Particulate macronutrient exports from tropical African montane catchments point to the impoverishment of agricultural soils

Catchments in the tropics often generate high concentrations of suspended sediments following the conversion of forests to agriculture. The eroded fine particles are generally enriched with carbon and nutrients 15 originating from the topsoil. Sediment-associated carbon and nutrients are an important loss to the terrestrial ecosystem and tightly connected to processes controlling riverine particulate carbon and nutrient export. Soil nutrient depletion can limit crop growth and yields, whereas an excess of nutrients in streams can cause eutrophication in freshwater systems. Streams in East Africa, with widespread land conversion are expected to receive high loads of sediment-associated carbon, nitrogen and phosphorus. In this study, we build the 20 knowledge base for particulate carbon, nitrogen and phosphorus for contrasting land uses. Suspended sediments (time-integrated, manual-event based and automatic-event based sediment samples) were analysed for total carbon, nitrogen and phosphorus concentrations collected at the outlet of a natural montane forest, a tea-tree plantation and a smallholder agriculture catchment in western Kenya during two sampling campaigns in 2018 and 2019. Particulate carbon and nutrient concentrations were up to three-fold higher (p<0.05) in the natural 25 forest catchment compared to fertilized agricultural catchments. The higher carbon and nutrient ratios in the natural forest suggest that the particulate nutrients are of organic origin due to tighter nutrient cycles, whereas lower ratios in both agricultural catchments suggest mineral sediment sources. The findings of this study imply that with the loss of natural forest, the inherent soil fertility is progressively lost under the current low fertilization rates and soil management strategies. 30 https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
In sub-Saharan Africa, streams often have high concentrations of suspended sediments mainly due to the anthropogenic disturbance of natural ecosystems (Mogaka et al., 2006;Penny, 2009). These sediment concentrations can be particularly high in the steep highlands of East Africa, where surface soil erosion in catchments dominated by agriculture generates significantly more suspended sediments than in forested 35 catchments (Brown et al., 1996;Stenfert Kroese et al., 2020b;Tamooh et al., 2014). The loss of organic carbon-(C) and nutrient-rich topsoil through surface runoff induced by soil erosion from agricultural surfaces (Powlson et al., 2011;Quinton et al., 2001) leads to further deterioration of tropical soils (Okalebo et al., 2005;Tully et al., 2015). Phosphorus (P) and nitrogen (N) are soil nutrients which can limit crop growth when they are in low supply, and therefore their loss from topsoils should be avoided (Pasley et al., 2019). The clays found in 40 many of the tropical soils of East Africa contain amorphous iron (Fe), which can limit plant P availability and consequently the application of fertilizers (inorganic N and P and/ or manure) is often required to improve agricultural productivity (Mutuo et al., 1999). Soil-forming processes tend to be much slower than the rates of soil loss through erosion (Amundson et al., 2015;Evans et al., 2020), which can lead to soil and plant nutrient deficits resulting in low and stagnant crop yields (Lederer et al., 2015;Saiz et al., 2016). Soil organic matter is a 45 critical determinant of soil fertility, providing nutrients to plants, and playing fundamental roles in soil carbon sequestration and soil water functions (Owuor et al., 2018;Weil and Brady, 2016). Processes affecting nutrient stocks and pools of soils are tightly connected to the processes controlling riverine C and nutrient fluxes.
Several studies investigating the effects of land use change and agricultural intensification on the water quality of river systems have focused on dissolved C and nutrients such as N and P (Bowes et al., 2018;Drewry et al., 60 2009;Smith et al., 2017). Others have shown that up to 95 % of the nutrient loads and up to 40 % of the C loads in streams are transported in particulate form, associated to suspended sediments (Moran et al., 2005;Rodríguez-Blanco et al., 2015;Scanlon et al., 2004). Sediment-associated nutrients are spatially variable (Harrington and Harrington, 2014;Withers et al., 2001) and their concentrations increase during storm events with the rising limb of the hydrograph (Drewry et al., 2009;López-Tarazón et al., 2016). Walling et al. (1997) 65 and Bender et al. (2018) observed that P loads mainly occur in particulate form, while N is mainly transported in dissolved form (Wang et al., 2015). Others have found that the major inputs to rivers occur as dissolved P and N loads (Harrington and Harrington, 2014). Currently, there are no studies for East Africa investigating sediment-associated carbon, nitrogen and phosphorus exports from different land uses. This is an important gap in the knowledge because the East Africa region has a very dynamic land use system, with intensive conversion 70 of natural ecosystems (forests, wetlands and grasslands) to subsistence and commercial agriculture (Carter et al., 2018).
This study aims to fill this knowledge gap by developing an improved understanding of the response of suspended sediments and carbon and nutrient fluxes associated with sediments under contrasting land use at catchment scale in the headwaters of the Sondu River Basin originating in the Mau Forest Complex, Kenya. 75 Earlier work in this area found much higher suspended sediment yields and dissolved nitrate exports in agricultural compared to forested land use (Jacobs et al., 2018;Stenfert Kroese et al., 2020b) and identified agricultural land as the main source of sediments within a smallholder agriculture catchment (Stenfert Kroese et al., 2020a). Therefore, we set out to test the hypothesis that the sediment-associated carbon and nutrient fluxes would be higher from the more intensively managed agricultural catchments than from a natural forest 80 catchment.

2
Materials and methods

Catchment characteristics
This study was conducted in nested catchments in the montane headwaters of the Sondu River Basin (3,470 km 2 ) located in the South-West Mau, Western Kenya, home to one of the largest closed-canopy montane 85 https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License.
forest of East Africa (the Mau Forest Complex), and where the inherent fertility of the soils allowed the development of a tea industry in parallel to subsistence agriculture (Binge, 1962). Over the last four decades, the western Highlands of Kenya have undergone significant land use changes, whereby 25 % of the Mau Forest Complex was converted to commercial and smallholder agriculture (Brandt et al., 2018). As an important headwater area to tributaries of Lake Victoria, the Mau Forest Complex is a critical catchment area that supplies 90 people (approximately 5 million), livestock, wildlife and the economy with fresh water (UNEP, 2008). Lake Victoria, an important freshwater lake for five countries, has shown increased signs of eutrophication over recent years (Zhou et al., 2014), stressing the need for mitigation and control of nutrient inputs.
The study catchments are characterized by distinct land uses: (1) natural forest (NF; 35.9 km 2 ), (2) tea-tree plantations (TTP; 33.3 km 2 ) and (3) smallholder agriculture (SHA; 27.2 km 2 ) ( Figure 1 & Figure 2). The mean 95 slope gradient of the catchments ranges between 11.6 and 15.7 %, up to a maximum of 72 % in the natural forest catchment. The streams are first-and second-order perennial streams merging to Sondu River (a sixth-order stream). The region has a bimodal rainfall pattern with a long rainy season (March-June) and a short rainy season (October-December) with a continued intermediate rainy season between the two wet seasons (July-September). The driest months are in January and February. The mean annual rainfall is 100 1,979 ± 325 mm yr -1 (period 1905-2019). Geology is composed of folded volcanics from the early Miocene.
https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. The natural forest catchment falls within the South-West Mau Forest part of the Mau Forest Complex. As an afromontane mixed forest, species such as Polyscias kikuyuensis, Macarangea kilimandscharica, Olea hochstetteri, Casearia battiscombei and Fagara spp. dominate the vegetation, transitioning to irregular patches of bamboo forest above 2,300 m above sea level (a.s.l.) (Binge, 1962). The riparian zone transits from the forest 115 vegetation containing an understorey with a dense cover of shrubs and tree ferns combined with tall indigenous tree species (Table 1).
The tea-tree plantation catchment borders the South-West Mau to the west. The catchment is characterised by tea (Camellia spp.) plantations alternated with Eucalyptus saligna and Cupressus lusitanica woodlots used for fuelwood for tea processing at the tea factories and timber production. The common practices to control soil 120 https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. erosion are mulching and interplanting rows of oat grass between rows of tea during the establishment of new tea fields, cover crop establishment with mature tea trees, and terracing and sited cut-off drains. A well maintained road system along with open culverts as road drainages connects surface runoff with the riparian buffer zone. Herbicides are commonly used to control weeds. Aerial application of inorganic fertilizer is conducted two to three times per year on the tea plantations (150-250 kg N ha -2 yr -1 and 8-13 kg P ha -2 yr -1 ) 125 (Jacobs et al., 2018). Riparian zones of up to 30 m from the river are commonly vegetated by native tree species densely covering the ground (Table 1).
In the upper smallholder agriculture catchment, subsistence farmers grow maize interspersed with beans, potatoes, millet, cabbage and tea (Camellia spp.) on farms usually smaller than one hectare. Grasslands for livestock and woodlots of Eucalyptus saligna, Cupressus lusitanica and Pinus patula are alternated with 130 agricultural land. A combination of hoeing and herbicide application is used for weed control. Inorganic fertilizer is applied manually on potatoes and maize (23-45 kg N ha -2 yr -1 and 12-23 kg P ha -2 yr -1 twice a year on potatoes and once a year on maize), while manure is commonly used for cabbage and other greens. Unpaved roads, frequently used by people, livestock and motorbikes, often develop into deeply incised gullies running down slope, coupling hillslopes with the stream network. Degraded river banks with a sparse or absent riparian 135 vegetation are prone to erosion, and in some places degraded riparian wetlands are found (Table 1).

Continuous field monitoring
The outlet of each catchment is equipped with an automatic gauging station to measure continuously (10 minute interval) water level (m) with a radar sensor (VEGAPULS WL61, VEGA Grieshaber KG, Schiltach, Germany) and turbidity (formazin turbidity unit=FTU) using a UV/Vis spectroscopy sensor (spectro::lyser, s::can Messtechnik GmbH, Vienna, Austria) ( Figure 1). Stream discharge (m 3 s -1 ) was obtained by using a site-specific 145 second-order polynomial water level to discharge rating curve (Jacobs et al., 2018). Specific discharge [mm day -1 ] was determined by integrating instantaneous discharge taken at 10 minute intervals over a day and relating it to the catchment area. In situ turbidity measurements were used to estimate suspended sediment concentrations (mg L -1 ), based on a rating curve between turbidity and suspended sediment concentrations established by Stenfert Kroese et al. (2020b). After obtaining continuous discharge and suspended sediment concentration 150 values, suspended sediment load was determined by multiplying suspended sediment concentration and https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. discharge for each 10 minute interval. Suspended sediment yield was calculated by integrating sediment load over time and relating it to the catchment area.
Precipitation was measured using eight automatic tipping bucket rain gauges calibrated to measure cumulative rainfall every 10 minutes with a 0.2 mm resolution (5 tipping bucket rain gauges: Theodor Friedrichs, 155 Schenefeld, Germany, and 3 weather stations: ECRN-100 high resolution rain gauge) ( Figure 1). Thiessen polygons weighted the contribution of rainfall of every tipping bucket to each catchment. A more detailed description of sampling sites and instrumentation can be found in Jacobs et al. (2018). This study uses hydrological and sedimentological data between January 2018 and December 2019.

Data quality assurance 160
For quality assurance the turbidity, discharge and rainfall datasets were checked for anomalities. Recorded values were flagged with Not-a-Number (NaN) during malfunctioning such as (i) sensor above water level, (ii) siltation of turbidity sensors, (iii) biofilm or debris on the measurement window due to malfunctioning of automatic cleaning with compressed air, (iv) measurement gaps due to incidents of power supply failure or (v) restricted counting of number of tips by the rain gauges by blocked funnel or spiderwebs. 165 Once anomalous values were replaced by NaN, local outliers were detected by the median absolute deviation (MAD) for discharge and suspended sediment: where is the whole dataset, 1 is the median of the dataset and 2 is the median of the absolute deviation from the dataset from its median. The standard deviation is estimated by the constant b set to 1.4826 for normal 170 distribution (Leys et al., 2013). A moving window of k=16 measurements around observation at time was used to detect local outliers with = ( − /2 … −1 , +1 … + /2 ): where a=6 is the threshold for outlier selection, , is the median and the , is the MAD for . There are gaps for discharge and suspended sediment data for the smallholder agriculture catchment in September until October 2019 due to theft of the power supply. Missing sediment data was integrated using a linear interpolation.  Table 2). The manual event-based 185 sampling was conducted next to the installed time-integrated samplers at each catchments outlet ( Figure 1).

Suspended sediment sampling
Time-integrated samplers were emptied of accumulated suspended sediment after three to five days. The storm event-based samples were retrieved during a storm manually with bulk river water samples (~10 L). The auto-sampling was only conducted at the outlet of the smallholder agriculture catchment, whereby samples were collected during the rising and falling stages of a storm event at 30 minute interval and composed to a bulk 190 sample. Sediment in suspension from all three sampling methods was extracted through settling and sedimentation followed by air-drying in aluminium trays.

Processing and qualitative analysis of suspended sediment
An aliquot of each sediment sample (>230 mg) was ground using a ball mill grinder for further laboratory 200 analysis. The ground samples were analysed for total carbon (TC), total nitrogen (TN) and total phosphorus (TP) concentrations. In this context, TC, TN and TP refer to the total C, N and P concentration in suspended sediments, which corresponds to particulate C, N and P (expressed in g kg -1 ). For TC and TN concentration measurements, a sub-sample of 30 mg of ground sediment was wrapped in tin capsules and combusted in an elemental micro-analyser (Elementar vario EL III, Elementar Analysensysteme GmbH, Langenselbold,205 Germany) at 950°C. For TP concentration measurements, a sub-sample of 200 mg of ground sediment was digested in 4.4 ml of sulphuric acid-hydrogen peroxide digest reagent and heated to 400°C for two hours (Allen https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. et al., 1974). After diluting the digestate twice, TP was determined using colorimetry based on a reaction with acidic molybdate in the presence of antimony which forms an antimony-phosphomolybdate complex. Ascorbic acid turns the complex in an intensely blue (phosphomolybdenum blue) which is measured 210 spectrophotometrically at 880 nm in a segmented flow analyser (Auto-analyser 3HQ, SEAL Analytical Ltd., Hampshire, United Kingdom). The remaining sediment samples were analysed for organic matter using gravimetric weight change by loss on ignition at 500°C for 4 hours in a muffle furnace.
To estimate nutrients and carbon concentrations in stream water (mg L -1 ), the mean suspended sediment concentrations (mg L -1 ) for each sampling period were multiplied with the particulate nutrients and carbon 215 concentrations (g kg -1 ) per dry weight sediments. For TC, TN and TP load calculations, the mean discharge and suspended sediment concentration was obtained for each sediment sampling period. The concentrations of TC, TN and TP (g kg -1 ) were multiplied to the mean discharge (m 3 s -1 ) and mean suspended sediment concentrations (mg L -1 ) to obtain the sediment-associated loads (t day -1 ). Annual suspended sediment-associated TC, TN and TP yields were calculated by integrating the mean of the daily loads to annual loads and relating it to the 220 catchment area.

Data analysis
All data were analysed for normality using the Shapiro-Wilk test. We tested significant differences on particulate nutrients and total carbon values among the different land uses between the two years using the Kruskal-Wallis test for analyses of variances. The effect of land use on particulate nutrients and total carbon 225 within each year were tested for significance using the pairwise Wilcoxon rank sum test. To test the linear relationship between two macroelements, the correlation coefficient r was identified between each two of the macroelements. All significant differences reported are at p<0.05.

Hydrological and suspended sediment responses 230
Mean annual rainfall for the study period (2018-2019) was 1,989, 2,006, and 1,671 mm yr -1 for the natural forest, tea-tree plantation and smallholder agriculture catchments, respectively. In the natural forest and the tea-tree plantations, 2019 was wetter than 2018, while the smallholder agriculture catchment was drier in 2019.
Mean annual specific discharge was highest in the natural forest catchment followed by the tea-tree plantation and the smallholder agriculture catchments with 806 (778-834) mm yr -1 , 678 (642-714) mm yr -1 and 658 235 (634-682) mm yr -1 . The catchment runoff coefficient was highest for the natural forest catchment (0.41) and smaller for the tea-tree plantations and the smallholder agriculture with a mean of 0.34 and 0.39, respectively.
The mean annual suspended sediment yield for the two years was highest in the smallholder agriculture with 231 (215-248) t km -2 yr -1 followed by the natural forest with 57 (54-62) t km -2 yr -1 and the tea-tree plantation catchments with 48 (44-52) t km -2 yr -1 (Table 3). 240 In all three catchments, discharge followed the rainfall pattern. The rising limb of the hydrograph was generally steep, followed by either steep or gentle falling limbs, depending on the magnitude of the storm event.
Discharge peaked during the long rainy season between April and July in 2018. In contrast, 2019 experienced a delayed onset of the rains and the highest discharge peaks occurred between October and December. Suspended 250 sediment peaks followed the same pattern as discharge (Figure 3).

Figure 4 Particulate TC, TN and TP concentrations [g kg sediment -1 ] of time-integrated, manual and automatic collected samples of the natural forest (NF), tea-tree plantation (TTP) and smallholder agriculture (SHA) catchments in the South-West Mau, Kenya in 2018 and 2019. Different letters indicate significant differences between land uses (p<0.05).
Particulate TC, TN and TP concentrations were significantly higher in the natural forest catchment than in the 270 tea-tree plantation and smallholder agriculture catchments in both years. The TC concentrations were significantly higher in the tea-tree plantations than in the smallholder agriculture in 2018, while there was no difference in 2019. TN and TP concentrations were significantly higher in the tea-tree plantation than in the https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. smallholder agriculture catchment in 2019, but concentrations were not significantly different between the tea-tree plantations and the smallholder agriculture in 2018 (Table 4). 275  The mean TC, TN and TP concentrations in the stream water for both years, estimated based on the particulate macronutrient concentrations and suspended sediment concentrations, were highest for the smallholder agriculture and lowest for the natural forest catchment and tea-tree plantations (Table 4). 285

Table 4 Mean ± standard deviation of particulate TC, TN and TP concentrations [g kg sediment -1 ] of time-integrated, manual and automatically collected samples, suspended sediment concentrations (TSS) [mg L -1 ] and TC, TN [mg L -1 ] and TP [µg L -1 ] concentrations in water at
The natural forest catchment had the highest percentage of organic matter in suspended sediments with 31 %, followed by the tea-tree plantations with 24 % and the lowest percentage was measured in the smallholder agriculture catchment with 16 %.

Stoichiometric C:N, C:P and N:P ratios and their relationships
The C:N ratio in the sediment from natural forest (12.6±1.0) was significantly higher (p<0.05) than those of the 290 tea-tree plantations (10.5±1.2) and the smallholder agriculture (10.6±1.3) for both years, while the C:N ratio was not significantly different between the tea-tree plantations and the smallholder agriculture (p>0.05). The C:P https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. ratio in the natural forest (269.8±106.4) was significantly higher than the tea-tree plantations (160.5±39.2) and the smallholder agriculture (126.7±22.6) in 2018, but the C:P ratio was not significantly different between the three catchments in 2019. The N:P ratio was significantly higher in the natural forest (19.9±6.9) than the 295 smallholder agriculture (11.8±3.0) in 2018, while 2019 showed no significant difference between the three catchments (p>0.05). The C:N ratio was significantly higher in 2018 than in 2019 in the natural forest and the tea-tree plantations, while the C:P and N:P ratio was significantly higher in 2019 than 2018 in the smallholder agriculture ( Figure 5). The sediment-associated C:N:P ratio was highest in the natural forest catchment with 225 C: 17N: 1P, followed by the smallholder agriculture with 172 C: 16N: 1P and the tea-tree plantations with 300 148 C:14 N: 1P, whereas the N:P ratio was similar for all three catchments. TC, TN and TP concentrations were correlated against each other for all three catchments ( Figure 6). The strongest relationship was observed between TC and TN within the tea-tree plantations (r=0.87), followed by the smallholder agriculture (r=0.71) and the natural forest catchment (r=0.64). The correlation between TN and TP was strongest in the tea-tree plantations (r=0.58). No significant relationships were found between TC and 310 TP in the natural forest and in the smallholder agriculture catchment with low correlation coefficient values ranging between -0.15 and 0.04, and between the correlation of TN and TP in the smallholder agriculture (r=-0.21) (p>0.05). The correlations of the natural forest catchment were in general higher than those of the tea-tree plantations and the smallholder agriculture ( Figure 6). https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License.

Sediment-associated TC, TN and TP loads 320
The mean daily suspended sediment loads for the sampling period were highest in the smallholder agriculture, followed by the natural forest and lowest in the tea-tree plantations (14.6±15.0 t day -1 , 4.9±2.5 t day -1 and 3.4±3.9 t day -1 , respectively). The mean daily TC and TN load in suspended sediment during the sampling periods in 2018 and 2019 was highest for the smallholder agriculture (668.9±661.9 kg day -1 and 60.5±59.5 kg day -1 ), followed by the natural forest (496.9±335.3 kg day -1 and 37.2±20.9 kg day -1 ) and the 325 tea-tree plantation catchment (193.4±178.6 kg day -1 and 18.1±18.5 kg day -1 ). For the sediment-associated TP, the highest daily loads were observed in the smallholder agriculture followed by the natural forest and the tea-tree plantations (5.5±5.1 kg day -1 , 1.9±1.3 kg day -1 , and 1.4±1.3 kg day -1 , respectively). The mean annual sediment-associated TC, TN and TP yields of the sampling period is estimated to be highest for the smallholder agriculture (9.0±8.9 t TC km -2 yr -1 , 0.8±0.8 t TN km -2 yr -1 and 73.9±68.9 kg TP km -2 yr -1 ), followed by the 330 natural forest and lowest for the tea-tree plantations (2.1±2.0 t TC km -2 yr -1 , 0.2±0.2 t TN km -2 yr -1 and 14.9±13.6 kg TP km -2 yr -1 ) ( Table 5).

Site
Year

Land use affects sediment-associated carbon and nutrients concentrations
This study shows that land use is a key control of suspended sediment-associated nutrients and carbon concentrations in the headwaters of the Sondu River Basin. Most significantly, and despite the application of 340 fertilizer to the tea-tree plantation and smallholder agriculture catchments, the natural forest catchment has much higher particulate TC, TN and TP concentrations than the agricultural catchments. These results suggest that the sediment from the forest catchment is comprised of organic material with a high C:N:P ratio either originating from the forest floor or falling directly into the river. In contrast, sediments from the agricultural catchments are mix of mineral and organic matter and fertilizer additions do not balance with the loss of carbon, nitrogen and 345 phosphorus from the system. This refutes our hypothesis that particulate-bound nutrients and carbon concentrations in sediment are highest in agricultural catchments where the majority of suspended sediment (77 %) originating from agricultural land (Stenfert Kroese et al., 2020a). Although concentrations of TC, TN and TP associated with sediments were lower in the smallholder agriculture catchment, because of higher sediment loads from the smallholder agriculture catchment the total sediment-associated loads of TC, TN and 350 TP were higher. In addition, higher sediment concentrations in the smallholder catchment meant that TC, TN and TP concentrations in the stream water were also higher from that catchment (Table 4, Table 5).
The low C and nutrient concentrations in sediments from agricultural soils are associated with soils much lower in nutrients and organic matter compared to nutrient-rich forest soils. The native forest vegetation was converted to smallholder farms during the last few decades (Brandt et al., 2018), conversion that leads to reduced organic 355 inputs and increased decomposition rates of soil organic matter following land management practices (Freibauer et al., 2004). Globally, an average decrease in soil C of around 30 % was found for soils after conversion of forests to croplands (Don et al., 2011;Murty et al., 2002), while a study of a converted native forest in the highlands of western Kenya showed a decrease by 30-40 % within the first 39 years after conversion (Nyberg et al., 2012). Similarly, a decline in soil organic carbon and nutrients was observed following conversion to 360 agricultural cultivation in the same catchments of the Mau Forest Complex (Arias-Navarro et al., 2017;Owuor et al., 2018;Wanyama et al., 2018). The highest TC and TN concentrations (81.1±24.2 g kg -1 and 4.9±2.3 g kg -1 , respectively) were measured in the surface soil (0-0.05 m) of the natural forest catchment, compared to lowest concentrations on smallholder croplands (56.9±11.1 g kg -1 and 2.1±1.2 g kg -1 , respectively) (Owuor et al., 2018). Chiti et al. (2018) found a significant decline in soil organic carbon as a consequence of forest degradation in 365 the same tropical montane forest. In their study, the organic horizon, the litter layer, had the highest TC concentration under primary forest (412.3±23.2 g kg -1 ) compared to a degraded forest (408.2±21.3 g kg -1 ) and to cypress (398.6±19.6 g kg -1 ) and tea plantations (381.7±17.6 g kg -1 ) that replaced the forest. With the reduction in the organic horizon a decline in soil TC was found in the uppermost mineral soil layer (<0.05 m) (Chiti et al., 2018). The suspended sediment is of mineral origin in the agricultural catchments and of organic origin in the 370 natural forest catchment, which is reflected in the differences in organic matter and the C concentrations in suspended sediment (Table 4).
The higher C:N:P ratios and the relationship between TC and TN in the natural forest catchment ( Figure 6) further imply that TN is of organic origin, where a lower and similar relationship in both agricultural catchments suggests a mix of mineral and organic sediment sources. A strong relationship was observed between TC and 375 TP for the agricultural catchments, which may indicate that the mineralization of soil organic matter contributes to the available phosphorous, as observed in the study of Maranguit et al. (2017). Similarly, the study of Johnson et al. (2017) in a forested catchment in the Piedmont region, Maryland (USA) showed that sediment sources from forest floor litter had the highest TC and TN concentrations compared to near-stream sources such as stream bed and stream banks. The fresher material from the forest floor was likely the least degraded 380 resulting in higher C and nutrient concentrations (Johnson et al., 2018). Old growth forests maintain a tight nutrient cycle through high and diverse aboveground biomass. They accumulate much more organic matter and https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License. nutrients by decomposition of fresh litter material and production of humus than managed land use types where carbon inputs are smaller and losses larger (Dawson and Smith, 2007).
When compared with other monitored catchments, both within and outside the tropics the sediments in the 385 forested catchment were more enriched with carbon, similar for nitrogen yet depleted in phosphorus, while the tea-tree plantation and smallholder agriculture catchments carbon and nitrogen fell within the range reported in other studies but phosphorus concentrations were lower ( Table 6). The TC concentrations of the natural forest catchment of our study were more than twice the concentrations recorded in the Congo basin under disturbed and undisturbed forest cover (14-95 %) (Coynel et al., 2005) and of disturbed agricultural catchments in 390 temperate regions (López-Tarazón et al., 2016;Walling et al., 2001). Other catchments of disturbed tropical basins under mixed land use (Tana River basin, Kenya and Ayeyarwady and Thanlwin River, Myanmar) had lower TC concentrations (Bird et al., 2008;Tamooh et al., 2012) than the concentrations of all three catchments of this study.
The TN concentrations of the natural forest catchment were in the same range as those of sub-catchments under 395 mixed land use of the Yangtze River, Jialing River and Wujian River, China, during the wet season, but exceeded the concentrations of intensified agricultural catchments in the USA, New Zealand and Spain by up to 12-fold (López-Tarazón et al., 2016;McDaniel et al., 2009;McDowell, 2015). The TN concentrations of the agricultural catchments of this study were within the ranges of concentrations in intensified temperate agricultural catchments (Pavanelli and Selli, 2013;Walling et al., 2001) (Table 6). 400 The particulate TP concentrations of the natural forest, tea-tree plantations and smallholder agriculture catchments were lower than those reported for a tobacco cultivated catchment in sub-tropical Brazil, with a TP range of 0.09-3.58 g P kg -1 , with a similar mean annual rainfall of 1,938 mm yr -1 (Bender et al., 2018). TP concentrations for agricultural catchments of temperate regions (Neal et al., 2006;Ramos et al., 2015;Sandström et al., 2020) and catchments in China during the wet season (Wang et al., 2015) exceeded the 405 concentrations of our study (Table 6).

Implications of nutrient losses from agricultural soils 415
The low sediment-bound concentrations (TC, TN and TP) in the smallholder agriculture catchment reflect soils with much lower organic matter and nutrients than the forest due to intensive cropping without appropriate practices for restoration of soil fertility. In addition to the decline in organic matter after conversion, soils cultivated on steep hillslopes with poor soil conservation practices and excessive surface runoff are prone to erosion and further nutrient depletion (Stenfert Kroese et al., 2020b). This may have been intensified by the cultivation of a major nutrient miner, Pyrethrum daisy (Chrysanthemum 420 cinerariifolium), following the clearance of the South-West Mau, which is known to promote surface soil erosion due to poor soil cover (Smaling et al., 1993). Similar soil nutrient losses have been observed in other densely populated tropical agricultural regions cultivated on steep hillslopes in Uganda (Lederer et al., 2015), Tigray (Ethiopia) (Girmay et al., 2009) or Kisii District (western Kenya) (Smaling et al., 1993) through nutrient losses, caused in particular by surface erosion and insufficient use of fertilizer application. Scarcity of land leads to an overexploitation of soil nutrients on agricultural land due 425 to the absence of fallow periods.
There is a need for soil management practices to be focused on the retention of soil organic matter through boosting agricultural productivity, mulching or cover crops in the agricultural catchments to maintain soil fertility. Increased retention of crop residues creates a positive feedback loop for the accumulation of organic matter (Nyberg et al., 2012). Incorporating crop residues into the soil is challenging for smallholder farmers, who prefer to feed them to their livestock (Castellanos-430 Navarrete et al., 2015). Other management practices such as vegetative buffer strips, erosion ditches or fanya juu terracing can prevent soil erosion (Conelly and Chaiken, 2000;Tiffen et al., 1994). Improved soil stability through soil organic matter will reduce soil erosion and suspended sediments and the further loss of soil organic carbon and nutrients, and therefore integrative practices that boost the production biomass and reduce bare soil are required.

5
Conclusions 435 This study conducted in catchments in the headwaters of the Sondu River Basin of the South-West Mau, Kenya, shows that cultivated land uses have led to a pronounced decline in sediment-associated carbon and nutrient concentrations compared to native forest. The high carbon and nitrogen ratio in the natural forest catchment reveals that most particulate nitrogen is of organic origin. The native forest is rich in biomass, with a tighter nutrient cycle and likely high organic matter inputs. This is in contrast to the smaller nutrient stocks in the agricultural land, reflecting nutrient losses and the impoverishment of 440 agricultural soils. The disturbance of soil through agricultural practices increases soil mineralization of soil organic matter, echoed in the lower organic matter content in suspended sediments. Despite the lower carbon and nutrient concentrations, the smallholder agriculture had the greatest particulate carbon and nutrient yields due to the higher export of suspended sediment yields passing the outlet. Elevated carbon and nutrient concentrations in sediments and in stream water contribute https://doi.org/10.5194/soil-2020-73 Preprint. Discussion started: 1 December 2020 c Author(s) 2020. CC BY 4.0 License.
to nutrient pollution and eutrophication impacting the downstream reaches such as Lake Victoria. Management practices 445 should focus on retaining and storing soil organic matter of agricultural topsoils to increase soil organic carbon, and soil nutrients. Practices to manage soil organic matter should also control soil erosion and consequently reduce suspended sediment concentrations and the loss of nutrient-rich topsoil.

Data availability
The raw data are available online https://doi.org/10.17635/lancaster/researchdata/387 hosted by Lancaster University, United 450 Kingdom.

Author contribution
JSK, MCR and JNQ designed the research. JSK conducted sampling, laboratory work and analysed results. JSK prepared the manuscript with input from MCR, JNQ, LB and SRJ.

Competing interest 455
The authors declare that they have no conflict of interest.