Flood Hazard Mapping Using Exploratory Regression Model in GIS Domain

In this study, attempt has been made to understand the spatial distribution of ood hazard and ood risk in the Awash basin, Ethiopia. Awash basin has been chosen because it’s in continuous threat spatially and temporally. Eight determinant factors of ood hazard (viz., elevation, slope, rainfall, drainage density, landuse, soil type, wetness index and lineament density) were studied as independent variables. Each factor was reclassied into four classes and each was weighted according to its susceptibility towards the hazard. For example topographic lows were given the highest weight of 4; whereas topographic highs were given the lowest weight of 1. All the independent variables were overlaid in GIS domain to get the nal spatial distribution of hotspots of ood hazard. Exploratory regression analysis showed that the existing landuse is the dominant factor inuencing the ood vulnerability. A total of 31 models were generated using Exploratory regression in GIS domain. Model number 31 was found to be the best t model with the highest Adjusted R 2 value of 0.839539 and the least Akaike’s Information criterion value of 1536.866. Spatial autocorrelation tool run on the standard residuals yields the Z score and p value of 0.742522 and 0.457771 respectively; indicating that the residuals were neither clustered nor dispersed. Spatial extents of ood hazard and ood risk along with the priority for spatial planning in the Awash basin is digested well in this study.


Introduction
Flood being the spatial hazard demands the spatial planning to mitigate its risk (Dewan, 2013; Khailani and Perera, 2013). Spatial technology is the most appropriate technique to handle with the ood disaster risks which helps in decision making (Manfreda et al., 2011;Albano et al., 2014;Samela et al., 2017). Several scholars have studied ood and generated ood hazard and ood hazard risk maps for reducing the ood damage (Marchi et al., 2010;Wondim, 2016). Hazard zoning is also documented by several workers (Adedeji et al., 2012;Santato et al., 2013) for spatial planning processes.
However, in developing countries like Ethiopia, spatial planning is limited (Wapwera and Egbu, 2013). About 80% of the rains are received in three months (between June and September) in Ethiopia. Torrential rains are common in most parts of the country which leads to ooding to the settlements in the topographic lows (Chibssa, 2007;Dessie and Tadesse, 2013;Getahun and Gebre, 2015). For long term disaster management and mitigation, geospatial technique is the most appropriate tool (Steinberg and Lind eld, 2012; Watson and Agbola, 2013; UNDP, 2015).
The paper focuses on ood inundation analyses in Awash basin, Ethiopia. According to Tsay (2013) ooding in the Awash River Basin is mainly due to seasonal rainfall which is of great concern. The Awash River ooding causes expansion of its oodplains and inundation of topographic lows and causes severe damages to the settlements (Abebe, 2007). The Awash basin is under the continuous threat varying in extents spatially and temporally. Coenraads (2005) testi ed that the Awash River Basin is ooded for short durations at upstream but the downstream remain inundated for months every year during the wet season which in uences the crop production on ood plains. According to IPCC (2007), the magnitude and frequency of oods may be compounded by climate change. As scholars have documented the increment of ash oods in entire Ethiopia (Paolo et al., 2013), hence the ood hazard mapping in Ethiopia especially the Awash basin becomes the need of an hour for decision making processes.

Study Area
The Awash basin located in the rift valley of the Ethiopia is chosen for ood hazard mapping. The elevation of the Awash basin ranges from 362.5 to 2,989.64 m above mean sea level (Fig. 1). The study area is rugged and undulating with slopes varying from 0 to 46.38 degrees. The mean annual rainfall in the study area varies from about 178.67-461.13 mmyr − 1 at Dubti, Tendaho and Mile towns in the northeast to 1170.41-1452.86 mmyr − 1 at Holota and Ginchi towns in southwest the study area. The lowlands of the Afar region comprise the major part of the Awash basin. The Awash river rises at southwest of the basin (near Ginchi town) ows along the rift valley and nally drains in the Lake Abbe situated at the north of the basin. The total length of the river is 1202.01 km. The river drains about 112244.88 km 2 of land with the perimeter of 1994.4 km. The climate of the Awash basin is partly in uenced by the movement of the Inter-Tropical Convergence Zone (ITCZ) and partly by Indian monsoon (Romill, 2010).
The main objective of this study is to analyze the spatial distribution of ood hazard, ood risk and evaluate the in uence of various factors towards the inundation of topographic lows in the Awash basin using ArcGIS 10.3.1 plateform.

Methodology
A total of eight determinant factors of ood hazard potential (viz., elevation, slope, rainfall, drainage density, landuse, soil type, topographic compound index and lineament density) has been studied and analyzed in GIS domain. Missing or no value data in the factors were computed using kriging interpolation technique. We choose the major semi axis and minor semi axis as 118661.1 and 111161.1 respectively along with the tolerance of 45 to accommodate the maximum neighborhood. The unknown value was determined by at least 18 neighboring values with different weights. Each factor has been given weightage from 1-5 scale (1, Very low hazard; 5; Very high hazard potential). These factors were overlaid in GIS domain to generate the ood vulnerability map of the study area. Exploratory regression analysis has been done to analyze the in uence of each factor towards the ood hazard vulnerability. Autocorrelation (Moran's I) tool was run to check if the standard residuals are normally distributed. Furthermore, ood hazard map has been overlaid on population of the study area map to generate the ood risk map for better spatial planning. The overall methodology used in this study is shown in Fig. 2.

Determinant factors
Elevation and slope The movement and direction of the ood is controlled by elevation of the terrain (Gigovi´c et al., 2017). The elevation of the study area varying from 362.5-2,989.64 m has been reclassi ed into ve classes.
Topographic lows (362.5-960.04 m) were found to cover 34.41% of the total area of the Awash basin. The topographic highs (2,309.68-2,989.64 m) were found to cover only 7.25% of the study area. Each class has been given weightage depending on its vulnerability towards ood. Low laying areas (for example 362. .04 m has been given the highest weight of 5 as these areas are more vulnerable to ood; whereas areas at the highest elevation of 2,309.68-2,989.64 m has been given the lowest weightage of 1 as such areas are the least vulnerable towards ooding. The weighted map of the elevation is shown in Fig. 3. Slope is another determinant factor for ood occurrence. The slope of the study area ranging from 0-46.38 0 have been reclassi ed into ve classes. Flat to gentle slopes are more vulnerable to overland ow accumulation compared to steep slopes. Most of the study area (49.32%) is underlain by at to gentle slopes (0-4.06 0 ).
Flat to gentle slopes (0 -4.06 0 ) has been assigned the highest weightage of 5, as such slopes are more vulnerable towards ooding. Whereas steep slopes (20.28-46.38 0 ) has been given the ood potential value of 1 because such slopes are the least vulnerable towards ooding. The weighted map of the slope is shown in Fig. 3.

Rainfall and drainage density
The rainfall (178.67-1452.86 m) is another important factor determining the ood vulnerability. The highest rainfall (1170.41-1452.86 mm) was found to cover 13.58% of the study area and the lowest rainfall (178.67-461.13 mm) was found to cover 26.65% of the area. More rainfall in any area means more proneness to oods (Guo et al., 2014). Accordingly suitable weightages have been given to different rainfall classes depending on their potential towards ood. For example, rainfall ranging between 178.67-461.13 mm was assigned the least weight of 1, whereas, rainfall ranging between 1170.41-1452.86 mm was assigned the highest weight of 5. The weighted rainfall map is shown in Fig. 4. The drainage density of the area ranges from 0.08 to 9.56 km/km 2 . About 22.08% of the area is underlain by the highest Dd values (5-9.56 km/km 2 ). Drainage densities of the study area have been reclassi ed according to their ood hazard potential. Drainage density of the range 5-9.56 km/km 2 is the highest contributor towards ood and hence the highest weightage of 5 has been given. Whereas, drainage density of 0.08-1.16 km/km 2 have been given the lowest weightage of 1 because low drainage densities are generally located at the high elevations. The weighted drainage density map is shown in the Fig. 4.

Landuse and soils
Water bodies, settlements, and wetlands were found to cover 1.29% of the total study area. The computed areal extent of forest, woodland, and afro alpine is 3.67%. The landuse map of the study area was analyzed with respect to the ood vulnerability. Accordingly suitable weightages have been given to the different landuse classes. For example, forest, woodland, and afro alpine landuse classes were given the least potential of 1. Whereas, landuse classes like water bodies, wetlands, and settlements has been assigned the highest weightage of 1. Landuse classes like bare land and urban land were given the weightage of 4 and landuse classes like cultivation land; and grassland were assigned a weightage of 3. The weighted landuse map is depicted in the Fig. 5. Different soils have different in ltration rates, permeability, and porosity. The soil characteristics de ne their potential with regard to ood potential (Rimba et al., 2017). Therefore, suitable weightages have been given to different soil types of the study area. Pellic and chromic vertisols have been given the highest weightages of 5 indicating their highest contribution towards ood due to their low in ltration rates. Conversely, chromic cambisols, eutric nitisols have been assigned the least weight of 1 due to their least susceptibility towards ooding. The weighted soil map of the study area is shown in the Lineament density (Ld) of the study area has been calculated using GIS. Ld of the study area varies from 6.81-0.43km/km 2 . Areas with higher Ld values have more channels to in ltrate water downwards.
Therefore areas with the highest Ld values (6.81-12.91 km/km 2 ) were assigned the lowest weightage of 1.  For computing the ood risk map, ood vulnerability map was overlaid with weighted map of population distribution and landuse map. The output ood risk map was prioritized accordingly to get a clear picture about which areas should be given preferences for ood management processes. The ood vulnerability and ood risk map is shown in Fig. 7. Areas such as Adi Gala, Asbuli, Gewane, Loqiya, Weranso, and Mile have been found in the highest ood vulnerability zone; hence these areas are kept in the rst priority for management process. Whereas, towns such as Alem Gena, Sendafa, Gina Ager, Ankober, Assagirt, Aliyu Amba, Debre Sina, Armanaya, Jimate, Ancharo, Karakore, Kombolcha, Boru, Wirgesa, Sirinka, Kobo, Hurso, Gota, and Teferi Ber fall in "very low" ood hazard; hence were giving the least priority for ood management process. The areal extent of each ood hazard zone along with the towns associated and their priority is shown in Table 2.   3 When this test is statistically signi cant (p < 0.01), the relationships modeled are not consistent (either due to non-stationarity or heteroskedasticity). 4 Large VIF (> 7.5, for example) indicates explanatory variable redundancy.

Discussion
All the thematic layers have been weighted in accordance with their susceptibility towards erosion in GIS domain. It is clear from the ood vulnerability and ood risk map that areas under threat mostly exist in the northeastern part of the study area which exists in topographic lows. Different determinant factors may in uence differently towards the ood susceptibility. In order to understand the in uence of various determinant factors towards the hotspots of ood hazard potential, exploratory regression analysis was done in GIS domain. Before running the exploratory regression, correlations between the variables have to done to exclude any variable with collinearity or redundancy. Hence seven factors (viz., rainfall, drainage density, landuse, lineament density, elevation, slope and compound topography) were selected out of nine factors. Among the seven factors, the highest value for adjusted R 2 (0.41192) was shown by landuse followed by rainfall (0.3762), elevation (0.314543), lineament density (0.104506), topography index (0.062832), slope (0.05857) and drainage density (0.035779). It can be inferred that the existing landuse of the study area is the dominant factor in uencing the ood hazard, while as, drainage density is not affecting much toward ood hazard. Therefore, shifting cultivation in the study area by just one unit may lead to increment in the hazard by 41.11%. All the seven factors showed statistically insigni cant Koenker (BP) values (p > 0.005). Among the seven factors, the best t factor observed is landuse re ected by its least Akaike's Information criterion value of 1872.95. A total of 31 models were generated by combing the individual factors differently to nd the best model t. There is no Exploratory variable redundancy in all the models as observed by their maximum variance in ation values (< 7). Furthermore Koenker (BP) statistic is statistically insigni cant (p > 0.005) in all the models. Among all the 31 models, model number 31 was found to be the best model t as re ected by its lowest Akaike's Information criterion value of 1536.866 (Table 2). In the model number 31, drainage density factor was omitted as it didn't contribute much towards the model.
After doing regression modeling, it's imperative to check if the standard residuals are normally distributed. If the standard residuals are clustered or dispersed, it implies the model is biased or invalid. Hence, spatial autocorrelation tool (Moran's I) was run on the standard residuals. The standard residuals were found to be distributed randomly (Fig. 7). The Moran's index value of 0.0219 was obtained with its Z score and p value of 0.742522 and 0.457771 respectively.

Conclusion
A total of eight determinant factors were chosen to study the hotspots of ood hazard in the Awash basin, Ethiopia. Each factor was weighted according to their ood potential. All the weighted factors were overlaid to get the spatial maps of ood hazard and ood risk. The areal extents of very low, low, moderate, high, and very high hazard zones are 6.96, 10.21, 33.09, 38.28 and 11.46 km 2 respectively. The output map depicts that northeast part of the study area exists in the "very high" ood potential as well as ood risk zone; therefore demands the rst priority for spatial planning. Priority table has been generated to highlight the areas demanding the highest and the lowest priority. Exploratory regression clearly showed that the existing landuse is the dominant factor in uencing the ood potential. Spatial autocorrelation tool showed the random distribution of the standard residuals; hence validating the model. This study proved the e ciency of exploratory regression model to be the best technique to understand the impact of various determinant thematic layers towards the mapping of ood hazard and ood risk and can be applied at continental level with some modi cation. Figure 1 Location of the Awash basin, Ethiopia. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 2
Methodology employed Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 3
Weighted layers of elevation (meters) and Slope (degrees) of the Awash basin, Ethiopia Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.

Figure 4
Weighted layers of Rainfall (mm) and Drainage density (km/km2) of the Awash basin, Ethiopia Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city Page 17/20 or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors. Weighted landuse and soils of the Awash basin, Ethiopia Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors. Weighted topographic wetness index and lineament density of the Awash basin, Ethiopia Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors. Spatial distribution of ood vulnerability and ood risk map of the Awash basin, Ethiopia Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.