2.3.2 WATEM/SEDEM input data
The rainfall erosivity factor (R) was derived from the relationship between annual rainfall and annual erosivity expressed in MJ mm m−2 h−1 y−1. R expresses the aggressivity of rainfall and kinetic energy (KE) is generally suggested to indicate the ability of a raindrop to detach soil particles from a soil mass (Nearing et al., 2005; Renard, 1997). R was calculated using the following formula proposed by Nearing et al. (2005); Renard (1997):
R=\(\frac{\sum_{i=1}^{j}{\left(E*I30\right)i}}{N}\) ( 8)
Where,
E = the total storm energy (MJ)
I30 = the maximum 30-minute intensity (mm h-1) of daily rainfall i,
j = the number of rain events in an N year period
N = the number of observation years.
Total storm energy was determined using:
E = 1.213 + 0.89 log I ( 9)
Where,
E = is kinetic energy of each rainstorm KJ m-2mm-1
I = is average intensity of daily rainfall i.
The erodibility (t h MJ-1 mm-1) of the soil was calculated from soil properties using the following equation proposed by N. Hudson (1993); Renard (1997); Wischmeier and Smith (1978):
K = [(2.1 * 10-4 M1.14 (12-OM)) + 3.25 (S-2) + 2.5 (P-3)] / 7.59 (10)
Where,
M (the textural factor) = (%silt + %sand) * (100 - %clay);
OM= % organic matter
S and P are field-determined average values of aggregate structure and permeability classes described as follows:
S = Aggregate/structural class with values (1-4): 1 for very fine, 2 for fine, 3 for medium coarse, 4 for massive structure respectively, and
P = Permeable class with values (1-6)]: 1 for fast, 2 for fast to moderate, 3 for moderate, 4 for low to moderate, 5 for low, and 6 for very low permeability respectively.
The remaining RUSLE parameters, crop management (C) factor and erosion control factor (P) factor, were consulted from the literature (H Hurni, 1985; Renard, 1997). Annual values of the C-factor were determined based on the land use types defined by a previous study (H Hurni, 1985), and spatially attributed based on the current land cover (Renald, 1997). The P–factor values were determined based on the types of SWC measures implemented in different areas (Table 1). The P-factor represents the ratio of soil loss with conservation measures to a reference plot without conservation measures: a value of one refers to a cultivated land without conservation practice.
Table 1 approximately here
The P-factor considers the interaction of LS (i,j) attributes and vegetation cover (C-factor), as well as the direction of flow and TC, which depends on the type and effectiveness of physical conservation measures (Foster, 2002; Renard, 1997). Therefore, P–factor and C-factor values were used in the WATEM/SEDEM model to verify the responses of different SWC scenarios to soil erosion and sediment yield in the study sub-watersheds. The P-factor values for the different supporting conservation practices were adoptable to local study sites environmental contexts (Foster, 2002; Renard, 1997; Wischmeier & Smith, 1978) and thus the compounding P-factor for different scenarios of SWC measures were calculated as follows:
P =PB * PCC * PSC * PGS (11)
Where PB, PCC, PSC and PGS are conservation practice sub-factors for bund structures, contour cultivation, strip cropping and grass strips respectively (Table 1). The Parcel trap efficiency in WATEM/SEDEM, Ptef, refers to the way each pixel’s runoff contribution to the upstream contributing area is reduced. This means that for different land use types, less runoff is simulated, thereby decreasing downstream LS (i,j) values and erosion rate by Ptef (Van Oost et al., 2000; A. J. Van Rompaey et al., 2001). Ptef values of 10 for cultivated lands and 75 for pasture and forests were chosen based on the optimal or KTC(h) during model calibration.
A DEM with 20m resolution was derived from global SRTM topographic data by resampling 30 m resolution and the land use map was created from Landsat GLSTM_2000 image. Besides, a parcel map was created by the model by combining the DEM, stream, road, forest, arable land, pasture and catchment area delineations to account for the effect of landscape structure on soil erosion and sedimentation processes (Van Oost et al., 2000). A parcel map is a reclassified land use map that takes a distinction between, arable land, forest, pasture, roads, infrastructures, rivers and build-up areas. This makes it possible to incorporate the effect of field borders on runoff diversion, runoff interception, erosion and sediment deposition (A. Van Rompaey, Krasa, & Dostal, 2007; A. J. Van Rompaey et al., 2001).
2.4 SWC scenarios
The SWC scenarios evaluated in this study were developed based on a biophysical inventory, farmers’ perception (Jemberu, Baartman, Fleskens, & Ritsema, 2018) and the regional government’s five year strategy program in the study sub-watersheds to select promising conservation strategies. Three conservation strategies, including physical (bund structures), agronomic (strip cropping and contour cultivation) and vegetative (grass strips) measures were created for the three sub-watersheds of Koga catchment to determine where each type of or combination of SWC measures can be implemented. Consequently, soil erosion, sediment deposition and sediment yields were simulated for five alternative scenarios of SWC measures described as follows. Scenario I: a baseline condition (present-day situation) including existing bund structures; Scenario II: existing bund structures and contour cultivation; Scenario III: combination of bunds, contour cultivation and strip cropping; Scenario IV: integrated use of bunds, contour cultivation, strip cropping and grass strips; and Scenario V: a scenario without SWC practices (Table 2).
Table 2 Approximately here
2.5 Model calibration
We calibrated and validated WATEM/SEDEM for this study based on sediment yields measured at the three study sub-watersheds of Koga catchment from 2016 to 2017. The sediment yield data of 2016 was used for calibration and that of 2017 to validate the performance of the model. The model was calibrated based on area-specific and absolute sediment yield (by minimizing the difference between measured and simulated values) since the objective of the study was to assess the effect of SWC measures on soil loss and sedimentary processes for various SWC scenarios at sub-watershed level. The model was first calibrated for the baseline scenario (scenario I) by changing the maximum and minimum values of the KTC or KTC (h) and KTC (l). The model was sensitive to KTC (h) and less sensitive to KTC(l) whereas the model was insensitive to the threshold KTC(t) value. The KTC(t) was set at 0.1 for arable and 0.01 for non-arable land uses for all sub-watersheds. After calibration, the model was run for the four remaining SWC scenarios.
The Nash-Sutcliff efficiency (NSE) statistic was applied to evaluate the efficiency of the model. NSE is a normalized statistic determining the relative magnitude of residual variance (between predicted and observed values) compared with the measured data variance. NSE values between 0 and 1.0 demonstrate model efficiency; the closer the value of NSE approaches 1, the more efficient is the model. A satisfactory model should have NSE >0.50 (Baartman, Jetten, Ritsema, & Vente, 2012). The accuracy of the model in predicting soil erosion for these scenarios was also assessed qualitatively by relating to previous studies on soil erosion (Jemberu et al., 2018) and available literature for similar areas (Bewket & Sterk, 2003; Herweg & Ludi, 1999; Mitiku et al., 2006; Nyssen et al., 2010).