Analysis of the impacts of land use land cover change on streamflow and surface water availability in Awash Basin, Ethiopia

Abstract

The current study analyzed the impact of Land Use Land Cover (LULC) change on streamflow and surface water availability in the Akaki catchment of the Awash Basin, Ethiopia. Soil and Water Assessment Tool (SWAT) and Climatic Water Balance (CWB) models were used. The SWAT model calibrated and validated the daily streamflow. The results indicated that from 1993 to 2016, built-up and barren land areas increased by 5.3% and 3.4%, respectively. The SWAT model over-simulated the peak flows and best simulated the medium and low flows. Based on the calibrated daily streamflow, the runoff was 236.01 mm and 272.59 mm under the 1993 and 2016 LULC scenarios. The total water yield of the catchment was 366.7 mm and 382.01 mm for the 1993 and 2016 LULC scenarios, respectively. The seasonal CWB of the catchment depicted decreasing trend except for the Tsedey (SON) season. Based on the current study’s findings, since LULC change enhanced surface runoff and total water yield in the catchment, proactive flood management is recommended to reduce the impact of the flood hazard on life and property.

Keywords:

• LULC change
• water balance
• surface water availability
• SWAT
• calibration
• validation

1. Introduction

The Land Use Land Cover (LULC) change caused by natural processes and anthropogenic alterations heavily influences the global hydrologic system (Kumar et al. 2018; Aragaw et al. 2021). These global hydrological processes can determine water availability in a river basin (Yang et al. 2021). The change affects soil and water interaction (Gashaw et al. 2018; Caballero et al. 2022). The impact can also affect surface and atmospheric systems, which are the main sources of surface water availability (Chaemiso et al. 2021; Senbore and Oke 2021). Hence, the LULC change’s impact on the hydrological processes at a river basin scale is still a timely topic of study worldwide (Muhammed et al. 2021).

The impacts of LULC change on river basins are difficult to manage in developing countries due to the ever-increasing population growth (Chaemiso et al. 2021; Daba and You 2022). The reasons for this are the high rate of LULC change due to the rural-urban migration, increasing demand for agricultural land, and a high rate of unsustainable human-nature interactions (Acebes et al. 2021; Degefu et al. 2021). The LULC change caused fluctuations in the amount, duration, and pattern of streamflow and hence water availability in the river basins in many regions of Africa (Chanapathi and Thatikonda 2020; Measho et al. 2020).

In Ethiopia, studies showed that LULC change affected streamflow and hydrological cycle processes in different basins, sub-basins, and catchments (Dibaba et al. 2020a; Affessa et al. 2022). For instance, a study by Hassen (2022) in Baro Basin indicated that converting forests, shrubs, and grasslands into agricultural lands affects the river’s streamflow and evapotranspiration. Leta et al. (2021) extrapolated that for the period 2019–2050, surface runoff increased by 4.23 percent during the short rain season and 2.0 percent during the dry season due to LULC alterations in the Nashe watershed of the Blue Nile Basin. Cultivated land enhances streamflow throughout the major rainy season and decreases evapotranspiration, but there is unpredictability during the short rainy and dry seasons (Adnan and Atkinson 2011; Hyandye et al. 2018). In Northwestern Ethiopia, according to Dibaba et al. (2020b), agricultural land expanded by 16 percent over the last 30 years, while forest land areas decreased by 12 percent. This indicated that significant changes in LULC change mean that the catchment’s water balance components are further disrupted since infiltration is reduced and surface runoff is increased (Chilagane et al. 2021; Kiprotich et al. 2021).

Awash is a highly utilized river in Ethiopia (Gedefaw et al. 2018). The sugar factories rely on the river’s water for their sugarcane plantation, and small-scale irrigation schemes that use the river’s water are common in the basin (Wendimu et al. 2016). Public and private investors own large-scale agricultural investments (Tufa et al. 2018). Moreover, due to the continuous population growth, agricultural land expansion, and urbanization, the Awash basin is known for its high rate of LULC change (Shawul and Chakma 2019; Tadese et al. 2020a; Damtew et al. 2022). These LULC changes increase the burden on the water in the basin, and the net impact is the fluctuation in water availability.

Global scale studies have been conducted on LULC change and its impact on streamflow. For example, Shahid et al. (2018) conducted a study on the impacts of climate change and human activities on streamflow using the ABCD hydrological model. The model was employed to examine how land use and climate change affected runoff generation at the basin level in Pakistan. Cong et al. (2017) also examined runoff record step change points and hydrological variable trends, attributing the changes to environmental variances in China. Previous studies in the Awash basin and elsewhere in Ethiopia identified the impacts of LULC change on surface runoff (Woldesenbet et al. 2017; Gashaw et al. 2018; Birhanu et al. 2019; Dinka and Klik 2019; Bulti and Abebe 2020; Dibaba et al. 2020a). These studies focused on analyzing the impact of climate change and land use change on runoff and runoff changes using diverse methodologies. Studies also tried to assess the impacts of LULC change on surface water availability (Kumar et al. 2018; Tadese et al. 2020b; Chaemiso et al. 2021). These studies contributed to the current understanding of the impact of LULC changes on surface runoff and the water availability in the watershed. However, they are presented in a disentangled way, and the exact relationship between the impact of LULC change on streamflow and hence the surface water availability is missing.

This can be achieved by applying the Soil and Water Assessment Tool (SWAT) to determine the impact of LULC change on surface water and quantify the surface water availability. The analysis of the LULC change on surface water and the subsequent surface water availability provides a complete picture of the LULC dynamics and related water use, irrigation expansion, and watershed management (Gebremicael et al. 2018; Aghsaei et al. 2020).

LULC change substantially impacts the hydrological processes and affects the process’s elements at the basin and sub-basin levels (Dinka and Klik 2019; Karakuş 2019). By affecting important water balance elements like groundwater recharge, interception, infiltration, and evaporation, LULC change modifies the precipitation path and basin’s water availability (Mekonnen et al. 2018; Wang et al. 2020; Nannawo et al. 2021). Hence, to fill the research gaps identified earlier, this study analyzed the impact of LULC change on streamflow using the SWAT model and, therefore, the surface water availability using the CWB model. The study was conducted in the Akaki catchment of the Awash basin, central Ethiopia

2. Materials and methods

2.1. Description of the study area

Akaki catchment is situated in the Awash basin, particularly in the upper part of the basin. The absolute location of the catchment is between 8^°50'01′'N and 9^°13'10′'N latitude and 38^°43'42′'E and 39^°00'26′'E longitude (Figure 1). The total area of the catchment is about 801.61 square kilometers. As shown in Figure 1, there are 15 streamflow outlets in the catchment.

Figure 1. Location map of the study area, stream outlets, stream reaches, and elevation.

The elevation of the Akaki catchment ranges between 2033 to 3217 meters above mean sea level. The topography undulates the catchment’s northern, western, and southwestern parts and creates a plateau. Rolling plains, steep river banks, valleys, hills, and mountains make up the physiographic elements of the area (Tolera and Chung 2021). The southern and southeastern parts of the catchment have gentle morphology and flat land regions (Zeberie 2019).

According to the Ministry of Agriculture (Ministry of Agriculture (MoA) Ethiopia, 1998), Ethiopia’s climate classification, which was based on temperature and moisture regime, Akaki catchment has a humid to subhumid climate in the highlands and a semiarid climate in the lowlands. The average annual temperature in humid and subhumid highlands is between 16 to 17 °C. The semiarid lowlands have a yearly mean temperature of 19 to 20 °C (Tessema et al. 2015). The catchment’s daily maximum temperature range is between 17.1 and 36.3oC, whereas the daily minimum temperature range is 0.6 to 26.1 °C. The precipitation is seasonal in the catchment. The major rainy season lasts from June to September and accounts for over 70% of the annual total precipitation (Tolera and Chung 2021). The daily range of total precipitation in the catchment is between 0 and 102 mm. The average annual precipitation in the catchment ranges from 800 to 1400 mm, depending on the elevation difference (Maru et al. 2021). Large-scale droughts and floods in the catchment are occasionally caused by the inter-annual precipitation variability in the Belg (MAM) and Kiremt (JJA) seasons (Shawul and Chakma 2019; Maru et al. 2022).

2.2. Trend analysis of streamflow, precipitation, and temperature

Before running the SWAT and Climatic Water Balance models, the time series trends of the hydro-meteorological data, such as streamflow, precipitation, and temperature, were conducted. The Mann Kendall Sen’s slope was used to analyze the trends of the main hydro-meteorological datasets. The Mann-Kendall test statistic 'S' is calculated using Equation 1(1) $$S=\mathrm{ }\sum _{i=1}^{n-1}\sum _{j=i+1}^n\mathit{sign}\left(x_j-x_i\right)$$(1) , based on Mann (1945) and Kendall (1948). (1) $$S=\mathrm{ }\sum _{i=1}^{n-1}\sum _{j=i+1}^n\mathit{sign}\left(x_j-x_i\right)$$(1) where $S$ is the Mann-Kendall statistics, $x_i$ is a time series where the trend is done, which is ranked from $i=1,2,\dots n-1$ and $x_j,$ ranked from $j=i+1,2,\dots n.$

Each of the data point $X_i$ is taken as a reference point which is compared with the rest of the data points $X_j$ so that: (2) $$\mathit{sgn}\left(X_j-X_i\right)=\{\begin{array}{l}+1\mathit{\text{ if }}\left(X_j-X_i\right)>0\\ 0\mathit{\text{ if }}\left(X_j-X_i\right)=0\\ -1\mathit{\text{ if }}\left(X_j-X_i\right)<0\end{array}$$(2) where $X_i$ and $X_j$ are the annual values in years $i$ and $j\mathrm{ }\left(j>i\right),$ respectively.

It has been documented that when the number of observations is more than 10, the statistic '$S$' is approximately normally distributed with the mean and $E\left(S\right),$ becomes 0 (Kendall 1948). In this case, the variance statistic is given as (Equation 3(3) $$\mathit{Var}\left(S\right)=\frac{n\left(n-1\right)\left(2n+5\right)-{\sum }_{t=1}^m{t}_1\left(t_1-1\right)\left({2t}_1+5\right)}{18}$$(3) ): (3) $$\mathit{Var}\left(S\right)=\frac{n\left(n-1\right)\left(2n+5\right)-{\sum }_{t=1}^m{t}_1\left(t_1-1\right)\left({2t}_1+5\right)}{18}$$(3) where $n$ is the number of observations and $t_i$ are the ties of the sample time series. The test statistics $Z_c$ is calculated as Equation 4(4) $$Z=\{\begin{array}{l}\frac{S-1}{\sigma }\mathit{\text{ if S}}>0\\ 0\mathit{\text{ if S}}=0\\ \frac{S+1}{\sigma }\mathit{\text{ if S}}<0\end{array}$$(4) : (4) $$Z=\{\begin{array}{l}\frac{S-1}{\sigma }\mathit{\text{ if S}}>0\\ 0\mathit{\text{ if S}}=0\\ \frac{S+1}{\sigma }\mathit{\text{ if S}}<0\end{array}$$(4)

2.3. Model description

Two models were used in the current study. The impact of the LULC change on streamflow was conducted using the SWAT model. The surface water availability was quantified using the Climatic Water Balance model.

2.3.1. Swat model

SWAT estimates major hydrological processes such as percolation, surface flow, infiltration, evapotranspiration (ET), aquifer flows, and shallow aquifer (Mapes and Pricope 2020). It is a popular model for simulating the influence of LULC change on streamflow at the catchment level. It can simulate bulk input data such as LULC change, soil, streamflow, precipitation, temperature, and digital elevation model (Anand et al. 2018). The other quality of the SWAT model is that it is integrated with ArcGIS software as an extension. This makes the model suitable for simulating LULC change and its impact on streamflow (Belihu et al. 2020).

The SWAT model uses the water balance equation to simulate the hydrological cycle presented in Equation 5(5) $${SW}_t=\mathrm{ }{SW}_0+\sum _{i=1}^t\left(R_{\mathit{day}}-\mathrm{ }{Q}_{\mathit{surf}}-\mathrm{ }{E}_a-\mathrm{ }{W}_{\mathit{seep}}-\mathrm{ }{Q}_{gw}\right)$$(5) (Neitsch et al. 2011). (5) $${SW}_t=\mathrm{ }{SW}_0+\sum _{i=1}^t\left(R_{\mathit{day}}-\mathrm{ }{Q}_{\mathit{surf}}-\mathrm{ }{E}_a-\mathrm{ }{W}_{\mathit{seep}}-\mathrm{ }{Q}_{gw}\right)$$(5) where ${SW}_{t }$ is the final soil water content (mm), ${SW}_0$ is the initial soil water content on day $i$ (mm), $R_{\mathit{day}}$ is the amount of precipitation on day $i$ (mm), $Q_{\mathit{surf}}$ is the amount of surface runoff on day $i$ (mm), $E_a$ is the amount of evapotranspiration on day $i$ (mm), $W_{\mathit{seep}}$ is the amount of water entering the vadose zone from the soil profile on day $i$ (mm), and $Q_{gw}$ is the amount of return flow on day $i$ (mm).

This study used the ArcMap extension of ArcSWAT and the computer program SWAT-CUP 12. The ArcSWAT was employed to set up the SWAT project, delineate the watershed, input DEM, LULC, soil, and slope data, integrate weather data, and run the SWAT model in Arc Map. The SWAT-CUP model was utilized to calibrate and validate the SWAT model using streamflow data. This model enables the SWAT model’s calibration, validation, sensitivity, and uncertainty analyses (Khalid et al. 2016).

In calibrating and validating the SWAT model, three performance metrics were applied. These are the R-Square Coefficient (R²), Nash–Sutcliffe Efficiency (NSE), and Percent bias (PBIAS). The R² is a metric that measures the strength of the association between the data and the fitted regression line (Bennour et al. 2022). It ranges between 0 and 1; the closer the R² value to 1, the less error variance is. The NSE compares the relative magnitudes of the residual ‘noise’ and the variance of the data. Its value is between -∞ and 1, where NSE > 0.5 is acceptable (Schuol and Abbaspour 2007). The relative bias, or PBIAS, determines whether the model data are greater or smaller than the observations. Its best value is 0 (Goshime et al. 2019).

2.3.2. Climatic water balance (CWB)

CWB characterizes the water availability in an area using precipitation and PET as factors (Hargreaves and Samani 1982). Before applying the CWB model, it is necessary to calculate the PET for a specific time. Although there are different methods of PET calculations based on the data availability, the Hargreaves and Samani (1982) method is the most convenient one. The method uses the minimum and maximum temperatures and solar radiation to estimate PET using Equation 6(6) $$\mathit{PET}=0.0023\left(T_{\mathit{mean}}+17.8\right)\sqrt{\left(T_{\mathit{max}}-T_{min}\right)R_{a,}}$$(6) . (6) $$\mathit{PET}=0.0023\left(T_{\mathit{mean}}+17.8\right)\sqrt{\left(T_{\mathit{max}}-T_{min}\right)R_{a,}}$$(6) where $\mathit{PET}$ is the potential evapotranspiration (mm/day), $T_{\mathit{mean}}$ is the average temperature (in ^oC), $T_{\max }$ and $T_{\min }$ are the maximum and minimum temperatures (in ^oC), respectively, and $R_a$ is extra-terrestrial radiation (in mm/day).

Then climatic water balance is the net difference between precipitation and potential evapotranspiration in a defined period (Equation 7(7) $$\mathit{CWB}=P-\mathit{PET}$$(7) ). (7) $$\mathit{CWB}=P-\mathit{PET}$$(7) where $\mathit{CWB}$ is climatic water balance (mm), $P$ is precipitation (mm), and $\mathit{PET}$ is potential evapotranspiration (mm/day). The climatic water balance was used to quantify the surface water availability seasonally in the catchment. The model produces the surface water availability in relation to the climate change impacts.

2.3.3. Image processing and classification

After acquiring the Landsat Mapper and Enhanced Thematic Mapper Plus (ETM+) from the United States Geological Survey (USGS) Earth Explorer, image pre-processing was done to increase the quality of the image for classification. Geometrical (spatial georeferencing) and radiometric (image enhancement, noise, and dark object removals) corrections were made to enhance the image. Stacking was done using ENVI 5.3 software to bring the image bands together. Then the corrected image was brought to ArcMap 10.5, masking the image with the Akaki catchment. Using ERDAS Imagine 2015, supervised image classification was performed. Supervised image classification was preferred because we have ground-truthing GPS points to verify the classified land uses on the image. The concept behind supervised classification is that a user can choose sample pixels from an image characteristic of particular classes and then instruct the image processing software to utilize these training sites as references for categorizing all other pixels in the image (Ahmed and Ahmed 2012). The high-resolution image of Google Earth linked with ERDAS Imagine and knowledge about the study area aided the classification. Repeated classifications were done to increase the accuracy using the signature editor tool.

Individual pixels on the images were used as validation units for the image accuracy assessment. User’s, producer’s, overall, and Kappa Coefficient assessments were done as part of the accuracy assessment. User’s accuracy measures commission mistakes that correspond to pixels from other classes that the classifier has categorized as belonging to the class of interest (Rwanga and Ndambuki 2017). The producer’s assessment indicates the number of omission mistakes corresponding to pixels belonging to the class of interest that the classifier missed (Thapa and Murayama 2009). The overall accuracy refers to the percentage of correctly classified samples (Story and Congalton 1986). The Kappa Coefficient quantifies the proportionate reduction in error caused by a classification algorithm (Fahsi et al. 2000).

Kappa Coefficient was calculated using Equation 8(8) $$\text{Kappa Coefficient }\left(\mathrm{T}\right)=\mathrm{ }\frac{\left(TS\times \mathit{TCS}\right)-\sum \left(CT\times RT\right)}{{TS}^2-\sum \left(CT-RT\right)}\times 100$$(8) (Congalton 1991). (8) $$\text{Kappa Coefficient }\left(\mathrm{T}\right)=\mathrm{ }\frac{\left(TS\times \mathit{TCS}\right)-\sum \left(CT\times RT\right)}{{TS}^2-\sum \left(CT-RT\right)}\times 100$$(8) where $TS=\mathit{Total}\mathrm{ }\mathit{Sample},$ $\mathit{TCS}=\mathit{Total}\mathrm{ }\mathit{Column}\mathrm{ }\mathit{Sample},\mathrm{ }CT=\mathit{Column}\mathrm{ }\mathit{Total},\mathrm{ }RT=\mathit{Row}\mathrm{ }\mathit{Total}$

2.3.4. Data types and sources

The current study simulated the SWAT model using the required datasets, including LULC, Digital Elevation Model (DEM), soil, 4 km × 4km gridded meteorological data (precipitation, minimum and maximum temperature, relative humidity, solar radiation, wind speed), and observed streamflow gauged at Akaki station.

The LULC map was developed for two different years. Hence, the LULC was done for 1993 and 2016. The years were selected based on a previous study by Maru et al. (2022). These years are non-drought and drought years, respectively, in the Awash basin. This was done to link the SWAT model outputs with the water availability in wet and dry years. The Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) were input data for the LULC maps of 1993 and 2016, respectively. The data were accessed from the USGS Earth Explorer site with a spatial resolution of 30 m × 30 m. Then, as shown in Figure 2, the LULC maps of 1993 (a) and 2016 (b) were developed to simulate the SWAT model.

Figure 2. The LULC map of 1993 (a) and 2016 (b).

The high-resolution ASTER Global Digital Elevation Model (ASTER-GDEM) was also obtained from the USGS Earth Explorer (Figure 3a). It was used as DEM input data for SWAT and to develop a slope map. The slope map of the study area was developed using the slope tool in ArcGIS using DEM as input data. The resolution of the ASTER-GDEM was 30 m × 30m, with the elevation ranges between 2033 and 3217 m above mean sea level (Figure 3b).

Figure 3. DEM (a), Slope map (b), and Soil type map (c) of Akaki catchment.

The soil data was found from the Food and Agriculture Organization (FAO) of the United Nations digital soil map of the world database and the Ministry of Water and Energy (MoWE). The soil types and codes were manually entered into the SWAT user soil database and their attributes (Figure 3c). There are six soil types in the Akaki catchment (Figure 3c). The Mollic Nitisols (Vp1-3a-283) is the dominant soil type in the catchment, with 58.1% of the catchment’s coverage. The Chromic Cambisols (Lc75-2b-3781) and Eutric Leptosols (Le19-3a-6578), covering 15.6% and 14.1% of the catchment’s area, are the second and third largest soil types, respectively. The other soil types of the catchment are Rhodic Nitisols (Lv2-3b-3535), Eutric Arenosols (Be9-3c-26), and Pellic Vertisols (Ne61-2-3a-5938), which cover the rest area of the catchment.

The meteorological data were obtained from two sources. Daily precipitation, minimum, and temperature (1990–2016) were accessed from Ethiopia’s National Meteorological Agency (NMA). Due to the absence of the data in expected spatio-temporal scales from the NMA, the study area’s daily relative humidity and wind speed were accessed from the mentioned-above USGS site by adding the location of stations. Ethiopia’s MoWE provided the observed daily streamflow data (1990–2016).

The general methodological flow of the study that combines the use of SWAT, SWAT CUP 12 and R software is presented in Figure 4.

Figure 4. The general methodological flow of the study.

3. Results and discussions

3.1. Trend analysis of hydro-meteorological data

The hydro-meteorological parameters are the core input datasets for the current study. Such parameters’ increasing or decreasing trend helps compare the outputs with the SWAT and CWB models. For this reason, annual-based Mann Kendall Sen’s slope trend analysis was done for 27 years (1990–2016). The results of the trend analysis are presented in Table 1. Accordingly, the annual streamflow trend in the Akaki catchment showed a statistically significant increasing trend at a 99% significance level over the study period. A previously conducted study in the Awash basin by Tadese et al. (2019) indicated a statistically significant increasing trend of streamflow based on station data. An increasing streamflow trend was also found in studies done in different parts of Ethiopia (Gebremicael et al. 2017; Gurara et al. 2021; Orke and Li 2021; Malede et al. 2022).

Table 1. Mann Kendall and Sen’s Slope trend analysis for hydro-meteorological data.

Parameters	P-value	Tau (direction)	Sen’s Slope
Streamflow (mm/day)	0.000***	0.266	0.0014
Total Precipitation (mm)	0.420	0.061	0.0000
Maximum Temperature (^oC)	0.000***	0.216	0.0003
Minimum Temperature (^oC)	0.000***	0.004	0.0001

^∗∗∗Significant at p < 0.01

As shown in Table 1, the total precipitation trend was statistically insignificant, with a p-value of 0.420. The precipitation trend was statistically insignificant in most parts of Ethiopia and the Awash basin (Mulugeta et al. 2019; Tadese et al. 2019). However, its seasonal variability was high (Esayas et al. 2018). In some cases, the precipitation appears during an unexpected season (Belay et al. 2017), is delayed from its usually expected season (Kassie et al. 2013), or shifts its appearance between two seasons (Orke and Li 2021). This high rate of variability affected the smallholder rainfed agricultural production and resulted in food insecurity and poverty in most rural areas in Ethiopia (Mekore and Yaekob 2018; Mekonnen et al. 2021).

The 27-year trends also indicated that the maximum and minimum temperatures showed a statistically increasing positive trend at 99% of the significance level. This result indicates that the temperature rises in the study catchment, implying high evapotranspiration rates that could affect the surface water availability. This is because temperature increment and evapotranspiration have a direct positive relationship (Al-Sudani 2019). The increasing precipitation trend in the catchment could result in an increasing trend of streamflow. This is because precipitation in the catchment is the direct source of streamflow. As recent studies indicate (Shawul and Chakma 2019; Daba and You 2020; Emiru et al. 2021), the long-term precipitation analysis in the Awash basin revealed an increasing trend.

3.2. LULC Classification accuracy assessment

For a supervised classification of different year images, the acceptable standard of accuracy results is above 85% overall accuracy level and over 85% Kappa coefficient (Fugara et al. 2009). We took 50 ground truth Global Positioning System (GPS) points for each of the seven land-use types in the Akaki catchment to validate the LULC classification on the 2016 image. The accuracy assessment results indicated a valid supervised image classification for the 2016 image with 96% of the user’s accuracy (calculated by dividing the row total by the total number of valid classifications for a certain class) and 96.38% of the producer’s accuracy (how often does the classified map accurately depict actual ground features). The overall accuracy was 96%, and the Kappa coefficient was 95% (Table 2).

Table 2. Classification Accuracy Assessment Matrix (2016).

	Grassland	Agriculture land	Built-up Area	Forest	Bush & Shrub Land	Barren Land	Water Body	Total User	Us. Acc. %
Grassland	47	3	0	0	0	0	0	50	94
Agriculture land	1	49	0	0	0	0	0	50	98
Built-up Area	0	0	49	0	0	1	0	50	98
Forest	1	0	0	48	1	0	0	50	96
Bush & Shrub Land	0	0	0	0	50	0	0	50	100
Barren Land	0	3	0	0	0	47	0	50	94
Water Body	0	4	0	0	0	0	46	50	92
Total Producer	49	59	49	48	51	48	46	350	96
Pr. Acc. %	95.9	83	100	100	98	97.8	100	96.38
Overall Acc. = 96%
Kappa = 0.95

Us. Acc. = User’s Accuracy; Pr. Acc. = Producer’s Accuracy; Acc. = Accuracy

3.3. LULC Change

Seven LULC classes, including Bush and Shrub Land (BSHL), Forest (FRST), Barren Land (BARR), Grassland (GRSL), Water Body (WATR), Agriculture Land (AGRL), and Built-up Area (BLTU), were classified from the supervised image classification to detect the changes. The classification was performed to detect changes in LULC between 1993 and 2016, with the results summarized in Table 3.

Table 3. LULC change between changes the 1993 and 2016.

S.N.		1993		2016		Gain	Loss
	LULC Class	LULC Area (sq.km)	%	LULC Area (sq.km)	%	%	%
1	Bush and Shrub Land	12.54	1.56	26.86	3.35	1.8	0
2	Forest	22.91	2.86	46.48	5.80	2.9	0
3	Barren Land	77.65	9.69	104.78	13.07	3.4	0
4	Grassland	159.44	19.89	57.82	7.21	0	12.7
5	Water Body	7.05	0.88	5.65	0.71	0	0.2
6	Agriculture Land	379.26	47.31	375.20	46.81	0	0.5
7	Built-up Area	142.75	17.81	184.83	23.06	5.3	0
	Total	801.6	100	801.6	100	13.4	13

The results indicated that bush and shrubland, forest, barren land, and built-up area increased from 1993 to 2016, while grassland, water body, and agricultural land decreased. The built-up area was the highest gain LULC class with an increment of 42.08 km² (5.3%) between 1993 and 2016. This is because of the alarming expansion of Addis Ababa on the surrounding grasslands and agricultural land, which showed a significant decrement of 101.62 km² (12.7%) and 4.06 km² (0.5%), respectively. Addis Ababa has been expanded to the surrounding areas mainly because of condominiums and related housing constructions (Koroso et al. 2020). The increment in the urban area implies a greater tendency for higher surface runoff (Leta et al. 2021; Demissie 2022).

Although its amount is small (0.2%), the LULC change of water bodies exhibited a shrinking trend over 27 years (Table 3). Many previously conducted studies in Ethiopia (Elias et al. 2019; Regasa et al. 2021) and the Awash basin (Tadese et al. 2020a; Tessema et al. 2020) have reported a shrinkage trend of water bodies. This could be due to the transformation of wetlands and swampy areas into bare land and the decrease in river water in the catchment.

On the other hand, the barren land and forest coverages showed an expansion trend in the Akaki catchment. In the Awash basin, as Damtew et al. (2022) reported, the barren land expanded in the basin mainly due to the expansion of fuelwood gathering and overgrazing. Area coverage of the forest land has shown an expansion trend of 5.8% between 1993 and 2016 (Table 3). This could be linked to the country’s recent recovery of forest resources (Betru et al. 2019) due to government and non-governmental organization-supported household and community-level reforestation and afforestation initiatives (Damtew et al. 2022).

The change of LULC in the catchment has many implications for the river’s streamflow and surface water availability. LULC type, soil parameters, and yearly precipitation directly relate to average annual watershed stream flows. For instance, due to the impermeable cover and low infiltration in specific catchment areas, urban land has the largest potential for increased runoff (Chaemiso et al. 2021). The reduced water storage capacity of the expanded barren land and the shrinkage in grassland as of 2016 resulted in increased surface water runoff and lateral flows in the catchment.

3.4. SWAT model results

3.4.1. Hydrological response unit (HRU) definition

In the SWAT simulation, the Akaki catchment was divided into 15 sub-basins. The sub-basins were further subdivided into HRUs based on the dominant land use, soil, and slope within each sub-basin. The HRUs are the results of the overlay between the generated slope map (Figure 3b) and the soil (Figure 3c) and land use maps (Figure 2). The combination of 12% land use, 15% soil, and 5% slope threshold was used to create the HRUs. Based on the overlay, 153 HRUs were created by the SWAT simulation. These HRUs were also used as input to calibrate the streamflow using SWATCUP12. The HRUs are the smallest divisions of the sub-basin, and an individual HRU does not necessarily reflect the characteristics of the sub-basin.

3.4.2. Sensitivity analysis

The first phase in the SWAT CUP simulation process is sensitivity analysis, which is used to find the model input parameters. Sensitivity analysis aims to evaluate the rate of change in a model’s output for changes in watersheds that cause a significant variation in hydrologic sensitivity (Wang et al. 2019). The model’s calibration necessitates identifying critical (sensitive) parameters and precision.

Ten parameters were used in the sensitivity analysis (Table 4). Parameters highly related to streamflow, such as temperature, precipitation, groundwater, land use, land cover, surface characteristics, and soil, were prioritized in the selection.

Table 4. Parameters used in the SWAT model sensitivity analysis.

S.N.	Parameter	Description	Rank
1	r__CN2.mgt	SCS runoff curve number	1
2	v__ALPHA_BF.gw	Baseflow alpha factor	2
3	v__GW_DELAY.gw	Groundwater delay	3
4	v__GW_QMN.gw	Threshold depth of water in the shallow aquifer required for return flow to occur	4
5	r__TLAPS.sub	Temperature lapse rate	5
6	r__OV_N.hru	Manning’s "n" value for overland flow	6
7	r__PLAPS.sub	Precipitation lapse rate	7
8	r__IGRO.mgt	Land cover status code	8
9	r__SOL_Z.sol	Depth from soil surface to bottom of layer	9
10	r__CH_L2.rte	Length of main channel	10

After the sensitivity analysis, the top 4 ranked parameters with better sensitivity were used to calibrate the streamflow (Figure 5). The sensitivity analyses for the parameters were conducted for 4 years in SWAT CUP using SUFI-2 global sensitivity analysis. The parameters’ sensitivity was evaluated using t-stat and p-value. The ratio between the coefficient of parameters and the standard error gives a t-stat. The parameter is sensitive when the standard error is less than the coefficient. During the sensitivity analysis of any parameter, the smaller the p-value, the more sensitive that parameter is to calibrate the model (Abbaspour 2015).

Figure 5. Dotty plot showing the best sensitive parameters in the model calibration (Y-axis = p-value, X-axis = t-stat).

3.4.3. Model calibration

Following adjusting the calibration inputs, observation, extraction, and objective functions in the SWAT Cup, the SWAT model was calibrated for streamflow for the Akaki catchment using the most sensitive parameters for 6 years (1990–1995) using the first 2 years as a warm-up period in Sufi-2 algorithm. Because the Arc SWAT calibration provided an unsatisfactory result, it was necessary to calibrate the results of the Arc SWAT simulations using the SWAT Cup interface. Accordingly, 100 simulations with 10 iterations were performed in the calibration process to get the best calibration outputs. The LULC map of 1993 was used to calibrate the daily streamflow data. Graphical comparison and statistical indices such as Nash Sutcliff Efficiency (NSE), R-Square Coefficient (R²), and Percentage Bias (PBIAS) to the standard deviation of measured data were used to evaluate the performance of the calibrated parameters. The observed and simulated streamflow calibration output coefficients were evaluated using the NSE, R², and PBIAS. The NSE, R^2, and PBIAS for the observed and simulated streamflow for the calibration period were 0.81, 0.83, and −0.54, respectively.

As presented in Figure 6, based on the statistical model evaluator calibration result, the daily observed and simulated streamflow agreed as predicted by the selected parameters. The SWAT model overestimated the daily high streamflow and underestimated the daily low streamflow. In the medium flow cases, the observed and simulated flows fit better. Abebe and Gebremariam (2019), in their SWAT-based streamflow and sediment yield study in the Kesem watershed of the Awash basin, indicated that SWAT overestimated the streamflow for some months, which had a higher record due to the rainy season. The reason is that SWAT streamflow simulation is highly sensitive to the high streamflow record period during the days of the Kiremt (JJA) season (Spruill et al. 2000). This indicates that the SWAT Cup model best simulated medium flow. Precise parameter adjustments and frequent iterations are needed to best simulate high and low flow. The calibration results suggest that using the SWAT model to simulate streamflow in the Akaki catchment is recommended, with a precise selection of sensitive streamflow parameters.

Figure 6. Calibrated daily streamflow hydrograph at Akaki gauging station (1993 LULC).

3.4.4. Model validation

Validation is the process of executing a model with parameters selected during the calibration stage and comparing the simulations to data not utilized in the calibration process (Arnold et al. 2012). SWAT model validation was conducted for 5 years (2012–2016) for the selected daily streamflow parameters. Due to the bulk nature of the daily streamflow data in a relatively larger study area, we preferred to calibrate and validate the SWAT model for limited years. The LULC map of 2016 was used to validate the streamflow. The model validation’s R², NSE, and PBIAS were 0.80, 0.79, and −0.24, respectively. As shown in Figure 7, the observed and simulated daily streamflow of the Akaki catchment depicted a good agreement, though the model overestimated the streamflow during Kiremt (JJA) time. This was due to the high variations in daily streamflow of the catchment during the high precipitation time (June to August).

Figure 7. Validated daily streamflow hydrograph at Akaki gauging station (2016 LULC).

The daily streamflow simulation between 2012 and 2016 also yielded a peak flow. According to the simulation, the peak flow of the observed and simulated daily streamflow in the Akaki catchment was 469.92 mm and 476.93 mm, respectively. The slight increment of the simulated flow compared to the observed agrees with the previously conducted studies in the Awash basin (Abebe and Gebremariam 2019; Tessema et al. 2021; Abdulahi et al. 2022).

3.5. Climatic water balance model results

The seasonal PET of the catchment was computed based on the 27 years (1990–2016) time series precipitation and temperature data. The reason behind the computation of the seasonal PET is that the water balance gives direct meaning when it is calculated, which directly correlates with the crop growing season. Hence, the PET was computed for the four seasons: Tsedey (SON), Bega (DJF), Belg (MAM), and Kiremt (JJA).

As presented in Figure 8, the PET was more than 140 mm/day for most years in the Belg (MAM) season of the Akaki catchment. The highest PET was 176 mm/day observed in the Belg (MAM) season. The Bega (DJF) recorded a relatively higher amount of PET than the Tsedey (SON) and Kiremt (JJA) seasons. Higher inter-seasonal variabilities of PET results were observed in the Kiremt (JJA) and Tsedey (SON) seasons. Evapotranspiration is an element of the hydrological process that can influence the amount and quality of surface water (Yadeta et al. 2020). Although increasing streamflow trends were observed in the Akaki catchment, the higher amount of calculated PET has influenced surface water availability.

Figure 8. Seasonal PET of Akaki catchment (1990–2016).

The Mann-Kendall trend analysis results also indicated a significant (95% confidence level) increase in PET trends for the Belg (MAM) and Kiremt (JJA) seasons. The Bega (DJF) season’s Pet al.so showed a significant (at 90% confidence level) increasing trend during the study period (Table 5). In the Awash basin, the temperature of Belg (MAM) and Kiremt (JJA) is relatively high (Shawul and Chakma 2019) due to the movement of the Intertropical Convergence Zone (ITCZ) to the northern hemisphere (Marriner et al. 2012). Belg (MAM) and Kiremt (JJA) are the main rainy seasons in the basin and Ethiopia when farmers fully engage in agricultural activities, especially in the rainfed farming system (Belihu et al. 2020).

Table 5. Mann Kendall and Sen’s Slope seasonal trend analysis for PET.

Season	P-value	Tau (direction)	Sen’s Slope
Tsedey (SON)	0.045	0.276	0.319
Bega (DJF)	0.050*	0.270	0.292
Belg (MAM)	0.002**	0.413	0.492
Kiremt (JJA)	0.005**	0.384	0.492

^∗∗Significant at p < 0.05; ^∗Significant at p < 0.1

The climatic water balance of the Akaki catchment showed variations between the four seasons. This concerns the precipitation and temperature variations across the seasons. As shown in Figure 8, the climatic water balance of the Bega (DJF) seasons was negative for 27 years, indicating the season’s dryness. Winters are dry in the catchment, so the water balance deficit is high. Positive and negative CWB were observed in the Tsedey (SON) and Belg (MAM) seasons, with the relative positives dominating. Tsedey (SON) is the transitional season from cultivation to harvest of the major crops, while Belg (MAM) is preparing the land for rainfed agriculture in most parts of the catchment.

In most parts of the Akaki catchment, Kiremt (JJA) is the main rain season. Hence, as shown in Figure 9, the CWB is high and positive except for three years compared to the other seasons. The highest CWB, more than 700 mm, was also registered this season. The water availability during this season is crucial for rainfed farming in the catchment (Gummadi et al. 2018). Most crops are planted during this season, and the absence of water during the Bega (DJF) is an indicator of drought in the coming harvesting seasons (Temam et al. 2019).

Figure 9. Seasonal CWB of Akaki catchment (1990–2016).

The recent years (2014–2016), CWB showed a negative value for all four seasons (Figure 9). This is due to the increment in PET and the reduction in rainfall. Maru et al. (2022) also indicated that recent years had been characterized by meteorological drought in the Awash basin. Dryness is being expanded in previously wet areas, and drought occurred in the basin’s previously non-drought agroecological zones.

The seasonal CWB Mann Kendall t-test results supported the above results. As presented in Table 6, except for the Tsedey (SON) season, statistically significant decreasing trends of CWB were observed in the three seasons. The Bega (DJF) CWB trend was significant at 95%, while the Belg (MAM) and Kiremt (JJA) trends were significant at 90%. The higher PET trends in the Bega (DJF) and Belg (MAM) seasons might have also contributed to the relatively higher decreasing trends in CWB. The results indicate that the seasonal CWB of the Akaki catchment showed a significant decrease in the 27 years. In relation to these results, the water availability in the catchment decreased from 1990 to 2016, implying the meteorological drought dominating the recent years.

Table 6. Mann Kendall and Sen’s Slope seasonal trend analysis for CWB.

Season	P-value	Tau (direction)	Sen’s Slope
Tsedey (SON)	0.966	0.008	0.012
Bega (DJF)	0.009**	−0.351	−2.772
Belg (MAM)	0.087*	−0.236	−4.650
Kiremt (JJA)	0.072*	−0.247	−6.598

^∗∗Significant at p < 0.05; ^∗Significant at p < 0.1

3.6. Implications of LULC change on surface water availability

One of the capabilities of the SWAT model is it can quantify the surface water using different parameters. Using LULC 1993 and LULC 2016 scenarios, the outputs of the SWAT model for the Akaki catchment were compared based on some parameters. These parameters can estimate the water availability in the catchment. The average surface runoff increased from 236.01 mm to 272.39 mm between 1993 and 2016 (Table 7). Between these years, the two major LULC classes were built-up areas and barren land. In the study period, the built-up area and barren land increased by 5.3% and 3.8%, respectively (Table 3). Addis Ababa is located in the study area, and its expansion at the expense of the surrounding rural areas is ever-increasing. The expansion of both land uses contributed positively to the lateral movement of water on the surface, increasing the runoff. According to Eshtawi et al. (2015), built-up and urban expansions yield a rapid lateral surface water flow and increase surface runoff. The increase in surface runoff was attributed mainly to the LULC change because other surface runoff providers, like total aquifer recharge, lateral soil flow, and percolation out of the soil, decreased in the catchment during the study period (Table 7).

Table 7. Water availability parameters with the 1993 and 2016 LULC scenarios in the Akaki catchment.

S.N.	Parameters	LULC Scenario 1993	LULC Scenario 2016
1	Surface Runoff (mm)	236.01	272.59
2	Lateral Soil Flow (mm)	17.06	13.8
3	Groundwater Shallow Aquifer Flow (mm)	107.09	90.01
4	Percolation Out of Soil (mm)	129.93	111.45
5	Evapotranspiration (ET) (mm/s)	428.6	414.2
6	Total Aquifer Recharge (mm)	129.95	111.47
7	Total Water Yield (mm)	366.7	382.01

The total water yield of the catchment also increased during the study period. The difference between the 1993 and 2016 LULC scenarios in total water yield is 15.31 mm (Table 7). The combined effect of the increasing surface runoff and total water yield in the catchment can be associated with an extreme climate event – flooding. In recent years, flooding has become one of the extreme climate events in the Akaki catchment (Zeberie 2019). For instance, Addis Ababa has expanded from 80.1 km² to 287.9 km² between 1984 and 2020 (Beshir and Song 2021), implying an increasing flood risk in the city. According to this study, the long-term precipitation trend of the city was not significant, and the increased surface runoff was the major contributor to the increasing flooding event.

4. Conclusions

The current study analyzed the impacts of LULC change on the streamflow of the Akaki catchment and the availability of surface water in the catchment using SWAT and Climatic Water Balance models. The study’s results indicated that the 27 years of Mann Kendall and Sen’s Slope trend analysis indicated that daily streamflow and maximum and minimum temperature increased in the catchment. On the other hand, between 1993 and 2016, the LULC of the catchment showed different changes. The major LULC changes during the study period were the increment of built-up area and barren land; and the decrement of grasslands. SWAT slightly over-simulated daily peak flows and best simulated the daily medium and low flows in the catchment. The over-simulated of the peak flows is related to the high variation in daily streamflow during dry and wet days. The results indicated the shift of the rainy season in the catchment from Belg (MAM) to Tsedey (SON).

One of the impacts of the LULC change on streamflow in the Akaki catchment is that surface water increased from 1993 to 2016. The increase in the surface runoff and total water yield could also lead to a climate extreme event – flooding in the catchment. The study also shows that surface water is enhanced in the catchment. There could be two options to manage the excess surface water: utilizing the water for productive purpose and draining it to its final destination properly. In the case of using the water for productive purposes, surface water harvesting and utilizing it through urban agriculture in terms of hydroponics and hydroponics could have a double benefit in utilizing the extra surface water before it causes the problem aiding urban food production. Enhancing the urban areas’ drainage and water disposal system to safely dispose of the generated surface flow to the appropriate area could be another consideration. Unless the surface water is utilized for productive purposes, it could cause flooding, especially in urban areas. In this case, proactive flood management is important to reduce the impacts of flooding on life and property.

Abbreviations
CWB	=	Climatic water balance
DJF	=	December January February
ETM	=	Enhanced thematic mapper
GPS	=	Global positioning system
HRU	=	Hydrological response unit
JJA	=	June July August
LULC	=	Land use land cover
MAM	=	March April May
MoWE	=	Ministry of water and energy
NMA	=	National Meteorology Agency
NSE	=	Nash–Sutcliffe Efficiency
SON	=	September October November
SWAT	=	Soil and Water Assessment Tool
USGS	=	United States Geological Survey

Acknowledgments

The authors are grateful and appreciate the provision of the gridded temperature and precipitation data from the National Meteorological Agency (NMA) of Ethiopia and its staff. We also would like to thank Water Land and Ecosystem (WLE), Addis Ababa University, and Wolaita Sodo University for their financial support for the corresponding author to conduct this study.

Data availability statement

The datasets supporting this study are available from the corresponding author upon reasonable request.

Disclosure statement

On behalf of all authors, the corresponding author declares that there is no conflict of interest.