Land Use/Land Cover Change Impact on Hydrological Process in the Upper Baro Basin, Ethiopia

Gambella University, College of Engineering and Technology, Department of Water Resource and Irrigation Engineering, Gambela, Ethiopia Institute of Technology, Hawassa University, Awassa, Ethiopia Wolkite University, Department of Natural Resource Management, P. O. Box No: 07, Wolkite, Ethiopia Gambella University, College of Agriculture and Natural Resource Management, Department of Plant Science, P.O. Box 126, Gambella, Ethiopia


Introduction
e response of hydrologic circulation is closely related to land use planning and management [1,2]. Land Use/Land Cover (LU/LC) change is one of the factors that directly affect the watershed hydrological cycle [3][4][5]. It has been widely accepted that human activities cause LU/LC change and have imposed a great impact on the hydrological processes and water resources of the watershed [6] For example, a study [1] showed that the water provision and the hydrological process decreased as consequence LU/LC change that was aggravated by significant increasing population pressure and development in Pennar Basin in India.
is can be either by increasing the water yield or by eliminating the flow in some circumstances, thereby increasing the sediment load and decreasing the groundwater [7,8].
LU/LC changes and population growth are the most common problems in developing countries such as Ethiopia since their economic development mainly depends on agriculture [9]. In the past 40 years, the increases in human activities resulted in the expansion of agricultural land, extraction of timber, and urbanization, thereby resulting in deforestation in southwest Ethiopia [10,11], while these changes in land cover affect the local watershed hydrological cycle and flood vulnerability of various subwatersheds [12]. e modifications/conversions of natural vegetation and physical soil conditions are usually the principal cause of changes in rainfall-runoff characteristics of the local catchment, which consequently change the river flow regimes [13,14]. Several studies show that the changes in vegetation cover, i.e., deforestation, lead to an increase in water yield and sedimentation [15,16].
Watershed studies can provide direct evidence of LU/ LC change impacts on runoff [17]. For these reasons, hydrological models are becoming important for studying LU/LC change effect on the hydrological cycle in a landscape [18,19]. Several studies were carried out on the hydrology of the watershed by utilizing LU/LC data in different regions of Ethiopia using the Soil and Water Assessment Tool (SWAT). ey indicated an increase in the wet seasonal flow, surface runoff, and water yield [20][21][22][23]. However, studies on the impact of the LU/LC dynamic on the hydrology regime of the Upper Baro Basin are limited.
e Upper Baro Basin is hurt from human-induced degradation mainly due to large-scale agricultural expansion, which leads to LU/LC dynamism. e impacts of such changes on local hydrology are poorly understood. Hence, systematic analyses and understanding the impacts of LU/ LC changes on the watershed hydrology are important for planning water resources management and maintaining the sustainable flow of water to the Baro-Akobo River.
is study also helps decision-makers know where to invest to conserve the hydrological services. erefore, the objectives of the study are as follows: (1) to assess the LU/LC change over the past 30 years on the hydrological response of Upper Baro Basin and (2) to check the applicability of SWAT model for the study area.

e Study Area.
e Upper Baro Basin is found in southwestern Ethiopia located at 766 km away from Addis Ababa, the capital city of Ethiopia, between latitudes 7.44°a nd 9.41°N and longitudes 34.52°and 36.31°E, covering an area of 23462 km 2 . e altitude ranged from 390 m to 3266 m above sea level (Figure 1). e Upper Baro basin has a long rainy season in the months of May-September. e area indicates a monomodal, bimodal, and triple-modal rainfall pattern. e annual average rainfall is varying from about 1163.03 mm to over 2258.36 mm with a monthly maximum rainfall record of 302.79 mm in August, while relative mean monthly rainfall for dry seasons occurs between December and March ( Figure 2). e annual rainfalls show pronounced annual and seasonal fluctuations ( Figure 3) and the maximum temperature of Upper Baro Basin ranges between 27.7°C (in the wet season) and 37.2°C (in the dry season) and occurs in June to August, while the minimum falls between 10.8°C and 13.7°C in the year. e topography of the Upper Baro Basin is characterized by variation from flat to mountainous terrain and falls in the altitude range of 390 m to over 3000 m a.m.s.l. with 42% in between 1000 m and 2000 m a.m.s.l. e eastern two-thirds of the basin area lies between 1000 m and 2400 m a.m.s.l. and a gently sloping plain lies in the west between 380 m and 500 m a.m.s.l. Dystric nitisols and dystric gleysols are the dominant soil type in the study area.

Meteorological Data.
e SWAT model requires daily meteorological (Weather) data such as daily precipitation, minimum and maximum air temperature, relative humidity, wind speed, dew point, and daily sunshine. e standard deviation of each parameter was calculated. iessen method is used to adjust for nonuniform meteorological station distribution by proportion to the area which is closer. Eleven meteorological stations were selected (Table 1) based on the quantity, quality, period, consistency, homogeneity, and their uniform distribution within and around the Upper Baro Basin. ese data were obtained from the National Meteorological Agency of Ethiopia (NMAE) over the period . e missing values were estimated from other stations around the missed record station by using both arithmetic mean method and normal ratio method and also consistency test which shows double mass.
e SWAT weather generator model (WGEN) was used to match the access table and lookup table. e daily precipitation and temperature of all gauging stations were prepared in comma-delimited (.csv) format. Solar radiation, relative humidity, and wind speed data were available only for principal Gore and Masha stations (synoptic stations).  (Figure 4), define the stream network, and determine sub-basin parameters such as slope area and slope length.          selections were in dry season fixed with the lowest percent or zero monthly cloud cover. erefore, all images are cloudfree (0%) in the study area. e images were already terrain corrected projected to the Universal Transverse Mercator (UTM) of WGS84 zone 37 N. ey were radiometrically calibrated to reflectance value and the Quick Atmospheric Correction algorithm was applied in ERDAS software. Radiometric and geometric corrections were applied to the removal of sensor noise, haze, correction data loss, and missing lines due to a solar position, satellite calibration [24,25]. A supervised image classification method was employed using Maximum Likelihood Algorithm classification. For supervised classification, the Region of Interest (ROI) was made a signature for different LU/LC categories. e reference data collected using a GPS receiver were first converted to a vector file, then the region of interest (ROI) was made in ERDAS software. Using the ROI, the spectral signature of each LU/LC type has been extracted. Further investigations were carried out by taking these sample points in the land use types using a handheld GPS receiver. Based on this analysis and the existing land use distribution information, the area was categorized into eight broad LU/LC classes: cultivated land, woodland, forest land, grazing land, bare land, shrubland, water bodies, and built-up (Table 3).

Soil
Overall accuracy and Kappa coefficient, the classify maps will check classification accuracy. According to [26,27], a good overall classification map accuracy is accepted if and only it possesses an accuracy of at least 85%. e LU/LC was an input of the SWAT model to describe the HRU of the subwatershed.
e SWAT Model has predetermined four-letter codes for each LU/LC category (Table 4). ese codes were used to link the LU/LC map of the study area with the SWAT land database. To make it compatible, LU/LC with the SWATmodel lookup table was prepared for each type of LU/LC.

SWAT Model Description.
SWAT is a physically based semidistributed continuous time-scale hydrological model, which works on a daily time step. is model can simulate runoff, sediment, nutrients, pesticide, and bacterial transport from agricultural watersheds [28]. It simulates the hydrological cycle parameters based on the water balance represented in equation (1) within the watershed [29]. More details on the information about the SWAT model can be found in the SWAT user manual [29]. e hydrologic cycle simulated by the SWAT is based on the water balance equation [29] as follows: where SW t and SW o are the final and initial soil water content, respectively, (mm), Q i � daily surface runoff (mm), R i � daily rainfall (mm), ET i � daily evapotranspiration (mm), R i � daily lateral flow (mm), and P i � daily percolation (mm).

Peak Runoff
Rate. e peak runoff rate is a result of the erosive power of a storm and is used to predict sediment loss. SWAT calculates the peak runoff rate with a modified rational method for each HRU using the following equation: where Q peak is peak runoff rate (m 3 /s), α tc the fraction of daily rainfall that occurs during the time of concentration, Q surf is the surface runoff (mm), A is the sub-basin area (km 2 ), and t conc is the time of concentration (hr).

Surface Runoff Generation.
e CN method was initially developed for small agricultural watersheds and the CN varies nonlinearly with the moisture content of the soil. It drops to zero as the soil approaches the wilting point and increases to be near 100 as the soil approaches saturation, with higher CNs associated with higher runoff potential watershed. is method is widely used [28].
According to USDA, 1985 [30], this can be mathematically expressed using the following equation:  where Q s is the accumulated runoff (mm), Ra is the rainfall depth for the day (mm), and where CN is the curve number for the day and its value is the function of land use practice, soil permeability and soil hydrologic group. Conceptual framework showing the components and relationships have been used as a framework for the analysis of hydrology in research ( Figure 5).

SWAT Model Simulation, Sensitivity Analysis, Calibration, and Validation.
e simulation result cannot be directly used for further analysis. Instead, the ability of the model to sufficiently predict the constituent streamflow should be evaluated through sensitivity analysis, model calibration, and model validation [31].
2.6.1. Sensitivity Analysis. Fifteen hydrological parameters related to streamflow were selected for sensitivity analysis in the study area. e sensitivity of each parameter was identified using the T test and P values. P values were used to determine the significance of the sensitivity. A value close to zero has more significance. T test provides a measure of sensitivity and the most sensitive parameters usually appear with P values less than the alpha level of 0.05 (larger absolute values are more sensitive) [32].

Model Calibration and Validation.
e SUFI-2 algorithm was applied and frequently used in this study of calibration for the SWAT model at a large scale due to its easy application and the reduced number of model runs needed to achieve a decent prediction [33,34]. In this study, automatic calibration was made every month from January 1, 1990, to December 31, 2002, until the average simulated value came closer to the measured value. Automatic calibration makes use of a numerical algorithm to increase the performance of the model and to optimize the numerical objective functions. After the simulation result for the calibration, the period had fulfilled the above statistical criteria, validation was performed for an independent period of records from January 1, 2003, to December 31, 2009, and the results were compared against an independent set of measured Upper Baro Basin.

Model Performance Evaluation.
Model performance was carried out to verify the robustness of the model to simulate hydrological processes. A model performance framework proposed by [35] was used in this study.
Nash and Sutcliffe simulation efficiency (NSE) indicates the degree of fitness of observed and simulated data, and it is calculated using the following equation: where N � number of compared values, Q obs is the observed data, Q obs is the observed mean, and Q sim is the simulated data. e NSE indicates how well the plot of observed versus simulated value fits the 1 : 1 line. e closer the model efficiency is to 1, the more accurate the model, and if it is found between 0 and 1, it indicates deviations between measured and predicted values. If NSE is negative, predictions are poor, and the average value of output is a better estimate than the model prediction [36].
Coefficient of determination (R 2 ) measures the ability of a model to predict or explain an outcome in the linear regression setting (Equation (6)). R 2 ranges from 0 to 1, with higher values indicating less error variance, and typically values greater than 0.5 are considered acceptable [35,37,38].
where X i is measured value, X av is average measured value, Y i is simulated value, and Y av is the average simulated value. Percent bias (PBIAS) measures the average difference between the simulated and measured values for a given quantity over a specified period (usually the entire calibration or validation period) and it is calculated using the following equation [39,40]: where X i is measured value and Y i is simulated value. Model evaluation classifies as follows: (i) 0.75 < NSE < 1.00, PBIAS < ±10-better predictor (ii) 0.65 < NSE ≤ 0.75, ±10 < PBIAS < ±15-good (iii) 0.50 < NSE ≤ 0.65, ±15 < PBIAS < ±25-satisfactory (iv) (NSE ≤ 0.50, PBIAS >±25)-unsatisfactory  (Table 5). e results obtained for 1987-2002 reveal that forest land covers 36.74% and followed by grassland, 19.77% and shrubland and woodland also cover 19.76% and 11.32%, respectively. Between 2002 and 2017, the distribution of forest cover demonstrates highest in the catchment, which covered 33.33%, followed by agricultural land and shrubland, with the value of 15.6%, 15.1%, respectively. Forest land and shrubland decreased after 1987-2002, with −3.41% and −4.65%, respectively (Figure 7). e area of forest land decreased due to of change of land use (conversion) to the expansion of agricultural land in the southwestern part of the catchment. e urban area gradually expanded from 2.01% to 15.51% over the entire study period. ese results indicate that government policy does not have a strong impact on land use.

LU/LC
In the last 30 years, the study displays an increase in agriculture, bare land, urban, and shrubland, whereas the forest, grassland, water, and wetland were noted declined. Agricultural land increased by 18.01% during the study period. is is due to the increase of population growth that causes the increase in demand for cultivation land for different agricultural products like cereal crops, field crops, industrial crops, whereas the forest area was diminished.
is might be due to deforestation activities that have taken place for agriculture and urban expansion.
Consequently, agricultural investment policies encouraged further deforestation that leads to land conversion [41] and various types of natural forests and woodlands in the Western part of Ethiopia, being cleared and replaced by commercial agriculture [42,43]. In the study basin, cultivated land has gained/increased by 18.01%, mainly converted from the grazing land by 15.64%. Forest and shrubland conversion also contributed to the increasing cultivated land.
In this study, the accuracy assessment was performed using the 340 referenced points collected through stratified sampling. Overall accuracies obtained were 92.14%, 94.63%, and 95.93% for the 1987, 2002, and 2017 images, respectively (Table 6). e overall accuracy was computed by dividing the correct classification by the total number of reference pixels in the error matrix.

Stream Flow Modeling Using SWAT
3.2.1. Sensitivity Analysis. Sensitivity analysis was carried out to identify which model parameter was the most important or sensitive. Flow sensitivity analysis was carried out for 13 years, the calibration period (from January 1, 1990, to December 31, 2002). e default and optimal parameters and their ranking in terms of the 10 highest relative sensitivity values used to develop, calibrate, and validate the model (Figure 8).  Applied and Environmental Soil Science

SWAT Model Calibration and Validation.
e simulated monthly flows demonstrated a good agreement with the monthly observed discharge of Upper Baro Basin with a coefficient of determination (R 2 � 0.86), Nash-Sutcliffe efficiency (NSE � 0.80), and percent bias (PBIAS � −17%) (Figure 9). However, the model overestimated the peak monthly flow for some years in the calibration period; it followed the trend of observed monthly Upper Baro basin discharge and presented a good response to extreme rainfall events, which resulted in high runoff volume.
SWAT model also successfully validated streamflow for an independent period (2003-2010) with R 2 � 0.84, NS � 0.76, and PBIAS � −4.5% during the validation period; the values fulfilled the statistical model performance criteria R 2 > 0.6 and NSE > 0.5 recommended by SWAT developer [37]. during calibration and validation periods, respectively, also fulfilled the statistical model performance criteria PBAIS> ±15 [35] (Table 7). e percentage of observations bracketed by the 95 PPU was 75% for calibration and 72% for validation. e r-factor, which is the average thickness of the 95PPU band divided by the standard deviation of the measured data, was 1.08 for calibration and 0.84 for validation. According to [32], a p-factor greater than 0.70 and r-factor less than 1.5 display that the model is accurate in simulating streamflow. us, the results displayed that the model accurately simulated the biophysical processes influencing flow in the study area.
Generally, the above information showed that the performance of the model was better during the calibration period more than the validation period (Table 8) and also the calibration validation period (Figures 9-12) shows there was good relation with observed and simulated streamflow.

Land Use/Land Cover Change Response to Hydrological
Process.
e simulated results of the Upper Baro Basin under different land use/land cover scenarios are shown in Table 9.

Mean Monthly Wet and Dry Month Streamflow
Simulation and eir Variability. Months July, August, and September were considered as wet periods and Jan, Feb, and March were considered dry months; these seasonal variabilities of streamflow were evaluated.

Spatial Variability of Surface Runoff over Sub-Basin.
Spatial variability of surface runoff shows that surface runoff is highly sensitive to the change of LU/LC over subwatersheds in the period of 1987, 2002, and 2017 ( Figure 13).

Comparison of Mean Monthly Surface Runoff and
Groundwater. Decreases in base flow had a potential effect on the change in annual river flow ( Figure 14). Results from LU/LC change scenarios, the mean monthly (wet and dry) flow has increased from 82.9 mm and decreased 20.54 mm concerning the change of LU/LC from 1987 to 2017 ( Figure 15). e results (Table 10) demonstrate that the mean dry monthly flow for 1987-2017 land cover decreased by −20.54 (m 3 /s) when compared to the land cover in 1987-2017. Changes in land use occurred in all parts of the Basin but mainly in the southwestern part. Vegetation cover within the catchment had changed considerably due to human activities, mainly through the conversion of natural vegetation to agriculture, Urban, and largescale investment. ese changes have altered water resources through biophysical and biogeochemical processes in the soil. rough these land-cover changes, the biophysical and morphological    [44,45] reported that an increase in annual surface runoff increased when forest lands were converted to the other classes especially cultivated land. e expansion of agricultural land by displacing forest and shrubland results in an increase in surface erosion generation following rainfall events and causes a reduction in soil moisture conditions and groundwater recharge to shallow aquifers. Similarly, studies suggested that grassland and lateral flow are positively correlated [46,47].        Figure 14: Mean monthly surface runoff (mm).

Conclusions
In the present study, the Upper Baro Basin has been simulated using the SWAT hydrological model for assessing the impact of LULC change on its hydrological regime successfully. Initially, the change in LULC between the years 1987, 2002, and 2017 has been assessed. It is evident that land cover changes had occurred between 1987 and 2017. Forest cover had decreased from 36.74% to 31.21%, attributable to logging for timber, firewood, and clearing for agricultural purposes. In contrast, the agricultural area increased over the years from 8.81% to 26.81%. Similarly, grasslands and bushes have also been converted to agricultural areas. Sensitivity analysis of the SWAT parameters indicates that runoff is most sensitive to curve number, an available water capacity of the soil layer, soil evaporation composition factor, manning's "n" value for the main channel, and saturated hydraulic conductivity. Seasonal streamflow variability due to the LU/LC was assessed and comparisons were made on surface runoff, lateral flow, and groundwater flow contributions to streamflow based on the three simulation outputs. Surface runoff was increased from 713.85 mm to 760.49 mm, while groundwater was decreased from 283.21 mm to 257.13 mm in the years 1987 and 2017. Also, surface runoff was increased from 737.67 mm to 760.49 mm, while groundwater was decreased from 266.32 mm to 257.13 mm in the year 2002 and 2017. Land use/land cover change over the last 30 years also affect the soil and soil properties in the study area significantly. For example, forest land declined and agricultural land is increased. is dynamism would maximize soil erosion, which leads to the loss of important soil nutrients. erefore, different soil and water conservation measures are required to reverse the challenges.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.