Flood Hazard Zoning of Upper Awash River Basin, Ethiopia, Using the Analytical Hierarchy Process (AHP) as Compared to Sensitivity Analysis

Floods and droughts have been two of the most devastating consequences of the climate crisis afecting billions of people in the world. However, unlike the other natural hazards, fooding is manageable through appropriate food management mechanisms. Tis study emphasizes on developing a food hazard zone for the Upper Awash River Basin (UARB), Ethiopia. Six relevant climate, physiographic, and biophysical factors were considered. Ten, a food hazard map was developed employing the analytic hierarchy process (AHP) method and further validated using sensitivity analysis and collected food marks. Te results revealed that drainage density, rainfall, and elevation have higher signifcance, while land use and soil permeability have a low impact in the process of food generation. Te map showed vulnerable areas at diferent levels and can serve as a valuable input for the decision makers to consider in the process of implementing emergency plans as well as long-term food mitigation options.


Introduction
A warmer climate, with its increased climate variability, will increase the risk of both foods and droughts [1][2][3][4][5][6]. Te efect of a warmer climate is more pronounced by urbanization and industrialization. Te urbanization usually leads to unprecedented deforestation and land use and land type changes. According to Stoy [7], deforestation often increases land-surface and near-surface temperatures and the severity of extreme heat. Tese results in a decline in land perviousness and increase in the amount of sunlight refected back from the earth's surface. As a result, the atmosphere warms up; a lot of water evaporates from seas, oceans, as well as from any water resources located on the earth, which in turn creates a feedback loop between global climate change and extreme hydrological events such as fooding and drought [8][9][10].
Floods cause serious harm to people and adversely afect socioeconomic development around the world, especially in urban areas where the high risks of fooding are the consequences of urbanization and industrialization [11][12][13][14][15].
Te World Bank report [16] states that over the last two decades, foods and droughts-two of the most devastating consequences of the climate crisis-have afected 3 billion people, with staggering costs in human sufering and economic loss. Another report of the United Nations (UN) reveals that foods alone afected 2.3 billion people from 1995 to 2015 in the world [17]. Tis contributes to 56% of the total afected people by weathered-related disasters. Te report also states that there were 3062 food disasters in those mentioned years that accounted for 47% of all geophysical hazards occurred in the globe. Rentschler et al. [4] also revealed that 1.81 billion people, which share 23% of the world population, are directly exposed to 1-in-100-year foods. Flood events are becoming more severe, and food frequency has been rising causing mainly a distraction of agricultural areas and food supply sectors and exacerbating malnutrition problems of the poorer areas of the world, such as Asia and Africa [14].
In Ethiopia, fash fooding, which is mostly anticipated in the areas of a river side as well as in areas with low water percolation capacity, has been pronounced in the southern, north-eastern, and eastern parts of the country [18]. Flood damaged and displaced hundreds thousands of people in Ethiopia (https://foodlist.com/tag/ethiopia). According to EM-DAT [19], foods and droughts were severe national disasters of the country during the last centuries causing huge loss of human lives and properties. Historical records on food data suggest that Ethiopia faced 47 major foods since 1900, which afected close to 2.2 million people [20]. Mamo et al. [21] reported that the frequency of food occurrences in the country increased from decade to decade with the 2001-10 decade being the most fooding decade with fve food years out of ten, whereas the last 2011-20 decade witnessed three food years.
Te Awash River Basin (ARB) with a total land area of 110,000 km 2 is one of the major river basins in Ethiopia that has series fooding problems [22][23][24][25]. An approximate area in the order of 200,000-250,000 ha is subjected to fooding during high fows of the Awash River. Te Upper Awash River Basin (UARB), which is the subject of this study, constitutes part of the ARB and is subjected to intense fooding for short durations after strong or prolonged rainfall events. Recurrent fooding that has occurred in UARB has been a critical problem. Te food event usually occurs in summer ("kiremt") season of the country, Ethiopia. According to a socioeconomic study conducted in the basin, in 2017/2018, foods afected 8,477 households, more than 18,996 ha of agricultural land, 25,087 live stokes, infrastructures, and health and educational institutions. Schools in the UARB often started late because the fooding and health centers are not functional in the rainy season of the country.
Unlike most types of disasters such as volcanoes and earthquakes, foods are preventable and manageable through proper implementation of an integrated food management approach and proper mitigation measures. Flood hazard zoning provides a starting point and useful resource for food risk management, mitigation actions, and governance. Moreover, food hazard zone maps are a valuable tool in planning the future development of the city, as well as identify areas that need development of infrastructure and food drainage [26].
Te main purpose of this study is to map food hazard zones in UARB by implementing the analytical hierarchy process (AHP) method. Te AHP method is a multicriteria analysis approach for organizing and analyzing complex decisions based on mathematics and psychology. It was developed by Saaty in the 1970s and has been extensively studied and refned ever since [27,28]. Te AHP approach for food hazard mapping is gaining wide range recognition in recent times. Hadjimitsis et al. [29] implemented the AHP to compare the diferent factors and their relative importance in assessing natural and anthropogenic risk of culture heritage in Cyprus. Fernandez and Lutz [30] zoned Bwana Argentina Yerba city in terms of food risk using GIS and a multicriteria decision-making system (AHP). Kazakis et al. [31] used the AHP approach to assess food hazard areas on a regional scale in north-eastern Greece, where recurring food events have appeared. Similar other studies can be found in the works of [32][33][34][35][36].
Besides the AHP process employed in this study, the research applied the sensitivity analysis method (i.e., map removal sensitivity analysis techniques as it was discussed in [35,36]), and results were validated against the collected food marks. In both AHP and sensitivity methods, six relevant climate, physiographic, and biophysical factors that were essential for food hazard zoning were identifed and used. Tese factors include rainfall amount, slope, elevation, river density, land use, and soil type-based permeability. Ten, these factors were used to rank the level of importance of each of them in the process of food generation. Te study is organized as follows: Section 2 presents the study region. Section 3 presents the data and methodology used in this study. Section 4 discusses the fndings. Finally, Section 5 gives the conclusion and recommendations of the work.

Study Area
Te Upper Awash River Basin (UARB) is the upper part of the Awash River Basin (i.e., one of the 12 major basins) of Ethiopia covering about 11,607 km 2 . Te basin is surrounded in the north by the Abbay basin, in the west and southwest by Omo Gibe and Rift valley lake basins, and in the east-south by the Wabi Shebele basin. Te river originates from the south of Mount Warqe at an altitude of 3000 m above mean sea level (m.s.l.) at a place specifcally called Elam close to the Ginchi town and runs up to Koka reservoir with altitude about 1500 m above m.s.l. Te Upper Awash River has meandering characteristics traveling about 200 km until it reaches to Koka reservoir ( Figure 1).
Te major tributaries of the river include Holeta, Alito, Teji, Gilo and Kelina, Kebena, Great and Little Akaki, and Mojo rivers. Te basin consists of eight food afected districts such as Ilu, Dawo, Sebeta Hawas, Welmera, Ejere Ejersa Lefo, Liben-Chuquala, and Bora districts ( Figure 1). It has a bimodal rainy season. Te main rainy season lasts from June to September and the second minor rain occurs from March to April. Te basin receives an average annual rainfall of 1052 mm where it varies from 400 mm to 1900 mm per year from place to place. Te mean annual temperature ranges from 20.8°C to 29°C at the Koka dam [25].
Recent history has shown that the basin is highly afected by recurrent foods as well as erosion [22,[39][40][41][42]. Increasing agricultural activities without land conservation and overgrazing leads to erosion and further aggravates fooding. Te cumulative efect of these hazards warrants for the need of proper planning for disaster reduction and sustainable mitigation plans.

Data and Methodology
In this study, six parameters such as rainfall, slope, elevation, river density, land use, and soil type-based permeability were considered to map the food hazard zones. Space Shuttle Radar Topography Mission (STRM) data were used to determine terrain elevation, and consequently slope and drainage density were extracted from the digital elevation model. Te land use and land cover (LULC) data of the study area were obtained from the Ethiopia Geospatial Mapping 2 Te Scientifc World Journal Agency (EGMA) for the year 2013. It showed that a large portion of the basin is covered by agricultural land such as annual and perennial cropland sharing 63.1% and 8.9%, respectively, while shrub land and grassland land cover 12.2% and 0.8% accordingly. Te land use type also includes dense forest (i.e., dense, moderate, and spares type), woodland, settlements, bare soil, and water bodies. Te soil map of the basin was collected from the Food and Agriculture Organization (FAO) soil database. Te dominant soil types of the study area consist of Pellic Vertisols (46.2%) and Vertic Cambisols (12.8%). Luvic Phaeozems, Eutric nitisols, and Orthic Solonchaks also shares 7.3%, 6.3%, and 6.2%, respectively. Te spatial variability of the collected data was prepared in raster format and further classifed accordingly. Elevation, drainage density, slope, and rainfall were classifed in accordance to get a uniform interval in between classes (Table 1). Similar research studies also applied uniform intervals to categorize classes for diferent factors [41][42][43][44]. Te land use type of the basin was condensed into fve classes such as forest, woodland, grass land, cultivated land, and water body where their level of infuences for food occurrence varies from very low, low, moderate, and high to very high, respectively. Similarly, according to the Soil Conservation Service (SCS) classifcation, the soil type was categorized into four hydrological soil groups (A, B, C, D) [45,46] and water bodies.

Analytical Hierarchy Process (AHP).
Flood hazard zoning is developed for UARB using the analytical hierarchy process (AHP) and further elaborated using the applying sensitivity analysis method. Te AHP method uses a multicriteria analysis approach, which was developed by Saaty in the 1970s [27,28], and has been extensively used ever since [41][42][43][44][47][48][49][50][51][52]. Tis allows for selecting potential parameters that cause fooding. When quantitative ratings are not available or difcult to rate factors, decision makers can still recognize whether one criterion is more important than another using pairwise comparison similar to what Saaty [28] developed. In this study, weight for each factor was also determined using a pairwise comparison matrix. Te AHP includes a comparison of importance between factors, normalization, and computing consistency ratio. Te relative importance of one factor over the others was defned based on the rating scale provided in Saaty [28], as given in Table 2. Te relative signifcance between the criteria is evaluated from 1 to 9, where 1 indicates equal important criteria and 9 represents much more important criteria.
Te pairwise comparative weight is highly dependent on expert judgment and should fulfll the consistency ratio (CR) criteria. Te normalization of weights assigned to each parameter is done using Eigen vector and further validated for consistency check employing CR formula (equation (1)). Te value obtained from the CR must be less than or equal to 0.1 [53]; as a result, any subjectivity which might be involved during prioritizing the level of importance of one factor over the other can be reduced.
where CI represents the consistency index (CI � average number of consistency vector − n/n − 1) and computed from the pairwise comparison matrix of all the parameters. Whereas, RI is the random consistency index as stated in Saaty [53], where value depends on the number of factors (n).

Sensitivity Analysis Method.
Te sensitivity analysis was also performed considering six food-generating factors such as rainfall, slope, elevation, river density, land use and soil type-based permeability, and weight overlay spatial analysis techniques. It is a biased-free map removal analysis Te Scientifc World Journal technique. Initially, all factors are considered as if they are equally important for generating foods. As a result, a basecase scenario was formulated providing equal weight for each factor (i.e., 16.67% weight for each). And then, an additional six scenarios were developed by turning of/removing one factor at a time. Consequently, a correspondent food hazard map was produced for each scenario. Tereafter, the food hazard maps of diferent scenarios were compared with the base-case scenario and the weights of each factor were determined accordingly. Te food-generating factors with classifed raster layers were then multiplied with correspondent weights that were obtained from the AHP method and later validated by the weights derived from the sensitivity method. Finally, these layers were overlain to generate food hazard zones following equation (2) and the steps involved in Figure 2.  Table 2: Fundamental scales of absolute numbers ("Saaty scale"), adapted from Saaty [53].
Level of importance Descriptions 1 Equal importance of both factors 3 Judgment slightly favors one factor over another (moderate diference of importance) 5 Judgment strongly favors one factor over another (strong diference of importance) 7 Very strong or demonstrated importance of one factor with respect to another 9 Evidence of extreme diference of importance of one factor with respect to another where FHI represents the food hazard index, R i are the corresponding weights, and F i are the food-generating factors.

Classifcations of Flood-Generating Layers.
Raster layers were prepared for each of the six food-generating factors at the start of the process. Tese raster layers were further classifed into fve classes based on their food-generating capability of the area and using an equal interval method of spatial analysis tool. Tus, a very high class was assigned a rate of 5, a high class rated as 4, a moderate class rated as 3, and low and very low classes rated as 2 and 1, respectively (Table 1 and Figure 3(a)).
Considering elevation as one of the food-generating factors, the lowest elevation values indicate the highest possibility of food-generating capability and hence are assigned the highest rate of fve. From Table 1, this value ranges from 1580 m to 1978 m. Te rest of the elevations between 1978 m and 3568 m were also categorized from high to very low depending on food-generating potential ( Table 1).
It is obvious that the rainfall amount has a direct relation with the amount of food generated. In UARB, as the long-year mean rainfall pattern indicates, there is high precipitation in the east highlands and northwest and southwest peripheries, while there is low rainfall in the west lowlands and central part of the river basin (Figure 3(b)). As a result, those areas with the highest rainfall intensity ranging from 1490 mm to 1655 mm are given the highest rate of fve and rated as very highly vulnerable areas (Table 1). Areas with rainfall amounts ranging from 831 mm to 1490 mm are accordingly classifed as very low to high depending on the rainfall amount generated (Figure 3(b)).
Te drainage density is the total length of all the streams and rivers in a drainage basin divided by the total area of the drainage basin. It was computed from the digital elevation model (DEM) of the basin applying Kernel density, and it varies from 0 to 1.31 for the basin (Table 1). Similarly, the drainage density layer was further classifed in into fve classes; a higher drainage density indicates a very high hazardous area and assigned a rate of fve, whereas an area having a smaller drainage density results in the minimum area to be afected by food and is ranked as very low (Figure 3(c)).
Te slope is also computed from the DEM of the basin (Figure 3(d)). Te general assumption followed was that the terrain with the steepest slope tends to retain the least amount of water, whereas terrains with the fattest slopes retain the more water. Accordingly, terrains with the fattest slope were identifed as highly vulnerable to food and assigned a rate of fve (i.e., slope from 0 to 14%). Other slope values ranging from 14% to 71% are classifed from high to very low rates.      Te Scientifc World Journal 7 Te LULC of a basin plays a signifcant role in rain water movement either by retarding or accelerating overland fow. LULC highly infuences the infltration rate and thus the water partitioning between surface and groundwater systems of a catchment. Forest land enhances the infltration capacity of the surface and as a result reduces fooding and hence is given the lowest rate of 1. Other land use types such as woodland, grassland, and cultivated land afect water percolation at diferent levels and are classifed as low (rate 2), moderate (rate 3), and high (rate 4), respectively. Settlements and water bodies highly aggravate overland fow and thus fooding; as a result, they are assigned with the highest rate (rate 5) (Figure 3(e) and Table 1).
Te soil type-permeability can amplify/extenuate the extent of food events. Diferent soil types have diferent capacities to infltrate water. Sandy soils have higher hydraulic conductivities than fne-textured soils because of the larger pore space between the soil particles. As such, the infltration rate of clayey soils is much lower than that of sandy soils. As mentioned in the methodology section, the soil of the Upper Awash Basin was divided into four hydrologic soil groups based on infltration rates (groups A, B, C, and D). Accordingly, the classifcation and rating of the soil factor were made as shown in Table 1.

Flood Hazard Map Using the AHP Method.
Te optimal pairwise diagonal matrix of the study is stated in Table 3. As mentioned in the methodology section, the pairwise diagonal matrix was developed by assigning the level of importance of one factor over the other. For instance, rainfall intensity is more important than land use and therefore assigned the value 7. Te row describes the importance of land use (Table 3). Terefore, the row has the inverse value of the rainfall in the pairwise comparison (i.e., land use is 1/7 th as signifcant as the rainfall intensity). Various comparison matrices were assumed and their corresponding consistency ratios were computed until a satisfactory result, which is a CR of less than or equal to 0.1, was achieved.
Tereafter, as mentioned in the methodology section, percentages of preference values were computed by dividing the individual preference value of each factor over the cumulative preferences' values in a column (Table 4). Consequently, the ratio of preference values with weight (i.e., which are determined in percentage preference matrix) gave us the weight value matrix from which the average number of consistency vectors was determined as 6.477 and results in a consistency index of 0.0955. Te random index (RI) depends on the size of the matrix, and when the matrix size is equal to 6, then RI will be 1.24 [53]. Following, by applying equation (1), CR is determined as 0.077 which is under the allowable threshold value.
Te weights of food-generating factors were then computed as described in Table 5. Te weight factors computed from AHP are applied for each of the foodgenerating layers in equation (2), and the weighted layers were overlain one after the other to generate a food hazard map of the basin ( Figure 4). As shown in Figure 4, the high fooding zone matches with the sample food marks collected in the years 2018 and 2019.
Te food hazard map (Figure 4) shows that the majority of the catchment has been subjected to a moderate food hazard amounting for an estimated area of about 8352 km 2 that accounts for about 76.42% of the total area of the UARB. Te fgure also shows that 1866 km 2 of the basin area is vulnerable to a high food hazard while 709 km 2 (6.48%) was prone to a low food hazard. Areas at the upstream of the basin (Ilu and Sebeta Hawas districts) as well as the fnal outlet of the river (upstream of Koka reservoir covering Bora and Liben districts) were more susceptible to high foods.
Excluding the "external" food-generating factor which is rainfall, the AHP result reveals that drainage density and elevation have a high infuence on food occurrence. Similarly, Figure 4 shows that places with low elevation range and high drainage density are more vulnerable to high foods.

Flood Hazard Map Using the Sensitivity Method.
Te food hazard zone developed using the AHP method was further validated with sensitivity analysis. In the sensitivity analysis, all factors were initially considered as if they are equally important for generating foods and have given equal weight. And hence, the base-case scenario with a food hazard map was developed given 16.67% weight for each of the food-generating factors ( Figure 5).
Tereafter, additional six scenarios were performed by turning of one factor at a time (Table 6). When one factor is turned of, the remaining fve factors will have a weight of 20% each. Ten, the correspondent food hazard maps were also determined for each scenario ( Figure 6). An evaluation was performed for identifying the impact of each factor against the others. Tis helps in a better understanding of the importance of each factor in identifying against the low impact factors for fooding.
Each food hazard map shown in Figure 6 exhibits the classifcation from very low to very high zones in spatial coverage. Consequently, the area coverages were computed from very low to very high ranges as given in Table 6. Examining the diference between food hazard maps of Scenario-1 ( Figure 5) and Scenario-2 ( Figure 6(a)) helps us to determine the weight factor of rainfall.
Similarly, weights (signifcance) of other remaining food-generating factors were computed referring to the base-case scenario and further validated against the GPStracked food mark of 2018 and 2019 food events.
Te sensitivity analysis showed that a very high food zone which leads to extreme events was observed due to drainage density (30%), rainfall intensity (25%), and elevations (15%), respectively, whereas LULC (10%) and soil types (5%) showed the lowest signifcance. Ten, the newly computed weights were applied to the corresponding layers of food-generating factors and resulted in a food hazard map, as shown in Figure 7. Te weighting values derived using sensitivity analysis were somehow similar resultant to those computed using the AHP method, showing rainfall, drainage density, and elevation have a higher infuence for causing food in both methods, respectively. Areas located in Ilu, Sebeta Hawasa, Bora, and Liben Chiquala districts are highly vulnerable to food, while very high food exposure exists in the upstream of Koka reservoir.
As mentioned, the sensitivity analysis is a biased-free analysis method and thus optimally enhances for computing reasonable weighting values. Te techniques used in this analysis method are logical and by itself leads to the solution. Te food map derived using this method matches well with the food hazard map derived by AHP. Tus, it can be used as a validating technique while developing food hazard maps.

Validation Process.
Te validation process was made by overlaying and comparing the traced peripheral of the fooded areas with the results achieved through the AHP method and sensitivity analysis method in the study area. A feld visit has been made, and the peripheral of food marks have been collected for the years 2018 and 2019 through Garmin hand GPS. A total of 174 food marks were collected (i.e., 24 and 150 points for the year 2018 and 2019, respectively). Te collection includes tracing the peripheral boundaries of fooded areas and taking their spatial location (Easting and Northing). Flood marks were collected for Bora, Liben Chiquala, Ilu, Dawo, Sebeta Hawasa, Ejere, and Ejere Lefo districts for the two successive years varying the spatial extent from the river channel. High food extent was recorded in Ilu and Sebeta Hawas districts. Tose boundaries of food points then overlaid in the generated food maps in ArcGIS environment ( Figure 8). As shown in Figures 8(a)  and 8(b), most of the food marks laid over the high food zone of the maps developed using AHP sensitivity analysis methods. From the total collected food marks, 146 points (83.9%) laid over high food zone of the AHP food map, while 28 points laid over moderate food zones. For the food map developed using the sensitivity method, 158 food marks, which is 90.8%, was laid over high food zones, while    Land use  Soil  Elevation  1  1/3  3  1/3  5  3  Drainage density  3  1  3  1/3  5  5  Slope  1/3  1/3  1  1/3  5  3  Rain fall  3  3  3  1  7  5

12
Te Scientifc World Journal 9.2% was spread over moderate food zones. Overall, more than 83% of the collected food marks match with high food zones of the developed maps which revealed their reliability.

Conclusion
Recurrent foods have increased from year to year in UARBs and have been causing huge economic losses and associated social impacts. Tis is mainly due to climate and other associated changes. Urbanization, agriculture, and grazing land have increased in the basin which signifcantly reduces the permeability of the land and enhances overland fow. Consequently, this resulted in fooding in the basin year after year. Tus, developing accurate food hazard zoning will favor the prevention and sustainable management of foods in the basin. In this study, various food-generating factors such as rainfall, slope, elevation, drainage density, land use, and soil type-based permeability were considered, and their corresponding infuence was quantifed using the AHP method and further validated by applying sensitivity analysis and previously collected food marks in order to develop an appropriate food hazard zone of the basin. Identifcation of the signifcant factor helps to select the type of measures to be employed while taking mitigation measures and thus enhances emergency food adaptation mechanism and favors extreme food management options.
Te study reveals that drainage density, rainfall, and elevation have higher impacts on generating foods relative to the other factors. However, land use and soil permeability have a lower infuence. Consequently, feasible food hazard maps showing diferent food zones such as high, moderate, low, and very low were developed and validated against food marks and sensitivity analysis. Te food hazard map showed that places with lower elevation range and high drainage density were more susceptible to fooding. Specifcally, places at Ilu, Sebeta Hawasa, Bora, and Liben Chiquala districts were more vulnerable to the recurrent food. Te results of the study can be instrumental for the decisionmaking parties for implementing emergency food adaptation mechanisms as well as for implementing longterm sustainable extreme food management options.

Data Availability
Te data used to support the fndings of this study are available upon reasonable request from the corresponding author.

Conflicts of Interest
Te authors declare that they have no conficts of interest. Te Scientifc World Journal 13