Soil erosion and sediment transport in Tanzania: Part I – sediment source tracing in three neighbouring river catchments

Water bodies in Tanzania are experiencing increased siltation, which is threatening water quality, ecosystem health, and livelihood security in the region. This phenomenon is caused by increasing rates of upstream soil erosion and downstream sediment transport. However, a lack of knowledge on the contributions from different catchment zones, land‐use types, and dominant erosion processes, to the transported sediment is undermining the mitigation of soil degradation at the source of the problem. In this context, complementary sediment source tracing techniques were applied in three Tanzanian river systems to further the understanding of the complex dynamics of soil erosion and sediment transport in the region. Analysis of the geochemical and biochemical fingerprints revealed a highly complex and variable soil system that could be grouped in distinct classes. These soil classes were unmixed against riverine sediment fingerprints using the Bayesian MixSIAR model, yielding proportionate source contributions for each catchment. This sediment source tracing indicated that hillslope erosion on the open rangelands and maize croplands in the mid‐zone contributed over 75% of the transported sediment load in all three river systems during the sampling time‐period. By integrating geochemical and biochemical fingerprints in sediment source tracing techniques, this study demonstrated links between land use, soil erosion and downstream sediment transport in Tanzania. This evidence can guide land managers in designing targeted interventions that safeguard both soil health and water quality.


| INTRODUCTION
River catchments in Tanzania have some of the highest sediment yields of sub-Saharan Africa, linked in part to a distinct topography and the semi-arid climate (Vanmaercke et al., 2014), but also to the effects of increasing land-use pressures (Borrelli et al., 2017;Wynants et al., 2019). The loss of permanent vegetation through deforestation, agricultural expansion and overgrazing is driving accelerating rates of erosion, which is causing a rapid depletion of soil resources, threatening food, water and livelihood security in the region (Fenta et al., 2020;Maitima et al., 2009). Furthermore, these processes are potentially amplified by natural rainfall variations (Ngecu & Mathu, 1999;Wynants et al., 2020) and projected increases in extreme climatic events . While soil resources are progressively being depleted, the Tanzanian population and their demand for the services the soil provides is increasing (FAO, 2019;UNDESA, 2017). Continued loss of productivity and arable land would be catastrophic for the agricultural sector in Tanzania, which currently employs about 75% of the working population, underpins the economy, and provides the basic caloric uptake for the majority of its inhabitants (FAO, 2019;Salami et al., 2010;Sanchez, 2002;Tengberg & Stocking, 1997). Besides these on-site impacts, increased downstream sediment transport also has major detrimental effects on aquatic ecosystems, water quality and energy security (Amasi et al., 2021;Dutton et al., 2019;Olago & Odada, 2007). There is an urgent need for science-based land-and water-management strategies in Tanzania to achieve sustainable intensification of agro-pastoral production and protect soil resources. However, a lacuna in environmental data and a lack of understanding on the complex dynamics of increased soil erosion and sediment transport in semi-arid East Africa impedes the development and application of sustainable land-and water-management plans (Blake et al., 2018b;Kelly et al., 2020).
In this context, sediment source tracing techniques are valuable tools for filling knowledge gaps and elucidating processes of soil erosion and sediment transport Walling, 2013). Soils and ecosystems co-evolve through a mutual interdependence on the balance between soil erosion and soil production through weathering (Lowdermilk, 1953). Differences in geology, climate, ecosystem structure, land use, and pedogenetic processes, give the resulting soils a characteristic geochemical and biochemical composition. Geochemical and biochemical fingerprinting can be used for grouping potential sources into different catchment zones, land-use types and soil depths (Gibbs, 2008;Motha et al., 2002;Reiffarth et al., 2016). Surface and subsurface erosion processes can detach soil and regolith particles, which are transported downstream from different catchment areas by hydrological processes as a mixture of sediment particles (Fryirs, 2013;Hoffmann, 2015;Kitch et al., 2019). During detachment, transport and deposition, sediment particles are, however, subject to sorting effects depending on their particle size (Laceby et al., 2017).
Moreover, during these processes, tracers are also potentially subject to chemical alterations (Belmont et al., 2014). The biochemical and geochemical composition of riverine sediments thus depends on the relative contributions of different sources, their physical and chemical properties, and the transport dynamics in the river system (Haddadchi et al., 2013;Walling, 2013;Walling & Woodward, 1995). Integrating multivariate source and mixture fingerprints within Bayesian mixing models (BMMs) allows a proportional attribution of soil sources to downstream sediment (Blake et al., 2018a;Collins et al., 2010Collins et al., , 2017Cooper et al., 2015). With respect to good practice, sediment source tracing can be a powerful tool for investigating the contributions of different catchment zones, erosion processes and land-use types to the sediment (Alewell et al., 2016;Blake et al., 2012;Owens et al., 2016).
By using complementary sediment source tracing techniques, this study aims to assess the dominant sources of transported sediment in Tanzanian river systems distinguishing between catchment zones, dominant land-use types and erosional processes. This is done by fingerprinting potential source material and river sediment in three neighbouring Tanzanian catchments using elemental geochemistry and the δ 13 C signature of plant-derived, long-chain (> C22), saturated fatty acids (FAs). The importance of land use, catchment zone, and erosion processes to the sediment fluxes is quantified using BMMs.
Estimations of source zone contributions to riverine sediment are particularly valuable for designing targeted management interventions to maintain both soil health and water quality. This article is the first part of an article pair, wherein the second article studies the changes in soil erosion and sediment transport in the region over the past 120 years .

| MATERIAL AND METHODS
The raw dataset, model inputs, model build, and model outputs are available as open access at https://doi.org/10.24382/9xmf-7e88.

| Study area
Sedimentation rates in Lake Manyara have increased significantly over the past 120 years , which threatens the ecosystem health and services of this National Park and UNESCO Man and Biosphere Reserve (Janssens de Bisthoven et al., 2020). Previous research of Wynants et al. (2020) attributed the observed increase in sedimentation in Lake Manyara in the past decades mainly to increased sediment delivery from the Makuyuni River. The Makuyuni system is spatially and hydrologically complex (Figure 1 and Supporting Information Figure S1), wherein its northern tributaries drain the Monduli, Lesimingore and Lepurko volcanic highlands, dominated by Andosols and Leptosols (Nachtergaele et al., 2008). They subsequently flow to the middle elevation zone, dominated by Chernozems, and converge with each other and with the southern tributaries further down in the drier Maasai steppe, from which the main river flows towards Lake Manyara.
The rainfall is seasonal and characterized by a bimodal rainy season with a short peak from November to December and a longer peak from February to May. Moreover, the rainfall is spatially variable in the catchment (Figure S1), with higher levels of annual precipitation at the higher elevations (Nicholson, 1996;Prins & Loth, 1988). Connectivity between tributaries is often not accomplished due to localized precipitation, loss of runoff water by infiltration and evapotranspiration, diverging flows, and the presence of sinks, such as reservoirs, between upland areas and the main river network (Guzha et al., 2018;Jacobs et al., 2018). This gives the Makuyuni a typical ephemeral character, with low or no flow in the drier periods and a high discharge during and after rainstorms. During these peak flows, the river can also spill into its connected flood plains. The conjunction of these climatic and hydrological processes also creates a natural vegetation continuum from drier lowland rangelands to upland forests (Prins & Loth, 1988;Wynants et al., 2018). Open savanna rangelands, bushlands, agriculture, and bare land dominate the land cover (Table 1).
Smaller pockets of forest and wetland vegetation are confined to the uplands and floodplains, respectively. However, a combination of unsustainable land-cover changes and a natural high vulnerability to soil erosion resulted in a marked increase in surface erosion, gully incision and land degradation in the area (Blake et al., 2018b;Kiunsi & Meadows, 2006;Maerker et al., 2015;Wynants et al., 2018). Therefore, three sub-catchments ( Figure 1 and Table 1) were selected for a detailed investigation into the sources of transported sediment: Nanja, Ardai and Musa. Due to their variability in soil types, geology, rainfall, altitude and dominant land use, both within and between the three catchments, they are a natural microcosm of the wider northern Tanzanian landscape, allowing a representable study of soil erosion and sediment transport in the region.
The dominant land-use groups (Table 1) were obtained from a previous land-cover reconstruction of the area by Wynants (2018), combined with ground observations made during fieldwork. The catchment was grouped into three approximate zones wherein the up-zone (> 1600 m) is generally characterized by younger soils developed on volcanic rocks, steeper slopes and higher rainfall. The lowzone (< 1400 m) is generally characterized by sandier soils developed on older metamorphic rocks, lower slopes and lower rainfall. The midzone (1400 m-1600 m) is an intermediate zone with often deeply F I G U R E 1 Location of: A, Nanja sub-catchment; B, Ardai sub-catchment; C, Musa sub-catchment depicting the sampling locations of source soils (black triangles) and riverine sediment (pink circles). The 1400 and 1600 indicate the proximate borders between the low-zone (< 1400), midzone (1400-1600) and up-zone (> 1600) of the catchment. The land cover of the catchments is given, as well as its geographical context within the Lake Manyara system and in Tanzania [Color figure can be viewed at wileyonlinelibrary.com] T A B L E 1 Characteristics of the studied river catchments and an overview of the amount and types of samples taken weathered soils, medium to steep slopes and variable rainfall levels.
The main distinction between erosion process was made between surface erosion, consisting of rill and interrill erosion, and subsurface erosion, consisting of gully and riverbank erosion. Photographs of the erosion features in the study area can be found in the discussion.

| Sampling strategy
Respectively 122, 66 and 63 soil samples were taken from 26, 8 and 8 sites in the Ardai, Nanja and Musa sub-catchments ( Figure 1 and Table 1). Sampling sites in each sub-catchment were selected to account for the variation in the three levels of interest: land use, catchment zone and erosion process. The specific sampling locations were further subject to accessibility, necessary permits and safety.  (Gellis & Noe, 2013;Wilkinson et al., 2013). Since the sampling of DS took place in the dry season, it was assumed that the DS samples integrated the sediment from multiple smaller flow events (Smith & Dragovich, 2008). If the rivers were in high flow during the wet season, suspended sediment (SS) samples were taken by collecting 3-5 bottles of 1.5 L river water and letting it settle. Due to the non-quantitative approach to SS collection, no estimations of SS load could be made. In the Ardai system, sediment was collected from two high flow events (SS) and two times at no flow in the dry season (DS). The Musa system was sampled in one flow event (SS) and two times in the dry season (DS). The Nanja outlet was not reachable in the wet season and was therefore only sampled two times in the dry season using the DS approach.

| Laboratory analysis
Sediment and soil samples were either freeze-dried or oven-dried at 40 C and subsequently disintegrated using a mortar and pestle.
Samples were analysed for major and minor element geochemistry by wavelength dispersive X-ray fluorescence (WD-XRF; OMNIAN application, Axios Max, Malvern PANalytical, Malvern, UK) as pressed pellets. Prior to analysis all samples were sieved to < 63 μm to limit particle size effects (Laceby et al., 2017;Motha et al., 2002) and because of the general focus on the detrimental fine sediment (Walling, 2013).
The sieved < 63 μm fraction was further homogenized by milling it for 20 min at 300 rpm in order to reduce shadowing effects and preferential analysis of finer particles (Willis et al., 2011). Measurements were validated using stream sediment certified reference material (GBW07318, LGC, Middlesex, UK). Triplicates were made of randomly selected samples to assess repeatability of the method. Instrument drift was assessed following internal quality control procedures using a multi-element glass sample. Only those elements returning measurements above the limit of detection for > 75% of the samples and with triplicate variability < 5% were used in further analysis.
Compound specific stable isotope analysis (CSIA) of FAs was per- Fatty acid methyl esters (FAME) concentrations were measured using gas chromatography with flame ionization detection (GC-FID; TRACE GC, Thermo Scientific, Waltham, MA, USA). The solvent volume was adapted to obtain ideal concentration for isotope determination with capillary gas chromatography-combustion-isotope ratio mass spectrometry (GC-C-IRMS; TRACE GC Ultra interfaced via a GC/C III to DeltaPLUS XP, Thermo Scientific). Isotope ratios were expressed as δ 13 C values in per mill relative to the VPDB standard. An in-house prepared Schimmelman FAME mix reference (C20-C30 FAs), traceable to IAEA-CH6, was injected every six samples for δ 13 C calibration.
The δ 13 C values of the analysed FAME were corrected for the methanol group that was added during derivation using the internal standard measurements. The short-chained δ 13 C-FA C16-C21 were omitted out of the further analysis because they are more susceptible to degradation and are predominately produced by microorganisms (Upadhayay et al., 2017).
On a selection of soil samples from representative locations over the catchment, the organic matter (OM) content and aggregate stability (AS) was estimated. OM was estimated using loss of ignition (LOI), wherein the percentage of mass lost after 24 h at 450 C was calculated (Heiri et al., 2001). AS was calculated using laboratory rainfall simulation on a 45 mm/h intensity with a mean drop size of 580 pm.
A mean rainfall simulation survival index (RSSI) was calculated based on the number of aggregates surviving at 5, 10, 15 and 20 min during the test (Ternan et al., 1996).

| Source grouping
Source grouping based solely on geospatial information might lead to grouping samples together with very different fingerprints, which would reduce the model efficacy. Vice versa, grouping samples solely based on unsupervised statistical techniques might lead to soil clusters that are not relevant from a geospatial and ultimate a land management point of view. Therefore, the source grouping in this study was done using a combination of geospatial analysis (GA) and unsupervised cluster analysis (CA) that were integrated using principal component analysis (PCA).
The first step in the source grouping was to perform a CA of the soil samples for each sub-catchment using the unsupervised 'K-means' method of Forgy (1965), where the only expert input is the number of cluster numbers, and the outcomes are soil groups solely based on the variation in the multivariate fingerprint. The number of clusters was selected based on the elbow technique, wherein the number of clusters was chosen so that adding another cluster did not reduce the intra-cluster variance significantly (Kassambara, 2017). This information was combined with expert-set expectations of source delineation based on the geospatial informa-   (Stock & Semmens, 2017) and adapted by Blake et al. (2018a) for river basin sediment transport. The covariance structure of MixSIAR handles redundancy so tracer selection by discriminant function analysis is not required (Stock et al., 2018). As If the mean tracer concentration of the mixture was found to be outside the mean concentrations of the different sources, the tracer was excluded out of the analysis. These observations can be linked to nonconservative behaviour through enrichment or depletion processes , but can also be due to pollution or sampling constraints in the case that certain tracers occur spatially concentrated in the catchment (Belmont et al., 2014;Yu & Oldfield, 1993).
Furthermore, if the range test demonstrated the intra-source variance to be higher than the inter-source variance for specific tracers or if the inter-mixture variance was too high, they were also removed.
Finally, the univariate tracer distributions of the riverine sediment were assessed for normality using the 'Shapiro-Wilk test' because the model assumes that mixture tracer data are normally distributed.
If the tracer mixtures were not normally distributed, they were also removed from the analysis.
Based on these multiple tests, 12 tracers were excluded in the  (Stock et al., 2018). If this was the case, the source grouping, explained in the previous section, was re-evaluated.
Model outcomes are probability plots, wherein the mean of the plots is given as the proportional contribution of the sources and the stan- Substantial overlap between those groups is evident, however, is mostly due to one 'outlier' low-zone maize sample.

| Nanja
Geochemical analysis of the Nanja soil samples indicates clear clustering by catchment zone, and to a lesser extent, soil depth and land use ( Figure 4A). One outlier gully sample (Arkaria gully 240-250 cm depth) was omitted from the analysis because it had unusually high concentrations of Ni, Cu, Zn and Cr, which might be due to a spatially isolated concentration of metals in the specific soil layer. Inclusion of this sample might skew the entire model away from gully contribution.
A slight distinction was observed between the eastern (Arkaria) and western (Lepurko)

| Musa
Analysis of the Musa soil samples shows distinct clustering that is mainly driven by geochemical differences between catchment zones on the one hand and soil depth on the other hand ( Figure 5A). The upzone surface soils are characterized by high concentrations of P 2 O 5 and SO 3 , as well as high concentrations of Br and CaO. The forest fingerprint slightly diverges from the up-zone agricultural fingerprint and can be grouped into two distinct sub-clusters (with four and two samples, respectively), wherein the latter seem to be driven by a weathering signal. Furthermore, the forest soils have higher concentration of SO 3 , while the agricultural soils have a higher concentration of SiO 2 , which could be a site-specific indication of a higher sand content in the latter. The up-zone agricultural soils form a distinct cluster with smaller sub-clusters between the different agricultural practices that slightly overlap.
The mid-zone surface soils also group into one distinct cluster with two sub-clusters of rangeland and maize cropland, and are mainly
It should be noted that the contributions from the mid-zone and sap-  (Figures 2A, 4A and 5A). In all three catchments, the low-zone soils are characterized by an evaporative signal (K 2 O, Na 2 O and MgO), which is due to their drier and hotter conditions that lead higher concentrations of salts on the soil surface, but also by higher SiO 2 concentrations that signal higher sand content (Horowitz, 1991). The wetter up-zone soils have a distinct detrital signal (P 2 O 5 , SO 3 and CaO) due to the higher amounts of OM. This also indicates that land-use/vegetation patterns are not independent to catchment zones. While these zonal trends dominate the differences in geochemical fingerprints, the remaining importance of location-specific pedogenetic processes is evident from the distinct differences between surface soils of the same catchment altitude zone (Little & Lee, 2010). In the Nanja and Ardai catchments, about half of the gullies and all river banks were characterized by less-F I G U R E 4 (A) PCA plot of soil and river geochemical fingerprint signals in the Nanja sub-catchment with hull areas drawn around the soil groups as classified from the source grouping. Each group is given a number: I = bedrock incision, II = eastern hillslope gullies, III = western hillslope gullies, IV = low-zone surface, V = mid-zone maize, VI = eastern mid-zone rangelands, VII = western mid-zone rangelands. Each soil sample has been given a unique colour and symbol depending on respectively the land use and catchment zone where it was sampled from. (B) The output from the BMM that unmixed the sediment mixture against the soil groups in the Nanja sub-catchment using their geochemical fingerprints. The boxplots represent the density distribution of the output from MCMC runs with median shown by central line, interquartile range by box, and range by whiskers. The boxplots thus indicate the contributions of each of the soil groups to the riverine sediment. The bigger the box and whiskers, the higher the probability range, meaning less certainty of the results. The numbers of the boxplot groups correspond with the numbers of the soil groups on the PCA plot [Color figure can be viewed at wileyonlinelibrary.com] weathered tracers (SiO 2 , Rb), indicating they are incising closer to the bedrock (Baskaran, 2011;Horowitz, 1991). Interestingly, the other half had similar fingerprints as the surface soils. Moreover, some of the deep gullies were characterized by a strong weathering signal (Ti, Fe 2 O 3 and Al 2 O 3 ), which seems counterintuitive as deeper soil layers are usually less weathered as they are closer to the bedrock. However, soils in East Africa can be deeply weathered (Jones et al., 2013) and incision into these deep saprolites (Figure 7) can increase the strength of the weathering signal because they are not 'diluted' with detrital or evaporative surface signals. This observed lack of distinction complicates the eventual attribution of eroded sediment into surface and subsurface soils, which is a major limitation of erosion process attribution in tropical areas. Moreover, while gullies can look similar, they can have originated from different hydrological processes.
Vice versa, the runoff processes that cause sheet erosion, are also responsible for hillslope gully incision. The distinction between surface and gully erosion is thus not clear-cut, and increased rates of sheet erosion can evolve into gully incision due to continued soil weakening and deepening rills (Figures 8 and 9). The ambivalence in gully fingerprints found in this study thus confirms existing knowledge about the complexity of gully erosion (Poesen, 2011;Valentin et al., 2005). Only in the Musa sub-catchment there was a clear distinction between the gully and surface fingerprints, wherein the former had a distinct Similar to the other two sub-catchments, the 'sheet erosion' section in the Musa sub-catchment is thus likely to include hillslope incision processes as well. This exposes an important limitation in the methodology related to the limited spatial extent of soil and gully samples in the sub-catchments (Haddadchi et al., 2019;Yu & Oldfield, 1993).
The δ 13 C-FA analysis of the Ardai soils accentuates that in East African Rift systems, the δ 13 C À FA fingerprint is mainly driven by the altitude-rainfall gradient, working through the C3/C4 metabolic and altitude effects on the plant δ 13 C signal (Upadhayay et al., 2020). In general, woody C3 plants with lower δ 13 C dominate the wetter upzone, while grass C4 plants with higher δ 13 C values dominate the drier low-zone (Osborne, 2008). As maize is a C4 plant, soils under the dominant maize cropping systems will also incorporate a C4 signal (Christensen et al., 2011). Besides the altitudinal C3-C4 gradient, δ 13 C-FA is additionally influenced by altitude through the effects of vapour pressure deficit, temperature, and CO 2 concentration on plant photosynthetic activity and stomatal conductance (Upadhayay et al., 2020). Nonetheless, as dominant land-use types also often correspond with catchment zones, CSIA is still a robust tool for land-use Partly overlapping with both up-zone groups was a cluster of closed and open rangeland, which probably signals a transition zone from woody to grass vegetation. Moreover, the biochemical fingerprint of 'open rangelands' and 'maize croplands' in the same catchment zone still diverged slightly, highlighting the power of using compound specific FA fingerprinting compared to bulk δ 13 C or δ 15 N. While substantial overlap was still evident between these two groups, this was mostly due to one outlier. Moreover, as shown by Wynants et al. (2018), rangelands are the major source of new maize cropping land. Vice versa, maize crops are often abandoned and return to rangeland. These changes in land cover leave a residual signal in the soil (Blake et al., 2012). Furthermore, the maize growing season is short and even during the growing season it is highly likely that the plots are re-colonized by natural grassland plants (Nassary et al., 2020).

| Sources of eroded sediment
In the three river systems, the sediment is dominated by mid-zone 'open rangelands' and mid-zone 'maize croplands' sources. This find-  (Kirkby, 1980). Third, overgrazing in the region prohibits natural vegetation recovery, leading to a higher vulnerability to soil erosion and runoff generation (Hein, 2006). Finally, the high contribution of 'open rangelands' could also be partly explained by their clearance for conversion into 'maize croplands', wherein land clearance increases erosion and thereby also the contribution of soil from the cleared land use to the sediment (Lal, 1996). The high contribution of 'maize croplands' is also not surprising given its high vulnerability to soil erosion.
The mid-zone and low-zone 'maize croplands' in the Makuyuni catchment are solely dependent on rainfall and are cleared for planting at the start of the rainy season (Traerup & Mertz, 2011). Furthermore, they only provide cover for a short period in the year and their superficial root system and row planting does not provide a solid buffer from erosion (Ngwira et al., 2013). Bedrock incision seems to only have a minor contribution to the total eroded sediment.
These findings in mid-sized catchments from Tanzania match those from agricultural headwater catchments in the highlands of southern Kenya, wherein agricultural surface soils were found to be the main source of riverine sediment (Kroese et al., 2020). However, it is important to note that the sediment samples were not taken continuously, and our results thus only represent the situation of the sampled time-period. Since river catchments systems are dynamic, they have variable amounts of discharge, sediment transport and source contributions (Lizaga et al., 2019(Lizaga et al., , 2020a. This is especially so in the context of semi-arid East Africa, which is further explored in the paired article of Wynants et al. (2021). The lack of high temporal resolution sampling is thus a major source of uncertainty in this study. The multivariate fingerprint analysis did reveal that the variance between river sediment sampled from different time-periods and modes of sampling was relatively small compared to the variance between the potential sources, adding some robustness to the model outputs.
Nonetheless, future sediment source tracing studies in East Africa will need to find innovate ways to obtain high temporal resolution sediment data from these highly unpredictable ephemeral systems. Moreover, as highlighted before, surface erosion and hillslope gully incision are not independent processes and were often found to resemble each other geochemically. The dominance of the hillslope source groups and low contribution from the bedrock incision source group should therefore not be interpreted as a lack of contribution from subsurface sources. Furthermore, in the context of catchment sediment connectivity, gullies also have an important effect on downstream sediment routing. While gully formation is usually mediated by farmers on private farms, limiting the downstream transport of eroded sediment, gullies often remain uncontrolled on the rangelands (Figures 7 and 9), speeding up hillslope degradation and downstream connectivity (Blake et al., 2018b). The higher contribution of sediment from the 'open rangelands' is thus not only caused by higher rates of erosion, but by a higher connectivity with the river system. Furthermore, the rangelands are often situated in the mid-zone of the catchment with significant runoff contribution from upstream agricultural lands. A study by Blake et al. (2020) has shown that increased runoff from upstream agricultural areas can lead to increased erosion and gully incision on the downstream rangelands. The observed lower contribution of 'upland agriculture' can be explained by longer growing seasons, a more diverse crop selection with better soil cover, higher soil OM content and aggregate stability ( Figure S8), and the presence of terraces and permanent vegetated buffer strips ( Figure 10). Finally, the low contributions of 'bushland' and 'upland forest' shows that natural vegetation remains the best buffer for soil erosion and sediment transport, especially since these land-use types are currently constrained to the steepest areas in the catchment.

| CONCLUSION
Analysis of the potential source materials in the Ardai, Nanja and Musa sub-catchments revealed a highly complex and variable earth surface system. Geochemical fingerprinting was shown to be a robust tool for distinguishing catchment zones. Biochemical δ 13 C-FA fingerprinting was also dominated by the catchment zone, however, specific By applying sediment source tracing, this study not only highlighted the dominant sources of eroded sediment in the specific catchments, but also elucidated some of the complex spatial dynamics of soil erosion and sediment transport in Tanzanian river systems.
Urgent mitigative strategies for both the rangelands and croplands are required to stop the further acceleration of soil erosion and sediment transport, wherein both the soil erodibility and the landscape connectivity needs to be reduced. Future sediment source tracing studies in East Africa should not only aim to quantify the contribution of hillslope gullies to the total sediment load, but also obtain a better understanding of the role of gully incision as a positive feedback loop in the processes of hillslope degradation and sediment connectivity. In this context, the study also highlighted the need for novel tracers that can better distinguish between the surface and subsurface deeply weathered soils.
IMIXSED project ID 644320), the Research Council UK Global Chal-