Stable isotope signatures of soil nitrogen on an environmental–geomorphic gradient within the Congo Basin

Nitrogen (N) availability can be highly variable in tropical forests on regional and local scales. While environmental gradients influence N cycling on a regional scale, topography is known to affect N availability on a local scale. We compared natural abundance of 15N isotopes of soil profiles in tropical lowland forest, tropical montane forest, and subtropical Miombo woodland within the Congo Basin as a proxy to assess ecosystem-level differences in N cycling. Soil δ15N profiles indicated that N cycling in the montane forest is relatively more closed and dominated by organic N turnover, whereas the lowland forest and Miombo woodland experienced a more open N cycle dominated by inorganic N. Furthermore, we examined the effect of slope gradient on soil δ15N within forest types to quantify local differences induced by topography. Our results show that slope gradient only affects the soil δ15N in the Miombo forest, which is prone to erosion due to a lower vegetation cover and intense rainfall at the onset of the wet season. Lowland forest, on the other hand, with a flat topography and protective vegetation cover, showed no influence of topography on soil δ15N in our study site. Despite the steep topography, slope angles do not affect soil δ15N in the montane forest, although stable isotope signatures exhibited higher variability within this ecosystem. A pan-tropical analysis of soil δ15N values (i.e., from our study and literature) reveals that soil δ15N in tropical forests is best explained by factors controlling erosion, namely mean annual precipitation, leaf area index, and slope gradient. Erosive forces vary immensely between different tropical forest ecosystems, and our results highlight the need for more spatial coverage of N cycling studies in tropical forests, to further elucidate the local impact of topography on N cycling in this biome. Published by Copernicus Publications on behalf of the European Geosciences Union. 84 S. Baumgartner et al.: Soil 15N signatures of the Congo Basin

. In addition, models show that the increase of primary production due to CO2 fertilization will be constrained by nutrient limitations (Wieder et al., 2015).
Therefore, understanding nutrient cycling in forests globally is key to assess present and future forest productivity Wieder et al., 2011). However, given their major role in the global C cycle (Lewis, 2006;Malhi and Grace, 2000), 35 quantifying nutrient supply and availability seems especially important for tropical forests.
Up-to-date, research has revealed high diversity in nutrient cycling with substantial variation from regional to local scales throughout the tropics. Old and highly weathered soils in tropical lowland forests are typically considered to be rich in N and poor in rock-derived P (e.g. Vitousek 1984). In contrast, tropical montane forests that lie on steep terrain are subjected to 40 geomorphic processes that rejuvenate soils with bedrock nutrients through uplift and erosion. In regions with a moderate uplift, P depletion is unlikely to occur due to this rejuvenation (Porder et al., 2007). As a result of reduced N supply in higher altitudes, due to slower N mineralization processes at lower temperatures, it is more likely that plant growth is limited by N availability in tropical montane forest (Bauters et al., 2017;Soethe et al., 2008;Tanner et al., 1998). It is important that nutrient availability in tropical forests are not being generalized and differentiations between different ecosystems within the tropics are being 45 made. Therefore, analysis of N cycling within ecological gradients are necessary. The "openness of the N cycle" is a good metric to describe cycling differences across ecosystems with contrasting N availabilities. While in N limited ecosystems the internal cycling and recycling of N is relatively more important than the absolute in-and outputs (closed N cycle), the opposite is true for ecosystems that are rich in N (open N cycle) (Boeckx et al., 2005).

50
Furthermore, at the local scale, topography has been identified as a main determinant for both biogeochemical and ecological variability in tropical forests Hofhansl et al., 2020) in part through its effect on geomorphic process activity.
In intact forest, slope gradient is the main control on physical erosion intensity at the hillslope scale (Roering et al., 2001).
Accumulated topsoil N has shown to be affected by physical erosion (Amundson et al., 2003;Hilton et al., 2013;Perakis et al., 2015;Weintraub et al., 2015), resulting in a close relation between nutrient availability and geomorphic process rates. 55 Therefore, while at regional scales mainly climatic and environmental gradients are responsible for differences in nutrient availabilities (Prescott, 2002;Read, 1991), hydro-geomorphic gradients seem to be a main determinant of N cycling at the local scale (Amundson et al., 2003). Recently, studies from Taiwan and Costa Rica have suggested that erosion has a significant control on N cycling (Hilton et al., 2013;Weintraub et al., 2015). They found indications of lower N availability and more closed N cycle in steeper sloping positions. This effect has only been reported from more geomorphic active sites of the tropics 60 (Costa Rica and Taiwan) and the magnitude of this effect in more stable landscapes is unknown calling for a consistent study across geomorphic gradients in the tropics.
The stable isotope composition of N (ẟ 15 N) is a useful proxy to assess various N cycling processes (Amundson et al., 2003; how soil δ 15 N values give insights in input and output pathways of N in the system (Högberg, 1990;Högberg and Johannisson, 1993). As such, climatic conditions such as rainfall (Austin & Vitousek, 1998;Boeckx et al., 2005), mean annual temperature (Boeckx et al., 2005) and soil moisture (Handley et al., 1999) all influence soil N isotopic signature. Generally, soil δ 15 N also signals the openness of the N cycle (Boeckx et al., 2005). On a local scale, soil erosion alters the δ 15 N signature and leads to more depleted values (Hilton et al., 2013;Weintraub et al., 2015). Because soil erosion is a non-fractionating process, the 70 combined fractionation of a system decreases when soil erosion becomes more important relative to fractionating N losses, such as denitrification. In this case, the isotopic signature of the remaining N in the system more closely resembles the signature of the N inputs compared to systems with less erosion.
The objectives of this study were 1) to assess the differences in N availability and N openness in three different forest types 75 along an environmental-geomorphic gradient in the Congo Basin and 2) evaluate the extent and 3) drivers of physical erosion and its effects on N availability in these forests. To answer these objectives, soil N content and the stable isotope compositions were measured in soil profiles taken from different topographical positions in three different forest ecotypes. We used δ 15 N signatures as a proxy that integrates N transformations for differences in N availability and the openness of a system. We hypothesized N availability and openness of N cycle would be highest in lowland tropical forest, that δ 15 N signatures would 80 be lower on steeper slopes, and that this effect would be more pronounced in the montane and Miombo forests, where the erosional forces are higher compared to the lowland forest.

Site description
We selected representative sites for the Congo Basin's major forest ecotypes (tropical montane forest, tropical lowland forest, 85 and the semi-dry Miombo woodland). The montane forest site is situated in the Kahuzi-Biéga National Park (2.344° S, 28.746° E), northwest of the city of Bukavu in the eastern part of Democratic Republic of the Congo (DRC). Two forest subtypes are present in the Park: montane mixed forest and monodominant bamboo forest. Soil samples were taken in the mixed forests.
Soils are classified as Ferralsols/Acrisols (Bauters et al., 2019) (Table 1). The topography of the sampled forest is characterized by steep slopes (Figure 1). Mean annual temperature is ~15°C and mean annual rainfall is ~1970 mm (Fick and Hijmans, 90 2017). Rainfall is seasonal with two peaks in April and October and a dry season that lasts from June to September. and mean annual rainfall is ~1760 mm (Fick and Hijmans, 2017) (Table 1). Yoko experiences a short wet season from March to May and a longer one from August to November.
For the Miombo woodland, a site 50 km north of the city of Lubumbashi was selected (11.212° S, 27.233° E). A pristine Miombo is characterized by deciduous woodland dominated by trees of the Brachystegia, Julbernadia and Isoberlinia genera 100 (Campbell, 1996). As charcoal production and seasonal fires are a common occurrence, soil sampling in areas with recent charcoal production activities was avoided. We classified the soils as Acrisols and Cambisols. The topography is mostly flat and the elevation ranges between 1260 and 1310 m a.s.l ( Figure 1). This region experiences a wet season from October to April, with a peak rainfall in December. Mean annual rainfall is ~1250 mm and mean temperature is ~21 °C (Fick and Hijmans, 2017) (Table 1). 105

Sampling
To analyze differences in soil N cycling between forest sites, soil samples were taken from soil profiles at different slope 120 positions in the catchments (slope, shoulder and ridge) in order to better represent the catchment as a whole. Nine soil profiles were dug and sampled in both the lowland and Miombo forests. Due to difficult terrain within the montane forest, only six soil profiles were sampled ( Figure 1). For each 1 m-profile, soil samples from nine different depths were taken (0-2 cm, 2-4 cm, 4-6 cm, 6-10 cm, 10-15 cm, 15-20 cm, 20-30 cm, 30-50 cm and 50-100cm). To calculate N stocks, bulk density of the soil was determined at five depths of each soil profile using a 100 cm 3 bulk density ring. The bulk density samples were dried at 105°C 125 for 48h and then weighed. To analyze the effect of surface slope angles on stable isotopic signature of soil δ 15 N, additional topsoil samples (bulk sample from the top 0-20 cm) were taken from throughout the catchments at locations with varying slopes (montane: n=27; lowland: n=24; Miombo: n=21) ( Figure 1). Profile-and topsoil sample positions were recorded with a GPS (GPSMAP 60CSx; Garmin; accuracy <5m) and slope angles were estimated using 30-m SRTM derived digital elevation models (DEM) (NASA JPL) which were smoothed with a low-pass filter. 130

Lab analysis
Soil samples were air dried and subsequently sieved through a 2 mm sieve to remove roots and organic residues. To homogenize, soil samples were milled and analyzed for C and N content and 15 N isotopes using an elemental analyzer (Automated Nitrogen Carbon Analyzer; SerCon; Crewe, UK) interfaced with an Isotope Ratios Mass Spectrometer (IRMS; 20-20, SerCon). Stable isotope ratios are expressed in delta notation against AIR-N2 standard. N stocks were calculated using 135 mass-based N content, multiplied by the bulk density of the respective depth and integrated over the whole profile depth.

Inter-site comparison
For a better inter-site comparison of the erosional effect we estimated a soil loss coefficient. In addition to slope, soil erosion is controlled by vegetation cover and rainfall characteristics. To incorporate these additional factors, we calculated an erosion coefficient (EC) for each surface soil sample. To calculate the EC, we adapted the numerical model for sediment detachment 140 (kg m -2 yr -1 ) from Pelletier (2012) MAP data were extracted from the WorldClim database and averaged for the years 2010-2018 (Fick and Hijmans, 2017). 145 For a pan-tropical comparison, data points from Costa Rica (Weintraub et al., 2015) and Taiwan (Hilton et al., 2013) were extracted from the supporting material of the representative studies.

Statistical analysis
To assess the effect of forest type and soil depth on δ 15 N and mass-based C:N ratio, linear mixed effects models were fitted, 150 controlling for topographic position via a random intercept. A third model was fitted for N-stocks. Since stocks are integrated over the whole soil profile, soil depth was omitted as an explanatory variable in the third model, including only the effect of forest type as a fixed effect, while controlling for topographic position via a random intercept. All models were fitted using maximum likelihood methods via the lmerTest package (Kuznetsova et al., 2017). To assess the relation between slope angle and δ 15 N of the spatial surface samples, linear regressions were applied. Statistical analyses were conducted using the R-155 software (R Development Core Team, 2019).
Furthermore, a structural equation model (SEM) was applied to examine the relative direct and indirect importance of different variables on the soil δ 15 N values of the five tropical forests. The SEM model was generated using the lavaan package version 0.6 (Rosseel, 2012) in the R-software.

Results 160
The montane forest showed the highest N stock with mean value (mean ± standard deviation (SD)) of 1.47 ± 0.61 t N ha -1 . N stocks were lower in the lowland forest and Miombo woodland, with 0.87 ± 0.11 and 0.73 ± 0.30 t N ha -1 , respectively ( Table   2). The montane forest and Miombo woodland soils exhibited similar C:N ratios with 11.4 ± 1.1 in the montane and 11.2 ± 1.8 in the Miombo, while the C:N ratio in the lowland forest was lower with a mean value of 7.4 ± 0.5 (Table 2)

Differences of N cycling between forest types
At present, relatively little information on δ 15 N data for soils in tropical forests is available. This holds especially for Central Africa, where almost no data is reported. Rütting et al. (2015) is one of the few studies that reported on δ 15 N from montane tropical forests in Rwanda. Their values for topsoils (0-5 cm) range between 3.6 and 4.6 ‰ and are similar to the values 190 reported in this study for the montane forest (mean of 4.4 ‰ for 0-6 cm depth in this study). δ 15 N values from tropical mountain forests of Taiwan are also in the same range, with a mean value of 4 ‰ for the upper 10 cm of the soil profile (Hilton et al., 2013). Jamaican soils from tropical montane forests are much more depleted in 15 N, with values from -1.6 to 1.5 ‰ for depths down to 20 cm (Brearley, 2013). Top soils from an altitudinal gradient in Borneo were also more depleted, with a mean value of 1.6 ‰ (Kitayama and Iwamoto, 2001). Values reported for tropical lowland forests are available for soils in Costa Rica 195 where top soils (0-10 cm) showed values of 4.7 ‰ (Osborne et al., 2017) and a mean value of 5.9 ‰ for soils from 0-20 cm depth (Weintraub et al., 2015). Higher erosion, due to high rainfall and uplift rates (1.7-8.5 mm yr -1 ), may explain why these values of lowland forest in Costa Rica are more depleted in the heavier N isotope compared to the lowland forest in Yoko from https://doi.org/10.5194/soil-2020-70 Preprint. Discussion started: 10 November 2020 c Author(s) 2020. CC BY 4.0 License.
this study (7.55 ± 0.69 ‰ for 0 -20 cm soil samples) because the residence time of soil N decreases with erosion and hence soil rejuvenation (Amundson et al., 2003). The δ 15 N reported here for the Miombo woodland (2.68 -4.32 ‰ between 0 -20 200 cm, Figure 2) falls within the same range as values reported by Wang et al. (2013) from a moist woodland savanna in Zambia (2.6 -4.8 ‰), but are lower compared to values reported from a dry forest in the Cerrado Savanna of Brazil (6.3 -10.8 ‰, Bustamante et al., 2004).
Soil δ 15 N is related to the residence time of ecosystem N. Higher rates of mineralization, nitrification and dentification leaves enriched N in the remaining soil organic matter (SOM), while the more depleted mineral N is removed via plant uptake or 205 denitrification and transported down into deeper soil layers by leaching (Handley et al., 1999;Hobbie and Ouimette, 2009).
Over the whole profile, the Miombo woodland showed highest depletion of 15 N (Figure 2), which indicates a slower N turnover and a more closed N cycle compared to the lowland and montane forest. However, topsoil δ 15 N is influenced by the isotopic signature of incoming litter fall, inputs through roots, and N fixation (Craine et al., 2015). Downward transport processes and transformation of SOM influences the δ 15 N in deeper soil layers (Baisden et al., 2002). Decomposition processes enrich the 210 remaining SOM in soils with the heavier 15 N isotope (Billings & Richter, 2006;Compton et al., 2007). In this study, the Miombo woodland showed slightly lower δ 15 N values for the topsoil (0-2 cm, 2.66 ‰) than the montane forest (2.94 ‰), while the lowland forest showed a more enriched signature (6.35 ‰). As topsoil 15 N signature is mostly governed by N inputs, the Miombo and montane forest are likely to receive depleted N inputs. Furthermore, the presence of N2-fixation depletes δ 15 N in top soils (Robinson, 2001), as it imports atmospheric N with an isotopic signature of 0‰, and might explain the low values in 215 the Miombo woodland. Indeed, previous studies have shown that N2-fixation is a more important process in tropical woodlands than in tropical forests (Hogberg & Alexander 1995), where less trees of the Fabacaea family are present (Malmer and Nyberg, 2008). Furthermore, it is reported that symbiotic N2-fixation is downregulated in old-growth tropical lowland forest (Bauters et al., 2016).
Soil δ 15 N profiles of the lowland forest and the Miombo woodland showed a δ 15 N peak at 10 cm and 30 cm depths, respectively, 220 indicating the highest microbial activity and/or leaching of soil N in this layer. Montane forest showed a similar peak at the same depth, indicating high turnover or leaching in this soil layer as well; however, from 20 cm onwards δ 15 N increased even further to the highest values at 100 cm depth. Hobbie & Ouimette (2009) presented a conceptual model for processes that influence δ 15 N values in a profile. According to this model, a soil profile with increasing 15 N values with depth (i.e. the case of the montane forest) is indicative of an N-limited ecosystem or dominated by organic N cycling. Although, we found highest 225 N stocks in the montane forest, higher foliar C:N ratios of tropical forests with increasing altitude (Bauters et al., 2017) are indicating, however, that this ecosystem experiences lower N availability and is dominated by organic N cycling. In Rwandan and Ecuadorian forests, Bauters et al. (2017) concluded from leaf δ 15 N altitudinal gradients that N-cycling becomes more closed with increasing elevation and hypothesized that organic N sources for plants become more important in ecosystems with lower temperatures. Furthermore, very high mineralization rates have been measured in the montane forests of  Biéga, which indicates a high organic matter turnover in these forests (Bauters et al., 2019). On the other hand, a profile with higher δ 15 N values in the root layers and decreasing δ 15 N in subsequent depths (similar to the lowland and Miombo profiles) https://doi.org/10.5194/soil-2020-70 Preprint. Discussion started: 10 November 2020 c Author(s) 2020. CC BY 4.0 License.
is more likely to occur in systems with less N limitation and more inorganic N cycling (Hobbie and Ouimette, 2009). Indeed, Bauters et al. (2019) found indication of mineral N excess in the Yoko lowland forest. In such systems, δ 15 N depleted Nspecies are transported downwards through the soil column, resulting in an accumulation of depleted 15 N in lower layers. 235 Understanding N cycles in tropical forests is important for a better assessment of the productivity of these ecosystems.
Although N stocks were highest in the montane forest, the δ 15 N profile in this forest indicate a more closed N cycling system with mainly organic N cycling, which is less bioavailable. On the other hand, the lowland forest and the Miombo woodland seem to have a more open N cycle due to excess of available N. Furthermore, lowland forest in Central Africa are reported to experience high atmospheric N depositions (Bauters et al., 2018) and the Miombo woodlands have more likely additional 240 inputs from N2-fixation.

Influence of topography on soil N
In contrast to the Costa Rican tropical lowland forests (Weintraub et al., 2015), slope angle was found to have no effect on δ 15 N in the lowland forest of this study (Figure 3a). Topography is one of the main factors controlling physical erosion 250 processes (Hilton et al., 2013;Weintraub et al., 2015). There are no steep slopes in the lowland forest (max. slope sampled 9°) and thus, there is a low potential for physical erosion. This is also indicated by a lower EC (1.22) in the lowlands compared to the coefficient in Costa Rica (2.39). Although the Miombo woodland is located in an area, with similar topography to the https://doi.org/10.5194/soil-2020-70 Preprint. Discussion started: 10 November 2020 c Author(s) 2020. CC BY 4.0 License. lowland forest (Figure 1), the EC from this area is much higher (20.11). This can be explained by the fact that the forest cover in the Miombo is less dense than the tropical forests and this increases the exposure of the soil surface to the energetic input 255 of rainfall. Furthermore, the loss of understory vegetation from fire, which occurs frequently during the dry season (Campbell, 1996), leaves the soil unprotected from erosion at the onset of the rainy season. The spatial topsoil samples of the Miombo woodland clearly indicate a significant effect of slope angles on δ 15 N (Figure 3a). The δ 15 N values in the montane forest, situated on steep terrain, tend to decrease with increasing slope angle, however, the visible trend is not statistically significant (p value of 0.57, Figure 3a). Furthermore, we found a higher variability in δ 15 N values of the soil samples in the montane forest 260 (CV of 22%) compared to the Miombo (CV of 15%) and lowland forest (CV of 9%). While slope angle is a commonly used predictor of soil erosion potential, the curvature of the land surface (i.e. convex or concave) has also been shown to affect soil erosion at a local scale (Stefano et al., 2000). The specific hydrology seems to control the erosive mechanism, as slope gradients have been found to influence the rate of soil erosion through overland flow, while curvatures affect diffusive processes, such as splash erosion (Stefano et al., 2000). However, while a trend of lower δ 15 N on more convex landscapes in the montane forest 265 is visible ( Figure S1), no statistical significance was found either, as samples were not equally distributed over the range of curvatures. In addition, because our sampling strategy was focused on the slopes, sample sites were biased towards concave curvatures.
Comparing the effect of slope angles on topsoil δ 15 N of the study sites from the Congo Basin with the effect observed in tropical forests of Costa Rica (Weintraub et al., 2015) and Taiwan (Hilton et al., 2013) shows that the magnitude of the effect 270 is related to the mean erosion coefficient in these ecosystems (Figure 3b). Sites with a high erosion coefficient show a higher slope of the regression δ 15 N ~slope angles compared to sites with a low erosion coefficient, implying a bigger effect on soil δ 15 N in more geomorphic active sites. As the dense tropical lowland forests of the Congo Basin do not represent high EC values, the local effect of erosion on soil δ 15 N signatures seems negligible. Although the montane forest shows a moderate EC, due to the steep topography, the high variability in soil N masks the potential effect of topography on δ 15 N in our dataset. The 275 estimation of EC based on equation 1 assumes that splash and overland flow are the dominant erosion processes but this may underestimate erosional forces at higher hillslope angles (Pelletier, 2012). At higher hillslope angles, landslides will be more important erosional processes than overland flow. This might indicate that at sites with high risks of landslides (as shown in the landslide hazard index (LSI) in Table 1) are even more prone to physical soil loss than the EC would indicate. In this case especially the forest in Taiwan (LHI of 4), the montane forest (LHI of 3.96) and the forest in Costa Rica (LHI of 3.65) 280 experience a high risk of landslides and total soil loss in this forest might even be higher than predicted by the EC.
We applied a SEM to assess the driving factors of the top soil δ 15 N variability between different tropical forest sites ( Figure   4). On a global scale, soil δ 15 N is reported to be mainly influenced by climatic factors (MAT and MAP) and soil properties, such as soil C and clay content (Craine et al., 2015a). Our results showed that soil C had only a marginal influence (lower coefficient compared to slope, MAP and LAI) on tropical soil δ 15 N and is only significantly influenced by MAT and not MAP. 285 Soil δ 15 N in these tropical forests seems to be primarily influenced by factors controlling erosion (MAP, LAI and slope). Beside the direct effect of MAP on soil δ 15 N, there is also a significant indirect effect visible, where sites with higher MAP experience https://doi.org/10.5194/soil-2020-70 Preprint. Discussion started: 10 November 2020 c Author(s) 2020. CC BY 4.0 License. also higher LAI values (Figure 4). It was not possible to analyze the effect of clay content on δ 15 N because the data were not available for all the samples. Although our data showed only for the Miombo a significant effect of topography on the soil δ 15 N values, we can highlight that especially in geomorphic active sites, erosion needs to be taken into account to explain local 290 differences in soil δ 15 N values.

Conclusion
The three tropical forest ecosystems in the Congo Basin each showed a distinct δ 15 N soil profile. The δ 15 N soil profiles in the montane forest indicate a closed N cycle, which supports the common theory of decreasing N availability at higher altitudes.
In contrast, the δ 15 N soil profiles indicated that the lowland forest and Miombo woodland tended to have more open N cycles, which were mostly dominated by inorganic N cycling. The Miombo woodland showed the lowest δ 15 N values, which is most 300 plausibly explained by enhanced N fixation due to the plant species assemblage. Furthermore, topsoil δ 15 N signatures across our and literature sites of the tropics suggests that a significant amount of the observed variability can be explained by erosional processes. Rainfall, vegetation cover, and topography are the main factors to explain δ 15 N variability between five different tropical forest sites. Within the Congo Basin, only the Miombo woodland showed an influence of slope on soil δ 15 N signature, whereas in the lowland forest, the comparably low MAP, dense vegetation, and flat topography likely limit erosion and resulted 305 in the lack of a correlation between the δ 15 N signature of soil N and slope angle. In general, this study highlights the possible importance of soil erosion on N cycling in tropical forests; however, not all sites are similarly impacted. Therefore, soil erosion has to be taken into account while interpreting of soil δ 15 N values as a proxy for N cycling. It underlines the importance of further spatial field studies in the tropics given the observed high variabilities between different tropical forest types.