Ijsini John Tekw , John Mthew Lflen , Aukr Mus Kundiri ,Adullhi Bl Alhssn

a Department of Soil Science & Land Resources Management, Federal University Wukari, P. M. B,1020, Wukari Taraba State, Nigeria

b USDA-ARS, National Soil Erosion Research Laboratory, Purdue University, Lafayette, IN, USA

c Department of Soil Science, Faculty of Agriculture, University of Maiduguri, P. M. B,1069, Borno State, Nigeria

Keywords:Empirical EGEM WEPP Ephemeral gully erosion Soil loss Mubi Northeast Nigeria

ABSTRACT Evaluation of prediction models is crucial to achieving valid information on erosion processes and their management choices. WEPP model efficiency in predicting ephemeral gully (EG) erosion was recently tested and compared with both EGEM and empirical models. The models abilities to predict EG erosion were validated using measured estimates at the 6 eroding locations around Mubi area in Northeast Nigeria between April 2008 and October 2009. Each location consisted of 3 watersheds where data on soils, climate, slope, management practices, EG shapes and dimensions were collected. Data on relevant soil properties were collected in the field and then analyzed in laboratory. The mass of soil loss (MSL)predicted by empirical,EGEM and WEPP models were compared with the measured using paired T-test,regression graphs (r2-values), error analysis, and analysis of variance (ANOVA) in a completely randomized design. The EG erosion losses varied significantly (P ≤0.05) between sites and years. No significant(P ≤0.05)differences were observed between measured and the empirically predicted aggregate MSL.The measured aggregate MSL strongly correlated with those predicted by empirical(r2=0.67),than with EGEM(r2=0.57),and WEPP(r2=0.53)models.Slight over and under-prediction instances against the measured erosion were noted with all the models. The WEPP model was found to slightly overpredict MSL when compared to either the empirical or EGEM model. The prediction quality of the models was generally impressive.Future works should focus more on local inputs such as climate,plants,management, and tillage data for use with WEPP.

1. Introduction

In recent decades,application of erosion prediction technologies and management strategies have increased and dominated larger parts of land degradation reports around the World. Despite these volumes of erosion reports,there is still insufficient information on ephemeral gully (EG) erosion processes in most parts of the sub-Saharan Africa, especially Nigeria. Presently, only few predictive studies using foreign models such as Water Erosion Prediction Project (WEPP) to predict EG erosion were carried out in Nigeria(Mbajiorgu&Ogbu,2011;Obeta&Adewumi,2013).Unfortunately,the few studies have not evaluated nor compared WEPP with other models in this part of the world as detailed by this study.The dual objectives of this work was firstly, to determine the WEPP model prediction efficiency using little EG site calibration, and, secondly,to compare its estimates with earlier erosion results obtained from both Ephemeral Gully Erosion Model (EGEM) (Tekwa et al., 2014)and empirically predicted erosion (Tekwa et al., 2016) in the same study area of Mubi, northeast Nigeria. The EG erosion is a recognized class of water erosion(Foster,1986,pp.90-118),which is also responsible for high losses of soils and soil fertility on agricultural farmlands (Laflen et al., 1986; Lal, 2001). The erosion process is widely influenced by climate,topography,and vegetation(Capra&Scicolone, 2002; Lal, 2001; Oygarden, 2003). Researches on EG erosion from different climates and land use practices shows that EG erosion accounts for between 10% and 94% of total soil loss on European farmlands(Poesen et al.,2003).Soil loss ranges from 2 to 90 m3ha-1in the Mediterranean areas annually (Capra &Scicolone, 2002). The choice of WEPP (Flanagan et al., 1995, pp.1-12; Foster & Lane,1987) was based on its simulation ability to precisely compute daily,weekly,monthly or annual values of runoff and soil erosion in environments (Laflen et al., 2004) such as the Mubi hilly terrains.WEPP model comprises of a climate generator,soil/tillage data,terrain,channel process information,plant growth,and management components that can predict hill-slope and channel erosion on watersheds (Flanagan et al., 2007; 2012).

The watershed version of a WEPP model was chosen among other recent models such as Revised Universal Soil Loss Equation(RUSLE 2), based on WEPP’s unique ability to predict individual storm erosion on daily basis than RUSLE 2. WEPP also has advantages over both EGEM and empirical models for its user friendliness, and wider range of applicability. Though, WEPP is a more complex model, but can suitably predict average annual erosion from concentrated flow channels as EGEM,which is restricted to 45 inches application depth. Tiwari et al. (2000) earlier evaluated the WEPP model efficiency using measured data and compared it against the popular Universal Soil Loss Equation (USLE) and its revised version(RUSLE).Results show that WEPP had slightly lower efficiency of 0.71 compared to 0.80 and 0.72, respectively for empirical (USLE) and RUSLE.

Conversely, the EGEM is a model that was developed in the United States (US) for use by the US Department of Agriculture(USDA) and National Resource Conservation Service (NRCS) to predict EG erosion (Woodward,1999), while the empirical model estimates was from the results of Tekwa et al. (2014). Most model evaluations show spurious correlations between measured and empirical EG erosion,than with EGEM estimates(Capra et al.,2004;Capra & Scicolone, 2002). Generally, empirical models perform better under their host environments, when compared to their adapted counterparts from foreign places.

It is our expectation that there will be several obstacles to using the rare WEPP model under local conditions of Mubi, Nigeria. It is hoped that model adaptation obstacles will be discovered and procedures developed for use of the WEPP model. This research is therefore, targeted at evaluating the scarcely applied WEPP model and to compare its efficiency against the tested EGEM and empirical models towards recommending the best adoptable EG erosion prediction tool for the Mubi area in Northeast (NE) Nigeria, and eventually to all of Africa.

2. Materials and methods

2.1. Description of the study area

The study was carried out at 6 selected erosion prone locations(Digil, Vimtim, and Muvur, then Gella, Lamorde and Madanya)(Figs. 2-4) in the Mubi area of Adamawa state, NE Nigeria(Fig.1),with comparable vegetation cover, topography, soils and land use types.Mubi area is characterized by dry and wet season conditions.The dry season starts from November to April,while the wet season starts from May and cease in October.The annual rainfall amount is between 700 and 1050 mm(Adebayo,2004,pp.32-38;Udo,1970,pp.195-197),while the minimum temperature averages 15.2°C in both December and January. The driest months are March and/or April with a maximum temperature of about 42°C (Adebayo &Tukur,1999, pp. 92).

The area is typical of savannah vegetation with notably grasslands and sparse trees (Adebayo, 2004, pp. 32-38; Adebayo &Tukur,1999, p. 92; Tekwa & Usman, 2006). Agricultural practices such as arable farming and cattle rearing are the most predominant activities around the Mubi area, and which are consequently menaced by water erosion.

2.2. Soil sampling and analysis

Soil samples were first taken from 3 sampling points at each of the 3 EG channels,then composited and tagged to represent each of the 3 sites selected from each of the 6 study locations in relation to their pattern of land-use and erosion types. Surface soil samples were taken at 0-15 cm depths in a transverse pattern using a bucket soil auger, when the soils were moist. The composite soil samples were kept separately in well labeled polythene bags.Firstly, the soil samples were air-dried, then crushed and later sieved through a 2 mm sieve before being analyzed for some relevant physical and chemical properties of the soils.

2.3. Determination of selected soil properties

The Bouyocous hydrometer method(Trout et al.,1987)was used to determine the distribution of soil particle sizes, while Clod method was employed to determine the soil bulk density (Wolf,2003). The soil erodibility indices (SEI) were computed from 18 soil textural classes with their corresponding SEI-values as reported by Mitchell and Bubnezer (1993). The potassium dichromate wetoxidation method of Walkley and Black (1934) and Wolf (2003)was used to determine the soil organic carbon (OC) content. Conversion of O.C into organic matter (OM) content was done by multiplying with a factor of 1.724(Wolf, 2003).

The estimate of soil shearing strength was obtained following a method reported by Smerdon and Beasley (1961), expressed as:

where τc = critical shear stress.

2.4. Collection of EG channel properties and rainfall amounts

The EG channel parameters studied between April 2008 and October 2009 are as shown in Fig.5 and Tables 1 and 2.Over time,the few conservation practices at the sites have significantly minimized changes in the EG channels. The actual EG lengths,channel widths and the maximum EG depths were measured with a meter tape in the field. The difference between the initial and maximum EG widths were also termed as the depleted width(Capra et al., 2004). The slope rates (SR) of the watersheds were determined with an Abney level as described by Tolu(1996,p.162).The EGEM software program was used to auto-generate the volume of run-off water received at each EG site from local rainfall data of the study area. The EG channel parameters (average lengths,widths,and depths)are shown in Table 1.The results show that the shortest EG length (71.68 m) was at Lamorde and the longest(173.94 m)and deepest(0.37 m)channels at Muvur site.The widest EG channel width (3.25 m) was at Vimtim and the narrowest(1.23 m) and shallowest (0.20 m) channels were at Gella site.Similar occurrence was observed at Muvur with the deepest and longest EG channels.The short nature of EG length at Lamorde with rockiness had its width expanded considerably. The soil textural class (silty clay loam) of Lamorde, perhaps, still increased the erosion, rather than just due to the EG length effect. This is usual with predicting erosion from rocky areas as earlier reported by Capra et al. (2004) and Tekwa et al. (2013; 2014 & 2015).

2.4.1. Measurement of observed MSL in the study area

The MSL was computed after both the area and volume of soil losses were determined.The areas of soil loss(ASL)were computed in terms of the EG cylindrical and cone shapes, respectively, and expressed (Equations (2) and (3)) as:

Fig.1. Map of the study area showing the 6 villages where EG erosion channels occurred.

Fig. 2. Map of EG channels at (a). Digil, and, (b). Vimtim study sites.

Fig. 3. Map of EG channels at (a). Muvur, and, (b). Gella study sites.

Fig. 4. Map of EG channels at (a). Lamorde, and, (b). Madanya study sites.

Fig.5. Schematic diagrams of EG erosion showing,(a).EG erosion channel formed on a sloping intersectional watershed areas,and,(b).erosion processes describing a developing EG channel with an actively migrating head-cut in the upstream direction.Source: Adapted from Watson et al. (1986); and Gordon et al. (2007).

where:h=perpendicular height of EG head from an imaginary axis(5 m adopted).

r = radius of cylindrical EG shapes, l = length of EG

r = radius of an EG head-cut area, π = constant of proportion,The total ASL=Net area of EG cylinder shapes+Net area of EG cone shapes

where:δb= bulk density.

2.4.2. Prediction of MSL using empirical equation

Linear empirical equation modeled from quantitative field data in the same Mubi area as reported by Tekwa et al. (2013 & 2014)was adopted in this study. The linear regression equation is as shown in equation (4).

Table 1 Site selection criteria and field characteristics.

Table 2 Mean values of EG erosion channel parameters.

YMSL= 2666.99-1899.59 (δb) - 6.93032 (clay) + 1124.52(SEI) + 17.2004 (PL)

where YMSL= predicted area of soil loss, OM = organic matter content,δb=bulk density,τc=shear strength,Clay=clay content,SR = site slope rate, SEI = erodibility index, Run-off = volume of run-off water, PL = plasticity index, r2= coefficient of determination.

2.4.3. Prediction of MSL using the EGEM model

The field identification component of an EGEM data input phase is comprised of information on the study location, altitude,topography, vegetation, climate, soil map, agriculture and human activities.The EGEM program also provides for entry of researcher’s identification data such as name of client, county, state, cultural practice,and date(Foster,1986,pp.90-118).The soil drainage was obtained from the curve number (CN) values. The Manning roughness coefficients were adapted from values used for tropical soils (SCS, 1992) in relation to soil type, clay content and tillage practices noticed in the study area.The particle detachment rates of the eroding soils were auto generated by EGEM software,while the volume of run-off water received at the EG sites were computed from rainfall data fed into the EGEM.

A field survey with a GPS modem was used to measure the watershed parameters such as watershed length (represents the maximum potential length of EG) (Nasri et al., 2008), while an Abney level device was used to measure the watershed slope(Tolu,2002). A meter tape was used to measure the Concentrated flow lengths (actual length of the EG) and the EG maximum depths.EGEM program limits maximum EG eroded depth to a tillage depth of 45 cm (18 inches). Depleted widths were obtained as the differences between the initial and maximum EG depths(Capra et al.,2004). The hydrologic soil group was determined using a method prescribed by USDA(Philips&Joubert,2009,pp.1-18).The rainfall data was the sum of the 24-h rainfall events received in the locations. Initially, EGEM used rain storm distributions (types I, IA, II and III)for different climates earlier designed for use mainly within the US environment (Capra et al., 2004).

The 24-h rainfall represents the amount of rainfall received in a manual rain gauge between 9.00 a.m. (the first day) and 9.00 a.m.(the next day)giving a total of 24 h duration.

The 2 year 24-h and 25 year 24-h rainfall events were computed using a method described by the Pennsylvania State Climatologist(PSC) (2009), expressed as:

where:X =extreme rainfall value.

ψ = average - γ β (where γ is Euler’s constant, approximately 0.557)

β = 0.78σ (where σ is the standard deviation)

F = (n - 1)/n (where F is the rainfall event period, and n = number of years)

ln = natural log.

The percent erosivity index (PEI) was obtained in line with the method reported by Lal (1983). The PEI was computed in terms of the percentage of days with erosive 24-h rainfall (＞20 mm)expressed over the total rainfall days in a season,and expressed as:

The EGEM model software (EGEM version 2.0) estimated soil loss in terms of voided area(acres)and eroded volume(tons)in this study. The soil losses are then computed in terms of the MSL.

2.4.4. Prediction of MSL using the WEPP model technology2.4.4.1. Description of the erosion driving forces and processes.

WEPP is a simulation model that computes daily values of runoff and soil erosion. It is comprised of a climate generator, plant growth,tillage, and residue management components. The WEPP model uses the steady state sediment continuity equation to estimate soil erosion and deposition processes.A description of these physically based processes occurring in channels, hillslopes or watersheds erosion when using a WEPP model are centered on 2 notable physical activities:soil detachments by shear stress;and sediment transportation by concentrated flow energy. Other factors such as plants and tillage or management methods are as well implied.

Firstly,erosion prediction in a WEPP model starts with detailing soil parameterization.Foster et al.(1995),Flanagan et al.(1995,pp.1-12)and Mbajiorgu and Ogbu(2011)suggested few mathematical equations for computing soil detachments and subsequent sediment deposition processes. Soil detachment by prevailing driving forces can be obtained using equation (7) (Foster et al. 1995)expressed as:

wherex= distance downslope (m), G = sediment load (kg/s/m2),Di=interrill erosion rate(kg/s/m2),andDf=rill erosion rate(kg/s/m2).

Hence, the soil sediment transport capacity by concentrated flow process was also given by Flanagan et al. (1995, pp.1-12) as expressed in equation (8).

where

Ktr

=constant rill erodibility parameter,

qw

=flow discharge per unit width (m

2

/s),

S

= slope (%).

In the instance where soil sediment load exceeds its transportation capacity,a net deposition occurs,and is calculated using equation (9) as suggested by Flanagan et al. (1995, pp.1-12).

whereVf= effective fall velocity for the sediment (m/s), and q = the flow discharge per unit width (m2/s).

Secondly,the hydrological components that moderate sediment deposition processes in a WEPP model are three folds:(a).effective rainfall intensity, and (b). effective runoff duration, and (c). peak runoff per unit area (Foster et al.,1995). As a process, the climate generator (CLIGEN) component of WEPP generates the rainfall intensity,while the hydrologic component of the WEPP computes the runoff peak and duration.The effective runoff duration is obtained using equation (10).

wheretr= effective runoff duration (s),Vt= total volume for rainfall event (m), andPr= peak runoff per unit area (m/s). The effective rainfall intensity used to compute interrill soil losses are given by equation(11)(Foster et al.,1995;Mbajiorgu&Ogbu,2011).

whereIe=effective rainfall intensity(m/s),t=time,te=total time(s) for which rainfall exceeds infiltration rates.

2.4.4.2. The WEPP model calibration.Since there were no existing databases for WEPP in NE Nigeria, the US databases for climate,soils,plants,tillage and managements were inspected to find those with the best matching conditions in terms of climate, soil, plants and tillage data for the Mubi area, Nigeria.

2.4.4.3. Climate data.The Temple Texas in the US was chosen,whose average climate, particularly temperature and annual rainfalls had a similar distribution to that of the Mubi area,NE Nigeria.The Cligen climate generator was used to generate 100 years of daily precipitation,temperature,solar radiation,and wind velocity for Temple, Texas and was used as an approximation of Mubi Climate in the WEPP program. WEPP runs for the EG erosion estimation were for 100 years long.Average annual computed channel erosion was used as the EG for the channel.

2.4.4.4. Soil data.The Temple Texas soil database inbuilt in WEPP model, with soils that had the same textural classification as the Nigerian soils were adapted as a fair match to the Mubi soils in this study.

2.4.4.5. Plants data.Plants modeled included Maize (Zemays),Cowpea(Vigna unguiculataL.),Guinea Corn(Sorghum bicolor), and Rice(Oriza sativaL).US parameters were used for Maize,but were adjusted for lower yields from options in the WEPP model that reflects yields found in the Mubi area. Sorghum was used for Guinea Corn, and a small grain was used as a substitute for Rice,while soybean was used for cowpea.These plants were grown in a 4 year rotation. Plant parameters such as stem girth, number of leaves and grains per plant were set to provide yields expected for the Mubi area. Planting and harvest dates were typical of local farming around the study area.

2.4.4.6. Tillage data.The tillage parameters used were typical of those used in the US corn-belt;disking and field cultivation usually between the months of May and June as occurred in Mubi area.These were expected to give similar residue burial and surface roughness values to those expected with the farming and tillage methods used in Nigeria.

The watershed version of WEPP was used. The EGs were modeled as channels that were supplied with water from both sides of the channel all along the length of the EG.The lengths and widths of the channels and the area along each side of the channel were based on the measurements reported by Tekwa et al. (2013).WEPP computes runoff volumes and rates to a channel, and the erosion due to detachment in a channel. It also computes erosion from the area along each side of the channel, and delivery to the channel as detailed in equations (7)-(11).

2.5. Data analysis

The data collected was analyzed using the generalized linear model in a completely randomized block design for the ANOVA using Statistix 9.1 (2012). Bar graphs (histographs) were used to compare the predicted MSL with the measured MSL from EG erosion. Regression graphs (scatter diagrams) were also used to validate the relationships between the measured and predicted erosion. In addition, analysis of errors in predicting the MSL by empirical,EGEM,and WEPP models,plus contrasts between WEPP versus EGEM and empirical model prediction efficiencies were determined using the standardized mean error (Mes) and root mean square error (Mse) as described by Capra et al. (2004), and expressed as:

where:S=standard deviation of the measured MSL,n=number of observations.

Zi=empirical MSL estimate,and,Zi*=measured MSL estimate.

3. Results and discussion

3.1. Erosion site characteristics

The results on erosion site characteristics are presented in Tables 1 and 2. The results show that the EG channels at the watersheds are widely heterogeneous with some of the EG channels eroding into a“V”or“U”shapes initiated from yearly channel floor incisions by run-off water on a largely rolling topography (Tekwa et al., 2013 & 2014). Results also reveal the presence of fewer trees and grass vegetation around Mubi North area (e.g. Vimtim and Digil)with verse farmlands compared to Mubi South area(e.g.Gella and Lamorde) with more rock outcrops which often limits agricultural tillage activities. Few erosion control practices such as terraces,rough tillage,vegetative barriers,and tied-ridges are used around Mubi area(Ekwue&Tashiwa,1992;Tekwa et al.,2014).The soils are predominantly sandy clay loam, except for Gella and Lamorde where the soils have sandy loam and silty clay loam textures,respectively.The soil OM content was low(＜2%)and appears inadequate to significantly limit erosion losses(Renard et al.,1997)as observed in the study area.

3.2. The MSL estimates in the study area

3.2.1. Measured MSL estimates

Fig.6 presents the results of measured versus the predicted EG erosion estimates using the EGEM, WEPP and Empirical models.The measured MSL generally differed significantly (P ≤ 0.05)among the study sites. The highest estimated MSL (1.75 tons/yr)was recorded at Lamorde, while the lowest (0.43 ton/yr) at Gella.The corresponding values recorded at Vimtim, Digil, Muvur and Madanya were 1.24,1.01,0.84 and 0.62 tons/yr,respectively.These values were higher than the MSL estimates reported on surface erosion by Obeta and Adewumi (2013). The observed variation could have been due to the differences in the soils and erosion features studied.

3.2.2. Predicted MSL (empirical, EGEM and WEPP model estimates)

The predicted MSL by the empirical, EGEM and WEPP models are presented in Fig. 6. The results showed that significant(P ＜0.05)difference exists in the MSL predicted using the 3 models.The highest estimate was still found at Lamorde with 1.43 tons/yr,followed by Vimtim (1.13 tons/yr) and Muvur (1.05 tons/yr). In addition, Digil (0.90 tons/yr), Madanya (0.63 tons/yr) and Gella(0.59 tons/yr) had relatively lesser erosion estimates. Statistically,no significant(P ≤0.05)differences were however,found between erosion values at Lamorde and Vimtim and between Gella and Madanya sites.These similarities in soil losses were earlier reported by Ekwue and Tashiwa(1992),and were perhaps influenced by the individual runoff amounts received in the sites.

Fig. 6. Estimates of measured and predicted (Empirical, EGEM and WEPP models) aggregate MSL in the 6 study sites.

The predicted MSL by the EGEM model exhibited significant(P ≤0.05)differences among the study sites as also shown in Fig.6.Lamorde still had the highest MSL (1.37 tons/yr), while Gella had the least estimates(0.37 tons/yr).MSL estimates at Muvur was 1.15 tons/yr, followed by Vimtim (0.84 tons/yr). The estimated MSL at both Digil(0.59 tons/yr)and Madanya (0.65 tons/yr)did not differ significantly(P ≤0.05)in this study.Nactergaele et al.(2001),Capra et al. (2004), and Tekwa et al. (2015) among others, similarly observed significant (P ≤0.05) differences between observed and EGEM estimates of MSL.

The MSL predicted by the WEPP model was the highest at Vimtim(1.90 tons/yr),while the lowest estimate(0.80 tons/yr)was recorded in Digil and Gella sites (Fig. 6). The predicted erosion at Lamorde trailed with 1.76 tons/yr,followed by Muvur(1.50 tons/yr)and Madanya (0.83 tons/yr). Unlike both EGEM and empirical models, no significant (P ≤0.05) difference was noted between WEPP predicted erosion at Vimtim and Lamorde,and at Madanya,Digil and Gella sites. The range of annual MSL observed(0.80-1.90 t/ha/yr) compares much higher than the average MSL(0.059 kg/m2) recorded for a single storm event from a plot in Nsukka,Nigeria(Mbajiorgu&Ogbu,2011).In addition,Tiwari et al.(2000) mentioned that WEPP model is not consistent in over or under prediction of soil loss on watersheds.

3.3. Validation of the predicted MSL

3.3.1. Validation of empirical MSL estimates

The results of student T-test analysis of measured and empirically predicted erosion are presented in Table 3. The differences between measured and empirical MSL were slightly over predicted by 0.01 tons/yr (or 2%), 0.16 tons/yr (or 37%) and 0.21 tons/yr (or 25%) respectively at Madanya, Gella and Muvur sites.Conversely, Digil, Vimtim, and Lamorde had their MSL slightly under predicted by 0.11 tons/yr (or 11%), 0.11 tons/yr (or 9%) and 0.32 tons/yr (or 18%), respectively. However, no significant(P ≤0.05) differences were observed between the measured and empirical MSL in all the sites.These results strongly underscore the high potentials of the empirical model in the study area.Nasri et al.(2008) reported that there exists a spurious correlation between actual and empirically predicted erosion.Similar relationships were as well reported by Nactergaele et al. (2001b) and Capra et al.(2004).

3.3.2. Validation of EGEM MSL estimates

Table 3 still presents the measured and EGEM predicted MSL mean differences and were between 0.03 tons/yr (or 5%) at Madanya and 0.42 tons/yr (or 42%) at Digil. The MSL was slightly over predicted at Madanya and Muvur by 0.03 tons/yr (or 5%) and 0.31 tons/yr (or 37%), respectively. All other sites had their MSL significantly(P ＜0.05)under predicted,except at Gella(0.06 tons/yr or 14%). Earlier work by Tekwa et al. (2014; 2015) reported similar over or under predictions with the EGEM model for the same Mubi area. The wide under and over prediction behavior of EGEM was as well observed by Capra and Scicolone (2002) and Foster et al. (1995).

Table 3 Efficiencies/validation of the 3 erosion models in predicting aggregate MSL using paired T-test and error analysis tools.

3.3.3. Validation of WEPP MSL estimates

The results of T-test analysis between measured and WEPP predicted estimates are presented in Table 3. The results reveal a widespread over-prediction behavior in all the sites,except at Digil with 0.21 tons/yr (or 21%) under prediction. There was no significant (P ＜0.05) differences between measured and WEPP model estimates at Digil,Lamorde,and Madanya,regardless of the over or under prediction incidences. In this study, estimates predicted by the WEPP model at Lamorde was the least difference(0.01 tons/yr or 1%),while Gella had the largest over prediction of about 86%.The mean difference between measured and WEPP estimates (1-86%)was higher than for both empirical(2-37%)and EGEM(5-42%)in this study. Similar event was reported by Tiwari et al. (2000) that when comparing measured and WEPP estimates of MSL,the WEPP overestimates low values and underestimates a majority of high values.Hence,Nearing(1998)earlier stated that the phenomena of erosion models overestimating small events and underestimating large events is inherent to all erosion models.

3.3.4. Validation of predicted aggregate MSL at all sites

A regression graph was used to test the relationships between measured and predicted erosion as presented in Fig. 7. The results showed that empirical estimates were better related to the measured erosion with r2= 0.67, followed by EGEM predictions with r2= 0.56. Even though, the relationship between measured and WEPP estimates was relatively lower (r2= 0.53), but was still strongly comparable with the EGEM estimates. This implies that WEPP model may have higher potential efficiency than EGEM,especially when indigenous variables such as climate, tillage and managements are fully used in the WEPP model. In a previous research by Tekwa et al.(2014),measured and empirical soil losses were strongly correlated (r2= 0.92), than with EGEM (r2= 0.55).Such relationships have been noticed with application of EGEM in other parts of the World (Capra et al., 2004; Nachtergaele et al.,2001; Nasri et al.,2008).

3.3.5. Validation of predicted MSL at the individual sites

Results on the relationship between measured and predicted erosion in the individual study sites are still shown in Table 3.The measured and empirical estimates exhibited a generally high relationship (87%-99%), except at Lamorde, which recorded moderate relationship (r2= 0.58). Empirical and measured estimates related well at both Digil and Vimtim by 99%, followed by Gella,Madanya and Muvur with 98%, 96% and 87% relationships.Nachtergaele et al.(2001)and Capra and Scicolone(2002)reported similar strong relationships between empirical and measured erosion in different Mediterranean and Belgium areas.

On the other hand, measured and EGEM estimates exhibited a better relationship than with both empirical and WEPP estimates in this study. Measured and EGEM estimates related by 99% at all 3 sites,in addition to 98%and 97%relationships at Madanya and Digil sites.Only Lamorde had the least relationship(r2=0.76)among the sites. This was perhaps, due to the site’s rockiness, which limits both of these foreign model accuracies. The observed relationship between measured and EGEM estimates was higher than the reported r2-values of between 0.40 and 0.61 by Capra et al. (2004).

The strongest (r2=1.00) and weakest (r2= 0.32) relationships between measured and WEPP predicted erosion estimates was respectively found at Digil and Vimtim sites. However, relationships between measured and WEPP estimates at Gella and Madanya sites also had high r2-values of 0.98 and 0.96, respectively.Likewise, both models were related by 87% (r2= 0.87) and 72%(r2= 0.72) at Muvur and Lamorde,respectively. WEPP predictions were however,inconsistent in contrast to those of both EGEM and empirical predictions in this study. This observation concurs with the reports of Tiwari et al.(2000)and Laflen et al.(2004)that while contrasting USLE, RUSLE and WEPP technologies, the WEPP predicted soil loss was better at few of the individual sites,than when averaged.

Fig. 7. Relationships between measured and predicted aggregate MSL by: (a). Empirical, (b). EGEM and (c). WEPP models in the 6 study sites.

3.4. Efficiency tests of the erosion models

3.4.1. Empirical model efficiency

Results on the prediction quality of the empirical model are also presented in Table 3. The models are theoretically consistent or efficient,if the standardized mean error(Mes)≡1.The root-meansquare-error (Mse) provides an effective measure of the accuracy(maximum accuracy), which was achieved with as low Msevalues as possible(Capra et al.,2004).The best prediction was at Madanya with Mseof 0.08, followed by Vimtim (Mse= 2.52) and Gella(Mse=9.48).Relatively lower prediction efficiencies were observed at Lamorde, Muvur and Digil with respective Mesvalues of 13.65,16.33 and 40.33. However, the test of maximum accuracy of the empirical model disclosed widespread of high reliability indices of between 0.01(at Madanya)and 0.18(at Lamorde).These Mesvalues confirm how empirical model best fitted to Mubi area.Earlier test of empirical model efficiency in the same Mubi area by Tekwa et al.(2016) reported similar efficiency with empirical models.Likewise, Tiwari et al. (2000) found a range of -2.88 -0.82,-10.54-0.85 and-18.05-0.94 in respect of RUSLE,USLE and WEPP models accuracies. Hence, the empirical models (USLE &RUSLE) were found to predict the average annual soil loss better,than the WEPP as was the case in this study.

3.4.2. EGEM model efficiency

The prediction efficiency of the EGEM model was generally lower than the empirical model as shown in Table 3. The EGEM prediction accuracy(Mes)indices ranged from 0.75(at Madanya)to 58.0 (at Digil). Madanya still compared with the best prediction efficiency as it was for the empirical model.Gella trailed with Mesof 1.33, followed by Lamorde with Mesof 19.25. Both Vimtim(Mes= 33.33) and Muvur (Mes= 35.59) had comparable EGEM prediction efficiencies,while the MSL at Digil(Mes=58.0)was the poorest predicted erosion using EGEM model. Conversely, the test of maximum accuracy (Mse) of EGEM predicted MSL were as well generally fair across the sites. The Msevalues ranged from 0.02 (at Lamorde) and 0.24 (at Digil). Madanya had its EGEM estimates predicted with an Mseof 0.02,followed by Gella with an Mseof 0.03.The Msevalues at Muvur trailed with 0.18,while Lamorde,Vimtim and Digil had comparable prediction efficiencies(Mse)of 0.22,0.23 and 0.24 respectively.Though,the EGEM prediction efficiency was lower than the empirical model, but it was still generally impressive. The observed EGEM efficiency appeals far better than those reported by Capra et al. (2004).

3.4.3. WEPP model efficiency

Results of test of WEPP model prediction efficiency presented in Table 3 and Fig.7 shows that the MSL prediction quality was lower than for both empirical and EGEM models. However, the highest and lowest prediction accuracies were respectively observed at Lamorde (0.01) and Muvur (161.33) in this study. Hence, the Mesvalues at Madanya,Gella,Vimtim and Digil were respectively 36.75,50.70,90.75 and 147.0,which were better than the indices reported by Capra et al. (2004). On the other hand, the test of maximum accuracy (Mse) of 0.01 at Lamorde (0.01) gave the best prediction.WEPP prediction accuracy at Digil and Madanya tallied with a better Mse=0.12 each,than for Vimtim and Muvur with accuracy of 0.38 each.Hence,Gella site had a prediction accuracy of Mse=0.21,being the intermediate prediction efficiency among sites studied.The observed prediction efficiencies compares closely with estimates of 0.58, 0.60, and 0.40 for USLE, RUSLE and WEPP respectively (Tiwari et al., 2000). Impressive efficiencies were however,noted with the use of these models, though; WEPP trailed behind both EGEM and the empirical models in terms of their combined annual averages. Nonetheless, Tiwari et al. (2000) similarly observed that WEPP exhibited better prediction efficiency, than both USLE and RUSLE in terms of average annual soil loss (0.71),compared to the yearly soil losses (0.40) with the most variability.

3.5. Evaluation of WEPP versus empirical and EGEM predictions

Results of this study considerably shows the potential capabilities of the adapted WEPP model in EG erosion predictions as was the previous case for EGEM trials in the same Mubi area.The study outcome generally appears as hypothetically envisaged. Perhaps,the Mubi area terrain,vegetation,soils and climate are suitable for such model applications.The followings are results of comparisons made to evaluate the WEPP model versus Empirical and the EGEM prediction efficiencies in the study area.

3.5.1. WEPP versus empirical model predictions

Results of paired T-test between WEPP and empirical model predictions presented in Table 4, shows that there were no significant (P ＜0.05) differences among EG erosion estimates in the study locations,except at Vimtim with a 0.77 t/ha(about 41%)overprediction by the WEPP model. There was generally slight overprediction of EG erosion by WEPP model in the sites. Relationship between WEPP and empirical erosion estimates at Vimtim,Muvur and Gella EG sites were weakly tied by 25%.Likewise,the aggregate EG erosion predicted by WEPP and empirical models was strongly related by 63%, than was between WEPP and the EGEM model(Fig. 8). The observed high relationship of WEPP with the locally fashioned empirical model; further implies WEPP’s potentials for efficient implementation around the Mubi area. Likewise, Tiwari et al. (2000) reported that even under local conditions, when USLE,RUSLE and WEPP results were compared for individual sites,WEPP was found to provide better predictions at few of the sites.

Table 4 Comparison between WEPP model versus Empirical and EGEM models in predicting aggregate MSL using paired T-test and R2-values.

3.5.2. WEPP versus EGEM predictions

Table 4 shows the results of paired T-test between WEPP and EGEM predicted EG erosion in the individual sites. Results show that there were no significant (P ＜0.05) differences between the erosion prediction types, except at Vimtim and Gella EG sites. The WEPP model slightly over-predicted EG erosion, than EGEM at these 2 sites by 56% and 54%, respectively. There were generally strong relationships (62%-97%) between WEPP and EGEM models across the sites,except at Vimtim site with the weakest relationship of only 23%.Fig.8 also presents the aggregate relationship between these models, which also shows moderate association between WEPP and EGEM predicted estimates with r2= 0.56 for the Mubi area. The relationship between these technologies portrays their potentials for good use,especially the WEPP model,when supplied with the locally fitted input datasets. Tiwari et al. (2000) also compared USLE, RUSLE and WEPP, and found out that, while the overall WEPP trials did not perform as well as the USLE, it did perform better at up to 85% of the 20 individual sites studied. Besides,Nearing(1998)previously mentioned that the phenomena of erosion models overestimating small events and underestimating large events is inherent to all erosion models. This could mean a success for both the WEPP and EGEM models, considering the influences of the biased datasets used in the empirical model that was found as best in this study.

Fig. 8. Comparison between WEPP model versus (a). Empirical, and, (b). EGEM predicted aggregate MSL estimates in the 6 study sites.

4. Conclusion

Results of prediction efficiency of EG erosion using WEPP versus EGEM and empirical models around the Mubi area in the NE Nigeria reveals that both the foreign EGEM(r2=0.57)and WEPP(r2=0.53)models are implementable around the Mubi area, despite they trailed behind the empirical model prediction efficiency(r2=0.67;Mse=0.01-0.18).Even with the over or under prediction behaviors noticed with the models, empirical estimates didn’t differ significantly (P ＜0.05) from the measured erosion and was the best erosion predicted, followed by those of EGEM with Mseof 0.02-0.24 and then WEPP model(Mseof 0.01-0.38).Notably,both WEPP and EGEM model performances were better at some individual sites,than when averaged(Tiwari et al.,2000).Nonetheless,the prediction accuracies of these models were generally better than the indices earlier reported by Capra et al.(2004).Sites such as Madanya and Digil conserved with vegetative barriers significantly reduced erosion more than at other sites.It suffices to conclude that the newly tried WEPP model could be improved especially when actual field variables such as local climate,tillage,management and soil datasets are directly used in the model as suggested by Laflen et al. (1987) and Flanagan et al. (2007).

It should also be noted that both WEPP and EGEM are models that were not well fitted to the variables measured in this study,and hence, are far more likely to satisfactorily perform in other situations than will the empirical model with the parameters and variable coefficients closely tied to the locations and conditions around the Mubi area.

5. Recommendations

It is recommended that potential users of these erosion prediction technologies should consider implementing the results of this work alongside appropriate managements around the Mubi area (Tekwa et al. (2013; 2014). Sufficient WEPP databases should be developed for use in the EG erosion predictions in future researches and applications using the WEPP model in this and other regions of Africa.The model’s efficiency should be tested over much wider areas especially at some parts or whole of Nigeria. In addition, more collaboration such as postdoctoral programmes on WEPP technology knowledge transfers with places outside of the US such as from the African sub-region is strongly recommended.