Upper Mwogo sub-catchment. The sediment composition at the outlet of the Upper Mwogo sub-catchment can differ from the river sediment composition upstream of these wetlands, thereby pointing again to the dynamic spatiotemporal nature of sediment composition. If the objective was to detect sources of erosion within the Upper Mwogo sub-catchment, it would be important to sample upstream of these large wetlands as well. In the present study this was not performed, as the main objectives were to identify the sources of sediment entering the NNYU Hydroelectric Project Reservoir and also leaving the NNYU. In addition, the study scope allowed a limited number of sampling points due to logistical difficulties in accessing sites in the roadless catchments. Increasing the number of sampling points within each sub-catchment (instead of just the sub-catchment outlet as performed in the present study) would yield more spatially detailed information on the dominant erosion sources within a sub-catchment.

4.2. Applying the Results: Pinpointing Priority Rehabilitation Focal Areas in NNYU

The Nyabarongo River is the lifeline of western and central Rwanda, including the most populated parts of the country around the capital, Kigali. It is also seen to have high hydropower potential, important for a country with no other major energy sources. However, the rapid ongoing siltation of hydropower reservoirs imperils the sustainability of hydropower generation, hence the Government of Rwanda’s impetus to locate erosion sources and understand sediment transport in this very complex and mostly inaccessible region.

This study’s results identify the geological types in each sub-catchment that contributed the highest levels of sediment over the sampling period. Locating these geological types on an administrative map (Figure 8) then indicates the cells (and their sectors and districts) that are the areas likely subject to the highest levels of erosion. These areas can then be visited to verify erosion, ascertain the reasons of erosion and thereby develop catchment rehabilitation strategies and programs to control erosion and sediment runoff. It is important to note that areas other than those identified as Levels 1–3 also contribute sediment, on account of the loss of the original native forest. However, areas at Levels 1–3 contribute anywhere between 50 and 80% of the sediment. Such maps greatly help to focus limited financial and manpower resources on rehabilitating areas with the gravest levels of soil erosion and sediment generation.

It should be noted that the source proportion results pertain to the sediment sampled at a certain location along a river, and that the sediment composition is likely to change upstream and/or downstream. For instance, Figure 8 shows much of the Rukarara and Mwogo catchments to be at Level 3, implying that these furthest headwater areas contribute relatively less sediment to the water sampled at the NNYU outlet, as compared to areas with Levels 1 and 2. This does not mean that the Rukarara and Mwogo sub-catchments have low levels of erosion—it just indicates that, at the NNYU outlet, the amount of sediment from these headwater catchments is present in relatively low amounts as compared to sediment from closer to the outlet. Hence, choosing a sediment sampling location depends on the identification of erosion-prone areas in the watershed as well as the river reaches affected by sedimentation.

4.3. Sediment Fingerprinting—A Monitoring Tool for Erosion Management

Land cover change and accelerating human activities have and will continue to result in soil erosion to varying degrees in various places. Hence, the periodic monitoring of watersheds for sediment sources enables keeping a track of actively eroding areas and is a prerequisite of ameliorative measures for natural resource management and ecosystem conservation. In the tropics, mountainous areas with forests as the natural form of vegetation (such as much of the NNYU) are highly vulnerable to soil erosion upon deforestation. Plantation forestry, while being a useful socioeconomic activity, still does not possess a dense multi-layered canopy to break the impact of rainfall upon the soil to the extent as under native primary forest. Hence, restoration and protection of native forest cover in the steepest slopes are important. Because the thin soils and steep slopes pose challenges for forest recovery, active restoration approaches are needed. The wide prevalence of agriculture in the NNYU implies that croplands will continue to be significant sources of sediment, as has been seen by a study in southern Brazil with similar catchment dynamics [12].

Reducing sediment loss also requires the continuation of existing soil conservation efforts such as agroforestry, terracing, mulching and contour trenching that also have the benefit of boosting long-term farm productivity, as seen in a Rwandan study [5]. Road building and mining pose special challenges, as these actively excavate and destabilize large areas of soil; hence, such activities need remediation programs that prevent the mass wasting of tailings and soil piles. Given the logistical challenges of undertaking these activities at the enormous spatial scale of the NNYU catchment, sediment fingerprinting can indicate the potential hotspots of erosion and thereby narrow down the critical targets for catchment rehabilitation in an effective and tiered manner. A similar prioritization exercise can be carried out for the Nyabarongo Hydroelectric Project Reservoir to discern the major sources of the sediment entering the reservoir. Once these sources are attended to by means of effective rehabilitation measures, a fresh set of suspended sediment samples can then indicate other sources of sediment via the sediment fingerprinting process. As erosion and sediment runoff processes are dynamic and changeable in nature, sediment fingerprinting can be an integral part of overall long-term monitoring to improve watershed management.

5. Conclusions

Sediment fingerprinting was utilized to identify erosional hotspots and sediment transport processes in the Nile Nyabarongo Upper Catchment in western Rwanda. The cells (areas) contributing the most suspended sediment were found to be Kabuga, Nganzo, Nyamirama and Kanyana. The dominance of a few geological types in suspended sediment composition at a sub-catchment outlet suggests localized disturbances such as open-pit mining or roadbuilding from areas with those geological types. On the other hand, if most geological types present in a sub-catchment are present in suspended sediment, this can indicate erosion from most of the sub-catchment, characteristic of diffuse human activities such as agriculture. This technique thus enables government agencies to visit and prioritize rehabilitation locations—a task that would otherwise be extremely arduous in extensive and rugged terrain undergoing swift land use change accompanied by extremely serious problems of erosion and sedimentation. Because sources of erosion change in time and space, sediment fingerprinting can be a valuable addition to an erosion and sediment transport monitoring program. This information then underpins a detailed understanding of critical zone soil–land cover–climate processes, necessary for developing region-specific watershed management and policy guidelines.