1. Introduction
Despite many years of effort to reduce their effects, soil erosion and sedimentation are critical problems in the Ethiopian highlands Bayabil, Tilahun, Collick, Yitaferu, and Steenhuis (2010); (Borselli, Cassi, & Torri, 2008; Nyssen, Poesen, Moeyersons, Haile, & Deckers, 2008; Setegn, Srinivasan, Dargahi, & Melesse, 2009; Taye et al., 2013). Soil erosion causes not only on-site degradation of land resources but also off-site problems such as downstream sedimentation and deposition in fields, plains and water bodies (Hans Hurni, Tato, & Zeleke, 2005; Yeshaneh, Eder, & Blöschl, 2014; Zeleke & Hurni, 2001). Loss of top soil and subsequent silting up of reservoirs degrades the environmental resources necessary for subsistence (Dagnew et al., 2015; P. F. Hudson, 2003; Nyssen et al., 2009; Steenhuis, Winchell, Rossing, Zollweg, & Walter, 1995). This problem extends to downstream countries, Sudan and Egypt, because the Blue Nile drains the Ethiopian highlands and contributes sediment to downstream areas (Bayabil et al., 2010; Nyssen et al., 2008; Setegn, Dargahi, Srinivasan, & Melesse, 2010; Tessema, 2006; Seifu Admassu Tilahun, 2012). To reverse land degradation, the government of Ethiopia launched a massive SWC program for the last three decades (Herweg & Ludi, 1999; Mitiku, Herweg, & Stillhardt, 2006; Shiferaw & Holden, 1998). The interventions were focused on physical SWC strategies with emphasis on reducing accelerated erosion and downstream sedimentation (Desta, 2000; Jemberu, Baartman, Fleskens, Selassie, & Ritsema, 2017; Zeleke & Hurni, 2001). However, sustainable land management (SLM) is not yet attained, with widespread failure of SWC measures (Gebrernichael et al., 2005; Herweg & Ludi, 1999; Tefera & Sterk, 2010).
Ex-ante determination of the effect of SWC strategies on soil erosion and sediment yield can support decision making about SLM (Baptista, Ritsema, Querido, Ferreira, & Geissen, 2015; Nyssen et al., 2008; Setegn et al., 2010). Some studies have indicated that the sediment load is reduced in the Ethiopian highlands by land use changes and widespread use of soil and water management strategies such as bund structures, check dams, flood-control ponds and water diversions (Adimassu, Mekonnen, Yirga, & Kessler, 2014; Gebreegziabher et al., 2009; Gebrernichael et al., 2005; Nyssen et al., 2010; Rust, Corstanje, Holman, & Milne, 2014). However, few measurements are available to quantify the impacts of SWM strategies on soil erosion and sedimentary processes in the Ethiopian highlands, and modelling the linkage of on-site soil erosion rates within a catchment to the sediment yield at the outlet is often lacking due to lack of input data (Adimassu, Kessler, & Hengsdijk, 2012; Grum et al., 2016; Haregeweyn et al., 2013; Nyssen et al., 2008; Steenhuis et al., 1995).
Empirical lumped-approaches have been used to estimate sediment yield using the average basin characteristics such as area, drainage density, slope, land cover, soil type, etc. (Lenzi & Marchi, 2000; Nyssen et al., 2008; Ritsema, Stolte, Oostindie, Van Den Elsen, & Van Dijk, 1996; Setegn et al., 2010; Zhao et al., 2015). However, the validity of the equations resulting from such approaches is limited to the specific areas for which they have been established (Feng, Wang, Chen, Fu, & Bai, 2010; Haregeweyn et al., 2008; Quiñonero‐Rubio, Nadeu, Boix‐Fayos, & Vente, 2016). Inherent to lumped approaches is that it is not possible to take into account the spatial structure of land use and topography within the catchment on erosion and sediment delivery (de Vente & Poesen, 2005; Haregeweyn et al., 2008; Seifu A Tilahun et al., 2015). This inherently limits their applicability to practical problems such as the evaluation of different SWC measures on soil erosion and sediment delivery (A. J. Van Rompaey, Verstraeten, Van Oost, Govers, & Poesen, 2001; Vandaele & Poesen, 1995; Zabaleta, Martínez, Uriarte, & Antigüedad, 2007). Likewise, the sediment yield measured at gauging stations of many river systems is only a fraction of the total sediment load delivered to reservoirs and dams in the downstream areas (Adimassu et al., 2014; N. E. Asselman, 1999; Bayabil et al., 2010; de Vente, Poesen, Bazzoffi, Rompaey, & Verstraeten, 2006; Haregeweyn et al., 2008). These problems can be overcome by using a spatially distributed model, whereby the eroded sediment is explicitly routed over the landscape towards the river system, allowing the evaluation of the effect of SWC measures on erosion and sedimentation processes (Romero-Díaz, Alonso-Sarriá, & Martínez-Lloris, 2007; Vandaele & Poesen, 1995; Wudneh, Erkossa, & Devi, 2014; Zabaleta et al., 2007; Zhao et al., 2015).
Spatially distributed models have been applied globally to support SWC decisions (Betrie, Mohamed, van Griensven, & Srinivasan, 2011; Boix‐Fayos, de Vente, Martínez‐Mena, Barberá, & Castillo, 2008; Fleskens & Stringer, 2014; Haregeweyn et al., 2013; Lemann et al., 2016b). The spatially distributed Water and Tillage Erosion / Sediment Delivery Model (WATEM/SEDEM) provides estimates of long-term mean annual soil erosion rates and sediment yield (Van Oost et al., 2005; A. Van Rompaey, Bazzoffi, Jones, & Montanarella, 2005; A. J. Van Rompaey et al., 2001; Verstraeten, Prosser, & Fogarty, 2007). Haregeweyn et al. (2013) used WATEM/SEDEM to assess sediment yield in Tigray region, Northern Ethiopia. Didoné, Minella, and Evrard (2017) applied WATEM/SEDEM for evaluating the impact of SWC scenarios in southern Brazil. In SE Spain, WATEM/SEDEM was used to assess the impacts of check dams, land use change and forest restoration on sediment yield Quiñonero‐Rubio et al. (2016).
Validation of the spatial pattern of erosion and sediment connectivity within the (treated) catchment is complicated and accurate prediction of sediment yield at outlets of sub-watersheds does not mean that the spatial patterns of erosion and sediment yields are also accurately predicted (Bracken, Turnbull, Wainwright, & Bogaart, 2015; Marchamalo, Hooke, & Sandercock, 2016). However, by simulating the spatial distribution of erosion, transport capacity, sediment routing and sediment deposition, the effects of SWC measures can be spatially evaluated (Feng et al., 2010; Quiñonero‐Rubio et al., 2016). In modelling SWC measures, the routing algorithms can alter the transport capacity and sediment deposition patterns, while causing little change in predicted total erosion and sediment yield (Takken et al., 1999; Takken et al., 2005; Vigiak, Sterk, Romanowicz, & Beven, 2006).
Based on previous experiences (Haregeweyn et al., 2013), the successful calibration of WATEM/SEDEM could be further used for evaluating of the effect of SWC measures on soil erosion and sediment delivery in the Ethiopian highlands. In this study, WATEM/SEDEM was applied to simulate the effect of alternative SWC scenarios at sub-watershed level to identify critical sediment source areas or erosion hotspots and to evaluate the effect of SWM strategies on soil erosion and sediment delivery. The objectives of the present study were: (1) to quantify the spatial distribution of soil erosion and sediment delivery at sub-watershed scale; (2) to evaluate the effect of different SWC measures on soil erosion and sediment yield and (3) to determine the most effective set of SWC strategies using scenario analysis.
  1. Materials and methods
2.1 Study area
Koga catchment is located south of Lake Tana, at the source of the Blue Nile, in the highlands of North-Western Ethiopia (37002” - 370 17” E longitude, 11010” - 110 25” N latitude; Figure 1). The Koga catchment is a narrow and elongated catchment, which has a concentrated networks of water divides, with highly variable and rugged topography. The total area of the catchment is 230 km2 with elevations ranging from 1860 to 3128 m a.s.l. In this catchment, the annual rainfall pattern is unimodal and rainfall mostly occurs between June and September. The average annual temperature and rainfall are 18.4oC and 1480 mm, respectively. Approximately 86% of Koga catchment is cultivated land, while around 12% is forest and the remaining part fallow and grazing land. Koga catchment represents a typical Ethiopian sub-humid highland environment where SWC measures have been implemented on a massive scale to reduce the effect of soil erosion and sedimentation in downstream areas and reservoirs (Mekonnen, Keesstra, Baartman, Ritsema, & Melesse, 2015). Three study sub-watersheds, Asanat, Debre Yakob and Rim, with drainage areas of 755, 303 and 1010 ha, were respectively selected at the upper, middle and lower reaches of Koga. Asanat is a hilly environment where more than 55% of the area has slopes of 15%-30% and ~11% of the area has slopes greater than 30%. In Debre Yakob 32% of the area has slopes of 15%-30% and about 33% of the area has slopes of less than 10%. Rim sub-watershed is relatively flat with 85% of the area having slopes of less than 10%. Approximately, 72% of Asanat, 64% of Debre Yakob and 72% of Rim is cultivated land while around 12%, 18% and 55% is used for grazing in Asanat, Debre Yakob and Rim respectively. The study sub-watersheds drain to Koga river which in turn drains into Lake Tana. The soil types in Koga are classified as Leptosols, Luvisols, Nitosols, Vertisols and Fluvisols. At the lower elevations of the catchments, Luvisols are the dominant soil type; these areas are well-suited for agricultural production. Leptosols are the predominant soil type in the upper part of the catchments; these soils are less suitable for crop production.
Fig. 1 approximately here
  1. Field sampling and measurements
Rainfall and stream discharge were automatically measured at the outlets of study sub-watersheds at 10-minutes intervals using Hobo data logger rain gauges and pressure transducer divers, respectively for periods 2016-2017. Thus stream discharge was estimated from depth measurements using rating curves calculated from the monitoring diver stations (Jemberu et al. sub.). One-litre SSC samples were taken manually for all rainfall events during the rainy season of 2016 and 2017. Sampling during a rainfall event started when the discharge developed and when the water at each outlet looked brown. About 3-8 representative samples were taken for each rain event based on fluctuation of flow depth. Due to large discharge, it was often impossible to sample sediment from an entire water column. However, since the flow was very turbulent during those events, a good mixing of sediments was observed from the brown colour of storm water for the rising and receding limbs of the flood so we assumed samples were representative for the full water column. Each sample was filtered using Whatman filter paper with a pore opening of 2.5 µm, oven dried and weighted to allow determination of dry sediment mass. A total of 101 one-litre samples for Asanat, 98 for Debre Yakob and 105 for Rim were taken during the rainy seasons of 2016 and 2017. The sediment yield (SY), in tonnes per day, for the stream’s cross-section was then obtained by multiplying the concentration, C (g/l) by the discharge Q (m3/s) (N. Asselman, 2000; Moliere, Evans, Saynor, & Erskine, 2004; Morehead, Syvitski, Hutton, & Peckham, 2003).
SY = C * Q * 86.4 (1)
Where 86.4 is a factor to convert to ton/ha.
2.3 WATEM/SEDEM model description
The WATEM/SEDEM model was developed to predict sediment yield for different catchment scales with limited data requirements (Van Oost, Govers, & Desmet, 2000; A. J. Van Rompaey, Govers, & Puttemans, 2002; A. J. Van Rompaey et al., 2001). WATEM/SEDEM is a sediment delivery model that calculates how much sediment is transported to the river channel on an annual basis (Van Oost et al., 2000; A. Van Rompaey et al., 2005; A. J. Van Rompaey et al., 2001). It is a spatially distributed model; for each grid cell, mean annual soil erosion and mean annual transport capacity are calculated (Van Oost et al., 2000; A. J. Van Rompaey et al., 2001; Verstraeten, Oost, Rompaey, Poesen, & Govers, 2002). WATEM/SEDEM comprises of soil erosion and sediment transport capacity assessment, and sediment routing processes (Van Oost et al., 2000; A. J. Van Rompaey et al., 2001). The model enables exploring the spatial pattern of sediment sources, erosion hotspot areas and annual sediment delivery. The effect of various existing or new SWC measures can be evaluated by the way they impact on spatial patterns, rates and processes of soil erosion (Didoné et al., 2017; A. J. Van Rompaey et al., 2002).
The model calculations are based on a spatially distributed assessment of mean annual soil erosion using the Revised Universal Soil Loss Equation (RUSLE) and mean annual sediment transport capacity (TC) (de Vente, Poesen, Verstraeten, Van Rompaey, & Govers, 2008; Desmet & Govers, 1996; A. Van Rompaey et al., 2005; Verstraeten & Poesen, 2000, 2001). An adapted version of the RUSLE (Renard, 1997) is used:
SE = R * K * LS (i, j) * C * P (2)
Where:
SE = Mean annual soil loss (kg m-2y-1)
R = Erosivity factor (MJ mm m-2h-1y-1)
K = Erodibility factor (kg h MJ-1mm-1)
LS (i, j) = Two-dimensional slope gradient and slope length factor of (i, j) coordinates
C = Crop management factor
P = Erosion control factor
The two dimensional slope length and steepness factor LS (i, j) were calculated based on an algorithm proposed by Desmet and Govers (1996):
LS(i,j) =[(Ai,j+D2)m+1-A(i,j)m+1(6.8Sg(i,j)0.8)]/Dm+2X(i,j)m(22.13)m(3)
X(i,j) =sinα(i,j)+cosα(i,j) (4)
Where:
A(i, j) is the runoff contributing area at the inlet of a grid cell (m2); D is the length of the side of a grid cell (m); Sg (i,j) is the slope gradient of the grid cell (i.j); α(i,j) is the aspect of the grid cell (i,j); and m is the slope length exponent.
This two-dimensional approach of the RUSLE not only accounts for inter-rill and rill erosion but also for smaller ephemeral gullies as the effects of flow convergence are explicitly accounted for (Desmet & Govers, 1996).