Identification of Erosion Hotspot Area using GIS and MCE Technique for Koga Watershed in the Upper Blue Nile Basin, Ethiopia

Corresponding Author: Tewodros T. Assefa Energy and Environmental System, North Carolina A and T State University, Greensboro, USA Email: tassefa@aggies.ncat.edu Abstract: Soil erosion is a serious threat in Ethiopian highlands. Continuous land degradation resulted in loss of fertile top soil leading to low agricultural productivity. In addition, excessive soil erosion from Koga Watershed in upper catchment to an artificial reservoir (Koga Dam reservoir) is substantially reducing its service life. Community participatory based effective watershed management strategies may have tremendous potential to reduce soil erosion. However, it is not practical to implement management interventions in the entire basin. This study aims to identify and map erosion hotspot areas in Koga Watershed to assist local government decision towards implementing watershed management strategies. Multi Criteria Evaluation (MCE) technique was integrated with Geographic Information System (GIS). For these analysis four major factors: Topography, soil, land use and potential location of gullies were considered. Each of these was processed and analyzed for its potential contribution to erosion on a pixel by pixel basis. The factors were weighted using pair-wise comparison matrix and weights were combined using Weighted Overlay Tool of ArcGIS Spatial Analyst Toolbox to obtain the final erosion hotspot map. The results found that 2% (440 ha) to be highly sensitive, 43% (9,460 ha) to be moderately sensitive, 16% (3,520 ha) to be marginally sensitive and 32% (7,040 ha) currently not sensitive. The remaining 7% of the watershed area (22,000 ha) was constraint to erosion. The lowland area near the dam was found to be found most sensitive for erosion and sedimentation.


Introduction
Excessive soil erosion in the highlands of Ethiopia brought reduced agricultural productivity. Furthermore, natural and artificial reservoirs are suffering from sedimentation and constructed dams are ceased functioning before their service life. Lack of effective watershed management system and poor land use practices played significant role in land degradation in the region (Setegn et al., 2009). According to the Ethiopian highland reclamation study (Yilma and Awulachew, 2009), in the mid 1980's, 27 million hectare or almost 50% of the high land area was significantly eroded, 14 million hectare seriously eroded and over 2 million hectare were beyond reclamation. Rate of soil erosion is increasing alarmingly necessitating design of proper watershed management strategies. Recently constructed Koga dam reservoir is one of the affected reservoirs in the basin. Its storage volume is substantially reducing because of siltation. The siltation rate is high (100-200 Mt/ha/yr.) and 50% is supplied from agricultural lands in the form of sheet and rill erosion (Reynolds, 2012). This adversely results in constrained irrigation supply to farmlands. Field visit witnessed inability of Koga dam reservoir to serve some farmlands which are originally included in the design.
Cultivation became infeasible in several parts of Koga Watershed; however and farmers continue to cultivate marginal lands (Mengstie, 2009). Current soil and water conservation practices in Koga Watershed are focused at the highland areas which are highly degraded and have limited production potential. However, this effort could not show significant erosion reduction to date (Mengstie, 2009). Strategic watershed management interventions should focus on erosion sensitive portion of the catchment to prevent further land degradation and siltation to the dam. Lack of finer resolution spatial data and closely installed hydrometeorological data hindered spatial analysis of erosion generation. Setegn et al. (2009) delineated erosion sensitive area for Lake Tana Basin considering steep slope as the main source of erosion (infiltration excess runoff mechanism). However, the saturated area could be the main source of erosion and the dominant runoff mechanism is uncertain for the wider basin.
Understanding the dominant runoff mechanisms (infiltration excess or saturation) is essential to identify watershed management priority areas. Both mechanisms could happen in the same watershed while one dominate the other depending on hydrological properties, moisture content, slope and other factors. In a recent study conducted to nearby experimental watershed, Debremawi (Tilahun et al., 2014) showed that the infiltration capacity were exceeded by rainfall intensity during only 1.5 and 4% of the time in 2010 and 2011, respectively, indicating the dominance of saturation excess runoff mechanism. These findings indicate that the study area (Koga Watershed) is also dominated by saturation excess runoff mechanism.
The present study utilized GIS capability to analyze topographical characteristics of the watershed in terms of wetness property of the land surface. Multi Criteria Evaluation (MCE) technique was integrated with GIS to rank/rate alternatives based on multiple criteria/factors that would affect erosion. MCE analysis of spatial information is an emerging approach and is efficient to analyze complex problems within a watershed (Setegn et al., 2009). MCE approach has been successfully used in many applications that involve decision making (Pereira and Duckstein, 1993;Malczewski, 2006;Hajkowicz and Higgins, 2008;Setegn et al., 2009;Greene et al., 2011). MCE technique is well assimilated in GIS environment. GIS is capable of efficiently storing, retrieving, transforming, displaying and analyzing spatial data. In this study, we used MCE technique within GIS environment to identify the actual source of erosion and map sensitive areas based on spatial dataset analysis. Weight of decision factors are assigned based on their relative effect to erosion process.

Description of Study Area
Koga River is a tributary to Gilgel Abay River which drains to Lake Tana in the Upper Blue Nile Basin. Koga dam (drainage area of around 22,000 ha) constructed downstream of the river, supply water for irrigating downstream farms. Koga dam is geographically located at 12 0 10'N latitude and 37 0 38'E longitude, within the Lake Tana Basin (Fig.  1). The annual rainfall is around 1480 mm based data available in an adjacent meteorological station, in Merawi. The dam was designed to store around 83 million metric cube of water to irrigate about 6,000 ha of command area (AfDB, 2001). The elevation of Koga watershed varies from 2,005 to 3,147 meter above sea level within the watershed based on Digital Elevation Model (DEM). Koga watershed falls under subtropical climate zone (Yeshaneh et al., 2013). The upstream of the watershed is narrow and mountainous while the downstream is wide and gentle slope.

Data Requirement and Sources
Data required for the MCE technique are spatial in nature. Four major types of data (Land use, DEM, Soil type and Rainfall) were collected. Data sources for each category presented in Table 1. There was no meteorological station found within the catchment. Nearby stations (Merawi, Adet, Dangila and Bahir Dar) were used to find the average rainfall using thiessen polygon method. The average rainfall was highly dominated (~97%) by a single adjacent station, Merawi. The variation of rainfall within the catchment was analyzed and found insignificant (with 95% confidence). This was because of absence of closely installed meteorological stations within the catchment. As a result, rainfall was not considered as one of MCE criteria in this research.
After collecting the required data, spatial analysis were made to obtain MCE criteria map. Landsat 8 image together with intensive field point data collection were used to perform supervised land use classification in ArcGIS environment. The output map was validated and used to produce land use criteria map based on land use suitability classes. Soil map from Ministry of Water and Energy (MoWE) were directly used to produce soil criteria map based on soil type suitability classes. DEM with 30 m resolution was used to produce two criteria maps: Topographic wetness index and potential location of gullies. The wetness of the catchment (topographic wetness index) was predicted based on flow accumulation and slope of the particular pixel. Potential location of gullies was predicted based on threshold concept of two criteria: Wetness index and stream power index. Koga dam reservoir was considered as constraint. Each factor map (topography, potential locations of gullies, land use and soil) were reclassified based on sensitivity classes. The sensitivity classes for each factor were determined through discussion with experts (personal communication, T. Steenhuis, professor at Cornell University). Those experts have been doing intensive researches and experiments long on erosion and sediment transport process close to the study area.
Relative weights were assigned to each factor depending on the relevance of each factor and experts opinions. The value (1 to 9 and its reciprocal) were assigned to each factors based on pairwise comparison criteria. Pairwise comparison method was used to get the final weight of each factor. Based on factors final weight, the reclassified map was overlaid to get the combined effect of all actors. The output map was multiplied with the constraint map to exclude the reservoir area and produce the final erosion hotspot map. Figure 2 below presented the procedure to obtain the final erosion hotspot map. The detail of sensitivity classes for each factor and criteria layer map discussed in result section.

Factors Classification Approach
Erosion sensitivity of the catchment is classified into two classes (i.e., sensitive and not-sensitive) based on FAO (1981) land classification framework for irrigation purpose. Sensitive classes are further classified depending on the degree of sensitivity while not sensitive classes are classified into two classes (currently not sensitive and permanently not sensitive) ( Table 2). In this study, permanently not sensitive classes are described as constraint to erosion.

Important Factors for Soil Erosion
Four significant factors that can potentially affect soil erosion were considered. The first factor was land cover which controls the detachability and transport of soil particles and infiltration of water in to soil. The second was topography in terms of the wetness of the land. Soil types are important factor to erosion and sediment transport process depending on their physical and chemical properties. Locations of gullies were also considered as one major factor promoting erosion.

Preparation of Criteria Maps
Field work was carried out to collect control points for Landsat 8 satellite image supervised land use classification. Accuracy assessments were performed on classified images to determine how well the classification process accomplished the task using error matrix. Ministry of Water and Energy (MoWE) soil classification comparatively has finer resolution for the region and hence chosen for this study. It has 5 soil classes for Koga Watershed. Digital Elevation Model (DEM) of 30 m resolution was used to extract topographical parameters and potential location of gullies as well. The potential locations of gullies were predicted when both of the following two conditions of the thresholds were satisfied (Lulseged and Vlek, 2005) where, A S is local upslope contributing area (m 2 ) from flow accumulation raster and is local slope angle (degree).
Area covered by Koga dam reservoir were considered constraint, area that could not promote erosion because of physical limitations. The constraint map has a value of 0 and 1 (Fig. 7), where 0 indicates constraint area. Factors that cannot support erosion The weight of factors are assigned based on their relative effect on erosion and ranked according to their relative importance order (Wale et al., 2012). Pairwise comparison based on (Saaty, 1977) was adopted to calculate the total weight of each factor. It provides a powerful and simplified selection criterion by reducing bias in decision making (Wale et al., 2012). It refers to the process of comparing factors in pairs to evaluate which one is preferred. Equal interval range method was applied to distribute the overall weight of factor maps. Weighted Overlay Tool in ArcGIS Spatial Analyst Toolbox environment was used to identify erosion sensitive areas (Nyerges and Jankowski, 2009).

Impact of Land Use on Erosion
Land use map is one of the most important factors that affect surface runoff and erosion in a watershed. It enables to assess the resistance of terrain unit to erosion as a result of surface protection. High erosion and quick response to rainfall are resulted from poor surface cover. From field visit, eight land use/land cover categories were distinguished followed by supervised land use classification method and maximum likelihood algorithm. Post-classification processing was applied to the classified image such as filtering and smoothing class boundaries. Filtering is the process used to remove fixed isolated pixels from classified image. Smoothing class boundaries is the process used to clump the classes and smooth's the ragged edge.
Accuracy assessments were performed on classified images to determine how well the classification process accomplished the task. It compares the classified image to an image which is considered to be correct (Google earth for this case) with the help of error matrix (Table 3). The error matrix compares the classified and correct image on a point by point basis. The points were generated randomly using ArcGIS random point generator tool (35 random points were generated for this case) within the boundary of Koga Watershed. The overall accuracy assessment resulted in acceptable range (~89%).
In Table 3, numbers in the first row and first column represent land use classes which are: 1 (Bare land), 2 (Bush land), 3 (Forest), 4 (Grass land), 5 (Intensively cultivated), 6 (Moderately cultivated), 7 (Urban/Village), 8 (Water). Similar row and column numbers combination indicates that the land use class in the classified image (Fig. 3a) is the same as the land use in the correct map (Google earth). For instance, (a) represents how many random points that are bare land in Google earth and bare land in the classified image. But (b) represents how many random points are bare lands in Google earth but bush land in the classified image. Finally, the overall accuracy is the total number of random points that are the same in both images per total number of sample random points: (31*100/35) ~89%. According to Anderson et al. (1976), 85% accuracy is considered acceptable for image classification.
Kappa coefficient provides an insight into land use classification whether or not better results are achieved than we would have achieved strictly by chance. Kappa coefficient is given below (Equation 3 (Table 4). Information from Table 3 was used to compute the value in Table 4. Column represented by "Total" from Table 3 multiplied with row represented by "Total" to produce the Kappa coefficient in Table 4. The product matrix was determined from the sum of the products of the diagonal matrix. Therefore, the Kappa coefficient is calculated as: Product matrix (8+6+2+4+32+140+1+81 = 274), cumulative sum product (1365), expected (274/1365 = 0.201) and finally Kappa coefficient (~86.5%). Kappa coefficient value interpreted as the classification was 86.5% better than it would have occurred strictly by chance. Percent distribution of land use/cover and sensitivity classes in Koga Watershed presented in Table 5.
The re-classified land use map (Fig. 3b) indicated that 36.6% of the land use is highly sensitive; 34.6% moderately sensitive; 19.8% marginally sensitive; 4.4% currently not sensitive and about 7% constraint to erosion.

Soil Type Impact on Erosion
Soil type is one of the important factors that affect erosion process depending on the physical and chemical characteristics. It controls detachability of soil, soil particle transport and infiltration of water into the soil (Setegn et al., 2009). Soil texture is an important property which contributes to soil's erodibility. The watershed is dominated by Haplic Luvisols (38.0%) followed by Haplic Alisols (24.3%) which has moderately well drainage classes (Table  6). Figure 4a presented soil types in Koga Watershed.
The reclassified soil map (Fig. 4b) indicated that 17.5% of the soil is highly sensitive; 16.7% is moderately sensitive; 59 marginally sensitive and the remaining (~7%) is constraint to erosion.

Impact of Topography on Erosion
Topography is the major surface parameter for soil erosion assessment. Although slope has a great impact on soil erosion, the presence of soil erosion and heavy runoff on gentle slopes indicate that this phenomenon can occur without any need for a steep slope; the action of rain is enough (Fauck, 1956). Topographic Wetness Index (TWI) was used to describe the effect of topography based on saturated excess runoff generation mechanism. It represents spatial distribution of surface saturation and surface runoff which is important factor for soil erosion simulation. The concept was originally developed in the TOPMODEL framework (Beven et al., 1984) and described in Equation 1. As the contributing area increases and slope gradient decreases, topographic wetness index and soil moisture increases, it has higher correlation with saturation (Easton et al., 2010). Figure 5a presented topographic index map of Koga watershed. Erosion sensitivity class for topographic wetness index is presented in Table 7. The reclassified TWI map (Fig. 5b) indicated that 12.6% is highly sensitive; 44.3% moderately sensitive; 36.1% marginally sensitive and about 7% constraint to erosion.

Impact of Gully on Erosion
Gullies are large and deep erosion depression or channels that normally occur in drainage ways and not much deep (Imasuen et al., 2011). It occurs where surface water flow has become trapped in a small concentrated stream and begins to erode channels in the ground surface. To predict the sensitivity of a particular field to gully formation, threshold concept has been adopted. Stream power is the rate of energy of flowing water which is exerted on the bed and bank of a channel. Stream Power Index (SPI) is the product of watershed area and slope that indicates possible source of erosion by concentrated flow detachment risk (Equation 2). The potential locations of gullies are predicted based on SPI and TWI combination (Fig. 6a).   Table 7. Topographic wetness index sensitivity class TWI Erosion sensittivity group Up to 11.5 S3 11.5 to 16.5 S2 1.5 to High S1 Gully formation follows stream route and the map clearly shows that small gullies (plot level) were not capture by the threshold. Gully locations were given high sensitive class (S1) while no gully location less sensitive class (S3). The sensitivity classes were used to reclassify gully map (Fig. 6b).
Constrain map (Fig. 7) was prepared considering Koga reservoir. Value zero was given to the reservoir area to indicate permanently not suitable class. However, the sensitivity class for value 1 varies depending on the weight of other factors.

Combined Effect of all Factors
Pairwise comparison matrix was prepared by comparing factors one to one based on Pairwise comparison scale which is broken down from 1 to 9. The highest value indicates absolute important and the reciprocal kept in the transpose position indicating absolute insignificant (Table 8). For example in Table 8 topographic index (on the left) is much more important from soil (on the top), then a value of 7 were assigned at their intersection topographic index row and soil column. Conversely, soil (on the left) of Table 8 is much less important than topographic index (on the top) therefore the reciprocal was assigned (i.e., 1/7). Among the major factors, soil type was considered as the least important factor. The eigenvector was calculated for each factor (rows). According to Podvezko (2009), eigenvector is defined as the n th root of product of rows. The weights of factors were computed after normalizing the Eigen vector by its cumulative and multiplied by 100%. The reliabilities of weights were checked by computing the consistency of comparison matrix which was found consistent.
The percent value of each factor was divided into four using equal interval method to assign value for each sensitivity classes (S1, S2, S3, S4). Each map was then summed up using raster calculator so that each pixel in that map had the sum of the four map values based on sensitivity classes (overlay analysis) (Fig. 8a). Finally, the total raster value was re-classified equally into four regions: Highly sensitive, moderately sensitive, marginally sensitive and currently insensitive (Fig. 8b). Zero value indicated constraint map.
The erosion sensitivity map indicated that 2% of the total watershed is highly sensitive (S1); 43% is moderately sensitive (S2); 16% marginally sensitive; 32% currently not sensitive and 7% constraint to erosion (reservoir) in Koga watershed. The result also indicated that major source of erosion is the low land area of the catchment. This is due to the gentle slope of catchment and especially areas near to the dam are flat and the ground water table is near to the surface (saturated areas). Saturated areas are found to be high source of erosion since the rainfall after saturation could not infiltrate more leading to high surface runoff. When there is high concentrated surface runoff the power of the stream will be higher to detach sediment particles and transport downstream. Furthermore, when the catchment is saturated the water inside has hydrostatic pressure head and needs only small reach like flow path to widen and slide the earth surface.  The upland part of the watershed is found to be the least source of erosion. This is due to negligible saturation at the upland of the watershed and there is also source limit in some cases due to degraded and rocky areas. Furthermore, most of the gully formation which significantly affects erosion initiation is related to saturation while gully formation is negligible at the highland part of the watershed. Hence, it is essential to consider the saturated areas (low land) during design of watershed management strategies and implementation to effectively prevent soil loss from the catchment.

Conclusion and Recommendation
Multi-Criteria Evaluation (MCE) technique integrated with GIS has the potential to identify erosion hotspot area. The results in combination with proper field validation provide more accurate erosion sensitivity prediction. The threshold values for assessing the potential location of gullies (stream power index and wetness index threshold) were found to be high and hence did not capture catchment level small gully formations and hence further research on threshold value to capture catchment level gully formation would improve the result. The result could also be improved using higher resolution DEM.
The overall research indicated that most erosion hotspots areas were found in the lowland (more than 75% of erosion hotspot area) of the catchment, indicating that it is extremely important to consider the saturated areas during design of watershed management strategies.