Land Use Evaluation over the Jema Watershed, in the Upper Blue Nile River Basin, Northwestern Highlands of Ethiopia

: Generating land capability class guidelines at a watershed scale has become a priority in sustainable agricultural land use. This study analyzed the area of cultivated land use situated on the non-arable land-capability class in the Jema watershed in the Upper Blue Nile River Basin. Soil surveys, meteorological ground observations, a digital elevation model (DEM) at 30 m, Meteosat at 10 km × 10 km and Landsat at 30 m were used to generate the sample soil texture class, average annual total rainfall (ATRF in mm), terrain, slope (%), elevation (m a.s.l) and land-use land cover (%). The land capability class was analyzed by considering raster layers of terrain, the average ATRF and soil texture. Geo-statistics was employed to ﬁt a surface of soil texture and average ATRF estimates. An overlay technique was used to compute the proportion of cultivated land placed on non-arable land. As per the results of the terrain analysis, the elevation (m a.s.l) of the watershed is in the range of 1895 to 3518 m. The slope was found to be in the range of 0 to 45%. The amount of estimated rainfall ranged from 1640 to 131 mm with value declined from the lower to the higher elevation. Clay loam, clay and heavy clay were found to be the major soil texture classes. Four land capability classes, i.e., II, III, IV (arable) and V (non-arable), were identiﬁed with proportions of 28.56%, 45.74%, 22.16% and 3.54%, respectively. Seven land-use land covers were identiﬁed, i.e., annual crop land, grazing land, bush land, bare land, settlement land, forestland and water bodies, with proportions of 42.1, 35.9, 8.90, 8.3, 2.6, 2.1, and 0.2, respectively. Around 1707.7 ha of land in the watershed is categorized under non-arable land that cannot be used for annual crop cultivation at any level of intensity. Around 437 ha (3.5%) of land was cultivated on non-arable land. To conclude, the observed unsustainable crop land use could maximize soil loss in upstream regions and siltation and ﬂooding downstream. The annual crop land use that was observed on non-arable land needs to be replaced with perennial crops, pasture and/or forest land uses.

Nevertheless, to date, no such evaluation exists for the study watershed. The proportion of cultivated land that was utilized in contradiction to LCC guidelines has not been examined and documented. To address this need, the current study attempts an estimation of the proportion of the Jema watershed that is currently used for annual crop cultivation in a manner inconsistent with LCC guidelines. This study shows the methods and procedures of how to capture limitation of farmland taking into account more appropriate geographical unit of analysis, agroclimatic zones of a watershed. The same method could be implemented in tropical highlands where an ecosystem is complex like that of the study site.

Description of the Study Site
The Jema watershed is located within the area 11 • 22 0 N to 11 • 3 30 N and 37 • 2 0 E to 37 • 23 30 E (digital elevation model (DEM) at 30 m, shuttle radar topographic mission (SRTM)). The watershed belongs to part of the north-western highlands of Ethiopia in the Lake Tana Sub-Basin in the Upper Blue Nile River Basin (Figure 1). Its geographical area covers~483 km 2 . The area is situated in the West (Mirab) Gojjam Administrative Zone of the Amhara National Regional State. The topography of the site is characterized by gentle slopes, hilly landscape, and steep slopes, and the geology is dominated by volcanic rocks [38] of Late-Tertiary to Quaternary age. The site is categorized under semi-arid highland [10,24]. On average, its ATRF is~1400 mm, whereas its MTmax and MTmin are~25 • C and 7 • C, respectively. About 80% of the total precipitation occurs during the summer (Kiremt) season in the months of June, July and August. The maximum temperature has been recorded in March, during the spring (Belg) season in the months of March, April and May [39]. The main soil orders are Alisols, Nitisols and Vertisols [40]. The population density was estimated to be~189.4 persons/km 2 [41]. The primary livelihood is subsistence-scale mixed farming, i.e., food crop production and livestock rearing [7]. Jema watershed that is currently used for annual crop cultivation in a manner inconsistent with LCC guidelines. This study shows the methods and procedures of how to capture limitation of farmland taking into account more appropriate geographical unit of analysis, agroclimatic zones of a watershed. The same method could be implemented in tropical highlands where an ecosystem is complex like that of the study site.

Description of the Study Site
The Jema watershed is located within the area 11°22′0″ N to 11°3′30″ N and 37°2′0″ E to 37°23′30″ E (digital elevation model (DEM) at 30 m, shuttle radar topographic mission (SRTM)). The watershed belongs to part of the north-western highlands of Ethiopia in the Lake Tana Sub-Basin in the Upper Blue Nile River Basin (Figure 1). Its geographical area covers ~483 km 2 . The area is situated in the West (Mirab) Gojjam Administrative Zone of the Amhara National Regional State. The topography of the site is characterized by gentle slopes, hilly landscape, and steep slopes, and the geology is dominated by volcanic rocks [38] of Late-Tertiary to Quaternary age. The site is categorized under semi-arid highland [10,24]. On average, its ATRF is ~1400 mm, whereas its MTmax and MTmin are ~25 °C and ~7 °C, respectively. About 80% of the total precipitation occurs during the summer (Kiremt) season in the months of June, July and August. The maximum temperature has been recorded in March, during the spring (Belg) season in the months of March, April and May [39]. The main soil orders are Alisols, Nitisols and Vertisols [40]. The population density was estimated to be ~189.4 persons/km 2 [41]. The primary livelihood is subsistence-scale mixed farming, i.e., food crop production and livestock rearing [7].

Analysis of Land Capability Class (LCC)
LCC was examined considering soil texture (% of soil particle size), slope (%) and elevation (m), and the range of ATRF (mm). As stated in the work on the USDA system suggested by [42], in LCC analysis, the critical point is considering the degree of limitation of certain environmental variables; that is, a given LCC analysis is not required to consider the limitation of all environmental variables. The proportion of cultivated land use was also examined. Ultimately, the proportion of cultivated land situated on non-arable classes was computed.

Spatial Variability of ATRF
The rainfall data of this study were extracted from the estimated (reconstructed) data sets generated by Meteosat. The data were gridded at a spatial scale of 10 km × 10 km, and made for the years from 1983 to 2010. The reconstruction was done with the reference of observed data from 550 meteorological stations situated over Ethiopia. The data were reconstructed by the National Meteorological Agency of Ethiopia in collaboration with the International Research Institute for Climate and Society, USA. The data calibration and validation of this reconstructed data set was undertaken by Reading University, United Kingdom. The result showed a strong correlation (i.e., r = 0.8) between the station and satellite-derived data (Ethiopian Meteorological Agency cited in [6]. From the gridded Meteosat dataset, the value of 36 grid cells located inside and around the study watershed was considered. The sample values of the average ATRF were interpolated to the entire watershed, employing the most appropriate geospatial interpolation models. The geospatial interpolation models were evaluated by comparing their relative committed root mean square error (RMSE) as in Equation (1). The analysis was done at the 95% confidence interval using ArcMap software. A model that might be found with the lowest value of RMSE was assumed to be the most optimal/accurate model to interpolate/estimate soil texture particles. The error value was expected to be in the range of the observed minimum and maximum values of an observed environmental attribute. The estimated value of a given environmental attribute with a model with the lowest RMSE value was expected to be closer to the measured values. In this estimation, the annual total rainfall (mm) data set was a predictand variable, and the elevation (meter) data set generated from the DEM 30 m was an auxiliary variable. Ordinary kriging (OK), simple kriging (SK), universal kriging (UK), disjunctive kriging (DK), simple cokriging (SCK), ordinary cokriging (OCK), universal cokriging (UK), inverse distance weighting (IDW), global polynomial interpolation (GPI), local polynomial interpolation (LPI), and radial basis function (RBF) were the interpolation models examined in the study. The first five models are stochastic, whereas the last four models are deterministic [43]. Unlike deterministic methods, stochastic (kriging and cokriging) interpolation methods can take into account the values of available relevant auxiliary variables to estimate the values of a given predictand variable. Since Ethiopian rainfall depends on topography, this study considered elevation as a co-variable. The application of these geo-statistical methods was recommended in previous studies [26,44,45].
where RMSE is the root mean square error; n = number observations or samples; o = observed value at place i; p = predicted/estimated value at place i; osi = standardized observed value at place i; and psi = standardized predicted/estimated value at place i.

Spatial Variability of Soil Texture
Soil sample sites and points were selected based on a two-stage cluster sampling technique accompanied by a stratified and proportionate sampling procedure. The sample strata were prepared considering the elevation and slope class variation of farmlands in the Jema watershed. Four elevation and seven slope sample strata were identified, and the numbers of soil and elevation classes was decided following the recommendation of [24,27]. The strata of slopes were decided mainly based on the severity of soil erosion, whereas elevation strata were based on agro-climatic factors including the length of growing period. Both the watershed boundary and the terrain class maps (raster layers) were delineated from a satellite-derived 30 m resolution SRTM DEM ( Figure 2). The accuracy of the map (DEM data) was verified by a ground truth field survey data collected by using a hand GPS (geographic positioning system) with a reported accuracy of 3-5 m. Once elevation and slope classes (the sample units or clusters) were identified, the specific farmland sample sites and points in each sample unit (that were assumed to be homogenous in terms of their biophysical setup) of soil were chosen depending on their accessibility. accuracy of the map (DEM data) was verified by a ground truth field survey data collected by using a hand GPS (geographic positioning system) with a reported accuracy of 3-5 m. Once elevation and slope classes (the sample units or clusters) were identified, the specific farmland sample sites and points in each sample unit (that were assumed to be homogenous in terms of their biophysical setup) of soil were chosen depending on their accessibility. A hand auger was used to collect sample soils. For testing soil texture, 36 composite (and disturbed) samples were collected from 36 farmland sites (sample units) situated in the stratified slope and elevation classes ( Figure 2). Each farmland site covers an area of ~12.7 km 2 ; that is, the average spatial area coverage of each sample unit (farmland site) where one composite sample was collected. The scale of the map is categorized under a large-scale map (i.e., 1 cm to 2 km). One composite (and disturbed) soil sample was collected from 3 to 4 discrete sample points in the same sample site. In some steep slope sites, it was difficult to get the top soil. Thus, to homogenize the soil horizon where samples need to be taken, the samples for texture analysis were collected at a depth of 0-60 cm. The samples were analyzed in a soil laboratory following the procedure of ISRIC [46]. The texture samples were then dried in an oven and sieved through a 2 mm sieve.
In order to estimate the surface soil texture (sand, clay and silt particles) across the entire Jema watershed, the sample was subjected to geostatistical analysis. The same statistical tool, technique and procedure that were employed in the case of the rainfall interpolation were employed. Since temperature and gradient affect the nature of soil texture, both elevation and slope were considered as covariables. Finally, the raster surfaces of sand, clay and silt soil particles were combined and mapped into appropriate soil texture classes to the entire watershed. The texture class was A hand auger was used to collect sample soils. For testing soil texture, 36 composite (and disturbed) samples were collected from 36 farmland sites (sample units) situated in the stratified slope and elevation classes ( Figure 2). Each farmland site covers an area of~12.7 km 2 ; that is, the average spatial area coverage of each sample unit (farmland site) where one composite sample was collected. The scale of the map is categorized under a large-scale map (i.e., 1 cm to 2 km). One composite (and disturbed) soil sample was collected from 3 to 4 discrete sample points in the same sample site. In some steep slope sites, it was difficult to get the top soil. Thus, to homogenize the soil horizon where samples need to be taken, the samples for texture analysis were collected at a depth of 0-60 cm. The samples were analyzed in a soil laboratory following the procedure of ISRIC [46]. The texture samples were then dried in an oven and sieved through a 2 mm sieve.
In order to estimate the surface soil texture (sand, clay and silt particles) across the entire Jema watershed, the sample was subjected to geostatistical analysis. The same statistical tool, technique and procedure that were employed in the case of the rainfall interpolation were employed. Since temperature and gradient affect the nature of soil texture, both elevation and slope were considered as covariables. Finally, the raster surfaces of sand, clay and silt soil particles were combined and mapped into appropriate soil texture classes to the entire watershed. The texture class was determined by referring to the USDA's soil texture triangle.
With respect to soil texture, a loam texture is associated with minimum limitation, and limitations increase as the texture diverges from loam in either direction [16]. Loam soil is believed to be the most optimal for plant growth with respect to air, water, mineral and nutrient circulation, which is vital for plant-soil interaction.
In general, as the slope of farmland increases, the limitation of land increases [16]. In the study region, the limitation of land also increases with elevation, as one moves from moist-cool to Alpine [24]. This is reflected in the terms "cultivable crop diversity" and "length of growing period" (LGP) [3,47]. Based on the preliminary focus group discussion conducted in the ACZs of Jema watershed, crop diversity was found to be lower in the upstream region. With regard to LGP, there is a positive spatial association with elevation. The annual average ATRF was considered as an environmental factor in the LCC analysis because the study site is located in a semi-arid tropical region which requires more rainfall amount to maximize moisture availability and enhance land productivity. Moreover, together with the incoming solar radiation to be received at different elevation zones, differences in the average ATRF that is received in the cropping season may create variation in terms of crop type and diversity to be adapted, and water harvesting potential [24].
In order to produce one comprehensive LCC layer for the watershed, it was first essential to generate the same type of layers for soil, slope and elevation and average ATRF features separately. Then, the raster layers of the three variables were combined using a balanced weight overlay technique.
The degree of influence of the environmental variables in LCC analysis might not be the same. To date, there is no scientific guideline that shows the weights of indicators in terms of agricultural LCC analysis in the context of tropical highlands.

Spatial Variability of Land-Use Land Cover
The spatial variability of land-use land cover (LULC) was analyzed from Landsat8 scanned in January 2018, which is the dry season in the study region. The imagery was acquired from the United States Geological Survey (USGS) online data archive. The basic features of the imagery are given in Table 2. A total of 380 sample polygons were collected with the help of GPS for seven land-use land cover classes (grazing land, cultivated land/crop land, bush land, bare land, settlement land, forestland and water body) in the four ACZs of the watershed. For sampling purposes, the proportion of LULC types was estimated by visual observation made with the help of a field survey and Google Earth imagery. Nearly two-thirds of the total sample were used for training purposes, and the rest were reserved for validation. The imagery was analyzed employing the maximum likelihood classifier in a supervised classification method. It is recommended to employ this method especially in cases where the researcher has prior experience about the physical features of the site and could access the place for ground truthing [48,49]. Under supervised classification, the maximum likelihood classifier tool has been the most recognized tool of image classification.
The accuracy of the LULC classification analysis of this study was evaluated by calculating a confusion matrix (error matrix), including overall accuracy, producer's accuracy, and user's accuracy. Taking the statistical output of the confusion matrix, the Kappa statistic was computed. [50] suggested a kappa value of <40% as poor, 40% to 55% fair, 55% to 70% good, 70% to 85% very good and > 85% as excellent.
Ultimately, the non-arable land (i.e., LCC-V) raster layer and cultivated land raster layer were overlaid to identify any part of cultivated land situated on non-arable land. All statistics were computed using map algebra in ArcMap.

Spatial Pattern of Land Capability Class
3.1.1. Surfacing the Spatial Distribution of the ATRF Among all tested interpolation models, the OCK model provided the lowest RMSE when interpolating the average ATRF (RMSE = 0.72 mm). In this regard, the model was found to be the most optimal/accurate one, i.e., the estimated value with this model was close to the measured value. In the entire sample site, the estimated amount of the average ATRF increases as we go from north-west to south-east ( Figure 3); that is, from the lower elevation to the higher elevation (Table 3).
Similarly, in the case of the specified study watershed, the amount of the average ATRF declines in the same direction. The value of the rainfall ranges from 1310 mm to 1640 mm in the study watershed. The figure shows the existence of a number of ups and downs in the values of the spatial distribution of the average ATRF in the whole sample site.  Similar to the case study of [51] and in contrast to the national observation of [10], the result of this study showed that the amount of the ATRF declines from north-west to south-east. This happened in a situation where elevation increases as one moves from north-west to south-east. Nevertheless, the same pattern of mean ATRF was not observed across the study corners that were included in the interpolated surface. The reason for the observed patterns of rainfall could be an interactive effect of the direction of moist wind movement and microclimate factors.
Thus, in a river basin, where there is a complex topography and wind convection system, accurate and precise mean ATRF estimation needs micro-scale analysis.
The total moisture obtained from the rainwater, predominantly in the summer season, can be considered adequate to grow a wide range of local annual crops including barley (Hordeum Vulgare), wheat (Triticum Spp.), teff (Eragrostis), maize (Zea Mays), finger millet (Eleusine Coracana), sorghum (Sorghum Bicolor), horse bean (Vicia Faba), pea (Pisum Sativum), lentil (Lens Esculenta), and potatoes (Solanum Tuberosum) [10]. However, apart from terrain and soil properties, the difference in the amount of temperature was important in determining the type of crop grown in the ACZs, as temperatures are cooler and the LGP is shorter at higher elevations.  Similar to the case study of [51] and in contrast to the national observation of [10], the result of this study showed that the amount of the ATRF declines from north-west to south-east. This happened in a situation where elevation increases as one moves from north-west to south-east. Nevertheless, the same pattern of mean ATRF was not observed across the study corners that were included in the interpolated surface. The reason for the observed patterns of rainfall could be an interactive effect of the direction of moist wind movement and microclimate factors.

Geo-Statistical Interpolation of Soil Texture
Thus, in a river basin, where there is a complex topography and wind convection system, accurate and precise mean ATRF estimation needs micro-scale analysis.
The total moisture obtained from the rainwater, predominantly in the summer season, can be considered adequate to grow a wide range of local annual crops including barley (Hordeum Vulgare), wheat (Triticum Spp.), teff (Eragrostis), maize (Zea Mays), finger millet (Eleusine Coracana), sorghum (Sorghum Bicolor), horse bean (Vicia Faba), pea (Pisum Sativum), lentil (Lens Esculenta), and potatoes (Solanum Tuberosum) [10]. However, apart from terrain and soil properties, the difference in the amount of temperature was important in determining the type of crop grown in the ACZs, as temperatures are cooler and the LGP is shorter at higher elevations.

Geo-Statistical Interpolation of Soil Texture
SCK (simple cokriging) of soil texture outperformed other kriging methods and deterministic interpolation models, as assessed by RMSE. SCK provided an accuracy of 86.5% for sand, 85.5% for clay, and 77% for silt. Heavy clay, clay, and clay loam were found as the major texture classes in the lower, middle and higher elevations, respectively (Table 4 and Figure 4). This indicates the complexity of the terrain in the study site. The result implies the existence of a spatial autocorrelation between soil texture and terrain attributes. Thus, the result was consistent with the theory of catenary soil development [52]. In the catenary soil development approach, there is a sequence of soils down a slope, created by the balance of processes such as precipitation, infiltration, and runoff. Catenary soil development is mainly true for relatively permanent soil properties such as texture [53]. SCK (simple cokriging) of soil texture outperformed other kriging methods and deterministic interpolation models, as assessed by RMSE. SCK provided an accuracy of 86.5% for sand, 85.5% for clay, and 77% for silt. Heavy clay, clay, and clay loam were found as the major texture classes in the lower, middle and higher elevations, respectively (Table 4 and Figure 4). This indicates the complexity of the terrain in the study site. The result implies the existence of a spatial autocorrelation between soil texture and terrain attributes. Thus, the result was consistent with the theory of catenary soil development [52]. In the catenary soil development approach, there is a sequence of soils down a slope, created by the balance of processes such as precipitation, infiltration, and runoff. Catenary soil development is mainly true for relatively permanent soil properties such as texture [53].  The findings of this study contrast with the results of [5] that was conducted at the national level. In that study, the texture class of the Jema watershed was condensed only to the clay texture class. Similarly, as inconsistent with the result of [5,24], the ATRF of the study watershed declines with elevation. The different findings might be attributed to the differences in the geographical scale of analysis and the complexity of the terrain in the north-western part of the country. Looking at the nature of soil texture, the water holding capacity of the soil can be regarded as low in the entire watershed and is especially limited in the downstream portion of the watershed. This signified the prevalence of erosion upstream and siltation downstream. The findings of this study contrast with the results of [5] that was conducted at the national level. In that study, the texture class of the Jema watershed was condensed only to the clay texture class. Similarly, as inconsistent with the result of [5,24], the ATRF of the study watershed declines with elevation. The different findings might be attributed to the differences in the geographical scale of analysis and the complexity of the terrain in the north-western part of the country. Looking at the nature of soil texture, the water holding capacity of the soil can be regarded as low in the entire watershed and is especially limited in the downstream portion of the watershed. This signified the prevalence of erosion upstream and siltation downstream.

Land Capability Class
Based on the result of LCC analysis that was conducted, considering slope (%), elevation (m), soil texture (%) and average ATRF (mm), four (II, III, IV and V) USDA LCC were identified ( Figure 5). Their area coverage was calculated for the entire Jema watershed (Table 5) and for each AEZ of the watershed ( Table 6). Most of the land in the watershed was characterized by LCC-III, while LCC-V was found only in the two ACZs (sub-Alpine and moist-cold) situated in the higher elevation. Based on the result of LCC analysis that was conducted, considering slope (%), elevation (m), soil texture (%) and average ATRF (mm), four (II, III, IV and V) USDA LCC were identified ( Figure  5). Their area coverage was calculated for the entire Jema watershed (Table 5) and for each AEZ of the watershed ( Table 6). Most of the land in the watershed was characterized by LCC-III, while LCC-V was found only in the two ACZs (sub-Alpine and moist-cold) situated in the higher elevation.  Area Coverage (%) in Each of the ACZ  Most (~96.5%) of the land in the Jema watershed could be categorized as some category of arable land; the remaining portion was non-arable. Though very intensive cultivation cannot be exercised in the entire watershed, limited cultivation with proper land management can be exercised all over the watershed except on LCC-V (3.5%) of the watershed. Much of the land with LCC-V was situated in the higher elevation areas, sub-Alpine and moist-cold ACZs. In the study conducted in the Guila-Abena watershed in northern Ethiopia [54], four major LCCs (II, III, IV and V) were identified. In the study of [14] conducted in the Gido watershed in northern Ethiopia, the capability of the land fell into four classes ranging from I to IV.

Spatial Variability of Land-Use Land Cover
The overall statistical error report of the image classification is shown in Table 7. The highlighted frequency values represent the main diagonal of the matrix that contain the cases where the class labels depicted in the image classification and ground dataset matching, whereas the off-diagonal elements contain those cases where a mismatch in the labels was indicated. Thus, with regard to overall accuracy, 85.48% of the classification was found to be accurate. With respect to reliability (Kappa Coefficient), the classification was 82.4% better than would have occurred strictly by chance. The result revealed that the proportion of grazing land, crop land (cultivated land), bush land, bare land, settlement land, forest land and water bodies were 42.09%, 35.90%, 8.95%, 8.27%, 2.58%, 2.05% and 0.17%, respectively ( Figure 6 and Table 8). Most of the area was used for and/or covered with grazing land and crop land. It was found out that a larger proportion of cultivable land is situated downstream.

Conclusions and Recommendations
Analysis of LCC taking into account the ACZs of a watershed was found to be more appropriate in terms of capturing the limitations of farming land. Accordingly, taking into account the limitations of land (soil texture, slope, elevation and mean ATRF), ~3.5% (1707.72 ha) of the watershed should not be used for cultivation at any level of intensity. The degree of land limitation is larger in the upstream portion of the watershed. At present, out of the total non-arable land, 437 ha was cultivated for crop production. To limit land degradation, farmers who settle on this non-arable land should change their way of using their land from annual crop production into perennial crops (fruit), pasture or forestry. These land uses can also be made to be more effective through the proper integration of forage legumes with agro-forestry systems, the expansion of perennial grasses, bunds, grass strips, contour leveling, terraces, shade trees and waterways. Through the expansion of farming practices such as the rotation of grazing and promotion of stall feeding, the practice of pasture land use and livestock rearing would be more effective. Meanwhile, the local government should be more committed to enforcing land use regulations through strengthening natural resource conservation bylaws.   Given the proportion of bare land and bush land (Table 8), the agro-ecosystem of the entire watershed could be regarded as sensitive to climate variability. In addition, since most of the forest cover was dominated by eucalyptus trees that have sparse canopies (local farmers-focus group discussion made in 2017), the existing limited forest would have little potential to reduce run-off [11]. Accordingly, cultivated lands are most likely to face severe soil loss, predominantly in parts of the watershed where the slope is steep. As shown by [31,55], such places would be subject to high surface runoff and low water retention.

Cultivated Land Use versus Arable Land Capability Class
Overlaying the LCC V layer extracted from the LCC layer ( Figure 5) over the cultivated land use layer extracted from the land-use land cover layer (Figure 6), it was possible to find~437 ha of cultivated land on potentially non-arable land in the upstream. This study proved that a good part of the land segments was not utilized in accordance with their comparative advantages, i.e., LCC guidelines. This may be due to the scarcity of farmland induced by the high annual total population growth rate (3% per year at regional level [2], the expansion of gullies in cultivated areas, the prevailing high soil loss rate (300 + tons/ha/year at the national level [24], and the low penetration of agricultural technologies. All of these pressures can lead farmers to encroach onto non-arable and steep slope land segments for cultivation and settlement purposes. General issues of low productivity and poverty [10] accompanied by weak land use and land administration institutions [1] can also contribute to this encroachment into marginal lands. Studies conducted in the Ethiopian highlands indicated that the way some farmers utilize and conserve their land and environment was not in line with the capability of the land [9,12]. If the same pattern of misuse of land continues, runoff [15], flood frequency [17], drought periods [9,19] could be intensified. As per the result of the current study, homogenizing a watershed into ACZs would enable researchers and policy makers to easily target specific sites where misuse of land is concentrated. Accordingly, unless the ongoing poverty reduction program introduces safety nets to enhance non-cultivation and alternative livelihood strategies, 3.5% (437 ha) of the land that is being used for crop cultivation will become more exposed to soil loss. Since this endangered land is situated in the upstream portion of the watershed, its adverse impacts would extend to downstream areas in various forms. For example, it could aggravate the problem of siltation and minimize the volume of surface stream flow. As a result, the productivity of farming systems would be affected negatively.

Conclusions and Recommendations
Analysis of LCC taking into account the ACZs of a watershed was found to be more appropriate in terms of capturing the limitations of farming land. Accordingly, taking into account the limitations of land (soil texture, slope, elevation and mean ATRF),~3.5% (1707.72 ha) of the watershed should not be used for cultivation at any level of intensity. The degree of land limitation is larger in the upstream portion of the watershed. At present, out of the total non-arable land, 437 ha was cultivated for crop production. To limit land degradation, farmers who settle on this non-arable land should change their way of using their land from annual crop production into perennial crops (fruit), pasture or forestry. These land uses can also be made to be more effective through the proper integration of forage legumes with agro-forestry systems, the expansion of perennial grasses, bunds, grass strips, contour leveling, terraces, shade trees and waterways. Through the expansion of farming practices such as the rotation of grazing and promotion of stall feeding, the practice of pasture land use and livestock rearing would be more effective. Meanwhile, the local government should be more committed to enforcing land use regulations through strengthening natural resource conservation bylaws. Acknowledgments: Florida International University, USA, is appreciated for hosting the senior author as a visiting scholar.